1 /*
    2  * Copyright (c) 1997, 2026, 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 "asm/assembler.hpp"
   26 #include "asm/assembler.inline.hpp"
   27 #include "code/aotCodeCache.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 "interpreter/interpreterRuntime.hpp"
   39 #include "jvm.h"
   40 #include "memory/resourceArea.hpp"
   41 #include "memory/universe.hpp"
   42 #include "oops/accessDecorators.hpp"
   43 #include "oops/compressedKlass.inline.hpp"
   44 #include "oops/compressedOops.inline.hpp"
   45 #include "oops/klass.inline.hpp"

   46 #include "prims/methodHandles.hpp"

   47 #include "runtime/continuation.hpp"
   48 #include "runtime/interfaceSupport.inline.hpp"
   49 #include "runtime/javaThread.hpp"
   50 #include "runtime/jniHandles.hpp"
   51 #include "runtime/objectMonitor.hpp"
   52 #include "runtime/os.hpp"
   53 #include "runtime/safepoint.hpp"
   54 #include "runtime/safepointMechanism.hpp"
   55 #include "runtime/sharedRuntime.hpp"

   56 #include "runtime/stubRoutines.hpp"
   57 #include "utilities/checkedCast.hpp"
   58 #include "utilities/globalDefinitions.hpp"
   59 #include "utilities/macros.hpp"




   60 
   61 #ifdef PRODUCT
   62 #define BLOCK_COMMENT(str) /* nothing */
   63 #define STOP(error) stop(error)
   64 #else
   65 #define BLOCK_COMMENT(str) block_comment(str)
   66 #define STOP(error) block_comment(error); stop(error)
   67 #endif
   68 
   69 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   70 
   71 #ifdef ASSERT
   72 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   73 #endif
   74 
   75 static const Assembler::Condition reverse[] = {
   76     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   77     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   78     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   79     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
   80     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
   81     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
   82     Assembler::above          /* belowEqual    = 0x6 */ ,
   83     Assembler::belowEqual     /* above         = 0x7 */ ,
   84     Assembler::positive       /* negative      = 0x8 */ ,
   85     Assembler::negative       /* positive      = 0x9 */ ,
   86     Assembler::noParity       /* parity        = 0xa */ ,
   87     Assembler::parity         /* noParity      = 0xb */ ,
   88     Assembler::greaterEqual   /* less          = 0xc */ ,
   89     Assembler::less           /* greaterEqual  = 0xd */ ,
   90     Assembler::greater        /* lessEqual     = 0xe */ ,
   91     Assembler::lessEqual      /* greater       = 0xf, */
   92 
   93 };
   94 
   95 
   96 // Implementation of MacroAssembler
   97 
   98 Address MacroAssembler::as_Address(AddressLiteral adr) {
   99   // amd64 always does this as a pc-rel
  100   // we can be absolute or disp based on the instruction type
  101   // jmp/call are displacements others are absolute
  102   assert(!adr.is_lval(), "must be rval");
  103   assert(reachable(adr), "must be");
  104   return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
  105 
  106 }
  107 
  108 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  109   AddressLiteral base = adr.base();
  110   lea(rscratch, base);
  111   Address index = adr.index();
  112   assert(index._disp == 0, "must not have disp"); // maybe it can?
  113   Address array(rscratch, index._index, index._scale, index._disp);
  114   return array;
  115 }
  116 
  117 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
  118   Label L, E;
  119 
  120 #ifdef _WIN64
  121   // Windows always allocates space for it's register args
  122   assert(num_args <= 4, "only register arguments supported");
  123   subq(rsp,  frame::arg_reg_save_area_bytes);
  124 #endif
  125 
  126   // Align stack if necessary
  127   testl(rsp, 15);
  128   jcc(Assembler::zero, L);
  129 
  130   subq(rsp, 8);
  131   call(RuntimeAddress(entry_point));
  132   addq(rsp, 8);
  133   jmp(E);
  134 
  135   bind(L);
  136   call(RuntimeAddress(entry_point));
  137 
  138   bind(E);
  139 
  140 #ifdef _WIN64
  141   // restore stack pointer
  142   addq(rsp, frame::arg_reg_save_area_bytes);
  143 #endif
  144 }
  145 
  146 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
  147   assert(!src2.is_lval(), "should use cmpptr");
  148   assert(rscratch != noreg || always_reachable(src2), "missing");
  149 
  150   if (reachable(src2)) {
  151     cmpq(src1, as_Address(src2));
  152   } else {
  153     lea(rscratch, src2);
  154     Assembler::cmpq(src1, Address(rscratch, 0));
  155   }
  156 }
  157 
  158 int MacroAssembler::corrected_idivq(Register reg) {
  159   // Full implementation of Java ldiv and lrem; checks for special
  160   // case as described in JVM spec., p.243 & p.271.  The function
  161   // returns the (pc) offset of the idivl instruction - may be needed
  162   // for implicit exceptions.
  163   //
  164   //         normal case                           special case
  165   //
  166   // input : rax: dividend                         min_long
  167   //         reg: divisor   (may not be eax/edx)   -1
  168   //
  169   // output: rax: quotient  (= rax idiv reg)       min_long
  170   //         rdx: remainder (= rax irem reg)       0
  171   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
  172   static const int64_t min_long = 0x8000000000000000;
  173   Label normal_case, special_case;
  174 
  175   // check for special case
  176   cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
  177   jcc(Assembler::notEqual, normal_case);
  178   xorl(rdx, rdx); // prepare rdx for possible special case (where
  179                   // remainder = 0)
  180   cmpq(reg, -1);
  181   jcc(Assembler::equal, special_case);
  182 
  183   // handle normal case
  184   bind(normal_case);
  185   cdqq();
  186   int idivq_offset = offset();
  187   idivq(reg);
  188 
  189   // normal and special case exit
  190   bind(special_case);
  191 
  192   return idivq_offset;
  193 }
  194 
  195 void MacroAssembler::decrementq(Register reg, int value) {
  196   if (value == min_jint) { subq(reg, value); return; }
  197   if (value <  0) { incrementq(reg, -value); return; }
  198   if (value == 0) {                        ; return; }
  199   if (value == 1 && UseIncDec) { decq(reg) ; return; }
  200   /* else */      { subq(reg, value)       ; return; }
  201 }
  202 
  203 void MacroAssembler::decrementq(Address dst, int value) {
  204   if (value == min_jint) { subq(dst, value); return; }
  205   if (value <  0) { incrementq(dst, -value); return; }
  206   if (value == 0) {                        ; return; }
  207   if (value == 1 && UseIncDec) { decq(dst) ; return; }
  208   /* else */      { subq(dst, value)       ; return; }
  209 }
  210 
  211 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
  212   assert(rscratch != noreg || always_reachable(dst), "missing");
  213 
  214   if (reachable(dst)) {
  215     incrementq(as_Address(dst));
  216   } else {
  217     lea(rscratch, dst);
  218     incrementq(Address(rscratch, 0));
  219   }
  220 }
  221 
  222 void MacroAssembler::incrementq(Register reg, int value) {
  223   if (value == min_jint) { addq(reg, value); return; }
  224   if (value <  0) { decrementq(reg, -value); return; }
  225   if (value == 0) {                        ; return; }
  226   if (value == 1 && UseIncDec) { incq(reg) ; return; }
  227   /* else */      { addq(reg, value)       ; return; }
  228 }
  229 
  230 void MacroAssembler::incrementq(Address dst, int value) {
  231   if (value == min_jint) { addq(dst, value); return; }
  232   if (value <  0) { decrementq(dst, -value); return; }
  233   if (value == 0) {                        ; return; }
  234   if (value == 1 && UseIncDec) { incq(dst) ; return; }
  235   /* else */      { addq(dst, value)       ; return; }
  236 }
  237 
  238 // 32bit can do a case table jump in one instruction but we no longer allow the base
  239 // to be installed in the Address class
  240 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  241   lea(rscratch, entry.base());
  242   Address dispatch = entry.index();
  243   assert(dispatch._base == noreg, "must be");
  244   dispatch._base = rscratch;
  245   jmp(dispatch);
  246 }
  247 
  248 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  249   ShouldNotReachHere(); // 64bit doesn't use two regs
  250   cmpq(x_lo, y_lo);
  251 }
  252 
  253 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  254   mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  255 }
  256 
  257 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  258   lea(rscratch, adr);
  259   movptr(dst, rscratch);
  260 }
  261 
  262 void MacroAssembler::leave() {
  263   // %%% is this really better? Why not on 32bit too?
  264   emit_int8((unsigned char)0xC9); // LEAVE
  265 }
  266 
  267 void MacroAssembler::lneg(Register hi, Register lo) {
  268   ShouldNotReachHere(); // 64bit doesn't use two regs
  269   negq(lo);
  270 }
  271 
  272 void MacroAssembler::movoop(Register dst, jobject obj) {
  273   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  274 }
  275 
  276 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  277   mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  278   movq(dst, rscratch);
  279 }
  280 
  281 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  282   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  283 }
  284 
  285 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  286   mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  287   movq(dst, rscratch);
  288 }
  289 
  290 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  291   if (src.is_lval()) {
  292     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  293   } else {
  294     if (reachable(src)) {
  295       movq(dst, as_Address(src));
  296     } else {
  297       lea(dst, src);
  298       movq(dst, Address(dst, 0));
  299     }
  300   }
  301 }
  302 
  303 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  304   movq(as_Address(dst, rscratch), src);
  305 }
  306 
  307 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  308   movq(dst, as_Address(src, dst /*rscratch*/));
  309 }
  310 
  311 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
  312 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  313   if (is_simm32(src)) {
  314     movptr(dst, checked_cast<int32_t>(src));
  315   } else {
  316     mov64(rscratch, src);
  317     movq(dst, rscratch);
  318   }
  319 }
  320 
  321 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  322   movoop(rscratch, obj);
  323   push(rscratch);
  324 }
  325 
  326 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  327   mov_metadata(rscratch, obj);
  328   push(rscratch);
  329 }
  330 
  331 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  332   lea(rscratch, src);
  333   if (src.is_lval()) {
  334     push(rscratch);
  335   } else {
  336     pushq(Address(rscratch, 0));
  337   }
  338 }
  339 
  340 static void pass_arg0(MacroAssembler* masm, Register arg) {
  341   if (c_rarg0 != arg ) {
  342     masm->mov(c_rarg0, arg);
  343   }
  344 }
  345 
  346 static void pass_arg1(MacroAssembler* masm, Register arg) {
  347   if (c_rarg1 != arg ) {
  348     masm->mov(c_rarg1, arg);
  349   }
  350 }
  351 
  352 static void pass_arg2(MacroAssembler* masm, Register arg) {
  353   if (c_rarg2 != arg ) {
  354     masm->mov(c_rarg2, arg);
  355   }
  356 }
  357 
  358 static void pass_arg3(MacroAssembler* masm, Register arg) {
  359   if (c_rarg3 != arg ) {
  360     masm->mov(c_rarg3, arg);
  361   }
  362 }
  363 
  364 void MacroAssembler::stop(const char* msg) {
  365   if (ShowMessageBoxOnError) {
  366     address rip = pc();
  367     pusha(); // get regs on stack
  368     lea(c_rarg1, InternalAddress(rip));
  369     movq(c_rarg2, rsp); // pass pointer to regs array
  370   }
  371   // Skip AOT caching C strings in scratch buffer.
  372   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  373   lea(c_rarg0, ExternalAddress((address) str));
  374   andq(rsp, -16); // align stack as required by ABI
  375   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
  376   hlt();
  377 }
  378 
  379 void MacroAssembler::warn(const char* msg) {
  380   push(rbp);
  381   movq(rbp, rsp);
  382   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  383   push_CPU_state();   // keeps alignment at 16 bytes
  384 
  385 #ifdef _WIN64
  386   // Windows always allocates space for its register args
  387   subq(rsp,  frame::arg_reg_save_area_bytes);
  388 #endif
  389   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  390   lea(c_rarg0, ExternalAddress((address) str));
  391   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  392 
  393 #ifdef _WIN64
  394   // restore stack pointer
  395   addq(rsp, frame::arg_reg_save_area_bytes);
  396 #endif
  397   pop_CPU_state();
  398   mov(rsp, rbp);
  399   pop(rbp);
  400 }
  401 
  402 void MacroAssembler::print_state() {
  403   address rip = pc();
  404   pusha();            // get regs on stack
  405   push(rbp);
  406   movq(rbp, rsp);
  407   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  408   push_CPU_state();   // keeps alignment at 16 bytes
  409 
  410   lea(c_rarg0, InternalAddress(rip));
  411   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  412   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  413 
  414   pop_CPU_state();
  415   mov(rsp, rbp);
  416   pop(rbp);
  417   popa();
  418 }
  419 
  420 #ifndef PRODUCT
  421 extern "C" void findpc(intptr_t x);
  422 #endif
  423 
  424 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  425   // In order to get locks to work, we need to fake a in_VM state
  426   if (ShowMessageBoxOnError) {
  427     JavaThread* thread = JavaThread::current();
  428     JavaThreadState saved_state = thread->thread_state();
  429     thread->set_thread_state(_thread_in_vm);
  430 #ifndef PRODUCT
  431     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  432       ttyLocker ttyl;
  433       BytecodeCounter::print();
  434     }
  435 #endif
  436     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  437     // XXX correct this offset for amd64
  438     // This is the value of eip which points to where verify_oop will return.
  439     if (os::message_box(msg, "Execution stopped, print registers?")) {
  440       print_state64(pc, regs);
  441       BREAKPOINT;
  442     }
  443   }
  444   fatal("DEBUG MESSAGE: %s", msg);
  445 }
  446 
  447 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  448   ttyLocker ttyl;
  449   DebuggingContext debugging{};
  450   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  451 #ifndef PRODUCT
  452   tty->cr();
  453   findpc(pc);
  454   tty->cr();
  455 #endif
  456 #define PRINT_REG(rax, value) \
  457   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  458   PRINT_REG(rax, regs[15]);
  459   PRINT_REG(rbx, regs[12]);
  460   PRINT_REG(rcx, regs[14]);
  461   PRINT_REG(rdx, regs[13]);
  462   PRINT_REG(rdi, regs[8]);
  463   PRINT_REG(rsi, regs[9]);
  464   PRINT_REG(rbp, regs[10]);
  465   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  466   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  467   PRINT_REG(r8 , regs[7]);
  468   PRINT_REG(r9 , regs[6]);
  469   PRINT_REG(r10, regs[5]);
  470   PRINT_REG(r11, regs[4]);
  471   PRINT_REG(r12, regs[3]);
  472   PRINT_REG(r13, regs[2]);
  473   PRINT_REG(r14, regs[1]);
  474   PRINT_REG(r15, regs[0]);
  475 #undef PRINT_REG
  476   // Print some words near the top of the stack.
  477   int64_t* rsp = &regs[16];
  478   int64_t* dump_sp = rsp;
  479   for (int col1 = 0; col1 < 8; col1++) {
  480     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  481     os::print_location(tty, *dump_sp++);
  482   }
  483   for (int row = 0; row < 25; row++) {
  484     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  485     for (int col = 0; col < 4; col++) {
  486       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  487     }
  488     tty->cr();
  489   }
  490   // Print some instructions around pc:
  491   Disassembler::decode((address)pc-64, (address)pc);
  492   tty->print_cr("--------");
  493   Disassembler::decode((address)pc, (address)pc+32);
  494 }
  495 
  496 // The java_calling_convention describes stack locations as ideal slots on
  497 // a frame with no abi restrictions. Since we must observe abi restrictions
  498 // (like the placement of the register window) the slots must be biased by
  499 // the following value.
  500 static int reg2offset_in(VMReg r) {
  501   // Account for saved rbp and return address
  502   // This should really be in_preserve_stack_slots
  503   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  504 }
  505 
  506 static int reg2offset_out(VMReg r) {
  507   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  508 }
  509 
  510 // A long move
  511 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  512 
  513   // The calling conventions assures us that each VMregpair is either
  514   // all really one physical register or adjacent stack slots.
  515 
  516   if (src.is_single_phys_reg() ) {
  517     if (dst.is_single_phys_reg()) {
  518       if (dst.first() != src.first()) {
  519         mov(dst.first()->as_Register(), src.first()->as_Register());
  520       }
  521     } else {
  522       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  523              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  524       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  525     }
  526   } else if (dst.is_single_phys_reg()) {
  527     assert(src.is_single_reg(),  "not a stack pair");
  528     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  529   } else {
  530     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  531     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  532     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  533   }
  534 }
  535 
  536 // A double move
  537 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  538 
  539   // The calling conventions assures us that each VMregpair is either
  540   // all really one physical register or adjacent stack slots.
  541 
  542   if (src.is_single_phys_reg() ) {
  543     if (dst.is_single_phys_reg()) {
  544       // In theory these overlap but the ordering is such that this is likely a nop
  545       if ( src.first() != dst.first()) {
  546         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  547       }
  548     } else {
  549       assert(dst.is_single_reg(), "not a stack pair");
  550       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  551     }
  552   } else if (dst.is_single_phys_reg()) {
  553     assert(src.is_single_reg(),  "not a stack pair");
  554     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  555   } else {
  556     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  557     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  558     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  559   }
  560 }
  561 
  562 
  563 // A float arg may have to do float reg int reg conversion
  564 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  565   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  566 
  567   // The calling conventions assures us that each VMregpair is either
  568   // all really one physical register or adjacent stack slots.
  569 
  570   if (src.first()->is_stack()) {
  571     if (dst.first()->is_stack()) {
  572       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  573       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  574     } else {
  575       // stack to reg
  576       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  577       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  578     }
  579   } else if (dst.first()->is_stack()) {
  580     // reg to stack
  581     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  582     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  583   } else {
  584     // reg to reg
  585     // In theory these overlap but the ordering is such that this is likely a nop
  586     if ( src.first() != dst.first()) {
  587       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  588     }
  589   }
  590 }
  591 
  592 // On 64 bit we will store integer like items to the stack as
  593 // 64 bits items (x86_32/64 abi) even though java would only store
  594 // 32bits for a parameter. On 32bit it will simply be 32 bits
  595 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  596 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  597   if (src.first()->is_stack()) {
  598     if (dst.first()->is_stack()) {
  599       // stack to stack
  600       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  601       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  602     } else {
  603       // stack to reg
  604       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  605     }
  606   } else if (dst.first()->is_stack()) {
  607     // reg to stack
  608     // Do we really have to sign extend???
  609     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  610     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  611   } else {
  612     // Do we really have to sign extend???
  613     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
  614     if (dst.first() != src.first()) {
  615       movq(dst.first()->as_Register(), src.first()->as_Register());
  616     }
  617   }
  618 }
  619 
  620 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
  621   if (src.first()->is_stack()) {
  622     if (dst.first()->is_stack()) {
  623       // stack to stack
  624       movq(rax, Address(rbp, reg2offset_in(src.first())));
  625       movq(Address(rsp, reg2offset_out(dst.first())), rax);
  626     } else {
  627       // stack to reg
  628       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
  629     }
  630   } else if (dst.first()->is_stack()) {
  631     // reg to stack
  632     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
  633   } else {
  634     if (dst.first() != src.first()) {
  635       movq(dst.first()->as_Register(), src.first()->as_Register());
  636     }
  637   }
  638 }
  639 
  640 // An oop arg. Must pass a handle not the oop itself
  641 void MacroAssembler::object_move(OopMap* map,
  642                         int oop_handle_offset,
  643                         int framesize_in_slots,
  644                         VMRegPair src,
  645                         VMRegPair dst,
  646                         bool is_receiver,
  647                         int* receiver_offset) {
  648 
  649   // must pass a handle. First figure out the location we use as a handle
  650 
  651   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
  652 
  653   // See if oop is null if it is we need no handle
  654 
  655   if (src.first()->is_stack()) {
  656 
  657     // Oop is already on the stack as an argument
  658     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
  659     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
  660     if (is_receiver) {
  661       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
  662     }
  663 
  664     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
  665     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
  666     // conditionally move a null
  667     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
  668   } else {
  669 
  670     // Oop is in a register we must store it to the space we reserve
  671     // on the stack for oop_handles and pass a handle if oop is non-null
  672 
  673     const Register rOop = src.first()->as_Register();
  674     int oop_slot;
  675     if (rOop == j_rarg0)
  676       oop_slot = 0;
  677     else if (rOop == j_rarg1)
  678       oop_slot = 1;
  679     else if (rOop == j_rarg2)
  680       oop_slot = 2;
  681     else if (rOop == j_rarg3)
  682       oop_slot = 3;
  683     else if (rOop == j_rarg4)
  684       oop_slot = 4;
  685     else {
  686       assert(rOop == j_rarg5, "wrong register");
  687       oop_slot = 5;
  688     }
  689 
  690     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
  691     int offset = oop_slot*VMRegImpl::stack_slot_size;
  692 
  693     map->set_oop(VMRegImpl::stack2reg(oop_slot));
  694     // Store oop in handle area, may be null
  695     movptr(Address(rsp, offset), rOop);
  696     if (is_receiver) {
  697       *receiver_offset = offset;
  698     }
  699 
  700     cmpptr(rOop, NULL_WORD);
  701     lea(rHandle, Address(rsp, offset));
  702     // conditionally move a null from the handle area where it was just stored
  703     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
  704   }
  705 
  706   // If arg is on the stack then place it otherwise it is already in correct reg.
  707   if (dst.first()->is_stack()) {
  708     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
  709   }
  710 }
  711 
  712 void MacroAssembler::addptr(Register dst, int32_t imm32) {
  713   addq(dst, imm32);
  714 }
  715 
  716 void MacroAssembler::addptr(Register dst, Register src) {
  717   addq(dst, src);
  718 }
  719 
  720 void MacroAssembler::addptr(Address dst, Register src) {
  721   addq(dst, src);
  722 }
  723 
  724 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  725   assert(rscratch != noreg || always_reachable(src), "missing");
  726 
  727   if (reachable(src)) {
  728     Assembler::addsd(dst, as_Address(src));
  729   } else {
  730     lea(rscratch, src);
  731     Assembler::addsd(dst, Address(rscratch, 0));
  732   }
  733 }
  734 
  735 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
  736   assert(rscratch != noreg || always_reachable(src), "missing");
  737 
  738   if (reachable(src)) {
  739     addss(dst, as_Address(src));
  740   } else {
  741     lea(rscratch, src);
  742     addss(dst, Address(rscratch, 0));
  743   }
  744 }
  745 
  746 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  747   assert(rscratch != noreg || always_reachable(src), "missing");
  748 
  749   if (reachable(src)) {
  750     Assembler::addpd(dst, as_Address(src));
  751   } else {
  752     lea(rscratch, src);
  753     Assembler::addpd(dst, Address(rscratch, 0));
  754   }
  755 }
  756 
  757 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
  758 // Stub code is generated once and never copied.
  759 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
  760 void MacroAssembler::align64() {
  761   align(64, (uint)(uintptr_t)pc());
  762 }
  763 
  764 void MacroAssembler::align32() {
  765   align(32, (uint)(uintptr_t)pc());
  766 }
  767 
  768 void MacroAssembler::align(uint modulus) {
  769   // 8273459: Ensure alignment is possible with current segment alignment
  770   assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
  771   align(modulus, offset());
  772 }
  773 
  774 void MacroAssembler::align(uint modulus, uint target) {
  775   if (target % modulus != 0) {
  776     nop(modulus - (target % modulus));
  777   }
  778 }
  779 
  780 void MacroAssembler::push_f(XMMRegister r) {
  781   subptr(rsp, wordSize);
  782   movflt(Address(rsp, 0), r);
  783 }
  784 
  785 void MacroAssembler::pop_f(XMMRegister r) {
  786   movflt(r, Address(rsp, 0));
  787   addptr(rsp, wordSize);
  788 }
  789 
  790 void MacroAssembler::push_d(XMMRegister r) {
  791   subptr(rsp, 2 * wordSize);
  792   movdbl(Address(rsp, 0), r);
  793 }
  794 
  795 void MacroAssembler::pop_d(XMMRegister r) {
  796   movdbl(r, Address(rsp, 0));
  797   addptr(rsp, 2 * Interpreter::stackElementSize);
  798 }
  799 
  800 void MacroAssembler::push_ppx(Register src) {
  801   if (VM_Version::supports_apx_f()) {
  802     pushp(src);
  803   } else {
  804     Assembler::push(src);
  805   }
  806 }
  807 
  808 void MacroAssembler::pop_ppx(Register dst) {
  809   if (VM_Version::supports_apx_f()) {
  810     popp(dst);
  811   } else {
  812     Assembler::pop(dst);
  813   }
  814 }
  815 
  816 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  817   // Used in sign-masking with aligned address.
  818   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  819   assert(rscratch != noreg || always_reachable(src), "missing");
  820 
  821   if (UseAVX > 2 &&
  822       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
  823       (dst->encoding() >= 16)) {
  824     vpand(dst, dst, src, AVX_512bit, rscratch);
  825   } else if (reachable(src)) {
  826     Assembler::andpd(dst, as_Address(src));
  827   } else {
  828     lea(rscratch, src);
  829     Assembler::andpd(dst, Address(rscratch, 0));
  830   }
  831 }
  832 
  833 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
  834   // Used in sign-masking with aligned address.
  835   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  836   assert(rscratch != noreg || always_reachable(src), "missing");
  837 
  838   if (reachable(src)) {
  839     Assembler::andps(dst, as_Address(src));
  840   } else {
  841     lea(rscratch, src);
  842     Assembler::andps(dst, Address(rscratch, 0));
  843   }
  844 }
  845 
  846 void MacroAssembler::andptr(Register dst, int32_t imm32) {
  847   andq(dst, imm32);
  848 }
  849 
  850 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
  851   assert(rscratch != noreg || always_reachable(src), "missing");
  852 
  853   if (reachable(src)) {
  854     andq(dst, as_Address(src));
  855   } else {
  856     lea(rscratch, src);
  857     andq(dst, Address(rscratch, 0));
  858   }
  859 }
  860 
  861 void MacroAssembler::atomic_incl(Address counter_addr) {
  862   lock();
  863   incrementl(counter_addr);
  864 }
  865 
  866 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
  867   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  868 
  869   if (reachable(counter_addr)) {
  870     atomic_incl(as_Address(counter_addr));
  871   } else {
  872     lea(rscratch, counter_addr);
  873     atomic_incl(Address(rscratch, 0));
  874   }
  875 }
  876 
  877 void MacroAssembler::atomic_incq(Address counter_addr) {
  878   lock();
  879   incrementq(counter_addr);
  880 }
  881 
  882 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
  883   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  884 
  885   if (reachable(counter_addr)) {
  886     atomic_incq(as_Address(counter_addr));
  887   } else {
  888     lea(rscratch, counter_addr);
  889     atomic_incq(Address(rscratch, 0));
  890   }
  891 }
  892 
  893 // Writes to stack successive pages until offset reached to check for
  894 // stack overflow + shadow pages.  This clobbers tmp.
  895 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
  896   movptr(tmp, rsp);
  897   // Bang stack for total size given plus shadow page size.
  898   // Bang one page at a time because large size can bang beyond yellow and
  899   // red zones.
  900   Label loop;
  901   bind(loop);
  902   movl(Address(tmp, (-(int)os::vm_page_size())), size );
  903   subptr(tmp, (int)os::vm_page_size());
  904   subl(size, (int)os::vm_page_size());
  905   jcc(Assembler::greater, loop);
  906 
  907   // Bang down shadow pages too.
  908   // At this point, (tmp-0) is the last address touched, so don't
  909   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
  910   // was post-decremented.)  Skip this address by starting at i=1, and
  911   // touch a few more pages below.  N.B.  It is important to touch all
  912   // the way down including all pages in the shadow zone.
  913   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
  914     // this could be any sized move but this is can be a debugging crumb
  915     // so the bigger the better.
  916     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
  917   }
  918 }
  919 
  920 void MacroAssembler::reserved_stack_check() {
  921   // testing if reserved zone needs to be enabled
  922   Label no_reserved_zone_enabling;
  923 
  924   cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
  925   jcc(Assembler::below, no_reserved_zone_enabling);
  926 
  927   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
  928   jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
  929   should_not_reach_here();
  930 
  931   bind(no_reserved_zone_enabling);
  932 }
  933 
  934 void MacroAssembler::c2bool(Register x) {
  935   // implements x == 0 ? 0 : 1
  936   // note: must only look at least-significant byte of x
  937   //       since C-style booleans are stored in one byte
  938   //       only! (was bug)
  939   andl(x, 0xFF);
  940   setb(Assembler::notZero, x);
  941 }
  942 
  943 // Wouldn't need if AddressLiteral version had new name
  944 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
  945   Assembler::call(L, rtype);
  946 }
  947 
  948 void MacroAssembler::call(Register entry) {
  949   Assembler::call(entry);
  950 }
  951 
  952 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
  953   assert(rscratch != noreg || always_reachable(entry), "missing");
  954 
  955   if (reachable(entry)) {
  956     Assembler::call_literal(entry.target(), entry.rspec());
  957   } else {
  958     lea(rscratch, entry);
  959     Assembler::call(rscratch);
  960   }
  961 }
  962 
  963 void MacroAssembler::ic_call(address entry, jint method_index) {
  964   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
  965   // Needs full 64-bit immediate for later patching.
  966   Assembler::mov64(rax, (int64_t)Universe::non_oop_word());
  967   call(AddressLiteral(entry, rh));
  968 }
  969 
  970 int MacroAssembler::ic_check_size() {
  971   return UseCompactObjectHeaders ? 17 : 14;
  972 }
  973 
  974 int MacroAssembler::ic_check(int end_alignment) {
  975   Register receiver = j_rarg0;
  976   Register data = rax;
  977   Register temp = rscratch1;
  978 
  979   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
  980   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
  981   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
  982   // before the inline cache check here, and not after
  983   align(end_alignment, offset() + ic_check_size());
  984 
  985   int uep_offset = offset();
  986 
  987   if (UseCompactObjectHeaders) {
  988     load_narrow_klass_compact(temp, receiver);
  989     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  990   } else {
  991     movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
  992     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  993   }
  994 
  995   // if inline cache check fails, then jump to runtime routine
  996   jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
  997   assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
  998 
  999   return uep_offset;
 1000 }
 1001 
 1002 void MacroAssembler::emit_static_call_stub() {
 1003   // Static stub relocation also tags the Method* in the code-stream.
 1004   mov_metadata(rbx, (Metadata*) nullptr);  // Method is zapped till fixup time.
 1005   // This is recognized as unresolved by relocs/nativeinst/ic code.
 1006   jump(RuntimeAddress(pc()));
 1007 }
 1008 
 1009 // Implementation of call_VM versions
 1010 
 1011 void MacroAssembler::call_VM(Register oop_result,
 1012                              address entry_point,
 1013                              bool check_exceptions) {
 1014   Label C, E;
 1015   call(C, relocInfo::none);
 1016   jmp(E);
 1017 
 1018   bind(C);
 1019   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
 1020   ret(0);
 1021 
 1022   bind(E);
 1023 }
 1024 
 1025 void MacroAssembler::call_VM(Register oop_result,
 1026                              address entry_point,
 1027                              Register arg_1,
 1028                              bool check_exceptions) {
 1029   Label C, E;
 1030   call(C, relocInfo::none);
 1031   jmp(E);
 1032 
 1033   bind(C);
 1034   pass_arg1(this, arg_1);
 1035   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
 1036   ret(0);
 1037 
 1038   bind(E);
 1039 }
 1040 
 1041 void MacroAssembler::call_VM(Register oop_result,
 1042                              address entry_point,
 1043                              Register arg_1,
 1044                              Register arg_2,
 1045                              bool check_exceptions) {
 1046   Label C, E;
 1047   call(C, relocInfo::none);
 1048   jmp(E);
 1049 
 1050   bind(C);
 1051 
 1052   assert_different_registers(arg_1, c_rarg2);
 1053 
 1054   pass_arg2(this, arg_2);
 1055   pass_arg1(this, arg_1);
 1056   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
 1057   ret(0);
 1058 
 1059   bind(E);
 1060 }
 1061 
 1062 void MacroAssembler::call_VM(Register oop_result,
 1063                              address entry_point,
 1064                              Register arg_1,
 1065                              Register arg_2,
 1066                              Register arg_3,
 1067                              bool check_exceptions) {
 1068   Label C, E;
 1069   call(C, relocInfo::none);
 1070   jmp(E);
 1071 
 1072   bind(C);
 1073 
 1074   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1075   assert_different_registers(arg_2, c_rarg3);
 1076   pass_arg3(this, arg_3);
 1077   pass_arg2(this, arg_2);
 1078   pass_arg1(this, arg_1);
 1079   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
 1080   ret(0);
 1081 
 1082   bind(E);
 1083 }
 1084 
 1085 void MacroAssembler::call_VM(Register oop_result,
 1086                              Register last_java_sp,
 1087                              address entry_point,
 1088                              int number_of_arguments,
 1089                              bool check_exceptions) {
 1090   call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1091 }
 1092 
 1093 void MacroAssembler::call_VM(Register oop_result,
 1094                              Register last_java_sp,
 1095                              address entry_point,
 1096                              Register arg_1,
 1097                              bool check_exceptions) {
 1098   pass_arg1(this, arg_1);
 1099   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1100 }
 1101 
 1102 void MacroAssembler::call_VM(Register oop_result,
 1103                              Register last_java_sp,
 1104                              address entry_point,
 1105                              Register arg_1,
 1106                              Register arg_2,
 1107                              bool check_exceptions) {
 1108 
 1109   assert_different_registers(arg_1, c_rarg2);
 1110   pass_arg2(this, arg_2);
 1111   pass_arg1(this, arg_1);
 1112   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1113 }
 1114 
 1115 void MacroAssembler::call_VM(Register oop_result,
 1116                              Register last_java_sp,
 1117                              address entry_point,
 1118                              Register arg_1,
 1119                              Register arg_2,
 1120                              Register arg_3,
 1121                              bool check_exceptions) {
 1122   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1123   assert_different_registers(arg_2, c_rarg3);
 1124   pass_arg3(this, arg_3);
 1125   pass_arg2(this, arg_2);
 1126   pass_arg1(this, arg_1);
 1127   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1128 }
 1129 
 1130 void MacroAssembler::super_call_VM(Register oop_result,
 1131                                    Register last_java_sp,
 1132                                    address entry_point,
 1133                                    int number_of_arguments,
 1134                                    bool check_exceptions) {
 1135   MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1136 }
 1137 
 1138 void MacroAssembler::super_call_VM(Register oop_result,
 1139                                    Register last_java_sp,
 1140                                    address entry_point,
 1141                                    Register arg_1,
 1142                                    bool check_exceptions) {
 1143   pass_arg1(this, arg_1);
 1144   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1145 }
 1146 
 1147 void MacroAssembler::super_call_VM(Register oop_result,
 1148                                    Register last_java_sp,
 1149                                    address entry_point,
 1150                                    Register arg_1,
 1151                                    Register arg_2,
 1152                                    bool check_exceptions) {
 1153 
 1154   assert_different_registers(arg_1, c_rarg2);
 1155   pass_arg2(this, arg_2);
 1156   pass_arg1(this, arg_1);
 1157   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1158 }
 1159 
 1160 void MacroAssembler::super_call_VM(Register oop_result,
 1161                                    Register last_java_sp,
 1162                                    address entry_point,
 1163                                    Register arg_1,
 1164                                    Register arg_2,
 1165                                    Register arg_3,
 1166                                    bool check_exceptions) {
 1167   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1168   assert_different_registers(arg_2, c_rarg3);
 1169   pass_arg3(this, arg_3);
 1170   pass_arg2(this, arg_2);
 1171   pass_arg1(this, arg_1);
 1172   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1173 }
 1174 
 1175 void MacroAssembler::call_VM_base(Register oop_result,
 1176                                   Register last_java_sp,
 1177                                   address  entry_point,
 1178                                   int      number_of_arguments,
 1179                                   bool     check_exceptions) {
 1180   Register java_thread = r15_thread;
 1181 
 1182   // determine last_java_sp register
 1183   if (!last_java_sp->is_valid()) {
 1184     last_java_sp = rsp;
 1185   }
 1186   // debugging support
 1187   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 1188 #ifdef ASSERT
 1189   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 1190   // r12 is the heapbase.
 1191   if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 1192 #endif // ASSERT
 1193 
 1194   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 1195   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 1196 
 1197   // push java thread (becomes first argument of C function)
 1198 
 1199   mov(c_rarg0, r15_thread);
 1200 
 1201   // set last Java frame before call
 1202   assert(last_java_sp != rbp, "can't use ebp/rbp");
 1203 
 1204   // Only interpreter should have to set fp
 1205   set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
 1206 
 1207   // do the call, remove parameters
 1208   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 1209 
 1210 #ifdef ASSERT
 1211   // Check that thread register is not clobbered.
 1212   guarantee(java_thread != rax, "change this code");
 1213   push(rax);
 1214   { Label L;
 1215     get_thread_slow(rax);
 1216     cmpptr(java_thread, rax);
 1217     jcc(Assembler::equal, L);
 1218     STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
 1219     bind(L);
 1220   }
 1221   pop(rax);
 1222 #endif
 1223 
 1224   // reset last Java frame
 1225   // Only interpreter should have to clear fp
 1226   reset_last_Java_frame(true);
 1227 
 1228    // C++ interp handles this in the interpreter
 1229   check_and_handle_popframe();
 1230   check_and_handle_earlyret();
 1231 
 1232   if (check_exceptions) {
 1233     // check for pending exceptions (java_thread is set upon return)
 1234     cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 1235     // This used to conditionally jump to forward_exception however it is
 1236     // possible if we relocate that the branch will not reach. So we must jump
 1237     // around so we can always reach
 1238 
 1239     Label ok;
 1240     jcc(Assembler::equal, ok);
 1241     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 1242     bind(ok);
 1243   }
 1244 
 1245   // get oop result if there is one and reset the value in the thread
 1246   if (oop_result->is_valid()) {
 1247     get_vm_result_oop(oop_result);
 1248   }
 1249 }
 1250 
 1251 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 1252   // Calculate the value for last_Java_sp somewhat subtle.
 1253   // call_VM does an intermediate call which places a return address on
 1254   // the stack just under the stack pointer as the user finished with it.
 1255   // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
 1256 
 1257   // We've pushed one address, correct last_Java_sp
 1258   lea(rax, Address(rsp, wordSize));
 1259 
 1260   call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
 1261 }
 1262 
 1263 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
 1264 void MacroAssembler::call_VM_leaf0(address entry_point) {
 1265   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 1266 }
 1267 
 1268 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
 1269   call_VM_leaf_base(entry_point, number_of_arguments);
 1270 }
 1271 
 1272 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
 1273   pass_arg0(this, arg_0);
 1274   call_VM_leaf(entry_point, 1);
 1275 }
 1276 
 1277 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1278 
 1279   assert_different_registers(arg_0, c_rarg1);
 1280   pass_arg1(this, arg_1);
 1281   pass_arg0(this, arg_0);
 1282   call_VM_leaf(entry_point, 2);
 1283 }
 1284 
 1285 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1286   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1287   assert_different_registers(arg_1, c_rarg2);
 1288   pass_arg2(this, arg_2);
 1289   pass_arg1(this, arg_1);
 1290   pass_arg0(this, arg_0);
 1291   call_VM_leaf(entry_point, 3);
 1292 }
 1293 
 1294 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1295   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1296   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1297   assert_different_registers(arg_2, c_rarg3);
 1298   pass_arg3(this, arg_3);
 1299   pass_arg2(this, arg_2);
 1300   pass_arg1(this, arg_1);
 1301   pass_arg0(this, arg_0);
 1302   call_VM_leaf(entry_point, 3);
 1303 }
 1304 




 1305 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1306   pass_arg0(this, arg_0);
 1307   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1308 }
 1309 
 1310 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1311   assert_different_registers(arg_0, c_rarg1);
 1312   pass_arg1(this, arg_1);
 1313   pass_arg0(this, arg_0);
 1314   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1315 }
 1316 
 1317 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1318   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1319   assert_different_registers(arg_1, c_rarg2);
 1320   pass_arg2(this, arg_2);
 1321   pass_arg1(this, arg_1);
 1322   pass_arg0(this, arg_0);
 1323   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1324 }
 1325 
 1326 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1327   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1328   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1329   assert_different_registers(arg_2, c_rarg3);
 1330   pass_arg3(this, arg_3);
 1331   pass_arg2(this, arg_2);
 1332   pass_arg1(this, arg_1);
 1333   pass_arg0(this, arg_0);
 1334   MacroAssembler::call_VM_leaf_base(entry_point, 4);
 1335 }
 1336 
 1337 void MacroAssembler::get_vm_result_oop(Register oop_result) {
 1338   movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
 1339   movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
 1340   verify_oop_msg(oop_result, "broken oop in call_VM_base");
 1341 }
 1342 
 1343 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
 1344   movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
 1345   movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
 1346 }
 1347 
 1348 void MacroAssembler::check_and_handle_earlyret() {
 1349 }
 1350 
 1351 void MacroAssembler::check_and_handle_popframe() {
 1352 }
 1353 
 1354 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
 1355   assert(rscratch != noreg || always_reachable(src1), "missing");
 1356 
 1357   if (reachable(src1)) {
 1358     cmpl(as_Address(src1), imm);
 1359   } else {
 1360     lea(rscratch, src1);
 1361     cmpl(Address(rscratch, 0), imm);
 1362   }
 1363 }
 1364 
 1365 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
 1366   assert(!src2.is_lval(), "use cmpptr");
 1367   assert(rscratch != noreg || always_reachable(src2), "missing");
 1368 
 1369   if (reachable(src2)) {
 1370     cmpl(src1, as_Address(src2));
 1371   } else {
 1372     lea(rscratch, src2);
 1373     cmpl(src1, Address(rscratch, 0));
 1374   }
 1375 }
 1376 
 1377 void MacroAssembler::cmp32(Register src1, int32_t imm) {
 1378   Assembler::cmpl(src1, imm);
 1379 }
 1380 
 1381 void MacroAssembler::cmp32(Register src1, Address src2) {
 1382   Assembler::cmpl(src1, src2);
 1383 }
 1384 
 1385 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1386   ucomisd(opr1, opr2);
 1387 
 1388   Label L;
 1389   if (unordered_is_less) {
 1390     movl(dst, -1);
 1391     jcc(Assembler::parity, L);
 1392     jcc(Assembler::below , L);
 1393     movl(dst, 0);
 1394     jcc(Assembler::equal , L);
 1395     increment(dst);
 1396   } else { // unordered is greater
 1397     movl(dst, 1);
 1398     jcc(Assembler::parity, L);
 1399     jcc(Assembler::above , L);
 1400     movl(dst, 0);
 1401     jcc(Assembler::equal , L);
 1402     decrementl(dst);
 1403   }
 1404   bind(L);
 1405 }
 1406 
 1407 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1408   ucomiss(opr1, opr2);
 1409 
 1410   Label L;
 1411   if (unordered_is_less) {
 1412     movl(dst, -1);
 1413     jcc(Assembler::parity, L);
 1414     jcc(Assembler::below , L);
 1415     movl(dst, 0);
 1416     jcc(Assembler::equal , L);
 1417     increment(dst);
 1418   } else { // unordered is greater
 1419     movl(dst, 1);
 1420     jcc(Assembler::parity, L);
 1421     jcc(Assembler::above , L);
 1422     movl(dst, 0);
 1423     jcc(Assembler::equal , L);
 1424     decrementl(dst);
 1425   }
 1426   bind(L);
 1427 }
 1428 
 1429 
 1430 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
 1431   assert(rscratch != noreg || always_reachable(src1), "missing");
 1432 
 1433   if (reachable(src1)) {
 1434     cmpb(as_Address(src1), imm);
 1435   } else {
 1436     lea(rscratch, src1);
 1437     cmpb(Address(rscratch, 0), imm);
 1438   }
 1439 }
 1440 
 1441 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
 1442   assert(rscratch != noreg || always_reachable(src2), "missing");
 1443 
 1444   if (src2.is_lval()) {
 1445     movptr(rscratch, src2);
 1446     Assembler::cmpq(src1, rscratch);
 1447   } else if (reachable(src2)) {
 1448     cmpq(src1, as_Address(src2));
 1449   } else {
 1450     lea(rscratch, src2);
 1451     Assembler::cmpq(src1, Address(rscratch, 0));
 1452   }
 1453 }
 1454 
 1455 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
 1456   assert(src2.is_lval(), "not a mem-mem compare");
 1457   // moves src2's literal address
 1458   movptr(rscratch, src2);
 1459   Assembler::cmpq(src1, rscratch);
 1460 }
 1461 
 1462 void MacroAssembler::cmpoop(Register src1, Register src2) {
 1463   cmpptr(src1, src2);
 1464 }
 1465 
 1466 void MacroAssembler::cmpoop(Register src1, Address src2) {
 1467   cmpptr(src1, src2);
 1468 }
 1469 
 1470 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
 1471   movoop(rscratch, src2);
 1472   cmpptr(src1, rscratch);
 1473 }
 1474 
 1475 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
 1476   assert(rscratch != noreg || always_reachable(adr), "missing");
 1477 
 1478   if (reachable(adr)) {
 1479     lock();
 1480     cmpxchgptr(reg, as_Address(adr));
 1481   } else {
 1482     lea(rscratch, adr);
 1483     lock();
 1484     cmpxchgptr(reg, Address(rscratch, 0));
 1485   }
 1486 }
 1487 
 1488 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
 1489   cmpxchgq(reg, adr);
 1490 }
 1491 
 1492 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1493   assert(rscratch != noreg || always_reachable(src), "missing");
 1494 
 1495   if (reachable(src)) {
 1496     Assembler::comisd(dst, as_Address(src));
 1497   } else {
 1498     lea(rscratch, src);
 1499     Assembler::comisd(dst, Address(rscratch, 0));
 1500   }
 1501 }
 1502 
 1503 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1504   assert(rscratch != noreg || always_reachable(src), "missing");
 1505 
 1506   if (reachable(src)) {
 1507     Assembler::comiss(dst, as_Address(src));
 1508   } else {
 1509     lea(rscratch, src);
 1510     Assembler::comiss(dst, Address(rscratch, 0));
 1511   }
 1512 }
 1513 
 1514 
 1515 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
 1516   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1517 
 1518   Condition negated_cond = negate_condition(cond);
 1519   Label L;
 1520   jcc(negated_cond, L);
 1521   pushf(); // Preserve flags
 1522   atomic_incl(counter_addr, rscratch);
 1523   popf();
 1524   bind(L);
 1525 }
 1526 
 1527 int MacroAssembler::corrected_idivl(Register reg) {
 1528   // Full implementation of Java idiv and irem; checks for
 1529   // special case as described in JVM spec., p.243 & p.271.
 1530   // The function returns the (pc) offset of the idivl
 1531   // instruction - may be needed for implicit exceptions.
 1532   //
 1533   //         normal case                           special case
 1534   //
 1535   // input : rax,: dividend                         min_int
 1536   //         reg: divisor   (may not be rax,/rdx)   -1
 1537   //
 1538   // output: rax,: quotient  (= rax, idiv reg)       min_int
 1539   //         rdx: remainder (= rax, irem reg)       0
 1540   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
 1541   const int min_int = 0x80000000;
 1542   Label normal_case, special_case;
 1543 
 1544   // check for special case
 1545   cmpl(rax, min_int);
 1546   jcc(Assembler::notEqual, normal_case);
 1547   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
 1548   cmpl(reg, -1);
 1549   jcc(Assembler::equal, special_case);
 1550 
 1551   // handle normal case
 1552   bind(normal_case);
 1553   cdql();
 1554   int idivl_offset = offset();
 1555   idivl(reg);
 1556 
 1557   // normal and special case exit
 1558   bind(special_case);
 1559 
 1560   return idivl_offset;
 1561 }
 1562 
 1563 
 1564 
 1565 void MacroAssembler::decrementl(Register reg, int value) {
 1566   if (value == min_jint) {subl(reg, value) ; return; }
 1567   if (value <  0) { incrementl(reg, -value); return; }
 1568   if (value == 0) {                        ; return; }
 1569   if (value == 1 && UseIncDec) { decl(reg) ; return; }
 1570   /* else */      { subl(reg, value)       ; return; }
 1571 }
 1572 
 1573 void MacroAssembler::decrementl(Address dst, int value) {
 1574   if (value == min_jint) {subl(dst, value) ; return; }
 1575   if (value <  0) { incrementl(dst, -value); return; }
 1576   if (value == 0) {                        ; return; }
 1577   if (value == 1 && UseIncDec) { decl(dst) ; return; }
 1578   /* else */      { subl(dst, value)       ; return; }
 1579 }
 1580 
 1581 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
 1582   assert(shift_value > 0, "illegal shift value");
 1583   Label _is_positive;
 1584   testl (reg, reg);
 1585   jcc (Assembler::positive, _is_positive);
 1586   int offset = (1 << shift_value) - 1 ;
 1587 
 1588   if (offset == 1) {
 1589     incrementl(reg);
 1590   } else {
 1591     addl(reg, offset);
 1592   }
 1593 
 1594   bind (_is_positive);
 1595   sarl(reg, shift_value);
 1596 }
 1597 
 1598 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1599   assert(rscratch != noreg || always_reachable(src), "missing");
 1600 
 1601   if (reachable(src)) {
 1602     Assembler::divsd(dst, as_Address(src));
 1603   } else {
 1604     lea(rscratch, src);
 1605     Assembler::divsd(dst, Address(rscratch, 0));
 1606   }
 1607 }
 1608 
 1609 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1610   assert(rscratch != noreg || always_reachable(src), "missing");
 1611 
 1612   if (reachable(src)) {
 1613     Assembler::divss(dst, as_Address(src));
 1614   } else {
 1615     lea(rscratch, src);
 1616     Assembler::divss(dst, Address(rscratch, 0));
 1617   }
 1618 }
 1619 
 1620 void MacroAssembler::enter() {
 1621   push(rbp);
 1622   mov(rbp, rsp);
 1623 }
 1624 
 1625 void MacroAssembler::post_call_nop() {
 1626   if (!Continuations::enabled()) {
 1627     return;
 1628   }
 1629   InstructionMark im(this);
 1630   relocate(post_call_nop_Relocation::spec());
 1631   InlineSkippedInstructionsCounter skipCounter(this);
 1632   emit_int8((uint8_t)0x0f);
 1633   emit_int8((uint8_t)0x1f);
 1634   emit_int8((uint8_t)0x84);
 1635   emit_int8((uint8_t)0x00);
 1636   emit_int32(0x00);
 1637 }
 1638 
 1639 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1640   assert(rscratch != noreg || always_reachable(src), "missing");
 1641   if (reachable(src)) {
 1642     Assembler::mulpd(dst, as_Address(src));
 1643   } else {
 1644     lea(rscratch, src);
 1645     Assembler::mulpd(dst, Address(rscratch, 0));
 1646   }
 1647 }
 1648 
 1649 // dst = c = a * b + c
 1650 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1651   Assembler::vfmadd231sd(c, a, b);
 1652   if (dst != c) {
 1653     movdbl(dst, c);
 1654   }
 1655 }
 1656 
 1657 // dst = c = a * b + c
 1658 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1659   Assembler::vfmadd231ss(c, a, b);
 1660   if (dst != c) {
 1661     movflt(dst, c);
 1662   }
 1663 }
 1664 
 1665 // dst = c = a * b + c
 1666 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1667   Assembler::vfmadd231pd(c, a, b, vector_len);
 1668   if (dst != c) {
 1669     vmovdqu(dst, c);
 1670   }
 1671 }
 1672 
 1673 // dst = c = a * b + c
 1674 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1675   Assembler::vfmadd231ps(c, a, b, vector_len);
 1676   if (dst != c) {
 1677     vmovdqu(dst, c);
 1678   }
 1679 }
 1680 
 1681 // dst = c = a * b + c
 1682 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1683   Assembler::vfmadd231pd(c, a, b, vector_len);
 1684   if (dst != c) {
 1685     vmovdqu(dst, c);
 1686   }
 1687 }
 1688 
 1689 // dst = c = a * b + c
 1690 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1691   Assembler::vfmadd231ps(c, a, b, vector_len);
 1692   if (dst != c) {
 1693     vmovdqu(dst, c);
 1694   }
 1695 }
 1696 
 1697 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
 1698   assert(rscratch != noreg || always_reachable(dst), "missing");
 1699 
 1700   if (reachable(dst)) {
 1701     incrementl(as_Address(dst));
 1702   } else {
 1703     lea(rscratch, dst);
 1704     incrementl(Address(rscratch, 0));
 1705   }
 1706 }
 1707 
 1708 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
 1709   incrementl(as_Address(dst, rscratch));
 1710 }
 1711 
 1712 void MacroAssembler::incrementl(Register reg, int value) {
 1713   if (value == min_jint) {addl(reg, value) ; return; }
 1714   if (value <  0) { decrementl(reg, -value); return; }
 1715   if (value == 0) {                        ; return; }
 1716   if (value == 1 && UseIncDec) { incl(reg) ; return; }
 1717   /* else */      { addl(reg, value)       ; return; }
 1718 }
 1719 
 1720 void MacroAssembler::incrementl(Address dst, int value) {
 1721   if (value == min_jint) {addl(dst, value) ; return; }
 1722   if (value <  0) { decrementl(dst, -value); return; }
 1723   if (value == 0) {                        ; return; }
 1724   if (value == 1 && UseIncDec) { incl(dst) ; return; }
 1725   /* else */      { addl(dst, value)       ; return; }
 1726 }
 1727 
 1728 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
 1729   assert(rscratch != noreg || always_reachable(dst), "missing");
 1730   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
 1731   if (reachable(dst)) {
 1732     jmp_literal(dst.target(), dst.rspec());
 1733   } else {
 1734     lea(rscratch, dst);
 1735     jmp(rscratch);
 1736   }
 1737 }
 1738 
 1739 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
 1740   assert(rscratch != noreg || always_reachable(dst), "missing");
 1741   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump_cc");
 1742   if (reachable(dst)) {
 1743     InstructionMark im(this);
 1744     relocate(dst.reloc());
 1745     const int short_size = 2;
 1746     const int long_size = 6;
 1747     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
 1748     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
 1749       // 0111 tttn #8-bit disp
 1750       emit_int8(0x70 | cc);
 1751       emit_int8((offs - short_size) & 0xFF);
 1752     } else {
 1753       // 0000 1111 1000 tttn #32-bit disp
 1754       emit_int8(0x0F);
 1755       emit_int8((unsigned char)(0x80 | cc));
 1756       emit_int32(offs - long_size);
 1757     }
 1758   } else {
 1759 #ifdef ASSERT
 1760     warning("reversing conditional branch");
 1761 #endif /* ASSERT */
 1762     Label skip;
 1763     jccb(reverse[cc], skip);
 1764     lea(rscratch, dst);
 1765     Assembler::jmp(rscratch);
 1766     bind(skip);
 1767   }
 1768 }
 1769 
 1770 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
 1771   ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 1772   assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
 1773 
 1774   stmxcsr(mxcsr_save);
 1775   movl(tmp, mxcsr_save);
 1776   if (EnableX86ECoreOpts) {
 1777     // The mxcsr_std has status bits set for performance on ECore
 1778     orl(tmp, 0x003f);
 1779   } else {
 1780     // Mask out status bits (only check control and mask bits)
 1781     andl(tmp, 0xFFC0);
 1782   }
 1783   cmp32(tmp, mxcsr_std, rscratch);
 1784 }
 1785 
 1786 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
 1787   assert(rscratch != noreg || always_reachable(src), "missing");
 1788 
 1789   if (reachable(src)) {
 1790     Assembler::ldmxcsr(as_Address(src));
 1791   } else {
 1792     lea(rscratch, src);
 1793     Assembler::ldmxcsr(Address(rscratch, 0));
 1794   }
 1795 }
 1796 
 1797 int MacroAssembler::load_signed_byte(Register dst, Address src) {
 1798   int off = offset();
 1799   movsbl(dst, src); // movsxb
 1800   return off;
 1801 }
 1802 
 1803 // Note: load_signed_short used to be called load_signed_word.
 1804 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
 1805 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
 1806 // The term "word" in HotSpot means a 32- or 64-bit machine word.
 1807 int MacroAssembler::load_signed_short(Register dst, Address src) {
 1808   // This is dubious to me since it seems safe to do a signed 16 => 64 bit
 1809   // version but this is what 64bit has always done. This seems to imply
 1810   // that users are only using 32bits worth.
 1811   int off = offset();
 1812   movswl(dst, src); // movsxw
 1813   return off;
 1814 }
 1815 
 1816 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
 1817   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1818   // and "3.9 Partial Register Penalties", p. 22).
 1819   int off = offset();
 1820   movzbl(dst, src); // movzxb
 1821   return off;
 1822 }
 1823 
 1824 // Note: load_unsigned_short used to be called load_unsigned_word.
 1825 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
 1826   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1827   // and "3.9 Partial Register Penalties", p. 22).
 1828   int off = offset();
 1829   movzwl(dst, src); // movzxw
 1830   return off;
 1831 }
 1832 
 1833 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
 1834   switch (size_in_bytes) {
 1835   case  8:  movq(dst, src); break;
 1836   case  4:  movl(dst, src); break;
 1837   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
 1838   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
 1839   default:  ShouldNotReachHere();
 1840   }
 1841 }
 1842 
 1843 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
 1844   switch (size_in_bytes) {
 1845   case  8:  movq(dst, src); break;
 1846   case  4:  movl(dst, src); break;
 1847   case  2:  movw(dst, src); break;
 1848   case  1:  movb(dst, src); break;
 1849   default:  ShouldNotReachHere();
 1850   }
 1851 }
 1852 
 1853 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
 1854   assert(rscratch != noreg || always_reachable(dst), "missing");
 1855 
 1856   if (reachable(dst)) {
 1857     movl(as_Address(dst), src);
 1858   } else {
 1859     lea(rscratch, dst);
 1860     movl(Address(rscratch, 0), src);
 1861   }
 1862 }
 1863 
 1864 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
 1865   if (reachable(src)) {
 1866     movl(dst, as_Address(src));
 1867   } else {
 1868     lea(dst, src);
 1869     movl(dst, Address(dst, 0));
 1870   }
 1871 }
 1872 
 1873 // C++ bool manipulation
 1874 
 1875 void MacroAssembler::movbool(Register dst, Address src) {
 1876   if(sizeof(bool) == 1)
 1877     movb(dst, src);
 1878   else if(sizeof(bool) == 2)
 1879     movw(dst, src);
 1880   else if(sizeof(bool) == 4)
 1881     movl(dst, src);
 1882   else
 1883     // unsupported
 1884     ShouldNotReachHere();
 1885 }
 1886 
 1887 void MacroAssembler::movbool(Address dst, bool boolconst) {
 1888   if(sizeof(bool) == 1)
 1889     movb(dst, (int) boolconst);
 1890   else if(sizeof(bool) == 2)
 1891     movw(dst, (int) boolconst);
 1892   else if(sizeof(bool) == 4)
 1893     movl(dst, (int) boolconst);
 1894   else
 1895     // unsupported
 1896     ShouldNotReachHere();
 1897 }
 1898 
 1899 void MacroAssembler::movbool(Address dst, Register src) {
 1900   if(sizeof(bool) == 1)
 1901     movb(dst, src);
 1902   else if(sizeof(bool) == 2)
 1903     movw(dst, src);
 1904   else if(sizeof(bool) == 4)
 1905     movl(dst, src);
 1906   else
 1907     // unsupported
 1908     ShouldNotReachHere();
 1909 }
 1910 
 1911 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1912   assert(rscratch != noreg || always_reachable(src), "missing");
 1913 
 1914   if (reachable(src)) {
 1915     movdl(dst, as_Address(src));
 1916   } else {
 1917     lea(rscratch, src);
 1918     movdl(dst, Address(rscratch, 0));
 1919   }
 1920 }
 1921 
 1922 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1923   assert(rscratch != noreg || always_reachable(src), "missing");
 1924 
 1925   if (reachable(src)) {
 1926     movq(dst, as_Address(src));
 1927   } else {
 1928     lea(rscratch, src);
 1929     movq(dst, Address(rscratch, 0));
 1930   }
 1931 }
 1932 
 1933 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1934   assert(rscratch != noreg || always_reachable(src), "missing");
 1935 
 1936   if (reachable(src)) {
 1937     if (UseXmmLoadAndClearUpper) {
 1938       movsd (dst, as_Address(src));
 1939     } else {
 1940       movlpd(dst, as_Address(src));
 1941     }
 1942   } else {
 1943     lea(rscratch, src);
 1944     if (UseXmmLoadAndClearUpper) {
 1945       movsd (dst, Address(rscratch, 0));
 1946     } else {
 1947       movlpd(dst, Address(rscratch, 0));
 1948     }
 1949   }
 1950 }
 1951 
 1952 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1953   assert(rscratch != noreg || always_reachable(src), "missing");
 1954 
 1955   if (reachable(src)) {
 1956     movss(dst, as_Address(src));
 1957   } else {
 1958     lea(rscratch, src);
 1959     movss(dst, Address(rscratch, 0));
 1960   }
 1961 }
 1962 
 1963 void MacroAssembler::movhlf(XMMRegister dst, XMMRegister src, Register rscratch) {
 1964   if (VM_Version::supports_avx10_2()) {
 1965     evmovw(dst, src);
 1966   } else {
 1967     assert(rscratch != noreg, "missing");
 1968     evmovw(rscratch, src);
 1969     evmovw(dst, rscratch);
 1970   }
 1971 }
 1972 
 1973 void MacroAssembler::mov64(Register dst, int64_t imm64) {
 1974   if (is_uimm32(imm64)) {
 1975     movl(dst, checked_cast<uint32_t>(imm64));
 1976   } else if (is_simm32(imm64)) {
 1977     movq(dst, checked_cast<int32_t>(imm64));
 1978   } else {
 1979     Assembler::mov64(dst, imm64);
 1980   }
 1981 }
 1982 
 1983 void MacroAssembler::mov64(Register dst, int64_t imm64, relocInfo::relocType rtype, int format) {
 1984   Assembler::mov64(dst, imm64, rtype, format);
 1985 }
 1986 
 1987 void MacroAssembler::movptr(Register dst, Register src) {
 1988   movq(dst, src);
 1989 }
 1990 
 1991 void MacroAssembler::movptr(Register dst, Address src) {
 1992   movq(dst, src);
 1993 }
 1994 
 1995 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 1996 void MacroAssembler::movptr(Register dst, intptr_t src) {
 1997   mov64(dst, src);
 1998 }
 1999 
 2000 void MacroAssembler::movptr(Address dst, Register src) {
 2001   movq(dst, src);
 2002 }
 2003 
 2004 void MacroAssembler::movptr(Address dst, int32_t src) {
 2005   movslq(dst, src);
 2006 }
 2007 
 2008 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 2009   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2010   Assembler::movdqu(dst, src);
 2011 }
 2012 
 2013 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 2014   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2015   Assembler::movdqu(dst, src);
 2016 }
 2017 
 2018 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2019   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2020   Assembler::movdqu(dst, src);
 2021 }
 2022 
 2023 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2024   assert(rscratch != noreg || always_reachable(src), "missing");
 2025 
 2026   if (reachable(src)) {
 2027     movdqu(dst, as_Address(src));
 2028   } else {
 2029     lea(rscratch, src);
 2030     movdqu(dst, Address(rscratch, 0));
 2031   }
 2032 }
 2033 
 2034 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2035   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2036   Assembler::vmovdqu(dst, src);
 2037 }
 2038 
 2039 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2040   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2041   Assembler::vmovdqu(dst, src);
 2042 }
 2043 
 2044 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2045   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2046   Assembler::vmovdqu(dst, src);
 2047 }
 2048 
 2049 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2050   assert(rscratch != noreg || always_reachable(src), "missing");
 2051 
 2052   if (reachable(src)) {
 2053     vmovdqu(dst, as_Address(src));
 2054   }
 2055   else {
 2056     lea(rscratch, src);
 2057     vmovdqu(dst, Address(rscratch, 0));
 2058   }
 2059 }
 2060 
 2061 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2062   assert(rscratch != noreg || always_reachable(src), "missing");
 2063 
 2064   if (vector_len == AVX_512bit) {
 2065     evmovdquq(dst, src, AVX_512bit, rscratch);
 2066   } else if (vector_len == AVX_256bit) {
 2067     vmovdqu(dst, src, rscratch);
 2068   } else {
 2069     movdqu(dst, src, rscratch);
 2070   }
 2071 }
 2072 
 2073 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
 2074   if (vector_len == AVX_512bit) {
 2075     evmovdquq(dst, src, AVX_512bit);
 2076   } else if (vector_len == AVX_256bit) {
 2077     vmovdqu(dst, src);
 2078   } else {
 2079     movdqu(dst, src);
 2080   }
 2081 }
 2082 
 2083 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
 2084   if (vector_len == AVX_512bit) {
 2085     evmovdquq(dst, src, AVX_512bit);
 2086   } else if (vector_len == AVX_256bit) {
 2087     vmovdqu(dst, src);
 2088   } else {
 2089     movdqu(dst, src);
 2090   }
 2091 }
 2092 
 2093 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
 2094   if (vector_len == AVX_512bit) {
 2095     evmovdquq(dst, src, AVX_512bit);
 2096   } else if (vector_len == AVX_256bit) {
 2097     vmovdqu(dst, src);
 2098   } else {
 2099     movdqu(dst, src);
 2100   }
 2101 }
 2102 
 2103 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2104   assert(rscratch != noreg || always_reachable(src), "missing");
 2105 
 2106   if (reachable(src)) {
 2107     vmovdqa(dst, as_Address(src));
 2108   }
 2109   else {
 2110     lea(rscratch, src);
 2111     vmovdqa(dst, Address(rscratch, 0));
 2112   }
 2113 }
 2114 
 2115 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2116   assert(rscratch != noreg || always_reachable(src), "missing");
 2117 
 2118   if (vector_len == AVX_512bit) {
 2119     evmovdqaq(dst, src, AVX_512bit, rscratch);
 2120   } else if (vector_len == AVX_256bit) {
 2121     vmovdqa(dst, src, rscratch);
 2122   } else {
 2123     movdqa(dst, src, rscratch);
 2124   }
 2125 }
 2126 
 2127 void MacroAssembler::kmov(KRegister dst, Address src) {
 2128   if (VM_Version::supports_avx512bw()) {
 2129     kmovql(dst, src);
 2130   } else {
 2131     assert(VM_Version::supports_evex(), "");
 2132     kmovwl(dst, src);
 2133   }
 2134 }
 2135 
 2136 void MacroAssembler::kmov(Address dst, KRegister src) {
 2137   if (VM_Version::supports_avx512bw()) {
 2138     kmovql(dst, src);
 2139   } else {
 2140     assert(VM_Version::supports_evex(), "");
 2141     kmovwl(dst, src);
 2142   }
 2143 }
 2144 
 2145 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2146   if (VM_Version::supports_avx512bw()) {
 2147     kmovql(dst, src);
 2148   } else {
 2149     assert(VM_Version::supports_evex(), "");
 2150     kmovwl(dst, src);
 2151   }
 2152 }
 2153 
 2154 void MacroAssembler::kmov(Register dst, KRegister src) {
 2155   if (VM_Version::supports_avx512bw()) {
 2156     kmovql(dst, src);
 2157   } else {
 2158     assert(VM_Version::supports_evex(), "");
 2159     kmovwl(dst, src);
 2160   }
 2161 }
 2162 
 2163 void MacroAssembler::kmov(KRegister dst, Register src) {
 2164   if (VM_Version::supports_avx512bw()) {
 2165     kmovql(dst, src);
 2166   } else {
 2167     assert(VM_Version::supports_evex(), "");
 2168     kmovwl(dst, src);
 2169   }
 2170 }
 2171 
 2172 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2173   assert(rscratch != noreg || always_reachable(src), "missing");
 2174 
 2175   if (reachable(src)) {
 2176     kmovql(dst, as_Address(src));
 2177   } else {
 2178     lea(rscratch, src);
 2179     kmovql(dst, Address(rscratch, 0));
 2180   }
 2181 }
 2182 
 2183 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2184   assert(rscratch != noreg || always_reachable(src), "missing");
 2185 
 2186   if (reachable(src)) {
 2187     kmovwl(dst, as_Address(src));
 2188   } else {
 2189     lea(rscratch, src);
 2190     kmovwl(dst, Address(rscratch, 0));
 2191   }
 2192 }
 2193 
 2194 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2195                                int vector_len, Register rscratch) {
 2196   assert(rscratch != noreg || always_reachable(src), "missing");
 2197 
 2198   if (reachable(src)) {
 2199     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2200   } else {
 2201     lea(rscratch, src);
 2202     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2203   }
 2204 }
 2205 
 2206 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2207                                int vector_len, Register rscratch) {
 2208   assert(rscratch != noreg || always_reachable(src), "missing");
 2209 
 2210   if (reachable(src)) {
 2211     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2212   } else {
 2213     lea(rscratch, src);
 2214     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2215   }
 2216 }
 2217 
 2218 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2219   assert(rscratch != noreg || always_reachable(src), "missing");
 2220 
 2221   if (reachable(src)) {
 2222     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2223   } else {
 2224     lea(rscratch, src);
 2225     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2226   }
 2227 }
 2228 
 2229 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2230   assert(rscratch != noreg || always_reachable(src), "missing");
 2231 
 2232   if (reachable(src)) {
 2233     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2234   } else {
 2235     lea(rscratch, src);
 2236     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2237   }
 2238 }
 2239 
 2240 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2241   assert(rscratch != noreg || always_reachable(src), "missing");
 2242 
 2243   if (reachable(src)) {
 2244     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2245   } else {
 2246     lea(rscratch, src);
 2247     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2248   }
 2249 }
 2250 
 2251 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2252   assert(rscratch != noreg || always_reachable(src), "missing");
 2253 
 2254   if (reachable(src)) {
 2255     Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
 2256   } else {
 2257     lea(rscratch, src);
 2258     Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2259   }
 2260 }
 2261 
 2262 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2263   assert(rscratch != noreg || always_reachable(src), "missing");
 2264 
 2265   if (reachable(src)) {
 2266     Assembler::evmovdqaq(dst, as_Address(src), vector_len);
 2267   } else {
 2268     lea(rscratch, src);
 2269     Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
 2270   }
 2271 }
 2272 
 2273 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2274   assert(rscratch != noreg || always_reachable(src), "missing");
 2275 
 2276   if (reachable(src)) {
 2277     Assembler::movapd(dst, as_Address(src));
 2278   } else {
 2279     lea(rscratch, src);
 2280     Assembler::movapd(dst, Address(rscratch, 0));
 2281   }
 2282 }
 2283 
 2284 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2285   assert(rscratch != noreg || always_reachable(src), "missing");
 2286 
 2287   if (reachable(src)) {
 2288     Assembler::movdqa(dst, as_Address(src));
 2289   } else {
 2290     lea(rscratch, src);
 2291     Assembler::movdqa(dst, Address(rscratch, 0));
 2292   }
 2293 }
 2294 
 2295 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2296   assert(rscratch != noreg || always_reachable(src), "missing");
 2297 
 2298   if (reachable(src)) {
 2299     Assembler::movsd(dst, as_Address(src));
 2300   } else {
 2301     lea(rscratch, src);
 2302     Assembler::movsd(dst, Address(rscratch, 0));
 2303   }
 2304 }
 2305 
 2306 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2307   assert(rscratch != noreg || always_reachable(src), "missing");
 2308 
 2309   if (reachable(src)) {
 2310     Assembler::movss(dst, as_Address(src));
 2311   } else {
 2312     lea(rscratch, src);
 2313     Assembler::movss(dst, Address(rscratch, 0));
 2314   }
 2315 }
 2316 
 2317 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2318   assert(rscratch != noreg || always_reachable(src), "missing");
 2319 
 2320   if (reachable(src)) {
 2321     Assembler::movddup(dst, as_Address(src));
 2322   } else {
 2323     lea(rscratch, src);
 2324     Assembler::movddup(dst, Address(rscratch, 0));
 2325   }
 2326 }
 2327 
 2328 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2329   assert(rscratch != noreg || always_reachable(src), "missing");
 2330 
 2331   if (reachable(src)) {
 2332     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2333   } else {
 2334     lea(rscratch, src);
 2335     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2336   }
 2337 }
 2338 
 2339 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2340   assert(rscratch != noreg || always_reachable(src), "missing");
 2341 
 2342   if (reachable(src)) {
 2343     Assembler::mulsd(dst, as_Address(src));
 2344   } else {
 2345     lea(rscratch, src);
 2346     Assembler::mulsd(dst, Address(rscratch, 0));
 2347   }
 2348 }
 2349 
 2350 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2351   assert(rscratch != noreg || always_reachable(src), "missing");
 2352 
 2353   if (reachable(src)) {
 2354     Assembler::mulss(dst, as_Address(src));
 2355   } else {
 2356     lea(rscratch, src);
 2357     Assembler::mulss(dst, Address(rscratch, 0));
 2358   }
 2359 }
 2360 
 2361 void MacroAssembler::null_check(Register reg, int offset) {
 2362   if (needs_explicit_null_check(offset)) {
 2363     // provoke OS null exception if reg is null by
 2364     // accessing M[reg] w/o changing any (non-CC) registers
 2365     // NOTE: cmpl is plenty here to provoke a segv
 2366     cmpptr(rax, Address(reg, 0));
 2367     // Note: should probably use testl(rax, Address(reg, 0));
 2368     //       may be shorter code (however, this version of
 2369     //       testl needs to be implemented first)
 2370   } else {
 2371     // nothing to do, (later) access of M[reg + offset]
 2372     // will provoke OS null exception if reg is null
 2373   }
 2374 }
 2375 











































































































 2376 void MacroAssembler::os_breakpoint() {
 2377   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2378   // (e.g., MSVC can't call ps() otherwise)
 2379   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2380 }
 2381 
 2382 void MacroAssembler::unimplemented(const char* what) {
 2383   const char* buf = nullptr;
 2384   {
 2385     ResourceMark rm;
 2386     stringStream ss;
 2387     ss.print("unimplemented: %s", what);
 2388     buf = code_string(ss.as_string());
 2389   }
 2390   stop(buf);
 2391 }
 2392 
 2393 #define XSTATE_BV 0x200
 2394 
 2395 void MacroAssembler::pop_CPU_state() {
 2396   pop_FPU_state();
 2397   pop_IU_state();
 2398 }
 2399 
 2400 void MacroAssembler::pop_FPU_state() {
 2401   fxrstor(Address(rsp, 0));
 2402   addptr(rsp, FPUStateSizeInWords * wordSize);
 2403 }
 2404 
 2405 void MacroAssembler::pop_IU_state() {
 2406   popa();
 2407   addq(rsp, 8);
 2408   popf();
 2409 }
 2410 
 2411 // Save Integer and Float state
 2412 // Warning: Stack must be 16 byte aligned (64bit)
 2413 void MacroAssembler::push_CPU_state() {
 2414   push_IU_state();
 2415   push_FPU_state();
 2416 }
 2417 
 2418 void MacroAssembler::push_FPU_state() {
 2419   subptr(rsp, FPUStateSizeInWords * wordSize);
 2420   fxsave(Address(rsp, 0));
 2421 }
 2422 
 2423 void MacroAssembler::push_IU_state() {
 2424   // Push flags first because pusha kills them
 2425   pushf();
 2426   // Make sure rsp stays 16-byte aligned
 2427   subq(rsp, 8);
 2428   pusha();
 2429 }
 2430 
 2431 void MacroAssembler::push_cont_fastpath() {
 2432   if (!Continuations::enabled()) return;
 2433 
 2434   Label L_done;
 2435   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2436   jccb(Assembler::belowEqual, L_done);
 2437   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2438   bind(L_done);
 2439 }
 2440 
 2441 void MacroAssembler::pop_cont_fastpath() {
 2442   if (!Continuations::enabled()) return;
 2443 
 2444   Label L_done;
 2445   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2446   jccb(Assembler::below, L_done);
 2447   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2448   bind(L_done);
 2449 }
 2450 
 2451 #ifdef ASSERT
 2452 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2453   Label no_cont;
 2454   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2455   testl(cont, cont);
 2456   jcc(Assembler::zero, no_cont);
 2457   stop(name);
 2458   bind(no_cont);
 2459 }
 2460 #endif
 2461 
 2462 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2463   // we must set sp to zero to clear frame
 2464   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2465   // must clear fp, so that compiled frames are not confused; it is
 2466   // possible that we need it only for debugging
 2467   if (clear_fp) {
 2468     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2469   }
 2470   // Always clear the pc because it could have been set by make_walkable()
 2471   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2472   vzeroupper();
 2473 }
 2474 
 2475 void MacroAssembler::round_to(Register reg, int modulus) {
 2476   addptr(reg, modulus - 1);
 2477   andptr(reg, -modulus);
 2478 }
 2479 
 2480 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2481   if (at_return) {
 2482     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2483     // we may safely use rsp instead to perform the stack watermark check.
 2484     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2485     jcc(Assembler::above, slow_path);
 2486     return;
 2487   }
 2488   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2489   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2490 }
 2491 
 2492 // Calls to C land
 2493 //
 2494 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2495 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2496 // has to be reset to 0. This is required to allow proper stack traversal.
 2497 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2498                                          Register last_java_fp,
 2499                                          address  last_java_pc,
 2500                                          Register rscratch) {
 2501   vzeroupper();
 2502   // determine last_java_sp register
 2503   if (!last_java_sp->is_valid()) {
 2504     last_java_sp = rsp;
 2505   }
 2506   // last_java_fp is optional
 2507   if (last_java_fp->is_valid()) {
 2508     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2509   }
 2510   // last_java_pc is optional
 2511   if (last_java_pc != nullptr) {
 2512     Address java_pc(r15_thread,
 2513                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2514     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2515   }
 2516   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2517 }
 2518 
 2519 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2520                                          Register last_java_fp,
 2521                                          Label &L,
 2522                                          Register scratch) {
 2523   lea(scratch, L);
 2524   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2525   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2526 }
 2527 
 2528 void MacroAssembler::shlptr(Register dst, int imm8) {
 2529   shlq(dst, imm8);
 2530 }
 2531 
 2532 void MacroAssembler::shrptr(Register dst, int imm8) {
 2533   shrq(dst, imm8);
 2534 }
 2535 
 2536 void MacroAssembler::sign_extend_byte(Register reg) {
 2537   movsbl(reg, reg); // movsxb
 2538 }
 2539 
 2540 void MacroAssembler::sign_extend_short(Register reg) {
 2541   movswl(reg, reg); // movsxw
 2542 }
 2543 
 2544 void MacroAssembler::narrow_subword_type(Register reg, BasicType bt) {
 2545   assert(is_subword_type(bt), "required");
 2546   switch (bt) {
 2547   case T_BOOLEAN: andl(reg, 1); break;
 2548   case T_BYTE:    movsbl(reg, reg); break;
 2549   case T_CHAR:    movzwl(reg, reg); break;
 2550   case T_SHORT:   movswl(reg, reg); break;
 2551   default:        ShouldNotReachHere();
 2552   }
 2553 }
 2554 
 2555 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2556   if (imm32 >= 0 && is8bit(imm32)) {
 2557     testb(dst, imm32);
 2558   } else {
 2559     Assembler::testl(dst, imm32);
 2560   }
 2561 }
 2562 
 2563 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2564   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2565     testb(dst, imm32);
 2566   } else {
 2567     Assembler::testl(dst, imm32);
 2568   }
 2569 }
 2570 
 2571 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2572   assert(always_reachable(src), "Address should be reachable");
 2573   testl(dst, as_Address(src));
 2574 }
 2575 
 2576 void MacroAssembler::testq(Address dst, int32_t imm32) {
 2577   if (imm32 >= 0) {
 2578     testl(dst, imm32);
 2579   } else {
 2580     Assembler::testq(dst, imm32);
 2581   }
 2582 }
 2583 
 2584 void MacroAssembler::testq(Register dst, int32_t imm32) {
 2585   if (imm32 >= 0) {
 2586     testl(dst, imm32);
 2587   } else {
 2588     Assembler::testq(dst, imm32);
 2589   }
 2590 }
 2591 
 2592 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2593   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2594   Assembler::pcmpeqb(dst, src);
 2595 }
 2596 
 2597 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2598   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2599   Assembler::pcmpeqw(dst, src);
 2600 }
 2601 
 2602 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2603   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2604   Assembler::pcmpestri(dst, src, imm8);
 2605 }
 2606 
 2607 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2608   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2609   Assembler::pcmpestri(dst, src, imm8);
 2610 }
 2611 
 2612 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2613   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2614   Assembler::pmovzxbw(dst, src);
 2615 }
 2616 
 2617 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2618   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2619   Assembler::pmovzxbw(dst, src);
 2620 }
 2621 
 2622 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2623   assert((src->encoding() < 16),"XMM register should be 0-15");
 2624   Assembler::pmovmskb(dst, src);
 2625 }
 2626 
 2627 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2628   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2629   Assembler::ptest(dst, src);
 2630 }
 2631 
 2632 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2633   assert(rscratch != noreg || always_reachable(src), "missing");
 2634 
 2635   if (reachable(src)) {
 2636     Assembler::sqrtss(dst, as_Address(src));
 2637   } else {
 2638     lea(rscratch, src);
 2639     Assembler::sqrtss(dst, Address(rscratch, 0));
 2640   }
 2641 }
 2642 
 2643 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2644   assert(rscratch != noreg || always_reachable(src), "missing");
 2645 
 2646   if (reachable(src)) {
 2647     Assembler::subsd(dst, as_Address(src));
 2648   } else {
 2649     lea(rscratch, src);
 2650     Assembler::subsd(dst, Address(rscratch, 0));
 2651   }
 2652 }
 2653 
 2654 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2655   assert(rscratch != noreg || always_reachable(src), "missing");
 2656 
 2657   if (reachable(src)) {
 2658     Assembler::roundsd(dst, as_Address(src), rmode);
 2659   } else {
 2660     lea(rscratch, src);
 2661     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2662   }
 2663 }
 2664 
 2665 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2666   assert(rscratch != noreg || always_reachable(src), "missing");
 2667 
 2668   if (reachable(src)) {
 2669     Assembler::subss(dst, as_Address(src));
 2670   } else {
 2671     lea(rscratch, src);
 2672     Assembler::subss(dst, Address(rscratch, 0));
 2673   }
 2674 }
 2675 
 2676 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2677   assert(rscratch != noreg || always_reachable(src), "missing");
 2678 
 2679   if (reachable(src)) {
 2680     Assembler::ucomisd(dst, as_Address(src));
 2681   } else {
 2682     lea(rscratch, src);
 2683     Assembler::ucomisd(dst, Address(rscratch, 0));
 2684   }
 2685 }
 2686 
 2687 void MacroAssembler::evucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2688   assert(rscratch != noreg || always_reachable(src), "missing");
 2689 
 2690   if (reachable(src)) {
 2691     Assembler::evucomxsd(dst, as_Address(src));
 2692   } else {
 2693     lea(rscratch, src);
 2694     Assembler::evucomxsd(dst, Address(rscratch, 0));
 2695   }
 2696 }
 2697 
 2698 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2699   assert(rscratch != noreg || always_reachable(src), "missing");
 2700 
 2701   if (reachable(src)) {
 2702     Assembler::ucomiss(dst, as_Address(src));
 2703   } else {
 2704     lea(rscratch, src);
 2705     Assembler::ucomiss(dst, Address(rscratch, 0));
 2706   }
 2707 }
 2708 
 2709 void MacroAssembler::evucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2710   assert(rscratch != noreg || always_reachable(src), "missing");
 2711 
 2712   if (reachable(src)) {
 2713     Assembler::evucomxss(dst, as_Address(src));
 2714   } else {
 2715     lea(rscratch, src);
 2716     Assembler::evucomxss(dst, Address(rscratch, 0));
 2717   }
 2718 }
 2719 
 2720 void MacroAssembler::evucomish(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2721   assert(rscratch != noreg || always_reachable(src), "missing");
 2722 
 2723   if (reachable(src)) {
 2724     Assembler::evucomish(dst, as_Address(src));
 2725   } else {
 2726     lea(rscratch, src);
 2727     Assembler::evucomish(dst, Address(rscratch, 0));
 2728   }
 2729 }
 2730 
 2731 void MacroAssembler::evucomxsh(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2732   assert(rscratch != noreg || always_reachable(src), "missing");
 2733 
 2734   if (reachable(src)) {
 2735     Assembler::evucomxsh(dst, as_Address(src));
 2736   } else {
 2737     lea(rscratch, src);
 2738     Assembler::evucomxsh(dst, Address(rscratch, 0));
 2739   }
 2740 }
 2741 
 2742 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2743   assert(rscratch != noreg || always_reachable(src), "missing");
 2744 
 2745   // Used in sign-bit flipping with aligned address.
 2746   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2747 
 2748   if (UseAVX > 2 &&
 2749       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2750       (dst->encoding() >= 16)) {
 2751     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2752   } else if (reachable(src)) {
 2753     Assembler::xorpd(dst, as_Address(src));
 2754   } else {
 2755     lea(rscratch, src);
 2756     Assembler::xorpd(dst, Address(rscratch, 0));
 2757   }
 2758 }
 2759 
 2760 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2761   if (UseAVX > 2 &&
 2762       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2763       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2764     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2765   } else {
 2766     Assembler::xorpd(dst, src);
 2767   }
 2768 }
 2769 
 2770 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2771   if (UseAVX > 2 &&
 2772       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2773       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2774     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2775   } else {
 2776     Assembler::xorps(dst, src);
 2777   }
 2778 }
 2779 
 2780 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2781   assert(rscratch != noreg || always_reachable(src), "missing");
 2782 
 2783   // Used in sign-bit flipping with aligned address.
 2784   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2785 
 2786   if (UseAVX > 2 &&
 2787       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2788       (dst->encoding() >= 16)) {
 2789     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2790   } else if (reachable(src)) {
 2791     Assembler::xorps(dst, as_Address(src));
 2792   } else {
 2793     lea(rscratch, src);
 2794     Assembler::xorps(dst, Address(rscratch, 0));
 2795   }
 2796 }
 2797 
 2798 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2799   assert(rscratch != noreg || always_reachable(src), "missing");
 2800 
 2801   // Used in sign-bit flipping with aligned address.
 2802   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2803   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2804   if (reachable(src)) {
 2805     Assembler::pshufb(dst, as_Address(src));
 2806   } else {
 2807     lea(rscratch, src);
 2808     Assembler::pshufb(dst, Address(rscratch, 0));
 2809   }
 2810 }
 2811 
 2812 // AVX 3-operands instructions
 2813 
 2814 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2815   assert(rscratch != noreg || always_reachable(src), "missing");
 2816 
 2817   if (reachable(src)) {
 2818     vaddsd(dst, nds, as_Address(src));
 2819   } else {
 2820     lea(rscratch, src);
 2821     vaddsd(dst, nds, Address(rscratch, 0));
 2822   }
 2823 }
 2824 
 2825 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2826   assert(rscratch != noreg || always_reachable(src), "missing");
 2827 
 2828   if (reachable(src)) {
 2829     vaddss(dst, nds, as_Address(src));
 2830   } else {
 2831     lea(rscratch, src);
 2832     vaddss(dst, nds, Address(rscratch, 0));
 2833   }
 2834 }
 2835 
 2836 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2837   assert(UseAVX > 0, "requires some form of AVX");
 2838   assert(rscratch != noreg || always_reachable(src), "missing");
 2839 
 2840   if (reachable(src)) {
 2841     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2842   } else {
 2843     lea(rscratch, src);
 2844     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2845   }
 2846 }
 2847 
 2848 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2849   assert(UseAVX > 0, "requires some form of AVX");
 2850   assert(rscratch != noreg || always_reachable(src), "missing");
 2851 
 2852   if (reachable(src)) {
 2853     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2854   } else {
 2855     lea(rscratch, src);
 2856     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2857   }
 2858 }
 2859 
 2860 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2861   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2862   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2863 
 2864   vandps(dst, nds, negate_field, vector_len, rscratch);
 2865 }
 2866 
 2867 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2868   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2869   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2870 
 2871   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2872 }
 2873 
 2874 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2875   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2876   Assembler::vpaddb(dst, nds, src, vector_len);
 2877 }
 2878 
 2879 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2880   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2881   Assembler::vpaddb(dst, nds, src, vector_len);
 2882 }
 2883 
 2884 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2885   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2886   Assembler::vpaddw(dst, nds, src, vector_len);
 2887 }
 2888 
 2889 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2890   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2891   Assembler::vpaddw(dst, nds, src, vector_len);
 2892 }
 2893 
 2894 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2895   assert(rscratch != noreg || always_reachable(src), "missing");
 2896 
 2897   if (reachable(src)) {
 2898     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 2899   } else {
 2900     lea(rscratch, src);
 2901     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 2902   }
 2903 }
 2904 
 2905 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2906   assert(rscratch != noreg || always_reachable(src), "missing");
 2907 
 2908   if (reachable(src)) {
 2909     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 2910   } else {
 2911     lea(rscratch, src);
 2912     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 2913   }
 2914 }
 2915 
 2916 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2917   assert(rscratch != noreg || always_reachable(src), "missing");
 2918 
 2919   if (reachable(src)) {
 2920     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 2921   } else {
 2922     lea(rscratch, src);
 2923     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 2924   }
 2925 }
 2926 
 2927 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2928   assert(rscratch != noreg || always_reachable(src), "missing");
 2929 
 2930   if (reachable(src)) {
 2931     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 2932   } else {
 2933     lea(rscratch, src);
 2934     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 2935   }
 2936 }
 2937 
 2938 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2939   assert(rscratch != noreg || always_reachable(src), "missing");
 2940 
 2941   if (reachable(src)) {
 2942     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 2943   } else {
 2944     lea(rscratch, src);
 2945     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 2946   }
 2947 }
 2948 
 2949 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2950   assert(rscratch != noreg || always_reachable(src), "missing");
 2951 
 2952   if (reachable(src)) {
 2953     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 2954   } else {
 2955     lea(rscratch, src);
 2956     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 2957   }
 2958 }
 2959 
 2960 // Vector float blend
 2961 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2962 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2963   // WARN: Allow dst == (src1|src2), mask == scratch
 2964   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2965                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2966   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 2967   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2968   if (blend_emulation && scratch_available && dst_available) {
 2969     if (compute_mask) {
 2970       vpsrad(scratch, mask, 32, vector_len);
 2971       mask = scratch;
 2972     }
 2973     if (dst == src1) {
 2974       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 2975       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 2976     } else {
 2977       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 2978       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 2979     }
 2980     vpor(dst, dst, scratch, vector_len);
 2981   } else {
 2982     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 2983   }
 2984 }
 2985 
 2986 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2987 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2988   // WARN: Allow dst == (src1|src2), mask == scratch
 2989   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2990                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2991   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 2992   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2993   if (blend_emulation && scratch_available && dst_available) {
 2994     if (compute_mask) {
 2995       vpxor(scratch, scratch, scratch, vector_len);
 2996       vpcmpgtq(scratch, scratch, mask, vector_len);
 2997       mask = scratch;
 2998     }
 2999     if (dst == src1) {
 3000       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 3001       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3002     } else {
 3003       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3004       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 3005     }
 3006     vpor(dst, dst, scratch, vector_len);
 3007   } else {
 3008     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 3009   }
 3010 }
 3011 
 3012 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3013   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3014   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 3015 }
 3016 
 3017 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 3018   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3019   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 3020 }
 3021 
 3022 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3023   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3024   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3025 }
 3026 
 3027 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3028   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3029   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3030 }
 3031 
 3032 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3033   assert(rscratch != noreg || always_reachable(src), "missing");
 3034 
 3035   if (reachable(src)) {
 3036     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 3037   } else {
 3038     lea(rscratch, src);
 3039     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 3040   }
 3041 }
 3042 
 3043 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3044                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3045   assert(rscratch != noreg || always_reachable(src), "missing");
 3046 
 3047   if (reachable(src)) {
 3048     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3049   } else {
 3050     lea(rscratch, src);
 3051     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3052   }
 3053 }
 3054 
 3055 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3056                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3057   assert(rscratch != noreg || always_reachable(src), "missing");
 3058 
 3059   if (reachable(src)) {
 3060     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3061   } else {
 3062     lea(rscratch, src);
 3063     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3064   }
 3065 }
 3066 
 3067 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3068                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3069   assert(rscratch != noreg || always_reachable(src), "missing");
 3070 
 3071   if (reachable(src)) {
 3072     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3073   } else {
 3074     lea(rscratch, src);
 3075     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3076   }
 3077 }
 3078 
 3079 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3080                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3081   assert(rscratch != noreg || always_reachable(src), "missing");
 3082 
 3083   if (reachable(src)) {
 3084     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3085   } else {
 3086     lea(rscratch, src);
 3087     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3088   }
 3089 }
 3090 
 3091 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3092   if (width == Assembler::Q) {
 3093     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3094   } else {
 3095     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3096   }
 3097 }
 3098 
 3099 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3100   int eq_cond_enc = 0x29;
 3101   int gt_cond_enc = 0x37;
 3102   if (width != Assembler::Q) {
 3103     eq_cond_enc = 0x74 + width;
 3104     gt_cond_enc = 0x64 + width;
 3105   }
 3106   switch (cond) {
 3107   case eq:
 3108     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3109     break;
 3110   case neq:
 3111     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3112     vallones(xtmp, vector_len);
 3113     vpxor(dst, xtmp, dst, vector_len);
 3114     break;
 3115   case le:
 3116     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3117     vallones(xtmp, vector_len);
 3118     vpxor(dst, xtmp, dst, vector_len);
 3119     break;
 3120   case nlt:
 3121     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3122     vallones(xtmp, vector_len);
 3123     vpxor(dst, xtmp, dst, vector_len);
 3124     break;
 3125   case lt:
 3126     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3127     break;
 3128   case nle:
 3129     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3130     break;
 3131   default:
 3132     assert(false, "Should not reach here");
 3133   }
 3134 }
 3135 
 3136 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3137   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3138   Assembler::vpmovzxbw(dst, src, vector_len);
 3139 }
 3140 
 3141 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3142   assert((src->encoding() < 16),"XMM register should be 0-15");
 3143   Assembler::vpmovmskb(dst, src, vector_len);
 3144 }
 3145 
 3146 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3147   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3148   Assembler::vpmullw(dst, nds, src, vector_len);
 3149 }
 3150 
 3151 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3152   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3153   Assembler::vpmullw(dst, nds, src, vector_len);
 3154 }
 3155 
 3156 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3157   assert((UseAVX > 0), "AVX support is needed");
 3158   assert(rscratch != noreg || always_reachable(src), "missing");
 3159 
 3160   if (reachable(src)) {
 3161     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3162   } else {
 3163     lea(rscratch, src);
 3164     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3165   }
 3166 }
 3167 
 3168 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3169   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3170   Assembler::vpsubb(dst, nds, src, vector_len);
 3171 }
 3172 
 3173 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3174   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3175   Assembler::vpsubb(dst, nds, src, vector_len);
 3176 }
 3177 
 3178 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3179   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3180   Assembler::vpsubw(dst, nds, src, vector_len);
 3181 }
 3182 
 3183 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3184   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3185   Assembler::vpsubw(dst, nds, src, vector_len);
 3186 }
 3187 
 3188 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3189   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3190   Assembler::vpsraw(dst, nds, shift, vector_len);
 3191 }
 3192 
 3193 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3194   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3195   Assembler::vpsraw(dst, nds, shift, vector_len);
 3196 }
 3197 
 3198 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3199   assert(UseAVX > 2,"");
 3200   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3201      vector_len = 2;
 3202   }
 3203   Assembler::evpsraq(dst, nds, shift, vector_len);
 3204 }
 3205 
 3206 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3207   assert(UseAVX > 2,"");
 3208   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3209      vector_len = 2;
 3210   }
 3211   Assembler::evpsraq(dst, nds, shift, vector_len);
 3212 }
 3213 
 3214 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3215   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3216   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3217 }
 3218 
 3219 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3220   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3221   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3222 }
 3223 
 3224 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3225   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3226   Assembler::vpsllw(dst, nds, shift, vector_len);
 3227 }
 3228 
 3229 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3230   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3231   Assembler::vpsllw(dst, nds, shift, vector_len);
 3232 }
 3233 
 3234 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3235   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3236   Assembler::vptest(dst, src);
 3237 }
 3238 
 3239 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3240   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3241   Assembler::punpcklbw(dst, src);
 3242 }
 3243 
 3244 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3245   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3246   Assembler::pshufd(dst, src, mode);
 3247 }
 3248 
 3249 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3250   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3251   Assembler::pshuflw(dst, src, mode);
 3252 }
 3253 
 3254 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3255   assert(rscratch != noreg || always_reachable(src), "missing");
 3256 
 3257   if (reachable(src)) {
 3258     vandpd(dst, nds, as_Address(src), vector_len);
 3259   } else {
 3260     lea(rscratch, src);
 3261     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3262   }
 3263 }
 3264 
 3265 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3266   assert(rscratch != noreg || always_reachable(src), "missing");
 3267 
 3268   if (reachable(src)) {
 3269     vandps(dst, nds, as_Address(src), vector_len);
 3270   } else {
 3271     lea(rscratch, src);
 3272     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3273   }
 3274 }
 3275 
 3276 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3277                             bool merge, int vector_len, Register rscratch) {
 3278   assert(rscratch != noreg || always_reachable(src), "missing");
 3279 
 3280   if (reachable(src)) {
 3281     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3282   } else {
 3283     lea(rscratch, src);
 3284     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3285   }
 3286 }
 3287 
 3288 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3289   assert(rscratch != noreg || always_reachable(src), "missing");
 3290 
 3291   if (reachable(src)) {
 3292     vdivsd(dst, nds, as_Address(src));
 3293   } else {
 3294     lea(rscratch, src);
 3295     vdivsd(dst, nds, Address(rscratch, 0));
 3296   }
 3297 }
 3298 
 3299 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3300   assert(rscratch != noreg || always_reachable(src), "missing");
 3301 
 3302   if (reachable(src)) {
 3303     vdivss(dst, nds, as_Address(src));
 3304   } else {
 3305     lea(rscratch, src);
 3306     vdivss(dst, nds, Address(rscratch, 0));
 3307   }
 3308 }
 3309 
 3310 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3311   assert(rscratch != noreg || always_reachable(src), "missing");
 3312 
 3313   if (reachable(src)) {
 3314     vmulsd(dst, nds, as_Address(src));
 3315   } else {
 3316     lea(rscratch, src);
 3317     vmulsd(dst, nds, Address(rscratch, 0));
 3318   }
 3319 }
 3320 
 3321 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3322   assert(rscratch != noreg || always_reachable(src), "missing");
 3323 
 3324   if (reachable(src)) {
 3325     vmulss(dst, nds, as_Address(src));
 3326   } else {
 3327     lea(rscratch, src);
 3328     vmulss(dst, nds, Address(rscratch, 0));
 3329   }
 3330 }
 3331 
 3332 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3333   assert(rscratch != noreg || always_reachable(src), "missing");
 3334 
 3335   if (reachable(src)) {
 3336     vsubsd(dst, nds, as_Address(src));
 3337   } else {
 3338     lea(rscratch, src);
 3339     vsubsd(dst, nds, Address(rscratch, 0));
 3340   }
 3341 }
 3342 
 3343 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3344   assert(rscratch != noreg || always_reachable(src), "missing");
 3345 
 3346   if (reachable(src)) {
 3347     vsubss(dst, nds, as_Address(src));
 3348   } else {
 3349     lea(rscratch, src);
 3350     vsubss(dst, nds, Address(rscratch, 0));
 3351   }
 3352 }
 3353 
 3354 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3355   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3356   assert(rscratch != noreg || always_reachable(src), "missing");
 3357 
 3358   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3359 }
 3360 
 3361 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3362   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3363   assert(rscratch != noreg || always_reachable(src), "missing");
 3364 
 3365   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3366 }
 3367 
 3368 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3369   assert(rscratch != noreg || always_reachable(src), "missing");
 3370 
 3371   if (reachable(src)) {
 3372     vxorpd(dst, nds, as_Address(src), vector_len);
 3373   } else {
 3374     lea(rscratch, src);
 3375     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3376   }
 3377 }
 3378 
 3379 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3380   assert(rscratch != noreg || always_reachable(src), "missing");
 3381 
 3382   if (reachable(src)) {
 3383     vxorps(dst, nds, as_Address(src), vector_len);
 3384   } else {
 3385     lea(rscratch, src);
 3386     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3387   }
 3388 }
 3389 
 3390 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3391   assert(rscratch != noreg || always_reachable(src), "missing");
 3392 
 3393   if (UseAVX > 1 || (vector_len < 1)) {
 3394     if (reachable(src)) {
 3395       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3396     } else {
 3397       lea(rscratch, src);
 3398       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3399     }
 3400   } else {
 3401     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3402   }
 3403 }
 3404 
 3405 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3406   assert(rscratch != noreg || always_reachable(src), "missing");
 3407 
 3408   if (reachable(src)) {
 3409     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3410   } else {
 3411     lea(rscratch, src);
 3412     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3413   }
 3414 }
 3415 
 3416 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3417   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3418   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3419   // The inverted mask is sign-extended
 3420   andptr(possibly_non_local, inverted_mask);
 3421 }
 3422 
 3423 void MacroAssembler::resolve_jobject(Register value,
 3424                                      Register tmp) {
 3425   Register thread = r15_thread;
 3426   assert_different_registers(value, thread, tmp);
 3427   Label done, tagged, weak_tagged;
 3428   testptr(value, value);
 3429   jcc(Assembler::zero, done);           // Use null as-is.
 3430   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3431   jcc(Assembler::notZero, tagged);
 3432 
 3433   // Resolve local handle
 3434   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3435   verify_oop(value);
 3436   jmp(done);
 3437 
 3438   bind(tagged);
 3439   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3440   jcc(Assembler::notZero, weak_tagged);
 3441 
 3442   // Resolve global handle
 3443   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3444   verify_oop(value);
 3445   jmp(done);
 3446 
 3447   bind(weak_tagged);
 3448   // Resolve jweak.
 3449   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3450                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3451   verify_oop(value);
 3452 
 3453   bind(done);
 3454 }
 3455 
 3456 void MacroAssembler::resolve_global_jobject(Register value,
 3457                                             Register tmp) {
 3458   Register thread = r15_thread;
 3459   assert_different_registers(value, thread, tmp);
 3460   Label done;
 3461 
 3462   testptr(value, value);
 3463   jcc(Assembler::zero, done);           // Use null as-is.
 3464 
 3465 #ifdef ASSERT
 3466   {
 3467     Label valid_global_tag;
 3468     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3469     jcc(Assembler::notZero, valid_global_tag);
 3470     stop("non global jobject using resolve_global_jobject");
 3471     bind(valid_global_tag);
 3472   }
 3473 #endif
 3474 
 3475   // Resolve global handle
 3476   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3477   verify_oop(value);
 3478 
 3479   bind(done);
 3480 }
 3481 
 3482 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3483   subq(dst, imm32);
 3484 }
 3485 
 3486 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3487 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3488   subq_imm32(dst, imm32);
 3489 }
 3490 
 3491 void MacroAssembler::subptr(Register dst, Register src) {
 3492   subq(dst, src);
 3493 }
 3494 
 3495 // C++ bool manipulation
 3496 void MacroAssembler::testbool(Register dst) {
 3497   if(sizeof(bool) == 1)
 3498     testb(dst, 0xff);
 3499   else if(sizeof(bool) == 2) {
 3500     // testw implementation needed for two byte bools
 3501     ShouldNotReachHere();
 3502   } else if(sizeof(bool) == 4)
 3503     testl(dst, dst);
 3504   else
 3505     // unsupported
 3506     ShouldNotReachHere();
 3507 }
 3508 
 3509 void MacroAssembler::testptr(Register dst, Register src) {
 3510   testq(dst, src);
 3511 }
 3512 






















































































































 3513 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3514 void MacroAssembler::tlab_allocate(Register obj,
 3515                                    Register var_size_in_bytes,
 3516                                    int con_size_in_bytes,
 3517                                    Register t1,
 3518                                    Register t2,
 3519                                    Label& slow_case) {
 3520   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3521   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3522 }
 3523 
 3524 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3525   RegSet regs;
 3526   regs += RegSet::of(rax, rcx, rdx);
 3527 #ifndef _WINDOWS
 3528   regs += RegSet::of(rsi, rdi);
 3529 #endif
 3530   regs += RegSet::range(r8, r11);
 3531   if (UseAPX) {
 3532     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3533   }
 3534   return regs;
 3535 }
 3536 
 3537 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3538   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3539 #if defined(_WINDOWS)
 3540   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3541   if (num_xmm_registers > 16) {
 3542      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3543   }
 3544   return result;
 3545 #else
 3546   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3547 #endif
 3548 }
 3549 
 3550 // C1 only ever uses the first double/float of the XMM register.
 3551 static int xmm_save_size() { return sizeof(double); }
 3552 
 3553 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3554   masm->movdbl(Address(rsp, offset), reg);
 3555 }
 3556 
 3557 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3558   masm->movdbl(reg, Address(rsp, offset));
 3559 }
 3560 
 3561 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3562                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3563 
 3564   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3565                          StackAlignmentInBytes);
 3566   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3567 
 3568   return gp_area_size + xmm_area_size;
 3569 }
 3570 
 3571 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3572   block_comment("push_call_clobbered_registers start");
 3573   // Regular registers
 3574   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3575 
 3576   int gp_area_size;
 3577   int xmm_area_size;
 3578   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3579                                                gp_area_size, xmm_area_size);
 3580   subptr(rsp, total_save_size);
 3581 
 3582   push_set(gp_registers_to_push, 0);
 3583 
 3584   if (save_fpu) {
 3585     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3586   }
 3587 
 3588   block_comment("push_call_clobbered_registers end");
 3589 }
 3590 
 3591 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3592   block_comment("pop_call_clobbered_registers start");
 3593 
 3594   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 3595 
 3596   int gp_area_size;
 3597   int xmm_area_size;
 3598   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 3599                                                gp_area_size, xmm_area_size);
 3600 
 3601   if (restore_fpu) {
 3602     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3603   }
 3604 
 3605   pop_set(gp_registers_to_pop, 0);
 3606 
 3607   addptr(rsp, total_save_size);
 3608 
 3609   vzeroupper();
 3610 
 3611   block_comment("pop_call_clobbered_registers end");
 3612 }
 3613 
 3614 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3615   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3616   int spill_offset = offset;
 3617 
 3618   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3619     save_xmm_register(this, spill_offset, *it);
 3620     spill_offset += xmm_save_size();
 3621   }
 3622 }
 3623 
 3624 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3625   int restore_size = set.size() * xmm_save_size();
 3626   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3627 
 3628   int restore_offset = offset + restore_size - xmm_save_size();
 3629 
 3630   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3631     restore_xmm_register(this, restore_offset, *it);
 3632     restore_offset -= xmm_save_size();
 3633   }
 3634 }
 3635 
 3636 void MacroAssembler::push_set(RegSet set, int offset) {
 3637   int spill_offset;
 3638   if (offset == -1) {
 3639     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3640     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3641     subptr(rsp, aligned_size);
 3642     spill_offset = 0;
 3643   } else {
 3644     spill_offset = offset;
 3645   }
 3646 
 3647   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3648     movptr(Address(rsp, spill_offset), *it);
 3649     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3650   }
 3651 }
 3652 
 3653 void MacroAssembler::pop_set(RegSet set, int offset) {
 3654 
 3655   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3656   int restore_size = set.size() * gp_reg_size;
 3657   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3658 
 3659   int restore_offset;
 3660   if (offset == -1) {
 3661     restore_offset = restore_size - gp_reg_size;
 3662   } else {
 3663     restore_offset = offset + restore_size - gp_reg_size;
 3664   }
 3665   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3666     movptr(*it, Address(rsp, restore_offset));
 3667     restore_offset -= gp_reg_size;
 3668   }
 3669 
 3670   if (offset == -1) {
 3671     addptr(rsp, aligned_size);
 3672   }
 3673 }
 3674 
 3675 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3676 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3677   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3678   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3679   Label done;
 3680 
 3681   testptr(length_in_bytes, length_in_bytes);
 3682   jcc(Assembler::zero, done);
 3683 
 3684   // initialize topmost word, divide index by 2, check if odd and test if zero
 3685   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3686 #ifdef ASSERT
 3687   {
 3688     Label L;
 3689     testptr(length_in_bytes, BytesPerWord - 1);
 3690     jcc(Assembler::zero, L);
 3691     stop("length must be a multiple of BytesPerWord");
 3692     bind(L);
 3693   }
 3694 #endif
 3695   Register index = length_in_bytes;
 3696   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3697   if (UseIncDec) {
 3698     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3699   } else {
 3700     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3701     shrptr(index, 1);
 3702   }
 3703 
 3704   // initialize remaining object fields: index is a multiple of 2 now
 3705   {
 3706     Label loop;
 3707     bind(loop);
 3708     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3709     decrement(index);
 3710     jcc(Assembler::notZero, loop);
 3711   }
 3712 
 3713   bind(done);
 3714 }
 3715 






















 3716 // Look up the method for a megamorphic invokeinterface call.
 3717 // The target method is determined by <intf_klass, itable_index>.
 3718 // The receiver klass is in recv_klass.
 3719 // On success, the result will be in method_result, and execution falls through.
 3720 // On failure, execution transfers to the given label.
 3721 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3722                                              Register intf_klass,
 3723                                              RegisterOrConstant itable_index,
 3724                                              Register method_result,
 3725                                              Register scan_temp,
 3726                                              Label& L_no_such_interface,
 3727                                              bool return_method) {
 3728   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3729   assert_different_registers(method_result, intf_klass, scan_temp);
 3730   assert(recv_klass != method_result || !return_method,
 3731          "recv_klass can be destroyed when method isn't needed");
 3732 
 3733   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3734          "caller must use same register for non-constant itable index as for method");
 3735 
 3736   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3737   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3738   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3739   int scan_step   = itableOffsetEntry::size() * wordSize;
 3740   int vte_size    = vtableEntry::size_in_bytes();
 3741   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3742   assert(vte_size == wordSize, "else adjust times_vte_scale");
 3743 
 3744   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3745 
 3746   // Could store the aligned, prescaled offset in the klass.
 3747   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 3748 
 3749   if (return_method) {
 3750     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 3751     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3752     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 3753   }
 3754 
 3755   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 3756   //   if (scan->interface() == intf) {
 3757   //     result = (klass + scan->offset() + itable_index);
 3758   //   }
 3759   // }
 3760   Label search, found_method;
 3761 
 3762   for (int peel = 1; peel >= 0; peel--) {
 3763     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 3764     cmpptr(intf_klass, method_result);
 3765 
 3766     if (peel) {
 3767       jccb(Assembler::equal, found_method);
 3768     } else {
 3769       jccb(Assembler::notEqual, search);
 3770       // (invert the test to fall through to found_method...)
 3771     }
 3772 
 3773     if (!peel)  break;
 3774 
 3775     bind(search);
 3776 
 3777     // Check that the previous entry is non-null.  A null entry means that
 3778     // the receiver class doesn't implement the interface, and wasn't the
 3779     // same as when the caller was compiled.
 3780     testptr(method_result, method_result);
 3781     jcc(Assembler::zero, L_no_such_interface);
 3782     addptr(scan_temp, scan_step);
 3783   }
 3784 
 3785   bind(found_method);
 3786 
 3787   if (return_method) {
 3788     // Got a hit.
 3789     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 3790     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 3791   }
 3792 }
 3793 
 3794 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 3795 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 3796 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 3797 // The target method is determined by <holder_klass, itable_index>.
 3798 // The receiver klass is in recv_klass.
 3799 // On success, the result will be in method_result, and execution falls through.
 3800 // On failure, execution transfers to the given label.
 3801 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 3802                                                   Register holder_klass,
 3803                                                   Register resolved_klass,
 3804                                                   Register method_result,
 3805                                                   Register scan_temp,
 3806                                                   Register temp_reg2,
 3807                                                   Register receiver,
 3808                                                   int itable_index,
 3809                                                   Label& L_no_such_interface) {
 3810   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 3811   Register temp_itbl_klass = method_result;
 3812   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 3813 
 3814   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3815   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3816   int scan_step = itableOffsetEntry::size() * wordSize;
 3817   int vte_size = vtableEntry::size_in_bytes();
 3818   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 3819   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 3820   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3821   assert(vte_size == wordSize, "adjust times_vte_scale");
 3822 
 3823   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 3824 
 3825   // temp_itbl_klass = recv_klass.itable[0]
 3826   // scan_temp = &recv_klass.itable[0] + step
 3827   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3828   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 3829   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 3830   xorptr(temp_reg, temp_reg);
 3831 
 3832   // Initial checks:
 3833   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 3834   //   - if (itable[0] == 0), no such interface
 3835   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 3836   cmpptr(holder_klass, resolved_klass);
 3837   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 3838   testptr(temp_itbl_klass, temp_itbl_klass);
 3839   jccb(Assembler::zero, L_no_such_interface);
 3840   cmpptr(holder_klass, temp_itbl_klass);
 3841   jccb(Assembler::equal, L_holder_found);
 3842 
 3843   // Loop: Look for holder_klass record in itable
 3844   //   do {
 3845   //     tmp = itable[index];
 3846   //     index += step;
 3847   //     if (tmp == holder_klass) {
 3848   //       goto L_holder_found; // Found!
 3849   //     }
 3850   //   } while (tmp != 0);
 3851   //   goto L_no_such_interface // Not found.
 3852   Label L_scan_holder;
 3853   bind(L_scan_holder);
 3854     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3855     addptr(scan_temp, scan_step);
 3856     cmpptr(holder_klass, temp_itbl_klass);
 3857     jccb(Assembler::equal, L_holder_found);
 3858     testptr(temp_itbl_klass, temp_itbl_klass);
 3859     jccb(Assembler::notZero, L_scan_holder);
 3860 
 3861   jmpb(L_no_such_interface);
 3862 
 3863   // Loop: Look for resolved_class record in itable
 3864   //   do {
 3865   //     tmp = itable[index];
 3866   //     index += step;
 3867   //     if (tmp == holder_klass) {
 3868   //        // Also check if we have met a holder klass
 3869   //        holder_tmp = itable[index-step-ioffset];
 3870   //     }
 3871   //     if (tmp == resolved_klass) {
 3872   //        goto L_resolved_found;  // Found!
 3873   //     }
 3874   //   } while (tmp != 0);
 3875   //   goto L_no_such_interface // Not found.
 3876   //
 3877   Label L_loop_scan_resolved;
 3878   bind(L_loop_scan_resolved);
 3879     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3880     addptr(scan_temp, scan_step);
 3881     bind(L_loop_scan_resolved_entry);
 3882     cmpptr(holder_klass, temp_itbl_klass);
 3883     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3884     cmpptr(resolved_klass, temp_itbl_klass);
 3885     jccb(Assembler::equal, L_resolved_found);
 3886     testptr(temp_itbl_klass, temp_itbl_klass);
 3887     jccb(Assembler::notZero, L_loop_scan_resolved);
 3888 
 3889   jmpb(L_no_such_interface);
 3890 
 3891   Label L_ready;
 3892 
 3893   // See if we already have a holder klass. If not, go and scan for it.
 3894   bind(L_resolved_found);
 3895   testptr(temp_reg, temp_reg);
 3896   jccb(Assembler::zero, L_scan_holder);
 3897   jmpb(L_ready);
 3898 
 3899   bind(L_holder_found);
 3900   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3901 
 3902   // Finally, temp_reg contains holder_klass vtable offset
 3903   bind(L_ready);
 3904   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3905   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 3906     load_klass(scan_temp, receiver, noreg);
 3907     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3908   } else {
 3909     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3910   }
 3911 }
 3912 
 3913 
 3914 // virtual method calling
 3915 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 3916                                            RegisterOrConstant vtable_index,
 3917                                            Register method_result) {
 3918   const ByteSize base = Klass::vtable_start_offset();
 3919   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 3920   Address vtable_entry_addr(recv_klass,
 3921                             vtable_index, Address::times_ptr,
 3922                             base + vtableEntry::method_offset());
 3923   movptr(method_result, vtable_entry_addr);
 3924 }
 3925 
 3926 
 3927 void MacroAssembler::check_klass_subtype(Register sub_klass,
 3928                            Register super_klass,
 3929                            Register temp_reg,
 3930                            Label& L_success) {
 3931   Label L_failure;
 3932   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 3933   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 3934   bind(L_failure);
 3935 }
 3936 
 3937 
 3938 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 3939                                                    Register super_klass,
 3940                                                    Register temp_reg,
 3941                                                    Label* L_success,
 3942                                                    Label* L_failure,
 3943                                                    Label* L_slow_path,
 3944                                         RegisterOrConstant super_check_offset) {
 3945   assert_different_registers(sub_klass, super_klass, temp_reg);
 3946   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 3947   if (super_check_offset.is_register()) {
 3948     assert_different_registers(sub_klass, super_klass,
 3949                                super_check_offset.as_register());
 3950   } else if (must_load_sco) {
 3951     assert(temp_reg != noreg, "supply either a temp or a register offset");
 3952   }
 3953 
 3954   Label L_fallthrough;
 3955   int label_nulls = 0;
 3956   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 3957   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 3958   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 3959   assert(label_nulls <= 1, "at most one null in the batch");
 3960 
 3961   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 3962   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 3963   Address super_check_offset_addr(super_klass, sco_offset);
 3964 
 3965   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 3966   // range of a jccb.  If this routine grows larger, reconsider at
 3967   // least some of these.
 3968 #define local_jcc(assembler_cond, label)                                \
 3969   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 3970   else                             jcc( assembler_cond, label) /*omit semi*/
 3971 
 3972   // Hacked jmp, which may only be used just before L_fallthrough.
 3973 #define final_jmp(label)                                                \
 3974   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 3975   else                            jmp(label)                /*omit semi*/
 3976 
 3977   // If the pointers are equal, we are done (e.g., String[] elements).
 3978   // This self-check enables sharing of secondary supertype arrays among
 3979   // non-primary types such as array-of-interface.  Otherwise, each such
 3980   // type would need its own customized SSA.
 3981   // We move this check to the front of the fast path because many
 3982   // type checks are in fact trivially successful in this manner,
 3983   // so we get a nicely predicted branch right at the start of the check.
 3984   cmpptr(sub_klass, super_klass);
 3985   local_jcc(Assembler::equal, *L_success);
 3986 
 3987   // Check the supertype display:
 3988   if (must_load_sco) {
 3989     // Positive movl does right thing on LP64.
 3990     movl(temp_reg, super_check_offset_addr);
 3991     super_check_offset = RegisterOrConstant(temp_reg);
 3992   }
 3993   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 3994   cmpptr(super_klass, super_check_addr); // load displayed supertype
 3995 
 3996   // This check has worked decisively for primary supers.
 3997   // Secondary supers are sought in the super_cache ('super_cache_addr').
 3998   // (Secondary supers are interfaces and very deeply nested subtypes.)
 3999   // This works in the same check above because of a tricky aliasing
 4000   // between the super_cache and the primary super display elements.
 4001   // (The 'super_check_addr' can address either, as the case requires.)
 4002   // Note that the cache is updated below if it does not help us find
 4003   // what we need immediately.
 4004   // So if it was a primary super, we can just fail immediately.
 4005   // Otherwise, it's the slow path for us (no success at this point).
 4006 
 4007   if (super_check_offset.is_register()) {
 4008     local_jcc(Assembler::equal, *L_success);
 4009     cmpl(super_check_offset.as_register(), sc_offset);
 4010     if (L_failure == &L_fallthrough) {
 4011       local_jcc(Assembler::equal, *L_slow_path);
 4012     } else {
 4013       local_jcc(Assembler::notEqual, *L_failure);
 4014       final_jmp(*L_slow_path);
 4015     }
 4016   } else if (super_check_offset.as_constant() == sc_offset) {
 4017     // Need a slow path; fast failure is impossible.
 4018     if (L_slow_path == &L_fallthrough) {
 4019       local_jcc(Assembler::equal, *L_success);
 4020     } else {
 4021       local_jcc(Assembler::notEqual, *L_slow_path);
 4022       final_jmp(*L_success);
 4023     }
 4024   } else {
 4025     // No slow path; it's a fast decision.
 4026     if (L_failure == &L_fallthrough) {
 4027       local_jcc(Assembler::equal, *L_success);
 4028     } else {
 4029       local_jcc(Assembler::notEqual, *L_failure);
 4030       final_jmp(*L_success);
 4031     }
 4032   }
 4033 
 4034   bind(L_fallthrough);
 4035 
 4036 #undef local_jcc
 4037 #undef final_jmp
 4038 }
 4039 
 4040 
 4041 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 4042                                                           Register super_klass,
 4043                                                           Register temp_reg,
 4044                                                           Register temp2_reg,
 4045                                                           Label* L_success,
 4046                                                           Label* L_failure,
 4047                                                           bool set_cond_codes) {
 4048   assert_different_registers(sub_klass, super_klass, temp_reg);
 4049   if (temp2_reg != noreg)
 4050     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4051 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4052 
 4053   Label L_fallthrough;
 4054   int label_nulls = 0;
 4055   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4056   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4057   assert(label_nulls <= 1, "at most one null in the batch");
 4058 
 4059   // a couple of useful fields in sub_klass:
 4060   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4061   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4062   Address secondary_supers_addr(sub_klass, ss_offset);
 4063   Address super_cache_addr(     sub_klass, sc_offset);
 4064 
 4065   // Do a linear scan of the secondary super-klass chain.
 4066   // This code is rarely used, so simplicity is a virtue here.
 4067   // The repne_scan instruction uses fixed registers, which we must spill.
 4068   // Don't worry too much about pre-existing connections with the input regs.
 4069 
 4070   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4071   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4072 
 4073   // Get super_klass value into rax (even if it was in rdi or rcx).
 4074   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4075   if (super_klass != rax) {
 4076     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4077     mov(rax, super_klass);
 4078   }
 4079   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4080   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4081 
 4082 #ifndef PRODUCT
 4083   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4084   ExternalAddress pst_counter_addr((address) pst_counter);
 4085   lea(rcx, pst_counter_addr);
 4086   incrementl(Address(rcx, 0));
 4087 #endif //PRODUCT
 4088 
 4089   // We will consult the secondary-super array.
 4090   movptr(rdi, secondary_supers_addr);
 4091   // Load the array length.  (Positive movl does right thing on LP64.)
 4092   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4093   // Skip to start of data.
 4094   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4095 
 4096   // Scan RCX words at [RDI] for an occurrence of RAX.
 4097   // Set NZ/Z based on last compare.
 4098   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4099   // not change flags (only scas instruction which is repeated sets flags).
 4100   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4101 
 4102     testptr(rax,rax); // Set Z = 0
 4103     repne_scan();
 4104 
 4105   // Unspill the temp. registers:
 4106   if (pushed_rdi)  pop(rdi);
 4107   if (pushed_rcx)  pop(rcx);
 4108   if (pushed_rax)  pop(rax);
 4109 
 4110   if (set_cond_codes) {
 4111     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4112     assert(!pushed_rdi, "rdi must be left non-null");
 4113     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4114   }
 4115 
 4116   if (L_failure == &L_fallthrough)
 4117         jccb(Assembler::notEqual, *L_failure);
 4118   else  jcc(Assembler::notEqual, *L_failure);
 4119 
 4120   // Success.  Cache the super we found and proceed in triumph.
 4121   movptr(super_cache_addr, super_klass);
 4122 
 4123   if (L_success != &L_fallthrough) {
 4124     jmp(*L_success);
 4125   }
 4126 
 4127 #undef IS_A_TEMP
 4128 
 4129   bind(L_fallthrough);
 4130 }
 4131 
 4132 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4133                                                    Register super_klass,
 4134                                                    Register temp_reg,
 4135                                                    Register temp2_reg,
 4136                                                    Label* L_success,
 4137                                                    Label* L_failure,
 4138                                                    bool set_cond_codes) {
 4139   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4140   check_klass_subtype_slow_path
 4141     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4142      L_success, L_failure);
 4143 }
 4144 
 4145 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4146                                                    Register super_klass,
 4147                                                    Register temp_reg,
 4148                                                    Register temp2_reg,
 4149                                                    Register temp3_reg,
 4150                                                    Register temp4_reg,
 4151                                                    Label* L_success,
 4152                                                    Label* L_failure) {
 4153   if (UseSecondarySupersTable) {
 4154     check_klass_subtype_slow_path_table
 4155       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4156        L_success, L_failure);
 4157   } else {
 4158     check_klass_subtype_slow_path_linear
 4159       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4160   }
 4161 }
 4162 
 4163 Register MacroAssembler::allocate_if_noreg(Register r,
 4164                                   RegSetIterator<Register> &available_regs,
 4165                                   RegSet &regs_to_push) {
 4166   if (!r->is_valid()) {
 4167     r = *available_regs++;
 4168     regs_to_push += r;
 4169   }
 4170   return r;
 4171 }
 4172 
 4173 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4174                                                          Register super_klass,
 4175                                                          Register temp_reg,
 4176                                                          Register temp2_reg,
 4177                                                          Register temp3_reg,
 4178                                                          Register result_reg,
 4179                                                          Label* L_success,
 4180                                                          Label* L_failure) {
 4181   // NB! Callers may assume that, when temp2_reg is a valid register,
 4182   // this code sets it to a nonzero value.
 4183   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4184 
 4185   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4186 
 4187   Label L_fallthrough;
 4188   int label_nulls = 0;
 4189   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4190   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4191   assert(label_nulls <= 1, "at most one null in the batch");
 4192 
 4193   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4194 
 4195   RegSetIterator<Register> available_regs
 4196     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4197 
 4198   RegSet pushed_regs;
 4199 
 4200   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4201   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4202   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4203   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4204   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4205 
 4206   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4207 
 4208   {
 4209 
 4210     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4211     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4212     subptr(rsp, aligned_size);
 4213     push_set(pushed_regs, 0);
 4214 
 4215     lookup_secondary_supers_table_var(sub_klass,
 4216                                       super_klass,
 4217                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4218     cmpq(result_reg, 0);
 4219 
 4220     // Unspill the temp. registers:
 4221     pop_set(pushed_regs, 0);
 4222     // Increment SP but do not clobber flags.
 4223     lea(rsp, Address(rsp, aligned_size));
 4224   }
 4225 
 4226   if (temp2_reg_was_valid) {
 4227     movq(temp2_reg, 1);
 4228   }
 4229 
 4230   jcc(Assembler::notEqual, *L_failure);
 4231 
 4232   if (L_success != &L_fallthrough) {
 4233     jmp(*L_success);
 4234   }
 4235 
 4236   bind(L_fallthrough);
 4237 }
 4238 
 4239 // population_count variant for running without the POPCNT
 4240 // instruction, which was introduced with SSE4.2 in 2008.
 4241 void MacroAssembler::population_count(Register dst, Register src,
 4242                                       Register scratch1, Register scratch2) {
 4243   assert_different_registers(src, scratch1, scratch2);
 4244   if (UsePopCountInstruction) {
 4245     Assembler::popcntq(dst, src);
 4246   } else {
 4247     assert_different_registers(src, scratch1, scratch2);
 4248     assert_different_registers(dst, scratch1, scratch2);
 4249     Label loop, done;
 4250 
 4251     mov(scratch1, src);
 4252     // dst = 0;
 4253     // while(scratch1 != 0) {
 4254     //   dst++;
 4255     //   scratch1 &= (scratch1 - 1);
 4256     // }
 4257     xorl(dst, dst);
 4258     testq(scratch1, scratch1);
 4259     jccb(Assembler::equal, done);
 4260     {
 4261       bind(loop);
 4262       incq(dst);
 4263       movq(scratch2, scratch1);
 4264       decq(scratch2);
 4265       andq(scratch1, scratch2);
 4266       jccb(Assembler::notEqual, loop);
 4267     }
 4268     bind(done);
 4269   }
 4270 #ifdef ASSERT
 4271   mov64(scratch1, 0xCafeBabeDeadBeef);
 4272   movq(scratch2, scratch1);
 4273 #endif
 4274 }
 4275 
 4276 // Ensure that the inline code and the stub are using the same registers.
 4277 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4278 do {                                                                 \
 4279   assert(r_super_klass  == rax, "mismatch");                         \
 4280   assert(r_array_base   == rbx, "mismatch");                         \
 4281   assert(r_array_length == rcx, "mismatch");                         \
 4282   assert(r_array_index  == rdx, "mismatch");                         \
 4283   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4284   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4285   assert(result         == rdi || result      == noreg, "mismatch"); \
 4286 } while(0)
 4287 
 4288 // Versions of salq and rorq that don't need count to be in rcx
 4289 
 4290 void MacroAssembler::salq(Register dest, Register count) {
 4291   if (count == rcx) {
 4292     Assembler::salq(dest);
 4293   } else {
 4294     assert_different_registers(rcx, dest);
 4295     xchgq(rcx, count);
 4296     Assembler::salq(dest);
 4297     xchgq(rcx, count);
 4298   }
 4299 }
 4300 
 4301 void MacroAssembler::rorq(Register dest, Register count) {
 4302   if (count == rcx) {
 4303     Assembler::rorq(dest);
 4304   } else {
 4305     assert_different_registers(rcx, dest);
 4306     xchgq(rcx, count);
 4307     Assembler::rorq(dest);
 4308     xchgq(rcx, count);
 4309   }
 4310 }
 4311 
 4312 // Return true: we succeeded in generating this code
 4313 //
 4314 // At runtime, return 0 in result if r_super_klass is a superclass of
 4315 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4316 // super_klass_slot of the class you're looking for. This is always
 4317 // the case for instanceof and checkcast.
 4318 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4319                                                          Register r_super_klass,
 4320                                                          Register temp1,
 4321                                                          Register temp2,
 4322                                                          Register temp3,
 4323                                                          Register temp4,
 4324                                                          Register result,
 4325                                                          u1 super_klass_slot) {
 4326   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4327 
 4328   Label L_fallthrough, L_success, L_failure;
 4329 
 4330   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4331 
 4332   const Register
 4333     r_array_index  = temp1,
 4334     r_array_length = temp2,
 4335     r_array_base   = temp3,
 4336     r_bitmap       = temp4;
 4337 
 4338   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4339 
 4340   xorq(result, result); // = 0
 4341 
 4342   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4343   movq(r_array_index, r_bitmap);
 4344 
 4345   // First check the bitmap to see if super_klass might be present. If
 4346   // the bit is zero, we are certain that super_klass is not one of
 4347   // the secondary supers.
 4348   u1 bit = super_klass_slot;
 4349   {
 4350     // NB: If the count in a x86 shift instruction is 0, the flags are
 4351     // not affected, so we do a testq instead.
 4352     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4353     if (shift_count != 0) {
 4354       salq(r_array_index, shift_count);
 4355     } else {
 4356       testq(r_array_index, r_array_index);
 4357     }
 4358   }
 4359   // We test the MSB of r_array_index, i.e. its sign bit
 4360   jcc(Assembler::positive, L_failure);
 4361 
 4362   // Get the first array index that can contain super_klass into r_array_index.
 4363   if (bit != 0) {
 4364     population_count(r_array_index, r_array_index, temp2, temp3);
 4365   } else {
 4366     movl(r_array_index, 1);
 4367   }
 4368   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4369 
 4370   // We will consult the secondary-super array.
 4371   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4372 
 4373   // We're asserting that the first word in an Array<Klass*> is the
 4374   // length, and the second word is the first word of the data. If
 4375   // that ever changes, r_array_base will have to be adjusted here.
 4376   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4377   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4378 
 4379   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4380   jccb(Assembler::equal, L_success);
 4381 
 4382   // Is there another entry to check? Consult the bitmap.
 4383   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4384   jccb(Assembler::carryClear, L_failure);
 4385 
 4386   // Linear probe. Rotate the bitmap so that the next bit to test is
 4387   // in Bit 1.
 4388   if (bit != 0) {
 4389     rorq(r_bitmap, bit);
 4390   }
 4391 
 4392   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4393   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4394   // Kills: r_array_length.
 4395   // Returns: result.
 4396   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4397   // Result (0/1) is in rdi
 4398   jmpb(L_fallthrough);
 4399 
 4400   bind(L_failure);
 4401   incq(result); // 0 => 1
 4402 
 4403   bind(L_success);
 4404   // result = 0;
 4405 
 4406   bind(L_fallthrough);
 4407   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4408 
 4409   if (VerifySecondarySupers) {
 4410     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4411                                   temp1, temp2, temp3);
 4412   }
 4413 }
 4414 
 4415 // At runtime, return 0 in result if r_super_klass is a superclass of
 4416 // r_sub_klass, otherwise return nonzero. Use this version of
 4417 // lookup_secondary_supers_table() if you don't know ahead of time
 4418 // which superclass will be searched for. Used by interpreter and
 4419 // runtime stubs. It is larger and has somewhat greater latency than
 4420 // the version above, which takes a constant super_klass_slot.
 4421 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4422                                                        Register r_super_klass,
 4423                                                        Register temp1,
 4424                                                        Register temp2,
 4425                                                        Register temp3,
 4426                                                        Register temp4,
 4427                                                        Register result) {
 4428   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4429   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4430   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4431 
 4432   Label L_fallthrough, L_success, L_failure;
 4433 
 4434   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4435 
 4436   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4437 
 4438   // FIXME. Once we are sure that all paths reaching this point really
 4439   // do pass rcx as one of our temps we can get rid of the following
 4440   // workaround.
 4441   assert(temps.contains(rcx), "fix this code");
 4442 
 4443   // We prefer to have our shift count in rcx. If rcx is one of our
 4444   // temps, use it for slot. If not, pick any of our temps.
 4445   Register slot;
 4446   if (!temps.contains(rcx)) {
 4447     slot = *available_regs++;
 4448   } else {
 4449     slot = rcx;
 4450   }
 4451 
 4452   const Register r_array_index = *available_regs++;
 4453   const Register r_bitmap      = *available_regs++;
 4454 
 4455   // The logic above guarantees this property, but we state it here.
 4456   assert_different_registers(r_array_index, r_bitmap, rcx);
 4457 
 4458   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4459   movq(r_array_index, r_bitmap);
 4460 
 4461   // First check the bitmap to see if super_klass might be present. If
 4462   // the bit is zero, we are certain that super_klass is not one of
 4463   // the secondary supers.
 4464   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4465   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4466   salq(r_array_index, slot);
 4467 
 4468   testq(r_array_index, r_array_index);
 4469   // We test the MSB of r_array_index, i.e. its sign bit
 4470   jcc(Assembler::positive, L_failure);
 4471 
 4472   const Register r_array_base = *available_regs++;
 4473 
 4474   // Get the first array index that can contain super_klass into r_array_index.
 4475   // Note: Clobbers r_array_base and slot.
 4476   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4477 
 4478   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4479 
 4480   // We will consult the secondary-super array.
 4481   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4482 
 4483   // We're asserting that the first word in an Array<Klass*> is the
 4484   // length, and the second word is the first word of the data. If
 4485   // that ever changes, r_array_base will have to be adjusted here.
 4486   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4487   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4488 
 4489   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4490   jccb(Assembler::equal, L_success);
 4491 
 4492   // Restore slot to its true value
 4493   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4494 
 4495   // Linear probe. Rotate the bitmap so that the next bit to test is
 4496   // in Bit 1.
 4497   rorq(r_bitmap, slot);
 4498 
 4499   // Is there another entry to check? Consult the bitmap.
 4500   btq(r_bitmap, 1);
 4501   jccb(Assembler::carryClear, L_failure);
 4502 
 4503   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4504   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4505   // Kills: r_array_length.
 4506   // Returns: result.
 4507   lookup_secondary_supers_table_slow_path(r_super_klass,
 4508                                           r_array_base,
 4509                                           r_array_index,
 4510                                           r_bitmap,
 4511                                           /*temp1*/result,
 4512                                           /*temp2*/slot,
 4513                                           &L_success,
 4514                                           nullptr);
 4515 
 4516   bind(L_failure);
 4517   movq(result, 1);
 4518   jmpb(L_fallthrough);
 4519 
 4520   bind(L_success);
 4521   xorq(result, result); // = 0
 4522 
 4523   bind(L_fallthrough);
 4524   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4525 
 4526   if (VerifySecondarySupers) {
 4527     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4528                                   temp1, temp2, temp3);
 4529   }
 4530 }
 4531 
 4532 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4533                                  Label* L_success, Label* L_failure) {
 4534   Label L_loop, L_fallthrough;
 4535   {
 4536     int label_nulls = 0;
 4537     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4538     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4539     assert(label_nulls <= 1, "at most one null in the batch");
 4540   }
 4541   bind(L_loop);
 4542   cmpq(value, Address(addr, count, Address::times_8));
 4543   jcc(Assembler::equal, *L_success);
 4544   addl(count, 1);
 4545   cmpl(count, limit);
 4546   jcc(Assembler::less, L_loop);
 4547 
 4548   if (&L_fallthrough != L_failure) {
 4549     jmp(*L_failure);
 4550   }
 4551   bind(L_fallthrough);
 4552 }
 4553 
 4554 // Called by code generated by check_klass_subtype_slow_path
 4555 // above. This is called when there is a collision in the hashed
 4556 // lookup in the secondary supers array.
 4557 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4558                                                              Register r_array_base,
 4559                                                              Register r_array_index,
 4560                                                              Register r_bitmap,
 4561                                                              Register temp1,
 4562                                                              Register temp2,
 4563                                                              Label* L_success,
 4564                                                              Label* L_failure) {
 4565   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4566 
 4567   const Register
 4568     r_array_length = temp1,
 4569     r_sub_klass    = noreg,
 4570     result         = noreg;
 4571 
 4572   Label L_fallthrough;
 4573   int label_nulls = 0;
 4574   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4575   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4576   assert(label_nulls <= 1, "at most one null in the batch");
 4577 
 4578   // Load the array length.
 4579   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4580   // And adjust the array base to point to the data.
 4581   // NB! Effectively increments current slot index by 1.
 4582   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4583   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4584 
 4585   // Linear probe
 4586   Label L_huge;
 4587 
 4588   // The bitmap is full to bursting.
 4589   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4590   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4591   jcc(Assembler::greater, L_huge);
 4592 
 4593   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4594   // current slot (at secondary_supers[r_array_index]) has not yet
 4595   // been inspected, and r_array_index may be out of bounds if we
 4596   // wrapped around the end of the array.
 4597 
 4598   { // This is conventional linear probing, but instead of terminating
 4599     // when a null entry is found in the table, we maintain a bitmap
 4600     // in which a 0 indicates missing entries.
 4601     // The check above guarantees there are 0s in the bitmap, so the loop
 4602     // eventually terminates.
 4603 
 4604     xorl(temp2, temp2); // = 0;
 4605 
 4606     Label L_again;
 4607     bind(L_again);
 4608 
 4609     // Check for array wraparound.
 4610     cmpl(r_array_index, r_array_length);
 4611     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4612 
 4613     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4614     jcc(Assembler::equal, *L_success);
 4615 
 4616     // If the next bit in bitmap is zero, we're done.
 4617     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4618     jcc(Assembler::carryClear, *L_failure);
 4619 
 4620     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4621     addl(r_array_index, 1);
 4622 
 4623     jmp(L_again);
 4624   }
 4625 
 4626   { // Degenerate case: more than 64 secondary supers.
 4627     // FIXME: We could do something smarter here, maybe a vectorized
 4628     // comparison or a binary search, but is that worth any added
 4629     // complexity?
 4630     bind(L_huge);
 4631     xorl(r_array_index, r_array_index); // = 0
 4632     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4633                 L_success,
 4634                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4635 
 4636     bind(L_fallthrough);
 4637   }
 4638 }
 4639 
 4640 struct VerifyHelperArguments {
 4641   Klass* _super;
 4642   Klass* _sub;
 4643   intptr_t _linear_result;
 4644   intptr_t _table_result;
 4645 };
 4646 
 4647 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4648   Klass::on_secondary_supers_verification_failure(args->_super,
 4649                                                   args->_sub,
 4650                                                   args->_linear_result,
 4651                                                   args->_table_result,
 4652                                                   msg);
 4653 }
 4654 
 4655 // Make sure that the hashed lookup and a linear scan agree.
 4656 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4657                                                    Register r_super_klass,
 4658                                                    Register result,
 4659                                                    Register temp1,
 4660                                                    Register temp2,
 4661                                                    Register temp3) {
 4662   const Register
 4663       r_array_index  = temp1,
 4664       r_array_length = temp2,
 4665       r_array_base   = temp3,
 4666       r_bitmap       = noreg;
 4667 
 4668   BLOCK_COMMENT("verify_secondary_supers_table {");
 4669 
 4670   Label L_success, L_failure, L_check, L_done;
 4671 
 4672   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4673   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4674   // And adjust the array base to point to the data.
 4675   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4676 
 4677   testl(r_array_length, r_array_length); // array_length == 0?
 4678   jcc(Assembler::zero, L_failure);
 4679 
 4680   movl(r_array_index, 0);
 4681   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4682   // fall through to L_failure
 4683 
 4684   const Register linear_result = r_array_index; // reuse temp1
 4685 
 4686   bind(L_failure); // not present
 4687   movl(linear_result, 1);
 4688   jmp(L_check);
 4689 
 4690   bind(L_success); // present
 4691   movl(linear_result, 0);
 4692 
 4693   bind(L_check);
 4694   cmpl(linear_result, result);
 4695   jcc(Assembler::equal, L_done);
 4696 
 4697   { // To avoid calling convention issues, build a record on the stack
 4698     // and pass the pointer to that instead.
 4699     push(result);
 4700     push(linear_result);
 4701     push(r_sub_klass);
 4702     push(r_super_klass);
 4703     movptr(c_rarg1, rsp);
 4704     movptr(c_rarg0, (uintptr_t) "mismatch");
 4705     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4706     should_not_reach_here();
 4707   }
 4708   bind(L_done);
 4709 
 4710   BLOCK_COMMENT("} verify_secondary_supers_table");
 4711 }
 4712 
 4713 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4714 
 4715 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4716   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4717 
 4718   Label L_fallthrough;
 4719   if (L_fast_path == nullptr) {
 4720     L_fast_path = &L_fallthrough;
 4721   } else if (L_slow_path == nullptr) {
 4722     L_slow_path = &L_fallthrough;
 4723   }
 4724 
 4725   // Fast path check: class is fully initialized.
 4726   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4727   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4728   jcc(Assembler::equal, *L_fast_path);
 4729 
 4730   // Fast path check: current thread is initializer thread
 4731   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4732   if (L_slow_path == &L_fallthrough) {
 4733     jcc(Assembler::equal, *L_fast_path);
 4734     bind(*L_slow_path);
 4735   } else if (L_fast_path == &L_fallthrough) {
 4736     jcc(Assembler::notEqual, *L_slow_path);
 4737     bind(*L_fast_path);
 4738   } else {
 4739     Unimplemented();
 4740   }
 4741 }
 4742 
 4743 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4744   if (VM_Version::supports_cmov()) {
 4745     cmovl(cc, dst, src);
 4746   } else {
 4747     Label L;
 4748     jccb(negate_condition(cc), L);
 4749     movl(dst, src);
 4750     bind(L);
 4751   }
 4752 }
 4753 
 4754 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4755   if (VM_Version::supports_cmov()) {
 4756     cmovl(cc, dst, src);
 4757   } else {
 4758     Label L;
 4759     jccb(negate_condition(cc), L);
 4760     movl(dst, src);
 4761     bind(L);
 4762   }
 4763 }
 4764 
 4765 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4766   if (!VerifyOops) return;




 4767 
 4768   BLOCK_COMMENT("verify_oop {");
 4769   push(rscratch1);
 4770   push(rax);                          // save rax
 4771   push(reg);                          // pass register argument
 4772 
 4773   // Pass register number to verify_oop_subroutine
 4774   const char* b = nullptr;
 4775   {
 4776     ResourceMark rm;
 4777     stringStream ss;
 4778     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4779     b = code_string(ss.as_string());
 4780   }
 4781   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4782   pushptr(buffer.addr(), rscratch1);
 4783 
 4784   // call indirectly to solve generation ordering problem
 4785   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4786   call(rax);
 4787   // Caller pops the arguments (oop, message) and restores rax, r10
 4788   BLOCK_COMMENT("} verify_oop");
 4789 }
 4790 
 4791 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 4792   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 4793     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 4794     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 4795     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 4796   } else if (VM_Version::supports_avx()) {
 4797     vpcmpeqd(dst, dst, dst, vector_len);
 4798   } else {
 4799     pcmpeqd(dst, dst);
 4800   }
 4801 }
 4802 
 4803 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 4804                                          int extra_slot_offset) {
 4805   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 4806   int stackElementSize = Interpreter::stackElementSize;
 4807   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 4808 #ifdef ASSERT
 4809   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 4810   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 4811 #endif
 4812   Register             scale_reg    = noreg;
 4813   Address::ScaleFactor scale_factor = Address::no_scale;
 4814   if (arg_slot.is_constant()) {
 4815     offset += arg_slot.as_constant() * stackElementSize;
 4816   } else {
 4817     scale_reg    = arg_slot.as_register();
 4818     scale_factor = Address::times(stackElementSize);
 4819   }
 4820   offset += wordSize;           // return PC is on stack
 4821   return Address(rsp, scale_reg, scale_factor, offset);
 4822 }
 4823 
 4824 // Handle the receiver type profile update given the "recv" klass.
 4825 //
 4826 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 4827 // If there are no matching or claimable receiver entries in RD, updates
 4828 // the polymorphic counter.
 4829 //
 4830 // This code expected to run by either the interpreter or JIT-ed code, without
 4831 // extra synchronization. For safety, receiver cells are claimed atomically, which
 4832 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 4833 // counter updates are not atomic.
 4834 //
 4835 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 4836   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 4837   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 4838   int poly_count_offset      = in_bytes(CounterData::count_offset());
 4839   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 4840   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 4841 
 4842   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 4843   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 4844   base_receiver_offset += mdp_offset;
 4845   end_receiver_offset  += mdp_offset;
 4846   poly_count_offset    += mdp_offset;
 4847 
 4848   // Scale down to optimize encoding. Slots are pointer-sized.
 4849   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 4850   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 4851   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 4852   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 4853   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 4854   base_receiver_offset   >>= LogBytesPerWord;
 4855   end_receiver_offset    >>= LogBytesPerWord;
 4856   poly_count_offset      >>= LogBytesPerWord;
 4857   receiver_step          >>= LogBytesPerWord;
 4858   receiver_to_count_step >>= LogBytesPerWord;
 4859 
 4860 #ifdef ASSERT
 4861   // We are about to walk the MDO slots without asking for offsets.
 4862   // Check that our math hits all the right spots.
 4863   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 4864     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 4865     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 4866     int offset = base_receiver_offset + receiver_step*c;
 4867     int count_offset = offset + receiver_to_count_step;
 4868     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 4869     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 4870   }
 4871   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 4872   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 4873 #endif
 4874 
 4875   // Corner case: no profile table. Increment poly counter and exit.
 4876   if (ReceiverTypeData::row_limit() == 0) {
 4877     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 4878     return;
 4879   }
 4880 
 4881   Register offset = rscratch1;
 4882 
 4883   Label L_loop_search_receiver, L_loop_search_empty;
 4884   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 4885 
 4886   // The code here recognizes three major cases:
 4887   //   A. Fastest: receiver found in the table
 4888   //   B. Fast: no receiver in the table, and the table is full
 4889   //   C. Slow: no receiver in the table, free slots in the table
 4890   //
 4891   // The case A performance is most important, as perfectly-behaved code would end up
 4892   // there, especially with larger TypeProfileWidth. The case B performance is
 4893   // important as well, this is where bulk of code would land for normally megamorphic
 4894   // cases. The case C performance is not essential, its job is to deal with installation
 4895   // races, we optimize for code density instead. Case C needs to make sure that receiver
 4896   // rows are only claimed once. This makes sure we never overwrite a row for another
 4897   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 4898   //
 4899   // It is very tempting to handle these cases in a single loop, and claim the first slot
 4900   // without checking the rest of the table. But, profiling code should tolerate free slots
 4901   // in the table, as class unloading can clear them. After such cleanup, the receiver
 4902   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 4903   // to complete, before trying to install new slots. Splitting the code in several tight
 4904   // loops also helpfully optimizes for cases A and B.
 4905   //
 4906   // This code is effectively:
 4907   //
 4908   // restart:
 4909   //   // Fastest: receiver is already installed
 4910   //   for (i = 0; i < receiver_count(); i++) {
 4911   //     if (receiver(i) == recv) goto found_recv(i);
 4912   //   }
 4913   //
 4914   //   // Fast: no receiver, but profile is full
 4915   //   for (i = 0; i < receiver_count(); i++) {
 4916   //     if (receiver(i) == null) goto found_null(i);
 4917   //   }
 4918   //   goto polymorphic
 4919   //
 4920   //   // Slow: try to install receiver
 4921   // found_null(i):
 4922   //   CAS(&receiver(i), null, recv);
 4923   //   goto restart
 4924   //
 4925   // polymorphic:
 4926   //   count++;
 4927   //   return
 4928   //
 4929   // found_recv(i):
 4930   //   *receiver_count(i)++
 4931   //
 4932 
 4933   bind(L_restart);
 4934 
 4935   // Fastest: receiver is already installed
 4936   movptr(offset, base_receiver_offset);
 4937   bind(L_loop_search_receiver);
 4938     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 4939     jccb(Assembler::equal, L_found_recv);
 4940   addptr(offset, receiver_step);
 4941   cmpptr(offset, end_receiver_offset);
 4942   jccb(Assembler::notEqual, L_loop_search_receiver);
 4943 
 4944   // Fast: no receiver, but profile is full
 4945   movptr(offset, base_receiver_offset);
 4946   bind(L_loop_search_empty);
 4947     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 4948     jccb(Assembler::equal, L_found_empty);
 4949   addptr(offset, receiver_step);
 4950   cmpptr(offset, end_receiver_offset);
 4951   jccb(Assembler::notEqual, L_loop_search_empty);
 4952   jmpb(L_polymorphic);
 4953 
 4954   // Slow: try to install receiver
 4955   bind(L_found_empty);
 4956 
 4957   // Atomically swing receiver slot: null -> recv.
 4958   //
 4959   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 4960   // other important registers untouched, as they form the address. Therefore, we need
 4961   // to shift any important registers from RAX into some other spare register. If we
 4962   // have a spare register, we are forced to save it on stack here.
 4963 
 4964   Register spare_reg = noreg;
 4965   Register shifted_mdp = mdp;
 4966   Register shifted_recv = recv;
 4967   if (recv == rax || mdp == rax) {
 4968     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 4969                 (recv != rcx && mdp != rcx) ? rcx :
 4970                 rdx;
 4971     assert_different_registers(mdp, recv, offset, spare_reg);
 4972 
 4973     push(spare_reg);
 4974     if (recv == rax) {
 4975       movptr(spare_reg, recv);
 4976       shifted_recv = spare_reg;
 4977     } else {
 4978       assert(mdp == rax, "Remaining case");
 4979       movptr(spare_reg, mdp);
 4980       shifted_mdp = spare_reg;
 4981     }
 4982   } else {
 4983     push(rax);
 4984   }
 4985 
 4986   // None of the important registers are in RAX after this shuffle.
 4987   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 4988 
 4989   xorptr(rax, rax);
 4990   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 4991 
 4992   // Unshift registers.
 4993   if (recv == rax || mdp == rax) {
 4994     movptr(rax, spare_reg);
 4995     pop(spare_reg);
 4996   } else {
 4997     pop(rax);
 4998   }
 4999 
 5000   // CAS success means the slot now has the receiver we want. CAS failure means
 5001   // something had claimed the slot concurrently: it can be the same receiver we want,
 5002   // or something else. Since this is a slow path, we can optimize for code density,
 5003   // and just restart the search from the beginning.
 5004   jmpb(L_restart);
 5005 
 5006   // Counter updates:
 5007 
 5008   // Increment polymorphic counter instead of receiver slot.
 5009   bind(L_polymorphic);
 5010   movptr(offset, poly_count_offset);
 5011   jmpb(L_count_update);
 5012 
 5013   // Found a receiver, convert its slot offset to corresponding count offset.
 5014   bind(L_found_recv);
 5015   addptr(offset, receiver_to_count_step);
 5016 
 5017   bind(L_count_update);
 5018   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 5019 }
 5020 
 5021 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5022   if (!VerifyOops) return;




 5023 
 5024   push(rscratch1);
 5025   push(rax); // save rax,
 5026   // addr may contain rsp so we will have to adjust it based on the push
 5027   // we just did (and on 64 bit we do two pushes)
 5028   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5029   // stores rax into addr which is backwards of what was intended.
 5030   if (addr.uses(rsp)) {
 5031     lea(rax, addr);
 5032     pushptr(Address(rax, 2 * BytesPerWord));
 5033   } else {
 5034     pushptr(addr);
 5035   }
 5036 
 5037   // Pass register number to verify_oop_subroutine
 5038   const char* b = nullptr;
 5039   {
 5040     ResourceMark rm;
 5041     stringStream ss;
 5042     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 5043     b = code_string(ss.as_string());
 5044   }
 5045   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5046   pushptr(buffer.addr(), rscratch1);
 5047 
 5048   // call indirectly to solve generation ordering problem
 5049   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5050   call(rax);
 5051   // Caller pops the arguments (addr, message) and restores rax, r10.
 5052 }
 5053 
 5054 void MacroAssembler::verify_tlab() {
 5055 #ifdef ASSERT
 5056   if (UseTLAB && VerifyOops) {
 5057     Label next, ok;
 5058     Register t1 = rsi;
 5059 
 5060     push(t1);
 5061 
 5062     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5063     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 5064     jcc(Assembler::aboveEqual, next);
 5065     STOP("assert(top >= start)");
 5066     should_not_reach_here();
 5067 
 5068     bind(next);
 5069     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 5070     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5071     jcc(Assembler::aboveEqual, ok);
 5072     STOP("assert(top <= end)");
 5073     should_not_reach_here();
 5074 
 5075     bind(ok);
 5076     pop(t1);
 5077   }
 5078 #endif
 5079 }
 5080 
 5081 class ControlWord {
 5082  public:
 5083   int32_t _value;
 5084 
 5085   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5086   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5087   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5088   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5089   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5090   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5091   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5092   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5093 
 5094   void print() const {
 5095     // rounding control
 5096     const char* rc;
 5097     switch (rounding_control()) {
 5098       case 0: rc = "round near"; break;
 5099       case 1: rc = "round down"; break;
 5100       case 2: rc = "round up  "; break;
 5101       case 3: rc = "chop      "; break;
 5102       default:
 5103         rc = nullptr; // silence compiler warnings
 5104         fatal("Unknown rounding control: %d", rounding_control());
 5105     };
 5106     // precision control
 5107     const char* pc;
 5108     switch (precision_control()) {
 5109       case 0: pc = "24 bits "; break;
 5110       case 1: pc = "reserved"; break;
 5111       case 2: pc = "53 bits "; break;
 5112       case 3: pc = "64 bits "; break;
 5113       default:
 5114         pc = nullptr; // silence compiler warnings
 5115         fatal("Unknown precision control: %d", precision_control());
 5116     };
 5117     // flags
 5118     char f[9];
 5119     f[0] = ' ';
 5120     f[1] = ' ';
 5121     f[2] = (precision   ()) ? 'P' : 'p';
 5122     f[3] = (underflow   ()) ? 'U' : 'u';
 5123     f[4] = (overflow    ()) ? 'O' : 'o';
 5124     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5125     f[6] = (denormalized()) ? 'D' : 'd';
 5126     f[7] = (invalid     ()) ? 'I' : 'i';
 5127     f[8] = '\x0';
 5128     // output
 5129     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5130   }
 5131 
 5132 };
 5133 
 5134 class StatusWord {
 5135  public:
 5136   int32_t _value;
 5137 
 5138   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5139   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5140   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5141   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5142   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5143   int  top() const                     { return  (_value >> 11) & 7      ; }
 5144   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5145   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5146   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5147   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5148   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5149   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5150   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5151   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5152 
 5153   void print() const {
 5154     // condition codes
 5155     char c[5];
 5156     c[0] = (C3()) ? '3' : '-';
 5157     c[1] = (C2()) ? '2' : '-';
 5158     c[2] = (C1()) ? '1' : '-';
 5159     c[3] = (C0()) ? '0' : '-';
 5160     c[4] = '\x0';
 5161     // flags
 5162     char f[9];
 5163     f[0] = (error_status()) ? 'E' : '-';
 5164     f[1] = (stack_fault ()) ? 'S' : '-';
 5165     f[2] = (precision   ()) ? 'P' : '-';
 5166     f[3] = (underflow   ()) ? 'U' : '-';
 5167     f[4] = (overflow    ()) ? 'O' : '-';
 5168     f[5] = (zero_divide ()) ? 'Z' : '-';
 5169     f[6] = (denormalized()) ? 'D' : '-';
 5170     f[7] = (invalid     ()) ? 'I' : '-';
 5171     f[8] = '\x0';
 5172     // output
 5173     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5174   }
 5175 
 5176 };
 5177 
 5178 class TagWord {
 5179  public:
 5180   int32_t _value;
 5181 
 5182   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5183 
 5184   void print() const {
 5185     printf("%04x", _value & 0xFFFF);
 5186   }
 5187 
 5188 };
 5189 
 5190 class FPU_Register {
 5191  public:
 5192   int32_t _m0;
 5193   int32_t _m1;
 5194   int16_t _ex;
 5195 
 5196   bool is_indefinite() const           {
 5197     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5198   }
 5199 
 5200   void print() const {
 5201     char  sign = (_ex < 0) ? '-' : '+';
 5202     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5203     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5204   };
 5205 
 5206 };
 5207 
 5208 class FPU_State {
 5209  public:
 5210   enum {
 5211     register_size       = 10,
 5212     number_of_registers =  8,
 5213     register_mask       =  7
 5214   };
 5215 
 5216   ControlWord  _control_word;
 5217   StatusWord   _status_word;
 5218   TagWord      _tag_word;
 5219   int32_t      _error_offset;
 5220   int32_t      _error_selector;
 5221   int32_t      _data_offset;
 5222   int32_t      _data_selector;
 5223   int8_t       _register[register_size * number_of_registers];
 5224 
 5225   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5226   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5227 
 5228   const char* tag_as_string(int tag) const {
 5229     switch (tag) {
 5230       case 0: return "valid";
 5231       case 1: return "zero";
 5232       case 2: return "special";
 5233       case 3: return "empty";
 5234     }
 5235     ShouldNotReachHere();
 5236     return nullptr;
 5237   }
 5238 
 5239   void print() const {
 5240     // print computation registers
 5241     { int t = _status_word.top();
 5242       for (int i = 0; i < number_of_registers; i++) {
 5243         int j = (i - t) & register_mask;
 5244         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5245         st(j)->print();
 5246         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5247       }
 5248     }
 5249     printf("\n");
 5250     // print control registers
 5251     printf("ctrl = "); _control_word.print(); printf("\n");
 5252     printf("stat = "); _status_word .print(); printf("\n");
 5253     printf("tags = "); _tag_word    .print(); printf("\n");
 5254   }
 5255 
 5256 };
 5257 
 5258 class Flag_Register {
 5259  public:
 5260   int32_t _value;
 5261 
 5262   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5263   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5264   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5265   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5266   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5267   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5268   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5269 
 5270   void print() const {
 5271     // flags
 5272     char f[8];
 5273     f[0] = (overflow       ()) ? 'O' : '-';
 5274     f[1] = (direction      ()) ? 'D' : '-';
 5275     f[2] = (sign           ()) ? 'S' : '-';
 5276     f[3] = (zero           ()) ? 'Z' : '-';
 5277     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5278     f[5] = (parity         ()) ? 'P' : '-';
 5279     f[6] = (carry          ()) ? 'C' : '-';
 5280     f[7] = '\x0';
 5281     // output
 5282     printf("%08x  flags = %s", _value, f);
 5283   }
 5284 
 5285 };
 5286 
 5287 class IU_Register {
 5288  public:
 5289   int32_t _value;
 5290 
 5291   void print() const {
 5292     printf("%08x  %11d", _value, _value);
 5293   }
 5294 
 5295 };
 5296 
 5297 class IU_State {
 5298  public:
 5299   Flag_Register _eflags;
 5300   IU_Register   _rdi;
 5301   IU_Register   _rsi;
 5302   IU_Register   _rbp;
 5303   IU_Register   _rsp;
 5304   IU_Register   _rbx;
 5305   IU_Register   _rdx;
 5306   IU_Register   _rcx;
 5307   IU_Register   _rax;
 5308 
 5309   void print() const {
 5310     // computation registers
 5311     printf("rax,  = "); _rax.print(); printf("\n");
 5312     printf("rbx,  = "); _rbx.print(); printf("\n");
 5313     printf("rcx  = "); _rcx.print(); printf("\n");
 5314     printf("rdx  = "); _rdx.print(); printf("\n");
 5315     printf("rdi  = "); _rdi.print(); printf("\n");
 5316     printf("rsi  = "); _rsi.print(); printf("\n");
 5317     printf("rbp,  = "); _rbp.print(); printf("\n");
 5318     printf("rsp  = "); _rsp.print(); printf("\n");
 5319     printf("\n");
 5320     // control registers
 5321     printf("flgs = "); _eflags.print(); printf("\n");
 5322   }
 5323 };
 5324 
 5325 
 5326 class CPU_State {
 5327  public:
 5328   FPU_State _fpu_state;
 5329   IU_State  _iu_state;
 5330 
 5331   void print() const {
 5332     printf("--------------------------------------------------\n");
 5333     _iu_state .print();
 5334     printf("\n");
 5335     _fpu_state.print();
 5336     printf("--------------------------------------------------\n");
 5337   }
 5338 
 5339 };
 5340 
 5341 
 5342 static void _print_CPU_state(CPU_State* state) {
 5343   state->print();
 5344 };
 5345 
 5346 
 5347 void MacroAssembler::print_CPU_state() {
 5348   push_CPU_state();
 5349   push(rsp);                // pass CPU state
 5350   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5351   addptr(rsp, wordSize);       // discard argument
 5352   pop_CPU_state();
 5353 }
 5354 
 5355 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5356   // Either restore the MXCSR register after returning from the JNI Call
 5357   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5358   if (VM_Version::supports_sse()) {
 5359     if (RestoreMXCSROnJNICalls) {
 5360       ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5361     } else if (CheckJNICalls) {
 5362       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5363     }
 5364   }
 5365   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5366   vzeroupper();
 5367 }
 5368 
 5369 // ((OopHandle)result).resolve();
 5370 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5371   assert_different_registers(result, tmp);
 5372 
 5373   // Only 64 bit platforms support GCs that require a tmp register
 5374   // Only IN_HEAP loads require a thread_tmp register
 5375   // OopHandle::resolve is an indirection like jobject.
 5376   access_load_at(T_OBJECT, IN_NATIVE,
 5377                  result, Address(result, 0), tmp);
 5378 }
 5379 
 5380 // ((WeakHandle)result).resolve();
 5381 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5382   assert_different_registers(rresult, rtmp);
 5383   Label resolved;
 5384 
 5385   // A null weak handle resolves to null.
 5386   cmpptr(rresult, 0);
 5387   jcc(Assembler::equal, resolved);
 5388 
 5389   // Only 64 bit platforms support GCs that require a tmp register
 5390   // Only IN_HEAP loads require a thread_tmp register
 5391   // WeakHandle::resolve is an indirection like jweak.
 5392   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5393                  rresult, Address(rresult, 0), rtmp);
 5394   bind(resolved);
 5395 }
 5396 
 5397 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5398   // get mirror
 5399   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5400   load_method_holder(mirror, method);
 5401   movptr(mirror, Address(mirror, mirror_offset));
 5402   resolve_oop_handle(mirror, tmp);
 5403 }
 5404 
 5405 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5406   load_method_holder(rresult, rmethod);
 5407   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5408 }
 5409 
 5410 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5411   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5412   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5413   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5414 }
 5415 








 5416 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5417   assert(UseCompactObjectHeaders, "expect compact object headers");
 5418   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5419   shrq(dst, markWord::klass_shift);
 5420 }
 5421 
 5422 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5423   assert_different_registers(src, tmp);
 5424   assert_different_registers(dst, tmp);
 5425 
 5426   if (UseCompactObjectHeaders) {
 5427     load_narrow_klass_compact(dst, src);
 5428     decode_klass_not_null(dst, tmp);
 5429   } else {
 5430     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5431     decode_klass_not_null(dst, tmp);
 5432   }
 5433 }
 5434 





 5435 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5436   assert(!UseCompactObjectHeaders, "not with compact headers");
 5437   assert_different_registers(src, tmp);
 5438   assert_different_registers(dst, tmp);
 5439   encode_klass_not_null(src, tmp);
 5440   movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5441 }
 5442 
 5443 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5444   if (UseCompactObjectHeaders) {
 5445     assert(tmp != noreg, "need tmp");
 5446     assert_different_registers(klass, obj, tmp);
 5447     load_narrow_klass_compact(tmp, obj);
 5448     cmpl(klass, tmp);
 5449   } else {
 5450     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5451   }
 5452 }
 5453 
 5454 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5455   if (UseCompactObjectHeaders) {
 5456     assert(tmp2 != noreg, "need tmp2");
 5457     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5458     load_narrow_klass_compact(tmp1, obj1);
 5459     load_narrow_klass_compact(tmp2, obj2);
 5460     cmpl(tmp1, tmp2);
 5461   } else {
 5462     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5463     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5464   }
 5465 }
 5466 
 5467 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5468                                     Register tmp1) {
 5469   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5470   decorators = AccessInternal::decorator_fixup(decorators, type);
 5471   bool as_raw = (decorators & AS_RAW) != 0;
 5472   if (as_raw) {
 5473     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5474   } else {
 5475     bs->load_at(this, decorators, type, dst, src, tmp1);
 5476   }
 5477 }
 5478 
 5479 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5480                                      Register tmp1, Register tmp2, Register tmp3) {
 5481   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5482   decorators = AccessInternal::decorator_fixup(decorators, type);
 5483   bool as_raw = (decorators & AS_RAW) != 0;
 5484   if (as_raw) {
 5485     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5486   } else {
 5487     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5488   }
 5489 }
 5490 








































 5491 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5492   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5493 }
 5494 
 5495 // Doesn't do verification, generates fixed size code
 5496 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5497   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5498 }
 5499 
 5500 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5501                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5502   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5503 }
 5504 
 5505 // Used for storing nulls.
 5506 void MacroAssembler::store_heap_oop_null(Address dst) {
 5507   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5508 }
 5509 
 5510 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5511   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5512   // Store to klass gap in destination
 5513   movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5514 }
 5515 
 5516 #ifdef ASSERT
 5517 void MacroAssembler::verify_heapbase(const char* msg) {
 5518   assert (UseCompressedOops, "should be compressed");
 5519   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5520   if (CheckCompressedOops) {
 5521     Label ok;
 5522     ExternalAddress src2(CompressedOops::base_addr());
 5523     const bool is_src2_reachable = reachable(src2);
 5524     if (!is_src2_reachable) {
 5525       push(rscratch1);  // cmpptr trashes rscratch1
 5526     }
 5527     cmpptr(r12_heapbase, src2, rscratch1);
 5528     jcc(Assembler::equal, ok);
 5529     STOP(msg);
 5530     bind(ok);
 5531     if (!is_src2_reachable) {
 5532       pop(rscratch1);
 5533     }
 5534   }
 5535 }
 5536 #endif
 5537 
 5538 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5539 void MacroAssembler::encode_heap_oop(Register r) {
 5540 #ifdef ASSERT
 5541   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5542 #endif
 5543   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5544   if (CompressedOops::base() == nullptr) {
 5545     if (CompressedOops::shift() != 0) {
 5546       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5547       shrq(r, LogMinObjAlignmentInBytes);
 5548     }
 5549     return;
 5550   }
 5551   testq(r, r);
 5552   cmovq(Assembler::equal, r, r12_heapbase);
 5553   subq(r, r12_heapbase);
 5554   shrq(r, LogMinObjAlignmentInBytes);
 5555 }
 5556 
 5557 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5558 #ifdef ASSERT
 5559   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5560   if (CheckCompressedOops) {
 5561     Label ok;
 5562     testq(r, r);
 5563     jcc(Assembler::notEqual, ok);
 5564     STOP("null oop passed to encode_heap_oop_not_null");
 5565     bind(ok);
 5566   }
 5567 #endif
 5568   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5569   if (CompressedOops::base() != nullptr) {
 5570     subq(r, r12_heapbase);
 5571   }
 5572   if (CompressedOops::shift() != 0) {
 5573     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5574     shrq(r, LogMinObjAlignmentInBytes);
 5575   }
 5576 }
 5577 
 5578 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5579 #ifdef ASSERT
 5580   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5581   if (CheckCompressedOops) {
 5582     Label ok;
 5583     testq(src, src);
 5584     jcc(Assembler::notEqual, ok);
 5585     STOP("null oop passed to encode_heap_oop_not_null2");
 5586     bind(ok);
 5587   }
 5588 #endif
 5589   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5590   if (dst != src) {
 5591     movq(dst, src);
 5592   }
 5593   if (CompressedOops::base() != nullptr) {
 5594     subq(dst, r12_heapbase);
 5595   }
 5596   if (CompressedOops::shift() != 0) {
 5597     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5598     shrq(dst, LogMinObjAlignmentInBytes);
 5599   }
 5600 }
 5601 
 5602 void  MacroAssembler::decode_heap_oop(Register r) {
 5603 #ifdef ASSERT
 5604   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5605 #endif
 5606   if (CompressedOops::base() == nullptr) {
 5607     if (CompressedOops::shift() != 0) {
 5608       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5609       shlq(r, LogMinObjAlignmentInBytes);
 5610     }
 5611   } else {
 5612     Label done;
 5613     shlq(r, LogMinObjAlignmentInBytes);
 5614     jccb(Assembler::equal, done);
 5615     addq(r, r12_heapbase);
 5616     bind(done);
 5617   }
 5618   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5619 }
 5620 
 5621 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5622   // Note: it will change flags
 5623   assert (UseCompressedOops, "should only be used for compressed headers");
 5624   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5625   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5626   // vtableStubs also counts instructions in pd_code_size_limit.
 5627   // Also do not verify_oop as this is called by verify_oop.
 5628   if (CompressedOops::shift() != 0) {
 5629     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5630     shlq(r, LogMinObjAlignmentInBytes);
 5631     if (CompressedOops::base() != nullptr) {
 5632       addq(r, r12_heapbase);
 5633     }
 5634   } else {
 5635     assert (CompressedOops::base() == nullptr, "sanity");
 5636   }
 5637 }
 5638 
 5639 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5640   // Note: it will change flags
 5641   assert (UseCompressedOops, "should only be used for compressed headers");
 5642   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5643   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5644   // vtableStubs also counts instructions in pd_code_size_limit.
 5645   // Also do not verify_oop as this is called by verify_oop.
 5646   if (CompressedOops::shift() != 0) {
 5647     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5648     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5649       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5650     } else {
 5651       if (dst != src) {
 5652         movq(dst, src);
 5653       }
 5654       shlq(dst, LogMinObjAlignmentInBytes);
 5655       if (CompressedOops::base() != nullptr) {
 5656         addq(dst, r12_heapbase);
 5657       }
 5658     }
 5659   } else {
 5660     assert (CompressedOops::base() == nullptr, "sanity");
 5661     if (dst != src) {
 5662       movq(dst, src);
 5663     }
 5664   }
 5665 }
 5666 
 5667 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5668   BLOCK_COMMENT("encode_klass_not_null {");
 5669   assert_different_registers(r, tmp);
 5670   if (CompressedKlassPointers::base() != nullptr) {
 5671     if (AOTCodeCache::is_on_for_dump()) {
 5672       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5673     } else {
 5674       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5675     }
 5676     subq(r, tmp);
 5677   }
 5678   if (CompressedKlassPointers::shift() != 0) {
 5679     shrq(r, CompressedKlassPointers::shift());
 5680   }
 5681   BLOCK_COMMENT("} encode_klass_not_null");
 5682 }
 5683 
 5684 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5685   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5686   assert_different_registers(src, dst);
 5687   if (CompressedKlassPointers::base() != nullptr) {
 5688     if (AOTCodeCache::is_on_for_dump()) {
 5689       movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5690       negq(dst);
 5691     } else {
 5692       movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5693     }
 5694     addq(dst, src);
 5695   } else {
 5696     movptr(dst, src);
 5697   }
 5698   if (CompressedKlassPointers::shift() != 0) {
 5699     shrq(dst, CompressedKlassPointers::shift());
 5700   }
 5701   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5702 }
 5703 
 5704 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5705   BLOCK_COMMENT("decode_klass_not_null {");
 5706   assert_different_registers(r, tmp);
 5707   // Note: it will change flags
 5708   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5709   // vtableStubs also counts instructions in pd_code_size_limit.
 5710   // Also do not verify_oop as this is called by verify_oop.
 5711   if (CompressedKlassPointers::shift() != 0) {
 5712     shlq(r, CompressedKlassPointers::shift());
 5713   }
 5714   if (CompressedKlassPointers::base() != nullptr) {
 5715     if (AOTCodeCache::is_on_for_dump()) {
 5716       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5717     } else {
 5718       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5719     }
 5720     addq(r, tmp);
 5721   }
 5722   BLOCK_COMMENT("} decode_klass_not_null");
 5723 }
 5724 
 5725 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5726   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 5727   assert_different_registers(src, dst);
 5728   // Note: it will change flags
 5729   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5730   // vtableStubs also counts instructions in pd_code_size_limit.
 5731   // Also do not verify_oop as this is called by verify_oop.
 5732 
 5733   if (CompressedKlassPointers::base() == nullptr &&
 5734       CompressedKlassPointers::shift() == 0) {
 5735     // The best case scenario is that there is no base or shift. Then it is already
 5736     // a pointer that needs nothing but a register rename.
 5737     movl(dst, src);
 5738   } else {
 5739     if (CompressedKlassPointers::shift() <= Address::times_8) {
 5740       if (CompressedKlassPointers::base() != nullptr) {
 5741         if (AOTCodeCache::is_on_for_dump()) {
 5742           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5743         } else {
 5744           movptr(dst, (intptr_t)CompressedKlassPointers::base());
 5745         }
 5746       } else {
 5747         xorq(dst, dst);
 5748       }
 5749       if (CompressedKlassPointers::shift() != 0) {
 5750         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 5751         leaq(dst, Address(dst, src, Address::times_8, 0));
 5752       } else {
 5753         addq(dst, src);
 5754       }
 5755     } else {
 5756       if (CompressedKlassPointers::base() != nullptr) {
 5757         if (AOTCodeCache::is_on_for_dump()) {
 5758           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5759           shrq(dst, CompressedKlassPointers::shift());
 5760         } else {
 5761           const intptr_t base_right_shifted =
 5762                (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 5763           movptr(dst, base_right_shifted);
 5764         }
 5765       } else {
 5766         xorq(dst, dst);
 5767       }
 5768       addq(dst, src);
 5769       shlq(dst, CompressedKlassPointers::shift());
 5770     }
 5771   }
 5772   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 5773 }
 5774 
 5775 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5776   assert (UseCompressedOops, "should only be used for compressed headers");
 5777   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5778   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5779   int oop_index = oop_recorder()->find_index(obj);
 5780   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5781   mov_narrow_oop(dst, oop_index, rspec);
 5782 }
 5783 
 5784 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5785   assert (UseCompressedOops, "should only be used for compressed headers");
 5786   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5787   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5788   int oop_index = oop_recorder()->find_index(obj);
 5789   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5790   mov_narrow_oop(dst, oop_index, rspec);
 5791 }
 5792 
 5793 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5794   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5795   int klass_index = oop_recorder()->find_index(k);
 5796   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5797   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5798 }
 5799 
 5800 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 5801   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5802   int klass_index = oop_recorder()->find_index(k);
 5803   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5804   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5805 }
 5806 
 5807 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 5808   assert (UseCompressedOops, "should only be used for compressed headers");
 5809   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5810   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5811   int oop_index = oop_recorder()->find_index(obj);
 5812   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5813   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5814 }
 5815 
 5816 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 5817   assert (UseCompressedOops, "should only be used for compressed headers");
 5818   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5819   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5820   int oop_index = oop_recorder()->find_index(obj);
 5821   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5822   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5823 }
 5824 
 5825 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 5826   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5827   int klass_index = oop_recorder()->find_index(k);
 5828   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5829   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5830 }
 5831 
 5832 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 5833   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5834   int klass_index = oop_recorder()->find_index(k);
 5835   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5836   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5837 }
 5838 
 5839 void MacroAssembler::reinit_heapbase() {
 5840   if (UseCompressedOops) {
 5841     if (Universe::heap() != nullptr && !AOTCodeCache::is_on_for_dump()) {
 5842       if (CompressedOops::base() == nullptr) {
 5843         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 5844       } else {
 5845         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 5846       }
 5847     } else {
 5848       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 5849     }
 5850   }
 5851 }
 5852 




































































































































































































































































































































































































































































































 5853 #if COMPILER2_OR_JVMCI
 5854 
 5855 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 5856 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5857   // cnt - number of qwords (8-byte words).
 5858   // base - start address, qword aligned.
 5859   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 5860   bool use64byteVector = (MaxVectorSize == 64) && (CopyAVX3Threshold == 0);
 5861   if (use64byteVector) {
 5862     vpxor(xtmp, xtmp, xtmp, AVX_512bit);
 5863   } else if (MaxVectorSize >= 32) {
 5864     vpxor(xtmp, xtmp, xtmp, AVX_256bit);


 5865   } else {
 5866     pxor(xtmp, xtmp);

 5867   }
 5868   jmp(L_zero_64_bytes);
 5869 
 5870   BIND(L_loop);
 5871   if (MaxVectorSize >= 32) {
 5872     fill64(base, 0, xtmp, use64byteVector);
 5873   } else {
 5874     movdqu(Address(base,  0), xtmp);
 5875     movdqu(Address(base, 16), xtmp);
 5876     movdqu(Address(base, 32), xtmp);
 5877     movdqu(Address(base, 48), xtmp);
 5878   }
 5879   addptr(base, 64);
 5880 
 5881   BIND(L_zero_64_bytes);
 5882   subptr(cnt, 8);
 5883   jccb(Assembler::greaterEqual, L_loop);
 5884 
 5885   // Copy trailing 64 bytes
 5886   if (use64byteVector) {
 5887     addptr(cnt, 8);
 5888     jccb(Assembler::equal, L_end);
 5889     fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
 5890     jmp(L_end);
 5891   } else {
 5892     addptr(cnt, 4);
 5893     jccb(Assembler::less, L_tail);
 5894     if (MaxVectorSize >= 32) {
 5895       vmovdqu(Address(base, 0), xtmp);
 5896     } else {
 5897       movdqu(Address(base,  0), xtmp);
 5898       movdqu(Address(base, 16), xtmp);
 5899     }
 5900   }
 5901   addptr(base, 32);
 5902   subptr(cnt, 4);
 5903 
 5904   BIND(L_tail);
 5905   addptr(cnt, 4);
 5906   jccb(Assembler::lessEqual, L_end);
 5907   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 5908     fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
 5909   } else {
 5910     decrement(cnt);
 5911 
 5912     BIND(L_sloop);
 5913     movq(Address(base, 0), xtmp);
 5914     addptr(base, 8);
 5915     decrement(cnt);
 5916     jccb(Assembler::greaterEqual, L_sloop);
 5917   }
 5918   BIND(L_end);
 5919 }
 5920 
 5921 // Clearing constant sized memory using YMM/ZMM registers.
 5922 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5923   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 5924   bool use64byteVector = (MaxVectorSize > 32) && (CopyAVX3Threshold == 0);
 5925 
 5926   int vector64_count = (cnt & (~0x7)) >> 3;
 5927   cnt = cnt & 0x7;
 5928   const int fill64_per_loop = 4;
 5929   const int max_unrolled_fill64 = 8;
 5930 
 5931   // 64 byte initialization loop.
 5932   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 5933   int start64 = 0;
 5934   if (vector64_count > max_unrolled_fill64) {
 5935     Label LOOP;
 5936     Register index = rtmp;
 5937 
 5938     start64 = vector64_count - (vector64_count % fill64_per_loop);
 5939 
 5940     movl(index, 0);
 5941     BIND(LOOP);
 5942     for (int i = 0; i < fill64_per_loop; i++) {
 5943       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 5944     }
 5945     addl(index, fill64_per_loop * 64);
 5946     cmpl(index, start64 * 64);
 5947     jccb(Assembler::less, LOOP);
 5948   }
 5949   for (int i = start64; i < vector64_count; i++) {
 5950     fill64(base, i * 64, xtmp, use64byteVector);
 5951   }
 5952 
 5953   // Clear remaining 64 byte tail.
 5954   int disp = vector64_count * 64;
 5955   if (cnt) {
 5956     switch (cnt) {
 5957       case 1:
 5958         movq(Address(base, disp), xtmp);
 5959         break;
 5960       case 2:
 5961         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 5962         break;
 5963       case 3:
 5964         movl(rtmp, 0x7);
 5965         kmovwl(mask, rtmp);
 5966         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 5967         break;
 5968       case 4:
 5969         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5970         break;
 5971       case 5:
 5972         if (use64byteVector) {
 5973           movl(rtmp, 0x1F);
 5974           kmovwl(mask, rtmp);
 5975           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5976         } else {
 5977           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5978           movq(Address(base, disp + 32), xtmp);
 5979         }
 5980         break;
 5981       case 6:
 5982         if (use64byteVector) {
 5983           movl(rtmp, 0x3F);
 5984           kmovwl(mask, rtmp);
 5985           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5986         } else {
 5987           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5988           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 5989         }
 5990         break;
 5991       case 7:
 5992         if (use64byteVector) {
 5993           movl(rtmp, 0x7F);
 5994           kmovwl(mask, rtmp);
 5995           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5996         } else {
 5997           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5998           movl(rtmp, 0x7);
 5999           kmovwl(mask, rtmp);
 6000           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 6001         }
 6002         break;
 6003       default:
 6004         fatal("Unexpected length : %d\n",cnt);
 6005         break;
 6006     }
 6007   }
 6008 }
 6009 
 6010 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
 6011                                bool is_large, KRegister mask) {
 6012   // cnt      - number of qwords (8-byte words).
 6013   // base     - start address, qword aligned.
 6014   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6015   assert(base==rdi, "base register must be edi for rep stos");
 6016   assert(tmp==rax,   "tmp register must be eax for rep stos");
 6017   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6018   assert(InitArrayShortSize % BytesPerLong == 0,
 6019     "InitArrayShortSize should be the multiple of BytesPerLong");
 6020 
 6021   Label DONE;
 6022   if (!is_large || !UseXMMForObjInit) {
 6023     xorptr(tmp, tmp);
 6024   }
 6025 
 6026   if (!is_large) {
 6027     Label LOOP, LONG;
 6028     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6029     jccb(Assembler::greater, LONG);
 6030 
 6031     decrement(cnt);
 6032     jccb(Assembler::negative, DONE); // Zero length
 6033 
 6034     // Use individual pointer-sized stores for small counts:
 6035     BIND(LOOP);
 6036     movptr(Address(base, cnt, Address::times_ptr), tmp);
 6037     decrement(cnt);
 6038     jccb(Assembler::greaterEqual, LOOP);
 6039     jmpb(DONE);
 6040 
 6041     BIND(LONG);
 6042   }
 6043 
 6044   // Use longer rep-prefixed ops for non-small counts:
 6045   if (UseFastStosb) {
 6046     shlptr(cnt, 3); // convert to number of bytes
 6047     rep_stosb();
 6048   } else if (UseXMMForObjInit) {
 6049     xmm_clear_mem(base, cnt, tmp, xtmp, mask);
 6050   } else {
 6051     rep_stos();
 6052   }
 6053 
 6054   BIND(DONE);
 6055 }
 6056 
 6057 #endif //COMPILER2_OR_JVMCI
 6058 
 6059 
 6060 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6061                                    Register to, Register value, Register count,
 6062                                    Register rtmp, XMMRegister xtmp) {
 6063   ShortBranchVerifier sbv(this);
 6064   assert_different_registers(to, value, count, rtmp);
 6065   Label L_exit;
 6066   Label L_fill_2_bytes, L_fill_4_bytes;
 6067 
 6068 #if defined(COMPILER2)
 6069   if(MaxVectorSize >=32 &&
 6070      VM_Version::supports_avx512vlbw() &&
 6071      VM_Version::supports_bmi2()) {
 6072     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6073     return;
 6074   }
 6075 #endif
 6076 
 6077   int shift = -1;
 6078   switch (t) {
 6079     case T_BYTE:
 6080       shift = 2;
 6081       break;
 6082     case T_SHORT:
 6083       shift = 1;
 6084       break;
 6085     case T_INT:
 6086       shift = 0;
 6087       break;
 6088     default: ShouldNotReachHere();
 6089   }
 6090 
 6091   if (t == T_BYTE) {
 6092     andl(value, 0xff);
 6093     movl(rtmp, value);
 6094     shll(rtmp, 8);
 6095     orl(value, rtmp);
 6096   }
 6097   if (t == T_SHORT) {
 6098     andl(value, 0xffff);
 6099   }
 6100   if (t == T_BYTE || t == T_SHORT) {
 6101     movl(rtmp, value);
 6102     shll(rtmp, 16);
 6103     orl(value, rtmp);
 6104   }
 6105 
 6106   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6107   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6108   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6109     Label L_skip_align2;
 6110     // align source address at 4 bytes address boundary
 6111     if (t == T_BYTE) {
 6112       Label L_skip_align1;
 6113       // One byte misalignment happens only for byte arrays
 6114       testptr(to, 1);
 6115       jccb(Assembler::zero, L_skip_align1);
 6116       movb(Address(to, 0), value);
 6117       increment(to);
 6118       decrement(count);
 6119       BIND(L_skip_align1);
 6120     }
 6121     // Two bytes misalignment happens only for byte and short (char) arrays
 6122     testptr(to, 2);
 6123     jccb(Assembler::zero, L_skip_align2);
 6124     movw(Address(to, 0), value);
 6125     addptr(to, 2);
 6126     subptr(count, 1<<(shift-1));
 6127     BIND(L_skip_align2);
 6128   }
 6129   {
 6130     Label L_fill_32_bytes;
 6131     if (!UseUnalignedLoadStores) {
 6132       // align to 8 bytes, we know we are 4 byte aligned to start
 6133       testptr(to, 4);
 6134       jccb(Assembler::zero, L_fill_32_bytes);
 6135       movl(Address(to, 0), value);
 6136       addptr(to, 4);
 6137       subptr(count, 1<<shift);
 6138     }
 6139     BIND(L_fill_32_bytes);
 6140     {
 6141       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6142       movdl(xtmp, value);
 6143       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6144         Label L_check_fill_32_bytes;
 6145         if (UseAVX > 2) {
 6146           // Fill 64-byte chunks
 6147           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6148 
 6149           // If number of bytes to fill < CopyAVX3Threshold, perform fill using AVX2
 6150           cmpptr(count, CopyAVX3Threshold);
 6151           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6152 
 6153           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6154 
 6155           subptr(count, 16 << shift);
 6156           jcc(Assembler::less, L_check_fill_32_bytes);
 6157           align(16);
 6158 
 6159           BIND(L_fill_64_bytes_loop_avx3);
 6160           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6161           addptr(to, 64);
 6162           subptr(count, 16 << shift);
 6163           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6164           jmpb(L_check_fill_32_bytes);
 6165 
 6166           BIND(L_check_fill_64_bytes_avx2);
 6167         }
 6168         // Fill 64-byte chunks
 6169         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6170 
 6171         subptr(count, 16 << shift);
 6172         jcc(Assembler::less, L_check_fill_32_bytes);
 6173 
 6174         // align data for 64-byte chunks
 6175         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6176         if (EnableX86ECoreOpts) {
 6177             // align 'big' arrays to cache lines to minimize split_stores
 6178             cmpptr(count, 96 << shift);
 6179             jcc(Assembler::below, L_fill_64_bytes_loop);
 6180 
 6181             // Find the bytes needed for alignment
 6182             movptr(rtmp, to);
 6183             andptr(rtmp, 0x1c);
 6184             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6185             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6186             addptr(rtmp, 32);
 6187             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6188             subptr(count, rtmp);    // adjust count by number of elements
 6189 
 6190             align(16);
 6191             BIND(L_align_64_bytes_loop);
 6192             movdl(Address(to, 0), xtmp);
 6193             addptr(to, 4);
 6194             subptr(rtmp, 1 << shift);
 6195             jcc(Assembler::greater, L_align_64_bytes_loop);
 6196         }
 6197 
 6198         align(16);
 6199         BIND(L_fill_64_bytes_loop);
 6200         vmovdqu(Address(to, 0), xtmp);
 6201         vmovdqu(Address(to, 32), xtmp);
 6202         addptr(to, 64);
 6203         subptr(count, 16 << shift);
 6204         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6205 
 6206         align(16);
 6207         BIND(L_check_fill_32_bytes);
 6208         addptr(count, 8 << shift);
 6209         jccb(Assembler::less, L_check_fill_8_bytes);
 6210         vmovdqu(Address(to, 0), xtmp);
 6211         addptr(to, 32);
 6212         subptr(count, 8 << shift);
 6213 
 6214         BIND(L_check_fill_8_bytes);
 6215         // clean upper bits of YMM registers
 6216         movdl(xtmp, value);
 6217         pshufd(xtmp, xtmp, 0);
 6218       } else {
 6219         // Fill 32-byte chunks
 6220         pshufd(xtmp, xtmp, 0);
 6221 
 6222         subptr(count, 8 << shift);
 6223         jcc(Assembler::less, L_check_fill_8_bytes);
 6224         align(16);
 6225 
 6226         BIND(L_fill_32_bytes_loop);
 6227 
 6228         if (UseUnalignedLoadStores) {
 6229           movdqu(Address(to, 0), xtmp);
 6230           movdqu(Address(to, 16), xtmp);
 6231         } else {
 6232           movq(Address(to, 0), xtmp);
 6233           movq(Address(to, 8), xtmp);
 6234           movq(Address(to, 16), xtmp);
 6235           movq(Address(to, 24), xtmp);
 6236         }
 6237 
 6238         addptr(to, 32);
 6239         subptr(count, 8 << shift);
 6240         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6241 
 6242         BIND(L_check_fill_8_bytes);
 6243       }
 6244       addptr(count, 8 << shift);
 6245       jccb(Assembler::zero, L_exit);
 6246       jmpb(L_fill_8_bytes);
 6247 
 6248       //
 6249       // length is too short, just fill qwords
 6250       //
 6251       align(16);
 6252       BIND(L_fill_8_bytes_loop);
 6253       movq(Address(to, 0), xtmp);
 6254       addptr(to, 8);
 6255       BIND(L_fill_8_bytes);
 6256       subptr(count, 1 << (shift + 1));
 6257       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6258     }
 6259   }
 6260 
 6261   Label L_fill_4_bytes_loop;
 6262   testl(count, 1 << shift);
 6263   jccb(Assembler::zero, L_fill_2_bytes);
 6264 
 6265   align(16);
 6266   BIND(L_fill_4_bytes_loop);
 6267   movl(Address(to, 0), value);
 6268   addptr(to, 4);
 6269 
 6270   BIND(L_fill_4_bytes);
 6271   subptr(count, 1 << shift);
 6272   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 6273 
 6274   if (t == T_BYTE || t == T_SHORT) {
 6275     Label L_fill_byte;
 6276     BIND(L_fill_2_bytes);
 6277     // fill trailing 2 bytes
 6278     testl(count, 1<<(shift-1));
 6279     jccb(Assembler::zero, L_fill_byte);
 6280     movw(Address(to, 0), value);
 6281     if (t == T_BYTE) {
 6282       addptr(to, 2);
 6283       BIND(L_fill_byte);
 6284       // fill trailing byte
 6285       testl(count, 1);
 6286       jccb(Assembler::zero, L_exit);
 6287       movb(Address(to, 0), value);
 6288     } else {
 6289       BIND(L_fill_byte);
 6290     }
 6291   } else {
 6292     BIND(L_fill_2_bytes);
 6293   }
 6294   BIND(L_exit);
 6295 }
 6296 
 6297 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6298   switch(type) {
 6299     case T_BYTE:
 6300     case T_BOOLEAN:
 6301       evpbroadcastb(dst, src, vector_len);
 6302       break;
 6303     case T_SHORT:
 6304     case T_CHAR:
 6305       evpbroadcastw(dst, src, vector_len);
 6306       break;
 6307     case T_INT:
 6308     case T_FLOAT:
 6309       evpbroadcastd(dst, src, vector_len);
 6310       break;
 6311     case T_LONG:
 6312     case T_DOUBLE:
 6313       evpbroadcastq(dst, src, vector_len);
 6314       break;
 6315     default:
 6316       fatal("Unhandled type : %s", type2name(type));
 6317       break;
 6318   }
 6319 }
 6320 
 6321 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 6322 //
 6323 // @IntrinsicCandidate
 6324 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 6325 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6326 //     int i = 0;
 6327 //     for (; i < len; i++) {
 6328 //         char c = sa[sp++];
 6329 //         if (c > '\u00FF')
 6330 //             break;
 6331 //         da[dp++] = (byte) c;
 6332 //     }
 6333 //     return i;
 6334 // }
 6335 //
 6336 // @IntrinsicCandidate
 6337 // int java.lang.StringCoding.encodeISOArray0(
 6338 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 6339 //   int i = 0;
 6340 //   for (; i < len; i++) {
 6341 //     char c = StringUTF16.getChar(sa, sp++);
 6342 //     if (c > '\u00FF')
 6343 //       break;
 6344 //     da[dp++] = (byte) c;
 6345 //   }
 6346 //   return i;
 6347 // }
 6348 //
 6349 // @IntrinsicCandidate
 6350 // int java.lang.StringCoding.encodeAsciiArray0(
 6351 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6352 //   int i = 0;
 6353 //   for (; i < len; i++) {
 6354 //     char c = sa[sp++];
 6355 //     if (c >= '\u0080')
 6356 //       break;
 6357 //     da[dp++] = (byte) c;
 6358 //   }
 6359 //   return i;
 6360 // }
 6361 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 6362   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 6363   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 6364   Register tmp5, Register result, bool ascii) {
 6365 
 6366   // rsi: src
 6367   // rdi: dst
 6368   // rdx: len
 6369   // rcx: tmp5
 6370   // rax: result
 6371   ShortBranchVerifier sbv(this);
 6372   assert_different_registers(src, dst, len, tmp5, result);
 6373   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 6374 
 6375   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 6376   int short_mask = ascii ? 0xff80 : 0xff00;
 6377 
 6378   // set result
 6379   xorl(result, result);
 6380   // check for zero length
 6381   testl(len, len);
 6382   jcc(Assembler::zero, L_done);
 6383 
 6384   movl(result, len);
 6385 
 6386   // Setup pointers
 6387   lea(src, Address(src, len, Address::times_2)); // char[]
 6388   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 6389   negptr(len);
 6390 
 6391   if (UseSSE42Intrinsics || UseAVX >= 2) {
 6392     Label L_copy_8_chars, L_copy_8_chars_exit;
 6393     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 6394 
 6395     if (UseAVX >= 2) {
 6396       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 6397       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6398       movdl(tmp1Reg, tmp5);
 6399       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 6400       jmp(L_chars_32_check);
 6401 
 6402       bind(L_copy_32_chars);
 6403       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 6404       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 6405       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6406       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6407       jccb(Assembler::notZero, L_copy_32_chars_exit);
 6408       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6409       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 6410       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 6411 
 6412       bind(L_chars_32_check);
 6413       addptr(len, 32);
 6414       jcc(Assembler::lessEqual, L_copy_32_chars);
 6415 
 6416       bind(L_copy_32_chars_exit);
 6417       subptr(len, 16);
 6418       jccb(Assembler::greater, L_copy_16_chars_exit);
 6419 
 6420     } else if (UseSSE42Intrinsics) {
 6421       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6422       movdl(tmp1Reg, tmp5);
 6423       pshufd(tmp1Reg, tmp1Reg, 0);
 6424       jmpb(L_chars_16_check);
 6425     }
 6426 
 6427     bind(L_copy_16_chars);
 6428     if (UseAVX >= 2) {
 6429       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 6430       vptest(tmp2Reg, tmp1Reg);
 6431       jcc(Assembler::notZero, L_copy_16_chars_exit);
 6432       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 6433       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 6434     } else {
 6435       if (UseAVX > 0) {
 6436         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6437         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6438         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 6439       } else {
 6440         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6441         por(tmp2Reg, tmp3Reg);
 6442         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6443         por(tmp2Reg, tmp4Reg);
 6444       }
 6445       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6446       jccb(Assembler::notZero, L_copy_16_chars_exit);
 6447       packuswb(tmp3Reg, tmp4Reg);
 6448     }
 6449     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 6450 
 6451     bind(L_chars_16_check);
 6452     addptr(len, 16);
 6453     jcc(Assembler::lessEqual, L_copy_16_chars);
 6454 
 6455     bind(L_copy_16_chars_exit);
 6456     if (UseAVX >= 2) {
 6457       // clean upper bits of YMM registers
 6458       vpxor(tmp2Reg, tmp2Reg);
 6459       vpxor(tmp3Reg, tmp3Reg);
 6460       vpxor(tmp4Reg, tmp4Reg);
 6461       movdl(tmp1Reg, tmp5);
 6462       pshufd(tmp1Reg, tmp1Reg, 0);
 6463     }
 6464     subptr(len, 8);
 6465     jccb(Assembler::greater, L_copy_8_chars_exit);
 6466 
 6467     bind(L_copy_8_chars);
 6468     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 6469     ptest(tmp3Reg, tmp1Reg);
 6470     jccb(Assembler::notZero, L_copy_8_chars_exit);
 6471     packuswb(tmp3Reg, tmp1Reg);
 6472     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 6473     addptr(len, 8);
 6474     jccb(Assembler::lessEqual, L_copy_8_chars);
 6475 
 6476     bind(L_copy_8_chars_exit);
 6477     subptr(len, 8);
 6478     jccb(Assembler::zero, L_done);
 6479   }
 6480 
 6481   bind(L_copy_1_char);
 6482   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 6483   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 6484   jccb(Assembler::notZero, L_copy_1_char_exit);
 6485   movb(Address(dst, len, Address::times_1, 0), tmp5);
 6486   addptr(len, 1);
 6487   jccb(Assembler::less, L_copy_1_char);
 6488 
 6489   bind(L_copy_1_char_exit);
 6490   addptr(result, len); // len is negative count of not processed elements
 6491 
 6492   bind(L_done);
 6493 }
 6494 
 6495 /**
 6496  * Helper for multiply_to_len().
 6497  */
 6498 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 6499   addq(dest_lo, src1);
 6500   adcq(dest_hi, 0);
 6501   addq(dest_lo, src2);
 6502   adcq(dest_hi, 0);
 6503 }
 6504 
 6505 /**
 6506  * Multiply 64 bit by 64 bit first loop.
 6507  */
 6508 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 6509                                            Register y, Register y_idx, Register z,
 6510                                            Register carry, Register product,
 6511                                            Register idx, Register kdx) {
 6512   //
 6513   //  jlong carry, x[], y[], z[];
 6514   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6515   //    huge_128 product = y[idx] * x[xstart] + carry;
 6516   //    z[kdx] = (jlong)product;
 6517   //    carry  = (jlong)(product >>> 64);
 6518   //  }
 6519   //  z[xstart] = carry;
 6520   //
 6521 
 6522   Label L_first_loop, L_first_loop_exit;
 6523   Label L_one_x, L_one_y, L_multiply;
 6524 
 6525   decrementl(xstart);
 6526   jcc(Assembler::negative, L_one_x);
 6527 
 6528   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6529   rorq(x_xstart, 32); // convert big-endian to little-endian
 6530 
 6531   bind(L_first_loop);
 6532   decrementl(idx);
 6533   jcc(Assembler::negative, L_first_loop_exit);
 6534   decrementl(idx);
 6535   jcc(Assembler::negative, L_one_y);
 6536   movq(y_idx, Address(y, idx, Address::times_4,  0));
 6537   rorq(y_idx, 32); // convert big-endian to little-endian
 6538   bind(L_multiply);
 6539   movq(product, x_xstart);
 6540   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 6541   addq(product, carry);
 6542   adcq(rdx, 0);
 6543   subl(kdx, 2);
 6544   movl(Address(z, kdx, Address::times_4,  4), product);
 6545   shrq(product, 32);
 6546   movl(Address(z, kdx, Address::times_4,  0), product);
 6547   movq(carry, rdx);
 6548   jmp(L_first_loop);
 6549 
 6550   bind(L_one_y);
 6551   movl(y_idx, Address(y,  0));
 6552   jmp(L_multiply);
 6553 
 6554   bind(L_one_x);
 6555   movl(x_xstart, Address(x,  0));
 6556   jmp(L_first_loop);
 6557 
 6558   bind(L_first_loop_exit);
 6559 }
 6560 
 6561 /**
 6562  * Multiply 64 bit by 64 bit and add 128 bit.
 6563  */
 6564 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 6565                                             Register yz_idx, Register idx,
 6566                                             Register carry, Register product, int offset) {
 6567   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 6568   //     z[kdx] = (jlong)product;
 6569 
 6570   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 6571   rorq(yz_idx, 32); // convert big-endian to little-endian
 6572   movq(product, x_xstart);
 6573   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 6574   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 6575   rorq(yz_idx, 32); // convert big-endian to little-endian
 6576 
 6577   add2_with_carry(rdx, product, carry, yz_idx);
 6578 
 6579   movl(Address(z, idx, Address::times_4,  offset+4), product);
 6580   shrq(product, 32);
 6581   movl(Address(z, idx, Address::times_4,  offset), product);
 6582 
 6583 }
 6584 
 6585 /**
 6586  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 6587  */
 6588 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 6589                                              Register yz_idx, Register idx, Register jdx,
 6590                                              Register carry, Register product,
 6591                                              Register carry2) {
 6592   //   jlong carry, x[], y[], z[];
 6593   //   int kdx = ystart+1;
 6594   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6595   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 6596   //     z[kdx+idx+1] = (jlong)product;
 6597   //     jlong carry2  = (jlong)(product >>> 64);
 6598   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 6599   //     z[kdx+idx] = (jlong)product;
 6600   //     carry  = (jlong)(product >>> 64);
 6601   //   }
 6602   //   idx += 2;
 6603   //   if (idx > 0) {
 6604   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 6605   //     z[kdx+idx] = (jlong)product;
 6606   //     carry  = (jlong)(product >>> 64);
 6607   //   }
 6608   //
 6609 
 6610   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6611 
 6612   movl(jdx, idx);
 6613   andl(jdx, 0xFFFFFFFC);
 6614   shrl(jdx, 2);
 6615 
 6616   bind(L_third_loop);
 6617   subl(jdx, 1);
 6618   jcc(Assembler::negative, L_third_loop_exit);
 6619   subl(idx, 4);
 6620 
 6621   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 6622   movq(carry2, rdx);
 6623 
 6624   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 6625   movq(carry, rdx);
 6626   jmp(L_third_loop);
 6627 
 6628   bind (L_third_loop_exit);
 6629 
 6630   andl (idx, 0x3);
 6631   jcc(Assembler::zero, L_post_third_loop_done);
 6632 
 6633   Label L_check_1;
 6634   subl(idx, 2);
 6635   jcc(Assembler::negative, L_check_1);
 6636 
 6637   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 6638   movq(carry, rdx);
 6639 
 6640   bind (L_check_1);
 6641   addl (idx, 0x2);
 6642   andl (idx, 0x1);
 6643   subl(idx, 1);
 6644   jcc(Assembler::negative, L_post_third_loop_done);
 6645 
 6646   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 6647   movq(product, x_xstart);
 6648   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 6649   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 6650 
 6651   add2_with_carry(rdx, product, yz_idx, carry);
 6652 
 6653   movl(Address(z, idx, Address::times_4,  0), product);
 6654   shrq(product, 32);
 6655 
 6656   shlq(rdx, 32);
 6657   orq(product, rdx);
 6658   movq(carry, product);
 6659 
 6660   bind(L_post_third_loop_done);
 6661 }
 6662 
 6663 /**
 6664  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 6665  *
 6666  */
 6667 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 6668                                                   Register carry, Register carry2,
 6669                                                   Register idx, Register jdx,
 6670                                                   Register yz_idx1, Register yz_idx2,
 6671                                                   Register tmp, Register tmp3, Register tmp4) {
 6672   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 6673 
 6674   //   jlong carry, x[], y[], z[];
 6675   //   int kdx = ystart+1;
 6676   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6677   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 6678   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 6679   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 6680   //     carry  = (jlong)(tmp4 >>> 64);
 6681   //     z[kdx+idx+1] = (jlong)tmp3;
 6682   //     z[kdx+idx] = (jlong)tmp4;
 6683   //   }
 6684   //   idx += 2;
 6685   //   if (idx > 0) {
 6686   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 6687   //     z[kdx+idx] = (jlong)yz_idx1;
 6688   //     carry  = (jlong)(yz_idx1 >>> 64);
 6689   //   }
 6690   //
 6691 
 6692   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6693 
 6694   movl(jdx, idx);
 6695   andl(jdx, 0xFFFFFFFC);
 6696   shrl(jdx, 2);
 6697 
 6698   bind(L_third_loop);
 6699   subl(jdx, 1);
 6700   jcc(Assembler::negative, L_third_loop_exit);
 6701   subl(idx, 4);
 6702 
 6703   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 6704   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 6705   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 6706   rorxq(yz_idx2, yz_idx2, 32);
 6707 
 6708   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 6709   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 6710 
 6711   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 6712   rorxq(yz_idx1, yz_idx1, 32);
 6713   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6714   rorxq(yz_idx2, yz_idx2, 32);
 6715 
 6716   if (VM_Version::supports_adx()) {
 6717     adcxq(tmp3, carry);
 6718     adoxq(tmp3, yz_idx1);
 6719 
 6720     adcxq(tmp4, tmp);
 6721     adoxq(tmp4, yz_idx2);
 6722 
 6723     movl(carry, 0); // does not affect flags
 6724     adcxq(carry2, carry);
 6725     adoxq(carry2, carry);
 6726   } else {
 6727     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 6728     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 6729   }
 6730   movq(carry, carry2);
 6731 
 6732   movl(Address(z, idx, Address::times_4, 12), tmp3);
 6733   shrq(tmp3, 32);
 6734   movl(Address(z, idx, Address::times_4,  8), tmp3);
 6735 
 6736   movl(Address(z, idx, Address::times_4,  4), tmp4);
 6737   shrq(tmp4, 32);
 6738   movl(Address(z, idx, Address::times_4,  0), tmp4);
 6739 
 6740   jmp(L_third_loop);
 6741 
 6742   bind (L_third_loop_exit);
 6743 
 6744   andl (idx, 0x3);
 6745   jcc(Assembler::zero, L_post_third_loop_done);
 6746 
 6747   Label L_check_1;
 6748   subl(idx, 2);
 6749   jcc(Assembler::negative, L_check_1);
 6750 
 6751   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 6752   rorxq(yz_idx1, yz_idx1, 32);
 6753   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 6754   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6755   rorxq(yz_idx2, yz_idx2, 32);
 6756 
 6757   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 6758 
 6759   movl(Address(z, idx, Address::times_4,  4), tmp3);
 6760   shrq(tmp3, 32);
 6761   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6762   movq(carry, tmp4);
 6763 
 6764   bind (L_check_1);
 6765   addl (idx, 0x2);
 6766   andl (idx, 0x1);
 6767   subl(idx, 1);
 6768   jcc(Assembler::negative, L_post_third_loop_done);
 6769   movl(tmp4, Address(y, idx, Address::times_4,  0));
 6770   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 6771   movl(tmp4, Address(z, idx, Address::times_4,  0));
 6772 
 6773   add2_with_carry(carry2, tmp3, tmp4, carry);
 6774 
 6775   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6776   shrq(tmp3, 32);
 6777 
 6778   shlq(carry2, 32);
 6779   orq(tmp3, carry2);
 6780   movq(carry, tmp3);
 6781 
 6782   bind(L_post_third_loop_done);
 6783 }
 6784 
 6785 /**
 6786  * Code for BigInteger::multiplyToLen() intrinsic.
 6787  *
 6788  * rdi: x
 6789  * rax: xlen
 6790  * rsi: y
 6791  * rcx: ylen
 6792  * r8:  z
 6793  * r11: tmp0
 6794  * r12: tmp1
 6795  * r13: tmp2
 6796  * r14: tmp3
 6797  * r15: tmp4
 6798  * rbx: tmp5
 6799  *
 6800  */
 6801 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 6802                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 6803   ShortBranchVerifier sbv(this);
 6804   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 6805 
 6806   push(tmp0);
 6807   push(tmp1);
 6808   push(tmp2);
 6809   push(tmp3);
 6810   push(tmp4);
 6811   push(tmp5);
 6812 
 6813   push(xlen);
 6814 
 6815   const Register idx = tmp1;
 6816   const Register kdx = tmp2;
 6817   const Register xstart = tmp3;
 6818 
 6819   const Register y_idx = tmp4;
 6820   const Register carry = tmp5;
 6821   const Register product  = xlen;
 6822   const Register x_xstart = tmp0;
 6823 
 6824   // First Loop.
 6825   //
 6826   //  final static long LONG_MASK = 0xffffffffL;
 6827   //  int xstart = xlen - 1;
 6828   //  int ystart = ylen - 1;
 6829   //  long carry = 0;
 6830   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6831   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 6832   //    z[kdx] = (int)product;
 6833   //    carry = product >>> 32;
 6834   //  }
 6835   //  z[xstart] = (int)carry;
 6836   //
 6837 
 6838   movl(idx, ylen);               // idx = ylen;
 6839   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 6840   xorq(carry, carry);            // carry = 0;
 6841 
 6842   Label L_done;
 6843 
 6844   movl(xstart, xlen);
 6845   decrementl(xstart);
 6846   jcc(Assembler::negative, L_done);
 6847 
 6848   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 6849 
 6850   Label L_second_loop;
 6851   testl(kdx, kdx);
 6852   jcc(Assembler::zero, L_second_loop);
 6853 
 6854   Label L_carry;
 6855   subl(kdx, 1);
 6856   jcc(Assembler::zero, L_carry);
 6857 
 6858   movl(Address(z, kdx, Address::times_4,  0), carry);
 6859   shrq(carry, 32);
 6860   subl(kdx, 1);
 6861 
 6862   bind(L_carry);
 6863   movl(Address(z, kdx, Address::times_4,  0), carry);
 6864 
 6865   // Second and third (nested) loops.
 6866   //
 6867   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 6868   //   carry = 0;
 6869   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 6870   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 6871   //                    (z[k] & LONG_MASK) + carry;
 6872   //     z[k] = (int)product;
 6873   //     carry = product >>> 32;
 6874   //   }
 6875   //   z[i] = (int)carry;
 6876   // }
 6877   //
 6878   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 6879 
 6880   const Register jdx = tmp1;
 6881 
 6882   bind(L_second_loop);
 6883   xorl(carry, carry);    // carry = 0;
 6884   movl(jdx, ylen);       // j = ystart+1
 6885 
 6886   subl(xstart, 1);       // i = xstart-1;
 6887   jcc(Assembler::negative, L_done);
 6888 
 6889   push (z);
 6890 
 6891   Label L_last_x;
 6892   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 6893   subl(xstart, 1);       // i = xstart-1;
 6894   jcc(Assembler::negative, L_last_x);
 6895 
 6896   if (UseBMI2Instructions) {
 6897     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 6898     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 6899   } else {
 6900     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6901     rorq(x_xstart, 32);  // convert big-endian to little-endian
 6902   }
 6903 
 6904   Label L_third_loop_prologue;
 6905   bind(L_third_loop_prologue);
 6906 
 6907   push (x);
 6908   push (xstart);
 6909   push (ylen);
 6910 
 6911 
 6912   if (UseBMI2Instructions) {
 6913     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 6914   } else { // !UseBMI2Instructions
 6915     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 6916   }
 6917 
 6918   pop(ylen);
 6919   pop(xlen);
 6920   pop(x);
 6921   pop(z);
 6922 
 6923   movl(tmp3, xlen);
 6924   addl(tmp3, 1);
 6925   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6926   subl(tmp3, 1);
 6927   jccb(Assembler::negative, L_done);
 6928 
 6929   shrq(carry, 32);
 6930   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6931   jmp(L_second_loop);
 6932 
 6933   // Next infrequent code is moved outside loops.
 6934   bind(L_last_x);
 6935   if (UseBMI2Instructions) {
 6936     movl(rdx, Address(x,  0));
 6937   } else {
 6938     movl(x_xstart, Address(x,  0));
 6939   }
 6940   jmp(L_third_loop_prologue);
 6941 
 6942   bind(L_done);
 6943 
 6944   pop(xlen);
 6945 
 6946   pop(tmp5);
 6947   pop(tmp4);
 6948   pop(tmp3);
 6949   pop(tmp2);
 6950   pop(tmp1);
 6951   pop(tmp0);
 6952 }
 6953 
 6954 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 6955   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 6956   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 6957   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 6958   Label VECTOR8_TAIL, VECTOR4_TAIL;
 6959   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 6960   Label SAME_TILL_END, DONE;
 6961   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 6962 
 6963   //scale is in rcx in both Win64 and Unix
 6964   ShortBranchVerifier sbv(this);
 6965 
 6966   shlq(length);
 6967   xorq(result, result);
 6968 
 6969   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 6970       VM_Version::supports_avx512vlbw()) {
 6971     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 6972 
 6973     cmpq(length, 64);
 6974     jcc(Assembler::less, VECTOR32_TAIL);
 6975 
 6976     movq(tmp1, length);
 6977     andq(tmp1, 0x3F);      // tail count
 6978     andq(length, ~(0x3F)); //vector count
 6979 
 6980     bind(VECTOR64_LOOP);
 6981     // AVX512 code to compare 64 byte vectors.
 6982     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 6983     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6984     kortestql(k7, k7);
 6985     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 6986     addq(result, 64);
 6987     subq(length, 64);
 6988     jccb(Assembler::notZero, VECTOR64_LOOP);
 6989 
 6990     //bind(VECTOR64_TAIL);
 6991     testq(tmp1, tmp1);
 6992     jcc(Assembler::zero, SAME_TILL_END);
 6993 
 6994     //bind(VECTOR64_TAIL);
 6995     // AVX512 code to compare up to 63 byte vectors.
 6996     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 6997     shlxq(tmp2, tmp2, tmp1);
 6998     notq(tmp2);
 6999     kmovql(k3, tmp2);
 7000 
 7001     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 7002     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7003 
 7004     ktestql(k7, k3);
 7005     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 7006 
 7007     bind(VECTOR64_NOT_EQUAL);
 7008     kmovql(tmp1, k7);
 7009     notq(tmp1);
 7010     tzcntq(tmp1, tmp1);
 7011     addq(result, tmp1);
 7012     shrq(result);
 7013     jmp(DONE);
 7014     bind(VECTOR32_TAIL);
 7015   }
 7016 
 7017   cmpq(length, 8);
 7018   jcc(Assembler::equal, VECTOR8_LOOP);
 7019   jcc(Assembler::less, VECTOR4_TAIL);
 7020 
 7021   if (UseAVX >= 2) {
 7022     Label VECTOR16_TAIL, VECTOR32_LOOP;
 7023 
 7024     cmpq(length, 16);
 7025     jcc(Assembler::equal, VECTOR16_LOOP);
 7026     jcc(Assembler::less, VECTOR8_LOOP);
 7027 
 7028     cmpq(length, 32);
 7029     jccb(Assembler::less, VECTOR16_TAIL);
 7030 
 7031     subq(length, 32);
 7032     bind(VECTOR32_LOOP);
 7033     vmovdqu(rymm0, Address(obja, result));
 7034     vmovdqu(rymm1, Address(objb, result));
 7035     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 7036     vptest(rymm2, rymm2);
 7037     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7038     addq(result, 32);
 7039     subq(length, 32);
 7040     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7041     addq(length, 32);
 7042     jcc(Assembler::equal, SAME_TILL_END);
 7043     //falling through if less than 32 bytes left //close the branch here.
 7044 
 7045     bind(VECTOR16_TAIL);
 7046     cmpq(length, 16);
 7047     jccb(Assembler::less, VECTOR8_TAIL);
 7048     bind(VECTOR16_LOOP);
 7049     movdqu(rymm0, Address(obja, result));
 7050     movdqu(rymm1, Address(objb, result));
 7051     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7052     ptest(rymm2, rymm2);
 7053     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7054     addq(result, 16);
 7055     subq(length, 16);
 7056     jcc(Assembler::equal, SAME_TILL_END);
 7057     //falling through if less than 16 bytes left
 7058   } else {//regular intrinsics
 7059 
 7060     cmpq(length, 16);
 7061     jccb(Assembler::less, VECTOR8_TAIL);
 7062 
 7063     subq(length, 16);
 7064     bind(VECTOR16_LOOP);
 7065     movdqu(rymm0, Address(obja, result));
 7066     movdqu(rymm1, Address(objb, result));
 7067     pxor(rymm0, rymm1);
 7068     ptest(rymm0, rymm0);
 7069     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7070     addq(result, 16);
 7071     subq(length, 16);
 7072     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7073     addq(length, 16);
 7074     jcc(Assembler::equal, SAME_TILL_END);
 7075     //falling through if less than 16 bytes left
 7076   }
 7077 
 7078   bind(VECTOR8_TAIL);
 7079   cmpq(length, 8);
 7080   jccb(Assembler::less, VECTOR4_TAIL);
 7081   bind(VECTOR8_LOOP);
 7082   movq(tmp1, Address(obja, result));
 7083   movq(tmp2, Address(objb, result));
 7084   xorq(tmp1, tmp2);
 7085   testq(tmp1, tmp1);
 7086   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7087   addq(result, 8);
 7088   subq(length, 8);
 7089   jcc(Assembler::equal, SAME_TILL_END);
 7090   //falling through if less than 8 bytes left
 7091 
 7092   bind(VECTOR4_TAIL);
 7093   cmpq(length, 4);
 7094   jccb(Assembler::less, BYTES_TAIL);
 7095   bind(VECTOR4_LOOP);
 7096   movl(tmp1, Address(obja, result));
 7097   xorl(tmp1, Address(objb, result));
 7098   testl(tmp1, tmp1);
 7099   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7100   addq(result, 4);
 7101   subq(length, 4);
 7102   jcc(Assembler::equal, SAME_TILL_END);
 7103   //falling through if less than 4 bytes left
 7104 
 7105   bind(BYTES_TAIL);
 7106   bind(BYTES_LOOP);
 7107   load_unsigned_byte(tmp1, Address(obja, result));
 7108   load_unsigned_byte(tmp2, Address(objb, result));
 7109   xorl(tmp1, tmp2);
 7110   testl(tmp1, tmp1);
 7111   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7112   decq(length);
 7113   jcc(Assembler::zero, SAME_TILL_END);
 7114   incq(result);
 7115   load_unsigned_byte(tmp1, Address(obja, result));
 7116   load_unsigned_byte(tmp2, Address(objb, result));
 7117   xorl(tmp1, tmp2);
 7118   testl(tmp1, tmp1);
 7119   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7120   decq(length);
 7121   jcc(Assembler::zero, SAME_TILL_END);
 7122   incq(result);
 7123   load_unsigned_byte(tmp1, Address(obja, result));
 7124   load_unsigned_byte(tmp2, Address(objb, result));
 7125   xorl(tmp1, tmp2);
 7126   testl(tmp1, tmp1);
 7127   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7128   jmp(SAME_TILL_END);
 7129 
 7130   if (UseAVX >= 2) {
 7131     bind(VECTOR32_NOT_EQUAL);
 7132     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7133     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7134     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7135     vpmovmskb(tmp1, rymm0);
 7136     bsfq(tmp1, tmp1);
 7137     addq(result, tmp1);
 7138     shrq(result);
 7139     jmp(DONE);
 7140   }
 7141 
 7142   bind(VECTOR16_NOT_EQUAL);
 7143   if (UseAVX >= 2) {
 7144     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7145     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7146     pxor(rymm0, rymm2);
 7147   } else {
 7148     pcmpeqb(rymm2, rymm2);
 7149     pxor(rymm0, rymm1);
 7150     pcmpeqb(rymm0, rymm1);
 7151     pxor(rymm0, rymm2);
 7152   }
 7153   pmovmskb(tmp1, rymm0);
 7154   bsfq(tmp1, tmp1);
 7155   addq(result, tmp1);
 7156   shrq(result);
 7157   jmpb(DONE);
 7158 
 7159   bind(VECTOR8_NOT_EQUAL);
 7160   bind(VECTOR4_NOT_EQUAL);
 7161   bsfq(tmp1, tmp1);
 7162   shrq(tmp1, 3);
 7163   addq(result, tmp1);
 7164   bind(BYTES_NOT_EQUAL);
 7165   shrq(result);
 7166   jmpb(DONE);
 7167 
 7168   bind(SAME_TILL_END);
 7169   mov64(result, -1);
 7170 
 7171   bind(DONE);
 7172 }
 7173 
 7174 //Helper functions for square_to_len()
 7175 
 7176 /**
 7177  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7178  * Preserves x and z and modifies rest of the registers.
 7179  */
 7180 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7181   // Perform square and right shift by 1
 7182   // Handle odd xlen case first, then for even xlen do the following
 7183   // jlong carry = 0;
 7184   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7185   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7186   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7187   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7188   //     carry = (jlong)product;
 7189   // }
 7190 
 7191   xorq(tmp5, tmp5);     // carry
 7192   xorq(rdxReg, rdxReg);
 7193   xorl(tmp1, tmp1);     // index for x
 7194   xorl(tmp4, tmp4);     // index for z
 7195 
 7196   Label L_first_loop, L_first_loop_exit;
 7197 
 7198   testl(xlen, 1);
 7199   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7200 
 7201   // Square and right shift by 1 the odd element using 32 bit multiply
 7202   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7203   imulq(raxReg, raxReg);
 7204   shrq(raxReg, 1);
 7205   adcq(tmp5, 0);
 7206   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7207   incrementl(tmp1);
 7208   addl(tmp4, 2);
 7209 
 7210   // Square and  right shift by 1 the rest using 64 bit multiply
 7211   bind(L_first_loop);
 7212   cmpptr(tmp1, xlen);
 7213   jccb(Assembler::equal, L_first_loop_exit);
 7214 
 7215   // Square
 7216   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7217   rorq(raxReg, 32);    // convert big-endian to little-endian
 7218   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7219 
 7220   // Right shift by 1 and save carry
 7221   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7222   rcrq(rdxReg, 1);
 7223   rcrq(raxReg, 1);
 7224   adcq(tmp5, 0);
 7225 
 7226   // Store result in z
 7227   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7228   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7229 
 7230   // Update indices for x and z
 7231   addl(tmp1, 2);
 7232   addl(tmp4, 4);
 7233   jmp(L_first_loop);
 7234 
 7235   bind(L_first_loop_exit);
 7236 }
 7237 
 7238 
 7239 /**
 7240  * Perform the following multiply add operation using BMI2 instructions
 7241  * carry:sum = sum + op1*op2 + carry
 7242  * op2 should be in rdx
 7243  * op2 is preserved, all other registers are modified
 7244  */
 7245 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7246   // assert op2 is rdx
 7247   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7248   addq(sum, carry);
 7249   adcq(tmp2, 0);
 7250   addq(sum, op1);
 7251   adcq(tmp2, 0);
 7252   movq(carry, tmp2);
 7253 }
 7254 
 7255 /**
 7256  * Perform the following multiply add operation:
 7257  * carry:sum = sum + op1*op2 + carry
 7258  * Preserves op1, op2 and modifies rest of registers
 7259  */
 7260 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7261   // rdx:rax = op1 * op2
 7262   movq(raxReg, op2);
 7263   mulq(op1);
 7264 
 7265   //  rdx:rax = sum + carry + rdx:rax
 7266   addq(sum, carry);
 7267   adcq(rdxReg, 0);
 7268   addq(sum, raxReg);
 7269   adcq(rdxReg, 0);
 7270 
 7271   // carry:sum = rdx:sum
 7272   movq(carry, rdxReg);
 7273 }
 7274 
 7275 /**
 7276  * Add 64 bit long carry into z[] with carry propagation.
 7277  * Preserves z and carry register values and modifies rest of registers.
 7278  *
 7279  */
 7280 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7281   Label L_fourth_loop, L_fourth_loop_exit;
 7282 
 7283   movl(tmp1, 1);
 7284   subl(zlen, 2);
 7285   addq(Address(z, zlen, Address::times_4, 0), carry);
 7286 
 7287   bind(L_fourth_loop);
 7288   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7289   subl(zlen, 2);
 7290   jccb(Assembler::negative, L_fourth_loop_exit);
 7291   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7292   jmp(L_fourth_loop);
 7293   bind(L_fourth_loop_exit);
 7294 }
 7295 
 7296 /**
 7297  * Shift z[] left by 1 bit.
 7298  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7299  *
 7300  */
 7301 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7302 
 7303   Label L_fifth_loop, L_fifth_loop_exit;
 7304 
 7305   // Fifth loop
 7306   // Perform primitiveLeftShift(z, zlen, 1)
 7307 
 7308   const Register prev_carry = tmp1;
 7309   const Register new_carry = tmp4;
 7310   const Register value = tmp2;
 7311   const Register zidx = tmp3;
 7312 
 7313   // int zidx, carry;
 7314   // long value;
 7315   // carry = 0;
 7316   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 7317   //    (carry:value)  = (z[i] << 1) | carry ;
 7318   //    z[i] = value;
 7319   // }
 7320 
 7321   movl(zidx, zlen);
 7322   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 7323 
 7324   bind(L_fifth_loop);
 7325   decl(zidx);  // Use decl to preserve carry flag
 7326   decl(zidx);
 7327   jccb(Assembler::negative, L_fifth_loop_exit);
 7328 
 7329   if (UseBMI2Instructions) {
 7330      movq(value, Address(z, zidx, Address::times_4, 0));
 7331      rclq(value, 1);
 7332      rorxq(value, value, 32);
 7333      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7334   }
 7335   else {
 7336     // clear new_carry
 7337     xorl(new_carry, new_carry);
 7338 
 7339     // Shift z[i] by 1, or in previous carry and save new carry
 7340     movq(value, Address(z, zidx, Address::times_4, 0));
 7341     shlq(value, 1);
 7342     adcl(new_carry, 0);
 7343 
 7344     orq(value, prev_carry);
 7345     rorq(value, 0x20);
 7346     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7347 
 7348     // Set previous carry = new carry
 7349     movl(prev_carry, new_carry);
 7350   }
 7351   jmp(L_fifth_loop);
 7352 
 7353   bind(L_fifth_loop_exit);
 7354 }
 7355 
 7356 
 7357 /**
 7358  * Code for BigInteger::squareToLen() intrinsic
 7359  *
 7360  * rdi: x
 7361  * rsi: len
 7362  * r8:  z
 7363  * rcx: zlen
 7364  * r12: tmp1
 7365  * r13: tmp2
 7366  * r14: tmp3
 7367  * r15: tmp4
 7368  * rbx: tmp5
 7369  *
 7370  */
 7371 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) {
 7372 
 7373   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 7374   push(tmp1);
 7375   push(tmp2);
 7376   push(tmp3);
 7377   push(tmp4);
 7378   push(tmp5);
 7379 
 7380   // First loop
 7381   // Store the squares, right shifted one bit (i.e., divided by 2).
 7382   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7383 
 7384   // Add in off-diagonal sums.
 7385   //
 7386   // Second, third (nested) and fourth loops.
 7387   // zlen +=2;
 7388   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 7389   //    carry = 0;
 7390   //    long op2 = x[xidx:xidx+1];
 7391   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 7392   //       k -= 2;
 7393   //       long op1 = x[j:j+1];
 7394   //       long sum = z[k:k+1];
 7395   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 7396   //       z[k:k+1] = sum;
 7397   //    }
 7398   //    add_one_64(z, k, carry, tmp_regs);
 7399   // }
 7400 
 7401   const Register carry = tmp5;
 7402   const Register sum = tmp3;
 7403   const Register op1 = tmp4;
 7404   Register op2 = tmp2;
 7405 
 7406   push(zlen);
 7407   push(len);
 7408   addl(zlen,2);
 7409   bind(L_second_loop);
 7410   xorq(carry, carry);
 7411   subl(zlen, 4);
 7412   subl(len, 2);
 7413   push(zlen);
 7414   push(len);
 7415   cmpl(len, 0);
 7416   jccb(Assembler::lessEqual, L_second_loop_exit);
 7417 
 7418   // Multiply an array by one 64 bit long.
 7419   if (UseBMI2Instructions) {
 7420     op2 = rdxReg;
 7421     movq(op2, Address(x, len, Address::times_4,  0));
 7422     rorxq(op2, op2, 32);
 7423   }
 7424   else {
 7425     movq(op2, Address(x, len, Address::times_4,  0));
 7426     rorq(op2, 32);
 7427   }
 7428 
 7429   bind(L_third_loop);
 7430   decrementl(len);
 7431   jccb(Assembler::negative, L_third_loop_exit);
 7432   decrementl(len);
 7433   jccb(Assembler::negative, L_last_x);
 7434 
 7435   movq(op1, Address(x, len, Address::times_4,  0));
 7436   rorq(op1, 32);
 7437 
 7438   bind(L_multiply);
 7439   subl(zlen, 2);
 7440   movq(sum, Address(z, zlen, Address::times_4,  0));
 7441 
 7442   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 7443   if (UseBMI2Instructions) {
 7444     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 7445   }
 7446   else {
 7447     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7448   }
 7449 
 7450   movq(Address(z, zlen, Address::times_4, 0), sum);
 7451 
 7452   jmp(L_third_loop);
 7453   bind(L_third_loop_exit);
 7454 
 7455   // Fourth loop
 7456   // Add 64 bit long carry into z with carry propagation.
 7457   // Uses offsetted zlen.
 7458   add_one_64(z, zlen, carry, tmp1);
 7459 
 7460   pop(len);
 7461   pop(zlen);
 7462   jmp(L_second_loop);
 7463 
 7464   // Next infrequent code is moved outside loops.
 7465   bind(L_last_x);
 7466   movl(op1, Address(x, 0));
 7467   jmp(L_multiply);
 7468 
 7469   bind(L_second_loop_exit);
 7470   pop(len);
 7471   pop(zlen);
 7472   pop(len);
 7473   pop(zlen);
 7474 
 7475   // Fifth loop
 7476   // Shift z left 1 bit.
 7477   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 7478 
 7479   // z[zlen-1] |= x[len-1] & 1;
 7480   movl(tmp3, Address(x, len, Address::times_4, -4));
 7481   andl(tmp3, 1);
 7482   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 7483 
 7484   pop(tmp5);
 7485   pop(tmp4);
 7486   pop(tmp3);
 7487   pop(tmp2);
 7488   pop(tmp1);
 7489 }
 7490 
 7491 /**
 7492  * Helper function for mul_add()
 7493  * Multiply the in[] by int k and add to out[] starting at offset offs using
 7494  * 128 bit by 32 bit multiply and return the carry in tmp5.
 7495  * Only quad int aligned length of in[] is operated on in this function.
 7496  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 7497  * This function preserves out, in and k registers.
 7498  * len and offset point to the appropriate index in "in" & "out" correspondingly
 7499  * tmp5 has the carry.
 7500  * other registers are temporary and are modified.
 7501  *
 7502  */
 7503 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 7504   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 7505   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7506 
 7507   Label L_first_loop, L_first_loop_exit;
 7508 
 7509   movl(tmp1, len);
 7510   shrl(tmp1, 2);
 7511 
 7512   bind(L_first_loop);
 7513   subl(tmp1, 1);
 7514   jccb(Assembler::negative, L_first_loop_exit);
 7515 
 7516   subl(len, 4);
 7517   subl(offset, 4);
 7518 
 7519   Register op2 = tmp2;
 7520   const Register sum = tmp3;
 7521   const Register op1 = tmp4;
 7522   const Register carry = tmp5;
 7523 
 7524   if (UseBMI2Instructions) {
 7525     op2 = rdxReg;
 7526   }
 7527 
 7528   movq(op1, Address(in, len, Address::times_4,  8));
 7529   rorq(op1, 32);
 7530   movq(sum, Address(out, offset, Address::times_4,  8));
 7531   rorq(sum, 32);
 7532   if (UseBMI2Instructions) {
 7533     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7534   }
 7535   else {
 7536     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7537   }
 7538   // Store back in big endian from little endian
 7539   rorq(sum, 0x20);
 7540   movq(Address(out, offset, Address::times_4,  8), sum);
 7541 
 7542   movq(op1, Address(in, len, Address::times_4,  0));
 7543   rorq(op1, 32);
 7544   movq(sum, Address(out, offset, Address::times_4,  0));
 7545   rorq(sum, 32);
 7546   if (UseBMI2Instructions) {
 7547     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7548   }
 7549   else {
 7550     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7551   }
 7552   // Store back in big endian from little endian
 7553   rorq(sum, 0x20);
 7554   movq(Address(out, offset, Address::times_4,  0), sum);
 7555 
 7556   jmp(L_first_loop);
 7557   bind(L_first_loop_exit);
 7558 }
 7559 
 7560 /**
 7561  * Code for BigInteger::mulAdd() intrinsic
 7562  *
 7563  * rdi: out
 7564  * rsi: in
 7565  * r11: offs (out.length - offset)
 7566  * rcx: len
 7567  * r8:  k
 7568  * r12: tmp1
 7569  * r13: tmp2
 7570  * r14: tmp3
 7571  * r15: tmp4
 7572  * rbx: tmp5
 7573  * Multiply the in[] by word k and add to out[], return the carry in rax
 7574  */
 7575 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 7576    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 7577    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7578 
 7579   Label L_carry, L_last_in, L_done;
 7580 
 7581 // carry = 0;
 7582 // for (int j=len-1; j >= 0; j--) {
 7583 //    long product = (in[j] & LONG_MASK) * kLong +
 7584 //                   (out[offs] & LONG_MASK) + carry;
 7585 //    out[offs--] = (int)product;
 7586 //    carry = product >>> 32;
 7587 // }
 7588 //
 7589   push(tmp1);
 7590   push(tmp2);
 7591   push(tmp3);
 7592   push(tmp4);
 7593   push(tmp5);
 7594 
 7595   Register op2 = tmp2;
 7596   const Register sum = tmp3;
 7597   const Register op1 = tmp4;
 7598   const Register carry =  tmp5;
 7599 
 7600   if (UseBMI2Instructions) {
 7601     op2 = rdxReg;
 7602     movl(op2, k);
 7603   }
 7604   else {
 7605     movl(op2, k);
 7606   }
 7607 
 7608   xorq(carry, carry);
 7609 
 7610   //First loop
 7611 
 7612   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 7613   //The carry is in tmp5
 7614   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7615 
 7616   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 7617   decrementl(len);
 7618   jccb(Assembler::negative, L_carry);
 7619   decrementl(len);
 7620   jccb(Assembler::negative, L_last_in);
 7621 
 7622   movq(op1, Address(in, len, Address::times_4,  0));
 7623   rorq(op1, 32);
 7624 
 7625   subl(offs, 2);
 7626   movq(sum, Address(out, offs, Address::times_4,  0));
 7627   rorq(sum, 32);
 7628 
 7629   if (UseBMI2Instructions) {
 7630     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7631   }
 7632   else {
 7633     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7634   }
 7635 
 7636   // Store back in big endian from little endian
 7637   rorq(sum, 0x20);
 7638   movq(Address(out, offs, Address::times_4,  0), sum);
 7639 
 7640   testl(len, len);
 7641   jccb(Assembler::zero, L_carry);
 7642 
 7643   //Multiply the last in[] entry, if any
 7644   bind(L_last_in);
 7645   movl(op1, Address(in, 0));
 7646   movl(sum, Address(out, offs, Address::times_4,  -4));
 7647 
 7648   movl(raxReg, k);
 7649   mull(op1); //tmp4 * eax -> edx:eax
 7650   addl(sum, carry);
 7651   adcl(rdxReg, 0);
 7652   addl(sum, raxReg);
 7653   adcl(rdxReg, 0);
 7654   movl(carry, rdxReg);
 7655 
 7656   movl(Address(out, offs, Address::times_4,  -4), sum);
 7657 
 7658   bind(L_carry);
 7659   //return tmp5/carry as carry in rax
 7660   movl(rax, carry);
 7661 
 7662   bind(L_done);
 7663   pop(tmp5);
 7664   pop(tmp4);
 7665   pop(tmp3);
 7666   pop(tmp2);
 7667   pop(tmp1);
 7668 }
 7669 
 7670 /**
 7671  * Emits code to update CRC-32 with a byte value according to constants in table
 7672  *
 7673  * @param [in,out]crc   Register containing the crc.
 7674  * @param [in]val       Register containing the byte to fold into the CRC.
 7675  * @param [in]table     Register containing the table of crc constants.
 7676  *
 7677  * uint32_t crc;
 7678  * val = crc_table[(val ^ crc) & 0xFF];
 7679  * crc = val ^ (crc >> 8);
 7680  *
 7681  */
 7682 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 7683   xorl(val, crc);
 7684   andl(val, 0xFF);
 7685   shrl(crc, 8); // unsigned shift
 7686   xorl(crc, Address(table, val, Address::times_4, 0));
 7687 }
 7688 
 7689 /**
 7690  * Fold 128-bit data chunk
 7691  */
 7692 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 7693   if (UseAVX > 0) {
 7694     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 7695     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 7696     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 7697     pxor(xcrc, xtmp);
 7698   } else {
 7699     movdqa(xtmp, xcrc);
 7700     pclmulhdq(xtmp, xK);   // [123:64]
 7701     pclmulldq(xcrc, xK);   // [63:0]
 7702     pxor(xcrc, xtmp);
 7703     movdqu(xtmp, Address(buf, offset));
 7704     pxor(xcrc, xtmp);
 7705   }
 7706 }
 7707 
 7708 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 7709   if (UseAVX > 0) {
 7710     vpclmulhdq(xtmp, xK, xcrc);
 7711     vpclmulldq(xcrc, xK, xcrc);
 7712     pxor(xcrc, xbuf);
 7713     pxor(xcrc, xtmp);
 7714   } else {
 7715     movdqa(xtmp, xcrc);
 7716     pclmulhdq(xtmp, xK);
 7717     pclmulldq(xcrc, xK);
 7718     pxor(xcrc, xbuf);
 7719     pxor(xcrc, xtmp);
 7720   }
 7721 }
 7722 
 7723 /**
 7724  * 8-bit folds to compute 32-bit CRC
 7725  *
 7726  * uint64_t xcrc;
 7727  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 7728  */
 7729 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 7730   movdl(tmp, xcrc);
 7731   andl(tmp, 0xFF);
 7732   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 7733   psrldq(xcrc, 1); // unsigned shift one byte
 7734   pxor(xcrc, xtmp);
 7735 }
 7736 
 7737 /**
 7738  * uint32_t crc;
 7739  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 7740  */
 7741 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 7742   movl(tmp, crc);
 7743   andl(tmp, 0xFF);
 7744   shrl(crc, 8);
 7745   xorl(crc, Address(table, tmp, Address::times_4, 0));
 7746 }
 7747 
 7748 /**
 7749  * @param crc   register containing existing CRC (32-bit)
 7750  * @param buf   register pointing to input byte buffer (byte*)
 7751  * @param len   register containing number of bytes
 7752  * @param table register that will contain address of CRC table
 7753  * @param tmp   scratch register
 7754  */
 7755 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 7756   assert_different_registers(crc, buf, len, table, tmp, rax);
 7757 
 7758   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 7759   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 7760 
 7761   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 7762   // context for the registers used, where all instructions below are using 128-bit mode
 7763   // On EVEX without VL and BW, these instructions will all be AVX.
 7764   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 7765   notl(crc); // ~crc
 7766   cmpl(len, 16);
 7767   jcc(Assembler::less, L_tail);
 7768 
 7769   // Align buffer to 16 bytes
 7770   movl(tmp, buf);
 7771   andl(tmp, 0xF);
 7772   jccb(Assembler::zero, L_aligned);
 7773   subl(tmp,  16);
 7774   addl(len, tmp);
 7775 
 7776   align(4);
 7777   BIND(L_align_loop);
 7778   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7779   update_byte_crc32(crc, rax, table);
 7780   increment(buf);
 7781   incrementl(tmp);
 7782   jccb(Assembler::less, L_align_loop);
 7783 
 7784   BIND(L_aligned);
 7785   movl(tmp, len); // save
 7786   shrl(len, 4);
 7787   jcc(Assembler::zero, L_tail_restore);
 7788 
 7789   // Fold crc into first bytes of vector
 7790   movdqa(xmm1, Address(buf, 0));
 7791   movdl(rax, xmm1);
 7792   xorl(crc, rax);
 7793   if (VM_Version::supports_sse4_1()) {
 7794     pinsrd(xmm1, crc, 0);
 7795   } else {
 7796     pinsrw(xmm1, crc, 0);
 7797     shrl(crc, 16);
 7798     pinsrw(xmm1, crc, 1);
 7799   }
 7800   addptr(buf, 16);
 7801   subl(len, 4); // len > 0
 7802   jcc(Assembler::less, L_fold_tail);
 7803 
 7804   movdqa(xmm2, Address(buf,  0));
 7805   movdqa(xmm3, Address(buf, 16));
 7806   movdqa(xmm4, Address(buf, 32));
 7807   addptr(buf, 48);
 7808   subl(len, 3);
 7809   jcc(Assembler::lessEqual, L_fold_512b);
 7810 
 7811   // Fold total 512 bits of polynomial on each iteration,
 7812   // 128 bits per each of 4 parallel streams.
 7813   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 7814 
 7815   align32();
 7816   BIND(L_fold_512b_loop);
 7817   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7818   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 7819   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 7820   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 7821   addptr(buf, 64);
 7822   subl(len, 4);
 7823   jcc(Assembler::greater, L_fold_512b_loop);
 7824 
 7825   // Fold 512 bits to 128 bits.
 7826   BIND(L_fold_512b);
 7827   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7828   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 7829   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 7830   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 7831 
 7832   // Fold the rest of 128 bits data chunks
 7833   BIND(L_fold_tail);
 7834   addl(len, 3);
 7835   jccb(Assembler::lessEqual, L_fold_128b);
 7836   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7837 
 7838   BIND(L_fold_tail_loop);
 7839   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7840   addptr(buf, 16);
 7841   decrementl(len);
 7842   jccb(Assembler::greater, L_fold_tail_loop);
 7843 
 7844   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 7845   BIND(L_fold_128b);
 7846   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 7847   if (UseAVX > 0) {
 7848     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 7849     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 7850     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 7851   } else {
 7852     movdqa(xmm2, xmm0);
 7853     pclmulqdq(xmm2, xmm1, 0x1);
 7854     movdqa(xmm3, xmm0);
 7855     pand(xmm3, xmm2);
 7856     pclmulqdq(xmm0, xmm3, 0x1);
 7857   }
 7858   psrldq(xmm1, 8);
 7859   psrldq(xmm2, 4);
 7860   pxor(xmm0, xmm1);
 7861   pxor(xmm0, xmm2);
 7862 
 7863   // 8 8-bit folds to compute 32-bit CRC.
 7864   for (int j = 0; j < 4; j++) {
 7865     fold_8bit_crc32(xmm0, table, xmm1, rax);
 7866   }
 7867   movdl(crc, xmm0); // mov 32 bits to general register
 7868   for (int j = 0; j < 4; j++) {
 7869     fold_8bit_crc32(crc, table, rax);
 7870   }
 7871 
 7872   BIND(L_tail_restore);
 7873   movl(len, tmp); // restore
 7874   BIND(L_tail);
 7875   andl(len, 0xf);
 7876   jccb(Assembler::zero, L_exit);
 7877 
 7878   // Fold the rest of bytes
 7879   align(4);
 7880   BIND(L_tail_loop);
 7881   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7882   update_byte_crc32(crc, rax, table);
 7883   increment(buf);
 7884   decrementl(len);
 7885   jccb(Assembler::greater, L_tail_loop);
 7886 
 7887   BIND(L_exit);
 7888   notl(crc); // ~c
 7889 }
 7890 
 7891 // Helper function for AVX 512 CRC32
 7892 // Fold 512-bit data chunks
 7893 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 7894                                              Register pos, int offset) {
 7895   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 7896   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 7897   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 7898   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 7899   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 7900 }
 7901 
 7902 // Helper function for AVX 512 CRC32
 7903 // Compute CRC32 for < 256B buffers
 7904 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 7905                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 7906                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 7907 
 7908   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 7909   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 7910   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 7911 
 7912   // check if there is enough buffer to be able to fold 16B at a time
 7913   cmpl(len, 32);
 7914   jcc(Assembler::less, L_less_than_32);
 7915 
 7916   // if there is, load the constants
 7917   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 7918   movdl(xmm0, crc);                        // get the initial crc value
 7919   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7920   pxor(xmm7, xmm0);
 7921 
 7922   // update the buffer pointer
 7923   addl(pos, 16);
 7924   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 7925   subl(len, 32);
 7926   jmp(L_16B_reduction_loop);
 7927 
 7928   bind(L_less_than_32);
 7929   //mov initial crc to the return value. this is necessary for zero - length buffers.
 7930   movl(rax, crc);
 7931   testl(len, len);
 7932   jcc(Assembler::equal, L_cleanup);
 7933 
 7934   movdl(xmm0, crc);                        //get the initial crc value
 7935 
 7936   cmpl(len, 16);
 7937   jcc(Assembler::equal, L_exact_16_left);
 7938   jcc(Assembler::less, L_less_than_16_left);
 7939 
 7940   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7941   pxor(xmm7, xmm0);                       //xor the initial crc value
 7942   addl(pos, 16);
 7943   subl(len, 16);
 7944   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 7945   jmp(L_get_last_two_xmms);
 7946 
 7947   bind(L_less_than_16_left);
 7948   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 7949   pxor(xmm1, xmm1);
 7950   movptr(tmp1, rsp);
 7951   movdqu(Address(tmp1, 0 * 16), xmm1);
 7952 
 7953   cmpl(len, 4);
 7954   jcc(Assembler::less, L_only_less_than_4);
 7955 
 7956   //backup the counter value
 7957   movl(tmp2, len);
 7958   cmpl(len, 8);
 7959   jcc(Assembler::less, L_less_than_8_left);
 7960 
 7961   //load 8 Bytes
 7962   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 7963   movq(Address(tmp1, 0 * 16), rax);
 7964   addptr(tmp1, 8);
 7965   subl(len, 8);
 7966   addl(pos, 8);
 7967 
 7968   bind(L_less_than_8_left);
 7969   cmpl(len, 4);
 7970   jcc(Assembler::less, L_less_than_4_left);
 7971 
 7972   //load 4 Bytes
 7973   movl(rax, Address(buf, pos, Address::times_1, 0));
 7974   movl(Address(tmp1, 0 * 16), rax);
 7975   addptr(tmp1, 4);
 7976   subl(len, 4);
 7977   addl(pos, 4);
 7978 
 7979   bind(L_less_than_4_left);
 7980   cmpl(len, 2);
 7981   jcc(Assembler::less, L_less_than_2_left);
 7982 
 7983   // load 2 Bytes
 7984   movw(rax, Address(buf, pos, Address::times_1, 0));
 7985   movl(Address(tmp1, 0 * 16), rax);
 7986   addptr(tmp1, 2);
 7987   subl(len, 2);
 7988   addl(pos, 2);
 7989 
 7990   bind(L_less_than_2_left);
 7991   cmpl(len, 1);
 7992   jcc(Assembler::less, L_zero_left);
 7993 
 7994   // load 1 Byte
 7995   movb(rax, Address(buf, pos, Address::times_1, 0));
 7996   movb(Address(tmp1, 0 * 16), rax);
 7997 
 7998   bind(L_zero_left);
 7999   movdqu(xmm7, Address(rsp, 0));
 8000   pxor(xmm7, xmm0);                       //xor the initial crc value
 8001 
 8002   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8003   movdqu(xmm0, Address(rax, tmp2));
 8004   pshufb(xmm7, xmm0);
 8005   jmp(L_128_done);
 8006 
 8007   bind(L_exact_16_left);
 8008   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 8009   pxor(xmm7, xmm0);                       //xor the initial crc value
 8010   jmp(L_128_done);
 8011 
 8012   bind(L_only_less_than_4);
 8013   cmpl(len, 3);
 8014   jcc(Assembler::less, L_only_less_than_3);
 8015 
 8016   // load 3 Bytes
 8017   movb(rax, Address(buf, pos, Address::times_1, 0));
 8018   movb(Address(tmp1, 0), rax);
 8019 
 8020   movb(rax, Address(buf, pos, Address::times_1, 1));
 8021   movb(Address(tmp1, 1), rax);
 8022 
 8023   movb(rax, Address(buf, pos, Address::times_1, 2));
 8024   movb(Address(tmp1, 2), rax);
 8025 
 8026   movdqu(xmm7, Address(rsp, 0));
 8027   pxor(xmm7, xmm0);                     //xor the initial crc value
 8028 
 8029   pslldq(xmm7, 0x5);
 8030   jmp(L_barrett);
 8031   bind(L_only_less_than_3);
 8032   cmpl(len, 2);
 8033   jcc(Assembler::less, L_only_less_than_2);
 8034 
 8035   // load 2 Bytes
 8036   movb(rax, Address(buf, pos, Address::times_1, 0));
 8037   movb(Address(tmp1, 0), rax);
 8038 
 8039   movb(rax, Address(buf, pos, Address::times_1, 1));
 8040   movb(Address(tmp1, 1), rax);
 8041 
 8042   movdqu(xmm7, Address(rsp, 0));
 8043   pxor(xmm7, xmm0);                     //xor the initial crc value
 8044 
 8045   pslldq(xmm7, 0x6);
 8046   jmp(L_barrett);
 8047 
 8048   bind(L_only_less_than_2);
 8049   //load 1 Byte
 8050   movb(rax, Address(buf, pos, Address::times_1, 0));
 8051   movb(Address(tmp1, 0), rax);
 8052 
 8053   movdqu(xmm7, Address(rsp, 0));
 8054   pxor(xmm7, xmm0);                     //xor the initial crc value
 8055 
 8056   pslldq(xmm7, 0x7);
 8057 }
 8058 
 8059 /**
 8060 * Compute CRC32 using AVX512 instructions
 8061 * param crc   register containing existing CRC (32-bit)
 8062 * param buf   register pointing to input byte buffer (byte*)
 8063 * param len   register containing number of bytes
 8064 * param table address of crc or crc32c table
 8065 * param tmp1  scratch register
 8066 * param tmp2  scratch register
 8067 * return rax  result register
 8068 *
 8069 * This routine is identical for crc32c with the exception of the precomputed constant
 8070 * table which will be passed as the table argument.  The calculation steps are
 8071 * the same for both variants.
 8072 */
 8073 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8074   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8075 
 8076   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8077   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8078   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8079   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8080   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8081 
 8082   const Register pos = r12;
 8083   push(r12);
 8084   subptr(rsp, 16 * 2 + 8);
 8085 
 8086   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8087   // context for the registers used, where all instructions below are using 128-bit mode
 8088   // On EVEX without VL and BW, these instructions will all be AVX.
 8089   movl(pos, 0);
 8090 
 8091   // check if smaller than 256B
 8092   cmpl(len, 256);
 8093   jcc(Assembler::less, L_less_than_256);
 8094 
 8095   // load the initial crc value
 8096   movdl(xmm10, crc);
 8097 
 8098   // receive the initial 64B data, xor the initial crc value
 8099   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8100   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8101   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8102   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8103 
 8104   subl(len, 256);
 8105   cmpl(len, 256);
 8106   jcc(Assembler::less, L_fold_128_B_loop);
 8107 
 8108   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8109   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8110   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8111   subl(len, 256);
 8112 
 8113   bind(L_fold_256_B_loop);
 8114   addl(pos, 256);
 8115   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8116   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8117   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8118   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8119 
 8120   subl(len, 256);
 8121   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8122 
 8123   // Fold 256 into 128
 8124   addl(pos, 256);
 8125   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8126   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8127   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8128 
 8129   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8130   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8131   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8132 
 8133   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8134   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8135 
 8136   addl(len, 128);
 8137   jmp(L_fold_128_B_register);
 8138 
 8139   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8140   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8141 
 8142   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8143   bind(L_fold_128_B_loop);
 8144   addl(pos, 128);
 8145   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8146   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8147 
 8148   subl(len, 128);
 8149   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8150 
 8151   addl(pos, 128);
 8152 
 8153   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8154   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8155   bind(L_fold_128_B_register);
 8156   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8157   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8158   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8159   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8160   // save last that has no multiplicand
 8161   vextracti64x2(xmm7, xmm4, 3);
 8162 
 8163   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8164   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8165   // Needed later in reduction loop
 8166   movdqu(xmm10, Address(table, 1 * 16));
 8167   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8168   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8169 
 8170   // Swap 1,0,3,2 - 01 00 11 10
 8171   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8172   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8173   vextracti128(xmm5, xmm8, 1);
 8174   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8175 
 8176   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8177   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8178   addl(len, 128 - 16);
 8179   jcc(Assembler::less, L_final_reduction_for_128);
 8180 
 8181   bind(L_16B_reduction_loop);
 8182   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8183   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8184   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8185   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8186   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8187   addl(pos, 16);
 8188   subl(len, 16);
 8189   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8190 
 8191   bind(L_final_reduction_for_128);
 8192   addl(len, 16);
 8193   jcc(Assembler::equal, L_128_done);
 8194 
 8195   bind(L_get_last_two_xmms);
 8196   movdqu(xmm2, xmm7);
 8197   addl(pos, len);
 8198   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8199   subl(pos, len);
 8200 
 8201   // get rid of the extra data that was loaded before
 8202   // load the shift constant
 8203   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8204   movdqu(xmm0, Address(rax, len));
 8205   addl(rax, len);
 8206 
 8207   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8208   //Change mask to 512
 8209   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8210   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8211 
 8212   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8213   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8214   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8215   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8216   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8217 
 8218   bind(L_128_done);
 8219   // compute crc of a 128-bit value
 8220   movdqu(xmm10, Address(table, 3 * 16));
 8221   movdqu(xmm0, xmm7);
 8222 
 8223   // 64b fold
 8224   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8225   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8226   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8227 
 8228   // 32b fold
 8229   movdqu(xmm0, xmm7);
 8230   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8231   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8232   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8233   jmp(L_barrett);
 8234 
 8235   bind(L_less_than_256);
 8236   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);
 8237 
 8238   //barrett reduction
 8239   bind(L_barrett);
 8240   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8241   movdqu(xmm1, xmm7);
 8242   movdqu(xmm2, xmm7);
 8243   movdqu(xmm10, Address(table, 4 * 16));
 8244 
 8245   pclmulqdq(xmm7, xmm10, 0x0);
 8246   pxor(xmm7, xmm2);
 8247   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8248   movdqu(xmm2, xmm7);
 8249   pclmulqdq(xmm7, xmm10, 0x10);
 8250   pxor(xmm7, xmm2);
 8251   pxor(xmm7, xmm1);
 8252   pextrd(crc, xmm7, 2);
 8253 
 8254   bind(L_cleanup);
 8255   addptr(rsp, 16 * 2 + 8);
 8256   pop(r12);
 8257 }
 8258 
 8259 // S. Gueron / Information Processing Letters 112 (2012) 184
 8260 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8261 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8262 // Output: the 64-bit carry-less product of B * CONST
 8263 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8264                                      Register tmp1, Register tmp2, Register tmp3) {
 8265   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8266   if (n > 0) {
 8267     addq(tmp3, n * 256 * 8);
 8268   }
 8269   //    Q1 = TABLEExt[n][B & 0xFF];
 8270   movl(tmp1, in);
 8271   andl(tmp1, 0x000000FF);
 8272   shll(tmp1, 3);
 8273   addq(tmp1, tmp3);
 8274   movq(tmp1, Address(tmp1, 0));
 8275 
 8276   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8277   movl(tmp2, in);
 8278   shrl(tmp2, 8);
 8279   andl(tmp2, 0x000000FF);
 8280   shll(tmp2, 3);
 8281   addq(tmp2, tmp3);
 8282   movq(tmp2, Address(tmp2, 0));
 8283 
 8284   shlq(tmp2, 8);
 8285   xorq(tmp1, tmp2);
 8286 
 8287   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8288   movl(tmp2, in);
 8289   shrl(tmp2, 16);
 8290   andl(tmp2, 0x000000FF);
 8291   shll(tmp2, 3);
 8292   addq(tmp2, tmp3);
 8293   movq(tmp2, Address(tmp2, 0));
 8294 
 8295   shlq(tmp2, 16);
 8296   xorq(tmp1, tmp2);
 8297 
 8298   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8299   shrl(in, 24);
 8300   andl(in, 0x000000FF);
 8301   shll(in, 3);
 8302   addq(in, tmp3);
 8303   movq(in, Address(in, 0));
 8304 
 8305   shlq(in, 24);
 8306   xorq(in, tmp1);
 8307   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8308 }
 8309 
 8310 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8311                                       Register in_out,
 8312                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8313                                       XMMRegister w_xtmp2,
 8314                                       Register tmp1,
 8315                                       Register n_tmp2, Register n_tmp3) {
 8316   if (is_pclmulqdq_supported) {
 8317     movdl(w_xtmp1, in_out); // modified blindly
 8318 
 8319     movl(tmp1, const_or_pre_comp_const_index);
 8320     movdl(w_xtmp2, tmp1);
 8321     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8322 
 8323     movdq(in_out, w_xtmp1);
 8324   } else {
 8325     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 8326   }
 8327 }
 8328 
 8329 // Recombination Alternative 2: No bit-reflections
 8330 // T1 = (CRC_A * U1) << 1
 8331 // T2 = (CRC_B * U2) << 1
 8332 // C1 = T1 >> 32
 8333 // C2 = T2 >> 32
 8334 // T1 = T1 & 0xFFFFFFFF
 8335 // T2 = T2 & 0xFFFFFFFF
 8336 // T1 = CRC32(0, T1)
 8337 // T2 = CRC32(0, T2)
 8338 // C1 = C1 ^ T1
 8339 // C2 = C2 ^ T2
 8340 // CRC = C1 ^ C2 ^ CRC_C
 8341 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,
 8342                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8343                                      Register tmp1, Register tmp2,
 8344                                      Register n_tmp3) {
 8345   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8346   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8347   shlq(in_out, 1);
 8348   movl(tmp1, in_out);
 8349   shrq(in_out, 32);
 8350   xorl(tmp2, tmp2);
 8351   crc32(tmp2, tmp1, 4);
 8352   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 8353   shlq(in1, 1);
 8354   movl(tmp1, in1);
 8355   shrq(in1, 32);
 8356   xorl(tmp2, tmp2);
 8357   crc32(tmp2, tmp1, 4);
 8358   xorl(in1, tmp2);
 8359   xorl(in_out, in1);
 8360   xorl(in_out, in2);
 8361 }
 8362 
 8363 // Set N to predefined value
 8364 // Subtract from a length of a buffer
 8365 // execute in a loop:
 8366 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 8367 // for i = 1 to N do
 8368 //  CRC_A = CRC32(CRC_A, A[i])
 8369 //  CRC_B = CRC32(CRC_B, B[i])
 8370 //  CRC_C = CRC32(CRC_C, C[i])
 8371 // end for
 8372 // Recombine
 8373 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,
 8374                                        Register in_out1, Register in_out2, Register in_out3,
 8375                                        Register tmp1, Register tmp2, Register tmp3,
 8376                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8377                                        Register tmp4, Register tmp5,
 8378                                        Register n_tmp6) {
 8379   Label L_processPartitions;
 8380   Label L_processPartition;
 8381   Label L_exit;
 8382 
 8383   bind(L_processPartitions);
 8384   cmpl(in_out1, 3 * size);
 8385   jcc(Assembler::less, L_exit);
 8386     xorl(tmp1, tmp1);
 8387     xorl(tmp2, tmp2);
 8388     movq(tmp3, in_out2);
 8389     addq(tmp3, size);
 8390 
 8391     bind(L_processPartition);
 8392       crc32(in_out3, Address(in_out2, 0), 8);
 8393       crc32(tmp1, Address(in_out2, size), 8);
 8394       crc32(tmp2, Address(in_out2, size * 2), 8);
 8395       addq(in_out2, 8);
 8396       cmpq(in_out2, tmp3);
 8397       jcc(Assembler::less, L_processPartition);
 8398     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 8399             w_xtmp1, w_xtmp2, w_xtmp3,
 8400             tmp4, tmp5,
 8401             n_tmp6);
 8402     addq(in_out2, 2 * size);
 8403     subl(in_out1, 3 * size);
 8404     jmp(L_processPartitions);
 8405 
 8406   bind(L_exit);
 8407 }
 8408 
 8409 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 8410 // Input: A buffer I of L bytes.
 8411 // Output: the CRC32C value of the buffer.
 8412 // Notations:
 8413 // Write L = 24N + r, with N = floor (L/24).
 8414 // r = L mod 24 (0 <= r < 24).
 8415 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 8416 // N quadwords, and R consists of r bytes.
 8417 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 8418 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 8419 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 8420 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 8421 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 8422                                           Register tmp1, Register tmp2, Register tmp3,
 8423                                           Register tmp4, Register tmp5, Register tmp6,
 8424                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8425                                           bool is_pclmulqdq_supported) {
 8426   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 8427   Label L_wordByWord;
 8428   Label L_byteByByteProlog;
 8429   Label L_byteByByte;
 8430   Label L_exit;
 8431 
 8432   if (is_pclmulqdq_supported ) {
 8433     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 8434     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 8435 
 8436     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 8437     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 8438 
 8439     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 8440     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 8441     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 8442   } else {
 8443     const_or_pre_comp_const_index[0] = 1;
 8444     const_or_pre_comp_const_index[1] = 0;
 8445 
 8446     const_or_pre_comp_const_index[2] = 3;
 8447     const_or_pre_comp_const_index[3] = 2;
 8448 
 8449     const_or_pre_comp_const_index[4] = 5;
 8450     const_or_pre_comp_const_index[5] = 4;
 8451    }
 8452   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 8453                     in2, in1, in_out,
 8454                     tmp1, tmp2, tmp3,
 8455                     w_xtmp1, w_xtmp2, w_xtmp3,
 8456                     tmp4, tmp5,
 8457                     tmp6);
 8458   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 8459                     in2, in1, in_out,
 8460                     tmp1, tmp2, tmp3,
 8461                     w_xtmp1, w_xtmp2, w_xtmp3,
 8462                     tmp4, tmp5,
 8463                     tmp6);
 8464   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 8465                     in2, in1, in_out,
 8466                     tmp1, tmp2, tmp3,
 8467                     w_xtmp1, w_xtmp2, w_xtmp3,
 8468                     tmp4, tmp5,
 8469                     tmp6);
 8470   movl(tmp1, in2);
 8471   andl(tmp1, 0x00000007);
 8472   negl(tmp1);
 8473   addl(tmp1, in2);
 8474   addq(tmp1, in1);
 8475 
 8476   cmpq(in1, tmp1);
 8477   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 8478   align(16);
 8479   BIND(L_wordByWord);
 8480     crc32(in_out, Address(in1, 0), 8);
 8481     addq(in1, 8);
 8482     cmpq(in1, tmp1);
 8483     jcc(Assembler::less, L_wordByWord);
 8484 
 8485   BIND(L_byteByByteProlog);
 8486   andl(in2, 0x00000007);
 8487   movl(tmp2, 1);
 8488 
 8489   cmpl(tmp2, in2);
 8490   jccb(Assembler::greater, L_exit);
 8491   BIND(L_byteByByte);
 8492     crc32(in_out, Address(in1, 0), 1);
 8493     incq(in1);
 8494     incl(tmp2);
 8495     cmpl(tmp2, in2);
 8496     jcc(Assembler::lessEqual, L_byteByByte);
 8497 
 8498   BIND(L_exit);
 8499 }
 8500 #undef BIND
 8501 #undef BLOCK_COMMENT
 8502 
 8503 // Compress char[] array to byte[].
 8504 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 8505 // Return the array length if every element in array can be encoded,
 8506 // otherwise, the index of first non-latin1 (> 0xff) character.
 8507 //   @IntrinsicCandidate
 8508 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 8509 //     for (int i = 0; i < len; i++) {
 8510 //       char c = src[srcOff];
 8511 //       if (c > 0xff) {
 8512 //           return i;  // return index of non-latin1 char
 8513 //       }
 8514 //       dst[dstOff] = (byte)c;
 8515 //       srcOff++;
 8516 //       dstOff++;
 8517 //     }
 8518 //     return len;
 8519 //   }
 8520 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 8521   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 8522   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 8523   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 8524   Label copy_chars_loop, done, reset_sp, copy_tail;
 8525 
 8526   // rsi: src
 8527   // rdi: dst
 8528   // rdx: len
 8529   // rcx: tmp5
 8530   // rax: result
 8531 
 8532   // rsi holds start addr of source char[] to be compressed
 8533   // rdi holds start addr of destination byte[]
 8534   // rdx holds length
 8535 
 8536   assert(len != result, "");
 8537 
 8538   // save length for return
 8539   movl(result, len);
 8540 
 8541   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 8542     VM_Version::supports_avx512vlbw() &&
 8543     VM_Version::supports_bmi2()) {
 8544 
 8545     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 8546 
 8547     // alignment
 8548     Label post_alignment;
 8549 
 8550     // if length of the string is less than 32, handle it the old fashioned way
 8551     testl(len, -32);
 8552     jcc(Assembler::zero, below_threshold);
 8553 
 8554     // First check whether a character is compressible ( <= 0xFF).
 8555     // Create mask to test for Unicode chars inside zmm vector
 8556     movl(tmp5, 0x00FF);
 8557     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 8558 
 8559     testl(len, -64);
 8560     jccb(Assembler::zero, post_alignment);
 8561 
 8562     movl(tmp5, dst);
 8563     andl(tmp5, (32 - 1));
 8564     negl(tmp5);
 8565     andl(tmp5, (32 - 1));
 8566 
 8567     // bail out when there is nothing to be done
 8568     testl(tmp5, 0xFFFFFFFF);
 8569     jccb(Assembler::zero, post_alignment);
 8570 
 8571     // ~(~0 << len), where len is the # of remaining elements to process
 8572     movl(len, 0xFFFFFFFF);
 8573     shlxl(len, len, tmp5);
 8574     notl(len);
 8575     kmovdl(mask2, len);
 8576     movl(len, result);
 8577 
 8578     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8579     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8580     ktestd(mask1, mask2);
 8581     jcc(Assembler::carryClear, copy_tail);
 8582 
 8583     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8584 
 8585     addptr(src, tmp5);
 8586     addptr(src, tmp5);
 8587     addptr(dst, tmp5);
 8588     subl(len, tmp5);
 8589 
 8590     bind(post_alignment);
 8591     // end of alignment
 8592 
 8593     movl(tmp5, len);
 8594     andl(tmp5, (32 - 1));    // tail count (in chars)
 8595     andl(len, ~(32 - 1));    // vector count (in chars)
 8596     jccb(Assembler::zero, copy_loop_tail);
 8597 
 8598     lea(src, Address(src, len, Address::times_2));
 8599     lea(dst, Address(dst, len, Address::times_1));
 8600     negptr(len);
 8601 
 8602     bind(copy_32_loop);
 8603     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 8604     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 8605     kortestdl(mask1, mask1);
 8606     jccb(Assembler::carryClear, reset_for_copy_tail);
 8607 
 8608     // All elements in current processed chunk are valid candidates for
 8609     // compression. Write a truncated byte elements to the memory.
 8610     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 8611     addptr(len, 32);
 8612     jccb(Assembler::notZero, copy_32_loop);
 8613 
 8614     bind(copy_loop_tail);
 8615     // bail out when there is nothing to be done
 8616     testl(tmp5, 0xFFFFFFFF);
 8617     jcc(Assembler::zero, done);
 8618 
 8619     movl(len, tmp5);
 8620 
 8621     // ~(~0 << len), where len is the # of remaining elements to process
 8622     movl(tmp5, 0xFFFFFFFF);
 8623     shlxl(tmp5, tmp5, len);
 8624     notl(tmp5);
 8625 
 8626     kmovdl(mask2, tmp5);
 8627 
 8628     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8629     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8630     ktestd(mask1, mask2);
 8631     jcc(Assembler::carryClear, copy_tail);
 8632 
 8633     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8634     jmp(done);
 8635 
 8636     bind(reset_for_copy_tail);
 8637     lea(src, Address(src, tmp5, Address::times_2));
 8638     lea(dst, Address(dst, tmp5, Address::times_1));
 8639     subptr(len, tmp5);
 8640     jmp(copy_chars_loop);
 8641 
 8642     bind(below_threshold);
 8643   }
 8644 
 8645   if (UseSSE42Intrinsics) {
 8646     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 8647 
 8648     // vectored compression
 8649     testl(len, 0xfffffff8);
 8650     jcc(Assembler::zero, copy_tail);
 8651 
 8652     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 8653     movdl(tmp1Reg, tmp5);
 8654     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 8655 
 8656     andl(len, 0xfffffff0);
 8657     jccb(Assembler::zero, copy_16);
 8658 
 8659     // compress 16 chars per iter
 8660     pxor(tmp4Reg, tmp4Reg);
 8661 
 8662     lea(src, Address(src, len, Address::times_2));
 8663     lea(dst, Address(dst, len, Address::times_1));
 8664     negptr(len);
 8665 
 8666     bind(copy_32_loop);
 8667     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 8668     por(tmp4Reg, tmp2Reg);
 8669     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 8670     por(tmp4Reg, tmp3Reg);
 8671     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 8672     jccb(Assembler::notZero, reset_for_copy_tail);
 8673     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 8674     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 8675     addptr(len, 16);
 8676     jccb(Assembler::notZero, copy_32_loop);
 8677 
 8678     // compress next vector of 8 chars (if any)
 8679     bind(copy_16);
 8680     // len = 0
 8681     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 8682     jccb(Assembler::zero, copy_tail_sse);
 8683 
 8684     pxor(tmp3Reg, tmp3Reg);
 8685 
 8686     movdqu(tmp2Reg, Address(src, 0));
 8687     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 8688     jccb(Assembler::notZero, reset_for_copy_tail);
 8689     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 8690     movq(Address(dst, 0), tmp2Reg);
 8691     addptr(src, 16);
 8692     addptr(dst, 8);
 8693     jmpb(copy_tail_sse);
 8694 
 8695     bind(reset_for_copy_tail);
 8696     movl(tmp5, result);
 8697     andl(tmp5, 0x0000000f);
 8698     lea(src, Address(src, tmp5, Address::times_2));
 8699     lea(dst, Address(dst, tmp5, Address::times_1));
 8700     subptr(len, tmp5);
 8701     jmpb(copy_chars_loop);
 8702 
 8703     bind(copy_tail_sse);
 8704     movl(len, result);
 8705     andl(len, 0x00000007);    // tail count (in chars)
 8706   }
 8707   // compress 1 char per iter
 8708   bind(copy_tail);
 8709   testl(len, len);
 8710   jccb(Assembler::zero, done);
 8711   lea(src, Address(src, len, Address::times_2));
 8712   lea(dst, Address(dst, len, Address::times_1));
 8713   negptr(len);
 8714 
 8715   bind(copy_chars_loop);
 8716   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 8717   testl(tmp5, 0xff00);      // check if Unicode char
 8718   jccb(Assembler::notZero, reset_sp);
 8719   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 8720   increment(len);
 8721   jccb(Assembler::notZero, copy_chars_loop);
 8722 
 8723   // add len then return (len will be zero if compress succeeded, otherwise negative)
 8724   bind(reset_sp);
 8725   addl(result, len);
 8726 
 8727   bind(done);
 8728 }
 8729 
 8730 // Inflate byte[] array to char[].
 8731 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 8732 //   @IntrinsicCandidate
 8733 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 8734 //     for (int i = 0; i < len; i++) {
 8735 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 8736 //     }
 8737 //   }
 8738 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 8739   XMMRegister tmp1, Register tmp2, KRegister mask) {
 8740   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 8741   // rsi: src
 8742   // rdi: dst
 8743   // rdx: len
 8744   // rcx: tmp2
 8745 
 8746   // rsi holds start addr of source byte[] to be inflated
 8747   // rdi holds start addr of destination char[]
 8748   // rdx holds length
 8749   assert_different_registers(src, dst, len, tmp2);
 8750   movl(tmp2, len);
 8751   if ((UseAVX > 2) && // AVX512
 8752     VM_Version::supports_avx512vlbw() &&
 8753     VM_Version::supports_bmi2()) {
 8754 
 8755     Label copy_32_loop, copy_tail;
 8756     Register tmp3_aliased = len;
 8757 
 8758     // if length of the string is less than 16, handle it in an old fashioned way
 8759     testl(len, -16);
 8760     jcc(Assembler::zero, below_threshold);
 8761 
 8762     testl(len, -1 * AVX3Threshold);
 8763     jcc(Assembler::zero, avx3_threshold);
 8764 
 8765     // In order to use only one arithmetic operation for the main loop we use
 8766     // this pre-calculation
 8767     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 8768     andl(len, -32);     // vector count
 8769     jccb(Assembler::zero, copy_tail);
 8770 
 8771     lea(src, Address(src, len, Address::times_1));
 8772     lea(dst, Address(dst, len, Address::times_2));
 8773     negptr(len);
 8774 
 8775 
 8776     // inflate 32 chars per iter
 8777     bind(copy_32_loop);
 8778     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 8779     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 8780     addptr(len, 32);
 8781     jcc(Assembler::notZero, copy_32_loop);
 8782 
 8783     bind(copy_tail);
 8784     // bail out when there is nothing to be done
 8785     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 8786     jcc(Assembler::zero, done);
 8787 
 8788     // ~(~0 << length), where length is the # of remaining elements to process
 8789     movl(tmp3_aliased, -1);
 8790     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 8791     notl(tmp3_aliased);
 8792     kmovdl(mask, tmp3_aliased);
 8793     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 8794     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 8795 
 8796     jmp(done);
 8797     bind(avx3_threshold);
 8798   }
 8799   if (UseSSE42Intrinsics) {
 8800     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 8801 
 8802     if (UseAVX > 1) {
 8803       andl(tmp2, (16 - 1));
 8804       andl(len, -16);
 8805       jccb(Assembler::zero, copy_new_tail);
 8806     } else {
 8807       andl(tmp2, 0x00000007);   // tail count (in chars)
 8808       andl(len, 0xfffffff8);    // vector count (in chars)
 8809       jccb(Assembler::zero, copy_tail);
 8810     }
 8811 
 8812     // vectored inflation
 8813     lea(src, Address(src, len, Address::times_1));
 8814     lea(dst, Address(dst, len, Address::times_2));
 8815     negptr(len);
 8816 
 8817     if (UseAVX > 1) {
 8818       bind(copy_16_loop);
 8819       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 8820       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 8821       addptr(len, 16);
 8822       jcc(Assembler::notZero, copy_16_loop);
 8823 
 8824       bind(below_threshold);
 8825       bind(copy_new_tail);
 8826       movl(len, tmp2);
 8827       andl(tmp2, 0x00000007);
 8828       andl(len, 0xFFFFFFF8);
 8829       jccb(Assembler::zero, copy_tail);
 8830 
 8831       pmovzxbw(tmp1, Address(src, 0));
 8832       movdqu(Address(dst, 0), tmp1);
 8833       addptr(src, 8);
 8834       addptr(dst, 2 * 8);
 8835 
 8836       jmp(copy_tail, true);
 8837     }
 8838 
 8839     // inflate 8 chars per iter
 8840     bind(copy_8_loop);
 8841     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 8842     movdqu(Address(dst, len, Address::times_2), tmp1);
 8843     addptr(len, 8);
 8844     jcc(Assembler::notZero, copy_8_loop);
 8845 
 8846     bind(copy_tail);
 8847     movl(len, tmp2);
 8848 
 8849     cmpl(len, 4);
 8850     jccb(Assembler::less, copy_bytes);
 8851 
 8852     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 8853     pmovzxbw(tmp1, tmp1);
 8854     movq(Address(dst, 0), tmp1);
 8855     subptr(len, 4);
 8856     addptr(src, 4);
 8857     addptr(dst, 8);
 8858 
 8859     bind(copy_bytes);
 8860   } else {
 8861     bind(below_threshold);
 8862   }
 8863 
 8864   testl(len, len);
 8865   jccb(Assembler::zero, done);
 8866   lea(src, Address(src, len, Address::times_1));
 8867   lea(dst, Address(dst, len, Address::times_2));
 8868   negptr(len);
 8869 
 8870   // inflate 1 char per iter
 8871   bind(copy_chars_loop);
 8872   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 8873   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 8874   increment(len);
 8875   jcc(Assembler::notZero, copy_chars_loop);
 8876 
 8877   bind(done);
 8878 }
 8879 
 8880 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 8881   switch(type) {
 8882     case T_BYTE:
 8883     case T_BOOLEAN:
 8884       evmovdqub(dst, kmask, src, merge, vector_len);
 8885       break;
 8886     case T_CHAR:
 8887     case T_SHORT:
 8888       evmovdquw(dst, kmask, src, merge, vector_len);
 8889       break;
 8890     case T_INT:
 8891     case T_FLOAT:
 8892       evmovdqul(dst, kmask, src, merge, vector_len);
 8893       break;
 8894     case T_LONG:
 8895     case T_DOUBLE:
 8896       evmovdquq(dst, kmask, src, merge, vector_len);
 8897       break;
 8898     default:
 8899       fatal("Unexpected type argument %s", type2name(type));
 8900       break;
 8901   }
 8902 }
 8903 
 8904 
 8905 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 8906   switch(type) {
 8907     case T_BYTE:
 8908     case T_BOOLEAN:
 8909       evmovdqub(dst, kmask, src, merge, vector_len);
 8910       break;
 8911     case T_CHAR:
 8912     case T_SHORT:
 8913       evmovdquw(dst, kmask, src, merge, vector_len);
 8914       break;
 8915     case T_INT:
 8916     case T_FLOAT:
 8917       evmovdqul(dst, kmask, src, merge, vector_len);
 8918       break;
 8919     case T_LONG:
 8920     case T_DOUBLE:
 8921       evmovdquq(dst, kmask, src, merge, vector_len);
 8922       break;
 8923     default:
 8924       fatal("Unexpected type argument %s", type2name(type));
 8925       break;
 8926   }
 8927 }
 8928 
 8929 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 8930   switch(type) {
 8931     case T_BYTE:
 8932     case T_BOOLEAN:
 8933       evmovdqub(dst, kmask, src, merge, vector_len);
 8934       break;
 8935     case T_CHAR:
 8936     case T_SHORT:
 8937       evmovdquw(dst, kmask, src, merge, vector_len);
 8938       break;
 8939     case T_INT:
 8940     case T_FLOAT:
 8941       evmovdqul(dst, kmask, src, merge, vector_len);
 8942       break;
 8943     case T_LONG:
 8944     case T_DOUBLE:
 8945       evmovdquq(dst, kmask, src, merge, vector_len);
 8946       break;
 8947     default:
 8948       fatal("Unexpected type argument %s", type2name(type));
 8949       break;
 8950   }
 8951 }
 8952 
 8953 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 8954   switch(masklen) {
 8955     case 2:
 8956        knotbl(dst, src);
 8957        movl(rtmp, 3);
 8958        kmovbl(ktmp, rtmp);
 8959        kandbl(dst, ktmp, dst);
 8960        break;
 8961     case 4:
 8962        knotbl(dst, src);
 8963        movl(rtmp, 15);
 8964        kmovbl(ktmp, rtmp);
 8965        kandbl(dst, ktmp, dst);
 8966        break;
 8967     case 8:
 8968        knotbl(dst, src);
 8969        break;
 8970     case 16:
 8971        knotwl(dst, src);
 8972        break;
 8973     case 32:
 8974        knotdl(dst, src);
 8975        break;
 8976     case 64:
 8977        knotql(dst, src);
 8978        break;
 8979     default:
 8980       fatal("Unexpected vector length %d", masklen);
 8981       break;
 8982   }
 8983 }
 8984 
 8985 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8986   switch(type) {
 8987     case T_BOOLEAN:
 8988     case T_BYTE:
 8989        kandbl(dst, src1, src2);
 8990        break;
 8991     case T_CHAR:
 8992     case T_SHORT:
 8993        kandwl(dst, src1, src2);
 8994        break;
 8995     case T_INT:
 8996     case T_FLOAT:
 8997        kanddl(dst, src1, src2);
 8998        break;
 8999     case T_LONG:
 9000     case T_DOUBLE:
 9001        kandql(dst, src1, src2);
 9002        break;
 9003     default:
 9004       fatal("Unexpected type argument %s", type2name(type));
 9005       break;
 9006   }
 9007 }
 9008 
 9009 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9010   switch(type) {
 9011     case T_BOOLEAN:
 9012     case T_BYTE:
 9013        korbl(dst, src1, src2);
 9014        break;
 9015     case T_CHAR:
 9016     case T_SHORT:
 9017        korwl(dst, src1, src2);
 9018        break;
 9019     case T_INT:
 9020     case T_FLOAT:
 9021        kordl(dst, src1, src2);
 9022        break;
 9023     case T_LONG:
 9024     case T_DOUBLE:
 9025        korql(dst, src1, src2);
 9026        break;
 9027     default:
 9028       fatal("Unexpected type argument %s", type2name(type));
 9029       break;
 9030   }
 9031 }
 9032 
 9033 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9034   switch(type) {
 9035     case T_BOOLEAN:
 9036     case T_BYTE:
 9037        kxorbl(dst, src1, src2);
 9038        break;
 9039     case T_CHAR:
 9040     case T_SHORT:
 9041        kxorwl(dst, src1, src2);
 9042        break;
 9043     case T_INT:
 9044     case T_FLOAT:
 9045        kxordl(dst, src1, src2);
 9046        break;
 9047     case T_LONG:
 9048     case T_DOUBLE:
 9049        kxorql(dst, src1, src2);
 9050        break;
 9051     default:
 9052       fatal("Unexpected type argument %s", type2name(type));
 9053       break;
 9054   }
 9055 }
 9056 
 9057 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9058   switch(type) {
 9059     case T_BOOLEAN:
 9060     case T_BYTE:
 9061       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9062     case T_CHAR:
 9063     case T_SHORT:
 9064       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9065     case T_INT:
 9066     case T_FLOAT:
 9067       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9068     case T_LONG:
 9069     case T_DOUBLE:
 9070       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9071     default:
 9072       fatal("Unexpected type argument %s", type2name(type)); break;
 9073   }
 9074 }
 9075 
 9076 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9077   switch(type) {
 9078     case T_BOOLEAN:
 9079     case T_BYTE:
 9080       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9081     case T_CHAR:
 9082     case T_SHORT:
 9083       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9084     case T_INT:
 9085     case T_FLOAT:
 9086       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9087     case T_LONG:
 9088     case T_DOUBLE:
 9089       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9090     default:
 9091       fatal("Unexpected type argument %s", type2name(type)); break;
 9092   }
 9093 }
 9094 
 9095 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9096   switch(type) {
 9097     case T_BYTE:
 9098       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9099     case T_SHORT:
 9100       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9101     case T_INT:
 9102       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9103     case T_LONG:
 9104       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9105     default:
 9106       fatal("Unexpected type argument %s", type2name(type)); break;
 9107   }
 9108 }
 9109 
 9110 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9111   switch(type) {
 9112     case T_BYTE:
 9113       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9114     case T_SHORT:
 9115       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9116     case T_INT:
 9117       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9118     case T_LONG:
 9119       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9120     default:
 9121       fatal("Unexpected type argument %s", type2name(type)); break;
 9122   }
 9123 }
 9124 
 9125 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9126   switch(type) {
 9127     case T_BYTE:
 9128       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9129     case T_SHORT:
 9130       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9131     case T_INT:
 9132       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9133     case T_LONG:
 9134       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9135     default:
 9136       fatal("Unexpected type argument %s", type2name(type)); break;
 9137   }
 9138 }
 9139 
 9140 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9141   switch(type) {
 9142     case T_BYTE:
 9143       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9144     case T_SHORT:
 9145       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9146     case T_INT:
 9147       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9148     case T_LONG:
 9149       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9150     default:
 9151       fatal("Unexpected type argument %s", type2name(type)); break;
 9152   }
 9153 }
 9154 
 9155 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9156   switch(type) {
 9157     case T_BYTE:
 9158       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9159     case T_SHORT:
 9160       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9161     case T_INT:
 9162       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9163     case T_LONG:
 9164       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9165     case T_FLOAT:
 9166       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9167     case T_DOUBLE:
 9168       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9169     default:
 9170       fatal("Unexpected type argument %s", type2name(type)); break;
 9171   }
 9172 }
 9173 
 9174 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9175   switch(type) {
 9176     case T_BYTE:
 9177       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9178     case T_SHORT:
 9179       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9180     case T_INT:
 9181       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9182     case T_LONG:
 9183       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9184     case T_FLOAT:
 9185       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9186     case T_DOUBLE:
 9187       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9188     default:
 9189       fatal("Unexpected type argument %s", type2name(type)); break;
 9190   }
 9191 }
 9192 
 9193 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9194   switch(type) {
 9195     case T_BYTE:
 9196       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9197     case T_SHORT:
 9198       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9199     case T_INT:
 9200       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9201     case T_LONG:
 9202       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9203     case T_FLOAT:
 9204       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9205     case T_DOUBLE:
 9206       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9207     default:
 9208       fatal("Unexpected type argument %s", type2name(type)); break;
 9209   }
 9210 }
 9211 
 9212 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9213   switch(type) {
 9214     case T_BYTE:
 9215       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9216     case T_SHORT:
 9217       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9218     case T_INT:
 9219       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9220     case T_LONG:
 9221       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9222     case T_FLOAT:
 9223       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9224     case T_DOUBLE:
 9225       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9226     default:
 9227       fatal("Unexpected type argument %s", type2name(type)); break;
 9228   }
 9229 }
 9230 
 9231 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9232   switch(type) {
 9233     case T_INT:
 9234       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9235     case T_LONG:
 9236       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9237     default:
 9238       fatal("Unexpected type argument %s", type2name(type)); break;
 9239   }
 9240 }
 9241 
 9242 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9243   switch(type) {
 9244     case T_INT:
 9245       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9246     case T_LONG:
 9247       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9248     default:
 9249       fatal("Unexpected type argument %s", type2name(type)); break;
 9250   }
 9251 }
 9252 
 9253 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9254   switch(type) {
 9255     case T_INT:
 9256       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9257     case T_LONG:
 9258       evporq(dst, mask, nds, src, merge, vector_len); break;
 9259     default:
 9260       fatal("Unexpected type argument %s", type2name(type)); break;
 9261   }
 9262 }
 9263 
 9264 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9265   switch(type) {
 9266     case T_INT:
 9267       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9268     case T_LONG:
 9269       evporq(dst, mask, nds, src, merge, vector_len); break;
 9270     default:
 9271       fatal("Unexpected type argument %s", type2name(type)); break;
 9272   }
 9273 }
 9274 
 9275 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9276   switch(type) {
 9277     case T_INT:
 9278       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9279     case T_LONG:
 9280       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9281     default:
 9282       fatal("Unexpected type argument %s", type2name(type)); break;
 9283   }
 9284 }
 9285 
 9286 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9287   switch(type) {
 9288     case T_INT:
 9289       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9290     case T_LONG:
 9291       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9292     default:
 9293       fatal("Unexpected type argument %s", type2name(type)); break;
 9294   }
 9295 }
 9296 
 9297 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9298   switch(masklen) {
 9299     case 8:
 9300        kortestbl(src1, src2);
 9301        break;
 9302     case 16:
 9303        kortestwl(src1, src2);
 9304        break;
 9305     case 32:
 9306        kortestdl(src1, src2);
 9307        break;
 9308     case 64:
 9309        kortestql(src1, src2);
 9310        break;
 9311     default:
 9312       fatal("Unexpected mask length %d", masklen);
 9313       break;
 9314   }
 9315 }
 9316 
 9317 
 9318 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
 9319   switch(masklen)  {
 9320     case 8:
 9321        ktestbl(src1, src2);
 9322        break;
 9323     case 16:
 9324        ktestwl(src1, src2);
 9325        break;
 9326     case 32:
 9327        ktestdl(src1, src2);
 9328        break;
 9329     case 64:
 9330        ktestql(src1, src2);
 9331        break;
 9332     default:
 9333       fatal("Unexpected mask length %d", masklen);
 9334       break;
 9335   }
 9336 }
 9337 
 9338 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9339   switch(type) {
 9340     case T_INT:
 9341       evprold(dst, mask, src, shift, merge, vlen_enc); break;
 9342     case T_LONG:
 9343       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
 9344     default:
 9345       fatal("Unexpected type argument %s", type2name(type)); break;
 9346       break;
 9347   }
 9348 }
 9349 
 9350 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9351   switch(type) {
 9352     case T_INT:
 9353       evprord(dst, mask, src, shift, merge, vlen_enc); break;
 9354     case T_LONG:
 9355       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
 9356     default:
 9357       fatal("Unexpected type argument %s", type2name(type)); break;
 9358   }
 9359 }
 9360 
 9361 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9362   switch(type) {
 9363     case T_INT:
 9364       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9365     case T_LONG:
 9366       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9367     default:
 9368       fatal("Unexpected type argument %s", type2name(type)); break;
 9369   }
 9370 }
 9371 
 9372 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9373   switch(type) {
 9374     case T_INT:
 9375       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9376     case T_LONG:
 9377       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9378     default:
 9379       fatal("Unexpected type argument %s", type2name(type)); break;
 9380   }
 9381 }
 9382 
 9383 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9384   assert(rscratch != noreg || always_reachable(src), "missing");
 9385 
 9386   if (reachable(src)) {
 9387     evpandq(dst, nds, as_Address(src), vector_len);
 9388   } else {
 9389     lea(rscratch, src);
 9390     evpandq(dst, nds, Address(rscratch, 0), vector_len);
 9391   }
 9392 }
 9393 
 9394 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 9395   assert(rscratch != noreg || always_reachable(src), "missing");
 9396 
 9397   if (reachable(src)) {
 9398     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
 9399   } else {
 9400     lea(rscratch, src);
 9401     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 9402   }
 9403 }
 9404 
 9405 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9406   assert(rscratch != noreg || always_reachable(src), "missing");
 9407 
 9408   if (reachable(src)) {
 9409     evporq(dst, nds, as_Address(src), vector_len);
 9410   } else {
 9411     lea(rscratch, src);
 9412     evporq(dst, nds, Address(rscratch, 0), vector_len);
 9413   }
 9414 }
 9415 
 9416 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9417   assert(rscratch != noreg || always_reachable(src), "missing");
 9418 
 9419   if (reachable(src)) {
 9420     vpshufb(dst, nds, as_Address(src), vector_len);
 9421   } else {
 9422     lea(rscratch, src);
 9423     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
 9424   }
 9425 }
 9426 
 9427 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9428   assert(rscratch != noreg || always_reachable(src), "missing");
 9429 
 9430   if (reachable(src)) {
 9431     Assembler::vpor(dst, nds, as_Address(src), vector_len);
 9432   } else {
 9433     lea(rscratch, src);
 9434     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
 9435   }
 9436 }
 9437 
 9438 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
 9439   assert(rscratch != noreg || always_reachable(src3), "missing");
 9440 
 9441   if (reachable(src3)) {
 9442     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
 9443   } else {
 9444     lea(rscratch, src3);
 9445     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
 9446   }
 9447 }
 9448 
 9449 #if COMPILER2_OR_JVMCI
 9450 
 9451 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
 9452                                  Register length, Register temp, int vec_enc) {
 9453   // Computing mask for predicated vector store.
 9454   movptr(temp, -1);
 9455   bzhiq(temp, temp, length);
 9456   kmov(mask, temp);
 9457   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
 9458 }
 9459 
 9460 // Set memory operation for length "less than" 64 bytes.
 9461 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
 9462                                        XMMRegister xmm, KRegister mask, Register length,
 9463                                        Register temp, bool use64byteVector) {
 9464   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9465   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9466   if (!use64byteVector) {
 9467     fill32(dst, disp, xmm);
 9468     subptr(length, 32 >> shift);
 9469     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
 9470   } else {
 9471     assert(MaxVectorSize == 64, "vector length != 64");
 9472     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
 9473   }
 9474 }
 9475 
 9476 
 9477 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
 9478                                        XMMRegister xmm, KRegister mask, Register length,
 9479                                        Register temp) {
 9480   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9481   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9482   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
 9483 }
 9484 
 9485 
 9486 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
 9487   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9488   vmovdqu(dst, xmm);
 9489 }
 9490 
 9491 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
 9492   fill32(Address(dst, disp), xmm);
 9493 }
 9494 
 9495 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
 9496   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9497   if (!use64byteVector) {
 9498     fill32(dst, xmm);
 9499     fill32(dst.plus_disp(32), xmm);
 9500   } else {
 9501     evmovdquq(dst, xmm, Assembler::AVX_512bit);
 9502   }
 9503 }
 9504 
 9505 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
 9506   fill64(Address(dst, disp), xmm, use64byteVector);
 9507 }
 9508 
 9509 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
 9510                                         Register count, Register rtmp, XMMRegister xtmp) {
 9511   Label L_exit;
 9512   Label L_fill_start;
 9513   Label L_fill_64_bytes;
 9514   Label L_fill_96_bytes;
 9515   Label L_fill_128_bytes;
 9516   Label L_fill_128_bytes_loop;
 9517   Label L_fill_128_loop_header;
 9518   Label L_fill_128_bytes_loop_header;
 9519   Label L_fill_128_bytes_loop_pre_header;
 9520   Label L_fill_zmm_sequence;
 9521 
 9522   int shift = -1;
 9523   switch(type) {
 9524     case T_BYTE:  shift = 0;
 9525       break;
 9526     case T_SHORT: shift = 1;
 9527       break;
 9528     case T_INT:   shift = 2;
 9529       break;
 9530     /* Uncomment when LONG fill stubs are supported.
 9531     case T_LONG:  shift = 3;
 9532       break;
 9533     */
 9534     default:
 9535       fatal("Unhandled type: %s\n", type2name(type));
 9536   }
 9537 
 9538   if ((CopyAVX3Threshold != 0)  || (MaxVectorSize == 32)) {
 9539 
 9540     if (MaxVectorSize == 64) {
 9541       cmpq(count, CopyAVX3Threshold >> shift);
 9542       jcc(Assembler::greater, L_fill_zmm_sequence);
 9543     }
 9544 
 9545     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
 9546 
 9547     bind(L_fill_start);
 9548 
 9549     cmpq(count, 32 >> shift);
 9550     jccb(Assembler::greater, L_fill_64_bytes);
 9551     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9552     jmp(L_exit);
 9553 
 9554     bind(L_fill_64_bytes);
 9555     cmpq(count, 64 >> shift);
 9556     jccb(Assembler::greater, L_fill_96_bytes);
 9557     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9558     jmp(L_exit);
 9559 
 9560     bind(L_fill_96_bytes);
 9561     cmpq(count, 96 >> shift);
 9562     jccb(Assembler::greater, L_fill_128_bytes);
 9563     fill64(to, 0, xtmp);
 9564     subq(count, 64 >> shift);
 9565     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
 9566     jmp(L_exit);
 9567 
 9568     bind(L_fill_128_bytes);
 9569     cmpq(count, 128 >> shift);
 9570     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
 9571     fill64(to, 0, xtmp);
 9572     fill32(to, 64, xtmp);
 9573     subq(count, 96 >> shift);
 9574     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
 9575     jmp(L_exit);
 9576 
 9577     bind(L_fill_128_bytes_loop_pre_header);
 9578     {
 9579       mov(rtmp, to);
 9580       andq(rtmp, 31);
 9581       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
 9582       negq(rtmp);
 9583       addq(rtmp, 32);
 9584       mov64(r8, -1L);
 9585       bzhiq(r8, r8, rtmp);
 9586       kmovql(k2, r8);
 9587       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
 9588       addq(to, rtmp);
 9589       shrq(rtmp, shift);
 9590       subq(count, rtmp);
 9591     }
 9592 
 9593     cmpq(count, 128 >> shift);
 9594     jcc(Assembler::less, L_fill_start);
 9595 
 9596     bind(L_fill_128_bytes_loop_header);
 9597     subq(count, 128 >> shift);
 9598 
 9599     align32();
 9600     bind(L_fill_128_bytes_loop);
 9601       fill64(to, 0, xtmp);
 9602       fill64(to, 64, xtmp);
 9603       addq(to, 128);
 9604       subq(count, 128 >> shift);
 9605       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
 9606 
 9607     addq(count, 128 >> shift);
 9608     jcc(Assembler::zero, L_exit);
 9609     jmp(L_fill_start);
 9610   }
 9611 
 9612   if (MaxVectorSize == 64) {
 9613     // Sequence using 64 byte ZMM register.
 9614     Label L_fill_128_bytes_zmm;
 9615     Label L_fill_192_bytes_zmm;
 9616     Label L_fill_192_bytes_loop_zmm;
 9617     Label L_fill_192_bytes_loop_header_zmm;
 9618     Label L_fill_192_bytes_loop_pre_header_zmm;
 9619     Label L_fill_start_zmm_sequence;
 9620 
 9621     bind(L_fill_zmm_sequence);
 9622     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
 9623 
 9624     bind(L_fill_start_zmm_sequence);
 9625     cmpq(count, 64 >> shift);
 9626     jccb(Assembler::greater, L_fill_128_bytes_zmm);
 9627     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
 9628     jmp(L_exit);
 9629 
 9630     bind(L_fill_128_bytes_zmm);
 9631     cmpq(count, 128 >> shift);
 9632     jccb(Assembler::greater, L_fill_192_bytes_zmm);
 9633     fill64(to, 0, xtmp, true);
 9634     subq(count, 64 >> shift);
 9635     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
 9636     jmp(L_exit);
 9637 
 9638     bind(L_fill_192_bytes_zmm);
 9639     cmpq(count, 192 >> shift);
 9640     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
 9641     fill64(to, 0, xtmp, true);
 9642     fill64(to, 64, xtmp, true);
 9643     subq(count, 128 >> shift);
 9644     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
 9645     jmp(L_exit);
 9646 
 9647     bind(L_fill_192_bytes_loop_pre_header_zmm);
 9648     {
 9649       movq(rtmp, to);
 9650       andq(rtmp, 63);
 9651       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
 9652       negq(rtmp);
 9653       addq(rtmp, 64);
 9654       mov64(r8, -1L);
 9655       bzhiq(r8, r8, rtmp);
 9656       kmovql(k2, r8);
 9657       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
 9658       addq(to, rtmp);
 9659       shrq(rtmp, shift);
 9660       subq(count, rtmp);
 9661     }
 9662 
 9663     cmpq(count, 192 >> shift);
 9664     jcc(Assembler::less, L_fill_start_zmm_sequence);
 9665 
 9666     bind(L_fill_192_bytes_loop_header_zmm);
 9667     subq(count, 192 >> shift);
 9668 
 9669     align32();
 9670     bind(L_fill_192_bytes_loop_zmm);
 9671       fill64(to, 0, xtmp, true);
 9672       fill64(to, 64, xtmp, true);
 9673       fill64(to, 128, xtmp, true);
 9674       addq(to, 192);
 9675       subq(count, 192 >> shift);
 9676       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
 9677 
 9678     addq(count, 192 >> shift);
 9679     jcc(Assembler::zero, L_exit);
 9680     jmp(L_fill_start_zmm_sequence);
 9681   }
 9682   bind(L_exit);
 9683 }
 9684 #endif //COMPILER2_OR_JVMCI
 9685 
 9686 
 9687 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
 9688   Label done;
 9689   cvttss2sil(dst, src);
 9690   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9691   cmpl(dst, 0x80000000); // float_sign_flip
 9692   jccb(Assembler::notEqual, done);
 9693   subptr(rsp, 8);
 9694   movflt(Address(rsp, 0), src);
 9695   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
 9696   pop(dst);
 9697   bind(done);
 9698 }
 9699 
 9700 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
 9701   Label done;
 9702   cvttsd2sil(dst, src);
 9703   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9704   cmpl(dst, 0x80000000); // float_sign_flip
 9705   jccb(Assembler::notEqual, done);
 9706   subptr(rsp, 8);
 9707   movdbl(Address(rsp, 0), src);
 9708   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
 9709   pop(dst);
 9710   bind(done);
 9711 }
 9712 
 9713 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
 9714   Label done;
 9715   cvttss2siq(dst, src);
 9716   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9717   jccb(Assembler::notEqual, done);
 9718   subptr(rsp, 8);
 9719   movflt(Address(rsp, 0), src);
 9720   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
 9721   pop(dst);
 9722   bind(done);
 9723 }
 9724 
 9725 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9726   // Following code is line by line assembly translation rounding algorithm.
 9727   // Please refer to java.lang.Math.round(float) algorithm for details.
 9728   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
 9729   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
 9730   const int32_t FloatConsts_EXP_BIAS = 127;
 9731   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
 9732   const int32_t MINUS_32 = 0xFFFFFFE0;
 9733   Label L_special_case, L_block1, L_exit;
 9734   movl(rtmp, FloatConsts_EXP_BIT_MASK);
 9735   movdl(dst, src);
 9736   andl(dst, rtmp);
 9737   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
 9738   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
 9739   subl(rtmp, dst);
 9740   movl(rcx, rtmp);
 9741   movl(dst, MINUS_32);
 9742   testl(rtmp, dst);
 9743   jccb(Assembler::notEqual, L_special_case);
 9744   movdl(dst, src);
 9745   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
 9746   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
 9747   movdl(rtmp, src);
 9748   testl(rtmp, rtmp);
 9749   jccb(Assembler::greaterEqual, L_block1);
 9750   negl(dst);
 9751   bind(L_block1);
 9752   sarl(dst);
 9753   addl(dst, 0x1);
 9754   sarl(dst, 0x1);
 9755   jmp(L_exit);
 9756   bind(L_special_case);
 9757   convert_f2i(dst, src);
 9758   bind(L_exit);
 9759 }
 9760 
 9761 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9762   // Following code is line by line assembly translation rounding algorithm.
 9763   // Please refer to java.lang.Math.round(double) algorithm for details.
 9764   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
 9765   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
 9766   const int64_t DoubleConsts_EXP_BIAS = 1023;
 9767   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
 9768   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
 9769   Label L_special_case, L_block1, L_exit;
 9770   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
 9771   movq(dst, src);
 9772   andq(dst, rtmp);
 9773   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
 9774   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
 9775   subq(rtmp, dst);
 9776   movq(rcx, rtmp);
 9777   mov64(dst, MINUS_64);
 9778   testq(rtmp, dst);
 9779   jccb(Assembler::notEqual, L_special_case);
 9780   movq(dst, src);
 9781   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
 9782   andq(dst, rtmp);
 9783   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
 9784   orq(dst, rtmp);
 9785   movq(rtmp, src);
 9786   testq(rtmp, rtmp);
 9787   jccb(Assembler::greaterEqual, L_block1);
 9788   negq(dst);
 9789   bind(L_block1);
 9790   sarq(dst);
 9791   addq(dst, 0x1);
 9792   sarq(dst, 0x1);
 9793   jmp(L_exit);
 9794   bind(L_special_case);
 9795   convert_d2l(dst, src);
 9796   bind(L_exit);
 9797 }
 9798 
 9799 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
 9800   Label done;
 9801   cvttsd2siq(dst, src);
 9802   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9803   jccb(Assembler::notEqual, done);
 9804   subptr(rsp, 8);
 9805   movdbl(Address(rsp, 0), src);
 9806   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
 9807   pop(dst);
 9808   bind(done);
 9809 }
 9810 
 9811 void MacroAssembler::cache_wb(Address line)
 9812 {
 9813   // 64 bit cpus always support clflush
 9814   assert(VM_Version::supports_clflush(), "clflush should be available");
 9815   bool optimized = VM_Version::supports_clflushopt();
 9816   bool no_evict = VM_Version::supports_clwb();
 9817 
 9818   // prefer clwb (writeback without evict) otherwise
 9819   // prefer clflushopt (potentially parallel writeback with evict)
 9820   // otherwise fallback on clflush (serial writeback with evict)
 9821 
 9822   if (optimized) {
 9823     if (no_evict) {
 9824       clwb(line);
 9825     } else {
 9826       clflushopt(line);
 9827     }
 9828   } else {
 9829     // no need for fence when using CLFLUSH
 9830     clflush(line);
 9831   }
 9832 }
 9833 
 9834 void MacroAssembler::cache_wbsync(bool is_pre)
 9835 {
 9836   assert(VM_Version::supports_clflush(), "clflush should be available");
 9837   bool optimized = VM_Version::supports_clflushopt();
 9838   bool no_evict = VM_Version::supports_clwb();
 9839 
 9840   // pick the correct implementation
 9841 
 9842   if (!is_pre && (optimized || no_evict)) {
 9843     // need an sfence for post flush when using clflushopt or clwb
 9844     // otherwise no no need for any synchroniaztion
 9845 
 9846     sfence();
 9847   }
 9848 }
 9849 
 9850 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
 9851   switch (cond) {
 9852     // Note some conditions are synonyms for others
 9853     case Assembler::zero:         return Assembler::notZero;
 9854     case Assembler::notZero:      return Assembler::zero;
 9855     case Assembler::less:         return Assembler::greaterEqual;
 9856     case Assembler::lessEqual:    return Assembler::greater;
 9857     case Assembler::greater:      return Assembler::lessEqual;
 9858     case Assembler::greaterEqual: return Assembler::less;
 9859     case Assembler::below:        return Assembler::aboveEqual;
 9860     case Assembler::belowEqual:   return Assembler::above;
 9861     case Assembler::above:        return Assembler::belowEqual;
 9862     case Assembler::aboveEqual:   return Assembler::below;
 9863     case Assembler::overflow:     return Assembler::noOverflow;
 9864     case Assembler::noOverflow:   return Assembler::overflow;
 9865     case Assembler::negative:     return Assembler::positive;
 9866     case Assembler::positive:     return Assembler::negative;
 9867     case Assembler::parity:       return Assembler::noParity;
 9868     case Assembler::noParity:     return Assembler::parity;
 9869   }
 9870   ShouldNotReachHere(); return Assembler::overflow;
 9871 }
 9872 
 9873 // This is simply a call to Thread::current()
 9874 void MacroAssembler::get_thread_slow(Register thread) {
 9875   if (thread != rax) {
 9876     push(rax);
 9877   }
 9878   push(rdi);
 9879   push(rsi);
 9880   push(rdx);
 9881   push(rcx);
 9882   push(r8);
 9883   push(r9);
 9884   push(r10);
 9885   push(r11);
 9886 
 9887   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
 9888 
 9889   pop(r11);
 9890   pop(r10);
 9891   pop(r9);
 9892   pop(r8);
 9893   pop(rcx);
 9894   pop(rdx);
 9895   pop(rsi);
 9896   pop(rdi);
 9897   if (thread != rax) {
 9898     mov(thread, rax);
 9899     pop(rax);
 9900   }
 9901 }
 9902 
 9903 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
 9904   Label L_stack_ok;
 9905   if (bias == 0) {
 9906     testptr(sp, 2 * wordSize - 1);
 9907   } else {
 9908     // lea(tmp, Address(rsp, bias);
 9909     mov(tmp, sp);
 9910     addptr(tmp, bias);
 9911     testptr(tmp, 2 * wordSize - 1);
 9912   }
 9913   jcc(Assembler::equal, L_stack_ok);
 9914   block_comment(msg);
 9915   stop(msg);
 9916   bind(L_stack_ok);
 9917 }
 9918 
 9919 // Implements fast-locking.
 9920 //
 9921 // obj: the object to be locked
 9922 // reg_rax: rax
 9923 // thread: the thread which attempts to lock obj
 9924 // tmp: a temporary register
 9925 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
 9926   Register thread = r15_thread;
 9927 
 9928   assert(reg_rax == rax, "");
 9929   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
 9930 
 9931   Label push;
 9932   const Register top = tmp;
 9933 
 9934   // Preload the markWord. It is important that this is the first
 9935   // instruction emitted as it is part of C1's null check semantics.
 9936   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9937 
 9938   if (UseObjectMonitorTable) {
 9939     // Clear cache in case fast locking succeeds or we need to take the slow-path.
 9940     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
 9941   }
 9942 
 9943   if (DiagnoseSyncOnValueBasedClasses != 0) {
 9944     load_klass(tmp, obj, rscratch1);
 9945     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
 9946     jcc(Assembler::notZero, slow);
 9947   }
 9948 
 9949   // Load top.
 9950   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9951 
 9952   // Check if the lock-stack is full.
 9953   cmpl(top, LockStack::end_offset());
 9954   jcc(Assembler::greaterEqual, slow);
 9955 
 9956   // Check for recursion.
 9957   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9958   jcc(Assembler::equal, push);
 9959 
 9960   // Check header for monitor (0b10).
 9961   testptr(reg_rax, markWord::monitor_value);
 9962   jcc(Assembler::notZero, slow);
 9963 
 9964   // Try to lock. Transition lock bits 0b01 => 0b00
 9965   movptr(tmp, reg_rax);
 9966   andptr(tmp, ~(int32_t)markWord::unlocked_value);
 9967   orptr(reg_rax, markWord::unlocked_value);



 9968   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9969   jcc(Assembler::notEqual, slow);
 9970 
 9971   // Restore top, CAS clobbers register.
 9972   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9973 
 9974   bind(push);
 9975   // After successful lock, push object on lock-stack.
 9976   movptr(Address(thread, top), obj);
 9977   incrementl(top, oopSize);
 9978   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
 9979 }
 9980 
 9981 // Implements fast-unlocking.
 9982 //
 9983 // obj: the object to be unlocked
 9984 // reg_rax: rax
 9985 // thread: the thread
 9986 // tmp: a temporary register
 9987 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
 9988   Register thread = r15_thread;
 9989 
 9990   assert(reg_rax == rax, "");
 9991   assert_different_registers(obj, reg_rax, thread, tmp);
 9992 
 9993   Label unlocked, push_and_slow;
 9994   const Register top = tmp;
 9995 
 9996   // Check if obj is top of lock-stack.
 9997   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9998   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9999   jcc(Assembler::notEqual, slow);
10000 
10001   // Pop lock-stack.
10002   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
10003   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10004 
10005   // Check if recursive.
10006   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
10007   jcc(Assembler::equal, unlocked);
10008 
10009   // Not recursive. Check header for monitor (0b10).
10010   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10011   testptr(reg_rax, markWord::monitor_value);
10012   jcc(Assembler::notZero, push_and_slow);
10013 
10014 #ifdef ASSERT
10015   // Check header not unlocked (0b01).
10016   Label not_unlocked;
10017   testptr(reg_rax, markWord::unlocked_value);
10018   jcc(Assembler::zero, not_unlocked);
10019   stop("fast_unlock already unlocked");
10020   bind(not_unlocked);
10021 #endif
10022 
10023   // Try to unlock. Transition lock bits 0b00 => 0b01
10024   movptr(tmp, reg_rax);
10025   orptr(tmp, markWord::unlocked_value);
10026   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10027   jcc(Assembler::equal, unlocked);
10028 
10029   bind(push_and_slow);
10030   // Restore lock-stack and handle the unlock in runtime.
10031 #ifdef ASSERT
10032   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10033   movptr(Address(thread, top), obj);
10034 #endif
10035   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10036   jmp(slow);
10037 
10038   bind(unlocked);
10039 }
10040 
10041 // Saves legacy GPRs state on stack.
10042 void MacroAssembler::save_legacy_gprs() {
10043   subq(rsp, 16 * wordSize);
10044   movq(Address(rsp, 15 * wordSize), rax);
10045   movq(Address(rsp, 14 * wordSize), rcx);
10046   movq(Address(rsp, 13 * wordSize), rdx);
10047   movq(Address(rsp, 12 * wordSize), rbx);
10048   movq(Address(rsp, 10 * wordSize), rbp);
10049   movq(Address(rsp, 9 * wordSize), rsi);
10050   movq(Address(rsp, 8 * wordSize), rdi);
10051   movq(Address(rsp, 7 * wordSize), r8);
10052   movq(Address(rsp, 6 * wordSize), r9);
10053   movq(Address(rsp, 5 * wordSize), r10);
10054   movq(Address(rsp, 4 * wordSize), r11);
10055   movq(Address(rsp, 3 * wordSize), r12);
10056   movq(Address(rsp, 2 * wordSize), r13);
10057   movq(Address(rsp, wordSize), r14);
10058   movq(Address(rsp, 0), r15);
10059 }
10060 
10061 // Resotres back legacy GPRs state from stack.
10062 void MacroAssembler::restore_legacy_gprs() {
10063   movq(r15, Address(rsp, 0));
10064   movq(r14, Address(rsp, wordSize));
10065   movq(r13, Address(rsp, 2 * wordSize));
10066   movq(r12, Address(rsp, 3 * wordSize));
10067   movq(r11, Address(rsp, 4 * wordSize));
10068   movq(r10, Address(rsp, 5 * wordSize));
10069   movq(r9,  Address(rsp, 6 * wordSize));
10070   movq(r8,  Address(rsp, 7 * wordSize));
10071   movq(rdi, Address(rsp, 8 * wordSize));
10072   movq(rsi, Address(rsp, 9 * wordSize));
10073   movq(rbp, Address(rsp, 10 * wordSize));
10074   movq(rbx, Address(rsp, 12 * wordSize));
10075   movq(rdx, Address(rsp, 13 * wordSize));
10076   movq(rcx, Address(rsp, 14 * wordSize));
10077   movq(rax, Address(rsp, 15 * wordSize));
10078   addq(rsp, 16 * wordSize);
10079 }
10080 
10081 void MacroAssembler::load_aotrc_address(Register reg, address a) {
10082 #if INCLUDE_CDS
10083   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
10084   if (AOTCodeCache::is_on_for_dump()) {
10085     // all aotrc field addresses should be registered in the AOTCodeCache address table
10086     lea(reg, ExternalAddress(a));
10087   } else {
10088     mov64(reg, (uint64_t)a);
10089   }
10090 #else
10091   ShouldNotReachHere();
10092 #endif
10093 }
10094 
10095 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10096   if (VM_Version::supports_apx_f()) {
10097     esetzucc(comparison, dst);
10098   } else {
10099     setb(comparison, dst);
10100     movzbl(dst, dst);
10101   }
10102 }
--- EOF ---