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/macros.hpp"
   59 
   60 #ifdef PRODUCT
   61 #define BLOCK_COMMENT(str) /* nothing */
   62 #define STOP(error) stop(error)
   63 #else
   64 #define BLOCK_COMMENT(str) block_comment(str)
   65 #define STOP(error) block_comment(error); stop(error)
   66 #endif
   67 
   68 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   69 
   70 #ifdef ASSERT
   71 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   72 #endif
   73 
   74 static const Assembler::Condition reverse[] = {
   75     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   76     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   77     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   78     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
   79     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
   80     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
   81     Assembler::above          /* belowEqual    = 0x6 */ ,
   82     Assembler::belowEqual     /* above         = 0x7 */ ,
   83     Assembler::positive       /* negative      = 0x8 */ ,
   84     Assembler::negative       /* positive      = 0x9 */ ,
   85     Assembler::noParity       /* parity        = 0xa */ ,
   86     Assembler::parity         /* noParity      = 0xb */ ,
   87     Assembler::greaterEqual   /* less          = 0xc */ ,
   88     Assembler::less           /* greaterEqual  = 0xd */ ,
   89     Assembler::greater        /* lessEqual     = 0xe */ ,
   90     Assembler::lessEqual      /* greater       = 0xf, */
   91 
   92 };
   93 
   94 
   95 // Implementation of MacroAssembler
   96 
   97 Address MacroAssembler::as_Address(AddressLiteral adr) {
   98   // amd64 always does this as a pc-rel
   99   // we can be absolute or disp based on the instruction type
  100   // jmp/call are displacements others are absolute
  101   assert(!adr.is_lval(), "must be rval");
  102   assert(reachable(adr), "must be");
  103   return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
  104 
  105 }
  106 
  107 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  108   AddressLiteral base = adr.base();
  109   lea(rscratch, base);
  110   Address index = adr.index();
  111   assert(index._disp == 0, "must not have disp"); // maybe it can?
  112   Address array(rscratch, index._index, index._scale, index._disp);
  113   return array;
  114 }
  115 
  116 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
  117   Label L, E;
  118 
  119 #ifdef _WIN64
  120   // Windows always allocates space for it's register args
  121   assert(num_args <= 4, "only register arguments supported");
  122   subq(rsp,  frame::arg_reg_save_area_bytes);
  123 #endif
  124 
  125   // Align stack if necessary
  126   testl(rsp, 15);
  127   jcc(Assembler::zero, L);
  128 
  129   subq(rsp, 8);
  130   call(RuntimeAddress(entry_point));
  131   addq(rsp, 8);
  132   jmp(E);
  133 
  134   bind(L);
  135   call(RuntimeAddress(entry_point));
  136 
  137   bind(E);
  138 
  139 #ifdef _WIN64
  140   // restore stack pointer
  141   addq(rsp, frame::arg_reg_save_area_bytes);
  142 #endif
  143 }
  144 
  145 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
  146   assert(!src2.is_lval(), "should use cmpptr");
  147   assert(rscratch != noreg || always_reachable(src2), "missing");
  148 
  149   if (reachable(src2)) {
  150     cmpq(src1, as_Address(src2));
  151   } else {
  152     lea(rscratch, src2);
  153     Assembler::cmpq(src1, Address(rscratch, 0));
  154   }
  155 }
  156 
  157 int MacroAssembler::corrected_idivq(Register reg) {
  158   // Full implementation of Java ldiv and lrem; checks for special
  159   // case as described in JVM spec., p.243 & p.271.  The function
  160   // returns the (pc) offset of the idivl instruction - may be needed
  161   // for implicit exceptions.
  162   //
  163   //         normal case                           special case
  164   //
  165   // input : rax: dividend                         min_long
  166   //         reg: divisor   (may not be eax/edx)   -1
  167   //
  168   // output: rax: quotient  (= rax idiv reg)       min_long
  169   //         rdx: remainder (= rax irem reg)       0
  170   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
  171   static const int64_t min_long = 0x8000000000000000;
  172   Label normal_case, special_case;
  173 
  174   // check for special case
  175   cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
  176   jcc(Assembler::notEqual, normal_case);
  177   xorl(rdx, rdx); // prepare rdx for possible special case (where
  178                   // remainder = 0)
  179   cmpq(reg, -1);
  180   jcc(Assembler::equal, special_case);
  181 
  182   // handle normal case
  183   bind(normal_case);
  184   cdqq();
  185   int idivq_offset = offset();
  186   idivq(reg);
  187 
  188   // normal and special case exit
  189   bind(special_case);
  190 
  191   return idivq_offset;
  192 }
  193 
  194 void MacroAssembler::decrementq(Register reg, int value) {
  195   if (value == min_jint) { subq(reg, value); return; }
  196   if (value <  0) { incrementq(reg, -value); return; }
  197   if (value == 0) {                        ; return; }
  198   if (value == 1 && UseIncDec) { decq(reg) ; return; }
  199   /* else */      { subq(reg, value)       ; return; }
  200 }
  201 
  202 void MacroAssembler::decrementq(Address dst, int value) {
  203   if (value == min_jint) { subq(dst, value); return; }
  204   if (value <  0) { incrementq(dst, -value); return; }
  205   if (value == 0) {                        ; return; }
  206   if (value == 1 && UseIncDec) { decq(dst) ; return; }
  207   /* else */      { subq(dst, value)       ; return; }
  208 }
  209 
  210 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
  211   assert(rscratch != noreg || always_reachable(dst), "missing");
  212 
  213   if (reachable(dst)) {
  214     incrementq(as_Address(dst));
  215   } else {
  216     lea(rscratch, dst);
  217     incrementq(Address(rscratch, 0));
  218   }
  219 }
  220 
  221 void MacroAssembler::incrementq(Register reg, int value) {
  222   if (value == min_jint) { addq(reg, value); return; }
  223   if (value <  0) { decrementq(reg, -value); return; }
  224   if (value == 0) {                        ; return; }
  225   if (value == 1 && UseIncDec) { incq(reg) ; return; }
  226   /* else */      { addq(reg, value)       ; return; }
  227 }
  228 
  229 void MacroAssembler::incrementq(Address dst, int value) {
  230   if (value == min_jint) { addq(dst, value); return; }
  231   if (value <  0) { decrementq(dst, -value); return; }
  232   if (value == 0) {                        ; return; }
  233   if (value == 1 && UseIncDec) { incq(dst) ; return; }
  234   /* else */      { addq(dst, value)       ; return; }
  235 }
  236 
  237 // 32bit can do a case table jump in one instruction but we no longer allow the base
  238 // to be installed in the Address class
  239 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  240   lea(rscratch, entry.base());
  241   Address dispatch = entry.index();
  242   assert(dispatch._base == noreg, "must be");
  243   dispatch._base = rscratch;
  244   jmp(dispatch);
  245 }
  246 
  247 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  248   ShouldNotReachHere(); // 64bit doesn't use two regs
  249   cmpq(x_lo, y_lo);
  250 }
  251 
  252 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  253   mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  254 }
  255 
  256 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  257   lea(rscratch, adr);
  258   movptr(dst, rscratch);
  259 }
  260 
  261 void MacroAssembler::leave() {
  262   // %%% is this really better? Why not on 32bit too?
  263   emit_int8((unsigned char)0xC9); // LEAVE
  264 }
  265 
  266 void MacroAssembler::lneg(Register hi, Register lo) {
  267   ShouldNotReachHere(); // 64bit doesn't use two regs
  268   negq(lo);
  269 }
  270 
  271 void MacroAssembler::movoop(Register dst, jobject obj) {
  272   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  273 }
  274 
  275 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  276   mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  277   movq(dst, rscratch);
  278 }
  279 
  280 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  281   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  282 }
  283 
  284 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  285   mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  286   movq(dst, rscratch);
  287 }
  288 
  289 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  290   if (src.is_lval()) {
  291     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  292   } else {
  293     if (reachable(src)) {
  294       movq(dst, as_Address(src));
  295     } else {
  296       lea(dst, src);
  297       movq(dst, Address(dst, 0));
  298     }
  299   }
  300 }
  301 
  302 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  303   movq(as_Address(dst, rscratch), src);
  304 }
  305 
  306 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  307   movq(dst, as_Address(src, dst /*rscratch*/));
  308 }
  309 
  310 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
  311 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  312   if (is_simm32(src)) {
  313     movptr(dst, checked_cast<int32_t>(src));
  314   } else {
  315     mov64(rscratch, src);
  316     movq(dst, rscratch);
  317   }
  318 }
  319 
  320 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  321   movoop(rscratch, obj);
  322   push(rscratch);
  323 }
  324 
  325 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  326   mov_metadata(rscratch, obj);
  327   push(rscratch);
  328 }
  329 
  330 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  331   lea(rscratch, src);
  332   if (src.is_lval()) {
  333     push(rscratch);
  334   } else {
  335     pushq(Address(rscratch, 0));
  336   }
  337 }
  338 
  339 static void pass_arg0(MacroAssembler* masm, Register arg) {
  340   if (c_rarg0 != arg ) {
  341     masm->mov(c_rarg0, arg);
  342   }
  343 }
  344 
  345 static void pass_arg1(MacroAssembler* masm, Register arg) {
  346   if (c_rarg1 != arg ) {
  347     masm->mov(c_rarg1, arg);
  348   }
  349 }
  350 
  351 static void pass_arg2(MacroAssembler* masm, Register arg) {
  352   if (c_rarg2 != arg ) {
  353     masm->mov(c_rarg2, arg);
  354   }
  355 }
  356 
  357 static void pass_arg3(MacroAssembler* masm, Register arg) {
  358   if (c_rarg3 != arg ) {
  359     masm->mov(c_rarg3, arg);
  360   }
  361 }
  362 
  363 void MacroAssembler::stop(const char* msg) {
  364   if (ShowMessageBoxOnError) {
  365     address rip = pc();
  366     pusha(); // get regs on stack
  367     lea(c_rarg1, InternalAddress(rip));
  368     movq(c_rarg2, rsp); // pass pointer to regs array
  369   }
  370   // Skip AOT caching C strings in scratch buffer.
  371   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  372   lea(c_rarg0, ExternalAddress((address) str));
  373   andq(rsp, -16); // align stack as required by ABI
  374   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
  375   hlt();
  376 }
  377 
  378 void MacroAssembler::warn(const char* msg) {
  379   push(rbp);
  380   movq(rbp, rsp);
  381   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  382   push_CPU_state();   // keeps alignment at 16 bytes
  383 
  384 #ifdef _WIN64
  385   // Windows always allocates space for its register args
  386   subq(rsp,  frame::arg_reg_save_area_bytes);
  387 #endif
  388   lea(c_rarg0, ExternalAddress((address) msg));
  389   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  390 
  391 #ifdef _WIN64
  392   // restore stack pointer
  393   addq(rsp, frame::arg_reg_save_area_bytes);
  394 #endif
  395   pop_CPU_state();
  396   mov(rsp, rbp);
  397   pop(rbp);
  398 }
  399 
  400 void MacroAssembler::print_state() {
  401   address rip = pc();
  402   pusha();            // get regs on stack
  403   push(rbp);
  404   movq(rbp, rsp);
  405   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  406   push_CPU_state();   // keeps alignment at 16 bytes
  407 
  408   lea(c_rarg0, InternalAddress(rip));
  409   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  410   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  411 
  412   pop_CPU_state();
  413   mov(rsp, rbp);
  414   pop(rbp);
  415   popa();
  416 }
  417 
  418 #ifndef PRODUCT
  419 extern "C" void findpc(intptr_t x);
  420 #endif
  421 
  422 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  423   // In order to get locks to work, we need to fake a in_VM state
  424   if (ShowMessageBoxOnError) {
  425     JavaThread* thread = JavaThread::current();
  426     JavaThreadState saved_state = thread->thread_state();
  427     thread->set_thread_state(_thread_in_vm);
  428 #ifndef PRODUCT
  429     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  430       ttyLocker ttyl;
  431       BytecodeCounter::print();
  432     }
  433 #endif
  434     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  435     // XXX correct this offset for amd64
  436     // This is the value of eip which points to where verify_oop will return.
  437     if (os::message_box(msg, "Execution stopped, print registers?")) {
  438       print_state64(pc, regs);
  439       BREAKPOINT;
  440     }
  441   }
  442   fatal("DEBUG MESSAGE: %s", msg);
  443 }
  444 
  445 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  446   ttyLocker ttyl;
  447   DebuggingContext debugging{};
  448   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  449 #ifndef PRODUCT
  450   tty->cr();
  451   findpc(pc);
  452   tty->cr();
  453 #endif
  454 #define PRINT_REG(rax, value) \
  455   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  456   PRINT_REG(rax, regs[15]);
  457   PRINT_REG(rbx, regs[12]);
  458   PRINT_REG(rcx, regs[14]);
  459   PRINT_REG(rdx, regs[13]);
  460   PRINT_REG(rdi, regs[8]);
  461   PRINT_REG(rsi, regs[9]);
  462   PRINT_REG(rbp, regs[10]);
  463   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  464   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  465   PRINT_REG(r8 , regs[7]);
  466   PRINT_REG(r9 , regs[6]);
  467   PRINT_REG(r10, regs[5]);
  468   PRINT_REG(r11, regs[4]);
  469   PRINT_REG(r12, regs[3]);
  470   PRINT_REG(r13, regs[2]);
  471   PRINT_REG(r14, regs[1]);
  472   PRINT_REG(r15, regs[0]);
  473 #undef PRINT_REG
  474   // Print some words near the top of the stack.
  475   int64_t* rsp = &regs[16];
  476   int64_t* dump_sp = rsp;
  477   for (int col1 = 0; col1 < 8; col1++) {
  478     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  479     os::print_location(tty, *dump_sp++);
  480   }
  481   for (int row = 0; row < 25; row++) {
  482     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  483     for (int col = 0; col < 4; col++) {
  484       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  485     }
  486     tty->cr();
  487   }
  488   // Print some instructions around pc:
  489   Disassembler::decode((address)pc-64, (address)pc);
  490   tty->print_cr("--------");
  491   Disassembler::decode((address)pc, (address)pc+32);
  492 }
  493 
  494 // The java_calling_convention describes stack locations as ideal slots on
  495 // a frame with no abi restrictions. Since we must observe abi restrictions
  496 // (like the placement of the register window) the slots must be biased by
  497 // the following value.
  498 static int reg2offset_in(VMReg r) {
  499   // Account for saved rbp and return address
  500   // This should really be in_preserve_stack_slots
  501   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  502 }
  503 
  504 static int reg2offset_out(VMReg r) {
  505   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  506 }
  507 
  508 // A long move
  509 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  510 
  511   // The calling conventions assures us that each VMregpair is either
  512   // all really one physical register or adjacent stack slots.
  513 
  514   if (src.is_single_phys_reg() ) {
  515     if (dst.is_single_phys_reg()) {
  516       if (dst.first() != src.first()) {
  517         mov(dst.first()->as_Register(), src.first()->as_Register());
  518       }
  519     } else {
  520       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  521              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  522       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  523     }
  524   } else if (dst.is_single_phys_reg()) {
  525     assert(src.is_single_reg(),  "not a stack pair");
  526     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  527   } else {
  528     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  529     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  530     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  531   }
  532 }
  533 
  534 // A double move
  535 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  536 
  537   // The calling conventions assures us that each VMregpair is either
  538   // all really one physical register or adjacent stack slots.
  539 
  540   if (src.is_single_phys_reg() ) {
  541     if (dst.is_single_phys_reg()) {
  542       // In theory these overlap but the ordering is such that this is likely a nop
  543       if ( src.first() != dst.first()) {
  544         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  545       }
  546     } else {
  547       assert(dst.is_single_reg(), "not a stack pair");
  548       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  549     }
  550   } else if (dst.is_single_phys_reg()) {
  551     assert(src.is_single_reg(),  "not a stack pair");
  552     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  553   } else {
  554     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  555     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  556     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  557   }
  558 }
  559 
  560 
  561 // A float arg may have to do float reg int reg conversion
  562 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  563   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  564 
  565   // The calling conventions assures us that each VMregpair is either
  566   // all really one physical register or adjacent stack slots.
  567 
  568   if (src.first()->is_stack()) {
  569     if (dst.first()->is_stack()) {
  570       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  571       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  572     } else {
  573       // stack to reg
  574       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  575       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  576     }
  577   } else if (dst.first()->is_stack()) {
  578     // reg to stack
  579     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  580     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  581   } else {
  582     // reg to reg
  583     // In theory these overlap but the ordering is such that this is likely a nop
  584     if ( src.first() != dst.first()) {
  585       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  586     }
  587   }
  588 }
  589 
  590 // On 64 bit we will store integer like items to the stack as
  591 // 64 bits items (x86_32/64 abi) even though java would only store
  592 // 32bits for a parameter. On 32bit it will simply be 32 bits
  593 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  594 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  595   if (src.first()->is_stack()) {
  596     if (dst.first()->is_stack()) {
  597       // stack to stack
  598       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  599       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  600     } else {
  601       // stack to reg
  602       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  603     }
  604   } else if (dst.first()->is_stack()) {
  605     // reg to stack
  606     // Do we really have to sign extend???
  607     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  608     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  609   } else {
  610     // Do we really have to sign extend???
  611     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
  612     if (dst.first() != src.first()) {
  613       movq(dst.first()->as_Register(), src.first()->as_Register());
  614     }
  615   }
  616 }
  617 
  618 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
  619   if (src.first()->is_stack()) {
  620     if (dst.first()->is_stack()) {
  621       // stack to stack
  622       movq(rax, Address(rbp, reg2offset_in(src.first())));
  623       movq(Address(rsp, reg2offset_out(dst.first())), rax);
  624     } else {
  625       // stack to reg
  626       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
  627     }
  628   } else if (dst.first()->is_stack()) {
  629     // reg to stack
  630     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
  631   } else {
  632     if (dst.first() != src.first()) {
  633       movq(dst.first()->as_Register(), src.first()->as_Register());
  634     }
  635   }
  636 }
  637 
  638 // An oop arg. Must pass a handle not the oop itself
  639 void MacroAssembler::object_move(OopMap* map,
  640                         int oop_handle_offset,
  641                         int framesize_in_slots,
  642                         VMRegPair src,
  643                         VMRegPair dst,
  644                         bool is_receiver,
  645                         int* receiver_offset) {
  646 
  647   // must pass a handle. First figure out the location we use as a handle
  648 
  649   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
  650 
  651   // See if oop is null if it is we need no handle
  652 
  653   if (src.first()->is_stack()) {
  654 
  655     // Oop is already on the stack as an argument
  656     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
  657     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
  658     if (is_receiver) {
  659       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
  660     }
  661 
  662     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
  663     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
  664     // conditionally move a null
  665     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
  666   } else {
  667 
  668     // Oop is in a register we must store it to the space we reserve
  669     // on the stack for oop_handles and pass a handle if oop is non-null
  670 
  671     const Register rOop = src.first()->as_Register();
  672     int oop_slot;
  673     if (rOop == j_rarg0)
  674       oop_slot = 0;
  675     else if (rOop == j_rarg1)
  676       oop_slot = 1;
  677     else if (rOop == j_rarg2)
  678       oop_slot = 2;
  679     else if (rOop == j_rarg3)
  680       oop_slot = 3;
  681     else if (rOop == j_rarg4)
  682       oop_slot = 4;
  683     else {
  684       assert(rOop == j_rarg5, "wrong register");
  685       oop_slot = 5;
  686     }
  687 
  688     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
  689     int offset = oop_slot*VMRegImpl::stack_slot_size;
  690 
  691     map->set_oop(VMRegImpl::stack2reg(oop_slot));
  692     // Store oop in handle area, may be null
  693     movptr(Address(rsp, offset), rOop);
  694     if (is_receiver) {
  695       *receiver_offset = offset;
  696     }
  697 
  698     cmpptr(rOop, NULL_WORD);
  699     lea(rHandle, Address(rsp, offset));
  700     // conditionally move a null from the handle area where it was just stored
  701     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
  702   }
  703 
  704   // If arg is on the stack then place it otherwise it is already in correct reg.
  705   if (dst.first()->is_stack()) {
  706     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
  707   }
  708 }
  709 
  710 void MacroAssembler::addptr(Register dst, int32_t imm32) {
  711   addq(dst, imm32);
  712 }
  713 
  714 void MacroAssembler::addptr(Register dst, Register src) {
  715   addq(dst, src);
  716 }
  717 
  718 void MacroAssembler::addptr(Address dst, Register src) {
  719   addq(dst, src);
  720 }
  721 
  722 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  723   assert(rscratch != noreg || always_reachable(src), "missing");
  724 
  725   if (reachable(src)) {
  726     Assembler::addsd(dst, as_Address(src));
  727   } else {
  728     lea(rscratch, src);
  729     Assembler::addsd(dst, Address(rscratch, 0));
  730   }
  731 }
  732 
  733 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
  734   assert(rscratch != noreg || always_reachable(src), "missing");
  735 
  736   if (reachable(src)) {
  737     addss(dst, as_Address(src));
  738   } else {
  739     lea(rscratch, src);
  740     addss(dst, Address(rscratch, 0));
  741   }
  742 }
  743 
  744 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  745   assert(rscratch != noreg || always_reachable(src), "missing");
  746 
  747   if (reachable(src)) {
  748     Assembler::addpd(dst, as_Address(src));
  749   } else {
  750     lea(rscratch, src);
  751     Assembler::addpd(dst, Address(rscratch, 0));
  752   }
  753 }
  754 
  755 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
  756 // Stub code is generated once and never copied.
  757 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
  758 void MacroAssembler::align64() {
  759   align(64, (uint)(uintptr_t)pc());
  760 }
  761 
  762 void MacroAssembler::align32() {
  763   align(32, (uint)(uintptr_t)pc());
  764 }
  765 
  766 void MacroAssembler::align(uint modulus) {
  767   // 8273459: Ensure alignment is possible with current segment alignment
  768   assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
  769   align(modulus, offset());
  770 }
  771 
  772 void MacroAssembler::align(uint modulus, uint target) {
  773   if (target % modulus != 0) {
  774     nop(modulus - (target % modulus));
  775   }
  776 }
  777 
  778 void MacroAssembler::push_f(XMMRegister r) {
  779   subptr(rsp, wordSize);
  780   movflt(Address(rsp, 0), r);
  781 }
  782 
  783 void MacroAssembler::pop_f(XMMRegister r) {
  784   movflt(r, Address(rsp, 0));
  785   addptr(rsp, wordSize);
  786 }
  787 
  788 void MacroAssembler::push_d(XMMRegister r) {
  789   subptr(rsp, 2 * wordSize);
  790   movdbl(Address(rsp, 0), r);
  791 }
  792 
  793 void MacroAssembler::pop_d(XMMRegister r) {
  794   movdbl(r, Address(rsp, 0));
  795   addptr(rsp, 2 * Interpreter::stackElementSize);
  796 }
  797 
  798 void MacroAssembler::push_ppx(Register src) {
  799   if (VM_Version::supports_apx_f()) {
  800     pushp(src);
  801   } else {
  802     Assembler::push(src);
  803   }
  804 }
  805 
  806 void MacroAssembler::pop_ppx(Register dst) {
  807   if (VM_Version::supports_apx_f()) {
  808     popp(dst);
  809   } else {
  810     Assembler::pop(dst);
  811   }
  812 }
  813 
  814 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  815   // Used in sign-masking with aligned address.
  816   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  817   assert(rscratch != noreg || always_reachable(src), "missing");
  818 
  819   if (UseAVX > 2 &&
  820       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
  821       (dst->encoding() >= 16)) {
  822     vpand(dst, dst, src, AVX_512bit, rscratch);
  823   } else if (reachable(src)) {
  824     Assembler::andpd(dst, as_Address(src));
  825   } else {
  826     lea(rscratch, src);
  827     Assembler::andpd(dst, Address(rscratch, 0));
  828   }
  829 }
  830 
  831 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
  832   // Used in sign-masking with aligned address.
  833   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  834   assert(rscratch != noreg || always_reachable(src), "missing");
  835 
  836   if (reachable(src)) {
  837     Assembler::andps(dst, as_Address(src));
  838   } else {
  839     lea(rscratch, src);
  840     Assembler::andps(dst, Address(rscratch, 0));
  841   }
  842 }
  843 
  844 void MacroAssembler::andptr(Register dst, int32_t imm32) {
  845   andq(dst, imm32);
  846 }
  847 
  848 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
  849   assert(rscratch != noreg || always_reachable(src), "missing");
  850 
  851   if (reachable(src)) {
  852     andq(dst, as_Address(src));
  853   } else {
  854     lea(rscratch, src);
  855     andq(dst, Address(rscratch, 0));
  856   }
  857 }
  858 
  859 void MacroAssembler::atomic_incl(Address counter_addr) {
  860   lock();
  861   incrementl(counter_addr);
  862 }
  863 
  864 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
  865   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  866 
  867   if (reachable(counter_addr)) {
  868     atomic_incl(as_Address(counter_addr));
  869   } else {
  870     lea(rscratch, counter_addr);
  871     atomic_incl(Address(rscratch, 0));
  872   }
  873 }
  874 
  875 void MacroAssembler::atomic_incq(Address counter_addr) {
  876   lock();
  877   incrementq(counter_addr);
  878 }
  879 
  880 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
  881   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  882 
  883   if (reachable(counter_addr)) {
  884     atomic_incq(as_Address(counter_addr));
  885   } else {
  886     lea(rscratch, counter_addr);
  887     atomic_incq(Address(rscratch, 0));
  888   }
  889 }
  890 
  891 // Writes to stack successive pages until offset reached to check for
  892 // stack overflow + shadow pages.  This clobbers tmp.
  893 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
  894   movptr(tmp, rsp);
  895   // Bang stack for total size given plus shadow page size.
  896   // Bang one page at a time because large size can bang beyond yellow and
  897   // red zones.
  898   Label loop;
  899   bind(loop);
  900   movl(Address(tmp, (-(int)os::vm_page_size())), size );
  901   subptr(tmp, (int)os::vm_page_size());
  902   subl(size, (int)os::vm_page_size());
  903   jcc(Assembler::greater, loop);
  904 
  905   // Bang down shadow pages too.
  906   // At this point, (tmp-0) is the last address touched, so don't
  907   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
  908   // was post-decremented.)  Skip this address by starting at i=1, and
  909   // touch a few more pages below.  N.B.  It is important to touch all
  910   // the way down including all pages in the shadow zone.
  911   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
  912     // this could be any sized move but this is can be a debugging crumb
  913     // so the bigger the better.
  914     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
  915   }
  916 }
  917 
  918 void MacroAssembler::reserved_stack_check() {
  919   // testing if reserved zone needs to be enabled
  920   Label no_reserved_zone_enabling;
  921 
  922   cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
  923   jcc(Assembler::below, no_reserved_zone_enabling);
  924 
  925   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
  926   jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
  927   should_not_reach_here();
  928 
  929   bind(no_reserved_zone_enabling);
  930 }
  931 
  932 void MacroAssembler::c2bool(Register x) {
  933   // implements x == 0 ? 0 : 1
  934   // note: must only look at least-significant byte of x
  935   //       since C-style booleans are stored in one byte
  936   //       only! (was bug)
  937   andl(x, 0xFF);
  938   setb(Assembler::notZero, x);
  939 }
  940 
  941 // Wouldn't need if AddressLiteral version had new name
  942 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
  943   Assembler::call(L, rtype);
  944 }
  945 
  946 void MacroAssembler::call(Register entry) {
  947   Assembler::call(entry);
  948 }
  949 
  950 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
  951   assert(rscratch != noreg || always_reachable(entry), "missing");
  952 
  953   if (reachable(entry)) {
  954     Assembler::call_literal(entry.target(), entry.rspec());
  955   } else {
  956     lea(rscratch, entry);
  957     Assembler::call(rscratch);
  958   }
  959 }
  960 
  961 void MacroAssembler::ic_call(address entry, jint method_index) {
  962   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
  963   // Needs full 64-bit immediate for later patching.
  964   Assembler::mov64(rax, (int64_t)Universe::non_oop_word());
  965   call(AddressLiteral(entry, rh));
  966 }
  967 
  968 int MacroAssembler::ic_check_size() {
  969   return UseCompactObjectHeaders ? 17 : 14;
  970 }
  971 
  972 int MacroAssembler::ic_check(int end_alignment) {
  973   Register receiver = j_rarg0;
  974   Register data = rax;
  975   Register temp = rscratch1;
  976 
  977   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
  978   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
  979   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
  980   // before the inline cache check here, and not after
  981   align(end_alignment, offset() + ic_check_size());
  982 
  983   int uep_offset = offset();
  984 
  985   if (UseCompactObjectHeaders) {
  986     load_narrow_klass_compact(temp, receiver);
  987     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  988   } else {
  989     movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
  990     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  991   }
  992 
  993   // if inline cache check fails, then jump to runtime routine
  994   jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
  995   assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
  996 
  997   return uep_offset;
  998 }
  999 
 1000 void MacroAssembler::emit_static_call_stub() {
 1001   // Static stub relocation also tags the Method* in the code-stream.
 1002   mov_metadata(rbx, (Metadata*) nullptr);  // Method is zapped till fixup time.
 1003   // This is recognized as unresolved by relocs/nativeinst/ic code.
 1004   jump(RuntimeAddress(pc()));
 1005 }
 1006 
 1007 // Implementation of call_VM versions
 1008 
 1009 void MacroAssembler::call_VM(Register oop_result,
 1010                              address entry_point,
 1011                              bool check_exceptions) {
 1012   Label C, E;
 1013   call(C, relocInfo::none);
 1014   jmp(E);
 1015 
 1016   bind(C);
 1017   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
 1018   ret(0);
 1019 
 1020   bind(E);
 1021 }
 1022 
 1023 void MacroAssembler::call_VM(Register oop_result,
 1024                              address entry_point,
 1025                              Register arg_1,
 1026                              bool check_exceptions) {
 1027   Label C, E;
 1028   call(C, relocInfo::none);
 1029   jmp(E);
 1030 
 1031   bind(C);
 1032   pass_arg1(this, arg_1);
 1033   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
 1034   ret(0);
 1035 
 1036   bind(E);
 1037 }
 1038 
 1039 void MacroAssembler::call_VM(Register oop_result,
 1040                              address entry_point,
 1041                              Register arg_1,
 1042                              Register arg_2,
 1043                              bool check_exceptions) {
 1044   Label C, E;
 1045   call(C, relocInfo::none);
 1046   jmp(E);
 1047 
 1048   bind(C);
 1049 
 1050   assert_different_registers(arg_1, c_rarg2);
 1051 
 1052   pass_arg2(this, arg_2);
 1053   pass_arg1(this, arg_1);
 1054   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
 1055   ret(0);
 1056 
 1057   bind(E);
 1058 }
 1059 
 1060 void MacroAssembler::call_VM(Register oop_result,
 1061                              address entry_point,
 1062                              Register arg_1,
 1063                              Register arg_2,
 1064                              Register arg_3,
 1065                              bool check_exceptions) {
 1066   Label C, E;
 1067   call(C, relocInfo::none);
 1068   jmp(E);
 1069 
 1070   bind(C);
 1071 
 1072   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1073   assert_different_registers(arg_2, c_rarg3);
 1074   pass_arg3(this, arg_3);
 1075   pass_arg2(this, arg_2);
 1076   pass_arg1(this, arg_1);
 1077   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
 1078   ret(0);
 1079 
 1080   bind(E);
 1081 }
 1082 
 1083 void MacroAssembler::call_VM(Register oop_result,
 1084                              Register last_java_sp,
 1085                              address entry_point,
 1086                              int number_of_arguments,
 1087                              bool check_exceptions) {
 1088   call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1089 }
 1090 
 1091 void MacroAssembler::call_VM(Register oop_result,
 1092                              Register last_java_sp,
 1093                              address entry_point,
 1094                              Register arg_1,
 1095                              bool check_exceptions) {
 1096   pass_arg1(this, arg_1);
 1097   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1098 }
 1099 
 1100 void MacroAssembler::call_VM(Register oop_result,
 1101                              Register last_java_sp,
 1102                              address entry_point,
 1103                              Register arg_1,
 1104                              Register arg_2,
 1105                              bool check_exceptions) {
 1106 
 1107   assert_different_registers(arg_1, c_rarg2);
 1108   pass_arg2(this, arg_2);
 1109   pass_arg1(this, arg_1);
 1110   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1111 }
 1112 
 1113 void MacroAssembler::call_VM(Register oop_result,
 1114                              Register last_java_sp,
 1115                              address entry_point,
 1116                              Register arg_1,
 1117                              Register arg_2,
 1118                              Register arg_3,
 1119                              bool check_exceptions) {
 1120   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1121   assert_different_registers(arg_2, c_rarg3);
 1122   pass_arg3(this, arg_3);
 1123   pass_arg2(this, arg_2);
 1124   pass_arg1(this, arg_1);
 1125   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1126 }
 1127 
 1128 void MacroAssembler::super_call_VM(Register oop_result,
 1129                                    Register last_java_sp,
 1130                                    address entry_point,
 1131                                    int number_of_arguments,
 1132                                    bool check_exceptions) {
 1133   MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1134 }
 1135 
 1136 void MacroAssembler::super_call_VM(Register oop_result,
 1137                                    Register last_java_sp,
 1138                                    address entry_point,
 1139                                    Register arg_1,
 1140                                    bool check_exceptions) {
 1141   pass_arg1(this, arg_1);
 1142   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1143 }
 1144 
 1145 void MacroAssembler::super_call_VM(Register oop_result,
 1146                                    Register last_java_sp,
 1147                                    address entry_point,
 1148                                    Register arg_1,
 1149                                    Register arg_2,
 1150                                    bool check_exceptions) {
 1151 
 1152   assert_different_registers(arg_1, c_rarg2);
 1153   pass_arg2(this, arg_2);
 1154   pass_arg1(this, arg_1);
 1155   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1156 }
 1157 
 1158 void MacroAssembler::super_call_VM(Register oop_result,
 1159                                    Register last_java_sp,
 1160                                    address entry_point,
 1161                                    Register arg_1,
 1162                                    Register arg_2,
 1163                                    Register arg_3,
 1164                                    bool check_exceptions) {
 1165   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1166   assert_different_registers(arg_2, c_rarg3);
 1167   pass_arg3(this, arg_3);
 1168   pass_arg2(this, arg_2);
 1169   pass_arg1(this, arg_1);
 1170   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1171 }
 1172 
 1173 void MacroAssembler::call_VM_base(Register oop_result,
 1174                                   Register last_java_sp,
 1175                                   address  entry_point,
 1176                                   int      number_of_arguments,
 1177                                   bool     check_exceptions) {
 1178   Register java_thread = r15_thread;
 1179 
 1180   // determine last_java_sp register
 1181   if (!last_java_sp->is_valid()) {
 1182     last_java_sp = rsp;
 1183   }
 1184   // debugging support
 1185   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 1186 #ifdef ASSERT
 1187   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 1188   // r12 is the heapbase.
 1189   if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 1190 #endif // ASSERT
 1191 
 1192   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 1193   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 1194 
 1195   // push java thread (becomes first argument of C function)
 1196 
 1197   mov(c_rarg0, r15_thread);
 1198 
 1199   // set last Java frame before call
 1200   assert(last_java_sp != rbp, "can't use ebp/rbp");
 1201 
 1202   // Only interpreter should have to set fp
 1203   set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
 1204 
 1205   // do the call, remove parameters
 1206   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 1207 
 1208 #ifdef ASSERT
 1209   // Check that thread register is not clobbered.
 1210   guarantee(java_thread != rax, "change this code");
 1211   push(rax);
 1212   { Label L;
 1213     get_thread_slow(rax);
 1214     cmpptr(java_thread, rax);
 1215     jcc(Assembler::equal, L);
 1216     STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
 1217     bind(L);
 1218   }
 1219   pop(rax);
 1220 #endif
 1221 
 1222   // reset last Java frame
 1223   // Only interpreter should have to clear fp
 1224   reset_last_Java_frame(true);
 1225 
 1226    // C++ interp handles this in the interpreter
 1227   check_and_handle_popframe();
 1228   check_and_handle_earlyret();
 1229 
 1230   if (check_exceptions) {
 1231     // check for pending exceptions (java_thread is set upon return)
 1232     cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 1233     // This used to conditionally jump to forward_exception however it is
 1234     // possible if we relocate that the branch will not reach. So we must jump
 1235     // around so we can always reach
 1236 
 1237     Label ok;
 1238     jcc(Assembler::equal, ok);
 1239     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 1240     bind(ok);
 1241   }
 1242 
 1243   // get oop result if there is one and reset the value in the thread
 1244   if (oop_result->is_valid()) {
 1245     get_vm_result_oop(oop_result);
 1246   }
 1247 }
 1248 
 1249 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 1250   // Calculate the value for last_Java_sp somewhat subtle.
 1251   // call_VM does an intermediate call which places a return address on
 1252   // the stack just under the stack pointer as the user finished with it.
 1253   // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
 1254 
 1255   // We've pushed one address, correct last_Java_sp
 1256   lea(rax, Address(rsp, wordSize));
 1257 
 1258   call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
 1259 }
 1260 
 1261 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
 1262 void MacroAssembler::call_VM_leaf0(address entry_point) {
 1263   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 1264 }
 1265 
 1266 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
 1267   call_VM_leaf_base(entry_point, number_of_arguments);
 1268 }
 1269 
 1270 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
 1271   pass_arg0(this, arg_0);
 1272   call_VM_leaf(entry_point, 1);
 1273 }
 1274 
 1275 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1276 
 1277   assert_different_registers(arg_0, c_rarg1);
 1278   pass_arg1(this, arg_1);
 1279   pass_arg0(this, arg_0);
 1280   call_VM_leaf(entry_point, 2);
 1281 }
 1282 
 1283 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1284   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1285   assert_different_registers(arg_1, c_rarg2);
 1286   pass_arg2(this, arg_2);
 1287   pass_arg1(this, arg_1);
 1288   pass_arg0(this, arg_0);
 1289   call_VM_leaf(entry_point, 3);
 1290 }
 1291 
 1292 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1293   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1294   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1295   assert_different_registers(arg_2, c_rarg3);
 1296   pass_arg3(this, arg_3);
 1297   pass_arg2(this, arg_2);
 1298   pass_arg1(this, arg_1);
 1299   pass_arg0(this, arg_0);
 1300   call_VM_leaf(entry_point, 3);
 1301 }
 1302 
 1303 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1304   pass_arg0(this, arg_0);
 1305   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1306 }
 1307 
 1308 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1309   assert_different_registers(arg_0, c_rarg1);
 1310   pass_arg1(this, arg_1);
 1311   pass_arg0(this, arg_0);
 1312   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1313 }
 1314 
 1315 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1316   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1317   assert_different_registers(arg_1, c_rarg2);
 1318   pass_arg2(this, arg_2);
 1319   pass_arg1(this, arg_1);
 1320   pass_arg0(this, arg_0);
 1321   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1322 }
 1323 
 1324 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1325   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1326   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1327   assert_different_registers(arg_2, c_rarg3);
 1328   pass_arg3(this, arg_3);
 1329   pass_arg2(this, arg_2);
 1330   pass_arg1(this, arg_1);
 1331   pass_arg0(this, arg_0);
 1332   MacroAssembler::call_VM_leaf_base(entry_point, 4);
 1333 }
 1334 
 1335 void MacroAssembler::get_vm_result_oop(Register oop_result) {
 1336   movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
 1337   movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
 1338   verify_oop_msg(oop_result, "broken oop in call_VM_base");
 1339 }
 1340 
 1341 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
 1342   movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
 1343   movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
 1344 }
 1345 
 1346 void MacroAssembler::check_and_handle_earlyret() {
 1347 }
 1348 
 1349 void MacroAssembler::check_and_handle_popframe() {
 1350 }
 1351 
 1352 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
 1353   assert(rscratch != noreg || always_reachable(src1), "missing");
 1354 
 1355   if (reachable(src1)) {
 1356     cmpl(as_Address(src1), imm);
 1357   } else {
 1358     lea(rscratch, src1);
 1359     cmpl(Address(rscratch, 0), imm);
 1360   }
 1361 }
 1362 
 1363 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
 1364   assert(!src2.is_lval(), "use cmpptr");
 1365   assert(rscratch != noreg || always_reachable(src2), "missing");
 1366 
 1367   if (reachable(src2)) {
 1368     cmpl(src1, as_Address(src2));
 1369   } else {
 1370     lea(rscratch, src2);
 1371     cmpl(src1, Address(rscratch, 0));
 1372   }
 1373 }
 1374 
 1375 void MacroAssembler::cmp32(Register src1, int32_t imm) {
 1376   Assembler::cmpl(src1, imm);
 1377 }
 1378 
 1379 void MacroAssembler::cmp32(Register src1, Address src2) {
 1380   Assembler::cmpl(src1, src2);
 1381 }
 1382 
 1383 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1384   ucomisd(opr1, opr2);
 1385 
 1386   Label L;
 1387   if (unordered_is_less) {
 1388     movl(dst, -1);
 1389     jcc(Assembler::parity, L);
 1390     jcc(Assembler::below , L);
 1391     movl(dst, 0);
 1392     jcc(Assembler::equal , L);
 1393     increment(dst);
 1394   } else { // unordered is greater
 1395     movl(dst, 1);
 1396     jcc(Assembler::parity, L);
 1397     jcc(Assembler::above , L);
 1398     movl(dst, 0);
 1399     jcc(Assembler::equal , L);
 1400     decrementl(dst);
 1401   }
 1402   bind(L);
 1403 }
 1404 
 1405 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1406   ucomiss(opr1, opr2);
 1407 
 1408   Label L;
 1409   if (unordered_is_less) {
 1410     movl(dst, -1);
 1411     jcc(Assembler::parity, L);
 1412     jcc(Assembler::below , L);
 1413     movl(dst, 0);
 1414     jcc(Assembler::equal , L);
 1415     increment(dst);
 1416   } else { // unordered is greater
 1417     movl(dst, 1);
 1418     jcc(Assembler::parity, L);
 1419     jcc(Assembler::above , L);
 1420     movl(dst, 0);
 1421     jcc(Assembler::equal , L);
 1422     decrementl(dst);
 1423   }
 1424   bind(L);
 1425 }
 1426 
 1427 
 1428 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
 1429   assert(rscratch != noreg || always_reachable(src1), "missing");
 1430 
 1431   if (reachable(src1)) {
 1432     cmpb(as_Address(src1), imm);
 1433   } else {
 1434     lea(rscratch, src1);
 1435     cmpb(Address(rscratch, 0), imm);
 1436   }
 1437 }
 1438 
 1439 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
 1440   assert(rscratch != noreg || always_reachable(src2), "missing");
 1441 
 1442   if (src2.is_lval()) {
 1443     movptr(rscratch, src2);
 1444     Assembler::cmpq(src1, rscratch);
 1445   } else if (reachable(src2)) {
 1446     cmpq(src1, as_Address(src2));
 1447   } else {
 1448     lea(rscratch, src2);
 1449     Assembler::cmpq(src1, Address(rscratch, 0));
 1450   }
 1451 }
 1452 
 1453 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
 1454   assert(src2.is_lval(), "not a mem-mem compare");
 1455   // moves src2's literal address
 1456   movptr(rscratch, src2);
 1457   Assembler::cmpq(src1, rscratch);
 1458 }
 1459 
 1460 void MacroAssembler::cmpoop(Register src1, Register src2) {
 1461   cmpptr(src1, src2);
 1462 }
 1463 
 1464 void MacroAssembler::cmpoop(Register src1, Address src2) {
 1465   cmpptr(src1, src2);
 1466 }
 1467 
 1468 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
 1469   movoop(rscratch, src2);
 1470   cmpptr(src1, rscratch);
 1471 }
 1472 
 1473 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
 1474   assert(rscratch != noreg || always_reachable(adr), "missing");
 1475 
 1476   if (reachable(adr)) {
 1477     lock();
 1478     cmpxchgptr(reg, as_Address(adr));
 1479   } else {
 1480     lea(rscratch, adr);
 1481     lock();
 1482     cmpxchgptr(reg, Address(rscratch, 0));
 1483   }
 1484 }
 1485 
 1486 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
 1487   cmpxchgq(reg, adr);
 1488 }
 1489 
 1490 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1491   assert(rscratch != noreg || always_reachable(src), "missing");
 1492 
 1493   if (reachable(src)) {
 1494     Assembler::comisd(dst, as_Address(src));
 1495   } else {
 1496     lea(rscratch, src);
 1497     Assembler::comisd(dst, Address(rscratch, 0));
 1498   }
 1499 }
 1500 
 1501 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1502   assert(rscratch != noreg || always_reachable(src), "missing");
 1503 
 1504   if (reachable(src)) {
 1505     Assembler::comiss(dst, as_Address(src));
 1506   } else {
 1507     lea(rscratch, src);
 1508     Assembler::comiss(dst, Address(rscratch, 0));
 1509   }
 1510 }
 1511 
 1512 
 1513 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
 1514   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1515 
 1516   Condition negated_cond = negate_condition(cond);
 1517   Label L;
 1518   jcc(negated_cond, L);
 1519   pushf(); // Preserve flags
 1520   atomic_incl(counter_addr, rscratch);
 1521   popf();
 1522   bind(L);
 1523 }
 1524 
 1525 int MacroAssembler::corrected_idivl(Register reg) {
 1526   // Full implementation of Java idiv and irem; checks for
 1527   // special case as described in JVM spec., p.243 & p.271.
 1528   // The function returns the (pc) offset of the idivl
 1529   // instruction - may be needed for implicit exceptions.
 1530   //
 1531   //         normal case                           special case
 1532   //
 1533   // input : rax,: dividend                         min_int
 1534   //         reg: divisor   (may not be rax,/rdx)   -1
 1535   //
 1536   // output: rax,: quotient  (= rax, idiv reg)       min_int
 1537   //         rdx: remainder (= rax, irem reg)       0
 1538   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
 1539   const int min_int = 0x80000000;
 1540   Label normal_case, special_case;
 1541 
 1542   // check for special case
 1543   cmpl(rax, min_int);
 1544   jcc(Assembler::notEqual, normal_case);
 1545   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
 1546   cmpl(reg, -1);
 1547   jcc(Assembler::equal, special_case);
 1548 
 1549   // handle normal case
 1550   bind(normal_case);
 1551   cdql();
 1552   int idivl_offset = offset();
 1553   idivl(reg);
 1554 
 1555   // normal and special case exit
 1556   bind(special_case);
 1557 
 1558   return idivl_offset;
 1559 }
 1560 
 1561 
 1562 
 1563 void MacroAssembler::decrementl(Register reg, int value) {
 1564   if (value == min_jint) {subl(reg, value) ; return; }
 1565   if (value <  0) { incrementl(reg, -value); return; }
 1566   if (value == 0) {                        ; return; }
 1567   if (value == 1 && UseIncDec) { decl(reg) ; return; }
 1568   /* else */      { subl(reg, value)       ; return; }
 1569 }
 1570 
 1571 void MacroAssembler::decrementl(Address dst, int value) {
 1572   if (value == min_jint) {subl(dst, value) ; return; }
 1573   if (value <  0) { incrementl(dst, -value); return; }
 1574   if (value == 0) {                        ; return; }
 1575   if (value == 1 && UseIncDec) { decl(dst) ; return; }
 1576   /* else */      { subl(dst, value)       ; return; }
 1577 }
 1578 
 1579 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
 1580   assert(shift_value > 0, "illegal shift value");
 1581   Label _is_positive;
 1582   testl (reg, reg);
 1583   jcc (Assembler::positive, _is_positive);
 1584   int offset = (1 << shift_value) - 1 ;
 1585 
 1586   if (offset == 1) {
 1587     incrementl(reg);
 1588   } else {
 1589     addl(reg, offset);
 1590   }
 1591 
 1592   bind (_is_positive);
 1593   sarl(reg, shift_value);
 1594 }
 1595 
 1596 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1597   assert(rscratch != noreg || always_reachable(src), "missing");
 1598 
 1599   if (reachable(src)) {
 1600     Assembler::divsd(dst, as_Address(src));
 1601   } else {
 1602     lea(rscratch, src);
 1603     Assembler::divsd(dst, Address(rscratch, 0));
 1604   }
 1605 }
 1606 
 1607 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1608   assert(rscratch != noreg || always_reachable(src), "missing");
 1609 
 1610   if (reachable(src)) {
 1611     Assembler::divss(dst, as_Address(src));
 1612   } else {
 1613     lea(rscratch, src);
 1614     Assembler::divss(dst, Address(rscratch, 0));
 1615   }
 1616 }
 1617 
 1618 void MacroAssembler::enter() {
 1619   push(rbp);
 1620   mov(rbp, rsp);
 1621 }
 1622 
 1623 void MacroAssembler::post_call_nop() {
 1624   if (!Continuations::enabled()) {
 1625     return;
 1626   }
 1627   InstructionMark im(this);
 1628   relocate(post_call_nop_Relocation::spec());
 1629   InlineSkippedInstructionsCounter skipCounter(this);
 1630   emit_int8((uint8_t)0x0f);
 1631   emit_int8((uint8_t)0x1f);
 1632   emit_int8((uint8_t)0x84);
 1633   emit_int8((uint8_t)0x00);
 1634   emit_int32(0x00);
 1635 }
 1636 
 1637 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1638   assert(rscratch != noreg || always_reachable(src), "missing");
 1639   if (reachable(src)) {
 1640     Assembler::mulpd(dst, as_Address(src));
 1641   } else {
 1642     lea(rscratch, src);
 1643     Assembler::mulpd(dst, Address(rscratch, 0));
 1644   }
 1645 }
 1646 
 1647 // dst = c = a * b + c
 1648 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1649   Assembler::vfmadd231sd(c, a, b);
 1650   if (dst != c) {
 1651     movdbl(dst, c);
 1652   }
 1653 }
 1654 
 1655 // dst = c = a * b + c
 1656 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1657   Assembler::vfmadd231ss(c, a, b);
 1658   if (dst != c) {
 1659     movflt(dst, c);
 1660   }
 1661 }
 1662 
 1663 // dst = c = a * b + c
 1664 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1665   Assembler::vfmadd231pd(c, a, b, vector_len);
 1666   if (dst != c) {
 1667     vmovdqu(dst, c);
 1668   }
 1669 }
 1670 
 1671 // dst = c = a * b + c
 1672 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1673   Assembler::vfmadd231ps(c, a, b, vector_len);
 1674   if (dst != c) {
 1675     vmovdqu(dst, c);
 1676   }
 1677 }
 1678 
 1679 // dst = c = a * b + c
 1680 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1681   Assembler::vfmadd231pd(c, a, b, vector_len);
 1682   if (dst != c) {
 1683     vmovdqu(dst, c);
 1684   }
 1685 }
 1686 
 1687 // dst = c = a * b + c
 1688 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1689   Assembler::vfmadd231ps(c, a, b, vector_len);
 1690   if (dst != c) {
 1691     vmovdqu(dst, c);
 1692   }
 1693 }
 1694 
 1695 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
 1696   assert(rscratch != noreg || always_reachable(dst), "missing");
 1697 
 1698   if (reachable(dst)) {
 1699     incrementl(as_Address(dst));
 1700   } else {
 1701     lea(rscratch, dst);
 1702     incrementl(Address(rscratch, 0));
 1703   }
 1704 }
 1705 
 1706 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
 1707   incrementl(as_Address(dst, rscratch));
 1708 }
 1709 
 1710 void MacroAssembler::incrementl(Register reg, int value) {
 1711   if (value == min_jint) {addl(reg, value) ; return; }
 1712   if (value <  0) { decrementl(reg, -value); return; }
 1713   if (value == 0) {                        ; return; }
 1714   if (value == 1 && UseIncDec) { incl(reg) ; return; }
 1715   /* else */      { addl(reg, value)       ; return; }
 1716 }
 1717 
 1718 void MacroAssembler::incrementl(Address dst, int value) {
 1719   if (value == min_jint) {addl(dst, value) ; return; }
 1720   if (value <  0) { decrementl(dst, -value); return; }
 1721   if (value == 0) {                        ; return; }
 1722   if (value == 1 && UseIncDec) { incl(dst) ; return; }
 1723   /* else */      { addl(dst, value)       ; return; }
 1724 }
 1725 
 1726 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
 1727   assert(rscratch != noreg || always_reachable(dst), "missing");
 1728   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
 1729   if (reachable(dst)) {
 1730     jmp_literal(dst.target(), dst.rspec());
 1731   } else {
 1732     lea(rscratch, dst);
 1733     jmp(rscratch);
 1734   }
 1735 }
 1736 
 1737 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
 1738   assert(rscratch != noreg || always_reachable(dst), "missing");
 1739   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump_cc");
 1740   if (reachable(dst)) {
 1741     InstructionMark im(this);
 1742     relocate(dst.reloc());
 1743     const int short_size = 2;
 1744     const int long_size = 6;
 1745     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
 1746     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
 1747       // 0111 tttn #8-bit disp
 1748       emit_int8(0x70 | cc);
 1749       emit_int8((offs - short_size) & 0xFF);
 1750     } else {
 1751       // 0000 1111 1000 tttn #32-bit disp
 1752       emit_int8(0x0F);
 1753       emit_int8((unsigned char)(0x80 | cc));
 1754       emit_int32(offs - long_size);
 1755     }
 1756   } else {
 1757 #ifdef ASSERT
 1758     warning("reversing conditional branch");
 1759 #endif /* ASSERT */
 1760     Label skip;
 1761     jccb(reverse[cc], skip);
 1762     lea(rscratch, dst);
 1763     Assembler::jmp(rscratch);
 1764     bind(skip);
 1765   }
 1766 }
 1767 
 1768 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
 1769   ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 1770   assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
 1771 
 1772   stmxcsr(mxcsr_save);
 1773   movl(tmp, mxcsr_save);
 1774   if (EnableX86ECoreOpts) {
 1775     // The mxcsr_std has status bits set for performance on ECore
 1776     orl(tmp, 0x003f);
 1777   } else {
 1778     // Mask out status bits (only check control and mask bits)
 1779     andl(tmp, 0xFFC0);
 1780   }
 1781   cmp32(tmp, mxcsr_std, rscratch);
 1782 }
 1783 
 1784 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
 1785   assert(rscratch != noreg || always_reachable(src), "missing");
 1786 
 1787   if (reachable(src)) {
 1788     Assembler::ldmxcsr(as_Address(src));
 1789   } else {
 1790     lea(rscratch, src);
 1791     Assembler::ldmxcsr(Address(rscratch, 0));
 1792   }
 1793 }
 1794 
 1795 int MacroAssembler::load_signed_byte(Register dst, Address src) {
 1796   int off = offset();
 1797   movsbl(dst, src); // movsxb
 1798   return off;
 1799 }
 1800 
 1801 // Note: load_signed_short used to be called load_signed_word.
 1802 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
 1803 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
 1804 // The term "word" in HotSpot means a 32- or 64-bit machine word.
 1805 int MacroAssembler::load_signed_short(Register dst, Address src) {
 1806   // This is dubious to me since it seems safe to do a signed 16 => 64 bit
 1807   // version but this is what 64bit has always done. This seems to imply
 1808   // that users are only using 32bits worth.
 1809   int off = offset();
 1810   movswl(dst, src); // movsxw
 1811   return off;
 1812 }
 1813 
 1814 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
 1815   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1816   // and "3.9 Partial Register Penalties", p. 22).
 1817   int off = offset();
 1818   movzbl(dst, src); // movzxb
 1819   return off;
 1820 }
 1821 
 1822 // Note: load_unsigned_short used to be called load_unsigned_word.
 1823 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
 1824   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1825   // and "3.9 Partial Register Penalties", p. 22).
 1826   int off = offset();
 1827   movzwl(dst, src); // movzxw
 1828   return off;
 1829 }
 1830 
 1831 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
 1832   switch (size_in_bytes) {
 1833   case  8:  movq(dst, src); break;
 1834   case  4:  movl(dst, src); break;
 1835   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
 1836   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
 1837   default:  ShouldNotReachHere();
 1838   }
 1839 }
 1840 
 1841 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
 1842   switch (size_in_bytes) {
 1843   case  8:  movq(dst, src); break;
 1844   case  4:  movl(dst, src); break;
 1845   case  2:  movw(dst, src); break;
 1846   case  1:  movb(dst, src); break;
 1847   default:  ShouldNotReachHere();
 1848   }
 1849 }
 1850 
 1851 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
 1852   assert(rscratch != noreg || always_reachable(dst), "missing");
 1853 
 1854   if (reachable(dst)) {
 1855     movl(as_Address(dst), src);
 1856   } else {
 1857     lea(rscratch, dst);
 1858     movl(Address(rscratch, 0), src);
 1859   }
 1860 }
 1861 
 1862 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
 1863   if (reachable(src)) {
 1864     movl(dst, as_Address(src));
 1865   } else {
 1866     lea(dst, src);
 1867     movl(dst, Address(dst, 0));
 1868   }
 1869 }
 1870 
 1871 // C++ bool manipulation
 1872 
 1873 void MacroAssembler::movbool(Register dst, Address src) {
 1874   if(sizeof(bool) == 1)
 1875     movb(dst, src);
 1876   else if(sizeof(bool) == 2)
 1877     movw(dst, src);
 1878   else if(sizeof(bool) == 4)
 1879     movl(dst, src);
 1880   else
 1881     // unsupported
 1882     ShouldNotReachHere();
 1883 }
 1884 
 1885 void MacroAssembler::movbool(Address dst, bool boolconst) {
 1886   if(sizeof(bool) == 1)
 1887     movb(dst, (int) boolconst);
 1888   else if(sizeof(bool) == 2)
 1889     movw(dst, (int) boolconst);
 1890   else if(sizeof(bool) == 4)
 1891     movl(dst, (int) boolconst);
 1892   else
 1893     // unsupported
 1894     ShouldNotReachHere();
 1895 }
 1896 
 1897 void MacroAssembler::movbool(Address dst, Register src) {
 1898   if(sizeof(bool) == 1)
 1899     movb(dst, src);
 1900   else if(sizeof(bool) == 2)
 1901     movw(dst, src);
 1902   else if(sizeof(bool) == 4)
 1903     movl(dst, src);
 1904   else
 1905     // unsupported
 1906     ShouldNotReachHere();
 1907 }
 1908 
 1909 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1910   assert(rscratch != noreg || always_reachable(src), "missing");
 1911 
 1912   if (reachable(src)) {
 1913     movdl(dst, as_Address(src));
 1914   } else {
 1915     lea(rscratch, src);
 1916     movdl(dst, Address(rscratch, 0));
 1917   }
 1918 }
 1919 
 1920 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1921   assert(rscratch != noreg || always_reachable(src), "missing");
 1922 
 1923   if (reachable(src)) {
 1924     movq(dst, as_Address(src));
 1925   } else {
 1926     lea(rscratch, src);
 1927     movq(dst, Address(rscratch, 0));
 1928   }
 1929 }
 1930 
 1931 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1932   assert(rscratch != noreg || always_reachable(src), "missing");
 1933 
 1934   if (reachable(src)) {
 1935     if (UseXmmLoadAndClearUpper) {
 1936       movsd (dst, as_Address(src));
 1937     } else {
 1938       movlpd(dst, as_Address(src));
 1939     }
 1940   } else {
 1941     lea(rscratch, src);
 1942     if (UseXmmLoadAndClearUpper) {
 1943       movsd (dst, Address(rscratch, 0));
 1944     } else {
 1945       movlpd(dst, Address(rscratch, 0));
 1946     }
 1947   }
 1948 }
 1949 
 1950 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1951   assert(rscratch != noreg || always_reachable(src), "missing");
 1952 
 1953   if (reachable(src)) {
 1954     movss(dst, as_Address(src));
 1955   } else {
 1956     lea(rscratch, src);
 1957     movss(dst, Address(rscratch, 0));
 1958   }
 1959 }
 1960 
 1961 void MacroAssembler::movhlf(XMMRegister dst, XMMRegister src, Register rscratch) {
 1962   if (VM_Version::supports_avx10_2()) {
 1963     evmovw(dst, src);
 1964   } else {
 1965     assert(rscratch != noreg, "missing");
 1966     evmovw(rscratch, src);
 1967     evmovw(dst, rscratch);
 1968   }
 1969 }
 1970 
 1971 void MacroAssembler::mov64(Register dst, int64_t imm64) {
 1972   if (is_uimm32(imm64)) {
 1973     movl(dst, checked_cast<uint32_t>(imm64));
 1974   } else if (is_simm32(imm64)) {
 1975     movq(dst, checked_cast<int32_t>(imm64));
 1976   } else {
 1977     Assembler::mov64(dst, imm64);
 1978   }
 1979 }
 1980 
 1981 void MacroAssembler::mov64(Register dst, int64_t imm64, relocInfo::relocType rtype, int format) {
 1982   Assembler::mov64(dst, imm64, rtype, format);
 1983 }
 1984 
 1985 void MacroAssembler::movptr(Register dst, Register src) {
 1986   movq(dst, src);
 1987 }
 1988 
 1989 void MacroAssembler::movptr(Register dst, Address src) {
 1990   movq(dst, src);
 1991 }
 1992 
 1993 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 1994 void MacroAssembler::movptr(Register dst, intptr_t src) {
 1995   mov64(dst, src);
 1996 }
 1997 
 1998 void MacroAssembler::movptr(Address dst, Register src) {
 1999   movq(dst, src);
 2000 }
 2001 
 2002 void MacroAssembler::movptr(Address dst, int32_t src) {
 2003   movslq(dst, src);
 2004 }
 2005 
 2006 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 2007   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2008   Assembler::movdqu(dst, src);
 2009 }
 2010 
 2011 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 2012   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2013   Assembler::movdqu(dst, src);
 2014 }
 2015 
 2016 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2017   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2018   Assembler::movdqu(dst, src);
 2019 }
 2020 
 2021 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2022   assert(rscratch != noreg || always_reachable(src), "missing");
 2023 
 2024   if (reachable(src)) {
 2025     movdqu(dst, as_Address(src));
 2026   } else {
 2027     lea(rscratch, src);
 2028     movdqu(dst, Address(rscratch, 0));
 2029   }
 2030 }
 2031 
 2032 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2033   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2034   Assembler::vmovdqu(dst, src);
 2035 }
 2036 
 2037 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2038   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2039   Assembler::vmovdqu(dst, src);
 2040 }
 2041 
 2042 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2043   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2044   Assembler::vmovdqu(dst, src);
 2045 }
 2046 
 2047 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2048   assert(rscratch != noreg || always_reachable(src), "missing");
 2049 
 2050   if (reachable(src)) {
 2051     vmovdqu(dst, as_Address(src));
 2052   }
 2053   else {
 2054     lea(rscratch, src);
 2055     vmovdqu(dst, Address(rscratch, 0));
 2056   }
 2057 }
 2058 
 2059 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2060   assert(rscratch != noreg || always_reachable(src), "missing");
 2061 
 2062   if (vector_len == AVX_512bit) {
 2063     evmovdquq(dst, src, AVX_512bit, rscratch);
 2064   } else if (vector_len == AVX_256bit) {
 2065     vmovdqu(dst, src, rscratch);
 2066   } else {
 2067     movdqu(dst, src, rscratch);
 2068   }
 2069 }
 2070 
 2071 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
 2072   if (vector_len == AVX_512bit) {
 2073     evmovdquq(dst, src, AVX_512bit);
 2074   } else if (vector_len == AVX_256bit) {
 2075     vmovdqu(dst, src);
 2076   } else {
 2077     movdqu(dst, src);
 2078   }
 2079 }
 2080 
 2081 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
 2082   if (vector_len == AVX_512bit) {
 2083     evmovdquq(dst, src, AVX_512bit);
 2084   } else if (vector_len == AVX_256bit) {
 2085     vmovdqu(dst, src);
 2086   } else {
 2087     movdqu(dst, src);
 2088   }
 2089 }
 2090 
 2091 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
 2092   if (vector_len == AVX_512bit) {
 2093     evmovdquq(dst, src, AVX_512bit);
 2094   } else if (vector_len == AVX_256bit) {
 2095     vmovdqu(dst, src);
 2096   } else {
 2097     movdqu(dst, src);
 2098   }
 2099 }
 2100 
 2101 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2102   assert(rscratch != noreg || always_reachable(src), "missing");
 2103 
 2104   if (reachable(src)) {
 2105     vmovdqa(dst, as_Address(src));
 2106   }
 2107   else {
 2108     lea(rscratch, src);
 2109     vmovdqa(dst, Address(rscratch, 0));
 2110   }
 2111 }
 2112 
 2113 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2114   assert(rscratch != noreg || always_reachable(src), "missing");
 2115 
 2116   if (vector_len == AVX_512bit) {
 2117     evmovdqaq(dst, src, AVX_512bit, rscratch);
 2118   } else if (vector_len == AVX_256bit) {
 2119     vmovdqa(dst, src, rscratch);
 2120   } else {
 2121     movdqa(dst, src, rscratch);
 2122   }
 2123 }
 2124 
 2125 void MacroAssembler::kmov(KRegister dst, Address src) {
 2126   if (VM_Version::supports_avx512bw()) {
 2127     kmovql(dst, src);
 2128   } else {
 2129     assert(VM_Version::supports_evex(), "");
 2130     kmovwl(dst, src);
 2131   }
 2132 }
 2133 
 2134 void MacroAssembler::kmov(Address dst, KRegister src) {
 2135   if (VM_Version::supports_avx512bw()) {
 2136     kmovql(dst, src);
 2137   } else {
 2138     assert(VM_Version::supports_evex(), "");
 2139     kmovwl(dst, src);
 2140   }
 2141 }
 2142 
 2143 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2144   if (VM_Version::supports_avx512bw()) {
 2145     kmovql(dst, src);
 2146   } else {
 2147     assert(VM_Version::supports_evex(), "");
 2148     kmovwl(dst, src);
 2149   }
 2150 }
 2151 
 2152 void MacroAssembler::kmov(Register dst, KRegister src) {
 2153   if (VM_Version::supports_avx512bw()) {
 2154     kmovql(dst, src);
 2155   } else {
 2156     assert(VM_Version::supports_evex(), "");
 2157     kmovwl(dst, src);
 2158   }
 2159 }
 2160 
 2161 void MacroAssembler::kmov(KRegister dst, Register src) {
 2162   if (VM_Version::supports_avx512bw()) {
 2163     kmovql(dst, src);
 2164   } else {
 2165     assert(VM_Version::supports_evex(), "");
 2166     kmovwl(dst, src);
 2167   }
 2168 }
 2169 
 2170 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2171   assert(rscratch != noreg || always_reachable(src), "missing");
 2172 
 2173   if (reachable(src)) {
 2174     kmovql(dst, as_Address(src));
 2175   } else {
 2176     lea(rscratch, src);
 2177     kmovql(dst, Address(rscratch, 0));
 2178   }
 2179 }
 2180 
 2181 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2182   assert(rscratch != noreg || always_reachable(src), "missing");
 2183 
 2184   if (reachable(src)) {
 2185     kmovwl(dst, as_Address(src));
 2186   } else {
 2187     lea(rscratch, src);
 2188     kmovwl(dst, Address(rscratch, 0));
 2189   }
 2190 }
 2191 
 2192 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2193                                int vector_len, Register rscratch) {
 2194   assert(rscratch != noreg || always_reachable(src), "missing");
 2195 
 2196   if (reachable(src)) {
 2197     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2198   } else {
 2199     lea(rscratch, src);
 2200     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2201   }
 2202 }
 2203 
 2204 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2205                                int vector_len, Register rscratch) {
 2206   assert(rscratch != noreg || always_reachable(src), "missing");
 2207 
 2208   if (reachable(src)) {
 2209     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2210   } else {
 2211     lea(rscratch, src);
 2212     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2213   }
 2214 }
 2215 
 2216 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2217   assert(rscratch != noreg || always_reachable(src), "missing");
 2218 
 2219   if (reachable(src)) {
 2220     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2221   } else {
 2222     lea(rscratch, src);
 2223     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2224   }
 2225 }
 2226 
 2227 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2228   assert(rscratch != noreg || always_reachable(src), "missing");
 2229 
 2230   if (reachable(src)) {
 2231     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2232   } else {
 2233     lea(rscratch, src);
 2234     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2235   }
 2236 }
 2237 
 2238 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2239   assert(rscratch != noreg || always_reachable(src), "missing");
 2240 
 2241   if (reachable(src)) {
 2242     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2243   } else {
 2244     lea(rscratch, src);
 2245     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2246   }
 2247 }
 2248 
 2249 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2250   assert(rscratch != noreg || always_reachable(src), "missing");
 2251 
 2252   if (reachable(src)) {
 2253     Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
 2254   } else {
 2255     lea(rscratch, src);
 2256     Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2257   }
 2258 }
 2259 
 2260 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2261   assert(rscratch != noreg || always_reachable(src), "missing");
 2262 
 2263   if (reachable(src)) {
 2264     Assembler::evmovdqaq(dst, as_Address(src), vector_len);
 2265   } else {
 2266     lea(rscratch, src);
 2267     Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
 2268   }
 2269 }
 2270 
 2271 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2272   assert(rscratch != noreg || always_reachable(src), "missing");
 2273 
 2274   if (reachable(src)) {
 2275     Assembler::movapd(dst, as_Address(src));
 2276   } else {
 2277     lea(rscratch, src);
 2278     Assembler::movapd(dst, Address(rscratch, 0));
 2279   }
 2280 }
 2281 
 2282 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2283   assert(rscratch != noreg || always_reachable(src), "missing");
 2284 
 2285   if (reachable(src)) {
 2286     Assembler::movdqa(dst, as_Address(src));
 2287   } else {
 2288     lea(rscratch, src);
 2289     Assembler::movdqa(dst, Address(rscratch, 0));
 2290   }
 2291 }
 2292 
 2293 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2294   assert(rscratch != noreg || always_reachable(src), "missing");
 2295 
 2296   if (reachable(src)) {
 2297     Assembler::movsd(dst, as_Address(src));
 2298   } else {
 2299     lea(rscratch, src);
 2300     Assembler::movsd(dst, Address(rscratch, 0));
 2301   }
 2302 }
 2303 
 2304 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2305   assert(rscratch != noreg || always_reachable(src), "missing");
 2306 
 2307   if (reachable(src)) {
 2308     Assembler::movss(dst, as_Address(src));
 2309   } else {
 2310     lea(rscratch, src);
 2311     Assembler::movss(dst, Address(rscratch, 0));
 2312   }
 2313 }
 2314 
 2315 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2316   assert(rscratch != noreg || always_reachable(src), "missing");
 2317 
 2318   if (reachable(src)) {
 2319     Assembler::movddup(dst, as_Address(src));
 2320   } else {
 2321     lea(rscratch, src);
 2322     Assembler::movddup(dst, Address(rscratch, 0));
 2323   }
 2324 }
 2325 
 2326 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2327   assert(rscratch != noreg || always_reachable(src), "missing");
 2328 
 2329   if (reachable(src)) {
 2330     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2331   } else {
 2332     lea(rscratch, src);
 2333     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2334   }
 2335 }
 2336 
 2337 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2338   assert(rscratch != noreg || always_reachable(src), "missing");
 2339 
 2340   if (reachable(src)) {
 2341     Assembler::mulsd(dst, as_Address(src));
 2342   } else {
 2343     lea(rscratch, src);
 2344     Assembler::mulsd(dst, Address(rscratch, 0));
 2345   }
 2346 }
 2347 
 2348 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2349   assert(rscratch != noreg || always_reachable(src), "missing");
 2350 
 2351   if (reachable(src)) {
 2352     Assembler::mulss(dst, as_Address(src));
 2353   } else {
 2354     lea(rscratch, src);
 2355     Assembler::mulss(dst, Address(rscratch, 0));
 2356   }
 2357 }
 2358 
 2359 void MacroAssembler::null_check(Register reg, int offset) {
 2360   if (needs_explicit_null_check(offset)) {
 2361     // provoke OS null exception if reg is null by
 2362     // accessing M[reg] w/o changing any (non-CC) registers
 2363     // NOTE: cmpl is plenty here to provoke a segv
 2364     cmpptr(rax, Address(reg, 0));
 2365     // Note: should probably use testl(rax, Address(reg, 0));
 2366     //       may be shorter code (however, this version of
 2367     //       testl needs to be implemented first)
 2368   } else {
 2369     // nothing to do, (later) access of M[reg + offset]
 2370     // will provoke OS null exception if reg is null
 2371   }
 2372 }
 2373 
 2374 void MacroAssembler::os_breakpoint() {
 2375   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2376   // (e.g., MSVC can't call ps() otherwise)
 2377   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2378 }
 2379 
 2380 void MacroAssembler::unimplemented(const char* what) {
 2381   const char* buf = nullptr;
 2382   {
 2383     ResourceMark rm;
 2384     stringStream ss;
 2385     ss.print("unimplemented: %s", what);
 2386     buf = code_string(ss.as_string());
 2387   }
 2388   stop(buf);
 2389 }
 2390 
 2391 #define XSTATE_BV 0x200
 2392 
 2393 void MacroAssembler::pop_CPU_state() {
 2394   pop_FPU_state();
 2395   pop_IU_state();
 2396 }
 2397 
 2398 void MacroAssembler::pop_FPU_state() {
 2399   fxrstor(Address(rsp, 0));
 2400   addptr(rsp, FPUStateSizeInWords * wordSize);
 2401 }
 2402 
 2403 void MacroAssembler::pop_IU_state() {
 2404   popa();
 2405   addq(rsp, 8);
 2406   popf();
 2407 }
 2408 
 2409 // Save Integer and Float state
 2410 // Warning: Stack must be 16 byte aligned (64bit)
 2411 void MacroAssembler::push_CPU_state() {
 2412   push_IU_state();
 2413   push_FPU_state();
 2414 }
 2415 
 2416 void MacroAssembler::push_FPU_state() {
 2417   subptr(rsp, FPUStateSizeInWords * wordSize);
 2418   fxsave(Address(rsp, 0));
 2419 }
 2420 
 2421 void MacroAssembler::push_IU_state() {
 2422   // Push flags first because pusha kills them
 2423   pushf();
 2424   // Make sure rsp stays 16-byte aligned
 2425   subq(rsp, 8);
 2426   pusha();
 2427 }
 2428 
 2429 void MacroAssembler::push_cont_fastpath() {
 2430   if (!Continuations::enabled()) return;
 2431 
 2432   Label L_done;
 2433   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2434   jccb(Assembler::belowEqual, L_done);
 2435   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2436   bind(L_done);
 2437 }
 2438 
 2439 void MacroAssembler::pop_cont_fastpath() {
 2440   if (!Continuations::enabled()) return;
 2441 
 2442   Label L_done;
 2443   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2444   jccb(Assembler::below, L_done);
 2445   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2446   bind(L_done);
 2447 }
 2448 
 2449 #ifdef ASSERT
 2450 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2451   Label no_cont;
 2452   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2453   testl(cont, cont);
 2454   jcc(Assembler::zero, no_cont);
 2455   stop(name);
 2456   bind(no_cont);
 2457 }
 2458 #endif
 2459 
 2460 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2461   // we must set sp to zero to clear frame
 2462   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2463   // must clear fp, so that compiled frames are not confused; it is
 2464   // possible that we need it only for debugging
 2465   if (clear_fp) {
 2466     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2467   }
 2468   // Always clear the pc because it could have been set by make_walkable()
 2469   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2470   vzeroupper();
 2471 }
 2472 
 2473 void MacroAssembler::round_to(Register reg, int modulus) {
 2474   addptr(reg, modulus - 1);
 2475   andptr(reg, -modulus);
 2476 }
 2477 
 2478 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2479   if (at_return) {
 2480     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2481     // we may safely use rsp instead to perform the stack watermark check.
 2482     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2483     jcc(Assembler::above, slow_path);
 2484     return;
 2485   }
 2486   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2487   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2488 }
 2489 
 2490 // Calls to C land
 2491 //
 2492 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2493 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2494 // has to be reset to 0. This is required to allow proper stack traversal.
 2495 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2496                                          Register last_java_fp,
 2497                                          address  last_java_pc,
 2498                                          Register rscratch) {
 2499   vzeroupper();
 2500   // determine last_java_sp register
 2501   if (!last_java_sp->is_valid()) {
 2502     last_java_sp = rsp;
 2503   }
 2504   // last_java_fp is optional
 2505   if (last_java_fp->is_valid()) {
 2506     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2507   }
 2508   // last_java_pc is optional
 2509   if (last_java_pc != nullptr) {
 2510     Address java_pc(r15_thread,
 2511                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2512     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2513   }
 2514   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2515 }
 2516 
 2517 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2518                                          Register last_java_fp,
 2519                                          Label &L,
 2520                                          Register scratch) {
 2521   lea(scratch, L);
 2522   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2523   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2524 }
 2525 
 2526 void MacroAssembler::shlptr(Register dst, int imm8) {
 2527   shlq(dst, imm8);
 2528 }
 2529 
 2530 void MacroAssembler::shrptr(Register dst, int imm8) {
 2531   shrq(dst, imm8);
 2532 }
 2533 
 2534 void MacroAssembler::sign_extend_byte(Register reg) {
 2535   movsbl(reg, reg); // movsxb
 2536 }
 2537 
 2538 void MacroAssembler::sign_extend_short(Register reg) {
 2539   movswl(reg, reg); // movsxw
 2540 }
 2541 
 2542 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2543   if (imm32 >= 0 && is8bit(imm32)) {
 2544     testb(dst, imm32);
 2545   } else {
 2546     Assembler::testl(dst, imm32);
 2547   }
 2548 }
 2549 
 2550 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2551   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2552     testb(dst, imm32);
 2553   } else {
 2554     Assembler::testl(dst, imm32);
 2555   }
 2556 }
 2557 
 2558 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2559   assert(always_reachable(src), "Address should be reachable");
 2560   testl(dst, as_Address(src));
 2561 }
 2562 
 2563 void MacroAssembler::testq(Address dst, int32_t imm32) {
 2564   if (imm32 >= 0) {
 2565     testl(dst, imm32);
 2566   } else {
 2567     Assembler::testq(dst, imm32);
 2568   }
 2569 }
 2570 
 2571 void MacroAssembler::testq(Register dst, int32_t imm32) {
 2572   if (imm32 >= 0) {
 2573     testl(dst, imm32);
 2574   } else {
 2575     Assembler::testq(dst, imm32);
 2576   }
 2577 }
 2578 
 2579 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2580   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2581   Assembler::pcmpeqb(dst, src);
 2582 }
 2583 
 2584 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2585   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2586   Assembler::pcmpeqw(dst, src);
 2587 }
 2588 
 2589 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2590   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2591   Assembler::pcmpestri(dst, src, imm8);
 2592 }
 2593 
 2594 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2595   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2596   Assembler::pcmpestri(dst, src, imm8);
 2597 }
 2598 
 2599 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2600   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2601   Assembler::pmovzxbw(dst, src);
 2602 }
 2603 
 2604 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2605   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2606   Assembler::pmovzxbw(dst, src);
 2607 }
 2608 
 2609 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2610   assert((src->encoding() < 16),"XMM register should be 0-15");
 2611   Assembler::pmovmskb(dst, src);
 2612 }
 2613 
 2614 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2615   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2616   Assembler::ptest(dst, src);
 2617 }
 2618 
 2619 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2620   assert(rscratch != noreg || always_reachable(src), "missing");
 2621 
 2622   if (reachable(src)) {
 2623     Assembler::sqrtss(dst, as_Address(src));
 2624   } else {
 2625     lea(rscratch, src);
 2626     Assembler::sqrtss(dst, Address(rscratch, 0));
 2627   }
 2628 }
 2629 
 2630 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2631   assert(rscratch != noreg || always_reachable(src), "missing");
 2632 
 2633   if (reachable(src)) {
 2634     Assembler::subsd(dst, as_Address(src));
 2635   } else {
 2636     lea(rscratch, src);
 2637     Assembler::subsd(dst, Address(rscratch, 0));
 2638   }
 2639 }
 2640 
 2641 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2642   assert(rscratch != noreg || always_reachable(src), "missing");
 2643 
 2644   if (reachable(src)) {
 2645     Assembler::roundsd(dst, as_Address(src), rmode);
 2646   } else {
 2647     lea(rscratch, src);
 2648     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2649   }
 2650 }
 2651 
 2652 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2653   assert(rscratch != noreg || always_reachable(src), "missing");
 2654 
 2655   if (reachable(src)) {
 2656     Assembler::subss(dst, as_Address(src));
 2657   } else {
 2658     lea(rscratch, src);
 2659     Assembler::subss(dst, Address(rscratch, 0));
 2660   }
 2661 }
 2662 
 2663 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2664   assert(rscratch != noreg || always_reachable(src), "missing");
 2665 
 2666   if (reachable(src)) {
 2667     Assembler::ucomisd(dst, as_Address(src));
 2668   } else {
 2669     lea(rscratch, src);
 2670     Assembler::ucomisd(dst, Address(rscratch, 0));
 2671   }
 2672 }
 2673 
 2674 void MacroAssembler::evucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2675   assert(rscratch != noreg || always_reachable(src), "missing");
 2676 
 2677   if (reachable(src)) {
 2678     Assembler::evucomxsd(dst, as_Address(src));
 2679   } else {
 2680     lea(rscratch, src);
 2681     Assembler::evucomxsd(dst, Address(rscratch, 0));
 2682   }
 2683 }
 2684 
 2685 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2686   assert(rscratch != noreg || always_reachable(src), "missing");
 2687 
 2688   if (reachable(src)) {
 2689     Assembler::ucomiss(dst, as_Address(src));
 2690   } else {
 2691     lea(rscratch, src);
 2692     Assembler::ucomiss(dst, Address(rscratch, 0));
 2693   }
 2694 }
 2695 
 2696 void MacroAssembler::evucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2697   assert(rscratch != noreg || always_reachable(src), "missing");
 2698 
 2699   if (reachable(src)) {
 2700     Assembler::evucomxss(dst, as_Address(src));
 2701   } else {
 2702     lea(rscratch, src);
 2703     Assembler::evucomxss(dst, Address(rscratch, 0));
 2704   }
 2705 }
 2706 
 2707 void MacroAssembler::evucomish(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2708   assert(rscratch != noreg || always_reachable(src), "missing");
 2709 
 2710   if (reachable(src)) {
 2711     Assembler::evucomish(dst, as_Address(src));
 2712   } else {
 2713     lea(rscratch, src);
 2714     Assembler::evucomish(dst, Address(rscratch, 0));
 2715   }
 2716 }
 2717 
 2718 void MacroAssembler::evucomxsh(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2719   assert(rscratch != noreg || always_reachable(src), "missing");
 2720 
 2721   if (reachable(src)) {
 2722     Assembler::evucomxsh(dst, as_Address(src));
 2723   } else {
 2724     lea(rscratch, src);
 2725     Assembler::evucomxsh(dst, Address(rscratch, 0));
 2726   }
 2727 }
 2728 
 2729 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2730   assert(rscratch != noreg || always_reachable(src), "missing");
 2731 
 2732   // Used in sign-bit flipping with aligned address.
 2733   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2734 
 2735   if (UseAVX > 2 &&
 2736       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2737       (dst->encoding() >= 16)) {
 2738     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2739   } else if (reachable(src)) {
 2740     Assembler::xorpd(dst, as_Address(src));
 2741   } else {
 2742     lea(rscratch, src);
 2743     Assembler::xorpd(dst, Address(rscratch, 0));
 2744   }
 2745 }
 2746 
 2747 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2748   if (UseAVX > 2 &&
 2749       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2750       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2751     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2752   } else {
 2753     Assembler::xorpd(dst, src);
 2754   }
 2755 }
 2756 
 2757 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2758   if (UseAVX > 2 &&
 2759       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2760       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2761     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2762   } else {
 2763     Assembler::xorps(dst, src);
 2764   }
 2765 }
 2766 
 2767 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2768   assert(rscratch != noreg || always_reachable(src), "missing");
 2769 
 2770   // Used in sign-bit flipping with aligned address.
 2771   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2772 
 2773   if (UseAVX > 2 &&
 2774       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2775       (dst->encoding() >= 16)) {
 2776     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2777   } else if (reachable(src)) {
 2778     Assembler::xorps(dst, as_Address(src));
 2779   } else {
 2780     lea(rscratch, src);
 2781     Assembler::xorps(dst, Address(rscratch, 0));
 2782   }
 2783 }
 2784 
 2785 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2786   assert(rscratch != noreg || always_reachable(src), "missing");
 2787 
 2788   // Used in sign-bit flipping with aligned address.
 2789   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2790   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2791   if (reachable(src)) {
 2792     Assembler::pshufb(dst, as_Address(src));
 2793   } else {
 2794     lea(rscratch, src);
 2795     Assembler::pshufb(dst, Address(rscratch, 0));
 2796   }
 2797 }
 2798 
 2799 // AVX 3-operands instructions
 2800 
 2801 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2802   assert(rscratch != noreg || always_reachable(src), "missing");
 2803 
 2804   if (reachable(src)) {
 2805     vaddsd(dst, nds, as_Address(src));
 2806   } else {
 2807     lea(rscratch, src);
 2808     vaddsd(dst, nds, Address(rscratch, 0));
 2809   }
 2810 }
 2811 
 2812 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2813   assert(rscratch != noreg || always_reachable(src), "missing");
 2814 
 2815   if (reachable(src)) {
 2816     vaddss(dst, nds, as_Address(src));
 2817   } else {
 2818     lea(rscratch, src);
 2819     vaddss(dst, nds, Address(rscratch, 0));
 2820   }
 2821 }
 2822 
 2823 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2824   assert(UseAVX > 0, "requires some form of AVX");
 2825   assert(rscratch != noreg || always_reachable(src), "missing");
 2826 
 2827   if (reachable(src)) {
 2828     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2829   } else {
 2830     lea(rscratch, src);
 2831     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2832   }
 2833 }
 2834 
 2835 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2836   assert(UseAVX > 0, "requires some form of AVX");
 2837   assert(rscratch != noreg || always_reachable(src), "missing");
 2838 
 2839   if (reachable(src)) {
 2840     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2841   } else {
 2842     lea(rscratch, src);
 2843     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2844   }
 2845 }
 2846 
 2847 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2848   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2849   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2850 
 2851   vandps(dst, nds, negate_field, vector_len, rscratch);
 2852 }
 2853 
 2854 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2855   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2856   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2857 
 2858   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2859 }
 2860 
 2861 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2862   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2863   Assembler::vpaddb(dst, nds, src, vector_len);
 2864 }
 2865 
 2866 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2867   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2868   Assembler::vpaddb(dst, nds, src, vector_len);
 2869 }
 2870 
 2871 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2872   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2873   Assembler::vpaddw(dst, nds, src, vector_len);
 2874 }
 2875 
 2876 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2877   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2878   Assembler::vpaddw(dst, nds, src, vector_len);
 2879 }
 2880 
 2881 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2882   assert(rscratch != noreg || always_reachable(src), "missing");
 2883 
 2884   if (reachable(src)) {
 2885     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 2886   } else {
 2887     lea(rscratch, src);
 2888     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 2889   }
 2890 }
 2891 
 2892 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2893   assert(rscratch != noreg || always_reachable(src), "missing");
 2894 
 2895   if (reachable(src)) {
 2896     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 2897   } else {
 2898     lea(rscratch, src);
 2899     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 2900   }
 2901 }
 2902 
 2903 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2904   assert(rscratch != noreg || always_reachable(src), "missing");
 2905 
 2906   if (reachable(src)) {
 2907     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 2908   } else {
 2909     lea(rscratch, src);
 2910     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 2911   }
 2912 }
 2913 
 2914 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2915   assert(rscratch != noreg || always_reachable(src), "missing");
 2916 
 2917   if (reachable(src)) {
 2918     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 2919   } else {
 2920     lea(rscratch, src);
 2921     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 2922   }
 2923 }
 2924 
 2925 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2926   assert(rscratch != noreg || always_reachable(src), "missing");
 2927 
 2928   if (reachable(src)) {
 2929     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 2930   } else {
 2931     lea(rscratch, src);
 2932     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 2933   }
 2934 }
 2935 
 2936 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2937   assert(rscratch != noreg || always_reachable(src), "missing");
 2938 
 2939   if (reachable(src)) {
 2940     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 2941   } else {
 2942     lea(rscratch, src);
 2943     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 2944   }
 2945 }
 2946 
 2947 // Vector float blend
 2948 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2949 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2950   // WARN: Allow dst == (src1|src2), mask == scratch
 2951   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2952                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2953   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 2954   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2955   if (blend_emulation && scratch_available && dst_available) {
 2956     if (compute_mask) {
 2957       vpsrad(scratch, mask, 32, vector_len);
 2958       mask = scratch;
 2959     }
 2960     if (dst == src1) {
 2961       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 2962       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 2963     } else {
 2964       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 2965       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 2966     }
 2967     vpor(dst, dst, scratch, vector_len);
 2968   } else {
 2969     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 2970   }
 2971 }
 2972 
 2973 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2974 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2975   // WARN: Allow dst == (src1|src2), mask == scratch
 2976   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2977                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2978   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 2979   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2980   if (blend_emulation && scratch_available && dst_available) {
 2981     if (compute_mask) {
 2982       vpxor(scratch, scratch, scratch, vector_len);
 2983       vpcmpgtq(scratch, scratch, mask, vector_len);
 2984       mask = scratch;
 2985     }
 2986     if (dst == src1) {
 2987       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 2988       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 2989     } else {
 2990       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 2991       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 2992     }
 2993     vpor(dst, dst, scratch, vector_len);
 2994   } else {
 2995     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 2996   }
 2997 }
 2998 
 2999 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3000   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3001   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 3002 }
 3003 
 3004 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 3005   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3006   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 3007 }
 3008 
 3009 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3010   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3011   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3012 }
 3013 
 3014 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3015   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3016   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3017 }
 3018 
 3019 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3020   assert(rscratch != noreg || always_reachable(src), "missing");
 3021 
 3022   if (reachable(src)) {
 3023     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 3024   } else {
 3025     lea(rscratch, src);
 3026     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 3027   }
 3028 }
 3029 
 3030 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3031                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3032   assert(rscratch != noreg || always_reachable(src), "missing");
 3033 
 3034   if (reachable(src)) {
 3035     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3036   } else {
 3037     lea(rscratch, src);
 3038     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3039   }
 3040 }
 3041 
 3042 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3043                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3044   assert(rscratch != noreg || always_reachable(src), "missing");
 3045 
 3046   if (reachable(src)) {
 3047     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3048   } else {
 3049     lea(rscratch, src);
 3050     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3051   }
 3052 }
 3053 
 3054 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3055                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3056   assert(rscratch != noreg || always_reachable(src), "missing");
 3057 
 3058   if (reachable(src)) {
 3059     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3060   } else {
 3061     lea(rscratch, src);
 3062     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3063   }
 3064 }
 3065 
 3066 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3067                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3068   assert(rscratch != noreg || always_reachable(src), "missing");
 3069 
 3070   if (reachable(src)) {
 3071     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3072   } else {
 3073     lea(rscratch, src);
 3074     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3075   }
 3076 }
 3077 
 3078 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3079   if (width == Assembler::Q) {
 3080     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3081   } else {
 3082     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3083   }
 3084 }
 3085 
 3086 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3087   int eq_cond_enc = 0x29;
 3088   int gt_cond_enc = 0x37;
 3089   if (width != Assembler::Q) {
 3090     eq_cond_enc = 0x74 + width;
 3091     gt_cond_enc = 0x64 + width;
 3092   }
 3093   switch (cond) {
 3094   case eq:
 3095     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3096     break;
 3097   case neq:
 3098     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3099     vallones(xtmp, vector_len);
 3100     vpxor(dst, xtmp, dst, vector_len);
 3101     break;
 3102   case le:
 3103     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3104     vallones(xtmp, vector_len);
 3105     vpxor(dst, xtmp, dst, vector_len);
 3106     break;
 3107   case nlt:
 3108     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3109     vallones(xtmp, vector_len);
 3110     vpxor(dst, xtmp, dst, vector_len);
 3111     break;
 3112   case lt:
 3113     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3114     break;
 3115   case nle:
 3116     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3117     break;
 3118   default:
 3119     assert(false, "Should not reach here");
 3120   }
 3121 }
 3122 
 3123 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3124   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3125   Assembler::vpmovzxbw(dst, src, vector_len);
 3126 }
 3127 
 3128 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3129   assert((src->encoding() < 16),"XMM register should be 0-15");
 3130   Assembler::vpmovmskb(dst, src, vector_len);
 3131 }
 3132 
 3133 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3134   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3135   Assembler::vpmullw(dst, nds, src, vector_len);
 3136 }
 3137 
 3138 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3139   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3140   Assembler::vpmullw(dst, nds, src, vector_len);
 3141 }
 3142 
 3143 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3144   assert((UseAVX > 0), "AVX support is needed");
 3145   assert(rscratch != noreg || always_reachable(src), "missing");
 3146 
 3147   if (reachable(src)) {
 3148     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3149   } else {
 3150     lea(rscratch, src);
 3151     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3152   }
 3153 }
 3154 
 3155 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3156   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3157   Assembler::vpsubb(dst, nds, src, vector_len);
 3158 }
 3159 
 3160 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3161   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3162   Assembler::vpsubb(dst, nds, src, vector_len);
 3163 }
 3164 
 3165 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3166   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3167   Assembler::vpsubw(dst, nds, src, vector_len);
 3168 }
 3169 
 3170 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3171   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3172   Assembler::vpsubw(dst, nds, src, vector_len);
 3173 }
 3174 
 3175 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3176   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3177   Assembler::vpsraw(dst, nds, shift, vector_len);
 3178 }
 3179 
 3180 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3181   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3182   Assembler::vpsraw(dst, nds, shift, vector_len);
 3183 }
 3184 
 3185 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3186   assert(UseAVX > 2,"");
 3187   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3188      vector_len = 2;
 3189   }
 3190   Assembler::evpsraq(dst, nds, shift, vector_len);
 3191 }
 3192 
 3193 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3194   assert(UseAVX > 2,"");
 3195   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3196      vector_len = 2;
 3197   }
 3198   Assembler::evpsraq(dst, nds, shift, vector_len);
 3199 }
 3200 
 3201 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3202   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3203   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3204 }
 3205 
 3206 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3207   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3208   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3209 }
 3210 
 3211 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3212   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3213   Assembler::vpsllw(dst, nds, shift, vector_len);
 3214 }
 3215 
 3216 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3217   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3218   Assembler::vpsllw(dst, nds, shift, vector_len);
 3219 }
 3220 
 3221 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3222   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3223   Assembler::vptest(dst, src);
 3224 }
 3225 
 3226 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3227   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3228   Assembler::punpcklbw(dst, src);
 3229 }
 3230 
 3231 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3232   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3233   Assembler::pshufd(dst, src, mode);
 3234 }
 3235 
 3236 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3237   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3238   Assembler::pshuflw(dst, src, mode);
 3239 }
 3240 
 3241 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3242   assert(rscratch != noreg || always_reachable(src), "missing");
 3243 
 3244   if (reachable(src)) {
 3245     vandpd(dst, nds, as_Address(src), vector_len);
 3246   } else {
 3247     lea(rscratch, src);
 3248     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3249   }
 3250 }
 3251 
 3252 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3253   assert(rscratch != noreg || always_reachable(src), "missing");
 3254 
 3255   if (reachable(src)) {
 3256     vandps(dst, nds, as_Address(src), vector_len);
 3257   } else {
 3258     lea(rscratch, src);
 3259     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3260   }
 3261 }
 3262 
 3263 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3264                             bool merge, int vector_len, Register rscratch) {
 3265   assert(rscratch != noreg || always_reachable(src), "missing");
 3266 
 3267   if (reachable(src)) {
 3268     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3269   } else {
 3270     lea(rscratch, src);
 3271     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3272   }
 3273 }
 3274 
 3275 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3276   assert(rscratch != noreg || always_reachable(src), "missing");
 3277 
 3278   if (reachable(src)) {
 3279     vdivsd(dst, nds, as_Address(src));
 3280   } else {
 3281     lea(rscratch, src);
 3282     vdivsd(dst, nds, Address(rscratch, 0));
 3283   }
 3284 }
 3285 
 3286 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3287   assert(rscratch != noreg || always_reachable(src), "missing");
 3288 
 3289   if (reachable(src)) {
 3290     vdivss(dst, nds, as_Address(src));
 3291   } else {
 3292     lea(rscratch, src);
 3293     vdivss(dst, nds, Address(rscratch, 0));
 3294   }
 3295 }
 3296 
 3297 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3298   assert(rscratch != noreg || always_reachable(src), "missing");
 3299 
 3300   if (reachable(src)) {
 3301     vmulsd(dst, nds, as_Address(src));
 3302   } else {
 3303     lea(rscratch, src);
 3304     vmulsd(dst, nds, Address(rscratch, 0));
 3305   }
 3306 }
 3307 
 3308 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3309   assert(rscratch != noreg || always_reachable(src), "missing");
 3310 
 3311   if (reachable(src)) {
 3312     vmulss(dst, nds, as_Address(src));
 3313   } else {
 3314     lea(rscratch, src);
 3315     vmulss(dst, nds, Address(rscratch, 0));
 3316   }
 3317 }
 3318 
 3319 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3320   assert(rscratch != noreg || always_reachable(src), "missing");
 3321 
 3322   if (reachable(src)) {
 3323     vsubsd(dst, nds, as_Address(src));
 3324   } else {
 3325     lea(rscratch, src);
 3326     vsubsd(dst, nds, Address(rscratch, 0));
 3327   }
 3328 }
 3329 
 3330 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3331   assert(rscratch != noreg || always_reachable(src), "missing");
 3332 
 3333   if (reachable(src)) {
 3334     vsubss(dst, nds, as_Address(src));
 3335   } else {
 3336     lea(rscratch, src);
 3337     vsubss(dst, nds, Address(rscratch, 0));
 3338   }
 3339 }
 3340 
 3341 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3342   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3343   assert(rscratch != noreg || always_reachable(src), "missing");
 3344 
 3345   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3346 }
 3347 
 3348 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3349   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3350   assert(rscratch != noreg || always_reachable(src), "missing");
 3351 
 3352   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3353 }
 3354 
 3355 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3356   assert(rscratch != noreg || always_reachable(src), "missing");
 3357 
 3358   if (reachable(src)) {
 3359     vxorpd(dst, nds, as_Address(src), vector_len);
 3360   } else {
 3361     lea(rscratch, src);
 3362     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3363   }
 3364 }
 3365 
 3366 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3367   assert(rscratch != noreg || always_reachable(src), "missing");
 3368 
 3369   if (reachable(src)) {
 3370     vxorps(dst, nds, as_Address(src), vector_len);
 3371   } else {
 3372     lea(rscratch, src);
 3373     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3374   }
 3375 }
 3376 
 3377 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3378   assert(rscratch != noreg || always_reachable(src), "missing");
 3379 
 3380   if (UseAVX > 1 || (vector_len < 1)) {
 3381     if (reachable(src)) {
 3382       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3383     } else {
 3384       lea(rscratch, src);
 3385       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3386     }
 3387   } else {
 3388     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3389   }
 3390 }
 3391 
 3392 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3393   assert(rscratch != noreg || always_reachable(src), "missing");
 3394 
 3395   if (reachable(src)) {
 3396     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3397   } else {
 3398     lea(rscratch, src);
 3399     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3400   }
 3401 }
 3402 
 3403 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3404   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3405   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3406   // The inverted mask is sign-extended
 3407   andptr(possibly_non_local, inverted_mask);
 3408 }
 3409 
 3410 void MacroAssembler::resolve_jobject(Register value,
 3411                                      Register tmp) {
 3412   Register thread = r15_thread;
 3413   assert_different_registers(value, thread, tmp);
 3414   Label done, tagged, weak_tagged;
 3415   testptr(value, value);
 3416   jcc(Assembler::zero, done);           // Use null as-is.
 3417   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3418   jcc(Assembler::notZero, tagged);
 3419 
 3420   // Resolve local handle
 3421   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3422   verify_oop(value);
 3423   jmp(done);
 3424 
 3425   bind(tagged);
 3426   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3427   jcc(Assembler::notZero, weak_tagged);
 3428 
 3429   // Resolve global handle
 3430   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3431   verify_oop(value);
 3432   jmp(done);
 3433 
 3434   bind(weak_tagged);
 3435   // Resolve jweak.
 3436   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3437                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3438   verify_oop(value);
 3439 
 3440   bind(done);
 3441 }
 3442 
 3443 void MacroAssembler::resolve_global_jobject(Register value,
 3444                                             Register tmp) {
 3445   Register thread = r15_thread;
 3446   assert_different_registers(value, thread, tmp);
 3447   Label done;
 3448 
 3449   testptr(value, value);
 3450   jcc(Assembler::zero, done);           // Use null as-is.
 3451 
 3452 #ifdef ASSERT
 3453   {
 3454     Label valid_global_tag;
 3455     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3456     jcc(Assembler::notZero, valid_global_tag);
 3457     stop("non global jobject using resolve_global_jobject");
 3458     bind(valid_global_tag);
 3459   }
 3460 #endif
 3461 
 3462   // Resolve global handle
 3463   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3464   verify_oop(value);
 3465 
 3466   bind(done);
 3467 }
 3468 
 3469 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3470   subq(dst, imm32);
 3471 }
 3472 
 3473 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3474 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3475   subq_imm32(dst, imm32);
 3476 }
 3477 
 3478 void MacroAssembler::subptr(Register dst, Register src) {
 3479   subq(dst, src);
 3480 }
 3481 
 3482 // C++ bool manipulation
 3483 void MacroAssembler::testbool(Register dst) {
 3484   if(sizeof(bool) == 1)
 3485     testb(dst, 0xff);
 3486   else if(sizeof(bool) == 2) {
 3487     // testw implementation needed for two byte bools
 3488     ShouldNotReachHere();
 3489   } else if(sizeof(bool) == 4)
 3490     testl(dst, dst);
 3491   else
 3492     // unsupported
 3493     ShouldNotReachHere();
 3494 }
 3495 
 3496 void MacroAssembler::testptr(Register dst, Register src) {
 3497   testq(dst, src);
 3498 }
 3499 
 3500 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3501 void MacroAssembler::tlab_allocate(Register obj,
 3502                                    Register var_size_in_bytes,
 3503                                    int con_size_in_bytes,
 3504                                    Register t1,
 3505                                    Register t2,
 3506                                    Label& slow_case) {
 3507   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3508   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3509 }
 3510 
 3511 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3512   RegSet regs;
 3513   regs += RegSet::of(rax, rcx, rdx);
 3514 #ifndef _WINDOWS
 3515   regs += RegSet::of(rsi, rdi);
 3516 #endif
 3517   regs += RegSet::range(r8, r11);
 3518   if (UseAPX) {
 3519     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3520   }
 3521   return regs;
 3522 }
 3523 
 3524 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3525   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3526 #if defined(_WINDOWS)
 3527   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3528   if (num_xmm_registers > 16) {
 3529      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3530   }
 3531   return result;
 3532 #else
 3533   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3534 #endif
 3535 }
 3536 
 3537 // C1 only ever uses the first double/float of the XMM register.
 3538 static int xmm_save_size() { return sizeof(double); }
 3539 
 3540 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3541   masm->movdbl(Address(rsp, offset), reg);
 3542 }
 3543 
 3544 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3545   masm->movdbl(reg, Address(rsp, offset));
 3546 }
 3547 
 3548 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3549                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3550 
 3551   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3552                          StackAlignmentInBytes);
 3553   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3554 
 3555   return gp_area_size + xmm_area_size;
 3556 }
 3557 
 3558 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3559   block_comment("push_call_clobbered_registers start");
 3560   // Regular registers
 3561   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3562 
 3563   int gp_area_size;
 3564   int xmm_area_size;
 3565   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3566                                                gp_area_size, xmm_area_size);
 3567   subptr(rsp, total_save_size);
 3568 
 3569   push_set(gp_registers_to_push, 0);
 3570 
 3571   if (save_fpu) {
 3572     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3573   }
 3574 
 3575   block_comment("push_call_clobbered_registers end");
 3576 }
 3577 
 3578 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3579   block_comment("pop_call_clobbered_registers start");
 3580 
 3581   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 3582 
 3583   int gp_area_size;
 3584   int xmm_area_size;
 3585   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 3586                                                gp_area_size, xmm_area_size);
 3587 
 3588   if (restore_fpu) {
 3589     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3590   }
 3591 
 3592   pop_set(gp_registers_to_pop, 0);
 3593 
 3594   addptr(rsp, total_save_size);
 3595 
 3596   vzeroupper();
 3597 
 3598   block_comment("pop_call_clobbered_registers end");
 3599 }
 3600 
 3601 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3602   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3603   int spill_offset = offset;
 3604 
 3605   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3606     save_xmm_register(this, spill_offset, *it);
 3607     spill_offset += xmm_save_size();
 3608   }
 3609 }
 3610 
 3611 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3612   int restore_size = set.size() * xmm_save_size();
 3613   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3614 
 3615   int restore_offset = offset + restore_size - xmm_save_size();
 3616 
 3617   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3618     restore_xmm_register(this, restore_offset, *it);
 3619     restore_offset -= xmm_save_size();
 3620   }
 3621 }
 3622 
 3623 void MacroAssembler::push_set(RegSet set, int offset) {
 3624   int spill_offset;
 3625   if (offset == -1) {
 3626     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3627     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3628     subptr(rsp, aligned_size);
 3629     spill_offset = 0;
 3630   } else {
 3631     spill_offset = offset;
 3632   }
 3633 
 3634   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3635     movptr(Address(rsp, spill_offset), *it);
 3636     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3637   }
 3638 }
 3639 
 3640 void MacroAssembler::pop_set(RegSet set, int offset) {
 3641 
 3642   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3643   int restore_size = set.size() * gp_reg_size;
 3644   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3645 
 3646   int restore_offset;
 3647   if (offset == -1) {
 3648     restore_offset = restore_size - gp_reg_size;
 3649   } else {
 3650     restore_offset = offset + restore_size - gp_reg_size;
 3651   }
 3652   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3653     movptr(*it, Address(rsp, restore_offset));
 3654     restore_offset -= gp_reg_size;
 3655   }
 3656 
 3657   if (offset == -1) {
 3658     addptr(rsp, aligned_size);
 3659   }
 3660 }
 3661 
 3662 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3663 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3664   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3665   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3666   Label done;
 3667 
 3668   testptr(length_in_bytes, length_in_bytes);
 3669   jcc(Assembler::zero, done);
 3670 
 3671   // initialize topmost word, divide index by 2, check if odd and test if zero
 3672   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3673 #ifdef ASSERT
 3674   {
 3675     Label L;
 3676     testptr(length_in_bytes, BytesPerWord - 1);
 3677     jcc(Assembler::zero, L);
 3678     stop("length must be a multiple of BytesPerWord");
 3679     bind(L);
 3680   }
 3681 #endif
 3682   Register index = length_in_bytes;
 3683   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3684   if (UseIncDec) {
 3685     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3686   } else {
 3687     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3688     shrptr(index, 1);
 3689   }
 3690 
 3691   // initialize remaining object fields: index is a multiple of 2 now
 3692   {
 3693     Label loop;
 3694     bind(loop);
 3695     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3696     decrement(index);
 3697     jcc(Assembler::notZero, loop);
 3698   }
 3699 
 3700   bind(done);
 3701 }
 3702 
 3703 // Look up the method for a megamorphic invokeinterface call.
 3704 // The target method is determined by <intf_klass, itable_index>.
 3705 // The receiver klass is in recv_klass.
 3706 // On success, the result will be in method_result, and execution falls through.
 3707 // On failure, execution transfers to the given label.
 3708 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3709                                              Register intf_klass,
 3710                                              RegisterOrConstant itable_index,
 3711                                              Register method_result,
 3712                                              Register scan_temp,
 3713                                              Label& L_no_such_interface,
 3714                                              bool return_method) {
 3715   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3716   assert_different_registers(method_result, intf_klass, scan_temp);
 3717   assert(recv_klass != method_result || !return_method,
 3718          "recv_klass can be destroyed when method isn't needed");
 3719 
 3720   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3721          "caller must use same register for non-constant itable index as for method");
 3722 
 3723   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3724   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3725   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3726   int scan_step   = itableOffsetEntry::size() * wordSize;
 3727   int vte_size    = vtableEntry::size_in_bytes();
 3728   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3729   assert(vte_size == wordSize, "else adjust times_vte_scale");
 3730 
 3731   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3732 
 3733   // Could store the aligned, prescaled offset in the klass.
 3734   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 3735 
 3736   if (return_method) {
 3737     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 3738     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3739     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 3740   }
 3741 
 3742   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 3743   //   if (scan->interface() == intf) {
 3744   //     result = (klass + scan->offset() + itable_index);
 3745   //   }
 3746   // }
 3747   Label search, found_method;
 3748 
 3749   for (int peel = 1; peel >= 0; peel--) {
 3750     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 3751     cmpptr(intf_klass, method_result);
 3752 
 3753     if (peel) {
 3754       jccb(Assembler::equal, found_method);
 3755     } else {
 3756       jccb(Assembler::notEqual, search);
 3757       // (invert the test to fall through to found_method...)
 3758     }
 3759 
 3760     if (!peel)  break;
 3761 
 3762     bind(search);
 3763 
 3764     // Check that the previous entry is non-null.  A null entry means that
 3765     // the receiver class doesn't implement the interface, and wasn't the
 3766     // same as when the caller was compiled.
 3767     testptr(method_result, method_result);
 3768     jcc(Assembler::zero, L_no_such_interface);
 3769     addptr(scan_temp, scan_step);
 3770   }
 3771 
 3772   bind(found_method);
 3773 
 3774   if (return_method) {
 3775     // Got a hit.
 3776     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 3777     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 3778   }
 3779 }
 3780 
 3781 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 3782 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 3783 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 3784 // The target method is determined by <holder_klass, itable_index>.
 3785 // The receiver klass is in recv_klass.
 3786 // On success, the result will be in method_result, and execution falls through.
 3787 // On failure, execution transfers to the given label.
 3788 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 3789                                                   Register holder_klass,
 3790                                                   Register resolved_klass,
 3791                                                   Register method_result,
 3792                                                   Register scan_temp,
 3793                                                   Register temp_reg2,
 3794                                                   Register receiver,
 3795                                                   int itable_index,
 3796                                                   Label& L_no_such_interface) {
 3797   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 3798   Register temp_itbl_klass = method_result;
 3799   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 3800 
 3801   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3802   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3803   int scan_step = itableOffsetEntry::size() * wordSize;
 3804   int vte_size = vtableEntry::size_in_bytes();
 3805   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 3806   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 3807   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3808   assert(vte_size == wordSize, "adjust times_vte_scale");
 3809 
 3810   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 3811 
 3812   // temp_itbl_klass = recv_klass.itable[0]
 3813   // scan_temp = &recv_klass.itable[0] + step
 3814   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3815   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 3816   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 3817   xorptr(temp_reg, temp_reg);
 3818 
 3819   // Initial checks:
 3820   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 3821   //   - if (itable[0] == 0), no such interface
 3822   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 3823   cmpptr(holder_klass, resolved_klass);
 3824   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 3825   testptr(temp_itbl_klass, temp_itbl_klass);
 3826   jccb(Assembler::zero, L_no_such_interface);
 3827   cmpptr(holder_klass, temp_itbl_klass);
 3828   jccb(Assembler::equal, L_holder_found);
 3829 
 3830   // Loop: Look for holder_klass record in itable
 3831   //   do {
 3832   //     tmp = itable[index];
 3833   //     index += step;
 3834   //     if (tmp == holder_klass) {
 3835   //       goto L_holder_found; // Found!
 3836   //     }
 3837   //   } while (tmp != 0);
 3838   //   goto L_no_such_interface // Not found.
 3839   Label L_scan_holder;
 3840   bind(L_scan_holder);
 3841     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3842     addptr(scan_temp, scan_step);
 3843     cmpptr(holder_klass, temp_itbl_klass);
 3844     jccb(Assembler::equal, L_holder_found);
 3845     testptr(temp_itbl_klass, temp_itbl_klass);
 3846     jccb(Assembler::notZero, L_scan_holder);
 3847 
 3848   jmpb(L_no_such_interface);
 3849 
 3850   // Loop: Look for resolved_class record in itable
 3851   //   do {
 3852   //     tmp = itable[index];
 3853   //     index += step;
 3854   //     if (tmp == holder_klass) {
 3855   //        // Also check if we have met a holder klass
 3856   //        holder_tmp = itable[index-step-ioffset];
 3857   //     }
 3858   //     if (tmp == resolved_klass) {
 3859   //        goto L_resolved_found;  // Found!
 3860   //     }
 3861   //   } while (tmp != 0);
 3862   //   goto L_no_such_interface // Not found.
 3863   //
 3864   Label L_loop_scan_resolved;
 3865   bind(L_loop_scan_resolved);
 3866     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3867     addptr(scan_temp, scan_step);
 3868     bind(L_loop_scan_resolved_entry);
 3869     cmpptr(holder_klass, temp_itbl_klass);
 3870     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3871     cmpptr(resolved_klass, temp_itbl_klass);
 3872     jccb(Assembler::equal, L_resolved_found);
 3873     testptr(temp_itbl_klass, temp_itbl_klass);
 3874     jccb(Assembler::notZero, L_loop_scan_resolved);
 3875 
 3876   jmpb(L_no_such_interface);
 3877 
 3878   Label L_ready;
 3879 
 3880   // See if we already have a holder klass. If not, go and scan for it.
 3881   bind(L_resolved_found);
 3882   testptr(temp_reg, temp_reg);
 3883   jccb(Assembler::zero, L_scan_holder);
 3884   jmpb(L_ready);
 3885 
 3886   bind(L_holder_found);
 3887   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3888 
 3889   // Finally, temp_reg contains holder_klass vtable offset
 3890   bind(L_ready);
 3891   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3892   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 3893     load_klass(scan_temp, receiver, noreg);
 3894     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3895   } else {
 3896     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3897   }
 3898 }
 3899 
 3900 
 3901 // virtual method calling
 3902 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 3903                                            RegisterOrConstant vtable_index,
 3904                                            Register method_result) {
 3905   const ByteSize base = Klass::vtable_start_offset();
 3906   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 3907   Address vtable_entry_addr(recv_klass,
 3908                             vtable_index, Address::times_ptr,
 3909                             base + vtableEntry::method_offset());
 3910   movptr(method_result, vtable_entry_addr);
 3911 }
 3912 
 3913 
 3914 void MacroAssembler::check_klass_subtype(Register sub_klass,
 3915                            Register super_klass,
 3916                            Register temp_reg,
 3917                            Label& L_success) {
 3918   Label L_failure;
 3919   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 3920   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 3921   bind(L_failure);
 3922 }
 3923 
 3924 
 3925 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 3926                                                    Register super_klass,
 3927                                                    Register temp_reg,
 3928                                                    Label* L_success,
 3929                                                    Label* L_failure,
 3930                                                    Label* L_slow_path,
 3931                                         RegisterOrConstant super_check_offset) {
 3932   assert_different_registers(sub_klass, super_klass, temp_reg);
 3933   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 3934   if (super_check_offset.is_register()) {
 3935     assert_different_registers(sub_klass, super_klass,
 3936                                super_check_offset.as_register());
 3937   } else if (must_load_sco) {
 3938     assert(temp_reg != noreg, "supply either a temp or a register offset");
 3939   }
 3940 
 3941   Label L_fallthrough;
 3942   int label_nulls = 0;
 3943   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 3944   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 3945   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 3946   assert(label_nulls <= 1, "at most one null in the batch");
 3947 
 3948   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 3949   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 3950   Address super_check_offset_addr(super_klass, sco_offset);
 3951 
 3952   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 3953   // range of a jccb.  If this routine grows larger, reconsider at
 3954   // least some of these.
 3955 #define local_jcc(assembler_cond, label)                                \
 3956   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 3957   else                             jcc( assembler_cond, label) /*omit semi*/
 3958 
 3959   // Hacked jmp, which may only be used just before L_fallthrough.
 3960 #define final_jmp(label)                                                \
 3961   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 3962   else                            jmp(label)                /*omit semi*/
 3963 
 3964   // If the pointers are equal, we are done (e.g., String[] elements).
 3965   // This self-check enables sharing of secondary supertype arrays among
 3966   // non-primary types such as array-of-interface.  Otherwise, each such
 3967   // type would need its own customized SSA.
 3968   // We move this check to the front of the fast path because many
 3969   // type checks are in fact trivially successful in this manner,
 3970   // so we get a nicely predicted branch right at the start of the check.
 3971   cmpptr(sub_klass, super_klass);
 3972   local_jcc(Assembler::equal, *L_success);
 3973 
 3974   // Check the supertype display:
 3975   if (must_load_sco) {
 3976     // Positive movl does right thing on LP64.
 3977     movl(temp_reg, super_check_offset_addr);
 3978     super_check_offset = RegisterOrConstant(temp_reg);
 3979   }
 3980   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 3981   cmpptr(super_klass, super_check_addr); // load displayed supertype
 3982 
 3983   // This check has worked decisively for primary supers.
 3984   // Secondary supers are sought in the super_cache ('super_cache_addr').
 3985   // (Secondary supers are interfaces and very deeply nested subtypes.)
 3986   // This works in the same check above because of a tricky aliasing
 3987   // between the super_cache and the primary super display elements.
 3988   // (The 'super_check_addr' can address either, as the case requires.)
 3989   // Note that the cache is updated below if it does not help us find
 3990   // what we need immediately.
 3991   // So if it was a primary super, we can just fail immediately.
 3992   // Otherwise, it's the slow path for us (no success at this point).
 3993 
 3994   if (super_check_offset.is_register()) {
 3995     local_jcc(Assembler::equal, *L_success);
 3996     cmpl(super_check_offset.as_register(), sc_offset);
 3997     if (L_failure == &L_fallthrough) {
 3998       local_jcc(Assembler::equal, *L_slow_path);
 3999     } else {
 4000       local_jcc(Assembler::notEqual, *L_failure);
 4001       final_jmp(*L_slow_path);
 4002     }
 4003   } else if (super_check_offset.as_constant() == sc_offset) {
 4004     // Need a slow path; fast failure is impossible.
 4005     if (L_slow_path == &L_fallthrough) {
 4006       local_jcc(Assembler::equal, *L_success);
 4007     } else {
 4008       local_jcc(Assembler::notEqual, *L_slow_path);
 4009       final_jmp(*L_success);
 4010     }
 4011   } else {
 4012     // No slow path; it's a fast decision.
 4013     if (L_failure == &L_fallthrough) {
 4014       local_jcc(Assembler::equal, *L_success);
 4015     } else {
 4016       local_jcc(Assembler::notEqual, *L_failure);
 4017       final_jmp(*L_success);
 4018     }
 4019   }
 4020 
 4021   bind(L_fallthrough);
 4022 
 4023 #undef local_jcc
 4024 #undef final_jmp
 4025 }
 4026 
 4027 
 4028 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 4029                                                           Register super_klass,
 4030                                                           Register temp_reg,
 4031                                                           Register temp2_reg,
 4032                                                           Label* L_success,
 4033                                                           Label* L_failure,
 4034                                                           bool set_cond_codes) {
 4035   assert_different_registers(sub_klass, super_klass, temp_reg);
 4036   if (temp2_reg != noreg)
 4037     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4038 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4039 
 4040   Label L_fallthrough;
 4041   int label_nulls = 0;
 4042   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4043   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4044   assert(label_nulls <= 1, "at most one null in the batch");
 4045 
 4046   // a couple of useful fields in sub_klass:
 4047   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4048   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4049   Address secondary_supers_addr(sub_klass, ss_offset);
 4050   Address super_cache_addr(     sub_klass, sc_offset);
 4051 
 4052   // Do a linear scan of the secondary super-klass chain.
 4053   // This code is rarely used, so simplicity is a virtue here.
 4054   // The repne_scan instruction uses fixed registers, which we must spill.
 4055   // Don't worry too much about pre-existing connections with the input regs.
 4056 
 4057   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4058   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4059 
 4060   // Get super_klass value into rax (even if it was in rdi or rcx).
 4061   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4062   if (super_klass != rax) {
 4063     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4064     mov(rax, super_klass);
 4065   }
 4066   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4067   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4068 
 4069 #ifndef PRODUCT
 4070   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4071   ExternalAddress pst_counter_addr((address) pst_counter);
 4072   lea(rcx, pst_counter_addr);
 4073   incrementl(Address(rcx, 0));
 4074 #endif //PRODUCT
 4075 
 4076   // We will consult the secondary-super array.
 4077   movptr(rdi, secondary_supers_addr);
 4078   // Load the array length.  (Positive movl does right thing on LP64.)
 4079   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4080   // Skip to start of data.
 4081   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4082 
 4083   // Scan RCX words at [RDI] for an occurrence of RAX.
 4084   // Set NZ/Z based on last compare.
 4085   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4086   // not change flags (only scas instruction which is repeated sets flags).
 4087   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4088 
 4089     testptr(rax,rax); // Set Z = 0
 4090     repne_scan();
 4091 
 4092   // Unspill the temp. registers:
 4093   if (pushed_rdi)  pop(rdi);
 4094   if (pushed_rcx)  pop(rcx);
 4095   if (pushed_rax)  pop(rax);
 4096 
 4097   if (set_cond_codes) {
 4098     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4099     assert(!pushed_rdi, "rdi must be left non-null");
 4100     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4101   }
 4102 
 4103   if (L_failure == &L_fallthrough)
 4104         jccb(Assembler::notEqual, *L_failure);
 4105   else  jcc(Assembler::notEqual, *L_failure);
 4106 
 4107   // Success.  Cache the super we found and proceed in triumph.
 4108   movptr(super_cache_addr, super_klass);
 4109 
 4110   if (L_success != &L_fallthrough) {
 4111     jmp(*L_success);
 4112   }
 4113 
 4114 #undef IS_A_TEMP
 4115 
 4116   bind(L_fallthrough);
 4117 }
 4118 
 4119 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4120                                                    Register super_klass,
 4121                                                    Register temp_reg,
 4122                                                    Register temp2_reg,
 4123                                                    Label* L_success,
 4124                                                    Label* L_failure,
 4125                                                    bool set_cond_codes) {
 4126   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4127   check_klass_subtype_slow_path
 4128     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4129      L_success, L_failure);
 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                                                    Register temp3_reg,
 4137                                                    Register temp4_reg,
 4138                                                    Label* L_success,
 4139                                                    Label* L_failure) {
 4140   if (UseSecondarySupersTable) {
 4141     check_klass_subtype_slow_path_table
 4142       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4143        L_success, L_failure);
 4144   } else {
 4145     check_klass_subtype_slow_path_linear
 4146       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4147   }
 4148 }
 4149 
 4150 Register MacroAssembler::allocate_if_noreg(Register r,
 4151                                   RegSetIterator<Register> &available_regs,
 4152                                   RegSet &regs_to_push) {
 4153   if (!r->is_valid()) {
 4154     r = *available_regs++;
 4155     regs_to_push += r;
 4156   }
 4157   return r;
 4158 }
 4159 
 4160 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4161                                                          Register super_klass,
 4162                                                          Register temp_reg,
 4163                                                          Register temp2_reg,
 4164                                                          Register temp3_reg,
 4165                                                          Register result_reg,
 4166                                                          Label* L_success,
 4167                                                          Label* L_failure) {
 4168   // NB! Callers may assume that, when temp2_reg is a valid register,
 4169   // this code sets it to a nonzero value.
 4170   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4171 
 4172   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4173 
 4174   Label L_fallthrough;
 4175   int label_nulls = 0;
 4176   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4177   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4178   assert(label_nulls <= 1, "at most one null in the batch");
 4179 
 4180   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4181 
 4182   RegSetIterator<Register> available_regs
 4183     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4184 
 4185   RegSet pushed_regs;
 4186 
 4187   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4188   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4189   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4190   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4191   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4192 
 4193   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4194 
 4195   {
 4196 
 4197     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4198     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4199     subptr(rsp, aligned_size);
 4200     push_set(pushed_regs, 0);
 4201 
 4202     lookup_secondary_supers_table_var(sub_klass,
 4203                                       super_klass,
 4204                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4205     cmpq(result_reg, 0);
 4206 
 4207     // Unspill the temp. registers:
 4208     pop_set(pushed_regs, 0);
 4209     // Increment SP but do not clobber flags.
 4210     lea(rsp, Address(rsp, aligned_size));
 4211   }
 4212 
 4213   if (temp2_reg_was_valid) {
 4214     movq(temp2_reg, 1);
 4215   }
 4216 
 4217   jcc(Assembler::notEqual, *L_failure);
 4218 
 4219   if (L_success != &L_fallthrough) {
 4220     jmp(*L_success);
 4221   }
 4222 
 4223   bind(L_fallthrough);
 4224 }
 4225 
 4226 // population_count variant for running without the POPCNT
 4227 // instruction, which was introduced with SSE4.2 in 2008.
 4228 void MacroAssembler::population_count(Register dst, Register src,
 4229                                       Register scratch1, Register scratch2) {
 4230   assert_different_registers(src, scratch1, scratch2);
 4231   if (UsePopCountInstruction) {
 4232     Assembler::popcntq(dst, src);
 4233   } else {
 4234     assert_different_registers(src, scratch1, scratch2);
 4235     assert_different_registers(dst, scratch1, scratch2);
 4236     Label loop, done;
 4237 
 4238     mov(scratch1, src);
 4239     // dst = 0;
 4240     // while(scratch1 != 0) {
 4241     //   dst++;
 4242     //   scratch1 &= (scratch1 - 1);
 4243     // }
 4244     xorl(dst, dst);
 4245     testq(scratch1, scratch1);
 4246     jccb(Assembler::equal, done);
 4247     {
 4248       bind(loop);
 4249       incq(dst);
 4250       movq(scratch2, scratch1);
 4251       decq(scratch2);
 4252       andq(scratch1, scratch2);
 4253       jccb(Assembler::notEqual, loop);
 4254     }
 4255     bind(done);
 4256   }
 4257 #ifdef ASSERT
 4258   mov64(scratch1, 0xCafeBabeDeadBeef);
 4259   movq(scratch2, scratch1);
 4260 #endif
 4261 }
 4262 
 4263 // Ensure that the inline code and the stub are using the same registers.
 4264 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4265 do {                                                                 \
 4266   assert(r_super_klass  == rax, "mismatch");                         \
 4267   assert(r_array_base   == rbx, "mismatch");                         \
 4268   assert(r_array_length == rcx, "mismatch");                         \
 4269   assert(r_array_index  == rdx, "mismatch");                         \
 4270   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4271   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4272   assert(result         == rdi || result      == noreg, "mismatch"); \
 4273 } while(0)
 4274 
 4275 // Versions of salq and rorq that don't need count to be in rcx
 4276 
 4277 void MacroAssembler::salq(Register dest, Register count) {
 4278   if (count == rcx) {
 4279     Assembler::salq(dest);
 4280   } else {
 4281     assert_different_registers(rcx, dest);
 4282     xchgq(rcx, count);
 4283     Assembler::salq(dest);
 4284     xchgq(rcx, count);
 4285   }
 4286 }
 4287 
 4288 void MacroAssembler::rorq(Register dest, Register count) {
 4289   if (count == rcx) {
 4290     Assembler::rorq(dest);
 4291   } else {
 4292     assert_different_registers(rcx, dest);
 4293     xchgq(rcx, count);
 4294     Assembler::rorq(dest);
 4295     xchgq(rcx, count);
 4296   }
 4297 }
 4298 
 4299 // Return true: we succeeded in generating this code
 4300 //
 4301 // At runtime, return 0 in result if r_super_klass is a superclass of
 4302 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4303 // super_klass_slot of the class you're looking for. This is always
 4304 // the case for instanceof and checkcast.
 4305 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4306                                                          Register r_super_klass,
 4307                                                          Register temp1,
 4308                                                          Register temp2,
 4309                                                          Register temp3,
 4310                                                          Register temp4,
 4311                                                          Register result,
 4312                                                          u1 super_klass_slot) {
 4313   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4314 
 4315   Label L_fallthrough, L_success, L_failure;
 4316 
 4317   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4318 
 4319   const Register
 4320     r_array_index  = temp1,
 4321     r_array_length = temp2,
 4322     r_array_base   = temp3,
 4323     r_bitmap       = temp4;
 4324 
 4325   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4326 
 4327   xorq(result, result); // = 0
 4328 
 4329   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4330   movq(r_array_index, r_bitmap);
 4331 
 4332   // First check the bitmap to see if super_klass might be present. If
 4333   // the bit is zero, we are certain that super_klass is not one of
 4334   // the secondary supers.
 4335   u1 bit = super_klass_slot;
 4336   {
 4337     // NB: If the count in a x86 shift instruction is 0, the flags are
 4338     // not affected, so we do a testq instead.
 4339     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4340     if (shift_count != 0) {
 4341       salq(r_array_index, shift_count);
 4342     } else {
 4343       testq(r_array_index, r_array_index);
 4344     }
 4345   }
 4346   // We test the MSB of r_array_index, i.e. its sign bit
 4347   jcc(Assembler::positive, L_failure);
 4348 
 4349   // Get the first array index that can contain super_klass into r_array_index.
 4350   if (bit != 0) {
 4351     population_count(r_array_index, r_array_index, temp2, temp3);
 4352   } else {
 4353     movl(r_array_index, 1);
 4354   }
 4355   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4356 
 4357   // We will consult the secondary-super array.
 4358   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4359 
 4360   // We're asserting that the first word in an Array<Klass*> is the
 4361   // length, and the second word is the first word of the data. If
 4362   // that ever changes, r_array_base will have to be adjusted here.
 4363   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4364   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4365 
 4366   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4367   jccb(Assembler::equal, L_success);
 4368 
 4369   // Is there another entry to check? Consult the bitmap.
 4370   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4371   jccb(Assembler::carryClear, L_failure);
 4372 
 4373   // Linear probe. Rotate the bitmap so that the next bit to test is
 4374   // in Bit 1.
 4375   if (bit != 0) {
 4376     rorq(r_bitmap, bit);
 4377   }
 4378 
 4379   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4380   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4381   // Kills: r_array_length.
 4382   // Returns: result.
 4383   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4384   // Result (0/1) is in rdi
 4385   jmpb(L_fallthrough);
 4386 
 4387   bind(L_failure);
 4388   incq(result); // 0 => 1
 4389 
 4390   bind(L_success);
 4391   // result = 0;
 4392 
 4393   bind(L_fallthrough);
 4394   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4395 
 4396   if (VerifySecondarySupers) {
 4397     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4398                                   temp1, temp2, temp3);
 4399   }
 4400 }
 4401 
 4402 // At runtime, return 0 in result if r_super_klass is a superclass of
 4403 // r_sub_klass, otherwise return nonzero. Use this version of
 4404 // lookup_secondary_supers_table() if you don't know ahead of time
 4405 // which superclass will be searched for. Used by interpreter and
 4406 // runtime stubs. It is larger and has somewhat greater latency than
 4407 // the version above, which takes a constant super_klass_slot.
 4408 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4409                                                        Register r_super_klass,
 4410                                                        Register temp1,
 4411                                                        Register temp2,
 4412                                                        Register temp3,
 4413                                                        Register temp4,
 4414                                                        Register result) {
 4415   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4416   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4417   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4418 
 4419   Label L_fallthrough, L_success, L_failure;
 4420 
 4421   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4422 
 4423   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4424 
 4425   // FIXME. Once we are sure that all paths reaching this point really
 4426   // do pass rcx as one of our temps we can get rid of the following
 4427   // workaround.
 4428   assert(temps.contains(rcx), "fix this code");
 4429 
 4430   // We prefer to have our shift count in rcx. If rcx is one of our
 4431   // temps, use it for slot. If not, pick any of our temps.
 4432   Register slot;
 4433   if (!temps.contains(rcx)) {
 4434     slot = *available_regs++;
 4435   } else {
 4436     slot = rcx;
 4437   }
 4438 
 4439   const Register r_array_index = *available_regs++;
 4440   const Register r_bitmap      = *available_regs++;
 4441 
 4442   // The logic above guarantees this property, but we state it here.
 4443   assert_different_registers(r_array_index, r_bitmap, rcx);
 4444 
 4445   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4446   movq(r_array_index, r_bitmap);
 4447 
 4448   // First check the bitmap to see if super_klass might be present. If
 4449   // the bit is zero, we are certain that super_klass is not one of
 4450   // the secondary supers.
 4451   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4452   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4453   salq(r_array_index, slot);
 4454 
 4455   testq(r_array_index, r_array_index);
 4456   // We test the MSB of r_array_index, i.e. its sign bit
 4457   jcc(Assembler::positive, L_failure);
 4458 
 4459   const Register r_array_base = *available_regs++;
 4460 
 4461   // Get the first array index that can contain super_klass into r_array_index.
 4462   // Note: Clobbers r_array_base and slot.
 4463   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4464 
 4465   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4466 
 4467   // We will consult the secondary-super array.
 4468   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4469 
 4470   // We're asserting that the first word in an Array<Klass*> is the
 4471   // length, and the second word is the first word of the data. If
 4472   // that ever changes, r_array_base will have to be adjusted here.
 4473   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4474   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4475 
 4476   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4477   jccb(Assembler::equal, L_success);
 4478 
 4479   // Restore slot to its true value
 4480   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4481 
 4482   // Linear probe. Rotate the bitmap so that the next bit to test is
 4483   // in Bit 1.
 4484   rorq(r_bitmap, slot);
 4485 
 4486   // Is there another entry to check? Consult the bitmap.
 4487   btq(r_bitmap, 1);
 4488   jccb(Assembler::carryClear, L_failure);
 4489 
 4490   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4491   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4492   // Kills: r_array_length.
 4493   // Returns: result.
 4494   lookup_secondary_supers_table_slow_path(r_super_klass,
 4495                                           r_array_base,
 4496                                           r_array_index,
 4497                                           r_bitmap,
 4498                                           /*temp1*/result,
 4499                                           /*temp2*/slot,
 4500                                           &L_success,
 4501                                           nullptr);
 4502 
 4503   bind(L_failure);
 4504   movq(result, 1);
 4505   jmpb(L_fallthrough);
 4506 
 4507   bind(L_success);
 4508   xorq(result, result); // = 0
 4509 
 4510   bind(L_fallthrough);
 4511   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4512 
 4513   if (VerifySecondarySupers) {
 4514     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4515                                   temp1, temp2, temp3);
 4516   }
 4517 }
 4518 
 4519 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4520                                  Label* L_success, Label* L_failure) {
 4521   Label L_loop, L_fallthrough;
 4522   {
 4523     int label_nulls = 0;
 4524     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4525     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4526     assert(label_nulls <= 1, "at most one null in the batch");
 4527   }
 4528   bind(L_loop);
 4529   cmpq(value, Address(addr, count, Address::times_8));
 4530   jcc(Assembler::equal, *L_success);
 4531   addl(count, 1);
 4532   cmpl(count, limit);
 4533   jcc(Assembler::less, L_loop);
 4534 
 4535   if (&L_fallthrough != L_failure) {
 4536     jmp(*L_failure);
 4537   }
 4538   bind(L_fallthrough);
 4539 }
 4540 
 4541 // Called by code generated by check_klass_subtype_slow_path
 4542 // above. This is called when there is a collision in the hashed
 4543 // lookup in the secondary supers array.
 4544 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4545                                                              Register r_array_base,
 4546                                                              Register r_array_index,
 4547                                                              Register r_bitmap,
 4548                                                              Register temp1,
 4549                                                              Register temp2,
 4550                                                              Label* L_success,
 4551                                                              Label* L_failure) {
 4552   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4553 
 4554   const Register
 4555     r_array_length = temp1,
 4556     r_sub_klass    = noreg,
 4557     result         = noreg;
 4558 
 4559   Label L_fallthrough;
 4560   int label_nulls = 0;
 4561   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4562   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4563   assert(label_nulls <= 1, "at most one null in the batch");
 4564 
 4565   // Load the array length.
 4566   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4567   // And adjust the array base to point to the data.
 4568   // NB! Effectively increments current slot index by 1.
 4569   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4570   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4571 
 4572   // Linear probe
 4573   Label L_huge;
 4574 
 4575   // The bitmap is full to bursting.
 4576   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4577   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4578   jcc(Assembler::greater, L_huge);
 4579 
 4580   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4581   // current slot (at secondary_supers[r_array_index]) has not yet
 4582   // been inspected, and r_array_index may be out of bounds if we
 4583   // wrapped around the end of the array.
 4584 
 4585   { // This is conventional linear probing, but instead of terminating
 4586     // when a null entry is found in the table, we maintain a bitmap
 4587     // in which a 0 indicates missing entries.
 4588     // The check above guarantees there are 0s in the bitmap, so the loop
 4589     // eventually terminates.
 4590 
 4591     xorl(temp2, temp2); // = 0;
 4592 
 4593     Label L_again;
 4594     bind(L_again);
 4595 
 4596     // Check for array wraparound.
 4597     cmpl(r_array_index, r_array_length);
 4598     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4599 
 4600     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4601     jcc(Assembler::equal, *L_success);
 4602 
 4603     // If the next bit in bitmap is zero, we're done.
 4604     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4605     jcc(Assembler::carryClear, *L_failure);
 4606 
 4607     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4608     addl(r_array_index, 1);
 4609 
 4610     jmp(L_again);
 4611   }
 4612 
 4613   { // Degenerate case: more than 64 secondary supers.
 4614     // FIXME: We could do something smarter here, maybe a vectorized
 4615     // comparison or a binary search, but is that worth any added
 4616     // complexity?
 4617     bind(L_huge);
 4618     xorl(r_array_index, r_array_index); // = 0
 4619     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4620                 L_success,
 4621                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4622 
 4623     bind(L_fallthrough);
 4624   }
 4625 }
 4626 
 4627 struct VerifyHelperArguments {
 4628   Klass* _super;
 4629   Klass* _sub;
 4630   intptr_t _linear_result;
 4631   intptr_t _table_result;
 4632 };
 4633 
 4634 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4635   Klass::on_secondary_supers_verification_failure(args->_super,
 4636                                                   args->_sub,
 4637                                                   args->_linear_result,
 4638                                                   args->_table_result,
 4639                                                   msg);
 4640 }
 4641 
 4642 // Make sure that the hashed lookup and a linear scan agree.
 4643 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4644                                                    Register r_super_klass,
 4645                                                    Register result,
 4646                                                    Register temp1,
 4647                                                    Register temp2,
 4648                                                    Register temp3) {
 4649   const Register
 4650       r_array_index  = temp1,
 4651       r_array_length = temp2,
 4652       r_array_base   = temp3,
 4653       r_bitmap       = noreg;
 4654 
 4655   BLOCK_COMMENT("verify_secondary_supers_table {");
 4656 
 4657   Label L_success, L_failure, L_check, L_done;
 4658 
 4659   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4660   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4661   // And adjust the array base to point to the data.
 4662   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4663 
 4664   testl(r_array_length, r_array_length); // array_length == 0?
 4665   jcc(Assembler::zero, L_failure);
 4666 
 4667   movl(r_array_index, 0);
 4668   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4669   // fall through to L_failure
 4670 
 4671   const Register linear_result = r_array_index; // reuse temp1
 4672 
 4673   bind(L_failure); // not present
 4674   movl(linear_result, 1);
 4675   jmp(L_check);
 4676 
 4677   bind(L_success); // present
 4678   movl(linear_result, 0);
 4679 
 4680   bind(L_check);
 4681   cmpl(linear_result, result);
 4682   jcc(Assembler::equal, L_done);
 4683 
 4684   { // To avoid calling convention issues, build a record on the stack
 4685     // and pass the pointer to that instead.
 4686     push(result);
 4687     push(linear_result);
 4688     push(r_sub_klass);
 4689     push(r_super_klass);
 4690     movptr(c_rarg1, rsp);
 4691     movptr(c_rarg0, (uintptr_t) "mismatch");
 4692     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4693     should_not_reach_here();
 4694   }
 4695   bind(L_done);
 4696 
 4697   BLOCK_COMMENT("} verify_secondary_supers_table");
 4698 }
 4699 
 4700 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4701 
 4702 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4703   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4704 
 4705   Label L_fallthrough;
 4706   if (L_fast_path == nullptr) {
 4707     L_fast_path = &L_fallthrough;
 4708   } else if (L_slow_path == nullptr) {
 4709     L_slow_path = &L_fallthrough;
 4710   }
 4711 
 4712   // Fast path check: class is fully initialized.
 4713   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4714   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4715   jcc(Assembler::equal, *L_fast_path);
 4716 
 4717   // Fast path check: current thread is initializer thread
 4718   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4719   if (L_slow_path == &L_fallthrough) {
 4720     jcc(Assembler::equal, *L_fast_path);
 4721     bind(*L_slow_path);
 4722   } else if (L_fast_path == &L_fallthrough) {
 4723     jcc(Assembler::notEqual, *L_slow_path);
 4724     bind(*L_fast_path);
 4725   } else {
 4726     Unimplemented();
 4727   }
 4728 }
 4729 
 4730 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4731   if (VM_Version::supports_cmov()) {
 4732     cmovl(cc, dst, src);
 4733   } else {
 4734     Label L;
 4735     jccb(negate_condition(cc), L);
 4736     movl(dst, src);
 4737     bind(L);
 4738   }
 4739 }
 4740 
 4741 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4742   if (VM_Version::supports_cmov()) {
 4743     cmovl(cc, dst, src);
 4744   } else {
 4745     Label L;
 4746     jccb(negate_condition(cc), L);
 4747     movl(dst, src);
 4748     bind(L);
 4749   }
 4750 }
 4751 
 4752 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4753   if (!VerifyOops) return;
 4754 
 4755   BLOCK_COMMENT("verify_oop {");
 4756   push(rscratch1);
 4757   push(rax);                          // save rax
 4758   push(reg);                          // pass register argument
 4759 
 4760   // Pass register number to verify_oop_subroutine
 4761   const char* b = nullptr;
 4762   {
 4763     ResourceMark rm;
 4764     stringStream ss;
 4765     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4766     b = code_string(ss.as_string());
 4767   }
 4768   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4769   pushptr(buffer.addr(), rscratch1);
 4770 
 4771   // call indirectly to solve generation ordering problem
 4772   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4773   call(rax);
 4774   // Caller pops the arguments (oop, message) and restores rax, r10
 4775   BLOCK_COMMENT("} verify_oop");
 4776 }
 4777 
 4778 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 4779   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 4780     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 4781     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 4782     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 4783   } else if (VM_Version::supports_avx()) {
 4784     vpcmpeqd(dst, dst, dst, vector_len);
 4785   } else {
 4786     pcmpeqd(dst, dst);
 4787   }
 4788 }
 4789 
 4790 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 4791                                          int extra_slot_offset) {
 4792   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 4793   int stackElementSize = Interpreter::stackElementSize;
 4794   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 4795 #ifdef ASSERT
 4796   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 4797   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 4798 #endif
 4799   Register             scale_reg    = noreg;
 4800   Address::ScaleFactor scale_factor = Address::no_scale;
 4801   if (arg_slot.is_constant()) {
 4802     offset += arg_slot.as_constant() * stackElementSize;
 4803   } else {
 4804     scale_reg    = arg_slot.as_register();
 4805     scale_factor = Address::times(stackElementSize);
 4806   }
 4807   offset += wordSize;           // return PC is on stack
 4808   return Address(rsp, scale_reg, scale_factor, offset);
 4809 }
 4810 
 4811 // Handle the receiver type profile update given the "recv" klass.
 4812 //
 4813 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 4814 // If there are no matching or claimable receiver entries in RD, updates
 4815 // the polymorphic counter.
 4816 //
 4817 // This code expected to run by either the interpreter or JIT-ed code, without
 4818 // extra synchronization. For safety, receiver cells are claimed atomically, which
 4819 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 4820 // counter updates are not atomic.
 4821 //
 4822 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 4823   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 4824   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 4825   int poly_count_offset      = in_bytes(CounterData::count_offset());
 4826   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 4827   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 4828 
 4829   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 4830   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 4831   base_receiver_offset += mdp_offset;
 4832   end_receiver_offset  += mdp_offset;
 4833   poly_count_offset    += mdp_offset;
 4834 
 4835   // Scale down to optimize encoding. Slots are pointer-sized.
 4836   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 4837   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 4838   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 4839   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 4840   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 4841   base_receiver_offset   >>= LogBytesPerWord;
 4842   end_receiver_offset    >>= LogBytesPerWord;
 4843   poly_count_offset      >>= LogBytesPerWord;
 4844   receiver_step          >>= LogBytesPerWord;
 4845   receiver_to_count_step >>= LogBytesPerWord;
 4846 
 4847 #ifdef ASSERT
 4848   // We are about to walk the MDO slots without asking for offsets.
 4849   // Check that our math hits all the right spots.
 4850   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 4851     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 4852     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 4853     int offset = base_receiver_offset + receiver_step*c;
 4854     int count_offset = offset + receiver_to_count_step;
 4855     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 4856     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 4857   }
 4858   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 4859   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 4860 #endif
 4861 
 4862   // Corner case: no profile table. Increment poly counter and exit.
 4863   if (ReceiverTypeData::row_limit() == 0) {
 4864     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 4865     return;
 4866   }
 4867 
 4868   Register offset = rscratch1;
 4869 
 4870   Label L_loop_search_receiver, L_loop_search_empty;
 4871   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 4872 
 4873   // The code here recognizes three major cases:
 4874   //   A. Fastest: receiver found in the table
 4875   //   B. Fast: no receiver in the table, and the table is full
 4876   //   C. Slow: no receiver in the table, free slots in the table
 4877   //
 4878   // The case A performance is most important, as perfectly-behaved code would end up
 4879   // there, especially with larger TypeProfileWidth. The case B performance is
 4880   // important as well, this is where bulk of code would land for normally megamorphic
 4881   // cases. The case C performance is not essential, its job is to deal with installation
 4882   // races, we optimize for code density instead. Case C needs to make sure that receiver
 4883   // rows are only claimed once. This makes sure we never overwrite a row for another
 4884   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 4885   //
 4886   // It is very tempting to handle these cases in a single loop, and claim the first slot
 4887   // without checking the rest of the table. But, profiling code should tolerate free slots
 4888   // in the table, as class unloading can clear them. After such cleanup, the receiver
 4889   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 4890   // to complete, before trying to install new slots. Splitting the code in several tight
 4891   // loops also helpfully optimizes for cases A and B.
 4892   //
 4893   // This code is effectively:
 4894   //
 4895   // restart:
 4896   //   // Fastest: receiver is already installed
 4897   //   for (i = 0; i < receiver_count(); i++) {
 4898   //     if (receiver(i) == recv) goto found_recv(i);
 4899   //   }
 4900   //
 4901   //   // Fast: no receiver, but profile is full
 4902   //   for (i = 0; i < receiver_count(); i++) {
 4903   //     if (receiver(i) == null) goto found_null(i);
 4904   //   }
 4905   //   goto polymorphic
 4906   //
 4907   //   // Slow: try to install receiver
 4908   // found_null(i):
 4909   //   CAS(&receiver(i), null, recv);
 4910   //   goto restart
 4911   //
 4912   // polymorphic:
 4913   //   count++;
 4914   //   return
 4915   //
 4916   // found_recv(i):
 4917   //   *receiver_count(i)++
 4918   //
 4919 
 4920   bind(L_restart);
 4921 
 4922   // Fastest: receiver is already installed
 4923   movptr(offset, base_receiver_offset);
 4924   bind(L_loop_search_receiver);
 4925     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 4926     jccb(Assembler::equal, L_found_recv);
 4927   addptr(offset, receiver_step);
 4928   cmpptr(offset, end_receiver_offset);
 4929   jccb(Assembler::notEqual, L_loop_search_receiver);
 4930 
 4931   // Fast: no receiver, but profile is full
 4932   movptr(offset, base_receiver_offset);
 4933   bind(L_loop_search_empty);
 4934     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 4935     jccb(Assembler::equal, L_found_empty);
 4936   addptr(offset, receiver_step);
 4937   cmpptr(offset, end_receiver_offset);
 4938   jccb(Assembler::notEqual, L_loop_search_empty);
 4939   jmpb(L_polymorphic);
 4940 
 4941   // Slow: try to install receiver
 4942   bind(L_found_empty);
 4943 
 4944   // Atomically swing receiver slot: null -> recv.
 4945   //
 4946   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 4947   // other important registers untouched, as they form the address. Therefore, we need
 4948   // to shift any important registers from RAX into some other spare register. If we
 4949   // have a spare register, we are forced to save it on stack here.
 4950 
 4951   Register spare_reg = noreg;
 4952   Register shifted_mdp = mdp;
 4953   Register shifted_recv = recv;
 4954   if (recv == rax || mdp == rax) {
 4955     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 4956                 (recv != rcx && mdp != rcx) ? rcx :
 4957                 rdx;
 4958     assert_different_registers(mdp, recv, offset, spare_reg);
 4959 
 4960     push(spare_reg);
 4961     if (recv == rax) {
 4962       movptr(spare_reg, recv);
 4963       shifted_recv = spare_reg;
 4964     } else {
 4965       assert(mdp == rax, "Remaining case");
 4966       movptr(spare_reg, mdp);
 4967       shifted_mdp = spare_reg;
 4968     }
 4969   } else {
 4970     push(rax);
 4971   }
 4972 
 4973   // None of the important registers are in RAX after this shuffle.
 4974   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 4975 
 4976   xorptr(rax, rax);
 4977   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 4978 
 4979   // Unshift registers.
 4980   if (recv == rax || mdp == rax) {
 4981     movptr(rax, spare_reg);
 4982     pop(spare_reg);
 4983   } else {
 4984     pop(rax);
 4985   }
 4986 
 4987   // CAS success means the slot now has the receiver we want. CAS failure means
 4988   // something had claimed the slot concurrently: it can be the same receiver we want,
 4989   // or something else. Since this is a slow path, we can optimize for code density,
 4990   // and just restart the search from the beginning.
 4991   jmpb(L_restart);
 4992 
 4993   // Counter updates:
 4994 
 4995   // Increment polymorphic counter instead of receiver slot.
 4996   bind(L_polymorphic);
 4997   movptr(offset, poly_count_offset);
 4998   jmpb(L_count_update);
 4999 
 5000   // Found a receiver, convert its slot offset to corresponding count offset.
 5001   bind(L_found_recv);
 5002   addptr(offset, receiver_to_count_step);
 5003 
 5004   bind(L_count_update);
 5005   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 5006 }
 5007 
 5008 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5009   if (!VerifyOops) return;
 5010 
 5011   push(rscratch1);
 5012   push(rax); // save rax,
 5013   // addr may contain rsp so we will have to adjust it based on the push
 5014   // we just did (and on 64 bit we do two pushes)
 5015   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5016   // stores rax into addr which is backwards of what was intended.
 5017   if (addr.uses(rsp)) {
 5018     lea(rax, addr);
 5019     pushptr(Address(rax, 2 * BytesPerWord));
 5020   } else {
 5021     pushptr(addr);
 5022   }
 5023 
 5024   // Pass register number to verify_oop_subroutine
 5025   const char* b = nullptr;
 5026   {
 5027     ResourceMark rm;
 5028     stringStream ss;
 5029     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 5030     b = code_string(ss.as_string());
 5031   }
 5032   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5033   pushptr(buffer.addr(), rscratch1);
 5034 
 5035   // call indirectly to solve generation ordering problem
 5036   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5037   call(rax);
 5038   // Caller pops the arguments (addr, message) and restores rax, r10.
 5039 }
 5040 
 5041 void MacroAssembler::verify_tlab() {
 5042 #ifdef ASSERT
 5043   if (UseTLAB && VerifyOops) {
 5044     Label next, ok;
 5045     Register t1 = rsi;
 5046 
 5047     push(t1);
 5048 
 5049     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5050     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 5051     jcc(Assembler::aboveEqual, next);
 5052     STOP("assert(top >= start)");
 5053     should_not_reach_here();
 5054 
 5055     bind(next);
 5056     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 5057     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5058     jcc(Assembler::aboveEqual, ok);
 5059     STOP("assert(top <= end)");
 5060     should_not_reach_here();
 5061 
 5062     bind(ok);
 5063     pop(t1);
 5064   }
 5065 #endif
 5066 }
 5067 
 5068 class ControlWord {
 5069  public:
 5070   int32_t _value;
 5071 
 5072   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5073   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5074   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5075   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5076   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5077   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5078   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5079   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5080 
 5081   void print() const {
 5082     // rounding control
 5083     const char* rc;
 5084     switch (rounding_control()) {
 5085       case 0: rc = "round near"; break;
 5086       case 1: rc = "round down"; break;
 5087       case 2: rc = "round up  "; break;
 5088       case 3: rc = "chop      "; break;
 5089       default:
 5090         rc = nullptr; // silence compiler warnings
 5091         fatal("Unknown rounding control: %d", rounding_control());
 5092     };
 5093     // precision control
 5094     const char* pc;
 5095     switch (precision_control()) {
 5096       case 0: pc = "24 bits "; break;
 5097       case 1: pc = "reserved"; break;
 5098       case 2: pc = "53 bits "; break;
 5099       case 3: pc = "64 bits "; break;
 5100       default:
 5101         pc = nullptr; // silence compiler warnings
 5102         fatal("Unknown precision control: %d", precision_control());
 5103     };
 5104     // flags
 5105     char f[9];
 5106     f[0] = ' ';
 5107     f[1] = ' ';
 5108     f[2] = (precision   ()) ? 'P' : 'p';
 5109     f[3] = (underflow   ()) ? 'U' : 'u';
 5110     f[4] = (overflow    ()) ? 'O' : 'o';
 5111     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5112     f[6] = (denormalized()) ? 'D' : 'd';
 5113     f[7] = (invalid     ()) ? 'I' : 'i';
 5114     f[8] = '\x0';
 5115     // output
 5116     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5117   }
 5118 
 5119 };
 5120 
 5121 class StatusWord {
 5122  public:
 5123   int32_t _value;
 5124 
 5125   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5126   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5127   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5128   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5129   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5130   int  top() const                     { return  (_value >> 11) & 7      ; }
 5131   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5132   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5133   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5134   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5135   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5136   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5137   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5138   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5139 
 5140   void print() const {
 5141     // condition codes
 5142     char c[5];
 5143     c[0] = (C3()) ? '3' : '-';
 5144     c[1] = (C2()) ? '2' : '-';
 5145     c[2] = (C1()) ? '1' : '-';
 5146     c[3] = (C0()) ? '0' : '-';
 5147     c[4] = '\x0';
 5148     // flags
 5149     char f[9];
 5150     f[0] = (error_status()) ? 'E' : '-';
 5151     f[1] = (stack_fault ()) ? 'S' : '-';
 5152     f[2] = (precision   ()) ? 'P' : '-';
 5153     f[3] = (underflow   ()) ? 'U' : '-';
 5154     f[4] = (overflow    ()) ? 'O' : '-';
 5155     f[5] = (zero_divide ()) ? 'Z' : '-';
 5156     f[6] = (denormalized()) ? 'D' : '-';
 5157     f[7] = (invalid     ()) ? 'I' : '-';
 5158     f[8] = '\x0';
 5159     // output
 5160     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5161   }
 5162 
 5163 };
 5164 
 5165 class TagWord {
 5166  public:
 5167   int32_t _value;
 5168 
 5169   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5170 
 5171   void print() const {
 5172     printf("%04x", _value & 0xFFFF);
 5173   }
 5174 
 5175 };
 5176 
 5177 class FPU_Register {
 5178  public:
 5179   int32_t _m0;
 5180   int32_t _m1;
 5181   int16_t _ex;
 5182 
 5183   bool is_indefinite() const           {
 5184     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5185   }
 5186 
 5187   void print() const {
 5188     char  sign = (_ex < 0) ? '-' : '+';
 5189     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5190     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5191   };
 5192 
 5193 };
 5194 
 5195 class FPU_State {
 5196  public:
 5197   enum {
 5198     register_size       = 10,
 5199     number_of_registers =  8,
 5200     register_mask       =  7
 5201   };
 5202 
 5203   ControlWord  _control_word;
 5204   StatusWord   _status_word;
 5205   TagWord      _tag_word;
 5206   int32_t      _error_offset;
 5207   int32_t      _error_selector;
 5208   int32_t      _data_offset;
 5209   int32_t      _data_selector;
 5210   int8_t       _register[register_size * number_of_registers];
 5211 
 5212   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5213   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5214 
 5215   const char* tag_as_string(int tag) const {
 5216     switch (tag) {
 5217       case 0: return "valid";
 5218       case 1: return "zero";
 5219       case 2: return "special";
 5220       case 3: return "empty";
 5221     }
 5222     ShouldNotReachHere();
 5223     return nullptr;
 5224   }
 5225 
 5226   void print() const {
 5227     // print computation registers
 5228     { int t = _status_word.top();
 5229       for (int i = 0; i < number_of_registers; i++) {
 5230         int j = (i - t) & register_mask;
 5231         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5232         st(j)->print();
 5233         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5234       }
 5235     }
 5236     printf("\n");
 5237     // print control registers
 5238     printf("ctrl = "); _control_word.print(); printf("\n");
 5239     printf("stat = "); _status_word .print(); printf("\n");
 5240     printf("tags = "); _tag_word    .print(); printf("\n");
 5241   }
 5242 
 5243 };
 5244 
 5245 class Flag_Register {
 5246  public:
 5247   int32_t _value;
 5248 
 5249   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5250   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5251   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5252   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5253   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5254   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5255   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5256 
 5257   void print() const {
 5258     // flags
 5259     char f[8];
 5260     f[0] = (overflow       ()) ? 'O' : '-';
 5261     f[1] = (direction      ()) ? 'D' : '-';
 5262     f[2] = (sign           ()) ? 'S' : '-';
 5263     f[3] = (zero           ()) ? 'Z' : '-';
 5264     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5265     f[5] = (parity         ()) ? 'P' : '-';
 5266     f[6] = (carry          ()) ? 'C' : '-';
 5267     f[7] = '\x0';
 5268     // output
 5269     printf("%08x  flags = %s", _value, f);
 5270   }
 5271 
 5272 };
 5273 
 5274 class IU_Register {
 5275  public:
 5276   int32_t _value;
 5277 
 5278   void print() const {
 5279     printf("%08x  %11d", _value, _value);
 5280   }
 5281 
 5282 };
 5283 
 5284 class IU_State {
 5285  public:
 5286   Flag_Register _eflags;
 5287   IU_Register   _rdi;
 5288   IU_Register   _rsi;
 5289   IU_Register   _rbp;
 5290   IU_Register   _rsp;
 5291   IU_Register   _rbx;
 5292   IU_Register   _rdx;
 5293   IU_Register   _rcx;
 5294   IU_Register   _rax;
 5295 
 5296   void print() const {
 5297     // computation registers
 5298     printf("rax,  = "); _rax.print(); printf("\n");
 5299     printf("rbx,  = "); _rbx.print(); printf("\n");
 5300     printf("rcx  = "); _rcx.print(); printf("\n");
 5301     printf("rdx  = "); _rdx.print(); printf("\n");
 5302     printf("rdi  = "); _rdi.print(); printf("\n");
 5303     printf("rsi  = "); _rsi.print(); printf("\n");
 5304     printf("rbp,  = "); _rbp.print(); printf("\n");
 5305     printf("rsp  = "); _rsp.print(); printf("\n");
 5306     printf("\n");
 5307     // control registers
 5308     printf("flgs = "); _eflags.print(); printf("\n");
 5309   }
 5310 };
 5311 
 5312 
 5313 class CPU_State {
 5314  public:
 5315   FPU_State _fpu_state;
 5316   IU_State  _iu_state;
 5317 
 5318   void print() const {
 5319     printf("--------------------------------------------------\n");
 5320     _iu_state .print();
 5321     printf("\n");
 5322     _fpu_state.print();
 5323     printf("--------------------------------------------------\n");
 5324   }
 5325 
 5326 };
 5327 
 5328 
 5329 static void _print_CPU_state(CPU_State* state) {
 5330   state->print();
 5331 };
 5332 
 5333 
 5334 void MacroAssembler::print_CPU_state() {
 5335   push_CPU_state();
 5336   push(rsp);                // pass CPU state
 5337   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5338   addptr(rsp, wordSize);       // discard argument
 5339   pop_CPU_state();
 5340 }
 5341 
 5342 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5343   // Either restore the MXCSR register after returning from the JNI Call
 5344   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5345   if (VM_Version::supports_sse()) {
 5346     if (RestoreMXCSROnJNICalls) {
 5347       ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5348     } else if (CheckJNICalls) {
 5349       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5350     }
 5351   }
 5352   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5353   vzeroupper();
 5354 }
 5355 
 5356 // ((OopHandle)result).resolve();
 5357 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5358   assert_different_registers(result, tmp);
 5359 
 5360   // Only 64 bit platforms support GCs that require a tmp register
 5361   // Only IN_HEAP loads require a thread_tmp register
 5362   // OopHandle::resolve is an indirection like jobject.
 5363   access_load_at(T_OBJECT, IN_NATIVE,
 5364                  result, Address(result, 0), tmp);
 5365 }
 5366 
 5367 // ((WeakHandle)result).resolve();
 5368 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5369   assert_different_registers(rresult, rtmp);
 5370   Label resolved;
 5371 
 5372   // A null weak handle resolves to null.
 5373   cmpptr(rresult, 0);
 5374   jcc(Assembler::equal, resolved);
 5375 
 5376   // Only 64 bit platforms support GCs that require a tmp register
 5377   // Only IN_HEAP loads require a thread_tmp register
 5378   // WeakHandle::resolve is an indirection like jweak.
 5379   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5380                  rresult, Address(rresult, 0), rtmp);
 5381   bind(resolved);
 5382 }
 5383 
 5384 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5385   // get mirror
 5386   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5387   load_method_holder(mirror, method);
 5388   movptr(mirror, Address(mirror, mirror_offset));
 5389   resolve_oop_handle(mirror, tmp);
 5390 }
 5391 
 5392 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5393   load_method_holder(rresult, rmethod);
 5394   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5395 }
 5396 
 5397 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5398   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5399   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5400   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5401 }
 5402 
 5403 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5404   assert(UseCompactObjectHeaders, "expect compact object headers");
 5405   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5406   shrq(dst, markWord::klass_shift);
 5407 }
 5408 
 5409 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5410   assert_different_registers(src, tmp);
 5411   assert_different_registers(dst, tmp);
 5412 
 5413   if (UseCompactObjectHeaders) {
 5414     load_narrow_klass_compact(dst, src);
 5415     decode_klass_not_null(dst, tmp);
 5416   } else {
 5417     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5418     decode_klass_not_null(dst, tmp);
 5419   }
 5420 }
 5421 
 5422 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5423   assert(!UseCompactObjectHeaders, "not with compact headers");
 5424   assert_different_registers(src, tmp);
 5425   assert_different_registers(dst, tmp);
 5426   encode_klass_not_null(src, tmp);
 5427   movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5428 }
 5429 
 5430 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5431   if (UseCompactObjectHeaders) {
 5432     assert(tmp != noreg, "need tmp");
 5433     assert_different_registers(klass, obj, tmp);
 5434     load_narrow_klass_compact(tmp, obj);
 5435     cmpl(klass, tmp);
 5436   } else {
 5437     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5438   }
 5439 }
 5440 
 5441 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5442   if (UseCompactObjectHeaders) {
 5443     assert(tmp2 != noreg, "need tmp2");
 5444     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5445     load_narrow_klass_compact(tmp1, obj1);
 5446     load_narrow_klass_compact(tmp2, obj2);
 5447     cmpl(tmp1, tmp2);
 5448   } else {
 5449     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5450     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5451   }
 5452 }
 5453 
 5454 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5455                                     Register tmp1) {
 5456   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5457   decorators = AccessInternal::decorator_fixup(decorators, type);
 5458   bool as_raw = (decorators & AS_RAW) != 0;
 5459   if (as_raw) {
 5460     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5461   } else {
 5462     bs->load_at(this, decorators, type, dst, src, tmp1);
 5463   }
 5464 }
 5465 
 5466 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5467                                      Register tmp1, Register tmp2, Register tmp3) {
 5468   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5469   decorators = AccessInternal::decorator_fixup(decorators, type);
 5470   bool as_raw = (decorators & AS_RAW) != 0;
 5471   if (as_raw) {
 5472     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5473   } else {
 5474     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5475   }
 5476 }
 5477 
 5478 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5479   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5480 }
 5481 
 5482 // Doesn't do verification, generates fixed size code
 5483 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5484   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5485 }
 5486 
 5487 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5488                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5489   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5490 }
 5491 
 5492 // Used for storing nulls.
 5493 void MacroAssembler::store_heap_oop_null(Address dst) {
 5494   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5495 }
 5496 
 5497 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5498   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5499   // Store to klass gap in destination
 5500   movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5501 }
 5502 
 5503 #ifdef ASSERT
 5504 void MacroAssembler::verify_heapbase(const char* msg) {
 5505   assert (UseCompressedOops, "should be compressed");
 5506   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5507   if (CheckCompressedOops) {
 5508     Label ok;
 5509     ExternalAddress src2(CompressedOops::base_addr());
 5510     const bool is_src2_reachable = reachable(src2);
 5511     if (!is_src2_reachable) {
 5512       push(rscratch1);  // cmpptr trashes rscratch1
 5513     }
 5514     cmpptr(r12_heapbase, src2, rscratch1);
 5515     jcc(Assembler::equal, ok);
 5516     STOP(msg);
 5517     bind(ok);
 5518     if (!is_src2_reachable) {
 5519       pop(rscratch1);
 5520     }
 5521   }
 5522 }
 5523 #endif
 5524 
 5525 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5526 void MacroAssembler::encode_heap_oop(Register r) {
 5527 #ifdef ASSERT
 5528   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5529 #endif
 5530   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5531   if (CompressedOops::base() == nullptr) {
 5532     if (CompressedOops::shift() != 0) {
 5533       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5534       shrq(r, LogMinObjAlignmentInBytes);
 5535     }
 5536     return;
 5537   }
 5538   testq(r, r);
 5539   cmovq(Assembler::equal, r, r12_heapbase);
 5540   subq(r, r12_heapbase);
 5541   shrq(r, LogMinObjAlignmentInBytes);
 5542 }
 5543 
 5544 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5545 #ifdef ASSERT
 5546   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5547   if (CheckCompressedOops) {
 5548     Label ok;
 5549     testq(r, r);
 5550     jcc(Assembler::notEqual, ok);
 5551     STOP("null oop passed to encode_heap_oop_not_null");
 5552     bind(ok);
 5553   }
 5554 #endif
 5555   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5556   if (CompressedOops::base() != nullptr) {
 5557     subq(r, r12_heapbase);
 5558   }
 5559   if (CompressedOops::shift() != 0) {
 5560     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5561     shrq(r, LogMinObjAlignmentInBytes);
 5562   }
 5563 }
 5564 
 5565 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5566 #ifdef ASSERT
 5567   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5568   if (CheckCompressedOops) {
 5569     Label ok;
 5570     testq(src, src);
 5571     jcc(Assembler::notEqual, ok);
 5572     STOP("null oop passed to encode_heap_oop_not_null2");
 5573     bind(ok);
 5574   }
 5575 #endif
 5576   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5577   if (dst != src) {
 5578     movq(dst, src);
 5579   }
 5580   if (CompressedOops::base() != nullptr) {
 5581     subq(dst, r12_heapbase);
 5582   }
 5583   if (CompressedOops::shift() != 0) {
 5584     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5585     shrq(dst, LogMinObjAlignmentInBytes);
 5586   }
 5587 }
 5588 
 5589 void  MacroAssembler::decode_heap_oop(Register r) {
 5590 #ifdef ASSERT
 5591   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5592 #endif
 5593   if (CompressedOops::base() == nullptr) {
 5594     if (CompressedOops::shift() != 0) {
 5595       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5596       shlq(r, LogMinObjAlignmentInBytes);
 5597     }
 5598   } else {
 5599     Label done;
 5600     shlq(r, LogMinObjAlignmentInBytes);
 5601     jccb(Assembler::equal, done);
 5602     addq(r, r12_heapbase);
 5603     bind(done);
 5604   }
 5605   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5606 }
 5607 
 5608 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5609   // Note: it will change flags
 5610   assert (UseCompressedOops, "should only be used for compressed headers");
 5611   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5612   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5613   // vtableStubs also counts instructions in pd_code_size_limit.
 5614   // Also do not verify_oop as this is called by verify_oop.
 5615   if (CompressedOops::shift() != 0) {
 5616     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5617     shlq(r, LogMinObjAlignmentInBytes);
 5618     if (CompressedOops::base() != nullptr) {
 5619       addq(r, r12_heapbase);
 5620     }
 5621   } else {
 5622     assert (CompressedOops::base() == nullptr, "sanity");
 5623   }
 5624 }
 5625 
 5626 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5627   // Note: it will change flags
 5628   assert (UseCompressedOops, "should only be used for compressed headers");
 5629   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5630   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5631   // vtableStubs also counts instructions in pd_code_size_limit.
 5632   // Also do not verify_oop as this is called by verify_oop.
 5633   if (CompressedOops::shift() != 0) {
 5634     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5635     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5636       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5637     } else {
 5638       if (dst != src) {
 5639         movq(dst, src);
 5640       }
 5641       shlq(dst, LogMinObjAlignmentInBytes);
 5642       if (CompressedOops::base() != nullptr) {
 5643         addq(dst, r12_heapbase);
 5644       }
 5645     }
 5646   } else {
 5647     assert (CompressedOops::base() == nullptr, "sanity");
 5648     if (dst != src) {
 5649       movq(dst, src);
 5650     }
 5651   }
 5652 }
 5653 
 5654 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5655   BLOCK_COMMENT("encode_klass_not_null {");
 5656   assert_different_registers(r, tmp);
 5657   if (CompressedKlassPointers::base() != nullptr) {
 5658     if (AOTCodeCache::is_on_for_dump()) {
 5659       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5660     } else {
 5661       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5662     }
 5663     subq(r, tmp);
 5664   }
 5665   if (CompressedKlassPointers::shift() != 0) {
 5666     shrq(r, CompressedKlassPointers::shift());
 5667   }
 5668   BLOCK_COMMENT("} encode_klass_not_null");
 5669 }
 5670 
 5671 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5672   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5673   assert_different_registers(src, dst);
 5674   if (CompressedKlassPointers::base() != nullptr) {
 5675     movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5676     addq(dst, src);
 5677   } else {
 5678     movptr(dst, src);
 5679   }
 5680   if (CompressedKlassPointers::shift() != 0) {
 5681     shrq(dst, CompressedKlassPointers::shift());
 5682   }
 5683   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5684 }
 5685 
 5686 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5687   BLOCK_COMMENT("decode_klass_not_null {");
 5688   assert_different_registers(r, tmp);
 5689   // Note: it will change flags
 5690   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5691   // vtableStubs also counts instructions in pd_code_size_limit.
 5692   // Also do not verify_oop as this is called by verify_oop.
 5693   if (CompressedKlassPointers::shift() != 0) {
 5694     shlq(r, CompressedKlassPointers::shift());
 5695   }
 5696   if (CompressedKlassPointers::base() != nullptr) {
 5697     if (AOTCodeCache::is_on_for_dump()) {
 5698       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5699     } else {
 5700       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5701     }
 5702     addq(r, tmp);
 5703   }
 5704   BLOCK_COMMENT("} decode_klass_not_null");
 5705 }
 5706 
 5707 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5708   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 5709   assert_different_registers(src, dst);
 5710   // Note: it will change flags
 5711   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5712   // vtableStubs also counts instructions in pd_code_size_limit.
 5713   // Also do not verify_oop as this is called by verify_oop.
 5714 
 5715   if (CompressedKlassPointers::base() == nullptr &&
 5716       CompressedKlassPointers::shift() == 0) {
 5717     // The best case scenario is that there is no base or shift. Then it is already
 5718     // a pointer that needs nothing but a register rename.
 5719     movl(dst, src);
 5720   } else {
 5721     if (CompressedKlassPointers::shift() <= Address::times_8) {
 5722       if (CompressedKlassPointers::base() != nullptr) {
 5723         movptr(dst, (intptr_t)CompressedKlassPointers::base());
 5724       } else {
 5725         xorq(dst, dst);
 5726       }
 5727       if (CompressedKlassPointers::shift() != 0) {
 5728         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 5729         leaq(dst, Address(dst, src, Address::times_8, 0));
 5730       } else {
 5731         addq(dst, src);
 5732       }
 5733     } else {
 5734       if (CompressedKlassPointers::base() != nullptr) {
 5735         const intptr_t base_right_shifted =
 5736             (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 5737         movptr(dst, base_right_shifted);
 5738       } else {
 5739         xorq(dst, dst);
 5740       }
 5741       addq(dst, src);
 5742       shlq(dst, CompressedKlassPointers::shift());
 5743     }
 5744   }
 5745   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 5746 }
 5747 
 5748 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5749   assert (UseCompressedOops, "should only be used for compressed headers");
 5750   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5751   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5752   int oop_index = oop_recorder()->find_index(obj);
 5753   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5754   mov_narrow_oop(dst, oop_index, rspec);
 5755 }
 5756 
 5757 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5758   assert (UseCompressedOops, "should only be used for compressed headers");
 5759   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5760   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5761   int oop_index = oop_recorder()->find_index(obj);
 5762   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5763   mov_narrow_oop(dst, oop_index, rspec);
 5764 }
 5765 
 5766 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5767   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5768   int klass_index = oop_recorder()->find_index(k);
 5769   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5770   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5771 }
 5772 
 5773 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 5774   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5775   int klass_index = oop_recorder()->find_index(k);
 5776   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5777   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5778 }
 5779 
 5780 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 5781   assert (UseCompressedOops, "should only be used for compressed headers");
 5782   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5783   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5784   int oop_index = oop_recorder()->find_index(obj);
 5785   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5786   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5787 }
 5788 
 5789 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 5790   assert (UseCompressedOops, "should only be used for compressed headers");
 5791   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5792   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5793   int oop_index = oop_recorder()->find_index(obj);
 5794   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5795   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5796 }
 5797 
 5798 void  MacroAssembler::cmp_narrow_klass(Register 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   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5803 }
 5804 
 5805 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 5806   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5807   int klass_index = oop_recorder()->find_index(k);
 5808   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5809   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5810 }
 5811 
 5812 void MacroAssembler::reinit_heapbase() {
 5813   if (UseCompressedOops) {
 5814     if (Universe::heap() != nullptr) {
 5815       if (CompressedOops::base() == nullptr) {
 5816         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 5817       } else {
 5818         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 5819       }
 5820     } else {
 5821       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 5822     }
 5823   }
 5824 }
 5825 
 5826 #if COMPILER2_OR_JVMCI
 5827 
 5828 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 5829 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5830   // cnt - number of qwords (8-byte words).
 5831   // base - start address, qword aligned.
 5832   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 5833   bool use64byteVector = (MaxVectorSize == 64) && (CopyAVX3Threshold == 0);
 5834   if (use64byteVector) {
 5835     vpxor(xtmp, xtmp, xtmp, AVX_512bit);
 5836   } else if (MaxVectorSize >= 32) {
 5837     vpxor(xtmp, xtmp, xtmp, AVX_256bit);
 5838   } else {
 5839     pxor(xtmp, xtmp);
 5840   }
 5841   jmp(L_zero_64_bytes);
 5842 
 5843   BIND(L_loop);
 5844   if (MaxVectorSize >= 32) {
 5845     fill64(base, 0, xtmp, use64byteVector);
 5846   } else {
 5847     movdqu(Address(base,  0), xtmp);
 5848     movdqu(Address(base, 16), xtmp);
 5849     movdqu(Address(base, 32), xtmp);
 5850     movdqu(Address(base, 48), xtmp);
 5851   }
 5852   addptr(base, 64);
 5853 
 5854   BIND(L_zero_64_bytes);
 5855   subptr(cnt, 8);
 5856   jccb(Assembler::greaterEqual, L_loop);
 5857 
 5858   // Copy trailing 64 bytes
 5859   if (use64byteVector) {
 5860     addptr(cnt, 8);
 5861     jccb(Assembler::equal, L_end);
 5862     fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
 5863     jmp(L_end);
 5864   } else {
 5865     addptr(cnt, 4);
 5866     jccb(Assembler::less, L_tail);
 5867     if (MaxVectorSize >= 32) {
 5868       vmovdqu(Address(base, 0), xtmp);
 5869     } else {
 5870       movdqu(Address(base,  0), xtmp);
 5871       movdqu(Address(base, 16), xtmp);
 5872     }
 5873   }
 5874   addptr(base, 32);
 5875   subptr(cnt, 4);
 5876 
 5877   BIND(L_tail);
 5878   addptr(cnt, 4);
 5879   jccb(Assembler::lessEqual, L_end);
 5880   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 5881     fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
 5882   } else {
 5883     decrement(cnt);
 5884 
 5885     BIND(L_sloop);
 5886     movq(Address(base, 0), xtmp);
 5887     addptr(base, 8);
 5888     decrement(cnt);
 5889     jccb(Assembler::greaterEqual, L_sloop);
 5890   }
 5891   BIND(L_end);
 5892 }
 5893 
 5894 // Clearing constant sized memory using YMM/ZMM registers.
 5895 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5896   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 5897   bool use64byteVector = (MaxVectorSize > 32) && (CopyAVX3Threshold == 0);
 5898 
 5899   int vector64_count = (cnt & (~0x7)) >> 3;
 5900   cnt = cnt & 0x7;
 5901   const int fill64_per_loop = 4;
 5902   const int max_unrolled_fill64 = 8;
 5903 
 5904   // 64 byte initialization loop.
 5905   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 5906   int start64 = 0;
 5907   if (vector64_count > max_unrolled_fill64) {
 5908     Label LOOP;
 5909     Register index = rtmp;
 5910 
 5911     start64 = vector64_count - (vector64_count % fill64_per_loop);
 5912 
 5913     movl(index, 0);
 5914     BIND(LOOP);
 5915     for (int i = 0; i < fill64_per_loop; i++) {
 5916       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 5917     }
 5918     addl(index, fill64_per_loop * 64);
 5919     cmpl(index, start64 * 64);
 5920     jccb(Assembler::less, LOOP);
 5921   }
 5922   for (int i = start64; i < vector64_count; i++) {
 5923     fill64(base, i * 64, xtmp, use64byteVector);
 5924   }
 5925 
 5926   // Clear remaining 64 byte tail.
 5927   int disp = vector64_count * 64;
 5928   if (cnt) {
 5929     switch (cnt) {
 5930       case 1:
 5931         movq(Address(base, disp), xtmp);
 5932         break;
 5933       case 2:
 5934         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 5935         break;
 5936       case 3:
 5937         movl(rtmp, 0x7);
 5938         kmovwl(mask, rtmp);
 5939         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 5940         break;
 5941       case 4:
 5942         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5943         break;
 5944       case 5:
 5945         if (use64byteVector) {
 5946           movl(rtmp, 0x1F);
 5947           kmovwl(mask, rtmp);
 5948           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5949         } else {
 5950           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5951           movq(Address(base, disp + 32), xtmp);
 5952         }
 5953         break;
 5954       case 6:
 5955         if (use64byteVector) {
 5956           movl(rtmp, 0x3F);
 5957           kmovwl(mask, rtmp);
 5958           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5959         } else {
 5960           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5961           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 5962         }
 5963         break;
 5964       case 7:
 5965         if (use64byteVector) {
 5966           movl(rtmp, 0x7F);
 5967           kmovwl(mask, rtmp);
 5968           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5969         } else {
 5970           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5971           movl(rtmp, 0x7);
 5972           kmovwl(mask, rtmp);
 5973           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 5974         }
 5975         break;
 5976       default:
 5977         fatal("Unexpected length : %d\n",cnt);
 5978         break;
 5979     }
 5980   }
 5981 }
 5982 
 5983 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
 5984                                bool is_large, KRegister mask) {
 5985   // cnt      - number of qwords (8-byte words).
 5986   // base     - start address, qword aligned.
 5987   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 5988   assert(base==rdi, "base register must be edi for rep stos");
 5989   assert(tmp==rax,   "tmp register must be eax for rep stos");
 5990   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 5991   assert(InitArrayShortSize % BytesPerLong == 0,
 5992     "InitArrayShortSize should be the multiple of BytesPerLong");
 5993 
 5994   Label DONE;
 5995   if (!is_large || !UseXMMForObjInit) {
 5996     xorptr(tmp, tmp);
 5997   }
 5998 
 5999   if (!is_large) {
 6000     Label LOOP, LONG;
 6001     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6002     jccb(Assembler::greater, LONG);
 6003 
 6004     decrement(cnt);
 6005     jccb(Assembler::negative, DONE); // Zero length
 6006 
 6007     // Use individual pointer-sized stores for small counts:
 6008     BIND(LOOP);
 6009     movptr(Address(base, cnt, Address::times_ptr), tmp);
 6010     decrement(cnt);
 6011     jccb(Assembler::greaterEqual, LOOP);
 6012     jmpb(DONE);
 6013 
 6014     BIND(LONG);
 6015   }
 6016 
 6017   // Use longer rep-prefixed ops for non-small counts:
 6018   if (UseFastStosb) {
 6019     shlptr(cnt, 3); // convert to number of bytes
 6020     rep_stosb();
 6021   } else if (UseXMMForObjInit) {
 6022     xmm_clear_mem(base, cnt, tmp, xtmp, mask);
 6023   } else {
 6024     rep_stos();
 6025   }
 6026 
 6027   BIND(DONE);
 6028 }
 6029 
 6030 #endif //COMPILER2_OR_JVMCI
 6031 
 6032 
 6033 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6034                                    Register to, Register value, Register count,
 6035                                    Register rtmp, XMMRegister xtmp) {
 6036   ShortBranchVerifier sbv(this);
 6037   assert_different_registers(to, value, count, rtmp);
 6038   Label L_exit;
 6039   Label L_fill_2_bytes, L_fill_4_bytes;
 6040 
 6041 #if defined(COMPILER2)
 6042   if(MaxVectorSize >=32 &&
 6043      VM_Version::supports_avx512vlbw() &&
 6044      VM_Version::supports_bmi2()) {
 6045     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6046     return;
 6047   }
 6048 #endif
 6049 
 6050   int shift = -1;
 6051   switch (t) {
 6052     case T_BYTE:
 6053       shift = 2;
 6054       break;
 6055     case T_SHORT:
 6056       shift = 1;
 6057       break;
 6058     case T_INT:
 6059       shift = 0;
 6060       break;
 6061     default: ShouldNotReachHere();
 6062   }
 6063 
 6064   if (t == T_BYTE) {
 6065     andl(value, 0xff);
 6066     movl(rtmp, value);
 6067     shll(rtmp, 8);
 6068     orl(value, rtmp);
 6069   }
 6070   if (t == T_SHORT) {
 6071     andl(value, 0xffff);
 6072   }
 6073   if (t == T_BYTE || t == T_SHORT) {
 6074     movl(rtmp, value);
 6075     shll(rtmp, 16);
 6076     orl(value, rtmp);
 6077   }
 6078 
 6079   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6080   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6081   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6082     Label L_skip_align2;
 6083     // align source address at 4 bytes address boundary
 6084     if (t == T_BYTE) {
 6085       Label L_skip_align1;
 6086       // One byte misalignment happens only for byte arrays
 6087       testptr(to, 1);
 6088       jccb(Assembler::zero, L_skip_align1);
 6089       movb(Address(to, 0), value);
 6090       increment(to);
 6091       decrement(count);
 6092       BIND(L_skip_align1);
 6093     }
 6094     // Two bytes misalignment happens only for byte and short (char) arrays
 6095     testptr(to, 2);
 6096     jccb(Assembler::zero, L_skip_align2);
 6097     movw(Address(to, 0), value);
 6098     addptr(to, 2);
 6099     subptr(count, 1<<(shift-1));
 6100     BIND(L_skip_align2);
 6101   }
 6102   {
 6103     Label L_fill_32_bytes;
 6104     if (!UseUnalignedLoadStores) {
 6105       // align to 8 bytes, we know we are 4 byte aligned to start
 6106       testptr(to, 4);
 6107       jccb(Assembler::zero, L_fill_32_bytes);
 6108       movl(Address(to, 0), value);
 6109       addptr(to, 4);
 6110       subptr(count, 1<<shift);
 6111     }
 6112     BIND(L_fill_32_bytes);
 6113     {
 6114       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6115       movdl(xtmp, value);
 6116       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6117         Label L_check_fill_32_bytes;
 6118         if (UseAVX > 2) {
 6119           // Fill 64-byte chunks
 6120           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6121 
 6122           // If number of bytes to fill < CopyAVX3Threshold, perform fill using AVX2
 6123           cmpptr(count, CopyAVX3Threshold);
 6124           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6125 
 6126           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6127 
 6128           subptr(count, 16 << shift);
 6129           jcc(Assembler::less, L_check_fill_32_bytes);
 6130           align(16);
 6131 
 6132           BIND(L_fill_64_bytes_loop_avx3);
 6133           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6134           addptr(to, 64);
 6135           subptr(count, 16 << shift);
 6136           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6137           jmpb(L_check_fill_32_bytes);
 6138 
 6139           BIND(L_check_fill_64_bytes_avx2);
 6140         }
 6141         // Fill 64-byte chunks
 6142         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6143 
 6144         subptr(count, 16 << shift);
 6145         jcc(Assembler::less, L_check_fill_32_bytes);
 6146 
 6147         // align data for 64-byte chunks
 6148         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6149         if (EnableX86ECoreOpts) {
 6150             // align 'big' arrays to cache lines to minimize split_stores
 6151             cmpptr(count, 96 << shift);
 6152             jcc(Assembler::below, L_fill_64_bytes_loop);
 6153 
 6154             // Find the bytes needed for alignment
 6155             movptr(rtmp, to);
 6156             andptr(rtmp, 0x1c);
 6157             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6158             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6159             addptr(rtmp, 32);
 6160             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6161             subptr(count, rtmp);    // adjust count by number of elements
 6162 
 6163             align(16);
 6164             BIND(L_align_64_bytes_loop);
 6165             movdl(Address(to, 0), xtmp);
 6166             addptr(to, 4);
 6167             subptr(rtmp, 1 << shift);
 6168             jcc(Assembler::greater, L_align_64_bytes_loop);
 6169         }
 6170 
 6171         align(16);
 6172         BIND(L_fill_64_bytes_loop);
 6173         vmovdqu(Address(to, 0), xtmp);
 6174         vmovdqu(Address(to, 32), xtmp);
 6175         addptr(to, 64);
 6176         subptr(count, 16 << shift);
 6177         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6178 
 6179         align(16);
 6180         BIND(L_check_fill_32_bytes);
 6181         addptr(count, 8 << shift);
 6182         jccb(Assembler::less, L_check_fill_8_bytes);
 6183         vmovdqu(Address(to, 0), xtmp);
 6184         addptr(to, 32);
 6185         subptr(count, 8 << shift);
 6186 
 6187         BIND(L_check_fill_8_bytes);
 6188         // clean upper bits of YMM registers
 6189         movdl(xtmp, value);
 6190         pshufd(xtmp, xtmp, 0);
 6191       } else {
 6192         // Fill 32-byte chunks
 6193         pshufd(xtmp, xtmp, 0);
 6194 
 6195         subptr(count, 8 << shift);
 6196         jcc(Assembler::less, L_check_fill_8_bytes);
 6197         align(16);
 6198 
 6199         BIND(L_fill_32_bytes_loop);
 6200 
 6201         if (UseUnalignedLoadStores) {
 6202           movdqu(Address(to, 0), xtmp);
 6203           movdqu(Address(to, 16), xtmp);
 6204         } else {
 6205           movq(Address(to, 0), xtmp);
 6206           movq(Address(to, 8), xtmp);
 6207           movq(Address(to, 16), xtmp);
 6208           movq(Address(to, 24), xtmp);
 6209         }
 6210 
 6211         addptr(to, 32);
 6212         subptr(count, 8 << shift);
 6213         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6214 
 6215         BIND(L_check_fill_8_bytes);
 6216       }
 6217       addptr(count, 8 << shift);
 6218       jccb(Assembler::zero, L_exit);
 6219       jmpb(L_fill_8_bytes);
 6220 
 6221       //
 6222       // length is too short, just fill qwords
 6223       //
 6224       align(16);
 6225       BIND(L_fill_8_bytes_loop);
 6226       movq(Address(to, 0), xtmp);
 6227       addptr(to, 8);
 6228       BIND(L_fill_8_bytes);
 6229       subptr(count, 1 << (shift + 1));
 6230       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6231     }
 6232   }
 6233 
 6234   Label L_fill_4_bytes_loop;
 6235   testl(count, 1 << shift);
 6236   jccb(Assembler::zero, L_fill_2_bytes);
 6237 
 6238   align(16);
 6239   BIND(L_fill_4_bytes_loop);
 6240   movl(Address(to, 0), value);
 6241   addptr(to, 4);
 6242 
 6243   BIND(L_fill_4_bytes);
 6244   subptr(count, 1 << shift);
 6245   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 6246 
 6247   if (t == T_BYTE || t == T_SHORT) {
 6248     Label L_fill_byte;
 6249     BIND(L_fill_2_bytes);
 6250     // fill trailing 2 bytes
 6251     testl(count, 1<<(shift-1));
 6252     jccb(Assembler::zero, L_fill_byte);
 6253     movw(Address(to, 0), value);
 6254     if (t == T_BYTE) {
 6255       addptr(to, 2);
 6256       BIND(L_fill_byte);
 6257       // fill trailing byte
 6258       testl(count, 1);
 6259       jccb(Assembler::zero, L_exit);
 6260       movb(Address(to, 0), value);
 6261     } else {
 6262       BIND(L_fill_byte);
 6263     }
 6264   } else {
 6265     BIND(L_fill_2_bytes);
 6266   }
 6267   BIND(L_exit);
 6268 }
 6269 
 6270 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6271   switch(type) {
 6272     case T_BYTE:
 6273     case T_BOOLEAN:
 6274       evpbroadcastb(dst, src, vector_len);
 6275       break;
 6276     case T_SHORT:
 6277     case T_CHAR:
 6278       evpbroadcastw(dst, src, vector_len);
 6279       break;
 6280     case T_INT:
 6281     case T_FLOAT:
 6282       evpbroadcastd(dst, src, vector_len);
 6283       break;
 6284     case T_LONG:
 6285     case T_DOUBLE:
 6286       evpbroadcastq(dst, src, vector_len);
 6287       break;
 6288     default:
 6289       fatal("Unhandled type : %s", type2name(type));
 6290       break;
 6291   }
 6292 }
 6293 
 6294 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 6295 //
 6296 // @IntrinsicCandidate
 6297 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 6298 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6299 //     int i = 0;
 6300 //     for (; i < len; i++) {
 6301 //         char c = sa[sp++];
 6302 //         if (c > '\u00FF')
 6303 //             break;
 6304 //         da[dp++] = (byte) c;
 6305 //     }
 6306 //     return i;
 6307 // }
 6308 //
 6309 // @IntrinsicCandidate
 6310 // int java.lang.StringCoding.encodeISOArray0(
 6311 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 6312 //   int i = 0;
 6313 //   for (; i < len; i++) {
 6314 //     char c = StringUTF16.getChar(sa, sp++);
 6315 //     if (c > '\u00FF')
 6316 //       break;
 6317 //     da[dp++] = (byte) c;
 6318 //   }
 6319 //   return i;
 6320 // }
 6321 //
 6322 // @IntrinsicCandidate
 6323 // int java.lang.StringCoding.encodeAsciiArray0(
 6324 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6325 //   int i = 0;
 6326 //   for (; i < len; i++) {
 6327 //     char c = sa[sp++];
 6328 //     if (c >= '\u0080')
 6329 //       break;
 6330 //     da[dp++] = (byte) c;
 6331 //   }
 6332 //   return i;
 6333 // }
 6334 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 6335   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 6336   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 6337   Register tmp5, Register result, bool ascii) {
 6338 
 6339   // rsi: src
 6340   // rdi: dst
 6341   // rdx: len
 6342   // rcx: tmp5
 6343   // rax: result
 6344   ShortBranchVerifier sbv(this);
 6345   assert_different_registers(src, dst, len, tmp5, result);
 6346   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 6347 
 6348   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 6349   int short_mask = ascii ? 0xff80 : 0xff00;
 6350 
 6351   // set result
 6352   xorl(result, result);
 6353   // check for zero length
 6354   testl(len, len);
 6355   jcc(Assembler::zero, L_done);
 6356 
 6357   movl(result, len);
 6358 
 6359   // Setup pointers
 6360   lea(src, Address(src, len, Address::times_2)); // char[]
 6361   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 6362   negptr(len);
 6363 
 6364   if (UseSSE42Intrinsics || UseAVX >= 2) {
 6365     Label L_copy_8_chars, L_copy_8_chars_exit;
 6366     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 6367 
 6368     if (UseAVX >= 2) {
 6369       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 6370       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6371       movdl(tmp1Reg, tmp5);
 6372       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 6373       jmp(L_chars_32_check);
 6374 
 6375       bind(L_copy_32_chars);
 6376       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 6377       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 6378       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6379       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6380       jccb(Assembler::notZero, L_copy_32_chars_exit);
 6381       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6382       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 6383       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 6384 
 6385       bind(L_chars_32_check);
 6386       addptr(len, 32);
 6387       jcc(Assembler::lessEqual, L_copy_32_chars);
 6388 
 6389       bind(L_copy_32_chars_exit);
 6390       subptr(len, 16);
 6391       jccb(Assembler::greater, L_copy_16_chars_exit);
 6392 
 6393     } else if (UseSSE42Intrinsics) {
 6394       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6395       movdl(tmp1Reg, tmp5);
 6396       pshufd(tmp1Reg, tmp1Reg, 0);
 6397       jmpb(L_chars_16_check);
 6398     }
 6399 
 6400     bind(L_copy_16_chars);
 6401     if (UseAVX >= 2) {
 6402       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 6403       vptest(tmp2Reg, tmp1Reg);
 6404       jcc(Assembler::notZero, L_copy_16_chars_exit);
 6405       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 6406       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 6407     } else {
 6408       if (UseAVX > 0) {
 6409         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6410         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6411         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 6412       } else {
 6413         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6414         por(tmp2Reg, tmp3Reg);
 6415         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6416         por(tmp2Reg, tmp4Reg);
 6417       }
 6418       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6419       jccb(Assembler::notZero, L_copy_16_chars_exit);
 6420       packuswb(tmp3Reg, tmp4Reg);
 6421     }
 6422     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 6423 
 6424     bind(L_chars_16_check);
 6425     addptr(len, 16);
 6426     jcc(Assembler::lessEqual, L_copy_16_chars);
 6427 
 6428     bind(L_copy_16_chars_exit);
 6429     if (UseAVX >= 2) {
 6430       // clean upper bits of YMM registers
 6431       vpxor(tmp2Reg, tmp2Reg);
 6432       vpxor(tmp3Reg, tmp3Reg);
 6433       vpxor(tmp4Reg, tmp4Reg);
 6434       movdl(tmp1Reg, tmp5);
 6435       pshufd(tmp1Reg, tmp1Reg, 0);
 6436     }
 6437     subptr(len, 8);
 6438     jccb(Assembler::greater, L_copy_8_chars_exit);
 6439 
 6440     bind(L_copy_8_chars);
 6441     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 6442     ptest(tmp3Reg, tmp1Reg);
 6443     jccb(Assembler::notZero, L_copy_8_chars_exit);
 6444     packuswb(tmp3Reg, tmp1Reg);
 6445     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 6446     addptr(len, 8);
 6447     jccb(Assembler::lessEqual, L_copy_8_chars);
 6448 
 6449     bind(L_copy_8_chars_exit);
 6450     subptr(len, 8);
 6451     jccb(Assembler::zero, L_done);
 6452   }
 6453 
 6454   bind(L_copy_1_char);
 6455   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 6456   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 6457   jccb(Assembler::notZero, L_copy_1_char_exit);
 6458   movb(Address(dst, len, Address::times_1, 0), tmp5);
 6459   addptr(len, 1);
 6460   jccb(Assembler::less, L_copy_1_char);
 6461 
 6462   bind(L_copy_1_char_exit);
 6463   addptr(result, len); // len is negative count of not processed elements
 6464 
 6465   bind(L_done);
 6466 }
 6467 
 6468 /**
 6469  * Helper for multiply_to_len().
 6470  */
 6471 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 6472   addq(dest_lo, src1);
 6473   adcq(dest_hi, 0);
 6474   addq(dest_lo, src2);
 6475   adcq(dest_hi, 0);
 6476 }
 6477 
 6478 /**
 6479  * Multiply 64 bit by 64 bit first loop.
 6480  */
 6481 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 6482                                            Register y, Register y_idx, Register z,
 6483                                            Register carry, Register product,
 6484                                            Register idx, Register kdx) {
 6485   //
 6486   //  jlong carry, x[], y[], z[];
 6487   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6488   //    huge_128 product = y[idx] * x[xstart] + carry;
 6489   //    z[kdx] = (jlong)product;
 6490   //    carry  = (jlong)(product >>> 64);
 6491   //  }
 6492   //  z[xstart] = carry;
 6493   //
 6494 
 6495   Label L_first_loop, L_first_loop_exit;
 6496   Label L_one_x, L_one_y, L_multiply;
 6497 
 6498   decrementl(xstart);
 6499   jcc(Assembler::negative, L_one_x);
 6500 
 6501   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6502   rorq(x_xstart, 32); // convert big-endian to little-endian
 6503 
 6504   bind(L_first_loop);
 6505   decrementl(idx);
 6506   jcc(Assembler::negative, L_first_loop_exit);
 6507   decrementl(idx);
 6508   jcc(Assembler::negative, L_one_y);
 6509   movq(y_idx, Address(y, idx, Address::times_4,  0));
 6510   rorq(y_idx, 32); // convert big-endian to little-endian
 6511   bind(L_multiply);
 6512   movq(product, x_xstart);
 6513   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 6514   addq(product, carry);
 6515   adcq(rdx, 0);
 6516   subl(kdx, 2);
 6517   movl(Address(z, kdx, Address::times_4,  4), product);
 6518   shrq(product, 32);
 6519   movl(Address(z, kdx, Address::times_4,  0), product);
 6520   movq(carry, rdx);
 6521   jmp(L_first_loop);
 6522 
 6523   bind(L_one_y);
 6524   movl(y_idx, Address(y,  0));
 6525   jmp(L_multiply);
 6526 
 6527   bind(L_one_x);
 6528   movl(x_xstart, Address(x,  0));
 6529   jmp(L_first_loop);
 6530 
 6531   bind(L_first_loop_exit);
 6532 }
 6533 
 6534 /**
 6535  * Multiply 64 bit by 64 bit and add 128 bit.
 6536  */
 6537 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 6538                                             Register yz_idx, Register idx,
 6539                                             Register carry, Register product, int offset) {
 6540   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 6541   //     z[kdx] = (jlong)product;
 6542 
 6543   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 6544   rorq(yz_idx, 32); // convert big-endian to little-endian
 6545   movq(product, x_xstart);
 6546   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 6547   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 6548   rorq(yz_idx, 32); // convert big-endian to little-endian
 6549 
 6550   add2_with_carry(rdx, product, carry, yz_idx);
 6551 
 6552   movl(Address(z, idx, Address::times_4,  offset+4), product);
 6553   shrq(product, 32);
 6554   movl(Address(z, idx, Address::times_4,  offset), product);
 6555 
 6556 }
 6557 
 6558 /**
 6559  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 6560  */
 6561 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 6562                                              Register yz_idx, Register idx, Register jdx,
 6563                                              Register carry, Register product,
 6564                                              Register carry2) {
 6565   //   jlong carry, x[], y[], z[];
 6566   //   int kdx = ystart+1;
 6567   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6568   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 6569   //     z[kdx+idx+1] = (jlong)product;
 6570   //     jlong carry2  = (jlong)(product >>> 64);
 6571   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 6572   //     z[kdx+idx] = (jlong)product;
 6573   //     carry  = (jlong)(product >>> 64);
 6574   //   }
 6575   //   idx += 2;
 6576   //   if (idx > 0) {
 6577   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 6578   //     z[kdx+idx] = (jlong)product;
 6579   //     carry  = (jlong)(product >>> 64);
 6580   //   }
 6581   //
 6582 
 6583   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6584 
 6585   movl(jdx, idx);
 6586   andl(jdx, 0xFFFFFFFC);
 6587   shrl(jdx, 2);
 6588 
 6589   bind(L_third_loop);
 6590   subl(jdx, 1);
 6591   jcc(Assembler::negative, L_third_loop_exit);
 6592   subl(idx, 4);
 6593 
 6594   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 6595   movq(carry2, rdx);
 6596 
 6597   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 6598   movq(carry, rdx);
 6599   jmp(L_third_loop);
 6600 
 6601   bind (L_third_loop_exit);
 6602 
 6603   andl (idx, 0x3);
 6604   jcc(Assembler::zero, L_post_third_loop_done);
 6605 
 6606   Label L_check_1;
 6607   subl(idx, 2);
 6608   jcc(Assembler::negative, L_check_1);
 6609 
 6610   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 6611   movq(carry, rdx);
 6612 
 6613   bind (L_check_1);
 6614   addl (idx, 0x2);
 6615   andl (idx, 0x1);
 6616   subl(idx, 1);
 6617   jcc(Assembler::negative, L_post_third_loop_done);
 6618 
 6619   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 6620   movq(product, x_xstart);
 6621   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 6622   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 6623 
 6624   add2_with_carry(rdx, product, yz_idx, carry);
 6625 
 6626   movl(Address(z, idx, Address::times_4,  0), product);
 6627   shrq(product, 32);
 6628 
 6629   shlq(rdx, 32);
 6630   orq(product, rdx);
 6631   movq(carry, product);
 6632 
 6633   bind(L_post_third_loop_done);
 6634 }
 6635 
 6636 /**
 6637  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 6638  *
 6639  */
 6640 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 6641                                                   Register carry, Register carry2,
 6642                                                   Register idx, Register jdx,
 6643                                                   Register yz_idx1, Register yz_idx2,
 6644                                                   Register tmp, Register tmp3, Register tmp4) {
 6645   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 6646 
 6647   //   jlong carry, x[], y[], z[];
 6648   //   int kdx = ystart+1;
 6649   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6650   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 6651   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 6652   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 6653   //     carry  = (jlong)(tmp4 >>> 64);
 6654   //     z[kdx+idx+1] = (jlong)tmp3;
 6655   //     z[kdx+idx] = (jlong)tmp4;
 6656   //   }
 6657   //   idx += 2;
 6658   //   if (idx > 0) {
 6659   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 6660   //     z[kdx+idx] = (jlong)yz_idx1;
 6661   //     carry  = (jlong)(yz_idx1 >>> 64);
 6662   //   }
 6663   //
 6664 
 6665   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6666 
 6667   movl(jdx, idx);
 6668   andl(jdx, 0xFFFFFFFC);
 6669   shrl(jdx, 2);
 6670 
 6671   bind(L_third_loop);
 6672   subl(jdx, 1);
 6673   jcc(Assembler::negative, L_third_loop_exit);
 6674   subl(idx, 4);
 6675 
 6676   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 6677   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 6678   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 6679   rorxq(yz_idx2, yz_idx2, 32);
 6680 
 6681   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 6682   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 6683 
 6684   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 6685   rorxq(yz_idx1, yz_idx1, 32);
 6686   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6687   rorxq(yz_idx2, yz_idx2, 32);
 6688 
 6689   if (VM_Version::supports_adx()) {
 6690     adcxq(tmp3, carry);
 6691     adoxq(tmp3, yz_idx1);
 6692 
 6693     adcxq(tmp4, tmp);
 6694     adoxq(tmp4, yz_idx2);
 6695 
 6696     movl(carry, 0); // does not affect flags
 6697     adcxq(carry2, carry);
 6698     adoxq(carry2, carry);
 6699   } else {
 6700     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 6701     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 6702   }
 6703   movq(carry, carry2);
 6704 
 6705   movl(Address(z, idx, Address::times_4, 12), tmp3);
 6706   shrq(tmp3, 32);
 6707   movl(Address(z, idx, Address::times_4,  8), tmp3);
 6708 
 6709   movl(Address(z, idx, Address::times_4,  4), tmp4);
 6710   shrq(tmp4, 32);
 6711   movl(Address(z, idx, Address::times_4,  0), tmp4);
 6712 
 6713   jmp(L_third_loop);
 6714 
 6715   bind (L_third_loop_exit);
 6716 
 6717   andl (idx, 0x3);
 6718   jcc(Assembler::zero, L_post_third_loop_done);
 6719 
 6720   Label L_check_1;
 6721   subl(idx, 2);
 6722   jcc(Assembler::negative, L_check_1);
 6723 
 6724   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 6725   rorxq(yz_idx1, yz_idx1, 32);
 6726   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 6727   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6728   rorxq(yz_idx2, yz_idx2, 32);
 6729 
 6730   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 6731 
 6732   movl(Address(z, idx, Address::times_4,  4), tmp3);
 6733   shrq(tmp3, 32);
 6734   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6735   movq(carry, tmp4);
 6736 
 6737   bind (L_check_1);
 6738   addl (idx, 0x2);
 6739   andl (idx, 0x1);
 6740   subl(idx, 1);
 6741   jcc(Assembler::negative, L_post_third_loop_done);
 6742   movl(tmp4, Address(y, idx, Address::times_4,  0));
 6743   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 6744   movl(tmp4, Address(z, idx, Address::times_4,  0));
 6745 
 6746   add2_with_carry(carry2, tmp3, tmp4, carry);
 6747 
 6748   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6749   shrq(tmp3, 32);
 6750 
 6751   shlq(carry2, 32);
 6752   orq(tmp3, carry2);
 6753   movq(carry, tmp3);
 6754 
 6755   bind(L_post_third_loop_done);
 6756 }
 6757 
 6758 /**
 6759  * Code for BigInteger::multiplyToLen() intrinsic.
 6760  *
 6761  * rdi: x
 6762  * rax: xlen
 6763  * rsi: y
 6764  * rcx: ylen
 6765  * r8:  z
 6766  * r11: tmp0
 6767  * r12: tmp1
 6768  * r13: tmp2
 6769  * r14: tmp3
 6770  * r15: tmp4
 6771  * rbx: tmp5
 6772  *
 6773  */
 6774 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 6775                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 6776   ShortBranchVerifier sbv(this);
 6777   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 6778 
 6779   push(tmp0);
 6780   push(tmp1);
 6781   push(tmp2);
 6782   push(tmp3);
 6783   push(tmp4);
 6784   push(tmp5);
 6785 
 6786   push(xlen);
 6787 
 6788   const Register idx = tmp1;
 6789   const Register kdx = tmp2;
 6790   const Register xstart = tmp3;
 6791 
 6792   const Register y_idx = tmp4;
 6793   const Register carry = tmp5;
 6794   const Register product  = xlen;
 6795   const Register x_xstart = tmp0;
 6796 
 6797   // First Loop.
 6798   //
 6799   //  final static long LONG_MASK = 0xffffffffL;
 6800   //  int xstart = xlen - 1;
 6801   //  int ystart = ylen - 1;
 6802   //  long carry = 0;
 6803   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6804   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 6805   //    z[kdx] = (int)product;
 6806   //    carry = product >>> 32;
 6807   //  }
 6808   //  z[xstart] = (int)carry;
 6809   //
 6810 
 6811   movl(idx, ylen);               // idx = ylen;
 6812   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 6813   xorq(carry, carry);            // carry = 0;
 6814 
 6815   Label L_done;
 6816 
 6817   movl(xstart, xlen);
 6818   decrementl(xstart);
 6819   jcc(Assembler::negative, L_done);
 6820 
 6821   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 6822 
 6823   Label L_second_loop;
 6824   testl(kdx, kdx);
 6825   jcc(Assembler::zero, L_second_loop);
 6826 
 6827   Label L_carry;
 6828   subl(kdx, 1);
 6829   jcc(Assembler::zero, L_carry);
 6830 
 6831   movl(Address(z, kdx, Address::times_4,  0), carry);
 6832   shrq(carry, 32);
 6833   subl(kdx, 1);
 6834 
 6835   bind(L_carry);
 6836   movl(Address(z, kdx, Address::times_4,  0), carry);
 6837 
 6838   // Second and third (nested) loops.
 6839   //
 6840   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 6841   //   carry = 0;
 6842   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 6843   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 6844   //                    (z[k] & LONG_MASK) + carry;
 6845   //     z[k] = (int)product;
 6846   //     carry = product >>> 32;
 6847   //   }
 6848   //   z[i] = (int)carry;
 6849   // }
 6850   //
 6851   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 6852 
 6853   const Register jdx = tmp1;
 6854 
 6855   bind(L_second_loop);
 6856   xorl(carry, carry);    // carry = 0;
 6857   movl(jdx, ylen);       // j = ystart+1
 6858 
 6859   subl(xstart, 1);       // i = xstart-1;
 6860   jcc(Assembler::negative, L_done);
 6861 
 6862   push (z);
 6863 
 6864   Label L_last_x;
 6865   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 6866   subl(xstart, 1);       // i = xstart-1;
 6867   jcc(Assembler::negative, L_last_x);
 6868 
 6869   if (UseBMI2Instructions) {
 6870     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 6871     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 6872   } else {
 6873     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6874     rorq(x_xstart, 32);  // convert big-endian to little-endian
 6875   }
 6876 
 6877   Label L_third_loop_prologue;
 6878   bind(L_third_loop_prologue);
 6879 
 6880   push (x);
 6881   push (xstart);
 6882   push (ylen);
 6883 
 6884 
 6885   if (UseBMI2Instructions) {
 6886     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 6887   } else { // !UseBMI2Instructions
 6888     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 6889   }
 6890 
 6891   pop(ylen);
 6892   pop(xlen);
 6893   pop(x);
 6894   pop(z);
 6895 
 6896   movl(tmp3, xlen);
 6897   addl(tmp3, 1);
 6898   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6899   subl(tmp3, 1);
 6900   jccb(Assembler::negative, L_done);
 6901 
 6902   shrq(carry, 32);
 6903   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6904   jmp(L_second_loop);
 6905 
 6906   // Next infrequent code is moved outside loops.
 6907   bind(L_last_x);
 6908   if (UseBMI2Instructions) {
 6909     movl(rdx, Address(x,  0));
 6910   } else {
 6911     movl(x_xstart, Address(x,  0));
 6912   }
 6913   jmp(L_third_loop_prologue);
 6914 
 6915   bind(L_done);
 6916 
 6917   pop(xlen);
 6918 
 6919   pop(tmp5);
 6920   pop(tmp4);
 6921   pop(tmp3);
 6922   pop(tmp2);
 6923   pop(tmp1);
 6924   pop(tmp0);
 6925 }
 6926 
 6927 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 6928   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 6929   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 6930   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 6931   Label VECTOR8_TAIL, VECTOR4_TAIL;
 6932   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 6933   Label SAME_TILL_END, DONE;
 6934   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 6935 
 6936   //scale is in rcx in both Win64 and Unix
 6937   ShortBranchVerifier sbv(this);
 6938 
 6939   shlq(length);
 6940   xorq(result, result);
 6941 
 6942   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 6943       VM_Version::supports_avx512vlbw()) {
 6944     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 6945 
 6946     cmpq(length, 64);
 6947     jcc(Assembler::less, VECTOR32_TAIL);
 6948 
 6949     movq(tmp1, length);
 6950     andq(tmp1, 0x3F);      // tail count
 6951     andq(length, ~(0x3F)); //vector count
 6952 
 6953     bind(VECTOR64_LOOP);
 6954     // AVX512 code to compare 64 byte vectors.
 6955     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 6956     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6957     kortestql(k7, k7);
 6958     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 6959     addq(result, 64);
 6960     subq(length, 64);
 6961     jccb(Assembler::notZero, VECTOR64_LOOP);
 6962 
 6963     //bind(VECTOR64_TAIL);
 6964     testq(tmp1, tmp1);
 6965     jcc(Assembler::zero, SAME_TILL_END);
 6966 
 6967     //bind(VECTOR64_TAIL);
 6968     // AVX512 code to compare up to 63 byte vectors.
 6969     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 6970     shlxq(tmp2, tmp2, tmp1);
 6971     notq(tmp2);
 6972     kmovql(k3, tmp2);
 6973 
 6974     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 6975     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6976 
 6977     ktestql(k7, k3);
 6978     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 6979 
 6980     bind(VECTOR64_NOT_EQUAL);
 6981     kmovql(tmp1, k7);
 6982     notq(tmp1);
 6983     tzcntq(tmp1, tmp1);
 6984     addq(result, tmp1);
 6985     shrq(result);
 6986     jmp(DONE);
 6987     bind(VECTOR32_TAIL);
 6988   }
 6989 
 6990   cmpq(length, 8);
 6991   jcc(Assembler::equal, VECTOR8_LOOP);
 6992   jcc(Assembler::less, VECTOR4_TAIL);
 6993 
 6994   if (UseAVX >= 2) {
 6995     Label VECTOR16_TAIL, VECTOR32_LOOP;
 6996 
 6997     cmpq(length, 16);
 6998     jcc(Assembler::equal, VECTOR16_LOOP);
 6999     jcc(Assembler::less, VECTOR8_LOOP);
 7000 
 7001     cmpq(length, 32);
 7002     jccb(Assembler::less, VECTOR16_TAIL);
 7003 
 7004     subq(length, 32);
 7005     bind(VECTOR32_LOOP);
 7006     vmovdqu(rymm0, Address(obja, result));
 7007     vmovdqu(rymm1, Address(objb, result));
 7008     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 7009     vptest(rymm2, rymm2);
 7010     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7011     addq(result, 32);
 7012     subq(length, 32);
 7013     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7014     addq(length, 32);
 7015     jcc(Assembler::equal, SAME_TILL_END);
 7016     //falling through if less than 32 bytes left //close the branch here.
 7017 
 7018     bind(VECTOR16_TAIL);
 7019     cmpq(length, 16);
 7020     jccb(Assembler::less, VECTOR8_TAIL);
 7021     bind(VECTOR16_LOOP);
 7022     movdqu(rymm0, Address(obja, result));
 7023     movdqu(rymm1, Address(objb, result));
 7024     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7025     ptest(rymm2, rymm2);
 7026     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7027     addq(result, 16);
 7028     subq(length, 16);
 7029     jcc(Assembler::equal, SAME_TILL_END);
 7030     //falling through if less than 16 bytes left
 7031   } else {//regular intrinsics
 7032 
 7033     cmpq(length, 16);
 7034     jccb(Assembler::less, VECTOR8_TAIL);
 7035 
 7036     subq(length, 16);
 7037     bind(VECTOR16_LOOP);
 7038     movdqu(rymm0, Address(obja, result));
 7039     movdqu(rymm1, Address(objb, result));
 7040     pxor(rymm0, rymm1);
 7041     ptest(rymm0, rymm0);
 7042     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7043     addq(result, 16);
 7044     subq(length, 16);
 7045     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7046     addq(length, 16);
 7047     jcc(Assembler::equal, SAME_TILL_END);
 7048     //falling through if less than 16 bytes left
 7049   }
 7050 
 7051   bind(VECTOR8_TAIL);
 7052   cmpq(length, 8);
 7053   jccb(Assembler::less, VECTOR4_TAIL);
 7054   bind(VECTOR8_LOOP);
 7055   movq(tmp1, Address(obja, result));
 7056   movq(tmp2, Address(objb, result));
 7057   xorq(tmp1, tmp2);
 7058   testq(tmp1, tmp1);
 7059   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7060   addq(result, 8);
 7061   subq(length, 8);
 7062   jcc(Assembler::equal, SAME_TILL_END);
 7063   //falling through if less than 8 bytes left
 7064 
 7065   bind(VECTOR4_TAIL);
 7066   cmpq(length, 4);
 7067   jccb(Assembler::less, BYTES_TAIL);
 7068   bind(VECTOR4_LOOP);
 7069   movl(tmp1, Address(obja, result));
 7070   xorl(tmp1, Address(objb, result));
 7071   testl(tmp1, tmp1);
 7072   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7073   addq(result, 4);
 7074   subq(length, 4);
 7075   jcc(Assembler::equal, SAME_TILL_END);
 7076   //falling through if less than 4 bytes left
 7077 
 7078   bind(BYTES_TAIL);
 7079   bind(BYTES_LOOP);
 7080   load_unsigned_byte(tmp1, Address(obja, result));
 7081   load_unsigned_byte(tmp2, Address(objb, result));
 7082   xorl(tmp1, tmp2);
 7083   testl(tmp1, tmp1);
 7084   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7085   decq(length);
 7086   jcc(Assembler::zero, SAME_TILL_END);
 7087   incq(result);
 7088   load_unsigned_byte(tmp1, Address(obja, result));
 7089   load_unsigned_byte(tmp2, Address(objb, result));
 7090   xorl(tmp1, tmp2);
 7091   testl(tmp1, tmp1);
 7092   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7093   decq(length);
 7094   jcc(Assembler::zero, SAME_TILL_END);
 7095   incq(result);
 7096   load_unsigned_byte(tmp1, Address(obja, result));
 7097   load_unsigned_byte(tmp2, Address(objb, result));
 7098   xorl(tmp1, tmp2);
 7099   testl(tmp1, tmp1);
 7100   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7101   jmp(SAME_TILL_END);
 7102 
 7103   if (UseAVX >= 2) {
 7104     bind(VECTOR32_NOT_EQUAL);
 7105     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7106     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7107     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7108     vpmovmskb(tmp1, rymm0);
 7109     bsfq(tmp1, tmp1);
 7110     addq(result, tmp1);
 7111     shrq(result);
 7112     jmp(DONE);
 7113   }
 7114 
 7115   bind(VECTOR16_NOT_EQUAL);
 7116   if (UseAVX >= 2) {
 7117     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7118     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7119     pxor(rymm0, rymm2);
 7120   } else {
 7121     pcmpeqb(rymm2, rymm2);
 7122     pxor(rymm0, rymm1);
 7123     pcmpeqb(rymm0, rymm1);
 7124     pxor(rymm0, rymm2);
 7125   }
 7126   pmovmskb(tmp1, rymm0);
 7127   bsfq(tmp1, tmp1);
 7128   addq(result, tmp1);
 7129   shrq(result);
 7130   jmpb(DONE);
 7131 
 7132   bind(VECTOR8_NOT_EQUAL);
 7133   bind(VECTOR4_NOT_EQUAL);
 7134   bsfq(tmp1, tmp1);
 7135   shrq(tmp1, 3);
 7136   addq(result, tmp1);
 7137   bind(BYTES_NOT_EQUAL);
 7138   shrq(result);
 7139   jmpb(DONE);
 7140 
 7141   bind(SAME_TILL_END);
 7142   mov64(result, -1);
 7143 
 7144   bind(DONE);
 7145 }
 7146 
 7147 //Helper functions for square_to_len()
 7148 
 7149 /**
 7150  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7151  * Preserves x and z and modifies rest of the registers.
 7152  */
 7153 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7154   // Perform square and right shift by 1
 7155   // Handle odd xlen case first, then for even xlen do the following
 7156   // jlong carry = 0;
 7157   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7158   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7159   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7160   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7161   //     carry = (jlong)product;
 7162   // }
 7163 
 7164   xorq(tmp5, tmp5);     // carry
 7165   xorq(rdxReg, rdxReg);
 7166   xorl(tmp1, tmp1);     // index for x
 7167   xorl(tmp4, tmp4);     // index for z
 7168 
 7169   Label L_first_loop, L_first_loop_exit;
 7170 
 7171   testl(xlen, 1);
 7172   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7173 
 7174   // Square and right shift by 1 the odd element using 32 bit multiply
 7175   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7176   imulq(raxReg, raxReg);
 7177   shrq(raxReg, 1);
 7178   adcq(tmp5, 0);
 7179   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7180   incrementl(tmp1);
 7181   addl(tmp4, 2);
 7182 
 7183   // Square and  right shift by 1 the rest using 64 bit multiply
 7184   bind(L_first_loop);
 7185   cmpptr(tmp1, xlen);
 7186   jccb(Assembler::equal, L_first_loop_exit);
 7187 
 7188   // Square
 7189   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7190   rorq(raxReg, 32);    // convert big-endian to little-endian
 7191   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7192 
 7193   // Right shift by 1 and save carry
 7194   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7195   rcrq(rdxReg, 1);
 7196   rcrq(raxReg, 1);
 7197   adcq(tmp5, 0);
 7198 
 7199   // Store result in z
 7200   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7201   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7202 
 7203   // Update indices for x and z
 7204   addl(tmp1, 2);
 7205   addl(tmp4, 4);
 7206   jmp(L_first_loop);
 7207 
 7208   bind(L_first_loop_exit);
 7209 }
 7210 
 7211 
 7212 /**
 7213  * Perform the following multiply add operation using BMI2 instructions
 7214  * carry:sum = sum + op1*op2 + carry
 7215  * op2 should be in rdx
 7216  * op2 is preserved, all other registers are modified
 7217  */
 7218 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7219   // assert op2 is rdx
 7220   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7221   addq(sum, carry);
 7222   adcq(tmp2, 0);
 7223   addq(sum, op1);
 7224   adcq(tmp2, 0);
 7225   movq(carry, tmp2);
 7226 }
 7227 
 7228 /**
 7229  * Perform the following multiply add operation:
 7230  * carry:sum = sum + op1*op2 + carry
 7231  * Preserves op1, op2 and modifies rest of registers
 7232  */
 7233 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7234   // rdx:rax = op1 * op2
 7235   movq(raxReg, op2);
 7236   mulq(op1);
 7237 
 7238   //  rdx:rax = sum + carry + rdx:rax
 7239   addq(sum, carry);
 7240   adcq(rdxReg, 0);
 7241   addq(sum, raxReg);
 7242   adcq(rdxReg, 0);
 7243 
 7244   // carry:sum = rdx:sum
 7245   movq(carry, rdxReg);
 7246 }
 7247 
 7248 /**
 7249  * Add 64 bit long carry into z[] with carry propagation.
 7250  * Preserves z and carry register values and modifies rest of registers.
 7251  *
 7252  */
 7253 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7254   Label L_fourth_loop, L_fourth_loop_exit;
 7255 
 7256   movl(tmp1, 1);
 7257   subl(zlen, 2);
 7258   addq(Address(z, zlen, Address::times_4, 0), carry);
 7259 
 7260   bind(L_fourth_loop);
 7261   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7262   subl(zlen, 2);
 7263   jccb(Assembler::negative, L_fourth_loop_exit);
 7264   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7265   jmp(L_fourth_loop);
 7266   bind(L_fourth_loop_exit);
 7267 }
 7268 
 7269 /**
 7270  * Shift z[] left by 1 bit.
 7271  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7272  *
 7273  */
 7274 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7275 
 7276   Label L_fifth_loop, L_fifth_loop_exit;
 7277 
 7278   // Fifth loop
 7279   // Perform primitiveLeftShift(z, zlen, 1)
 7280 
 7281   const Register prev_carry = tmp1;
 7282   const Register new_carry = tmp4;
 7283   const Register value = tmp2;
 7284   const Register zidx = tmp3;
 7285 
 7286   // int zidx, carry;
 7287   // long value;
 7288   // carry = 0;
 7289   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 7290   //    (carry:value)  = (z[i] << 1) | carry ;
 7291   //    z[i] = value;
 7292   // }
 7293 
 7294   movl(zidx, zlen);
 7295   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 7296 
 7297   bind(L_fifth_loop);
 7298   decl(zidx);  // Use decl to preserve carry flag
 7299   decl(zidx);
 7300   jccb(Assembler::negative, L_fifth_loop_exit);
 7301 
 7302   if (UseBMI2Instructions) {
 7303      movq(value, Address(z, zidx, Address::times_4, 0));
 7304      rclq(value, 1);
 7305      rorxq(value, value, 32);
 7306      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7307   }
 7308   else {
 7309     // clear new_carry
 7310     xorl(new_carry, new_carry);
 7311 
 7312     // Shift z[i] by 1, or in previous carry and save new carry
 7313     movq(value, Address(z, zidx, Address::times_4, 0));
 7314     shlq(value, 1);
 7315     adcl(new_carry, 0);
 7316 
 7317     orq(value, prev_carry);
 7318     rorq(value, 0x20);
 7319     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7320 
 7321     // Set previous carry = new carry
 7322     movl(prev_carry, new_carry);
 7323   }
 7324   jmp(L_fifth_loop);
 7325 
 7326   bind(L_fifth_loop_exit);
 7327 }
 7328 
 7329 
 7330 /**
 7331  * Code for BigInteger::squareToLen() intrinsic
 7332  *
 7333  * rdi: x
 7334  * rsi: len
 7335  * r8:  z
 7336  * rcx: zlen
 7337  * r12: tmp1
 7338  * r13: tmp2
 7339  * r14: tmp3
 7340  * r15: tmp4
 7341  * rbx: tmp5
 7342  *
 7343  */
 7344 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) {
 7345 
 7346   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 7347   push(tmp1);
 7348   push(tmp2);
 7349   push(tmp3);
 7350   push(tmp4);
 7351   push(tmp5);
 7352 
 7353   // First loop
 7354   // Store the squares, right shifted one bit (i.e., divided by 2).
 7355   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7356 
 7357   // Add in off-diagonal sums.
 7358   //
 7359   // Second, third (nested) and fourth loops.
 7360   // zlen +=2;
 7361   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 7362   //    carry = 0;
 7363   //    long op2 = x[xidx:xidx+1];
 7364   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 7365   //       k -= 2;
 7366   //       long op1 = x[j:j+1];
 7367   //       long sum = z[k:k+1];
 7368   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 7369   //       z[k:k+1] = sum;
 7370   //    }
 7371   //    add_one_64(z, k, carry, tmp_regs);
 7372   // }
 7373 
 7374   const Register carry = tmp5;
 7375   const Register sum = tmp3;
 7376   const Register op1 = tmp4;
 7377   Register op2 = tmp2;
 7378 
 7379   push(zlen);
 7380   push(len);
 7381   addl(zlen,2);
 7382   bind(L_second_loop);
 7383   xorq(carry, carry);
 7384   subl(zlen, 4);
 7385   subl(len, 2);
 7386   push(zlen);
 7387   push(len);
 7388   cmpl(len, 0);
 7389   jccb(Assembler::lessEqual, L_second_loop_exit);
 7390 
 7391   // Multiply an array by one 64 bit long.
 7392   if (UseBMI2Instructions) {
 7393     op2 = rdxReg;
 7394     movq(op2, Address(x, len, Address::times_4,  0));
 7395     rorxq(op2, op2, 32);
 7396   }
 7397   else {
 7398     movq(op2, Address(x, len, Address::times_4,  0));
 7399     rorq(op2, 32);
 7400   }
 7401 
 7402   bind(L_third_loop);
 7403   decrementl(len);
 7404   jccb(Assembler::negative, L_third_loop_exit);
 7405   decrementl(len);
 7406   jccb(Assembler::negative, L_last_x);
 7407 
 7408   movq(op1, Address(x, len, Address::times_4,  0));
 7409   rorq(op1, 32);
 7410 
 7411   bind(L_multiply);
 7412   subl(zlen, 2);
 7413   movq(sum, Address(z, zlen, Address::times_4,  0));
 7414 
 7415   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 7416   if (UseBMI2Instructions) {
 7417     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 7418   }
 7419   else {
 7420     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7421   }
 7422 
 7423   movq(Address(z, zlen, Address::times_4, 0), sum);
 7424 
 7425   jmp(L_third_loop);
 7426   bind(L_third_loop_exit);
 7427 
 7428   // Fourth loop
 7429   // Add 64 bit long carry into z with carry propagation.
 7430   // Uses offsetted zlen.
 7431   add_one_64(z, zlen, carry, tmp1);
 7432 
 7433   pop(len);
 7434   pop(zlen);
 7435   jmp(L_second_loop);
 7436 
 7437   // Next infrequent code is moved outside loops.
 7438   bind(L_last_x);
 7439   movl(op1, Address(x, 0));
 7440   jmp(L_multiply);
 7441 
 7442   bind(L_second_loop_exit);
 7443   pop(len);
 7444   pop(zlen);
 7445   pop(len);
 7446   pop(zlen);
 7447 
 7448   // Fifth loop
 7449   // Shift z left 1 bit.
 7450   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 7451 
 7452   // z[zlen-1] |= x[len-1] & 1;
 7453   movl(tmp3, Address(x, len, Address::times_4, -4));
 7454   andl(tmp3, 1);
 7455   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 7456 
 7457   pop(tmp5);
 7458   pop(tmp4);
 7459   pop(tmp3);
 7460   pop(tmp2);
 7461   pop(tmp1);
 7462 }
 7463 
 7464 /**
 7465  * Helper function for mul_add()
 7466  * Multiply the in[] by int k and add to out[] starting at offset offs using
 7467  * 128 bit by 32 bit multiply and return the carry in tmp5.
 7468  * Only quad int aligned length of in[] is operated on in this function.
 7469  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 7470  * This function preserves out, in and k registers.
 7471  * len and offset point to the appropriate index in "in" & "out" correspondingly
 7472  * tmp5 has the carry.
 7473  * other registers are temporary and are modified.
 7474  *
 7475  */
 7476 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 7477   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 7478   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7479 
 7480   Label L_first_loop, L_first_loop_exit;
 7481 
 7482   movl(tmp1, len);
 7483   shrl(tmp1, 2);
 7484 
 7485   bind(L_first_loop);
 7486   subl(tmp1, 1);
 7487   jccb(Assembler::negative, L_first_loop_exit);
 7488 
 7489   subl(len, 4);
 7490   subl(offset, 4);
 7491 
 7492   Register op2 = tmp2;
 7493   const Register sum = tmp3;
 7494   const Register op1 = tmp4;
 7495   const Register carry = tmp5;
 7496 
 7497   if (UseBMI2Instructions) {
 7498     op2 = rdxReg;
 7499   }
 7500 
 7501   movq(op1, Address(in, len, Address::times_4,  8));
 7502   rorq(op1, 32);
 7503   movq(sum, Address(out, offset, Address::times_4,  8));
 7504   rorq(sum, 32);
 7505   if (UseBMI2Instructions) {
 7506     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7507   }
 7508   else {
 7509     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7510   }
 7511   // Store back in big endian from little endian
 7512   rorq(sum, 0x20);
 7513   movq(Address(out, offset, Address::times_4,  8), sum);
 7514 
 7515   movq(op1, Address(in, len, Address::times_4,  0));
 7516   rorq(op1, 32);
 7517   movq(sum, Address(out, offset, Address::times_4,  0));
 7518   rorq(sum, 32);
 7519   if (UseBMI2Instructions) {
 7520     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7521   }
 7522   else {
 7523     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7524   }
 7525   // Store back in big endian from little endian
 7526   rorq(sum, 0x20);
 7527   movq(Address(out, offset, Address::times_4,  0), sum);
 7528 
 7529   jmp(L_first_loop);
 7530   bind(L_first_loop_exit);
 7531 }
 7532 
 7533 /**
 7534  * Code for BigInteger::mulAdd() intrinsic
 7535  *
 7536  * rdi: out
 7537  * rsi: in
 7538  * r11: offs (out.length - offset)
 7539  * rcx: len
 7540  * r8:  k
 7541  * r12: tmp1
 7542  * r13: tmp2
 7543  * r14: tmp3
 7544  * r15: tmp4
 7545  * rbx: tmp5
 7546  * Multiply the in[] by word k and add to out[], return the carry in rax
 7547  */
 7548 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 7549    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 7550    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7551 
 7552   Label L_carry, L_last_in, L_done;
 7553 
 7554 // carry = 0;
 7555 // for (int j=len-1; j >= 0; j--) {
 7556 //    long product = (in[j] & LONG_MASK) * kLong +
 7557 //                   (out[offs] & LONG_MASK) + carry;
 7558 //    out[offs--] = (int)product;
 7559 //    carry = product >>> 32;
 7560 // }
 7561 //
 7562   push(tmp1);
 7563   push(tmp2);
 7564   push(tmp3);
 7565   push(tmp4);
 7566   push(tmp5);
 7567 
 7568   Register op2 = tmp2;
 7569   const Register sum = tmp3;
 7570   const Register op1 = tmp4;
 7571   const Register carry =  tmp5;
 7572 
 7573   if (UseBMI2Instructions) {
 7574     op2 = rdxReg;
 7575     movl(op2, k);
 7576   }
 7577   else {
 7578     movl(op2, k);
 7579   }
 7580 
 7581   xorq(carry, carry);
 7582 
 7583   //First loop
 7584 
 7585   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 7586   //The carry is in tmp5
 7587   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7588 
 7589   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 7590   decrementl(len);
 7591   jccb(Assembler::negative, L_carry);
 7592   decrementl(len);
 7593   jccb(Assembler::negative, L_last_in);
 7594 
 7595   movq(op1, Address(in, len, Address::times_4,  0));
 7596   rorq(op1, 32);
 7597 
 7598   subl(offs, 2);
 7599   movq(sum, Address(out, offs, Address::times_4,  0));
 7600   rorq(sum, 32);
 7601 
 7602   if (UseBMI2Instructions) {
 7603     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7604   }
 7605   else {
 7606     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7607   }
 7608 
 7609   // Store back in big endian from little endian
 7610   rorq(sum, 0x20);
 7611   movq(Address(out, offs, Address::times_4,  0), sum);
 7612 
 7613   testl(len, len);
 7614   jccb(Assembler::zero, L_carry);
 7615 
 7616   //Multiply the last in[] entry, if any
 7617   bind(L_last_in);
 7618   movl(op1, Address(in, 0));
 7619   movl(sum, Address(out, offs, Address::times_4,  -4));
 7620 
 7621   movl(raxReg, k);
 7622   mull(op1); //tmp4 * eax -> edx:eax
 7623   addl(sum, carry);
 7624   adcl(rdxReg, 0);
 7625   addl(sum, raxReg);
 7626   adcl(rdxReg, 0);
 7627   movl(carry, rdxReg);
 7628 
 7629   movl(Address(out, offs, Address::times_4,  -4), sum);
 7630 
 7631   bind(L_carry);
 7632   //return tmp5/carry as carry in rax
 7633   movl(rax, carry);
 7634 
 7635   bind(L_done);
 7636   pop(tmp5);
 7637   pop(tmp4);
 7638   pop(tmp3);
 7639   pop(tmp2);
 7640   pop(tmp1);
 7641 }
 7642 
 7643 /**
 7644  * Emits code to update CRC-32 with a byte value according to constants in table
 7645  *
 7646  * @param [in,out]crc   Register containing the crc.
 7647  * @param [in]val       Register containing the byte to fold into the CRC.
 7648  * @param [in]table     Register containing the table of crc constants.
 7649  *
 7650  * uint32_t crc;
 7651  * val = crc_table[(val ^ crc) & 0xFF];
 7652  * crc = val ^ (crc >> 8);
 7653  *
 7654  */
 7655 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 7656   xorl(val, crc);
 7657   andl(val, 0xFF);
 7658   shrl(crc, 8); // unsigned shift
 7659   xorl(crc, Address(table, val, Address::times_4, 0));
 7660 }
 7661 
 7662 /**
 7663  * Fold 128-bit data chunk
 7664  */
 7665 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 7666   if (UseAVX > 0) {
 7667     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 7668     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 7669     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 7670     pxor(xcrc, xtmp);
 7671   } else {
 7672     movdqa(xtmp, xcrc);
 7673     pclmulhdq(xtmp, xK);   // [123:64]
 7674     pclmulldq(xcrc, xK);   // [63:0]
 7675     pxor(xcrc, xtmp);
 7676     movdqu(xtmp, Address(buf, offset));
 7677     pxor(xcrc, xtmp);
 7678   }
 7679 }
 7680 
 7681 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 7682   if (UseAVX > 0) {
 7683     vpclmulhdq(xtmp, xK, xcrc);
 7684     vpclmulldq(xcrc, xK, xcrc);
 7685     pxor(xcrc, xbuf);
 7686     pxor(xcrc, xtmp);
 7687   } else {
 7688     movdqa(xtmp, xcrc);
 7689     pclmulhdq(xtmp, xK);
 7690     pclmulldq(xcrc, xK);
 7691     pxor(xcrc, xbuf);
 7692     pxor(xcrc, xtmp);
 7693   }
 7694 }
 7695 
 7696 /**
 7697  * 8-bit folds to compute 32-bit CRC
 7698  *
 7699  * uint64_t xcrc;
 7700  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 7701  */
 7702 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 7703   movdl(tmp, xcrc);
 7704   andl(tmp, 0xFF);
 7705   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 7706   psrldq(xcrc, 1); // unsigned shift one byte
 7707   pxor(xcrc, xtmp);
 7708 }
 7709 
 7710 /**
 7711  * uint32_t crc;
 7712  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 7713  */
 7714 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 7715   movl(tmp, crc);
 7716   andl(tmp, 0xFF);
 7717   shrl(crc, 8);
 7718   xorl(crc, Address(table, tmp, Address::times_4, 0));
 7719 }
 7720 
 7721 /**
 7722  * @param crc   register containing existing CRC (32-bit)
 7723  * @param buf   register pointing to input byte buffer (byte*)
 7724  * @param len   register containing number of bytes
 7725  * @param table register that will contain address of CRC table
 7726  * @param tmp   scratch register
 7727  */
 7728 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 7729   assert_different_registers(crc, buf, len, table, tmp, rax);
 7730 
 7731   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 7732   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 7733 
 7734   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 7735   // context for the registers used, where all instructions below are using 128-bit mode
 7736   // On EVEX without VL and BW, these instructions will all be AVX.
 7737   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 7738   notl(crc); // ~crc
 7739   cmpl(len, 16);
 7740   jcc(Assembler::less, L_tail);
 7741 
 7742   // Align buffer to 16 bytes
 7743   movl(tmp, buf);
 7744   andl(tmp, 0xF);
 7745   jccb(Assembler::zero, L_aligned);
 7746   subl(tmp,  16);
 7747   addl(len, tmp);
 7748 
 7749   align(4);
 7750   BIND(L_align_loop);
 7751   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7752   update_byte_crc32(crc, rax, table);
 7753   increment(buf);
 7754   incrementl(tmp);
 7755   jccb(Assembler::less, L_align_loop);
 7756 
 7757   BIND(L_aligned);
 7758   movl(tmp, len); // save
 7759   shrl(len, 4);
 7760   jcc(Assembler::zero, L_tail_restore);
 7761 
 7762   // Fold crc into first bytes of vector
 7763   movdqa(xmm1, Address(buf, 0));
 7764   movdl(rax, xmm1);
 7765   xorl(crc, rax);
 7766   if (VM_Version::supports_sse4_1()) {
 7767     pinsrd(xmm1, crc, 0);
 7768   } else {
 7769     pinsrw(xmm1, crc, 0);
 7770     shrl(crc, 16);
 7771     pinsrw(xmm1, crc, 1);
 7772   }
 7773   addptr(buf, 16);
 7774   subl(len, 4); // len > 0
 7775   jcc(Assembler::less, L_fold_tail);
 7776 
 7777   movdqa(xmm2, Address(buf,  0));
 7778   movdqa(xmm3, Address(buf, 16));
 7779   movdqa(xmm4, Address(buf, 32));
 7780   addptr(buf, 48);
 7781   subl(len, 3);
 7782   jcc(Assembler::lessEqual, L_fold_512b);
 7783 
 7784   // Fold total 512 bits of polynomial on each iteration,
 7785   // 128 bits per each of 4 parallel streams.
 7786   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 7787 
 7788   align32();
 7789   BIND(L_fold_512b_loop);
 7790   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7791   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 7792   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 7793   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 7794   addptr(buf, 64);
 7795   subl(len, 4);
 7796   jcc(Assembler::greater, L_fold_512b_loop);
 7797 
 7798   // Fold 512 bits to 128 bits.
 7799   BIND(L_fold_512b);
 7800   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7801   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 7802   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 7803   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 7804 
 7805   // Fold the rest of 128 bits data chunks
 7806   BIND(L_fold_tail);
 7807   addl(len, 3);
 7808   jccb(Assembler::lessEqual, L_fold_128b);
 7809   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7810 
 7811   BIND(L_fold_tail_loop);
 7812   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7813   addptr(buf, 16);
 7814   decrementl(len);
 7815   jccb(Assembler::greater, L_fold_tail_loop);
 7816 
 7817   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 7818   BIND(L_fold_128b);
 7819   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 7820   if (UseAVX > 0) {
 7821     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 7822     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 7823     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 7824   } else {
 7825     movdqa(xmm2, xmm0);
 7826     pclmulqdq(xmm2, xmm1, 0x1);
 7827     movdqa(xmm3, xmm0);
 7828     pand(xmm3, xmm2);
 7829     pclmulqdq(xmm0, xmm3, 0x1);
 7830   }
 7831   psrldq(xmm1, 8);
 7832   psrldq(xmm2, 4);
 7833   pxor(xmm0, xmm1);
 7834   pxor(xmm0, xmm2);
 7835 
 7836   // 8 8-bit folds to compute 32-bit CRC.
 7837   for (int j = 0; j < 4; j++) {
 7838     fold_8bit_crc32(xmm0, table, xmm1, rax);
 7839   }
 7840   movdl(crc, xmm0); // mov 32 bits to general register
 7841   for (int j = 0; j < 4; j++) {
 7842     fold_8bit_crc32(crc, table, rax);
 7843   }
 7844 
 7845   BIND(L_tail_restore);
 7846   movl(len, tmp); // restore
 7847   BIND(L_tail);
 7848   andl(len, 0xf);
 7849   jccb(Assembler::zero, L_exit);
 7850 
 7851   // Fold the rest of bytes
 7852   align(4);
 7853   BIND(L_tail_loop);
 7854   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7855   update_byte_crc32(crc, rax, table);
 7856   increment(buf);
 7857   decrementl(len);
 7858   jccb(Assembler::greater, L_tail_loop);
 7859 
 7860   BIND(L_exit);
 7861   notl(crc); // ~c
 7862 }
 7863 
 7864 // Helper function for AVX 512 CRC32
 7865 // Fold 512-bit data chunks
 7866 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 7867                                              Register pos, int offset) {
 7868   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 7869   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 7870   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 7871   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 7872   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 7873 }
 7874 
 7875 // Helper function for AVX 512 CRC32
 7876 // Compute CRC32 for < 256B buffers
 7877 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 7878                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 7879                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 7880 
 7881   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 7882   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 7883   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 7884 
 7885   // check if there is enough buffer to be able to fold 16B at a time
 7886   cmpl(len, 32);
 7887   jcc(Assembler::less, L_less_than_32);
 7888 
 7889   // if there is, load the constants
 7890   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 7891   movdl(xmm0, crc);                        // get the initial crc value
 7892   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7893   pxor(xmm7, xmm0);
 7894 
 7895   // update the buffer pointer
 7896   addl(pos, 16);
 7897   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 7898   subl(len, 32);
 7899   jmp(L_16B_reduction_loop);
 7900 
 7901   bind(L_less_than_32);
 7902   //mov initial crc to the return value. this is necessary for zero - length buffers.
 7903   movl(rax, crc);
 7904   testl(len, len);
 7905   jcc(Assembler::equal, L_cleanup);
 7906 
 7907   movdl(xmm0, crc);                        //get the initial crc value
 7908 
 7909   cmpl(len, 16);
 7910   jcc(Assembler::equal, L_exact_16_left);
 7911   jcc(Assembler::less, L_less_than_16_left);
 7912 
 7913   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7914   pxor(xmm7, xmm0);                       //xor the initial crc value
 7915   addl(pos, 16);
 7916   subl(len, 16);
 7917   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 7918   jmp(L_get_last_two_xmms);
 7919 
 7920   bind(L_less_than_16_left);
 7921   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 7922   pxor(xmm1, xmm1);
 7923   movptr(tmp1, rsp);
 7924   movdqu(Address(tmp1, 0 * 16), xmm1);
 7925 
 7926   cmpl(len, 4);
 7927   jcc(Assembler::less, L_only_less_than_4);
 7928 
 7929   //backup the counter value
 7930   movl(tmp2, len);
 7931   cmpl(len, 8);
 7932   jcc(Assembler::less, L_less_than_8_left);
 7933 
 7934   //load 8 Bytes
 7935   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 7936   movq(Address(tmp1, 0 * 16), rax);
 7937   addptr(tmp1, 8);
 7938   subl(len, 8);
 7939   addl(pos, 8);
 7940 
 7941   bind(L_less_than_8_left);
 7942   cmpl(len, 4);
 7943   jcc(Assembler::less, L_less_than_4_left);
 7944 
 7945   //load 4 Bytes
 7946   movl(rax, Address(buf, pos, Address::times_1, 0));
 7947   movl(Address(tmp1, 0 * 16), rax);
 7948   addptr(tmp1, 4);
 7949   subl(len, 4);
 7950   addl(pos, 4);
 7951 
 7952   bind(L_less_than_4_left);
 7953   cmpl(len, 2);
 7954   jcc(Assembler::less, L_less_than_2_left);
 7955 
 7956   // load 2 Bytes
 7957   movw(rax, Address(buf, pos, Address::times_1, 0));
 7958   movl(Address(tmp1, 0 * 16), rax);
 7959   addptr(tmp1, 2);
 7960   subl(len, 2);
 7961   addl(pos, 2);
 7962 
 7963   bind(L_less_than_2_left);
 7964   cmpl(len, 1);
 7965   jcc(Assembler::less, L_zero_left);
 7966 
 7967   // load 1 Byte
 7968   movb(rax, Address(buf, pos, Address::times_1, 0));
 7969   movb(Address(tmp1, 0 * 16), rax);
 7970 
 7971   bind(L_zero_left);
 7972   movdqu(xmm7, Address(rsp, 0));
 7973   pxor(xmm7, xmm0);                       //xor the initial crc value
 7974 
 7975   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 7976   movdqu(xmm0, Address(rax, tmp2));
 7977   pshufb(xmm7, xmm0);
 7978   jmp(L_128_done);
 7979 
 7980   bind(L_exact_16_left);
 7981   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 7982   pxor(xmm7, xmm0);                       //xor the initial crc value
 7983   jmp(L_128_done);
 7984 
 7985   bind(L_only_less_than_4);
 7986   cmpl(len, 3);
 7987   jcc(Assembler::less, L_only_less_than_3);
 7988 
 7989   // load 3 Bytes
 7990   movb(rax, Address(buf, pos, Address::times_1, 0));
 7991   movb(Address(tmp1, 0), rax);
 7992 
 7993   movb(rax, Address(buf, pos, Address::times_1, 1));
 7994   movb(Address(tmp1, 1), rax);
 7995 
 7996   movb(rax, Address(buf, pos, Address::times_1, 2));
 7997   movb(Address(tmp1, 2), rax);
 7998 
 7999   movdqu(xmm7, Address(rsp, 0));
 8000   pxor(xmm7, xmm0);                     //xor the initial crc value
 8001 
 8002   pslldq(xmm7, 0x5);
 8003   jmp(L_barrett);
 8004   bind(L_only_less_than_3);
 8005   cmpl(len, 2);
 8006   jcc(Assembler::less, L_only_less_than_2);
 8007 
 8008   // load 2 Bytes
 8009   movb(rax, Address(buf, pos, Address::times_1, 0));
 8010   movb(Address(tmp1, 0), rax);
 8011 
 8012   movb(rax, Address(buf, pos, Address::times_1, 1));
 8013   movb(Address(tmp1, 1), rax);
 8014 
 8015   movdqu(xmm7, Address(rsp, 0));
 8016   pxor(xmm7, xmm0);                     //xor the initial crc value
 8017 
 8018   pslldq(xmm7, 0x6);
 8019   jmp(L_barrett);
 8020 
 8021   bind(L_only_less_than_2);
 8022   //load 1 Byte
 8023   movb(rax, Address(buf, pos, Address::times_1, 0));
 8024   movb(Address(tmp1, 0), rax);
 8025 
 8026   movdqu(xmm7, Address(rsp, 0));
 8027   pxor(xmm7, xmm0);                     //xor the initial crc value
 8028 
 8029   pslldq(xmm7, 0x7);
 8030 }
 8031 
 8032 /**
 8033 * Compute CRC32 using AVX512 instructions
 8034 * param crc   register containing existing CRC (32-bit)
 8035 * param buf   register pointing to input byte buffer (byte*)
 8036 * param len   register containing number of bytes
 8037 * param table address of crc or crc32c table
 8038 * param tmp1  scratch register
 8039 * param tmp2  scratch register
 8040 * return rax  result register
 8041 *
 8042 * This routine is identical for crc32c with the exception of the precomputed constant
 8043 * table which will be passed as the table argument.  The calculation steps are
 8044 * the same for both variants.
 8045 */
 8046 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8047   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8048 
 8049   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8050   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8051   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8052   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8053   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8054 
 8055   const Register pos = r12;
 8056   push(r12);
 8057   subptr(rsp, 16 * 2 + 8);
 8058 
 8059   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8060   // context for the registers used, where all instructions below are using 128-bit mode
 8061   // On EVEX without VL and BW, these instructions will all be AVX.
 8062   movl(pos, 0);
 8063 
 8064   // check if smaller than 256B
 8065   cmpl(len, 256);
 8066   jcc(Assembler::less, L_less_than_256);
 8067 
 8068   // load the initial crc value
 8069   movdl(xmm10, crc);
 8070 
 8071   // receive the initial 64B data, xor the initial crc value
 8072   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8073   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8074   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8075   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8076 
 8077   subl(len, 256);
 8078   cmpl(len, 256);
 8079   jcc(Assembler::less, L_fold_128_B_loop);
 8080 
 8081   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8082   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8083   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8084   subl(len, 256);
 8085 
 8086   bind(L_fold_256_B_loop);
 8087   addl(pos, 256);
 8088   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8089   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8090   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8091   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8092 
 8093   subl(len, 256);
 8094   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8095 
 8096   // Fold 256 into 128
 8097   addl(pos, 256);
 8098   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8099   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8100   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8101 
 8102   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8103   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8104   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8105 
 8106   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8107   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8108 
 8109   addl(len, 128);
 8110   jmp(L_fold_128_B_register);
 8111 
 8112   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8113   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8114 
 8115   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8116   bind(L_fold_128_B_loop);
 8117   addl(pos, 128);
 8118   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8119   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8120 
 8121   subl(len, 128);
 8122   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8123 
 8124   addl(pos, 128);
 8125 
 8126   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8127   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8128   bind(L_fold_128_B_register);
 8129   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8130   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8131   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8132   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8133   // save last that has no multiplicand
 8134   vextracti64x2(xmm7, xmm4, 3);
 8135 
 8136   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8137   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8138   // Needed later in reduction loop
 8139   movdqu(xmm10, Address(table, 1 * 16));
 8140   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8141   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8142 
 8143   // Swap 1,0,3,2 - 01 00 11 10
 8144   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8145   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8146   vextracti128(xmm5, xmm8, 1);
 8147   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8148 
 8149   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8150   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8151   addl(len, 128 - 16);
 8152   jcc(Assembler::less, L_final_reduction_for_128);
 8153 
 8154   bind(L_16B_reduction_loop);
 8155   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8156   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8157   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8158   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8159   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8160   addl(pos, 16);
 8161   subl(len, 16);
 8162   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8163 
 8164   bind(L_final_reduction_for_128);
 8165   addl(len, 16);
 8166   jcc(Assembler::equal, L_128_done);
 8167 
 8168   bind(L_get_last_two_xmms);
 8169   movdqu(xmm2, xmm7);
 8170   addl(pos, len);
 8171   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8172   subl(pos, len);
 8173 
 8174   // get rid of the extra data that was loaded before
 8175   // load the shift constant
 8176   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8177   movdqu(xmm0, Address(rax, len));
 8178   addl(rax, len);
 8179 
 8180   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8181   //Change mask to 512
 8182   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8183   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8184 
 8185   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8186   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8187   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8188   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8189   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8190 
 8191   bind(L_128_done);
 8192   // compute crc of a 128-bit value
 8193   movdqu(xmm10, Address(table, 3 * 16));
 8194   movdqu(xmm0, xmm7);
 8195 
 8196   // 64b fold
 8197   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8198   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8199   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8200 
 8201   // 32b fold
 8202   movdqu(xmm0, xmm7);
 8203   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8204   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8205   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8206   jmp(L_barrett);
 8207 
 8208   bind(L_less_than_256);
 8209   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);
 8210 
 8211   //barrett reduction
 8212   bind(L_barrett);
 8213   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8214   movdqu(xmm1, xmm7);
 8215   movdqu(xmm2, xmm7);
 8216   movdqu(xmm10, Address(table, 4 * 16));
 8217 
 8218   pclmulqdq(xmm7, xmm10, 0x0);
 8219   pxor(xmm7, xmm2);
 8220   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8221   movdqu(xmm2, xmm7);
 8222   pclmulqdq(xmm7, xmm10, 0x10);
 8223   pxor(xmm7, xmm2);
 8224   pxor(xmm7, xmm1);
 8225   pextrd(crc, xmm7, 2);
 8226 
 8227   bind(L_cleanup);
 8228   addptr(rsp, 16 * 2 + 8);
 8229   pop(r12);
 8230 }
 8231 
 8232 // S. Gueron / Information Processing Letters 112 (2012) 184
 8233 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8234 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8235 // Output: the 64-bit carry-less product of B * CONST
 8236 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8237                                      Register tmp1, Register tmp2, Register tmp3) {
 8238   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8239   if (n > 0) {
 8240     addq(tmp3, n * 256 * 8);
 8241   }
 8242   //    Q1 = TABLEExt[n][B & 0xFF];
 8243   movl(tmp1, in);
 8244   andl(tmp1, 0x000000FF);
 8245   shll(tmp1, 3);
 8246   addq(tmp1, tmp3);
 8247   movq(tmp1, Address(tmp1, 0));
 8248 
 8249   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8250   movl(tmp2, in);
 8251   shrl(tmp2, 8);
 8252   andl(tmp2, 0x000000FF);
 8253   shll(tmp2, 3);
 8254   addq(tmp2, tmp3);
 8255   movq(tmp2, Address(tmp2, 0));
 8256 
 8257   shlq(tmp2, 8);
 8258   xorq(tmp1, tmp2);
 8259 
 8260   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8261   movl(tmp2, in);
 8262   shrl(tmp2, 16);
 8263   andl(tmp2, 0x000000FF);
 8264   shll(tmp2, 3);
 8265   addq(tmp2, tmp3);
 8266   movq(tmp2, Address(tmp2, 0));
 8267 
 8268   shlq(tmp2, 16);
 8269   xorq(tmp1, tmp2);
 8270 
 8271   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8272   shrl(in, 24);
 8273   andl(in, 0x000000FF);
 8274   shll(in, 3);
 8275   addq(in, tmp3);
 8276   movq(in, Address(in, 0));
 8277 
 8278   shlq(in, 24);
 8279   xorq(in, tmp1);
 8280   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8281 }
 8282 
 8283 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8284                                       Register in_out,
 8285                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8286                                       XMMRegister w_xtmp2,
 8287                                       Register tmp1,
 8288                                       Register n_tmp2, Register n_tmp3) {
 8289   if (is_pclmulqdq_supported) {
 8290     movdl(w_xtmp1, in_out); // modified blindly
 8291 
 8292     movl(tmp1, const_or_pre_comp_const_index);
 8293     movdl(w_xtmp2, tmp1);
 8294     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8295 
 8296     movdq(in_out, w_xtmp1);
 8297   } else {
 8298     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 8299   }
 8300 }
 8301 
 8302 // Recombination Alternative 2: No bit-reflections
 8303 // T1 = (CRC_A * U1) << 1
 8304 // T2 = (CRC_B * U2) << 1
 8305 // C1 = T1 >> 32
 8306 // C2 = T2 >> 32
 8307 // T1 = T1 & 0xFFFFFFFF
 8308 // T2 = T2 & 0xFFFFFFFF
 8309 // T1 = CRC32(0, T1)
 8310 // T2 = CRC32(0, T2)
 8311 // C1 = C1 ^ T1
 8312 // C2 = C2 ^ T2
 8313 // CRC = C1 ^ C2 ^ CRC_C
 8314 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,
 8315                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8316                                      Register tmp1, Register tmp2,
 8317                                      Register n_tmp3) {
 8318   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8319   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8320   shlq(in_out, 1);
 8321   movl(tmp1, in_out);
 8322   shrq(in_out, 32);
 8323   xorl(tmp2, tmp2);
 8324   crc32(tmp2, tmp1, 4);
 8325   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 8326   shlq(in1, 1);
 8327   movl(tmp1, in1);
 8328   shrq(in1, 32);
 8329   xorl(tmp2, tmp2);
 8330   crc32(tmp2, tmp1, 4);
 8331   xorl(in1, tmp2);
 8332   xorl(in_out, in1);
 8333   xorl(in_out, in2);
 8334 }
 8335 
 8336 // Set N to predefined value
 8337 // Subtract from a length of a buffer
 8338 // execute in a loop:
 8339 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 8340 // for i = 1 to N do
 8341 //  CRC_A = CRC32(CRC_A, A[i])
 8342 //  CRC_B = CRC32(CRC_B, B[i])
 8343 //  CRC_C = CRC32(CRC_C, C[i])
 8344 // end for
 8345 // Recombine
 8346 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,
 8347                                        Register in_out1, Register in_out2, Register in_out3,
 8348                                        Register tmp1, Register tmp2, Register tmp3,
 8349                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8350                                        Register tmp4, Register tmp5,
 8351                                        Register n_tmp6) {
 8352   Label L_processPartitions;
 8353   Label L_processPartition;
 8354   Label L_exit;
 8355 
 8356   bind(L_processPartitions);
 8357   cmpl(in_out1, 3 * size);
 8358   jcc(Assembler::less, L_exit);
 8359     xorl(tmp1, tmp1);
 8360     xorl(tmp2, tmp2);
 8361     movq(tmp3, in_out2);
 8362     addq(tmp3, size);
 8363 
 8364     bind(L_processPartition);
 8365       crc32(in_out3, Address(in_out2, 0), 8);
 8366       crc32(tmp1, Address(in_out2, size), 8);
 8367       crc32(tmp2, Address(in_out2, size * 2), 8);
 8368       addq(in_out2, 8);
 8369       cmpq(in_out2, tmp3);
 8370       jcc(Assembler::less, L_processPartition);
 8371     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 8372             w_xtmp1, w_xtmp2, w_xtmp3,
 8373             tmp4, tmp5,
 8374             n_tmp6);
 8375     addq(in_out2, 2 * size);
 8376     subl(in_out1, 3 * size);
 8377     jmp(L_processPartitions);
 8378 
 8379   bind(L_exit);
 8380 }
 8381 
 8382 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 8383 // Input: A buffer I of L bytes.
 8384 // Output: the CRC32C value of the buffer.
 8385 // Notations:
 8386 // Write L = 24N + r, with N = floor (L/24).
 8387 // r = L mod 24 (0 <= r < 24).
 8388 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 8389 // N quadwords, and R consists of r bytes.
 8390 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 8391 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 8392 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 8393 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 8394 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 8395                                           Register tmp1, Register tmp2, Register tmp3,
 8396                                           Register tmp4, Register tmp5, Register tmp6,
 8397                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8398                                           bool is_pclmulqdq_supported) {
 8399   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 8400   Label L_wordByWord;
 8401   Label L_byteByByteProlog;
 8402   Label L_byteByByte;
 8403   Label L_exit;
 8404 
 8405   if (is_pclmulqdq_supported ) {
 8406     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 8407     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 8408 
 8409     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 8410     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 8411 
 8412     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 8413     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 8414     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 8415   } else {
 8416     const_or_pre_comp_const_index[0] = 1;
 8417     const_or_pre_comp_const_index[1] = 0;
 8418 
 8419     const_or_pre_comp_const_index[2] = 3;
 8420     const_or_pre_comp_const_index[3] = 2;
 8421 
 8422     const_or_pre_comp_const_index[4] = 5;
 8423     const_or_pre_comp_const_index[5] = 4;
 8424    }
 8425   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 8426                     in2, in1, in_out,
 8427                     tmp1, tmp2, tmp3,
 8428                     w_xtmp1, w_xtmp2, w_xtmp3,
 8429                     tmp4, tmp5,
 8430                     tmp6);
 8431   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 8432                     in2, in1, in_out,
 8433                     tmp1, tmp2, tmp3,
 8434                     w_xtmp1, w_xtmp2, w_xtmp3,
 8435                     tmp4, tmp5,
 8436                     tmp6);
 8437   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 8438                     in2, in1, in_out,
 8439                     tmp1, tmp2, tmp3,
 8440                     w_xtmp1, w_xtmp2, w_xtmp3,
 8441                     tmp4, tmp5,
 8442                     tmp6);
 8443   movl(tmp1, in2);
 8444   andl(tmp1, 0x00000007);
 8445   negl(tmp1);
 8446   addl(tmp1, in2);
 8447   addq(tmp1, in1);
 8448 
 8449   cmpq(in1, tmp1);
 8450   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 8451   align(16);
 8452   BIND(L_wordByWord);
 8453     crc32(in_out, Address(in1, 0), 8);
 8454     addq(in1, 8);
 8455     cmpq(in1, tmp1);
 8456     jcc(Assembler::less, L_wordByWord);
 8457 
 8458   BIND(L_byteByByteProlog);
 8459   andl(in2, 0x00000007);
 8460   movl(tmp2, 1);
 8461 
 8462   cmpl(tmp2, in2);
 8463   jccb(Assembler::greater, L_exit);
 8464   BIND(L_byteByByte);
 8465     crc32(in_out, Address(in1, 0), 1);
 8466     incq(in1);
 8467     incl(tmp2);
 8468     cmpl(tmp2, in2);
 8469     jcc(Assembler::lessEqual, L_byteByByte);
 8470 
 8471   BIND(L_exit);
 8472 }
 8473 #undef BIND
 8474 #undef BLOCK_COMMENT
 8475 
 8476 // Compress char[] array to byte[].
 8477 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 8478 // Return the array length if every element in array can be encoded,
 8479 // otherwise, the index of first non-latin1 (> 0xff) character.
 8480 //   @IntrinsicCandidate
 8481 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 8482 //     for (int i = 0; i < len; i++) {
 8483 //       char c = src[srcOff];
 8484 //       if (c > 0xff) {
 8485 //           return i;  // return index of non-latin1 char
 8486 //       }
 8487 //       dst[dstOff] = (byte)c;
 8488 //       srcOff++;
 8489 //       dstOff++;
 8490 //     }
 8491 //     return len;
 8492 //   }
 8493 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 8494   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 8495   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 8496   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 8497   Label copy_chars_loop, done, reset_sp, copy_tail;
 8498 
 8499   // rsi: src
 8500   // rdi: dst
 8501   // rdx: len
 8502   // rcx: tmp5
 8503   // rax: result
 8504 
 8505   // rsi holds start addr of source char[] to be compressed
 8506   // rdi holds start addr of destination byte[]
 8507   // rdx holds length
 8508 
 8509   assert(len != result, "");
 8510 
 8511   // save length for return
 8512   movl(result, len);
 8513 
 8514   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 8515     VM_Version::supports_avx512vlbw() &&
 8516     VM_Version::supports_bmi2()) {
 8517 
 8518     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 8519 
 8520     // alignment
 8521     Label post_alignment;
 8522 
 8523     // if length of the string is less than 32, handle it the old fashioned way
 8524     testl(len, -32);
 8525     jcc(Assembler::zero, below_threshold);
 8526 
 8527     // First check whether a character is compressible ( <= 0xFF).
 8528     // Create mask to test for Unicode chars inside zmm vector
 8529     movl(tmp5, 0x00FF);
 8530     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 8531 
 8532     testl(len, -64);
 8533     jccb(Assembler::zero, post_alignment);
 8534 
 8535     movl(tmp5, dst);
 8536     andl(tmp5, (32 - 1));
 8537     negl(tmp5);
 8538     andl(tmp5, (32 - 1));
 8539 
 8540     // bail out when there is nothing to be done
 8541     testl(tmp5, 0xFFFFFFFF);
 8542     jccb(Assembler::zero, post_alignment);
 8543 
 8544     // ~(~0 << len), where len is the # of remaining elements to process
 8545     movl(len, 0xFFFFFFFF);
 8546     shlxl(len, len, tmp5);
 8547     notl(len);
 8548     kmovdl(mask2, len);
 8549     movl(len, result);
 8550 
 8551     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8552     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8553     ktestd(mask1, mask2);
 8554     jcc(Assembler::carryClear, copy_tail);
 8555 
 8556     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8557 
 8558     addptr(src, tmp5);
 8559     addptr(src, tmp5);
 8560     addptr(dst, tmp5);
 8561     subl(len, tmp5);
 8562 
 8563     bind(post_alignment);
 8564     // end of alignment
 8565 
 8566     movl(tmp5, len);
 8567     andl(tmp5, (32 - 1));    // tail count (in chars)
 8568     andl(len, ~(32 - 1));    // vector count (in chars)
 8569     jccb(Assembler::zero, copy_loop_tail);
 8570 
 8571     lea(src, Address(src, len, Address::times_2));
 8572     lea(dst, Address(dst, len, Address::times_1));
 8573     negptr(len);
 8574 
 8575     bind(copy_32_loop);
 8576     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 8577     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 8578     kortestdl(mask1, mask1);
 8579     jccb(Assembler::carryClear, reset_for_copy_tail);
 8580 
 8581     // All elements in current processed chunk are valid candidates for
 8582     // compression. Write a truncated byte elements to the memory.
 8583     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 8584     addptr(len, 32);
 8585     jccb(Assembler::notZero, copy_32_loop);
 8586 
 8587     bind(copy_loop_tail);
 8588     // bail out when there is nothing to be done
 8589     testl(tmp5, 0xFFFFFFFF);
 8590     jcc(Assembler::zero, done);
 8591 
 8592     movl(len, tmp5);
 8593 
 8594     // ~(~0 << len), where len is the # of remaining elements to process
 8595     movl(tmp5, 0xFFFFFFFF);
 8596     shlxl(tmp5, tmp5, len);
 8597     notl(tmp5);
 8598 
 8599     kmovdl(mask2, tmp5);
 8600 
 8601     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8602     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8603     ktestd(mask1, mask2);
 8604     jcc(Assembler::carryClear, copy_tail);
 8605 
 8606     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8607     jmp(done);
 8608 
 8609     bind(reset_for_copy_tail);
 8610     lea(src, Address(src, tmp5, Address::times_2));
 8611     lea(dst, Address(dst, tmp5, Address::times_1));
 8612     subptr(len, tmp5);
 8613     jmp(copy_chars_loop);
 8614 
 8615     bind(below_threshold);
 8616   }
 8617 
 8618   if (UseSSE42Intrinsics) {
 8619     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 8620 
 8621     // vectored compression
 8622     testl(len, 0xfffffff8);
 8623     jcc(Assembler::zero, copy_tail);
 8624 
 8625     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 8626     movdl(tmp1Reg, tmp5);
 8627     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 8628 
 8629     andl(len, 0xfffffff0);
 8630     jccb(Assembler::zero, copy_16);
 8631 
 8632     // compress 16 chars per iter
 8633     pxor(tmp4Reg, tmp4Reg);
 8634 
 8635     lea(src, Address(src, len, Address::times_2));
 8636     lea(dst, Address(dst, len, Address::times_1));
 8637     negptr(len);
 8638 
 8639     bind(copy_32_loop);
 8640     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 8641     por(tmp4Reg, tmp2Reg);
 8642     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 8643     por(tmp4Reg, tmp3Reg);
 8644     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 8645     jccb(Assembler::notZero, reset_for_copy_tail);
 8646     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 8647     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 8648     addptr(len, 16);
 8649     jccb(Assembler::notZero, copy_32_loop);
 8650 
 8651     // compress next vector of 8 chars (if any)
 8652     bind(copy_16);
 8653     // len = 0
 8654     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 8655     jccb(Assembler::zero, copy_tail_sse);
 8656 
 8657     pxor(tmp3Reg, tmp3Reg);
 8658 
 8659     movdqu(tmp2Reg, Address(src, 0));
 8660     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 8661     jccb(Assembler::notZero, reset_for_copy_tail);
 8662     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 8663     movq(Address(dst, 0), tmp2Reg);
 8664     addptr(src, 16);
 8665     addptr(dst, 8);
 8666     jmpb(copy_tail_sse);
 8667 
 8668     bind(reset_for_copy_tail);
 8669     movl(tmp5, result);
 8670     andl(tmp5, 0x0000000f);
 8671     lea(src, Address(src, tmp5, Address::times_2));
 8672     lea(dst, Address(dst, tmp5, Address::times_1));
 8673     subptr(len, tmp5);
 8674     jmpb(copy_chars_loop);
 8675 
 8676     bind(copy_tail_sse);
 8677     movl(len, result);
 8678     andl(len, 0x00000007);    // tail count (in chars)
 8679   }
 8680   // compress 1 char per iter
 8681   bind(copy_tail);
 8682   testl(len, len);
 8683   jccb(Assembler::zero, done);
 8684   lea(src, Address(src, len, Address::times_2));
 8685   lea(dst, Address(dst, len, Address::times_1));
 8686   negptr(len);
 8687 
 8688   bind(copy_chars_loop);
 8689   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 8690   testl(tmp5, 0xff00);      // check if Unicode char
 8691   jccb(Assembler::notZero, reset_sp);
 8692   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 8693   increment(len);
 8694   jccb(Assembler::notZero, copy_chars_loop);
 8695 
 8696   // add len then return (len will be zero if compress succeeded, otherwise negative)
 8697   bind(reset_sp);
 8698   addl(result, len);
 8699 
 8700   bind(done);
 8701 }
 8702 
 8703 // Inflate byte[] array to char[].
 8704 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 8705 //   @IntrinsicCandidate
 8706 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 8707 //     for (int i = 0; i < len; i++) {
 8708 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 8709 //     }
 8710 //   }
 8711 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 8712   XMMRegister tmp1, Register tmp2, KRegister mask) {
 8713   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 8714   // rsi: src
 8715   // rdi: dst
 8716   // rdx: len
 8717   // rcx: tmp2
 8718 
 8719   // rsi holds start addr of source byte[] to be inflated
 8720   // rdi holds start addr of destination char[]
 8721   // rdx holds length
 8722   assert_different_registers(src, dst, len, tmp2);
 8723   movl(tmp2, len);
 8724   if ((UseAVX > 2) && // AVX512
 8725     VM_Version::supports_avx512vlbw() &&
 8726     VM_Version::supports_bmi2()) {
 8727 
 8728     Label copy_32_loop, copy_tail;
 8729     Register tmp3_aliased = len;
 8730 
 8731     // if length of the string is less than 16, handle it in an old fashioned way
 8732     testl(len, -16);
 8733     jcc(Assembler::zero, below_threshold);
 8734 
 8735     testl(len, -1 * AVX3Threshold);
 8736     jcc(Assembler::zero, avx3_threshold);
 8737 
 8738     // In order to use only one arithmetic operation for the main loop we use
 8739     // this pre-calculation
 8740     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 8741     andl(len, -32);     // vector count
 8742     jccb(Assembler::zero, copy_tail);
 8743 
 8744     lea(src, Address(src, len, Address::times_1));
 8745     lea(dst, Address(dst, len, Address::times_2));
 8746     negptr(len);
 8747 
 8748 
 8749     // inflate 32 chars per iter
 8750     bind(copy_32_loop);
 8751     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 8752     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 8753     addptr(len, 32);
 8754     jcc(Assembler::notZero, copy_32_loop);
 8755 
 8756     bind(copy_tail);
 8757     // bail out when there is nothing to be done
 8758     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 8759     jcc(Assembler::zero, done);
 8760 
 8761     // ~(~0 << length), where length is the # of remaining elements to process
 8762     movl(tmp3_aliased, -1);
 8763     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 8764     notl(tmp3_aliased);
 8765     kmovdl(mask, tmp3_aliased);
 8766     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 8767     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 8768 
 8769     jmp(done);
 8770     bind(avx3_threshold);
 8771   }
 8772   if (UseSSE42Intrinsics) {
 8773     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 8774 
 8775     if (UseAVX > 1) {
 8776       andl(tmp2, (16 - 1));
 8777       andl(len, -16);
 8778       jccb(Assembler::zero, copy_new_tail);
 8779     } else {
 8780       andl(tmp2, 0x00000007);   // tail count (in chars)
 8781       andl(len, 0xfffffff8);    // vector count (in chars)
 8782       jccb(Assembler::zero, copy_tail);
 8783     }
 8784 
 8785     // vectored inflation
 8786     lea(src, Address(src, len, Address::times_1));
 8787     lea(dst, Address(dst, len, Address::times_2));
 8788     negptr(len);
 8789 
 8790     if (UseAVX > 1) {
 8791       bind(copy_16_loop);
 8792       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 8793       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 8794       addptr(len, 16);
 8795       jcc(Assembler::notZero, copy_16_loop);
 8796 
 8797       bind(below_threshold);
 8798       bind(copy_new_tail);
 8799       movl(len, tmp2);
 8800       andl(tmp2, 0x00000007);
 8801       andl(len, 0xFFFFFFF8);
 8802       jccb(Assembler::zero, copy_tail);
 8803 
 8804       pmovzxbw(tmp1, Address(src, 0));
 8805       movdqu(Address(dst, 0), tmp1);
 8806       addptr(src, 8);
 8807       addptr(dst, 2 * 8);
 8808 
 8809       jmp(copy_tail, true);
 8810     }
 8811 
 8812     // inflate 8 chars per iter
 8813     bind(copy_8_loop);
 8814     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 8815     movdqu(Address(dst, len, Address::times_2), tmp1);
 8816     addptr(len, 8);
 8817     jcc(Assembler::notZero, copy_8_loop);
 8818 
 8819     bind(copy_tail);
 8820     movl(len, tmp2);
 8821 
 8822     cmpl(len, 4);
 8823     jccb(Assembler::less, copy_bytes);
 8824 
 8825     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 8826     pmovzxbw(tmp1, tmp1);
 8827     movq(Address(dst, 0), tmp1);
 8828     subptr(len, 4);
 8829     addptr(src, 4);
 8830     addptr(dst, 8);
 8831 
 8832     bind(copy_bytes);
 8833   } else {
 8834     bind(below_threshold);
 8835   }
 8836 
 8837   testl(len, len);
 8838   jccb(Assembler::zero, done);
 8839   lea(src, Address(src, len, Address::times_1));
 8840   lea(dst, Address(dst, len, Address::times_2));
 8841   negptr(len);
 8842 
 8843   // inflate 1 char per iter
 8844   bind(copy_chars_loop);
 8845   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 8846   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 8847   increment(len);
 8848   jcc(Assembler::notZero, copy_chars_loop);
 8849 
 8850   bind(done);
 8851 }
 8852 
 8853 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 8854   switch(type) {
 8855     case T_BYTE:
 8856     case T_BOOLEAN:
 8857       evmovdqub(dst, kmask, src, merge, vector_len);
 8858       break;
 8859     case T_CHAR:
 8860     case T_SHORT:
 8861       evmovdquw(dst, kmask, src, merge, vector_len);
 8862       break;
 8863     case T_INT:
 8864     case T_FLOAT:
 8865       evmovdqul(dst, kmask, src, merge, vector_len);
 8866       break;
 8867     case T_LONG:
 8868     case T_DOUBLE:
 8869       evmovdquq(dst, kmask, src, merge, vector_len);
 8870       break;
 8871     default:
 8872       fatal("Unexpected type argument %s", type2name(type));
 8873       break;
 8874   }
 8875 }
 8876 
 8877 
 8878 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address 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 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 8903   switch(type) {
 8904     case T_BYTE:
 8905     case T_BOOLEAN:
 8906       evmovdqub(dst, kmask, src, merge, vector_len);
 8907       break;
 8908     case T_CHAR:
 8909     case T_SHORT:
 8910       evmovdquw(dst, kmask, src, merge, vector_len);
 8911       break;
 8912     case T_INT:
 8913     case T_FLOAT:
 8914       evmovdqul(dst, kmask, src, merge, vector_len);
 8915       break;
 8916     case T_LONG:
 8917     case T_DOUBLE:
 8918       evmovdquq(dst, kmask, src, merge, vector_len);
 8919       break;
 8920     default:
 8921       fatal("Unexpected type argument %s", type2name(type));
 8922       break;
 8923   }
 8924 }
 8925 
 8926 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 8927   switch(masklen) {
 8928     case 2:
 8929        knotbl(dst, src);
 8930        movl(rtmp, 3);
 8931        kmovbl(ktmp, rtmp);
 8932        kandbl(dst, ktmp, dst);
 8933        break;
 8934     case 4:
 8935        knotbl(dst, src);
 8936        movl(rtmp, 15);
 8937        kmovbl(ktmp, rtmp);
 8938        kandbl(dst, ktmp, dst);
 8939        break;
 8940     case 8:
 8941        knotbl(dst, src);
 8942        break;
 8943     case 16:
 8944        knotwl(dst, src);
 8945        break;
 8946     case 32:
 8947        knotdl(dst, src);
 8948        break;
 8949     case 64:
 8950        knotql(dst, src);
 8951        break;
 8952     default:
 8953       fatal("Unexpected vector length %d", masklen);
 8954       break;
 8955   }
 8956 }
 8957 
 8958 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8959   switch(type) {
 8960     case T_BOOLEAN:
 8961     case T_BYTE:
 8962        kandbl(dst, src1, src2);
 8963        break;
 8964     case T_CHAR:
 8965     case T_SHORT:
 8966        kandwl(dst, src1, src2);
 8967        break;
 8968     case T_INT:
 8969     case T_FLOAT:
 8970        kanddl(dst, src1, src2);
 8971        break;
 8972     case T_LONG:
 8973     case T_DOUBLE:
 8974        kandql(dst, src1, src2);
 8975        break;
 8976     default:
 8977       fatal("Unexpected type argument %s", type2name(type));
 8978       break;
 8979   }
 8980 }
 8981 
 8982 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8983   switch(type) {
 8984     case T_BOOLEAN:
 8985     case T_BYTE:
 8986        korbl(dst, src1, src2);
 8987        break;
 8988     case T_CHAR:
 8989     case T_SHORT:
 8990        korwl(dst, src1, src2);
 8991        break;
 8992     case T_INT:
 8993     case T_FLOAT:
 8994        kordl(dst, src1, src2);
 8995        break;
 8996     case T_LONG:
 8997     case T_DOUBLE:
 8998        korql(dst, src1, src2);
 8999        break;
 9000     default:
 9001       fatal("Unexpected type argument %s", type2name(type));
 9002       break;
 9003   }
 9004 }
 9005 
 9006 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9007   switch(type) {
 9008     case T_BOOLEAN:
 9009     case T_BYTE:
 9010        kxorbl(dst, src1, src2);
 9011        break;
 9012     case T_CHAR:
 9013     case T_SHORT:
 9014        kxorwl(dst, src1, src2);
 9015        break;
 9016     case T_INT:
 9017     case T_FLOAT:
 9018        kxordl(dst, src1, src2);
 9019        break;
 9020     case T_LONG:
 9021     case T_DOUBLE:
 9022        kxorql(dst, src1, src2);
 9023        break;
 9024     default:
 9025       fatal("Unexpected type argument %s", type2name(type));
 9026       break;
 9027   }
 9028 }
 9029 
 9030 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9031   switch(type) {
 9032     case T_BOOLEAN:
 9033     case T_BYTE:
 9034       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9035     case T_CHAR:
 9036     case T_SHORT:
 9037       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9038     case T_INT:
 9039     case T_FLOAT:
 9040       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9041     case T_LONG:
 9042     case T_DOUBLE:
 9043       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9044     default:
 9045       fatal("Unexpected type argument %s", type2name(type)); break;
 9046   }
 9047 }
 9048 
 9049 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9050   switch(type) {
 9051     case T_BOOLEAN:
 9052     case T_BYTE:
 9053       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9054     case T_CHAR:
 9055     case T_SHORT:
 9056       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9057     case T_INT:
 9058     case T_FLOAT:
 9059       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9060     case T_LONG:
 9061     case T_DOUBLE:
 9062       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9063     default:
 9064       fatal("Unexpected type argument %s", type2name(type)); break;
 9065   }
 9066 }
 9067 
 9068 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9069   switch(type) {
 9070     case T_BYTE:
 9071       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9072     case T_SHORT:
 9073       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9074     case T_INT:
 9075       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9076     case T_LONG:
 9077       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9078     default:
 9079       fatal("Unexpected type argument %s", type2name(type)); break;
 9080   }
 9081 }
 9082 
 9083 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9084   switch(type) {
 9085     case T_BYTE:
 9086       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9087     case T_SHORT:
 9088       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9089     case T_INT:
 9090       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9091     case T_LONG:
 9092       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9093     default:
 9094       fatal("Unexpected type argument %s", type2name(type)); break;
 9095   }
 9096 }
 9097 
 9098 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9099   switch(type) {
 9100     case T_BYTE:
 9101       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9102     case T_SHORT:
 9103       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9104     case T_INT:
 9105       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9106     case T_LONG:
 9107       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9108     default:
 9109       fatal("Unexpected type argument %s", type2name(type)); break;
 9110   }
 9111 }
 9112 
 9113 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9114   switch(type) {
 9115     case T_BYTE:
 9116       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9117     case T_SHORT:
 9118       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9119     case T_INT:
 9120       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9121     case T_LONG:
 9122       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9123     default:
 9124       fatal("Unexpected type argument %s", type2name(type)); break;
 9125   }
 9126 }
 9127 
 9128 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9129   switch(type) {
 9130     case T_BYTE:
 9131       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9132     case T_SHORT:
 9133       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9134     case T_INT:
 9135       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9136     case T_LONG:
 9137       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9138     case T_FLOAT:
 9139       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9140     case T_DOUBLE:
 9141       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9142     default:
 9143       fatal("Unexpected type argument %s", type2name(type)); break;
 9144   }
 9145 }
 9146 
 9147 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9148   switch(type) {
 9149     case T_BYTE:
 9150       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9151     case T_SHORT:
 9152       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9153     case T_INT:
 9154       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9155     case T_LONG:
 9156       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9157     case T_FLOAT:
 9158       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9159     case T_DOUBLE:
 9160       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9161     default:
 9162       fatal("Unexpected type argument %s", type2name(type)); break;
 9163   }
 9164 }
 9165 
 9166 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9167   switch(type) {
 9168     case T_BYTE:
 9169       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9170     case T_SHORT:
 9171       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9172     case T_INT:
 9173       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9174     case T_LONG:
 9175       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9176     case T_FLOAT:
 9177       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9178     case T_DOUBLE:
 9179       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9180     default:
 9181       fatal("Unexpected type argument %s", type2name(type)); break;
 9182   }
 9183 }
 9184 
 9185 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9186   switch(type) {
 9187     case T_BYTE:
 9188       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9189     case T_SHORT:
 9190       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9191     case T_INT:
 9192       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9193     case T_LONG:
 9194       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9195     case T_FLOAT:
 9196       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9197     case T_DOUBLE:
 9198       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9199     default:
 9200       fatal("Unexpected type argument %s", type2name(type)); break;
 9201   }
 9202 }
 9203 
 9204 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9205   switch(type) {
 9206     case T_INT:
 9207       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9208     case T_LONG:
 9209       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9210     default:
 9211       fatal("Unexpected type argument %s", type2name(type)); break;
 9212   }
 9213 }
 9214 
 9215 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9216   switch(type) {
 9217     case T_INT:
 9218       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9219     case T_LONG:
 9220       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9221     default:
 9222       fatal("Unexpected type argument %s", type2name(type)); break;
 9223   }
 9224 }
 9225 
 9226 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9227   switch(type) {
 9228     case T_INT:
 9229       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9230     case T_LONG:
 9231       evporq(dst, mask, nds, src, merge, vector_len); break;
 9232     default:
 9233       fatal("Unexpected type argument %s", type2name(type)); break;
 9234   }
 9235 }
 9236 
 9237 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9238   switch(type) {
 9239     case T_INT:
 9240       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9241     case T_LONG:
 9242       evporq(dst, mask, nds, src, merge, vector_len); break;
 9243     default:
 9244       fatal("Unexpected type argument %s", type2name(type)); break;
 9245   }
 9246 }
 9247 
 9248 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9249   switch(type) {
 9250     case T_INT:
 9251       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9252     case T_LONG:
 9253       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9254     default:
 9255       fatal("Unexpected type argument %s", type2name(type)); break;
 9256   }
 9257 }
 9258 
 9259 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9260   switch(type) {
 9261     case T_INT:
 9262       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9263     case T_LONG:
 9264       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9265     default:
 9266       fatal("Unexpected type argument %s", type2name(type)); break;
 9267   }
 9268 }
 9269 
 9270 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9271   switch(masklen) {
 9272     case 8:
 9273        kortestbl(src1, src2);
 9274        break;
 9275     case 16:
 9276        kortestwl(src1, src2);
 9277        break;
 9278     case 32:
 9279        kortestdl(src1, src2);
 9280        break;
 9281     case 64:
 9282        kortestql(src1, src2);
 9283        break;
 9284     default:
 9285       fatal("Unexpected mask length %d", masklen);
 9286       break;
 9287   }
 9288 }
 9289 
 9290 
 9291 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
 9292   switch(masklen)  {
 9293     case 8:
 9294        ktestbl(src1, src2);
 9295        break;
 9296     case 16:
 9297        ktestwl(src1, src2);
 9298        break;
 9299     case 32:
 9300        ktestdl(src1, src2);
 9301        break;
 9302     case 64:
 9303        ktestql(src1, src2);
 9304        break;
 9305     default:
 9306       fatal("Unexpected mask length %d", masklen);
 9307       break;
 9308   }
 9309 }
 9310 
 9311 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9312   switch(type) {
 9313     case T_INT:
 9314       evprold(dst, mask, src, shift, merge, vlen_enc); break;
 9315     case T_LONG:
 9316       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
 9317     default:
 9318       fatal("Unexpected type argument %s", type2name(type)); break;
 9319       break;
 9320   }
 9321 }
 9322 
 9323 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9324   switch(type) {
 9325     case T_INT:
 9326       evprord(dst, mask, src, shift, merge, vlen_enc); break;
 9327     case T_LONG:
 9328       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
 9329     default:
 9330       fatal("Unexpected type argument %s", type2name(type)); break;
 9331   }
 9332 }
 9333 
 9334 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9335   switch(type) {
 9336     case T_INT:
 9337       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9338     case T_LONG:
 9339       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9340     default:
 9341       fatal("Unexpected type argument %s", type2name(type)); break;
 9342   }
 9343 }
 9344 
 9345 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9346   switch(type) {
 9347     case T_INT:
 9348       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9349     case T_LONG:
 9350       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9351     default:
 9352       fatal("Unexpected type argument %s", type2name(type)); break;
 9353   }
 9354 }
 9355 
 9356 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9357   assert(rscratch != noreg || always_reachable(src), "missing");
 9358 
 9359   if (reachable(src)) {
 9360     evpandq(dst, nds, as_Address(src), vector_len);
 9361   } else {
 9362     lea(rscratch, src);
 9363     evpandq(dst, nds, Address(rscratch, 0), vector_len);
 9364   }
 9365 }
 9366 
 9367 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 9368   assert(rscratch != noreg || always_reachable(src), "missing");
 9369 
 9370   if (reachable(src)) {
 9371     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
 9372   } else {
 9373     lea(rscratch, src);
 9374     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 9375   }
 9376 }
 9377 
 9378 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9379   assert(rscratch != noreg || always_reachable(src), "missing");
 9380 
 9381   if (reachable(src)) {
 9382     evporq(dst, nds, as_Address(src), vector_len);
 9383   } else {
 9384     lea(rscratch, src);
 9385     evporq(dst, nds, Address(rscratch, 0), vector_len);
 9386   }
 9387 }
 9388 
 9389 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9390   assert(rscratch != noreg || always_reachable(src), "missing");
 9391 
 9392   if (reachable(src)) {
 9393     vpshufb(dst, nds, as_Address(src), vector_len);
 9394   } else {
 9395     lea(rscratch, src);
 9396     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
 9397   }
 9398 }
 9399 
 9400 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9401   assert(rscratch != noreg || always_reachable(src), "missing");
 9402 
 9403   if (reachable(src)) {
 9404     Assembler::vpor(dst, nds, as_Address(src), vector_len);
 9405   } else {
 9406     lea(rscratch, src);
 9407     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
 9408   }
 9409 }
 9410 
 9411 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
 9412   assert(rscratch != noreg || always_reachable(src3), "missing");
 9413 
 9414   if (reachable(src3)) {
 9415     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
 9416   } else {
 9417     lea(rscratch, src3);
 9418     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
 9419   }
 9420 }
 9421 
 9422 #if COMPILER2_OR_JVMCI
 9423 
 9424 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
 9425                                  Register length, Register temp, int vec_enc) {
 9426   // Computing mask for predicated vector store.
 9427   movptr(temp, -1);
 9428   bzhiq(temp, temp, length);
 9429   kmov(mask, temp);
 9430   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
 9431 }
 9432 
 9433 // Set memory operation for length "less than" 64 bytes.
 9434 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
 9435                                        XMMRegister xmm, KRegister mask, Register length,
 9436                                        Register temp, bool use64byteVector) {
 9437   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9438   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9439   if (!use64byteVector) {
 9440     fill32(dst, disp, xmm);
 9441     subptr(length, 32 >> shift);
 9442     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
 9443   } else {
 9444     assert(MaxVectorSize == 64, "vector length != 64");
 9445     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
 9446   }
 9447 }
 9448 
 9449 
 9450 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
 9451                                        XMMRegister xmm, KRegister mask, Register length,
 9452                                        Register temp) {
 9453   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9454   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9455   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
 9456 }
 9457 
 9458 
 9459 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
 9460   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9461   vmovdqu(dst, xmm);
 9462 }
 9463 
 9464 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
 9465   fill32(Address(dst, disp), xmm);
 9466 }
 9467 
 9468 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
 9469   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9470   if (!use64byteVector) {
 9471     fill32(dst, xmm);
 9472     fill32(dst.plus_disp(32), xmm);
 9473   } else {
 9474     evmovdquq(dst, xmm, Assembler::AVX_512bit);
 9475   }
 9476 }
 9477 
 9478 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
 9479   fill64(Address(dst, disp), xmm, use64byteVector);
 9480 }
 9481 
 9482 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
 9483                                         Register count, Register rtmp, XMMRegister xtmp) {
 9484   Label L_exit;
 9485   Label L_fill_start;
 9486   Label L_fill_64_bytes;
 9487   Label L_fill_96_bytes;
 9488   Label L_fill_128_bytes;
 9489   Label L_fill_128_bytes_loop;
 9490   Label L_fill_128_loop_header;
 9491   Label L_fill_128_bytes_loop_header;
 9492   Label L_fill_128_bytes_loop_pre_header;
 9493   Label L_fill_zmm_sequence;
 9494 
 9495   int shift = -1;
 9496   switch(type) {
 9497     case T_BYTE:  shift = 0;
 9498       break;
 9499     case T_SHORT: shift = 1;
 9500       break;
 9501     case T_INT:   shift = 2;
 9502       break;
 9503     /* Uncomment when LONG fill stubs are supported.
 9504     case T_LONG:  shift = 3;
 9505       break;
 9506     */
 9507     default:
 9508       fatal("Unhandled type: %s\n", type2name(type));
 9509   }
 9510 
 9511   if ((CopyAVX3Threshold != 0)  || (MaxVectorSize == 32)) {
 9512 
 9513     if (MaxVectorSize == 64) {
 9514       cmpq(count, CopyAVX3Threshold >> shift);
 9515       jcc(Assembler::greater, L_fill_zmm_sequence);
 9516     }
 9517 
 9518     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
 9519 
 9520     bind(L_fill_start);
 9521 
 9522     cmpq(count, 32 >> shift);
 9523     jccb(Assembler::greater, L_fill_64_bytes);
 9524     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9525     jmp(L_exit);
 9526 
 9527     bind(L_fill_64_bytes);
 9528     cmpq(count, 64 >> shift);
 9529     jccb(Assembler::greater, L_fill_96_bytes);
 9530     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9531     jmp(L_exit);
 9532 
 9533     bind(L_fill_96_bytes);
 9534     cmpq(count, 96 >> shift);
 9535     jccb(Assembler::greater, L_fill_128_bytes);
 9536     fill64(to, 0, xtmp);
 9537     subq(count, 64 >> shift);
 9538     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
 9539     jmp(L_exit);
 9540 
 9541     bind(L_fill_128_bytes);
 9542     cmpq(count, 128 >> shift);
 9543     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
 9544     fill64(to, 0, xtmp);
 9545     fill32(to, 64, xtmp);
 9546     subq(count, 96 >> shift);
 9547     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
 9548     jmp(L_exit);
 9549 
 9550     bind(L_fill_128_bytes_loop_pre_header);
 9551     {
 9552       mov(rtmp, to);
 9553       andq(rtmp, 31);
 9554       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
 9555       negq(rtmp);
 9556       addq(rtmp, 32);
 9557       mov64(r8, -1L);
 9558       bzhiq(r8, r8, rtmp);
 9559       kmovql(k2, r8);
 9560       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
 9561       addq(to, rtmp);
 9562       shrq(rtmp, shift);
 9563       subq(count, rtmp);
 9564     }
 9565 
 9566     cmpq(count, 128 >> shift);
 9567     jcc(Assembler::less, L_fill_start);
 9568 
 9569     bind(L_fill_128_bytes_loop_header);
 9570     subq(count, 128 >> shift);
 9571 
 9572     align32();
 9573     bind(L_fill_128_bytes_loop);
 9574       fill64(to, 0, xtmp);
 9575       fill64(to, 64, xtmp);
 9576       addq(to, 128);
 9577       subq(count, 128 >> shift);
 9578       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
 9579 
 9580     addq(count, 128 >> shift);
 9581     jcc(Assembler::zero, L_exit);
 9582     jmp(L_fill_start);
 9583   }
 9584 
 9585   if (MaxVectorSize == 64) {
 9586     // Sequence using 64 byte ZMM register.
 9587     Label L_fill_128_bytes_zmm;
 9588     Label L_fill_192_bytes_zmm;
 9589     Label L_fill_192_bytes_loop_zmm;
 9590     Label L_fill_192_bytes_loop_header_zmm;
 9591     Label L_fill_192_bytes_loop_pre_header_zmm;
 9592     Label L_fill_start_zmm_sequence;
 9593 
 9594     bind(L_fill_zmm_sequence);
 9595     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
 9596 
 9597     bind(L_fill_start_zmm_sequence);
 9598     cmpq(count, 64 >> shift);
 9599     jccb(Assembler::greater, L_fill_128_bytes_zmm);
 9600     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
 9601     jmp(L_exit);
 9602 
 9603     bind(L_fill_128_bytes_zmm);
 9604     cmpq(count, 128 >> shift);
 9605     jccb(Assembler::greater, L_fill_192_bytes_zmm);
 9606     fill64(to, 0, xtmp, true);
 9607     subq(count, 64 >> shift);
 9608     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
 9609     jmp(L_exit);
 9610 
 9611     bind(L_fill_192_bytes_zmm);
 9612     cmpq(count, 192 >> shift);
 9613     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
 9614     fill64(to, 0, xtmp, true);
 9615     fill64(to, 64, xtmp, true);
 9616     subq(count, 128 >> shift);
 9617     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
 9618     jmp(L_exit);
 9619 
 9620     bind(L_fill_192_bytes_loop_pre_header_zmm);
 9621     {
 9622       movq(rtmp, to);
 9623       andq(rtmp, 63);
 9624       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
 9625       negq(rtmp);
 9626       addq(rtmp, 64);
 9627       mov64(r8, -1L);
 9628       bzhiq(r8, r8, rtmp);
 9629       kmovql(k2, r8);
 9630       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
 9631       addq(to, rtmp);
 9632       shrq(rtmp, shift);
 9633       subq(count, rtmp);
 9634     }
 9635 
 9636     cmpq(count, 192 >> shift);
 9637     jcc(Assembler::less, L_fill_start_zmm_sequence);
 9638 
 9639     bind(L_fill_192_bytes_loop_header_zmm);
 9640     subq(count, 192 >> shift);
 9641 
 9642     align32();
 9643     bind(L_fill_192_bytes_loop_zmm);
 9644       fill64(to, 0, xtmp, true);
 9645       fill64(to, 64, xtmp, true);
 9646       fill64(to, 128, xtmp, true);
 9647       addq(to, 192);
 9648       subq(count, 192 >> shift);
 9649       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
 9650 
 9651     addq(count, 192 >> shift);
 9652     jcc(Assembler::zero, L_exit);
 9653     jmp(L_fill_start_zmm_sequence);
 9654   }
 9655   bind(L_exit);
 9656 }
 9657 #endif //COMPILER2_OR_JVMCI
 9658 
 9659 
 9660 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
 9661   Label done;
 9662   cvttss2sil(dst, src);
 9663   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9664   cmpl(dst, 0x80000000); // float_sign_flip
 9665   jccb(Assembler::notEqual, done);
 9666   subptr(rsp, 8);
 9667   movflt(Address(rsp, 0), src);
 9668   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
 9669   pop(dst);
 9670   bind(done);
 9671 }
 9672 
 9673 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
 9674   Label done;
 9675   cvttsd2sil(dst, src);
 9676   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9677   cmpl(dst, 0x80000000); // float_sign_flip
 9678   jccb(Assembler::notEqual, done);
 9679   subptr(rsp, 8);
 9680   movdbl(Address(rsp, 0), src);
 9681   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
 9682   pop(dst);
 9683   bind(done);
 9684 }
 9685 
 9686 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
 9687   Label done;
 9688   cvttss2siq(dst, src);
 9689   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_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::f2l_fixup())));
 9694   pop(dst);
 9695   bind(done);
 9696 }
 9697 
 9698 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9699   // Following code is line by line assembly translation rounding algorithm.
 9700   // Please refer to java.lang.Math.round(float) algorithm for details.
 9701   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
 9702   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
 9703   const int32_t FloatConsts_EXP_BIAS = 127;
 9704   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
 9705   const int32_t MINUS_32 = 0xFFFFFFE0;
 9706   Label L_special_case, L_block1, L_exit;
 9707   movl(rtmp, FloatConsts_EXP_BIT_MASK);
 9708   movdl(dst, src);
 9709   andl(dst, rtmp);
 9710   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
 9711   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
 9712   subl(rtmp, dst);
 9713   movl(rcx, rtmp);
 9714   movl(dst, MINUS_32);
 9715   testl(rtmp, dst);
 9716   jccb(Assembler::notEqual, L_special_case);
 9717   movdl(dst, src);
 9718   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
 9719   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
 9720   movdl(rtmp, src);
 9721   testl(rtmp, rtmp);
 9722   jccb(Assembler::greaterEqual, L_block1);
 9723   negl(dst);
 9724   bind(L_block1);
 9725   sarl(dst);
 9726   addl(dst, 0x1);
 9727   sarl(dst, 0x1);
 9728   jmp(L_exit);
 9729   bind(L_special_case);
 9730   convert_f2i(dst, src);
 9731   bind(L_exit);
 9732 }
 9733 
 9734 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9735   // Following code is line by line assembly translation rounding algorithm.
 9736   // Please refer to java.lang.Math.round(double) algorithm for details.
 9737   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
 9738   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
 9739   const int64_t DoubleConsts_EXP_BIAS = 1023;
 9740   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
 9741   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
 9742   Label L_special_case, L_block1, L_exit;
 9743   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
 9744   movq(dst, src);
 9745   andq(dst, rtmp);
 9746   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
 9747   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
 9748   subq(rtmp, dst);
 9749   movq(rcx, rtmp);
 9750   mov64(dst, MINUS_64);
 9751   testq(rtmp, dst);
 9752   jccb(Assembler::notEqual, L_special_case);
 9753   movq(dst, src);
 9754   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
 9755   andq(dst, rtmp);
 9756   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
 9757   orq(dst, rtmp);
 9758   movq(rtmp, src);
 9759   testq(rtmp, rtmp);
 9760   jccb(Assembler::greaterEqual, L_block1);
 9761   negq(dst);
 9762   bind(L_block1);
 9763   sarq(dst);
 9764   addq(dst, 0x1);
 9765   sarq(dst, 0x1);
 9766   jmp(L_exit);
 9767   bind(L_special_case);
 9768   convert_d2l(dst, src);
 9769   bind(L_exit);
 9770 }
 9771 
 9772 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
 9773   Label done;
 9774   cvttsd2siq(dst, src);
 9775   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9776   jccb(Assembler::notEqual, done);
 9777   subptr(rsp, 8);
 9778   movdbl(Address(rsp, 0), src);
 9779   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
 9780   pop(dst);
 9781   bind(done);
 9782 }
 9783 
 9784 void MacroAssembler::cache_wb(Address line)
 9785 {
 9786   // 64 bit cpus always support clflush
 9787   assert(VM_Version::supports_clflush(), "clflush should be available");
 9788   bool optimized = VM_Version::supports_clflushopt();
 9789   bool no_evict = VM_Version::supports_clwb();
 9790 
 9791   // prefer clwb (writeback without evict) otherwise
 9792   // prefer clflushopt (potentially parallel writeback with evict)
 9793   // otherwise fallback on clflush (serial writeback with evict)
 9794 
 9795   if (optimized) {
 9796     if (no_evict) {
 9797       clwb(line);
 9798     } else {
 9799       clflushopt(line);
 9800     }
 9801   } else {
 9802     // no need for fence when using CLFLUSH
 9803     clflush(line);
 9804   }
 9805 }
 9806 
 9807 void MacroAssembler::cache_wbsync(bool is_pre)
 9808 {
 9809   assert(VM_Version::supports_clflush(), "clflush should be available");
 9810   bool optimized = VM_Version::supports_clflushopt();
 9811   bool no_evict = VM_Version::supports_clwb();
 9812 
 9813   // pick the correct implementation
 9814 
 9815   if (!is_pre && (optimized || no_evict)) {
 9816     // need an sfence for post flush when using clflushopt or clwb
 9817     // otherwise no no need for any synchroniaztion
 9818 
 9819     sfence();
 9820   }
 9821 }
 9822 
 9823 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
 9824   switch (cond) {
 9825     // Note some conditions are synonyms for others
 9826     case Assembler::zero:         return Assembler::notZero;
 9827     case Assembler::notZero:      return Assembler::zero;
 9828     case Assembler::less:         return Assembler::greaterEqual;
 9829     case Assembler::lessEqual:    return Assembler::greater;
 9830     case Assembler::greater:      return Assembler::lessEqual;
 9831     case Assembler::greaterEqual: return Assembler::less;
 9832     case Assembler::below:        return Assembler::aboveEqual;
 9833     case Assembler::belowEqual:   return Assembler::above;
 9834     case Assembler::above:        return Assembler::belowEqual;
 9835     case Assembler::aboveEqual:   return Assembler::below;
 9836     case Assembler::overflow:     return Assembler::noOverflow;
 9837     case Assembler::noOverflow:   return Assembler::overflow;
 9838     case Assembler::negative:     return Assembler::positive;
 9839     case Assembler::positive:     return Assembler::negative;
 9840     case Assembler::parity:       return Assembler::noParity;
 9841     case Assembler::noParity:     return Assembler::parity;
 9842   }
 9843   ShouldNotReachHere(); return Assembler::overflow;
 9844 }
 9845 
 9846 // This is simply a call to Thread::current()
 9847 void MacroAssembler::get_thread_slow(Register thread) {
 9848   if (thread != rax) {
 9849     push(rax);
 9850   }
 9851   push(rdi);
 9852   push(rsi);
 9853   push(rdx);
 9854   push(rcx);
 9855   push(r8);
 9856   push(r9);
 9857   push(r10);
 9858   push(r11);
 9859 
 9860   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
 9861 
 9862   pop(r11);
 9863   pop(r10);
 9864   pop(r9);
 9865   pop(r8);
 9866   pop(rcx);
 9867   pop(rdx);
 9868   pop(rsi);
 9869   pop(rdi);
 9870   if (thread != rax) {
 9871     mov(thread, rax);
 9872     pop(rax);
 9873   }
 9874 }
 9875 
 9876 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
 9877   Label L_stack_ok;
 9878   if (bias == 0) {
 9879     testptr(sp, 2 * wordSize - 1);
 9880   } else {
 9881     // lea(tmp, Address(rsp, bias);
 9882     mov(tmp, sp);
 9883     addptr(tmp, bias);
 9884     testptr(tmp, 2 * wordSize - 1);
 9885   }
 9886   jcc(Assembler::equal, L_stack_ok);
 9887   block_comment(msg);
 9888   stop(msg);
 9889   bind(L_stack_ok);
 9890 }
 9891 
 9892 // Implements fast-locking.
 9893 //
 9894 // obj: the object to be locked
 9895 // reg_rax: rax
 9896 // thread: the thread which attempts to lock obj
 9897 // tmp: a temporary register
 9898 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
 9899   Register thread = r15_thread;
 9900 
 9901   assert(reg_rax == rax, "");
 9902   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
 9903 
 9904   Label push;
 9905   const Register top = tmp;
 9906 
 9907   // Preload the markWord. It is important that this is the first
 9908   // instruction emitted as it is part of C1's null check semantics.
 9909   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9910 
 9911   if (UseObjectMonitorTable) {
 9912     // Clear cache in case fast locking succeeds or we need to take the slow-path.
 9913     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
 9914   }
 9915 
 9916   if (DiagnoseSyncOnValueBasedClasses != 0) {
 9917     load_klass(tmp, obj, rscratch1);
 9918     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
 9919     jcc(Assembler::notZero, slow);
 9920   }
 9921 
 9922   // Load top.
 9923   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9924 
 9925   // Check if the lock-stack is full.
 9926   cmpl(top, LockStack::end_offset());
 9927   jcc(Assembler::greaterEqual, slow);
 9928 
 9929   // Check for recursion.
 9930   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9931   jcc(Assembler::equal, push);
 9932 
 9933   // Check header for monitor (0b10).
 9934   testptr(reg_rax, markWord::monitor_value);
 9935   jcc(Assembler::notZero, slow);
 9936 
 9937   // Try to lock. Transition lock bits 0b01 => 0b00
 9938   movptr(tmp, reg_rax);
 9939   andptr(tmp, ~(int32_t)markWord::unlocked_value);
 9940   orptr(reg_rax, markWord::unlocked_value);
 9941   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9942   jcc(Assembler::notEqual, slow);
 9943 
 9944   // Restore top, CAS clobbers register.
 9945   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9946 
 9947   bind(push);
 9948   // After successful lock, push object on lock-stack.
 9949   movptr(Address(thread, top), obj);
 9950   incrementl(top, oopSize);
 9951   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
 9952 }
 9953 
 9954 // Implements fast-unlocking.
 9955 //
 9956 // obj: the object to be unlocked
 9957 // reg_rax: rax
 9958 // thread: the thread
 9959 // tmp: a temporary register
 9960 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
 9961   Register thread = r15_thread;
 9962 
 9963   assert(reg_rax == rax, "");
 9964   assert_different_registers(obj, reg_rax, thread, tmp);
 9965 
 9966   Label unlocked, push_and_slow;
 9967   const Register top = tmp;
 9968 
 9969   // Check if obj is top of lock-stack.
 9970   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9971   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9972   jcc(Assembler::notEqual, slow);
 9973 
 9974   // Pop lock-stack.
 9975   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
 9976   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
 9977 
 9978   // Check if recursive.
 9979   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
 9980   jcc(Assembler::equal, unlocked);
 9981 
 9982   // Not recursive. Check header for monitor (0b10).
 9983   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9984   testptr(reg_rax, markWord::monitor_value);
 9985   jcc(Assembler::notZero, push_and_slow);
 9986 
 9987 #ifdef ASSERT
 9988   // Check header not unlocked (0b01).
 9989   Label not_unlocked;
 9990   testptr(reg_rax, markWord::unlocked_value);
 9991   jcc(Assembler::zero, not_unlocked);
 9992   stop("fast_unlock already unlocked");
 9993   bind(not_unlocked);
 9994 #endif
 9995 
 9996   // Try to unlock. Transition lock bits 0b00 => 0b01
 9997   movptr(tmp, reg_rax);
 9998   orptr(tmp, markWord::unlocked_value);
 9999   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10000   jcc(Assembler::equal, unlocked);
10001 
10002   bind(push_and_slow);
10003   // Restore lock-stack and handle the unlock in runtime.
10004 #ifdef ASSERT
10005   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10006   movptr(Address(thread, top), obj);
10007 #endif
10008   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10009   jmp(slow);
10010 
10011   bind(unlocked);
10012 }
10013 
10014 // Saves legacy GPRs state on stack.
10015 void MacroAssembler::save_legacy_gprs() {
10016   subq(rsp, 16 * wordSize);
10017   movq(Address(rsp, 15 * wordSize), rax);
10018   movq(Address(rsp, 14 * wordSize), rcx);
10019   movq(Address(rsp, 13 * wordSize), rdx);
10020   movq(Address(rsp, 12 * wordSize), rbx);
10021   movq(Address(rsp, 10 * wordSize), rbp);
10022   movq(Address(rsp, 9 * wordSize), rsi);
10023   movq(Address(rsp, 8 * wordSize), rdi);
10024   movq(Address(rsp, 7 * wordSize), r8);
10025   movq(Address(rsp, 6 * wordSize), r9);
10026   movq(Address(rsp, 5 * wordSize), r10);
10027   movq(Address(rsp, 4 * wordSize), r11);
10028   movq(Address(rsp, 3 * wordSize), r12);
10029   movq(Address(rsp, 2 * wordSize), r13);
10030   movq(Address(rsp, wordSize), r14);
10031   movq(Address(rsp, 0), r15);
10032 }
10033 
10034 // Resotres back legacy GPRs state from stack.
10035 void MacroAssembler::restore_legacy_gprs() {
10036   movq(r15, Address(rsp, 0));
10037   movq(r14, Address(rsp, wordSize));
10038   movq(r13, Address(rsp, 2 * wordSize));
10039   movq(r12, Address(rsp, 3 * wordSize));
10040   movq(r11, Address(rsp, 4 * wordSize));
10041   movq(r10, Address(rsp, 5 * wordSize));
10042   movq(r9,  Address(rsp, 6 * wordSize));
10043   movq(r8,  Address(rsp, 7 * wordSize));
10044   movq(rdi, Address(rsp, 8 * wordSize));
10045   movq(rsi, Address(rsp, 9 * wordSize));
10046   movq(rbp, Address(rsp, 10 * wordSize));
10047   movq(rbx, Address(rsp, 12 * wordSize));
10048   movq(rdx, Address(rsp, 13 * wordSize));
10049   movq(rcx, Address(rsp, 14 * wordSize));
10050   movq(rax, Address(rsp, 15 * wordSize));
10051   addq(rsp, 16 * wordSize);
10052 }
10053 
10054 void MacroAssembler::load_aotrc_address(Register reg, address a) {
10055 #if INCLUDE_CDS
10056   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
10057   if (AOTCodeCache::is_on_for_dump()) {
10058     // all aotrc field addresses should be registered in the AOTCodeCache address table
10059     lea(reg, ExternalAddress(a));
10060   } else {
10061     mov64(reg, (uint64_t)a);
10062   }
10063 #else
10064   ShouldNotReachHere();
10065 #endif
10066 }
10067 
10068 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10069   if (VM_Version::supports_apx_f()) {
10070     esetzucc(comparison, dst);
10071   } else {
10072     setb(comparison, dst);
10073     movzbl(dst, dst);
10074   }
10075 }