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 "ci/ciInlineKlass.hpp"
   32 #include "crc32c.h"
   33 #include "gc/shared/barrierSet.hpp"
   34 #include "gc/shared/barrierSetAssembler.hpp"
   35 #include "gc/shared/collectedHeap.inline.hpp"
   36 #include "gc/shared/tlab_globals.hpp"
   37 #include "interpreter/bytecodeHistogram.hpp"
   38 #include "interpreter/interpreter.hpp"
   39 #include "interpreter/interpreterRuntime.hpp"
   40 #include "jvm.h"
   41 #include "memory/resourceArea.hpp"
   42 #include "memory/universe.hpp"
   43 #include "oops/accessDecorators.hpp"
   44 #include "oops/compressedKlass.inline.hpp"
   45 #include "oops/compressedOops.inline.hpp"
   46 #include "oops/klass.inline.hpp"
   47 #include "oops/resolvedFieldEntry.hpp"
   48 #include "prims/methodHandles.hpp"
   49 #include "runtime/arguments.hpp"
   50 #include "runtime/continuation.hpp"
   51 #include "runtime/interfaceSupport.inline.hpp"
   52 #include "runtime/javaThread.hpp"
   53 #include "runtime/jniHandles.hpp"
   54 #include "runtime/objectMonitor.hpp"
   55 #include "runtime/os.hpp"
   56 #include "runtime/safepoint.hpp"
   57 #include "runtime/safepointMechanism.hpp"
   58 #include "runtime/sharedRuntime.hpp"
   59 #include "runtime/signature_cc.hpp"
   60 #include "runtime/stubRoutines.hpp"
   61 #include "utilities/checkedCast.hpp"
   62 #include "utilities/globalDefinitions.hpp"
   63 #include "utilities/macros.hpp"
   64 #include "vmreg_x86.inline.hpp"
   65 #ifdef COMPILER2
   66 #include "opto/output.hpp"
   67 #endif
   68 
   69 #ifdef PRODUCT
   70 #define BLOCK_COMMENT(str) /* nothing */
   71 #define STOP(error) stop(error)
   72 #else
   73 #define BLOCK_COMMENT(str) block_comment(str)
   74 #define STOP(error) block_comment(error); stop(error)
   75 #endif
   76 
   77 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   78 
   79 #ifdef ASSERT
   80 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   81 #endif
   82 
   83 static const Assembler::Condition reverse[] = {
   84     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   85     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   86     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   87     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
   88     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
   89     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
   90     Assembler::above          /* belowEqual    = 0x6 */ ,
   91     Assembler::belowEqual     /* above         = 0x7 */ ,
   92     Assembler::positive       /* negative      = 0x8 */ ,
   93     Assembler::negative       /* positive      = 0x9 */ ,
   94     Assembler::noParity       /* parity        = 0xa */ ,
   95     Assembler::parity         /* noParity      = 0xb */ ,
   96     Assembler::greaterEqual   /* less          = 0xc */ ,
   97     Assembler::less           /* greaterEqual  = 0xd */ ,
   98     Assembler::greater        /* lessEqual     = 0xe */ ,
   99     Assembler::lessEqual      /* greater       = 0xf, */
  100 
  101 };
  102 
  103 
  104 // Implementation of MacroAssembler
  105 
  106 Address MacroAssembler::as_Address(AddressLiteral adr) {
  107   // amd64 always does this as a pc-rel
  108   // we can be absolute or disp based on the instruction type
  109   // jmp/call are displacements others are absolute
  110   assert(!adr.is_lval(), "must be rval");
  111   assert(reachable(adr), "must be");
  112   return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
  113 
  114 }
  115 
  116 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  117   AddressLiteral base = adr.base();
  118   lea(rscratch, base);
  119   Address index = adr.index();
  120   assert(index._disp == 0, "must not have disp"); // maybe it can?
  121   Address array(rscratch, index._index, index._scale, index._disp);
  122   return array;
  123 }
  124 
  125 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
  126   Label L, E;
  127 
  128 #ifdef _WIN64
  129   // Windows always allocates space for it's register args
  130   assert(num_args <= 4, "only register arguments supported");
  131   subq(rsp,  frame::arg_reg_save_area_bytes);
  132 #endif
  133 
  134   // Align stack if necessary
  135   testl(rsp, 15);
  136   jcc(Assembler::zero, L);
  137 
  138   subq(rsp, 8);
  139   call(RuntimeAddress(entry_point));
  140   addq(rsp, 8);
  141   jmp(E);
  142 
  143   bind(L);
  144   call(RuntimeAddress(entry_point));
  145 
  146   bind(E);
  147 
  148 #ifdef _WIN64
  149   // restore stack pointer
  150   addq(rsp, frame::arg_reg_save_area_bytes);
  151 #endif
  152 }
  153 
  154 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
  155   assert(!src2.is_lval(), "should use cmpptr");
  156   assert(rscratch != noreg || always_reachable(src2), "missing");
  157 
  158   if (reachable(src2)) {
  159     cmpq(src1, as_Address(src2));
  160   } else {
  161     lea(rscratch, src2);
  162     Assembler::cmpq(src1, Address(rscratch, 0));
  163   }
  164 }
  165 
  166 int MacroAssembler::corrected_idivq(Register reg) {
  167   // Full implementation of Java ldiv and lrem; checks for special
  168   // case as described in JVM spec., p.243 & p.271.  The function
  169   // returns the (pc) offset of the idivl instruction - may be needed
  170   // for implicit exceptions.
  171   //
  172   //         normal case                           special case
  173   //
  174   // input : rax: dividend                         min_long
  175   //         reg: divisor   (may not be eax/edx)   -1
  176   //
  177   // output: rax: quotient  (= rax idiv reg)       min_long
  178   //         rdx: remainder (= rax irem reg)       0
  179   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
  180   static const int64_t min_long = 0x8000000000000000;
  181   Label normal_case, special_case;
  182 
  183   // check for special case
  184   cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
  185   jcc(Assembler::notEqual, normal_case);
  186   xorl(rdx, rdx); // prepare rdx for possible special case (where
  187                   // remainder = 0)
  188   cmpq(reg, -1);
  189   jcc(Assembler::equal, special_case);
  190 
  191   // handle normal case
  192   bind(normal_case);
  193   cdqq();
  194   int idivq_offset = offset();
  195   idivq(reg);
  196 
  197   // normal and special case exit
  198   bind(special_case);
  199 
  200   return idivq_offset;
  201 }
  202 
  203 void MacroAssembler::decrementq(Register reg, int value) {
  204   if (value == min_jint) { subq(reg, value); return; }
  205   if (value <  0) { incrementq(reg, -value); return; }
  206   if (value == 0) {                        ; return; }
  207   if (value == 1 && UseIncDec) { decq(reg) ; return; }
  208   /* else */      { subq(reg, value)       ; return; }
  209 }
  210 
  211 void MacroAssembler::decrementq(Address dst, int value) {
  212   if (value == min_jint) { subq(dst, value); return; }
  213   if (value <  0) { incrementq(dst, -value); return; }
  214   if (value == 0) {                        ; return; }
  215   if (value == 1 && UseIncDec) { decq(dst) ; return; }
  216   /* else */      { subq(dst, value)       ; return; }
  217 }
  218 
  219 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
  220   assert(rscratch != noreg || always_reachable(dst), "missing");
  221 
  222   if (reachable(dst)) {
  223     incrementq(as_Address(dst));
  224   } else {
  225     lea(rscratch, dst);
  226     incrementq(Address(rscratch, 0));
  227   }
  228 }
  229 
  230 void MacroAssembler::incrementq(Register reg, int value) {
  231   if (value == min_jint) { addq(reg, value); return; }
  232   if (value <  0) { decrementq(reg, -value); return; }
  233   if (value == 0) {                        ; return; }
  234   if (value == 1 && UseIncDec) { incq(reg) ; return; }
  235   /* else */      { addq(reg, value)       ; return; }
  236 }
  237 
  238 void MacroAssembler::incrementq(Address dst, int value) {
  239   if (value == min_jint) { addq(dst, value); return; }
  240   if (value <  0) { decrementq(dst, -value); return; }
  241   if (value == 0) {                        ; return; }
  242   if (value == 1 && UseIncDec) { incq(dst) ; return; }
  243   /* else */      { addq(dst, value)       ; return; }
  244 }
  245 
  246 // 32bit can do a case table jump in one instruction but we no longer allow the base
  247 // to be installed in the Address class
  248 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  249   lea(rscratch, entry.base());
  250   Address dispatch = entry.index();
  251   assert(dispatch._base == noreg, "must be");
  252   dispatch._base = rscratch;
  253   jmp(dispatch);
  254 }
  255 
  256 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  257   ShouldNotReachHere(); // 64bit doesn't use two regs
  258   cmpq(x_lo, y_lo);
  259 }
  260 
  261 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  262   mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  263 }
  264 
  265 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  266   lea(rscratch, adr);
  267   movptr(dst, rscratch);
  268 }
  269 
  270 void MacroAssembler::leave() {
  271   // %%% is this really better? Why not on 32bit too?
  272   emit_int8((unsigned char)0xC9); // LEAVE
  273 }
  274 
  275 void MacroAssembler::lneg(Register hi, Register lo) {
  276   ShouldNotReachHere(); // 64bit doesn't use two regs
  277   negq(lo);
  278 }
  279 
  280 void MacroAssembler::movoop(Register dst, jobject obj) {
  281   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  282 }
  283 
  284 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  285   mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  286   movq(dst, rscratch);
  287 }
  288 
  289 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  290   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  291 }
  292 
  293 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  294   mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  295   movq(dst, rscratch);
  296 }
  297 
  298 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  299   if (src.is_lval()) {
  300     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  301   } else {
  302     if (reachable(src)) {
  303       movq(dst, as_Address(src));
  304     } else {
  305       lea(dst, src);
  306       movq(dst, Address(dst, 0));
  307     }
  308   }
  309 }
  310 
  311 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  312   movq(as_Address(dst, rscratch), src);
  313 }
  314 
  315 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  316   movq(dst, as_Address(src, dst /*rscratch*/));
  317 }
  318 
  319 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
  320 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  321   if (is_simm32(src)) {
  322     movptr(dst, checked_cast<int32_t>(src));
  323   } else {
  324     mov64(rscratch, src);
  325     movq(dst, rscratch);
  326   }
  327 }
  328 
  329 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  330   movoop(rscratch, obj);
  331   push(rscratch);
  332 }
  333 
  334 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  335   mov_metadata(rscratch, obj);
  336   push(rscratch);
  337 }
  338 
  339 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  340   lea(rscratch, src);
  341   if (src.is_lval()) {
  342     push(rscratch);
  343   } else {
  344     pushq(Address(rscratch, 0));
  345   }
  346 }
  347 
  348 static void pass_arg0(MacroAssembler* masm, Register arg) {
  349   if (c_rarg0 != arg ) {
  350     masm->mov(c_rarg0, arg);
  351   }
  352 }
  353 
  354 static void pass_arg1(MacroAssembler* masm, Register arg) {
  355   if (c_rarg1 != arg ) {
  356     masm->mov(c_rarg1, arg);
  357   }
  358 }
  359 
  360 static void pass_arg2(MacroAssembler* masm, Register arg) {
  361   if (c_rarg2 != arg ) {
  362     masm->mov(c_rarg2, arg);
  363   }
  364 }
  365 
  366 static void pass_arg3(MacroAssembler* masm, Register arg) {
  367   if (c_rarg3 != arg ) {
  368     masm->mov(c_rarg3, arg);
  369   }
  370 }
  371 
  372 void MacroAssembler::stop(const char* msg) {
  373   if (ShowMessageBoxOnError) {
  374     address rip = pc();
  375     pusha(); // get regs on stack
  376     lea(c_rarg1, InternalAddress(rip));
  377     movq(c_rarg2, rsp); // pass pointer to regs array
  378   }
  379   // Skip AOT caching C strings in scratch buffer.
  380   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  381   lea(c_rarg0, ExternalAddress((address) str));
  382   andq(rsp, -16); // align stack as required by ABI
  383   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
  384   hlt();
  385 }
  386 
  387 void MacroAssembler::warn(const char* msg) {
  388   push(rbp);
  389   movq(rbp, rsp);
  390   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  391   push_CPU_state();   // keeps alignment at 16 bytes
  392 
  393 #ifdef _WIN64
  394   // Windows always allocates space for its register args
  395   subq(rsp,  frame::arg_reg_save_area_bytes);
  396 #endif
  397   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  398   lea(c_rarg0, ExternalAddress((address) str));
  399   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  400 
  401 #ifdef _WIN64
  402   // restore stack pointer
  403   addq(rsp, frame::arg_reg_save_area_bytes);
  404 #endif
  405   pop_CPU_state();
  406   mov(rsp, rbp);
  407   pop(rbp);
  408 }
  409 
  410 void MacroAssembler::print_state() {
  411   address rip = pc();
  412   pusha();            // get regs on stack
  413   push(rbp);
  414   movq(rbp, rsp);
  415   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  416   push_CPU_state();   // keeps alignment at 16 bytes
  417 
  418   lea(c_rarg0, InternalAddress(rip));
  419   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  420   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  421 
  422   pop_CPU_state();
  423   mov(rsp, rbp);
  424   pop(rbp);
  425   popa();
  426 }
  427 
  428 #ifndef PRODUCT
  429 extern "C" void findpc(intptr_t x);
  430 #endif
  431 
  432 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  433   // In order to get locks to work, we need to fake a in_VM state
  434   if (ShowMessageBoxOnError) {
  435     JavaThread* thread = JavaThread::current();
  436     JavaThreadState saved_state = thread->thread_state();
  437     thread->set_thread_state(_thread_in_vm);
  438 #ifndef PRODUCT
  439     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  440       ttyLocker ttyl;
  441       BytecodeCounter::print();
  442     }
  443 #endif
  444     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  445     // XXX correct this offset for amd64
  446     // This is the value of eip which points to where verify_oop will return.
  447     if (os::message_box(msg, "Execution stopped, print registers?")) {
  448       print_state64(pc, regs);
  449       BREAKPOINT;
  450     }
  451   }
  452   fatal("DEBUG MESSAGE: %s", msg);
  453 }
  454 
  455 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  456   ttyLocker ttyl;
  457   DebuggingContext debugging{};
  458   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  459 #ifndef PRODUCT
  460   tty->cr();
  461   findpc(pc);
  462   tty->cr();
  463 #endif
  464 #define PRINT_REG(rax, value) \
  465   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  466   PRINT_REG(rax, regs[15]);
  467   PRINT_REG(rbx, regs[12]);
  468   PRINT_REG(rcx, regs[14]);
  469   PRINT_REG(rdx, regs[13]);
  470   PRINT_REG(rdi, regs[8]);
  471   PRINT_REG(rsi, regs[9]);
  472   PRINT_REG(rbp, regs[10]);
  473   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  474   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  475   PRINT_REG(r8 , regs[7]);
  476   PRINT_REG(r9 , regs[6]);
  477   PRINT_REG(r10, regs[5]);
  478   PRINT_REG(r11, regs[4]);
  479   PRINT_REG(r12, regs[3]);
  480   PRINT_REG(r13, regs[2]);
  481   PRINT_REG(r14, regs[1]);
  482   PRINT_REG(r15, regs[0]);
  483 #undef PRINT_REG
  484   // Print some words near the top of the stack.
  485   int64_t* rsp = &regs[16];
  486   int64_t* dump_sp = rsp;
  487   for (int col1 = 0; col1 < 8; col1++) {
  488     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  489     os::print_location(tty, *dump_sp++);
  490   }
  491   for (int row = 0; row < 25; row++) {
  492     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  493     for (int col = 0; col < 4; col++) {
  494       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  495     }
  496     tty->cr();
  497   }
  498   // Print some instructions around pc:
  499   Disassembler::decode((address)pc-64, (address)pc);
  500   tty->print_cr("--------");
  501   Disassembler::decode((address)pc, (address)pc+32);
  502 }
  503 
  504 // The java_calling_convention describes stack locations as ideal slots on
  505 // a frame with no abi restrictions. Since we must observe abi restrictions
  506 // (like the placement of the register window) the slots must be biased by
  507 // the following value.
  508 static int reg2offset_in(VMReg r) {
  509   // Account for saved rbp and return address
  510   // This should really be in_preserve_stack_slots
  511   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  512 }
  513 
  514 static int reg2offset_out(VMReg r) {
  515   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  516 }
  517 
  518 // A long move
  519 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  520 
  521   // The calling conventions assures us that each VMregpair is either
  522   // all really one physical register or adjacent stack slots.
  523 
  524   if (src.is_single_phys_reg() ) {
  525     if (dst.is_single_phys_reg()) {
  526       if (dst.first() != src.first()) {
  527         mov(dst.first()->as_Register(), src.first()->as_Register());
  528       }
  529     } else {
  530       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  531              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  532       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  533     }
  534   } else if (dst.is_single_phys_reg()) {
  535     assert(src.is_single_reg(),  "not a stack pair");
  536     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  537   } else {
  538     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  539     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  540     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  541   }
  542 }
  543 
  544 // A double move
  545 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  546 
  547   // The calling conventions assures us that each VMregpair is either
  548   // all really one physical register or adjacent stack slots.
  549 
  550   if (src.is_single_phys_reg() ) {
  551     if (dst.is_single_phys_reg()) {
  552       // In theory these overlap but the ordering is such that this is likely a nop
  553       if ( src.first() != dst.first()) {
  554         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  555       }
  556     } else {
  557       assert(dst.is_single_reg(), "not a stack pair");
  558       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  559     }
  560   } else if (dst.is_single_phys_reg()) {
  561     assert(src.is_single_reg(),  "not a stack pair");
  562     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  563   } else {
  564     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  565     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  566     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  567   }
  568 }
  569 
  570 
  571 // A float arg may have to do float reg int reg conversion
  572 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  573   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  574 
  575   // The calling conventions assures us that each VMregpair is either
  576   // all really one physical register or adjacent stack slots.
  577 
  578   if (src.first()->is_stack()) {
  579     if (dst.first()->is_stack()) {
  580       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  581       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  582     } else {
  583       // stack to reg
  584       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  585       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  586     }
  587   } else if (dst.first()->is_stack()) {
  588     // reg to stack
  589     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  590     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  591   } else {
  592     // reg to reg
  593     // In theory these overlap but the ordering is such that this is likely a nop
  594     if ( src.first() != dst.first()) {
  595       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  596     }
  597   }
  598 }
  599 
  600 // On 64 bit we will store integer like items to the stack as
  601 // 64 bits items (x86_32/64 abi) even though java would only store
  602 // 32bits for a parameter. On 32bit it will simply be 32 bits
  603 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  604 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  605   if (src.first()->is_stack()) {
  606     if (dst.first()->is_stack()) {
  607       // stack to stack
  608       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  609       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  610     } else {
  611       // stack to reg
  612       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  613     }
  614   } else if (dst.first()->is_stack()) {
  615     // reg to stack
  616     // Do we really have to sign extend???
  617     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  618     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  619   } else {
  620     // Do we really have to sign extend???
  621     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
  622     if (dst.first() != src.first()) {
  623       movq(dst.first()->as_Register(), src.first()->as_Register());
  624     }
  625   }
  626 }
  627 
  628 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
  629   if (src.first()->is_stack()) {
  630     if (dst.first()->is_stack()) {
  631       // stack to stack
  632       movq(rax, Address(rbp, reg2offset_in(src.first())));
  633       movq(Address(rsp, reg2offset_out(dst.first())), rax);
  634     } else {
  635       // stack to reg
  636       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
  637     }
  638   } else if (dst.first()->is_stack()) {
  639     // reg to stack
  640     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
  641   } else {
  642     if (dst.first() != src.first()) {
  643       movq(dst.first()->as_Register(), src.first()->as_Register());
  644     }
  645   }
  646 }
  647 
  648 // An oop arg. Must pass a handle not the oop itself
  649 void MacroAssembler::object_move(OopMap* map,
  650                         int oop_handle_offset,
  651                         int framesize_in_slots,
  652                         VMRegPair src,
  653                         VMRegPair dst,
  654                         bool is_receiver,
  655                         int* receiver_offset) {
  656 
  657   // must pass a handle. First figure out the location we use as a handle
  658 
  659   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
  660 
  661   // See if oop is null if it is we need no handle
  662 
  663   if (src.first()->is_stack()) {
  664 
  665     // Oop is already on the stack as an argument
  666     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
  667     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
  668     if (is_receiver) {
  669       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
  670     }
  671 
  672     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
  673     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
  674     // conditionally move a null
  675     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
  676   } else {
  677 
  678     // Oop is in a register we must store it to the space we reserve
  679     // on the stack for oop_handles and pass a handle if oop is non-null
  680 
  681     const Register rOop = src.first()->as_Register();
  682     int oop_slot;
  683     if (rOop == j_rarg0)
  684       oop_slot = 0;
  685     else if (rOop == j_rarg1)
  686       oop_slot = 1;
  687     else if (rOop == j_rarg2)
  688       oop_slot = 2;
  689     else if (rOop == j_rarg3)
  690       oop_slot = 3;
  691     else if (rOop == j_rarg4)
  692       oop_slot = 4;
  693     else {
  694       assert(rOop == j_rarg5, "wrong register");
  695       oop_slot = 5;
  696     }
  697 
  698     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
  699     int offset = oop_slot*VMRegImpl::stack_slot_size;
  700 
  701     map->set_oop(VMRegImpl::stack2reg(oop_slot));
  702     // Store oop in handle area, may be null
  703     movptr(Address(rsp, offset), rOop);
  704     if (is_receiver) {
  705       *receiver_offset = offset;
  706     }
  707 
  708     cmpptr(rOop, NULL_WORD);
  709     lea(rHandle, Address(rsp, offset));
  710     // conditionally move a null from the handle area where it was just stored
  711     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
  712   }
  713 
  714   // If arg is on the stack then place it otherwise it is already in correct reg.
  715   if (dst.first()->is_stack()) {
  716     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
  717   }
  718 }
  719 
  720 void MacroAssembler::addptr(Register dst, int32_t imm32) {
  721   addq(dst, imm32);
  722 }
  723 
  724 void MacroAssembler::addptr(Register dst, Register src) {
  725   addq(dst, src);
  726 }
  727 
  728 void MacroAssembler::addptr(Address dst, Register src) {
  729   addq(dst, src);
  730 }
  731 
  732 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  733   assert(rscratch != noreg || always_reachable(src), "missing");
  734 
  735   if (reachable(src)) {
  736     Assembler::addsd(dst, as_Address(src));
  737   } else {
  738     lea(rscratch, src);
  739     Assembler::addsd(dst, Address(rscratch, 0));
  740   }
  741 }
  742 
  743 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
  744   assert(rscratch != noreg || always_reachable(src), "missing");
  745 
  746   if (reachable(src)) {
  747     addss(dst, as_Address(src));
  748   } else {
  749     lea(rscratch, src);
  750     addss(dst, Address(rscratch, 0));
  751   }
  752 }
  753 
  754 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  755   assert(rscratch != noreg || always_reachable(src), "missing");
  756 
  757   if (reachable(src)) {
  758     Assembler::addpd(dst, as_Address(src));
  759   } else {
  760     lea(rscratch, src);
  761     Assembler::addpd(dst, Address(rscratch, 0));
  762   }
  763 }
  764 
  765 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
  766 // Stub code is generated once and never copied.
  767 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
  768 void MacroAssembler::align64() {
  769   align(64, (uint)(uintptr_t)pc());
  770 }
  771 
  772 void MacroAssembler::align32() {
  773   align(32, (uint)(uintptr_t)pc());
  774 }
  775 
  776 void MacroAssembler::align(uint modulus) {
  777   // 8273459: Ensure alignment is possible with current segment alignment
  778   assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
  779   align(modulus, offset());
  780 }
  781 
  782 void MacroAssembler::align(uint modulus, uint target) {
  783   if (target % modulus != 0) {
  784     nop(modulus - (target % modulus));
  785   }
  786 }
  787 
  788 void MacroAssembler::push_f(XMMRegister r) {
  789   subptr(rsp, wordSize);
  790   movflt(Address(rsp, 0), r);
  791 }
  792 
  793 void MacroAssembler::pop_f(XMMRegister r) {
  794   movflt(r, Address(rsp, 0));
  795   addptr(rsp, wordSize);
  796 }
  797 
  798 void MacroAssembler::push_d(XMMRegister r) {
  799   subptr(rsp, 2 * wordSize);
  800   movdbl(Address(rsp, 0), r);
  801 }
  802 
  803 void MacroAssembler::pop_d(XMMRegister r) {
  804   movdbl(r, Address(rsp, 0));
  805   addptr(rsp, 2 * Interpreter::stackElementSize);
  806 }
  807 
  808 void MacroAssembler::push_ppx(Register src) {
  809   if (VM_Version::supports_apx_f()) {
  810     pushp(src);
  811   } else {
  812     Assembler::push(src);
  813   }
  814 }
  815 
  816 void MacroAssembler::pop_ppx(Register dst) {
  817   if (VM_Version::supports_apx_f()) {
  818     popp(dst);
  819   } else {
  820     Assembler::pop(dst);
  821   }
  822 }
  823 
  824 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  825   // Used in sign-masking with aligned address.
  826   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  827   assert(rscratch != noreg || always_reachable(src), "missing");
  828 
  829   if (UseAVX > 2 &&
  830       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
  831       (dst->encoding() >= 16)) {
  832     vpand(dst, dst, src, AVX_512bit, rscratch);
  833   } else if (reachable(src)) {
  834     Assembler::andpd(dst, as_Address(src));
  835   } else {
  836     lea(rscratch, src);
  837     Assembler::andpd(dst, Address(rscratch, 0));
  838   }
  839 }
  840 
  841 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
  842   // Used in sign-masking with aligned address.
  843   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  844   assert(rscratch != noreg || always_reachable(src), "missing");
  845 
  846   if (reachable(src)) {
  847     Assembler::andps(dst, as_Address(src));
  848   } else {
  849     lea(rscratch, src);
  850     Assembler::andps(dst, Address(rscratch, 0));
  851   }
  852 }
  853 
  854 void MacroAssembler::andptr(Register dst, int32_t imm32) {
  855   andq(dst, imm32);
  856 }
  857 
  858 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
  859   assert(rscratch != noreg || always_reachable(src), "missing");
  860 
  861   if (reachable(src)) {
  862     andq(dst, as_Address(src));
  863   } else {
  864     lea(rscratch, src);
  865     andq(dst, Address(rscratch, 0));
  866   }
  867 }
  868 
  869 void MacroAssembler::atomic_incl(Address counter_addr) {
  870   lock();
  871   incrementl(counter_addr);
  872 }
  873 
  874 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
  875   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  876 
  877   if (reachable(counter_addr)) {
  878     atomic_incl(as_Address(counter_addr));
  879   } else {
  880     lea(rscratch, counter_addr);
  881     atomic_incl(Address(rscratch, 0));
  882   }
  883 }
  884 
  885 void MacroAssembler::atomic_incq(Address counter_addr) {
  886   lock();
  887   incrementq(counter_addr);
  888 }
  889 
  890 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
  891   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  892 
  893   if (reachable(counter_addr)) {
  894     atomic_incq(as_Address(counter_addr));
  895   } else {
  896     lea(rscratch, counter_addr);
  897     atomic_incq(Address(rscratch, 0));
  898   }
  899 }
  900 
  901 // Writes to stack successive pages until offset reached to check for
  902 // stack overflow + shadow pages.  This clobbers tmp.
  903 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
  904   movptr(tmp, rsp);
  905   // Bang stack for total size given plus shadow page size.
  906   // Bang one page at a time because large size can bang beyond yellow and
  907   // red zones.
  908   Label loop;
  909   bind(loop);
  910   movl(Address(tmp, (-(int)os::vm_page_size())), size );
  911   subptr(tmp, (int)os::vm_page_size());
  912   subl(size, (int)os::vm_page_size());
  913   jcc(Assembler::greater, loop);
  914 
  915   // Bang down shadow pages too.
  916   // At this point, (tmp-0) is the last address touched, so don't
  917   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
  918   // was post-decremented.)  Skip this address by starting at i=1, and
  919   // touch a few more pages below.  N.B.  It is important to touch all
  920   // the way down including all pages in the shadow zone.
  921   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
  922     // this could be any sized move but this is can be a debugging crumb
  923     // so the bigger the better.
  924     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
  925   }
  926 }
  927 
  928 void MacroAssembler::reserved_stack_check() {
  929   // testing if reserved zone needs to be enabled
  930   Label no_reserved_zone_enabling;
  931 
  932   cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
  933   jcc(Assembler::below, no_reserved_zone_enabling);
  934 
  935   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
  936   jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
  937   should_not_reach_here();
  938 
  939   bind(no_reserved_zone_enabling);
  940 }
  941 
  942 void MacroAssembler::c2bool(Register x) {
  943   // implements x == 0 ? 0 : 1
  944   // note: must only look at least-significant byte of x
  945   //       since C-style booleans are stored in one byte
  946   //       only! (was bug)
  947   andl(x, 0xFF);
  948   setb(Assembler::notZero, x);
  949 }
  950 
  951 // Wouldn't need if AddressLiteral version had new name
  952 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
  953   Assembler::call(L, rtype);
  954 }
  955 
  956 void MacroAssembler::call(Register entry) {
  957   Assembler::call(entry);
  958 }
  959 
  960 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
  961   assert(rscratch != noreg || always_reachable(entry), "missing");
  962 
  963   if (reachable(entry)) {
  964     Assembler::call_literal(entry.target(), entry.rspec());
  965   } else {
  966     lea(rscratch, entry);
  967     Assembler::call(rscratch);
  968   }
  969 }
  970 
  971 void MacroAssembler::ic_call(address entry, jint method_index) {
  972   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
  973   // Needs full 64-bit immediate for later patching.
  974   Assembler::mov64(rax, (int64_t)Universe::non_oop_word());
  975   call(AddressLiteral(entry, rh));
  976 }
  977 
  978 int MacroAssembler::ic_check_size() {
  979   return UseCompactObjectHeaders ? 17 : 14;
  980 }
  981 
  982 int MacroAssembler::ic_check(int end_alignment) {
  983   Register receiver = j_rarg0;
  984   Register data = rax;
  985   Register temp = rscratch1;
  986 
  987   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
  988   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
  989   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
  990   // before the inline cache check here, and not after
  991   align(end_alignment, offset() + ic_check_size());
  992 
  993   int uep_offset = offset();
  994 
  995   if (UseCompactObjectHeaders) {
  996     load_narrow_klass_compact(temp, receiver);
  997     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  998   } else {
  999     movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
 1000     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
 1001   }
 1002 
 1003   // if inline cache check fails, then jump to runtime routine
 1004   jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 1005   assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
 1006 
 1007   return uep_offset;
 1008 }
 1009 
 1010 void MacroAssembler::emit_static_call_stub() {
 1011   // Static stub relocation also tags the Method* in the code-stream.
 1012   mov_metadata(rbx, (Metadata*) nullptr);  // Method is zapped till fixup time.
 1013   // This is recognized as unresolved by relocs/nativeinst/ic code.
 1014   jump(RuntimeAddress(pc()));
 1015 }
 1016 
 1017 // Implementation of call_VM versions
 1018 
 1019 void MacroAssembler::call_VM(Register oop_result,
 1020                              address entry_point,
 1021                              bool check_exceptions) {
 1022   Label C, E;
 1023   call(C, relocInfo::none);
 1024   jmp(E);
 1025 
 1026   bind(C);
 1027   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
 1028   ret(0);
 1029 
 1030   bind(E);
 1031 }
 1032 
 1033 void MacroAssembler::call_VM(Register oop_result,
 1034                              address entry_point,
 1035                              Register arg_1,
 1036                              bool check_exceptions) {
 1037   Label C, E;
 1038   call(C, relocInfo::none);
 1039   jmp(E);
 1040 
 1041   bind(C);
 1042   pass_arg1(this, arg_1);
 1043   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
 1044   ret(0);
 1045 
 1046   bind(E);
 1047 }
 1048 
 1049 void MacroAssembler::call_VM(Register oop_result,
 1050                              address entry_point,
 1051                              Register arg_1,
 1052                              Register arg_2,
 1053                              bool check_exceptions) {
 1054   Label C, E;
 1055   call(C, relocInfo::none);
 1056   jmp(E);
 1057 
 1058   bind(C);
 1059 
 1060   assert_different_registers(arg_1, c_rarg2);
 1061 
 1062   pass_arg2(this, arg_2);
 1063   pass_arg1(this, arg_1);
 1064   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
 1065   ret(0);
 1066 
 1067   bind(E);
 1068 }
 1069 
 1070 void MacroAssembler::call_VM(Register oop_result,
 1071                              address entry_point,
 1072                              Register arg_1,
 1073                              Register arg_2,
 1074                              Register arg_3,
 1075                              bool check_exceptions) {
 1076   Label C, E;
 1077   call(C, relocInfo::none);
 1078   jmp(E);
 1079 
 1080   bind(C);
 1081 
 1082   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1083   assert_different_registers(arg_2, c_rarg3);
 1084   pass_arg3(this, arg_3);
 1085   pass_arg2(this, arg_2);
 1086   pass_arg1(this, arg_1);
 1087   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
 1088   ret(0);
 1089 
 1090   bind(E);
 1091 }
 1092 
 1093 void MacroAssembler::call_VM(Register oop_result,
 1094                              Register last_java_sp,
 1095                              address entry_point,
 1096                              int number_of_arguments,
 1097                              bool check_exceptions) {
 1098   call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1099 }
 1100 
 1101 void MacroAssembler::call_VM(Register oop_result,
 1102                              Register last_java_sp,
 1103                              address entry_point,
 1104                              Register arg_1,
 1105                              bool check_exceptions) {
 1106   pass_arg1(this, arg_1);
 1107   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1108 }
 1109 
 1110 void MacroAssembler::call_VM(Register oop_result,
 1111                              Register last_java_sp,
 1112                              address entry_point,
 1113                              Register arg_1,
 1114                              Register arg_2,
 1115                              bool check_exceptions) {
 1116 
 1117   assert_different_registers(arg_1, c_rarg2);
 1118   pass_arg2(this, arg_2);
 1119   pass_arg1(this, arg_1);
 1120   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1121 }
 1122 
 1123 void MacroAssembler::call_VM(Register oop_result,
 1124                              Register last_java_sp,
 1125                              address entry_point,
 1126                              Register arg_1,
 1127                              Register arg_2,
 1128                              Register arg_3,
 1129                              bool check_exceptions) {
 1130   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1131   assert_different_registers(arg_2, c_rarg3);
 1132   pass_arg3(this, arg_3);
 1133   pass_arg2(this, arg_2);
 1134   pass_arg1(this, arg_1);
 1135   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1136 }
 1137 
 1138 void MacroAssembler::super_call_VM(Register oop_result,
 1139                                    Register last_java_sp,
 1140                                    address entry_point,
 1141                                    int number_of_arguments,
 1142                                    bool check_exceptions) {
 1143   MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1144 }
 1145 
 1146 void MacroAssembler::super_call_VM(Register oop_result,
 1147                                    Register last_java_sp,
 1148                                    address entry_point,
 1149                                    Register arg_1,
 1150                                    bool check_exceptions) {
 1151   pass_arg1(this, arg_1);
 1152   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1153 }
 1154 
 1155 void MacroAssembler::super_call_VM(Register oop_result,
 1156                                    Register last_java_sp,
 1157                                    address entry_point,
 1158                                    Register arg_1,
 1159                                    Register arg_2,
 1160                                    bool check_exceptions) {
 1161 
 1162   assert_different_registers(arg_1, c_rarg2);
 1163   pass_arg2(this, arg_2);
 1164   pass_arg1(this, arg_1);
 1165   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1166 }
 1167 
 1168 void MacroAssembler::super_call_VM(Register oop_result,
 1169                                    Register last_java_sp,
 1170                                    address entry_point,
 1171                                    Register arg_1,
 1172                                    Register arg_2,
 1173                                    Register arg_3,
 1174                                    bool check_exceptions) {
 1175   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1176   assert_different_registers(arg_2, c_rarg3);
 1177   pass_arg3(this, arg_3);
 1178   pass_arg2(this, arg_2);
 1179   pass_arg1(this, arg_1);
 1180   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1181 }
 1182 
 1183 void MacroAssembler::call_VM_base(Register oop_result,
 1184                                   Register last_java_sp,
 1185                                   address  entry_point,
 1186                                   int      number_of_arguments,
 1187                                   bool     check_exceptions) {
 1188   Register java_thread = r15_thread;
 1189 
 1190   // determine last_java_sp register
 1191   if (!last_java_sp->is_valid()) {
 1192     last_java_sp = rsp;
 1193   }
 1194   // debugging support
 1195   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 1196 #ifdef ASSERT
 1197   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 1198   // r12 is the heapbase.
 1199   if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 1200 #endif // ASSERT
 1201 
 1202   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 1203   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 1204 
 1205   // push java thread (becomes first argument of C function)
 1206 
 1207   mov(c_rarg0, r15_thread);
 1208 
 1209   // set last Java frame before call
 1210   assert(last_java_sp != rbp, "can't use ebp/rbp");
 1211 
 1212   // Only interpreter should have to set fp
 1213   set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
 1214 
 1215   // do the call, remove parameters
 1216   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 1217 
 1218 #ifdef ASSERT
 1219   // Check that thread register is not clobbered.
 1220   guarantee(java_thread != rax, "change this code");
 1221   push(rax);
 1222   { Label L;
 1223     get_thread_slow(rax);
 1224     cmpptr(java_thread, rax);
 1225     jcc(Assembler::equal, L);
 1226     STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
 1227     bind(L);
 1228   }
 1229   pop(rax);
 1230 #endif
 1231 
 1232   // reset last Java frame
 1233   // Only interpreter should have to clear fp
 1234   reset_last_Java_frame(true);
 1235 
 1236    // C++ interp handles this in the interpreter
 1237   check_and_handle_popframe();
 1238   check_and_handle_earlyret();
 1239 
 1240   if (check_exceptions) {
 1241     // check for pending exceptions (java_thread is set upon return)
 1242     cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 1243     // This used to conditionally jump to forward_exception however it is
 1244     // possible if we relocate that the branch will not reach. So we must jump
 1245     // around so we can always reach
 1246 
 1247     Label ok;
 1248     jcc(Assembler::equal, ok);
 1249     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 1250     bind(ok);
 1251   }
 1252 
 1253   // get oop result if there is one and reset the value in the thread
 1254   if (oop_result->is_valid()) {
 1255     get_vm_result_oop(oop_result);
 1256   }
 1257 }
 1258 
 1259 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 1260   // Calculate the value for last_Java_sp somewhat subtle.
 1261   // call_VM does an intermediate call which places a return address on
 1262   // the stack just under the stack pointer as the user finished with it.
 1263   // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
 1264 
 1265   // We've pushed one address, correct last_Java_sp
 1266   lea(rax, Address(rsp, wordSize));
 1267 
 1268   call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
 1269 }
 1270 
 1271 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
 1272 void MacroAssembler::call_VM_leaf0(address entry_point) {
 1273   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 1274 }
 1275 
 1276 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
 1277   call_VM_leaf_base(entry_point, number_of_arguments);
 1278 }
 1279 
 1280 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
 1281   pass_arg0(this, arg_0);
 1282   call_VM_leaf(entry_point, 1);
 1283 }
 1284 
 1285 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1286 
 1287   assert_different_registers(arg_0, c_rarg1);
 1288   pass_arg1(this, arg_1);
 1289   pass_arg0(this, arg_0);
 1290   call_VM_leaf(entry_point, 2);
 1291 }
 1292 
 1293 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1294   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1295   assert_different_registers(arg_1, c_rarg2);
 1296   pass_arg2(this, arg_2);
 1297   pass_arg1(this, arg_1);
 1298   pass_arg0(this, arg_0);
 1299   call_VM_leaf(entry_point, 3);
 1300 }
 1301 
 1302 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1303   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1304   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1305   assert_different_registers(arg_2, c_rarg3);
 1306   pass_arg3(this, arg_3);
 1307   pass_arg2(this, arg_2);
 1308   pass_arg1(this, arg_1);
 1309   pass_arg0(this, arg_0);
 1310   call_VM_leaf(entry_point, 3);
 1311 }
 1312 
 1313 void MacroAssembler::super_call_VM_leaf(address entry_point) {
 1314   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1315 }
 1316 
 1317 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1318   pass_arg0(this, arg_0);
 1319   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1320 }
 1321 
 1322 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1323   assert_different_registers(arg_0, c_rarg1);
 1324   pass_arg1(this, arg_1);
 1325   pass_arg0(this, arg_0);
 1326   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1327 }
 1328 
 1329 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1330   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1331   assert_different_registers(arg_1, c_rarg2);
 1332   pass_arg2(this, arg_2);
 1333   pass_arg1(this, arg_1);
 1334   pass_arg0(this, arg_0);
 1335   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1336 }
 1337 
 1338 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1339   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1340   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1341   assert_different_registers(arg_2, c_rarg3);
 1342   pass_arg3(this, arg_3);
 1343   pass_arg2(this, arg_2);
 1344   pass_arg1(this, arg_1);
 1345   pass_arg0(this, arg_0);
 1346   MacroAssembler::call_VM_leaf_base(entry_point, 4);
 1347 }
 1348 
 1349 void MacroAssembler::get_vm_result_oop(Register oop_result) {
 1350   movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
 1351   movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
 1352   verify_oop_msg(oop_result, "broken oop in call_VM_base");
 1353 }
 1354 
 1355 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
 1356   movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
 1357   movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
 1358 }
 1359 
 1360 void MacroAssembler::check_and_handle_earlyret() {
 1361 }
 1362 
 1363 void MacroAssembler::check_and_handle_popframe() {
 1364 }
 1365 
 1366 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
 1367   assert(rscratch != noreg || always_reachable(src1), "missing");
 1368 
 1369   if (reachable(src1)) {
 1370     cmpl(as_Address(src1), imm);
 1371   } else {
 1372     lea(rscratch, src1);
 1373     cmpl(Address(rscratch, 0), imm);
 1374   }
 1375 }
 1376 
 1377 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
 1378   assert(!src2.is_lval(), "use cmpptr");
 1379   assert(rscratch != noreg || always_reachable(src2), "missing");
 1380 
 1381   if (reachable(src2)) {
 1382     cmpl(src1, as_Address(src2));
 1383   } else {
 1384     lea(rscratch, src2);
 1385     cmpl(src1, Address(rscratch, 0));
 1386   }
 1387 }
 1388 
 1389 void MacroAssembler::cmp32(Register src1, int32_t imm) {
 1390   Assembler::cmpl(src1, imm);
 1391 }
 1392 
 1393 void MacroAssembler::cmp32(Register src1, Address src2) {
 1394   Assembler::cmpl(src1, src2);
 1395 }
 1396 
 1397 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1398   ucomisd(opr1, opr2);
 1399 
 1400   Label L;
 1401   if (unordered_is_less) {
 1402     movl(dst, -1);
 1403     jcc(Assembler::parity, L);
 1404     jcc(Assembler::below , L);
 1405     movl(dst, 0);
 1406     jcc(Assembler::equal , L);
 1407     increment(dst);
 1408   } else { // unordered is greater
 1409     movl(dst, 1);
 1410     jcc(Assembler::parity, L);
 1411     jcc(Assembler::above , L);
 1412     movl(dst, 0);
 1413     jcc(Assembler::equal , L);
 1414     decrementl(dst);
 1415   }
 1416   bind(L);
 1417 }
 1418 
 1419 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1420   ucomiss(opr1, opr2);
 1421 
 1422   Label L;
 1423   if (unordered_is_less) {
 1424     movl(dst, -1);
 1425     jcc(Assembler::parity, L);
 1426     jcc(Assembler::below , L);
 1427     movl(dst, 0);
 1428     jcc(Assembler::equal , L);
 1429     increment(dst);
 1430   } else { // unordered is greater
 1431     movl(dst, 1);
 1432     jcc(Assembler::parity, L);
 1433     jcc(Assembler::above , L);
 1434     movl(dst, 0);
 1435     jcc(Assembler::equal , L);
 1436     decrementl(dst);
 1437   }
 1438   bind(L);
 1439 }
 1440 
 1441 
 1442 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
 1443   assert(rscratch != noreg || always_reachable(src1), "missing");
 1444 
 1445   if (reachable(src1)) {
 1446     cmpb(as_Address(src1), imm);
 1447   } else {
 1448     lea(rscratch, src1);
 1449     cmpb(Address(rscratch, 0), imm);
 1450   }
 1451 }
 1452 
 1453 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
 1454   assert(rscratch != noreg || always_reachable(src2), "missing");
 1455 
 1456   if (src2.is_lval()) {
 1457     movptr(rscratch, src2);
 1458     Assembler::cmpq(src1, rscratch);
 1459   } else if (reachable(src2)) {
 1460     cmpq(src1, as_Address(src2));
 1461   } else {
 1462     lea(rscratch, src2);
 1463     Assembler::cmpq(src1, Address(rscratch, 0));
 1464   }
 1465 }
 1466 
 1467 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
 1468   assert(src2.is_lval(), "not a mem-mem compare");
 1469   // moves src2's literal address
 1470   movptr(rscratch, src2);
 1471   Assembler::cmpq(src1, rscratch);
 1472 }
 1473 
 1474 void MacroAssembler::cmpoop(Register src1, Register src2) {
 1475   cmpptr(src1, src2);
 1476 }
 1477 
 1478 void MacroAssembler::cmpoop(Register src1, Address src2) {
 1479   cmpptr(src1, src2);
 1480 }
 1481 
 1482 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
 1483   movoop(rscratch, src2);
 1484   cmpptr(src1, rscratch);
 1485 }
 1486 
 1487 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
 1488   assert(rscratch != noreg || always_reachable(adr), "missing");
 1489 
 1490   if (reachable(adr)) {
 1491     lock();
 1492     cmpxchgptr(reg, as_Address(adr));
 1493   } else {
 1494     lea(rscratch, adr);
 1495     lock();
 1496     cmpxchgptr(reg, Address(rscratch, 0));
 1497   }
 1498 }
 1499 
 1500 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
 1501   cmpxchgq(reg, adr);
 1502 }
 1503 
 1504 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1505   assert(rscratch != noreg || always_reachable(src), "missing");
 1506 
 1507   if (reachable(src)) {
 1508     Assembler::comisd(dst, as_Address(src));
 1509   } else {
 1510     lea(rscratch, src);
 1511     Assembler::comisd(dst, Address(rscratch, 0));
 1512   }
 1513 }
 1514 
 1515 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1516   assert(rscratch != noreg || always_reachable(src), "missing");
 1517 
 1518   if (reachable(src)) {
 1519     Assembler::comiss(dst, as_Address(src));
 1520   } else {
 1521     lea(rscratch, src);
 1522     Assembler::comiss(dst, Address(rscratch, 0));
 1523   }
 1524 }
 1525 
 1526 
 1527 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
 1528   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1529 
 1530   Condition negated_cond = negate_condition(cond);
 1531   Label L;
 1532   jcc(negated_cond, L);
 1533   pushf(); // Preserve flags
 1534   atomic_incl(counter_addr, rscratch);
 1535   popf();
 1536   bind(L);
 1537 }
 1538 
 1539 int MacroAssembler::corrected_idivl(Register reg) {
 1540   // Full implementation of Java idiv and irem; checks for
 1541   // special case as described in JVM spec., p.243 & p.271.
 1542   // The function returns the (pc) offset of the idivl
 1543   // instruction - may be needed for implicit exceptions.
 1544   //
 1545   //         normal case                           special case
 1546   //
 1547   // input : rax,: dividend                         min_int
 1548   //         reg: divisor   (may not be rax,/rdx)   -1
 1549   //
 1550   // output: rax,: quotient  (= rax, idiv reg)       min_int
 1551   //         rdx: remainder (= rax, irem reg)       0
 1552   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
 1553   const int min_int = 0x80000000;
 1554   Label normal_case, special_case;
 1555 
 1556   // check for special case
 1557   cmpl(rax, min_int);
 1558   jcc(Assembler::notEqual, normal_case);
 1559   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
 1560   cmpl(reg, -1);
 1561   jcc(Assembler::equal, special_case);
 1562 
 1563   // handle normal case
 1564   bind(normal_case);
 1565   cdql();
 1566   int idivl_offset = offset();
 1567   idivl(reg);
 1568 
 1569   // normal and special case exit
 1570   bind(special_case);
 1571 
 1572   return idivl_offset;
 1573 }
 1574 
 1575 
 1576 
 1577 void MacroAssembler::decrementl(Register reg, int value) {
 1578   if (value == min_jint) {subl(reg, value) ; return; }
 1579   if (value <  0) { incrementl(reg, -value); return; }
 1580   if (value == 0) {                        ; return; }
 1581   if (value == 1 && UseIncDec) { decl(reg) ; return; }
 1582   /* else */      { subl(reg, value)       ; return; }
 1583 }
 1584 
 1585 void MacroAssembler::decrementl(Address dst, int value) {
 1586   if (value == min_jint) {subl(dst, value) ; return; }
 1587   if (value <  0) { incrementl(dst, -value); return; }
 1588   if (value == 0) {                        ; return; }
 1589   if (value == 1 && UseIncDec) { decl(dst) ; return; }
 1590   /* else */      { subl(dst, value)       ; return; }
 1591 }
 1592 
 1593 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
 1594   assert(shift_value > 0, "illegal shift value");
 1595   Label _is_positive;
 1596   testl (reg, reg);
 1597   jcc (Assembler::positive, _is_positive);
 1598   int offset = (1 << shift_value) - 1 ;
 1599 
 1600   if (offset == 1) {
 1601     incrementl(reg);
 1602   } else {
 1603     addl(reg, offset);
 1604   }
 1605 
 1606   bind (_is_positive);
 1607   sarl(reg, shift_value);
 1608 }
 1609 
 1610 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1611   assert(rscratch != noreg || always_reachable(src), "missing");
 1612 
 1613   if (reachable(src)) {
 1614     Assembler::divsd(dst, as_Address(src));
 1615   } else {
 1616     lea(rscratch, src);
 1617     Assembler::divsd(dst, Address(rscratch, 0));
 1618   }
 1619 }
 1620 
 1621 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1622   assert(rscratch != noreg || always_reachable(src), "missing");
 1623 
 1624   if (reachable(src)) {
 1625     Assembler::divss(dst, as_Address(src));
 1626   } else {
 1627     lea(rscratch, src);
 1628     Assembler::divss(dst, Address(rscratch, 0));
 1629   }
 1630 }
 1631 
 1632 void MacroAssembler::enter() {
 1633   push(rbp);
 1634   mov(rbp, rsp);
 1635 }
 1636 
 1637 void MacroAssembler::post_call_nop() {
 1638   if (!Continuations::enabled()) {
 1639     return;
 1640   }
 1641   InstructionMark im(this);
 1642   relocate(post_call_nop_Relocation::spec());
 1643   InlineSkippedInstructionsCounter skipCounter(this);
 1644   emit_int8((uint8_t)0x0f);
 1645   emit_int8((uint8_t)0x1f);
 1646   emit_int8((uint8_t)0x84);
 1647   emit_int8((uint8_t)0x00);
 1648   emit_int32(0x00);
 1649 }
 1650 
 1651 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1652   assert(rscratch != noreg || always_reachable(src), "missing");
 1653   if (reachable(src)) {
 1654     Assembler::mulpd(dst, as_Address(src));
 1655   } else {
 1656     lea(rscratch, src);
 1657     Assembler::mulpd(dst, Address(rscratch, 0));
 1658   }
 1659 }
 1660 
 1661 // dst = c = a * b + c
 1662 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1663   Assembler::vfmadd231sd(c, a, b);
 1664   if (dst != c) {
 1665     movdbl(dst, c);
 1666   }
 1667 }
 1668 
 1669 // dst = c = a * b + c
 1670 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1671   Assembler::vfmadd231ss(c, a, b);
 1672   if (dst != c) {
 1673     movflt(dst, c);
 1674   }
 1675 }
 1676 
 1677 // dst = c = a * b + c
 1678 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1679   Assembler::vfmadd231pd(c, a, b, vector_len);
 1680   if (dst != c) {
 1681     vmovdqu(dst, c);
 1682   }
 1683 }
 1684 
 1685 // dst = c = a * b + c
 1686 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1687   Assembler::vfmadd231ps(c, a, b, vector_len);
 1688   if (dst != c) {
 1689     vmovdqu(dst, c);
 1690   }
 1691 }
 1692 
 1693 // dst = c = a * b + c
 1694 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1695   Assembler::vfmadd231pd(c, a, b, vector_len);
 1696   if (dst != c) {
 1697     vmovdqu(dst, c);
 1698   }
 1699 }
 1700 
 1701 // dst = c = a * b + c
 1702 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1703   Assembler::vfmadd231ps(c, a, b, vector_len);
 1704   if (dst != c) {
 1705     vmovdqu(dst, c);
 1706   }
 1707 }
 1708 
 1709 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
 1710   assert(rscratch != noreg || always_reachable(dst), "missing");
 1711 
 1712   if (reachable(dst)) {
 1713     incrementl(as_Address(dst));
 1714   } else {
 1715     lea(rscratch, dst);
 1716     incrementl(Address(rscratch, 0));
 1717   }
 1718 }
 1719 
 1720 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
 1721   incrementl(as_Address(dst, rscratch));
 1722 }
 1723 
 1724 void MacroAssembler::incrementl(Register reg, int value) {
 1725   if (value == min_jint) {addl(reg, value) ; return; }
 1726   if (value <  0) { decrementl(reg, -value); return; }
 1727   if (value == 0) {                        ; return; }
 1728   if (value == 1 && UseIncDec) { incl(reg) ; return; }
 1729   /* else */      { addl(reg, value)       ; return; }
 1730 }
 1731 
 1732 void MacroAssembler::incrementl(Address dst, int value) {
 1733   if (value == min_jint) {addl(dst, value) ; return; }
 1734   if (value <  0) { decrementl(dst, -value); return; }
 1735   if (value == 0) {                        ; return; }
 1736   if (value == 1 && UseIncDec) { incl(dst) ; return; }
 1737   /* else */      { addl(dst, value)       ; return; }
 1738 }
 1739 
 1740 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
 1741   assert(rscratch != noreg || always_reachable(dst), "missing");
 1742   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
 1743   if (reachable(dst)) {
 1744     jmp_literal(dst.target(), dst.rspec());
 1745   } else {
 1746     lea(rscratch, dst);
 1747     jmp(rscratch);
 1748   }
 1749 }
 1750 
 1751 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
 1752   assert(rscratch != noreg || always_reachable(dst), "missing");
 1753   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump_cc");
 1754   if (reachable(dst)) {
 1755     InstructionMark im(this);
 1756     relocate(dst.reloc());
 1757     const int short_size = 2;
 1758     const int long_size = 6;
 1759     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
 1760     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
 1761       // 0111 tttn #8-bit disp
 1762       emit_int8(0x70 | cc);
 1763       emit_int8((offs - short_size) & 0xFF);
 1764     } else {
 1765       // 0000 1111 1000 tttn #32-bit disp
 1766       emit_int8(0x0F);
 1767       emit_int8((unsigned char)(0x80 | cc));
 1768       emit_int32(offs - long_size);
 1769     }
 1770   } else {
 1771 #ifdef ASSERT
 1772     warning("reversing conditional branch");
 1773 #endif /* ASSERT */
 1774     Label skip;
 1775     jccb(reverse[cc], skip);
 1776     lea(rscratch, dst);
 1777     Assembler::jmp(rscratch);
 1778     bind(skip);
 1779   }
 1780 }
 1781 
 1782 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
 1783   ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 1784   assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
 1785 
 1786   stmxcsr(mxcsr_save);
 1787   movl(tmp, mxcsr_save);
 1788   if (EnableX86ECoreOpts) {
 1789     // The mxcsr_std has status bits set for performance on ECore
 1790     orl(tmp, 0x003f);
 1791   } else {
 1792     // Mask out status bits (only check control and mask bits)
 1793     andl(tmp, 0xFFC0);
 1794   }
 1795   cmp32(tmp, mxcsr_std, rscratch);
 1796 }
 1797 
 1798 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
 1799   assert(rscratch != noreg || always_reachable(src), "missing");
 1800 
 1801   if (reachable(src)) {
 1802     Assembler::ldmxcsr(as_Address(src));
 1803   } else {
 1804     lea(rscratch, src);
 1805     Assembler::ldmxcsr(Address(rscratch, 0));
 1806   }
 1807 }
 1808 
 1809 int MacroAssembler::load_signed_byte(Register dst, Address src) {
 1810   int off = offset();
 1811   movsbl(dst, src); // movsxb
 1812   return off;
 1813 }
 1814 
 1815 // Note: load_signed_short used to be called load_signed_word.
 1816 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
 1817 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
 1818 // The term "word" in HotSpot means a 32- or 64-bit machine word.
 1819 int MacroAssembler::load_signed_short(Register dst, Address src) {
 1820   // This is dubious to me since it seems safe to do a signed 16 => 64 bit
 1821   // version but this is what 64bit has always done. This seems to imply
 1822   // that users are only using 32bits worth.
 1823   int off = offset();
 1824   movswl(dst, src); // movsxw
 1825   return off;
 1826 }
 1827 
 1828 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
 1829   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1830   // and "3.9 Partial Register Penalties", p. 22).
 1831   int off = offset();
 1832   movzbl(dst, src); // movzxb
 1833   return off;
 1834 }
 1835 
 1836 // Note: load_unsigned_short used to be called load_unsigned_word.
 1837 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
 1838   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1839   // and "3.9 Partial Register Penalties", p. 22).
 1840   int off = offset();
 1841   movzwl(dst, src); // movzxw
 1842   return off;
 1843 }
 1844 
 1845 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
 1846   switch (size_in_bytes) {
 1847   case  8:  movq(dst, src); break;
 1848   case  4:  movl(dst, src); break;
 1849   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
 1850   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
 1851   default:  ShouldNotReachHere();
 1852   }
 1853 }
 1854 
 1855 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
 1856   switch (size_in_bytes) {
 1857   case  8:  movq(dst, src); break;
 1858   case  4:  movl(dst, src); break;
 1859   case  2:  movw(dst, src); break;
 1860   case  1:  movb(dst, src); break;
 1861   default:  ShouldNotReachHere();
 1862   }
 1863 }
 1864 
 1865 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
 1866   assert(rscratch != noreg || always_reachable(dst), "missing");
 1867 
 1868   if (reachable(dst)) {
 1869     movl(as_Address(dst), src);
 1870   } else {
 1871     lea(rscratch, dst);
 1872     movl(Address(rscratch, 0), src);
 1873   }
 1874 }
 1875 
 1876 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
 1877   if (reachable(src)) {
 1878     movl(dst, as_Address(src));
 1879   } else {
 1880     lea(dst, src);
 1881     movl(dst, Address(dst, 0));
 1882   }
 1883 }
 1884 
 1885 // C++ bool manipulation
 1886 
 1887 void MacroAssembler::movbool(Register dst, Address src) {
 1888   if(sizeof(bool) == 1)
 1889     movb(dst, src);
 1890   else if(sizeof(bool) == 2)
 1891     movw(dst, src);
 1892   else if(sizeof(bool) == 4)
 1893     movl(dst, src);
 1894   else
 1895     // unsupported
 1896     ShouldNotReachHere();
 1897 }
 1898 
 1899 void MacroAssembler::movbool(Address dst, bool boolconst) {
 1900   if(sizeof(bool) == 1)
 1901     movb(dst, (int) boolconst);
 1902   else if(sizeof(bool) == 2)
 1903     movw(dst, (int) boolconst);
 1904   else if(sizeof(bool) == 4)
 1905     movl(dst, (int) boolconst);
 1906   else
 1907     // unsupported
 1908     ShouldNotReachHere();
 1909 }
 1910 
 1911 void MacroAssembler::movbool(Address dst, Register src) {
 1912   if(sizeof(bool) == 1)
 1913     movb(dst, src);
 1914   else if(sizeof(bool) == 2)
 1915     movw(dst, src);
 1916   else if(sizeof(bool) == 4)
 1917     movl(dst, src);
 1918   else
 1919     // unsupported
 1920     ShouldNotReachHere();
 1921 }
 1922 
 1923 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1924   assert(rscratch != noreg || always_reachable(src), "missing");
 1925 
 1926   if (reachable(src)) {
 1927     movdl(dst, as_Address(src));
 1928   } else {
 1929     lea(rscratch, src);
 1930     movdl(dst, Address(rscratch, 0));
 1931   }
 1932 }
 1933 
 1934 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1935   assert(rscratch != noreg || always_reachable(src), "missing");
 1936 
 1937   if (reachable(src)) {
 1938     movq(dst, as_Address(src));
 1939   } else {
 1940     lea(rscratch, src);
 1941     movq(dst, Address(rscratch, 0));
 1942   }
 1943 }
 1944 
 1945 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1946   assert(rscratch != noreg || always_reachable(src), "missing");
 1947 
 1948   if (reachable(src)) {
 1949     if (UseXmmLoadAndClearUpper) {
 1950       movsd (dst, as_Address(src));
 1951     } else {
 1952       movlpd(dst, as_Address(src));
 1953     }
 1954   } else {
 1955     lea(rscratch, src);
 1956     if (UseXmmLoadAndClearUpper) {
 1957       movsd (dst, Address(rscratch, 0));
 1958     } else {
 1959       movlpd(dst, Address(rscratch, 0));
 1960     }
 1961   }
 1962 }
 1963 
 1964 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1965   assert(rscratch != noreg || always_reachable(src), "missing");
 1966 
 1967   if (reachable(src)) {
 1968     movss(dst, as_Address(src));
 1969   } else {
 1970     lea(rscratch, src);
 1971     movss(dst, Address(rscratch, 0));
 1972   }
 1973 }
 1974 
 1975 void MacroAssembler::movhlf(XMMRegister dst, XMMRegister src, Register rscratch) {
 1976   if (VM_Version::supports_avx10_2()) {
 1977     evmovw(dst, src);
 1978   } else {
 1979     assert(rscratch != noreg, "missing");
 1980     evmovw(rscratch, src);
 1981     evmovw(dst, rscratch);
 1982   }
 1983 }
 1984 
 1985 void MacroAssembler::mov64(Register dst, int64_t imm64) {
 1986   if (is_uimm32(imm64)) {
 1987     movl(dst, checked_cast<uint32_t>(imm64));
 1988   } else if (is_simm32(imm64)) {
 1989     movq(dst, checked_cast<int32_t>(imm64));
 1990   } else {
 1991     Assembler::mov64(dst, imm64);
 1992   }
 1993 }
 1994 
 1995 void MacroAssembler::mov64(Register dst, int64_t imm64, relocInfo::relocType rtype, int format) {
 1996   Assembler::mov64(dst, imm64, rtype, format);
 1997 }
 1998 
 1999 void MacroAssembler::movptr(Register dst, Register src) {
 2000   movq(dst, src);
 2001 }
 2002 
 2003 void MacroAssembler::movptr(Register dst, Address src) {
 2004   movq(dst, src);
 2005 }
 2006 
 2007 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 2008 void MacroAssembler::movptr(Register dst, intptr_t src) {
 2009   mov64(dst, src);
 2010 }
 2011 
 2012 void MacroAssembler::movptr(Address dst, Register src) {
 2013   movq(dst, src);
 2014 }
 2015 
 2016 void MacroAssembler::movptr(Address dst, int32_t src) {
 2017   movslq(dst, src);
 2018 }
 2019 
 2020 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 2021   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2022   Assembler::movdqu(dst, src);
 2023 }
 2024 
 2025 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 2026   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2027   Assembler::movdqu(dst, src);
 2028 }
 2029 
 2030 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2031   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2032   Assembler::movdqu(dst, src);
 2033 }
 2034 
 2035 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2036   assert(rscratch != noreg || always_reachable(src), "missing");
 2037 
 2038   if (reachable(src)) {
 2039     movdqu(dst, as_Address(src));
 2040   } else {
 2041     lea(rscratch, src);
 2042     movdqu(dst, Address(rscratch, 0));
 2043   }
 2044 }
 2045 
 2046 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2047   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2048   Assembler::vmovdqu(dst, src);
 2049 }
 2050 
 2051 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2052   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2053   Assembler::vmovdqu(dst, src);
 2054 }
 2055 
 2056 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2057   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2058   Assembler::vmovdqu(dst, src);
 2059 }
 2060 
 2061 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2062   assert(rscratch != noreg || always_reachable(src), "missing");
 2063 
 2064   if (reachable(src)) {
 2065     vmovdqu(dst, as_Address(src));
 2066   }
 2067   else {
 2068     lea(rscratch, src);
 2069     vmovdqu(dst, Address(rscratch, 0));
 2070   }
 2071 }
 2072 
 2073 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2074   assert(rscratch != noreg || always_reachable(src), "missing");
 2075 
 2076   if (vector_len == AVX_512bit) {
 2077     evmovdquq(dst, src, AVX_512bit, rscratch);
 2078   } else if (vector_len == AVX_256bit) {
 2079     vmovdqu(dst, src, rscratch);
 2080   } else {
 2081     movdqu(dst, src, rscratch);
 2082   }
 2083 }
 2084 
 2085 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
 2086   if (vector_len == AVX_512bit) {
 2087     evmovdquq(dst, src, AVX_512bit);
 2088   } else if (vector_len == AVX_256bit) {
 2089     vmovdqu(dst, src);
 2090   } else {
 2091     movdqu(dst, src);
 2092   }
 2093 }
 2094 
 2095 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
 2096   if (vector_len == AVX_512bit) {
 2097     evmovdquq(dst, src, AVX_512bit);
 2098   } else if (vector_len == AVX_256bit) {
 2099     vmovdqu(dst, src);
 2100   } else {
 2101     movdqu(dst, src);
 2102   }
 2103 }
 2104 
 2105 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
 2106   if (vector_len == AVX_512bit) {
 2107     evmovdquq(dst, src, AVX_512bit);
 2108   } else if (vector_len == AVX_256bit) {
 2109     vmovdqu(dst, src);
 2110   } else {
 2111     movdqu(dst, src);
 2112   }
 2113 }
 2114 
 2115 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2116   assert(rscratch != noreg || always_reachable(src), "missing");
 2117 
 2118   if (reachable(src)) {
 2119     vmovdqa(dst, as_Address(src));
 2120   }
 2121   else {
 2122     lea(rscratch, src);
 2123     vmovdqa(dst, Address(rscratch, 0));
 2124   }
 2125 }
 2126 
 2127 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2128   assert(rscratch != noreg || always_reachable(src), "missing");
 2129 
 2130   if (vector_len == AVX_512bit) {
 2131     evmovdqaq(dst, src, AVX_512bit, rscratch);
 2132   } else if (vector_len == AVX_256bit) {
 2133     vmovdqa(dst, src, rscratch);
 2134   } else {
 2135     movdqa(dst, src, rscratch);
 2136   }
 2137 }
 2138 
 2139 void MacroAssembler::kmov(KRegister dst, Address src) {
 2140   if (VM_Version::supports_avx512bw()) {
 2141     kmovql(dst, src);
 2142   } else {
 2143     assert(VM_Version::supports_evex(), "");
 2144     kmovwl(dst, src);
 2145   }
 2146 }
 2147 
 2148 void MacroAssembler::kmov(Address dst, KRegister src) {
 2149   if (VM_Version::supports_avx512bw()) {
 2150     kmovql(dst, src);
 2151   } else {
 2152     assert(VM_Version::supports_evex(), "");
 2153     kmovwl(dst, src);
 2154   }
 2155 }
 2156 
 2157 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2158   if (VM_Version::supports_avx512bw()) {
 2159     kmovql(dst, src);
 2160   } else {
 2161     assert(VM_Version::supports_evex(), "");
 2162     kmovwl(dst, src);
 2163   }
 2164 }
 2165 
 2166 void MacroAssembler::kmov(Register dst, KRegister src) {
 2167   if (VM_Version::supports_avx512bw()) {
 2168     kmovql(dst, src);
 2169   } else {
 2170     assert(VM_Version::supports_evex(), "");
 2171     kmovwl(dst, src);
 2172   }
 2173 }
 2174 
 2175 void MacroAssembler::kmov(KRegister dst, Register src) {
 2176   if (VM_Version::supports_avx512bw()) {
 2177     kmovql(dst, src);
 2178   } else {
 2179     assert(VM_Version::supports_evex(), "");
 2180     kmovwl(dst, src);
 2181   }
 2182 }
 2183 
 2184 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2185   assert(rscratch != noreg || always_reachable(src), "missing");
 2186 
 2187   if (reachable(src)) {
 2188     kmovql(dst, as_Address(src));
 2189   } else {
 2190     lea(rscratch, src);
 2191     kmovql(dst, Address(rscratch, 0));
 2192   }
 2193 }
 2194 
 2195 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2196   assert(rscratch != noreg || always_reachable(src), "missing");
 2197 
 2198   if (reachable(src)) {
 2199     kmovwl(dst, as_Address(src));
 2200   } else {
 2201     lea(rscratch, src);
 2202     kmovwl(dst, Address(rscratch, 0));
 2203   }
 2204 }
 2205 
 2206 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2207                                int vector_len, Register rscratch) {
 2208   assert(rscratch != noreg || always_reachable(src), "missing");
 2209 
 2210   if (reachable(src)) {
 2211     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2212   } else {
 2213     lea(rscratch, src);
 2214     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2215   }
 2216 }
 2217 
 2218 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2219                                int vector_len, Register rscratch) {
 2220   assert(rscratch != noreg || always_reachable(src), "missing");
 2221 
 2222   if (reachable(src)) {
 2223     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2224   } else {
 2225     lea(rscratch, src);
 2226     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2227   }
 2228 }
 2229 
 2230 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2231   assert(rscratch != noreg || always_reachable(src), "missing");
 2232 
 2233   if (reachable(src)) {
 2234     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2235   } else {
 2236     lea(rscratch, src);
 2237     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2238   }
 2239 }
 2240 
 2241 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2242   assert(rscratch != noreg || always_reachable(src), "missing");
 2243 
 2244   if (reachable(src)) {
 2245     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2246   } else {
 2247     lea(rscratch, src);
 2248     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2249   }
 2250 }
 2251 
 2252 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2253   assert(rscratch != noreg || always_reachable(src), "missing");
 2254 
 2255   if (reachable(src)) {
 2256     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2257   } else {
 2258     lea(rscratch, src);
 2259     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2260   }
 2261 }
 2262 
 2263 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2264   assert(rscratch != noreg || always_reachable(src), "missing");
 2265 
 2266   if (reachable(src)) {
 2267     Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
 2268   } else {
 2269     lea(rscratch, src);
 2270     Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2271   }
 2272 }
 2273 
 2274 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2275   assert(rscratch != noreg || always_reachable(src), "missing");
 2276 
 2277   if (reachable(src)) {
 2278     Assembler::evmovdqaq(dst, as_Address(src), vector_len);
 2279   } else {
 2280     lea(rscratch, src);
 2281     Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
 2282   }
 2283 }
 2284 
 2285 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2286   assert(rscratch != noreg || always_reachable(src), "missing");
 2287 
 2288   if (reachable(src)) {
 2289     Assembler::movapd(dst, as_Address(src));
 2290   } else {
 2291     lea(rscratch, src);
 2292     Assembler::movapd(dst, Address(rscratch, 0));
 2293   }
 2294 }
 2295 
 2296 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2297   assert(rscratch != noreg || always_reachable(src), "missing");
 2298 
 2299   if (reachable(src)) {
 2300     Assembler::movdqa(dst, as_Address(src));
 2301   } else {
 2302     lea(rscratch, src);
 2303     Assembler::movdqa(dst, Address(rscratch, 0));
 2304   }
 2305 }
 2306 
 2307 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2308   assert(rscratch != noreg || always_reachable(src), "missing");
 2309 
 2310   if (reachable(src)) {
 2311     Assembler::movsd(dst, as_Address(src));
 2312   } else {
 2313     lea(rscratch, src);
 2314     Assembler::movsd(dst, Address(rscratch, 0));
 2315   }
 2316 }
 2317 
 2318 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2319   assert(rscratch != noreg || always_reachable(src), "missing");
 2320 
 2321   if (reachable(src)) {
 2322     Assembler::movss(dst, as_Address(src));
 2323   } else {
 2324     lea(rscratch, src);
 2325     Assembler::movss(dst, Address(rscratch, 0));
 2326   }
 2327 }
 2328 
 2329 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2330   assert(rscratch != noreg || always_reachable(src), "missing");
 2331 
 2332   if (reachable(src)) {
 2333     Assembler::movddup(dst, as_Address(src));
 2334   } else {
 2335     lea(rscratch, src);
 2336     Assembler::movddup(dst, Address(rscratch, 0));
 2337   }
 2338 }
 2339 
 2340 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2341   assert(rscratch != noreg || always_reachable(src), "missing");
 2342 
 2343   if (reachable(src)) {
 2344     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2345   } else {
 2346     lea(rscratch, src);
 2347     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2348   }
 2349 }
 2350 
 2351 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2352   assert(rscratch != noreg || always_reachable(src), "missing");
 2353 
 2354   if (reachable(src)) {
 2355     Assembler::mulsd(dst, as_Address(src));
 2356   } else {
 2357     lea(rscratch, src);
 2358     Assembler::mulsd(dst, Address(rscratch, 0));
 2359   }
 2360 }
 2361 
 2362 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2363   assert(rscratch != noreg || always_reachable(src), "missing");
 2364 
 2365   if (reachable(src)) {
 2366     Assembler::mulss(dst, as_Address(src));
 2367   } else {
 2368     lea(rscratch, src);
 2369     Assembler::mulss(dst, Address(rscratch, 0));
 2370   }
 2371 }
 2372 
 2373 void MacroAssembler::null_check(Register reg, int offset) {
 2374   if (needs_explicit_null_check(offset)) {
 2375     // provoke OS null exception if reg is null by
 2376     // accessing M[reg] w/o changing any (non-CC) registers
 2377     // NOTE: cmpl is plenty here to provoke a segv
 2378     cmpptr(rax, Address(reg, 0));
 2379     // Note: should probably use testl(rax, Address(reg, 0));
 2380     //       may be shorter code (however, this version of
 2381     //       testl needs to be implemented first)
 2382   } else {
 2383     // nothing to do, (later) access of M[reg + offset]
 2384     // will provoke OS null exception if reg is null
 2385   }
 2386 }
 2387 
 2388 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
 2389   andptr(markword, markWord::inline_type_pattern_mask);
 2390   cmpptr(markword, markWord::inline_type_pattern);
 2391   jcc(Assembler::equal, is_inline_type);
 2392 }
 2393 
 2394 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
 2395   if (can_be_null) {
 2396     testptr(object, object);
 2397     jcc(Assembler::zero, not_inline_type);
 2398   }
 2399   const int is_inline_type_mask = markWord::inline_type_pattern;
 2400   movptr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
 2401   andptr(tmp, is_inline_type_mask);
 2402   cmpptr(tmp, is_inline_type_mask);
 2403   jcc(Assembler::notEqual, not_inline_type);
 2404 }
 2405 
 2406 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
 2407   movl(temp_reg, flags);
 2408   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2409   jcc(Assembler::notEqual, is_null_free_inline_type);
 2410 }
 2411 
 2412 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
 2413   movl(temp_reg, flags);
 2414   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2415   jcc(Assembler::equal, not_null_free_inline_type);
 2416 }
 2417 
 2418 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
 2419   movl(temp_reg, flags);
 2420   testl(temp_reg, 1 << ResolvedFieldEntry::is_flat_shift);
 2421   jcc(Assembler::notEqual, is_flat);
 2422 }
 2423 
 2424 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
 2425   movl(temp_reg, flags);
 2426   testl(temp_reg, 1 << ResolvedFieldEntry::has_null_marker_shift);
 2427   jcc(Assembler::notEqual, has_null_marker);
 2428 }
 2429 
 2430 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
 2431   Label test_mark_word;
 2432   // load mark word
 2433   movptr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
 2434   // check displaced
 2435   testl(temp_reg, markWord::unlocked_value);
 2436   jccb(Assembler::notZero, test_mark_word);
 2437   // slow path use klass prototype
 2438   push(rscratch1);
 2439   load_prototype_header(temp_reg, oop, rscratch1);
 2440   pop(rscratch1);
 2441 
 2442   bind(test_mark_word);
 2443   testl(temp_reg, test_bit);
 2444   jcc((jmp_set) ? Assembler::notZero : Assembler::zero, jmp_label);
 2445 }
 2446 
 2447 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg,
 2448                                          Label& is_flat_array) {
 2449 #ifdef _LP64
 2450   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
 2451 #else
 2452   load_klass(temp_reg, oop, noreg);
 2453   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 2454   test_flat_array_layout(temp_reg, is_flat_array);
 2455 #endif
 2456 }
 2457 
 2458 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
 2459                                              Label& is_non_flat_array) {
 2460 #ifdef _LP64
 2461   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
 2462 #else
 2463   load_klass(temp_reg, oop, noreg);
 2464   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 2465   test_non_flat_array_layout(temp_reg, is_non_flat_array);
 2466 #endif
 2467 }
 2468 
 2469 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label&is_null_free_array) {
 2470 #ifdef _LP64
 2471   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
 2472 #else
 2473   Unimplemented();
 2474 #endif
 2475 }
 2476 
 2477 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
 2478 #ifdef _LP64
 2479   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
 2480 #else
 2481   Unimplemented();
 2482 #endif
 2483 }
 2484 
 2485 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
 2486   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2487   jcc(Assembler::notZero, is_flat_array);
 2488 }
 2489 
 2490 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
 2491   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2492   jcc(Assembler::zero, is_non_flat_array);
 2493 }
 2494 
 2495 void MacroAssembler::os_breakpoint() {
 2496   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2497   // (e.g., MSVC can't call ps() otherwise)
 2498   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2499 }
 2500 
 2501 void MacroAssembler::unimplemented(const char* what) {
 2502   const char* buf = nullptr;
 2503   {
 2504     ResourceMark rm;
 2505     stringStream ss;
 2506     ss.print("unimplemented: %s", what);
 2507     buf = code_string(ss.as_string());
 2508   }
 2509   stop(buf);
 2510 }
 2511 
 2512 #define XSTATE_BV 0x200
 2513 
 2514 void MacroAssembler::pop_CPU_state() {
 2515   pop_FPU_state();
 2516   pop_IU_state();
 2517 }
 2518 
 2519 void MacroAssembler::pop_FPU_state() {
 2520   fxrstor(Address(rsp, 0));
 2521   addptr(rsp, FPUStateSizeInWords * wordSize);
 2522 }
 2523 
 2524 void MacroAssembler::pop_IU_state() {
 2525   popa();
 2526   addq(rsp, 8);
 2527   popf();
 2528 }
 2529 
 2530 // Save Integer and Float state
 2531 // Warning: Stack must be 16 byte aligned (64bit)
 2532 void MacroAssembler::push_CPU_state() {
 2533   push_IU_state();
 2534   push_FPU_state();
 2535 }
 2536 
 2537 void MacroAssembler::push_FPU_state() {
 2538   subptr(rsp, FPUStateSizeInWords * wordSize);
 2539   fxsave(Address(rsp, 0));
 2540 }
 2541 
 2542 void MacroAssembler::push_IU_state() {
 2543   // Push flags first because pusha kills them
 2544   pushf();
 2545   // Make sure rsp stays 16-byte aligned
 2546   subq(rsp, 8);
 2547   pusha();
 2548 }
 2549 
 2550 void MacroAssembler::push_cont_fastpath() {
 2551   if (!Continuations::enabled()) return;
 2552 
 2553   Label L_done;
 2554   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2555   jccb(Assembler::belowEqual, L_done);
 2556   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2557   bind(L_done);
 2558 }
 2559 
 2560 void MacroAssembler::pop_cont_fastpath() {
 2561   if (!Continuations::enabled()) return;
 2562 
 2563   Label L_done;
 2564   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2565   jccb(Assembler::below, L_done);
 2566   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2567   bind(L_done);
 2568 }
 2569 
 2570 #ifdef ASSERT
 2571 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2572   Label no_cont;
 2573   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2574   testl(cont, cont);
 2575   jcc(Assembler::zero, no_cont);
 2576   stop(name);
 2577   bind(no_cont);
 2578 }
 2579 #endif
 2580 
 2581 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2582   // we must set sp to zero to clear frame
 2583   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2584   // must clear fp, so that compiled frames are not confused; it is
 2585   // possible that we need it only for debugging
 2586   if (clear_fp) {
 2587     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2588   }
 2589   // Always clear the pc because it could have been set by make_walkable()
 2590   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2591   vzeroupper();
 2592 }
 2593 
 2594 void MacroAssembler::round_to(Register reg, int modulus) {
 2595   addptr(reg, modulus - 1);
 2596   andptr(reg, -modulus);
 2597 }
 2598 
 2599 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2600   if (at_return) {
 2601     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2602     // we may safely use rsp instead to perform the stack watermark check.
 2603     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2604     jcc(Assembler::above, slow_path);
 2605     return;
 2606   }
 2607   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2608   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2609 }
 2610 
 2611 // Calls to C land
 2612 //
 2613 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2614 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2615 // has to be reset to 0. This is required to allow proper stack traversal.
 2616 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2617                                          Register last_java_fp,
 2618                                          address  last_java_pc,
 2619                                          Register rscratch) {
 2620   vzeroupper();
 2621   // determine last_java_sp register
 2622   if (!last_java_sp->is_valid()) {
 2623     last_java_sp = rsp;
 2624   }
 2625   // last_java_fp is optional
 2626   if (last_java_fp->is_valid()) {
 2627     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2628   }
 2629   // last_java_pc is optional
 2630   if (last_java_pc != nullptr) {
 2631     Address java_pc(r15_thread,
 2632                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2633     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2634   }
 2635   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2636 }
 2637 
 2638 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2639                                          Register last_java_fp,
 2640                                          Label &L,
 2641                                          Register scratch) {
 2642   lea(scratch, L);
 2643   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2644   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2645 }
 2646 
 2647 void MacroAssembler::shlptr(Register dst, int imm8) {
 2648   shlq(dst, imm8);
 2649 }
 2650 
 2651 void MacroAssembler::shrptr(Register dst, int imm8) {
 2652   shrq(dst, imm8);
 2653 }
 2654 
 2655 void MacroAssembler::sign_extend_byte(Register reg) {
 2656   movsbl(reg, reg); // movsxb
 2657 }
 2658 
 2659 void MacroAssembler::sign_extend_short(Register reg) {
 2660   movswl(reg, reg); // movsxw
 2661 }
 2662 
 2663 void MacroAssembler::narrow_subword_type(Register reg, BasicType bt) {
 2664   assert(is_subword_type(bt), "required");
 2665   switch (bt) {
 2666   case T_BOOLEAN: andl(reg, 1); break;
 2667   case T_BYTE:    movsbl(reg, reg); break;
 2668   case T_CHAR:    movzwl(reg, reg); break;
 2669   case T_SHORT:   movswl(reg, reg); break;
 2670   default:        ShouldNotReachHere();
 2671   }
 2672 }
 2673 
 2674 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2675   if (imm32 >= 0 && is8bit(imm32)) {
 2676     testb(dst, imm32);
 2677   } else {
 2678     Assembler::testl(dst, imm32);
 2679   }
 2680 }
 2681 
 2682 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2683   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2684     testb(dst, imm32);
 2685   } else {
 2686     Assembler::testl(dst, imm32);
 2687   }
 2688 }
 2689 
 2690 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2691   assert(always_reachable(src), "Address should be reachable");
 2692   testl(dst, as_Address(src));
 2693 }
 2694 
 2695 void MacroAssembler::testq(Address dst, int32_t imm32) {
 2696   if (imm32 >= 0) {
 2697     testl(dst, imm32);
 2698   } else {
 2699     Assembler::testq(dst, imm32);
 2700   }
 2701 }
 2702 
 2703 void MacroAssembler::testq(Register dst, int32_t imm32) {
 2704   if (imm32 >= 0) {
 2705     testl(dst, imm32);
 2706   } else {
 2707     Assembler::testq(dst, imm32);
 2708   }
 2709 }
 2710 
 2711 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2712   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2713   Assembler::pcmpeqb(dst, src);
 2714 }
 2715 
 2716 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2717   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2718   Assembler::pcmpeqw(dst, src);
 2719 }
 2720 
 2721 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2722   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2723   Assembler::pcmpestri(dst, src, imm8);
 2724 }
 2725 
 2726 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2727   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2728   Assembler::pcmpestri(dst, src, imm8);
 2729 }
 2730 
 2731 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2732   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2733   Assembler::pmovzxbw(dst, src);
 2734 }
 2735 
 2736 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2737   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2738   Assembler::pmovzxbw(dst, src);
 2739 }
 2740 
 2741 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2742   assert((src->encoding() < 16),"XMM register should be 0-15");
 2743   Assembler::pmovmskb(dst, src);
 2744 }
 2745 
 2746 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2747   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2748   Assembler::ptest(dst, src);
 2749 }
 2750 
 2751 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2752   assert(rscratch != noreg || always_reachable(src), "missing");
 2753 
 2754   if (reachable(src)) {
 2755     Assembler::sqrtss(dst, as_Address(src));
 2756   } else {
 2757     lea(rscratch, src);
 2758     Assembler::sqrtss(dst, Address(rscratch, 0));
 2759   }
 2760 }
 2761 
 2762 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2763   assert(rscratch != noreg || always_reachable(src), "missing");
 2764 
 2765   if (reachable(src)) {
 2766     Assembler::subsd(dst, as_Address(src));
 2767   } else {
 2768     lea(rscratch, src);
 2769     Assembler::subsd(dst, Address(rscratch, 0));
 2770   }
 2771 }
 2772 
 2773 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2774   assert(rscratch != noreg || always_reachable(src), "missing");
 2775 
 2776   if (reachable(src)) {
 2777     Assembler::roundsd(dst, as_Address(src), rmode);
 2778   } else {
 2779     lea(rscratch, src);
 2780     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2781   }
 2782 }
 2783 
 2784 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2785   assert(rscratch != noreg || always_reachable(src), "missing");
 2786 
 2787   if (reachable(src)) {
 2788     Assembler::subss(dst, as_Address(src));
 2789   } else {
 2790     lea(rscratch, src);
 2791     Assembler::subss(dst, Address(rscratch, 0));
 2792   }
 2793 }
 2794 
 2795 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2796   assert(rscratch != noreg || always_reachable(src), "missing");
 2797 
 2798   if (reachable(src)) {
 2799     Assembler::ucomisd(dst, as_Address(src));
 2800   } else {
 2801     lea(rscratch, src);
 2802     Assembler::ucomisd(dst, Address(rscratch, 0));
 2803   }
 2804 }
 2805 
 2806 void MacroAssembler::evucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2807   assert(rscratch != noreg || always_reachable(src), "missing");
 2808 
 2809   if (reachable(src)) {
 2810     Assembler::evucomxsd(dst, as_Address(src));
 2811   } else {
 2812     lea(rscratch, src);
 2813     Assembler::evucomxsd(dst, Address(rscratch, 0));
 2814   }
 2815 }
 2816 
 2817 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2818   assert(rscratch != noreg || always_reachable(src), "missing");
 2819 
 2820   if (reachable(src)) {
 2821     Assembler::ucomiss(dst, as_Address(src));
 2822   } else {
 2823     lea(rscratch, src);
 2824     Assembler::ucomiss(dst, Address(rscratch, 0));
 2825   }
 2826 }
 2827 
 2828 void MacroAssembler::evucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2829   assert(rscratch != noreg || always_reachable(src), "missing");
 2830 
 2831   if (reachable(src)) {
 2832     Assembler::evucomxss(dst, as_Address(src));
 2833   } else {
 2834     lea(rscratch, src);
 2835     Assembler::evucomxss(dst, Address(rscratch, 0));
 2836   }
 2837 }
 2838 
 2839 void MacroAssembler::evucomish(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2840   assert(rscratch != noreg || always_reachable(src), "missing");
 2841 
 2842   if (reachable(src)) {
 2843     Assembler::evucomish(dst, as_Address(src));
 2844   } else {
 2845     lea(rscratch, src);
 2846     Assembler::evucomish(dst, Address(rscratch, 0));
 2847   }
 2848 }
 2849 
 2850 void MacroAssembler::evucomxsh(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2851   assert(rscratch != noreg || always_reachable(src), "missing");
 2852 
 2853   if (reachable(src)) {
 2854     Assembler::evucomxsh(dst, as_Address(src));
 2855   } else {
 2856     lea(rscratch, src);
 2857     Assembler::evucomxsh(dst, Address(rscratch, 0));
 2858   }
 2859 }
 2860 
 2861 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2862   assert(rscratch != noreg || always_reachable(src), "missing");
 2863 
 2864   // Used in sign-bit flipping with aligned address.
 2865   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2866 
 2867   if (UseAVX > 2 &&
 2868       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2869       (dst->encoding() >= 16)) {
 2870     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2871   } else if (reachable(src)) {
 2872     Assembler::xorpd(dst, as_Address(src));
 2873   } else {
 2874     lea(rscratch, src);
 2875     Assembler::xorpd(dst, Address(rscratch, 0));
 2876   }
 2877 }
 2878 
 2879 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2880   if (UseAVX > 2 &&
 2881       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2882       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2883     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2884   } else {
 2885     Assembler::xorpd(dst, src);
 2886   }
 2887 }
 2888 
 2889 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2890   if (UseAVX > 2 &&
 2891       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2892       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2893     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2894   } else {
 2895     Assembler::xorps(dst, src);
 2896   }
 2897 }
 2898 
 2899 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2900   assert(rscratch != noreg || always_reachable(src), "missing");
 2901 
 2902   // Used in sign-bit flipping with aligned address.
 2903   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2904 
 2905   if (UseAVX > 2 &&
 2906       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2907       (dst->encoding() >= 16)) {
 2908     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2909   } else if (reachable(src)) {
 2910     Assembler::xorps(dst, as_Address(src));
 2911   } else {
 2912     lea(rscratch, src);
 2913     Assembler::xorps(dst, Address(rscratch, 0));
 2914   }
 2915 }
 2916 
 2917 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2918   assert(rscratch != noreg || always_reachable(src), "missing");
 2919 
 2920   // Used in sign-bit flipping with aligned address.
 2921   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2922   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2923   if (reachable(src)) {
 2924     Assembler::pshufb(dst, as_Address(src));
 2925   } else {
 2926     lea(rscratch, src);
 2927     Assembler::pshufb(dst, Address(rscratch, 0));
 2928   }
 2929 }
 2930 
 2931 // AVX 3-operands instructions
 2932 
 2933 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2934   assert(rscratch != noreg || always_reachable(src), "missing");
 2935 
 2936   if (reachable(src)) {
 2937     vaddsd(dst, nds, as_Address(src));
 2938   } else {
 2939     lea(rscratch, src);
 2940     vaddsd(dst, nds, Address(rscratch, 0));
 2941   }
 2942 }
 2943 
 2944 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2945   assert(rscratch != noreg || always_reachable(src), "missing");
 2946 
 2947   if (reachable(src)) {
 2948     vaddss(dst, nds, as_Address(src));
 2949   } else {
 2950     lea(rscratch, src);
 2951     vaddss(dst, nds, Address(rscratch, 0));
 2952   }
 2953 }
 2954 
 2955 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2956   assert(UseAVX > 0, "requires some form of AVX");
 2957   assert(rscratch != noreg || always_reachable(src), "missing");
 2958 
 2959   if (reachable(src)) {
 2960     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2961   } else {
 2962     lea(rscratch, src);
 2963     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2964   }
 2965 }
 2966 
 2967 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2968   assert(UseAVX > 0, "requires some form of AVX");
 2969   assert(rscratch != noreg || always_reachable(src), "missing");
 2970 
 2971   if (reachable(src)) {
 2972     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2973   } else {
 2974     lea(rscratch, src);
 2975     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2976   }
 2977 }
 2978 
 2979 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2980   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2981   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2982 
 2983   vandps(dst, nds, negate_field, vector_len, rscratch);
 2984 }
 2985 
 2986 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2987   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2988   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2989 
 2990   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2991 }
 2992 
 2993 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2994   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2995   Assembler::vpaddb(dst, nds, src, vector_len);
 2996 }
 2997 
 2998 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2999   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3000   Assembler::vpaddb(dst, nds, src, vector_len);
 3001 }
 3002 
 3003 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3004   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3005   Assembler::vpaddw(dst, nds, src, vector_len);
 3006 }
 3007 
 3008 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3009   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3010   Assembler::vpaddw(dst, nds, src, vector_len);
 3011 }
 3012 
 3013 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3014   assert(rscratch != noreg || always_reachable(src), "missing");
 3015 
 3016   if (reachable(src)) {
 3017     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 3018   } else {
 3019     lea(rscratch, src);
 3020     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 3021   }
 3022 }
 3023 
 3024 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3025   assert(rscratch != noreg || always_reachable(src), "missing");
 3026 
 3027   if (reachable(src)) {
 3028     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 3029   } else {
 3030     lea(rscratch, src);
 3031     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 3032   }
 3033 }
 3034 
 3035 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3036   assert(rscratch != noreg || always_reachable(src), "missing");
 3037 
 3038   if (reachable(src)) {
 3039     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 3040   } else {
 3041     lea(rscratch, src);
 3042     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 3043   }
 3044 }
 3045 
 3046 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3047   assert(rscratch != noreg || always_reachable(src), "missing");
 3048 
 3049   if (reachable(src)) {
 3050     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 3051   } else {
 3052     lea(rscratch, src);
 3053     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 3054   }
 3055 }
 3056 
 3057 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3058   assert(rscratch != noreg || always_reachable(src), "missing");
 3059 
 3060   if (reachable(src)) {
 3061     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 3062   } else {
 3063     lea(rscratch, src);
 3064     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 3065   }
 3066 }
 3067 
 3068 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3069   assert(rscratch != noreg || always_reachable(src), "missing");
 3070 
 3071   if (reachable(src)) {
 3072     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 3073   } else {
 3074     lea(rscratch, src);
 3075     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 3076   }
 3077 }
 3078 
 3079 // Vector float blend
 3080 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3081 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3082   // WARN: Allow dst == (src1|src2), mask == scratch
 3083   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3084                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3085   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 3086   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3087   if (blend_emulation && scratch_available && dst_available) {
 3088     if (compute_mask) {
 3089       vpsrad(scratch, mask, 32, vector_len);
 3090       mask = scratch;
 3091     }
 3092     if (dst == src1) {
 3093       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 3094       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3095     } else {
 3096       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3097       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 3098     }
 3099     vpor(dst, dst, scratch, vector_len);
 3100   } else {
 3101     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 3102   }
 3103 }
 3104 
 3105 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3106 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3107   // WARN: Allow dst == (src1|src2), mask == scratch
 3108   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3109                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3110   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 3111   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3112   if (blend_emulation && scratch_available && dst_available) {
 3113     if (compute_mask) {
 3114       vpxor(scratch, scratch, scratch, vector_len);
 3115       vpcmpgtq(scratch, scratch, mask, vector_len);
 3116       mask = scratch;
 3117     }
 3118     if (dst == src1) {
 3119       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 3120       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3121     } else {
 3122       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3123       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 3124     }
 3125     vpor(dst, dst, scratch, vector_len);
 3126   } else {
 3127     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 3128   }
 3129 }
 3130 
 3131 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3132   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3133   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 3134 }
 3135 
 3136 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 3137   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3138   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 3139 }
 3140 
 3141 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3142   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3143   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3144 }
 3145 
 3146 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3147   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3148   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3149 }
 3150 
 3151 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3152   assert(rscratch != noreg || always_reachable(src), "missing");
 3153 
 3154   if (reachable(src)) {
 3155     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 3156   } else {
 3157     lea(rscratch, src);
 3158     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 3159   }
 3160 }
 3161 
 3162 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3163                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3164   assert(rscratch != noreg || always_reachable(src), "missing");
 3165 
 3166   if (reachable(src)) {
 3167     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3168   } else {
 3169     lea(rscratch, src);
 3170     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3171   }
 3172 }
 3173 
 3174 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3175                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3176   assert(rscratch != noreg || always_reachable(src), "missing");
 3177 
 3178   if (reachable(src)) {
 3179     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3180   } else {
 3181     lea(rscratch, src);
 3182     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3183   }
 3184 }
 3185 
 3186 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3187                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3188   assert(rscratch != noreg || always_reachable(src), "missing");
 3189 
 3190   if (reachable(src)) {
 3191     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3192   } else {
 3193     lea(rscratch, src);
 3194     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3195   }
 3196 }
 3197 
 3198 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3199                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3200   assert(rscratch != noreg || always_reachable(src), "missing");
 3201 
 3202   if (reachable(src)) {
 3203     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3204   } else {
 3205     lea(rscratch, src);
 3206     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3207   }
 3208 }
 3209 
 3210 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3211   if (width == Assembler::Q) {
 3212     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3213   } else {
 3214     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3215   }
 3216 }
 3217 
 3218 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3219   int eq_cond_enc = 0x29;
 3220   int gt_cond_enc = 0x37;
 3221   if (width != Assembler::Q) {
 3222     eq_cond_enc = 0x74 + width;
 3223     gt_cond_enc = 0x64 + width;
 3224   }
 3225   switch (cond) {
 3226   case eq:
 3227     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3228     break;
 3229   case neq:
 3230     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3231     vallones(xtmp, vector_len);
 3232     vpxor(dst, xtmp, dst, vector_len);
 3233     break;
 3234   case le:
 3235     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3236     vallones(xtmp, vector_len);
 3237     vpxor(dst, xtmp, dst, vector_len);
 3238     break;
 3239   case nlt:
 3240     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3241     vallones(xtmp, vector_len);
 3242     vpxor(dst, xtmp, dst, vector_len);
 3243     break;
 3244   case lt:
 3245     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3246     break;
 3247   case nle:
 3248     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3249     break;
 3250   default:
 3251     assert(false, "Should not reach here");
 3252   }
 3253 }
 3254 
 3255 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3256   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3257   Assembler::vpmovzxbw(dst, src, vector_len);
 3258 }
 3259 
 3260 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3261   assert((src->encoding() < 16),"XMM register should be 0-15");
 3262   Assembler::vpmovmskb(dst, src, vector_len);
 3263 }
 3264 
 3265 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3266   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3267   Assembler::vpmullw(dst, nds, src, vector_len);
 3268 }
 3269 
 3270 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3271   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3272   Assembler::vpmullw(dst, nds, src, vector_len);
 3273 }
 3274 
 3275 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3276   assert((UseAVX > 0), "AVX support is needed");
 3277   assert(rscratch != noreg || always_reachable(src), "missing");
 3278 
 3279   if (reachable(src)) {
 3280     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3281   } else {
 3282     lea(rscratch, src);
 3283     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3284   }
 3285 }
 3286 
 3287 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3288   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3289   Assembler::vpsubb(dst, nds, src, vector_len);
 3290 }
 3291 
 3292 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3293   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3294   Assembler::vpsubb(dst, nds, src, vector_len);
 3295 }
 3296 
 3297 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3298   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3299   Assembler::vpsubw(dst, nds, src, vector_len);
 3300 }
 3301 
 3302 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3303   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3304   Assembler::vpsubw(dst, nds, src, vector_len);
 3305 }
 3306 
 3307 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3308   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3309   Assembler::vpsraw(dst, nds, shift, vector_len);
 3310 }
 3311 
 3312 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3313   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3314   Assembler::vpsraw(dst, nds, shift, vector_len);
 3315 }
 3316 
 3317 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3318   assert(UseAVX > 2,"");
 3319   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3320      vector_len = 2;
 3321   }
 3322   Assembler::evpsraq(dst, nds, shift, vector_len);
 3323 }
 3324 
 3325 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3326   assert(UseAVX > 2,"");
 3327   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3328      vector_len = 2;
 3329   }
 3330   Assembler::evpsraq(dst, nds, shift, vector_len);
 3331 }
 3332 
 3333 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3334   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3335   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3336 }
 3337 
 3338 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3339   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3340   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3341 }
 3342 
 3343 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3344   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3345   Assembler::vpsllw(dst, nds, shift, vector_len);
 3346 }
 3347 
 3348 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3349   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3350   Assembler::vpsllw(dst, nds, shift, vector_len);
 3351 }
 3352 
 3353 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3354   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3355   Assembler::vptest(dst, src);
 3356 }
 3357 
 3358 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3359   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3360   Assembler::punpcklbw(dst, src);
 3361 }
 3362 
 3363 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3364   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3365   Assembler::pshufd(dst, src, mode);
 3366 }
 3367 
 3368 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3369   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3370   Assembler::pshuflw(dst, src, mode);
 3371 }
 3372 
 3373 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3374   assert(rscratch != noreg || always_reachable(src), "missing");
 3375 
 3376   if (reachable(src)) {
 3377     vandpd(dst, nds, as_Address(src), vector_len);
 3378   } else {
 3379     lea(rscratch, src);
 3380     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3381   }
 3382 }
 3383 
 3384 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3385   assert(rscratch != noreg || always_reachable(src), "missing");
 3386 
 3387   if (reachable(src)) {
 3388     vandps(dst, nds, as_Address(src), vector_len);
 3389   } else {
 3390     lea(rscratch, src);
 3391     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3392   }
 3393 }
 3394 
 3395 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3396                             bool merge, int vector_len, Register rscratch) {
 3397   assert(rscratch != noreg || always_reachable(src), "missing");
 3398 
 3399   if (reachable(src)) {
 3400     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3401   } else {
 3402     lea(rscratch, src);
 3403     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3404   }
 3405 }
 3406 
 3407 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3408   assert(rscratch != noreg || always_reachable(src), "missing");
 3409 
 3410   if (reachable(src)) {
 3411     vdivsd(dst, nds, as_Address(src));
 3412   } else {
 3413     lea(rscratch, src);
 3414     vdivsd(dst, nds, Address(rscratch, 0));
 3415   }
 3416 }
 3417 
 3418 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3419   assert(rscratch != noreg || always_reachable(src), "missing");
 3420 
 3421   if (reachable(src)) {
 3422     vdivss(dst, nds, as_Address(src));
 3423   } else {
 3424     lea(rscratch, src);
 3425     vdivss(dst, nds, Address(rscratch, 0));
 3426   }
 3427 }
 3428 
 3429 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3430   assert(rscratch != noreg || always_reachable(src), "missing");
 3431 
 3432   if (reachable(src)) {
 3433     vmulsd(dst, nds, as_Address(src));
 3434   } else {
 3435     lea(rscratch, src);
 3436     vmulsd(dst, nds, Address(rscratch, 0));
 3437   }
 3438 }
 3439 
 3440 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3441   assert(rscratch != noreg || always_reachable(src), "missing");
 3442 
 3443   if (reachable(src)) {
 3444     vmulss(dst, nds, as_Address(src));
 3445   } else {
 3446     lea(rscratch, src);
 3447     vmulss(dst, nds, Address(rscratch, 0));
 3448   }
 3449 }
 3450 
 3451 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3452   assert(rscratch != noreg || always_reachable(src), "missing");
 3453 
 3454   if (reachable(src)) {
 3455     vsubsd(dst, nds, as_Address(src));
 3456   } else {
 3457     lea(rscratch, src);
 3458     vsubsd(dst, nds, Address(rscratch, 0));
 3459   }
 3460 }
 3461 
 3462 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3463   assert(rscratch != noreg || always_reachable(src), "missing");
 3464 
 3465   if (reachable(src)) {
 3466     vsubss(dst, nds, as_Address(src));
 3467   } else {
 3468     lea(rscratch, src);
 3469     vsubss(dst, nds, Address(rscratch, 0));
 3470   }
 3471 }
 3472 
 3473 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3474   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3475   assert(rscratch != noreg || always_reachable(src), "missing");
 3476 
 3477   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3478 }
 3479 
 3480 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3481   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3482   assert(rscratch != noreg || always_reachable(src), "missing");
 3483 
 3484   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3485 }
 3486 
 3487 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3488   assert(rscratch != noreg || always_reachable(src), "missing");
 3489 
 3490   if (reachable(src)) {
 3491     vxorpd(dst, nds, as_Address(src), vector_len);
 3492   } else {
 3493     lea(rscratch, src);
 3494     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3495   }
 3496 }
 3497 
 3498 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3499   assert(rscratch != noreg || always_reachable(src), "missing");
 3500 
 3501   if (reachable(src)) {
 3502     vxorps(dst, nds, as_Address(src), vector_len);
 3503   } else {
 3504     lea(rscratch, src);
 3505     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3506   }
 3507 }
 3508 
 3509 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3510   assert(rscratch != noreg || always_reachable(src), "missing");
 3511 
 3512   if (UseAVX > 1 || (vector_len < 1)) {
 3513     if (reachable(src)) {
 3514       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3515     } else {
 3516       lea(rscratch, src);
 3517       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3518     }
 3519   } else {
 3520     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3521   }
 3522 }
 3523 
 3524 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3525   assert(rscratch != noreg || always_reachable(src), "missing");
 3526 
 3527   if (reachable(src)) {
 3528     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3529   } else {
 3530     lea(rscratch, src);
 3531     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3532   }
 3533 }
 3534 
 3535 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3536   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3537   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3538   // The inverted mask is sign-extended
 3539   andptr(possibly_non_local, inverted_mask);
 3540 }
 3541 
 3542 void MacroAssembler::resolve_jobject(Register value,
 3543                                      Register tmp) {
 3544   Register thread = r15_thread;
 3545   assert_different_registers(value, thread, tmp);
 3546   Label done, tagged, weak_tagged;
 3547   testptr(value, value);
 3548   jcc(Assembler::zero, done);           // Use null as-is.
 3549   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3550   jcc(Assembler::notZero, tagged);
 3551 
 3552   // Resolve local handle
 3553   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3554   verify_oop(value);
 3555   jmp(done);
 3556 
 3557   bind(tagged);
 3558   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3559   jcc(Assembler::notZero, weak_tagged);
 3560 
 3561   // Resolve global handle
 3562   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3563   verify_oop(value);
 3564   jmp(done);
 3565 
 3566   bind(weak_tagged);
 3567   // Resolve jweak.
 3568   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3569                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3570   verify_oop(value);
 3571 
 3572   bind(done);
 3573 }
 3574 
 3575 void MacroAssembler::resolve_global_jobject(Register value,
 3576                                             Register tmp) {
 3577   Register thread = r15_thread;
 3578   assert_different_registers(value, thread, tmp);
 3579   Label done;
 3580 
 3581   testptr(value, value);
 3582   jcc(Assembler::zero, done);           // Use null as-is.
 3583 
 3584 #ifdef ASSERT
 3585   {
 3586     Label valid_global_tag;
 3587     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3588     jcc(Assembler::notZero, valid_global_tag);
 3589     stop("non global jobject using resolve_global_jobject");
 3590     bind(valid_global_tag);
 3591   }
 3592 #endif
 3593 
 3594   // Resolve global handle
 3595   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3596   verify_oop(value);
 3597 
 3598   bind(done);
 3599 }
 3600 
 3601 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3602   subq(dst, imm32);
 3603 }
 3604 
 3605 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3606 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3607   subq_imm32(dst, imm32);
 3608 }
 3609 
 3610 void MacroAssembler::subptr(Register dst, Register src) {
 3611   subq(dst, src);
 3612 }
 3613 
 3614 // C++ bool manipulation
 3615 void MacroAssembler::testbool(Register dst) {
 3616   if(sizeof(bool) == 1)
 3617     testb(dst, 0xff);
 3618   else if(sizeof(bool) == 2) {
 3619     // testw implementation needed for two byte bools
 3620     ShouldNotReachHere();
 3621   } else if(sizeof(bool) == 4)
 3622     testl(dst, dst);
 3623   else
 3624     // unsupported
 3625     ShouldNotReachHere();
 3626 }
 3627 
 3628 void MacroAssembler::testptr(Register dst, Register src) {
 3629   testq(dst, src);
 3630 }
 3631 
 3632 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3633 void MacroAssembler::tlab_allocate(Register obj,
 3634                                    Register var_size_in_bytes,
 3635                                    int con_size_in_bytes,
 3636                                    Register t1,
 3637                                    Register t2,
 3638                                    Label& slow_case) {
 3639   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3640   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3641 }
 3642 
 3643 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3644   RegSet regs;
 3645   regs += RegSet::of(rax, rcx, rdx);
 3646 #ifndef _WINDOWS
 3647   regs += RegSet::of(rsi, rdi);
 3648 #endif
 3649   regs += RegSet::range(r8, r11);
 3650   if (UseAPX) {
 3651     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3652   }
 3653   return regs;
 3654 }
 3655 
 3656 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3657   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3658 #if defined(_WINDOWS)
 3659   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3660   if (num_xmm_registers > 16) {
 3661      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3662   }
 3663   return result;
 3664 #else
 3665   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3666 #endif
 3667 }
 3668 
 3669 // C1 only ever uses the first double/float of the XMM register.
 3670 static int xmm_save_size() { return sizeof(double); }
 3671 
 3672 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3673   masm->movdbl(Address(rsp, offset), reg);
 3674 }
 3675 
 3676 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3677   masm->movdbl(reg, Address(rsp, offset));
 3678 }
 3679 
 3680 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3681                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3682 
 3683   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3684                          StackAlignmentInBytes);
 3685   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3686 
 3687   return gp_area_size + xmm_area_size;
 3688 }
 3689 
 3690 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3691   block_comment("push_call_clobbered_registers start");
 3692   // Regular registers
 3693   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3694 
 3695   int gp_area_size;
 3696   int xmm_area_size;
 3697   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3698                                                gp_area_size, xmm_area_size);
 3699   subptr(rsp, total_save_size);
 3700 
 3701   push_set(gp_registers_to_push, 0);
 3702 
 3703   if (save_fpu) {
 3704     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3705   }
 3706 
 3707   block_comment("push_call_clobbered_registers end");
 3708 }
 3709 
 3710 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3711   block_comment("pop_call_clobbered_registers start");
 3712 
 3713   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 3714 
 3715   int gp_area_size;
 3716   int xmm_area_size;
 3717   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 3718                                                gp_area_size, xmm_area_size);
 3719 
 3720   if (restore_fpu) {
 3721     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3722   }
 3723 
 3724   pop_set(gp_registers_to_pop, 0);
 3725 
 3726   addptr(rsp, total_save_size);
 3727 
 3728   vzeroupper();
 3729 
 3730   block_comment("pop_call_clobbered_registers end");
 3731 }
 3732 
 3733 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3734   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3735   int spill_offset = offset;
 3736 
 3737   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3738     save_xmm_register(this, spill_offset, *it);
 3739     spill_offset += xmm_save_size();
 3740   }
 3741 }
 3742 
 3743 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3744   int restore_size = set.size() * xmm_save_size();
 3745   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3746 
 3747   int restore_offset = offset + restore_size - xmm_save_size();
 3748 
 3749   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3750     restore_xmm_register(this, restore_offset, *it);
 3751     restore_offset -= xmm_save_size();
 3752   }
 3753 }
 3754 
 3755 void MacroAssembler::push_set(RegSet set, int offset) {
 3756   int spill_offset;
 3757   if (offset == -1) {
 3758     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3759     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3760     subptr(rsp, aligned_size);
 3761     spill_offset = 0;
 3762   } else {
 3763     spill_offset = offset;
 3764   }
 3765 
 3766   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3767     movptr(Address(rsp, spill_offset), *it);
 3768     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3769   }
 3770 }
 3771 
 3772 void MacroAssembler::pop_set(RegSet set, int offset) {
 3773 
 3774   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3775   int restore_size = set.size() * gp_reg_size;
 3776   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3777 
 3778   int restore_offset;
 3779   if (offset == -1) {
 3780     restore_offset = restore_size - gp_reg_size;
 3781   } else {
 3782     restore_offset = offset + restore_size - gp_reg_size;
 3783   }
 3784   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3785     movptr(*it, Address(rsp, restore_offset));
 3786     restore_offset -= gp_reg_size;
 3787   }
 3788 
 3789   if (offset == -1) {
 3790     addptr(rsp, aligned_size);
 3791   }
 3792 }
 3793 
 3794 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3795 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3796   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3797   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3798   Label done;
 3799 
 3800   testptr(length_in_bytes, length_in_bytes);
 3801   jcc(Assembler::zero, done);
 3802 
 3803   // initialize topmost word, divide index by 2, check if odd and test if zero
 3804   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3805 #ifdef ASSERT
 3806   {
 3807     Label L;
 3808     testptr(length_in_bytes, BytesPerWord - 1);
 3809     jcc(Assembler::zero, L);
 3810     stop("length must be a multiple of BytesPerWord");
 3811     bind(L);
 3812   }
 3813 #endif
 3814   Register index = length_in_bytes;
 3815   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3816   if (UseIncDec) {
 3817     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3818   } else {
 3819     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3820     shrptr(index, 1);
 3821   }
 3822 
 3823   // initialize remaining object fields: index is a multiple of 2 now
 3824   {
 3825     Label loop;
 3826     bind(loop);
 3827     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3828     decrement(index);
 3829     jcc(Assembler::notZero, loop);
 3830   }
 3831 
 3832   bind(done);
 3833 }
 3834 
 3835 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
 3836   movptr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
 3837 #ifdef ASSERT
 3838   {
 3839     Label done;
 3840     cmpptr(layout_info, 0);
 3841     jcc(Assembler::notEqual, done);
 3842     stop("inline_layout_info_array is null");
 3843     bind(done);
 3844   }
 3845 #endif
 3846 
 3847   InlineLayoutInfo array[2];
 3848   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
 3849   if (is_power_of_2(size)) {
 3850     shll(index, log2i_exact(size)); // Scale index by power of 2
 3851   } else {
 3852     imull(index, index, size); // Scale the index to be the entry index * array_element_size
 3853   }
 3854   lea(layout_info, Address(layout_info, index, Address::times_1, Array<InlineLayoutInfo>::base_offset_in_bytes()));
 3855 }
 3856 
 3857 // Look up the method for a megamorphic invokeinterface call.
 3858 // The target method is determined by <intf_klass, itable_index>.
 3859 // The receiver klass is in recv_klass.
 3860 // On success, the result will be in method_result, and execution falls through.
 3861 // On failure, execution transfers to the given label.
 3862 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3863                                              Register intf_klass,
 3864                                              RegisterOrConstant itable_index,
 3865                                              Register method_result,
 3866                                              Register scan_temp,
 3867                                              Label& L_no_such_interface,
 3868                                              bool return_method) {
 3869   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3870   assert_different_registers(method_result, intf_klass, scan_temp);
 3871   assert(recv_klass != method_result || !return_method,
 3872          "recv_klass can be destroyed when method isn't needed");
 3873 
 3874   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3875          "caller must use same register for non-constant itable index as for method");
 3876 
 3877   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3878   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3879   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3880   int scan_step   = itableOffsetEntry::size() * wordSize;
 3881   int vte_size    = vtableEntry::size_in_bytes();
 3882   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3883   assert(vte_size == wordSize, "else adjust times_vte_scale");
 3884 
 3885   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3886 
 3887   // Could store the aligned, prescaled offset in the klass.
 3888   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 3889 
 3890   if (return_method) {
 3891     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 3892     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3893     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 3894   }
 3895 
 3896   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 3897   //   if (scan->interface() == intf) {
 3898   //     result = (klass + scan->offset() + itable_index);
 3899   //   }
 3900   // }
 3901   Label search, found_method;
 3902 
 3903   for (int peel = 1; peel >= 0; peel--) {
 3904     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 3905     cmpptr(intf_klass, method_result);
 3906 
 3907     if (peel) {
 3908       jccb(Assembler::equal, found_method);
 3909     } else {
 3910       jccb(Assembler::notEqual, search);
 3911       // (invert the test to fall through to found_method...)
 3912     }
 3913 
 3914     if (!peel)  break;
 3915 
 3916     bind(search);
 3917 
 3918     // Check that the previous entry is non-null.  A null entry means that
 3919     // the receiver class doesn't implement the interface, and wasn't the
 3920     // same as when the caller was compiled.
 3921     testptr(method_result, method_result);
 3922     jcc(Assembler::zero, L_no_such_interface);
 3923     addptr(scan_temp, scan_step);
 3924   }
 3925 
 3926   bind(found_method);
 3927 
 3928   if (return_method) {
 3929     // Got a hit.
 3930     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 3931     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 3932   }
 3933 }
 3934 
 3935 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 3936 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 3937 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 3938 // The target method is determined by <holder_klass, itable_index>.
 3939 // The receiver klass is in recv_klass.
 3940 // On success, the result will be in method_result, and execution falls through.
 3941 // On failure, execution transfers to the given label.
 3942 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 3943                                                   Register holder_klass,
 3944                                                   Register resolved_klass,
 3945                                                   Register method_result,
 3946                                                   Register scan_temp,
 3947                                                   Register temp_reg2,
 3948                                                   Register receiver,
 3949                                                   int itable_index,
 3950                                                   Label& L_no_such_interface) {
 3951   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 3952   Register temp_itbl_klass = method_result;
 3953   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 3954 
 3955   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3956   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3957   int scan_step = itableOffsetEntry::size() * wordSize;
 3958   int vte_size = vtableEntry::size_in_bytes();
 3959   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 3960   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 3961   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3962   assert(vte_size == wordSize, "adjust times_vte_scale");
 3963 
 3964   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 3965 
 3966   // temp_itbl_klass = recv_klass.itable[0]
 3967   // scan_temp = &recv_klass.itable[0] + step
 3968   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3969   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 3970   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 3971   xorptr(temp_reg, temp_reg);
 3972 
 3973   // Initial checks:
 3974   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 3975   //   - if (itable[0] == 0), no such interface
 3976   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 3977   cmpptr(holder_klass, resolved_klass);
 3978   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 3979   testptr(temp_itbl_klass, temp_itbl_klass);
 3980   jccb(Assembler::zero, L_no_such_interface);
 3981   cmpptr(holder_klass, temp_itbl_klass);
 3982   jccb(Assembler::equal, L_holder_found);
 3983 
 3984   // Loop: Look for holder_klass record in itable
 3985   //   do {
 3986   //     tmp = itable[index];
 3987   //     index += step;
 3988   //     if (tmp == holder_klass) {
 3989   //       goto L_holder_found; // Found!
 3990   //     }
 3991   //   } while (tmp != 0);
 3992   //   goto L_no_such_interface // Not found.
 3993   Label L_scan_holder;
 3994   bind(L_scan_holder);
 3995     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3996     addptr(scan_temp, scan_step);
 3997     cmpptr(holder_klass, temp_itbl_klass);
 3998     jccb(Assembler::equal, L_holder_found);
 3999     testptr(temp_itbl_klass, temp_itbl_klass);
 4000     jccb(Assembler::notZero, L_scan_holder);
 4001 
 4002   jmpb(L_no_such_interface);
 4003 
 4004   // Loop: Look for resolved_class record in itable
 4005   //   do {
 4006   //     tmp = itable[index];
 4007   //     index += step;
 4008   //     if (tmp == holder_klass) {
 4009   //        // Also check if we have met a holder klass
 4010   //        holder_tmp = itable[index-step-ioffset];
 4011   //     }
 4012   //     if (tmp == resolved_klass) {
 4013   //        goto L_resolved_found;  // Found!
 4014   //     }
 4015   //   } while (tmp != 0);
 4016   //   goto L_no_such_interface // Not found.
 4017   //
 4018   Label L_loop_scan_resolved;
 4019   bind(L_loop_scan_resolved);
 4020     movptr(temp_itbl_klass, Address(scan_temp, 0));
 4021     addptr(scan_temp, scan_step);
 4022     bind(L_loop_scan_resolved_entry);
 4023     cmpptr(holder_klass, temp_itbl_klass);
 4024     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4025     cmpptr(resolved_klass, temp_itbl_klass);
 4026     jccb(Assembler::equal, L_resolved_found);
 4027     testptr(temp_itbl_klass, temp_itbl_klass);
 4028     jccb(Assembler::notZero, L_loop_scan_resolved);
 4029 
 4030   jmpb(L_no_such_interface);
 4031 
 4032   Label L_ready;
 4033 
 4034   // See if we already have a holder klass. If not, go and scan for it.
 4035   bind(L_resolved_found);
 4036   testptr(temp_reg, temp_reg);
 4037   jccb(Assembler::zero, L_scan_holder);
 4038   jmpb(L_ready);
 4039 
 4040   bind(L_holder_found);
 4041   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4042 
 4043   // Finally, temp_reg contains holder_klass vtable offset
 4044   bind(L_ready);
 4045   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 4046   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 4047     load_klass(scan_temp, receiver, noreg);
 4048     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4049   } else {
 4050     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4051   }
 4052 }
 4053 
 4054 
 4055 // virtual method calling
 4056 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 4057                                            RegisterOrConstant vtable_index,
 4058                                            Register method_result) {
 4059   const ByteSize base = Klass::vtable_start_offset();
 4060   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 4061   Address vtable_entry_addr(recv_klass,
 4062                             vtable_index, Address::times_ptr,
 4063                             base + vtableEntry::method_offset());
 4064   movptr(method_result, vtable_entry_addr);
 4065 }
 4066 
 4067 
 4068 void MacroAssembler::check_klass_subtype(Register sub_klass,
 4069                            Register super_klass,
 4070                            Register temp_reg,
 4071                            Label& L_success) {
 4072   Label L_failure;
 4073   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 4074   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 4075   bind(L_failure);
 4076 }
 4077 
 4078 
 4079 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 4080                                                    Register super_klass,
 4081                                                    Register temp_reg,
 4082                                                    Label* L_success,
 4083                                                    Label* L_failure,
 4084                                                    Label* L_slow_path,
 4085                                         RegisterOrConstant super_check_offset) {
 4086   assert_different_registers(sub_klass, super_klass, temp_reg);
 4087   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 4088   if (super_check_offset.is_register()) {
 4089     assert_different_registers(sub_klass, super_klass,
 4090                                super_check_offset.as_register());
 4091   } else if (must_load_sco) {
 4092     assert(temp_reg != noreg, "supply either a temp or a register offset");
 4093   }
 4094 
 4095   Label L_fallthrough;
 4096   int label_nulls = 0;
 4097   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4098   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4099   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 4100   assert(label_nulls <= 1, "at most one null in the batch");
 4101 
 4102   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4103   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 4104   Address super_check_offset_addr(super_klass, sco_offset);
 4105 
 4106   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 4107   // range of a jccb.  If this routine grows larger, reconsider at
 4108   // least some of these.
 4109 #define local_jcc(assembler_cond, label)                                \
 4110   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 4111   else                             jcc( assembler_cond, label) /*omit semi*/
 4112 
 4113   // Hacked jmp, which may only be used just before L_fallthrough.
 4114 #define final_jmp(label)                                                \
 4115   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 4116   else                            jmp(label)                /*omit semi*/
 4117 
 4118   // If the pointers are equal, we are done (e.g., String[] elements).
 4119   // This self-check enables sharing of secondary supertype arrays among
 4120   // non-primary types such as array-of-interface.  Otherwise, each such
 4121   // type would need its own customized SSA.
 4122   // We move this check to the front of the fast path because many
 4123   // type checks are in fact trivially successful in this manner,
 4124   // so we get a nicely predicted branch right at the start of the check.
 4125   cmpptr(sub_klass, super_klass);
 4126   local_jcc(Assembler::equal, *L_success);
 4127 
 4128   // Check the supertype display:
 4129   if (must_load_sco) {
 4130     // Positive movl does right thing on LP64.
 4131     movl(temp_reg, super_check_offset_addr);
 4132     super_check_offset = RegisterOrConstant(temp_reg);
 4133   }
 4134   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 4135   cmpptr(super_klass, super_check_addr); // load displayed supertype
 4136 
 4137   // This check has worked decisively for primary supers.
 4138   // Secondary supers are sought in the super_cache ('super_cache_addr').
 4139   // (Secondary supers are interfaces and very deeply nested subtypes.)
 4140   // This works in the same check above because of a tricky aliasing
 4141   // between the super_cache and the primary super display elements.
 4142   // (The 'super_check_addr' can address either, as the case requires.)
 4143   // Note that the cache is updated below if it does not help us find
 4144   // what we need immediately.
 4145   // So if it was a primary super, we can just fail immediately.
 4146   // Otherwise, it's the slow path for us (no success at this point).
 4147 
 4148   if (super_check_offset.is_register()) {
 4149     local_jcc(Assembler::equal, *L_success);
 4150     cmpl(super_check_offset.as_register(), sc_offset);
 4151     if (L_failure == &L_fallthrough) {
 4152       local_jcc(Assembler::equal, *L_slow_path);
 4153     } else {
 4154       local_jcc(Assembler::notEqual, *L_failure);
 4155       final_jmp(*L_slow_path);
 4156     }
 4157   } else if (super_check_offset.as_constant() == sc_offset) {
 4158     // Need a slow path; fast failure is impossible.
 4159     if (L_slow_path == &L_fallthrough) {
 4160       local_jcc(Assembler::equal, *L_success);
 4161     } else {
 4162       local_jcc(Assembler::notEqual, *L_slow_path);
 4163       final_jmp(*L_success);
 4164     }
 4165   } else {
 4166     // No slow path; it's a fast decision.
 4167     if (L_failure == &L_fallthrough) {
 4168       local_jcc(Assembler::equal, *L_success);
 4169     } else {
 4170       local_jcc(Assembler::notEqual, *L_failure);
 4171       final_jmp(*L_success);
 4172     }
 4173   }
 4174 
 4175   bind(L_fallthrough);
 4176 
 4177 #undef local_jcc
 4178 #undef final_jmp
 4179 }
 4180 
 4181 
 4182 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 4183                                                           Register super_klass,
 4184                                                           Register temp_reg,
 4185                                                           Register temp2_reg,
 4186                                                           Label* L_success,
 4187                                                           Label* L_failure,
 4188                                                           bool set_cond_codes) {
 4189   assert_different_registers(sub_klass, super_klass, temp_reg);
 4190   if (temp2_reg != noreg)
 4191     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4192 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4193 
 4194   Label L_fallthrough;
 4195   int label_nulls = 0;
 4196   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4197   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4198   assert(label_nulls <= 1, "at most one null in the batch");
 4199 
 4200   // a couple of useful fields in sub_klass:
 4201   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4202   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4203   Address secondary_supers_addr(sub_klass, ss_offset);
 4204   Address super_cache_addr(     sub_klass, sc_offset);
 4205 
 4206   // Do a linear scan of the secondary super-klass chain.
 4207   // This code is rarely used, so simplicity is a virtue here.
 4208   // The repne_scan instruction uses fixed registers, which we must spill.
 4209   // Don't worry too much about pre-existing connections with the input regs.
 4210 
 4211   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4212   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4213 
 4214   // Get super_klass value into rax (even if it was in rdi or rcx).
 4215   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4216   if (super_klass != rax) {
 4217     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4218     mov(rax, super_klass);
 4219   }
 4220   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4221   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4222 
 4223 #ifndef PRODUCT
 4224   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4225   ExternalAddress pst_counter_addr((address) pst_counter);
 4226   lea(rcx, pst_counter_addr);
 4227   incrementl(Address(rcx, 0));
 4228 #endif //PRODUCT
 4229 
 4230   // We will consult the secondary-super array.
 4231   movptr(rdi, secondary_supers_addr);
 4232   // Load the array length.  (Positive movl does right thing on LP64.)
 4233   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4234   // Skip to start of data.
 4235   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4236 
 4237   // Scan RCX words at [RDI] for an occurrence of RAX.
 4238   // Set NZ/Z based on last compare.
 4239   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4240   // not change flags (only scas instruction which is repeated sets flags).
 4241   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4242 
 4243     testptr(rax,rax); // Set Z = 0
 4244     repne_scan();
 4245 
 4246   // Unspill the temp. registers:
 4247   if (pushed_rdi)  pop(rdi);
 4248   if (pushed_rcx)  pop(rcx);
 4249   if (pushed_rax)  pop(rax);
 4250 
 4251   if (set_cond_codes) {
 4252     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4253     assert(!pushed_rdi, "rdi must be left non-null");
 4254     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4255   }
 4256 
 4257   if (L_failure == &L_fallthrough)
 4258         jccb(Assembler::notEqual, *L_failure);
 4259   else  jcc(Assembler::notEqual, *L_failure);
 4260 
 4261   // Success.  Cache the super we found and proceed in triumph.
 4262   movptr(super_cache_addr, super_klass);
 4263 
 4264   if (L_success != &L_fallthrough) {
 4265     jmp(*L_success);
 4266   }
 4267 
 4268 #undef IS_A_TEMP
 4269 
 4270   bind(L_fallthrough);
 4271 }
 4272 
 4273 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4274                                                    Register super_klass,
 4275                                                    Register temp_reg,
 4276                                                    Register temp2_reg,
 4277                                                    Label* L_success,
 4278                                                    Label* L_failure,
 4279                                                    bool set_cond_codes) {
 4280   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4281   check_klass_subtype_slow_path
 4282     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4283      L_success, L_failure);
 4284 }
 4285 
 4286 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4287                                                    Register super_klass,
 4288                                                    Register temp_reg,
 4289                                                    Register temp2_reg,
 4290                                                    Register temp3_reg,
 4291                                                    Register temp4_reg,
 4292                                                    Label* L_success,
 4293                                                    Label* L_failure) {
 4294   if (UseSecondarySupersTable) {
 4295     check_klass_subtype_slow_path_table
 4296       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4297        L_success, L_failure);
 4298   } else {
 4299     check_klass_subtype_slow_path_linear
 4300       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4301   }
 4302 }
 4303 
 4304 Register MacroAssembler::allocate_if_noreg(Register r,
 4305                                   RegSetIterator<Register> &available_regs,
 4306                                   RegSet &regs_to_push) {
 4307   if (!r->is_valid()) {
 4308     r = *available_regs++;
 4309     regs_to_push += r;
 4310   }
 4311   return r;
 4312 }
 4313 
 4314 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4315                                                          Register super_klass,
 4316                                                          Register temp_reg,
 4317                                                          Register temp2_reg,
 4318                                                          Register temp3_reg,
 4319                                                          Register result_reg,
 4320                                                          Label* L_success,
 4321                                                          Label* L_failure) {
 4322   // NB! Callers may assume that, when temp2_reg is a valid register,
 4323   // this code sets it to a nonzero value.
 4324   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4325 
 4326   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4327 
 4328   Label L_fallthrough;
 4329   int label_nulls = 0;
 4330   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4331   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4332   assert(label_nulls <= 1, "at most one null in the batch");
 4333 
 4334   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4335 
 4336   RegSetIterator<Register> available_regs
 4337     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4338 
 4339   RegSet pushed_regs;
 4340 
 4341   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4342   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4343   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4344   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4345   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4346 
 4347   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4348 
 4349   {
 4350 
 4351     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4352     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4353     subptr(rsp, aligned_size);
 4354     push_set(pushed_regs, 0);
 4355 
 4356     lookup_secondary_supers_table_var(sub_klass,
 4357                                       super_klass,
 4358                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4359     cmpq(result_reg, 0);
 4360 
 4361     // Unspill the temp. registers:
 4362     pop_set(pushed_regs, 0);
 4363     // Increment SP but do not clobber flags.
 4364     lea(rsp, Address(rsp, aligned_size));
 4365   }
 4366 
 4367   if (temp2_reg_was_valid) {
 4368     movq(temp2_reg, 1);
 4369   }
 4370 
 4371   jcc(Assembler::notEqual, *L_failure);
 4372 
 4373   if (L_success != &L_fallthrough) {
 4374     jmp(*L_success);
 4375   }
 4376 
 4377   bind(L_fallthrough);
 4378 }
 4379 
 4380 // population_count variant for running without the POPCNT
 4381 // instruction, which was introduced with SSE4.2 in 2008.
 4382 void MacroAssembler::population_count(Register dst, Register src,
 4383                                       Register scratch1, Register scratch2) {
 4384   assert_different_registers(src, scratch1, scratch2);
 4385   if (UsePopCountInstruction) {
 4386     Assembler::popcntq(dst, src);
 4387   } else {
 4388     assert_different_registers(src, scratch1, scratch2);
 4389     assert_different_registers(dst, scratch1, scratch2);
 4390     Label loop, done;
 4391 
 4392     mov(scratch1, src);
 4393     // dst = 0;
 4394     // while(scratch1 != 0) {
 4395     //   dst++;
 4396     //   scratch1 &= (scratch1 - 1);
 4397     // }
 4398     xorl(dst, dst);
 4399     testq(scratch1, scratch1);
 4400     jccb(Assembler::equal, done);
 4401     {
 4402       bind(loop);
 4403       incq(dst);
 4404       movq(scratch2, scratch1);
 4405       decq(scratch2);
 4406       andq(scratch1, scratch2);
 4407       jccb(Assembler::notEqual, loop);
 4408     }
 4409     bind(done);
 4410   }
 4411 #ifdef ASSERT
 4412   mov64(scratch1, 0xCafeBabeDeadBeef);
 4413   movq(scratch2, scratch1);
 4414 #endif
 4415 }
 4416 
 4417 // Ensure that the inline code and the stub are using the same registers.
 4418 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4419 do {                                                                 \
 4420   assert(r_super_klass  == rax, "mismatch");                         \
 4421   assert(r_array_base   == rbx, "mismatch");                         \
 4422   assert(r_array_length == rcx, "mismatch");                         \
 4423   assert(r_array_index  == rdx, "mismatch");                         \
 4424   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4425   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4426   assert(result         == rdi || result      == noreg, "mismatch"); \
 4427 } while(0)
 4428 
 4429 // Versions of salq and rorq that don't need count to be in rcx
 4430 
 4431 void MacroAssembler::salq(Register dest, Register count) {
 4432   if (count == rcx) {
 4433     Assembler::salq(dest);
 4434   } else {
 4435     assert_different_registers(rcx, dest);
 4436     xchgq(rcx, count);
 4437     Assembler::salq(dest);
 4438     xchgq(rcx, count);
 4439   }
 4440 }
 4441 
 4442 void MacroAssembler::rorq(Register dest, Register count) {
 4443   if (count == rcx) {
 4444     Assembler::rorq(dest);
 4445   } else {
 4446     assert_different_registers(rcx, dest);
 4447     xchgq(rcx, count);
 4448     Assembler::rorq(dest);
 4449     xchgq(rcx, count);
 4450   }
 4451 }
 4452 
 4453 // Return true: we succeeded in generating this code
 4454 //
 4455 // At runtime, return 0 in result if r_super_klass is a superclass of
 4456 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4457 // super_klass_slot of the class you're looking for. This is always
 4458 // the case for instanceof and checkcast.
 4459 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4460                                                          Register r_super_klass,
 4461                                                          Register temp1,
 4462                                                          Register temp2,
 4463                                                          Register temp3,
 4464                                                          Register temp4,
 4465                                                          Register result,
 4466                                                          u1 super_klass_slot) {
 4467   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4468 
 4469   Label L_fallthrough, L_success, L_failure;
 4470 
 4471   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4472 
 4473   const Register
 4474     r_array_index  = temp1,
 4475     r_array_length = temp2,
 4476     r_array_base   = temp3,
 4477     r_bitmap       = temp4;
 4478 
 4479   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4480 
 4481   xorq(result, result); // = 0
 4482 
 4483   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4484   movq(r_array_index, r_bitmap);
 4485 
 4486   // First check the bitmap to see if super_klass might be present. If
 4487   // the bit is zero, we are certain that super_klass is not one of
 4488   // the secondary supers.
 4489   u1 bit = super_klass_slot;
 4490   {
 4491     // NB: If the count in a x86 shift instruction is 0, the flags are
 4492     // not affected, so we do a testq instead.
 4493     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4494     if (shift_count != 0) {
 4495       salq(r_array_index, shift_count);
 4496     } else {
 4497       testq(r_array_index, r_array_index);
 4498     }
 4499   }
 4500   // We test the MSB of r_array_index, i.e. its sign bit
 4501   jcc(Assembler::positive, L_failure);
 4502 
 4503   // Get the first array index that can contain super_klass into r_array_index.
 4504   if (bit != 0) {
 4505     population_count(r_array_index, r_array_index, temp2, temp3);
 4506   } else {
 4507     movl(r_array_index, 1);
 4508   }
 4509   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4510 
 4511   // We will consult the secondary-super array.
 4512   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4513 
 4514   // We're asserting that the first word in an Array<Klass*> is the
 4515   // length, and the second word is the first word of the data. If
 4516   // that ever changes, r_array_base will have to be adjusted here.
 4517   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4518   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4519 
 4520   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4521   jccb(Assembler::equal, L_success);
 4522 
 4523   // Is there another entry to check? Consult the bitmap.
 4524   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4525   jccb(Assembler::carryClear, L_failure);
 4526 
 4527   // Linear probe. Rotate the bitmap so that the next bit to test is
 4528   // in Bit 1.
 4529   if (bit != 0) {
 4530     rorq(r_bitmap, bit);
 4531   }
 4532 
 4533   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4534   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4535   // Kills: r_array_length.
 4536   // Returns: result.
 4537   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4538   // Result (0/1) is in rdi
 4539   jmpb(L_fallthrough);
 4540 
 4541   bind(L_failure);
 4542   incq(result); // 0 => 1
 4543 
 4544   bind(L_success);
 4545   // result = 0;
 4546 
 4547   bind(L_fallthrough);
 4548   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4549 
 4550   if (VerifySecondarySupers) {
 4551     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4552                                   temp1, temp2, temp3);
 4553   }
 4554 }
 4555 
 4556 // At runtime, return 0 in result if r_super_klass is a superclass of
 4557 // r_sub_klass, otherwise return nonzero. Use this version of
 4558 // lookup_secondary_supers_table() if you don't know ahead of time
 4559 // which superclass will be searched for. Used by interpreter and
 4560 // runtime stubs. It is larger and has somewhat greater latency than
 4561 // the version above, which takes a constant super_klass_slot.
 4562 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4563                                                        Register r_super_klass,
 4564                                                        Register temp1,
 4565                                                        Register temp2,
 4566                                                        Register temp3,
 4567                                                        Register temp4,
 4568                                                        Register result) {
 4569   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4570   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4571   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4572 
 4573   Label L_fallthrough, L_success, L_failure;
 4574 
 4575   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4576 
 4577   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4578 
 4579   // FIXME. Once we are sure that all paths reaching this point really
 4580   // do pass rcx as one of our temps we can get rid of the following
 4581   // workaround.
 4582   assert(temps.contains(rcx), "fix this code");
 4583 
 4584   // We prefer to have our shift count in rcx. If rcx is one of our
 4585   // temps, use it for slot. If not, pick any of our temps.
 4586   Register slot;
 4587   if (!temps.contains(rcx)) {
 4588     slot = *available_regs++;
 4589   } else {
 4590     slot = rcx;
 4591   }
 4592 
 4593   const Register r_array_index = *available_regs++;
 4594   const Register r_bitmap      = *available_regs++;
 4595 
 4596   // The logic above guarantees this property, but we state it here.
 4597   assert_different_registers(r_array_index, r_bitmap, rcx);
 4598 
 4599   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4600   movq(r_array_index, r_bitmap);
 4601 
 4602   // First check the bitmap to see if super_klass might be present. If
 4603   // the bit is zero, we are certain that super_klass is not one of
 4604   // the secondary supers.
 4605   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4606   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4607   salq(r_array_index, slot);
 4608 
 4609   testq(r_array_index, r_array_index);
 4610   // We test the MSB of r_array_index, i.e. its sign bit
 4611   jcc(Assembler::positive, L_failure);
 4612 
 4613   const Register r_array_base = *available_regs++;
 4614 
 4615   // Get the first array index that can contain super_klass into r_array_index.
 4616   // Note: Clobbers r_array_base and slot.
 4617   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4618 
 4619   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4620 
 4621   // We will consult the secondary-super array.
 4622   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4623 
 4624   // We're asserting that the first word in an Array<Klass*> is the
 4625   // length, and the second word is the first word of the data. If
 4626   // that ever changes, r_array_base will have to be adjusted here.
 4627   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4628   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4629 
 4630   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4631   jccb(Assembler::equal, L_success);
 4632 
 4633   // Restore slot to its true value
 4634   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4635 
 4636   // Linear probe. Rotate the bitmap so that the next bit to test is
 4637   // in Bit 1.
 4638   rorq(r_bitmap, slot);
 4639 
 4640   // Is there another entry to check? Consult the bitmap.
 4641   btq(r_bitmap, 1);
 4642   jccb(Assembler::carryClear, L_failure);
 4643 
 4644   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4645   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4646   // Kills: r_array_length.
 4647   // Returns: result.
 4648   lookup_secondary_supers_table_slow_path(r_super_klass,
 4649                                           r_array_base,
 4650                                           r_array_index,
 4651                                           r_bitmap,
 4652                                           /*temp1*/result,
 4653                                           /*temp2*/slot,
 4654                                           &L_success,
 4655                                           nullptr);
 4656 
 4657   bind(L_failure);
 4658   movq(result, 1);
 4659   jmpb(L_fallthrough);
 4660 
 4661   bind(L_success);
 4662   xorq(result, result); // = 0
 4663 
 4664   bind(L_fallthrough);
 4665   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4666 
 4667   if (VerifySecondarySupers) {
 4668     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4669                                   temp1, temp2, temp3);
 4670   }
 4671 }
 4672 
 4673 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4674                                  Label* L_success, Label* L_failure) {
 4675   Label L_loop, L_fallthrough;
 4676   {
 4677     int label_nulls = 0;
 4678     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4679     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4680     assert(label_nulls <= 1, "at most one null in the batch");
 4681   }
 4682   bind(L_loop);
 4683   cmpq(value, Address(addr, count, Address::times_8));
 4684   jcc(Assembler::equal, *L_success);
 4685   addl(count, 1);
 4686   cmpl(count, limit);
 4687   jcc(Assembler::less, L_loop);
 4688 
 4689   if (&L_fallthrough != L_failure) {
 4690     jmp(*L_failure);
 4691   }
 4692   bind(L_fallthrough);
 4693 }
 4694 
 4695 // Called by code generated by check_klass_subtype_slow_path
 4696 // above. This is called when there is a collision in the hashed
 4697 // lookup in the secondary supers array.
 4698 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4699                                                              Register r_array_base,
 4700                                                              Register r_array_index,
 4701                                                              Register r_bitmap,
 4702                                                              Register temp1,
 4703                                                              Register temp2,
 4704                                                              Label* L_success,
 4705                                                              Label* L_failure) {
 4706   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4707 
 4708   const Register
 4709     r_array_length = temp1,
 4710     r_sub_klass    = noreg,
 4711     result         = noreg;
 4712 
 4713   Label L_fallthrough;
 4714   int label_nulls = 0;
 4715   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4716   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4717   assert(label_nulls <= 1, "at most one null in the batch");
 4718 
 4719   // Load the array length.
 4720   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4721   // And adjust the array base to point to the data.
 4722   // NB! Effectively increments current slot index by 1.
 4723   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4724   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4725 
 4726   // Linear probe
 4727   Label L_huge;
 4728 
 4729   // The bitmap is full to bursting.
 4730   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4731   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4732   jcc(Assembler::greater, L_huge);
 4733 
 4734   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4735   // current slot (at secondary_supers[r_array_index]) has not yet
 4736   // been inspected, and r_array_index may be out of bounds if we
 4737   // wrapped around the end of the array.
 4738 
 4739   { // This is conventional linear probing, but instead of terminating
 4740     // when a null entry is found in the table, we maintain a bitmap
 4741     // in which a 0 indicates missing entries.
 4742     // The check above guarantees there are 0s in the bitmap, so the loop
 4743     // eventually terminates.
 4744 
 4745     xorl(temp2, temp2); // = 0;
 4746 
 4747     Label L_again;
 4748     bind(L_again);
 4749 
 4750     // Check for array wraparound.
 4751     cmpl(r_array_index, r_array_length);
 4752     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4753 
 4754     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4755     jcc(Assembler::equal, *L_success);
 4756 
 4757     // If the next bit in bitmap is zero, we're done.
 4758     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4759     jcc(Assembler::carryClear, *L_failure);
 4760 
 4761     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4762     addl(r_array_index, 1);
 4763 
 4764     jmp(L_again);
 4765   }
 4766 
 4767   { // Degenerate case: more than 64 secondary supers.
 4768     // FIXME: We could do something smarter here, maybe a vectorized
 4769     // comparison or a binary search, but is that worth any added
 4770     // complexity?
 4771     bind(L_huge);
 4772     xorl(r_array_index, r_array_index); // = 0
 4773     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4774                 L_success,
 4775                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4776 
 4777     bind(L_fallthrough);
 4778   }
 4779 }
 4780 
 4781 struct VerifyHelperArguments {
 4782   Klass* _super;
 4783   Klass* _sub;
 4784   intptr_t _linear_result;
 4785   intptr_t _table_result;
 4786 };
 4787 
 4788 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4789   Klass::on_secondary_supers_verification_failure(args->_super,
 4790                                                   args->_sub,
 4791                                                   args->_linear_result,
 4792                                                   args->_table_result,
 4793                                                   msg);
 4794 }
 4795 
 4796 // Make sure that the hashed lookup and a linear scan agree.
 4797 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4798                                                    Register r_super_klass,
 4799                                                    Register result,
 4800                                                    Register temp1,
 4801                                                    Register temp2,
 4802                                                    Register temp3) {
 4803   const Register
 4804       r_array_index  = temp1,
 4805       r_array_length = temp2,
 4806       r_array_base   = temp3,
 4807       r_bitmap       = noreg;
 4808 
 4809   BLOCK_COMMENT("verify_secondary_supers_table {");
 4810 
 4811   Label L_success, L_failure, L_check, L_done;
 4812 
 4813   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4814   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4815   // And adjust the array base to point to the data.
 4816   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4817 
 4818   testl(r_array_length, r_array_length); // array_length == 0?
 4819   jcc(Assembler::zero, L_failure);
 4820 
 4821   movl(r_array_index, 0);
 4822   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4823   // fall through to L_failure
 4824 
 4825   const Register linear_result = r_array_index; // reuse temp1
 4826 
 4827   bind(L_failure); // not present
 4828   movl(linear_result, 1);
 4829   jmp(L_check);
 4830 
 4831   bind(L_success); // present
 4832   movl(linear_result, 0);
 4833 
 4834   bind(L_check);
 4835   cmpl(linear_result, result);
 4836   jcc(Assembler::equal, L_done);
 4837 
 4838   { // To avoid calling convention issues, build a record on the stack
 4839     // and pass the pointer to that instead.
 4840     push(result);
 4841     push(linear_result);
 4842     push(r_sub_klass);
 4843     push(r_super_klass);
 4844     movptr(c_rarg1, rsp);
 4845     movptr(c_rarg0, (uintptr_t) "mismatch");
 4846     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4847     should_not_reach_here();
 4848   }
 4849   bind(L_done);
 4850 
 4851   BLOCK_COMMENT("} verify_secondary_supers_table");
 4852 }
 4853 
 4854 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4855 
 4856 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4857   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4858 
 4859   Label L_fallthrough;
 4860   if (L_fast_path == nullptr) {
 4861     L_fast_path = &L_fallthrough;
 4862   } else if (L_slow_path == nullptr) {
 4863     L_slow_path = &L_fallthrough;
 4864   }
 4865 
 4866   // Fast path check: class is fully initialized.
 4867   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4868   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4869   jcc(Assembler::equal, *L_fast_path);
 4870 
 4871   // Fast path check: current thread is initializer thread
 4872   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4873   if (L_slow_path == &L_fallthrough) {
 4874     jcc(Assembler::equal, *L_fast_path);
 4875     bind(*L_slow_path);
 4876   } else if (L_fast_path == &L_fallthrough) {
 4877     jcc(Assembler::notEqual, *L_slow_path);
 4878     bind(*L_fast_path);
 4879   } else {
 4880     Unimplemented();
 4881   }
 4882 }
 4883 
 4884 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4885   if (VM_Version::supports_cmov()) {
 4886     cmovl(cc, dst, src);
 4887   } else {
 4888     Label L;
 4889     jccb(negate_condition(cc), L);
 4890     movl(dst, src);
 4891     bind(L);
 4892   }
 4893 }
 4894 
 4895 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4896   if (VM_Version::supports_cmov()) {
 4897     cmovl(cc, dst, src);
 4898   } else {
 4899     Label L;
 4900     jccb(negate_condition(cc), L);
 4901     movl(dst, src);
 4902     bind(L);
 4903   }
 4904 }
 4905 
 4906 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4907   if (!VerifyOops || VerifyAdapterSharing) {
 4908     // Below address of the code string confuses VerifyAdapterSharing
 4909     // because it may differ between otherwise equivalent adapters.
 4910     return;
 4911   }
 4912 
 4913   BLOCK_COMMENT("verify_oop {");
 4914   push(rscratch1);
 4915   push(rax);                          // save rax
 4916   push(reg);                          // pass register argument
 4917 
 4918   // Pass register number to verify_oop_subroutine
 4919   const char* b = nullptr;
 4920   {
 4921     ResourceMark rm;
 4922     stringStream ss;
 4923     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4924     b = code_string(ss.as_string());
 4925   }
 4926   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4927   pushptr(buffer.addr(), rscratch1);
 4928 
 4929   // call indirectly to solve generation ordering problem
 4930   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4931   call(rax);
 4932   // Caller pops the arguments (oop, message) and restores rax, r10
 4933   BLOCK_COMMENT("} verify_oop");
 4934 }
 4935 
 4936 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 4937   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 4938     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 4939     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 4940     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 4941   } else if (VM_Version::supports_avx()) {
 4942     vpcmpeqd(dst, dst, dst, vector_len);
 4943   } else {
 4944     pcmpeqd(dst, dst);
 4945   }
 4946 }
 4947 
 4948 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 4949                                          int extra_slot_offset) {
 4950   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 4951   int stackElementSize = Interpreter::stackElementSize;
 4952   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 4953 #ifdef ASSERT
 4954   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 4955   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 4956 #endif
 4957   Register             scale_reg    = noreg;
 4958   Address::ScaleFactor scale_factor = Address::no_scale;
 4959   if (arg_slot.is_constant()) {
 4960     offset += arg_slot.as_constant() * stackElementSize;
 4961   } else {
 4962     scale_reg    = arg_slot.as_register();
 4963     scale_factor = Address::times(stackElementSize);
 4964   }
 4965   offset += wordSize;           // return PC is on stack
 4966   return Address(rsp, scale_reg, scale_factor, offset);
 4967 }
 4968 
 4969 // Handle the receiver type profile update given the "recv" klass.
 4970 //
 4971 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 4972 // If there are no matching or claimable receiver entries in RD, updates
 4973 // the polymorphic counter.
 4974 //
 4975 // This code expected to run by either the interpreter or JIT-ed code, without
 4976 // extra synchronization. For safety, receiver cells are claimed atomically, which
 4977 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 4978 // counter updates are not atomic.
 4979 //
 4980 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 4981   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 4982   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 4983   int poly_count_offset      = in_bytes(CounterData::count_offset());
 4984   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 4985   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 4986 
 4987   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 4988   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 4989   base_receiver_offset += mdp_offset;
 4990   end_receiver_offset  += mdp_offset;
 4991   poly_count_offset    += mdp_offset;
 4992 
 4993   // Scale down to optimize encoding. Slots are pointer-sized.
 4994   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 4995   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 4996   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 4997   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 4998   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 4999   base_receiver_offset   >>= LogBytesPerWord;
 5000   end_receiver_offset    >>= LogBytesPerWord;
 5001   poly_count_offset      >>= LogBytesPerWord;
 5002   receiver_step          >>= LogBytesPerWord;
 5003   receiver_to_count_step >>= LogBytesPerWord;
 5004 
 5005 #ifdef ASSERT
 5006   // We are about to walk the MDO slots without asking for offsets.
 5007   // Check that our math hits all the right spots.
 5008   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 5009     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 5010     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 5011     int offset = base_receiver_offset + receiver_step*c;
 5012     int count_offset = offset + receiver_to_count_step;
 5013     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 5014     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 5015   }
 5016   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 5017   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 5018 #endif
 5019 
 5020   // Corner case: no profile table. Increment poly counter and exit.
 5021   if (ReceiverTypeData::row_limit() == 0) {
 5022     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 5023     return;
 5024   }
 5025 
 5026   Register offset = rscratch1;
 5027 
 5028   Label L_loop_search_receiver, L_loop_search_empty;
 5029   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 5030 
 5031   // The code here recognizes three major cases:
 5032   //   A. Fastest: receiver found in the table
 5033   //   B. Fast: no receiver in the table, and the table is full
 5034   //   C. Slow: no receiver in the table, free slots in the table
 5035   //
 5036   // The case A performance is most important, as perfectly-behaved code would end up
 5037   // there, especially with larger TypeProfileWidth. The case B performance is
 5038   // important as well, this is where bulk of code would land for normally megamorphic
 5039   // cases. The case C performance is not essential, its job is to deal with installation
 5040   // races, we optimize for code density instead. Case C needs to make sure that receiver
 5041   // rows are only claimed once. This makes sure we never overwrite a row for another
 5042   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 5043   //
 5044   // It is very tempting to handle these cases in a single loop, and claim the first slot
 5045   // without checking the rest of the table. But, profiling code should tolerate free slots
 5046   // in the table, as class unloading can clear them. After such cleanup, the receiver
 5047   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 5048   // to complete, before trying to install new slots. Splitting the code in several tight
 5049   // loops also helpfully optimizes for cases A and B.
 5050   //
 5051   // This code is effectively:
 5052   //
 5053   // restart:
 5054   //   // Fastest: receiver is already installed
 5055   //   for (i = 0; i < receiver_count(); i++) {
 5056   //     if (receiver(i) == recv) goto found_recv(i);
 5057   //   }
 5058   //
 5059   //   // Fast: no receiver, but profile is full
 5060   //   for (i = 0; i < receiver_count(); i++) {
 5061   //     if (receiver(i) == null) goto found_null(i);
 5062   //   }
 5063   //   goto polymorphic
 5064   //
 5065   //   // Slow: try to install receiver
 5066   // found_null(i):
 5067   //   CAS(&receiver(i), null, recv);
 5068   //   goto restart
 5069   //
 5070   // polymorphic:
 5071   //   count++;
 5072   //   return
 5073   //
 5074   // found_recv(i):
 5075   //   *receiver_count(i)++
 5076   //
 5077 
 5078   bind(L_restart);
 5079 
 5080   // Fastest: receiver is already installed
 5081   movptr(offset, base_receiver_offset);
 5082   bind(L_loop_search_receiver);
 5083     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 5084     jccb(Assembler::equal, L_found_recv);
 5085   addptr(offset, receiver_step);
 5086   cmpptr(offset, end_receiver_offset);
 5087   jccb(Assembler::notEqual, L_loop_search_receiver);
 5088 
 5089   // Fast: no receiver, but profile is full
 5090   movptr(offset, base_receiver_offset);
 5091   bind(L_loop_search_empty);
 5092     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 5093     jccb(Assembler::equal, L_found_empty);
 5094   addptr(offset, receiver_step);
 5095   cmpptr(offset, end_receiver_offset);
 5096   jccb(Assembler::notEqual, L_loop_search_empty);
 5097   jmpb(L_polymorphic);
 5098 
 5099   // Slow: try to install receiver
 5100   bind(L_found_empty);
 5101 
 5102   // Atomically swing receiver slot: null -> recv.
 5103   //
 5104   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 5105   // other important registers untouched, as they form the address. Therefore, we need
 5106   // to shift any important registers from RAX into some other spare register. If we
 5107   // have a spare register, we are forced to save it on stack here.
 5108 
 5109   Register spare_reg = noreg;
 5110   Register shifted_mdp = mdp;
 5111   Register shifted_recv = recv;
 5112   if (recv == rax || mdp == rax) {
 5113     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 5114                 (recv != rcx && mdp != rcx) ? rcx :
 5115                 rdx;
 5116     assert_different_registers(mdp, recv, offset, spare_reg);
 5117 
 5118     push(spare_reg);
 5119     if (recv == rax) {
 5120       movptr(spare_reg, recv);
 5121       shifted_recv = spare_reg;
 5122     } else {
 5123       assert(mdp == rax, "Remaining case");
 5124       movptr(spare_reg, mdp);
 5125       shifted_mdp = spare_reg;
 5126     }
 5127   } else {
 5128     push(rax);
 5129   }
 5130 
 5131   // None of the important registers are in RAX after this shuffle.
 5132   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 5133 
 5134   xorptr(rax, rax);
 5135   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 5136 
 5137   // Unshift registers.
 5138   if (recv == rax || mdp == rax) {
 5139     movptr(rax, spare_reg);
 5140     pop(spare_reg);
 5141   } else {
 5142     pop(rax);
 5143   }
 5144 
 5145   // CAS success means the slot now has the receiver we want. CAS failure means
 5146   // something had claimed the slot concurrently: it can be the same receiver we want,
 5147   // or something else. Since this is a slow path, we can optimize for code density,
 5148   // and just restart the search from the beginning.
 5149   jmpb(L_restart);
 5150 
 5151   // Counter updates:
 5152 
 5153   // Increment polymorphic counter instead of receiver slot.
 5154   bind(L_polymorphic);
 5155   movptr(offset, poly_count_offset);
 5156   jmpb(L_count_update);
 5157 
 5158   // Found a receiver, convert its slot offset to corresponding count offset.
 5159   bind(L_found_recv);
 5160   addptr(offset, receiver_to_count_step);
 5161 
 5162   bind(L_count_update);
 5163   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 5164 }
 5165 
 5166 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5167   if (!VerifyOops || VerifyAdapterSharing) {
 5168     // Below address of the code string confuses VerifyAdapterSharing
 5169     // because it may differ between otherwise equivalent adapters.
 5170     return;
 5171   }
 5172 
 5173   push(rscratch1);
 5174   push(rax); // save rax,
 5175   // addr may contain rsp so we will have to adjust it based on the push
 5176   // we just did (and on 64 bit we do two pushes)
 5177   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5178   // stores rax into addr which is backwards of what was intended.
 5179   if (addr.uses(rsp)) {
 5180     lea(rax, addr);
 5181     pushptr(Address(rax, 2 * BytesPerWord));
 5182   } else {
 5183     pushptr(addr);
 5184   }
 5185 
 5186   // Pass register number to verify_oop_subroutine
 5187   const char* b = nullptr;
 5188   {
 5189     ResourceMark rm;
 5190     stringStream ss;
 5191     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 5192     b = code_string(ss.as_string());
 5193   }
 5194   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5195   pushptr(buffer.addr(), rscratch1);
 5196 
 5197   // call indirectly to solve generation ordering problem
 5198   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5199   call(rax);
 5200   // Caller pops the arguments (addr, message) and restores rax, r10.
 5201 }
 5202 
 5203 void MacroAssembler::verify_tlab() {
 5204 #ifdef ASSERT
 5205   if (UseTLAB && VerifyOops) {
 5206     Label next, ok;
 5207     Register t1 = rsi;
 5208 
 5209     push(t1);
 5210 
 5211     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5212     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 5213     jcc(Assembler::aboveEqual, next);
 5214     STOP("assert(top >= start)");
 5215     should_not_reach_here();
 5216 
 5217     bind(next);
 5218     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 5219     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5220     jcc(Assembler::aboveEqual, ok);
 5221     STOP("assert(top <= end)");
 5222     should_not_reach_here();
 5223 
 5224     bind(ok);
 5225     pop(t1);
 5226   }
 5227 #endif
 5228 }
 5229 
 5230 class ControlWord {
 5231  public:
 5232   int32_t _value;
 5233 
 5234   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5235   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5236   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5237   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5238   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5239   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5240   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5241   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5242 
 5243   void print() const {
 5244     // rounding control
 5245     const char* rc;
 5246     switch (rounding_control()) {
 5247       case 0: rc = "round near"; break;
 5248       case 1: rc = "round down"; break;
 5249       case 2: rc = "round up  "; break;
 5250       case 3: rc = "chop      "; break;
 5251       default:
 5252         rc = nullptr; // silence compiler warnings
 5253         fatal("Unknown rounding control: %d", rounding_control());
 5254     };
 5255     // precision control
 5256     const char* pc;
 5257     switch (precision_control()) {
 5258       case 0: pc = "24 bits "; break;
 5259       case 1: pc = "reserved"; break;
 5260       case 2: pc = "53 bits "; break;
 5261       case 3: pc = "64 bits "; break;
 5262       default:
 5263         pc = nullptr; // silence compiler warnings
 5264         fatal("Unknown precision control: %d", precision_control());
 5265     };
 5266     // flags
 5267     char f[9];
 5268     f[0] = ' ';
 5269     f[1] = ' ';
 5270     f[2] = (precision   ()) ? 'P' : 'p';
 5271     f[3] = (underflow   ()) ? 'U' : 'u';
 5272     f[4] = (overflow    ()) ? 'O' : 'o';
 5273     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5274     f[6] = (denormalized()) ? 'D' : 'd';
 5275     f[7] = (invalid     ()) ? 'I' : 'i';
 5276     f[8] = '\x0';
 5277     // output
 5278     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5279   }
 5280 
 5281 };
 5282 
 5283 class StatusWord {
 5284  public:
 5285   int32_t _value;
 5286 
 5287   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5288   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5289   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5290   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5291   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5292   int  top() const                     { return  (_value >> 11) & 7      ; }
 5293   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5294   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5295   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5296   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5297   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5298   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5299   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5300   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5301 
 5302   void print() const {
 5303     // condition codes
 5304     char c[5];
 5305     c[0] = (C3()) ? '3' : '-';
 5306     c[1] = (C2()) ? '2' : '-';
 5307     c[2] = (C1()) ? '1' : '-';
 5308     c[3] = (C0()) ? '0' : '-';
 5309     c[4] = '\x0';
 5310     // flags
 5311     char f[9];
 5312     f[0] = (error_status()) ? 'E' : '-';
 5313     f[1] = (stack_fault ()) ? 'S' : '-';
 5314     f[2] = (precision   ()) ? 'P' : '-';
 5315     f[3] = (underflow   ()) ? 'U' : '-';
 5316     f[4] = (overflow    ()) ? 'O' : '-';
 5317     f[5] = (zero_divide ()) ? 'Z' : '-';
 5318     f[6] = (denormalized()) ? 'D' : '-';
 5319     f[7] = (invalid     ()) ? 'I' : '-';
 5320     f[8] = '\x0';
 5321     // output
 5322     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5323   }
 5324 
 5325 };
 5326 
 5327 class TagWord {
 5328  public:
 5329   int32_t _value;
 5330 
 5331   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5332 
 5333   void print() const {
 5334     printf("%04x", _value & 0xFFFF);
 5335   }
 5336 
 5337 };
 5338 
 5339 class FPU_Register {
 5340  public:
 5341   int32_t _m0;
 5342   int32_t _m1;
 5343   int16_t _ex;
 5344 
 5345   bool is_indefinite() const           {
 5346     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5347   }
 5348 
 5349   void print() const {
 5350     char  sign = (_ex < 0) ? '-' : '+';
 5351     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5352     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5353   };
 5354 
 5355 };
 5356 
 5357 class FPU_State {
 5358  public:
 5359   enum {
 5360     register_size       = 10,
 5361     number_of_registers =  8,
 5362     register_mask       =  7
 5363   };
 5364 
 5365   ControlWord  _control_word;
 5366   StatusWord   _status_word;
 5367   TagWord      _tag_word;
 5368   int32_t      _error_offset;
 5369   int32_t      _error_selector;
 5370   int32_t      _data_offset;
 5371   int32_t      _data_selector;
 5372   int8_t       _register[register_size * number_of_registers];
 5373 
 5374   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5375   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5376 
 5377   const char* tag_as_string(int tag) const {
 5378     switch (tag) {
 5379       case 0: return "valid";
 5380       case 1: return "zero";
 5381       case 2: return "special";
 5382       case 3: return "empty";
 5383     }
 5384     ShouldNotReachHere();
 5385     return nullptr;
 5386   }
 5387 
 5388   void print() const {
 5389     // print computation registers
 5390     { int t = _status_word.top();
 5391       for (int i = 0; i < number_of_registers; i++) {
 5392         int j = (i - t) & register_mask;
 5393         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5394         st(j)->print();
 5395         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5396       }
 5397     }
 5398     printf("\n");
 5399     // print control registers
 5400     printf("ctrl = "); _control_word.print(); printf("\n");
 5401     printf("stat = "); _status_word .print(); printf("\n");
 5402     printf("tags = "); _tag_word    .print(); printf("\n");
 5403   }
 5404 
 5405 };
 5406 
 5407 class Flag_Register {
 5408  public:
 5409   int32_t _value;
 5410 
 5411   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5412   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5413   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5414   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5415   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5416   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5417   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5418 
 5419   void print() const {
 5420     // flags
 5421     char f[8];
 5422     f[0] = (overflow       ()) ? 'O' : '-';
 5423     f[1] = (direction      ()) ? 'D' : '-';
 5424     f[2] = (sign           ()) ? 'S' : '-';
 5425     f[3] = (zero           ()) ? 'Z' : '-';
 5426     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5427     f[5] = (parity         ()) ? 'P' : '-';
 5428     f[6] = (carry          ()) ? 'C' : '-';
 5429     f[7] = '\x0';
 5430     // output
 5431     printf("%08x  flags = %s", _value, f);
 5432   }
 5433 
 5434 };
 5435 
 5436 class IU_Register {
 5437  public:
 5438   int32_t _value;
 5439 
 5440   void print() const {
 5441     printf("%08x  %11d", _value, _value);
 5442   }
 5443 
 5444 };
 5445 
 5446 class IU_State {
 5447  public:
 5448   Flag_Register _eflags;
 5449   IU_Register   _rdi;
 5450   IU_Register   _rsi;
 5451   IU_Register   _rbp;
 5452   IU_Register   _rsp;
 5453   IU_Register   _rbx;
 5454   IU_Register   _rdx;
 5455   IU_Register   _rcx;
 5456   IU_Register   _rax;
 5457 
 5458   void print() const {
 5459     // computation registers
 5460     printf("rax,  = "); _rax.print(); printf("\n");
 5461     printf("rbx,  = "); _rbx.print(); printf("\n");
 5462     printf("rcx  = "); _rcx.print(); printf("\n");
 5463     printf("rdx  = "); _rdx.print(); printf("\n");
 5464     printf("rdi  = "); _rdi.print(); printf("\n");
 5465     printf("rsi  = "); _rsi.print(); printf("\n");
 5466     printf("rbp,  = "); _rbp.print(); printf("\n");
 5467     printf("rsp  = "); _rsp.print(); printf("\n");
 5468     printf("\n");
 5469     // control registers
 5470     printf("flgs = "); _eflags.print(); printf("\n");
 5471   }
 5472 };
 5473 
 5474 
 5475 class CPU_State {
 5476  public:
 5477   FPU_State _fpu_state;
 5478   IU_State  _iu_state;
 5479 
 5480   void print() const {
 5481     printf("--------------------------------------------------\n");
 5482     _iu_state .print();
 5483     printf("\n");
 5484     _fpu_state.print();
 5485     printf("--------------------------------------------------\n");
 5486   }
 5487 
 5488 };
 5489 
 5490 
 5491 static void _print_CPU_state(CPU_State* state) {
 5492   state->print();
 5493 };
 5494 
 5495 
 5496 void MacroAssembler::print_CPU_state() {
 5497   push_CPU_state();
 5498   push(rsp);                // pass CPU state
 5499   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5500   addptr(rsp, wordSize);       // discard argument
 5501   pop_CPU_state();
 5502 }
 5503 
 5504 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5505   // Either restore the MXCSR register after returning from the JNI Call
 5506   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5507   if (VM_Version::supports_sse()) {
 5508     if (RestoreMXCSROnJNICalls) {
 5509       ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5510     } else if (CheckJNICalls) {
 5511       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5512     }
 5513   }
 5514   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5515   vzeroupper();
 5516 }
 5517 
 5518 // ((OopHandle)result).resolve();
 5519 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5520   assert_different_registers(result, tmp);
 5521 
 5522   // Only 64 bit platforms support GCs that require a tmp register
 5523   // Only IN_HEAP loads require a thread_tmp register
 5524   // OopHandle::resolve is an indirection like jobject.
 5525   access_load_at(T_OBJECT, IN_NATIVE,
 5526                  result, Address(result, 0), tmp);
 5527 }
 5528 
 5529 // ((WeakHandle)result).resolve();
 5530 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5531   assert_different_registers(rresult, rtmp);
 5532   Label resolved;
 5533 
 5534   // A null weak handle resolves to null.
 5535   cmpptr(rresult, 0);
 5536   jcc(Assembler::equal, resolved);
 5537 
 5538   // Only 64 bit platforms support GCs that require a tmp register
 5539   // Only IN_HEAP loads require a thread_tmp register
 5540   // WeakHandle::resolve is an indirection like jweak.
 5541   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5542                  rresult, Address(rresult, 0), rtmp);
 5543   bind(resolved);
 5544 }
 5545 
 5546 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5547   // get mirror
 5548   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5549   load_method_holder(mirror, method);
 5550   movptr(mirror, Address(mirror, mirror_offset));
 5551   resolve_oop_handle(mirror, tmp);
 5552 }
 5553 
 5554 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5555   load_method_holder(rresult, rmethod);
 5556   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5557 }
 5558 
 5559 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5560   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5561   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5562   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5563 }
 5564 
 5565 void MacroAssembler::load_metadata(Register dst, Register src) {
 5566   if (UseCompactObjectHeaders) {
 5567     load_narrow_klass_compact(dst, src);
 5568   } else {
 5569     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5570   }
 5571 }
 5572 
 5573 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5574   assert(UseCompactObjectHeaders, "expect compact object headers");
 5575   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5576   shrq(dst, markWord::klass_shift);
 5577 }
 5578 
 5579 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5580   assert_different_registers(src, tmp);
 5581   assert_different_registers(dst, tmp);
 5582 
 5583   if (UseCompactObjectHeaders) {
 5584     load_narrow_klass_compact(dst, src);
 5585     decode_klass_not_null(dst, tmp);
 5586   } else {
 5587     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5588     decode_klass_not_null(dst, tmp);
 5589   }
 5590 }
 5591 
 5592 void MacroAssembler::load_prototype_header(Register dst, Register src, Register tmp) {
 5593   load_klass(dst, src, tmp);
 5594   movptr(dst, Address(dst, Klass::prototype_header_offset()));
 5595 }
 5596 
 5597 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5598   assert(!UseCompactObjectHeaders, "not with compact headers");
 5599   assert_different_registers(src, tmp);
 5600   assert_different_registers(dst, tmp);
 5601   encode_klass_not_null(src, tmp);
 5602   movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5603 }
 5604 
 5605 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5606   if (UseCompactObjectHeaders) {
 5607     assert(tmp != noreg, "need tmp");
 5608     assert_different_registers(klass, obj, tmp);
 5609     load_narrow_klass_compact(tmp, obj);
 5610     cmpl(klass, tmp);
 5611   } else {
 5612     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5613   }
 5614 }
 5615 
 5616 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5617   if (UseCompactObjectHeaders) {
 5618     assert(tmp2 != noreg, "need tmp2");
 5619     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5620     load_narrow_klass_compact(tmp1, obj1);
 5621     load_narrow_klass_compact(tmp2, obj2);
 5622     cmpl(tmp1, tmp2);
 5623   } else {
 5624     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5625     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5626   }
 5627 }
 5628 
 5629 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5630                                     Register tmp1) {
 5631   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5632   decorators = AccessInternal::decorator_fixup(decorators, type);
 5633   bool as_raw = (decorators & AS_RAW) != 0;
 5634   if (as_raw) {
 5635     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5636   } else {
 5637     bs->load_at(this, decorators, type, dst, src, tmp1);
 5638   }
 5639 }
 5640 
 5641 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5642                                      Register tmp1, Register tmp2, Register tmp3) {
 5643   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5644   decorators = AccessInternal::decorator_fixup(decorators, type);
 5645   bool as_raw = (decorators & AS_RAW) != 0;
 5646   if (as_raw) {
 5647     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5648   } else {
 5649     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5650   }
 5651 }
 5652 
 5653 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
 5654                                      Register inline_layout_info) {
 5655   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5656   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
 5657 }
 5658 
 5659 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
 5660   movptr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
 5661   movl(offset, Address(offset, InlineKlass::payload_offset_offset()));
 5662 }
 5663 
 5664 void MacroAssembler::payload_addr(Register oop, Register data, Register inline_klass) {
 5665   // ((address) (void*) o) + vk->payload_offset();
 5666   Register offset = (data == oop) ? rscratch1 : data;
 5667   payload_offset(inline_klass, offset);
 5668   if (data == oop) {
 5669     addptr(data, offset);
 5670   } else {
 5671     lea(data, Address(oop, offset));
 5672   }
 5673 }
 5674 
 5675 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
 5676                                                 Register index, Register data) {
 5677   assert(index != rcx, "index needs to shift by rcx");
 5678   assert_different_registers(array, array_klass, index);
 5679   assert_different_registers(rcx, array, index);
 5680 
 5681   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
 5682   movl(rcx, Address(array_klass, Klass::layout_helper_offset()));
 5683 
 5684   // Klass::layout_helper_log2_element_size(lh)
 5685   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
 5686   shrl(rcx, Klass::_lh_log2_element_size_shift);
 5687   andl(rcx, Klass::_lh_log2_element_size_mask);
 5688   shlptr(index); // index << rcx
 5689 
 5690   lea(data, Address(array, index, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT)));
 5691 }
 5692 
 5693 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5694   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5695 }
 5696 
 5697 // Doesn't do verification, generates fixed size code
 5698 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5699   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5700 }
 5701 
 5702 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5703                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5704   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5705 }
 5706 
 5707 // Used for storing nulls.
 5708 void MacroAssembler::store_heap_oop_null(Address dst) {
 5709   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5710 }
 5711 
 5712 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5713   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5714   // Store to klass gap in destination
 5715   movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5716 }
 5717 
 5718 #ifdef ASSERT
 5719 void MacroAssembler::verify_heapbase(const char* msg) {
 5720   assert (UseCompressedOops, "should be compressed");
 5721   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5722   if (CheckCompressedOops) {
 5723     Label ok;
 5724     ExternalAddress src2(CompressedOops::base_addr());
 5725     const bool is_src2_reachable = reachable(src2);
 5726     if (!is_src2_reachable) {
 5727       push(rscratch1);  // cmpptr trashes rscratch1
 5728     }
 5729     cmpptr(r12_heapbase, src2, rscratch1);
 5730     jcc(Assembler::equal, ok);
 5731     STOP(msg);
 5732     bind(ok);
 5733     if (!is_src2_reachable) {
 5734       pop(rscratch1);
 5735     }
 5736   }
 5737 }
 5738 #endif
 5739 
 5740 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5741 void MacroAssembler::encode_heap_oop(Register r) {
 5742 #ifdef ASSERT
 5743   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5744 #endif
 5745   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5746   if (CompressedOops::base() == nullptr) {
 5747     if (CompressedOops::shift() != 0) {
 5748       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5749       shrq(r, LogMinObjAlignmentInBytes);
 5750     }
 5751     return;
 5752   }
 5753   testq(r, r);
 5754   cmovq(Assembler::equal, r, r12_heapbase);
 5755   subq(r, r12_heapbase);
 5756   shrq(r, LogMinObjAlignmentInBytes);
 5757 }
 5758 
 5759 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5760 #ifdef ASSERT
 5761   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5762   if (CheckCompressedOops) {
 5763     Label ok;
 5764     testq(r, r);
 5765     jcc(Assembler::notEqual, ok);
 5766     STOP("null oop passed to encode_heap_oop_not_null");
 5767     bind(ok);
 5768   }
 5769 #endif
 5770   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5771   if (CompressedOops::base() != nullptr) {
 5772     subq(r, r12_heapbase);
 5773   }
 5774   if (CompressedOops::shift() != 0) {
 5775     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5776     shrq(r, LogMinObjAlignmentInBytes);
 5777   }
 5778 }
 5779 
 5780 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5781 #ifdef ASSERT
 5782   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5783   if (CheckCompressedOops) {
 5784     Label ok;
 5785     testq(src, src);
 5786     jcc(Assembler::notEqual, ok);
 5787     STOP("null oop passed to encode_heap_oop_not_null2");
 5788     bind(ok);
 5789   }
 5790 #endif
 5791   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5792   if (dst != src) {
 5793     movq(dst, src);
 5794   }
 5795   if (CompressedOops::base() != nullptr) {
 5796     subq(dst, r12_heapbase);
 5797   }
 5798   if (CompressedOops::shift() != 0) {
 5799     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5800     shrq(dst, LogMinObjAlignmentInBytes);
 5801   }
 5802 }
 5803 
 5804 void  MacroAssembler::decode_heap_oop(Register r) {
 5805 #ifdef ASSERT
 5806   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5807 #endif
 5808   if (CompressedOops::base() == nullptr) {
 5809     if (CompressedOops::shift() != 0) {
 5810       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5811       shlq(r, LogMinObjAlignmentInBytes);
 5812     }
 5813   } else {
 5814     Label done;
 5815     shlq(r, LogMinObjAlignmentInBytes);
 5816     jccb(Assembler::equal, done);
 5817     addq(r, r12_heapbase);
 5818     bind(done);
 5819   }
 5820   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5821 }
 5822 
 5823 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5824   // Note: it will change flags
 5825   assert (UseCompressedOops, "should only be used for compressed headers");
 5826   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5827   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5828   // vtableStubs also counts instructions in pd_code_size_limit.
 5829   // Also do not verify_oop as this is called by verify_oop.
 5830   if (CompressedOops::shift() != 0) {
 5831     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5832     shlq(r, LogMinObjAlignmentInBytes);
 5833     if (CompressedOops::base() != nullptr) {
 5834       addq(r, r12_heapbase);
 5835     }
 5836   } else {
 5837     assert (CompressedOops::base() == nullptr, "sanity");
 5838   }
 5839 }
 5840 
 5841 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5842   // Note: it will change flags
 5843   assert (UseCompressedOops, "should only be used for compressed headers");
 5844   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5845   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5846   // vtableStubs also counts instructions in pd_code_size_limit.
 5847   // Also do not verify_oop as this is called by verify_oop.
 5848   if (CompressedOops::shift() != 0) {
 5849     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5850     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5851       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5852     } else {
 5853       if (dst != src) {
 5854         movq(dst, src);
 5855       }
 5856       shlq(dst, LogMinObjAlignmentInBytes);
 5857       if (CompressedOops::base() != nullptr) {
 5858         addq(dst, r12_heapbase);
 5859       }
 5860     }
 5861   } else {
 5862     assert (CompressedOops::base() == nullptr, "sanity");
 5863     if (dst != src) {
 5864       movq(dst, src);
 5865     }
 5866   }
 5867 }
 5868 
 5869 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5870   BLOCK_COMMENT("encode_klass_not_null {");
 5871   assert_different_registers(r, tmp);
 5872   if (CompressedKlassPointers::base() != nullptr) {
 5873     if (AOTCodeCache::is_on_for_dump()) {
 5874       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5875     } else {
 5876       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5877     }
 5878     subq(r, tmp);
 5879   }
 5880   if (CompressedKlassPointers::shift() != 0) {
 5881     shrq(r, CompressedKlassPointers::shift());
 5882   }
 5883   BLOCK_COMMENT("} encode_klass_not_null");
 5884 }
 5885 
 5886 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5887   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5888   assert_different_registers(src, dst);
 5889   if (CompressedKlassPointers::base() != nullptr) {
 5890     if (AOTCodeCache::is_on_for_dump()) {
 5891       movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5892       negq(dst);
 5893     } else {
 5894       movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5895     }
 5896     addq(dst, src);
 5897   } else {
 5898     movptr(dst, src);
 5899   }
 5900   if (CompressedKlassPointers::shift() != 0) {
 5901     shrq(dst, CompressedKlassPointers::shift());
 5902   }
 5903   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5904 }
 5905 
 5906 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5907   BLOCK_COMMENT("decode_klass_not_null {");
 5908   assert_different_registers(r, tmp);
 5909   // Note: it will change flags
 5910   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5911   // vtableStubs also counts instructions in pd_code_size_limit.
 5912   // Also do not verify_oop as this is called by verify_oop.
 5913   if (CompressedKlassPointers::shift() != 0) {
 5914     shlq(r, CompressedKlassPointers::shift());
 5915   }
 5916   if (CompressedKlassPointers::base() != nullptr) {
 5917     if (AOTCodeCache::is_on_for_dump()) {
 5918       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5919     } else {
 5920       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5921     }
 5922     addq(r, tmp);
 5923   }
 5924   BLOCK_COMMENT("} decode_klass_not_null");
 5925 }
 5926 
 5927 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5928   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 5929   assert_different_registers(src, dst);
 5930   // Note: it will change flags
 5931   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5932   // vtableStubs also counts instructions in pd_code_size_limit.
 5933   // Also do not verify_oop as this is called by verify_oop.
 5934 
 5935   if (CompressedKlassPointers::base() == nullptr &&
 5936       CompressedKlassPointers::shift() == 0) {
 5937     // The best case scenario is that there is no base or shift. Then it is already
 5938     // a pointer that needs nothing but a register rename.
 5939     movl(dst, src);
 5940   } else {
 5941     if (CompressedKlassPointers::shift() <= Address::times_8) {
 5942       if (CompressedKlassPointers::base() != nullptr) {
 5943         if (AOTCodeCache::is_on_for_dump()) {
 5944           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5945         } else {
 5946           movptr(dst, (intptr_t)CompressedKlassPointers::base());
 5947         }
 5948       } else {
 5949         xorq(dst, dst);
 5950       }
 5951       if (CompressedKlassPointers::shift() != 0) {
 5952         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 5953         leaq(dst, Address(dst, src, Address::times_8, 0));
 5954       } else {
 5955         addq(dst, src);
 5956       }
 5957     } else {
 5958       if (CompressedKlassPointers::base() != nullptr) {
 5959         if (AOTCodeCache::is_on_for_dump()) {
 5960           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5961           shrq(dst, CompressedKlassPointers::shift());
 5962         } else {
 5963           const intptr_t base_right_shifted =
 5964                (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 5965           movptr(dst, base_right_shifted);
 5966         }
 5967       } else {
 5968         xorq(dst, dst);
 5969       }
 5970       addq(dst, src);
 5971       shlq(dst, CompressedKlassPointers::shift());
 5972     }
 5973   }
 5974   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 5975 }
 5976 
 5977 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5978   assert (UseCompressedOops, "should only be used for compressed headers");
 5979   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5980   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5981   int oop_index = oop_recorder()->find_index(obj);
 5982   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5983   mov_narrow_oop(dst, oop_index, rspec);
 5984 }
 5985 
 5986 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5987   assert (UseCompressedOops, "should only be used for compressed headers");
 5988   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5989   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5990   int oop_index = oop_recorder()->find_index(obj);
 5991   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5992   mov_narrow_oop(dst, oop_index, rspec);
 5993 }
 5994 
 5995 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5996   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5997   int klass_index = oop_recorder()->find_index(k);
 5998   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5999   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6000 }
 6001 
 6002 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 6003   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6004   int klass_index = oop_recorder()->find_index(k);
 6005   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6006   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6007 }
 6008 
 6009 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 6010   assert (UseCompressedOops, "should only be used for compressed headers");
 6011   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6012   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6013   int oop_index = oop_recorder()->find_index(obj);
 6014   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6015   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 6016 }
 6017 
 6018 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 6019   assert (UseCompressedOops, "should only be used for compressed headers");
 6020   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6021   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6022   int oop_index = oop_recorder()->find_index(obj);
 6023   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6024   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 6025 }
 6026 
 6027 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 6028   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6029   int klass_index = oop_recorder()->find_index(k);
 6030   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6031   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6032 }
 6033 
 6034 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 6035   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6036   int klass_index = oop_recorder()->find_index(k);
 6037   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6038   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6039 }
 6040 
 6041 void MacroAssembler::reinit_heapbase() {
 6042   if (UseCompressedOops) {
 6043     if (Universe::heap() != nullptr && !AOTCodeCache::is_on_for_dump()) {
 6044       if (CompressedOops::base() == nullptr) {
 6045         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 6046       } else {
 6047         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 6048       }
 6049     } else {
 6050       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 6051     }
 6052   }
 6053 }
 6054 
 6055 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
 6056   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
 6057   // An inline type might be returned. If fields are in registers we
 6058   // need to allocate an inline type instance and initialize it with
 6059   // the value of the fields.
 6060   Label skip;
 6061   // We only need a new buffered inline type if a new one is not returned
 6062   testptr(rax, 1);
 6063   jcc(Assembler::zero, skip);
 6064   int call_offset = -1;
 6065 
 6066 #ifdef _LP64
 6067   // The following code is similar to allocation code in TemplateTable::_new but has some slight differences,
 6068   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
 6069   // allocating is not necessary if vk != nullptr, etc.
 6070   Label slow_case;
 6071   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
 6072   mov(rscratch1, rax); // save rax for slow_case since *_allocate may corrupt it when allocation failed
 6073   if (vk != nullptr) {
 6074     // Called from C1, where the return type is statically known.
 6075     movptr(rbx, (intptr_t)vk->get_InlineKlass());
 6076     jint lh = vk->layout_helper();
 6077     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
 6078     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
 6079       tlab_allocate(rax, noreg, lh, r13, r14, slow_case);
 6080     } else {
 6081       jmp(slow_case);
 6082     }
 6083   } else {
 6084     // Call from interpreter. RAX contains ((the InlineKlass* of the return type) | 0x01)
 6085     mov(rbx, rax);
 6086     andptr(rbx, -2);
 6087     if (UseTLAB) {
 6088       movl(r14, Address(rbx, Klass::layout_helper_offset()));
 6089       testl(r14, Klass::_lh_instance_slow_path_bit);
 6090       jcc(Assembler::notZero, slow_case);
 6091       tlab_allocate(rax, r14, 0, r13, r14, slow_case);
 6092     } else {
 6093       jmp(slow_case);
 6094     }
 6095   }
 6096   if (UseTLAB) {
 6097     // 2. Initialize buffered inline instance header
 6098     Register buffer_obj = rax;
 6099     Register klass = rbx;
 6100     if (UseCompactObjectHeaders) {
 6101       Register mark_word = r13;
 6102       movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 6103       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), mark_word);
 6104     } else {
 6105       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), (intptr_t)markWord::inline_type_prototype().value());
 6106       xorl(r13, r13);
 6107       store_klass_gap(buffer_obj, r13);
 6108       if (vk == nullptr) {
 6109         // store_klass corrupts rbx(klass), so save it in r13 for later use (interpreter case only).
 6110         mov(r13, klass);
 6111       }
 6112       store_klass(buffer_obj, klass, rscratch1);
 6113       klass = r13;
 6114     }
 6115     // 3. Initialize its fields with an inline class specific handler
 6116     if (vk != nullptr) {
 6117       call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
 6118     } else {
 6119       movptr(rbx, Address(klass, InlineKlass::adr_members_offset()));
 6120       movptr(rbx, Address(rbx, InlineKlass::pack_handler_offset()));
 6121       call(rbx);
 6122     }
 6123     jmp(skip);
 6124   }
 6125   bind(slow_case);
 6126   // We failed to allocate a new inline type, fall back to a runtime
 6127   // call. Some oop field may be live in some registers but we can't
 6128   // tell. That runtime call will take care of preserving them
 6129   // across a GC if there's one.
 6130   mov(rax, rscratch1);
 6131 #endif
 6132 
 6133   if (from_interpreter) {
 6134     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
 6135   } else {
 6136     call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
 6137     call_offset = offset();
 6138   }
 6139 
 6140   bind(skip);
 6141   return call_offset;
 6142 }
 6143 
 6144 // Move a value between registers/stack slots and update the reg_state
 6145 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
 6146   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
 6147   if (reg_state[to->value()] == reg_written) {
 6148     return true; // Already written
 6149   }
 6150   if (from != to && bt != T_VOID) {
 6151     if (reg_state[to->value()] == reg_readonly) {
 6152       return false; // Not yet writable
 6153     }
 6154     if (from->is_reg()) {
 6155       if (to->is_reg()) {
 6156         if (from->is_XMMRegister()) {
 6157           if (bt == T_DOUBLE) {
 6158             movdbl(to->as_XMMRegister(), from->as_XMMRegister());
 6159           } else {
 6160             assert(bt == T_FLOAT, "must be float");
 6161             movflt(to->as_XMMRegister(), from->as_XMMRegister());
 6162           }
 6163         } else {
 6164           movq(to->as_Register(), from->as_Register());
 6165         }
 6166       } else {
 6167         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6168         Address to_addr = Address(rsp, st_off);
 6169         if (from->is_XMMRegister()) {
 6170           if (bt == T_DOUBLE) {
 6171             movdbl(to_addr, from->as_XMMRegister());
 6172           } else {
 6173             assert(bt == T_FLOAT, "must be float");
 6174             movflt(to_addr, from->as_XMMRegister());
 6175           }
 6176         } else {
 6177           movq(to_addr, from->as_Register());
 6178         }
 6179       }
 6180     } else {
 6181       Address from_addr = Address(rsp, from->reg2stack() * VMRegImpl::stack_slot_size + wordSize);
 6182       if (to->is_reg()) {
 6183         if (to->is_XMMRegister()) {
 6184           if (bt == T_DOUBLE) {
 6185             movdbl(to->as_XMMRegister(), from_addr);
 6186           } else {
 6187             assert(bt == T_FLOAT, "must be float");
 6188             movflt(to->as_XMMRegister(), from_addr);
 6189           }
 6190         } else {
 6191           movq(to->as_Register(), from_addr);
 6192         }
 6193       } else {
 6194         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6195         movq(r13, from_addr);
 6196         movq(Address(rsp, st_off), r13);
 6197       }
 6198     }
 6199   }
 6200   // Update register states
 6201   reg_state[from->value()] = reg_writable;
 6202   reg_state[to->value()] = reg_written;
 6203   return true;
 6204 }
 6205 
 6206 // Calculate the extra stack space required for packing or unpacking inline
 6207 // args and adjust the stack pointer (see MacroAssembler::remove_frame).
 6208 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
 6209   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
 6210   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
 6211   assert(sp_inc > 0, "sanity");
 6212   // Two additional slots to account for return address
 6213   sp_inc +=  2 * VMRegImpl::stack_slot_size;
 6214 
 6215   push(rbp);
 6216   subptr(rsp, sp_inc);
 6217 #ifdef ASSERT
 6218   movl(Address(rsp, 0), badRegWordVal);
 6219   movl(Address(rsp, VMRegImpl::stack_slot_size), badRegWordVal);
 6220 #endif
 6221   return sp_inc + wordSize; // account for rbp space
 6222 }
 6223 
 6224 // Read all fields from an inline type buffer and store the field values in registers/stack slots.
 6225 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
 6226                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
 6227                                           RegState reg_state[]) {
 6228   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
 6229   assert(from->is_valid(), "source must be valid");
 6230   bool progress = false;
 6231 #ifdef ASSERT
 6232   const int start_offset = offset();
 6233 #endif
 6234 
 6235   Label L_null, L_notNull;
 6236   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
 6237   Register tmp1 = r10;
 6238   Register tmp2 = r13;
 6239   Register fromReg = noreg;
 6240   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
 6241   bool done = true;
 6242   bool mark_done = true;
 6243   VMReg toReg;
 6244   BasicType bt;
 6245   // Check if argument requires a null check
 6246   bool null_check = false;
 6247   VMReg nullCheckReg;
 6248   while (stream.next(nullCheckReg, bt)) {
 6249     if (sig->at(stream.sig_index())._offset == -1) {
 6250       null_check = true;
 6251       break;
 6252     }
 6253   }
 6254   stream.reset(sig_index, to_index);
 6255   while (stream.next(toReg, bt)) {
 6256     assert(toReg->is_valid(), "destination must be valid");
 6257     int idx = (int)toReg->value();
 6258     if (reg_state[idx] == reg_readonly) {
 6259       if (idx != from->value()) {
 6260         mark_done = false;
 6261       }
 6262       done = false;
 6263       continue;
 6264     } else if (reg_state[idx] == reg_written) {
 6265       continue;
 6266     }
 6267     assert(reg_state[idx] == reg_writable, "must be writable");
 6268     reg_state[idx] = reg_written;
 6269     progress = true;
 6270 
 6271     if (fromReg == noreg) {
 6272       if (from->is_reg()) {
 6273         fromReg = from->as_Register();
 6274       } else {
 6275         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6276         movq(tmp1, Address(rsp, st_off));
 6277         fromReg = tmp1;
 6278       }
 6279       if (null_check) {
 6280         // Nullable inline type argument, emit null check
 6281         testptr(fromReg, fromReg);
 6282         jcc(Assembler::zero, L_null);
 6283       }
 6284     }
 6285     int off = sig->at(stream.sig_index())._offset;
 6286     if (off == -1) {
 6287       assert(null_check, "Missing null check at");
 6288       if (toReg->is_stack()) {
 6289         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6290         movq(Address(rsp, st_off), 1);
 6291       } else {
 6292         movq(toReg->as_Register(), 1);
 6293       }
 6294       continue;
 6295     }
 6296     if (sig->at(stream.sig_index())._vt_oop) {
 6297       if (toReg->is_stack()) {
 6298         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6299         movq(Address(rsp, st_off), fromReg);
 6300       } else {
 6301         movq(toReg->as_Register(), fromReg);
 6302       }
 6303       continue;
 6304     }
 6305     assert(off > 0, "offset in object should be positive");
 6306     Address fromAddr = Address(fromReg, off);
 6307     if (!toReg->is_XMMRegister()) {
 6308       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
 6309       if (is_reference_type(bt)) {
 6310         load_heap_oop(dst, fromAddr);
 6311       } else {
 6312         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
 6313         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
 6314       }
 6315       if (toReg->is_stack()) {
 6316         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6317         movq(Address(rsp, st_off), dst);
 6318       }
 6319     } else if (bt == T_DOUBLE) {
 6320       movdbl(toReg->as_XMMRegister(), fromAddr);
 6321     } else {
 6322       assert(bt == T_FLOAT, "must be float");
 6323       movflt(toReg->as_XMMRegister(), fromAddr);
 6324     }
 6325   }
 6326   if (progress && null_check) {
 6327     if (done) {
 6328       jmp(L_notNull);
 6329       bind(L_null);
 6330       // Set null marker to zero to signal that the argument is null.
 6331       // Also set all fields to zero since the runtime requires a canonical
 6332       // representation of a flat null.
 6333       stream.reset(sig_index, to_index);
 6334       while (stream.next(toReg, bt)) {
 6335         if (toReg->is_stack()) {
 6336           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6337           movq(Address(rsp, st_off), 0);
 6338         } else if (toReg->is_XMMRegister()) {
 6339           xorps(toReg->as_XMMRegister(), toReg->as_XMMRegister());
 6340         } else {
 6341           xorl(toReg->as_Register(), toReg->as_Register());
 6342         }
 6343       }
 6344       bind(L_notNull);
 6345     } else {
 6346       bind(L_null);
 6347     }
 6348   }
 6349 
 6350   sig_index = stream.sig_index();
 6351   to_index = stream.regs_index();
 6352 
 6353   if (mark_done && reg_state[from->value()] != reg_written) {
 6354     // This is okay because no one else will write to that slot
 6355     reg_state[from->value()] = reg_writable;
 6356   }
 6357   from_index--;
 6358   assert(progress || (start_offset == offset()), "should not emit code");
 6359   return done;
 6360 }
 6361 
 6362 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
 6363                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
 6364                                         RegState reg_state[], Register val_array) {
 6365   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
 6366   assert(to->is_valid(), "destination must be valid");
 6367 
 6368   if (reg_state[to->value()] == reg_written) {
 6369     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6370     return true; // Already written
 6371   }
 6372 
 6373   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
 6374   Register val_obj_tmp = r11;
 6375   Register from_reg_tmp = r14;
 6376   Register tmp1 = r10;
 6377   Register tmp2 = r13;
 6378   Register tmp3 = rbx;
 6379   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
 6380 
 6381   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
 6382 
 6383   if (reg_state[to->value()] == reg_readonly) {
 6384     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
 6385       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6386       return false; // Not yet writable
 6387     }
 6388     val_obj = val_obj_tmp;
 6389   }
 6390 
 6391   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
 6392   VMReg fromReg;
 6393   BasicType bt;
 6394   Label L_null;
 6395   while (stream.next(fromReg, bt)) {
 6396     assert(fromReg->is_valid(), "source must be valid");
 6397     reg_state[fromReg->value()] = reg_writable;
 6398 
 6399     int off = sig->at(stream.sig_index())._offset;
 6400     if (off == -1) {
 6401       // Nullable inline type argument, emit null check
 6402       Label L_notNull;
 6403       if (fromReg->is_stack()) {
 6404         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6405         testb(Address(rsp, ld_off), 1);
 6406       } else {
 6407         testb(fromReg->as_Register(), 1);
 6408       }
 6409       jcc(Assembler::notZero, L_notNull);
 6410       movptr(val_obj, 0);
 6411       jmp(L_null);
 6412       bind(L_notNull);
 6413       continue;
 6414     }
 6415     if (sig->at(stream.sig_index())._vt_oop) {
 6416       // buffer argument: use if non null
 6417       if (fromReg->is_stack()) {
 6418         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6419         movptr(val_obj, Address(rsp, ld_off));
 6420       } else {
 6421         movptr(val_obj, fromReg->as_Register());
 6422       }
 6423       testptr(val_obj, val_obj);
 6424       jcc(Assembler::notEqual, L_null);
 6425       // otherwise get the buffer from the just allocated pool of buffers
 6426       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
 6427       load_heap_oop(val_obj, Address(val_array, index));
 6428       continue;
 6429     }
 6430 
 6431     assert(off > 0, "offset in object should be positive");
 6432     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
 6433 
 6434     // Pack the scalarized field into the value object.
 6435     Address dst(val_obj, off);
 6436     if (!fromReg->is_XMMRegister()) {
 6437       Register src;
 6438       if (fromReg->is_stack()) {
 6439         src = from_reg_tmp;
 6440         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6441         load_sized_value(src, Address(rsp, ld_off), size_in_bytes, /* is_signed */ false);
 6442       } else {
 6443         src = fromReg->as_Register();
 6444       }
 6445       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
 6446       if (is_reference_type(bt)) {
 6447         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
 6448         mov(tmp3, val_obj);
 6449         Address dst_with_tmp3(tmp3, off);
 6450         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
 6451       } else {
 6452         store_sized_value(dst, src, size_in_bytes);
 6453       }
 6454     } else if (bt == T_DOUBLE) {
 6455       movdbl(dst, fromReg->as_XMMRegister());
 6456     } else {
 6457       assert(bt == T_FLOAT, "must be float");
 6458       movflt(dst, fromReg->as_XMMRegister());
 6459     }
 6460   }
 6461   bind(L_null);
 6462   sig_index = stream.sig_index();
 6463   from_index = stream.regs_index();
 6464 
 6465   assert(reg_state[to->value()] == reg_writable, "must have already been read");
 6466   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
 6467   assert(success, "to register must be writeable");
 6468   return true;
 6469 }
 6470 
 6471 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
 6472   return reg->is_XMMRegister() ? xmm8->as_VMReg() : r14->as_VMReg();
 6473 }
 6474 
 6475 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
 6476   assert((initial_framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 6477   if (needs_stack_repair) {
 6478     // The method has a scalarized entry point (where fields of value object arguments
 6479     // are passed through registers and stack), and a non-scalarized entry point (where
 6480     // value object arguments are given as oops). The non-scalarized entry point will
 6481     // first load each field of value object arguments and store them in registers and on
 6482     // the stack in a way compatible with the scalarized entry point. To do so, some extra
 6483     // stack space might be reserved (if argument registers are not enough). On leaving the
 6484     // method, this space must be freed.
 6485     //
 6486     // In case we used the non-scalarized entry point the stack looks like this:
 6487     //
 6488     // | Arguments from caller     |
 6489     // |---------------------------|  <-- caller's SP
 6490     // | Return address #1         |
 6491     // | Saved RBP #1              |
 6492     // |---------------------------|
 6493     // | Extension space for       |
 6494     // |   inline arg (un)packing  |
 6495     // |---------------------------|  <-- start of this method's frame
 6496     // | Return address #2         |
 6497     // | Saved RBP #2              |
 6498     // |---------------------------|  <-- RBP (with -XX:+PreserveFramePointer)
 6499     // | sp_inc                    |
 6500     // | method locals             |
 6501     // |---------------------------|  <-- SP
 6502     //
 6503     // Space for the return pc and saved rbp is reserved twice. But only the #1 copies
 6504     // contain the real values of return pc and saved rbp. The #2 copies are not reliable
 6505     // and should not be used. They are mostly needed to add space between the extension
 6506     // space and the locals, as there would be between the real arguments and the locals
 6507     // if we don't need to do unpacking (from the scalarized entry point).
 6508     //
 6509     // When leaving, one must load RBP #1 into RBP, and use the copy #1 of the return address,
 6510     // while keeping in mind that from the scalarized entry point, there will be only one
 6511     // copy. Indeed, in the case we used the scalarized calling convention, the stack looks like this:
 6512     //
 6513     // | Arguments from caller     |
 6514     // |---------------------------|  <-- caller's SP
 6515     // | Return address            |
 6516     // | Saved RBP                 |
 6517     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
 6518     // | sp_inc                    |
 6519     // | method locals             |
 6520     // |---------------------------|  <-- SP
 6521     //
 6522     // The sp_inc stack slot holds the total size of the frame, including the extension
 6523     // space and copies #2 of the return address and the saved RBP (but never the copies
 6524     // #1 of the return address and saved RBP). That is how to find the copies #1 of the
 6525     // return address and saved rbp. This size is expressed in bytes. Be careful when using
 6526     // it from C++ in pointer arithmetic you might need to divide it by wordSize.
 6527 
 6528     // The stack increment resides just below the saved rbp
 6529     addq(rsp, Address(rsp, initial_framesize - wordSize));
 6530     pop(rbp);
 6531   } else {
 6532     if (initial_framesize > 0) {
 6533       addq(rsp, initial_framesize);
 6534     }
 6535     pop(rbp);
 6536   }
 6537 }
 6538 
 6539 #if COMPILER2_OR_JVMCI
 6540 
 6541 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 6542 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp, KRegister mask) {
 6543   // cnt - number of qwords (8-byte words).
 6544   // base - start address, qword aligned.
 6545   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 6546   bool use64byteVector = (MaxVectorSize == 64) && (CopyAVX3Threshold == 0);
 6547   if (use64byteVector) {
 6548     evpbroadcastq(xtmp, val, AVX_512bit);
 6549   } else if (MaxVectorSize >= 32) {
 6550     movdq(xtmp, val);
 6551     punpcklqdq(xtmp, xtmp);
 6552     vinserti128_high(xtmp, xtmp);
 6553   } else {
 6554     movdq(xtmp, val);
 6555     punpcklqdq(xtmp, xtmp);
 6556   }
 6557   jmp(L_zero_64_bytes);
 6558 
 6559   BIND(L_loop);
 6560   if (MaxVectorSize >= 32) {
 6561     fill64(base, 0, xtmp, use64byteVector);
 6562   } else {
 6563     movdqu(Address(base,  0), xtmp);
 6564     movdqu(Address(base, 16), xtmp);
 6565     movdqu(Address(base, 32), xtmp);
 6566     movdqu(Address(base, 48), xtmp);
 6567   }
 6568   addptr(base, 64);
 6569 
 6570   BIND(L_zero_64_bytes);
 6571   subptr(cnt, 8);
 6572   jccb(Assembler::greaterEqual, L_loop);
 6573 
 6574   // Copy trailing 64 bytes
 6575   if (use64byteVector) {
 6576     addptr(cnt, 8);
 6577     jccb(Assembler::equal, L_end);
 6578     fill64_masked(3, base, 0, xtmp, mask, cnt, val, true);
 6579     jmp(L_end);
 6580   } else {
 6581     addptr(cnt, 4);
 6582     jccb(Assembler::less, L_tail);
 6583     if (MaxVectorSize >= 32) {
 6584       vmovdqu(Address(base, 0), xtmp);
 6585     } else {
 6586       movdqu(Address(base,  0), xtmp);
 6587       movdqu(Address(base, 16), xtmp);
 6588     }
 6589   }
 6590   addptr(base, 32);
 6591   subptr(cnt, 4);
 6592 
 6593   BIND(L_tail);
 6594   addptr(cnt, 4);
 6595   jccb(Assembler::lessEqual, L_end);
 6596   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 6597     fill32_masked(3, base, 0, xtmp, mask, cnt, val);
 6598   } else {
 6599     decrement(cnt);
 6600 
 6601     BIND(L_sloop);
 6602     movq(Address(base, 0), xtmp);
 6603     addptr(base, 8);
 6604     decrement(cnt);
 6605     jccb(Assembler::greaterEqual, L_sloop);
 6606   }
 6607   BIND(L_end);
 6608 }
 6609 
 6610 // Clearing constant sized memory using YMM/ZMM registers.
 6611 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 6612   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 6613   bool use64byteVector = (MaxVectorSize > 32) && (CopyAVX3Threshold == 0);
 6614 
 6615   int vector64_count = (cnt & (~0x7)) >> 3;
 6616   cnt = cnt & 0x7;
 6617   const int fill64_per_loop = 4;
 6618   const int max_unrolled_fill64 = 8;
 6619 
 6620   // 64 byte initialization loop.
 6621   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 6622   int start64 = 0;
 6623   if (vector64_count > max_unrolled_fill64) {
 6624     Label LOOP;
 6625     Register index = rtmp;
 6626 
 6627     start64 = vector64_count - (vector64_count % fill64_per_loop);
 6628 
 6629     movl(index, 0);
 6630     BIND(LOOP);
 6631     for (int i = 0; i < fill64_per_loop; i++) {
 6632       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 6633     }
 6634     addl(index, fill64_per_loop * 64);
 6635     cmpl(index, start64 * 64);
 6636     jccb(Assembler::less, LOOP);
 6637   }
 6638   for (int i = start64; i < vector64_count; i++) {
 6639     fill64(base, i * 64, xtmp, use64byteVector);
 6640   }
 6641 
 6642   // Clear remaining 64 byte tail.
 6643   int disp = vector64_count * 64;
 6644   if (cnt) {
 6645     switch (cnt) {
 6646       case 1:
 6647         movq(Address(base, disp), xtmp);
 6648         break;
 6649       case 2:
 6650         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 6651         break;
 6652       case 3:
 6653         movl(rtmp, 0x7);
 6654         kmovwl(mask, rtmp);
 6655         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 6656         break;
 6657       case 4:
 6658         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6659         break;
 6660       case 5:
 6661         if (use64byteVector) {
 6662           movl(rtmp, 0x1F);
 6663           kmovwl(mask, rtmp);
 6664           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6665         } else {
 6666           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6667           movq(Address(base, disp + 32), xtmp);
 6668         }
 6669         break;
 6670       case 6:
 6671         if (use64byteVector) {
 6672           movl(rtmp, 0x3F);
 6673           kmovwl(mask, rtmp);
 6674           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6675         } else {
 6676           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6677           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 6678         }
 6679         break;
 6680       case 7:
 6681         if (use64byteVector) {
 6682           movl(rtmp, 0x7F);
 6683           kmovwl(mask, rtmp);
 6684           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6685         } else {
 6686           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6687           movl(rtmp, 0x7);
 6688           kmovwl(mask, rtmp);
 6689           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 6690         }
 6691         break;
 6692       default:
 6693         fatal("Unexpected length : %d\n",cnt);
 6694         break;
 6695     }
 6696   }
 6697 }
 6698 
 6699 void MacroAssembler::clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp,
 6700                                bool is_large, bool word_copy_only, KRegister mask) {
 6701   // cnt      - number of qwords (8-byte words).
 6702   // base     - start address, qword aligned.
 6703   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6704   assert(base==rdi, "base register must be edi for rep stos");
 6705   assert(val==rax,   "val register must be eax for rep stos");
 6706   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6707   assert(InitArrayShortSize % BytesPerLong == 0,
 6708     "InitArrayShortSize should be the multiple of BytesPerLong");
 6709 
 6710   Label DONE;
 6711 
 6712   if (!is_large) {
 6713     Label LOOP, LONG;
 6714     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6715     jccb(Assembler::greater, LONG);
 6716 
 6717     decrement(cnt);
 6718     jccb(Assembler::negative, DONE); // Zero length
 6719 
 6720     // Use individual pointer-sized stores for small counts:
 6721     BIND(LOOP);
 6722     movptr(Address(base, cnt, Address::times_ptr), val);
 6723     decrement(cnt);
 6724     jccb(Assembler::greaterEqual, LOOP);
 6725     jmpb(DONE);
 6726 
 6727     BIND(LONG);
 6728   }
 6729 
 6730   // Use longer rep-prefixed ops for non-small counts:
 6731   if (UseFastStosb && !word_copy_only) {
 6732     shlptr(cnt, 3); // convert to number of bytes
 6733     rep_stosb();
 6734   } else if (UseXMMForObjInit) {
 6735     xmm_clear_mem(base, cnt, val, xtmp, mask);
 6736   } else {
 6737     rep_stos();
 6738   }
 6739 
 6740   BIND(DONE);
 6741 }
 6742 
 6743 #endif //COMPILER2_OR_JVMCI
 6744 
 6745 
 6746 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6747                                    Register to, Register value, Register count,
 6748                                    Register rtmp, XMMRegister xtmp) {
 6749   ShortBranchVerifier sbv(this);
 6750   assert_different_registers(to, value, count, rtmp);
 6751   Label L_exit;
 6752   Label L_fill_2_bytes, L_fill_4_bytes;
 6753 
 6754 #if defined(COMPILER2)
 6755   if(MaxVectorSize >=32 &&
 6756      VM_Version::supports_avx512vlbw() &&
 6757      VM_Version::supports_bmi2()) {
 6758     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6759     return;
 6760   }
 6761 #endif
 6762 
 6763   int shift = -1;
 6764   switch (t) {
 6765     case T_BYTE:
 6766       shift = 2;
 6767       break;
 6768     case T_SHORT:
 6769       shift = 1;
 6770       break;
 6771     case T_INT:
 6772       shift = 0;
 6773       break;
 6774     default: ShouldNotReachHere();
 6775   }
 6776 
 6777   if (t == T_BYTE) {
 6778     andl(value, 0xff);
 6779     movl(rtmp, value);
 6780     shll(rtmp, 8);
 6781     orl(value, rtmp);
 6782   }
 6783   if (t == T_SHORT) {
 6784     andl(value, 0xffff);
 6785   }
 6786   if (t == T_BYTE || t == T_SHORT) {
 6787     movl(rtmp, value);
 6788     shll(rtmp, 16);
 6789     orl(value, rtmp);
 6790   }
 6791 
 6792   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6793   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6794   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6795     Label L_skip_align2;
 6796     // align source address at 4 bytes address boundary
 6797     if (t == T_BYTE) {
 6798       Label L_skip_align1;
 6799       // One byte misalignment happens only for byte arrays
 6800       testptr(to, 1);
 6801       jccb(Assembler::zero, L_skip_align1);
 6802       movb(Address(to, 0), value);
 6803       increment(to);
 6804       decrement(count);
 6805       BIND(L_skip_align1);
 6806     }
 6807     // Two bytes misalignment happens only for byte and short (char) arrays
 6808     testptr(to, 2);
 6809     jccb(Assembler::zero, L_skip_align2);
 6810     movw(Address(to, 0), value);
 6811     addptr(to, 2);
 6812     subptr(count, 1<<(shift-1));
 6813     BIND(L_skip_align2);
 6814   }
 6815   {
 6816     Label L_fill_32_bytes;
 6817     if (!UseUnalignedLoadStores) {
 6818       // align to 8 bytes, we know we are 4 byte aligned to start
 6819       testptr(to, 4);
 6820       jccb(Assembler::zero, L_fill_32_bytes);
 6821       movl(Address(to, 0), value);
 6822       addptr(to, 4);
 6823       subptr(count, 1<<shift);
 6824     }
 6825     BIND(L_fill_32_bytes);
 6826     {
 6827       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6828       movdl(xtmp, value);
 6829       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6830         Label L_check_fill_32_bytes;
 6831         if (UseAVX > 2) {
 6832           // Fill 64-byte chunks
 6833           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6834 
 6835           // If number of bytes to fill < CopyAVX3Threshold, perform fill using AVX2
 6836           cmpptr(count, CopyAVX3Threshold);
 6837           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6838 
 6839           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6840 
 6841           subptr(count, 16 << shift);
 6842           jcc(Assembler::less, L_check_fill_32_bytes);
 6843           align(16);
 6844 
 6845           BIND(L_fill_64_bytes_loop_avx3);
 6846           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6847           addptr(to, 64);
 6848           subptr(count, 16 << shift);
 6849           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6850           jmpb(L_check_fill_32_bytes);
 6851 
 6852           BIND(L_check_fill_64_bytes_avx2);
 6853         }
 6854         // Fill 64-byte chunks
 6855         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6856 
 6857         subptr(count, 16 << shift);
 6858         jcc(Assembler::less, L_check_fill_32_bytes);
 6859 
 6860         // align data for 64-byte chunks
 6861         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6862         if (EnableX86ECoreOpts) {
 6863             // align 'big' arrays to cache lines to minimize split_stores
 6864             cmpptr(count, 96 << shift);
 6865             jcc(Assembler::below, L_fill_64_bytes_loop);
 6866 
 6867             // Find the bytes needed for alignment
 6868             movptr(rtmp, to);
 6869             andptr(rtmp, 0x1c);
 6870             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6871             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6872             addptr(rtmp, 32);
 6873             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6874             subptr(count, rtmp);    // adjust count by number of elements
 6875 
 6876             align(16);
 6877             BIND(L_align_64_bytes_loop);
 6878             movdl(Address(to, 0), xtmp);
 6879             addptr(to, 4);
 6880             subptr(rtmp, 1 << shift);
 6881             jcc(Assembler::greater, L_align_64_bytes_loop);
 6882         }
 6883 
 6884         align(16);
 6885         BIND(L_fill_64_bytes_loop);
 6886         vmovdqu(Address(to, 0), xtmp);
 6887         vmovdqu(Address(to, 32), xtmp);
 6888         addptr(to, 64);
 6889         subptr(count, 16 << shift);
 6890         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6891 
 6892         align(16);
 6893         BIND(L_check_fill_32_bytes);
 6894         addptr(count, 8 << shift);
 6895         jccb(Assembler::less, L_check_fill_8_bytes);
 6896         vmovdqu(Address(to, 0), xtmp);
 6897         addptr(to, 32);
 6898         subptr(count, 8 << shift);
 6899 
 6900         BIND(L_check_fill_8_bytes);
 6901         // clean upper bits of YMM registers
 6902         movdl(xtmp, value);
 6903         pshufd(xtmp, xtmp, 0);
 6904       } else {
 6905         // Fill 32-byte chunks
 6906         pshufd(xtmp, xtmp, 0);
 6907 
 6908         subptr(count, 8 << shift);
 6909         jcc(Assembler::less, L_check_fill_8_bytes);
 6910         align(16);
 6911 
 6912         BIND(L_fill_32_bytes_loop);
 6913 
 6914         if (UseUnalignedLoadStores) {
 6915           movdqu(Address(to, 0), xtmp);
 6916           movdqu(Address(to, 16), xtmp);
 6917         } else {
 6918           movq(Address(to, 0), xtmp);
 6919           movq(Address(to, 8), xtmp);
 6920           movq(Address(to, 16), xtmp);
 6921           movq(Address(to, 24), xtmp);
 6922         }
 6923 
 6924         addptr(to, 32);
 6925         subptr(count, 8 << shift);
 6926         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6927 
 6928         BIND(L_check_fill_8_bytes);
 6929       }
 6930       addptr(count, 8 << shift);
 6931       jccb(Assembler::zero, L_exit);
 6932       jmpb(L_fill_8_bytes);
 6933 
 6934       //
 6935       // length is too short, just fill qwords
 6936       //
 6937       align(16);
 6938       BIND(L_fill_8_bytes_loop);
 6939       movq(Address(to, 0), xtmp);
 6940       addptr(to, 8);
 6941       BIND(L_fill_8_bytes);
 6942       subptr(count, 1 << (shift + 1));
 6943       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6944     }
 6945   }
 6946 
 6947   Label L_fill_4_bytes_loop;
 6948   testl(count, 1 << shift);
 6949   jccb(Assembler::zero, L_fill_2_bytes);
 6950 
 6951   align(16);
 6952   BIND(L_fill_4_bytes_loop);
 6953   movl(Address(to, 0), value);
 6954   addptr(to, 4);
 6955 
 6956   BIND(L_fill_4_bytes);
 6957   subptr(count, 1 << shift);
 6958   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 6959 
 6960   if (t == T_BYTE || t == T_SHORT) {
 6961     Label L_fill_byte;
 6962     BIND(L_fill_2_bytes);
 6963     // fill trailing 2 bytes
 6964     testl(count, 1<<(shift-1));
 6965     jccb(Assembler::zero, L_fill_byte);
 6966     movw(Address(to, 0), value);
 6967     if (t == T_BYTE) {
 6968       addptr(to, 2);
 6969       BIND(L_fill_byte);
 6970       // fill trailing byte
 6971       testl(count, 1);
 6972       jccb(Assembler::zero, L_exit);
 6973       movb(Address(to, 0), value);
 6974     } else {
 6975       BIND(L_fill_byte);
 6976     }
 6977   } else {
 6978     BIND(L_fill_2_bytes);
 6979   }
 6980   BIND(L_exit);
 6981 }
 6982 
 6983 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6984   switch(type) {
 6985     case T_BYTE:
 6986     case T_BOOLEAN:
 6987       evpbroadcastb(dst, src, vector_len);
 6988       break;
 6989     case T_SHORT:
 6990     case T_CHAR:
 6991       evpbroadcastw(dst, src, vector_len);
 6992       break;
 6993     case T_INT:
 6994     case T_FLOAT:
 6995       evpbroadcastd(dst, src, vector_len);
 6996       break;
 6997     case T_LONG:
 6998     case T_DOUBLE:
 6999       evpbroadcastq(dst, src, vector_len);
 7000       break;
 7001     default:
 7002       fatal("Unhandled type : %s", type2name(type));
 7003       break;
 7004   }
 7005 }
 7006 
 7007 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 7008 //
 7009 // @IntrinsicCandidate
 7010 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 7011 //         char[] sa, int sp, byte[] da, int dp, int len) {
 7012 //     int i = 0;
 7013 //     for (; i < len; i++) {
 7014 //         char c = sa[sp++];
 7015 //         if (c > '\u00FF')
 7016 //             break;
 7017 //         da[dp++] = (byte) c;
 7018 //     }
 7019 //     return i;
 7020 // }
 7021 //
 7022 // @IntrinsicCandidate
 7023 // int java.lang.StringCoding.encodeISOArray0(
 7024 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 7025 //   int i = 0;
 7026 //   for (; i < len; i++) {
 7027 //     char c = StringUTF16.getChar(sa, sp++);
 7028 //     if (c > '\u00FF')
 7029 //       break;
 7030 //     da[dp++] = (byte) c;
 7031 //   }
 7032 //   return i;
 7033 // }
 7034 //
 7035 // @IntrinsicCandidate
 7036 // int java.lang.StringCoding.encodeAsciiArray0(
 7037 //         char[] sa, int sp, byte[] da, int dp, int len) {
 7038 //   int i = 0;
 7039 //   for (; i < len; i++) {
 7040 //     char c = sa[sp++];
 7041 //     if (c >= '\u0080')
 7042 //       break;
 7043 //     da[dp++] = (byte) c;
 7044 //   }
 7045 //   return i;
 7046 // }
 7047 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 7048   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 7049   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 7050   Register tmp5, Register result, bool ascii) {
 7051 
 7052   // rsi: src
 7053   // rdi: dst
 7054   // rdx: len
 7055   // rcx: tmp5
 7056   // rax: result
 7057   ShortBranchVerifier sbv(this);
 7058   assert_different_registers(src, dst, len, tmp5, result);
 7059   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 7060 
 7061   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 7062   int short_mask = ascii ? 0xff80 : 0xff00;
 7063 
 7064   // set result
 7065   xorl(result, result);
 7066   // check for zero length
 7067   testl(len, len);
 7068   jcc(Assembler::zero, L_done);
 7069 
 7070   movl(result, len);
 7071 
 7072   // Setup pointers
 7073   lea(src, Address(src, len, Address::times_2)); // char[]
 7074   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 7075   negptr(len);
 7076 
 7077   if (UseSSE42Intrinsics || UseAVX >= 2) {
 7078     Label L_copy_8_chars, L_copy_8_chars_exit;
 7079     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 7080 
 7081     if (UseAVX >= 2) {
 7082       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 7083       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7084       movdl(tmp1Reg, tmp5);
 7085       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 7086       jmp(L_chars_32_check);
 7087 
 7088       bind(L_copy_32_chars);
 7089       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 7090       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 7091       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7092       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7093       jccb(Assembler::notZero, L_copy_32_chars_exit);
 7094       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7095       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 7096       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 7097 
 7098       bind(L_chars_32_check);
 7099       addptr(len, 32);
 7100       jcc(Assembler::lessEqual, L_copy_32_chars);
 7101 
 7102       bind(L_copy_32_chars_exit);
 7103       subptr(len, 16);
 7104       jccb(Assembler::greater, L_copy_16_chars_exit);
 7105 
 7106     } else if (UseSSE42Intrinsics) {
 7107       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7108       movdl(tmp1Reg, tmp5);
 7109       pshufd(tmp1Reg, tmp1Reg, 0);
 7110       jmpb(L_chars_16_check);
 7111     }
 7112 
 7113     bind(L_copy_16_chars);
 7114     if (UseAVX >= 2) {
 7115       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 7116       vptest(tmp2Reg, tmp1Reg);
 7117       jcc(Assembler::notZero, L_copy_16_chars_exit);
 7118       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 7119       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 7120     } else {
 7121       if (UseAVX > 0) {
 7122         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7123         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7124         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 7125       } else {
 7126         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7127         por(tmp2Reg, tmp3Reg);
 7128         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7129         por(tmp2Reg, tmp4Reg);
 7130       }
 7131       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7132       jccb(Assembler::notZero, L_copy_16_chars_exit);
 7133       packuswb(tmp3Reg, tmp4Reg);
 7134     }
 7135     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 7136 
 7137     bind(L_chars_16_check);
 7138     addptr(len, 16);
 7139     jcc(Assembler::lessEqual, L_copy_16_chars);
 7140 
 7141     bind(L_copy_16_chars_exit);
 7142     if (UseAVX >= 2) {
 7143       // clean upper bits of YMM registers
 7144       vpxor(tmp2Reg, tmp2Reg);
 7145       vpxor(tmp3Reg, tmp3Reg);
 7146       vpxor(tmp4Reg, tmp4Reg);
 7147       movdl(tmp1Reg, tmp5);
 7148       pshufd(tmp1Reg, tmp1Reg, 0);
 7149     }
 7150     subptr(len, 8);
 7151     jccb(Assembler::greater, L_copy_8_chars_exit);
 7152 
 7153     bind(L_copy_8_chars);
 7154     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 7155     ptest(tmp3Reg, tmp1Reg);
 7156     jccb(Assembler::notZero, L_copy_8_chars_exit);
 7157     packuswb(tmp3Reg, tmp1Reg);
 7158     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 7159     addptr(len, 8);
 7160     jccb(Assembler::lessEqual, L_copy_8_chars);
 7161 
 7162     bind(L_copy_8_chars_exit);
 7163     subptr(len, 8);
 7164     jccb(Assembler::zero, L_done);
 7165   }
 7166 
 7167   bind(L_copy_1_char);
 7168   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 7169   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 7170   jccb(Assembler::notZero, L_copy_1_char_exit);
 7171   movb(Address(dst, len, Address::times_1, 0), tmp5);
 7172   addptr(len, 1);
 7173   jccb(Assembler::less, L_copy_1_char);
 7174 
 7175   bind(L_copy_1_char_exit);
 7176   addptr(result, len); // len is negative count of not processed elements
 7177 
 7178   bind(L_done);
 7179 }
 7180 
 7181 /**
 7182  * Helper for multiply_to_len().
 7183  */
 7184 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 7185   addq(dest_lo, src1);
 7186   adcq(dest_hi, 0);
 7187   addq(dest_lo, src2);
 7188   adcq(dest_hi, 0);
 7189 }
 7190 
 7191 /**
 7192  * Multiply 64 bit by 64 bit first loop.
 7193  */
 7194 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 7195                                            Register y, Register y_idx, Register z,
 7196                                            Register carry, Register product,
 7197                                            Register idx, Register kdx) {
 7198   //
 7199   //  jlong carry, x[], y[], z[];
 7200   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7201   //    huge_128 product = y[idx] * x[xstart] + carry;
 7202   //    z[kdx] = (jlong)product;
 7203   //    carry  = (jlong)(product >>> 64);
 7204   //  }
 7205   //  z[xstart] = carry;
 7206   //
 7207 
 7208   Label L_first_loop, L_first_loop_exit;
 7209   Label L_one_x, L_one_y, L_multiply;
 7210 
 7211   decrementl(xstart);
 7212   jcc(Assembler::negative, L_one_x);
 7213 
 7214   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7215   rorq(x_xstart, 32); // convert big-endian to little-endian
 7216 
 7217   bind(L_first_loop);
 7218   decrementl(idx);
 7219   jcc(Assembler::negative, L_first_loop_exit);
 7220   decrementl(idx);
 7221   jcc(Assembler::negative, L_one_y);
 7222   movq(y_idx, Address(y, idx, Address::times_4,  0));
 7223   rorq(y_idx, 32); // convert big-endian to little-endian
 7224   bind(L_multiply);
 7225   movq(product, x_xstart);
 7226   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 7227   addq(product, carry);
 7228   adcq(rdx, 0);
 7229   subl(kdx, 2);
 7230   movl(Address(z, kdx, Address::times_4,  4), product);
 7231   shrq(product, 32);
 7232   movl(Address(z, kdx, Address::times_4,  0), product);
 7233   movq(carry, rdx);
 7234   jmp(L_first_loop);
 7235 
 7236   bind(L_one_y);
 7237   movl(y_idx, Address(y,  0));
 7238   jmp(L_multiply);
 7239 
 7240   bind(L_one_x);
 7241   movl(x_xstart, Address(x,  0));
 7242   jmp(L_first_loop);
 7243 
 7244   bind(L_first_loop_exit);
 7245 }
 7246 
 7247 /**
 7248  * Multiply 64 bit by 64 bit and add 128 bit.
 7249  */
 7250 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 7251                                             Register yz_idx, Register idx,
 7252                                             Register carry, Register product, int offset) {
 7253   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 7254   //     z[kdx] = (jlong)product;
 7255 
 7256   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 7257   rorq(yz_idx, 32); // convert big-endian to little-endian
 7258   movq(product, x_xstart);
 7259   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 7260   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 7261   rorq(yz_idx, 32); // convert big-endian to little-endian
 7262 
 7263   add2_with_carry(rdx, product, carry, yz_idx);
 7264 
 7265   movl(Address(z, idx, Address::times_4,  offset+4), product);
 7266   shrq(product, 32);
 7267   movl(Address(z, idx, Address::times_4,  offset), product);
 7268 
 7269 }
 7270 
 7271 /**
 7272  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 7273  */
 7274 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 7275                                              Register yz_idx, Register idx, Register jdx,
 7276                                              Register carry, Register product,
 7277                                              Register carry2) {
 7278   //   jlong carry, x[], y[], z[];
 7279   //   int kdx = ystart+1;
 7280   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7281   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 7282   //     z[kdx+idx+1] = (jlong)product;
 7283   //     jlong carry2  = (jlong)(product >>> 64);
 7284   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 7285   //     z[kdx+idx] = (jlong)product;
 7286   //     carry  = (jlong)(product >>> 64);
 7287   //   }
 7288   //   idx += 2;
 7289   //   if (idx > 0) {
 7290   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 7291   //     z[kdx+idx] = (jlong)product;
 7292   //     carry  = (jlong)(product >>> 64);
 7293   //   }
 7294   //
 7295 
 7296   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7297 
 7298   movl(jdx, idx);
 7299   andl(jdx, 0xFFFFFFFC);
 7300   shrl(jdx, 2);
 7301 
 7302   bind(L_third_loop);
 7303   subl(jdx, 1);
 7304   jcc(Assembler::negative, L_third_loop_exit);
 7305   subl(idx, 4);
 7306 
 7307   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 7308   movq(carry2, rdx);
 7309 
 7310   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 7311   movq(carry, rdx);
 7312   jmp(L_third_loop);
 7313 
 7314   bind (L_third_loop_exit);
 7315 
 7316   andl (idx, 0x3);
 7317   jcc(Assembler::zero, L_post_third_loop_done);
 7318 
 7319   Label L_check_1;
 7320   subl(idx, 2);
 7321   jcc(Assembler::negative, L_check_1);
 7322 
 7323   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 7324   movq(carry, rdx);
 7325 
 7326   bind (L_check_1);
 7327   addl (idx, 0x2);
 7328   andl (idx, 0x1);
 7329   subl(idx, 1);
 7330   jcc(Assembler::negative, L_post_third_loop_done);
 7331 
 7332   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 7333   movq(product, x_xstart);
 7334   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 7335   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 7336 
 7337   add2_with_carry(rdx, product, yz_idx, carry);
 7338 
 7339   movl(Address(z, idx, Address::times_4,  0), product);
 7340   shrq(product, 32);
 7341 
 7342   shlq(rdx, 32);
 7343   orq(product, rdx);
 7344   movq(carry, product);
 7345 
 7346   bind(L_post_third_loop_done);
 7347 }
 7348 
 7349 /**
 7350  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 7351  *
 7352  */
 7353 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 7354                                                   Register carry, Register carry2,
 7355                                                   Register idx, Register jdx,
 7356                                                   Register yz_idx1, Register yz_idx2,
 7357                                                   Register tmp, Register tmp3, Register tmp4) {
 7358   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 7359 
 7360   //   jlong carry, x[], y[], z[];
 7361   //   int kdx = ystart+1;
 7362   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7363   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 7364   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 7365   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 7366   //     carry  = (jlong)(tmp4 >>> 64);
 7367   //     z[kdx+idx+1] = (jlong)tmp3;
 7368   //     z[kdx+idx] = (jlong)tmp4;
 7369   //   }
 7370   //   idx += 2;
 7371   //   if (idx > 0) {
 7372   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 7373   //     z[kdx+idx] = (jlong)yz_idx1;
 7374   //     carry  = (jlong)(yz_idx1 >>> 64);
 7375   //   }
 7376   //
 7377 
 7378   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7379 
 7380   movl(jdx, idx);
 7381   andl(jdx, 0xFFFFFFFC);
 7382   shrl(jdx, 2);
 7383 
 7384   bind(L_third_loop);
 7385   subl(jdx, 1);
 7386   jcc(Assembler::negative, L_third_loop_exit);
 7387   subl(idx, 4);
 7388 
 7389   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 7390   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 7391   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 7392   rorxq(yz_idx2, yz_idx2, 32);
 7393 
 7394   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 7395   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 7396 
 7397   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 7398   rorxq(yz_idx1, yz_idx1, 32);
 7399   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7400   rorxq(yz_idx2, yz_idx2, 32);
 7401 
 7402   if (VM_Version::supports_adx()) {
 7403     adcxq(tmp3, carry);
 7404     adoxq(tmp3, yz_idx1);
 7405 
 7406     adcxq(tmp4, tmp);
 7407     adoxq(tmp4, yz_idx2);
 7408 
 7409     movl(carry, 0); // does not affect flags
 7410     adcxq(carry2, carry);
 7411     adoxq(carry2, carry);
 7412   } else {
 7413     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 7414     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 7415   }
 7416   movq(carry, carry2);
 7417 
 7418   movl(Address(z, idx, Address::times_4, 12), tmp3);
 7419   shrq(tmp3, 32);
 7420   movl(Address(z, idx, Address::times_4,  8), tmp3);
 7421 
 7422   movl(Address(z, idx, Address::times_4,  4), tmp4);
 7423   shrq(tmp4, 32);
 7424   movl(Address(z, idx, Address::times_4,  0), tmp4);
 7425 
 7426   jmp(L_third_loop);
 7427 
 7428   bind (L_third_loop_exit);
 7429 
 7430   andl (idx, 0x3);
 7431   jcc(Assembler::zero, L_post_third_loop_done);
 7432 
 7433   Label L_check_1;
 7434   subl(idx, 2);
 7435   jcc(Assembler::negative, L_check_1);
 7436 
 7437   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 7438   rorxq(yz_idx1, yz_idx1, 32);
 7439   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 7440   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7441   rorxq(yz_idx2, yz_idx2, 32);
 7442 
 7443   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 7444 
 7445   movl(Address(z, idx, Address::times_4,  4), tmp3);
 7446   shrq(tmp3, 32);
 7447   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7448   movq(carry, tmp4);
 7449 
 7450   bind (L_check_1);
 7451   addl (idx, 0x2);
 7452   andl (idx, 0x1);
 7453   subl(idx, 1);
 7454   jcc(Assembler::negative, L_post_third_loop_done);
 7455   movl(tmp4, Address(y, idx, Address::times_4,  0));
 7456   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 7457   movl(tmp4, Address(z, idx, Address::times_4,  0));
 7458 
 7459   add2_with_carry(carry2, tmp3, tmp4, carry);
 7460 
 7461   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7462   shrq(tmp3, 32);
 7463 
 7464   shlq(carry2, 32);
 7465   orq(tmp3, carry2);
 7466   movq(carry, tmp3);
 7467 
 7468   bind(L_post_third_loop_done);
 7469 }
 7470 
 7471 /**
 7472  * Code for BigInteger::multiplyToLen() intrinsic.
 7473  *
 7474  * rdi: x
 7475  * rax: xlen
 7476  * rsi: y
 7477  * rcx: ylen
 7478  * r8:  z
 7479  * r11: tmp0
 7480  * r12: tmp1
 7481  * r13: tmp2
 7482  * r14: tmp3
 7483  * r15: tmp4
 7484  * rbx: tmp5
 7485  *
 7486  */
 7487 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 7488                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 7489   ShortBranchVerifier sbv(this);
 7490   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 7491 
 7492   push(tmp0);
 7493   push(tmp1);
 7494   push(tmp2);
 7495   push(tmp3);
 7496   push(tmp4);
 7497   push(tmp5);
 7498 
 7499   push(xlen);
 7500 
 7501   const Register idx = tmp1;
 7502   const Register kdx = tmp2;
 7503   const Register xstart = tmp3;
 7504 
 7505   const Register y_idx = tmp4;
 7506   const Register carry = tmp5;
 7507   const Register product  = xlen;
 7508   const Register x_xstart = tmp0;
 7509 
 7510   // First Loop.
 7511   //
 7512   //  final static long LONG_MASK = 0xffffffffL;
 7513   //  int xstart = xlen - 1;
 7514   //  int ystart = ylen - 1;
 7515   //  long carry = 0;
 7516   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7517   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 7518   //    z[kdx] = (int)product;
 7519   //    carry = product >>> 32;
 7520   //  }
 7521   //  z[xstart] = (int)carry;
 7522   //
 7523 
 7524   movl(idx, ylen);               // idx = ylen;
 7525   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 7526   xorq(carry, carry);            // carry = 0;
 7527 
 7528   Label L_done;
 7529 
 7530   movl(xstart, xlen);
 7531   decrementl(xstart);
 7532   jcc(Assembler::negative, L_done);
 7533 
 7534   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 7535 
 7536   Label L_second_loop;
 7537   testl(kdx, kdx);
 7538   jcc(Assembler::zero, L_second_loop);
 7539 
 7540   Label L_carry;
 7541   subl(kdx, 1);
 7542   jcc(Assembler::zero, L_carry);
 7543 
 7544   movl(Address(z, kdx, Address::times_4,  0), carry);
 7545   shrq(carry, 32);
 7546   subl(kdx, 1);
 7547 
 7548   bind(L_carry);
 7549   movl(Address(z, kdx, Address::times_4,  0), carry);
 7550 
 7551   // Second and third (nested) loops.
 7552   //
 7553   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 7554   //   carry = 0;
 7555   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 7556   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 7557   //                    (z[k] & LONG_MASK) + carry;
 7558   //     z[k] = (int)product;
 7559   //     carry = product >>> 32;
 7560   //   }
 7561   //   z[i] = (int)carry;
 7562   // }
 7563   //
 7564   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 7565 
 7566   const Register jdx = tmp1;
 7567 
 7568   bind(L_second_loop);
 7569   xorl(carry, carry);    // carry = 0;
 7570   movl(jdx, ylen);       // j = ystart+1
 7571 
 7572   subl(xstart, 1);       // i = xstart-1;
 7573   jcc(Assembler::negative, L_done);
 7574 
 7575   push (z);
 7576 
 7577   Label L_last_x;
 7578   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 7579   subl(xstart, 1);       // i = xstart-1;
 7580   jcc(Assembler::negative, L_last_x);
 7581 
 7582   if (UseBMI2Instructions) {
 7583     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 7584     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 7585   } else {
 7586     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7587     rorq(x_xstart, 32);  // convert big-endian to little-endian
 7588   }
 7589 
 7590   Label L_third_loop_prologue;
 7591   bind(L_third_loop_prologue);
 7592 
 7593   push (x);
 7594   push (xstart);
 7595   push (ylen);
 7596 
 7597 
 7598   if (UseBMI2Instructions) {
 7599     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 7600   } else { // !UseBMI2Instructions
 7601     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 7602   }
 7603 
 7604   pop(ylen);
 7605   pop(xlen);
 7606   pop(x);
 7607   pop(z);
 7608 
 7609   movl(tmp3, xlen);
 7610   addl(tmp3, 1);
 7611   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7612   subl(tmp3, 1);
 7613   jccb(Assembler::negative, L_done);
 7614 
 7615   shrq(carry, 32);
 7616   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7617   jmp(L_second_loop);
 7618 
 7619   // Next infrequent code is moved outside loops.
 7620   bind(L_last_x);
 7621   if (UseBMI2Instructions) {
 7622     movl(rdx, Address(x,  0));
 7623   } else {
 7624     movl(x_xstart, Address(x,  0));
 7625   }
 7626   jmp(L_third_loop_prologue);
 7627 
 7628   bind(L_done);
 7629 
 7630   pop(xlen);
 7631 
 7632   pop(tmp5);
 7633   pop(tmp4);
 7634   pop(tmp3);
 7635   pop(tmp2);
 7636   pop(tmp1);
 7637   pop(tmp0);
 7638 }
 7639 
 7640 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 7641   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 7642   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 7643   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 7644   Label VECTOR8_TAIL, VECTOR4_TAIL;
 7645   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 7646   Label SAME_TILL_END, DONE;
 7647   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 7648 
 7649   //scale is in rcx in both Win64 and Unix
 7650   ShortBranchVerifier sbv(this);
 7651 
 7652   shlq(length);
 7653   xorq(result, result);
 7654 
 7655   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 7656       VM_Version::supports_avx512vlbw() && UseCountTrailingZerosInstruction) {
 7657     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 7658 
 7659     cmpq(length, 64);
 7660     jcc(Assembler::less, VECTOR32_TAIL);
 7661 
 7662     movq(tmp1, length);
 7663     andq(tmp1, 0x3F);      // tail count
 7664     andq(length, ~(0x3F)); //vector count
 7665 
 7666     bind(VECTOR64_LOOP);
 7667     // AVX512 code to compare 64 byte vectors.
 7668     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 7669     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7670     kortestql(k7, k7);
 7671     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 7672     addq(result, 64);
 7673     subq(length, 64);
 7674     jccb(Assembler::notZero, VECTOR64_LOOP);
 7675 
 7676     //bind(VECTOR64_TAIL);
 7677     testq(tmp1, tmp1);
 7678     jcc(Assembler::zero, SAME_TILL_END);
 7679 
 7680     //bind(VECTOR64_TAIL);
 7681     // AVX512 code to compare up to 63 byte vectors.
 7682     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 7683     shlxq(tmp2, tmp2, tmp1);
 7684     notq(tmp2);
 7685     kmovql(k3, tmp2);
 7686 
 7687     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 7688     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7689 
 7690     ktestql(k7, k3);
 7691     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 7692 
 7693     bind(VECTOR64_NOT_EQUAL);
 7694     kmovql(tmp1, k7);
 7695     notq(tmp1);
 7696     tzcntq(tmp1, tmp1);
 7697     addq(result, tmp1);
 7698     shrq(result);
 7699     jmp(DONE);
 7700     bind(VECTOR32_TAIL);
 7701   }
 7702 
 7703   cmpq(length, 8);
 7704   jcc(Assembler::equal, VECTOR8_LOOP);
 7705   jcc(Assembler::less, VECTOR4_TAIL);
 7706 
 7707   if (UseAVX >= 2) {
 7708     Label VECTOR16_TAIL, VECTOR32_LOOP;
 7709 
 7710     cmpq(length, 16);
 7711     jcc(Assembler::equal, VECTOR16_LOOP);
 7712     jcc(Assembler::less, VECTOR8_LOOP);
 7713 
 7714     cmpq(length, 32);
 7715     jccb(Assembler::less, VECTOR16_TAIL);
 7716 
 7717     subq(length, 32);
 7718     bind(VECTOR32_LOOP);
 7719     vmovdqu(rymm0, Address(obja, result));
 7720     vmovdqu(rymm1, Address(objb, result));
 7721     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 7722     vptest(rymm2, rymm2);
 7723     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7724     addq(result, 32);
 7725     subq(length, 32);
 7726     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7727     addq(length, 32);
 7728     jcc(Assembler::equal, SAME_TILL_END);
 7729     //falling through if less than 32 bytes left //close the branch here.
 7730 
 7731     bind(VECTOR16_TAIL);
 7732     cmpq(length, 16);
 7733     jccb(Assembler::less, VECTOR8_TAIL);
 7734     bind(VECTOR16_LOOP);
 7735     movdqu(rymm0, Address(obja, result));
 7736     movdqu(rymm1, Address(objb, result));
 7737     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7738     ptest(rymm2, rymm2);
 7739     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7740     addq(result, 16);
 7741     subq(length, 16);
 7742     jcc(Assembler::equal, SAME_TILL_END);
 7743     //falling through if less than 16 bytes left
 7744   } else {//regular intrinsics
 7745 
 7746     cmpq(length, 16);
 7747     jccb(Assembler::less, VECTOR8_TAIL);
 7748 
 7749     subq(length, 16);
 7750     bind(VECTOR16_LOOP);
 7751     movdqu(rymm0, Address(obja, result));
 7752     movdqu(rymm1, Address(objb, result));
 7753     pxor(rymm0, rymm1);
 7754     ptest(rymm0, rymm0);
 7755     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7756     addq(result, 16);
 7757     subq(length, 16);
 7758     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7759     addq(length, 16);
 7760     jcc(Assembler::equal, SAME_TILL_END);
 7761     //falling through if less than 16 bytes left
 7762   }
 7763 
 7764   bind(VECTOR8_TAIL);
 7765   cmpq(length, 8);
 7766   jccb(Assembler::less, VECTOR4_TAIL);
 7767   bind(VECTOR8_LOOP);
 7768   movq(tmp1, Address(obja, result));
 7769   movq(tmp2, Address(objb, result));
 7770   xorq(tmp1, tmp2);
 7771   testq(tmp1, tmp1);
 7772   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7773   addq(result, 8);
 7774   subq(length, 8);
 7775   jcc(Assembler::equal, SAME_TILL_END);
 7776   //falling through if less than 8 bytes left
 7777 
 7778   bind(VECTOR4_TAIL);
 7779   cmpq(length, 4);
 7780   jccb(Assembler::less, BYTES_TAIL);
 7781   bind(VECTOR4_LOOP);
 7782   movl(tmp1, Address(obja, result));
 7783   xorl(tmp1, Address(objb, result));
 7784   testl(tmp1, tmp1);
 7785   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7786   addq(result, 4);
 7787   subq(length, 4);
 7788   jcc(Assembler::equal, SAME_TILL_END);
 7789   //falling through if less than 4 bytes left
 7790 
 7791   bind(BYTES_TAIL);
 7792   bind(BYTES_LOOP);
 7793   load_unsigned_byte(tmp1, Address(obja, result));
 7794   load_unsigned_byte(tmp2, Address(objb, result));
 7795   xorl(tmp1, tmp2);
 7796   testl(tmp1, tmp1);
 7797   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7798   decq(length);
 7799   jcc(Assembler::zero, SAME_TILL_END);
 7800   incq(result);
 7801   load_unsigned_byte(tmp1, Address(obja, result));
 7802   load_unsigned_byte(tmp2, Address(objb, result));
 7803   xorl(tmp1, tmp2);
 7804   testl(tmp1, tmp1);
 7805   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7806   decq(length);
 7807   jcc(Assembler::zero, SAME_TILL_END);
 7808   incq(result);
 7809   load_unsigned_byte(tmp1, Address(obja, result));
 7810   load_unsigned_byte(tmp2, Address(objb, result));
 7811   xorl(tmp1, tmp2);
 7812   testl(tmp1, tmp1);
 7813   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7814   jmp(SAME_TILL_END);
 7815 
 7816   if (UseAVX >= 2) {
 7817     bind(VECTOR32_NOT_EQUAL);
 7818     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7819     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7820     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7821     vpmovmskb(tmp1, rymm0);
 7822     bsfq(tmp1, tmp1);
 7823     addq(result, tmp1);
 7824     shrq(result);
 7825     jmp(DONE);
 7826   }
 7827 
 7828   bind(VECTOR16_NOT_EQUAL);
 7829   if (UseAVX >= 2) {
 7830     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7831     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7832     pxor(rymm0, rymm2);
 7833   } else {
 7834     pcmpeqb(rymm2, rymm2);
 7835     pxor(rymm0, rymm1);
 7836     pcmpeqb(rymm0, rymm1);
 7837     pxor(rymm0, rymm2);
 7838   }
 7839   pmovmskb(tmp1, rymm0);
 7840   bsfq(tmp1, tmp1);
 7841   addq(result, tmp1);
 7842   shrq(result);
 7843   jmpb(DONE);
 7844 
 7845   bind(VECTOR8_NOT_EQUAL);
 7846   bind(VECTOR4_NOT_EQUAL);
 7847   bsfq(tmp1, tmp1);
 7848   shrq(tmp1, 3);
 7849   addq(result, tmp1);
 7850   bind(BYTES_NOT_EQUAL);
 7851   shrq(result);
 7852   jmpb(DONE);
 7853 
 7854   bind(SAME_TILL_END);
 7855   mov64(result, -1);
 7856 
 7857   bind(DONE);
 7858 }
 7859 
 7860 //Helper functions for square_to_len()
 7861 
 7862 /**
 7863  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7864  * Preserves x and z and modifies rest of the registers.
 7865  */
 7866 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7867   // Perform square and right shift by 1
 7868   // Handle odd xlen case first, then for even xlen do the following
 7869   // jlong carry = 0;
 7870   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7871   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7872   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7873   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7874   //     carry = (jlong)product;
 7875   // }
 7876 
 7877   xorq(tmp5, tmp5);     // carry
 7878   xorq(rdxReg, rdxReg);
 7879   xorl(tmp1, tmp1);     // index for x
 7880   xorl(tmp4, tmp4);     // index for z
 7881 
 7882   Label L_first_loop, L_first_loop_exit;
 7883 
 7884   testl(xlen, 1);
 7885   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7886 
 7887   // Square and right shift by 1 the odd element using 32 bit multiply
 7888   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7889   imulq(raxReg, raxReg);
 7890   shrq(raxReg, 1);
 7891   adcq(tmp5, 0);
 7892   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7893   incrementl(tmp1);
 7894   addl(tmp4, 2);
 7895 
 7896   // Square and  right shift by 1 the rest using 64 bit multiply
 7897   bind(L_first_loop);
 7898   cmpptr(tmp1, xlen);
 7899   jccb(Assembler::equal, L_first_loop_exit);
 7900 
 7901   // Square
 7902   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7903   rorq(raxReg, 32);    // convert big-endian to little-endian
 7904   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7905 
 7906   // Right shift by 1 and save carry
 7907   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7908   rcrq(rdxReg, 1);
 7909   rcrq(raxReg, 1);
 7910   adcq(tmp5, 0);
 7911 
 7912   // Store result in z
 7913   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7914   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7915 
 7916   // Update indices for x and z
 7917   addl(tmp1, 2);
 7918   addl(tmp4, 4);
 7919   jmp(L_first_loop);
 7920 
 7921   bind(L_first_loop_exit);
 7922 }
 7923 
 7924 
 7925 /**
 7926  * Perform the following multiply add operation using BMI2 instructions
 7927  * carry:sum = sum + op1*op2 + carry
 7928  * op2 should be in rdx
 7929  * op2 is preserved, all other registers are modified
 7930  */
 7931 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7932   // assert op2 is rdx
 7933   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7934   addq(sum, carry);
 7935   adcq(tmp2, 0);
 7936   addq(sum, op1);
 7937   adcq(tmp2, 0);
 7938   movq(carry, tmp2);
 7939 }
 7940 
 7941 /**
 7942  * Perform the following multiply add operation:
 7943  * carry:sum = sum + op1*op2 + carry
 7944  * Preserves op1, op2 and modifies rest of registers
 7945  */
 7946 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7947   // rdx:rax = op1 * op2
 7948   movq(raxReg, op2);
 7949   mulq(op1);
 7950 
 7951   //  rdx:rax = sum + carry + rdx:rax
 7952   addq(sum, carry);
 7953   adcq(rdxReg, 0);
 7954   addq(sum, raxReg);
 7955   adcq(rdxReg, 0);
 7956 
 7957   // carry:sum = rdx:sum
 7958   movq(carry, rdxReg);
 7959 }
 7960 
 7961 /**
 7962  * Add 64 bit long carry into z[] with carry propagation.
 7963  * Preserves z and carry register values and modifies rest of registers.
 7964  *
 7965  */
 7966 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7967   Label L_fourth_loop, L_fourth_loop_exit;
 7968 
 7969   movl(tmp1, 1);
 7970   subl(zlen, 2);
 7971   addq(Address(z, zlen, Address::times_4, 0), carry);
 7972 
 7973   bind(L_fourth_loop);
 7974   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7975   subl(zlen, 2);
 7976   jccb(Assembler::negative, L_fourth_loop_exit);
 7977   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7978   jmp(L_fourth_loop);
 7979   bind(L_fourth_loop_exit);
 7980 }
 7981 
 7982 /**
 7983  * Shift z[] left by 1 bit.
 7984  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7985  *
 7986  */
 7987 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7988 
 7989   Label L_fifth_loop, L_fifth_loop_exit;
 7990 
 7991   // Fifth loop
 7992   // Perform primitiveLeftShift(z, zlen, 1)
 7993 
 7994   const Register prev_carry = tmp1;
 7995   const Register new_carry = tmp4;
 7996   const Register value = tmp2;
 7997   const Register zidx = tmp3;
 7998 
 7999   // int zidx, carry;
 8000   // long value;
 8001   // carry = 0;
 8002   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 8003   //    (carry:value)  = (z[i] << 1) | carry ;
 8004   //    z[i] = value;
 8005   // }
 8006 
 8007   movl(zidx, zlen);
 8008   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 8009 
 8010   bind(L_fifth_loop);
 8011   decl(zidx);  // Use decl to preserve carry flag
 8012   decl(zidx);
 8013   jccb(Assembler::negative, L_fifth_loop_exit);
 8014 
 8015   if (UseBMI2Instructions) {
 8016      movq(value, Address(z, zidx, Address::times_4, 0));
 8017      rclq(value, 1);
 8018      rorxq(value, value, 32);
 8019      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 8020   }
 8021   else {
 8022     // clear new_carry
 8023     xorl(new_carry, new_carry);
 8024 
 8025     // Shift z[i] by 1, or in previous carry and save new carry
 8026     movq(value, Address(z, zidx, Address::times_4, 0));
 8027     shlq(value, 1);
 8028     adcl(new_carry, 0);
 8029 
 8030     orq(value, prev_carry);
 8031     rorq(value, 0x20);
 8032     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 8033 
 8034     // Set previous carry = new carry
 8035     movl(prev_carry, new_carry);
 8036   }
 8037   jmp(L_fifth_loop);
 8038 
 8039   bind(L_fifth_loop_exit);
 8040 }
 8041 
 8042 
 8043 /**
 8044  * Code for BigInteger::squareToLen() intrinsic
 8045  *
 8046  * rdi: x
 8047  * rsi: len
 8048  * r8:  z
 8049  * rcx: zlen
 8050  * r12: tmp1
 8051  * r13: tmp2
 8052  * r14: tmp3
 8053  * r15: tmp4
 8054  * rbx: tmp5
 8055  *
 8056  */
 8057 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) {
 8058 
 8059   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 8060   push(tmp1);
 8061   push(tmp2);
 8062   push(tmp3);
 8063   push(tmp4);
 8064   push(tmp5);
 8065 
 8066   // First loop
 8067   // Store the squares, right shifted one bit (i.e., divided by 2).
 8068   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8069 
 8070   // Add in off-diagonal sums.
 8071   //
 8072   // Second, third (nested) and fourth loops.
 8073   // zlen +=2;
 8074   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 8075   //    carry = 0;
 8076   //    long op2 = x[xidx:xidx+1];
 8077   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 8078   //       k -= 2;
 8079   //       long op1 = x[j:j+1];
 8080   //       long sum = z[k:k+1];
 8081   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 8082   //       z[k:k+1] = sum;
 8083   //    }
 8084   //    add_one_64(z, k, carry, tmp_regs);
 8085   // }
 8086 
 8087   const Register carry = tmp5;
 8088   const Register sum = tmp3;
 8089   const Register op1 = tmp4;
 8090   Register op2 = tmp2;
 8091 
 8092   push(zlen);
 8093   push(len);
 8094   addl(zlen,2);
 8095   bind(L_second_loop);
 8096   xorq(carry, carry);
 8097   subl(zlen, 4);
 8098   subl(len, 2);
 8099   push(zlen);
 8100   push(len);
 8101   cmpl(len, 0);
 8102   jccb(Assembler::lessEqual, L_second_loop_exit);
 8103 
 8104   // Multiply an array by one 64 bit long.
 8105   if (UseBMI2Instructions) {
 8106     op2 = rdxReg;
 8107     movq(op2, Address(x, len, Address::times_4,  0));
 8108     rorxq(op2, op2, 32);
 8109   }
 8110   else {
 8111     movq(op2, Address(x, len, Address::times_4,  0));
 8112     rorq(op2, 32);
 8113   }
 8114 
 8115   bind(L_third_loop);
 8116   decrementl(len);
 8117   jccb(Assembler::negative, L_third_loop_exit);
 8118   decrementl(len);
 8119   jccb(Assembler::negative, L_last_x);
 8120 
 8121   movq(op1, Address(x, len, Address::times_4,  0));
 8122   rorq(op1, 32);
 8123 
 8124   bind(L_multiply);
 8125   subl(zlen, 2);
 8126   movq(sum, Address(z, zlen, Address::times_4,  0));
 8127 
 8128   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 8129   if (UseBMI2Instructions) {
 8130     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 8131   }
 8132   else {
 8133     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8134   }
 8135 
 8136   movq(Address(z, zlen, Address::times_4, 0), sum);
 8137 
 8138   jmp(L_third_loop);
 8139   bind(L_third_loop_exit);
 8140 
 8141   // Fourth loop
 8142   // Add 64 bit long carry into z with carry propagation.
 8143   // Uses offsetted zlen.
 8144   add_one_64(z, zlen, carry, tmp1);
 8145 
 8146   pop(len);
 8147   pop(zlen);
 8148   jmp(L_second_loop);
 8149 
 8150   // Next infrequent code is moved outside loops.
 8151   bind(L_last_x);
 8152   movl(op1, Address(x, 0));
 8153   jmp(L_multiply);
 8154 
 8155   bind(L_second_loop_exit);
 8156   pop(len);
 8157   pop(zlen);
 8158   pop(len);
 8159   pop(zlen);
 8160 
 8161   // Fifth loop
 8162   // Shift z left 1 bit.
 8163   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 8164 
 8165   // z[zlen-1] |= x[len-1] & 1;
 8166   movl(tmp3, Address(x, len, Address::times_4, -4));
 8167   andl(tmp3, 1);
 8168   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 8169 
 8170   pop(tmp5);
 8171   pop(tmp4);
 8172   pop(tmp3);
 8173   pop(tmp2);
 8174   pop(tmp1);
 8175 }
 8176 
 8177 /**
 8178  * Helper function for mul_add()
 8179  * Multiply the in[] by int k and add to out[] starting at offset offs using
 8180  * 128 bit by 32 bit multiply and return the carry in tmp5.
 8181  * Only quad int aligned length of in[] is operated on in this function.
 8182  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 8183  * This function preserves out, in and k registers.
 8184  * len and offset point to the appropriate index in "in" & "out" correspondingly
 8185  * tmp5 has the carry.
 8186  * other registers are temporary and are modified.
 8187  *
 8188  */
 8189 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 8190   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 8191   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8192 
 8193   Label L_first_loop, L_first_loop_exit;
 8194 
 8195   movl(tmp1, len);
 8196   shrl(tmp1, 2);
 8197 
 8198   bind(L_first_loop);
 8199   subl(tmp1, 1);
 8200   jccb(Assembler::negative, L_first_loop_exit);
 8201 
 8202   subl(len, 4);
 8203   subl(offset, 4);
 8204 
 8205   Register op2 = tmp2;
 8206   const Register sum = tmp3;
 8207   const Register op1 = tmp4;
 8208   const Register carry = tmp5;
 8209 
 8210   if (UseBMI2Instructions) {
 8211     op2 = rdxReg;
 8212   }
 8213 
 8214   movq(op1, Address(in, len, Address::times_4,  8));
 8215   rorq(op1, 32);
 8216   movq(sum, Address(out, offset, Address::times_4,  8));
 8217   rorq(sum, 32);
 8218   if (UseBMI2Instructions) {
 8219     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8220   }
 8221   else {
 8222     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8223   }
 8224   // Store back in big endian from little endian
 8225   rorq(sum, 0x20);
 8226   movq(Address(out, offset, Address::times_4,  8), sum);
 8227 
 8228   movq(op1, Address(in, len, Address::times_4,  0));
 8229   rorq(op1, 32);
 8230   movq(sum, Address(out, offset, Address::times_4,  0));
 8231   rorq(sum, 32);
 8232   if (UseBMI2Instructions) {
 8233     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8234   }
 8235   else {
 8236     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8237   }
 8238   // Store back in big endian from little endian
 8239   rorq(sum, 0x20);
 8240   movq(Address(out, offset, Address::times_4,  0), sum);
 8241 
 8242   jmp(L_first_loop);
 8243   bind(L_first_loop_exit);
 8244 }
 8245 
 8246 /**
 8247  * Code for BigInteger::mulAdd() intrinsic
 8248  *
 8249  * rdi: out
 8250  * rsi: in
 8251  * r11: offs (out.length - offset)
 8252  * rcx: len
 8253  * r8:  k
 8254  * r12: tmp1
 8255  * r13: tmp2
 8256  * r14: tmp3
 8257  * r15: tmp4
 8258  * rbx: tmp5
 8259  * Multiply the in[] by word k and add to out[], return the carry in rax
 8260  */
 8261 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 8262    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 8263    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8264 
 8265   Label L_carry, L_last_in, L_done;
 8266 
 8267 // carry = 0;
 8268 // for (int j=len-1; j >= 0; j--) {
 8269 //    long product = (in[j] & LONG_MASK) * kLong +
 8270 //                   (out[offs] & LONG_MASK) + carry;
 8271 //    out[offs--] = (int)product;
 8272 //    carry = product >>> 32;
 8273 // }
 8274 //
 8275   push(tmp1);
 8276   push(tmp2);
 8277   push(tmp3);
 8278   push(tmp4);
 8279   push(tmp5);
 8280 
 8281   Register op2 = tmp2;
 8282   const Register sum = tmp3;
 8283   const Register op1 = tmp4;
 8284   const Register carry =  tmp5;
 8285 
 8286   if (UseBMI2Instructions) {
 8287     op2 = rdxReg;
 8288     movl(op2, k);
 8289   }
 8290   else {
 8291     movl(op2, k);
 8292   }
 8293 
 8294   xorq(carry, carry);
 8295 
 8296   //First loop
 8297 
 8298   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 8299   //The carry is in tmp5
 8300   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8301 
 8302   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 8303   decrementl(len);
 8304   jccb(Assembler::negative, L_carry);
 8305   decrementl(len);
 8306   jccb(Assembler::negative, L_last_in);
 8307 
 8308   movq(op1, Address(in, len, Address::times_4,  0));
 8309   rorq(op1, 32);
 8310 
 8311   subl(offs, 2);
 8312   movq(sum, Address(out, offs, Address::times_4,  0));
 8313   rorq(sum, 32);
 8314 
 8315   if (UseBMI2Instructions) {
 8316     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8317   }
 8318   else {
 8319     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8320   }
 8321 
 8322   // Store back in big endian from little endian
 8323   rorq(sum, 0x20);
 8324   movq(Address(out, offs, Address::times_4,  0), sum);
 8325 
 8326   testl(len, len);
 8327   jccb(Assembler::zero, L_carry);
 8328 
 8329   //Multiply the last in[] entry, if any
 8330   bind(L_last_in);
 8331   movl(op1, Address(in, 0));
 8332   movl(sum, Address(out, offs, Address::times_4,  -4));
 8333 
 8334   movl(raxReg, k);
 8335   mull(op1); //tmp4 * eax -> edx:eax
 8336   addl(sum, carry);
 8337   adcl(rdxReg, 0);
 8338   addl(sum, raxReg);
 8339   adcl(rdxReg, 0);
 8340   movl(carry, rdxReg);
 8341 
 8342   movl(Address(out, offs, Address::times_4,  -4), sum);
 8343 
 8344   bind(L_carry);
 8345   //return tmp5/carry as carry in rax
 8346   movl(rax, carry);
 8347 
 8348   bind(L_done);
 8349   pop(tmp5);
 8350   pop(tmp4);
 8351   pop(tmp3);
 8352   pop(tmp2);
 8353   pop(tmp1);
 8354 }
 8355 
 8356 /**
 8357  * Emits code to update CRC-32 with a byte value according to constants in table
 8358  *
 8359  * @param [in,out]crc   Register containing the crc.
 8360  * @param [in]val       Register containing the byte to fold into the CRC.
 8361  * @param [in]table     Register containing the table of crc constants.
 8362  *
 8363  * uint32_t crc;
 8364  * val = crc_table[(val ^ crc) & 0xFF];
 8365  * crc = val ^ (crc >> 8);
 8366  *
 8367  */
 8368 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 8369   xorl(val, crc);
 8370   andl(val, 0xFF);
 8371   shrl(crc, 8); // unsigned shift
 8372   xorl(crc, Address(table, val, Address::times_4, 0));
 8373 }
 8374 
 8375 /**
 8376  * Fold 128-bit data chunk
 8377  */
 8378 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 8379   if (UseAVX > 0) {
 8380     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 8381     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 8382     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 8383     pxor(xcrc, xtmp);
 8384   } else {
 8385     movdqa(xtmp, xcrc);
 8386     pclmulhdq(xtmp, xK);   // [123:64]
 8387     pclmulldq(xcrc, xK);   // [63:0]
 8388     pxor(xcrc, xtmp);
 8389     movdqu(xtmp, Address(buf, offset));
 8390     pxor(xcrc, xtmp);
 8391   }
 8392 }
 8393 
 8394 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 8395   if (UseAVX > 0) {
 8396     vpclmulhdq(xtmp, xK, xcrc);
 8397     vpclmulldq(xcrc, xK, xcrc);
 8398     pxor(xcrc, xbuf);
 8399     pxor(xcrc, xtmp);
 8400   } else {
 8401     movdqa(xtmp, xcrc);
 8402     pclmulhdq(xtmp, xK);
 8403     pclmulldq(xcrc, xK);
 8404     pxor(xcrc, xbuf);
 8405     pxor(xcrc, xtmp);
 8406   }
 8407 }
 8408 
 8409 /**
 8410  * 8-bit folds to compute 32-bit CRC
 8411  *
 8412  * uint64_t xcrc;
 8413  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 8414  */
 8415 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 8416   movdl(tmp, xcrc);
 8417   andl(tmp, 0xFF);
 8418   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 8419   psrldq(xcrc, 1); // unsigned shift one byte
 8420   pxor(xcrc, xtmp);
 8421 }
 8422 
 8423 /**
 8424  * uint32_t crc;
 8425  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 8426  */
 8427 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 8428   movl(tmp, crc);
 8429   andl(tmp, 0xFF);
 8430   shrl(crc, 8);
 8431   xorl(crc, Address(table, tmp, Address::times_4, 0));
 8432 }
 8433 
 8434 /**
 8435  * @param crc   register containing existing CRC (32-bit)
 8436  * @param buf   register pointing to input byte buffer (byte*)
 8437  * @param len   register containing number of bytes
 8438  * @param table register that will contain address of CRC table
 8439  * @param tmp   scratch register
 8440  */
 8441 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 8442   assert_different_registers(crc, buf, len, table, tmp, rax);
 8443 
 8444   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8445   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8446 
 8447   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8448   // context for the registers used, where all instructions below are using 128-bit mode
 8449   // On EVEX without VL and BW, these instructions will all be AVX.
 8450   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 8451   notl(crc); // ~crc
 8452   cmpl(len, 16);
 8453   jcc(Assembler::less, L_tail);
 8454 
 8455   // Align buffer to 16 bytes
 8456   movl(tmp, buf);
 8457   andl(tmp, 0xF);
 8458   jccb(Assembler::zero, L_aligned);
 8459   subl(tmp,  16);
 8460   addl(len, tmp);
 8461 
 8462   align(4);
 8463   BIND(L_align_loop);
 8464   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8465   update_byte_crc32(crc, rax, table);
 8466   increment(buf);
 8467   incrementl(tmp);
 8468   jccb(Assembler::less, L_align_loop);
 8469 
 8470   BIND(L_aligned);
 8471   movl(tmp, len); // save
 8472   shrl(len, 4);
 8473   jcc(Assembler::zero, L_tail_restore);
 8474 
 8475   // Fold crc into first bytes of vector
 8476   movdqa(xmm1, Address(buf, 0));
 8477   movdl(rax, xmm1);
 8478   xorl(crc, rax);
 8479   if (VM_Version::supports_sse4_1()) {
 8480     pinsrd(xmm1, crc, 0);
 8481   } else {
 8482     pinsrw(xmm1, crc, 0);
 8483     shrl(crc, 16);
 8484     pinsrw(xmm1, crc, 1);
 8485   }
 8486   addptr(buf, 16);
 8487   subl(len, 4); // len > 0
 8488   jcc(Assembler::less, L_fold_tail);
 8489 
 8490   movdqa(xmm2, Address(buf,  0));
 8491   movdqa(xmm3, Address(buf, 16));
 8492   movdqa(xmm4, Address(buf, 32));
 8493   addptr(buf, 48);
 8494   subl(len, 3);
 8495   jcc(Assembler::lessEqual, L_fold_512b);
 8496 
 8497   // Fold total 512 bits of polynomial on each iteration,
 8498   // 128 bits per each of 4 parallel streams.
 8499   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 8500 
 8501   align32();
 8502   BIND(L_fold_512b_loop);
 8503   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8504   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 8505   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 8506   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 8507   addptr(buf, 64);
 8508   subl(len, 4);
 8509   jcc(Assembler::greater, L_fold_512b_loop);
 8510 
 8511   // Fold 512 bits to 128 bits.
 8512   BIND(L_fold_512b);
 8513   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8514   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 8515   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 8516   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 8517 
 8518   // Fold the rest of 128 bits data chunks
 8519   BIND(L_fold_tail);
 8520   addl(len, 3);
 8521   jccb(Assembler::lessEqual, L_fold_128b);
 8522   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8523 
 8524   BIND(L_fold_tail_loop);
 8525   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8526   addptr(buf, 16);
 8527   decrementl(len);
 8528   jccb(Assembler::greater, L_fold_tail_loop);
 8529 
 8530   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 8531   BIND(L_fold_128b);
 8532   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 8533   if (UseAVX > 0) {
 8534     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 8535     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 8536     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 8537   } else {
 8538     movdqa(xmm2, xmm0);
 8539     pclmulqdq(xmm2, xmm1, 0x1);
 8540     movdqa(xmm3, xmm0);
 8541     pand(xmm3, xmm2);
 8542     pclmulqdq(xmm0, xmm3, 0x1);
 8543   }
 8544   psrldq(xmm1, 8);
 8545   psrldq(xmm2, 4);
 8546   pxor(xmm0, xmm1);
 8547   pxor(xmm0, xmm2);
 8548 
 8549   // 8 8-bit folds to compute 32-bit CRC.
 8550   for (int j = 0; j < 4; j++) {
 8551     fold_8bit_crc32(xmm0, table, xmm1, rax);
 8552   }
 8553   movdl(crc, xmm0); // mov 32 bits to general register
 8554   for (int j = 0; j < 4; j++) {
 8555     fold_8bit_crc32(crc, table, rax);
 8556   }
 8557 
 8558   BIND(L_tail_restore);
 8559   movl(len, tmp); // restore
 8560   BIND(L_tail);
 8561   andl(len, 0xf);
 8562   jccb(Assembler::zero, L_exit);
 8563 
 8564   // Fold the rest of bytes
 8565   align(4);
 8566   BIND(L_tail_loop);
 8567   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8568   update_byte_crc32(crc, rax, table);
 8569   increment(buf);
 8570   decrementl(len);
 8571   jccb(Assembler::greater, L_tail_loop);
 8572 
 8573   BIND(L_exit);
 8574   notl(crc); // ~c
 8575 }
 8576 
 8577 // Helper function for AVX 512 CRC32
 8578 // Fold 512-bit data chunks
 8579 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 8580                                              Register pos, int offset) {
 8581   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 8582   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 8583   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 8584   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 8585   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 8586 }
 8587 
 8588 // Helper function for AVX 512 CRC32
 8589 // Compute CRC32 for < 256B buffers
 8590 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 8591                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 8592                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 8593 
 8594   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 8595   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 8596   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 8597 
 8598   // check if there is enough buffer to be able to fold 16B at a time
 8599   cmpl(len, 32);
 8600   jcc(Assembler::less, L_less_than_32);
 8601 
 8602   // if there is, load the constants
 8603   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 8604   movdl(xmm0, crc);                        // get the initial crc value
 8605   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8606   pxor(xmm7, xmm0);
 8607 
 8608   // update the buffer pointer
 8609   addl(pos, 16);
 8610   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 8611   subl(len, 32);
 8612   jmp(L_16B_reduction_loop);
 8613 
 8614   bind(L_less_than_32);
 8615   //mov initial crc to the return value. this is necessary for zero - length buffers.
 8616   movl(rax, crc);
 8617   testl(len, len);
 8618   jcc(Assembler::equal, L_cleanup);
 8619 
 8620   movdl(xmm0, crc);                        //get the initial crc value
 8621 
 8622   cmpl(len, 16);
 8623   jcc(Assembler::equal, L_exact_16_left);
 8624   jcc(Assembler::less, L_less_than_16_left);
 8625 
 8626   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8627   pxor(xmm7, xmm0);                       //xor the initial crc value
 8628   addl(pos, 16);
 8629   subl(len, 16);
 8630   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 8631   jmp(L_get_last_two_xmms);
 8632 
 8633   bind(L_less_than_16_left);
 8634   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 8635   pxor(xmm1, xmm1);
 8636   movptr(tmp1, rsp);
 8637   movdqu(Address(tmp1, 0 * 16), xmm1);
 8638 
 8639   cmpl(len, 4);
 8640   jcc(Assembler::less, L_only_less_than_4);
 8641 
 8642   //backup the counter value
 8643   movl(tmp2, len);
 8644   cmpl(len, 8);
 8645   jcc(Assembler::less, L_less_than_8_left);
 8646 
 8647   //load 8 Bytes
 8648   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 8649   movq(Address(tmp1, 0 * 16), rax);
 8650   addptr(tmp1, 8);
 8651   subl(len, 8);
 8652   addl(pos, 8);
 8653 
 8654   bind(L_less_than_8_left);
 8655   cmpl(len, 4);
 8656   jcc(Assembler::less, L_less_than_4_left);
 8657 
 8658   //load 4 Bytes
 8659   movl(rax, Address(buf, pos, Address::times_1, 0));
 8660   movl(Address(tmp1, 0 * 16), rax);
 8661   addptr(tmp1, 4);
 8662   subl(len, 4);
 8663   addl(pos, 4);
 8664 
 8665   bind(L_less_than_4_left);
 8666   cmpl(len, 2);
 8667   jcc(Assembler::less, L_less_than_2_left);
 8668 
 8669   // load 2 Bytes
 8670   movw(rax, Address(buf, pos, Address::times_1, 0));
 8671   movl(Address(tmp1, 0 * 16), rax);
 8672   addptr(tmp1, 2);
 8673   subl(len, 2);
 8674   addl(pos, 2);
 8675 
 8676   bind(L_less_than_2_left);
 8677   cmpl(len, 1);
 8678   jcc(Assembler::less, L_zero_left);
 8679 
 8680   // load 1 Byte
 8681   movb(rax, Address(buf, pos, Address::times_1, 0));
 8682   movb(Address(tmp1, 0 * 16), rax);
 8683 
 8684   bind(L_zero_left);
 8685   movdqu(xmm7, Address(rsp, 0));
 8686   pxor(xmm7, xmm0);                       //xor the initial crc value
 8687 
 8688   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8689   movdqu(xmm0, Address(rax, tmp2));
 8690   pshufb(xmm7, xmm0);
 8691   jmp(L_128_done);
 8692 
 8693   bind(L_exact_16_left);
 8694   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 8695   pxor(xmm7, xmm0);                       //xor the initial crc value
 8696   jmp(L_128_done);
 8697 
 8698   bind(L_only_less_than_4);
 8699   cmpl(len, 3);
 8700   jcc(Assembler::less, L_only_less_than_3);
 8701 
 8702   // load 3 Bytes
 8703   movb(rax, Address(buf, pos, Address::times_1, 0));
 8704   movb(Address(tmp1, 0), rax);
 8705 
 8706   movb(rax, Address(buf, pos, Address::times_1, 1));
 8707   movb(Address(tmp1, 1), rax);
 8708 
 8709   movb(rax, Address(buf, pos, Address::times_1, 2));
 8710   movb(Address(tmp1, 2), rax);
 8711 
 8712   movdqu(xmm7, Address(rsp, 0));
 8713   pxor(xmm7, xmm0);                     //xor the initial crc value
 8714 
 8715   pslldq(xmm7, 0x5);
 8716   jmp(L_barrett);
 8717   bind(L_only_less_than_3);
 8718   cmpl(len, 2);
 8719   jcc(Assembler::less, L_only_less_than_2);
 8720 
 8721   // load 2 Bytes
 8722   movb(rax, Address(buf, pos, Address::times_1, 0));
 8723   movb(Address(tmp1, 0), rax);
 8724 
 8725   movb(rax, Address(buf, pos, Address::times_1, 1));
 8726   movb(Address(tmp1, 1), rax);
 8727 
 8728   movdqu(xmm7, Address(rsp, 0));
 8729   pxor(xmm7, xmm0);                     //xor the initial crc value
 8730 
 8731   pslldq(xmm7, 0x6);
 8732   jmp(L_barrett);
 8733 
 8734   bind(L_only_less_than_2);
 8735   //load 1 Byte
 8736   movb(rax, Address(buf, pos, Address::times_1, 0));
 8737   movb(Address(tmp1, 0), rax);
 8738 
 8739   movdqu(xmm7, Address(rsp, 0));
 8740   pxor(xmm7, xmm0);                     //xor the initial crc value
 8741 
 8742   pslldq(xmm7, 0x7);
 8743 }
 8744 
 8745 /**
 8746 * Compute CRC32 using AVX512 instructions
 8747 * param crc   register containing existing CRC (32-bit)
 8748 * param buf   register pointing to input byte buffer (byte*)
 8749 * param len   register containing number of bytes
 8750 * param table address of crc or crc32c table
 8751 * param tmp1  scratch register
 8752 * param tmp2  scratch register
 8753 * return rax  result register
 8754 *
 8755 * This routine is identical for crc32c with the exception of the precomputed constant
 8756 * table which will be passed as the table argument.  The calculation steps are
 8757 * the same for both variants.
 8758 */
 8759 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8760   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8761 
 8762   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8763   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8764   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8765   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8766   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8767 
 8768   const Register pos = r12;
 8769   push(r12);
 8770   subptr(rsp, 16 * 2 + 8);
 8771 
 8772   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8773   // context for the registers used, where all instructions below are using 128-bit mode
 8774   // On EVEX without VL and BW, these instructions will all be AVX.
 8775   movl(pos, 0);
 8776 
 8777   // check if smaller than 256B
 8778   cmpl(len, 256);
 8779   jcc(Assembler::less, L_less_than_256);
 8780 
 8781   // load the initial crc value
 8782   movdl(xmm10, crc);
 8783 
 8784   // receive the initial 64B data, xor the initial crc value
 8785   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8786   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8787   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8788   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8789 
 8790   subl(len, 256);
 8791   cmpl(len, 256);
 8792   jcc(Assembler::less, L_fold_128_B_loop);
 8793 
 8794   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8795   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8796   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8797   subl(len, 256);
 8798 
 8799   bind(L_fold_256_B_loop);
 8800   addl(pos, 256);
 8801   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8802   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8803   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8804   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8805 
 8806   subl(len, 256);
 8807   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8808 
 8809   // Fold 256 into 128
 8810   addl(pos, 256);
 8811   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8812   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8813   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8814 
 8815   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8816   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8817   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8818 
 8819   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8820   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8821 
 8822   addl(len, 128);
 8823   jmp(L_fold_128_B_register);
 8824 
 8825   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8826   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8827 
 8828   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8829   bind(L_fold_128_B_loop);
 8830   addl(pos, 128);
 8831   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8832   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8833 
 8834   subl(len, 128);
 8835   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8836 
 8837   addl(pos, 128);
 8838 
 8839   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8840   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8841   bind(L_fold_128_B_register);
 8842   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8843   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8844   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8845   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8846   // save last that has no multiplicand
 8847   vextracti64x2(xmm7, xmm4, 3);
 8848 
 8849   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8850   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8851   // Needed later in reduction loop
 8852   movdqu(xmm10, Address(table, 1 * 16));
 8853   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8854   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8855 
 8856   // Swap 1,0,3,2 - 01 00 11 10
 8857   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8858   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8859   vextracti128(xmm5, xmm8, 1);
 8860   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8861 
 8862   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8863   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8864   addl(len, 128 - 16);
 8865   jcc(Assembler::less, L_final_reduction_for_128);
 8866 
 8867   bind(L_16B_reduction_loop);
 8868   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8869   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8870   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8871   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8872   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8873   addl(pos, 16);
 8874   subl(len, 16);
 8875   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8876 
 8877   bind(L_final_reduction_for_128);
 8878   addl(len, 16);
 8879   jcc(Assembler::equal, L_128_done);
 8880 
 8881   bind(L_get_last_two_xmms);
 8882   movdqu(xmm2, xmm7);
 8883   addl(pos, len);
 8884   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8885   subl(pos, len);
 8886 
 8887   // get rid of the extra data that was loaded before
 8888   // load the shift constant
 8889   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8890   movdqu(xmm0, Address(rax, len));
 8891   addl(rax, len);
 8892 
 8893   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8894   //Change mask to 512
 8895   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8896   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8897 
 8898   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8899   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8900   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8901   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8902   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8903 
 8904   bind(L_128_done);
 8905   // compute crc of a 128-bit value
 8906   movdqu(xmm10, Address(table, 3 * 16));
 8907   movdqu(xmm0, xmm7);
 8908 
 8909   // 64b fold
 8910   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8911   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8912   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8913 
 8914   // 32b fold
 8915   movdqu(xmm0, xmm7);
 8916   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8917   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8918   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8919   jmp(L_barrett);
 8920 
 8921   bind(L_less_than_256);
 8922   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);
 8923 
 8924   //barrett reduction
 8925   bind(L_barrett);
 8926   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8927   movdqu(xmm1, xmm7);
 8928   movdqu(xmm2, xmm7);
 8929   movdqu(xmm10, Address(table, 4 * 16));
 8930 
 8931   pclmulqdq(xmm7, xmm10, 0x0);
 8932   pxor(xmm7, xmm2);
 8933   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8934   movdqu(xmm2, xmm7);
 8935   pclmulqdq(xmm7, xmm10, 0x10);
 8936   pxor(xmm7, xmm2);
 8937   pxor(xmm7, xmm1);
 8938   pextrd(crc, xmm7, 2);
 8939 
 8940   bind(L_cleanup);
 8941   addptr(rsp, 16 * 2 + 8);
 8942   pop(r12);
 8943 }
 8944 
 8945 // S. Gueron / Information Processing Letters 112 (2012) 184
 8946 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8947 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8948 // Output: the 64-bit carry-less product of B * CONST
 8949 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8950                                      Register tmp1, Register tmp2, Register tmp3) {
 8951   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8952   if (n > 0) {
 8953     addq(tmp3, n * 256 * 8);
 8954   }
 8955   //    Q1 = TABLEExt[n][B & 0xFF];
 8956   movl(tmp1, in);
 8957   andl(tmp1, 0x000000FF);
 8958   shll(tmp1, 3);
 8959   addq(tmp1, tmp3);
 8960   movq(tmp1, Address(tmp1, 0));
 8961 
 8962   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8963   movl(tmp2, in);
 8964   shrl(tmp2, 8);
 8965   andl(tmp2, 0x000000FF);
 8966   shll(tmp2, 3);
 8967   addq(tmp2, tmp3);
 8968   movq(tmp2, Address(tmp2, 0));
 8969 
 8970   shlq(tmp2, 8);
 8971   xorq(tmp1, tmp2);
 8972 
 8973   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8974   movl(tmp2, in);
 8975   shrl(tmp2, 16);
 8976   andl(tmp2, 0x000000FF);
 8977   shll(tmp2, 3);
 8978   addq(tmp2, tmp3);
 8979   movq(tmp2, Address(tmp2, 0));
 8980 
 8981   shlq(tmp2, 16);
 8982   xorq(tmp1, tmp2);
 8983 
 8984   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8985   shrl(in, 24);
 8986   andl(in, 0x000000FF);
 8987   shll(in, 3);
 8988   addq(in, tmp3);
 8989   movq(in, Address(in, 0));
 8990 
 8991   shlq(in, 24);
 8992   xorq(in, tmp1);
 8993   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8994 }
 8995 
 8996 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8997                                       Register in_out,
 8998                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8999                                       XMMRegister w_xtmp2,
 9000                                       Register tmp1,
 9001                                       Register n_tmp2, Register n_tmp3) {
 9002   if (is_pclmulqdq_supported) {
 9003     movdl(w_xtmp1, in_out); // modified blindly
 9004 
 9005     movl(tmp1, const_or_pre_comp_const_index);
 9006     movdl(w_xtmp2, tmp1);
 9007     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 9008 
 9009     movdq(in_out, w_xtmp1);
 9010   } else {
 9011     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 9012   }
 9013 }
 9014 
 9015 // Recombination Alternative 2: No bit-reflections
 9016 // T1 = (CRC_A * U1) << 1
 9017 // T2 = (CRC_B * U2) << 1
 9018 // C1 = T1 >> 32
 9019 // C2 = T2 >> 32
 9020 // T1 = T1 & 0xFFFFFFFF
 9021 // T2 = T2 & 0xFFFFFFFF
 9022 // T1 = CRC32(0, T1)
 9023 // T2 = CRC32(0, T2)
 9024 // C1 = C1 ^ T1
 9025 // C2 = C2 ^ T2
 9026 // CRC = C1 ^ C2 ^ CRC_C
 9027 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,
 9028                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9029                                      Register tmp1, Register tmp2,
 9030                                      Register n_tmp3) {
 9031   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 9032   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 9033   shlq(in_out, 1);
 9034   movl(tmp1, in_out);
 9035   shrq(in_out, 32);
 9036   xorl(tmp2, tmp2);
 9037   crc32(tmp2, tmp1, 4);
 9038   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 9039   shlq(in1, 1);
 9040   movl(tmp1, in1);
 9041   shrq(in1, 32);
 9042   xorl(tmp2, tmp2);
 9043   crc32(tmp2, tmp1, 4);
 9044   xorl(in1, tmp2);
 9045   xorl(in_out, in1);
 9046   xorl(in_out, in2);
 9047 }
 9048 
 9049 // Set N to predefined value
 9050 // Subtract from a length of a buffer
 9051 // execute in a loop:
 9052 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 9053 // for i = 1 to N do
 9054 //  CRC_A = CRC32(CRC_A, A[i])
 9055 //  CRC_B = CRC32(CRC_B, B[i])
 9056 //  CRC_C = CRC32(CRC_C, C[i])
 9057 // end for
 9058 // Recombine
 9059 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,
 9060                                        Register in_out1, Register in_out2, Register in_out3,
 9061                                        Register tmp1, Register tmp2, Register tmp3,
 9062                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9063                                        Register tmp4, Register tmp5,
 9064                                        Register n_tmp6) {
 9065   Label L_processPartitions;
 9066   Label L_processPartition;
 9067   Label L_exit;
 9068 
 9069   bind(L_processPartitions);
 9070   cmpl(in_out1, 3 * size);
 9071   jcc(Assembler::less, L_exit);
 9072     xorl(tmp1, tmp1);
 9073     xorl(tmp2, tmp2);
 9074     movq(tmp3, in_out2);
 9075     addq(tmp3, size);
 9076 
 9077     bind(L_processPartition);
 9078       crc32(in_out3, Address(in_out2, 0), 8);
 9079       crc32(tmp1, Address(in_out2, size), 8);
 9080       crc32(tmp2, Address(in_out2, size * 2), 8);
 9081       addq(in_out2, 8);
 9082       cmpq(in_out2, tmp3);
 9083       jcc(Assembler::less, L_processPartition);
 9084     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 9085             w_xtmp1, w_xtmp2, w_xtmp3,
 9086             tmp4, tmp5,
 9087             n_tmp6);
 9088     addq(in_out2, 2 * size);
 9089     subl(in_out1, 3 * size);
 9090     jmp(L_processPartitions);
 9091 
 9092   bind(L_exit);
 9093 }
 9094 
 9095 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 9096 // Input: A buffer I of L bytes.
 9097 // Output: the CRC32C value of the buffer.
 9098 // Notations:
 9099 // Write L = 24N + r, with N = floor (L/24).
 9100 // r = L mod 24 (0 <= r < 24).
 9101 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 9102 // N quadwords, and R consists of r bytes.
 9103 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 9104 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 9105 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 9106 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 9107 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 9108                                           Register tmp1, Register tmp2, Register tmp3,
 9109                                           Register tmp4, Register tmp5, Register tmp6,
 9110                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9111                                           bool is_pclmulqdq_supported) {
 9112   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 9113   Label L_wordByWord;
 9114   Label L_byteByByteProlog;
 9115   Label L_byteByByte;
 9116   Label L_exit;
 9117 
 9118   if (is_pclmulqdq_supported ) {
 9119     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 9120     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 9121 
 9122     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 9123     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 9124 
 9125     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 9126     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 9127     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 9128   } else {
 9129     const_or_pre_comp_const_index[0] = 1;
 9130     const_or_pre_comp_const_index[1] = 0;
 9131 
 9132     const_or_pre_comp_const_index[2] = 3;
 9133     const_or_pre_comp_const_index[3] = 2;
 9134 
 9135     const_or_pre_comp_const_index[4] = 5;
 9136     const_or_pre_comp_const_index[5] = 4;
 9137    }
 9138   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 9139                     in2, in1, in_out,
 9140                     tmp1, tmp2, tmp3,
 9141                     w_xtmp1, w_xtmp2, w_xtmp3,
 9142                     tmp4, tmp5,
 9143                     tmp6);
 9144   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 9145                     in2, in1, in_out,
 9146                     tmp1, tmp2, tmp3,
 9147                     w_xtmp1, w_xtmp2, w_xtmp3,
 9148                     tmp4, tmp5,
 9149                     tmp6);
 9150   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 9151                     in2, in1, in_out,
 9152                     tmp1, tmp2, tmp3,
 9153                     w_xtmp1, w_xtmp2, w_xtmp3,
 9154                     tmp4, tmp5,
 9155                     tmp6);
 9156   movl(tmp1, in2);
 9157   andl(tmp1, 0x00000007);
 9158   negl(tmp1);
 9159   addl(tmp1, in2);
 9160   addq(tmp1, in1);
 9161 
 9162   cmpq(in1, tmp1);
 9163   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 9164   align(16);
 9165   BIND(L_wordByWord);
 9166     crc32(in_out, Address(in1, 0), 8);
 9167     addq(in1, 8);
 9168     cmpq(in1, tmp1);
 9169     jcc(Assembler::less, L_wordByWord);
 9170 
 9171   BIND(L_byteByByteProlog);
 9172   andl(in2, 0x00000007);
 9173   movl(tmp2, 1);
 9174 
 9175   cmpl(tmp2, in2);
 9176   jccb(Assembler::greater, L_exit);
 9177   BIND(L_byteByByte);
 9178     crc32(in_out, Address(in1, 0), 1);
 9179     incq(in1);
 9180     incl(tmp2);
 9181     cmpl(tmp2, in2);
 9182     jcc(Assembler::lessEqual, L_byteByByte);
 9183 
 9184   BIND(L_exit);
 9185 }
 9186 #undef BIND
 9187 #undef BLOCK_COMMENT
 9188 
 9189 // Compress char[] array to byte[].
 9190 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 9191 // Return the array length if every element in array can be encoded,
 9192 // otherwise, the index of first non-latin1 (> 0xff) character.
 9193 //   @IntrinsicCandidate
 9194 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 9195 //     for (int i = 0; i < len; i++) {
 9196 //       char c = src[srcOff];
 9197 //       if (c > 0xff) {
 9198 //           return i;  // return index of non-latin1 char
 9199 //       }
 9200 //       dst[dstOff] = (byte)c;
 9201 //       srcOff++;
 9202 //       dstOff++;
 9203 //     }
 9204 //     return len;
 9205 //   }
 9206 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 9207   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 9208   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 9209   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 9210   Label copy_chars_loop, done, reset_sp, copy_tail;
 9211 
 9212   // rsi: src
 9213   // rdi: dst
 9214   // rdx: len
 9215   // rcx: tmp5
 9216   // rax: result
 9217 
 9218   // rsi holds start addr of source char[] to be compressed
 9219   // rdi holds start addr of destination byte[]
 9220   // rdx holds length
 9221 
 9222   assert(len != result, "");
 9223 
 9224   // save length for return
 9225   movl(result, len);
 9226 
 9227   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 9228     VM_Version::supports_avx512vlbw() &&
 9229     VM_Version::supports_bmi2()) {
 9230 
 9231     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 9232 
 9233     // alignment
 9234     Label post_alignment;
 9235 
 9236     // if length of the string is less than 32, handle it the old fashioned way
 9237     testl(len, -32);
 9238     jcc(Assembler::zero, below_threshold);
 9239 
 9240     // First check whether a character is compressible ( <= 0xFF).
 9241     // Create mask to test for Unicode chars inside zmm vector
 9242     movl(tmp5, 0x00FF);
 9243     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 9244 
 9245     testl(len, -64);
 9246     jccb(Assembler::zero, post_alignment);
 9247 
 9248     movl(tmp5, dst);
 9249     andl(tmp5, (32 - 1));
 9250     negl(tmp5);
 9251     andl(tmp5, (32 - 1));
 9252 
 9253     // bail out when there is nothing to be done
 9254     testl(tmp5, 0xFFFFFFFF);
 9255     jccb(Assembler::zero, post_alignment);
 9256 
 9257     // ~(~0 << len), where len is the # of remaining elements to process
 9258     movl(len, 0xFFFFFFFF);
 9259     shlxl(len, len, tmp5);
 9260     notl(len);
 9261     kmovdl(mask2, len);
 9262     movl(len, result);
 9263 
 9264     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9265     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9266     ktestd(mask1, mask2);
 9267     jcc(Assembler::carryClear, copy_tail);
 9268 
 9269     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9270 
 9271     addptr(src, tmp5);
 9272     addptr(src, tmp5);
 9273     addptr(dst, tmp5);
 9274     subl(len, tmp5);
 9275 
 9276     bind(post_alignment);
 9277     // end of alignment
 9278 
 9279     movl(tmp5, len);
 9280     andl(tmp5, (32 - 1));    // tail count (in chars)
 9281     andl(len, ~(32 - 1));    // vector count (in chars)
 9282     jccb(Assembler::zero, copy_loop_tail);
 9283 
 9284     lea(src, Address(src, len, Address::times_2));
 9285     lea(dst, Address(dst, len, Address::times_1));
 9286     negptr(len);
 9287 
 9288     bind(copy_32_loop);
 9289     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 9290     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 9291     kortestdl(mask1, mask1);
 9292     jccb(Assembler::carryClear, reset_for_copy_tail);
 9293 
 9294     // All elements in current processed chunk are valid candidates for
 9295     // compression. Write a truncated byte elements to the memory.
 9296     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 9297     addptr(len, 32);
 9298     jccb(Assembler::notZero, copy_32_loop);
 9299 
 9300     bind(copy_loop_tail);
 9301     // bail out when there is nothing to be done
 9302     testl(tmp5, 0xFFFFFFFF);
 9303     jcc(Assembler::zero, done);
 9304 
 9305     movl(len, tmp5);
 9306 
 9307     // ~(~0 << len), where len is the # of remaining elements to process
 9308     movl(tmp5, 0xFFFFFFFF);
 9309     shlxl(tmp5, tmp5, len);
 9310     notl(tmp5);
 9311 
 9312     kmovdl(mask2, tmp5);
 9313 
 9314     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9315     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9316     ktestd(mask1, mask2);
 9317     jcc(Assembler::carryClear, copy_tail);
 9318 
 9319     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9320     jmp(done);
 9321 
 9322     bind(reset_for_copy_tail);
 9323     lea(src, Address(src, tmp5, Address::times_2));
 9324     lea(dst, Address(dst, tmp5, Address::times_1));
 9325     subptr(len, tmp5);
 9326     jmp(copy_chars_loop);
 9327 
 9328     bind(below_threshold);
 9329   }
 9330 
 9331   if (UseSSE42Intrinsics) {
 9332     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 9333 
 9334     // vectored compression
 9335     testl(len, 0xfffffff8);
 9336     jcc(Assembler::zero, copy_tail);
 9337 
 9338     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 9339     movdl(tmp1Reg, tmp5);
 9340     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 9341 
 9342     andl(len, 0xfffffff0);
 9343     jccb(Assembler::zero, copy_16);
 9344 
 9345     // compress 16 chars per iter
 9346     pxor(tmp4Reg, tmp4Reg);
 9347 
 9348     lea(src, Address(src, len, Address::times_2));
 9349     lea(dst, Address(dst, len, Address::times_1));
 9350     negptr(len);
 9351 
 9352     bind(copy_32_loop);
 9353     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 9354     por(tmp4Reg, tmp2Reg);
 9355     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 9356     por(tmp4Reg, tmp3Reg);
 9357     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 9358     jccb(Assembler::notZero, reset_for_copy_tail);
 9359     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 9360     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 9361     addptr(len, 16);
 9362     jccb(Assembler::notZero, copy_32_loop);
 9363 
 9364     // compress next vector of 8 chars (if any)
 9365     bind(copy_16);
 9366     // len = 0
 9367     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 9368     jccb(Assembler::zero, copy_tail_sse);
 9369 
 9370     pxor(tmp3Reg, tmp3Reg);
 9371 
 9372     movdqu(tmp2Reg, Address(src, 0));
 9373     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 9374     jccb(Assembler::notZero, reset_for_copy_tail);
 9375     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 9376     movq(Address(dst, 0), tmp2Reg);
 9377     addptr(src, 16);
 9378     addptr(dst, 8);
 9379     jmpb(copy_tail_sse);
 9380 
 9381     bind(reset_for_copy_tail);
 9382     movl(tmp5, result);
 9383     andl(tmp5, 0x0000000f);
 9384     lea(src, Address(src, tmp5, Address::times_2));
 9385     lea(dst, Address(dst, tmp5, Address::times_1));
 9386     subptr(len, tmp5);
 9387     jmpb(copy_chars_loop);
 9388 
 9389     bind(copy_tail_sse);
 9390     movl(len, result);
 9391     andl(len, 0x00000007);    // tail count (in chars)
 9392   }
 9393   // compress 1 char per iter
 9394   bind(copy_tail);
 9395   testl(len, len);
 9396   jccb(Assembler::zero, done);
 9397   lea(src, Address(src, len, Address::times_2));
 9398   lea(dst, Address(dst, len, Address::times_1));
 9399   negptr(len);
 9400 
 9401   bind(copy_chars_loop);
 9402   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 9403   testl(tmp5, 0xff00);      // check if Unicode char
 9404   jccb(Assembler::notZero, reset_sp);
 9405   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 9406   increment(len);
 9407   jccb(Assembler::notZero, copy_chars_loop);
 9408 
 9409   // add len then return (len will be zero if compress succeeded, otherwise negative)
 9410   bind(reset_sp);
 9411   addl(result, len);
 9412 
 9413   bind(done);
 9414 }
 9415 
 9416 // Inflate byte[] array to char[].
 9417 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 9418 //   @IntrinsicCandidate
 9419 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 9420 //     for (int i = 0; i < len; i++) {
 9421 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 9422 //     }
 9423 //   }
 9424 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 9425   XMMRegister tmp1, Register tmp2, KRegister mask) {
 9426   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 9427   // rsi: src
 9428   // rdi: dst
 9429   // rdx: len
 9430   // rcx: tmp2
 9431 
 9432   // rsi holds start addr of source byte[] to be inflated
 9433   // rdi holds start addr of destination char[]
 9434   // rdx holds length
 9435   assert_different_registers(src, dst, len, tmp2);
 9436   movl(tmp2, len);
 9437   if ((UseAVX > 2) && // AVX512
 9438     VM_Version::supports_avx512vlbw() &&
 9439     VM_Version::supports_bmi2()) {
 9440 
 9441     Label copy_32_loop, copy_tail;
 9442     Register tmp3_aliased = len;
 9443 
 9444     // if length of the string is less than 16, handle it in an old fashioned way
 9445     testl(len, -16);
 9446     jcc(Assembler::zero, below_threshold);
 9447 
 9448     testl(len, -1 * AVX3Threshold);
 9449     jcc(Assembler::zero, avx3_threshold);
 9450 
 9451     // In order to use only one arithmetic operation for the main loop we use
 9452     // this pre-calculation
 9453     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 9454     andl(len, -32);     // vector count
 9455     jccb(Assembler::zero, copy_tail);
 9456 
 9457     lea(src, Address(src, len, Address::times_1));
 9458     lea(dst, Address(dst, len, Address::times_2));
 9459     negptr(len);
 9460 
 9461 
 9462     // inflate 32 chars per iter
 9463     bind(copy_32_loop);
 9464     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 9465     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 9466     addptr(len, 32);
 9467     jcc(Assembler::notZero, copy_32_loop);
 9468 
 9469     bind(copy_tail);
 9470     // bail out when there is nothing to be done
 9471     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 9472     jcc(Assembler::zero, done);
 9473 
 9474     // ~(~0 << length), where length is the # of remaining elements to process
 9475     movl(tmp3_aliased, -1);
 9476     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 9477     notl(tmp3_aliased);
 9478     kmovdl(mask, tmp3_aliased);
 9479     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 9480     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 9481 
 9482     jmp(done);
 9483     bind(avx3_threshold);
 9484   }
 9485   if (UseSSE42Intrinsics) {
 9486     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 9487 
 9488     if (UseAVX > 1) {
 9489       andl(tmp2, (16 - 1));
 9490       andl(len, -16);
 9491       jccb(Assembler::zero, copy_new_tail);
 9492     } else {
 9493       andl(tmp2, 0x00000007);   // tail count (in chars)
 9494       andl(len, 0xfffffff8);    // vector count (in chars)
 9495       jccb(Assembler::zero, copy_tail);
 9496     }
 9497 
 9498     // vectored inflation
 9499     lea(src, Address(src, len, Address::times_1));
 9500     lea(dst, Address(dst, len, Address::times_2));
 9501     negptr(len);
 9502 
 9503     if (UseAVX > 1) {
 9504       bind(copy_16_loop);
 9505       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 9506       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 9507       addptr(len, 16);
 9508       jcc(Assembler::notZero, copy_16_loop);
 9509 
 9510       bind(below_threshold);
 9511       bind(copy_new_tail);
 9512       movl(len, tmp2);
 9513       andl(tmp2, 0x00000007);
 9514       andl(len, 0xFFFFFFF8);
 9515       jccb(Assembler::zero, copy_tail);
 9516 
 9517       pmovzxbw(tmp1, Address(src, 0));
 9518       movdqu(Address(dst, 0), tmp1);
 9519       addptr(src, 8);
 9520       addptr(dst, 2 * 8);
 9521 
 9522       jmp(copy_tail, true);
 9523     }
 9524 
 9525     // inflate 8 chars per iter
 9526     bind(copy_8_loop);
 9527     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 9528     movdqu(Address(dst, len, Address::times_2), tmp1);
 9529     addptr(len, 8);
 9530     jcc(Assembler::notZero, copy_8_loop);
 9531 
 9532     bind(copy_tail);
 9533     movl(len, tmp2);
 9534 
 9535     cmpl(len, 4);
 9536     jccb(Assembler::less, copy_bytes);
 9537 
 9538     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 9539     pmovzxbw(tmp1, tmp1);
 9540     movq(Address(dst, 0), tmp1);
 9541     subptr(len, 4);
 9542     addptr(src, 4);
 9543     addptr(dst, 8);
 9544 
 9545     bind(copy_bytes);
 9546   } else {
 9547     bind(below_threshold);
 9548   }
 9549 
 9550   testl(len, len);
 9551   jccb(Assembler::zero, done);
 9552   lea(src, Address(src, len, Address::times_1));
 9553   lea(dst, Address(dst, len, Address::times_2));
 9554   negptr(len);
 9555 
 9556   // inflate 1 char per iter
 9557   bind(copy_chars_loop);
 9558   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 9559   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 9560   increment(len);
 9561   jcc(Assembler::notZero, copy_chars_loop);
 9562 
 9563   bind(done);
 9564 }
 9565 
 9566 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 9567   switch(type) {
 9568     case T_BYTE:
 9569     case T_BOOLEAN:
 9570       evmovdqub(dst, kmask, src, merge, vector_len);
 9571       break;
 9572     case T_CHAR:
 9573     case T_SHORT:
 9574       evmovdquw(dst, kmask, src, merge, vector_len);
 9575       break;
 9576     case T_INT:
 9577     case T_FLOAT:
 9578       evmovdqul(dst, kmask, src, merge, vector_len);
 9579       break;
 9580     case T_LONG:
 9581     case T_DOUBLE:
 9582       evmovdquq(dst, kmask, src, merge, vector_len);
 9583       break;
 9584     default:
 9585       fatal("Unexpected type argument %s", type2name(type));
 9586       break;
 9587   }
 9588 }
 9589 
 9590 
 9591 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 9592   switch(type) {
 9593     case T_BYTE:
 9594     case T_BOOLEAN:
 9595       evmovdqub(dst, kmask, src, merge, vector_len);
 9596       break;
 9597     case T_CHAR:
 9598     case T_SHORT:
 9599       evmovdquw(dst, kmask, src, merge, vector_len);
 9600       break;
 9601     case T_INT:
 9602     case T_FLOAT:
 9603       evmovdqul(dst, kmask, src, merge, vector_len);
 9604       break;
 9605     case T_LONG:
 9606     case T_DOUBLE:
 9607       evmovdquq(dst, kmask, src, merge, vector_len);
 9608       break;
 9609     default:
 9610       fatal("Unexpected type argument %s", type2name(type));
 9611       break;
 9612   }
 9613 }
 9614 
 9615 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 9616   switch(type) {
 9617     case T_BYTE:
 9618     case T_BOOLEAN:
 9619       evmovdqub(dst, kmask, src, merge, vector_len);
 9620       break;
 9621     case T_CHAR:
 9622     case T_SHORT:
 9623       evmovdquw(dst, kmask, src, merge, vector_len);
 9624       break;
 9625     case T_INT:
 9626     case T_FLOAT:
 9627       evmovdqul(dst, kmask, src, merge, vector_len);
 9628       break;
 9629     case T_LONG:
 9630     case T_DOUBLE:
 9631       evmovdquq(dst, kmask, src, merge, vector_len);
 9632       break;
 9633     default:
 9634       fatal("Unexpected type argument %s", type2name(type));
 9635       break;
 9636   }
 9637 }
 9638 
 9639 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 9640   switch(masklen) {
 9641     case 2:
 9642        knotbl(dst, src);
 9643        movl(rtmp, 3);
 9644        kmovbl(ktmp, rtmp);
 9645        kandbl(dst, ktmp, dst);
 9646        break;
 9647     case 4:
 9648        knotbl(dst, src);
 9649        movl(rtmp, 15);
 9650        kmovbl(ktmp, rtmp);
 9651        kandbl(dst, ktmp, dst);
 9652        break;
 9653     case 8:
 9654        knotbl(dst, src);
 9655        break;
 9656     case 16:
 9657        knotwl(dst, src);
 9658        break;
 9659     case 32:
 9660        knotdl(dst, src);
 9661        break;
 9662     case 64:
 9663        knotql(dst, src);
 9664        break;
 9665     default:
 9666       fatal("Unexpected vector length %d", masklen);
 9667       break;
 9668   }
 9669 }
 9670 
 9671 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9672   switch(type) {
 9673     case T_BOOLEAN:
 9674     case T_BYTE:
 9675        kandbl(dst, src1, src2);
 9676        break;
 9677     case T_CHAR:
 9678     case T_SHORT:
 9679        kandwl(dst, src1, src2);
 9680        break;
 9681     case T_INT:
 9682     case T_FLOAT:
 9683        kanddl(dst, src1, src2);
 9684        break;
 9685     case T_LONG:
 9686     case T_DOUBLE:
 9687        kandql(dst, src1, src2);
 9688        break;
 9689     default:
 9690       fatal("Unexpected type argument %s", type2name(type));
 9691       break;
 9692   }
 9693 }
 9694 
 9695 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9696   switch(type) {
 9697     case T_BOOLEAN:
 9698     case T_BYTE:
 9699        korbl(dst, src1, src2);
 9700        break;
 9701     case T_CHAR:
 9702     case T_SHORT:
 9703        korwl(dst, src1, src2);
 9704        break;
 9705     case T_INT:
 9706     case T_FLOAT:
 9707        kordl(dst, src1, src2);
 9708        break;
 9709     case T_LONG:
 9710     case T_DOUBLE:
 9711        korql(dst, src1, src2);
 9712        break;
 9713     default:
 9714       fatal("Unexpected type argument %s", type2name(type));
 9715       break;
 9716   }
 9717 }
 9718 
 9719 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9720   switch(type) {
 9721     case T_BOOLEAN:
 9722     case T_BYTE:
 9723        kxorbl(dst, src1, src2);
 9724        break;
 9725     case T_CHAR:
 9726     case T_SHORT:
 9727        kxorwl(dst, src1, src2);
 9728        break;
 9729     case T_INT:
 9730     case T_FLOAT:
 9731        kxordl(dst, src1, src2);
 9732        break;
 9733     case T_LONG:
 9734     case T_DOUBLE:
 9735        kxorql(dst, src1, src2);
 9736        break;
 9737     default:
 9738       fatal("Unexpected type argument %s", type2name(type));
 9739       break;
 9740   }
 9741 }
 9742 
 9743 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9744   switch(type) {
 9745     case T_BOOLEAN:
 9746     case T_BYTE:
 9747       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9748     case T_CHAR:
 9749     case T_SHORT:
 9750       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9751     case T_INT:
 9752     case T_FLOAT:
 9753       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9754     case T_LONG:
 9755     case T_DOUBLE:
 9756       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9757     default:
 9758       fatal("Unexpected type argument %s", type2name(type)); break;
 9759   }
 9760 }
 9761 
 9762 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9763   switch(type) {
 9764     case T_BOOLEAN:
 9765     case T_BYTE:
 9766       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9767     case T_CHAR:
 9768     case T_SHORT:
 9769       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9770     case T_INT:
 9771     case T_FLOAT:
 9772       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9773     case T_LONG:
 9774     case T_DOUBLE:
 9775       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9776     default:
 9777       fatal("Unexpected type argument %s", type2name(type)); break;
 9778   }
 9779 }
 9780 
 9781 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9782   switch(type) {
 9783     case T_BYTE:
 9784       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9785     case T_SHORT:
 9786       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9787     case T_INT:
 9788       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9789     case T_LONG:
 9790       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9791     default:
 9792       fatal("Unexpected type argument %s", type2name(type)); break;
 9793   }
 9794 }
 9795 
 9796 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9797   switch(type) {
 9798     case T_BYTE:
 9799       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9800     case T_SHORT:
 9801       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9802     case T_INT:
 9803       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9804     case T_LONG:
 9805       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9806     default:
 9807       fatal("Unexpected type argument %s", type2name(type)); break;
 9808   }
 9809 }
 9810 
 9811 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9812   switch(type) {
 9813     case T_BYTE:
 9814       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9815     case T_SHORT:
 9816       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9817     case T_INT:
 9818       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9819     case T_LONG:
 9820       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9821     default:
 9822       fatal("Unexpected type argument %s", type2name(type)); break;
 9823   }
 9824 }
 9825 
 9826 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9827   switch(type) {
 9828     case T_BYTE:
 9829       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9830     case T_SHORT:
 9831       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9832     case T_INT:
 9833       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9834     case T_LONG:
 9835       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9836     default:
 9837       fatal("Unexpected type argument %s", type2name(type)); break;
 9838   }
 9839 }
 9840 
 9841 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9842   switch(type) {
 9843     case T_BYTE:
 9844       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9845     case T_SHORT:
 9846       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9847     case T_INT:
 9848       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9849     case T_LONG:
 9850       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9851     case T_FLOAT:
 9852       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9853     case T_DOUBLE:
 9854       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9855     default:
 9856       fatal("Unexpected type argument %s", type2name(type)); break;
 9857   }
 9858 }
 9859 
 9860 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9861   switch(type) {
 9862     case T_BYTE:
 9863       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9864     case T_SHORT:
 9865       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9866     case T_INT:
 9867       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9868     case T_LONG:
 9869       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9870     case T_FLOAT:
 9871       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9872     case T_DOUBLE:
 9873       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9874     default:
 9875       fatal("Unexpected type argument %s", type2name(type)); break;
 9876   }
 9877 }
 9878 
 9879 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9880   switch(type) {
 9881     case T_BYTE:
 9882       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9883     case T_SHORT:
 9884       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9885     case T_INT:
 9886       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9887     case T_LONG:
 9888       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9889     case T_FLOAT:
 9890       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9891     case T_DOUBLE:
 9892       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9893     default:
 9894       fatal("Unexpected type argument %s", type2name(type)); break;
 9895   }
 9896 }
 9897 
 9898 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9899   switch(type) {
 9900     case T_BYTE:
 9901       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9902     case T_SHORT:
 9903       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9904     case T_INT:
 9905       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9906     case T_LONG:
 9907       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9908     case T_FLOAT:
 9909       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9910     case T_DOUBLE:
 9911       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9912     default:
 9913       fatal("Unexpected type argument %s", type2name(type)); break;
 9914   }
 9915 }
 9916 
 9917 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9918   switch(type) {
 9919     case T_INT:
 9920       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9921     case T_LONG:
 9922       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9923     default:
 9924       fatal("Unexpected type argument %s", type2name(type)); break;
 9925   }
 9926 }
 9927 
 9928 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9929   switch(type) {
 9930     case T_INT:
 9931       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9932     case T_LONG:
 9933       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9934     default:
 9935       fatal("Unexpected type argument %s", type2name(type)); break;
 9936   }
 9937 }
 9938 
 9939 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9940   switch(type) {
 9941     case T_INT:
 9942       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9943     case T_LONG:
 9944       evporq(dst, mask, nds, src, merge, vector_len); break;
 9945     default:
 9946       fatal("Unexpected type argument %s", type2name(type)); break;
 9947   }
 9948 }
 9949 
 9950 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9951   switch(type) {
 9952     case T_INT:
 9953       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9954     case T_LONG:
 9955       evporq(dst, mask, nds, src, merge, vector_len); break;
 9956     default:
 9957       fatal("Unexpected type argument %s", type2name(type)); break;
 9958   }
 9959 }
 9960 
 9961 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9962   switch(type) {
 9963     case T_INT:
 9964       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9965     case T_LONG:
 9966       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9967     default:
 9968       fatal("Unexpected type argument %s", type2name(type)); break;
 9969   }
 9970 }
 9971 
 9972 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9973   switch(type) {
 9974     case T_INT:
 9975       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9976     case T_LONG:
 9977       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9978     default:
 9979       fatal("Unexpected type argument %s", type2name(type)); break;
 9980   }
 9981 }
 9982 
 9983 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9984   switch(masklen) {
 9985     case 8:
 9986        kortestbl(src1, src2);
 9987        break;
 9988     case 16:
 9989        kortestwl(src1, src2);
 9990        break;
 9991     case 32:
 9992        kortestdl(src1, src2);
 9993        break;
 9994     case 64:
 9995        kortestql(src1, src2);
 9996        break;
 9997     default:
 9998       fatal("Unexpected mask length %d", masklen);
 9999       break;
10000   }
10001 }
10002 
10003 
10004 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
10005   switch(masklen)  {
10006     case 8:
10007        ktestbl(src1, src2);
10008        break;
10009     case 16:
10010        ktestwl(src1, src2);
10011        break;
10012     case 32:
10013        ktestdl(src1, src2);
10014        break;
10015     case 64:
10016        ktestql(src1, src2);
10017        break;
10018     default:
10019       fatal("Unexpected mask length %d", masklen);
10020       break;
10021   }
10022 }
10023 
10024 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
10025   switch(type) {
10026     case T_INT:
10027       evprold(dst, mask, src, shift, merge, vlen_enc); break;
10028     case T_LONG:
10029       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
10030     default:
10031       fatal("Unexpected type argument %s", type2name(type)); break;
10032       break;
10033   }
10034 }
10035 
10036 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
10037   switch(type) {
10038     case T_INT:
10039       evprord(dst, mask, src, shift, merge, vlen_enc); break;
10040     case T_LONG:
10041       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
10042     default:
10043       fatal("Unexpected type argument %s", type2name(type)); break;
10044   }
10045 }
10046 
10047 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
10048   switch(type) {
10049     case T_INT:
10050       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
10051     case T_LONG:
10052       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
10053     default:
10054       fatal("Unexpected type argument %s", type2name(type)); break;
10055   }
10056 }
10057 
10058 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
10059   switch(type) {
10060     case T_INT:
10061       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
10062     case T_LONG:
10063       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
10064     default:
10065       fatal("Unexpected type argument %s", type2name(type)); break;
10066   }
10067 }
10068 
10069 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10070   assert(rscratch != noreg || always_reachable(src), "missing");
10071 
10072   if (reachable(src)) {
10073     evpandq(dst, nds, as_Address(src), vector_len);
10074   } else {
10075     lea(rscratch, src);
10076     evpandq(dst, nds, Address(rscratch, 0), vector_len);
10077   }
10078 }
10079 
10080 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
10081   assert(rscratch != noreg || always_reachable(src), "missing");
10082 
10083   if (reachable(src)) {
10084     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
10085   } else {
10086     lea(rscratch, src);
10087     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
10088   }
10089 }
10090 
10091 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10092   assert(rscratch != noreg || always_reachable(src), "missing");
10093 
10094   if (reachable(src)) {
10095     evporq(dst, nds, as_Address(src), vector_len);
10096   } else {
10097     lea(rscratch, src);
10098     evporq(dst, nds, Address(rscratch, 0), vector_len);
10099   }
10100 }
10101 
10102 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10103   assert(rscratch != noreg || always_reachable(src), "missing");
10104 
10105   if (reachable(src)) {
10106     vpshufb(dst, nds, as_Address(src), vector_len);
10107   } else {
10108     lea(rscratch, src);
10109     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
10110   }
10111 }
10112 
10113 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10114   assert(rscratch != noreg || always_reachable(src), "missing");
10115 
10116   if (reachable(src)) {
10117     Assembler::vpor(dst, nds, as_Address(src), vector_len);
10118   } else {
10119     lea(rscratch, src);
10120     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
10121   }
10122 }
10123 
10124 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
10125   assert(rscratch != noreg || always_reachable(src3), "missing");
10126 
10127   if (reachable(src3)) {
10128     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
10129   } else {
10130     lea(rscratch, src3);
10131     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
10132   }
10133 }
10134 
10135 #if COMPILER2_OR_JVMCI
10136 
10137 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
10138                                  Register length, Register temp, int vec_enc) {
10139   // Computing mask for predicated vector store.
10140   movptr(temp, -1);
10141   bzhiq(temp, temp, length);
10142   kmov(mask, temp);
10143   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
10144 }
10145 
10146 // Set memory operation for length "less than" 64 bytes.
10147 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
10148                                        XMMRegister xmm, KRegister mask, Register length,
10149                                        Register temp, bool use64byteVector) {
10150   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10151   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10152   if (!use64byteVector) {
10153     fill32(dst, disp, xmm);
10154     subptr(length, 32 >> shift);
10155     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
10156   } else {
10157     assert(MaxVectorSize == 64, "vector length != 64");
10158     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
10159   }
10160 }
10161 
10162 
10163 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
10164                                        XMMRegister xmm, KRegister mask, Register length,
10165                                        Register temp) {
10166   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10167   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10168   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
10169 }
10170 
10171 
10172 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
10173   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10174   vmovdqu(dst, xmm);
10175 }
10176 
10177 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
10178   fill32(Address(dst, disp), xmm);
10179 }
10180 
10181 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
10182   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10183   if (!use64byteVector) {
10184     fill32(dst, xmm);
10185     fill32(dst.plus_disp(32), xmm);
10186   } else {
10187     evmovdquq(dst, xmm, Assembler::AVX_512bit);
10188   }
10189 }
10190 
10191 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
10192   fill64(Address(dst, disp), xmm, use64byteVector);
10193 }
10194 
10195 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
10196                                         Register count, Register rtmp, XMMRegister xtmp) {
10197   Label L_exit;
10198   Label L_fill_start;
10199   Label L_fill_64_bytes;
10200   Label L_fill_96_bytes;
10201   Label L_fill_128_bytes;
10202   Label L_fill_128_bytes_loop;
10203   Label L_fill_128_loop_header;
10204   Label L_fill_128_bytes_loop_header;
10205   Label L_fill_128_bytes_loop_pre_header;
10206   Label L_fill_zmm_sequence;
10207 
10208   int shift = -1;
10209   switch(type) {
10210     case T_BYTE:  shift = 0;
10211       break;
10212     case T_SHORT: shift = 1;
10213       break;
10214     case T_INT:   shift = 2;
10215       break;
10216     /* Uncomment when LONG fill stubs are supported.
10217     case T_LONG:  shift = 3;
10218       break;
10219     */
10220     default:
10221       fatal("Unhandled type: %s\n", type2name(type));
10222   }
10223 
10224   if ((CopyAVX3Threshold != 0)  || (MaxVectorSize == 32)) {
10225 
10226     if (MaxVectorSize == 64) {
10227       cmpq(count, CopyAVX3Threshold >> shift);
10228       jcc(Assembler::greater, L_fill_zmm_sequence);
10229     }
10230 
10231     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
10232 
10233     bind(L_fill_start);
10234 
10235     cmpq(count, 32 >> shift);
10236     jccb(Assembler::greater, L_fill_64_bytes);
10237     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
10238     jmp(L_exit);
10239 
10240     bind(L_fill_64_bytes);
10241     cmpq(count, 64 >> shift);
10242     jccb(Assembler::greater, L_fill_96_bytes);
10243     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
10244     jmp(L_exit);
10245 
10246     bind(L_fill_96_bytes);
10247     cmpq(count, 96 >> shift);
10248     jccb(Assembler::greater, L_fill_128_bytes);
10249     fill64(to, 0, xtmp);
10250     subq(count, 64 >> shift);
10251     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
10252     jmp(L_exit);
10253 
10254     bind(L_fill_128_bytes);
10255     cmpq(count, 128 >> shift);
10256     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
10257     fill64(to, 0, xtmp);
10258     fill32(to, 64, xtmp);
10259     subq(count, 96 >> shift);
10260     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
10261     jmp(L_exit);
10262 
10263     bind(L_fill_128_bytes_loop_pre_header);
10264     {
10265       mov(rtmp, to);
10266       andq(rtmp, 31);
10267       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
10268       negq(rtmp);
10269       addq(rtmp, 32);
10270       mov64(r8, -1L);
10271       bzhiq(r8, r8, rtmp);
10272       kmovql(k2, r8);
10273       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
10274       addq(to, rtmp);
10275       shrq(rtmp, shift);
10276       subq(count, rtmp);
10277     }
10278 
10279     cmpq(count, 128 >> shift);
10280     jcc(Assembler::less, L_fill_start);
10281 
10282     bind(L_fill_128_bytes_loop_header);
10283     subq(count, 128 >> shift);
10284 
10285     align32();
10286     bind(L_fill_128_bytes_loop);
10287       fill64(to, 0, xtmp);
10288       fill64(to, 64, xtmp);
10289       addq(to, 128);
10290       subq(count, 128 >> shift);
10291       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
10292 
10293     addq(count, 128 >> shift);
10294     jcc(Assembler::zero, L_exit);
10295     jmp(L_fill_start);
10296   }
10297 
10298   if (MaxVectorSize == 64) {
10299     // Sequence using 64 byte ZMM register.
10300     Label L_fill_128_bytes_zmm;
10301     Label L_fill_192_bytes_zmm;
10302     Label L_fill_192_bytes_loop_zmm;
10303     Label L_fill_192_bytes_loop_header_zmm;
10304     Label L_fill_192_bytes_loop_pre_header_zmm;
10305     Label L_fill_start_zmm_sequence;
10306 
10307     bind(L_fill_zmm_sequence);
10308     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
10309 
10310     bind(L_fill_start_zmm_sequence);
10311     cmpq(count, 64 >> shift);
10312     jccb(Assembler::greater, L_fill_128_bytes_zmm);
10313     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
10314     jmp(L_exit);
10315 
10316     bind(L_fill_128_bytes_zmm);
10317     cmpq(count, 128 >> shift);
10318     jccb(Assembler::greater, L_fill_192_bytes_zmm);
10319     fill64(to, 0, xtmp, true);
10320     subq(count, 64 >> shift);
10321     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
10322     jmp(L_exit);
10323 
10324     bind(L_fill_192_bytes_zmm);
10325     cmpq(count, 192 >> shift);
10326     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
10327     fill64(to, 0, xtmp, true);
10328     fill64(to, 64, xtmp, true);
10329     subq(count, 128 >> shift);
10330     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
10331     jmp(L_exit);
10332 
10333     bind(L_fill_192_bytes_loop_pre_header_zmm);
10334     {
10335       movq(rtmp, to);
10336       andq(rtmp, 63);
10337       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
10338       negq(rtmp);
10339       addq(rtmp, 64);
10340       mov64(r8, -1L);
10341       bzhiq(r8, r8, rtmp);
10342       kmovql(k2, r8);
10343       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
10344       addq(to, rtmp);
10345       shrq(rtmp, shift);
10346       subq(count, rtmp);
10347     }
10348 
10349     cmpq(count, 192 >> shift);
10350     jcc(Assembler::less, L_fill_start_zmm_sequence);
10351 
10352     bind(L_fill_192_bytes_loop_header_zmm);
10353     subq(count, 192 >> shift);
10354 
10355     align32();
10356     bind(L_fill_192_bytes_loop_zmm);
10357       fill64(to, 0, xtmp, true);
10358       fill64(to, 64, xtmp, true);
10359       fill64(to, 128, xtmp, true);
10360       addq(to, 192);
10361       subq(count, 192 >> shift);
10362       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
10363 
10364     addq(count, 192 >> shift);
10365     jcc(Assembler::zero, L_exit);
10366     jmp(L_fill_start_zmm_sequence);
10367   }
10368   bind(L_exit);
10369 }
10370 #endif //COMPILER2_OR_JVMCI
10371 
10372 
10373 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
10374   Label done;
10375   cvttss2sil(dst, src);
10376   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10377   cmpl(dst, 0x80000000); // float_sign_flip
10378   jccb(Assembler::notEqual, done);
10379   subptr(rsp, 8);
10380   movflt(Address(rsp, 0), src);
10381   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
10382   pop(dst);
10383   bind(done);
10384 }
10385 
10386 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
10387   Label done;
10388   cvttsd2sil(dst, src);
10389   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10390   cmpl(dst, 0x80000000); // float_sign_flip
10391   jccb(Assembler::notEqual, done);
10392   subptr(rsp, 8);
10393   movdbl(Address(rsp, 0), src);
10394   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
10395   pop(dst);
10396   bind(done);
10397 }
10398 
10399 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
10400   Label done;
10401   cvttss2siq(dst, src);
10402   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10403   jccb(Assembler::notEqual, done);
10404   subptr(rsp, 8);
10405   movflt(Address(rsp, 0), src);
10406   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
10407   pop(dst);
10408   bind(done);
10409 }
10410 
10411 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10412   // Following code is line by line assembly translation rounding algorithm.
10413   // Please refer to java.lang.Math.round(float) algorithm for details.
10414   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
10415   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
10416   const int32_t FloatConsts_EXP_BIAS = 127;
10417   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
10418   const int32_t MINUS_32 = 0xFFFFFFE0;
10419   Label L_special_case, L_block1, L_exit;
10420   movl(rtmp, FloatConsts_EXP_BIT_MASK);
10421   movdl(dst, src);
10422   andl(dst, rtmp);
10423   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
10424   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
10425   subl(rtmp, dst);
10426   movl(rcx, rtmp);
10427   movl(dst, MINUS_32);
10428   testl(rtmp, dst);
10429   jccb(Assembler::notEqual, L_special_case);
10430   movdl(dst, src);
10431   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
10432   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
10433   movdl(rtmp, src);
10434   testl(rtmp, rtmp);
10435   jccb(Assembler::greaterEqual, L_block1);
10436   negl(dst);
10437   bind(L_block1);
10438   sarl(dst);
10439   addl(dst, 0x1);
10440   sarl(dst, 0x1);
10441   jmp(L_exit);
10442   bind(L_special_case);
10443   convert_f2i(dst, src);
10444   bind(L_exit);
10445 }
10446 
10447 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10448   // Following code is line by line assembly translation rounding algorithm.
10449   // Please refer to java.lang.Math.round(double) algorithm for details.
10450   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
10451   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
10452   const int64_t DoubleConsts_EXP_BIAS = 1023;
10453   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
10454   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
10455   Label L_special_case, L_block1, L_exit;
10456   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
10457   movq(dst, src);
10458   andq(dst, rtmp);
10459   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
10460   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
10461   subq(rtmp, dst);
10462   movq(rcx, rtmp);
10463   mov64(dst, MINUS_64);
10464   testq(rtmp, dst);
10465   jccb(Assembler::notEqual, L_special_case);
10466   movq(dst, src);
10467   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
10468   andq(dst, rtmp);
10469   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
10470   orq(dst, rtmp);
10471   movq(rtmp, src);
10472   testq(rtmp, rtmp);
10473   jccb(Assembler::greaterEqual, L_block1);
10474   negq(dst);
10475   bind(L_block1);
10476   sarq(dst);
10477   addq(dst, 0x1);
10478   sarq(dst, 0x1);
10479   jmp(L_exit);
10480   bind(L_special_case);
10481   convert_d2l(dst, src);
10482   bind(L_exit);
10483 }
10484 
10485 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
10486   Label done;
10487   cvttsd2siq(dst, src);
10488   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10489   jccb(Assembler::notEqual, done);
10490   subptr(rsp, 8);
10491   movdbl(Address(rsp, 0), src);
10492   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
10493   pop(dst);
10494   bind(done);
10495 }
10496 
10497 void MacroAssembler::cache_wb(Address line)
10498 {
10499   // 64 bit cpus always support clflush
10500   assert(VM_Version::supports_clflush(), "clflush should be available");
10501   bool optimized = VM_Version::supports_clflushopt();
10502   bool no_evict = VM_Version::supports_clwb();
10503 
10504   // prefer clwb (writeback without evict) otherwise
10505   // prefer clflushopt (potentially parallel writeback with evict)
10506   // otherwise fallback on clflush (serial writeback with evict)
10507 
10508   if (optimized) {
10509     if (no_evict) {
10510       clwb(line);
10511     } else {
10512       clflushopt(line);
10513     }
10514   } else {
10515     // no need for fence when using CLFLUSH
10516     clflush(line);
10517   }
10518 }
10519 
10520 void MacroAssembler::cache_wbsync(bool is_pre)
10521 {
10522   assert(VM_Version::supports_clflush(), "clflush should be available");
10523   bool optimized = VM_Version::supports_clflushopt();
10524   bool no_evict = VM_Version::supports_clwb();
10525 
10526   // pick the correct implementation
10527 
10528   if (!is_pre && (optimized || no_evict)) {
10529     // need an sfence for post flush when using clflushopt or clwb
10530     // otherwise no no need for any synchroniaztion
10531 
10532     sfence();
10533   }
10534 }
10535 
10536 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
10537   switch (cond) {
10538     // Note some conditions are synonyms for others
10539     case Assembler::zero:         return Assembler::notZero;
10540     case Assembler::notZero:      return Assembler::zero;
10541     case Assembler::less:         return Assembler::greaterEqual;
10542     case Assembler::lessEqual:    return Assembler::greater;
10543     case Assembler::greater:      return Assembler::lessEqual;
10544     case Assembler::greaterEqual: return Assembler::less;
10545     case Assembler::below:        return Assembler::aboveEqual;
10546     case Assembler::belowEqual:   return Assembler::above;
10547     case Assembler::above:        return Assembler::belowEqual;
10548     case Assembler::aboveEqual:   return Assembler::below;
10549     case Assembler::overflow:     return Assembler::noOverflow;
10550     case Assembler::noOverflow:   return Assembler::overflow;
10551     case Assembler::negative:     return Assembler::positive;
10552     case Assembler::positive:     return Assembler::negative;
10553     case Assembler::parity:       return Assembler::noParity;
10554     case Assembler::noParity:     return Assembler::parity;
10555   }
10556   ShouldNotReachHere(); return Assembler::overflow;
10557 }
10558 
10559 // This is simply a call to Thread::current()
10560 void MacroAssembler::get_thread_slow(Register thread) {
10561   if (thread != rax) {
10562     push(rax);
10563   }
10564   push(rdi);
10565   push(rsi);
10566   push(rdx);
10567   push(rcx);
10568   push(r8);
10569   push(r9);
10570   push(r10);
10571   push(r11);
10572 
10573   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
10574 
10575   pop(r11);
10576   pop(r10);
10577   pop(r9);
10578   pop(r8);
10579   pop(rcx);
10580   pop(rdx);
10581   pop(rsi);
10582   pop(rdi);
10583   if (thread != rax) {
10584     mov(thread, rax);
10585     pop(rax);
10586   }
10587 }
10588 
10589 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
10590   Label L_stack_ok;
10591   if (bias == 0) {
10592     testptr(sp, 2 * wordSize - 1);
10593   } else {
10594     // lea(tmp, Address(rsp, bias);
10595     mov(tmp, sp);
10596     addptr(tmp, bias);
10597     testptr(tmp, 2 * wordSize - 1);
10598   }
10599   jcc(Assembler::equal, L_stack_ok);
10600   block_comment(msg);
10601   stop(msg);
10602   bind(L_stack_ok);
10603 }
10604 
10605 // Implements fast-locking.
10606 //
10607 // obj: the object to be locked
10608 // reg_rax: rax
10609 // thread: the thread which attempts to lock obj
10610 // tmp: a temporary register
10611 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
10612   Register thread = r15_thread;
10613 
10614   assert(reg_rax == rax, "");
10615   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
10616 
10617   Label push;
10618   const Register top = tmp;
10619 
10620   // Preload the markWord. It is important that this is the first
10621   // instruction emitted as it is part of C1's null check semantics.
10622   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10623 
10624   if (UseObjectMonitorTable) {
10625     // Clear cache in case fast locking succeeds or we need to take the slow-path.
10626     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
10627   }
10628 
10629   if (DiagnoseSyncOnValueBasedClasses != 0) {
10630     load_klass(tmp, obj, rscratch1);
10631     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
10632     jcc(Assembler::notZero, slow);
10633   }
10634 
10635   // Load top.
10636   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10637 
10638   // Check if the lock-stack is full.
10639   cmpl(top, LockStack::end_offset());
10640   jcc(Assembler::greaterEqual, slow);
10641 
10642   // Check for recursion.
10643   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10644   jcc(Assembler::equal, push);
10645 
10646   // Check header for monitor (0b10).
10647   testptr(reg_rax, markWord::monitor_value);
10648   jcc(Assembler::notZero, slow);
10649 
10650   // Try to lock. Transition lock bits 0b01 => 0b00
10651   movptr(tmp, reg_rax);
10652   andptr(tmp, ~(int32_t)markWord::unlocked_value);
10653   orptr(reg_rax, markWord::unlocked_value);
10654   // Mask inline_type bit such that we go to the slow path if object is an inline type
10655   andptr(reg_rax, ~((int) markWord::inline_type_bit_in_place));
10656 
10657   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10658   jcc(Assembler::notEqual, slow);
10659 
10660   // Restore top, CAS clobbers register.
10661   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10662 
10663   bind(push);
10664   // After successful lock, push object on lock-stack.
10665   movptr(Address(thread, top), obj);
10666   incrementl(top, oopSize);
10667   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
10668 }
10669 
10670 // Implements fast-unlocking.
10671 //
10672 // obj: the object to be unlocked
10673 // reg_rax: rax
10674 // thread: the thread
10675 // tmp: a temporary register
10676 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
10677   Register thread = r15_thread;
10678 
10679   assert(reg_rax == rax, "");
10680   assert_different_registers(obj, reg_rax, thread, tmp);
10681 
10682   Label unlocked, push_and_slow;
10683   const Register top = tmp;
10684 
10685   // Check if obj is top of lock-stack.
10686   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10687   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10688   jcc(Assembler::notEqual, slow);
10689 
10690   // Pop lock-stack.
10691   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
10692   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10693 
10694   // Check if recursive.
10695   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
10696   jcc(Assembler::equal, unlocked);
10697 
10698   // Not recursive. Check header for monitor (0b10).
10699   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10700   testptr(reg_rax, markWord::monitor_value);
10701   jcc(Assembler::notZero, push_and_slow);
10702 
10703 #ifdef ASSERT
10704   // Check header not unlocked (0b01).
10705   Label not_unlocked;
10706   testptr(reg_rax, markWord::unlocked_value);
10707   jcc(Assembler::zero, not_unlocked);
10708   stop("fast_unlock already unlocked");
10709   bind(not_unlocked);
10710 #endif
10711 
10712   // Try to unlock. Transition lock bits 0b00 => 0b01
10713   movptr(tmp, reg_rax);
10714   orptr(tmp, markWord::unlocked_value);
10715   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10716   jcc(Assembler::equal, unlocked);
10717 
10718   bind(push_and_slow);
10719   // Restore lock-stack and handle the unlock in runtime.
10720 #ifdef ASSERT
10721   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10722   movptr(Address(thread, top), obj);
10723 #endif
10724   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10725   jmp(slow);
10726 
10727   bind(unlocked);
10728 }
10729 
10730 // Saves legacy GPRs state on stack.
10731 void MacroAssembler::save_legacy_gprs() {
10732   subq(rsp, 16 * wordSize);
10733   movq(Address(rsp, 15 * wordSize), rax);
10734   movq(Address(rsp, 14 * wordSize), rcx);
10735   movq(Address(rsp, 13 * wordSize), rdx);
10736   movq(Address(rsp, 12 * wordSize), rbx);
10737   movq(Address(rsp, 10 * wordSize), rbp);
10738   movq(Address(rsp, 9 * wordSize), rsi);
10739   movq(Address(rsp, 8 * wordSize), rdi);
10740   movq(Address(rsp, 7 * wordSize), r8);
10741   movq(Address(rsp, 6 * wordSize), r9);
10742   movq(Address(rsp, 5 * wordSize), r10);
10743   movq(Address(rsp, 4 * wordSize), r11);
10744   movq(Address(rsp, 3 * wordSize), r12);
10745   movq(Address(rsp, 2 * wordSize), r13);
10746   movq(Address(rsp, wordSize), r14);
10747   movq(Address(rsp, 0), r15);
10748 }
10749 
10750 // Resotres back legacy GPRs state from stack.
10751 void MacroAssembler::restore_legacy_gprs() {
10752   movq(r15, Address(rsp, 0));
10753   movq(r14, Address(rsp, wordSize));
10754   movq(r13, Address(rsp, 2 * wordSize));
10755   movq(r12, Address(rsp, 3 * wordSize));
10756   movq(r11, Address(rsp, 4 * wordSize));
10757   movq(r10, Address(rsp, 5 * wordSize));
10758   movq(r9,  Address(rsp, 6 * wordSize));
10759   movq(r8,  Address(rsp, 7 * wordSize));
10760   movq(rdi, Address(rsp, 8 * wordSize));
10761   movq(rsi, Address(rsp, 9 * wordSize));
10762   movq(rbp, Address(rsp, 10 * wordSize));
10763   movq(rbx, Address(rsp, 12 * wordSize));
10764   movq(rdx, Address(rsp, 13 * wordSize));
10765   movq(rcx, Address(rsp, 14 * wordSize));
10766   movq(rax, Address(rsp, 15 * wordSize));
10767   addq(rsp, 16 * wordSize);
10768 }
10769 
10770 void MacroAssembler::load_aotrc_address(Register reg, address a) {
10771 #if INCLUDE_CDS
10772   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
10773   if (AOTCodeCache::is_on_for_dump()) {
10774     // all aotrc field addresses should be registered in the AOTCodeCache address table
10775     lea(reg, ExternalAddress(a));
10776   } else {
10777     mov64(reg, (uint64_t)a);
10778   }
10779 #else
10780   ShouldNotReachHere();
10781 #endif
10782 }
10783 
10784 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10785   if (VM_Version::supports_apx_f()) {
10786     esetzucc(comparison, dst);
10787   } else {
10788     setb(comparison, dst);
10789     movzbl(dst, dst);
10790   }
10791 }