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   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
 2450 }
 2451 
 2452 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
 2453                                              Label& is_non_flat_array) {
 2454   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
 2455 }
 2456 
 2457 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label&is_null_free_array) {
 2458   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
 2459 }
 2460 
 2461 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
 2462   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
 2463 }
 2464 
 2465 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
 2466   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2467   jcc(Assembler::notZero, is_flat_array);
 2468 }
 2469 
 2470 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
 2471   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2472   jcc(Assembler::zero, is_non_flat_array);
 2473 }
 2474 
 2475 void MacroAssembler::os_breakpoint() {
 2476   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2477   // (e.g., MSVC can't call ps() otherwise)
 2478   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2479 }
 2480 
 2481 void MacroAssembler::unimplemented(const char* what) {
 2482   const char* buf = nullptr;
 2483   {
 2484     ResourceMark rm;
 2485     stringStream ss;
 2486     ss.print("unimplemented: %s", what);
 2487     buf = code_string(ss.as_string());
 2488   }
 2489   stop(buf);
 2490 }
 2491 
 2492 #define XSTATE_BV 0x200
 2493 
 2494 void MacroAssembler::pop_CPU_state() {
 2495   pop_FPU_state();
 2496   pop_IU_state();
 2497 }
 2498 
 2499 void MacroAssembler::pop_FPU_state() {
 2500   fxrstor(Address(rsp, 0));
 2501   addptr(rsp, FPUStateSizeInWords * wordSize);
 2502 }
 2503 
 2504 void MacroAssembler::pop_IU_state() {
 2505   popa();
 2506   addq(rsp, 8);
 2507   popf();
 2508 }
 2509 
 2510 // Save Integer and Float state
 2511 // Warning: Stack must be 16 byte aligned (64bit)
 2512 void MacroAssembler::push_CPU_state() {
 2513   push_IU_state();
 2514   push_FPU_state();
 2515 }
 2516 
 2517 void MacroAssembler::push_FPU_state() {
 2518   subptr(rsp, FPUStateSizeInWords * wordSize);
 2519   fxsave(Address(rsp, 0));
 2520 }
 2521 
 2522 void MacroAssembler::push_IU_state() {
 2523   // Push flags first because pusha kills them
 2524   pushf();
 2525   // Make sure rsp stays 16-byte aligned
 2526   subq(rsp, 8);
 2527   pusha();
 2528 }
 2529 
 2530 void MacroAssembler::push_cont_fastpath() {
 2531   if (!Continuations::enabled()) return;
 2532 
 2533   Label L_done;
 2534   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2535   jccb(Assembler::belowEqual, L_done);
 2536   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2537   bind(L_done);
 2538 }
 2539 
 2540 void MacroAssembler::pop_cont_fastpath() {
 2541   if (!Continuations::enabled()) return;
 2542 
 2543   Label L_done;
 2544   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2545   jccb(Assembler::below, L_done);
 2546   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2547   bind(L_done);
 2548 }
 2549 
 2550 #ifdef ASSERT
 2551 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2552   Label no_cont;
 2553   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2554   testl(cont, cont);
 2555   jcc(Assembler::zero, no_cont);
 2556   stop(name);
 2557   bind(no_cont);
 2558 }
 2559 #endif
 2560 
 2561 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2562   // we must set sp to zero to clear frame
 2563   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2564   // must clear fp, so that compiled frames are not confused; it is
 2565   // possible that we need it only for debugging
 2566   if (clear_fp) {
 2567     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2568   }
 2569   // Always clear the pc because it could have been set by make_walkable()
 2570   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2571   vzeroupper();
 2572 }
 2573 
 2574 void MacroAssembler::round_to(Register reg, int modulus) {
 2575   addptr(reg, modulus - 1);
 2576   andptr(reg, -modulus);
 2577 }
 2578 
 2579 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2580   if (at_return) {
 2581     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2582     // we may safely use rsp instead to perform the stack watermark check.
 2583     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2584     jcc(Assembler::above, slow_path);
 2585     return;
 2586   }
 2587   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2588   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2589 }
 2590 
 2591 // Calls to C land
 2592 //
 2593 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2594 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2595 // has to be reset to 0. This is required to allow proper stack traversal.
 2596 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2597                                          Register last_java_fp,
 2598                                          address  last_java_pc,
 2599                                          Register rscratch) {
 2600   vzeroupper();
 2601   // determine last_java_sp register
 2602   if (!last_java_sp->is_valid()) {
 2603     last_java_sp = rsp;
 2604   }
 2605   // last_java_fp is optional
 2606   if (last_java_fp->is_valid()) {
 2607     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2608   }
 2609   // last_java_pc is optional
 2610   if (last_java_pc != nullptr) {
 2611     Address java_pc(r15_thread,
 2612                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2613     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2614   }
 2615   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2616 }
 2617 
 2618 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2619                                          Register last_java_fp,
 2620                                          Label &L,
 2621                                          Register scratch) {
 2622   lea(scratch, L);
 2623   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2624   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2625 }
 2626 
 2627 void MacroAssembler::shlptr(Register dst, int imm8) {
 2628   shlq(dst, imm8);
 2629 }
 2630 
 2631 void MacroAssembler::shrptr(Register dst, int imm8) {
 2632   shrq(dst, imm8);
 2633 }
 2634 
 2635 void MacroAssembler::sign_extend_byte(Register reg) {
 2636   movsbl(reg, reg); // movsxb
 2637 }
 2638 
 2639 void MacroAssembler::sign_extend_short(Register reg) {
 2640   movswl(reg, reg); // movsxw
 2641 }
 2642 
 2643 void MacroAssembler::narrow_subword_type(Register reg, BasicType bt) {
 2644   assert(is_subword_type(bt), "required");
 2645   switch (bt) {
 2646   case T_BOOLEAN: andl(reg, 1); break;
 2647   case T_BYTE:    movsbl(reg, reg); break;
 2648   case T_CHAR:    movzwl(reg, reg); break;
 2649   case T_SHORT:   movswl(reg, reg); break;
 2650   default:        ShouldNotReachHere();
 2651   }
 2652 }
 2653 
 2654 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2655   if (imm32 >= 0 && is8bit(imm32)) {
 2656     testb(dst, imm32);
 2657   } else {
 2658     Assembler::testl(dst, imm32);
 2659   }
 2660 }
 2661 
 2662 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2663   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2664     testb(dst, imm32);
 2665   } else {
 2666     Assembler::testl(dst, imm32);
 2667   }
 2668 }
 2669 
 2670 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2671   assert(always_reachable(src), "Address should be reachable");
 2672   testl(dst, as_Address(src));
 2673 }
 2674 
 2675 void MacroAssembler::testq(Address dst, int32_t imm32) {
 2676   if (imm32 >= 0) {
 2677     testl(dst, imm32);
 2678   } else {
 2679     Assembler::testq(dst, imm32);
 2680   }
 2681 }
 2682 
 2683 void MacroAssembler::testq(Register dst, int32_t imm32) {
 2684   if (imm32 >= 0) {
 2685     testl(dst, imm32);
 2686   } else {
 2687     Assembler::testq(dst, imm32);
 2688   }
 2689 }
 2690 
 2691 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2692   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2693   Assembler::pcmpeqb(dst, src);
 2694 }
 2695 
 2696 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2697   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2698   Assembler::pcmpeqw(dst, src);
 2699 }
 2700 
 2701 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2702   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2703   Assembler::pcmpestri(dst, src, imm8);
 2704 }
 2705 
 2706 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2707   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2708   Assembler::pcmpestri(dst, src, imm8);
 2709 }
 2710 
 2711 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2712   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2713   Assembler::pmovzxbw(dst, src);
 2714 }
 2715 
 2716 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2717   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2718   Assembler::pmovzxbw(dst, src);
 2719 }
 2720 
 2721 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2722   assert((src->encoding() < 16),"XMM register should be 0-15");
 2723   Assembler::pmovmskb(dst, src);
 2724 }
 2725 
 2726 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2727   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2728   Assembler::ptest(dst, src);
 2729 }
 2730 
 2731 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2732   assert(rscratch != noreg || always_reachable(src), "missing");
 2733 
 2734   if (reachable(src)) {
 2735     Assembler::sqrtss(dst, as_Address(src));
 2736   } else {
 2737     lea(rscratch, src);
 2738     Assembler::sqrtss(dst, Address(rscratch, 0));
 2739   }
 2740 }
 2741 
 2742 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2743   assert(rscratch != noreg || always_reachable(src), "missing");
 2744 
 2745   if (reachable(src)) {
 2746     Assembler::subsd(dst, as_Address(src));
 2747   } else {
 2748     lea(rscratch, src);
 2749     Assembler::subsd(dst, Address(rscratch, 0));
 2750   }
 2751 }
 2752 
 2753 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2754   assert(rscratch != noreg || always_reachable(src), "missing");
 2755 
 2756   if (reachable(src)) {
 2757     Assembler::roundsd(dst, as_Address(src), rmode);
 2758   } else {
 2759     lea(rscratch, src);
 2760     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2761   }
 2762 }
 2763 
 2764 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2765   assert(rscratch != noreg || always_reachable(src), "missing");
 2766 
 2767   if (reachable(src)) {
 2768     Assembler::subss(dst, as_Address(src));
 2769   } else {
 2770     lea(rscratch, src);
 2771     Assembler::subss(dst, Address(rscratch, 0));
 2772   }
 2773 }
 2774 
 2775 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2776   assert(rscratch != noreg || always_reachable(src), "missing");
 2777 
 2778   if (reachable(src)) {
 2779     Assembler::ucomisd(dst, as_Address(src));
 2780   } else {
 2781     lea(rscratch, src);
 2782     Assembler::ucomisd(dst, Address(rscratch, 0));
 2783   }
 2784 }
 2785 
 2786 void MacroAssembler::evucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2787   assert(rscratch != noreg || always_reachable(src), "missing");
 2788 
 2789   if (reachable(src)) {
 2790     Assembler::evucomxsd(dst, as_Address(src));
 2791   } else {
 2792     lea(rscratch, src);
 2793     Assembler::evucomxsd(dst, Address(rscratch, 0));
 2794   }
 2795 }
 2796 
 2797 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2798   assert(rscratch != noreg || always_reachable(src), "missing");
 2799 
 2800   if (reachable(src)) {
 2801     Assembler::ucomiss(dst, as_Address(src));
 2802   } else {
 2803     lea(rscratch, src);
 2804     Assembler::ucomiss(dst, Address(rscratch, 0));
 2805   }
 2806 }
 2807 
 2808 void MacroAssembler::evucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2809   assert(rscratch != noreg || always_reachable(src), "missing");
 2810 
 2811   if (reachable(src)) {
 2812     Assembler::evucomxss(dst, as_Address(src));
 2813   } else {
 2814     lea(rscratch, src);
 2815     Assembler::evucomxss(dst, Address(rscratch, 0));
 2816   }
 2817 }
 2818 
 2819 void MacroAssembler::evucomish(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2820   assert(rscratch != noreg || always_reachable(src), "missing");
 2821 
 2822   if (reachable(src)) {
 2823     Assembler::evucomish(dst, as_Address(src));
 2824   } else {
 2825     lea(rscratch, src);
 2826     Assembler::evucomish(dst, Address(rscratch, 0));
 2827   }
 2828 }
 2829 
 2830 void MacroAssembler::evucomxsh(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2831   assert(rscratch != noreg || always_reachable(src), "missing");
 2832 
 2833   if (reachable(src)) {
 2834     Assembler::evucomxsh(dst, as_Address(src));
 2835   } else {
 2836     lea(rscratch, src);
 2837     Assembler::evucomxsh(dst, Address(rscratch, 0));
 2838   }
 2839 }
 2840 
 2841 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2842   assert(rscratch != noreg || always_reachable(src), "missing");
 2843 
 2844   // Used in sign-bit flipping with aligned address.
 2845   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2846 
 2847   if (UseAVX > 2 &&
 2848       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2849       (dst->encoding() >= 16)) {
 2850     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2851   } else if (reachable(src)) {
 2852     Assembler::xorpd(dst, as_Address(src));
 2853   } else {
 2854     lea(rscratch, src);
 2855     Assembler::xorpd(dst, Address(rscratch, 0));
 2856   }
 2857 }
 2858 
 2859 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2860   if (UseAVX > 2 &&
 2861       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2862       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2863     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2864   } else {
 2865     Assembler::xorpd(dst, src);
 2866   }
 2867 }
 2868 
 2869 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2870   if (UseAVX > 2 &&
 2871       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2872       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2873     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2874   } else {
 2875     Assembler::xorps(dst, src);
 2876   }
 2877 }
 2878 
 2879 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2880   assert(rscratch != noreg || always_reachable(src), "missing");
 2881 
 2882   // Used in sign-bit flipping with aligned address.
 2883   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2884 
 2885   if (UseAVX > 2 &&
 2886       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2887       (dst->encoding() >= 16)) {
 2888     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2889   } else if (reachable(src)) {
 2890     Assembler::xorps(dst, as_Address(src));
 2891   } else {
 2892     lea(rscratch, src);
 2893     Assembler::xorps(dst, Address(rscratch, 0));
 2894   }
 2895 }
 2896 
 2897 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2898   assert(rscratch != noreg || always_reachable(src), "missing");
 2899 
 2900   // Used in sign-bit flipping with aligned address.
 2901   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2902   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2903   if (reachable(src)) {
 2904     Assembler::pshufb(dst, as_Address(src));
 2905   } else {
 2906     lea(rscratch, src);
 2907     Assembler::pshufb(dst, Address(rscratch, 0));
 2908   }
 2909 }
 2910 
 2911 // AVX 3-operands instructions
 2912 
 2913 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2914   assert(rscratch != noreg || always_reachable(src), "missing");
 2915 
 2916   if (reachable(src)) {
 2917     vaddsd(dst, nds, as_Address(src));
 2918   } else {
 2919     lea(rscratch, src);
 2920     vaddsd(dst, nds, Address(rscratch, 0));
 2921   }
 2922 }
 2923 
 2924 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2925   assert(rscratch != noreg || always_reachable(src), "missing");
 2926 
 2927   if (reachable(src)) {
 2928     vaddss(dst, nds, as_Address(src));
 2929   } else {
 2930     lea(rscratch, src);
 2931     vaddss(dst, nds, Address(rscratch, 0));
 2932   }
 2933 }
 2934 
 2935 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2936   assert(UseAVX > 0, "requires some form of AVX");
 2937   assert(rscratch != noreg || always_reachable(src), "missing");
 2938 
 2939   if (reachable(src)) {
 2940     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2941   } else {
 2942     lea(rscratch, src);
 2943     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2944   }
 2945 }
 2946 
 2947 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2948   assert(UseAVX > 0, "requires some form of AVX");
 2949   assert(rscratch != noreg || always_reachable(src), "missing");
 2950 
 2951   if (reachable(src)) {
 2952     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2953   } else {
 2954     lea(rscratch, src);
 2955     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2956   }
 2957 }
 2958 
 2959 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2960   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2961   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2962 
 2963   vandps(dst, nds, negate_field, vector_len, rscratch);
 2964 }
 2965 
 2966 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2967   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2968   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2969 
 2970   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2971 }
 2972 
 2973 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2974   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2975   Assembler::vpaddb(dst, nds, src, vector_len);
 2976 }
 2977 
 2978 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2979   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2980   Assembler::vpaddb(dst, nds, src, vector_len);
 2981 }
 2982 
 2983 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2984   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2985   Assembler::vpaddw(dst, nds, src, vector_len);
 2986 }
 2987 
 2988 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2989   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2990   Assembler::vpaddw(dst, nds, src, vector_len);
 2991 }
 2992 
 2993 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2994   assert(rscratch != noreg || always_reachable(src), "missing");
 2995 
 2996   if (reachable(src)) {
 2997     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 2998   } else {
 2999     lea(rscratch, src);
 3000     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 3001   }
 3002 }
 3003 
 3004 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3005   assert(rscratch != noreg || always_reachable(src), "missing");
 3006 
 3007   if (reachable(src)) {
 3008     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 3009   } else {
 3010     lea(rscratch, src);
 3011     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 3012   }
 3013 }
 3014 
 3015 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3016   assert(rscratch != noreg || always_reachable(src), "missing");
 3017 
 3018   if (reachable(src)) {
 3019     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 3020   } else {
 3021     lea(rscratch, src);
 3022     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 3023   }
 3024 }
 3025 
 3026 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3027   assert(rscratch != noreg || always_reachable(src), "missing");
 3028 
 3029   if (reachable(src)) {
 3030     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 3031   } else {
 3032     lea(rscratch, src);
 3033     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 3034   }
 3035 }
 3036 
 3037 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3038   assert(rscratch != noreg || always_reachable(src), "missing");
 3039 
 3040   if (reachable(src)) {
 3041     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 3042   } else {
 3043     lea(rscratch, src);
 3044     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 3045   }
 3046 }
 3047 
 3048 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3049   assert(rscratch != noreg || always_reachable(src), "missing");
 3050 
 3051   if (reachable(src)) {
 3052     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 3053   } else {
 3054     lea(rscratch, src);
 3055     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 3056   }
 3057 }
 3058 
 3059 // Vector float blend
 3060 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3061 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3062   // WARN: Allow dst == (src1|src2), mask == scratch
 3063   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3064                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3065   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 3066   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3067   if (blend_emulation && scratch_available && dst_available) {
 3068     if (compute_mask) {
 3069       vpsrad(scratch, mask, 32, vector_len);
 3070       mask = scratch;
 3071     }
 3072     if (dst == src1) {
 3073       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 3074       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3075     } else {
 3076       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3077       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 3078     }
 3079     vpor(dst, dst, scratch, vector_len);
 3080   } else {
 3081     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 3082   }
 3083 }
 3084 
 3085 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3086 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3087   // WARN: Allow dst == (src1|src2), mask == scratch
 3088   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3089                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3090   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 3091   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3092   if (blend_emulation && scratch_available && dst_available) {
 3093     if (compute_mask) {
 3094       vpxor(scratch, scratch, scratch, vector_len);
 3095       vpcmpgtq(scratch, scratch, mask, vector_len);
 3096       mask = scratch;
 3097     }
 3098     if (dst == src1) {
 3099       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 3100       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3101     } else {
 3102       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3103       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 3104     }
 3105     vpor(dst, dst, scratch, vector_len);
 3106   } else {
 3107     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 3108   }
 3109 }
 3110 
 3111 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3112   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3113   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 3114 }
 3115 
 3116 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 3117   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3118   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 3119 }
 3120 
 3121 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3122   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3123   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3124 }
 3125 
 3126 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3127   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3128   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3129 }
 3130 
 3131 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3132   assert(rscratch != noreg || always_reachable(src), "missing");
 3133 
 3134   if (reachable(src)) {
 3135     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 3136   } else {
 3137     lea(rscratch, src);
 3138     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 3139   }
 3140 }
 3141 
 3142 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3143                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3144   assert(rscratch != noreg || always_reachable(src), "missing");
 3145 
 3146   if (reachable(src)) {
 3147     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3148   } else {
 3149     lea(rscratch, src);
 3150     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3151   }
 3152 }
 3153 
 3154 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3155                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3156   assert(rscratch != noreg || always_reachable(src), "missing");
 3157 
 3158   if (reachable(src)) {
 3159     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3160   } else {
 3161     lea(rscratch, src);
 3162     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3163   }
 3164 }
 3165 
 3166 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3167                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3168   assert(rscratch != noreg || always_reachable(src), "missing");
 3169 
 3170   if (reachable(src)) {
 3171     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3172   } else {
 3173     lea(rscratch, src);
 3174     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3175   }
 3176 }
 3177 
 3178 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3179                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3180   assert(rscratch != noreg || always_reachable(src), "missing");
 3181 
 3182   if (reachable(src)) {
 3183     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3184   } else {
 3185     lea(rscratch, src);
 3186     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3187   }
 3188 }
 3189 
 3190 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3191   if (width == Assembler::Q) {
 3192     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3193   } else {
 3194     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3195   }
 3196 }
 3197 
 3198 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3199   int eq_cond_enc = 0x29;
 3200   int gt_cond_enc = 0x37;
 3201   if (width != Assembler::Q) {
 3202     eq_cond_enc = 0x74 + width;
 3203     gt_cond_enc = 0x64 + width;
 3204   }
 3205   switch (cond) {
 3206   case eq:
 3207     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3208     break;
 3209   case neq:
 3210     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3211     vallones(xtmp, vector_len);
 3212     vpxor(dst, xtmp, dst, vector_len);
 3213     break;
 3214   case le:
 3215     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3216     vallones(xtmp, vector_len);
 3217     vpxor(dst, xtmp, dst, vector_len);
 3218     break;
 3219   case nlt:
 3220     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3221     vallones(xtmp, vector_len);
 3222     vpxor(dst, xtmp, dst, vector_len);
 3223     break;
 3224   case lt:
 3225     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3226     break;
 3227   case nle:
 3228     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3229     break;
 3230   default:
 3231     assert(false, "Should not reach here");
 3232   }
 3233 }
 3234 
 3235 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3236   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3237   Assembler::vpmovzxbw(dst, src, vector_len);
 3238 }
 3239 
 3240 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3241   assert((src->encoding() < 16),"XMM register should be 0-15");
 3242   Assembler::vpmovmskb(dst, src, vector_len);
 3243 }
 3244 
 3245 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3246   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3247   Assembler::vpmullw(dst, nds, src, vector_len);
 3248 }
 3249 
 3250 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3251   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3252   Assembler::vpmullw(dst, nds, src, vector_len);
 3253 }
 3254 
 3255 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3256   assert((UseAVX > 0), "AVX support is needed");
 3257   assert(rscratch != noreg || always_reachable(src), "missing");
 3258 
 3259   if (reachable(src)) {
 3260     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3261   } else {
 3262     lea(rscratch, src);
 3263     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3264   }
 3265 }
 3266 
 3267 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3268   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3269   Assembler::vpsubb(dst, nds, src, vector_len);
 3270 }
 3271 
 3272 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3273   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3274   Assembler::vpsubb(dst, nds, src, vector_len);
 3275 }
 3276 
 3277 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3278   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3279   Assembler::vpsubw(dst, nds, src, vector_len);
 3280 }
 3281 
 3282 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3283   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3284   Assembler::vpsubw(dst, nds, src, vector_len);
 3285 }
 3286 
 3287 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3288   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3289   Assembler::vpsraw(dst, nds, shift, vector_len);
 3290 }
 3291 
 3292 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3293   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3294   Assembler::vpsraw(dst, nds, shift, vector_len);
 3295 }
 3296 
 3297 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3298   assert(UseAVX > 2,"");
 3299   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3300      vector_len = 2;
 3301   }
 3302   Assembler::evpsraq(dst, nds, shift, vector_len);
 3303 }
 3304 
 3305 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3306   assert(UseAVX > 2,"");
 3307   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3308      vector_len = 2;
 3309   }
 3310   Assembler::evpsraq(dst, nds, shift, vector_len);
 3311 }
 3312 
 3313 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3314   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3315   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3316 }
 3317 
 3318 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3319   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3320   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3321 }
 3322 
 3323 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3324   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3325   Assembler::vpsllw(dst, nds, shift, vector_len);
 3326 }
 3327 
 3328 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3329   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3330   Assembler::vpsllw(dst, nds, shift, vector_len);
 3331 }
 3332 
 3333 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3334   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3335   Assembler::vptest(dst, src);
 3336 }
 3337 
 3338 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3339   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3340   Assembler::punpcklbw(dst, src);
 3341 }
 3342 
 3343 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3344   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3345   Assembler::pshufd(dst, src, mode);
 3346 }
 3347 
 3348 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3349   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3350   Assembler::pshuflw(dst, src, mode);
 3351 }
 3352 
 3353 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3354   assert(rscratch != noreg || always_reachable(src), "missing");
 3355 
 3356   if (reachable(src)) {
 3357     vandpd(dst, nds, as_Address(src), vector_len);
 3358   } else {
 3359     lea(rscratch, src);
 3360     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3361   }
 3362 }
 3363 
 3364 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3365   assert(rscratch != noreg || always_reachable(src), "missing");
 3366 
 3367   if (reachable(src)) {
 3368     vandps(dst, nds, as_Address(src), vector_len);
 3369   } else {
 3370     lea(rscratch, src);
 3371     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3372   }
 3373 }
 3374 
 3375 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3376                             bool merge, int vector_len, Register rscratch) {
 3377   assert(rscratch != noreg || always_reachable(src), "missing");
 3378 
 3379   if (reachable(src)) {
 3380     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3381   } else {
 3382     lea(rscratch, src);
 3383     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3384   }
 3385 }
 3386 
 3387 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3388   assert(rscratch != noreg || always_reachable(src), "missing");
 3389 
 3390   if (reachable(src)) {
 3391     vdivsd(dst, nds, as_Address(src));
 3392   } else {
 3393     lea(rscratch, src);
 3394     vdivsd(dst, nds, Address(rscratch, 0));
 3395   }
 3396 }
 3397 
 3398 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3399   assert(rscratch != noreg || always_reachable(src), "missing");
 3400 
 3401   if (reachable(src)) {
 3402     vdivss(dst, nds, as_Address(src));
 3403   } else {
 3404     lea(rscratch, src);
 3405     vdivss(dst, nds, Address(rscratch, 0));
 3406   }
 3407 }
 3408 
 3409 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3410   assert(rscratch != noreg || always_reachable(src), "missing");
 3411 
 3412   if (reachable(src)) {
 3413     vmulsd(dst, nds, as_Address(src));
 3414   } else {
 3415     lea(rscratch, src);
 3416     vmulsd(dst, nds, Address(rscratch, 0));
 3417   }
 3418 }
 3419 
 3420 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3421   assert(rscratch != noreg || always_reachable(src), "missing");
 3422 
 3423   if (reachable(src)) {
 3424     vmulss(dst, nds, as_Address(src));
 3425   } else {
 3426     lea(rscratch, src);
 3427     vmulss(dst, nds, Address(rscratch, 0));
 3428   }
 3429 }
 3430 
 3431 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3432   assert(rscratch != noreg || always_reachable(src), "missing");
 3433 
 3434   if (reachable(src)) {
 3435     vsubsd(dst, nds, as_Address(src));
 3436   } else {
 3437     lea(rscratch, src);
 3438     vsubsd(dst, nds, Address(rscratch, 0));
 3439   }
 3440 }
 3441 
 3442 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3443   assert(rscratch != noreg || always_reachable(src), "missing");
 3444 
 3445   if (reachable(src)) {
 3446     vsubss(dst, nds, as_Address(src));
 3447   } else {
 3448     lea(rscratch, src);
 3449     vsubss(dst, nds, Address(rscratch, 0));
 3450   }
 3451 }
 3452 
 3453 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3454   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3455   assert(rscratch != noreg || always_reachable(src), "missing");
 3456 
 3457   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3458 }
 3459 
 3460 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3461   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3462   assert(rscratch != noreg || always_reachable(src), "missing");
 3463 
 3464   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3465 }
 3466 
 3467 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3468   assert(rscratch != noreg || always_reachable(src), "missing");
 3469 
 3470   if (reachable(src)) {
 3471     vxorpd(dst, nds, as_Address(src), vector_len);
 3472   } else {
 3473     lea(rscratch, src);
 3474     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3475   }
 3476 }
 3477 
 3478 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3479   assert(rscratch != noreg || always_reachable(src), "missing");
 3480 
 3481   if (reachable(src)) {
 3482     vxorps(dst, nds, as_Address(src), vector_len);
 3483   } else {
 3484     lea(rscratch, src);
 3485     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3486   }
 3487 }
 3488 
 3489 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3490   assert(rscratch != noreg || always_reachable(src), "missing");
 3491 
 3492   if (UseAVX > 1 || (vector_len < 1)) {
 3493     if (reachable(src)) {
 3494       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3495     } else {
 3496       lea(rscratch, src);
 3497       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3498     }
 3499   } else {
 3500     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3501   }
 3502 }
 3503 
 3504 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3505   assert(rscratch != noreg || always_reachable(src), "missing");
 3506 
 3507   if (reachable(src)) {
 3508     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3509   } else {
 3510     lea(rscratch, src);
 3511     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3512   }
 3513 }
 3514 
 3515 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3516   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3517   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3518   // The inverted mask is sign-extended
 3519   andptr(possibly_non_local, inverted_mask);
 3520 }
 3521 
 3522 void MacroAssembler::resolve_jobject(Register value,
 3523                                      Register tmp) {
 3524   Register thread = r15_thread;
 3525   assert_different_registers(value, thread, tmp);
 3526   Label done, tagged, weak_tagged;
 3527   testptr(value, value);
 3528   jcc(Assembler::zero, done);           // Use null as-is.
 3529   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3530   jcc(Assembler::notZero, tagged);
 3531 
 3532   // Resolve local handle
 3533   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3534   verify_oop(value);
 3535   jmp(done);
 3536 
 3537   bind(tagged);
 3538   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3539   jcc(Assembler::notZero, weak_tagged);
 3540 
 3541   // Resolve global handle
 3542   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3543   verify_oop(value);
 3544   jmp(done);
 3545 
 3546   bind(weak_tagged);
 3547   // Resolve jweak.
 3548   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3549                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3550   verify_oop(value);
 3551 
 3552   bind(done);
 3553 }
 3554 
 3555 void MacroAssembler::resolve_global_jobject(Register value,
 3556                                             Register tmp) {
 3557   Register thread = r15_thread;
 3558   assert_different_registers(value, thread, tmp);
 3559   Label done;
 3560 
 3561   testptr(value, value);
 3562   jcc(Assembler::zero, done);           // Use null as-is.
 3563 
 3564 #ifdef ASSERT
 3565   {
 3566     Label valid_global_tag;
 3567     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3568     jcc(Assembler::notZero, valid_global_tag);
 3569     stop("non global jobject using resolve_global_jobject");
 3570     bind(valid_global_tag);
 3571   }
 3572 #endif
 3573 
 3574   // Resolve global handle
 3575   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3576   verify_oop(value);
 3577 
 3578   bind(done);
 3579 }
 3580 
 3581 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3582   subq(dst, imm32);
 3583 }
 3584 
 3585 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3586 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3587   subq_imm32(dst, imm32);
 3588 }
 3589 
 3590 void MacroAssembler::subptr(Register dst, Register src) {
 3591   subq(dst, src);
 3592 }
 3593 
 3594 // C++ bool manipulation
 3595 void MacroAssembler::testbool(Register dst) {
 3596   if(sizeof(bool) == 1)
 3597     testb(dst, 0xff);
 3598   else if(sizeof(bool) == 2) {
 3599     // testw implementation needed for two byte bools
 3600     ShouldNotReachHere();
 3601   } else if(sizeof(bool) == 4)
 3602     testl(dst, dst);
 3603   else
 3604     // unsupported
 3605     ShouldNotReachHere();
 3606 }
 3607 
 3608 void MacroAssembler::testptr(Register dst, Register src) {
 3609   testq(dst, src);
 3610 }
 3611 
 3612 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3613 void MacroAssembler::tlab_allocate(Register obj,
 3614                                    Register var_size_in_bytes,
 3615                                    int con_size_in_bytes,
 3616                                    Register t1,
 3617                                    Register t2,
 3618                                    Label& slow_case) {
 3619   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3620   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3621 }
 3622 
 3623 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3624   RegSet regs;
 3625   regs += RegSet::of(rax, rcx, rdx);
 3626 #ifndef _WINDOWS
 3627   regs += RegSet::of(rsi, rdi);
 3628 #endif
 3629   regs += RegSet::range(r8, r11);
 3630   if (UseAPX) {
 3631     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3632   }
 3633   return regs;
 3634 }
 3635 
 3636 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3637   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3638 #if defined(_WINDOWS)
 3639   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3640   if (num_xmm_registers > 16) {
 3641      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3642   }
 3643   return result;
 3644 #else
 3645   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3646 #endif
 3647 }
 3648 
 3649 // C1 only ever uses the first double/float of the XMM register.
 3650 static int xmm_save_size() { return sizeof(double); }
 3651 
 3652 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3653   masm->movdbl(Address(rsp, offset), reg);
 3654 }
 3655 
 3656 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3657   masm->movdbl(reg, Address(rsp, offset));
 3658 }
 3659 
 3660 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3661                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3662 
 3663   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3664                          StackAlignmentInBytes);
 3665   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3666 
 3667   return gp_area_size + xmm_area_size;
 3668 }
 3669 
 3670 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3671   block_comment("push_call_clobbered_registers start");
 3672   // Regular registers
 3673   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3674 
 3675   int gp_area_size;
 3676   int xmm_area_size;
 3677   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3678                                                gp_area_size, xmm_area_size);
 3679   subptr(rsp, total_save_size);
 3680 
 3681   push_set(gp_registers_to_push, 0);
 3682 
 3683   if (save_fpu) {
 3684     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3685   }
 3686 
 3687   block_comment("push_call_clobbered_registers end");
 3688 }
 3689 
 3690 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3691   block_comment("pop_call_clobbered_registers start");
 3692 
 3693   RegSet gp_registers_to_pop = 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_pop, call_clobbered_xmm_registers(), restore_fpu,
 3698                                                gp_area_size, xmm_area_size);
 3699 
 3700   if (restore_fpu) {
 3701     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3702   }
 3703 
 3704   pop_set(gp_registers_to_pop, 0);
 3705 
 3706   addptr(rsp, total_save_size);
 3707 
 3708   vzeroupper();
 3709 
 3710   block_comment("pop_call_clobbered_registers end");
 3711 }
 3712 
 3713 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3714   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3715   int spill_offset = offset;
 3716 
 3717   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3718     save_xmm_register(this, spill_offset, *it);
 3719     spill_offset += xmm_save_size();
 3720   }
 3721 }
 3722 
 3723 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3724   int restore_size = set.size() * xmm_save_size();
 3725   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3726 
 3727   int restore_offset = offset + restore_size - xmm_save_size();
 3728 
 3729   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3730     restore_xmm_register(this, restore_offset, *it);
 3731     restore_offset -= xmm_save_size();
 3732   }
 3733 }
 3734 
 3735 void MacroAssembler::push_set(RegSet set, int offset) {
 3736   int spill_offset;
 3737   if (offset == -1) {
 3738     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3739     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3740     subptr(rsp, aligned_size);
 3741     spill_offset = 0;
 3742   } else {
 3743     spill_offset = offset;
 3744   }
 3745 
 3746   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3747     movptr(Address(rsp, spill_offset), *it);
 3748     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3749   }
 3750 }
 3751 
 3752 void MacroAssembler::pop_set(RegSet set, int offset) {
 3753 
 3754   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3755   int restore_size = set.size() * gp_reg_size;
 3756   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3757 
 3758   int restore_offset;
 3759   if (offset == -1) {
 3760     restore_offset = restore_size - gp_reg_size;
 3761   } else {
 3762     restore_offset = offset + restore_size - gp_reg_size;
 3763   }
 3764   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3765     movptr(*it, Address(rsp, restore_offset));
 3766     restore_offset -= gp_reg_size;
 3767   }
 3768 
 3769   if (offset == -1) {
 3770     addptr(rsp, aligned_size);
 3771   }
 3772 }
 3773 
 3774 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3775 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3776   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3777   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3778   Label done;
 3779 
 3780   testptr(length_in_bytes, length_in_bytes);
 3781   jcc(Assembler::zero, done);
 3782 
 3783   // initialize topmost word, divide index by 2, check if odd and test if zero
 3784   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3785 #ifdef ASSERT
 3786   {
 3787     Label L;
 3788     testptr(length_in_bytes, BytesPerWord - 1);
 3789     jcc(Assembler::zero, L);
 3790     stop("length must be a multiple of BytesPerWord");
 3791     bind(L);
 3792   }
 3793 #endif
 3794   Register index = length_in_bytes;
 3795   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3796   if (UseIncDec) {
 3797     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3798   } else {
 3799     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3800     shrptr(index, 1);
 3801   }
 3802 
 3803   // initialize remaining object fields: index is a multiple of 2 now
 3804   {
 3805     Label loop;
 3806     bind(loop);
 3807     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3808     decrement(index);
 3809     jcc(Assembler::notZero, loop);
 3810   }
 3811 
 3812   bind(done);
 3813 }
 3814 
 3815 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
 3816   movptr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
 3817 #ifdef ASSERT
 3818   {
 3819     Label done;
 3820     cmpptr(layout_info, 0);
 3821     jcc(Assembler::notEqual, done);
 3822     stop("inline_layout_info_array is null");
 3823     bind(done);
 3824   }
 3825 #endif
 3826 
 3827   InlineLayoutInfo array[2];
 3828   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
 3829   if (is_power_of_2(size)) {
 3830     shll(index, log2i_exact(size)); // Scale index by power of 2
 3831   } else {
 3832     imull(index, index, size); // Scale the index to be the entry index * array_element_size
 3833   }
 3834   lea(layout_info, Address(layout_info, index, Address::times_1, Array<InlineLayoutInfo>::base_offset_in_bytes()));
 3835 }
 3836 
 3837 // Look up the method for a megamorphic invokeinterface call.
 3838 // The target method is determined by <intf_klass, itable_index>.
 3839 // The receiver klass is in recv_klass.
 3840 // On success, the result will be in method_result, and execution falls through.
 3841 // On failure, execution transfers to the given label.
 3842 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3843                                              Register intf_klass,
 3844                                              RegisterOrConstant itable_index,
 3845                                              Register method_result,
 3846                                              Register scan_temp,
 3847                                              Label& L_no_such_interface,
 3848                                              bool return_method) {
 3849   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3850   assert_different_registers(method_result, intf_klass, scan_temp);
 3851   assert(recv_klass != method_result || !return_method,
 3852          "recv_klass can be destroyed when method isn't needed");
 3853 
 3854   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3855          "caller must use same register for non-constant itable index as for method");
 3856 
 3857   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3858   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3859   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3860   int scan_step   = itableOffsetEntry::size() * wordSize;
 3861   int vte_size    = vtableEntry::size_in_bytes();
 3862   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3863   assert(vte_size == wordSize, "else adjust times_vte_scale");
 3864 
 3865   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3866 
 3867   // Could store the aligned, prescaled offset in the klass.
 3868   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 3869 
 3870   if (return_method) {
 3871     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 3872     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3873     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 3874   }
 3875 
 3876   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 3877   //   if (scan->interface() == intf) {
 3878   //     result = (klass + scan->offset() + itable_index);
 3879   //   }
 3880   // }
 3881   Label search, found_method;
 3882 
 3883   for (int peel = 1; peel >= 0; peel--) {
 3884     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 3885     cmpptr(intf_klass, method_result);
 3886 
 3887     if (peel) {
 3888       jccb(Assembler::equal, found_method);
 3889     } else {
 3890       jccb(Assembler::notEqual, search);
 3891       // (invert the test to fall through to found_method...)
 3892     }
 3893 
 3894     if (!peel)  break;
 3895 
 3896     bind(search);
 3897 
 3898     // Check that the previous entry is non-null.  A null entry means that
 3899     // the receiver class doesn't implement the interface, and wasn't the
 3900     // same as when the caller was compiled.
 3901     testptr(method_result, method_result);
 3902     jcc(Assembler::zero, L_no_such_interface);
 3903     addptr(scan_temp, scan_step);
 3904   }
 3905 
 3906   bind(found_method);
 3907 
 3908   if (return_method) {
 3909     // Got a hit.
 3910     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 3911     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 3912   }
 3913 }
 3914 
 3915 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 3916 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 3917 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 3918 // The target method is determined by <holder_klass, itable_index>.
 3919 // The receiver klass is in recv_klass.
 3920 // On success, the result will be in method_result, and execution falls through.
 3921 // On failure, execution transfers to the given label.
 3922 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 3923                                                   Register holder_klass,
 3924                                                   Register resolved_klass,
 3925                                                   Register method_result,
 3926                                                   Register scan_temp,
 3927                                                   Register temp_reg2,
 3928                                                   Register receiver,
 3929                                                   int itable_index,
 3930                                                   Label& L_no_such_interface) {
 3931   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 3932   Register temp_itbl_klass = method_result;
 3933   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 3934 
 3935   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3936   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3937   int scan_step = itableOffsetEntry::size() * wordSize;
 3938   int vte_size = vtableEntry::size_in_bytes();
 3939   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 3940   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 3941   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3942   assert(vte_size == wordSize, "adjust times_vte_scale");
 3943 
 3944   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 3945 
 3946   // temp_itbl_klass = recv_klass.itable[0]
 3947   // scan_temp = &recv_klass.itable[0] + step
 3948   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3949   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 3950   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 3951   xorptr(temp_reg, temp_reg);
 3952 
 3953   // Initial checks:
 3954   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 3955   //   - if (itable[0] == 0), no such interface
 3956   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 3957   cmpptr(holder_klass, resolved_klass);
 3958   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 3959   testptr(temp_itbl_klass, temp_itbl_klass);
 3960   jccb(Assembler::zero, L_no_such_interface);
 3961   cmpptr(holder_klass, temp_itbl_klass);
 3962   jccb(Assembler::equal, L_holder_found);
 3963 
 3964   // Loop: Look for holder_klass record in itable
 3965   //   do {
 3966   //     tmp = itable[index];
 3967   //     index += step;
 3968   //     if (tmp == holder_klass) {
 3969   //       goto L_holder_found; // Found!
 3970   //     }
 3971   //   } while (tmp != 0);
 3972   //   goto L_no_such_interface // Not found.
 3973   Label L_scan_holder;
 3974   bind(L_scan_holder);
 3975     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3976     addptr(scan_temp, scan_step);
 3977     cmpptr(holder_klass, temp_itbl_klass);
 3978     jccb(Assembler::equal, L_holder_found);
 3979     testptr(temp_itbl_klass, temp_itbl_klass);
 3980     jccb(Assembler::notZero, L_scan_holder);
 3981 
 3982   jmpb(L_no_such_interface);
 3983 
 3984   // Loop: Look for resolved_class record in itable
 3985   //   do {
 3986   //     tmp = itable[index];
 3987   //     index += step;
 3988   //     if (tmp == holder_klass) {
 3989   //        // Also check if we have met a holder klass
 3990   //        holder_tmp = itable[index-step-ioffset];
 3991   //     }
 3992   //     if (tmp == resolved_klass) {
 3993   //        goto L_resolved_found;  // Found!
 3994   //     }
 3995   //   } while (tmp != 0);
 3996   //   goto L_no_such_interface // Not found.
 3997   //
 3998   Label L_loop_scan_resolved;
 3999   bind(L_loop_scan_resolved);
 4000     movptr(temp_itbl_klass, Address(scan_temp, 0));
 4001     addptr(scan_temp, scan_step);
 4002     bind(L_loop_scan_resolved_entry);
 4003     cmpptr(holder_klass, temp_itbl_klass);
 4004     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4005     cmpptr(resolved_klass, temp_itbl_klass);
 4006     jccb(Assembler::equal, L_resolved_found);
 4007     testptr(temp_itbl_klass, temp_itbl_klass);
 4008     jccb(Assembler::notZero, L_loop_scan_resolved);
 4009 
 4010   jmpb(L_no_such_interface);
 4011 
 4012   Label L_ready;
 4013 
 4014   // See if we already have a holder klass. If not, go and scan for it.
 4015   bind(L_resolved_found);
 4016   testptr(temp_reg, temp_reg);
 4017   jccb(Assembler::zero, L_scan_holder);
 4018   jmpb(L_ready);
 4019 
 4020   bind(L_holder_found);
 4021   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4022 
 4023   // Finally, temp_reg contains holder_klass vtable offset
 4024   bind(L_ready);
 4025   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 4026   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 4027     load_klass(scan_temp, receiver, noreg);
 4028     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4029   } else {
 4030     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4031   }
 4032 }
 4033 
 4034 
 4035 // virtual method calling
 4036 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 4037                                            RegisterOrConstant vtable_index,
 4038                                            Register method_result) {
 4039   const ByteSize base = Klass::vtable_start_offset();
 4040   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 4041   Address vtable_entry_addr(recv_klass,
 4042                             vtable_index, Address::times_ptr,
 4043                             base + vtableEntry::method_offset());
 4044   movptr(method_result, vtable_entry_addr);
 4045 }
 4046 
 4047 
 4048 void MacroAssembler::check_klass_subtype(Register sub_klass,
 4049                            Register super_klass,
 4050                            Register temp_reg,
 4051                            Label& L_success) {
 4052   Label L_failure;
 4053   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 4054   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 4055   bind(L_failure);
 4056 }
 4057 
 4058 
 4059 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 4060                                                    Register super_klass,
 4061                                                    Register temp_reg,
 4062                                                    Label* L_success,
 4063                                                    Label* L_failure,
 4064                                                    Label* L_slow_path,
 4065                                         RegisterOrConstant super_check_offset) {
 4066   assert_different_registers(sub_klass, super_klass, temp_reg);
 4067   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 4068   if (super_check_offset.is_register()) {
 4069     assert_different_registers(sub_klass, super_klass,
 4070                                super_check_offset.as_register());
 4071   } else if (must_load_sco) {
 4072     assert(temp_reg != noreg, "supply either a temp or a register offset");
 4073   }
 4074 
 4075   Label L_fallthrough;
 4076   int label_nulls = 0;
 4077   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4078   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4079   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 4080   assert(label_nulls <= 1, "at most one null in the batch");
 4081 
 4082   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4083   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 4084   Address super_check_offset_addr(super_klass, sco_offset);
 4085 
 4086   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 4087   // range of a jccb.  If this routine grows larger, reconsider at
 4088   // least some of these.
 4089 #define local_jcc(assembler_cond, label)                                \
 4090   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 4091   else                             jcc( assembler_cond, label) /*omit semi*/
 4092 
 4093   // Hacked jmp, which may only be used just before L_fallthrough.
 4094 #define final_jmp(label)                                                \
 4095   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 4096   else                            jmp(label)                /*omit semi*/
 4097 
 4098   // If the pointers are equal, we are done (e.g., String[] elements).
 4099   // This self-check enables sharing of secondary supertype arrays among
 4100   // non-primary types such as array-of-interface.  Otherwise, each such
 4101   // type would need its own customized SSA.
 4102   // We move this check to the front of the fast path because many
 4103   // type checks are in fact trivially successful in this manner,
 4104   // so we get a nicely predicted branch right at the start of the check.
 4105   cmpptr(sub_klass, super_klass);
 4106   local_jcc(Assembler::equal, *L_success);
 4107 
 4108   // Check the supertype display:
 4109   if (must_load_sco) {
 4110     // Positive movl does right thing on LP64.
 4111     movl(temp_reg, super_check_offset_addr);
 4112     super_check_offset = RegisterOrConstant(temp_reg);
 4113   }
 4114   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 4115   cmpptr(super_klass, super_check_addr); // load displayed supertype
 4116 
 4117   // This check has worked decisively for primary supers.
 4118   // Secondary supers are sought in the super_cache ('super_cache_addr').
 4119   // (Secondary supers are interfaces and very deeply nested subtypes.)
 4120   // This works in the same check above because of a tricky aliasing
 4121   // between the super_cache and the primary super display elements.
 4122   // (The 'super_check_addr' can address either, as the case requires.)
 4123   // Note that the cache is updated below if it does not help us find
 4124   // what we need immediately.
 4125   // So if it was a primary super, we can just fail immediately.
 4126   // Otherwise, it's the slow path for us (no success at this point).
 4127 
 4128   if (super_check_offset.is_register()) {
 4129     local_jcc(Assembler::equal, *L_success);
 4130     cmpl(super_check_offset.as_register(), sc_offset);
 4131     if (L_failure == &L_fallthrough) {
 4132       local_jcc(Assembler::equal, *L_slow_path);
 4133     } else {
 4134       local_jcc(Assembler::notEqual, *L_failure);
 4135       final_jmp(*L_slow_path);
 4136     }
 4137   } else if (super_check_offset.as_constant() == sc_offset) {
 4138     // Need a slow path; fast failure is impossible.
 4139     if (L_slow_path == &L_fallthrough) {
 4140       local_jcc(Assembler::equal, *L_success);
 4141     } else {
 4142       local_jcc(Assembler::notEqual, *L_slow_path);
 4143       final_jmp(*L_success);
 4144     }
 4145   } else {
 4146     // No slow path; it's a fast decision.
 4147     if (L_failure == &L_fallthrough) {
 4148       local_jcc(Assembler::equal, *L_success);
 4149     } else {
 4150       local_jcc(Assembler::notEqual, *L_failure);
 4151       final_jmp(*L_success);
 4152     }
 4153   }
 4154 
 4155   bind(L_fallthrough);
 4156 
 4157 #undef local_jcc
 4158 #undef final_jmp
 4159 }
 4160 
 4161 
 4162 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 4163                                                           Register super_klass,
 4164                                                           Register temp_reg,
 4165                                                           Register temp2_reg,
 4166                                                           Label* L_success,
 4167                                                           Label* L_failure,
 4168                                                           bool set_cond_codes) {
 4169   assert_different_registers(sub_klass, super_klass, temp_reg);
 4170   if (temp2_reg != noreg)
 4171     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4172 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4173 
 4174   Label L_fallthrough;
 4175   int label_nulls = 0;
 4176   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4177   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4178   assert(label_nulls <= 1, "at most one null in the batch");
 4179 
 4180   // a couple of useful fields in sub_klass:
 4181   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4182   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4183   Address secondary_supers_addr(sub_klass, ss_offset);
 4184   Address super_cache_addr(     sub_klass, sc_offset);
 4185 
 4186   // Do a linear scan of the secondary super-klass chain.
 4187   // This code is rarely used, so simplicity is a virtue here.
 4188   // The repne_scan instruction uses fixed registers, which we must spill.
 4189   // Don't worry too much about pre-existing connections with the input regs.
 4190 
 4191   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4192   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4193 
 4194   // Get super_klass value into rax (even if it was in rdi or rcx).
 4195   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4196   if (super_klass != rax) {
 4197     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4198     mov(rax, super_klass);
 4199   }
 4200   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4201   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4202 
 4203 #ifndef PRODUCT
 4204   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4205   ExternalAddress pst_counter_addr((address) pst_counter);
 4206   lea(rcx, pst_counter_addr);
 4207   incrementl(Address(rcx, 0));
 4208 #endif //PRODUCT
 4209 
 4210   // We will consult the secondary-super array.
 4211   movptr(rdi, secondary_supers_addr);
 4212   // Load the array length.  (Positive movl does right thing on LP64.)
 4213   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4214   // Skip to start of data.
 4215   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4216 
 4217   // Scan RCX words at [RDI] for an occurrence of RAX.
 4218   // Set NZ/Z based on last compare.
 4219   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4220   // not change flags (only scas instruction which is repeated sets flags).
 4221   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4222 
 4223     testptr(rax,rax); // Set Z = 0
 4224     repne_scan();
 4225 
 4226   // Unspill the temp. registers:
 4227   if (pushed_rdi)  pop(rdi);
 4228   if (pushed_rcx)  pop(rcx);
 4229   if (pushed_rax)  pop(rax);
 4230 
 4231   if (set_cond_codes) {
 4232     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4233     assert(!pushed_rdi, "rdi must be left non-null");
 4234     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4235   }
 4236 
 4237   if (L_failure == &L_fallthrough)
 4238         jccb(Assembler::notEqual, *L_failure);
 4239   else  jcc(Assembler::notEqual, *L_failure);
 4240 
 4241   // Success.  Cache the super we found and proceed in triumph.
 4242   movptr(super_cache_addr, super_klass);
 4243 
 4244   if (L_success != &L_fallthrough) {
 4245     jmp(*L_success);
 4246   }
 4247 
 4248 #undef IS_A_TEMP
 4249 
 4250   bind(L_fallthrough);
 4251 }
 4252 
 4253 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4254                                                    Register super_klass,
 4255                                                    Register temp_reg,
 4256                                                    Register temp2_reg,
 4257                                                    Label* L_success,
 4258                                                    Label* L_failure,
 4259                                                    bool set_cond_codes) {
 4260   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4261   check_klass_subtype_slow_path
 4262     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4263      L_success, L_failure);
 4264 }
 4265 
 4266 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4267                                                    Register super_klass,
 4268                                                    Register temp_reg,
 4269                                                    Register temp2_reg,
 4270                                                    Register temp3_reg,
 4271                                                    Register temp4_reg,
 4272                                                    Label* L_success,
 4273                                                    Label* L_failure) {
 4274   if (UseSecondarySupersTable) {
 4275     check_klass_subtype_slow_path_table
 4276       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4277        L_success, L_failure);
 4278   } else {
 4279     check_klass_subtype_slow_path_linear
 4280       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4281   }
 4282 }
 4283 
 4284 Register MacroAssembler::allocate_if_noreg(Register r,
 4285                                   RegSetIterator<Register> &available_regs,
 4286                                   RegSet &regs_to_push) {
 4287   if (!r->is_valid()) {
 4288     r = *available_regs++;
 4289     regs_to_push += r;
 4290   }
 4291   return r;
 4292 }
 4293 
 4294 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4295                                                          Register super_klass,
 4296                                                          Register temp_reg,
 4297                                                          Register temp2_reg,
 4298                                                          Register temp3_reg,
 4299                                                          Register result_reg,
 4300                                                          Label* L_success,
 4301                                                          Label* L_failure) {
 4302   // NB! Callers may assume that, when temp2_reg is a valid register,
 4303   // this code sets it to a nonzero value.
 4304   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4305 
 4306   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4307 
 4308   Label L_fallthrough;
 4309   int label_nulls = 0;
 4310   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4311   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4312   assert(label_nulls <= 1, "at most one null in the batch");
 4313 
 4314   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4315 
 4316   RegSetIterator<Register> available_regs
 4317     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4318 
 4319   RegSet pushed_regs;
 4320 
 4321   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4322   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4323   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4324   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4325   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4326 
 4327   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4328 
 4329   {
 4330 
 4331     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4332     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4333     subptr(rsp, aligned_size);
 4334     push_set(pushed_regs, 0);
 4335 
 4336     lookup_secondary_supers_table_var(sub_klass,
 4337                                       super_klass,
 4338                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4339     cmpq(result_reg, 0);
 4340 
 4341     // Unspill the temp. registers:
 4342     pop_set(pushed_regs, 0);
 4343     // Increment SP but do not clobber flags.
 4344     lea(rsp, Address(rsp, aligned_size));
 4345   }
 4346 
 4347   if (temp2_reg_was_valid) {
 4348     movq(temp2_reg, 1);
 4349   }
 4350 
 4351   jcc(Assembler::notEqual, *L_failure);
 4352 
 4353   if (L_success != &L_fallthrough) {
 4354     jmp(*L_success);
 4355   }
 4356 
 4357   bind(L_fallthrough);
 4358 }
 4359 
 4360 // population_count variant for running without the POPCNT
 4361 // instruction, which was introduced with SSE4.2 in 2008.
 4362 void MacroAssembler::population_count(Register dst, Register src,
 4363                                       Register scratch1, Register scratch2) {
 4364   assert_different_registers(src, scratch1, scratch2);
 4365   if (UsePopCountInstruction) {
 4366     Assembler::popcntq(dst, src);
 4367   } else {
 4368     assert_different_registers(src, scratch1, scratch2);
 4369     assert_different_registers(dst, scratch1, scratch2);
 4370     Label loop, done;
 4371 
 4372     mov(scratch1, src);
 4373     // dst = 0;
 4374     // while(scratch1 != 0) {
 4375     //   dst++;
 4376     //   scratch1 &= (scratch1 - 1);
 4377     // }
 4378     xorl(dst, dst);
 4379     testq(scratch1, scratch1);
 4380     jccb(Assembler::equal, done);
 4381     {
 4382       bind(loop);
 4383       incq(dst);
 4384       movq(scratch2, scratch1);
 4385       decq(scratch2);
 4386       andq(scratch1, scratch2);
 4387       jccb(Assembler::notEqual, loop);
 4388     }
 4389     bind(done);
 4390   }
 4391 #ifdef ASSERT
 4392   mov64(scratch1, 0xCafeBabeDeadBeef);
 4393   movq(scratch2, scratch1);
 4394 #endif
 4395 }
 4396 
 4397 // Ensure that the inline code and the stub are using the same registers.
 4398 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4399 do {                                                                 \
 4400   assert(r_super_klass  == rax, "mismatch");                         \
 4401   assert(r_array_base   == rbx, "mismatch");                         \
 4402   assert(r_array_length == rcx, "mismatch");                         \
 4403   assert(r_array_index  == rdx, "mismatch");                         \
 4404   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4405   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4406   assert(result         == rdi || result      == noreg, "mismatch"); \
 4407 } while(0)
 4408 
 4409 // Versions of salq and rorq that don't need count to be in rcx
 4410 
 4411 void MacroAssembler::salq(Register dest, Register count) {
 4412   if (count == rcx) {
 4413     Assembler::salq(dest);
 4414   } else {
 4415     assert_different_registers(rcx, dest);
 4416     xchgq(rcx, count);
 4417     Assembler::salq(dest);
 4418     xchgq(rcx, count);
 4419   }
 4420 }
 4421 
 4422 void MacroAssembler::rorq(Register dest, Register count) {
 4423   if (count == rcx) {
 4424     Assembler::rorq(dest);
 4425   } else {
 4426     assert_different_registers(rcx, dest);
 4427     xchgq(rcx, count);
 4428     Assembler::rorq(dest);
 4429     xchgq(rcx, count);
 4430   }
 4431 }
 4432 
 4433 // Return true: we succeeded in generating this code
 4434 //
 4435 // At runtime, return 0 in result if r_super_klass is a superclass of
 4436 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4437 // super_klass_slot of the class you're looking for. This is always
 4438 // the case for instanceof and checkcast.
 4439 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4440                                                          Register r_super_klass,
 4441                                                          Register temp1,
 4442                                                          Register temp2,
 4443                                                          Register temp3,
 4444                                                          Register temp4,
 4445                                                          Register result,
 4446                                                          u1 super_klass_slot) {
 4447   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4448 
 4449   Label L_fallthrough, L_success, L_failure;
 4450 
 4451   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4452 
 4453   const Register
 4454     r_array_index  = temp1,
 4455     r_array_length = temp2,
 4456     r_array_base   = temp3,
 4457     r_bitmap       = temp4;
 4458 
 4459   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4460 
 4461   xorq(result, result); // = 0
 4462 
 4463   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4464   movq(r_array_index, r_bitmap);
 4465 
 4466   // First check the bitmap to see if super_klass might be present. If
 4467   // the bit is zero, we are certain that super_klass is not one of
 4468   // the secondary supers.
 4469   u1 bit = super_klass_slot;
 4470   {
 4471     // NB: If the count in a x86 shift instruction is 0, the flags are
 4472     // not affected, so we do a testq instead.
 4473     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4474     if (shift_count != 0) {
 4475       salq(r_array_index, shift_count);
 4476     } else {
 4477       testq(r_array_index, r_array_index);
 4478     }
 4479   }
 4480   // We test the MSB of r_array_index, i.e. its sign bit
 4481   jcc(Assembler::positive, L_failure);
 4482 
 4483   // Get the first array index that can contain super_klass into r_array_index.
 4484   if (bit != 0) {
 4485     population_count(r_array_index, r_array_index, temp2, temp3);
 4486   } else {
 4487     movl(r_array_index, 1);
 4488   }
 4489   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4490 
 4491   // We will consult the secondary-super array.
 4492   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4493 
 4494   // We're asserting that the first word in an Array<Klass*> is the
 4495   // length, and the second word is the first word of the data. If
 4496   // that ever changes, r_array_base will have to be adjusted here.
 4497   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4498   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4499 
 4500   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4501   jccb(Assembler::equal, L_success);
 4502 
 4503   // Is there another entry to check? Consult the bitmap.
 4504   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4505   jccb(Assembler::carryClear, L_failure);
 4506 
 4507   // Linear probe. Rotate the bitmap so that the next bit to test is
 4508   // in Bit 1.
 4509   if (bit != 0) {
 4510     rorq(r_bitmap, bit);
 4511   }
 4512 
 4513   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4514   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4515   // Kills: r_array_length.
 4516   // Returns: result.
 4517   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4518   // Result (0/1) is in rdi
 4519   jmpb(L_fallthrough);
 4520 
 4521   bind(L_failure);
 4522   incq(result); // 0 => 1
 4523 
 4524   bind(L_success);
 4525   // result = 0;
 4526 
 4527   bind(L_fallthrough);
 4528   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4529 
 4530   if (VerifySecondarySupers) {
 4531     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4532                                   temp1, temp2, temp3);
 4533   }
 4534 }
 4535 
 4536 // At runtime, return 0 in result if r_super_klass is a superclass of
 4537 // r_sub_klass, otherwise return nonzero. Use this version of
 4538 // lookup_secondary_supers_table() if you don't know ahead of time
 4539 // which superclass will be searched for. Used by interpreter and
 4540 // runtime stubs. It is larger and has somewhat greater latency than
 4541 // the version above, which takes a constant super_klass_slot.
 4542 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4543                                                        Register r_super_klass,
 4544                                                        Register temp1,
 4545                                                        Register temp2,
 4546                                                        Register temp3,
 4547                                                        Register temp4,
 4548                                                        Register result) {
 4549   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4550   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4551   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4552 
 4553   Label L_fallthrough, L_success, L_failure;
 4554 
 4555   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4556 
 4557   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4558 
 4559   // FIXME. Once we are sure that all paths reaching this point really
 4560   // do pass rcx as one of our temps we can get rid of the following
 4561   // workaround.
 4562   assert(temps.contains(rcx), "fix this code");
 4563 
 4564   // We prefer to have our shift count in rcx. If rcx is one of our
 4565   // temps, use it for slot. If not, pick any of our temps.
 4566   Register slot;
 4567   if (!temps.contains(rcx)) {
 4568     slot = *available_regs++;
 4569   } else {
 4570     slot = rcx;
 4571   }
 4572 
 4573   const Register r_array_index = *available_regs++;
 4574   const Register r_bitmap      = *available_regs++;
 4575 
 4576   // The logic above guarantees this property, but we state it here.
 4577   assert_different_registers(r_array_index, r_bitmap, rcx);
 4578 
 4579   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4580   movq(r_array_index, r_bitmap);
 4581 
 4582   // First check the bitmap to see if super_klass might be present. If
 4583   // the bit is zero, we are certain that super_klass is not one of
 4584   // the secondary supers.
 4585   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4586   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4587   salq(r_array_index, slot);
 4588 
 4589   testq(r_array_index, r_array_index);
 4590   // We test the MSB of r_array_index, i.e. its sign bit
 4591   jcc(Assembler::positive, L_failure);
 4592 
 4593   const Register r_array_base = *available_regs++;
 4594 
 4595   // Get the first array index that can contain super_klass into r_array_index.
 4596   // Note: Clobbers r_array_base and slot.
 4597   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4598 
 4599   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4600 
 4601   // We will consult the secondary-super array.
 4602   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4603 
 4604   // We're asserting that the first word in an Array<Klass*> is the
 4605   // length, and the second word is the first word of the data. If
 4606   // that ever changes, r_array_base will have to be adjusted here.
 4607   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4608   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4609 
 4610   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4611   jccb(Assembler::equal, L_success);
 4612 
 4613   // Restore slot to its true value
 4614   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4615 
 4616   // Linear probe. Rotate the bitmap so that the next bit to test is
 4617   // in Bit 1.
 4618   rorq(r_bitmap, slot);
 4619 
 4620   // Is there another entry to check? Consult the bitmap.
 4621   btq(r_bitmap, 1);
 4622   jccb(Assembler::carryClear, L_failure);
 4623 
 4624   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4625   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4626   // Kills: r_array_length.
 4627   // Returns: result.
 4628   lookup_secondary_supers_table_slow_path(r_super_klass,
 4629                                           r_array_base,
 4630                                           r_array_index,
 4631                                           r_bitmap,
 4632                                           /*temp1*/result,
 4633                                           /*temp2*/slot,
 4634                                           &L_success,
 4635                                           nullptr);
 4636 
 4637   bind(L_failure);
 4638   movq(result, 1);
 4639   jmpb(L_fallthrough);
 4640 
 4641   bind(L_success);
 4642   xorq(result, result); // = 0
 4643 
 4644   bind(L_fallthrough);
 4645   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4646 
 4647   if (VerifySecondarySupers) {
 4648     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4649                                   temp1, temp2, temp3);
 4650   }
 4651 }
 4652 
 4653 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4654                                  Label* L_success, Label* L_failure) {
 4655   Label L_loop, L_fallthrough;
 4656   {
 4657     int label_nulls = 0;
 4658     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4659     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4660     assert(label_nulls <= 1, "at most one null in the batch");
 4661   }
 4662   bind(L_loop);
 4663   cmpq(value, Address(addr, count, Address::times_8));
 4664   jcc(Assembler::equal, *L_success);
 4665   addl(count, 1);
 4666   cmpl(count, limit);
 4667   jcc(Assembler::less, L_loop);
 4668 
 4669   if (&L_fallthrough != L_failure) {
 4670     jmp(*L_failure);
 4671   }
 4672   bind(L_fallthrough);
 4673 }
 4674 
 4675 // Called by code generated by check_klass_subtype_slow_path
 4676 // above. This is called when there is a collision in the hashed
 4677 // lookup in the secondary supers array.
 4678 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4679                                                              Register r_array_base,
 4680                                                              Register r_array_index,
 4681                                                              Register r_bitmap,
 4682                                                              Register temp1,
 4683                                                              Register temp2,
 4684                                                              Label* L_success,
 4685                                                              Label* L_failure) {
 4686   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4687 
 4688   const Register
 4689     r_array_length = temp1,
 4690     r_sub_klass    = noreg,
 4691     result         = noreg;
 4692 
 4693   Label L_fallthrough;
 4694   int label_nulls = 0;
 4695   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4696   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4697   assert(label_nulls <= 1, "at most one null in the batch");
 4698 
 4699   // Load the array length.
 4700   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4701   // And adjust the array base to point to the data.
 4702   // NB! Effectively increments current slot index by 1.
 4703   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4704   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4705 
 4706   // Linear probe
 4707   Label L_huge;
 4708 
 4709   // The bitmap is full to bursting.
 4710   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4711   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4712   jcc(Assembler::greater, L_huge);
 4713 
 4714   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4715   // current slot (at secondary_supers[r_array_index]) has not yet
 4716   // been inspected, and r_array_index may be out of bounds if we
 4717   // wrapped around the end of the array.
 4718 
 4719   { // This is conventional linear probing, but instead of terminating
 4720     // when a null entry is found in the table, we maintain a bitmap
 4721     // in which a 0 indicates missing entries.
 4722     // The check above guarantees there are 0s in the bitmap, so the loop
 4723     // eventually terminates.
 4724 
 4725     xorl(temp2, temp2); // = 0;
 4726 
 4727     Label L_again;
 4728     bind(L_again);
 4729 
 4730     // Check for array wraparound.
 4731     cmpl(r_array_index, r_array_length);
 4732     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4733 
 4734     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4735     jcc(Assembler::equal, *L_success);
 4736 
 4737     // If the next bit in bitmap is zero, we're done.
 4738     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4739     jcc(Assembler::carryClear, *L_failure);
 4740 
 4741     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4742     addl(r_array_index, 1);
 4743 
 4744     jmp(L_again);
 4745   }
 4746 
 4747   { // Degenerate case: more than 64 secondary supers.
 4748     // FIXME: We could do something smarter here, maybe a vectorized
 4749     // comparison or a binary search, but is that worth any added
 4750     // complexity?
 4751     bind(L_huge);
 4752     xorl(r_array_index, r_array_index); // = 0
 4753     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4754                 L_success,
 4755                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4756 
 4757     bind(L_fallthrough);
 4758   }
 4759 }
 4760 
 4761 struct VerifyHelperArguments {
 4762   Klass* _super;
 4763   Klass* _sub;
 4764   intptr_t _linear_result;
 4765   intptr_t _table_result;
 4766 };
 4767 
 4768 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4769   Klass::on_secondary_supers_verification_failure(args->_super,
 4770                                                   args->_sub,
 4771                                                   args->_linear_result,
 4772                                                   args->_table_result,
 4773                                                   msg);
 4774 }
 4775 
 4776 // Make sure that the hashed lookup and a linear scan agree.
 4777 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4778                                                    Register r_super_klass,
 4779                                                    Register result,
 4780                                                    Register temp1,
 4781                                                    Register temp2,
 4782                                                    Register temp3) {
 4783   const Register
 4784       r_array_index  = temp1,
 4785       r_array_length = temp2,
 4786       r_array_base   = temp3,
 4787       r_bitmap       = noreg;
 4788 
 4789   BLOCK_COMMENT("verify_secondary_supers_table {");
 4790 
 4791   Label L_success, L_failure, L_check, L_done;
 4792 
 4793   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4794   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4795   // And adjust the array base to point to the data.
 4796   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4797 
 4798   testl(r_array_length, r_array_length); // array_length == 0?
 4799   jcc(Assembler::zero, L_failure);
 4800 
 4801   movl(r_array_index, 0);
 4802   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4803   // fall through to L_failure
 4804 
 4805   const Register linear_result = r_array_index; // reuse temp1
 4806 
 4807   bind(L_failure); // not present
 4808   movl(linear_result, 1);
 4809   jmp(L_check);
 4810 
 4811   bind(L_success); // present
 4812   movl(linear_result, 0);
 4813 
 4814   bind(L_check);
 4815   cmpl(linear_result, result);
 4816   jcc(Assembler::equal, L_done);
 4817 
 4818   { // To avoid calling convention issues, build a record on the stack
 4819     // and pass the pointer to that instead.
 4820     push(result);
 4821     push(linear_result);
 4822     push(r_sub_klass);
 4823     push(r_super_klass);
 4824     movptr(c_rarg1, rsp);
 4825     movptr(c_rarg0, (uintptr_t) "mismatch");
 4826     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4827     should_not_reach_here();
 4828   }
 4829   bind(L_done);
 4830 
 4831   BLOCK_COMMENT("} verify_secondary_supers_table");
 4832 }
 4833 
 4834 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4835 
 4836 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4837   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4838 
 4839   Label L_fallthrough;
 4840   if (L_fast_path == nullptr) {
 4841     L_fast_path = &L_fallthrough;
 4842   } else if (L_slow_path == nullptr) {
 4843     L_slow_path = &L_fallthrough;
 4844   }
 4845 
 4846   // Fast path check: class is fully initialized.
 4847   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4848   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4849   jcc(Assembler::equal, *L_fast_path);
 4850 
 4851   // Fast path check: current thread is initializer thread
 4852   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4853   if (L_slow_path == &L_fallthrough) {
 4854     jcc(Assembler::equal, *L_fast_path);
 4855     bind(*L_slow_path);
 4856   } else if (L_fast_path == &L_fallthrough) {
 4857     jcc(Assembler::notEqual, *L_slow_path);
 4858     bind(*L_fast_path);
 4859   } else {
 4860     Unimplemented();
 4861   }
 4862 }
 4863 
 4864 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4865   if (VM_Version::supports_cmov()) {
 4866     cmovl(cc, dst, src);
 4867   } else {
 4868     Label L;
 4869     jccb(negate_condition(cc), L);
 4870     movl(dst, src);
 4871     bind(L);
 4872   }
 4873 }
 4874 
 4875 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4876   if (VM_Version::supports_cmov()) {
 4877     cmovl(cc, dst, src);
 4878   } else {
 4879     Label L;
 4880     jccb(negate_condition(cc), L);
 4881     movl(dst, src);
 4882     bind(L);
 4883   }
 4884 }
 4885 
 4886 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4887   if (!VerifyOops || VerifyAdapterSharing) {
 4888     // Below address of the code string confuses VerifyAdapterSharing
 4889     // because it may differ between otherwise equivalent adapters.
 4890     return;
 4891   }
 4892 
 4893   BLOCK_COMMENT("verify_oop {");
 4894   push(rscratch1);
 4895   push(rax);                          // save rax
 4896   push(reg);                          // pass register argument
 4897 
 4898   // Pass register number to verify_oop_subroutine
 4899   const char* b = nullptr;
 4900   {
 4901     ResourceMark rm;
 4902     stringStream ss;
 4903     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4904     b = code_string(ss.as_string());
 4905   }
 4906   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4907   pushptr(buffer.addr(), rscratch1);
 4908 
 4909   // call indirectly to solve generation ordering problem
 4910   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4911   call(rax);
 4912   // Caller pops the arguments (oop, message) and restores rax, r10
 4913   BLOCK_COMMENT("} verify_oop");
 4914 }
 4915 
 4916 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 4917   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 4918     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 4919     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 4920     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 4921   } else if (VM_Version::supports_avx()) {
 4922     vpcmpeqd(dst, dst, dst, vector_len);
 4923   } else {
 4924     pcmpeqd(dst, dst);
 4925   }
 4926 }
 4927 
 4928 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 4929                                          int extra_slot_offset) {
 4930   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 4931   int stackElementSize = Interpreter::stackElementSize;
 4932   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 4933 #ifdef ASSERT
 4934   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 4935   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 4936 #endif
 4937   Register             scale_reg    = noreg;
 4938   Address::ScaleFactor scale_factor = Address::no_scale;
 4939   if (arg_slot.is_constant()) {
 4940     offset += arg_slot.as_constant() * stackElementSize;
 4941   } else {
 4942     scale_reg    = arg_slot.as_register();
 4943     scale_factor = Address::times(stackElementSize);
 4944   }
 4945   offset += wordSize;           // return PC is on stack
 4946   return Address(rsp, scale_reg, scale_factor, offset);
 4947 }
 4948 
 4949 // Handle the receiver type profile update given the "recv" klass.
 4950 //
 4951 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 4952 // If there are no matching or claimable receiver entries in RD, updates
 4953 // the polymorphic counter.
 4954 //
 4955 // This code expected to run by either the interpreter or JIT-ed code, without
 4956 // extra synchronization. For safety, receiver cells are claimed atomically, which
 4957 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 4958 // counter updates are not atomic.
 4959 //
 4960 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 4961   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 4962   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 4963   int poly_count_offset      = in_bytes(CounterData::count_offset());
 4964   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 4965   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 4966 
 4967   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 4968   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 4969   base_receiver_offset += mdp_offset;
 4970   end_receiver_offset  += mdp_offset;
 4971   poly_count_offset    += mdp_offset;
 4972 
 4973   // Scale down to optimize encoding. Slots are pointer-sized.
 4974   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 4975   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 4976   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 4977   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 4978   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 4979   base_receiver_offset   >>= LogBytesPerWord;
 4980   end_receiver_offset    >>= LogBytesPerWord;
 4981   poly_count_offset      >>= LogBytesPerWord;
 4982   receiver_step          >>= LogBytesPerWord;
 4983   receiver_to_count_step >>= LogBytesPerWord;
 4984 
 4985 #ifdef ASSERT
 4986   // We are about to walk the MDO slots without asking for offsets.
 4987   // Check that our math hits all the right spots.
 4988   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 4989     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 4990     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 4991     int offset = base_receiver_offset + receiver_step*c;
 4992     int count_offset = offset + receiver_to_count_step;
 4993     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 4994     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 4995   }
 4996   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 4997   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 4998 #endif
 4999 
 5000   // Corner case: no profile table. Increment poly counter and exit.
 5001   if (ReceiverTypeData::row_limit() == 0) {
 5002     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 5003     return;
 5004   }
 5005 
 5006   Register offset = rscratch1;
 5007 
 5008   Label L_loop_search_receiver, L_loop_search_empty;
 5009   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 5010 
 5011   // The code here recognizes three major cases:
 5012   //   A. Fastest: receiver found in the table
 5013   //   B. Fast: no receiver in the table, and the table is full
 5014   //   C. Slow: no receiver in the table, free slots in the table
 5015   //
 5016   // The case A performance is most important, as perfectly-behaved code would end up
 5017   // there, especially with larger TypeProfileWidth. The case B performance is
 5018   // important as well, this is where bulk of code would land for normally megamorphic
 5019   // cases. The case C performance is not essential, its job is to deal with installation
 5020   // races, we optimize for code density instead. Case C needs to make sure that receiver
 5021   // rows are only claimed once. This makes sure we never overwrite a row for another
 5022   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 5023   //
 5024   // It is very tempting to handle these cases in a single loop, and claim the first slot
 5025   // without checking the rest of the table. But, profiling code should tolerate free slots
 5026   // in the table, as class unloading can clear them. After such cleanup, the receiver
 5027   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 5028   // to complete, before trying to install new slots. Splitting the code in several tight
 5029   // loops also helpfully optimizes for cases A and B.
 5030   //
 5031   // This code is effectively:
 5032   //
 5033   // restart:
 5034   //   // Fastest: receiver is already installed
 5035   //   for (i = 0; i < receiver_count(); i++) {
 5036   //     if (receiver(i) == recv) goto found_recv(i);
 5037   //   }
 5038   //
 5039   //   // Fast: no receiver, but profile is full
 5040   //   for (i = 0; i < receiver_count(); i++) {
 5041   //     if (receiver(i) == null) goto found_null(i);
 5042   //   }
 5043   //   goto polymorphic
 5044   //
 5045   //   // Slow: try to install receiver
 5046   // found_null(i):
 5047   //   CAS(&receiver(i), null, recv);
 5048   //   goto restart
 5049   //
 5050   // polymorphic:
 5051   //   count++;
 5052   //   return
 5053   //
 5054   // found_recv(i):
 5055   //   *receiver_count(i)++
 5056   //
 5057 
 5058   bind(L_restart);
 5059 
 5060   // Fastest: receiver is already installed
 5061   movptr(offset, base_receiver_offset);
 5062   bind(L_loop_search_receiver);
 5063     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 5064     jccb(Assembler::equal, L_found_recv);
 5065   addptr(offset, receiver_step);
 5066   cmpptr(offset, end_receiver_offset);
 5067   jccb(Assembler::notEqual, L_loop_search_receiver);
 5068 
 5069   // Fast: no receiver, but profile is full
 5070   movptr(offset, base_receiver_offset);
 5071   bind(L_loop_search_empty);
 5072     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 5073     jccb(Assembler::equal, L_found_empty);
 5074   addptr(offset, receiver_step);
 5075   cmpptr(offset, end_receiver_offset);
 5076   jccb(Assembler::notEqual, L_loop_search_empty);
 5077   jmpb(L_polymorphic);
 5078 
 5079   // Slow: try to install receiver
 5080   bind(L_found_empty);
 5081 
 5082   // Atomically swing receiver slot: null -> recv.
 5083   //
 5084   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 5085   // other important registers untouched, as they form the address. Therefore, we need
 5086   // to shift any important registers from RAX into some other spare register. If we
 5087   // have a spare register, we are forced to save it on stack here.
 5088 
 5089   Register spare_reg = noreg;
 5090   Register shifted_mdp = mdp;
 5091   Register shifted_recv = recv;
 5092   if (recv == rax || mdp == rax) {
 5093     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 5094                 (recv != rcx && mdp != rcx) ? rcx :
 5095                 rdx;
 5096     assert_different_registers(mdp, recv, offset, spare_reg);
 5097 
 5098     push(spare_reg);
 5099     if (recv == rax) {
 5100       movptr(spare_reg, recv);
 5101       shifted_recv = spare_reg;
 5102     } else {
 5103       assert(mdp == rax, "Remaining case");
 5104       movptr(spare_reg, mdp);
 5105       shifted_mdp = spare_reg;
 5106     }
 5107   } else {
 5108     push(rax);
 5109   }
 5110 
 5111   // None of the important registers are in RAX after this shuffle.
 5112   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 5113 
 5114   xorptr(rax, rax);
 5115   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 5116 
 5117   // Unshift registers.
 5118   if (recv == rax || mdp == rax) {
 5119     movptr(rax, spare_reg);
 5120     pop(spare_reg);
 5121   } else {
 5122     pop(rax);
 5123   }
 5124 
 5125   // CAS success means the slot now has the receiver we want. CAS failure means
 5126   // something had claimed the slot concurrently: it can be the same receiver we want,
 5127   // or something else. Since this is a slow path, we can optimize for code density,
 5128   // and just restart the search from the beginning.
 5129   jmpb(L_restart);
 5130 
 5131   // Counter updates:
 5132 
 5133   // Increment polymorphic counter instead of receiver slot.
 5134   bind(L_polymorphic);
 5135   movptr(offset, poly_count_offset);
 5136   jmpb(L_count_update);
 5137 
 5138   // Found a receiver, convert its slot offset to corresponding count offset.
 5139   bind(L_found_recv);
 5140   addptr(offset, receiver_to_count_step);
 5141 
 5142   bind(L_count_update);
 5143   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 5144 }
 5145 
 5146 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5147   if (!VerifyOops || VerifyAdapterSharing) {
 5148     // Below address of the code string confuses VerifyAdapterSharing
 5149     // because it may differ between otherwise equivalent adapters.
 5150     return;
 5151   }
 5152 
 5153   push(rscratch1);
 5154   push(rax); // save rax,
 5155   // addr may contain rsp so we will have to adjust it based on the push
 5156   // we just did (and on 64 bit we do two pushes)
 5157   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5158   // stores rax into addr which is backwards of what was intended.
 5159   if (addr.uses(rsp)) {
 5160     lea(rax, addr);
 5161     pushptr(Address(rax, 2 * BytesPerWord));
 5162   } else {
 5163     pushptr(addr);
 5164   }
 5165 
 5166   // Pass register number to verify_oop_subroutine
 5167   const char* b = nullptr;
 5168   {
 5169     ResourceMark rm;
 5170     stringStream ss;
 5171     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 5172     b = code_string(ss.as_string());
 5173   }
 5174   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5175   pushptr(buffer.addr(), rscratch1);
 5176 
 5177   // call indirectly to solve generation ordering problem
 5178   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5179   call(rax);
 5180   // Caller pops the arguments (addr, message) and restores rax, r10.
 5181 }
 5182 
 5183 void MacroAssembler::verify_tlab() {
 5184 #ifdef ASSERT
 5185   if (UseTLAB && VerifyOops) {
 5186     Label next, ok;
 5187     Register t1 = rsi;
 5188 
 5189     push(t1);
 5190 
 5191     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5192     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 5193     jcc(Assembler::aboveEqual, next);
 5194     STOP("assert(top >= start)");
 5195     should_not_reach_here();
 5196 
 5197     bind(next);
 5198     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 5199     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5200     jcc(Assembler::aboveEqual, ok);
 5201     STOP("assert(top <= end)");
 5202     should_not_reach_here();
 5203 
 5204     bind(ok);
 5205     pop(t1);
 5206   }
 5207 #endif
 5208 }
 5209 
 5210 class ControlWord {
 5211  public:
 5212   int32_t _value;
 5213 
 5214   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5215   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5216   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5217   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5218   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5219   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5220   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5221   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5222 
 5223   void print() const {
 5224     // rounding control
 5225     const char* rc;
 5226     switch (rounding_control()) {
 5227       case 0: rc = "round near"; break;
 5228       case 1: rc = "round down"; break;
 5229       case 2: rc = "round up  "; break;
 5230       case 3: rc = "chop      "; break;
 5231       default:
 5232         rc = nullptr; // silence compiler warnings
 5233         fatal("Unknown rounding control: %d", rounding_control());
 5234     };
 5235     // precision control
 5236     const char* pc;
 5237     switch (precision_control()) {
 5238       case 0: pc = "24 bits "; break;
 5239       case 1: pc = "reserved"; break;
 5240       case 2: pc = "53 bits "; break;
 5241       case 3: pc = "64 bits "; break;
 5242       default:
 5243         pc = nullptr; // silence compiler warnings
 5244         fatal("Unknown precision control: %d", precision_control());
 5245     };
 5246     // flags
 5247     char f[9];
 5248     f[0] = ' ';
 5249     f[1] = ' ';
 5250     f[2] = (precision   ()) ? 'P' : 'p';
 5251     f[3] = (underflow   ()) ? 'U' : 'u';
 5252     f[4] = (overflow    ()) ? 'O' : 'o';
 5253     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5254     f[6] = (denormalized()) ? 'D' : 'd';
 5255     f[7] = (invalid     ()) ? 'I' : 'i';
 5256     f[8] = '\x0';
 5257     // output
 5258     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5259   }
 5260 
 5261 };
 5262 
 5263 class StatusWord {
 5264  public:
 5265   int32_t _value;
 5266 
 5267   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5268   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5269   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5270   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5271   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5272   int  top() const                     { return  (_value >> 11) & 7      ; }
 5273   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5274   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5275   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5276   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5277   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5278   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5279   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5280   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5281 
 5282   void print() const {
 5283     // condition codes
 5284     char c[5];
 5285     c[0] = (C3()) ? '3' : '-';
 5286     c[1] = (C2()) ? '2' : '-';
 5287     c[2] = (C1()) ? '1' : '-';
 5288     c[3] = (C0()) ? '0' : '-';
 5289     c[4] = '\x0';
 5290     // flags
 5291     char f[9];
 5292     f[0] = (error_status()) ? 'E' : '-';
 5293     f[1] = (stack_fault ()) ? 'S' : '-';
 5294     f[2] = (precision   ()) ? 'P' : '-';
 5295     f[3] = (underflow   ()) ? 'U' : '-';
 5296     f[4] = (overflow    ()) ? 'O' : '-';
 5297     f[5] = (zero_divide ()) ? 'Z' : '-';
 5298     f[6] = (denormalized()) ? 'D' : '-';
 5299     f[7] = (invalid     ()) ? 'I' : '-';
 5300     f[8] = '\x0';
 5301     // output
 5302     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5303   }
 5304 
 5305 };
 5306 
 5307 class TagWord {
 5308  public:
 5309   int32_t _value;
 5310 
 5311   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5312 
 5313   void print() const {
 5314     printf("%04x", _value & 0xFFFF);
 5315   }
 5316 
 5317 };
 5318 
 5319 class FPU_Register {
 5320  public:
 5321   int32_t _m0;
 5322   int32_t _m1;
 5323   int16_t _ex;
 5324 
 5325   bool is_indefinite() const           {
 5326     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5327   }
 5328 
 5329   void print() const {
 5330     char  sign = (_ex < 0) ? '-' : '+';
 5331     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5332     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5333   };
 5334 
 5335 };
 5336 
 5337 class FPU_State {
 5338  public:
 5339   enum {
 5340     register_size       = 10,
 5341     number_of_registers =  8,
 5342     register_mask       =  7
 5343   };
 5344 
 5345   ControlWord  _control_word;
 5346   StatusWord   _status_word;
 5347   TagWord      _tag_word;
 5348   int32_t      _error_offset;
 5349   int32_t      _error_selector;
 5350   int32_t      _data_offset;
 5351   int32_t      _data_selector;
 5352   int8_t       _register[register_size * number_of_registers];
 5353 
 5354   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5355   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5356 
 5357   const char* tag_as_string(int tag) const {
 5358     switch (tag) {
 5359       case 0: return "valid";
 5360       case 1: return "zero";
 5361       case 2: return "special";
 5362       case 3: return "empty";
 5363     }
 5364     ShouldNotReachHere();
 5365     return nullptr;
 5366   }
 5367 
 5368   void print() const {
 5369     // print computation registers
 5370     { int t = _status_word.top();
 5371       for (int i = 0; i < number_of_registers; i++) {
 5372         int j = (i - t) & register_mask;
 5373         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5374         st(j)->print();
 5375         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5376       }
 5377     }
 5378     printf("\n");
 5379     // print control registers
 5380     printf("ctrl = "); _control_word.print(); printf("\n");
 5381     printf("stat = "); _status_word .print(); printf("\n");
 5382     printf("tags = "); _tag_word    .print(); printf("\n");
 5383   }
 5384 
 5385 };
 5386 
 5387 class Flag_Register {
 5388  public:
 5389   int32_t _value;
 5390 
 5391   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5392   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5393   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5394   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5395   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5396   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5397   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5398 
 5399   void print() const {
 5400     // flags
 5401     char f[8];
 5402     f[0] = (overflow       ()) ? 'O' : '-';
 5403     f[1] = (direction      ()) ? 'D' : '-';
 5404     f[2] = (sign           ()) ? 'S' : '-';
 5405     f[3] = (zero           ()) ? 'Z' : '-';
 5406     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5407     f[5] = (parity         ()) ? 'P' : '-';
 5408     f[6] = (carry          ()) ? 'C' : '-';
 5409     f[7] = '\x0';
 5410     // output
 5411     printf("%08x  flags = %s", _value, f);
 5412   }
 5413 
 5414 };
 5415 
 5416 class IU_Register {
 5417  public:
 5418   int32_t _value;
 5419 
 5420   void print() const {
 5421     printf("%08x  %11d", _value, _value);
 5422   }
 5423 
 5424 };
 5425 
 5426 class IU_State {
 5427  public:
 5428   Flag_Register _eflags;
 5429   IU_Register   _rdi;
 5430   IU_Register   _rsi;
 5431   IU_Register   _rbp;
 5432   IU_Register   _rsp;
 5433   IU_Register   _rbx;
 5434   IU_Register   _rdx;
 5435   IU_Register   _rcx;
 5436   IU_Register   _rax;
 5437 
 5438   void print() const {
 5439     // computation registers
 5440     printf("rax,  = "); _rax.print(); printf("\n");
 5441     printf("rbx,  = "); _rbx.print(); printf("\n");
 5442     printf("rcx  = "); _rcx.print(); printf("\n");
 5443     printf("rdx  = "); _rdx.print(); printf("\n");
 5444     printf("rdi  = "); _rdi.print(); printf("\n");
 5445     printf("rsi  = "); _rsi.print(); printf("\n");
 5446     printf("rbp,  = "); _rbp.print(); printf("\n");
 5447     printf("rsp  = "); _rsp.print(); printf("\n");
 5448     printf("\n");
 5449     // control registers
 5450     printf("flgs = "); _eflags.print(); printf("\n");
 5451   }
 5452 };
 5453 
 5454 
 5455 class CPU_State {
 5456  public:
 5457   FPU_State _fpu_state;
 5458   IU_State  _iu_state;
 5459 
 5460   void print() const {
 5461     printf("--------------------------------------------------\n");
 5462     _iu_state .print();
 5463     printf("\n");
 5464     _fpu_state.print();
 5465     printf("--------------------------------------------------\n");
 5466   }
 5467 
 5468 };
 5469 
 5470 
 5471 static void _print_CPU_state(CPU_State* state) {
 5472   state->print();
 5473 };
 5474 
 5475 
 5476 void MacroAssembler::print_CPU_state() {
 5477   push_CPU_state();
 5478   push(rsp);                // pass CPU state
 5479   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5480   addptr(rsp, wordSize);       // discard argument
 5481   pop_CPU_state();
 5482 }
 5483 
 5484 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5485   // Either restore the MXCSR register after returning from the JNI Call
 5486   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5487   if (RestoreMXCSROnJNICalls) {
 5488     ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5489   } else if (CheckJNICalls) {
 5490     call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5491   }
 5492   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5493   vzeroupper();
 5494 }
 5495 
 5496 // ((OopHandle)result).resolve();
 5497 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5498   assert_different_registers(result, tmp);
 5499 
 5500   // Only 64 bit platforms support GCs that require a tmp register
 5501   // Only IN_HEAP loads require a thread_tmp register
 5502   // OopHandle::resolve is an indirection like jobject.
 5503   access_load_at(T_OBJECT, IN_NATIVE,
 5504                  result, Address(result, 0), tmp);
 5505 }
 5506 
 5507 // ((WeakHandle)result).resolve();
 5508 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5509   assert_different_registers(rresult, rtmp);
 5510   Label resolved;
 5511 
 5512   // A null weak handle resolves to null.
 5513   cmpptr(rresult, 0);
 5514   jcc(Assembler::equal, resolved);
 5515 
 5516   // Only 64 bit platforms support GCs that require a tmp register
 5517   // Only IN_HEAP loads require a thread_tmp register
 5518   // WeakHandle::resolve is an indirection like jweak.
 5519   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5520                  rresult, Address(rresult, 0), rtmp);
 5521   bind(resolved);
 5522 }
 5523 
 5524 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5525   // get mirror
 5526   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5527   load_method_holder(mirror, method);
 5528   movptr(mirror, Address(mirror, mirror_offset));
 5529   resolve_oop_handle(mirror, tmp);
 5530 }
 5531 
 5532 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5533   load_method_holder(rresult, rmethod);
 5534   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5535 }
 5536 
 5537 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5538   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5539   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5540   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5541 }
 5542 
 5543 void MacroAssembler::load_metadata(Register dst, Register src) {
 5544   if (UseCompactObjectHeaders) {
 5545     load_narrow_klass_compact(dst, src);
 5546   } else {
 5547     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5548   }
 5549 }
 5550 
 5551 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5552   assert(UseCompactObjectHeaders, "expect compact object headers");
 5553   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5554   shrq(dst, markWord::klass_shift);
 5555 }
 5556 
 5557 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5558   assert_different_registers(src, tmp);
 5559   assert_different_registers(dst, tmp);
 5560 
 5561   if (UseCompactObjectHeaders) {
 5562     load_narrow_klass_compact(dst, src);
 5563     decode_klass_not_null(dst, tmp);
 5564   } else {
 5565     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5566     decode_klass_not_null(dst, tmp);
 5567   }
 5568 }
 5569 
 5570 void MacroAssembler::load_prototype_header(Register dst, Register src, Register tmp) {
 5571   load_klass(dst, src, tmp);
 5572   movptr(dst, Address(dst, Klass::prototype_header_offset()));
 5573 }
 5574 
 5575 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5576   assert(!UseCompactObjectHeaders, "not with compact headers");
 5577   assert_different_registers(src, tmp);
 5578   assert_different_registers(dst, tmp);
 5579   encode_klass_not_null(src, tmp);
 5580   movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5581 }
 5582 
 5583 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5584   if (UseCompactObjectHeaders) {
 5585     assert(tmp != noreg, "need tmp");
 5586     assert_different_registers(klass, obj, tmp);
 5587     load_narrow_klass_compact(tmp, obj);
 5588     cmpl(klass, tmp);
 5589   } else {
 5590     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5591   }
 5592 }
 5593 
 5594 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5595   if (UseCompactObjectHeaders) {
 5596     assert(tmp2 != noreg, "need tmp2");
 5597     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5598     load_narrow_klass_compact(tmp1, obj1);
 5599     load_narrow_klass_compact(tmp2, obj2);
 5600     cmpl(tmp1, tmp2);
 5601   } else {
 5602     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5603     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5604   }
 5605 }
 5606 
 5607 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5608                                     Register tmp1) {
 5609   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5610   decorators = AccessInternal::decorator_fixup(decorators, type);
 5611   bool as_raw = (decorators & AS_RAW) != 0;
 5612   if (as_raw) {
 5613     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5614   } else {
 5615     bs->load_at(this, decorators, type, dst, src, tmp1);
 5616   }
 5617 }
 5618 
 5619 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5620                                      Register tmp1, Register tmp2, Register tmp3) {
 5621   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5622   decorators = AccessInternal::decorator_fixup(decorators, type);
 5623   bool as_raw = (decorators & AS_RAW) != 0;
 5624   if (as_raw) {
 5625     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5626   } else {
 5627     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5628   }
 5629 }
 5630 
 5631 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
 5632                                      Register inline_layout_info) {
 5633   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5634   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
 5635 }
 5636 
 5637 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
 5638   movptr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
 5639   movl(offset, Address(offset, InlineKlass::payload_offset_offset()));
 5640 }
 5641 
 5642 void MacroAssembler::payload_addr(Register oop, Register data, Register inline_klass) {
 5643   // ((address) (void*) o) + vk->payload_offset();
 5644   Register offset = (data == oop) ? rscratch1 : data;
 5645   payload_offset(inline_klass, offset);
 5646   if (data == oop) {
 5647     addptr(data, offset);
 5648   } else {
 5649     lea(data, Address(oop, offset));
 5650   }
 5651 }
 5652 
 5653 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
 5654                                                 Register index, Register data) {
 5655   assert(index != rcx, "index needs to shift by rcx");
 5656   assert_different_registers(array, array_klass, index);
 5657   assert_different_registers(rcx, array, index);
 5658 
 5659   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
 5660   movl(rcx, Address(array_klass, Klass::layout_helper_offset()));
 5661 
 5662   // Klass::layout_helper_log2_element_size(lh)
 5663   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
 5664   shrl(rcx, Klass::_lh_log2_element_size_shift);
 5665   andl(rcx, Klass::_lh_log2_element_size_mask);
 5666   shlptr(index); // index << rcx
 5667 
 5668   lea(data, Address(array, index, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT)));
 5669 }
 5670 
 5671 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5672   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5673 }
 5674 
 5675 // Doesn't do verification, generates fixed size code
 5676 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5677   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5678 }
 5679 
 5680 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5681                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5682   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5683 }
 5684 
 5685 // Used for storing nulls.
 5686 void MacroAssembler::store_heap_oop_null(Address dst) {
 5687   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5688 }
 5689 
 5690 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5691   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5692   // Store to klass gap in destination
 5693   movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5694 }
 5695 
 5696 #ifdef ASSERT
 5697 void MacroAssembler::verify_heapbase(const char* msg) {
 5698   assert (UseCompressedOops, "should be compressed");
 5699   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5700   if (CheckCompressedOops) {
 5701     Label ok;
 5702     ExternalAddress src2(CompressedOops::base_addr());
 5703     const bool is_src2_reachable = reachable(src2);
 5704     if (!is_src2_reachable) {
 5705       push(rscratch1);  // cmpptr trashes rscratch1
 5706     }
 5707     cmpptr(r12_heapbase, src2, rscratch1);
 5708     jcc(Assembler::equal, ok);
 5709     STOP(msg);
 5710     bind(ok);
 5711     if (!is_src2_reachable) {
 5712       pop(rscratch1);
 5713     }
 5714   }
 5715 }
 5716 #endif
 5717 
 5718 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5719 void MacroAssembler::encode_heap_oop(Register r) {
 5720 #ifdef ASSERT
 5721   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5722 #endif
 5723   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5724   if (CompressedOops::base() == nullptr) {
 5725     if (CompressedOops::shift() != 0) {
 5726       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5727       shrq(r, LogMinObjAlignmentInBytes);
 5728     }
 5729     return;
 5730   }
 5731   testq(r, r);
 5732   cmovq(Assembler::equal, r, r12_heapbase);
 5733   subq(r, r12_heapbase);
 5734   shrq(r, LogMinObjAlignmentInBytes);
 5735 }
 5736 
 5737 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5738 #ifdef ASSERT
 5739   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5740   if (CheckCompressedOops) {
 5741     Label ok;
 5742     testq(r, r);
 5743     jcc(Assembler::notEqual, ok);
 5744     STOP("null oop passed to encode_heap_oop_not_null");
 5745     bind(ok);
 5746   }
 5747 #endif
 5748   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5749   if (CompressedOops::base() != nullptr) {
 5750     subq(r, r12_heapbase);
 5751   }
 5752   if (CompressedOops::shift() != 0) {
 5753     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5754     shrq(r, LogMinObjAlignmentInBytes);
 5755   }
 5756 }
 5757 
 5758 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5759 #ifdef ASSERT
 5760   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5761   if (CheckCompressedOops) {
 5762     Label ok;
 5763     testq(src, src);
 5764     jcc(Assembler::notEqual, ok);
 5765     STOP("null oop passed to encode_heap_oop_not_null2");
 5766     bind(ok);
 5767   }
 5768 #endif
 5769   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5770   if (dst != src) {
 5771     movq(dst, src);
 5772   }
 5773   if (CompressedOops::base() != nullptr) {
 5774     subq(dst, r12_heapbase);
 5775   }
 5776   if (CompressedOops::shift() != 0) {
 5777     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5778     shrq(dst, LogMinObjAlignmentInBytes);
 5779   }
 5780 }
 5781 
 5782 void  MacroAssembler::decode_heap_oop(Register r) {
 5783 #ifdef ASSERT
 5784   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5785 #endif
 5786   if (CompressedOops::base() == nullptr) {
 5787     if (CompressedOops::shift() != 0) {
 5788       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5789       shlq(r, LogMinObjAlignmentInBytes);
 5790     }
 5791   } else {
 5792     Label done;
 5793     shlq(r, LogMinObjAlignmentInBytes);
 5794     jccb(Assembler::equal, done);
 5795     addq(r, r12_heapbase);
 5796     bind(done);
 5797   }
 5798   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5799 }
 5800 
 5801 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5802   // Note: it will change flags
 5803   assert (UseCompressedOops, "should only be used for compressed headers");
 5804   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5805   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5806   // vtableStubs also counts instructions in pd_code_size_limit.
 5807   // Also do not verify_oop as this is called by verify_oop.
 5808   if (CompressedOops::shift() != 0) {
 5809     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5810     shlq(r, LogMinObjAlignmentInBytes);
 5811     if (CompressedOops::base() != nullptr) {
 5812       addq(r, r12_heapbase);
 5813     }
 5814   } else {
 5815     assert (CompressedOops::base() == nullptr, "sanity");
 5816   }
 5817 }
 5818 
 5819 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5820   // Note: it will change flags
 5821   assert (UseCompressedOops, "should only be used for compressed headers");
 5822   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5823   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5824   // vtableStubs also counts instructions in pd_code_size_limit.
 5825   // Also do not verify_oop as this is called by verify_oop.
 5826   if (CompressedOops::shift() != 0) {
 5827     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5828     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5829       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5830     } else {
 5831       if (dst != src) {
 5832         movq(dst, src);
 5833       }
 5834       shlq(dst, LogMinObjAlignmentInBytes);
 5835       if (CompressedOops::base() != nullptr) {
 5836         addq(dst, r12_heapbase);
 5837       }
 5838     }
 5839   } else {
 5840     assert (CompressedOops::base() == nullptr, "sanity");
 5841     if (dst != src) {
 5842       movq(dst, src);
 5843     }
 5844   }
 5845 }
 5846 
 5847 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5848   BLOCK_COMMENT("encode_klass_not_null {");
 5849   assert_different_registers(r, tmp);
 5850   if (CompressedKlassPointers::base() != nullptr) {
 5851     if (AOTCodeCache::is_on_for_dump()) {
 5852       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5853     } else {
 5854       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5855     }
 5856     subq(r, tmp);
 5857   }
 5858   if (CompressedKlassPointers::shift() != 0) {
 5859     shrq(r, CompressedKlassPointers::shift());
 5860   }
 5861   BLOCK_COMMENT("} encode_klass_not_null");
 5862 }
 5863 
 5864 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5865   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5866   assert_different_registers(src, dst);
 5867   if (CompressedKlassPointers::base() != nullptr) {
 5868     if (AOTCodeCache::is_on_for_dump()) {
 5869       movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5870       negq(dst);
 5871     } else {
 5872       movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5873     }
 5874     addq(dst, src);
 5875   } else {
 5876     movptr(dst, src);
 5877   }
 5878   if (CompressedKlassPointers::shift() != 0) {
 5879     shrq(dst, CompressedKlassPointers::shift());
 5880   }
 5881   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5882 }
 5883 
 5884 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5885   BLOCK_COMMENT("decode_klass_not_null {");
 5886   assert_different_registers(r, tmp);
 5887   // Note: it will change flags
 5888   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5889   // vtableStubs also counts instructions in pd_code_size_limit.
 5890   // Also do not verify_oop as this is called by verify_oop.
 5891   if (CompressedKlassPointers::shift() != 0) {
 5892     shlq(r, CompressedKlassPointers::shift());
 5893   }
 5894   if (CompressedKlassPointers::base() != nullptr) {
 5895     if (AOTCodeCache::is_on_for_dump()) {
 5896       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5897     } else {
 5898       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5899     }
 5900     addq(r, tmp);
 5901   }
 5902   BLOCK_COMMENT("} decode_klass_not_null");
 5903 }
 5904 
 5905 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5906   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 5907   assert_different_registers(src, dst);
 5908   // Note: it will change flags
 5909   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5910   // vtableStubs also counts instructions in pd_code_size_limit.
 5911   // Also do not verify_oop as this is called by verify_oop.
 5912 
 5913   if (CompressedKlassPointers::base() == nullptr &&
 5914       CompressedKlassPointers::shift() == 0) {
 5915     // The best case scenario is that there is no base or shift. Then it is already
 5916     // a pointer that needs nothing but a register rename.
 5917     movl(dst, src);
 5918   } else {
 5919     if (CompressedKlassPointers::shift() <= Address::times_8) {
 5920       if (CompressedKlassPointers::base() != nullptr) {
 5921         if (AOTCodeCache::is_on_for_dump()) {
 5922           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5923         } else {
 5924           movptr(dst, (intptr_t)CompressedKlassPointers::base());
 5925         }
 5926       } else {
 5927         xorq(dst, dst);
 5928       }
 5929       if (CompressedKlassPointers::shift() != 0) {
 5930         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 5931         leaq(dst, Address(dst, src, Address::times_8, 0));
 5932       } else {
 5933         addq(dst, src);
 5934       }
 5935     } else {
 5936       if (CompressedKlassPointers::base() != nullptr) {
 5937         if (AOTCodeCache::is_on_for_dump()) {
 5938           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5939           shrq(dst, CompressedKlassPointers::shift());
 5940         } else {
 5941           const intptr_t base_right_shifted =
 5942                (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 5943           movptr(dst, base_right_shifted);
 5944         }
 5945       } else {
 5946         xorq(dst, dst);
 5947       }
 5948       addq(dst, src);
 5949       shlq(dst, CompressedKlassPointers::shift());
 5950     }
 5951   }
 5952   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 5953 }
 5954 
 5955 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5956   assert (UseCompressedOops, "should only be used for compressed headers");
 5957   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5958   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5959   int oop_index = oop_recorder()->find_index(obj);
 5960   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5961   mov_narrow_oop(dst, oop_index, rspec);
 5962 }
 5963 
 5964 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5965   assert (UseCompressedOops, "should only be used for compressed headers");
 5966   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5967   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5968   int oop_index = oop_recorder()->find_index(obj);
 5969   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5970   mov_narrow_oop(dst, oop_index, rspec);
 5971 }
 5972 
 5973 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5974   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5975   int klass_index = oop_recorder()->find_index(k);
 5976   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5977   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5978 }
 5979 
 5980 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 5981   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5982   int klass_index = oop_recorder()->find_index(k);
 5983   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5984   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5985 }
 5986 
 5987 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 5988   assert (UseCompressedOops, "should only be used for compressed headers");
 5989   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5990   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5991   int oop_index = oop_recorder()->find_index(obj);
 5992   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5993   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5994 }
 5995 
 5996 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 5997   assert (UseCompressedOops, "should only be used for compressed headers");
 5998   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5999   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6000   int oop_index = oop_recorder()->find_index(obj);
 6001   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6002   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 6003 }
 6004 
 6005 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 6006   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6007   int klass_index = oop_recorder()->find_index(k);
 6008   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6009   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6010 }
 6011 
 6012 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 6013   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6014   int klass_index = oop_recorder()->find_index(k);
 6015   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6016   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6017 }
 6018 
 6019 void MacroAssembler::reinit_heapbase() {
 6020   if (UseCompressedOops) {
 6021     if (Universe::heap() != nullptr && !AOTCodeCache::is_on_for_dump()) {
 6022       if (CompressedOops::base() == nullptr) {
 6023         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 6024       } else {
 6025         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 6026       }
 6027     } else {
 6028       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 6029     }
 6030   }
 6031 }
 6032 
 6033 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
 6034   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
 6035   // An inline type might be returned. If fields are in registers we
 6036   // need to allocate an inline type instance and initialize it with
 6037   // the value of the fields.
 6038   Label skip;
 6039   // We only need a new buffered inline type if a new one is not returned
 6040   testptr(rax, 1);
 6041   jcc(Assembler::zero, skip);
 6042   int call_offset = -1;
 6043 
 6044   // The following code is similar to allocation code in TemplateTable::_new but has some slight differences,
 6045   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
 6046   // allocating is not necessary if vk != nullptr, etc.
 6047   Label slow_case;
 6048   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
 6049   mov(rscratch1, rax); // save rax for slow_case since *_allocate may corrupt it when allocation failed
 6050   if (vk != nullptr) {
 6051     // Called from C1, where the return type is statically known.
 6052     movptr(rbx, (intptr_t)vk->get_InlineKlass());
 6053     jint lh = vk->layout_helper();
 6054     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
 6055     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
 6056       tlab_allocate(rax, noreg, lh, r13, r14, slow_case);
 6057     } else {
 6058       jmp(slow_case);
 6059     }
 6060   } else {
 6061     // Call from interpreter. RAX contains ((the InlineKlass* of the return type) | 0x01)
 6062     mov(rbx, rax);
 6063     andptr(rbx, -2);
 6064     if (UseTLAB) {
 6065       movl(r14, Address(rbx, Klass::layout_helper_offset()));
 6066       testl(r14, Klass::_lh_instance_slow_path_bit);
 6067       jcc(Assembler::notZero, slow_case);
 6068       tlab_allocate(rax, r14, 0, r13, r14, slow_case);
 6069     } else {
 6070       jmp(slow_case);
 6071     }
 6072   }
 6073   if (UseTLAB) {
 6074     // 2. Initialize buffered inline instance header
 6075     Register buffer_obj = rax;
 6076     Register klass = rbx;
 6077     if (UseCompactObjectHeaders) {
 6078       Register mark_word = r13;
 6079       movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 6080       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), mark_word);
 6081     } else {
 6082       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), (intptr_t)markWord::inline_type_prototype().value());
 6083       xorl(r13, r13);
 6084       store_klass_gap(buffer_obj, r13);
 6085       if (vk == nullptr) {
 6086         // store_klass corrupts rbx(klass), so save it in r13 for later use (interpreter case only).
 6087         mov(r13, klass);
 6088       }
 6089       store_klass(buffer_obj, klass, rscratch1);
 6090       klass = r13;
 6091     }
 6092     // 3. Initialize its fields with an inline class specific handler
 6093     if (vk != nullptr) {
 6094       call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
 6095     } else {
 6096       movptr(rbx, Address(klass, InlineKlass::adr_members_offset()));
 6097       movptr(rbx, Address(rbx, InlineKlass::pack_handler_offset()));
 6098       call(rbx);
 6099     }
 6100     jmp(skip);
 6101   }
 6102   bind(slow_case);
 6103   // We failed to allocate a new inline type, fall back to a runtime
 6104   // call. Some oop field may be live in some registers but we can't
 6105   // tell. That runtime call will take care of preserving them
 6106   // across a GC if there's one.
 6107   mov(rax, rscratch1);
 6108 
 6109   if (from_interpreter) {
 6110     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
 6111   } else {
 6112     call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
 6113     call_offset = offset();
 6114   }
 6115 
 6116   bind(skip);
 6117   return call_offset;
 6118 }
 6119 
 6120 // Move a value between registers/stack slots and update the reg_state
 6121 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
 6122   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
 6123   if (reg_state[to->value()] == reg_written) {
 6124     return true; // Already written
 6125   }
 6126   if (from != to && bt != T_VOID) {
 6127     if (reg_state[to->value()] == reg_readonly) {
 6128       return false; // Not yet writable
 6129     }
 6130     if (from->is_reg()) {
 6131       if (to->is_reg()) {
 6132         if (from->is_XMMRegister()) {
 6133           if (bt == T_DOUBLE) {
 6134             movdbl(to->as_XMMRegister(), from->as_XMMRegister());
 6135           } else {
 6136             assert(bt == T_FLOAT, "must be float");
 6137             movflt(to->as_XMMRegister(), from->as_XMMRegister());
 6138           }
 6139         } else {
 6140           movq(to->as_Register(), from->as_Register());
 6141         }
 6142       } else {
 6143         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6144         Address to_addr = Address(rsp, st_off);
 6145         if (from->is_XMMRegister()) {
 6146           if (bt == T_DOUBLE) {
 6147             movdbl(to_addr, from->as_XMMRegister());
 6148           } else {
 6149             assert(bt == T_FLOAT, "must be float");
 6150             movflt(to_addr, from->as_XMMRegister());
 6151           }
 6152         } else {
 6153           movq(to_addr, from->as_Register());
 6154         }
 6155       }
 6156     } else {
 6157       Address from_addr = Address(rsp, from->reg2stack() * VMRegImpl::stack_slot_size + wordSize);
 6158       if (to->is_reg()) {
 6159         if (to->is_XMMRegister()) {
 6160           if (bt == T_DOUBLE) {
 6161             movdbl(to->as_XMMRegister(), from_addr);
 6162           } else {
 6163             assert(bt == T_FLOAT, "must be float");
 6164             movflt(to->as_XMMRegister(), from_addr);
 6165           }
 6166         } else {
 6167           movq(to->as_Register(), from_addr);
 6168         }
 6169       } else {
 6170         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6171         movq(r13, from_addr);
 6172         movq(Address(rsp, st_off), r13);
 6173       }
 6174     }
 6175   }
 6176   // Update register states
 6177   reg_state[from->value()] = reg_writable;
 6178   reg_state[to->value()] = reg_written;
 6179   return true;
 6180 }
 6181 
 6182 // Calculate the extra stack space required for packing or unpacking inline
 6183 // args and adjust the stack pointer (see MacroAssembler::remove_frame).
 6184 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
 6185   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
 6186   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
 6187   assert(sp_inc > 0, "sanity");
 6188   // Two additional slots to account for return address
 6189   sp_inc +=  2 * VMRegImpl::stack_slot_size;
 6190 
 6191   push(rbp);
 6192   subptr(rsp, sp_inc);
 6193 #ifdef ASSERT
 6194   movl(Address(rsp, 0), badRegWordVal);
 6195   movl(Address(rsp, VMRegImpl::stack_slot_size), badRegWordVal);
 6196 #endif
 6197   return sp_inc + wordSize; // account for rbp space
 6198 }
 6199 
 6200 // Read all fields from an inline type buffer and store the field values in registers/stack slots.
 6201 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
 6202                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
 6203                                           RegState reg_state[]) {
 6204   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
 6205   assert(from->is_valid(), "source must be valid");
 6206   bool progress = false;
 6207 #ifdef ASSERT
 6208   const int start_offset = offset();
 6209 #endif
 6210 
 6211   Label L_null, L_notNull;
 6212   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
 6213   Register tmp1 = r10;
 6214   Register tmp2 = r13;
 6215   Register fromReg = noreg;
 6216   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
 6217   bool done = true;
 6218   bool mark_done = true;
 6219   VMReg toReg;
 6220   BasicType bt;
 6221   // Check if argument requires a null check
 6222   bool null_check = false;
 6223   VMReg nullCheckReg;
 6224   while (stream.next(nullCheckReg, bt)) {
 6225     if (sig->at(stream.sig_index())._offset == -1) {
 6226       null_check = true;
 6227       break;
 6228     }
 6229   }
 6230   stream.reset(sig_index, to_index);
 6231   while (stream.next(toReg, bt)) {
 6232     assert(toReg->is_valid(), "destination must be valid");
 6233     int idx = (int)toReg->value();
 6234     if (reg_state[idx] == reg_readonly) {
 6235       if (idx != from->value()) {
 6236         mark_done = false;
 6237       }
 6238       done = false;
 6239       continue;
 6240     } else if (reg_state[idx] == reg_written) {
 6241       continue;
 6242     }
 6243     assert(reg_state[idx] == reg_writable, "must be writable");
 6244     reg_state[idx] = reg_written;
 6245     progress = true;
 6246 
 6247     if (fromReg == noreg) {
 6248       if (from->is_reg()) {
 6249         fromReg = from->as_Register();
 6250       } else {
 6251         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6252         movq(tmp1, Address(rsp, st_off));
 6253         fromReg = tmp1;
 6254       }
 6255       if (null_check) {
 6256         // Nullable inline type argument, emit null check
 6257         testptr(fromReg, fromReg);
 6258         jcc(Assembler::zero, L_null);
 6259       }
 6260     }
 6261     int off = sig->at(stream.sig_index())._offset;
 6262     if (off == -1) {
 6263       assert(null_check, "Missing null check at");
 6264       if (toReg->is_stack()) {
 6265         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6266         movq(Address(rsp, st_off), 1);
 6267       } else {
 6268         movq(toReg->as_Register(), 1);
 6269       }
 6270       continue;
 6271     }
 6272     if (sig->at(stream.sig_index())._vt_oop) {
 6273       if (toReg->is_stack()) {
 6274         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6275         movq(Address(rsp, st_off), fromReg);
 6276       } else {
 6277         movq(toReg->as_Register(), fromReg);
 6278       }
 6279       continue;
 6280     }
 6281     assert(off > 0, "offset in object should be positive");
 6282     Address fromAddr = Address(fromReg, off);
 6283     if (!toReg->is_XMMRegister()) {
 6284       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
 6285       if (is_reference_type(bt)) {
 6286         load_heap_oop(dst, fromAddr);
 6287       } else {
 6288         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
 6289         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
 6290       }
 6291       if (toReg->is_stack()) {
 6292         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6293         movq(Address(rsp, st_off), dst);
 6294       }
 6295     } else if (bt == T_DOUBLE) {
 6296       movdbl(toReg->as_XMMRegister(), fromAddr);
 6297     } else {
 6298       assert(bt == T_FLOAT, "must be float");
 6299       movflt(toReg->as_XMMRegister(), fromAddr);
 6300     }
 6301   }
 6302   if (progress && null_check) {
 6303     if (done) {
 6304       jmp(L_notNull);
 6305       bind(L_null);
 6306       // Set null marker to zero to signal that the argument is null.
 6307       // Also set all fields to zero since the runtime requires a canonical
 6308       // representation of a flat null.
 6309       stream.reset(sig_index, to_index);
 6310       while (stream.next(toReg, bt)) {
 6311         if (toReg->is_stack()) {
 6312           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6313           movq(Address(rsp, st_off), 0);
 6314         } else if (toReg->is_XMMRegister()) {
 6315           xorps(toReg->as_XMMRegister(), toReg->as_XMMRegister());
 6316         } else {
 6317           xorl(toReg->as_Register(), toReg->as_Register());
 6318         }
 6319       }
 6320       bind(L_notNull);
 6321     } else {
 6322       bind(L_null);
 6323     }
 6324   }
 6325 
 6326   sig_index = stream.sig_index();
 6327   to_index = stream.regs_index();
 6328 
 6329   if (mark_done && reg_state[from->value()] != reg_written) {
 6330     // This is okay because no one else will write to that slot
 6331     reg_state[from->value()] = reg_writable;
 6332   }
 6333   from_index--;
 6334   assert(progress || (start_offset == offset()), "should not emit code");
 6335   return done;
 6336 }
 6337 
 6338 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
 6339                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
 6340                                         RegState reg_state[], Register val_array) {
 6341   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
 6342   assert(to->is_valid(), "destination must be valid");
 6343 
 6344   if (reg_state[to->value()] == reg_written) {
 6345     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6346     return true; // Already written
 6347   }
 6348 
 6349   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
 6350   Register val_obj_tmp = r11;
 6351   Register from_reg_tmp = r14;
 6352   Register tmp1 = r10;
 6353   Register tmp2 = r13;
 6354   Register tmp3 = rbx;
 6355   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
 6356 
 6357   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
 6358 
 6359   if (reg_state[to->value()] == reg_readonly) {
 6360     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
 6361       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6362       return false; // Not yet writable
 6363     }
 6364     val_obj = val_obj_tmp;
 6365   }
 6366 
 6367   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
 6368   VMReg fromReg;
 6369   BasicType bt;
 6370   Label L_null;
 6371   while (stream.next(fromReg, bt)) {
 6372     assert(fromReg->is_valid(), "source must be valid");
 6373     reg_state[fromReg->value()] = reg_writable;
 6374 
 6375     int off = sig->at(stream.sig_index())._offset;
 6376     if (off == -1) {
 6377       // Nullable inline type argument, emit null check
 6378       Label L_notNull;
 6379       if (fromReg->is_stack()) {
 6380         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6381         testb(Address(rsp, ld_off), 1);
 6382       } else {
 6383         testb(fromReg->as_Register(), 1);
 6384       }
 6385       jcc(Assembler::notZero, L_notNull);
 6386       movptr(val_obj, 0);
 6387       jmp(L_null);
 6388       bind(L_notNull);
 6389       continue;
 6390     }
 6391     if (sig->at(stream.sig_index())._vt_oop) {
 6392       // buffer argument: use if non null
 6393       if (fromReg->is_stack()) {
 6394         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6395         movptr(val_obj, Address(rsp, ld_off));
 6396       } else {
 6397         movptr(val_obj, fromReg->as_Register());
 6398       }
 6399       testptr(val_obj, val_obj);
 6400       jcc(Assembler::notEqual, L_null);
 6401       // otherwise get the buffer from the just allocated pool of buffers
 6402       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
 6403       load_heap_oop(val_obj, Address(val_array, index));
 6404       continue;
 6405     }
 6406 
 6407     assert(off > 0, "offset in object should be positive");
 6408     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
 6409 
 6410     // Pack the scalarized field into the value object.
 6411     Address dst(val_obj, off);
 6412     if (!fromReg->is_XMMRegister()) {
 6413       Register src;
 6414       if (fromReg->is_stack()) {
 6415         src = from_reg_tmp;
 6416         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6417         load_sized_value(src, Address(rsp, ld_off), size_in_bytes, /* is_signed */ false);
 6418       } else {
 6419         src = fromReg->as_Register();
 6420       }
 6421       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
 6422       if (is_reference_type(bt)) {
 6423         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
 6424         mov(tmp3, val_obj);
 6425         Address dst_with_tmp3(tmp3, off);
 6426         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
 6427       } else {
 6428         store_sized_value(dst, src, size_in_bytes);
 6429       }
 6430     } else if (bt == T_DOUBLE) {
 6431       movdbl(dst, fromReg->as_XMMRegister());
 6432     } else {
 6433       assert(bt == T_FLOAT, "must be float");
 6434       movflt(dst, fromReg->as_XMMRegister());
 6435     }
 6436   }
 6437   bind(L_null);
 6438   sig_index = stream.sig_index();
 6439   from_index = stream.regs_index();
 6440 
 6441   assert(reg_state[to->value()] == reg_writable, "must have already been read");
 6442   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
 6443   assert(success, "to register must be writeable");
 6444   return true;
 6445 }
 6446 
 6447 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
 6448   return reg->is_XMMRegister() ? xmm8->as_VMReg() : r14->as_VMReg();
 6449 }
 6450 
 6451 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
 6452   assert((initial_framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 6453   if (needs_stack_repair) {
 6454     // The method has a scalarized entry point (where fields of value object arguments
 6455     // are passed through registers and stack), and a non-scalarized entry point (where
 6456     // value object arguments are given as oops). The non-scalarized entry point will
 6457     // first load each field of value object arguments and store them in registers and on
 6458     // the stack in a way compatible with the scalarized entry point. To do so, some extra
 6459     // stack space might be reserved (if argument registers are not enough). On leaving the
 6460     // method, this space must be freed.
 6461     //
 6462     // In case we used the non-scalarized entry point the stack looks like this:
 6463     //
 6464     // | Arguments from caller     |
 6465     // |---------------------------|  <-- caller's SP
 6466     // | Return address #1         |
 6467     // | Saved RBP #1              |
 6468     // |---------------------------|
 6469     // | Extension space for       |
 6470     // |   inline arg (un)packing  |
 6471     // |---------------------------|  <-- start of this method's frame
 6472     // | Return address #2         |
 6473     // | Saved RBP #2              |
 6474     // |---------------------------|  <-- RBP (with -XX:+PreserveFramePointer)
 6475     // | sp_inc                    |
 6476     // | method locals             |
 6477     // |---------------------------|  <-- SP
 6478     //
 6479     // Space for the return pc and saved rbp is reserved twice. But only the #1 copies
 6480     // contain the real values of return pc and saved rbp. The #2 copies are not reliable
 6481     // and should not be used. They are mostly needed to add space between the extension
 6482     // space and the locals, as there would be between the real arguments and the locals
 6483     // if we don't need to do unpacking (from the scalarized entry point).
 6484     //
 6485     // When leaving, one must load RBP #1 into RBP, and use the copy #1 of the return address,
 6486     // while keeping in mind that from the scalarized entry point, there will be only one
 6487     // copy. Indeed, in the case we used the scalarized calling convention, the stack looks like this:
 6488     //
 6489     // | Arguments from caller     |
 6490     // |---------------------------|  <-- caller's SP
 6491     // | Return address            |
 6492     // | Saved RBP                 |
 6493     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
 6494     // | sp_inc                    |
 6495     // | method locals             |
 6496     // |---------------------------|  <-- SP
 6497     //
 6498     // The sp_inc stack slot holds the total size of the frame, including the extension
 6499     // space and copies #2 of the return address and the saved RBP (but never the copies
 6500     // #1 of the return address and saved RBP). That is how to find the copies #1 of the
 6501     // return address and saved rbp. This size is expressed in bytes. Be careful when using
 6502     // it from C++ in pointer arithmetic you might need to divide it by wordSize.
 6503 
 6504     // The stack increment resides just below the saved rbp
 6505     addq(rsp, Address(rsp, initial_framesize - wordSize));
 6506     pop(rbp);
 6507   } else {
 6508     if (initial_framesize > 0) {
 6509       addq(rsp, initial_framesize);
 6510     }
 6511     pop(rbp);
 6512   }
 6513 }
 6514 
 6515 #if COMPILER2_OR_JVMCI
 6516 
 6517 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 6518 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp, KRegister mask) {
 6519   // cnt - number of qwords (8-byte words).
 6520   // base - start address, qword aligned.
 6521   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 6522   bool use64byteVector = (MaxVectorSize == 64) && (CopyAVX3Threshold == 0);
 6523   if (use64byteVector) {
 6524     evpbroadcastq(xtmp, val, AVX_512bit);
 6525   } else if (MaxVectorSize >= 32) {
 6526     movdq(xtmp, val);
 6527     punpcklqdq(xtmp, xtmp);
 6528     vinserti128_high(xtmp, xtmp);
 6529   } else {
 6530     movdq(xtmp, val);
 6531     punpcklqdq(xtmp, xtmp);
 6532   }
 6533   jmp(L_zero_64_bytes);
 6534 
 6535   BIND(L_loop);
 6536   if (MaxVectorSize >= 32) {
 6537     fill64(base, 0, xtmp, use64byteVector);
 6538   } else {
 6539     movdqu(Address(base,  0), xtmp);
 6540     movdqu(Address(base, 16), xtmp);
 6541     movdqu(Address(base, 32), xtmp);
 6542     movdqu(Address(base, 48), xtmp);
 6543   }
 6544   addptr(base, 64);
 6545 
 6546   BIND(L_zero_64_bytes);
 6547   subptr(cnt, 8);
 6548   jccb(Assembler::greaterEqual, L_loop);
 6549 
 6550   // Copy trailing 64 bytes
 6551   if (use64byteVector) {
 6552     addptr(cnt, 8);
 6553     jccb(Assembler::equal, L_end);
 6554     fill64_masked(3, base, 0, xtmp, mask, cnt, val, true);
 6555     jmp(L_end);
 6556   } else {
 6557     addptr(cnt, 4);
 6558     jccb(Assembler::less, L_tail);
 6559     if (MaxVectorSize >= 32) {
 6560       vmovdqu(Address(base, 0), xtmp);
 6561     } else {
 6562       movdqu(Address(base,  0), xtmp);
 6563       movdqu(Address(base, 16), xtmp);
 6564     }
 6565   }
 6566   addptr(base, 32);
 6567   subptr(cnt, 4);
 6568 
 6569   BIND(L_tail);
 6570   addptr(cnt, 4);
 6571   jccb(Assembler::lessEqual, L_end);
 6572   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 6573     fill32_masked(3, base, 0, xtmp, mask, cnt, val);
 6574   } else {
 6575     decrement(cnt);
 6576 
 6577     BIND(L_sloop);
 6578     movq(Address(base, 0), xtmp);
 6579     addptr(base, 8);
 6580     decrement(cnt);
 6581     jccb(Assembler::greaterEqual, L_sloop);
 6582   }
 6583   BIND(L_end);
 6584 }
 6585 
 6586 // Clearing constant sized memory using YMM/ZMM registers.
 6587 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 6588   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 6589   bool use64byteVector = (MaxVectorSize > 32) && (CopyAVX3Threshold == 0);
 6590 
 6591   int vector64_count = (cnt & (~0x7)) >> 3;
 6592   cnt = cnt & 0x7;
 6593   const int fill64_per_loop = 4;
 6594   const int max_unrolled_fill64 = 8;
 6595 
 6596   // 64 byte initialization loop.
 6597   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 6598   int start64 = 0;
 6599   if (vector64_count > max_unrolled_fill64) {
 6600     Label LOOP;
 6601     Register index = rtmp;
 6602 
 6603     start64 = vector64_count - (vector64_count % fill64_per_loop);
 6604 
 6605     movl(index, 0);
 6606     BIND(LOOP);
 6607     for (int i = 0; i < fill64_per_loop; i++) {
 6608       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 6609     }
 6610     addl(index, fill64_per_loop * 64);
 6611     cmpl(index, start64 * 64);
 6612     jccb(Assembler::less, LOOP);
 6613   }
 6614   for (int i = start64; i < vector64_count; i++) {
 6615     fill64(base, i * 64, xtmp, use64byteVector);
 6616   }
 6617 
 6618   // Clear remaining 64 byte tail.
 6619   int disp = vector64_count * 64;
 6620   if (cnt) {
 6621     switch (cnt) {
 6622       case 1:
 6623         movq(Address(base, disp), xtmp);
 6624         break;
 6625       case 2:
 6626         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 6627         break;
 6628       case 3:
 6629         movl(rtmp, 0x7);
 6630         kmovwl(mask, rtmp);
 6631         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 6632         break;
 6633       case 4:
 6634         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6635         break;
 6636       case 5:
 6637         if (use64byteVector) {
 6638           movl(rtmp, 0x1F);
 6639           kmovwl(mask, rtmp);
 6640           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6641         } else {
 6642           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6643           movq(Address(base, disp + 32), xtmp);
 6644         }
 6645         break;
 6646       case 6:
 6647         if (use64byteVector) {
 6648           movl(rtmp, 0x3F);
 6649           kmovwl(mask, rtmp);
 6650           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6651         } else {
 6652           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6653           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 6654         }
 6655         break;
 6656       case 7:
 6657         if (use64byteVector) {
 6658           movl(rtmp, 0x7F);
 6659           kmovwl(mask, rtmp);
 6660           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6661         } else {
 6662           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6663           movl(rtmp, 0x7);
 6664           kmovwl(mask, rtmp);
 6665           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 6666         }
 6667         break;
 6668       default:
 6669         fatal("Unexpected length : %d\n",cnt);
 6670         break;
 6671     }
 6672   }
 6673 }
 6674 
 6675 void MacroAssembler::clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp,
 6676                                bool is_large, bool word_copy_only, KRegister mask) {
 6677   // cnt      - number of qwords (8-byte words).
 6678   // base     - start address, qword aligned.
 6679   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6680   assert(base==rdi, "base register must be edi for rep stos");
 6681   assert(val==rax,   "val register must be eax for rep stos");
 6682   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6683   assert(InitArrayShortSize % BytesPerLong == 0,
 6684     "InitArrayShortSize should be the multiple of BytesPerLong");
 6685 
 6686   Label DONE;
 6687 
 6688   if (!is_large) {
 6689     Label LOOP, LONG;
 6690     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6691     jccb(Assembler::greater, LONG);
 6692 
 6693     decrement(cnt);
 6694     jccb(Assembler::negative, DONE); // Zero length
 6695 
 6696     // Use individual pointer-sized stores for small counts:
 6697     BIND(LOOP);
 6698     movptr(Address(base, cnt, Address::times_ptr), val);
 6699     decrement(cnt);
 6700     jccb(Assembler::greaterEqual, LOOP);
 6701     jmpb(DONE);
 6702 
 6703     BIND(LONG);
 6704   }
 6705 
 6706   // Use longer rep-prefixed ops for non-small counts:
 6707   if (UseFastStosb && !word_copy_only) {
 6708     shlptr(cnt, 3); // convert to number of bytes
 6709     rep_stosb();
 6710   } else if (UseXMMForObjInit) {
 6711     xmm_clear_mem(base, cnt, val, xtmp, mask);
 6712   } else {
 6713     rep_stos();
 6714   }
 6715 
 6716   BIND(DONE);
 6717 }
 6718 
 6719 #endif //COMPILER2_OR_JVMCI
 6720 
 6721 
 6722 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6723                                    Register to, Register value, Register count,
 6724                                    Register rtmp, XMMRegister xtmp) {
 6725   ShortBranchVerifier sbv(this);
 6726   assert_different_registers(to, value, count, rtmp);
 6727   Label L_exit;
 6728   Label L_fill_2_bytes, L_fill_4_bytes;
 6729 
 6730 #if defined(COMPILER2)
 6731   if(MaxVectorSize >=32 &&
 6732      VM_Version::supports_avx512vlbw() &&
 6733      VM_Version::supports_bmi2()) {
 6734     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6735     return;
 6736   }
 6737 #endif
 6738 
 6739   int shift = -1;
 6740   switch (t) {
 6741     case T_BYTE:
 6742       shift = 2;
 6743       break;
 6744     case T_SHORT:
 6745       shift = 1;
 6746       break;
 6747     case T_INT:
 6748       shift = 0;
 6749       break;
 6750     default: ShouldNotReachHere();
 6751   }
 6752 
 6753   if (t == T_BYTE) {
 6754     andl(value, 0xff);
 6755     movl(rtmp, value);
 6756     shll(rtmp, 8);
 6757     orl(value, rtmp);
 6758   }
 6759   if (t == T_SHORT) {
 6760     andl(value, 0xffff);
 6761   }
 6762   if (t == T_BYTE || t == T_SHORT) {
 6763     movl(rtmp, value);
 6764     shll(rtmp, 16);
 6765     orl(value, rtmp);
 6766   }
 6767 
 6768   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6769   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6770   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6771     Label L_skip_align2;
 6772     // align source address at 4 bytes address boundary
 6773     if (t == T_BYTE) {
 6774       Label L_skip_align1;
 6775       // One byte misalignment happens only for byte arrays
 6776       testptr(to, 1);
 6777       jccb(Assembler::zero, L_skip_align1);
 6778       movb(Address(to, 0), value);
 6779       increment(to);
 6780       decrement(count);
 6781       BIND(L_skip_align1);
 6782     }
 6783     // Two bytes misalignment happens only for byte and short (char) arrays
 6784     testptr(to, 2);
 6785     jccb(Assembler::zero, L_skip_align2);
 6786     movw(Address(to, 0), value);
 6787     addptr(to, 2);
 6788     subptr(count, 1<<(shift-1));
 6789     BIND(L_skip_align2);
 6790   }
 6791   {
 6792     Label L_fill_32_bytes;
 6793     if (!UseUnalignedLoadStores) {
 6794       // align to 8 bytes, we know we are 4 byte aligned to start
 6795       testptr(to, 4);
 6796       jccb(Assembler::zero, L_fill_32_bytes);
 6797       movl(Address(to, 0), value);
 6798       addptr(to, 4);
 6799       subptr(count, 1<<shift);
 6800     }
 6801     BIND(L_fill_32_bytes);
 6802     {
 6803       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6804       movdl(xtmp, value);
 6805       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6806         Label L_check_fill_32_bytes;
 6807         if (UseAVX > 2) {
 6808           // Fill 64-byte chunks
 6809           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6810 
 6811           // If number of bytes to fill < CopyAVX3Threshold, perform fill using AVX2
 6812           cmpptr(count, CopyAVX3Threshold);
 6813           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6814 
 6815           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6816 
 6817           subptr(count, 16 << shift);
 6818           jcc(Assembler::less, L_check_fill_32_bytes);
 6819           align(16);
 6820 
 6821           BIND(L_fill_64_bytes_loop_avx3);
 6822           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6823           addptr(to, 64);
 6824           subptr(count, 16 << shift);
 6825           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6826           jmpb(L_check_fill_32_bytes);
 6827 
 6828           BIND(L_check_fill_64_bytes_avx2);
 6829         }
 6830         // Fill 64-byte chunks
 6831         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6832 
 6833         subptr(count, 16 << shift);
 6834         jcc(Assembler::less, L_check_fill_32_bytes);
 6835 
 6836         // align data for 64-byte chunks
 6837         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6838         if (EnableX86ECoreOpts) {
 6839             // align 'big' arrays to cache lines to minimize split_stores
 6840             cmpptr(count, 96 << shift);
 6841             jcc(Assembler::below, L_fill_64_bytes_loop);
 6842 
 6843             // Find the bytes needed for alignment
 6844             movptr(rtmp, to);
 6845             andptr(rtmp, 0x1c);
 6846             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6847             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6848             addptr(rtmp, 32);
 6849             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6850             subptr(count, rtmp);    // adjust count by number of elements
 6851 
 6852             align(16);
 6853             BIND(L_align_64_bytes_loop);
 6854             movdl(Address(to, 0), xtmp);
 6855             addptr(to, 4);
 6856             subptr(rtmp, 1 << shift);
 6857             jcc(Assembler::greater, L_align_64_bytes_loop);
 6858         }
 6859 
 6860         align(16);
 6861         BIND(L_fill_64_bytes_loop);
 6862         vmovdqu(Address(to, 0), xtmp);
 6863         vmovdqu(Address(to, 32), xtmp);
 6864         addptr(to, 64);
 6865         subptr(count, 16 << shift);
 6866         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6867 
 6868         align(16);
 6869         BIND(L_check_fill_32_bytes);
 6870         addptr(count, 8 << shift);
 6871         jccb(Assembler::less, L_check_fill_8_bytes);
 6872         vmovdqu(Address(to, 0), xtmp);
 6873         addptr(to, 32);
 6874         subptr(count, 8 << shift);
 6875 
 6876         BIND(L_check_fill_8_bytes);
 6877         // clean upper bits of YMM registers
 6878         movdl(xtmp, value);
 6879         pshufd(xtmp, xtmp, 0);
 6880       } else {
 6881         // Fill 32-byte chunks
 6882         pshufd(xtmp, xtmp, 0);
 6883 
 6884         subptr(count, 8 << shift);
 6885         jcc(Assembler::less, L_check_fill_8_bytes);
 6886         align(16);
 6887 
 6888         BIND(L_fill_32_bytes_loop);
 6889 
 6890         if (UseUnalignedLoadStores) {
 6891           movdqu(Address(to, 0), xtmp);
 6892           movdqu(Address(to, 16), xtmp);
 6893         } else {
 6894           movq(Address(to, 0), xtmp);
 6895           movq(Address(to, 8), xtmp);
 6896           movq(Address(to, 16), xtmp);
 6897           movq(Address(to, 24), xtmp);
 6898         }
 6899 
 6900         addptr(to, 32);
 6901         subptr(count, 8 << shift);
 6902         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6903 
 6904         BIND(L_check_fill_8_bytes);
 6905       }
 6906       addptr(count, 8 << shift);
 6907       jccb(Assembler::zero, L_exit);
 6908       jmpb(L_fill_8_bytes);
 6909 
 6910       //
 6911       // length is too short, just fill qwords
 6912       //
 6913       align(16);
 6914       BIND(L_fill_8_bytes_loop);
 6915       movq(Address(to, 0), xtmp);
 6916       addptr(to, 8);
 6917       BIND(L_fill_8_bytes);
 6918       subptr(count, 1 << (shift + 1));
 6919       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6920     }
 6921   }
 6922 
 6923   Label L_fill_4_bytes_loop;
 6924   testl(count, 1 << shift);
 6925   jccb(Assembler::zero, L_fill_2_bytes);
 6926 
 6927   align(16);
 6928   BIND(L_fill_4_bytes_loop);
 6929   movl(Address(to, 0), value);
 6930   addptr(to, 4);
 6931 
 6932   BIND(L_fill_4_bytes);
 6933   subptr(count, 1 << shift);
 6934   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 6935 
 6936   if (t == T_BYTE || t == T_SHORT) {
 6937     Label L_fill_byte;
 6938     BIND(L_fill_2_bytes);
 6939     // fill trailing 2 bytes
 6940     testl(count, 1<<(shift-1));
 6941     jccb(Assembler::zero, L_fill_byte);
 6942     movw(Address(to, 0), value);
 6943     if (t == T_BYTE) {
 6944       addptr(to, 2);
 6945       BIND(L_fill_byte);
 6946       // fill trailing byte
 6947       testl(count, 1);
 6948       jccb(Assembler::zero, L_exit);
 6949       movb(Address(to, 0), value);
 6950     } else {
 6951       BIND(L_fill_byte);
 6952     }
 6953   } else {
 6954     BIND(L_fill_2_bytes);
 6955   }
 6956   BIND(L_exit);
 6957 }
 6958 
 6959 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6960   switch(type) {
 6961     case T_BYTE:
 6962     case T_BOOLEAN:
 6963       evpbroadcastb(dst, src, vector_len);
 6964       break;
 6965     case T_SHORT:
 6966     case T_CHAR:
 6967       evpbroadcastw(dst, src, vector_len);
 6968       break;
 6969     case T_INT:
 6970     case T_FLOAT:
 6971       evpbroadcastd(dst, src, vector_len);
 6972       break;
 6973     case T_LONG:
 6974     case T_DOUBLE:
 6975       evpbroadcastq(dst, src, vector_len);
 6976       break;
 6977     default:
 6978       fatal("Unhandled type : %s", type2name(type));
 6979       break;
 6980   }
 6981 }
 6982 
 6983 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 6984 //
 6985 // @IntrinsicCandidate
 6986 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 6987 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6988 //     int i = 0;
 6989 //     for (; i < len; i++) {
 6990 //         char c = sa[sp++];
 6991 //         if (c > '\u00FF')
 6992 //             break;
 6993 //         da[dp++] = (byte) c;
 6994 //     }
 6995 //     return i;
 6996 // }
 6997 //
 6998 // @IntrinsicCandidate
 6999 // int java.lang.StringCoding.encodeISOArray0(
 7000 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 7001 //   int i = 0;
 7002 //   for (; i < len; i++) {
 7003 //     char c = StringUTF16.getChar(sa, sp++);
 7004 //     if (c > '\u00FF')
 7005 //       break;
 7006 //     da[dp++] = (byte) c;
 7007 //   }
 7008 //   return i;
 7009 // }
 7010 //
 7011 // @IntrinsicCandidate
 7012 // int java.lang.StringCoding.encodeAsciiArray0(
 7013 //         char[] sa, int sp, byte[] da, int dp, int len) {
 7014 //   int i = 0;
 7015 //   for (; i < len; i++) {
 7016 //     char c = sa[sp++];
 7017 //     if (c >= '\u0080')
 7018 //       break;
 7019 //     da[dp++] = (byte) c;
 7020 //   }
 7021 //   return i;
 7022 // }
 7023 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 7024   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 7025   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 7026   Register tmp5, Register result, bool ascii) {
 7027 
 7028   // rsi: src
 7029   // rdi: dst
 7030   // rdx: len
 7031   // rcx: tmp5
 7032   // rax: result
 7033   ShortBranchVerifier sbv(this);
 7034   assert_different_registers(src, dst, len, tmp5, result);
 7035   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 7036 
 7037   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 7038   int short_mask = ascii ? 0xff80 : 0xff00;
 7039 
 7040   // set result
 7041   xorl(result, result);
 7042   // check for zero length
 7043   testl(len, len);
 7044   jcc(Assembler::zero, L_done);
 7045 
 7046   movl(result, len);
 7047 
 7048   // Setup pointers
 7049   lea(src, Address(src, len, Address::times_2)); // char[]
 7050   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 7051   negptr(len);
 7052 
 7053   if (UseSSE42Intrinsics || UseAVX >= 2) {
 7054     Label L_copy_8_chars, L_copy_8_chars_exit;
 7055     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 7056 
 7057     if (UseAVX >= 2) {
 7058       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 7059       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7060       movdl(tmp1Reg, tmp5);
 7061       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 7062       jmp(L_chars_32_check);
 7063 
 7064       bind(L_copy_32_chars);
 7065       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 7066       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 7067       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7068       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7069       jccb(Assembler::notZero, L_copy_32_chars_exit);
 7070       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7071       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 7072       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 7073 
 7074       bind(L_chars_32_check);
 7075       addptr(len, 32);
 7076       jcc(Assembler::lessEqual, L_copy_32_chars);
 7077 
 7078       bind(L_copy_32_chars_exit);
 7079       subptr(len, 16);
 7080       jccb(Assembler::greater, L_copy_16_chars_exit);
 7081 
 7082     } else if (UseSSE42Intrinsics) {
 7083       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7084       movdl(tmp1Reg, tmp5);
 7085       pshufd(tmp1Reg, tmp1Reg, 0);
 7086       jmpb(L_chars_16_check);
 7087     }
 7088 
 7089     bind(L_copy_16_chars);
 7090     if (UseAVX >= 2) {
 7091       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 7092       vptest(tmp2Reg, tmp1Reg);
 7093       jcc(Assembler::notZero, L_copy_16_chars_exit);
 7094       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 7095       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 7096     } else {
 7097       if (UseAVX > 0) {
 7098         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7099         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7100         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 7101       } else {
 7102         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7103         por(tmp2Reg, tmp3Reg);
 7104         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7105         por(tmp2Reg, tmp4Reg);
 7106       }
 7107       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7108       jccb(Assembler::notZero, L_copy_16_chars_exit);
 7109       packuswb(tmp3Reg, tmp4Reg);
 7110     }
 7111     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 7112 
 7113     bind(L_chars_16_check);
 7114     addptr(len, 16);
 7115     jcc(Assembler::lessEqual, L_copy_16_chars);
 7116 
 7117     bind(L_copy_16_chars_exit);
 7118     if (UseAVX >= 2) {
 7119       // clean upper bits of YMM registers
 7120       vpxor(tmp2Reg, tmp2Reg);
 7121       vpxor(tmp3Reg, tmp3Reg);
 7122       vpxor(tmp4Reg, tmp4Reg);
 7123       movdl(tmp1Reg, tmp5);
 7124       pshufd(tmp1Reg, tmp1Reg, 0);
 7125     }
 7126     subptr(len, 8);
 7127     jccb(Assembler::greater, L_copy_8_chars_exit);
 7128 
 7129     bind(L_copy_8_chars);
 7130     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 7131     ptest(tmp3Reg, tmp1Reg);
 7132     jccb(Assembler::notZero, L_copy_8_chars_exit);
 7133     packuswb(tmp3Reg, tmp1Reg);
 7134     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 7135     addptr(len, 8);
 7136     jccb(Assembler::lessEqual, L_copy_8_chars);
 7137 
 7138     bind(L_copy_8_chars_exit);
 7139     subptr(len, 8);
 7140     jccb(Assembler::zero, L_done);
 7141   }
 7142 
 7143   bind(L_copy_1_char);
 7144   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 7145   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 7146   jccb(Assembler::notZero, L_copy_1_char_exit);
 7147   movb(Address(dst, len, Address::times_1, 0), tmp5);
 7148   addptr(len, 1);
 7149   jccb(Assembler::less, L_copy_1_char);
 7150 
 7151   bind(L_copy_1_char_exit);
 7152   addptr(result, len); // len is negative count of not processed elements
 7153 
 7154   bind(L_done);
 7155 }
 7156 
 7157 /**
 7158  * Helper for multiply_to_len().
 7159  */
 7160 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 7161   addq(dest_lo, src1);
 7162   adcq(dest_hi, 0);
 7163   addq(dest_lo, src2);
 7164   adcq(dest_hi, 0);
 7165 }
 7166 
 7167 /**
 7168  * Multiply 64 bit by 64 bit first loop.
 7169  */
 7170 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 7171                                            Register y, Register y_idx, Register z,
 7172                                            Register carry, Register product,
 7173                                            Register idx, Register kdx) {
 7174   //
 7175   //  jlong carry, x[], y[], z[];
 7176   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7177   //    huge_128 product = y[idx] * x[xstart] + carry;
 7178   //    z[kdx] = (jlong)product;
 7179   //    carry  = (jlong)(product >>> 64);
 7180   //  }
 7181   //  z[xstart] = carry;
 7182   //
 7183 
 7184   Label L_first_loop, L_first_loop_exit;
 7185   Label L_one_x, L_one_y, L_multiply;
 7186 
 7187   decrementl(xstart);
 7188   jcc(Assembler::negative, L_one_x);
 7189 
 7190   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7191   rorq(x_xstart, 32); // convert big-endian to little-endian
 7192 
 7193   bind(L_first_loop);
 7194   decrementl(idx);
 7195   jcc(Assembler::negative, L_first_loop_exit);
 7196   decrementl(idx);
 7197   jcc(Assembler::negative, L_one_y);
 7198   movq(y_idx, Address(y, idx, Address::times_4,  0));
 7199   rorq(y_idx, 32); // convert big-endian to little-endian
 7200   bind(L_multiply);
 7201   movq(product, x_xstart);
 7202   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 7203   addq(product, carry);
 7204   adcq(rdx, 0);
 7205   subl(kdx, 2);
 7206   movl(Address(z, kdx, Address::times_4,  4), product);
 7207   shrq(product, 32);
 7208   movl(Address(z, kdx, Address::times_4,  0), product);
 7209   movq(carry, rdx);
 7210   jmp(L_first_loop);
 7211 
 7212   bind(L_one_y);
 7213   movl(y_idx, Address(y,  0));
 7214   jmp(L_multiply);
 7215 
 7216   bind(L_one_x);
 7217   movl(x_xstart, Address(x,  0));
 7218   jmp(L_first_loop);
 7219 
 7220   bind(L_first_loop_exit);
 7221 }
 7222 
 7223 /**
 7224  * Multiply 64 bit by 64 bit and add 128 bit.
 7225  */
 7226 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 7227                                             Register yz_idx, Register idx,
 7228                                             Register carry, Register product, int offset) {
 7229   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 7230   //     z[kdx] = (jlong)product;
 7231 
 7232   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 7233   rorq(yz_idx, 32); // convert big-endian to little-endian
 7234   movq(product, x_xstart);
 7235   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 7236   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 7237   rorq(yz_idx, 32); // convert big-endian to little-endian
 7238 
 7239   add2_with_carry(rdx, product, carry, yz_idx);
 7240 
 7241   movl(Address(z, idx, Address::times_4,  offset+4), product);
 7242   shrq(product, 32);
 7243   movl(Address(z, idx, Address::times_4,  offset), product);
 7244 
 7245 }
 7246 
 7247 /**
 7248  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 7249  */
 7250 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 7251                                              Register yz_idx, Register idx, Register jdx,
 7252                                              Register carry, Register product,
 7253                                              Register carry2) {
 7254   //   jlong carry, x[], y[], z[];
 7255   //   int kdx = ystart+1;
 7256   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7257   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 7258   //     z[kdx+idx+1] = (jlong)product;
 7259   //     jlong carry2  = (jlong)(product >>> 64);
 7260   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 7261   //     z[kdx+idx] = (jlong)product;
 7262   //     carry  = (jlong)(product >>> 64);
 7263   //   }
 7264   //   idx += 2;
 7265   //   if (idx > 0) {
 7266   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 7267   //     z[kdx+idx] = (jlong)product;
 7268   //     carry  = (jlong)(product >>> 64);
 7269   //   }
 7270   //
 7271 
 7272   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7273 
 7274   movl(jdx, idx);
 7275   andl(jdx, 0xFFFFFFFC);
 7276   shrl(jdx, 2);
 7277 
 7278   bind(L_third_loop);
 7279   subl(jdx, 1);
 7280   jcc(Assembler::negative, L_third_loop_exit);
 7281   subl(idx, 4);
 7282 
 7283   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 7284   movq(carry2, rdx);
 7285 
 7286   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 7287   movq(carry, rdx);
 7288   jmp(L_third_loop);
 7289 
 7290   bind (L_third_loop_exit);
 7291 
 7292   andl (idx, 0x3);
 7293   jcc(Assembler::zero, L_post_third_loop_done);
 7294 
 7295   Label L_check_1;
 7296   subl(idx, 2);
 7297   jcc(Assembler::negative, L_check_1);
 7298 
 7299   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 7300   movq(carry, rdx);
 7301 
 7302   bind (L_check_1);
 7303   addl (idx, 0x2);
 7304   andl (idx, 0x1);
 7305   subl(idx, 1);
 7306   jcc(Assembler::negative, L_post_third_loop_done);
 7307 
 7308   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 7309   movq(product, x_xstart);
 7310   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 7311   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 7312 
 7313   add2_with_carry(rdx, product, yz_idx, carry);
 7314 
 7315   movl(Address(z, idx, Address::times_4,  0), product);
 7316   shrq(product, 32);
 7317 
 7318   shlq(rdx, 32);
 7319   orq(product, rdx);
 7320   movq(carry, product);
 7321 
 7322   bind(L_post_third_loop_done);
 7323 }
 7324 
 7325 /**
 7326  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 7327  *
 7328  */
 7329 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 7330                                                   Register carry, Register carry2,
 7331                                                   Register idx, Register jdx,
 7332                                                   Register yz_idx1, Register yz_idx2,
 7333                                                   Register tmp, Register tmp3, Register tmp4) {
 7334   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 7335 
 7336   //   jlong carry, x[], y[], z[];
 7337   //   int kdx = ystart+1;
 7338   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7339   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 7340   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 7341   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 7342   //     carry  = (jlong)(tmp4 >>> 64);
 7343   //     z[kdx+idx+1] = (jlong)tmp3;
 7344   //     z[kdx+idx] = (jlong)tmp4;
 7345   //   }
 7346   //   idx += 2;
 7347   //   if (idx > 0) {
 7348   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 7349   //     z[kdx+idx] = (jlong)yz_idx1;
 7350   //     carry  = (jlong)(yz_idx1 >>> 64);
 7351   //   }
 7352   //
 7353 
 7354   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7355 
 7356   movl(jdx, idx);
 7357   andl(jdx, 0xFFFFFFFC);
 7358   shrl(jdx, 2);
 7359 
 7360   bind(L_third_loop);
 7361   subl(jdx, 1);
 7362   jcc(Assembler::negative, L_third_loop_exit);
 7363   subl(idx, 4);
 7364 
 7365   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 7366   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 7367   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 7368   rorxq(yz_idx2, yz_idx2, 32);
 7369 
 7370   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 7371   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 7372 
 7373   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 7374   rorxq(yz_idx1, yz_idx1, 32);
 7375   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7376   rorxq(yz_idx2, yz_idx2, 32);
 7377 
 7378   if (VM_Version::supports_adx()) {
 7379     adcxq(tmp3, carry);
 7380     adoxq(tmp3, yz_idx1);
 7381 
 7382     adcxq(tmp4, tmp);
 7383     adoxq(tmp4, yz_idx2);
 7384 
 7385     movl(carry, 0); // does not affect flags
 7386     adcxq(carry2, carry);
 7387     adoxq(carry2, carry);
 7388   } else {
 7389     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 7390     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 7391   }
 7392   movq(carry, carry2);
 7393 
 7394   movl(Address(z, idx, Address::times_4, 12), tmp3);
 7395   shrq(tmp3, 32);
 7396   movl(Address(z, idx, Address::times_4,  8), tmp3);
 7397 
 7398   movl(Address(z, idx, Address::times_4,  4), tmp4);
 7399   shrq(tmp4, 32);
 7400   movl(Address(z, idx, Address::times_4,  0), tmp4);
 7401 
 7402   jmp(L_third_loop);
 7403 
 7404   bind (L_third_loop_exit);
 7405 
 7406   andl (idx, 0x3);
 7407   jcc(Assembler::zero, L_post_third_loop_done);
 7408 
 7409   Label L_check_1;
 7410   subl(idx, 2);
 7411   jcc(Assembler::negative, L_check_1);
 7412 
 7413   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 7414   rorxq(yz_idx1, yz_idx1, 32);
 7415   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 7416   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7417   rorxq(yz_idx2, yz_idx2, 32);
 7418 
 7419   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 7420 
 7421   movl(Address(z, idx, Address::times_4,  4), tmp3);
 7422   shrq(tmp3, 32);
 7423   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7424   movq(carry, tmp4);
 7425 
 7426   bind (L_check_1);
 7427   addl (idx, 0x2);
 7428   andl (idx, 0x1);
 7429   subl(idx, 1);
 7430   jcc(Assembler::negative, L_post_third_loop_done);
 7431   movl(tmp4, Address(y, idx, Address::times_4,  0));
 7432   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 7433   movl(tmp4, Address(z, idx, Address::times_4,  0));
 7434 
 7435   add2_with_carry(carry2, tmp3, tmp4, carry);
 7436 
 7437   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7438   shrq(tmp3, 32);
 7439 
 7440   shlq(carry2, 32);
 7441   orq(tmp3, carry2);
 7442   movq(carry, tmp3);
 7443 
 7444   bind(L_post_third_loop_done);
 7445 }
 7446 
 7447 /**
 7448  * Code for BigInteger::multiplyToLen() intrinsic.
 7449  *
 7450  * rdi: x
 7451  * rax: xlen
 7452  * rsi: y
 7453  * rcx: ylen
 7454  * r8:  z
 7455  * r11: tmp0
 7456  * r12: tmp1
 7457  * r13: tmp2
 7458  * r14: tmp3
 7459  * r15: tmp4
 7460  * rbx: tmp5
 7461  *
 7462  */
 7463 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 7464                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 7465   ShortBranchVerifier sbv(this);
 7466   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 7467 
 7468   push(tmp0);
 7469   push(tmp1);
 7470   push(tmp2);
 7471   push(tmp3);
 7472   push(tmp4);
 7473   push(tmp5);
 7474 
 7475   push(xlen);
 7476 
 7477   const Register idx = tmp1;
 7478   const Register kdx = tmp2;
 7479   const Register xstart = tmp3;
 7480 
 7481   const Register y_idx = tmp4;
 7482   const Register carry = tmp5;
 7483   const Register product  = xlen;
 7484   const Register x_xstart = tmp0;
 7485 
 7486   // First Loop.
 7487   //
 7488   //  final static long LONG_MASK = 0xffffffffL;
 7489   //  int xstart = xlen - 1;
 7490   //  int ystart = ylen - 1;
 7491   //  long carry = 0;
 7492   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7493   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 7494   //    z[kdx] = (int)product;
 7495   //    carry = product >>> 32;
 7496   //  }
 7497   //  z[xstart] = (int)carry;
 7498   //
 7499 
 7500   movl(idx, ylen);               // idx = ylen;
 7501   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 7502   xorq(carry, carry);            // carry = 0;
 7503 
 7504   Label L_done;
 7505 
 7506   movl(xstart, xlen);
 7507   decrementl(xstart);
 7508   jcc(Assembler::negative, L_done);
 7509 
 7510   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 7511 
 7512   Label L_second_loop;
 7513   testl(kdx, kdx);
 7514   jcc(Assembler::zero, L_second_loop);
 7515 
 7516   Label L_carry;
 7517   subl(kdx, 1);
 7518   jcc(Assembler::zero, L_carry);
 7519 
 7520   movl(Address(z, kdx, Address::times_4,  0), carry);
 7521   shrq(carry, 32);
 7522   subl(kdx, 1);
 7523 
 7524   bind(L_carry);
 7525   movl(Address(z, kdx, Address::times_4,  0), carry);
 7526 
 7527   // Second and third (nested) loops.
 7528   //
 7529   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 7530   //   carry = 0;
 7531   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 7532   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 7533   //                    (z[k] & LONG_MASK) + carry;
 7534   //     z[k] = (int)product;
 7535   //     carry = product >>> 32;
 7536   //   }
 7537   //   z[i] = (int)carry;
 7538   // }
 7539   //
 7540   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 7541 
 7542   const Register jdx = tmp1;
 7543 
 7544   bind(L_second_loop);
 7545   xorl(carry, carry);    // carry = 0;
 7546   movl(jdx, ylen);       // j = ystart+1
 7547 
 7548   subl(xstart, 1);       // i = xstart-1;
 7549   jcc(Assembler::negative, L_done);
 7550 
 7551   push (z);
 7552 
 7553   Label L_last_x;
 7554   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 7555   subl(xstart, 1);       // i = xstart-1;
 7556   jcc(Assembler::negative, L_last_x);
 7557 
 7558   if (UseBMI2Instructions) {
 7559     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 7560     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 7561   } else {
 7562     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7563     rorq(x_xstart, 32);  // convert big-endian to little-endian
 7564   }
 7565 
 7566   Label L_third_loop_prologue;
 7567   bind(L_third_loop_prologue);
 7568 
 7569   push (x);
 7570   push (xstart);
 7571   push (ylen);
 7572 
 7573 
 7574   if (UseBMI2Instructions) {
 7575     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 7576   } else { // !UseBMI2Instructions
 7577     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 7578   }
 7579 
 7580   pop(ylen);
 7581   pop(xlen);
 7582   pop(x);
 7583   pop(z);
 7584 
 7585   movl(tmp3, xlen);
 7586   addl(tmp3, 1);
 7587   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7588   subl(tmp3, 1);
 7589   jccb(Assembler::negative, L_done);
 7590 
 7591   shrq(carry, 32);
 7592   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7593   jmp(L_second_loop);
 7594 
 7595   // Next infrequent code is moved outside loops.
 7596   bind(L_last_x);
 7597   if (UseBMI2Instructions) {
 7598     movl(rdx, Address(x,  0));
 7599   } else {
 7600     movl(x_xstart, Address(x,  0));
 7601   }
 7602   jmp(L_third_loop_prologue);
 7603 
 7604   bind(L_done);
 7605 
 7606   pop(xlen);
 7607 
 7608   pop(tmp5);
 7609   pop(tmp4);
 7610   pop(tmp3);
 7611   pop(tmp2);
 7612   pop(tmp1);
 7613   pop(tmp0);
 7614 }
 7615 
 7616 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 7617   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 7618   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 7619   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 7620   Label VECTOR8_TAIL, VECTOR4_TAIL;
 7621   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 7622   Label SAME_TILL_END, DONE;
 7623   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 7624 
 7625   //scale is in rcx in both Win64 and Unix
 7626   ShortBranchVerifier sbv(this);
 7627 
 7628   shlq(length);
 7629   xorq(result, result);
 7630 
 7631   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 7632       VM_Version::supports_avx512vlbw() && UseCountTrailingZerosInstruction) {
 7633     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 7634 
 7635     cmpq(length, 64);
 7636     jcc(Assembler::less, VECTOR32_TAIL);
 7637 
 7638     movq(tmp1, length);
 7639     andq(tmp1, 0x3F);      // tail count
 7640     andq(length, ~(0x3F)); //vector count
 7641 
 7642     bind(VECTOR64_LOOP);
 7643     // AVX512 code to compare 64 byte vectors.
 7644     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 7645     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7646     kortestql(k7, k7);
 7647     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 7648     addq(result, 64);
 7649     subq(length, 64);
 7650     jccb(Assembler::notZero, VECTOR64_LOOP);
 7651 
 7652     //bind(VECTOR64_TAIL);
 7653     testq(tmp1, tmp1);
 7654     jcc(Assembler::zero, SAME_TILL_END);
 7655 
 7656     //bind(VECTOR64_TAIL);
 7657     // AVX512 code to compare up to 63 byte vectors.
 7658     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 7659     shlxq(tmp2, tmp2, tmp1);
 7660     notq(tmp2);
 7661     kmovql(k3, tmp2);
 7662 
 7663     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 7664     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7665 
 7666     ktestql(k7, k3);
 7667     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 7668 
 7669     bind(VECTOR64_NOT_EQUAL);
 7670     kmovql(tmp1, k7);
 7671     notq(tmp1);
 7672     tzcntq(tmp1, tmp1);
 7673     addq(result, tmp1);
 7674     shrq(result);
 7675     jmp(DONE);
 7676     bind(VECTOR32_TAIL);
 7677   }
 7678 
 7679   cmpq(length, 8);
 7680   jcc(Assembler::equal, VECTOR8_LOOP);
 7681   jcc(Assembler::less, VECTOR4_TAIL);
 7682 
 7683   if (UseAVX >= 2) {
 7684     Label VECTOR16_TAIL, VECTOR32_LOOP;
 7685 
 7686     cmpq(length, 16);
 7687     jcc(Assembler::equal, VECTOR16_LOOP);
 7688     jcc(Assembler::less, VECTOR8_LOOP);
 7689 
 7690     cmpq(length, 32);
 7691     jccb(Assembler::less, VECTOR16_TAIL);
 7692 
 7693     subq(length, 32);
 7694     bind(VECTOR32_LOOP);
 7695     vmovdqu(rymm0, Address(obja, result));
 7696     vmovdqu(rymm1, Address(objb, result));
 7697     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 7698     vptest(rymm2, rymm2);
 7699     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7700     addq(result, 32);
 7701     subq(length, 32);
 7702     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7703     addq(length, 32);
 7704     jcc(Assembler::equal, SAME_TILL_END);
 7705     //falling through if less than 32 bytes left //close the branch here.
 7706 
 7707     bind(VECTOR16_TAIL);
 7708     cmpq(length, 16);
 7709     jccb(Assembler::less, VECTOR8_TAIL);
 7710     bind(VECTOR16_LOOP);
 7711     movdqu(rymm0, Address(obja, result));
 7712     movdqu(rymm1, Address(objb, result));
 7713     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7714     ptest(rymm2, rymm2);
 7715     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7716     addq(result, 16);
 7717     subq(length, 16);
 7718     jcc(Assembler::equal, SAME_TILL_END);
 7719     //falling through if less than 16 bytes left
 7720   } else {//regular intrinsics
 7721 
 7722     cmpq(length, 16);
 7723     jccb(Assembler::less, VECTOR8_TAIL);
 7724 
 7725     subq(length, 16);
 7726     bind(VECTOR16_LOOP);
 7727     movdqu(rymm0, Address(obja, result));
 7728     movdqu(rymm1, Address(objb, result));
 7729     pxor(rymm0, rymm1);
 7730     ptest(rymm0, rymm0);
 7731     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7732     addq(result, 16);
 7733     subq(length, 16);
 7734     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7735     addq(length, 16);
 7736     jcc(Assembler::equal, SAME_TILL_END);
 7737     //falling through if less than 16 bytes left
 7738   }
 7739 
 7740   bind(VECTOR8_TAIL);
 7741   cmpq(length, 8);
 7742   jccb(Assembler::less, VECTOR4_TAIL);
 7743   bind(VECTOR8_LOOP);
 7744   movq(tmp1, Address(obja, result));
 7745   movq(tmp2, Address(objb, result));
 7746   xorq(tmp1, tmp2);
 7747   testq(tmp1, tmp1);
 7748   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7749   addq(result, 8);
 7750   subq(length, 8);
 7751   jcc(Assembler::equal, SAME_TILL_END);
 7752   //falling through if less than 8 bytes left
 7753 
 7754   bind(VECTOR4_TAIL);
 7755   cmpq(length, 4);
 7756   jccb(Assembler::less, BYTES_TAIL);
 7757   bind(VECTOR4_LOOP);
 7758   movl(tmp1, Address(obja, result));
 7759   xorl(tmp1, Address(objb, result));
 7760   testl(tmp1, tmp1);
 7761   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7762   addq(result, 4);
 7763   subq(length, 4);
 7764   jcc(Assembler::equal, SAME_TILL_END);
 7765   //falling through if less than 4 bytes left
 7766 
 7767   bind(BYTES_TAIL);
 7768   bind(BYTES_LOOP);
 7769   load_unsigned_byte(tmp1, Address(obja, result));
 7770   load_unsigned_byte(tmp2, Address(objb, result));
 7771   xorl(tmp1, tmp2);
 7772   testl(tmp1, tmp1);
 7773   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7774   decq(length);
 7775   jcc(Assembler::zero, SAME_TILL_END);
 7776   incq(result);
 7777   load_unsigned_byte(tmp1, Address(obja, result));
 7778   load_unsigned_byte(tmp2, Address(objb, result));
 7779   xorl(tmp1, tmp2);
 7780   testl(tmp1, tmp1);
 7781   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7782   decq(length);
 7783   jcc(Assembler::zero, SAME_TILL_END);
 7784   incq(result);
 7785   load_unsigned_byte(tmp1, Address(obja, result));
 7786   load_unsigned_byte(tmp2, Address(objb, result));
 7787   xorl(tmp1, tmp2);
 7788   testl(tmp1, tmp1);
 7789   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7790   jmp(SAME_TILL_END);
 7791 
 7792   if (UseAVX >= 2) {
 7793     bind(VECTOR32_NOT_EQUAL);
 7794     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7795     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7796     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7797     vpmovmskb(tmp1, rymm0);
 7798     bsfq(tmp1, tmp1);
 7799     addq(result, tmp1);
 7800     shrq(result);
 7801     jmp(DONE);
 7802   }
 7803 
 7804   bind(VECTOR16_NOT_EQUAL);
 7805   if (UseAVX >= 2) {
 7806     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7807     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7808     pxor(rymm0, rymm2);
 7809   } else {
 7810     pcmpeqb(rymm2, rymm2);
 7811     pxor(rymm0, rymm1);
 7812     pcmpeqb(rymm0, rymm1);
 7813     pxor(rymm0, rymm2);
 7814   }
 7815   pmovmskb(tmp1, rymm0);
 7816   bsfq(tmp1, tmp1);
 7817   addq(result, tmp1);
 7818   shrq(result);
 7819   jmpb(DONE);
 7820 
 7821   bind(VECTOR8_NOT_EQUAL);
 7822   bind(VECTOR4_NOT_EQUAL);
 7823   bsfq(tmp1, tmp1);
 7824   shrq(tmp1, 3);
 7825   addq(result, tmp1);
 7826   bind(BYTES_NOT_EQUAL);
 7827   shrq(result);
 7828   jmpb(DONE);
 7829 
 7830   bind(SAME_TILL_END);
 7831   mov64(result, -1);
 7832 
 7833   bind(DONE);
 7834 }
 7835 
 7836 //Helper functions for square_to_len()
 7837 
 7838 /**
 7839  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7840  * Preserves x and z and modifies rest of the registers.
 7841  */
 7842 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7843   // Perform square and right shift by 1
 7844   // Handle odd xlen case first, then for even xlen do the following
 7845   // jlong carry = 0;
 7846   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7847   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7848   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7849   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7850   //     carry = (jlong)product;
 7851   // }
 7852 
 7853   xorq(tmp5, tmp5);     // carry
 7854   xorq(rdxReg, rdxReg);
 7855   xorl(tmp1, tmp1);     // index for x
 7856   xorl(tmp4, tmp4);     // index for z
 7857 
 7858   Label L_first_loop, L_first_loop_exit;
 7859 
 7860   testl(xlen, 1);
 7861   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7862 
 7863   // Square and right shift by 1 the odd element using 32 bit multiply
 7864   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7865   imulq(raxReg, raxReg);
 7866   shrq(raxReg, 1);
 7867   adcq(tmp5, 0);
 7868   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7869   incrementl(tmp1);
 7870   addl(tmp4, 2);
 7871 
 7872   // Square and  right shift by 1 the rest using 64 bit multiply
 7873   bind(L_first_loop);
 7874   cmpptr(tmp1, xlen);
 7875   jccb(Assembler::equal, L_first_loop_exit);
 7876 
 7877   // Square
 7878   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7879   rorq(raxReg, 32);    // convert big-endian to little-endian
 7880   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7881 
 7882   // Right shift by 1 and save carry
 7883   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7884   rcrq(rdxReg, 1);
 7885   rcrq(raxReg, 1);
 7886   adcq(tmp5, 0);
 7887 
 7888   // Store result in z
 7889   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7890   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7891 
 7892   // Update indices for x and z
 7893   addl(tmp1, 2);
 7894   addl(tmp4, 4);
 7895   jmp(L_first_loop);
 7896 
 7897   bind(L_first_loop_exit);
 7898 }
 7899 
 7900 
 7901 /**
 7902  * Perform the following multiply add operation using BMI2 instructions
 7903  * carry:sum = sum + op1*op2 + carry
 7904  * op2 should be in rdx
 7905  * op2 is preserved, all other registers are modified
 7906  */
 7907 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7908   // assert op2 is rdx
 7909   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7910   addq(sum, carry);
 7911   adcq(tmp2, 0);
 7912   addq(sum, op1);
 7913   adcq(tmp2, 0);
 7914   movq(carry, tmp2);
 7915 }
 7916 
 7917 /**
 7918  * Perform the following multiply add operation:
 7919  * carry:sum = sum + op1*op2 + carry
 7920  * Preserves op1, op2 and modifies rest of registers
 7921  */
 7922 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7923   // rdx:rax = op1 * op2
 7924   movq(raxReg, op2);
 7925   mulq(op1);
 7926 
 7927   //  rdx:rax = sum + carry + rdx:rax
 7928   addq(sum, carry);
 7929   adcq(rdxReg, 0);
 7930   addq(sum, raxReg);
 7931   adcq(rdxReg, 0);
 7932 
 7933   // carry:sum = rdx:sum
 7934   movq(carry, rdxReg);
 7935 }
 7936 
 7937 /**
 7938  * Add 64 bit long carry into z[] with carry propagation.
 7939  * Preserves z and carry register values and modifies rest of registers.
 7940  *
 7941  */
 7942 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7943   Label L_fourth_loop, L_fourth_loop_exit;
 7944 
 7945   movl(tmp1, 1);
 7946   subl(zlen, 2);
 7947   addq(Address(z, zlen, Address::times_4, 0), carry);
 7948 
 7949   bind(L_fourth_loop);
 7950   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7951   subl(zlen, 2);
 7952   jccb(Assembler::negative, L_fourth_loop_exit);
 7953   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7954   jmp(L_fourth_loop);
 7955   bind(L_fourth_loop_exit);
 7956 }
 7957 
 7958 /**
 7959  * Shift z[] left by 1 bit.
 7960  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7961  *
 7962  */
 7963 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7964 
 7965   Label L_fifth_loop, L_fifth_loop_exit;
 7966 
 7967   // Fifth loop
 7968   // Perform primitiveLeftShift(z, zlen, 1)
 7969 
 7970   const Register prev_carry = tmp1;
 7971   const Register new_carry = tmp4;
 7972   const Register value = tmp2;
 7973   const Register zidx = tmp3;
 7974 
 7975   // int zidx, carry;
 7976   // long value;
 7977   // carry = 0;
 7978   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 7979   //    (carry:value)  = (z[i] << 1) | carry ;
 7980   //    z[i] = value;
 7981   // }
 7982 
 7983   movl(zidx, zlen);
 7984   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 7985 
 7986   bind(L_fifth_loop);
 7987   decl(zidx);  // Use decl to preserve carry flag
 7988   decl(zidx);
 7989   jccb(Assembler::negative, L_fifth_loop_exit);
 7990 
 7991   if (UseBMI2Instructions) {
 7992      movq(value, Address(z, zidx, Address::times_4, 0));
 7993      rclq(value, 1);
 7994      rorxq(value, value, 32);
 7995      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7996   }
 7997   else {
 7998     // clear new_carry
 7999     xorl(new_carry, new_carry);
 8000 
 8001     // Shift z[i] by 1, or in previous carry and save new carry
 8002     movq(value, Address(z, zidx, Address::times_4, 0));
 8003     shlq(value, 1);
 8004     adcl(new_carry, 0);
 8005 
 8006     orq(value, prev_carry);
 8007     rorq(value, 0x20);
 8008     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 8009 
 8010     // Set previous carry = new carry
 8011     movl(prev_carry, new_carry);
 8012   }
 8013   jmp(L_fifth_loop);
 8014 
 8015   bind(L_fifth_loop_exit);
 8016 }
 8017 
 8018 
 8019 /**
 8020  * Code for BigInteger::squareToLen() intrinsic
 8021  *
 8022  * rdi: x
 8023  * rsi: len
 8024  * r8:  z
 8025  * rcx: zlen
 8026  * r12: tmp1
 8027  * r13: tmp2
 8028  * r14: tmp3
 8029  * r15: tmp4
 8030  * rbx: tmp5
 8031  *
 8032  */
 8033 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) {
 8034 
 8035   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 8036   push(tmp1);
 8037   push(tmp2);
 8038   push(tmp3);
 8039   push(tmp4);
 8040   push(tmp5);
 8041 
 8042   // First loop
 8043   // Store the squares, right shifted one bit (i.e., divided by 2).
 8044   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8045 
 8046   // Add in off-diagonal sums.
 8047   //
 8048   // Second, third (nested) and fourth loops.
 8049   // zlen +=2;
 8050   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 8051   //    carry = 0;
 8052   //    long op2 = x[xidx:xidx+1];
 8053   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 8054   //       k -= 2;
 8055   //       long op1 = x[j:j+1];
 8056   //       long sum = z[k:k+1];
 8057   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 8058   //       z[k:k+1] = sum;
 8059   //    }
 8060   //    add_one_64(z, k, carry, tmp_regs);
 8061   // }
 8062 
 8063   const Register carry = tmp5;
 8064   const Register sum = tmp3;
 8065   const Register op1 = tmp4;
 8066   Register op2 = tmp2;
 8067 
 8068   push(zlen);
 8069   push(len);
 8070   addl(zlen,2);
 8071   bind(L_second_loop);
 8072   xorq(carry, carry);
 8073   subl(zlen, 4);
 8074   subl(len, 2);
 8075   push(zlen);
 8076   push(len);
 8077   cmpl(len, 0);
 8078   jccb(Assembler::lessEqual, L_second_loop_exit);
 8079 
 8080   // Multiply an array by one 64 bit long.
 8081   if (UseBMI2Instructions) {
 8082     op2 = rdxReg;
 8083     movq(op2, Address(x, len, Address::times_4,  0));
 8084     rorxq(op2, op2, 32);
 8085   }
 8086   else {
 8087     movq(op2, Address(x, len, Address::times_4,  0));
 8088     rorq(op2, 32);
 8089   }
 8090 
 8091   bind(L_third_loop);
 8092   decrementl(len);
 8093   jccb(Assembler::negative, L_third_loop_exit);
 8094   decrementl(len);
 8095   jccb(Assembler::negative, L_last_x);
 8096 
 8097   movq(op1, Address(x, len, Address::times_4,  0));
 8098   rorq(op1, 32);
 8099 
 8100   bind(L_multiply);
 8101   subl(zlen, 2);
 8102   movq(sum, Address(z, zlen, Address::times_4,  0));
 8103 
 8104   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 8105   if (UseBMI2Instructions) {
 8106     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 8107   }
 8108   else {
 8109     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8110   }
 8111 
 8112   movq(Address(z, zlen, Address::times_4, 0), sum);
 8113 
 8114   jmp(L_third_loop);
 8115   bind(L_third_loop_exit);
 8116 
 8117   // Fourth loop
 8118   // Add 64 bit long carry into z with carry propagation.
 8119   // Uses offsetted zlen.
 8120   add_one_64(z, zlen, carry, tmp1);
 8121 
 8122   pop(len);
 8123   pop(zlen);
 8124   jmp(L_second_loop);
 8125 
 8126   // Next infrequent code is moved outside loops.
 8127   bind(L_last_x);
 8128   movl(op1, Address(x, 0));
 8129   jmp(L_multiply);
 8130 
 8131   bind(L_second_loop_exit);
 8132   pop(len);
 8133   pop(zlen);
 8134   pop(len);
 8135   pop(zlen);
 8136 
 8137   // Fifth loop
 8138   // Shift z left 1 bit.
 8139   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 8140 
 8141   // z[zlen-1] |= x[len-1] & 1;
 8142   movl(tmp3, Address(x, len, Address::times_4, -4));
 8143   andl(tmp3, 1);
 8144   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 8145 
 8146   pop(tmp5);
 8147   pop(tmp4);
 8148   pop(tmp3);
 8149   pop(tmp2);
 8150   pop(tmp1);
 8151 }
 8152 
 8153 /**
 8154  * Helper function for mul_add()
 8155  * Multiply the in[] by int k and add to out[] starting at offset offs using
 8156  * 128 bit by 32 bit multiply and return the carry in tmp5.
 8157  * Only quad int aligned length of in[] is operated on in this function.
 8158  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 8159  * This function preserves out, in and k registers.
 8160  * len and offset point to the appropriate index in "in" & "out" correspondingly
 8161  * tmp5 has the carry.
 8162  * other registers are temporary and are modified.
 8163  *
 8164  */
 8165 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 8166   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 8167   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8168 
 8169   Label L_first_loop, L_first_loop_exit;
 8170 
 8171   movl(tmp1, len);
 8172   shrl(tmp1, 2);
 8173 
 8174   bind(L_first_loop);
 8175   subl(tmp1, 1);
 8176   jccb(Assembler::negative, L_first_loop_exit);
 8177 
 8178   subl(len, 4);
 8179   subl(offset, 4);
 8180 
 8181   Register op2 = tmp2;
 8182   const Register sum = tmp3;
 8183   const Register op1 = tmp4;
 8184   const Register carry = tmp5;
 8185 
 8186   if (UseBMI2Instructions) {
 8187     op2 = rdxReg;
 8188   }
 8189 
 8190   movq(op1, Address(in, len, Address::times_4,  8));
 8191   rorq(op1, 32);
 8192   movq(sum, Address(out, offset, Address::times_4,  8));
 8193   rorq(sum, 32);
 8194   if (UseBMI2Instructions) {
 8195     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8196   }
 8197   else {
 8198     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8199   }
 8200   // Store back in big endian from little endian
 8201   rorq(sum, 0x20);
 8202   movq(Address(out, offset, Address::times_4,  8), sum);
 8203 
 8204   movq(op1, Address(in, len, Address::times_4,  0));
 8205   rorq(op1, 32);
 8206   movq(sum, Address(out, offset, Address::times_4,  0));
 8207   rorq(sum, 32);
 8208   if (UseBMI2Instructions) {
 8209     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8210   }
 8211   else {
 8212     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8213   }
 8214   // Store back in big endian from little endian
 8215   rorq(sum, 0x20);
 8216   movq(Address(out, offset, Address::times_4,  0), sum);
 8217 
 8218   jmp(L_first_loop);
 8219   bind(L_first_loop_exit);
 8220 }
 8221 
 8222 /**
 8223  * Code for BigInteger::mulAdd() intrinsic
 8224  *
 8225  * rdi: out
 8226  * rsi: in
 8227  * r11: offs (out.length - offset)
 8228  * rcx: len
 8229  * r8:  k
 8230  * r12: tmp1
 8231  * r13: tmp2
 8232  * r14: tmp3
 8233  * r15: tmp4
 8234  * rbx: tmp5
 8235  * Multiply the in[] by word k and add to out[], return the carry in rax
 8236  */
 8237 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 8238    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 8239    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8240 
 8241   Label L_carry, L_last_in, L_done;
 8242 
 8243 // carry = 0;
 8244 // for (int j=len-1; j >= 0; j--) {
 8245 //    long product = (in[j] & LONG_MASK) * kLong +
 8246 //                   (out[offs] & LONG_MASK) + carry;
 8247 //    out[offs--] = (int)product;
 8248 //    carry = product >>> 32;
 8249 // }
 8250 //
 8251   push(tmp1);
 8252   push(tmp2);
 8253   push(tmp3);
 8254   push(tmp4);
 8255   push(tmp5);
 8256 
 8257   Register op2 = tmp2;
 8258   const Register sum = tmp3;
 8259   const Register op1 = tmp4;
 8260   const Register carry =  tmp5;
 8261 
 8262   if (UseBMI2Instructions) {
 8263     op2 = rdxReg;
 8264     movl(op2, k);
 8265   }
 8266   else {
 8267     movl(op2, k);
 8268   }
 8269 
 8270   xorq(carry, carry);
 8271 
 8272   //First loop
 8273 
 8274   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 8275   //The carry is in tmp5
 8276   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8277 
 8278   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 8279   decrementl(len);
 8280   jccb(Assembler::negative, L_carry);
 8281   decrementl(len);
 8282   jccb(Assembler::negative, L_last_in);
 8283 
 8284   movq(op1, Address(in, len, Address::times_4,  0));
 8285   rorq(op1, 32);
 8286 
 8287   subl(offs, 2);
 8288   movq(sum, Address(out, offs, Address::times_4,  0));
 8289   rorq(sum, 32);
 8290 
 8291   if (UseBMI2Instructions) {
 8292     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8293   }
 8294   else {
 8295     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8296   }
 8297 
 8298   // Store back in big endian from little endian
 8299   rorq(sum, 0x20);
 8300   movq(Address(out, offs, Address::times_4,  0), sum);
 8301 
 8302   testl(len, len);
 8303   jccb(Assembler::zero, L_carry);
 8304 
 8305   //Multiply the last in[] entry, if any
 8306   bind(L_last_in);
 8307   movl(op1, Address(in, 0));
 8308   movl(sum, Address(out, offs, Address::times_4,  -4));
 8309 
 8310   movl(raxReg, k);
 8311   mull(op1); //tmp4 * eax -> edx:eax
 8312   addl(sum, carry);
 8313   adcl(rdxReg, 0);
 8314   addl(sum, raxReg);
 8315   adcl(rdxReg, 0);
 8316   movl(carry, rdxReg);
 8317 
 8318   movl(Address(out, offs, Address::times_4,  -4), sum);
 8319 
 8320   bind(L_carry);
 8321   //return tmp5/carry as carry in rax
 8322   movl(rax, carry);
 8323 
 8324   bind(L_done);
 8325   pop(tmp5);
 8326   pop(tmp4);
 8327   pop(tmp3);
 8328   pop(tmp2);
 8329   pop(tmp1);
 8330 }
 8331 
 8332 /**
 8333  * Emits code to update CRC-32 with a byte value according to constants in table
 8334  *
 8335  * @param [in,out]crc   Register containing the crc.
 8336  * @param [in]val       Register containing the byte to fold into the CRC.
 8337  * @param [in]table     Register containing the table of crc constants.
 8338  *
 8339  * uint32_t crc;
 8340  * val = crc_table[(val ^ crc) & 0xFF];
 8341  * crc = val ^ (crc >> 8);
 8342  *
 8343  */
 8344 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 8345   xorl(val, crc);
 8346   andl(val, 0xFF);
 8347   shrl(crc, 8); // unsigned shift
 8348   xorl(crc, Address(table, val, Address::times_4, 0));
 8349 }
 8350 
 8351 /**
 8352  * Fold 128-bit data chunk
 8353  */
 8354 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 8355   if (UseAVX > 0) {
 8356     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 8357     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 8358     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 8359     pxor(xcrc, xtmp);
 8360   } else {
 8361     movdqa(xtmp, xcrc);
 8362     pclmulhdq(xtmp, xK);   // [123:64]
 8363     pclmulldq(xcrc, xK);   // [63:0]
 8364     pxor(xcrc, xtmp);
 8365     movdqu(xtmp, Address(buf, offset));
 8366     pxor(xcrc, xtmp);
 8367   }
 8368 }
 8369 
 8370 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 8371   if (UseAVX > 0) {
 8372     vpclmulhdq(xtmp, xK, xcrc);
 8373     vpclmulldq(xcrc, xK, xcrc);
 8374     pxor(xcrc, xbuf);
 8375     pxor(xcrc, xtmp);
 8376   } else {
 8377     movdqa(xtmp, xcrc);
 8378     pclmulhdq(xtmp, xK);
 8379     pclmulldq(xcrc, xK);
 8380     pxor(xcrc, xbuf);
 8381     pxor(xcrc, xtmp);
 8382   }
 8383 }
 8384 
 8385 /**
 8386  * 8-bit folds to compute 32-bit CRC
 8387  *
 8388  * uint64_t xcrc;
 8389  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 8390  */
 8391 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 8392   movdl(tmp, xcrc);
 8393   andl(tmp, 0xFF);
 8394   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 8395   psrldq(xcrc, 1); // unsigned shift one byte
 8396   pxor(xcrc, xtmp);
 8397 }
 8398 
 8399 /**
 8400  * uint32_t crc;
 8401  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 8402  */
 8403 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 8404   movl(tmp, crc);
 8405   andl(tmp, 0xFF);
 8406   shrl(crc, 8);
 8407   xorl(crc, Address(table, tmp, Address::times_4, 0));
 8408 }
 8409 
 8410 /**
 8411  * @param crc   register containing existing CRC (32-bit)
 8412  * @param buf   register pointing to input byte buffer (byte*)
 8413  * @param len   register containing number of bytes
 8414  * @param table register that will contain address of CRC table
 8415  * @param tmp   scratch register
 8416  */
 8417 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 8418   assert_different_registers(crc, buf, len, table, tmp, rax);
 8419 
 8420   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8421   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8422 
 8423   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8424   // context for the registers used, where all instructions below are using 128-bit mode
 8425   // On EVEX without VL and BW, these instructions will all be AVX.
 8426   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 8427   notl(crc); // ~crc
 8428   cmpl(len, 16);
 8429   jcc(Assembler::less, L_tail);
 8430 
 8431   // Align buffer to 16 bytes
 8432   movl(tmp, buf);
 8433   andl(tmp, 0xF);
 8434   jccb(Assembler::zero, L_aligned);
 8435   subl(tmp,  16);
 8436   addl(len, tmp);
 8437 
 8438   align(4);
 8439   BIND(L_align_loop);
 8440   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8441   update_byte_crc32(crc, rax, table);
 8442   increment(buf);
 8443   incrementl(tmp);
 8444   jccb(Assembler::less, L_align_loop);
 8445 
 8446   BIND(L_aligned);
 8447   movl(tmp, len); // save
 8448   shrl(len, 4);
 8449   jcc(Assembler::zero, L_tail_restore);
 8450 
 8451   // Fold crc into first bytes of vector
 8452   movdqa(xmm1, Address(buf, 0));
 8453   movdl(rax, xmm1);
 8454   xorl(crc, rax);
 8455   if (VM_Version::supports_sse4_1()) {
 8456     pinsrd(xmm1, crc, 0);
 8457   } else {
 8458     pinsrw(xmm1, crc, 0);
 8459     shrl(crc, 16);
 8460     pinsrw(xmm1, crc, 1);
 8461   }
 8462   addptr(buf, 16);
 8463   subl(len, 4); // len > 0
 8464   jcc(Assembler::less, L_fold_tail);
 8465 
 8466   movdqa(xmm2, Address(buf,  0));
 8467   movdqa(xmm3, Address(buf, 16));
 8468   movdqa(xmm4, Address(buf, 32));
 8469   addptr(buf, 48);
 8470   subl(len, 3);
 8471   jcc(Assembler::lessEqual, L_fold_512b);
 8472 
 8473   // Fold total 512 bits of polynomial on each iteration,
 8474   // 128 bits per each of 4 parallel streams.
 8475   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 8476 
 8477   align32();
 8478   BIND(L_fold_512b_loop);
 8479   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8480   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 8481   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 8482   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 8483   addptr(buf, 64);
 8484   subl(len, 4);
 8485   jcc(Assembler::greater, L_fold_512b_loop);
 8486 
 8487   // Fold 512 bits to 128 bits.
 8488   BIND(L_fold_512b);
 8489   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8490   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 8491   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 8492   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 8493 
 8494   // Fold the rest of 128 bits data chunks
 8495   BIND(L_fold_tail);
 8496   addl(len, 3);
 8497   jccb(Assembler::lessEqual, L_fold_128b);
 8498   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8499 
 8500   BIND(L_fold_tail_loop);
 8501   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8502   addptr(buf, 16);
 8503   decrementl(len);
 8504   jccb(Assembler::greater, L_fold_tail_loop);
 8505 
 8506   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 8507   BIND(L_fold_128b);
 8508   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 8509   if (UseAVX > 0) {
 8510     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 8511     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 8512     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 8513   } else {
 8514     movdqa(xmm2, xmm0);
 8515     pclmulqdq(xmm2, xmm1, 0x1);
 8516     movdqa(xmm3, xmm0);
 8517     pand(xmm3, xmm2);
 8518     pclmulqdq(xmm0, xmm3, 0x1);
 8519   }
 8520   psrldq(xmm1, 8);
 8521   psrldq(xmm2, 4);
 8522   pxor(xmm0, xmm1);
 8523   pxor(xmm0, xmm2);
 8524 
 8525   // 8 8-bit folds to compute 32-bit CRC.
 8526   for (int j = 0; j < 4; j++) {
 8527     fold_8bit_crc32(xmm0, table, xmm1, rax);
 8528   }
 8529   movdl(crc, xmm0); // mov 32 bits to general register
 8530   for (int j = 0; j < 4; j++) {
 8531     fold_8bit_crc32(crc, table, rax);
 8532   }
 8533 
 8534   BIND(L_tail_restore);
 8535   movl(len, tmp); // restore
 8536   BIND(L_tail);
 8537   andl(len, 0xf);
 8538   jccb(Assembler::zero, L_exit);
 8539 
 8540   // Fold the rest of bytes
 8541   align(4);
 8542   BIND(L_tail_loop);
 8543   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8544   update_byte_crc32(crc, rax, table);
 8545   increment(buf);
 8546   decrementl(len);
 8547   jccb(Assembler::greater, L_tail_loop);
 8548 
 8549   BIND(L_exit);
 8550   notl(crc); // ~c
 8551 }
 8552 
 8553 // Helper function for AVX 512 CRC32
 8554 // Fold 512-bit data chunks
 8555 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 8556                                              Register pos, int offset) {
 8557   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 8558   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 8559   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 8560   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 8561   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 8562 }
 8563 
 8564 // Helper function for AVX 512 CRC32
 8565 // Compute CRC32 for < 256B buffers
 8566 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 8567                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 8568                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 8569 
 8570   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 8571   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 8572   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 8573 
 8574   // check if there is enough buffer to be able to fold 16B at a time
 8575   cmpl(len, 32);
 8576   jcc(Assembler::less, L_less_than_32);
 8577 
 8578   // if there is, load the constants
 8579   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 8580   movdl(xmm0, crc);                        // get the initial crc value
 8581   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8582   pxor(xmm7, xmm0);
 8583 
 8584   // update the buffer pointer
 8585   addl(pos, 16);
 8586   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 8587   subl(len, 32);
 8588   jmp(L_16B_reduction_loop);
 8589 
 8590   bind(L_less_than_32);
 8591   //mov initial crc to the return value. this is necessary for zero - length buffers.
 8592   movl(rax, crc);
 8593   testl(len, len);
 8594   jcc(Assembler::equal, L_cleanup);
 8595 
 8596   movdl(xmm0, crc);                        //get the initial crc value
 8597 
 8598   cmpl(len, 16);
 8599   jcc(Assembler::equal, L_exact_16_left);
 8600   jcc(Assembler::less, L_less_than_16_left);
 8601 
 8602   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8603   pxor(xmm7, xmm0);                       //xor the initial crc value
 8604   addl(pos, 16);
 8605   subl(len, 16);
 8606   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 8607   jmp(L_get_last_two_xmms);
 8608 
 8609   bind(L_less_than_16_left);
 8610   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 8611   pxor(xmm1, xmm1);
 8612   movptr(tmp1, rsp);
 8613   movdqu(Address(tmp1, 0 * 16), xmm1);
 8614 
 8615   cmpl(len, 4);
 8616   jcc(Assembler::less, L_only_less_than_4);
 8617 
 8618   //backup the counter value
 8619   movl(tmp2, len);
 8620   cmpl(len, 8);
 8621   jcc(Assembler::less, L_less_than_8_left);
 8622 
 8623   //load 8 Bytes
 8624   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 8625   movq(Address(tmp1, 0 * 16), rax);
 8626   addptr(tmp1, 8);
 8627   subl(len, 8);
 8628   addl(pos, 8);
 8629 
 8630   bind(L_less_than_8_left);
 8631   cmpl(len, 4);
 8632   jcc(Assembler::less, L_less_than_4_left);
 8633 
 8634   //load 4 Bytes
 8635   movl(rax, Address(buf, pos, Address::times_1, 0));
 8636   movl(Address(tmp1, 0 * 16), rax);
 8637   addptr(tmp1, 4);
 8638   subl(len, 4);
 8639   addl(pos, 4);
 8640 
 8641   bind(L_less_than_4_left);
 8642   cmpl(len, 2);
 8643   jcc(Assembler::less, L_less_than_2_left);
 8644 
 8645   // load 2 Bytes
 8646   movw(rax, Address(buf, pos, Address::times_1, 0));
 8647   movl(Address(tmp1, 0 * 16), rax);
 8648   addptr(tmp1, 2);
 8649   subl(len, 2);
 8650   addl(pos, 2);
 8651 
 8652   bind(L_less_than_2_left);
 8653   cmpl(len, 1);
 8654   jcc(Assembler::less, L_zero_left);
 8655 
 8656   // load 1 Byte
 8657   movb(rax, Address(buf, pos, Address::times_1, 0));
 8658   movb(Address(tmp1, 0 * 16), rax);
 8659 
 8660   bind(L_zero_left);
 8661   movdqu(xmm7, Address(rsp, 0));
 8662   pxor(xmm7, xmm0);                       //xor the initial crc value
 8663 
 8664   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8665   movdqu(xmm0, Address(rax, tmp2));
 8666   pshufb(xmm7, xmm0);
 8667   jmp(L_128_done);
 8668 
 8669   bind(L_exact_16_left);
 8670   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 8671   pxor(xmm7, xmm0);                       //xor the initial crc value
 8672   jmp(L_128_done);
 8673 
 8674   bind(L_only_less_than_4);
 8675   cmpl(len, 3);
 8676   jcc(Assembler::less, L_only_less_than_3);
 8677 
 8678   // load 3 Bytes
 8679   movb(rax, Address(buf, pos, Address::times_1, 0));
 8680   movb(Address(tmp1, 0), rax);
 8681 
 8682   movb(rax, Address(buf, pos, Address::times_1, 1));
 8683   movb(Address(tmp1, 1), rax);
 8684 
 8685   movb(rax, Address(buf, pos, Address::times_1, 2));
 8686   movb(Address(tmp1, 2), rax);
 8687 
 8688   movdqu(xmm7, Address(rsp, 0));
 8689   pxor(xmm7, xmm0);                     //xor the initial crc value
 8690 
 8691   pslldq(xmm7, 0x5);
 8692   jmp(L_barrett);
 8693   bind(L_only_less_than_3);
 8694   cmpl(len, 2);
 8695   jcc(Assembler::less, L_only_less_than_2);
 8696 
 8697   // load 2 Bytes
 8698   movb(rax, Address(buf, pos, Address::times_1, 0));
 8699   movb(Address(tmp1, 0), rax);
 8700 
 8701   movb(rax, Address(buf, pos, Address::times_1, 1));
 8702   movb(Address(tmp1, 1), rax);
 8703 
 8704   movdqu(xmm7, Address(rsp, 0));
 8705   pxor(xmm7, xmm0);                     //xor the initial crc value
 8706 
 8707   pslldq(xmm7, 0x6);
 8708   jmp(L_barrett);
 8709 
 8710   bind(L_only_less_than_2);
 8711   //load 1 Byte
 8712   movb(rax, Address(buf, pos, Address::times_1, 0));
 8713   movb(Address(tmp1, 0), rax);
 8714 
 8715   movdqu(xmm7, Address(rsp, 0));
 8716   pxor(xmm7, xmm0);                     //xor the initial crc value
 8717 
 8718   pslldq(xmm7, 0x7);
 8719 }
 8720 
 8721 /**
 8722 * Compute CRC32 using AVX512 instructions
 8723 * param crc   register containing existing CRC (32-bit)
 8724 * param buf   register pointing to input byte buffer (byte*)
 8725 * param len   register containing number of bytes
 8726 * param table address of crc or crc32c table
 8727 * param tmp1  scratch register
 8728 * param tmp2  scratch register
 8729 * return rax  result register
 8730 *
 8731 * This routine is identical for crc32c with the exception of the precomputed constant
 8732 * table which will be passed as the table argument.  The calculation steps are
 8733 * the same for both variants.
 8734 */
 8735 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8736   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8737 
 8738   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8739   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8740   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8741   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8742   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8743 
 8744   const Register pos = r12;
 8745   push(r12);
 8746   subptr(rsp, 16 * 2 + 8);
 8747 
 8748   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8749   // context for the registers used, where all instructions below are using 128-bit mode
 8750   // On EVEX without VL and BW, these instructions will all be AVX.
 8751   movl(pos, 0);
 8752 
 8753   // check if smaller than 256B
 8754   cmpl(len, 256);
 8755   jcc(Assembler::less, L_less_than_256);
 8756 
 8757   // load the initial crc value
 8758   movdl(xmm10, crc);
 8759 
 8760   // receive the initial 64B data, xor the initial crc value
 8761   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8762   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8763   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8764   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8765 
 8766   subl(len, 256);
 8767   cmpl(len, 256);
 8768   jcc(Assembler::less, L_fold_128_B_loop);
 8769 
 8770   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8771   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8772   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8773   subl(len, 256);
 8774 
 8775   bind(L_fold_256_B_loop);
 8776   addl(pos, 256);
 8777   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8778   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8779   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8780   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8781 
 8782   subl(len, 256);
 8783   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8784 
 8785   // Fold 256 into 128
 8786   addl(pos, 256);
 8787   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8788   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8789   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8790 
 8791   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8792   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8793   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8794 
 8795   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8796   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8797 
 8798   addl(len, 128);
 8799   jmp(L_fold_128_B_register);
 8800 
 8801   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8802   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8803 
 8804   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8805   bind(L_fold_128_B_loop);
 8806   addl(pos, 128);
 8807   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8808   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8809 
 8810   subl(len, 128);
 8811   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8812 
 8813   addl(pos, 128);
 8814 
 8815   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8816   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8817   bind(L_fold_128_B_register);
 8818   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8819   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8820   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8821   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8822   // save last that has no multiplicand
 8823   vextracti64x2(xmm7, xmm4, 3);
 8824 
 8825   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8826   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8827   // Needed later in reduction loop
 8828   movdqu(xmm10, Address(table, 1 * 16));
 8829   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8830   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8831 
 8832   // Swap 1,0,3,2 - 01 00 11 10
 8833   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8834   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8835   vextracti128(xmm5, xmm8, 1);
 8836   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8837 
 8838   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8839   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8840   addl(len, 128 - 16);
 8841   jcc(Assembler::less, L_final_reduction_for_128);
 8842 
 8843   bind(L_16B_reduction_loop);
 8844   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8845   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8846   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8847   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8848   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8849   addl(pos, 16);
 8850   subl(len, 16);
 8851   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8852 
 8853   bind(L_final_reduction_for_128);
 8854   addl(len, 16);
 8855   jcc(Assembler::equal, L_128_done);
 8856 
 8857   bind(L_get_last_two_xmms);
 8858   movdqu(xmm2, xmm7);
 8859   addl(pos, len);
 8860   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8861   subl(pos, len);
 8862 
 8863   // get rid of the extra data that was loaded before
 8864   // load the shift constant
 8865   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8866   movdqu(xmm0, Address(rax, len));
 8867   addl(rax, len);
 8868 
 8869   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8870   //Change mask to 512
 8871   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8872   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8873 
 8874   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8875   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8876   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8877   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8878   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8879 
 8880   bind(L_128_done);
 8881   // compute crc of a 128-bit value
 8882   movdqu(xmm10, Address(table, 3 * 16));
 8883   movdqu(xmm0, xmm7);
 8884 
 8885   // 64b fold
 8886   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8887   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8888   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8889 
 8890   // 32b fold
 8891   movdqu(xmm0, xmm7);
 8892   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8893   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8894   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8895   jmp(L_barrett);
 8896 
 8897   bind(L_less_than_256);
 8898   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);
 8899 
 8900   //barrett reduction
 8901   bind(L_barrett);
 8902   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8903   movdqu(xmm1, xmm7);
 8904   movdqu(xmm2, xmm7);
 8905   movdqu(xmm10, Address(table, 4 * 16));
 8906 
 8907   pclmulqdq(xmm7, xmm10, 0x0);
 8908   pxor(xmm7, xmm2);
 8909   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8910   movdqu(xmm2, xmm7);
 8911   pclmulqdq(xmm7, xmm10, 0x10);
 8912   pxor(xmm7, xmm2);
 8913   pxor(xmm7, xmm1);
 8914   pextrd(crc, xmm7, 2);
 8915 
 8916   bind(L_cleanup);
 8917   addptr(rsp, 16 * 2 + 8);
 8918   pop(r12);
 8919 }
 8920 
 8921 // S. Gueron / Information Processing Letters 112 (2012) 184
 8922 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8923 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8924 // Output: the 64-bit carry-less product of B * CONST
 8925 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8926                                      Register tmp1, Register tmp2, Register tmp3) {
 8927   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8928   if (n > 0) {
 8929     addq(tmp3, n * 256 * 8);
 8930   }
 8931   //    Q1 = TABLEExt[n][B & 0xFF];
 8932   movl(tmp1, in);
 8933   andl(tmp1, 0x000000FF);
 8934   shll(tmp1, 3);
 8935   addq(tmp1, tmp3);
 8936   movq(tmp1, Address(tmp1, 0));
 8937 
 8938   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8939   movl(tmp2, in);
 8940   shrl(tmp2, 8);
 8941   andl(tmp2, 0x000000FF);
 8942   shll(tmp2, 3);
 8943   addq(tmp2, tmp3);
 8944   movq(tmp2, Address(tmp2, 0));
 8945 
 8946   shlq(tmp2, 8);
 8947   xorq(tmp1, tmp2);
 8948 
 8949   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8950   movl(tmp2, in);
 8951   shrl(tmp2, 16);
 8952   andl(tmp2, 0x000000FF);
 8953   shll(tmp2, 3);
 8954   addq(tmp2, tmp3);
 8955   movq(tmp2, Address(tmp2, 0));
 8956 
 8957   shlq(tmp2, 16);
 8958   xorq(tmp1, tmp2);
 8959 
 8960   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8961   shrl(in, 24);
 8962   andl(in, 0x000000FF);
 8963   shll(in, 3);
 8964   addq(in, tmp3);
 8965   movq(in, Address(in, 0));
 8966 
 8967   shlq(in, 24);
 8968   xorq(in, tmp1);
 8969   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8970 }
 8971 
 8972 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8973                                       Register in_out,
 8974                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8975                                       XMMRegister w_xtmp2,
 8976                                       Register tmp1,
 8977                                       Register n_tmp2, Register n_tmp3) {
 8978   if (is_pclmulqdq_supported) {
 8979     movdl(w_xtmp1, in_out); // modified blindly
 8980 
 8981     movl(tmp1, const_or_pre_comp_const_index);
 8982     movdl(w_xtmp2, tmp1);
 8983     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8984 
 8985     movdq(in_out, w_xtmp1);
 8986   } else {
 8987     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 8988   }
 8989 }
 8990 
 8991 // Recombination Alternative 2: No bit-reflections
 8992 // T1 = (CRC_A * U1) << 1
 8993 // T2 = (CRC_B * U2) << 1
 8994 // C1 = T1 >> 32
 8995 // C2 = T2 >> 32
 8996 // T1 = T1 & 0xFFFFFFFF
 8997 // T2 = T2 & 0xFFFFFFFF
 8998 // T1 = CRC32(0, T1)
 8999 // T2 = CRC32(0, T2)
 9000 // C1 = C1 ^ T1
 9001 // C2 = C2 ^ T2
 9002 // CRC = C1 ^ C2 ^ CRC_C
 9003 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,
 9004                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9005                                      Register tmp1, Register tmp2,
 9006                                      Register n_tmp3) {
 9007   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 9008   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 9009   shlq(in_out, 1);
 9010   movl(tmp1, in_out);
 9011   shrq(in_out, 32);
 9012   xorl(tmp2, tmp2);
 9013   crc32(tmp2, tmp1, 4);
 9014   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 9015   shlq(in1, 1);
 9016   movl(tmp1, in1);
 9017   shrq(in1, 32);
 9018   xorl(tmp2, tmp2);
 9019   crc32(tmp2, tmp1, 4);
 9020   xorl(in1, tmp2);
 9021   xorl(in_out, in1);
 9022   xorl(in_out, in2);
 9023 }
 9024 
 9025 // Set N to predefined value
 9026 // Subtract from a length of a buffer
 9027 // execute in a loop:
 9028 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 9029 // for i = 1 to N do
 9030 //  CRC_A = CRC32(CRC_A, A[i])
 9031 //  CRC_B = CRC32(CRC_B, B[i])
 9032 //  CRC_C = CRC32(CRC_C, C[i])
 9033 // end for
 9034 // Recombine
 9035 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,
 9036                                        Register in_out1, Register in_out2, Register in_out3,
 9037                                        Register tmp1, Register tmp2, Register tmp3,
 9038                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9039                                        Register tmp4, Register tmp5,
 9040                                        Register n_tmp6) {
 9041   Label L_processPartitions;
 9042   Label L_processPartition;
 9043   Label L_exit;
 9044 
 9045   bind(L_processPartitions);
 9046   cmpl(in_out1, 3 * size);
 9047   jcc(Assembler::less, L_exit);
 9048     xorl(tmp1, tmp1);
 9049     xorl(tmp2, tmp2);
 9050     movq(tmp3, in_out2);
 9051     addq(tmp3, size);
 9052 
 9053     bind(L_processPartition);
 9054       crc32(in_out3, Address(in_out2, 0), 8);
 9055       crc32(tmp1, Address(in_out2, size), 8);
 9056       crc32(tmp2, Address(in_out2, size * 2), 8);
 9057       addq(in_out2, 8);
 9058       cmpq(in_out2, tmp3);
 9059       jcc(Assembler::less, L_processPartition);
 9060     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 9061             w_xtmp1, w_xtmp2, w_xtmp3,
 9062             tmp4, tmp5,
 9063             n_tmp6);
 9064     addq(in_out2, 2 * size);
 9065     subl(in_out1, 3 * size);
 9066     jmp(L_processPartitions);
 9067 
 9068   bind(L_exit);
 9069 }
 9070 
 9071 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 9072 // Input: A buffer I of L bytes.
 9073 // Output: the CRC32C value of the buffer.
 9074 // Notations:
 9075 // Write L = 24N + r, with N = floor (L/24).
 9076 // r = L mod 24 (0 <= r < 24).
 9077 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 9078 // N quadwords, and R consists of r bytes.
 9079 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 9080 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 9081 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 9082 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 9083 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 9084                                           Register tmp1, Register tmp2, Register tmp3,
 9085                                           Register tmp4, Register tmp5, Register tmp6,
 9086                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9087                                           bool is_pclmulqdq_supported) {
 9088   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 9089   Label L_wordByWord;
 9090   Label L_byteByByteProlog;
 9091   Label L_byteByByte;
 9092   Label L_exit;
 9093 
 9094   if (is_pclmulqdq_supported ) {
 9095     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 9096     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 9097 
 9098     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 9099     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 9100 
 9101     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 9102     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 9103     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 9104   } else {
 9105     const_or_pre_comp_const_index[0] = 1;
 9106     const_or_pre_comp_const_index[1] = 0;
 9107 
 9108     const_or_pre_comp_const_index[2] = 3;
 9109     const_or_pre_comp_const_index[3] = 2;
 9110 
 9111     const_or_pre_comp_const_index[4] = 5;
 9112     const_or_pre_comp_const_index[5] = 4;
 9113    }
 9114   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 9115                     in2, in1, in_out,
 9116                     tmp1, tmp2, tmp3,
 9117                     w_xtmp1, w_xtmp2, w_xtmp3,
 9118                     tmp4, tmp5,
 9119                     tmp6);
 9120   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 9121                     in2, in1, in_out,
 9122                     tmp1, tmp2, tmp3,
 9123                     w_xtmp1, w_xtmp2, w_xtmp3,
 9124                     tmp4, tmp5,
 9125                     tmp6);
 9126   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 9127                     in2, in1, in_out,
 9128                     tmp1, tmp2, tmp3,
 9129                     w_xtmp1, w_xtmp2, w_xtmp3,
 9130                     tmp4, tmp5,
 9131                     tmp6);
 9132   movl(tmp1, in2);
 9133   andl(tmp1, 0x00000007);
 9134   negl(tmp1);
 9135   addl(tmp1, in2);
 9136   addq(tmp1, in1);
 9137 
 9138   cmpq(in1, tmp1);
 9139   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 9140   align(16);
 9141   BIND(L_wordByWord);
 9142     crc32(in_out, Address(in1, 0), 8);
 9143     addq(in1, 8);
 9144     cmpq(in1, tmp1);
 9145     jcc(Assembler::less, L_wordByWord);
 9146 
 9147   BIND(L_byteByByteProlog);
 9148   andl(in2, 0x00000007);
 9149   movl(tmp2, 1);
 9150 
 9151   cmpl(tmp2, in2);
 9152   jccb(Assembler::greater, L_exit);
 9153   BIND(L_byteByByte);
 9154     crc32(in_out, Address(in1, 0), 1);
 9155     incq(in1);
 9156     incl(tmp2);
 9157     cmpl(tmp2, in2);
 9158     jcc(Assembler::lessEqual, L_byteByByte);
 9159 
 9160   BIND(L_exit);
 9161 }
 9162 #undef BIND
 9163 #undef BLOCK_COMMENT
 9164 
 9165 // Compress char[] array to byte[].
 9166 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 9167 // Return the array length if every element in array can be encoded,
 9168 // otherwise, the index of first non-latin1 (> 0xff) character.
 9169 //   @IntrinsicCandidate
 9170 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 9171 //     for (int i = 0; i < len; i++) {
 9172 //       char c = src[srcOff];
 9173 //       if (c > 0xff) {
 9174 //           return i;  // return index of non-latin1 char
 9175 //       }
 9176 //       dst[dstOff] = (byte)c;
 9177 //       srcOff++;
 9178 //       dstOff++;
 9179 //     }
 9180 //     return len;
 9181 //   }
 9182 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 9183   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 9184   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 9185   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 9186   Label copy_chars_loop, done, reset_sp, copy_tail;
 9187 
 9188   // rsi: src
 9189   // rdi: dst
 9190   // rdx: len
 9191   // rcx: tmp5
 9192   // rax: result
 9193 
 9194   // rsi holds start addr of source char[] to be compressed
 9195   // rdi holds start addr of destination byte[]
 9196   // rdx holds length
 9197 
 9198   assert(len != result, "");
 9199 
 9200   // save length for return
 9201   movl(result, len);
 9202 
 9203   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 9204     VM_Version::supports_avx512vlbw() &&
 9205     VM_Version::supports_bmi2()) {
 9206 
 9207     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 9208 
 9209     // alignment
 9210     Label post_alignment;
 9211 
 9212     // if length of the string is less than 32, handle it the old fashioned way
 9213     testl(len, -32);
 9214     jcc(Assembler::zero, below_threshold);
 9215 
 9216     // First check whether a character is compressible ( <= 0xFF).
 9217     // Create mask to test for Unicode chars inside zmm vector
 9218     movl(tmp5, 0x00FF);
 9219     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 9220 
 9221     testl(len, -64);
 9222     jccb(Assembler::zero, post_alignment);
 9223 
 9224     movl(tmp5, dst);
 9225     andl(tmp5, (32 - 1));
 9226     negl(tmp5);
 9227     andl(tmp5, (32 - 1));
 9228 
 9229     // bail out when there is nothing to be done
 9230     testl(tmp5, 0xFFFFFFFF);
 9231     jccb(Assembler::zero, post_alignment);
 9232 
 9233     // ~(~0 << len), where len is the # of remaining elements to process
 9234     movl(len, 0xFFFFFFFF);
 9235     shlxl(len, len, tmp5);
 9236     notl(len);
 9237     kmovdl(mask2, len);
 9238     movl(len, result);
 9239 
 9240     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9241     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9242     ktestd(mask1, mask2);
 9243     jcc(Assembler::carryClear, copy_tail);
 9244 
 9245     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9246 
 9247     addptr(src, tmp5);
 9248     addptr(src, tmp5);
 9249     addptr(dst, tmp5);
 9250     subl(len, tmp5);
 9251 
 9252     bind(post_alignment);
 9253     // end of alignment
 9254 
 9255     movl(tmp5, len);
 9256     andl(tmp5, (32 - 1));    // tail count (in chars)
 9257     andl(len, ~(32 - 1));    // vector count (in chars)
 9258     jccb(Assembler::zero, copy_loop_tail);
 9259 
 9260     lea(src, Address(src, len, Address::times_2));
 9261     lea(dst, Address(dst, len, Address::times_1));
 9262     negptr(len);
 9263 
 9264     bind(copy_32_loop);
 9265     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 9266     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 9267     kortestdl(mask1, mask1);
 9268     jccb(Assembler::carryClear, reset_for_copy_tail);
 9269 
 9270     // All elements in current processed chunk are valid candidates for
 9271     // compression. Write a truncated byte elements to the memory.
 9272     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 9273     addptr(len, 32);
 9274     jccb(Assembler::notZero, copy_32_loop);
 9275 
 9276     bind(copy_loop_tail);
 9277     // bail out when there is nothing to be done
 9278     testl(tmp5, 0xFFFFFFFF);
 9279     jcc(Assembler::zero, done);
 9280 
 9281     movl(len, tmp5);
 9282 
 9283     // ~(~0 << len), where len is the # of remaining elements to process
 9284     movl(tmp5, 0xFFFFFFFF);
 9285     shlxl(tmp5, tmp5, len);
 9286     notl(tmp5);
 9287 
 9288     kmovdl(mask2, tmp5);
 9289 
 9290     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9291     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9292     ktestd(mask1, mask2);
 9293     jcc(Assembler::carryClear, copy_tail);
 9294 
 9295     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9296     jmp(done);
 9297 
 9298     bind(reset_for_copy_tail);
 9299     lea(src, Address(src, tmp5, Address::times_2));
 9300     lea(dst, Address(dst, tmp5, Address::times_1));
 9301     subptr(len, tmp5);
 9302     jmp(copy_chars_loop);
 9303 
 9304     bind(below_threshold);
 9305   }
 9306 
 9307   if (UseSSE42Intrinsics) {
 9308     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 9309 
 9310     // vectored compression
 9311     testl(len, 0xfffffff8);
 9312     jcc(Assembler::zero, copy_tail);
 9313 
 9314     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 9315     movdl(tmp1Reg, tmp5);
 9316     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 9317 
 9318     andl(len, 0xfffffff0);
 9319     jccb(Assembler::zero, copy_16);
 9320 
 9321     // compress 16 chars per iter
 9322     pxor(tmp4Reg, tmp4Reg);
 9323 
 9324     lea(src, Address(src, len, Address::times_2));
 9325     lea(dst, Address(dst, len, Address::times_1));
 9326     negptr(len);
 9327 
 9328     bind(copy_32_loop);
 9329     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 9330     por(tmp4Reg, tmp2Reg);
 9331     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 9332     por(tmp4Reg, tmp3Reg);
 9333     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 9334     jccb(Assembler::notZero, reset_for_copy_tail);
 9335     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 9336     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 9337     addptr(len, 16);
 9338     jccb(Assembler::notZero, copy_32_loop);
 9339 
 9340     // compress next vector of 8 chars (if any)
 9341     bind(copy_16);
 9342     // len = 0
 9343     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 9344     jccb(Assembler::zero, copy_tail_sse);
 9345 
 9346     pxor(tmp3Reg, tmp3Reg);
 9347 
 9348     movdqu(tmp2Reg, Address(src, 0));
 9349     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 9350     jccb(Assembler::notZero, reset_for_copy_tail);
 9351     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 9352     movq(Address(dst, 0), tmp2Reg);
 9353     addptr(src, 16);
 9354     addptr(dst, 8);
 9355     jmpb(copy_tail_sse);
 9356 
 9357     bind(reset_for_copy_tail);
 9358     movl(tmp5, result);
 9359     andl(tmp5, 0x0000000f);
 9360     lea(src, Address(src, tmp5, Address::times_2));
 9361     lea(dst, Address(dst, tmp5, Address::times_1));
 9362     subptr(len, tmp5);
 9363     jmpb(copy_chars_loop);
 9364 
 9365     bind(copy_tail_sse);
 9366     movl(len, result);
 9367     andl(len, 0x00000007);    // tail count (in chars)
 9368   }
 9369   // compress 1 char per iter
 9370   bind(copy_tail);
 9371   testl(len, len);
 9372   jccb(Assembler::zero, done);
 9373   lea(src, Address(src, len, Address::times_2));
 9374   lea(dst, Address(dst, len, Address::times_1));
 9375   negptr(len);
 9376 
 9377   bind(copy_chars_loop);
 9378   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 9379   testl(tmp5, 0xff00);      // check if Unicode char
 9380   jccb(Assembler::notZero, reset_sp);
 9381   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 9382   increment(len);
 9383   jccb(Assembler::notZero, copy_chars_loop);
 9384 
 9385   // add len then return (len will be zero if compress succeeded, otherwise negative)
 9386   bind(reset_sp);
 9387   addl(result, len);
 9388 
 9389   bind(done);
 9390 }
 9391 
 9392 // Inflate byte[] array to char[].
 9393 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 9394 //   @IntrinsicCandidate
 9395 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 9396 //     for (int i = 0; i < len; i++) {
 9397 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 9398 //     }
 9399 //   }
 9400 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 9401   XMMRegister tmp1, Register tmp2, KRegister mask) {
 9402   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 9403   // rsi: src
 9404   // rdi: dst
 9405   // rdx: len
 9406   // rcx: tmp2
 9407 
 9408   // rsi holds start addr of source byte[] to be inflated
 9409   // rdi holds start addr of destination char[]
 9410   // rdx holds length
 9411   assert_different_registers(src, dst, len, tmp2);
 9412   movl(tmp2, len);
 9413   if ((UseAVX > 2) && // AVX512
 9414     VM_Version::supports_avx512vlbw() &&
 9415     VM_Version::supports_bmi2()) {
 9416 
 9417     Label copy_32_loop, copy_tail;
 9418     Register tmp3_aliased = len;
 9419 
 9420     // if length of the string is less than 16, handle it in an old fashioned way
 9421     testl(len, -16);
 9422     jcc(Assembler::zero, below_threshold);
 9423 
 9424     testl(len, -1 * AVX3Threshold);
 9425     jcc(Assembler::zero, avx3_threshold);
 9426 
 9427     // In order to use only one arithmetic operation for the main loop we use
 9428     // this pre-calculation
 9429     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 9430     andl(len, -32);     // vector count
 9431     jccb(Assembler::zero, copy_tail);
 9432 
 9433     lea(src, Address(src, len, Address::times_1));
 9434     lea(dst, Address(dst, len, Address::times_2));
 9435     negptr(len);
 9436 
 9437 
 9438     // inflate 32 chars per iter
 9439     bind(copy_32_loop);
 9440     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 9441     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 9442     addptr(len, 32);
 9443     jcc(Assembler::notZero, copy_32_loop);
 9444 
 9445     bind(copy_tail);
 9446     // bail out when there is nothing to be done
 9447     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 9448     jcc(Assembler::zero, done);
 9449 
 9450     // ~(~0 << length), where length is the # of remaining elements to process
 9451     movl(tmp3_aliased, -1);
 9452     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 9453     notl(tmp3_aliased);
 9454     kmovdl(mask, tmp3_aliased);
 9455     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 9456     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 9457 
 9458     jmp(done);
 9459     bind(avx3_threshold);
 9460   }
 9461   if (UseSSE42Intrinsics) {
 9462     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 9463 
 9464     if (UseAVX > 1) {
 9465       andl(tmp2, (16 - 1));
 9466       andl(len, -16);
 9467       jccb(Assembler::zero, copy_new_tail);
 9468     } else {
 9469       andl(tmp2, 0x00000007);   // tail count (in chars)
 9470       andl(len, 0xfffffff8);    // vector count (in chars)
 9471       jccb(Assembler::zero, copy_tail);
 9472     }
 9473 
 9474     // vectored inflation
 9475     lea(src, Address(src, len, Address::times_1));
 9476     lea(dst, Address(dst, len, Address::times_2));
 9477     negptr(len);
 9478 
 9479     if (UseAVX > 1) {
 9480       bind(copy_16_loop);
 9481       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 9482       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 9483       addptr(len, 16);
 9484       jcc(Assembler::notZero, copy_16_loop);
 9485 
 9486       bind(below_threshold);
 9487       bind(copy_new_tail);
 9488       movl(len, tmp2);
 9489       andl(tmp2, 0x00000007);
 9490       andl(len, 0xFFFFFFF8);
 9491       jccb(Assembler::zero, copy_tail);
 9492 
 9493       pmovzxbw(tmp1, Address(src, 0));
 9494       movdqu(Address(dst, 0), tmp1);
 9495       addptr(src, 8);
 9496       addptr(dst, 2 * 8);
 9497 
 9498       jmp(copy_tail, true);
 9499     }
 9500 
 9501     // inflate 8 chars per iter
 9502     bind(copy_8_loop);
 9503     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 9504     movdqu(Address(dst, len, Address::times_2), tmp1);
 9505     addptr(len, 8);
 9506     jcc(Assembler::notZero, copy_8_loop);
 9507 
 9508     bind(copy_tail);
 9509     movl(len, tmp2);
 9510 
 9511     cmpl(len, 4);
 9512     jccb(Assembler::less, copy_bytes);
 9513 
 9514     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 9515     pmovzxbw(tmp1, tmp1);
 9516     movq(Address(dst, 0), tmp1);
 9517     subptr(len, 4);
 9518     addptr(src, 4);
 9519     addptr(dst, 8);
 9520 
 9521     bind(copy_bytes);
 9522   } else {
 9523     bind(below_threshold);
 9524   }
 9525 
 9526   testl(len, len);
 9527   jccb(Assembler::zero, done);
 9528   lea(src, Address(src, len, Address::times_1));
 9529   lea(dst, Address(dst, len, Address::times_2));
 9530   negptr(len);
 9531 
 9532   // inflate 1 char per iter
 9533   bind(copy_chars_loop);
 9534   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 9535   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 9536   increment(len);
 9537   jcc(Assembler::notZero, copy_chars_loop);
 9538 
 9539   bind(done);
 9540 }
 9541 
 9542 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 9543   switch(type) {
 9544     case T_BYTE:
 9545     case T_BOOLEAN:
 9546       evmovdqub(dst, kmask, src, merge, vector_len);
 9547       break;
 9548     case T_CHAR:
 9549     case T_SHORT:
 9550       evmovdquw(dst, kmask, src, merge, vector_len);
 9551       break;
 9552     case T_INT:
 9553     case T_FLOAT:
 9554       evmovdqul(dst, kmask, src, merge, vector_len);
 9555       break;
 9556     case T_LONG:
 9557     case T_DOUBLE:
 9558       evmovdquq(dst, kmask, src, merge, vector_len);
 9559       break;
 9560     default:
 9561       fatal("Unexpected type argument %s", type2name(type));
 9562       break;
 9563   }
 9564 }
 9565 
 9566 
 9567 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 9568   switch(type) {
 9569     case T_BYTE:
 9570     case T_BOOLEAN:
 9571       evmovdqub(dst, kmask, src, merge, vector_len);
 9572       break;
 9573     case T_CHAR:
 9574     case T_SHORT:
 9575       evmovdquw(dst, kmask, src, merge, vector_len);
 9576       break;
 9577     case T_INT:
 9578     case T_FLOAT:
 9579       evmovdqul(dst, kmask, src, merge, vector_len);
 9580       break;
 9581     case T_LONG:
 9582     case T_DOUBLE:
 9583       evmovdquq(dst, kmask, src, merge, vector_len);
 9584       break;
 9585     default:
 9586       fatal("Unexpected type argument %s", type2name(type));
 9587       break;
 9588   }
 9589 }
 9590 
 9591 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister 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::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 9616   switch(masklen) {
 9617     case 2:
 9618        knotbl(dst, src);
 9619        movl(rtmp, 3);
 9620        kmovbl(ktmp, rtmp);
 9621        kandbl(dst, ktmp, dst);
 9622        break;
 9623     case 4:
 9624        knotbl(dst, src);
 9625        movl(rtmp, 15);
 9626        kmovbl(ktmp, rtmp);
 9627        kandbl(dst, ktmp, dst);
 9628        break;
 9629     case 8:
 9630        knotbl(dst, src);
 9631        break;
 9632     case 16:
 9633        knotwl(dst, src);
 9634        break;
 9635     case 32:
 9636        knotdl(dst, src);
 9637        break;
 9638     case 64:
 9639        knotql(dst, src);
 9640        break;
 9641     default:
 9642       fatal("Unexpected vector length %d", masklen);
 9643       break;
 9644   }
 9645 }
 9646 
 9647 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9648   switch(type) {
 9649     case T_BOOLEAN:
 9650     case T_BYTE:
 9651        kandbl(dst, src1, src2);
 9652        break;
 9653     case T_CHAR:
 9654     case T_SHORT:
 9655        kandwl(dst, src1, src2);
 9656        break;
 9657     case T_INT:
 9658     case T_FLOAT:
 9659        kanddl(dst, src1, src2);
 9660        break;
 9661     case T_LONG:
 9662     case T_DOUBLE:
 9663        kandql(dst, src1, src2);
 9664        break;
 9665     default:
 9666       fatal("Unexpected type argument %s", type2name(type));
 9667       break;
 9668   }
 9669 }
 9670 
 9671 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9672   switch(type) {
 9673     case T_BOOLEAN:
 9674     case T_BYTE:
 9675        korbl(dst, src1, src2);
 9676        break;
 9677     case T_CHAR:
 9678     case T_SHORT:
 9679        korwl(dst, src1, src2);
 9680        break;
 9681     case T_INT:
 9682     case T_FLOAT:
 9683        kordl(dst, src1, src2);
 9684        break;
 9685     case T_LONG:
 9686     case T_DOUBLE:
 9687        korql(dst, src1, src2);
 9688        break;
 9689     default:
 9690       fatal("Unexpected type argument %s", type2name(type));
 9691       break;
 9692   }
 9693 }
 9694 
 9695 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9696   switch(type) {
 9697     case T_BOOLEAN:
 9698     case T_BYTE:
 9699        kxorbl(dst, src1, src2);
 9700        break;
 9701     case T_CHAR:
 9702     case T_SHORT:
 9703        kxorwl(dst, src1, src2);
 9704        break;
 9705     case T_INT:
 9706     case T_FLOAT:
 9707        kxordl(dst, src1, src2);
 9708        break;
 9709     case T_LONG:
 9710     case T_DOUBLE:
 9711        kxorql(dst, src1, src2);
 9712        break;
 9713     default:
 9714       fatal("Unexpected type argument %s", type2name(type));
 9715       break;
 9716   }
 9717 }
 9718 
 9719 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9720   switch(type) {
 9721     case T_BOOLEAN:
 9722     case T_BYTE:
 9723       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9724     case T_CHAR:
 9725     case T_SHORT:
 9726       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9727     case T_INT:
 9728     case T_FLOAT:
 9729       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9730     case T_LONG:
 9731     case T_DOUBLE:
 9732       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9733     default:
 9734       fatal("Unexpected type argument %s", type2name(type)); break;
 9735   }
 9736 }
 9737 
 9738 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9739   switch(type) {
 9740     case T_BOOLEAN:
 9741     case T_BYTE:
 9742       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9743     case T_CHAR:
 9744     case T_SHORT:
 9745       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9746     case T_INT:
 9747     case T_FLOAT:
 9748       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9749     case T_LONG:
 9750     case T_DOUBLE:
 9751       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9752     default:
 9753       fatal("Unexpected type argument %s", type2name(type)); break;
 9754   }
 9755 }
 9756 
 9757 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9758   switch(type) {
 9759     case T_BYTE:
 9760       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9761     case T_SHORT:
 9762       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9763     case T_INT:
 9764       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9765     case T_LONG:
 9766       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9767     default:
 9768       fatal("Unexpected type argument %s", type2name(type)); break;
 9769   }
 9770 }
 9771 
 9772 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9773   switch(type) {
 9774     case T_BYTE:
 9775       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9776     case T_SHORT:
 9777       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9778     case T_INT:
 9779       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9780     case T_LONG:
 9781       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9782     default:
 9783       fatal("Unexpected type argument %s", type2name(type)); break;
 9784   }
 9785 }
 9786 
 9787 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9788   switch(type) {
 9789     case T_BYTE:
 9790       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9791     case T_SHORT:
 9792       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9793     case T_INT:
 9794       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9795     case T_LONG:
 9796       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9797     default:
 9798       fatal("Unexpected type argument %s", type2name(type)); break;
 9799   }
 9800 }
 9801 
 9802 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9803   switch(type) {
 9804     case T_BYTE:
 9805       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9806     case T_SHORT:
 9807       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9808     case T_INT:
 9809       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9810     case T_LONG:
 9811       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9812     default:
 9813       fatal("Unexpected type argument %s", type2name(type)); break;
 9814   }
 9815 }
 9816 
 9817 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9818   switch(type) {
 9819     case T_BYTE:
 9820       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9821     case T_SHORT:
 9822       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9823     case T_INT:
 9824       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9825     case T_LONG:
 9826       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9827     case T_FLOAT:
 9828       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9829     case T_DOUBLE:
 9830       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9831     default:
 9832       fatal("Unexpected type argument %s", type2name(type)); break;
 9833   }
 9834 }
 9835 
 9836 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9837   switch(type) {
 9838     case T_BYTE:
 9839       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9840     case T_SHORT:
 9841       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9842     case T_INT:
 9843       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9844     case T_LONG:
 9845       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9846     case T_FLOAT:
 9847       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9848     case T_DOUBLE:
 9849       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9850     default:
 9851       fatal("Unexpected type argument %s", type2name(type)); break;
 9852   }
 9853 }
 9854 
 9855 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9856   switch(type) {
 9857     case T_BYTE:
 9858       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9859     case T_SHORT:
 9860       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9861     case T_INT:
 9862       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9863     case T_LONG:
 9864       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9865     case T_FLOAT:
 9866       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9867     case T_DOUBLE:
 9868       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9869     default:
 9870       fatal("Unexpected type argument %s", type2name(type)); break;
 9871   }
 9872 }
 9873 
 9874 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9875   switch(type) {
 9876     case T_BYTE:
 9877       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9878     case T_SHORT:
 9879       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9880     case T_INT:
 9881       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9882     case T_LONG:
 9883       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9884     case T_FLOAT:
 9885       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9886     case T_DOUBLE:
 9887       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9888     default:
 9889       fatal("Unexpected type argument %s", type2name(type)); break;
 9890   }
 9891 }
 9892 
 9893 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9894   switch(type) {
 9895     case T_INT:
 9896       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9897     case T_LONG:
 9898       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9899     default:
 9900       fatal("Unexpected type argument %s", type2name(type)); break;
 9901   }
 9902 }
 9903 
 9904 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9905   switch(type) {
 9906     case T_INT:
 9907       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9908     case T_LONG:
 9909       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9910     default:
 9911       fatal("Unexpected type argument %s", type2name(type)); break;
 9912   }
 9913 }
 9914 
 9915 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9916   switch(type) {
 9917     case T_INT:
 9918       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9919     case T_LONG:
 9920       evporq(dst, mask, nds, src, merge, vector_len); break;
 9921     default:
 9922       fatal("Unexpected type argument %s", type2name(type)); break;
 9923   }
 9924 }
 9925 
 9926 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9927   switch(type) {
 9928     case T_INT:
 9929       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9930     case T_LONG:
 9931       evporq(dst, mask, nds, src, merge, vector_len); break;
 9932     default:
 9933       fatal("Unexpected type argument %s", type2name(type)); break;
 9934   }
 9935 }
 9936 
 9937 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9938   switch(type) {
 9939     case T_INT:
 9940       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9941     case T_LONG:
 9942       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9943     default:
 9944       fatal("Unexpected type argument %s", type2name(type)); break;
 9945   }
 9946 }
 9947 
 9948 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9949   switch(type) {
 9950     case T_INT:
 9951       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9952     case T_LONG:
 9953       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9954     default:
 9955       fatal("Unexpected type argument %s", type2name(type)); break;
 9956   }
 9957 }
 9958 
 9959 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9960   switch(masklen) {
 9961     case 8:
 9962        kortestbl(src1, src2);
 9963        break;
 9964     case 16:
 9965        kortestwl(src1, src2);
 9966        break;
 9967     case 32:
 9968        kortestdl(src1, src2);
 9969        break;
 9970     case 64:
 9971        kortestql(src1, src2);
 9972        break;
 9973     default:
 9974       fatal("Unexpected mask length %d", masklen);
 9975       break;
 9976   }
 9977 }
 9978 
 9979 
 9980 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
 9981   switch(masklen)  {
 9982     case 8:
 9983        ktestbl(src1, src2);
 9984        break;
 9985     case 16:
 9986        ktestwl(src1, src2);
 9987        break;
 9988     case 32:
 9989        ktestdl(src1, src2);
 9990        break;
 9991     case 64:
 9992        ktestql(src1, src2);
 9993        break;
 9994     default:
 9995       fatal("Unexpected mask length %d", masklen);
 9996       break;
 9997   }
 9998 }
 9999 
10000 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
10001   switch(type) {
10002     case T_INT:
10003       evprold(dst, mask, src, shift, merge, vlen_enc); break;
10004     case T_LONG:
10005       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
10006     default:
10007       fatal("Unexpected type argument %s", type2name(type)); break;
10008       break;
10009   }
10010 }
10011 
10012 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
10013   switch(type) {
10014     case T_INT:
10015       evprord(dst, mask, src, shift, merge, vlen_enc); break;
10016     case T_LONG:
10017       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
10018     default:
10019       fatal("Unexpected type argument %s", type2name(type)); break;
10020   }
10021 }
10022 
10023 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
10024   switch(type) {
10025     case T_INT:
10026       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
10027     case T_LONG:
10028       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
10029     default:
10030       fatal("Unexpected type argument %s", type2name(type)); break;
10031   }
10032 }
10033 
10034 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
10035   switch(type) {
10036     case T_INT:
10037       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
10038     case T_LONG:
10039       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
10040     default:
10041       fatal("Unexpected type argument %s", type2name(type)); break;
10042   }
10043 }
10044 
10045 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10046   assert(rscratch != noreg || always_reachable(src), "missing");
10047 
10048   if (reachable(src)) {
10049     evpandq(dst, nds, as_Address(src), vector_len);
10050   } else {
10051     lea(rscratch, src);
10052     evpandq(dst, nds, Address(rscratch, 0), vector_len);
10053   }
10054 }
10055 
10056 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
10057   assert(rscratch != noreg || always_reachable(src), "missing");
10058 
10059   if (reachable(src)) {
10060     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
10061   } else {
10062     lea(rscratch, src);
10063     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
10064   }
10065 }
10066 
10067 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10068   assert(rscratch != noreg || always_reachable(src), "missing");
10069 
10070   if (reachable(src)) {
10071     evporq(dst, nds, as_Address(src), vector_len);
10072   } else {
10073     lea(rscratch, src);
10074     evporq(dst, nds, Address(rscratch, 0), vector_len);
10075   }
10076 }
10077 
10078 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10079   assert(rscratch != noreg || always_reachable(src), "missing");
10080 
10081   if (reachable(src)) {
10082     vpshufb(dst, nds, as_Address(src), vector_len);
10083   } else {
10084     lea(rscratch, src);
10085     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
10086   }
10087 }
10088 
10089 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10090   assert(rscratch != noreg || always_reachable(src), "missing");
10091 
10092   if (reachable(src)) {
10093     Assembler::vpor(dst, nds, as_Address(src), vector_len);
10094   } else {
10095     lea(rscratch, src);
10096     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
10097   }
10098 }
10099 
10100 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
10101   assert(rscratch != noreg || always_reachable(src3), "missing");
10102 
10103   if (reachable(src3)) {
10104     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
10105   } else {
10106     lea(rscratch, src3);
10107     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
10108   }
10109 }
10110 
10111 #if COMPILER2_OR_JVMCI
10112 
10113 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
10114                                  Register length, Register temp, int vec_enc) {
10115   // Computing mask for predicated vector store.
10116   movptr(temp, -1);
10117   bzhiq(temp, temp, length);
10118   kmov(mask, temp);
10119   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
10120 }
10121 
10122 // Set memory operation for length "less than" 64 bytes.
10123 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
10124                                        XMMRegister xmm, KRegister mask, Register length,
10125                                        Register temp, bool use64byteVector) {
10126   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10127   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10128   if (!use64byteVector) {
10129     fill32(dst, disp, xmm);
10130     subptr(length, 32 >> shift);
10131     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
10132   } else {
10133     assert(MaxVectorSize == 64, "vector length != 64");
10134     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
10135   }
10136 }
10137 
10138 
10139 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
10140                                        XMMRegister xmm, KRegister mask, Register length,
10141                                        Register temp) {
10142   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10143   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10144   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
10145 }
10146 
10147 
10148 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
10149   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10150   vmovdqu(dst, xmm);
10151 }
10152 
10153 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
10154   fill32(Address(dst, disp), xmm);
10155 }
10156 
10157 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
10158   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10159   if (!use64byteVector) {
10160     fill32(dst, xmm);
10161     fill32(dst.plus_disp(32), xmm);
10162   } else {
10163     evmovdquq(dst, xmm, Assembler::AVX_512bit);
10164   }
10165 }
10166 
10167 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
10168   fill64(Address(dst, disp), xmm, use64byteVector);
10169 }
10170 
10171 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
10172                                         Register count, Register rtmp, XMMRegister xtmp) {
10173   Label L_exit;
10174   Label L_fill_start;
10175   Label L_fill_64_bytes;
10176   Label L_fill_96_bytes;
10177   Label L_fill_128_bytes;
10178   Label L_fill_128_bytes_loop;
10179   Label L_fill_128_loop_header;
10180   Label L_fill_128_bytes_loop_header;
10181   Label L_fill_128_bytes_loop_pre_header;
10182   Label L_fill_zmm_sequence;
10183 
10184   int shift = -1;
10185   switch(type) {
10186     case T_BYTE:  shift = 0;
10187       break;
10188     case T_SHORT: shift = 1;
10189       break;
10190     case T_INT:   shift = 2;
10191       break;
10192     /* Uncomment when LONG fill stubs are supported.
10193     case T_LONG:  shift = 3;
10194       break;
10195     */
10196     default:
10197       fatal("Unhandled type: %s\n", type2name(type));
10198   }
10199 
10200   if ((CopyAVX3Threshold != 0)  || (MaxVectorSize == 32)) {
10201 
10202     if (MaxVectorSize == 64) {
10203       cmpq(count, CopyAVX3Threshold >> shift);
10204       jcc(Assembler::greater, L_fill_zmm_sequence);
10205     }
10206 
10207     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
10208 
10209     bind(L_fill_start);
10210 
10211     cmpq(count, 32 >> shift);
10212     jccb(Assembler::greater, L_fill_64_bytes);
10213     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
10214     jmp(L_exit);
10215 
10216     bind(L_fill_64_bytes);
10217     cmpq(count, 64 >> shift);
10218     jccb(Assembler::greater, L_fill_96_bytes);
10219     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
10220     jmp(L_exit);
10221 
10222     bind(L_fill_96_bytes);
10223     cmpq(count, 96 >> shift);
10224     jccb(Assembler::greater, L_fill_128_bytes);
10225     fill64(to, 0, xtmp);
10226     subq(count, 64 >> shift);
10227     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
10228     jmp(L_exit);
10229 
10230     bind(L_fill_128_bytes);
10231     cmpq(count, 128 >> shift);
10232     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
10233     fill64(to, 0, xtmp);
10234     fill32(to, 64, xtmp);
10235     subq(count, 96 >> shift);
10236     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
10237     jmp(L_exit);
10238 
10239     bind(L_fill_128_bytes_loop_pre_header);
10240     {
10241       mov(rtmp, to);
10242       andq(rtmp, 31);
10243       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
10244       negq(rtmp);
10245       addq(rtmp, 32);
10246       mov64(r8, -1L);
10247       bzhiq(r8, r8, rtmp);
10248       kmovql(k2, r8);
10249       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
10250       addq(to, rtmp);
10251       shrq(rtmp, shift);
10252       subq(count, rtmp);
10253     }
10254 
10255     cmpq(count, 128 >> shift);
10256     jcc(Assembler::less, L_fill_start);
10257 
10258     bind(L_fill_128_bytes_loop_header);
10259     subq(count, 128 >> shift);
10260 
10261     align32();
10262     bind(L_fill_128_bytes_loop);
10263       fill64(to, 0, xtmp);
10264       fill64(to, 64, xtmp);
10265       addq(to, 128);
10266       subq(count, 128 >> shift);
10267       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
10268 
10269     addq(count, 128 >> shift);
10270     jcc(Assembler::zero, L_exit);
10271     jmp(L_fill_start);
10272   }
10273 
10274   if (MaxVectorSize == 64) {
10275     // Sequence using 64 byte ZMM register.
10276     Label L_fill_128_bytes_zmm;
10277     Label L_fill_192_bytes_zmm;
10278     Label L_fill_192_bytes_loop_zmm;
10279     Label L_fill_192_bytes_loop_header_zmm;
10280     Label L_fill_192_bytes_loop_pre_header_zmm;
10281     Label L_fill_start_zmm_sequence;
10282 
10283     bind(L_fill_zmm_sequence);
10284     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
10285 
10286     bind(L_fill_start_zmm_sequence);
10287     cmpq(count, 64 >> shift);
10288     jccb(Assembler::greater, L_fill_128_bytes_zmm);
10289     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
10290     jmp(L_exit);
10291 
10292     bind(L_fill_128_bytes_zmm);
10293     cmpq(count, 128 >> shift);
10294     jccb(Assembler::greater, L_fill_192_bytes_zmm);
10295     fill64(to, 0, xtmp, true);
10296     subq(count, 64 >> shift);
10297     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
10298     jmp(L_exit);
10299 
10300     bind(L_fill_192_bytes_zmm);
10301     cmpq(count, 192 >> shift);
10302     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
10303     fill64(to, 0, xtmp, true);
10304     fill64(to, 64, xtmp, true);
10305     subq(count, 128 >> shift);
10306     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
10307     jmp(L_exit);
10308 
10309     bind(L_fill_192_bytes_loop_pre_header_zmm);
10310     {
10311       movq(rtmp, to);
10312       andq(rtmp, 63);
10313       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
10314       negq(rtmp);
10315       addq(rtmp, 64);
10316       mov64(r8, -1L);
10317       bzhiq(r8, r8, rtmp);
10318       kmovql(k2, r8);
10319       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
10320       addq(to, rtmp);
10321       shrq(rtmp, shift);
10322       subq(count, rtmp);
10323     }
10324 
10325     cmpq(count, 192 >> shift);
10326     jcc(Assembler::less, L_fill_start_zmm_sequence);
10327 
10328     bind(L_fill_192_bytes_loop_header_zmm);
10329     subq(count, 192 >> shift);
10330 
10331     align32();
10332     bind(L_fill_192_bytes_loop_zmm);
10333       fill64(to, 0, xtmp, true);
10334       fill64(to, 64, xtmp, true);
10335       fill64(to, 128, xtmp, true);
10336       addq(to, 192);
10337       subq(count, 192 >> shift);
10338       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
10339 
10340     addq(count, 192 >> shift);
10341     jcc(Assembler::zero, L_exit);
10342     jmp(L_fill_start_zmm_sequence);
10343   }
10344   bind(L_exit);
10345 }
10346 #endif //COMPILER2_OR_JVMCI
10347 
10348 
10349 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
10350   Label done;
10351   cvttss2sil(dst, src);
10352   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10353   cmpl(dst, 0x80000000); // float_sign_flip
10354   jccb(Assembler::notEqual, done);
10355   subptr(rsp, 8);
10356   movflt(Address(rsp, 0), src);
10357   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
10358   pop(dst);
10359   bind(done);
10360 }
10361 
10362 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
10363   Label done;
10364   cvttsd2sil(dst, src);
10365   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10366   cmpl(dst, 0x80000000); // float_sign_flip
10367   jccb(Assembler::notEqual, done);
10368   subptr(rsp, 8);
10369   movdbl(Address(rsp, 0), src);
10370   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
10371   pop(dst);
10372   bind(done);
10373 }
10374 
10375 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
10376   Label done;
10377   cvttss2siq(dst, src);
10378   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10379   jccb(Assembler::notEqual, done);
10380   subptr(rsp, 8);
10381   movflt(Address(rsp, 0), src);
10382   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
10383   pop(dst);
10384   bind(done);
10385 }
10386 
10387 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10388   // Following code is line by line assembly translation rounding algorithm.
10389   // Please refer to java.lang.Math.round(float) algorithm for details.
10390   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
10391   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
10392   const int32_t FloatConsts_EXP_BIAS = 127;
10393   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
10394   const int32_t MINUS_32 = 0xFFFFFFE0;
10395   Label L_special_case, L_block1, L_exit;
10396   movl(rtmp, FloatConsts_EXP_BIT_MASK);
10397   movdl(dst, src);
10398   andl(dst, rtmp);
10399   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
10400   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
10401   subl(rtmp, dst);
10402   movl(rcx, rtmp);
10403   movl(dst, MINUS_32);
10404   testl(rtmp, dst);
10405   jccb(Assembler::notEqual, L_special_case);
10406   movdl(dst, src);
10407   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
10408   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
10409   movdl(rtmp, src);
10410   testl(rtmp, rtmp);
10411   jccb(Assembler::greaterEqual, L_block1);
10412   negl(dst);
10413   bind(L_block1);
10414   sarl(dst);
10415   addl(dst, 0x1);
10416   sarl(dst, 0x1);
10417   jmp(L_exit);
10418   bind(L_special_case);
10419   convert_f2i(dst, src);
10420   bind(L_exit);
10421 }
10422 
10423 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10424   // Following code is line by line assembly translation rounding algorithm.
10425   // Please refer to java.lang.Math.round(double) algorithm for details.
10426   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
10427   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
10428   const int64_t DoubleConsts_EXP_BIAS = 1023;
10429   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
10430   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
10431   Label L_special_case, L_block1, L_exit;
10432   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
10433   movq(dst, src);
10434   andq(dst, rtmp);
10435   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
10436   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
10437   subq(rtmp, dst);
10438   movq(rcx, rtmp);
10439   mov64(dst, MINUS_64);
10440   testq(rtmp, dst);
10441   jccb(Assembler::notEqual, L_special_case);
10442   movq(dst, src);
10443   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
10444   andq(dst, rtmp);
10445   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
10446   orq(dst, rtmp);
10447   movq(rtmp, src);
10448   testq(rtmp, rtmp);
10449   jccb(Assembler::greaterEqual, L_block1);
10450   negq(dst);
10451   bind(L_block1);
10452   sarq(dst);
10453   addq(dst, 0x1);
10454   sarq(dst, 0x1);
10455   jmp(L_exit);
10456   bind(L_special_case);
10457   convert_d2l(dst, src);
10458   bind(L_exit);
10459 }
10460 
10461 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
10462   Label done;
10463   cvttsd2siq(dst, src);
10464   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10465   jccb(Assembler::notEqual, done);
10466   subptr(rsp, 8);
10467   movdbl(Address(rsp, 0), src);
10468   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
10469   pop(dst);
10470   bind(done);
10471 }
10472 
10473 void MacroAssembler::cache_wb(Address line)
10474 {
10475   // 64 bit cpus always support clflush
10476   bool optimized = VM_Version::supports_clflushopt();
10477   bool no_evict = VM_Version::supports_clwb();
10478 
10479   // prefer clwb (writeback without evict) otherwise
10480   // prefer clflushopt (potentially parallel writeback with evict)
10481   // otherwise fallback on clflush (serial writeback with evict)
10482 
10483   if (optimized) {
10484     if (no_evict) {
10485       clwb(line);
10486     } else {
10487       clflushopt(line);
10488     }
10489   } else {
10490     // no need for fence when using CLFLUSH
10491     clflush(line);
10492   }
10493 }
10494 
10495 void MacroAssembler::cache_wbsync(bool is_pre)
10496 {
10497   bool optimized = VM_Version::supports_clflushopt();
10498   bool no_evict = VM_Version::supports_clwb();
10499 
10500   // pick the correct implementation
10501 
10502   if (!is_pre && (optimized || no_evict)) {
10503     // need an sfence for post flush when using clflushopt or clwb
10504     // otherwise no no need for any synchroniaztion
10505 
10506     sfence();
10507   }
10508 }
10509 
10510 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
10511   switch (cond) {
10512     // Note some conditions are synonyms for others
10513     case Assembler::zero:         return Assembler::notZero;
10514     case Assembler::notZero:      return Assembler::zero;
10515     case Assembler::less:         return Assembler::greaterEqual;
10516     case Assembler::lessEqual:    return Assembler::greater;
10517     case Assembler::greater:      return Assembler::lessEqual;
10518     case Assembler::greaterEqual: return Assembler::less;
10519     case Assembler::below:        return Assembler::aboveEqual;
10520     case Assembler::belowEqual:   return Assembler::above;
10521     case Assembler::above:        return Assembler::belowEqual;
10522     case Assembler::aboveEqual:   return Assembler::below;
10523     case Assembler::overflow:     return Assembler::noOverflow;
10524     case Assembler::noOverflow:   return Assembler::overflow;
10525     case Assembler::negative:     return Assembler::positive;
10526     case Assembler::positive:     return Assembler::negative;
10527     case Assembler::parity:       return Assembler::noParity;
10528     case Assembler::noParity:     return Assembler::parity;
10529   }
10530   ShouldNotReachHere(); return Assembler::overflow;
10531 }
10532 
10533 // This is simply a call to Thread::current()
10534 void MacroAssembler::get_thread_slow(Register thread) {
10535   if (thread != rax) {
10536     push(rax);
10537   }
10538   push(rdi);
10539   push(rsi);
10540   push(rdx);
10541   push(rcx);
10542   push(r8);
10543   push(r9);
10544   push(r10);
10545   push(r11);
10546 
10547   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
10548 
10549   pop(r11);
10550   pop(r10);
10551   pop(r9);
10552   pop(r8);
10553   pop(rcx);
10554   pop(rdx);
10555   pop(rsi);
10556   pop(rdi);
10557   if (thread != rax) {
10558     mov(thread, rax);
10559     pop(rax);
10560   }
10561 }
10562 
10563 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
10564   Label L_stack_ok;
10565   if (bias == 0) {
10566     testptr(sp, 2 * wordSize - 1);
10567   } else {
10568     // lea(tmp, Address(rsp, bias);
10569     mov(tmp, sp);
10570     addptr(tmp, bias);
10571     testptr(tmp, 2 * wordSize - 1);
10572   }
10573   jcc(Assembler::equal, L_stack_ok);
10574   block_comment(msg);
10575   stop(msg);
10576   bind(L_stack_ok);
10577 }
10578 
10579 // Implements fast-locking.
10580 //
10581 // obj: the object to be locked
10582 // reg_rax: rax
10583 // thread: the thread which attempts to lock obj
10584 // tmp: a temporary register
10585 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
10586   Register thread = r15_thread;
10587 
10588   assert(reg_rax == rax, "");
10589   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
10590 
10591   Label push;
10592   const Register top = tmp;
10593 
10594   // Preload the markWord. It is important that this is the first
10595   // instruction emitted as it is part of C1's null check semantics.
10596   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10597 
10598   if (UseObjectMonitorTable) {
10599     // Clear cache in case fast locking succeeds or we need to take the slow-path.
10600     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
10601   }
10602 
10603   if (DiagnoseSyncOnValueBasedClasses != 0) {
10604     load_klass(tmp, obj, rscratch1);
10605     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
10606     jcc(Assembler::notZero, slow);
10607   }
10608 
10609   // Load top.
10610   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10611 
10612   // Check if the lock-stack is full.
10613   cmpl(top, LockStack::end_offset());
10614   jcc(Assembler::greaterEqual, slow);
10615 
10616   // Check for recursion.
10617   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10618   jcc(Assembler::equal, push);
10619 
10620   // Check header for monitor (0b10).
10621   testptr(reg_rax, markWord::monitor_value);
10622   jcc(Assembler::notZero, slow);
10623 
10624   // Try to lock. Transition lock bits 0b01 => 0b00
10625   movptr(tmp, reg_rax);
10626   andptr(tmp, ~(int32_t)markWord::unlocked_value);
10627   orptr(reg_rax, markWord::unlocked_value);
10628   // Mask inline_type bit such that we go to the slow path if object is an inline type
10629   andptr(reg_rax, ~((int) markWord::inline_type_bit_in_place));
10630 
10631   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10632   jcc(Assembler::notEqual, slow);
10633 
10634   // Restore top, CAS clobbers register.
10635   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10636 
10637   bind(push);
10638   // After successful lock, push object on lock-stack.
10639   movptr(Address(thread, top), obj);
10640   incrementl(top, oopSize);
10641   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
10642 }
10643 
10644 // Implements fast-unlocking.
10645 //
10646 // obj: the object to be unlocked
10647 // reg_rax: rax
10648 // thread: the thread
10649 // tmp: a temporary register
10650 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
10651   Register thread = r15_thread;
10652 
10653   assert(reg_rax == rax, "");
10654   assert_different_registers(obj, reg_rax, thread, tmp);
10655 
10656   Label unlocked, push_and_slow;
10657   const Register top = tmp;
10658 
10659   // Check if obj is top of lock-stack.
10660   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10661   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10662   jcc(Assembler::notEqual, slow);
10663 
10664   // Pop lock-stack.
10665   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
10666   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10667 
10668   // Check if recursive.
10669   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
10670   jcc(Assembler::equal, unlocked);
10671 
10672   // Not recursive. Check header for monitor (0b10).
10673   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10674   testptr(reg_rax, markWord::monitor_value);
10675   jcc(Assembler::notZero, push_and_slow);
10676 
10677 #ifdef ASSERT
10678   // Check header not unlocked (0b01).
10679   Label not_unlocked;
10680   testptr(reg_rax, markWord::unlocked_value);
10681   jcc(Assembler::zero, not_unlocked);
10682   stop("fast_unlock already unlocked");
10683   bind(not_unlocked);
10684 #endif
10685 
10686   // Try to unlock. Transition lock bits 0b00 => 0b01
10687   movptr(tmp, reg_rax);
10688   orptr(tmp, markWord::unlocked_value);
10689   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10690   jcc(Assembler::equal, unlocked);
10691 
10692   bind(push_and_slow);
10693   // Restore lock-stack and handle the unlock in runtime.
10694 #ifdef ASSERT
10695   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10696   movptr(Address(thread, top), obj);
10697 #endif
10698   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10699   jmp(slow);
10700 
10701   bind(unlocked);
10702 }
10703 
10704 // Saves legacy GPRs state on stack.
10705 void MacroAssembler::save_legacy_gprs() {
10706   subq(rsp, 16 * wordSize);
10707   movq(Address(rsp, 15 * wordSize), rax);
10708   movq(Address(rsp, 14 * wordSize), rcx);
10709   movq(Address(rsp, 13 * wordSize), rdx);
10710   movq(Address(rsp, 12 * wordSize), rbx);
10711   movq(Address(rsp, 10 * wordSize), rbp);
10712   movq(Address(rsp, 9 * wordSize), rsi);
10713   movq(Address(rsp, 8 * wordSize), rdi);
10714   movq(Address(rsp, 7 * wordSize), r8);
10715   movq(Address(rsp, 6 * wordSize), r9);
10716   movq(Address(rsp, 5 * wordSize), r10);
10717   movq(Address(rsp, 4 * wordSize), r11);
10718   movq(Address(rsp, 3 * wordSize), r12);
10719   movq(Address(rsp, 2 * wordSize), r13);
10720   movq(Address(rsp, wordSize), r14);
10721   movq(Address(rsp, 0), r15);
10722 }
10723 
10724 // Resotres back legacy GPRs state from stack.
10725 void MacroAssembler::restore_legacy_gprs() {
10726   movq(r15, Address(rsp, 0));
10727   movq(r14, Address(rsp, wordSize));
10728   movq(r13, Address(rsp, 2 * wordSize));
10729   movq(r12, Address(rsp, 3 * wordSize));
10730   movq(r11, Address(rsp, 4 * wordSize));
10731   movq(r10, Address(rsp, 5 * wordSize));
10732   movq(r9,  Address(rsp, 6 * wordSize));
10733   movq(r8,  Address(rsp, 7 * wordSize));
10734   movq(rdi, Address(rsp, 8 * wordSize));
10735   movq(rsi, Address(rsp, 9 * wordSize));
10736   movq(rbp, Address(rsp, 10 * wordSize));
10737   movq(rbx, Address(rsp, 12 * wordSize));
10738   movq(rdx, Address(rsp, 13 * wordSize));
10739   movq(rcx, Address(rsp, 14 * wordSize));
10740   movq(rax, Address(rsp, 15 * wordSize));
10741   addq(rsp, 16 * wordSize);
10742 }
10743 
10744 void MacroAssembler::load_aotrc_address(Register reg, address a) {
10745 #if INCLUDE_CDS
10746   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
10747   if (AOTCodeCache::is_on_for_dump()) {
10748     // all aotrc field addresses should be registered in the AOTCodeCache address table
10749     lea(reg, ExternalAddress(a));
10750   } else {
10751     mov64(reg, (uint64_t)a);
10752   }
10753 #else
10754   ShouldNotReachHere();
10755 #endif
10756 }
10757 
10758 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10759   if (VM_Version::supports_apx_f()) {
10760     esetzucc(comparison, dst);
10761   } else {
10762     setb(comparison, dst);
10763     movzbl(dst, dst);
10764   }
10765 }