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 (RestoreMXCSROnJNICalls) {
 5359     ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5360   } else if (CheckJNICalls) {
 5361     call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5362   }
 5363   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5364   vzeroupper();
 5365 }
 5366 
 5367 // ((OopHandle)result).resolve();
 5368 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5369   assert_different_registers(result, tmp);
 5370 
 5371   // Only 64 bit platforms support GCs that require a tmp register
 5372   // Only IN_HEAP loads require a thread_tmp register
 5373   // OopHandle::resolve is an indirection like jobject.
 5374   access_load_at(T_OBJECT, IN_NATIVE,
 5375                  result, Address(result, 0), tmp);
 5376 }
 5377 
 5378 // ((WeakHandle)result).resolve();
 5379 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5380   assert_different_registers(rresult, rtmp);
 5381   Label resolved;
 5382 
 5383   // A null weak handle resolves to null.
 5384   cmpptr(rresult, 0);
 5385   jcc(Assembler::equal, resolved);
 5386 
 5387   // Only 64 bit platforms support GCs that require a tmp register
 5388   // Only IN_HEAP loads require a thread_tmp register
 5389   // WeakHandle::resolve is an indirection like jweak.
 5390   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5391                  rresult, Address(rresult, 0), rtmp);
 5392   bind(resolved);
 5393 }
 5394 
 5395 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5396   // get mirror
 5397   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5398   load_method_holder(mirror, method);
 5399   movptr(mirror, Address(mirror, mirror_offset));
 5400   resolve_oop_handle(mirror, tmp);
 5401 }
 5402 
 5403 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5404   load_method_holder(rresult, rmethod);
 5405   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5406 }
 5407 
 5408 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5409   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5410   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5411   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5412 }
 5413 








 5414 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5415   assert(UseCompactObjectHeaders, "expect compact object headers");
 5416   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5417   shrq(dst, markWord::klass_shift);
 5418 }
 5419 
 5420 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5421   assert_different_registers(src, tmp);
 5422   assert_different_registers(dst, tmp);
 5423 
 5424   if (UseCompactObjectHeaders) {
 5425     load_narrow_klass_compact(dst, src);
 5426     decode_klass_not_null(dst, tmp);
 5427   } else {
 5428     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5429     decode_klass_not_null(dst, tmp);
 5430   }
 5431 }
 5432 





 5433 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5434   assert(!UseCompactObjectHeaders, "not with compact headers");
 5435   assert_different_registers(src, tmp);
 5436   assert_different_registers(dst, tmp);
 5437   encode_klass_not_null(src, tmp);
 5438   movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5439 }
 5440 
 5441 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5442   if (UseCompactObjectHeaders) {
 5443     assert(tmp != noreg, "need tmp");
 5444     assert_different_registers(klass, obj, tmp);
 5445     load_narrow_klass_compact(tmp, obj);
 5446     cmpl(klass, tmp);
 5447   } else {
 5448     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5449   }
 5450 }
 5451 
 5452 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5453   if (UseCompactObjectHeaders) {
 5454     assert(tmp2 != noreg, "need tmp2");
 5455     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5456     load_narrow_klass_compact(tmp1, obj1);
 5457     load_narrow_klass_compact(tmp2, obj2);
 5458     cmpl(tmp1, tmp2);
 5459   } else {
 5460     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5461     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5462   }
 5463 }
 5464 
 5465 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5466                                     Register tmp1) {
 5467   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5468   decorators = AccessInternal::decorator_fixup(decorators, type);
 5469   bool as_raw = (decorators & AS_RAW) != 0;
 5470   if (as_raw) {
 5471     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5472   } else {
 5473     bs->load_at(this, decorators, type, dst, src, tmp1);
 5474   }
 5475 }
 5476 
 5477 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5478                                      Register tmp1, Register tmp2, Register tmp3) {
 5479   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5480   decorators = AccessInternal::decorator_fixup(decorators, type);
 5481   bool as_raw = (decorators & AS_RAW) != 0;
 5482   if (as_raw) {
 5483     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5484   } else {
 5485     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5486   }
 5487 }
 5488 








































 5489 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5490   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5491 }
 5492 
 5493 // Doesn't do verification, generates fixed size code
 5494 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5495   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5496 }
 5497 
 5498 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5499                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5500   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5501 }
 5502 
 5503 // Used for storing nulls.
 5504 void MacroAssembler::store_heap_oop_null(Address dst) {
 5505   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5506 }
 5507 
 5508 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5509   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5510   // Store to klass gap in destination
 5511   movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5512 }
 5513 
 5514 #ifdef ASSERT
 5515 void MacroAssembler::verify_heapbase(const char* msg) {
 5516   assert (UseCompressedOops, "should be compressed");
 5517   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5518   if (CheckCompressedOops) {
 5519     Label ok;
 5520     ExternalAddress src2(CompressedOops::base_addr());
 5521     const bool is_src2_reachable = reachable(src2);
 5522     if (!is_src2_reachable) {
 5523       push(rscratch1);  // cmpptr trashes rscratch1
 5524     }
 5525     cmpptr(r12_heapbase, src2, rscratch1);
 5526     jcc(Assembler::equal, ok);
 5527     STOP(msg);
 5528     bind(ok);
 5529     if (!is_src2_reachable) {
 5530       pop(rscratch1);
 5531     }
 5532   }
 5533 }
 5534 #endif
 5535 
 5536 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5537 void MacroAssembler::encode_heap_oop(Register r) {
 5538 #ifdef ASSERT
 5539   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5540 #endif
 5541   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5542   if (CompressedOops::base() == nullptr) {
 5543     if (CompressedOops::shift() != 0) {
 5544       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5545       shrq(r, LogMinObjAlignmentInBytes);
 5546     }
 5547     return;
 5548   }
 5549   testq(r, r);
 5550   cmovq(Assembler::equal, r, r12_heapbase);
 5551   subq(r, r12_heapbase);
 5552   shrq(r, LogMinObjAlignmentInBytes);
 5553 }
 5554 
 5555 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5556 #ifdef ASSERT
 5557   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5558   if (CheckCompressedOops) {
 5559     Label ok;
 5560     testq(r, r);
 5561     jcc(Assembler::notEqual, ok);
 5562     STOP("null oop passed to encode_heap_oop_not_null");
 5563     bind(ok);
 5564   }
 5565 #endif
 5566   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5567   if (CompressedOops::base() != nullptr) {
 5568     subq(r, r12_heapbase);
 5569   }
 5570   if (CompressedOops::shift() != 0) {
 5571     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5572     shrq(r, LogMinObjAlignmentInBytes);
 5573   }
 5574 }
 5575 
 5576 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5577 #ifdef ASSERT
 5578   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5579   if (CheckCompressedOops) {
 5580     Label ok;
 5581     testq(src, src);
 5582     jcc(Assembler::notEqual, ok);
 5583     STOP("null oop passed to encode_heap_oop_not_null2");
 5584     bind(ok);
 5585   }
 5586 #endif
 5587   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5588   if (dst != src) {
 5589     movq(dst, src);
 5590   }
 5591   if (CompressedOops::base() != nullptr) {
 5592     subq(dst, r12_heapbase);
 5593   }
 5594   if (CompressedOops::shift() != 0) {
 5595     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5596     shrq(dst, LogMinObjAlignmentInBytes);
 5597   }
 5598 }
 5599 
 5600 void  MacroAssembler::decode_heap_oop(Register r) {
 5601 #ifdef ASSERT
 5602   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5603 #endif
 5604   if (CompressedOops::base() == nullptr) {
 5605     if (CompressedOops::shift() != 0) {
 5606       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5607       shlq(r, LogMinObjAlignmentInBytes);
 5608     }
 5609   } else {
 5610     Label done;
 5611     shlq(r, LogMinObjAlignmentInBytes);
 5612     jccb(Assembler::equal, done);
 5613     addq(r, r12_heapbase);
 5614     bind(done);
 5615   }
 5616   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5617 }
 5618 
 5619 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5620   // Note: it will change flags
 5621   assert (UseCompressedOops, "should only be used for compressed headers");
 5622   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5623   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5624   // vtableStubs also counts instructions in pd_code_size_limit.
 5625   // Also do not verify_oop as this is called by verify_oop.
 5626   if (CompressedOops::shift() != 0) {
 5627     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5628     shlq(r, LogMinObjAlignmentInBytes);
 5629     if (CompressedOops::base() != nullptr) {
 5630       addq(r, r12_heapbase);
 5631     }
 5632   } else {
 5633     assert (CompressedOops::base() == nullptr, "sanity");
 5634   }
 5635 }
 5636 
 5637 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5638   // Note: it will change flags
 5639   assert (UseCompressedOops, "should only be used for compressed headers");
 5640   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5641   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5642   // vtableStubs also counts instructions in pd_code_size_limit.
 5643   // Also do not verify_oop as this is called by verify_oop.
 5644   if (CompressedOops::shift() != 0) {
 5645     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5646     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5647       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5648     } else {
 5649       if (dst != src) {
 5650         movq(dst, src);
 5651       }
 5652       shlq(dst, LogMinObjAlignmentInBytes);
 5653       if (CompressedOops::base() != nullptr) {
 5654         addq(dst, r12_heapbase);
 5655       }
 5656     }
 5657   } else {
 5658     assert (CompressedOops::base() == nullptr, "sanity");
 5659     if (dst != src) {
 5660       movq(dst, src);
 5661     }
 5662   }
 5663 }
 5664 
 5665 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5666   BLOCK_COMMENT("encode_klass_not_null {");
 5667   assert_different_registers(r, tmp);
 5668   if (CompressedKlassPointers::base() != nullptr) {
 5669     if (AOTCodeCache::is_on_for_dump()) {
 5670       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5671     } else {
 5672       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5673     }
 5674     subq(r, tmp);
 5675   }
 5676   if (CompressedKlassPointers::shift() != 0) {
 5677     shrq(r, CompressedKlassPointers::shift());
 5678   }
 5679   BLOCK_COMMENT("} encode_klass_not_null");
 5680 }
 5681 
 5682 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5683   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5684   assert_different_registers(src, dst);
 5685   if (CompressedKlassPointers::base() != nullptr) {
 5686     if (AOTCodeCache::is_on_for_dump()) {
 5687       movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5688       negq(dst);
 5689     } else {
 5690       movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5691     }
 5692     addq(dst, src);
 5693   } else {
 5694     movptr(dst, src);
 5695   }
 5696   if (CompressedKlassPointers::shift() != 0) {
 5697     shrq(dst, CompressedKlassPointers::shift());
 5698   }
 5699   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5700 }
 5701 
 5702 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5703   BLOCK_COMMENT("decode_klass_not_null {");
 5704   assert_different_registers(r, tmp);
 5705   // Note: it will change flags
 5706   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5707   // vtableStubs also counts instructions in pd_code_size_limit.
 5708   // Also do not verify_oop as this is called by verify_oop.
 5709   if (CompressedKlassPointers::shift() != 0) {
 5710     shlq(r, CompressedKlassPointers::shift());
 5711   }
 5712   if (CompressedKlassPointers::base() != nullptr) {
 5713     if (AOTCodeCache::is_on_for_dump()) {
 5714       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5715     } else {
 5716       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5717     }
 5718     addq(r, tmp);
 5719   }
 5720   BLOCK_COMMENT("} decode_klass_not_null");
 5721 }
 5722 
 5723 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5724   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 5725   assert_different_registers(src, dst);
 5726   // Note: it will change flags
 5727   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5728   // vtableStubs also counts instructions in pd_code_size_limit.
 5729   // Also do not verify_oop as this is called by verify_oop.
 5730 
 5731   if (CompressedKlassPointers::base() == nullptr &&
 5732       CompressedKlassPointers::shift() == 0) {
 5733     // The best case scenario is that there is no base or shift. Then it is already
 5734     // a pointer that needs nothing but a register rename.
 5735     movl(dst, src);
 5736   } else {
 5737     if (CompressedKlassPointers::shift() <= Address::times_8) {
 5738       if (CompressedKlassPointers::base() != nullptr) {
 5739         if (AOTCodeCache::is_on_for_dump()) {
 5740           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5741         } else {
 5742           movptr(dst, (intptr_t)CompressedKlassPointers::base());
 5743         }
 5744       } else {
 5745         xorq(dst, dst);
 5746       }
 5747       if (CompressedKlassPointers::shift() != 0) {
 5748         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 5749         leaq(dst, Address(dst, src, Address::times_8, 0));
 5750       } else {
 5751         addq(dst, src);
 5752       }
 5753     } else {
 5754       if (CompressedKlassPointers::base() != nullptr) {
 5755         if (AOTCodeCache::is_on_for_dump()) {
 5756           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5757           shrq(dst, CompressedKlassPointers::shift());
 5758         } else {
 5759           const intptr_t base_right_shifted =
 5760                (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 5761           movptr(dst, base_right_shifted);
 5762         }
 5763       } else {
 5764         xorq(dst, dst);
 5765       }
 5766       addq(dst, src);
 5767       shlq(dst, CompressedKlassPointers::shift());
 5768     }
 5769   }
 5770   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 5771 }
 5772 
 5773 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5774   assert (UseCompressedOops, "should only be used for compressed headers");
 5775   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5776   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5777   int oop_index = oop_recorder()->find_index(obj);
 5778   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5779   mov_narrow_oop(dst, oop_index, rspec);
 5780 }
 5781 
 5782 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5783   assert (UseCompressedOops, "should only be used for compressed headers");
 5784   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5785   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5786   int oop_index = oop_recorder()->find_index(obj);
 5787   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5788   mov_narrow_oop(dst, oop_index, rspec);
 5789 }
 5790 
 5791 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5792   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5793   int klass_index = oop_recorder()->find_index(k);
 5794   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5795   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5796 }
 5797 
 5798 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 5799   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5800   int klass_index = oop_recorder()->find_index(k);
 5801   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5802   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5803 }
 5804 
 5805 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 5806   assert (UseCompressedOops, "should only be used for compressed headers");
 5807   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5808   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5809   int oop_index = oop_recorder()->find_index(obj);
 5810   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5811   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5812 }
 5813 
 5814 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 5815   assert (UseCompressedOops, "should only be used for compressed headers");
 5816   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5817   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5818   int oop_index = oop_recorder()->find_index(obj);
 5819   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5820   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5821 }
 5822 
 5823 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 5824   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5825   int klass_index = oop_recorder()->find_index(k);
 5826   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5827   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5828 }
 5829 
 5830 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 5831   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5832   int klass_index = oop_recorder()->find_index(k);
 5833   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5834   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5835 }
 5836 
 5837 void MacroAssembler::reinit_heapbase() {
 5838   if (UseCompressedOops) {
 5839     if (Universe::heap() != nullptr && !AOTCodeCache::is_on_for_dump()) {
 5840       if (CompressedOops::base() == nullptr) {
 5841         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 5842       } else {
 5843         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 5844       }
 5845     } else {
 5846       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 5847     }
 5848   }
 5849 }
 5850 




































































































































































































































































































































































































































































































 5851 #if COMPILER2_OR_JVMCI
 5852 
 5853 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 5854 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5855   // cnt - number of qwords (8-byte words).
 5856   // base - start address, qword aligned.
 5857   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 5858   bool use64byteVector = (MaxVectorSize == 64) && (CopyAVX3Threshold == 0);
 5859   if (use64byteVector) {
 5860     vpxor(xtmp, xtmp, xtmp, AVX_512bit);
 5861   } else if (MaxVectorSize >= 32) {
 5862     vpxor(xtmp, xtmp, xtmp, AVX_256bit);


 5863   } else {
 5864     pxor(xtmp, xtmp);

 5865   }
 5866   jmp(L_zero_64_bytes);
 5867 
 5868   BIND(L_loop);
 5869   if (MaxVectorSize >= 32) {
 5870     fill64(base, 0, xtmp, use64byteVector);
 5871   } else {
 5872     movdqu(Address(base,  0), xtmp);
 5873     movdqu(Address(base, 16), xtmp);
 5874     movdqu(Address(base, 32), xtmp);
 5875     movdqu(Address(base, 48), xtmp);
 5876   }
 5877   addptr(base, 64);
 5878 
 5879   BIND(L_zero_64_bytes);
 5880   subptr(cnt, 8);
 5881   jccb(Assembler::greaterEqual, L_loop);
 5882 
 5883   // Copy trailing 64 bytes
 5884   if (use64byteVector) {
 5885     addptr(cnt, 8);
 5886     jccb(Assembler::equal, L_end);
 5887     fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
 5888     jmp(L_end);
 5889   } else {
 5890     addptr(cnt, 4);
 5891     jccb(Assembler::less, L_tail);
 5892     if (MaxVectorSize >= 32) {
 5893       vmovdqu(Address(base, 0), xtmp);
 5894     } else {
 5895       movdqu(Address(base,  0), xtmp);
 5896       movdqu(Address(base, 16), xtmp);
 5897     }
 5898   }
 5899   addptr(base, 32);
 5900   subptr(cnt, 4);
 5901 
 5902   BIND(L_tail);
 5903   addptr(cnt, 4);
 5904   jccb(Assembler::lessEqual, L_end);
 5905   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 5906     fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
 5907   } else {
 5908     decrement(cnt);
 5909 
 5910     BIND(L_sloop);
 5911     movq(Address(base, 0), xtmp);
 5912     addptr(base, 8);
 5913     decrement(cnt);
 5914     jccb(Assembler::greaterEqual, L_sloop);
 5915   }
 5916   BIND(L_end);
 5917 }
 5918 
 5919 // Clearing constant sized memory using YMM/ZMM registers.
 5920 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5921   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 5922   bool use64byteVector = (MaxVectorSize > 32) && (CopyAVX3Threshold == 0);
 5923 
 5924   int vector64_count = (cnt & (~0x7)) >> 3;
 5925   cnt = cnt & 0x7;
 5926   const int fill64_per_loop = 4;
 5927   const int max_unrolled_fill64 = 8;
 5928 
 5929   // 64 byte initialization loop.
 5930   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 5931   int start64 = 0;
 5932   if (vector64_count > max_unrolled_fill64) {
 5933     Label LOOP;
 5934     Register index = rtmp;
 5935 
 5936     start64 = vector64_count - (vector64_count % fill64_per_loop);
 5937 
 5938     movl(index, 0);
 5939     BIND(LOOP);
 5940     for (int i = 0; i < fill64_per_loop; i++) {
 5941       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 5942     }
 5943     addl(index, fill64_per_loop * 64);
 5944     cmpl(index, start64 * 64);
 5945     jccb(Assembler::less, LOOP);
 5946   }
 5947   for (int i = start64; i < vector64_count; i++) {
 5948     fill64(base, i * 64, xtmp, use64byteVector);
 5949   }
 5950 
 5951   // Clear remaining 64 byte tail.
 5952   int disp = vector64_count * 64;
 5953   if (cnt) {
 5954     switch (cnt) {
 5955       case 1:
 5956         movq(Address(base, disp), xtmp);
 5957         break;
 5958       case 2:
 5959         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 5960         break;
 5961       case 3:
 5962         movl(rtmp, 0x7);
 5963         kmovwl(mask, rtmp);
 5964         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 5965         break;
 5966       case 4:
 5967         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5968         break;
 5969       case 5:
 5970         if (use64byteVector) {
 5971           movl(rtmp, 0x1F);
 5972           kmovwl(mask, rtmp);
 5973           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5974         } else {
 5975           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5976           movq(Address(base, disp + 32), xtmp);
 5977         }
 5978         break;
 5979       case 6:
 5980         if (use64byteVector) {
 5981           movl(rtmp, 0x3F);
 5982           kmovwl(mask, rtmp);
 5983           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5984         } else {
 5985           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5986           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 5987         }
 5988         break;
 5989       case 7:
 5990         if (use64byteVector) {
 5991           movl(rtmp, 0x7F);
 5992           kmovwl(mask, rtmp);
 5993           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5994         } else {
 5995           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5996           movl(rtmp, 0x7);
 5997           kmovwl(mask, rtmp);
 5998           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 5999         }
 6000         break;
 6001       default:
 6002         fatal("Unexpected length : %d\n",cnt);
 6003         break;
 6004     }
 6005   }
 6006 }
 6007 
 6008 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
 6009                                bool is_large, KRegister mask) {
 6010   // cnt      - number of qwords (8-byte words).
 6011   // base     - start address, qword aligned.
 6012   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6013   assert(base==rdi, "base register must be edi for rep stos");
 6014   assert(tmp==rax,   "tmp register must be eax for rep stos");
 6015   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6016   assert(InitArrayShortSize % BytesPerLong == 0,
 6017     "InitArrayShortSize should be the multiple of BytesPerLong");
 6018 
 6019   Label DONE;
 6020   if (!is_large || !UseXMMForObjInit) {
 6021     xorptr(tmp, tmp);
 6022   }
 6023 
 6024   if (!is_large) {
 6025     Label LOOP, LONG;
 6026     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6027     jccb(Assembler::greater, LONG);
 6028 
 6029     decrement(cnt);
 6030     jccb(Assembler::negative, DONE); // Zero length
 6031 
 6032     // Use individual pointer-sized stores for small counts:
 6033     BIND(LOOP);
 6034     movptr(Address(base, cnt, Address::times_ptr), tmp);
 6035     decrement(cnt);
 6036     jccb(Assembler::greaterEqual, LOOP);
 6037     jmpb(DONE);
 6038 
 6039     BIND(LONG);
 6040   }
 6041 
 6042   // Use longer rep-prefixed ops for non-small counts:
 6043   if (UseFastStosb) {
 6044     shlptr(cnt, 3); // convert to number of bytes
 6045     rep_stosb();
 6046   } else if (UseXMMForObjInit) {
 6047     xmm_clear_mem(base, cnt, tmp, xtmp, mask);
 6048   } else {
 6049     rep_stos();
 6050   }
 6051 
 6052   BIND(DONE);
 6053 }
 6054 
 6055 #endif //COMPILER2_OR_JVMCI
 6056 
 6057 
 6058 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6059                                    Register to, Register value, Register count,
 6060                                    Register rtmp, XMMRegister xtmp) {
 6061   ShortBranchVerifier sbv(this);
 6062   assert_different_registers(to, value, count, rtmp);
 6063   Label L_exit;
 6064   Label L_fill_2_bytes, L_fill_4_bytes;
 6065 
 6066 #if defined(COMPILER2)
 6067   if(MaxVectorSize >=32 &&
 6068      VM_Version::supports_avx512vlbw() &&
 6069      VM_Version::supports_bmi2()) {
 6070     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6071     return;
 6072   }
 6073 #endif
 6074 
 6075   int shift = -1;
 6076   switch (t) {
 6077     case T_BYTE:
 6078       shift = 2;
 6079       break;
 6080     case T_SHORT:
 6081       shift = 1;
 6082       break;
 6083     case T_INT:
 6084       shift = 0;
 6085       break;
 6086     default: ShouldNotReachHere();
 6087   }
 6088 
 6089   if (t == T_BYTE) {
 6090     andl(value, 0xff);
 6091     movl(rtmp, value);
 6092     shll(rtmp, 8);
 6093     orl(value, rtmp);
 6094   }
 6095   if (t == T_SHORT) {
 6096     andl(value, 0xffff);
 6097   }
 6098   if (t == T_BYTE || t == T_SHORT) {
 6099     movl(rtmp, value);
 6100     shll(rtmp, 16);
 6101     orl(value, rtmp);
 6102   }
 6103 
 6104   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6105   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6106   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6107     Label L_skip_align2;
 6108     // align source address at 4 bytes address boundary
 6109     if (t == T_BYTE) {
 6110       Label L_skip_align1;
 6111       // One byte misalignment happens only for byte arrays
 6112       testptr(to, 1);
 6113       jccb(Assembler::zero, L_skip_align1);
 6114       movb(Address(to, 0), value);
 6115       increment(to);
 6116       decrement(count);
 6117       BIND(L_skip_align1);
 6118     }
 6119     // Two bytes misalignment happens only for byte and short (char) arrays
 6120     testptr(to, 2);
 6121     jccb(Assembler::zero, L_skip_align2);
 6122     movw(Address(to, 0), value);
 6123     addptr(to, 2);
 6124     subptr(count, 1<<(shift-1));
 6125     BIND(L_skip_align2);
 6126   }
 6127   {
 6128     Label L_fill_32_bytes;
 6129     if (!UseUnalignedLoadStores) {
 6130       // align to 8 bytes, we know we are 4 byte aligned to start
 6131       testptr(to, 4);
 6132       jccb(Assembler::zero, L_fill_32_bytes);
 6133       movl(Address(to, 0), value);
 6134       addptr(to, 4);
 6135       subptr(count, 1<<shift);
 6136     }
 6137     BIND(L_fill_32_bytes);
 6138     {
 6139       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6140       movdl(xtmp, value);
 6141       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6142         Label L_check_fill_32_bytes;
 6143         if (UseAVX > 2) {
 6144           // Fill 64-byte chunks
 6145           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6146 
 6147           // If number of bytes to fill < CopyAVX3Threshold, perform fill using AVX2
 6148           cmpptr(count, CopyAVX3Threshold);
 6149           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6150 
 6151           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6152 
 6153           subptr(count, 16 << shift);
 6154           jcc(Assembler::less, L_check_fill_32_bytes);
 6155           align(16);
 6156 
 6157           BIND(L_fill_64_bytes_loop_avx3);
 6158           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6159           addptr(to, 64);
 6160           subptr(count, 16 << shift);
 6161           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6162           jmpb(L_check_fill_32_bytes);
 6163 
 6164           BIND(L_check_fill_64_bytes_avx2);
 6165         }
 6166         // Fill 64-byte chunks
 6167         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6168 
 6169         subptr(count, 16 << shift);
 6170         jcc(Assembler::less, L_check_fill_32_bytes);
 6171 
 6172         // align data for 64-byte chunks
 6173         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6174         if (EnableX86ECoreOpts) {
 6175             // align 'big' arrays to cache lines to minimize split_stores
 6176             cmpptr(count, 96 << shift);
 6177             jcc(Assembler::below, L_fill_64_bytes_loop);
 6178 
 6179             // Find the bytes needed for alignment
 6180             movptr(rtmp, to);
 6181             andptr(rtmp, 0x1c);
 6182             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6183             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6184             addptr(rtmp, 32);
 6185             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6186             subptr(count, rtmp);    // adjust count by number of elements
 6187 
 6188             align(16);
 6189             BIND(L_align_64_bytes_loop);
 6190             movdl(Address(to, 0), xtmp);
 6191             addptr(to, 4);
 6192             subptr(rtmp, 1 << shift);
 6193             jcc(Assembler::greater, L_align_64_bytes_loop);
 6194         }
 6195 
 6196         align(16);
 6197         BIND(L_fill_64_bytes_loop);
 6198         vmovdqu(Address(to, 0), xtmp);
 6199         vmovdqu(Address(to, 32), xtmp);
 6200         addptr(to, 64);
 6201         subptr(count, 16 << shift);
 6202         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6203 
 6204         align(16);
 6205         BIND(L_check_fill_32_bytes);
 6206         addptr(count, 8 << shift);
 6207         jccb(Assembler::less, L_check_fill_8_bytes);
 6208         vmovdqu(Address(to, 0), xtmp);
 6209         addptr(to, 32);
 6210         subptr(count, 8 << shift);
 6211 
 6212         BIND(L_check_fill_8_bytes);
 6213         // clean upper bits of YMM registers
 6214         movdl(xtmp, value);
 6215         pshufd(xtmp, xtmp, 0);
 6216       } else {
 6217         // Fill 32-byte chunks
 6218         pshufd(xtmp, xtmp, 0);
 6219 
 6220         subptr(count, 8 << shift);
 6221         jcc(Assembler::less, L_check_fill_8_bytes);
 6222         align(16);
 6223 
 6224         BIND(L_fill_32_bytes_loop);
 6225 
 6226         if (UseUnalignedLoadStores) {
 6227           movdqu(Address(to, 0), xtmp);
 6228           movdqu(Address(to, 16), xtmp);
 6229         } else {
 6230           movq(Address(to, 0), xtmp);
 6231           movq(Address(to, 8), xtmp);
 6232           movq(Address(to, 16), xtmp);
 6233           movq(Address(to, 24), xtmp);
 6234         }
 6235 
 6236         addptr(to, 32);
 6237         subptr(count, 8 << shift);
 6238         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6239 
 6240         BIND(L_check_fill_8_bytes);
 6241       }
 6242       addptr(count, 8 << shift);
 6243       jccb(Assembler::zero, L_exit);
 6244       jmpb(L_fill_8_bytes);
 6245 
 6246       //
 6247       // length is too short, just fill qwords
 6248       //
 6249       align(16);
 6250       BIND(L_fill_8_bytes_loop);
 6251       movq(Address(to, 0), xtmp);
 6252       addptr(to, 8);
 6253       BIND(L_fill_8_bytes);
 6254       subptr(count, 1 << (shift + 1));
 6255       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6256     }
 6257   }
 6258 
 6259   Label L_fill_4_bytes_loop;
 6260   testl(count, 1 << shift);
 6261   jccb(Assembler::zero, L_fill_2_bytes);
 6262 
 6263   align(16);
 6264   BIND(L_fill_4_bytes_loop);
 6265   movl(Address(to, 0), value);
 6266   addptr(to, 4);
 6267 
 6268   BIND(L_fill_4_bytes);
 6269   subptr(count, 1 << shift);
 6270   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 6271 
 6272   if (t == T_BYTE || t == T_SHORT) {
 6273     Label L_fill_byte;
 6274     BIND(L_fill_2_bytes);
 6275     // fill trailing 2 bytes
 6276     testl(count, 1<<(shift-1));
 6277     jccb(Assembler::zero, L_fill_byte);
 6278     movw(Address(to, 0), value);
 6279     if (t == T_BYTE) {
 6280       addptr(to, 2);
 6281       BIND(L_fill_byte);
 6282       // fill trailing byte
 6283       testl(count, 1);
 6284       jccb(Assembler::zero, L_exit);
 6285       movb(Address(to, 0), value);
 6286     } else {
 6287       BIND(L_fill_byte);
 6288     }
 6289   } else {
 6290     BIND(L_fill_2_bytes);
 6291   }
 6292   BIND(L_exit);
 6293 }
 6294 
 6295 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6296   switch(type) {
 6297     case T_BYTE:
 6298     case T_BOOLEAN:
 6299       evpbroadcastb(dst, src, vector_len);
 6300       break;
 6301     case T_SHORT:
 6302     case T_CHAR:
 6303       evpbroadcastw(dst, src, vector_len);
 6304       break;
 6305     case T_INT:
 6306     case T_FLOAT:
 6307       evpbroadcastd(dst, src, vector_len);
 6308       break;
 6309     case T_LONG:
 6310     case T_DOUBLE:
 6311       evpbroadcastq(dst, src, vector_len);
 6312       break;
 6313     default:
 6314       fatal("Unhandled type : %s", type2name(type));
 6315       break;
 6316   }
 6317 }
 6318 
 6319 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 6320 //
 6321 // @IntrinsicCandidate
 6322 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 6323 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6324 //     int i = 0;
 6325 //     for (; i < len; i++) {
 6326 //         char c = sa[sp++];
 6327 //         if (c > '\u00FF')
 6328 //             break;
 6329 //         da[dp++] = (byte) c;
 6330 //     }
 6331 //     return i;
 6332 // }
 6333 //
 6334 // @IntrinsicCandidate
 6335 // int java.lang.StringCoding.encodeISOArray0(
 6336 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 6337 //   int i = 0;
 6338 //   for (; i < len; i++) {
 6339 //     char c = StringUTF16.getChar(sa, sp++);
 6340 //     if (c > '\u00FF')
 6341 //       break;
 6342 //     da[dp++] = (byte) c;
 6343 //   }
 6344 //   return i;
 6345 // }
 6346 //
 6347 // @IntrinsicCandidate
 6348 // int java.lang.StringCoding.encodeAsciiArray0(
 6349 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6350 //   int i = 0;
 6351 //   for (; i < len; i++) {
 6352 //     char c = sa[sp++];
 6353 //     if (c >= '\u0080')
 6354 //       break;
 6355 //     da[dp++] = (byte) c;
 6356 //   }
 6357 //   return i;
 6358 // }
 6359 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 6360   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 6361   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 6362   Register tmp5, Register result, bool ascii) {
 6363 
 6364   // rsi: src
 6365   // rdi: dst
 6366   // rdx: len
 6367   // rcx: tmp5
 6368   // rax: result
 6369   ShortBranchVerifier sbv(this);
 6370   assert_different_registers(src, dst, len, tmp5, result);
 6371   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 6372 
 6373   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 6374   int short_mask = ascii ? 0xff80 : 0xff00;
 6375 
 6376   // set result
 6377   xorl(result, result);
 6378   // check for zero length
 6379   testl(len, len);
 6380   jcc(Assembler::zero, L_done);
 6381 
 6382   movl(result, len);
 6383 
 6384   // Setup pointers
 6385   lea(src, Address(src, len, Address::times_2)); // char[]
 6386   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 6387   negptr(len);
 6388 
 6389   if (UseSSE42Intrinsics || UseAVX >= 2) {
 6390     Label L_copy_8_chars, L_copy_8_chars_exit;
 6391     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 6392 
 6393     if (UseAVX >= 2) {
 6394       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 6395       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6396       movdl(tmp1Reg, tmp5);
 6397       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 6398       jmp(L_chars_32_check);
 6399 
 6400       bind(L_copy_32_chars);
 6401       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 6402       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 6403       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6404       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6405       jccb(Assembler::notZero, L_copy_32_chars_exit);
 6406       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6407       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 6408       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 6409 
 6410       bind(L_chars_32_check);
 6411       addptr(len, 32);
 6412       jcc(Assembler::lessEqual, L_copy_32_chars);
 6413 
 6414       bind(L_copy_32_chars_exit);
 6415       subptr(len, 16);
 6416       jccb(Assembler::greater, L_copy_16_chars_exit);
 6417 
 6418     } else if (UseSSE42Intrinsics) {
 6419       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6420       movdl(tmp1Reg, tmp5);
 6421       pshufd(tmp1Reg, tmp1Reg, 0);
 6422       jmpb(L_chars_16_check);
 6423     }
 6424 
 6425     bind(L_copy_16_chars);
 6426     if (UseAVX >= 2) {
 6427       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 6428       vptest(tmp2Reg, tmp1Reg);
 6429       jcc(Assembler::notZero, L_copy_16_chars_exit);
 6430       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 6431       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 6432     } else {
 6433       if (UseAVX > 0) {
 6434         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6435         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6436         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 6437       } else {
 6438         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6439         por(tmp2Reg, tmp3Reg);
 6440         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6441         por(tmp2Reg, tmp4Reg);
 6442       }
 6443       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6444       jccb(Assembler::notZero, L_copy_16_chars_exit);
 6445       packuswb(tmp3Reg, tmp4Reg);
 6446     }
 6447     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 6448 
 6449     bind(L_chars_16_check);
 6450     addptr(len, 16);
 6451     jcc(Assembler::lessEqual, L_copy_16_chars);
 6452 
 6453     bind(L_copy_16_chars_exit);
 6454     if (UseAVX >= 2) {
 6455       // clean upper bits of YMM registers
 6456       vpxor(tmp2Reg, tmp2Reg);
 6457       vpxor(tmp3Reg, tmp3Reg);
 6458       vpxor(tmp4Reg, tmp4Reg);
 6459       movdl(tmp1Reg, tmp5);
 6460       pshufd(tmp1Reg, tmp1Reg, 0);
 6461     }
 6462     subptr(len, 8);
 6463     jccb(Assembler::greater, L_copy_8_chars_exit);
 6464 
 6465     bind(L_copy_8_chars);
 6466     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 6467     ptest(tmp3Reg, tmp1Reg);
 6468     jccb(Assembler::notZero, L_copy_8_chars_exit);
 6469     packuswb(tmp3Reg, tmp1Reg);
 6470     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 6471     addptr(len, 8);
 6472     jccb(Assembler::lessEqual, L_copy_8_chars);
 6473 
 6474     bind(L_copy_8_chars_exit);
 6475     subptr(len, 8);
 6476     jccb(Assembler::zero, L_done);
 6477   }
 6478 
 6479   bind(L_copy_1_char);
 6480   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 6481   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 6482   jccb(Assembler::notZero, L_copy_1_char_exit);
 6483   movb(Address(dst, len, Address::times_1, 0), tmp5);
 6484   addptr(len, 1);
 6485   jccb(Assembler::less, L_copy_1_char);
 6486 
 6487   bind(L_copy_1_char_exit);
 6488   addptr(result, len); // len is negative count of not processed elements
 6489 
 6490   bind(L_done);
 6491 }
 6492 
 6493 /**
 6494  * Helper for multiply_to_len().
 6495  */
 6496 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 6497   addq(dest_lo, src1);
 6498   adcq(dest_hi, 0);
 6499   addq(dest_lo, src2);
 6500   adcq(dest_hi, 0);
 6501 }
 6502 
 6503 /**
 6504  * Multiply 64 bit by 64 bit first loop.
 6505  */
 6506 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 6507                                            Register y, Register y_idx, Register z,
 6508                                            Register carry, Register product,
 6509                                            Register idx, Register kdx) {
 6510   //
 6511   //  jlong carry, x[], y[], z[];
 6512   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6513   //    huge_128 product = y[idx] * x[xstart] + carry;
 6514   //    z[kdx] = (jlong)product;
 6515   //    carry  = (jlong)(product >>> 64);
 6516   //  }
 6517   //  z[xstart] = carry;
 6518   //
 6519 
 6520   Label L_first_loop, L_first_loop_exit;
 6521   Label L_one_x, L_one_y, L_multiply;
 6522 
 6523   decrementl(xstart);
 6524   jcc(Assembler::negative, L_one_x);
 6525 
 6526   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6527   rorq(x_xstart, 32); // convert big-endian to little-endian
 6528 
 6529   bind(L_first_loop);
 6530   decrementl(idx);
 6531   jcc(Assembler::negative, L_first_loop_exit);
 6532   decrementl(idx);
 6533   jcc(Assembler::negative, L_one_y);
 6534   movq(y_idx, Address(y, idx, Address::times_4,  0));
 6535   rorq(y_idx, 32); // convert big-endian to little-endian
 6536   bind(L_multiply);
 6537   movq(product, x_xstart);
 6538   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 6539   addq(product, carry);
 6540   adcq(rdx, 0);
 6541   subl(kdx, 2);
 6542   movl(Address(z, kdx, Address::times_4,  4), product);
 6543   shrq(product, 32);
 6544   movl(Address(z, kdx, Address::times_4,  0), product);
 6545   movq(carry, rdx);
 6546   jmp(L_first_loop);
 6547 
 6548   bind(L_one_y);
 6549   movl(y_idx, Address(y,  0));
 6550   jmp(L_multiply);
 6551 
 6552   bind(L_one_x);
 6553   movl(x_xstart, Address(x,  0));
 6554   jmp(L_first_loop);
 6555 
 6556   bind(L_first_loop_exit);
 6557 }
 6558 
 6559 /**
 6560  * Multiply 64 bit by 64 bit and add 128 bit.
 6561  */
 6562 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 6563                                             Register yz_idx, Register idx,
 6564                                             Register carry, Register product, int offset) {
 6565   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 6566   //     z[kdx] = (jlong)product;
 6567 
 6568   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 6569   rorq(yz_idx, 32); // convert big-endian to little-endian
 6570   movq(product, x_xstart);
 6571   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 6572   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 6573   rorq(yz_idx, 32); // convert big-endian to little-endian
 6574 
 6575   add2_with_carry(rdx, product, carry, yz_idx);
 6576 
 6577   movl(Address(z, idx, Address::times_4,  offset+4), product);
 6578   shrq(product, 32);
 6579   movl(Address(z, idx, Address::times_4,  offset), product);
 6580 
 6581 }
 6582 
 6583 /**
 6584  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 6585  */
 6586 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 6587                                              Register yz_idx, Register idx, Register jdx,
 6588                                              Register carry, Register product,
 6589                                              Register carry2) {
 6590   //   jlong carry, x[], y[], z[];
 6591   //   int kdx = ystart+1;
 6592   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6593   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 6594   //     z[kdx+idx+1] = (jlong)product;
 6595   //     jlong carry2  = (jlong)(product >>> 64);
 6596   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 6597   //     z[kdx+idx] = (jlong)product;
 6598   //     carry  = (jlong)(product >>> 64);
 6599   //   }
 6600   //   idx += 2;
 6601   //   if (idx > 0) {
 6602   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 6603   //     z[kdx+idx] = (jlong)product;
 6604   //     carry  = (jlong)(product >>> 64);
 6605   //   }
 6606   //
 6607 
 6608   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6609 
 6610   movl(jdx, idx);
 6611   andl(jdx, 0xFFFFFFFC);
 6612   shrl(jdx, 2);
 6613 
 6614   bind(L_third_loop);
 6615   subl(jdx, 1);
 6616   jcc(Assembler::negative, L_third_loop_exit);
 6617   subl(idx, 4);
 6618 
 6619   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 6620   movq(carry2, rdx);
 6621 
 6622   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 6623   movq(carry, rdx);
 6624   jmp(L_third_loop);
 6625 
 6626   bind (L_third_loop_exit);
 6627 
 6628   andl (idx, 0x3);
 6629   jcc(Assembler::zero, L_post_third_loop_done);
 6630 
 6631   Label L_check_1;
 6632   subl(idx, 2);
 6633   jcc(Assembler::negative, L_check_1);
 6634 
 6635   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 6636   movq(carry, rdx);
 6637 
 6638   bind (L_check_1);
 6639   addl (idx, 0x2);
 6640   andl (idx, 0x1);
 6641   subl(idx, 1);
 6642   jcc(Assembler::negative, L_post_third_loop_done);
 6643 
 6644   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 6645   movq(product, x_xstart);
 6646   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 6647   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 6648 
 6649   add2_with_carry(rdx, product, yz_idx, carry);
 6650 
 6651   movl(Address(z, idx, Address::times_4,  0), product);
 6652   shrq(product, 32);
 6653 
 6654   shlq(rdx, 32);
 6655   orq(product, rdx);
 6656   movq(carry, product);
 6657 
 6658   bind(L_post_third_loop_done);
 6659 }
 6660 
 6661 /**
 6662  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 6663  *
 6664  */
 6665 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 6666                                                   Register carry, Register carry2,
 6667                                                   Register idx, Register jdx,
 6668                                                   Register yz_idx1, Register yz_idx2,
 6669                                                   Register tmp, Register tmp3, Register tmp4) {
 6670   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 6671 
 6672   //   jlong carry, x[], y[], z[];
 6673   //   int kdx = ystart+1;
 6674   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6675   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 6676   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 6677   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 6678   //     carry  = (jlong)(tmp4 >>> 64);
 6679   //     z[kdx+idx+1] = (jlong)tmp3;
 6680   //     z[kdx+idx] = (jlong)tmp4;
 6681   //   }
 6682   //   idx += 2;
 6683   //   if (idx > 0) {
 6684   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 6685   //     z[kdx+idx] = (jlong)yz_idx1;
 6686   //     carry  = (jlong)(yz_idx1 >>> 64);
 6687   //   }
 6688   //
 6689 
 6690   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6691 
 6692   movl(jdx, idx);
 6693   andl(jdx, 0xFFFFFFFC);
 6694   shrl(jdx, 2);
 6695 
 6696   bind(L_third_loop);
 6697   subl(jdx, 1);
 6698   jcc(Assembler::negative, L_third_loop_exit);
 6699   subl(idx, 4);
 6700 
 6701   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 6702   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 6703   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 6704   rorxq(yz_idx2, yz_idx2, 32);
 6705 
 6706   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 6707   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 6708 
 6709   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 6710   rorxq(yz_idx1, yz_idx1, 32);
 6711   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6712   rorxq(yz_idx2, yz_idx2, 32);
 6713 
 6714   if (VM_Version::supports_adx()) {
 6715     adcxq(tmp3, carry);
 6716     adoxq(tmp3, yz_idx1);
 6717 
 6718     adcxq(tmp4, tmp);
 6719     adoxq(tmp4, yz_idx2);
 6720 
 6721     movl(carry, 0); // does not affect flags
 6722     adcxq(carry2, carry);
 6723     adoxq(carry2, carry);
 6724   } else {
 6725     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 6726     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 6727   }
 6728   movq(carry, carry2);
 6729 
 6730   movl(Address(z, idx, Address::times_4, 12), tmp3);
 6731   shrq(tmp3, 32);
 6732   movl(Address(z, idx, Address::times_4,  8), tmp3);
 6733 
 6734   movl(Address(z, idx, Address::times_4,  4), tmp4);
 6735   shrq(tmp4, 32);
 6736   movl(Address(z, idx, Address::times_4,  0), tmp4);
 6737 
 6738   jmp(L_third_loop);
 6739 
 6740   bind (L_third_loop_exit);
 6741 
 6742   andl (idx, 0x3);
 6743   jcc(Assembler::zero, L_post_third_loop_done);
 6744 
 6745   Label L_check_1;
 6746   subl(idx, 2);
 6747   jcc(Assembler::negative, L_check_1);
 6748 
 6749   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 6750   rorxq(yz_idx1, yz_idx1, 32);
 6751   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 6752   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6753   rorxq(yz_idx2, yz_idx2, 32);
 6754 
 6755   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 6756 
 6757   movl(Address(z, idx, Address::times_4,  4), tmp3);
 6758   shrq(tmp3, 32);
 6759   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6760   movq(carry, tmp4);
 6761 
 6762   bind (L_check_1);
 6763   addl (idx, 0x2);
 6764   andl (idx, 0x1);
 6765   subl(idx, 1);
 6766   jcc(Assembler::negative, L_post_third_loop_done);
 6767   movl(tmp4, Address(y, idx, Address::times_4,  0));
 6768   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 6769   movl(tmp4, Address(z, idx, Address::times_4,  0));
 6770 
 6771   add2_with_carry(carry2, tmp3, tmp4, carry);
 6772 
 6773   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6774   shrq(tmp3, 32);
 6775 
 6776   shlq(carry2, 32);
 6777   orq(tmp3, carry2);
 6778   movq(carry, tmp3);
 6779 
 6780   bind(L_post_third_loop_done);
 6781 }
 6782 
 6783 /**
 6784  * Code for BigInteger::multiplyToLen() intrinsic.
 6785  *
 6786  * rdi: x
 6787  * rax: xlen
 6788  * rsi: y
 6789  * rcx: ylen
 6790  * r8:  z
 6791  * r11: tmp0
 6792  * r12: tmp1
 6793  * r13: tmp2
 6794  * r14: tmp3
 6795  * r15: tmp4
 6796  * rbx: tmp5
 6797  *
 6798  */
 6799 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 6800                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 6801   ShortBranchVerifier sbv(this);
 6802   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 6803 
 6804   push(tmp0);
 6805   push(tmp1);
 6806   push(tmp2);
 6807   push(tmp3);
 6808   push(tmp4);
 6809   push(tmp5);
 6810 
 6811   push(xlen);
 6812 
 6813   const Register idx = tmp1;
 6814   const Register kdx = tmp2;
 6815   const Register xstart = tmp3;
 6816 
 6817   const Register y_idx = tmp4;
 6818   const Register carry = tmp5;
 6819   const Register product  = xlen;
 6820   const Register x_xstart = tmp0;
 6821 
 6822   // First Loop.
 6823   //
 6824   //  final static long LONG_MASK = 0xffffffffL;
 6825   //  int xstart = xlen - 1;
 6826   //  int ystart = ylen - 1;
 6827   //  long carry = 0;
 6828   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6829   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 6830   //    z[kdx] = (int)product;
 6831   //    carry = product >>> 32;
 6832   //  }
 6833   //  z[xstart] = (int)carry;
 6834   //
 6835 
 6836   movl(idx, ylen);               // idx = ylen;
 6837   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 6838   xorq(carry, carry);            // carry = 0;
 6839 
 6840   Label L_done;
 6841 
 6842   movl(xstart, xlen);
 6843   decrementl(xstart);
 6844   jcc(Assembler::negative, L_done);
 6845 
 6846   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 6847 
 6848   Label L_second_loop;
 6849   testl(kdx, kdx);
 6850   jcc(Assembler::zero, L_second_loop);
 6851 
 6852   Label L_carry;
 6853   subl(kdx, 1);
 6854   jcc(Assembler::zero, L_carry);
 6855 
 6856   movl(Address(z, kdx, Address::times_4,  0), carry);
 6857   shrq(carry, 32);
 6858   subl(kdx, 1);
 6859 
 6860   bind(L_carry);
 6861   movl(Address(z, kdx, Address::times_4,  0), carry);
 6862 
 6863   // Second and third (nested) loops.
 6864   //
 6865   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 6866   //   carry = 0;
 6867   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 6868   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 6869   //                    (z[k] & LONG_MASK) + carry;
 6870   //     z[k] = (int)product;
 6871   //     carry = product >>> 32;
 6872   //   }
 6873   //   z[i] = (int)carry;
 6874   // }
 6875   //
 6876   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 6877 
 6878   const Register jdx = tmp1;
 6879 
 6880   bind(L_second_loop);
 6881   xorl(carry, carry);    // carry = 0;
 6882   movl(jdx, ylen);       // j = ystart+1
 6883 
 6884   subl(xstart, 1);       // i = xstart-1;
 6885   jcc(Assembler::negative, L_done);
 6886 
 6887   push (z);
 6888 
 6889   Label L_last_x;
 6890   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 6891   subl(xstart, 1);       // i = xstart-1;
 6892   jcc(Assembler::negative, L_last_x);
 6893 
 6894   if (UseBMI2Instructions) {
 6895     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 6896     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 6897   } else {
 6898     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6899     rorq(x_xstart, 32);  // convert big-endian to little-endian
 6900   }
 6901 
 6902   Label L_third_loop_prologue;
 6903   bind(L_third_loop_prologue);
 6904 
 6905   push (x);
 6906   push (xstart);
 6907   push (ylen);
 6908 
 6909 
 6910   if (UseBMI2Instructions) {
 6911     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 6912   } else { // !UseBMI2Instructions
 6913     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 6914   }
 6915 
 6916   pop(ylen);
 6917   pop(xlen);
 6918   pop(x);
 6919   pop(z);
 6920 
 6921   movl(tmp3, xlen);
 6922   addl(tmp3, 1);
 6923   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6924   subl(tmp3, 1);
 6925   jccb(Assembler::negative, L_done);
 6926 
 6927   shrq(carry, 32);
 6928   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6929   jmp(L_second_loop);
 6930 
 6931   // Next infrequent code is moved outside loops.
 6932   bind(L_last_x);
 6933   if (UseBMI2Instructions) {
 6934     movl(rdx, Address(x,  0));
 6935   } else {
 6936     movl(x_xstart, Address(x,  0));
 6937   }
 6938   jmp(L_third_loop_prologue);
 6939 
 6940   bind(L_done);
 6941 
 6942   pop(xlen);
 6943 
 6944   pop(tmp5);
 6945   pop(tmp4);
 6946   pop(tmp3);
 6947   pop(tmp2);
 6948   pop(tmp1);
 6949   pop(tmp0);
 6950 }
 6951 
 6952 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 6953   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 6954   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 6955   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 6956   Label VECTOR8_TAIL, VECTOR4_TAIL;
 6957   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 6958   Label SAME_TILL_END, DONE;
 6959   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 6960 
 6961   //scale is in rcx in both Win64 and Unix
 6962   ShortBranchVerifier sbv(this);
 6963 
 6964   shlq(length);
 6965   xorq(result, result);
 6966 
 6967   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 6968       VM_Version::supports_avx512vlbw() && UseCountTrailingZerosInstruction) {
 6969     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 6970 
 6971     cmpq(length, 64);
 6972     jcc(Assembler::less, VECTOR32_TAIL);
 6973 
 6974     movq(tmp1, length);
 6975     andq(tmp1, 0x3F);      // tail count
 6976     andq(length, ~(0x3F)); //vector count
 6977 
 6978     bind(VECTOR64_LOOP);
 6979     // AVX512 code to compare 64 byte vectors.
 6980     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 6981     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6982     kortestql(k7, k7);
 6983     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 6984     addq(result, 64);
 6985     subq(length, 64);
 6986     jccb(Assembler::notZero, VECTOR64_LOOP);
 6987 
 6988     //bind(VECTOR64_TAIL);
 6989     testq(tmp1, tmp1);
 6990     jcc(Assembler::zero, SAME_TILL_END);
 6991 
 6992     //bind(VECTOR64_TAIL);
 6993     // AVX512 code to compare up to 63 byte vectors.
 6994     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 6995     shlxq(tmp2, tmp2, tmp1);
 6996     notq(tmp2);
 6997     kmovql(k3, tmp2);
 6998 
 6999     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 7000     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7001 
 7002     ktestql(k7, k3);
 7003     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 7004 
 7005     bind(VECTOR64_NOT_EQUAL);
 7006     kmovql(tmp1, k7);
 7007     notq(tmp1);
 7008     tzcntq(tmp1, tmp1);
 7009     addq(result, tmp1);
 7010     shrq(result);
 7011     jmp(DONE);
 7012     bind(VECTOR32_TAIL);
 7013   }
 7014 
 7015   cmpq(length, 8);
 7016   jcc(Assembler::equal, VECTOR8_LOOP);
 7017   jcc(Assembler::less, VECTOR4_TAIL);
 7018 
 7019   if (UseAVX >= 2) {
 7020     Label VECTOR16_TAIL, VECTOR32_LOOP;
 7021 
 7022     cmpq(length, 16);
 7023     jcc(Assembler::equal, VECTOR16_LOOP);
 7024     jcc(Assembler::less, VECTOR8_LOOP);
 7025 
 7026     cmpq(length, 32);
 7027     jccb(Assembler::less, VECTOR16_TAIL);
 7028 
 7029     subq(length, 32);
 7030     bind(VECTOR32_LOOP);
 7031     vmovdqu(rymm0, Address(obja, result));
 7032     vmovdqu(rymm1, Address(objb, result));
 7033     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 7034     vptest(rymm2, rymm2);
 7035     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7036     addq(result, 32);
 7037     subq(length, 32);
 7038     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7039     addq(length, 32);
 7040     jcc(Assembler::equal, SAME_TILL_END);
 7041     //falling through if less than 32 bytes left //close the branch here.
 7042 
 7043     bind(VECTOR16_TAIL);
 7044     cmpq(length, 16);
 7045     jccb(Assembler::less, VECTOR8_TAIL);
 7046     bind(VECTOR16_LOOP);
 7047     movdqu(rymm0, Address(obja, result));
 7048     movdqu(rymm1, Address(objb, result));
 7049     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7050     ptest(rymm2, rymm2);
 7051     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7052     addq(result, 16);
 7053     subq(length, 16);
 7054     jcc(Assembler::equal, SAME_TILL_END);
 7055     //falling through if less than 16 bytes left
 7056   } else {//regular intrinsics
 7057 
 7058     cmpq(length, 16);
 7059     jccb(Assembler::less, VECTOR8_TAIL);
 7060 
 7061     subq(length, 16);
 7062     bind(VECTOR16_LOOP);
 7063     movdqu(rymm0, Address(obja, result));
 7064     movdqu(rymm1, Address(objb, result));
 7065     pxor(rymm0, rymm1);
 7066     ptest(rymm0, rymm0);
 7067     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7068     addq(result, 16);
 7069     subq(length, 16);
 7070     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7071     addq(length, 16);
 7072     jcc(Assembler::equal, SAME_TILL_END);
 7073     //falling through if less than 16 bytes left
 7074   }
 7075 
 7076   bind(VECTOR8_TAIL);
 7077   cmpq(length, 8);
 7078   jccb(Assembler::less, VECTOR4_TAIL);
 7079   bind(VECTOR8_LOOP);
 7080   movq(tmp1, Address(obja, result));
 7081   movq(tmp2, Address(objb, result));
 7082   xorq(tmp1, tmp2);
 7083   testq(tmp1, tmp1);
 7084   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7085   addq(result, 8);
 7086   subq(length, 8);
 7087   jcc(Assembler::equal, SAME_TILL_END);
 7088   //falling through if less than 8 bytes left
 7089 
 7090   bind(VECTOR4_TAIL);
 7091   cmpq(length, 4);
 7092   jccb(Assembler::less, BYTES_TAIL);
 7093   bind(VECTOR4_LOOP);
 7094   movl(tmp1, Address(obja, result));
 7095   xorl(tmp1, Address(objb, result));
 7096   testl(tmp1, tmp1);
 7097   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7098   addq(result, 4);
 7099   subq(length, 4);
 7100   jcc(Assembler::equal, SAME_TILL_END);
 7101   //falling through if less than 4 bytes left
 7102 
 7103   bind(BYTES_TAIL);
 7104   bind(BYTES_LOOP);
 7105   load_unsigned_byte(tmp1, Address(obja, result));
 7106   load_unsigned_byte(tmp2, Address(objb, result));
 7107   xorl(tmp1, tmp2);
 7108   testl(tmp1, tmp1);
 7109   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7110   decq(length);
 7111   jcc(Assembler::zero, SAME_TILL_END);
 7112   incq(result);
 7113   load_unsigned_byte(tmp1, Address(obja, result));
 7114   load_unsigned_byte(tmp2, Address(objb, result));
 7115   xorl(tmp1, tmp2);
 7116   testl(tmp1, tmp1);
 7117   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7118   decq(length);
 7119   jcc(Assembler::zero, SAME_TILL_END);
 7120   incq(result);
 7121   load_unsigned_byte(tmp1, Address(obja, result));
 7122   load_unsigned_byte(tmp2, Address(objb, result));
 7123   xorl(tmp1, tmp2);
 7124   testl(tmp1, tmp1);
 7125   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7126   jmp(SAME_TILL_END);
 7127 
 7128   if (UseAVX >= 2) {
 7129     bind(VECTOR32_NOT_EQUAL);
 7130     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7131     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7132     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7133     vpmovmskb(tmp1, rymm0);
 7134     bsfq(tmp1, tmp1);
 7135     addq(result, tmp1);
 7136     shrq(result);
 7137     jmp(DONE);
 7138   }
 7139 
 7140   bind(VECTOR16_NOT_EQUAL);
 7141   if (UseAVX >= 2) {
 7142     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7143     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7144     pxor(rymm0, rymm2);
 7145   } else {
 7146     pcmpeqb(rymm2, rymm2);
 7147     pxor(rymm0, rymm1);
 7148     pcmpeqb(rymm0, rymm1);
 7149     pxor(rymm0, rymm2);
 7150   }
 7151   pmovmskb(tmp1, rymm0);
 7152   bsfq(tmp1, tmp1);
 7153   addq(result, tmp1);
 7154   shrq(result);
 7155   jmpb(DONE);
 7156 
 7157   bind(VECTOR8_NOT_EQUAL);
 7158   bind(VECTOR4_NOT_EQUAL);
 7159   bsfq(tmp1, tmp1);
 7160   shrq(tmp1, 3);
 7161   addq(result, tmp1);
 7162   bind(BYTES_NOT_EQUAL);
 7163   shrq(result);
 7164   jmpb(DONE);
 7165 
 7166   bind(SAME_TILL_END);
 7167   mov64(result, -1);
 7168 
 7169   bind(DONE);
 7170 }
 7171 
 7172 //Helper functions for square_to_len()
 7173 
 7174 /**
 7175  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7176  * Preserves x and z and modifies rest of the registers.
 7177  */
 7178 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7179   // Perform square and right shift by 1
 7180   // Handle odd xlen case first, then for even xlen do the following
 7181   // jlong carry = 0;
 7182   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7183   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7184   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7185   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7186   //     carry = (jlong)product;
 7187   // }
 7188 
 7189   xorq(tmp5, tmp5);     // carry
 7190   xorq(rdxReg, rdxReg);
 7191   xorl(tmp1, tmp1);     // index for x
 7192   xorl(tmp4, tmp4);     // index for z
 7193 
 7194   Label L_first_loop, L_first_loop_exit;
 7195 
 7196   testl(xlen, 1);
 7197   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7198 
 7199   // Square and right shift by 1 the odd element using 32 bit multiply
 7200   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7201   imulq(raxReg, raxReg);
 7202   shrq(raxReg, 1);
 7203   adcq(tmp5, 0);
 7204   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7205   incrementl(tmp1);
 7206   addl(tmp4, 2);
 7207 
 7208   // Square and  right shift by 1 the rest using 64 bit multiply
 7209   bind(L_first_loop);
 7210   cmpptr(tmp1, xlen);
 7211   jccb(Assembler::equal, L_first_loop_exit);
 7212 
 7213   // Square
 7214   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7215   rorq(raxReg, 32);    // convert big-endian to little-endian
 7216   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7217 
 7218   // Right shift by 1 and save carry
 7219   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7220   rcrq(rdxReg, 1);
 7221   rcrq(raxReg, 1);
 7222   adcq(tmp5, 0);
 7223 
 7224   // Store result in z
 7225   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7226   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7227 
 7228   // Update indices for x and z
 7229   addl(tmp1, 2);
 7230   addl(tmp4, 4);
 7231   jmp(L_first_loop);
 7232 
 7233   bind(L_first_loop_exit);
 7234 }
 7235 
 7236 
 7237 /**
 7238  * Perform the following multiply add operation using BMI2 instructions
 7239  * carry:sum = sum + op1*op2 + carry
 7240  * op2 should be in rdx
 7241  * op2 is preserved, all other registers are modified
 7242  */
 7243 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7244   // assert op2 is rdx
 7245   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7246   addq(sum, carry);
 7247   adcq(tmp2, 0);
 7248   addq(sum, op1);
 7249   adcq(tmp2, 0);
 7250   movq(carry, tmp2);
 7251 }
 7252 
 7253 /**
 7254  * Perform the following multiply add operation:
 7255  * carry:sum = sum + op1*op2 + carry
 7256  * Preserves op1, op2 and modifies rest of registers
 7257  */
 7258 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7259   // rdx:rax = op1 * op2
 7260   movq(raxReg, op2);
 7261   mulq(op1);
 7262 
 7263   //  rdx:rax = sum + carry + rdx:rax
 7264   addq(sum, carry);
 7265   adcq(rdxReg, 0);
 7266   addq(sum, raxReg);
 7267   adcq(rdxReg, 0);
 7268 
 7269   // carry:sum = rdx:sum
 7270   movq(carry, rdxReg);
 7271 }
 7272 
 7273 /**
 7274  * Add 64 bit long carry into z[] with carry propagation.
 7275  * Preserves z and carry register values and modifies rest of registers.
 7276  *
 7277  */
 7278 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7279   Label L_fourth_loop, L_fourth_loop_exit;
 7280 
 7281   movl(tmp1, 1);
 7282   subl(zlen, 2);
 7283   addq(Address(z, zlen, Address::times_4, 0), carry);
 7284 
 7285   bind(L_fourth_loop);
 7286   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7287   subl(zlen, 2);
 7288   jccb(Assembler::negative, L_fourth_loop_exit);
 7289   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7290   jmp(L_fourth_loop);
 7291   bind(L_fourth_loop_exit);
 7292 }
 7293 
 7294 /**
 7295  * Shift z[] left by 1 bit.
 7296  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7297  *
 7298  */
 7299 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7300 
 7301   Label L_fifth_loop, L_fifth_loop_exit;
 7302 
 7303   // Fifth loop
 7304   // Perform primitiveLeftShift(z, zlen, 1)
 7305 
 7306   const Register prev_carry = tmp1;
 7307   const Register new_carry = tmp4;
 7308   const Register value = tmp2;
 7309   const Register zidx = tmp3;
 7310 
 7311   // int zidx, carry;
 7312   // long value;
 7313   // carry = 0;
 7314   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 7315   //    (carry:value)  = (z[i] << 1) | carry ;
 7316   //    z[i] = value;
 7317   // }
 7318 
 7319   movl(zidx, zlen);
 7320   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 7321 
 7322   bind(L_fifth_loop);
 7323   decl(zidx);  // Use decl to preserve carry flag
 7324   decl(zidx);
 7325   jccb(Assembler::negative, L_fifth_loop_exit);
 7326 
 7327   if (UseBMI2Instructions) {
 7328      movq(value, Address(z, zidx, Address::times_4, 0));
 7329      rclq(value, 1);
 7330      rorxq(value, value, 32);
 7331      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7332   }
 7333   else {
 7334     // clear new_carry
 7335     xorl(new_carry, new_carry);
 7336 
 7337     // Shift z[i] by 1, or in previous carry and save new carry
 7338     movq(value, Address(z, zidx, Address::times_4, 0));
 7339     shlq(value, 1);
 7340     adcl(new_carry, 0);
 7341 
 7342     orq(value, prev_carry);
 7343     rorq(value, 0x20);
 7344     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7345 
 7346     // Set previous carry = new carry
 7347     movl(prev_carry, new_carry);
 7348   }
 7349   jmp(L_fifth_loop);
 7350 
 7351   bind(L_fifth_loop_exit);
 7352 }
 7353 
 7354 
 7355 /**
 7356  * Code for BigInteger::squareToLen() intrinsic
 7357  *
 7358  * rdi: x
 7359  * rsi: len
 7360  * r8:  z
 7361  * rcx: zlen
 7362  * r12: tmp1
 7363  * r13: tmp2
 7364  * r14: tmp3
 7365  * r15: tmp4
 7366  * rbx: tmp5
 7367  *
 7368  */
 7369 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) {
 7370 
 7371   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 7372   push(tmp1);
 7373   push(tmp2);
 7374   push(tmp3);
 7375   push(tmp4);
 7376   push(tmp5);
 7377 
 7378   // First loop
 7379   // Store the squares, right shifted one bit (i.e., divided by 2).
 7380   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7381 
 7382   // Add in off-diagonal sums.
 7383   //
 7384   // Second, third (nested) and fourth loops.
 7385   // zlen +=2;
 7386   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 7387   //    carry = 0;
 7388   //    long op2 = x[xidx:xidx+1];
 7389   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 7390   //       k -= 2;
 7391   //       long op1 = x[j:j+1];
 7392   //       long sum = z[k:k+1];
 7393   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 7394   //       z[k:k+1] = sum;
 7395   //    }
 7396   //    add_one_64(z, k, carry, tmp_regs);
 7397   // }
 7398 
 7399   const Register carry = tmp5;
 7400   const Register sum = tmp3;
 7401   const Register op1 = tmp4;
 7402   Register op2 = tmp2;
 7403 
 7404   push(zlen);
 7405   push(len);
 7406   addl(zlen,2);
 7407   bind(L_second_loop);
 7408   xorq(carry, carry);
 7409   subl(zlen, 4);
 7410   subl(len, 2);
 7411   push(zlen);
 7412   push(len);
 7413   cmpl(len, 0);
 7414   jccb(Assembler::lessEqual, L_second_loop_exit);
 7415 
 7416   // Multiply an array by one 64 bit long.
 7417   if (UseBMI2Instructions) {
 7418     op2 = rdxReg;
 7419     movq(op2, Address(x, len, Address::times_4,  0));
 7420     rorxq(op2, op2, 32);
 7421   }
 7422   else {
 7423     movq(op2, Address(x, len, Address::times_4,  0));
 7424     rorq(op2, 32);
 7425   }
 7426 
 7427   bind(L_third_loop);
 7428   decrementl(len);
 7429   jccb(Assembler::negative, L_third_loop_exit);
 7430   decrementl(len);
 7431   jccb(Assembler::negative, L_last_x);
 7432 
 7433   movq(op1, Address(x, len, Address::times_4,  0));
 7434   rorq(op1, 32);
 7435 
 7436   bind(L_multiply);
 7437   subl(zlen, 2);
 7438   movq(sum, Address(z, zlen, Address::times_4,  0));
 7439 
 7440   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 7441   if (UseBMI2Instructions) {
 7442     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 7443   }
 7444   else {
 7445     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7446   }
 7447 
 7448   movq(Address(z, zlen, Address::times_4, 0), sum);
 7449 
 7450   jmp(L_third_loop);
 7451   bind(L_third_loop_exit);
 7452 
 7453   // Fourth loop
 7454   // Add 64 bit long carry into z with carry propagation.
 7455   // Uses offsetted zlen.
 7456   add_one_64(z, zlen, carry, tmp1);
 7457 
 7458   pop(len);
 7459   pop(zlen);
 7460   jmp(L_second_loop);
 7461 
 7462   // Next infrequent code is moved outside loops.
 7463   bind(L_last_x);
 7464   movl(op1, Address(x, 0));
 7465   jmp(L_multiply);
 7466 
 7467   bind(L_second_loop_exit);
 7468   pop(len);
 7469   pop(zlen);
 7470   pop(len);
 7471   pop(zlen);
 7472 
 7473   // Fifth loop
 7474   // Shift z left 1 bit.
 7475   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 7476 
 7477   // z[zlen-1] |= x[len-1] & 1;
 7478   movl(tmp3, Address(x, len, Address::times_4, -4));
 7479   andl(tmp3, 1);
 7480   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 7481 
 7482   pop(tmp5);
 7483   pop(tmp4);
 7484   pop(tmp3);
 7485   pop(tmp2);
 7486   pop(tmp1);
 7487 }
 7488 
 7489 /**
 7490  * Helper function for mul_add()
 7491  * Multiply the in[] by int k and add to out[] starting at offset offs using
 7492  * 128 bit by 32 bit multiply and return the carry in tmp5.
 7493  * Only quad int aligned length of in[] is operated on in this function.
 7494  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 7495  * This function preserves out, in and k registers.
 7496  * len and offset point to the appropriate index in "in" & "out" correspondingly
 7497  * tmp5 has the carry.
 7498  * other registers are temporary and are modified.
 7499  *
 7500  */
 7501 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 7502   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 7503   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7504 
 7505   Label L_first_loop, L_first_loop_exit;
 7506 
 7507   movl(tmp1, len);
 7508   shrl(tmp1, 2);
 7509 
 7510   bind(L_first_loop);
 7511   subl(tmp1, 1);
 7512   jccb(Assembler::negative, L_first_loop_exit);
 7513 
 7514   subl(len, 4);
 7515   subl(offset, 4);
 7516 
 7517   Register op2 = tmp2;
 7518   const Register sum = tmp3;
 7519   const Register op1 = tmp4;
 7520   const Register carry = tmp5;
 7521 
 7522   if (UseBMI2Instructions) {
 7523     op2 = rdxReg;
 7524   }
 7525 
 7526   movq(op1, Address(in, len, Address::times_4,  8));
 7527   rorq(op1, 32);
 7528   movq(sum, Address(out, offset, Address::times_4,  8));
 7529   rorq(sum, 32);
 7530   if (UseBMI2Instructions) {
 7531     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7532   }
 7533   else {
 7534     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7535   }
 7536   // Store back in big endian from little endian
 7537   rorq(sum, 0x20);
 7538   movq(Address(out, offset, Address::times_4,  8), sum);
 7539 
 7540   movq(op1, Address(in, len, Address::times_4,  0));
 7541   rorq(op1, 32);
 7542   movq(sum, Address(out, offset, Address::times_4,  0));
 7543   rorq(sum, 32);
 7544   if (UseBMI2Instructions) {
 7545     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7546   }
 7547   else {
 7548     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7549   }
 7550   // Store back in big endian from little endian
 7551   rorq(sum, 0x20);
 7552   movq(Address(out, offset, Address::times_4,  0), sum);
 7553 
 7554   jmp(L_first_loop);
 7555   bind(L_first_loop_exit);
 7556 }
 7557 
 7558 /**
 7559  * Code for BigInteger::mulAdd() intrinsic
 7560  *
 7561  * rdi: out
 7562  * rsi: in
 7563  * r11: offs (out.length - offset)
 7564  * rcx: len
 7565  * r8:  k
 7566  * r12: tmp1
 7567  * r13: tmp2
 7568  * r14: tmp3
 7569  * r15: tmp4
 7570  * rbx: tmp5
 7571  * Multiply the in[] by word k and add to out[], return the carry in rax
 7572  */
 7573 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 7574    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 7575    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7576 
 7577   Label L_carry, L_last_in, L_done;
 7578 
 7579 // carry = 0;
 7580 // for (int j=len-1; j >= 0; j--) {
 7581 //    long product = (in[j] & LONG_MASK) * kLong +
 7582 //                   (out[offs] & LONG_MASK) + carry;
 7583 //    out[offs--] = (int)product;
 7584 //    carry = product >>> 32;
 7585 // }
 7586 //
 7587   push(tmp1);
 7588   push(tmp2);
 7589   push(tmp3);
 7590   push(tmp4);
 7591   push(tmp5);
 7592 
 7593   Register op2 = tmp2;
 7594   const Register sum = tmp3;
 7595   const Register op1 = tmp4;
 7596   const Register carry =  tmp5;
 7597 
 7598   if (UseBMI2Instructions) {
 7599     op2 = rdxReg;
 7600     movl(op2, k);
 7601   }
 7602   else {
 7603     movl(op2, k);
 7604   }
 7605 
 7606   xorq(carry, carry);
 7607 
 7608   //First loop
 7609 
 7610   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 7611   //The carry is in tmp5
 7612   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7613 
 7614   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 7615   decrementl(len);
 7616   jccb(Assembler::negative, L_carry);
 7617   decrementl(len);
 7618   jccb(Assembler::negative, L_last_in);
 7619 
 7620   movq(op1, Address(in, len, Address::times_4,  0));
 7621   rorq(op1, 32);
 7622 
 7623   subl(offs, 2);
 7624   movq(sum, Address(out, offs, Address::times_4,  0));
 7625   rorq(sum, 32);
 7626 
 7627   if (UseBMI2Instructions) {
 7628     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7629   }
 7630   else {
 7631     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7632   }
 7633 
 7634   // Store back in big endian from little endian
 7635   rorq(sum, 0x20);
 7636   movq(Address(out, offs, Address::times_4,  0), sum);
 7637 
 7638   testl(len, len);
 7639   jccb(Assembler::zero, L_carry);
 7640 
 7641   //Multiply the last in[] entry, if any
 7642   bind(L_last_in);
 7643   movl(op1, Address(in, 0));
 7644   movl(sum, Address(out, offs, Address::times_4,  -4));
 7645 
 7646   movl(raxReg, k);
 7647   mull(op1); //tmp4 * eax -> edx:eax
 7648   addl(sum, carry);
 7649   adcl(rdxReg, 0);
 7650   addl(sum, raxReg);
 7651   adcl(rdxReg, 0);
 7652   movl(carry, rdxReg);
 7653 
 7654   movl(Address(out, offs, Address::times_4,  -4), sum);
 7655 
 7656   bind(L_carry);
 7657   //return tmp5/carry as carry in rax
 7658   movl(rax, carry);
 7659 
 7660   bind(L_done);
 7661   pop(tmp5);
 7662   pop(tmp4);
 7663   pop(tmp3);
 7664   pop(tmp2);
 7665   pop(tmp1);
 7666 }
 7667 
 7668 /**
 7669  * Emits code to update CRC-32 with a byte value according to constants in table
 7670  *
 7671  * @param [in,out]crc   Register containing the crc.
 7672  * @param [in]val       Register containing the byte to fold into the CRC.
 7673  * @param [in]table     Register containing the table of crc constants.
 7674  *
 7675  * uint32_t crc;
 7676  * val = crc_table[(val ^ crc) & 0xFF];
 7677  * crc = val ^ (crc >> 8);
 7678  *
 7679  */
 7680 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 7681   xorl(val, crc);
 7682   andl(val, 0xFF);
 7683   shrl(crc, 8); // unsigned shift
 7684   xorl(crc, Address(table, val, Address::times_4, 0));
 7685 }
 7686 
 7687 /**
 7688  * Fold 128-bit data chunk
 7689  */
 7690 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 7691   if (UseAVX > 0) {
 7692     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 7693     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 7694     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 7695     pxor(xcrc, xtmp);
 7696   } else {
 7697     movdqa(xtmp, xcrc);
 7698     pclmulhdq(xtmp, xK);   // [123:64]
 7699     pclmulldq(xcrc, xK);   // [63:0]
 7700     pxor(xcrc, xtmp);
 7701     movdqu(xtmp, Address(buf, offset));
 7702     pxor(xcrc, xtmp);
 7703   }
 7704 }
 7705 
 7706 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 7707   if (UseAVX > 0) {
 7708     vpclmulhdq(xtmp, xK, xcrc);
 7709     vpclmulldq(xcrc, xK, xcrc);
 7710     pxor(xcrc, xbuf);
 7711     pxor(xcrc, xtmp);
 7712   } else {
 7713     movdqa(xtmp, xcrc);
 7714     pclmulhdq(xtmp, xK);
 7715     pclmulldq(xcrc, xK);
 7716     pxor(xcrc, xbuf);
 7717     pxor(xcrc, xtmp);
 7718   }
 7719 }
 7720 
 7721 /**
 7722  * 8-bit folds to compute 32-bit CRC
 7723  *
 7724  * uint64_t xcrc;
 7725  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 7726  */
 7727 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 7728   movdl(tmp, xcrc);
 7729   andl(tmp, 0xFF);
 7730   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 7731   psrldq(xcrc, 1); // unsigned shift one byte
 7732   pxor(xcrc, xtmp);
 7733 }
 7734 
 7735 /**
 7736  * uint32_t crc;
 7737  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 7738  */
 7739 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 7740   movl(tmp, crc);
 7741   andl(tmp, 0xFF);
 7742   shrl(crc, 8);
 7743   xorl(crc, Address(table, tmp, Address::times_4, 0));
 7744 }
 7745 
 7746 /**
 7747  * @param crc   register containing existing CRC (32-bit)
 7748  * @param buf   register pointing to input byte buffer (byte*)
 7749  * @param len   register containing number of bytes
 7750  * @param table register that will contain address of CRC table
 7751  * @param tmp   scratch register
 7752  */
 7753 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 7754   assert_different_registers(crc, buf, len, table, tmp, rax);
 7755 
 7756   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 7757   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 7758 
 7759   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 7760   // context for the registers used, where all instructions below are using 128-bit mode
 7761   // On EVEX without VL and BW, these instructions will all be AVX.
 7762   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 7763   notl(crc); // ~crc
 7764   cmpl(len, 16);
 7765   jcc(Assembler::less, L_tail);
 7766 
 7767   // Align buffer to 16 bytes
 7768   movl(tmp, buf);
 7769   andl(tmp, 0xF);
 7770   jccb(Assembler::zero, L_aligned);
 7771   subl(tmp,  16);
 7772   addl(len, tmp);
 7773 
 7774   align(4);
 7775   BIND(L_align_loop);
 7776   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7777   update_byte_crc32(crc, rax, table);
 7778   increment(buf);
 7779   incrementl(tmp);
 7780   jccb(Assembler::less, L_align_loop);
 7781 
 7782   BIND(L_aligned);
 7783   movl(tmp, len); // save
 7784   shrl(len, 4);
 7785   jcc(Assembler::zero, L_tail_restore);
 7786 
 7787   // Fold crc into first bytes of vector
 7788   movdqa(xmm1, Address(buf, 0));
 7789   movdl(rax, xmm1);
 7790   xorl(crc, rax);
 7791   if (VM_Version::supports_sse4_1()) {
 7792     pinsrd(xmm1, crc, 0);
 7793   } else {
 7794     pinsrw(xmm1, crc, 0);
 7795     shrl(crc, 16);
 7796     pinsrw(xmm1, crc, 1);
 7797   }
 7798   addptr(buf, 16);
 7799   subl(len, 4); // len > 0
 7800   jcc(Assembler::less, L_fold_tail);
 7801 
 7802   movdqa(xmm2, Address(buf,  0));
 7803   movdqa(xmm3, Address(buf, 16));
 7804   movdqa(xmm4, Address(buf, 32));
 7805   addptr(buf, 48);
 7806   subl(len, 3);
 7807   jcc(Assembler::lessEqual, L_fold_512b);
 7808 
 7809   // Fold total 512 bits of polynomial on each iteration,
 7810   // 128 bits per each of 4 parallel streams.
 7811   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 7812 
 7813   align32();
 7814   BIND(L_fold_512b_loop);
 7815   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7816   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 7817   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 7818   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 7819   addptr(buf, 64);
 7820   subl(len, 4);
 7821   jcc(Assembler::greater, L_fold_512b_loop);
 7822 
 7823   // Fold 512 bits to 128 bits.
 7824   BIND(L_fold_512b);
 7825   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7826   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 7827   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 7828   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 7829 
 7830   // Fold the rest of 128 bits data chunks
 7831   BIND(L_fold_tail);
 7832   addl(len, 3);
 7833   jccb(Assembler::lessEqual, L_fold_128b);
 7834   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7835 
 7836   BIND(L_fold_tail_loop);
 7837   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7838   addptr(buf, 16);
 7839   decrementl(len);
 7840   jccb(Assembler::greater, L_fold_tail_loop);
 7841 
 7842   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 7843   BIND(L_fold_128b);
 7844   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 7845   if (UseAVX > 0) {
 7846     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 7847     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 7848     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 7849   } else {
 7850     movdqa(xmm2, xmm0);
 7851     pclmulqdq(xmm2, xmm1, 0x1);
 7852     movdqa(xmm3, xmm0);
 7853     pand(xmm3, xmm2);
 7854     pclmulqdq(xmm0, xmm3, 0x1);
 7855   }
 7856   psrldq(xmm1, 8);
 7857   psrldq(xmm2, 4);
 7858   pxor(xmm0, xmm1);
 7859   pxor(xmm0, xmm2);
 7860 
 7861   // 8 8-bit folds to compute 32-bit CRC.
 7862   for (int j = 0; j < 4; j++) {
 7863     fold_8bit_crc32(xmm0, table, xmm1, rax);
 7864   }
 7865   movdl(crc, xmm0); // mov 32 bits to general register
 7866   for (int j = 0; j < 4; j++) {
 7867     fold_8bit_crc32(crc, table, rax);
 7868   }
 7869 
 7870   BIND(L_tail_restore);
 7871   movl(len, tmp); // restore
 7872   BIND(L_tail);
 7873   andl(len, 0xf);
 7874   jccb(Assembler::zero, L_exit);
 7875 
 7876   // Fold the rest of bytes
 7877   align(4);
 7878   BIND(L_tail_loop);
 7879   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7880   update_byte_crc32(crc, rax, table);
 7881   increment(buf);
 7882   decrementl(len);
 7883   jccb(Assembler::greater, L_tail_loop);
 7884 
 7885   BIND(L_exit);
 7886   notl(crc); // ~c
 7887 }
 7888 
 7889 // Helper function for AVX 512 CRC32
 7890 // Fold 512-bit data chunks
 7891 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 7892                                              Register pos, int offset) {
 7893   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 7894   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 7895   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 7896   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 7897   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 7898 }
 7899 
 7900 // Helper function for AVX 512 CRC32
 7901 // Compute CRC32 for < 256B buffers
 7902 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 7903                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 7904                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 7905 
 7906   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 7907   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 7908   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 7909 
 7910   // check if there is enough buffer to be able to fold 16B at a time
 7911   cmpl(len, 32);
 7912   jcc(Assembler::less, L_less_than_32);
 7913 
 7914   // if there is, load the constants
 7915   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 7916   movdl(xmm0, crc);                        // get the initial crc value
 7917   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7918   pxor(xmm7, xmm0);
 7919 
 7920   // update the buffer pointer
 7921   addl(pos, 16);
 7922   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 7923   subl(len, 32);
 7924   jmp(L_16B_reduction_loop);
 7925 
 7926   bind(L_less_than_32);
 7927   //mov initial crc to the return value. this is necessary for zero - length buffers.
 7928   movl(rax, crc);
 7929   testl(len, len);
 7930   jcc(Assembler::equal, L_cleanup);
 7931 
 7932   movdl(xmm0, crc);                        //get the initial crc value
 7933 
 7934   cmpl(len, 16);
 7935   jcc(Assembler::equal, L_exact_16_left);
 7936   jcc(Assembler::less, L_less_than_16_left);
 7937 
 7938   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7939   pxor(xmm7, xmm0);                       //xor the initial crc value
 7940   addl(pos, 16);
 7941   subl(len, 16);
 7942   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 7943   jmp(L_get_last_two_xmms);
 7944 
 7945   bind(L_less_than_16_left);
 7946   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 7947   pxor(xmm1, xmm1);
 7948   movptr(tmp1, rsp);
 7949   movdqu(Address(tmp1, 0 * 16), xmm1);
 7950 
 7951   cmpl(len, 4);
 7952   jcc(Assembler::less, L_only_less_than_4);
 7953 
 7954   //backup the counter value
 7955   movl(tmp2, len);
 7956   cmpl(len, 8);
 7957   jcc(Assembler::less, L_less_than_8_left);
 7958 
 7959   //load 8 Bytes
 7960   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 7961   movq(Address(tmp1, 0 * 16), rax);
 7962   addptr(tmp1, 8);
 7963   subl(len, 8);
 7964   addl(pos, 8);
 7965 
 7966   bind(L_less_than_8_left);
 7967   cmpl(len, 4);
 7968   jcc(Assembler::less, L_less_than_4_left);
 7969 
 7970   //load 4 Bytes
 7971   movl(rax, Address(buf, pos, Address::times_1, 0));
 7972   movl(Address(tmp1, 0 * 16), rax);
 7973   addptr(tmp1, 4);
 7974   subl(len, 4);
 7975   addl(pos, 4);
 7976 
 7977   bind(L_less_than_4_left);
 7978   cmpl(len, 2);
 7979   jcc(Assembler::less, L_less_than_2_left);
 7980 
 7981   // load 2 Bytes
 7982   movw(rax, Address(buf, pos, Address::times_1, 0));
 7983   movl(Address(tmp1, 0 * 16), rax);
 7984   addptr(tmp1, 2);
 7985   subl(len, 2);
 7986   addl(pos, 2);
 7987 
 7988   bind(L_less_than_2_left);
 7989   cmpl(len, 1);
 7990   jcc(Assembler::less, L_zero_left);
 7991 
 7992   // load 1 Byte
 7993   movb(rax, Address(buf, pos, Address::times_1, 0));
 7994   movb(Address(tmp1, 0 * 16), rax);
 7995 
 7996   bind(L_zero_left);
 7997   movdqu(xmm7, Address(rsp, 0));
 7998   pxor(xmm7, xmm0);                       //xor the initial crc value
 7999 
 8000   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8001   movdqu(xmm0, Address(rax, tmp2));
 8002   pshufb(xmm7, xmm0);
 8003   jmp(L_128_done);
 8004 
 8005   bind(L_exact_16_left);
 8006   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 8007   pxor(xmm7, xmm0);                       //xor the initial crc value
 8008   jmp(L_128_done);
 8009 
 8010   bind(L_only_less_than_4);
 8011   cmpl(len, 3);
 8012   jcc(Assembler::less, L_only_less_than_3);
 8013 
 8014   // load 3 Bytes
 8015   movb(rax, Address(buf, pos, Address::times_1, 0));
 8016   movb(Address(tmp1, 0), rax);
 8017 
 8018   movb(rax, Address(buf, pos, Address::times_1, 1));
 8019   movb(Address(tmp1, 1), rax);
 8020 
 8021   movb(rax, Address(buf, pos, Address::times_1, 2));
 8022   movb(Address(tmp1, 2), rax);
 8023 
 8024   movdqu(xmm7, Address(rsp, 0));
 8025   pxor(xmm7, xmm0);                     //xor the initial crc value
 8026 
 8027   pslldq(xmm7, 0x5);
 8028   jmp(L_barrett);
 8029   bind(L_only_less_than_3);
 8030   cmpl(len, 2);
 8031   jcc(Assembler::less, L_only_less_than_2);
 8032 
 8033   // load 2 Bytes
 8034   movb(rax, Address(buf, pos, Address::times_1, 0));
 8035   movb(Address(tmp1, 0), rax);
 8036 
 8037   movb(rax, Address(buf, pos, Address::times_1, 1));
 8038   movb(Address(tmp1, 1), rax);
 8039 
 8040   movdqu(xmm7, Address(rsp, 0));
 8041   pxor(xmm7, xmm0);                     //xor the initial crc value
 8042 
 8043   pslldq(xmm7, 0x6);
 8044   jmp(L_barrett);
 8045 
 8046   bind(L_only_less_than_2);
 8047   //load 1 Byte
 8048   movb(rax, Address(buf, pos, Address::times_1, 0));
 8049   movb(Address(tmp1, 0), rax);
 8050 
 8051   movdqu(xmm7, Address(rsp, 0));
 8052   pxor(xmm7, xmm0);                     //xor the initial crc value
 8053 
 8054   pslldq(xmm7, 0x7);
 8055 }
 8056 
 8057 /**
 8058 * Compute CRC32 using AVX512 instructions
 8059 * param crc   register containing existing CRC (32-bit)
 8060 * param buf   register pointing to input byte buffer (byte*)
 8061 * param len   register containing number of bytes
 8062 * param table address of crc or crc32c table
 8063 * param tmp1  scratch register
 8064 * param tmp2  scratch register
 8065 * return rax  result register
 8066 *
 8067 * This routine is identical for crc32c with the exception of the precomputed constant
 8068 * table which will be passed as the table argument.  The calculation steps are
 8069 * the same for both variants.
 8070 */
 8071 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8072   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8073 
 8074   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8075   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8076   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8077   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8078   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8079 
 8080   const Register pos = r12;
 8081   push(r12);
 8082   subptr(rsp, 16 * 2 + 8);
 8083 
 8084   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8085   // context for the registers used, where all instructions below are using 128-bit mode
 8086   // On EVEX without VL and BW, these instructions will all be AVX.
 8087   movl(pos, 0);
 8088 
 8089   // check if smaller than 256B
 8090   cmpl(len, 256);
 8091   jcc(Assembler::less, L_less_than_256);
 8092 
 8093   // load the initial crc value
 8094   movdl(xmm10, crc);
 8095 
 8096   // receive the initial 64B data, xor the initial crc value
 8097   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8098   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8099   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8100   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8101 
 8102   subl(len, 256);
 8103   cmpl(len, 256);
 8104   jcc(Assembler::less, L_fold_128_B_loop);
 8105 
 8106   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8107   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8108   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8109   subl(len, 256);
 8110 
 8111   bind(L_fold_256_B_loop);
 8112   addl(pos, 256);
 8113   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8114   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8115   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8116   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8117 
 8118   subl(len, 256);
 8119   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8120 
 8121   // Fold 256 into 128
 8122   addl(pos, 256);
 8123   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8124   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8125   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8126 
 8127   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8128   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8129   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8130 
 8131   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8132   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8133 
 8134   addl(len, 128);
 8135   jmp(L_fold_128_B_register);
 8136 
 8137   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8138   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8139 
 8140   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8141   bind(L_fold_128_B_loop);
 8142   addl(pos, 128);
 8143   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8144   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8145 
 8146   subl(len, 128);
 8147   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8148 
 8149   addl(pos, 128);
 8150 
 8151   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8152   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8153   bind(L_fold_128_B_register);
 8154   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8155   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8156   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8157   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8158   // save last that has no multiplicand
 8159   vextracti64x2(xmm7, xmm4, 3);
 8160 
 8161   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8162   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8163   // Needed later in reduction loop
 8164   movdqu(xmm10, Address(table, 1 * 16));
 8165   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8166   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8167 
 8168   // Swap 1,0,3,2 - 01 00 11 10
 8169   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8170   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8171   vextracti128(xmm5, xmm8, 1);
 8172   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8173 
 8174   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8175   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8176   addl(len, 128 - 16);
 8177   jcc(Assembler::less, L_final_reduction_for_128);
 8178 
 8179   bind(L_16B_reduction_loop);
 8180   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8181   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8182   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8183   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8184   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8185   addl(pos, 16);
 8186   subl(len, 16);
 8187   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8188 
 8189   bind(L_final_reduction_for_128);
 8190   addl(len, 16);
 8191   jcc(Assembler::equal, L_128_done);
 8192 
 8193   bind(L_get_last_two_xmms);
 8194   movdqu(xmm2, xmm7);
 8195   addl(pos, len);
 8196   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8197   subl(pos, len);
 8198 
 8199   // get rid of the extra data that was loaded before
 8200   // load the shift constant
 8201   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8202   movdqu(xmm0, Address(rax, len));
 8203   addl(rax, len);
 8204 
 8205   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8206   //Change mask to 512
 8207   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8208   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8209 
 8210   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8211   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8212   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8213   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8214   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8215 
 8216   bind(L_128_done);
 8217   // compute crc of a 128-bit value
 8218   movdqu(xmm10, Address(table, 3 * 16));
 8219   movdqu(xmm0, xmm7);
 8220 
 8221   // 64b fold
 8222   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8223   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8224   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8225 
 8226   // 32b fold
 8227   movdqu(xmm0, xmm7);
 8228   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8229   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8230   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8231   jmp(L_barrett);
 8232 
 8233   bind(L_less_than_256);
 8234   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);
 8235 
 8236   //barrett reduction
 8237   bind(L_barrett);
 8238   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8239   movdqu(xmm1, xmm7);
 8240   movdqu(xmm2, xmm7);
 8241   movdqu(xmm10, Address(table, 4 * 16));
 8242 
 8243   pclmulqdq(xmm7, xmm10, 0x0);
 8244   pxor(xmm7, xmm2);
 8245   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8246   movdqu(xmm2, xmm7);
 8247   pclmulqdq(xmm7, xmm10, 0x10);
 8248   pxor(xmm7, xmm2);
 8249   pxor(xmm7, xmm1);
 8250   pextrd(crc, xmm7, 2);
 8251 
 8252   bind(L_cleanup);
 8253   addptr(rsp, 16 * 2 + 8);
 8254   pop(r12);
 8255 }
 8256 
 8257 // S. Gueron / Information Processing Letters 112 (2012) 184
 8258 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8259 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8260 // Output: the 64-bit carry-less product of B * CONST
 8261 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8262                                      Register tmp1, Register tmp2, Register tmp3) {
 8263   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8264   if (n > 0) {
 8265     addq(tmp3, n * 256 * 8);
 8266   }
 8267   //    Q1 = TABLEExt[n][B & 0xFF];
 8268   movl(tmp1, in);
 8269   andl(tmp1, 0x000000FF);
 8270   shll(tmp1, 3);
 8271   addq(tmp1, tmp3);
 8272   movq(tmp1, Address(tmp1, 0));
 8273 
 8274   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8275   movl(tmp2, in);
 8276   shrl(tmp2, 8);
 8277   andl(tmp2, 0x000000FF);
 8278   shll(tmp2, 3);
 8279   addq(tmp2, tmp3);
 8280   movq(tmp2, Address(tmp2, 0));
 8281 
 8282   shlq(tmp2, 8);
 8283   xorq(tmp1, tmp2);
 8284 
 8285   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8286   movl(tmp2, in);
 8287   shrl(tmp2, 16);
 8288   andl(tmp2, 0x000000FF);
 8289   shll(tmp2, 3);
 8290   addq(tmp2, tmp3);
 8291   movq(tmp2, Address(tmp2, 0));
 8292 
 8293   shlq(tmp2, 16);
 8294   xorq(tmp1, tmp2);
 8295 
 8296   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8297   shrl(in, 24);
 8298   andl(in, 0x000000FF);
 8299   shll(in, 3);
 8300   addq(in, tmp3);
 8301   movq(in, Address(in, 0));
 8302 
 8303   shlq(in, 24);
 8304   xorq(in, tmp1);
 8305   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8306 }
 8307 
 8308 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8309                                       Register in_out,
 8310                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8311                                       XMMRegister w_xtmp2,
 8312                                       Register tmp1,
 8313                                       Register n_tmp2, Register n_tmp3) {
 8314   if (is_pclmulqdq_supported) {
 8315     movdl(w_xtmp1, in_out); // modified blindly
 8316 
 8317     movl(tmp1, const_or_pre_comp_const_index);
 8318     movdl(w_xtmp2, tmp1);
 8319     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8320 
 8321     movdq(in_out, w_xtmp1);
 8322   } else {
 8323     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 8324   }
 8325 }
 8326 
 8327 // Recombination Alternative 2: No bit-reflections
 8328 // T1 = (CRC_A * U1) << 1
 8329 // T2 = (CRC_B * U2) << 1
 8330 // C1 = T1 >> 32
 8331 // C2 = T2 >> 32
 8332 // T1 = T1 & 0xFFFFFFFF
 8333 // T2 = T2 & 0xFFFFFFFF
 8334 // T1 = CRC32(0, T1)
 8335 // T2 = CRC32(0, T2)
 8336 // C1 = C1 ^ T1
 8337 // C2 = C2 ^ T2
 8338 // CRC = C1 ^ C2 ^ CRC_C
 8339 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,
 8340                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8341                                      Register tmp1, Register tmp2,
 8342                                      Register n_tmp3) {
 8343   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8344   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8345   shlq(in_out, 1);
 8346   movl(tmp1, in_out);
 8347   shrq(in_out, 32);
 8348   xorl(tmp2, tmp2);
 8349   crc32(tmp2, tmp1, 4);
 8350   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 8351   shlq(in1, 1);
 8352   movl(tmp1, in1);
 8353   shrq(in1, 32);
 8354   xorl(tmp2, tmp2);
 8355   crc32(tmp2, tmp1, 4);
 8356   xorl(in1, tmp2);
 8357   xorl(in_out, in1);
 8358   xorl(in_out, in2);
 8359 }
 8360 
 8361 // Set N to predefined value
 8362 // Subtract from a length of a buffer
 8363 // execute in a loop:
 8364 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 8365 // for i = 1 to N do
 8366 //  CRC_A = CRC32(CRC_A, A[i])
 8367 //  CRC_B = CRC32(CRC_B, B[i])
 8368 //  CRC_C = CRC32(CRC_C, C[i])
 8369 // end for
 8370 // Recombine
 8371 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,
 8372                                        Register in_out1, Register in_out2, Register in_out3,
 8373                                        Register tmp1, Register tmp2, Register tmp3,
 8374                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8375                                        Register tmp4, Register tmp5,
 8376                                        Register n_tmp6) {
 8377   Label L_processPartitions;
 8378   Label L_processPartition;
 8379   Label L_exit;
 8380 
 8381   bind(L_processPartitions);
 8382   cmpl(in_out1, 3 * size);
 8383   jcc(Assembler::less, L_exit);
 8384     xorl(tmp1, tmp1);
 8385     xorl(tmp2, tmp2);
 8386     movq(tmp3, in_out2);
 8387     addq(tmp3, size);
 8388 
 8389     bind(L_processPartition);
 8390       crc32(in_out3, Address(in_out2, 0), 8);
 8391       crc32(tmp1, Address(in_out2, size), 8);
 8392       crc32(tmp2, Address(in_out2, size * 2), 8);
 8393       addq(in_out2, 8);
 8394       cmpq(in_out2, tmp3);
 8395       jcc(Assembler::less, L_processPartition);
 8396     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 8397             w_xtmp1, w_xtmp2, w_xtmp3,
 8398             tmp4, tmp5,
 8399             n_tmp6);
 8400     addq(in_out2, 2 * size);
 8401     subl(in_out1, 3 * size);
 8402     jmp(L_processPartitions);
 8403 
 8404   bind(L_exit);
 8405 }
 8406 
 8407 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 8408 // Input: A buffer I of L bytes.
 8409 // Output: the CRC32C value of the buffer.
 8410 // Notations:
 8411 // Write L = 24N + r, with N = floor (L/24).
 8412 // r = L mod 24 (0 <= r < 24).
 8413 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 8414 // N quadwords, and R consists of r bytes.
 8415 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 8416 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 8417 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 8418 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 8419 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 8420                                           Register tmp1, Register tmp2, Register tmp3,
 8421                                           Register tmp4, Register tmp5, Register tmp6,
 8422                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8423                                           bool is_pclmulqdq_supported) {
 8424   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 8425   Label L_wordByWord;
 8426   Label L_byteByByteProlog;
 8427   Label L_byteByByte;
 8428   Label L_exit;
 8429 
 8430   if (is_pclmulqdq_supported ) {
 8431     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 8432     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 8433 
 8434     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 8435     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 8436 
 8437     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 8438     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 8439     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 8440   } else {
 8441     const_or_pre_comp_const_index[0] = 1;
 8442     const_or_pre_comp_const_index[1] = 0;
 8443 
 8444     const_or_pre_comp_const_index[2] = 3;
 8445     const_or_pre_comp_const_index[3] = 2;
 8446 
 8447     const_or_pre_comp_const_index[4] = 5;
 8448     const_or_pre_comp_const_index[5] = 4;
 8449    }
 8450   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 8451                     in2, in1, in_out,
 8452                     tmp1, tmp2, tmp3,
 8453                     w_xtmp1, w_xtmp2, w_xtmp3,
 8454                     tmp4, tmp5,
 8455                     tmp6);
 8456   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 8457                     in2, in1, in_out,
 8458                     tmp1, tmp2, tmp3,
 8459                     w_xtmp1, w_xtmp2, w_xtmp3,
 8460                     tmp4, tmp5,
 8461                     tmp6);
 8462   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 8463                     in2, in1, in_out,
 8464                     tmp1, tmp2, tmp3,
 8465                     w_xtmp1, w_xtmp2, w_xtmp3,
 8466                     tmp4, tmp5,
 8467                     tmp6);
 8468   movl(tmp1, in2);
 8469   andl(tmp1, 0x00000007);
 8470   negl(tmp1);
 8471   addl(tmp1, in2);
 8472   addq(tmp1, in1);
 8473 
 8474   cmpq(in1, tmp1);
 8475   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 8476   align(16);
 8477   BIND(L_wordByWord);
 8478     crc32(in_out, Address(in1, 0), 8);
 8479     addq(in1, 8);
 8480     cmpq(in1, tmp1);
 8481     jcc(Assembler::less, L_wordByWord);
 8482 
 8483   BIND(L_byteByByteProlog);
 8484   andl(in2, 0x00000007);
 8485   movl(tmp2, 1);
 8486 
 8487   cmpl(tmp2, in2);
 8488   jccb(Assembler::greater, L_exit);
 8489   BIND(L_byteByByte);
 8490     crc32(in_out, Address(in1, 0), 1);
 8491     incq(in1);
 8492     incl(tmp2);
 8493     cmpl(tmp2, in2);
 8494     jcc(Assembler::lessEqual, L_byteByByte);
 8495 
 8496   BIND(L_exit);
 8497 }
 8498 #undef BIND
 8499 #undef BLOCK_COMMENT
 8500 
 8501 // Compress char[] array to byte[].
 8502 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 8503 // Return the array length if every element in array can be encoded,
 8504 // otherwise, the index of first non-latin1 (> 0xff) character.
 8505 //   @IntrinsicCandidate
 8506 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 8507 //     for (int i = 0; i < len; i++) {
 8508 //       char c = src[srcOff];
 8509 //       if (c > 0xff) {
 8510 //           return i;  // return index of non-latin1 char
 8511 //       }
 8512 //       dst[dstOff] = (byte)c;
 8513 //       srcOff++;
 8514 //       dstOff++;
 8515 //     }
 8516 //     return len;
 8517 //   }
 8518 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 8519   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 8520   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 8521   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 8522   Label copy_chars_loop, done, reset_sp, copy_tail;
 8523 
 8524   // rsi: src
 8525   // rdi: dst
 8526   // rdx: len
 8527   // rcx: tmp5
 8528   // rax: result
 8529 
 8530   // rsi holds start addr of source char[] to be compressed
 8531   // rdi holds start addr of destination byte[]
 8532   // rdx holds length
 8533 
 8534   assert(len != result, "");
 8535 
 8536   // save length for return
 8537   movl(result, len);
 8538 
 8539   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 8540     VM_Version::supports_avx512vlbw() &&
 8541     VM_Version::supports_bmi2()) {
 8542 
 8543     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 8544 
 8545     // alignment
 8546     Label post_alignment;
 8547 
 8548     // if length of the string is less than 32, handle it the old fashioned way
 8549     testl(len, -32);
 8550     jcc(Assembler::zero, below_threshold);
 8551 
 8552     // First check whether a character is compressible ( <= 0xFF).
 8553     // Create mask to test for Unicode chars inside zmm vector
 8554     movl(tmp5, 0x00FF);
 8555     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 8556 
 8557     testl(len, -64);
 8558     jccb(Assembler::zero, post_alignment);
 8559 
 8560     movl(tmp5, dst);
 8561     andl(tmp5, (32 - 1));
 8562     negl(tmp5);
 8563     andl(tmp5, (32 - 1));
 8564 
 8565     // bail out when there is nothing to be done
 8566     testl(tmp5, 0xFFFFFFFF);
 8567     jccb(Assembler::zero, post_alignment);
 8568 
 8569     // ~(~0 << len), where len is the # of remaining elements to process
 8570     movl(len, 0xFFFFFFFF);
 8571     shlxl(len, len, tmp5);
 8572     notl(len);
 8573     kmovdl(mask2, len);
 8574     movl(len, result);
 8575 
 8576     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8577     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8578     ktestd(mask1, mask2);
 8579     jcc(Assembler::carryClear, copy_tail);
 8580 
 8581     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8582 
 8583     addptr(src, tmp5);
 8584     addptr(src, tmp5);
 8585     addptr(dst, tmp5);
 8586     subl(len, tmp5);
 8587 
 8588     bind(post_alignment);
 8589     // end of alignment
 8590 
 8591     movl(tmp5, len);
 8592     andl(tmp5, (32 - 1));    // tail count (in chars)
 8593     andl(len, ~(32 - 1));    // vector count (in chars)
 8594     jccb(Assembler::zero, copy_loop_tail);
 8595 
 8596     lea(src, Address(src, len, Address::times_2));
 8597     lea(dst, Address(dst, len, Address::times_1));
 8598     negptr(len);
 8599 
 8600     bind(copy_32_loop);
 8601     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 8602     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 8603     kortestdl(mask1, mask1);
 8604     jccb(Assembler::carryClear, reset_for_copy_tail);
 8605 
 8606     // All elements in current processed chunk are valid candidates for
 8607     // compression. Write a truncated byte elements to the memory.
 8608     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 8609     addptr(len, 32);
 8610     jccb(Assembler::notZero, copy_32_loop);
 8611 
 8612     bind(copy_loop_tail);
 8613     // bail out when there is nothing to be done
 8614     testl(tmp5, 0xFFFFFFFF);
 8615     jcc(Assembler::zero, done);
 8616 
 8617     movl(len, tmp5);
 8618 
 8619     // ~(~0 << len), where len is the # of remaining elements to process
 8620     movl(tmp5, 0xFFFFFFFF);
 8621     shlxl(tmp5, tmp5, len);
 8622     notl(tmp5);
 8623 
 8624     kmovdl(mask2, tmp5);
 8625 
 8626     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8627     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8628     ktestd(mask1, mask2);
 8629     jcc(Assembler::carryClear, copy_tail);
 8630 
 8631     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8632     jmp(done);
 8633 
 8634     bind(reset_for_copy_tail);
 8635     lea(src, Address(src, tmp5, Address::times_2));
 8636     lea(dst, Address(dst, tmp5, Address::times_1));
 8637     subptr(len, tmp5);
 8638     jmp(copy_chars_loop);
 8639 
 8640     bind(below_threshold);
 8641   }
 8642 
 8643   if (UseSSE42Intrinsics) {
 8644     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 8645 
 8646     // vectored compression
 8647     testl(len, 0xfffffff8);
 8648     jcc(Assembler::zero, copy_tail);
 8649 
 8650     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 8651     movdl(tmp1Reg, tmp5);
 8652     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 8653 
 8654     andl(len, 0xfffffff0);
 8655     jccb(Assembler::zero, copy_16);
 8656 
 8657     // compress 16 chars per iter
 8658     pxor(tmp4Reg, tmp4Reg);
 8659 
 8660     lea(src, Address(src, len, Address::times_2));
 8661     lea(dst, Address(dst, len, Address::times_1));
 8662     negptr(len);
 8663 
 8664     bind(copy_32_loop);
 8665     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 8666     por(tmp4Reg, tmp2Reg);
 8667     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 8668     por(tmp4Reg, tmp3Reg);
 8669     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 8670     jccb(Assembler::notZero, reset_for_copy_tail);
 8671     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 8672     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 8673     addptr(len, 16);
 8674     jccb(Assembler::notZero, copy_32_loop);
 8675 
 8676     // compress next vector of 8 chars (if any)
 8677     bind(copy_16);
 8678     // len = 0
 8679     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 8680     jccb(Assembler::zero, copy_tail_sse);
 8681 
 8682     pxor(tmp3Reg, tmp3Reg);
 8683 
 8684     movdqu(tmp2Reg, Address(src, 0));
 8685     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 8686     jccb(Assembler::notZero, reset_for_copy_tail);
 8687     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 8688     movq(Address(dst, 0), tmp2Reg);
 8689     addptr(src, 16);
 8690     addptr(dst, 8);
 8691     jmpb(copy_tail_sse);
 8692 
 8693     bind(reset_for_copy_tail);
 8694     movl(tmp5, result);
 8695     andl(tmp5, 0x0000000f);
 8696     lea(src, Address(src, tmp5, Address::times_2));
 8697     lea(dst, Address(dst, tmp5, Address::times_1));
 8698     subptr(len, tmp5);
 8699     jmpb(copy_chars_loop);
 8700 
 8701     bind(copy_tail_sse);
 8702     movl(len, result);
 8703     andl(len, 0x00000007);    // tail count (in chars)
 8704   }
 8705   // compress 1 char per iter
 8706   bind(copy_tail);
 8707   testl(len, len);
 8708   jccb(Assembler::zero, done);
 8709   lea(src, Address(src, len, Address::times_2));
 8710   lea(dst, Address(dst, len, Address::times_1));
 8711   negptr(len);
 8712 
 8713   bind(copy_chars_loop);
 8714   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 8715   testl(tmp5, 0xff00);      // check if Unicode char
 8716   jccb(Assembler::notZero, reset_sp);
 8717   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 8718   increment(len);
 8719   jccb(Assembler::notZero, copy_chars_loop);
 8720 
 8721   // add len then return (len will be zero if compress succeeded, otherwise negative)
 8722   bind(reset_sp);
 8723   addl(result, len);
 8724 
 8725   bind(done);
 8726 }
 8727 
 8728 // Inflate byte[] array to char[].
 8729 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 8730 //   @IntrinsicCandidate
 8731 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 8732 //     for (int i = 0; i < len; i++) {
 8733 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 8734 //     }
 8735 //   }
 8736 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 8737   XMMRegister tmp1, Register tmp2, KRegister mask) {
 8738   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 8739   // rsi: src
 8740   // rdi: dst
 8741   // rdx: len
 8742   // rcx: tmp2
 8743 
 8744   // rsi holds start addr of source byte[] to be inflated
 8745   // rdi holds start addr of destination char[]
 8746   // rdx holds length
 8747   assert_different_registers(src, dst, len, tmp2);
 8748   movl(tmp2, len);
 8749   if ((UseAVX > 2) && // AVX512
 8750     VM_Version::supports_avx512vlbw() &&
 8751     VM_Version::supports_bmi2()) {
 8752 
 8753     Label copy_32_loop, copy_tail;
 8754     Register tmp3_aliased = len;
 8755 
 8756     // if length of the string is less than 16, handle it in an old fashioned way
 8757     testl(len, -16);
 8758     jcc(Assembler::zero, below_threshold);
 8759 
 8760     testl(len, -1 * AVX3Threshold);
 8761     jcc(Assembler::zero, avx3_threshold);
 8762 
 8763     // In order to use only one arithmetic operation for the main loop we use
 8764     // this pre-calculation
 8765     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 8766     andl(len, -32);     // vector count
 8767     jccb(Assembler::zero, copy_tail);
 8768 
 8769     lea(src, Address(src, len, Address::times_1));
 8770     lea(dst, Address(dst, len, Address::times_2));
 8771     negptr(len);
 8772 
 8773 
 8774     // inflate 32 chars per iter
 8775     bind(copy_32_loop);
 8776     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 8777     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 8778     addptr(len, 32);
 8779     jcc(Assembler::notZero, copy_32_loop);
 8780 
 8781     bind(copy_tail);
 8782     // bail out when there is nothing to be done
 8783     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 8784     jcc(Assembler::zero, done);
 8785 
 8786     // ~(~0 << length), where length is the # of remaining elements to process
 8787     movl(tmp3_aliased, -1);
 8788     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 8789     notl(tmp3_aliased);
 8790     kmovdl(mask, tmp3_aliased);
 8791     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 8792     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 8793 
 8794     jmp(done);
 8795     bind(avx3_threshold);
 8796   }
 8797   if (UseSSE42Intrinsics) {
 8798     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 8799 
 8800     if (UseAVX > 1) {
 8801       andl(tmp2, (16 - 1));
 8802       andl(len, -16);
 8803       jccb(Assembler::zero, copy_new_tail);
 8804     } else {
 8805       andl(tmp2, 0x00000007);   // tail count (in chars)
 8806       andl(len, 0xfffffff8);    // vector count (in chars)
 8807       jccb(Assembler::zero, copy_tail);
 8808     }
 8809 
 8810     // vectored inflation
 8811     lea(src, Address(src, len, Address::times_1));
 8812     lea(dst, Address(dst, len, Address::times_2));
 8813     negptr(len);
 8814 
 8815     if (UseAVX > 1) {
 8816       bind(copy_16_loop);
 8817       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 8818       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 8819       addptr(len, 16);
 8820       jcc(Assembler::notZero, copy_16_loop);
 8821 
 8822       bind(below_threshold);
 8823       bind(copy_new_tail);
 8824       movl(len, tmp2);
 8825       andl(tmp2, 0x00000007);
 8826       andl(len, 0xFFFFFFF8);
 8827       jccb(Assembler::zero, copy_tail);
 8828 
 8829       pmovzxbw(tmp1, Address(src, 0));
 8830       movdqu(Address(dst, 0), tmp1);
 8831       addptr(src, 8);
 8832       addptr(dst, 2 * 8);
 8833 
 8834       jmp(copy_tail, true);
 8835     }
 8836 
 8837     // inflate 8 chars per iter
 8838     bind(copy_8_loop);
 8839     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 8840     movdqu(Address(dst, len, Address::times_2), tmp1);
 8841     addptr(len, 8);
 8842     jcc(Assembler::notZero, copy_8_loop);
 8843 
 8844     bind(copy_tail);
 8845     movl(len, tmp2);
 8846 
 8847     cmpl(len, 4);
 8848     jccb(Assembler::less, copy_bytes);
 8849 
 8850     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 8851     pmovzxbw(tmp1, tmp1);
 8852     movq(Address(dst, 0), tmp1);
 8853     subptr(len, 4);
 8854     addptr(src, 4);
 8855     addptr(dst, 8);
 8856 
 8857     bind(copy_bytes);
 8858   } else {
 8859     bind(below_threshold);
 8860   }
 8861 
 8862   testl(len, len);
 8863   jccb(Assembler::zero, done);
 8864   lea(src, Address(src, len, Address::times_1));
 8865   lea(dst, Address(dst, len, Address::times_2));
 8866   negptr(len);
 8867 
 8868   // inflate 1 char per iter
 8869   bind(copy_chars_loop);
 8870   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 8871   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 8872   increment(len);
 8873   jcc(Assembler::notZero, copy_chars_loop);
 8874 
 8875   bind(done);
 8876 }
 8877 
 8878 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 8879   switch(type) {
 8880     case T_BYTE:
 8881     case T_BOOLEAN:
 8882       evmovdqub(dst, kmask, src, merge, vector_len);
 8883       break;
 8884     case T_CHAR:
 8885     case T_SHORT:
 8886       evmovdquw(dst, kmask, src, merge, vector_len);
 8887       break;
 8888     case T_INT:
 8889     case T_FLOAT:
 8890       evmovdqul(dst, kmask, src, merge, vector_len);
 8891       break;
 8892     case T_LONG:
 8893     case T_DOUBLE:
 8894       evmovdquq(dst, kmask, src, merge, vector_len);
 8895       break;
 8896     default:
 8897       fatal("Unexpected type argument %s", type2name(type));
 8898       break;
 8899   }
 8900 }
 8901 
 8902 
 8903 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 8904   switch(type) {
 8905     case T_BYTE:
 8906     case T_BOOLEAN:
 8907       evmovdqub(dst, kmask, src, merge, vector_len);
 8908       break;
 8909     case T_CHAR:
 8910     case T_SHORT:
 8911       evmovdquw(dst, kmask, src, merge, vector_len);
 8912       break;
 8913     case T_INT:
 8914     case T_FLOAT:
 8915       evmovdqul(dst, kmask, src, merge, vector_len);
 8916       break;
 8917     case T_LONG:
 8918     case T_DOUBLE:
 8919       evmovdquq(dst, kmask, src, merge, vector_len);
 8920       break;
 8921     default:
 8922       fatal("Unexpected type argument %s", type2name(type));
 8923       break;
 8924   }
 8925 }
 8926 
 8927 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 8928   switch(type) {
 8929     case T_BYTE:
 8930     case T_BOOLEAN:
 8931       evmovdqub(dst, kmask, src, merge, vector_len);
 8932       break;
 8933     case T_CHAR:
 8934     case T_SHORT:
 8935       evmovdquw(dst, kmask, src, merge, vector_len);
 8936       break;
 8937     case T_INT:
 8938     case T_FLOAT:
 8939       evmovdqul(dst, kmask, src, merge, vector_len);
 8940       break;
 8941     case T_LONG:
 8942     case T_DOUBLE:
 8943       evmovdquq(dst, kmask, src, merge, vector_len);
 8944       break;
 8945     default:
 8946       fatal("Unexpected type argument %s", type2name(type));
 8947       break;
 8948   }
 8949 }
 8950 
 8951 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 8952   switch(masklen) {
 8953     case 2:
 8954        knotbl(dst, src);
 8955        movl(rtmp, 3);
 8956        kmovbl(ktmp, rtmp);
 8957        kandbl(dst, ktmp, dst);
 8958        break;
 8959     case 4:
 8960        knotbl(dst, src);
 8961        movl(rtmp, 15);
 8962        kmovbl(ktmp, rtmp);
 8963        kandbl(dst, ktmp, dst);
 8964        break;
 8965     case 8:
 8966        knotbl(dst, src);
 8967        break;
 8968     case 16:
 8969        knotwl(dst, src);
 8970        break;
 8971     case 32:
 8972        knotdl(dst, src);
 8973        break;
 8974     case 64:
 8975        knotql(dst, src);
 8976        break;
 8977     default:
 8978       fatal("Unexpected vector length %d", masklen);
 8979       break;
 8980   }
 8981 }
 8982 
 8983 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8984   switch(type) {
 8985     case T_BOOLEAN:
 8986     case T_BYTE:
 8987        kandbl(dst, src1, src2);
 8988        break;
 8989     case T_CHAR:
 8990     case T_SHORT:
 8991        kandwl(dst, src1, src2);
 8992        break;
 8993     case T_INT:
 8994     case T_FLOAT:
 8995        kanddl(dst, src1, src2);
 8996        break;
 8997     case T_LONG:
 8998     case T_DOUBLE:
 8999        kandql(dst, src1, src2);
 9000        break;
 9001     default:
 9002       fatal("Unexpected type argument %s", type2name(type));
 9003       break;
 9004   }
 9005 }
 9006 
 9007 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9008   switch(type) {
 9009     case T_BOOLEAN:
 9010     case T_BYTE:
 9011        korbl(dst, src1, src2);
 9012        break;
 9013     case T_CHAR:
 9014     case T_SHORT:
 9015        korwl(dst, src1, src2);
 9016        break;
 9017     case T_INT:
 9018     case T_FLOAT:
 9019        kordl(dst, src1, src2);
 9020        break;
 9021     case T_LONG:
 9022     case T_DOUBLE:
 9023        korql(dst, src1, src2);
 9024        break;
 9025     default:
 9026       fatal("Unexpected type argument %s", type2name(type));
 9027       break;
 9028   }
 9029 }
 9030 
 9031 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9032   switch(type) {
 9033     case T_BOOLEAN:
 9034     case T_BYTE:
 9035        kxorbl(dst, src1, src2);
 9036        break;
 9037     case T_CHAR:
 9038     case T_SHORT:
 9039        kxorwl(dst, src1, src2);
 9040        break;
 9041     case T_INT:
 9042     case T_FLOAT:
 9043        kxordl(dst, src1, src2);
 9044        break;
 9045     case T_LONG:
 9046     case T_DOUBLE:
 9047        kxorql(dst, src1, src2);
 9048        break;
 9049     default:
 9050       fatal("Unexpected type argument %s", type2name(type));
 9051       break;
 9052   }
 9053 }
 9054 
 9055 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9056   switch(type) {
 9057     case T_BOOLEAN:
 9058     case T_BYTE:
 9059       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9060     case T_CHAR:
 9061     case T_SHORT:
 9062       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9063     case T_INT:
 9064     case T_FLOAT:
 9065       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9066     case T_LONG:
 9067     case T_DOUBLE:
 9068       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9069     default:
 9070       fatal("Unexpected type argument %s", type2name(type)); break;
 9071   }
 9072 }
 9073 
 9074 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9075   switch(type) {
 9076     case T_BOOLEAN:
 9077     case T_BYTE:
 9078       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9079     case T_CHAR:
 9080     case T_SHORT:
 9081       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9082     case T_INT:
 9083     case T_FLOAT:
 9084       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9085     case T_LONG:
 9086     case T_DOUBLE:
 9087       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9088     default:
 9089       fatal("Unexpected type argument %s", type2name(type)); break;
 9090   }
 9091 }
 9092 
 9093 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9094   switch(type) {
 9095     case T_BYTE:
 9096       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9097     case T_SHORT:
 9098       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9099     case T_INT:
 9100       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9101     case T_LONG:
 9102       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9103     default:
 9104       fatal("Unexpected type argument %s", type2name(type)); break;
 9105   }
 9106 }
 9107 
 9108 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9109   switch(type) {
 9110     case T_BYTE:
 9111       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9112     case T_SHORT:
 9113       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9114     case T_INT:
 9115       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9116     case T_LONG:
 9117       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9118     default:
 9119       fatal("Unexpected type argument %s", type2name(type)); break;
 9120   }
 9121 }
 9122 
 9123 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9124   switch(type) {
 9125     case T_BYTE:
 9126       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9127     case T_SHORT:
 9128       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9129     case T_INT:
 9130       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9131     case T_LONG:
 9132       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9133     default:
 9134       fatal("Unexpected type argument %s", type2name(type)); break;
 9135   }
 9136 }
 9137 
 9138 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9139   switch(type) {
 9140     case T_BYTE:
 9141       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9142     case T_SHORT:
 9143       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9144     case T_INT:
 9145       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9146     case T_LONG:
 9147       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9148     default:
 9149       fatal("Unexpected type argument %s", type2name(type)); break;
 9150   }
 9151 }
 9152 
 9153 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9154   switch(type) {
 9155     case T_BYTE:
 9156       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9157     case T_SHORT:
 9158       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9159     case T_INT:
 9160       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9161     case T_LONG:
 9162       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9163     case T_FLOAT:
 9164       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9165     case T_DOUBLE:
 9166       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9167     default:
 9168       fatal("Unexpected type argument %s", type2name(type)); break;
 9169   }
 9170 }
 9171 
 9172 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9173   switch(type) {
 9174     case T_BYTE:
 9175       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9176     case T_SHORT:
 9177       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9178     case T_INT:
 9179       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9180     case T_LONG:
 9181       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9182     case T_FLOAT:
 9183       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9184     case T_DOUBLE:
 9185       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9186     default:
 9187       fatal("Unexpected type argument %s", type2name(type)); break;
 9188   }
 9189 }
 9190 
 9191 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9192   switch(type) {
 9193     case T_BYTE:
 9194       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9195     case T_SHORT:
 9196       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9197     case T_INT:
 9198       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9199     case T_LONG:
 9200       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9201     case T_FLOAT:
 9202       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9203     case T_DOUBLE:
 9204       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9205     default:
 9206       fatal("Unexpected type argument %s", type2name(type)); break;
 9207   }
 9208 }
 9209 
 9210 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9211   switch(type) {
 9212     case T_BYTE:
 9213       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9214     case T_SHORT:
 9215       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9216     case T_INT:
 9217       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9218     case T_LONG:
 9219       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9220     case T_FLOAT:
 9221       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9222     case T_DOUBLE:
 9223       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9224     default:
 9225       fatal("Unexpected type argument %s", type2name(type)); break;
 9226   }
 9227 }
 9228 
 9229 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9230   switch(type) {
 9231     case T_INT:
 9232       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9233     case T_LONG:
 9234       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9235     default:
 9236       fatal("Unexpected type argument %s", type2name(type)); break;
 9237   }
 9238 }
 9239 
 9240 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9241   switch(type) {
 9242     case T_INT:
 9243       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9244     case T_LONG:
 9245       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9246     default:
 9247       fatal("Unexpected type argument %s", type2name(type)); break;
 9248   }
 9249 }
 9250 
 9251 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9252   switch(type) {
 9253     case T_INT:
 9254       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9255     case T_LONG:
 9256       evporq(dst, mask, nds, src, merge, vector_len); break;
 9257     default:
 9258       fatal("Unexpected type argument %s", type2name(type)); break;
 9259   }
 9260 }
 9261 
 9262 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9263   switch(type) {
 9264     case T_INT:
 9265       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9266     case T_LONG:
 9267       evporq(dst, mask, nds, src, merge, vector_len); break;
 9268     default:
 9269       fatal("Unexpected type argument %s", type2name(type)); break;
 9270   }
 9271 }
 9272 
 9273 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9274   switch(type) {
 9275     case T_INT:
 9276       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9277     case T_LONG:
 9278       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9279     default:
 9280       fatal("Unexpected type argument %s", type2name(type)); break;
 9281   }
 9282 }
 9283 
 9284 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9285   switch(type) {
 9286     case T_INT:
 9287       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9288     case T_LONG:
 9289       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9290     default:
 9291       fatal("Unexpected type argument %s", type2name(type)); break;
 9292   }
 9293 }
 9294 
 9295 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9296   switch(masklen) {
 9297     case 8:
 9298        kortestbl(src1, src2);
 9299        break;
 9300     case 16:
 9301        kortestwl(src1, src2);
 9302        break;
 9303     case 32:
 9304        kortestdl(src1, src2);
 9305        break;
 9306     case 64:
 9307        kortestql(src1, src2);
 9308        break;
 9309     default:
 9310       fatal("Unexpected mask length %d", masklen);
 9311       break;
 9312   }
 9313 }
 9314 
 9315 
 9316 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
 9317   switch(masklen)  {
 9318     case 8:
 9319        ktestbl(src1, src2);
 9320        break;
 9321     case 16:
 9322        ktestwl(src1, src2);
 9323        break;
 9324     case 32:
 9325        ktestdl(src1, src2);
 9326        break;
 9327     case 64:
 9328        ktestql(src1, src2);
 9329        break;
 9330     default:
 9331       fatal("Unexpected mask length %d", masklen);
 9332       break;
 9333   }
 9334 }
 9335 
 9336 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9337   switch(type) {
 9338     case T_INT:
 9339       evprold(dst, mask, src, shift, merge, vlen_enc); break;
 9340     case T_LONG:
 9341       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
 9342     default:
 9343       fatal("Unexpected type argument %s", type2name(type)); break;
 9344       break;
 9345   }
 9346 }
 9347 
 9348 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9349   switch(type) {
 9350     case T_INT:
 9351       evprord(dst, mask, src, shift, merge, vlen_enc); break;
 9352     case T_LONG:
 9353       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
 9354     default:
 9355       fatal("Unexpected type argument %s", type2name(type)); break;
 9356   }
 9357 }
 9358 
 9359 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9360   switch(type) {
 9361     case T_INT:
 9362       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9363     case T_LONG:
 9364       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9365     default:
 9366       fatal("Unexpected type argument %s", type2name(type)); break;
 9367   }
 9368 }
 9369 
 9370 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9371   switch(type) {
 9372     case T_INT:
 9373       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9374     case T_LONG:
 9375       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9376     default:
 9377       fatal("Unexpected type argument %s", type2name(type)); break;
 9378   }
 9379 }
 9380 
 9381 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9382   assert(rscratch != noreg || always_reachable(src), "missing");
 9383 
 9384   if (reachable(src)) {
 9385     evpandq(dst, nds, as_Address(src), vector_len);
 9386   } else {
 9387     lea(rscratch, src);
 9388     evpandq(dst, nds, Address(rscratch, 0), vector_len);
 9389   }
 9390 }
 9391 
 9392 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 9393   assert(rscratch != noreg || always_reachable(src), "missing");
 9394 
 9395   if (reachable(src)) {
 9396     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
 9397   } else {
 9398     lea(rscratch, src);
 9399     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 9400   }
 9401 }
 9402 
 9403 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9404   assert(rscratch != noreg || always_reachable(src), "missing");
 9405 
 9406   if (reachable(src)) {
 9407     evporq(dst, nds, as_Address(src), vector_len);
 9408   } else {
 9409     lea(rscratch, src);
 9410     evporq(dst, nds, Address(rscratch, 0), vector_len);
 9411   }
 9412 }
 9413 
 9414 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9415   assert(rscratch != noreg || always_reachable(src), "missing");
 9416 
 9417   if (reachable(src)) {
 9418     vpshufb(dst, nds, as_Address(src), vector_len);
 9419   } else {
 9420     lea(rscratch, src);
 9421     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
 9422   }
 9423 }
 9424 
 9425 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9426   assert(rscratch != noreg || always_reachable(src), "missing");
 9427 
 9428   if (reachable(src)) {
 9429     Assembler::vpor(dst, nds, as_Address(src), vector_len);
 9430   } else {
 9431     lea(rscratch, src);
 9432     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
 9433   }
 9434 }
 9435 
 9436 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
 9437   assert(rscratch != noreg || always_reachable(src3), "missing");
 9438 
 9439   if (reachable(src3)) {
 9440     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
 9441   } else {
 9442     lea(rscratch, src3);
 9443     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
 9444   }
 9445 }
 9446 
 9447 #if COMPILER2_OR_JVMCI
 9448 
 9449 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
 9450                                  Register length, Register temp, int vec_enc) {
 9451   // Computing mask for predicated vector store.
 9452   movptr(temp, -1);
 9453   bzhiq(temp, temp, length);
 9454   kmov(mask, temp);
 9455   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
 9456 }
 9457 
 9458 // Set memory operation for length "less than" 64 bytes.
 9459 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
 9460                                        XMMRegister xmm, KRegister mask, Register length,
 9461                                        Register temp, bool use64byteVector) {
 9462   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9463   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9464   if (!use64byteVector) {
 9465     fill32(dst, disp, xmm);
 9466     subptr(length, 32 >> shift);
 9467     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
 9468   } else {
 9469     assert(MaxVectorSize == 64, "vector length != 64");
 9470     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
 9471   }
 9472 }
 9473 
 9474 
 9475 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
 9476                                        XMMRegister xmm, KRegister mask, Register length,
 9477                                        Register temp) {
 9478   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9479   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9480   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
 9481 }
 9482 
 9483 
 9484 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
 9485   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9486   vmovdqu(dst, xmm);
 9487 }
 9488 
 9489 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
 9490   fill32(Address(dst, disp), xmm);
 9491 }
 9492 
 9493 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
 9494   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9495   if (!use64byteVector) {
 9496     fill32(dst, xmm);
 9497     fill32(dst.plus_disp(32), xmm);
 9498   } else {
 9499     evmovdquq(dst, xmm, Assembler::AVX_512bit);
 9500   }
 9501 }
 9502 
 9503 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
 9504   fill64(Address(dst, disp), xmm, use64byteVector);
 9505 }
 9506 
 9507 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
 9508                                         Register count, Register rtmp, XMMRegister xtmp) {
 9509   Label L_exit;
 9510   Label L_fill_start;
 9511   Label L_fill_64_bytes;
 9512   Label L_fill_96_bytes;
 9513   Label L_fill_128_bytes;
 9514   Label L_fill_128_bytes_loop;
 9515   Label L_fill_128_loop_header;
 9516   Label L_fill_128_bytes_loop_header;
 9517   Label L_fill_128_bytes_loop_pre_header;
 9518   Label L_fill_zmm_sequence;
 9519 
 9520   int shift = -1;
 9521   switch(type) {
 9522     case T_BYTE:  shift = 0;
 9523       break;
 9524     case T_SHORT: shift = 1;
 9525       break;
 9526     case T_INT:   shift = 2;
 9527       break;
 9528     /* Uncomment when LONG fill stubs are supported.
 9529     case T_LONG:  shift = 3;
 9530       break;
 9531     */
 9532     default:
 9533       fatal("Unhandled type: %s\n", type2name(type));
 9534   }
 9535 
 9536   if ((CopyAVX3Threshold != 0)  || (MaxVectorSize == 32)) {
 9537 
 9538     if (MaxVectorSize == 64) {
 9539       cmpq(count, CopyAVX3Threshold >> shift);
 9540       jcc(Assembler::greater, L_fill_zmm_sequence);
 9541     }
 9542 
 9543     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
 9544 
 9545     bind(L_fill_start);
 9546 
 9547     cmpq(count, 32 >> shift);
 9548     jccb(Assembler::greater, L_fill_64_bytes);
 9549     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9550     jmp(L_exit);
 9551 
 9552     bind(L_fill_64_bytes);
 9553     cmpq(count, 64 >> shift);
 9554     jccb(Assembler::greater, L_fill_96_bytes);
 9555     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9556     jmp(L_exit);
 9557 
 9558     bind(L_fill_96_bytes);
 9559     cmpq(count, 96 >> shift);
 9560     jccb(Assembler::greater, L_fill_128_bytes);
 9561     fill64(to, 0, xtmp);
 9562     subq(count, 64 >> shift);
 9563     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
 9564     jmp(L_exit);
 9565 
 9566     bind(L_fill_128_bytes);
 9567     cmpq(count, 128 >> shift);
 9568     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
 9569     fill64(to, 0, xtmp);
 9570     fill32(to, 64, xtmp);
 9571     subq(count, 96 >> shift);
 9572     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
 9573     jmp(L_exit);
 9574 
 9575     bind(L_fill_128_bytes_loop_pre_header);
 9576     {
 9577       mov(rtmp, to);
 9578       andq(rtmp, 31);
 9579       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
 9580       negq(rtmp);
 9581       addq(rtmp, 32);
 9582       mov64(r8, -1L);
 9583       bzhiq(r8, r8, rtmp);
 9584       kmovql(k2, r8);
 9585       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
 9586       addq(to, rtmp);
 9587       shrq(rtmp, shift);
 9588       subq(count, rtmp);
 9589     }
 9590 
 9591     cmpq(count, 128 >> shift);
 9592     jcc(Assembler::less, L_fill_start);
 9593 
 9594     bind(L_fill_128_bytes_loop_header);
 9595     subq(count, 128 >> shift);
 9596 
 9597     align32();
 9598     bind(L_fill_128_bytes_loop);
 9599       fill64(to, 0, xtmp);
 9600       fill64(to, 64, xtmp);
 9601       addq(to, 128);
 9602       subq(count, 128 >> shift);
 9603       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
 9604 
 9605     addq(count, 128 >> shift);
 9606     jcc(Assembler::zero, L_exit);
 9607     jmp(L_fill_start);
 9608   }
 9609 
 9610   if (MaxVectorSize == 64) {
 9611     // Sequence using 64 byte ZMM register.
 9612     Label L_fill_128_bytes_zmm;
 9613     Label L_fill_192_bytes_zmm;
 9614     Label L_fill_192_bytes_loop_zmm;
 9615     Label L_fill_192_bytes_loop_header_zmm;
 9616     Label L_fill_192_bytes_loop_pre_header_zmm;
 9617     Label L_fill_start_zmm_sequence;
 9618 
 9619     bind(L_fill_zmm_sequence);
 9620     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
 9621 
 9622     bind(L_fill_start_zmm_sequence);
 9623     cmpq(count, 64 >> shift);
 9624     jccb(Assembler::greater, L_fill_128_bytes_zmm);
 9625     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
 9626     jmp(L_exit);
 9627 
 9628     bind(L_fill_128_bytes_zmm);
 9629     cmpq(count, 128 >> shift);
 9630     jccb(Assembler::greater, L_fill_192_bytes_zmm);
 9631     fill64(to, 0, xtmp, true);
 9632     subq(count, 64 >> shift);
 9633     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
 9634     jmp(L_exit);
 9635 
 9636     bind(L_fill_192_bytes_zmm);
 9637     cmpq(count, 192 >> shift);
 9638     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
 9639     fill64(to, 0, xtmp, true);
 9640     fill64(to, 64, xtmp, true);
 9641     subq(count, 128 >> shift);
 9642     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
 9643     jmp(L_exit);
 9644 
 9645     bind(L_fill_192_bytes_loop_pre_header_zmm);
 9646     {
 9647       movq(rtmp, to);
 9648       andq(rtmp, 63);
 9649       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
 9650       negq(rtmp);
 9651       addq(rtmp, 64);
 9652       mov64(r8, -1L);
 9653       bzhiq(r8, r8, rtmp);
 9654       kmovql(k2, r8);
 9655       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
 9656       addq(to, rtmp);
 9657       shrq(rtmp, shift);
 9658       subq(count, rtmp);
 9659     }
 9660 
 9661     cmpq(count, 192 >> shift);
 9662     jcc(Assembler::less, L_fill_start_zmm_sequence);
 9663 
 9664     bind(L_fill_192_bytes_loop_header_zmm);
 9665     subq(count, 192 >> shift);
 9666 
 9667     align32();
 9668     bind(L_fill_192_bytes_loop_zmm);
 9669       fill64(to, 0, xtmp, true);
 9670       fill64(to, 64, xtmp, true);
 9671       fill64(to, 128, xtmp, true);
 9672       addq(to, 192);
 9673       subq(count, 192 >> shift);
 9674       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
 9675 
 9676     addq(count, 192 >> shift);
 9677     jcc(Assembler::zero, L_exit);
 9678     jmp(L_fill_start_zmm_sequence);
 9679   }
 9680   bind(L_exit);
 9681 }
 9682 #endif //COMPILER2_OR_JVMCI
 9683 
 9684 
 9685 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
 9686   Label done;
 9687   cvttss2sil(dst, src);
 9688   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9689   cmpl(dst, 0x80000000); // float_sign_flip
 9690   jccb(Assembler::notEqual, done);
 9691   subptr(rsp, 8);
 9692   movflt(Address(rsp, 0), src);
 9693   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
 9694   pop(dst);
 9695   bind(done);
 9696 }
 9697 
 9698 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
 9699   Label done;
 9700   cvttsd2sil(dst, src);
 9701   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9702   cmpl(dst, 0x80000000); // float_sign_flip
 9703   jccb(Assembler::notEqual, done);
 9704   subptr(rsp, 8);
 9705   movdbl(Address(rsp, 0), src);
 9706   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
 9707   pop(dst);
 9708   bind(done);
 9709 }
 9710 
 9711 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
 9712   Label done;
 9713   cvttss2siq(dst, src);
 9714   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9715   jccb(Assembler::notEqual, done);
 9716   subptr(rsp, 8);
 9717   movflt(Address(rsp, 0), src);
 9718   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
 9719   pop(dst);
 9720   bind(done);
 9721 }
 9722 
 9723 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9724   // Following code is line by line assembly translation rounding algorithm.
 9725   // Please refer to java.lang.Math.round(float) algorithm for details.
 9726   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
 9727   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
 9728   const int32_t FloatConsts_EXP_BIAS = 127;
 9729   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
 9730   const int32_t MINUS_32 = 0xFFFFFFE0;
 9731   Label L_special_case, L_block1, L_exit;
 9732   movl(rtmp, FloatConsts_EXP_BIT_MASK);
 9733   movdl(dst, src);
 9734   andl(dst, rtmp);
 9735   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
 9736   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
 9737   subl(rtmp, dst);
 9738   movl(rcx, rtmp);
 9739   movl(dst, MINUS_32);
 9740   testl(rtmp, dst);
 9741   jccb(Assembler::notEqual, L_special_case);
 9742   movdl(dst, src);
 9743   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
 9744   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
 9745   movdl(rtmp, src);
 9746   testl(rtmp, rtmp);
 9747   jccb(Assembler::greaterEqual, L_block1);
 9748   negl(dst);
 9749   bind(L_block1);
 9750   sarl(dst);
 9751   addl(dst, 0x1);
 9752   sarl(dst, 0x1);
 9753   jmp(L_exit);
 9754   bind(L_special_case);
 9755   convert_f2i(dst, src);
 9756   bind(L_exit);
 9757 }
 9758 
 9759 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9760   // Following code is line by line assembly translation rounding algorithm.
 9761   // Please refer to java.lang.Math.round(double) algorithm for details.
 9762   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
 9763   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
 9764   const int64_t DoubleConsts_EXP_BIAS = 1023;
 9765   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
 9766   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
 9767   Label L_special_case, L_block1, L_exit;
 9768   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
 9769   movq(dst, src);
 9770   andq(dst, rtmp);
 9771   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
 9772   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
 9773   subq(rtmp, dst);
 9774   movq(rcx, rtmp);
 9775   mov64(dst, MINUS_64);
 9776   testq(rtmp, dst);
 9777   jccb(Assembler::notEqual, L_special_case);
 9778   movq(dst, src);
 9779   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
 9780   andq(dst, rtmp);
 9781   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
 9782   orq(dst, rtmp);
 9783   movq(rtmp, src);
 9784   testq(rtmp, rtmp);
 9785   jccb(Assembler::greaterEqual, L_block1);
 9786   negq(dst);
 9787   bind(L_block1);
 9788   sarq(dst);
 9789   addq(dst, 0x1);
 9790   sarq(dst, 0x1);
 9791   jmp(L_exit);
 9792   bind(L_special_case);
 9793   convert_d2l(dst, src);
 9794   bind(L_exit);
 9795 }
 9796 
 9797 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
 9798   Label done;
 9799   cvttsd2siq(dst, src);
 9800   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9801   jccb(Assembler::notEqual, done);
 9802   subptr(rsp, 8);
 9803   movdbl(Address(rsp, 0), src);
 9804   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
 9805   pop(dst);
 9806   bind(done);
 9807 }
 9808 
 9809 void MacroAssembler::cache_wb(Address line)
 9810 {
 9811   // 64 bit cpus always support clflush
 9812   bool optimized = VM_Version::supports_clflushopt();
 9813   bool no_evict = VM_Version::supports_clwb();
 9814 
 9815   // prefer clwb (writeback without evict) otherwise
 9816   // prefer clflushopt (potentially parallel writeback with evict)
 9817   // otherwise fallback on clflush (serial writeback with evict)
 9818 
 9819   if (optimized) {
 9820     if (no_evict) {
 9821       clwb(line);
 9822     } else {
 9823       clflushopt(line);
 9824     }
 9825   } else {
 9826     // no need for fence when using CLFLUSH
 9827     clflush(line);
 9828   }
 9829 }
 9830 
 9831 void MacroAssembler::cache_wbsync(bool is_pre)
 9832 {
 9833   bool optimized = VM_Version::supports_clflushopt();
 9834   bool no_evict = VM_Version::supports_clwb();
 9835 
 9836   // pick the correct implementation
 9837 
 9838   if (!is_pre && (optimized || no_evict)) {
 9839     // need an sfence for post flush when using clflushopt or clwb
 9840     // otherwise no no need for any synchroniaztion
 9841 
 9842     sfence();
 9843   }
 9844 }
 9845 
 9846 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
 9847   switch (cond) {
 9848     // Note some conditions are synonyms for others
 9849     case Assembler::zero:         return Assembler::notZero;
 9850     case Assembler::notZero:      return Assembler::zero;
 9851     case Assembler::less:         return Assembler::greaterEqual;
 9852     case Assembler::lessEqual:    return Assembler::greater;
 9853     case Assembler::greater:      return Assembler::lessEqual;
 9854     case Assembler::greaterEqual: return Assembler::less;
 9855     case Assembler::below:        return Assembler::aboveEqual;
 9856     case Assembler::belowEqual:   return Assembler::above;
 9857     case Assembler::above:        return Assembler::belowEqual;
 9858     case Assembler::aboveEqual:   return Assembler::below;
 9859     case Assembler::overflow:     return Assembler::noOverflow;
 9860     case Assembler::noOverflow:   return Assembler::overflow;
 9861     case Assembler::negative:     return Assembler::positive;
 9862     case Assembler::positive:     return Assembler::negative;
 9863     case Assembler::parity:       return Assembler::noParity;
 9864     case Assembler::noParity:     return Assembler::parity;
 9865   }
 9866   ShouldNotReachHere(); return Assembler::overflow;
 9867 }
 9868 
 9869 // This is simply a call to Thread::current()
 9870 void MacroAssembler::get_thread_slow(Register thread) {
 9871   if (thread != rax) {
 9872     push(rax);
 9873   }
 9874   push(rdi);
 9875   push(rsi);
 9876   push(rdx);
 9877   push(rcx);
 9878   push(r8);
 9879   push(r9);
 9880   push(r10);
 9881   push(r11);
 9882 
 9883   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
 9884 
 9885   pop(r11);
 9886   pop(r10);
 9887   pop(r9);
 9888   pop(r8);
 9889   pop(rcx);
 9890   pop(rdx);
 9891   pop(rsi);
 9892   pop(rdi);
 9893   if (thread != rax) {
 9894     mov(thread, rax);
 9895     pop(rax);
 9896   }
 9897 }
 9898 
 9899 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
 9900   Label L_stack_ok;
 9901   if (bias == 0) {
 9902     testptr(sp, 2 * wordSize - 1);
 9903   } else {
 9904     // lea(tmp, Address(rsp, bias);
 9905     mov(tmp, sp);
 9906     addptr(tmp, bias);
 9907     testptr(tmp, 2 * wordSize - 1);
 9908   }
 9909   jcc(Assembler::equal, L_stack_ok);
 9910   block_comment(msg);
 9911   stop(msg);
 9912   bind(L_stack_ok);
 9913 }
 9914 
 9915 // Implements fast-locking.
 9916 //
 9917 // obj: the object to be locked
 9918 // reg_rax: rax
 9919 // thread: the thread which attempts to lock obj
 9920 // tmp: a temporary register
 9921 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
 9922   Register thread = r15_thread;
 9923 
 9924   assert(reg_rax == rax, "");
 9925   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
 9926 
 9927   Label push;
 9928   const Register top = tmp;
 9929 
 9930   // Preload the markWord. It is important that this is the first
 9931   // instruction emitted as it is part of C1's null check semantics.
 9932   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9933 
 9934   if (UseObjectMonitorTable) {
 9935     // Clear cache in case fast locking succeeds or we need to take the slow-path.
 9936     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
 9937   }
 9938 
 9939   if (DiagnoseSyncOnValueBasedClasses != 0) {
 9940     load_klass(tmp, obj, rscratch1);
 9941     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
 9942     jcc(Assembler::notZero, slow);
 9943   }
 9944 
 9945   // Load top.
 9946   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9947 
 9948   // Check if the lock-stack is full.
 9949   cmpl(top, LockStack::end_offset());
 9950   jcc(Assembler::greaterEqual, slow);
 9951 
 9952   // Check for recursion.
 9953   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9954   jcc(Assembler::equal, push);
 9955 
 9956   // Check header for monitor (0b10).
 9957   testptr(reg_rax, markWord::monitor_value);
 9958   jcc(Assembler::notZero, slow);
 9959 
 9960   // Try to lock. Transition lock bits 0b01 => 0b00
 9961   movptr(tmp, reg_rax);
 9962   andptr(tmp, ~(int32_t)markWord::unlocked_value);
 9963   orptr(reg_rax, markWord::unlocked_value);



 9964   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9965   jcc(Assembler::notEqual, slow);
 9966 
 9967   // Restore top, CAS clobbers register.
 9968   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9969 
 9970   bind(push);
 9971   // After successful lock, push object on lock-stack.
 9972   movptr(Address(thread, top), obj);
 9973   incrementl(top, oopSize);
 9974   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
 9975 }
 9976 
 9977 // Implements fast-unlocking.
 9978 //
 9979 // obj: the object to be unlocked
 9980 // reg_rax: rax
 9981 // thread: the thread
 9982 // tmp: a temporary register
 9983 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
 9984   Register thread = r15_thread;
 9985 
 9986   assert(reg_rax == rax, "");
 9987   assert_different_registers(obj, reg_rax, thread, tmp);
 9988 
 9989   Label unlocked, push_and_slow;
 9990   const Register top = tmp;
 9991 
 9992   // Check if obj is top of lock-stack.
 9993   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9994   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9995   jcc(Assembler::notEqual, slow);
 9996 
 9997   // Pop lock-stack.
 9998   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
 9999   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10000 
10001   // Check if recursive.
10002   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
10003   jcc(Assembler::equal, unlocked);
10004 
10005   // Not recursive. Check header for monitor (0b10).
10006   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10007   testptr(reg_rax, markWord::monitor_value);
10008   jcc(Assembler::notZero, push_and_slow);
10009 
10010 #ifdef ASSERT
10011   // Check header not unlocked (0b01).
10012   Label not_unlocked;
10013   testptr(reg_rax, markWord::unlocked_value);
10014   jcc(Assembler::zero, not_unlocked);
10015   stop("fast_unlock already unlocked");
10016   bind(not_unlocked);
10017 #endif
10018 
10019   // Try to unlock. Transition lock bits 0b00 => 0b01
10020   movptr(tmp, reg_rax);
10021   orptr(tmp, markWord::unlocked_value);
10022   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10023   jcc(Assembler::equal, unlocked);
10024 
10025   bind(push_and_slow);
10026   // Restore lock-stack and handle the unlock in runtime.
10027 #ifdef ASSERT
10028   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10029   movptr(Address(thread, top), obj);
10030 #endif
10031   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10032   jmp(slow);
10033 
10034   bind(unlocked);
10035 }
10036 
10037 // Saves legacy GPRs state on stack.
10038 void MacroAssembler::save_legacy_gprs() {
10039   subq(rsp, 16 * wordSize);
10040   movq(Address(rsp, 15 * wordSize), rax);
10041   movq(Address(rsp, 14 * wordSize), rcx);
10042   movq(Address(rsp, 13 * wordSize), rdx);
10043   movq(Address(rsp, 12 * wordSize), rbx);
10044   movq(Address(rsp, 10 * wordSize), rbp);
10045   movq(Address(rsp, 9 * wordSize), rsi);
10046   movq(Address(rsp, 8 * wordSize), rdi);
10047   movq(Address(rsp, 7 * wordSize), r8);
10048   movq(Address(rsp, 6 * wordSize), r9);
10049   movq(Address(rsp, 5 * wordSize), r10);
10050   movq(Address(rsp, 4 * wordSize), r11);
10051   movq(Address(rsp, 3 * wordSize), r12);
10052   movq(Address(rsp, 2 * wordSize), r13);
10053   movq(Address(rsp, wordSize), r14);
10054   movq(Address(rsp, 0), r15);
10055 }
10056 
10057 // Resotres back legacy GPRs state from stack.
10058 void MacroAssembler::restore_legacy_gprs() {
10059   movq(r15, Address(rsp, 0));
10060   movq(r14, Address(rsp, wordSize));
10061   movq(r13, Address(rsp, 2 * wordSize));
10062   movq(r12, Address(rsp, 3 * wordSize));
10063   movq(r11, Address(rsp, 4 * wordSize));
10064   movq(r10, Address(rsp, 5 * wordSize));
10065   movq(r9,  Address(rsp, 6 * wordSize));
10066   movq(r8,  Address(rsp, 7 * wordSize));
10067   movq(rdi, Address(rsp, 8 * wordSize));
10068   movq(rsi, Address(rsp, 9 * wordSize));
10069   movq(rbp, Address(rsp, 10 * wordSize));
10070   movq(rbx, Address(rsp, 12 * wordSize));
10071   movq(rdx, Address(rsp, 13 * wordSize));
10072   movq(rcx, Address(rsp, 14 * wordSize));
10073   movq(rax, Address(rsp, 15 * wordSize));
10074   addq(rsp, 16 * wordSize);
10075 }
10076 
10077 void MacroAssembler::load_aotrc_address(Register reg, address a) {
10078 #if INCLUDE_CDS
10079   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
10080   if (AOTCodeCache::is_on_for_dump()) {
10081     // all aotrc field addresses should be registered in the AOTCodeCache address table
10082     lea(reg, ExternalAddress(a));
10083   } else {
10084     mov64(reg, (uint64_t)a);
10085   }
10086 #else
10087   ShouldNotReachHere();
10088 #endif
10089 }
10090 
10091 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10092   if (VM_Version::supports_apx_f()) {
10093     esetzucc(comparison, dst);
10094   } else {
10095     setb(comparison, dst);
10096     movzbl(dst, dst);
10097   }
10098 }
--- EOF ---