1 /*
    2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
    3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
    4  *
    5  * This code is free software; you can redistribute it and/or modify it
    6  * under the terms of the GNU General Public License version 2 only, as
    7  * published by the Free Software Foundation.
    8  *
    9  * This code is distributed in the hope that it will be useful, but WITHOUT
   10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
   11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
   12  * version 2 for more details (a copy is included in the LICENSE file that
   13  * accompanied this code).
   14  *
   15  * You should have received a copy of the GNU General Public License version
   16  * 2 along with this work; if not, write to the Free Software Foundation,
   17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
   18  *
   19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
   20  * or visit www.oracle.com if you need additional information or have any
   21  * questions.
   22  *
   23  */
   24 
   25 #include "asm/assembler.hpp"
   26 #include "asm/assembler.inline.hpp"
   27 #include "code/aotCodeCache.hpp"
   28 #include "code/compiledIC.hpp"
   29 #include "compiler/compiler_globals.hpp"
   30 #include "compiler/disassembler.hpp"
   31 #include "ci/ciInlineKlass.hpp"
   32 #include "crc32c.h"
   33 #include "gc/shared/barrierSet.hpp"
   34 #include "gc/shared/barrierSetAssembler.hpp"
   35 #include "gc/shared/collectedHeap.inline.hpp"
   36 #include "gc/shared/tlab_globals.hpp"
   37 #include "interpreter/bytecodeHistogram.hpp"
   38 #include "interpreter/interpreter.hpp"
   39 #include "interpreter/interpreterRuntime.hpp"
   40 #include "jvm.h"
   41 #include "memory/resourceArea.hpp"
   42 #include "memory/universe.hpp"
   43 #include "oops/accessDecorators.hpp"
   44 #include "oops/compressedKlass.inline.hpp"
   45 #include "oops/compressedOops.inline.hpp"
   46 #include "oops/klass.inline.hpp"
   47 #include "oops/resolvedFieldEntry.hpp"
   48 #include "prims/methodHandles.hpp"
   49 #include "runtime/arguments.hpp"
   50 #include "runtime/continuation.hpp"
   51 #include "runtime/interfaceSupport.inline.hpp"
   52 #include "runtime/javaThread.hpp"
   53 #include "runtime/jniHandles.hpp"
   54 #include "runtime/objectMonitor.hpp"
   55 #include "runtime/os.hpp"
   56 #include "runtime/safepoint.hpp"
   57 #include "runtime/safepointMechanism.hpp"
   58 #include "runtime/sharedRuntime.hpp"
   59 #include "runtime/signature_cc.hpp"
   60 #include "runtime/stubRoutines.hpp"
   61 #include "utilities/checkedCast.hpp"
   62 #include "utilities/globalDefinitions.hpp"
   63 #include "utilities/macros.hpp"
   64 #include "vmreg_x86.inline.hpp"
   65 #ifdef COMPILER2
   66 #include "opto/output.hpp"
   67 #endif
   68 
   69 #ifdef PRODUCT
   70 #define BLOCK_COMMENT(str) /* nothing */
   71 #define STOP(error) stop(error)
   72 #else
   73 #define BLOCK_COMMENT(str) block_comment(str)
   74 #define STOP(error) block_comment(error); stop(error)
   75 #endif
   76 
   77 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   78 
   79 #ifdef ASSERT
   80 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   81 #endif
   82 
   83 static const Assembler::Condition reverse[] = {
   84     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   85     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   86     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   87     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
   88     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
   89     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
   90     Assembler::above          /* belowEqual    = 0x6 */ ,
   91     Assembler::belowEqual     /* above         = 0x7 */ ,
   92     Assembler::positive       /* negative      = 0x8 */ ,
   93     Assembler::negative       /* positive      = 0x9 */ ,
   94     Assembler::noParity       /* parity        = 0xa */ ,
   95     Assembler::parity         /* noParity      = 0xb */ ,
   96     Assembler::greaterEqual   /* less          = 0xc */ ,
   97     Assembler::less           /* greaterEqual  = 0xd */ ,
   98     Assembler::greater        /* lessEqual     = 0xe */ ,
   99     Assembler::lessEqual      /* greater       = 0xf, */
  100 
  101 };
  102 
  103 
  104 // Implementation of MacroAssembler
  105 
  106 Address MacroAssembler::as_Address(AddressLiteral adr) {
  107   // amd64 always does this as a pc-rel
  108   // we can be absolute or disp based on the instruction type
  109   // jmp/call are displacements others are absolute
  110   assert(!adr.is_lval(), "must be rval");
  111   assert(reachable(adr), "must be");
  112   return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
  113 
  114 }
  115 
  116 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  117   AddressLiteral base = adr.base();
  118   lea(rscratch, base);
  119   Address index = adr.index();
  120   assert(index._disp == 0, "must not have disp"); // maybe it can?
  121   Address array(rscratch, index._index, index._scale, index._disp);
  122   return array;
  123 }
  124 
  125 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
  126   Label L, E;
  127 
  128 #ifdef _WIN64
  129   // Windows always allocates space for it's register args
  130   assert(num_args <= 4, "only register arguments supported");
  131   subq(rsp,  frame::arg_reg_save_area_bytes);
  132 #endif
  133 
  134   // Align stack if necessary
  135   testl(rsp, 15);
  136   jcc(Assembler::zero, L);
  137 
  138   subq(rsp, 8);
  139   call(RuntimeAddress(entry_point));
  140   addq(rsp, 8);
  141   jmp(E);
  142 
  143   bind(L);
  144   call(RuntimeAddress(entry_point));
  145 
  146   bind(E);
  147 
  148 #ifdef _WIN64
  149   // restore stack pointer
  150   addq(rsp, frame::arg_reg_save_area_bytes);
  151 #endif
  152 }
  153 
  154 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
  155   assert(!src2.is_lval(), "should use cmpptr");
  156   assert(rscratch != noreg || always_reachable(src2), "missing");
  157 
  158   if (reachable(src2)) {
  159     cmpq(src1, as_Address(src2));
  160   } else {
  161     lea(rscratch, src2);
  162     Assembler::cmpq(src1, Address(rscratch, 0));
  163   }
  164 }
  165 
  166 int MacroAssembler::corrected_idivq(Register reg) {
  167   // Full implementation of Java ldiv and lrem; checks for special
  168   // case as described in JVM spec., p.243 & p.271.  The function
  169   // returns the (pc) offset of the idivl instruction - may be needed
  170   // for implicit exceptions.
  171   //
  172   //         normal case                           special case
  173   //
  174   // input : rax: dividend                         min_long
  175   //         reg: divisor   (may not be eax/edx)   -1
  176   //
  177   // output: rax: quotient  (= rax idiv reg)       min_long
  178   //         rdx: remainder (= rax irem reg)       0
  179   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
  180   static const int64_t min_long = 0x8000000000000000;
  181   Label normal_case, special_case;
  182 
  183   // check for special case
  184   cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
  185   jcc(Assembler::notEqual, normal_case);
  186   xorl(rdx, rdx); // prepare rdx for possible special case (where
  187                   // remainder = 0)
  188   cmpq(reg, -1);
  189   jcc(Assembler::equal, special_case);
  190 
  191   // handle normal case
  192   bind(normal_case);
  193   cdqq();
  194   int idivq_offset = offset();
  195   idivq(reg);
  196 
  197   // normal and special case exit
  198   bind(special_case);
  199 
  200   return idivq_offset;
  201 }
  202 
  203 void MacroAssembler::decrementq(Register reg, int value) {
  204   if (value == min_jint) { subq(reg, value); return; }
  205   if (value <  0) { incrementq(reg, -value); return; }
  206   if (value == 0) {                        ; return; }
  207   if (value == 1 && UseIncDec) { decq(reg) ; return; }
  208   /* else */      { subq(reg, value)       ; return; }
  209 }
  210 
  211 void MacroAssembler::decrementq(Address dst, int value) {
  212   if (value == min_jint) { subq(dst, value); return; }
  213   if (value <  0) { incrementq(dst, -value); return; }
  214   if (value == 0) {                        ; return; }
  215   if (value == 1 && UseIncDec) { decq(dst) ; return; }
  216   /* else */      { subq(dst, value)       ; return; }
  217 }
  218 
  219 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
  220   assert(rscratch != noreg || always_reachable(dst), "missing");
  221 
  222   if (reachable(dst)) {
  223     incrementq(as_Address(dst));
  224   } else {
  225     lea(rscratch, dst);
  226     incrementq(Address(rscratch, 0));
  227   }
  228 }
  229 
  230 void MacroAssembler::incrementq(Register reg, int value) {
  231   if (value == min_jint) { addq(reg, value); return; }
  232   if (value <  0) { decrementq(reg, -value); return; }
  233   if (value == 0) {                        ; return; }
  234   if (value == 1 && UseIncDec) { incq(reg) ; return; }
  235   /* else */      { addq(reg, value)       ; return; }
  236 }
  237 
  238 void MacroAssembler::incrementq(Address dst, int value) {
  239   if (value == min_jint) { addq(dst, value); return; }
  240   if (value <  0) { decrementq(dst, -value); return; }
  241   if (value == 0) {                        ; return; }
  242   if (value == 1 && UseIncDec) { incq(dst) ; return; }
  243   /* else */      { addq(dst, value)       ; return; }
  244 }
  245 
  246 // 32bit can do a case table jump in one instruction but we no longer allow the base
  247 // to be installed in the Address class
  248 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  249   lea(rscratch, entry.base());
  250   Address dispatch = entry.index();
  251   assert(dispatch._base == noreg, "must be");
  252   dispatch._base = rscratch;
  253   jmp(dispatch);
  254 }
  255 
  256 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  257   ShouldNotReachHere(); // 64bit doesn't use two regs
  258   cmpq(x_lo, y_lo);
  259 }
  260 
  261 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  262   mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  263 }
  264 
  265 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  266   lea(rscratch, adr);
  267   movptr(dst, rscratch);
  268 }
  269 
  270 void MacroAssembler::leave() {
  271   // %%% is this really better? Why not on 32bit too?
  272   emit_int8((unsigned char)0xC9); // LEAVE
  273 }
  274 
  275 void MacroAssembler::lneg(Register hi, Register lo) {
  276   ShouldNotReachHere(); // 64bit doesn't use two regs
  277   negq(lo);
  278 }
  279 
  280 void MacroAssembler::movoop(Register dst, jobject obj) {
  281   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  282 }
  283 
  284 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  285   mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  286   movq(dst, rscratch);
  287 }
  288 
  289 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  290   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  291 }
  292 
  293 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  294   mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  295   movq(dst, rscratch);
  296 }
  297 
  298 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  299   if (src.is_lval()) {
  300     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  301   } else {
  302     if (reachable(src)) {
  303       movq(dst, as_Address(src));
  304     } else {
  305       lea(dst, src);
  306       movq(dst, Address(dst, 0));
  307     }
  308   }
  309 }
  310 
  311 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  312   movq(as_Address(dst, rscratch), src);
  313 }
  314 
  315 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  316   movq(dst, as_Address(src, dst /*rscratch*/));
  317 }
  318 
  319 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
  320 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  321   if (is_simm32(src)) {
  322     movptr(dst, checked_cast<int32_t>(src));
  323   } else {
  324     mov64(rscratch, src);
  325     movq(dst, rscratch);
  326   }
  327 }
  328 
  329 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  330   movoop(rscratch, obj);
  331   push(rscratch);
  332 }
  333 
  334 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  335   mov_metadata(rscratch, obj);
  336   push(rscratch);
  337 }
  338 
  339 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  340   lea(rscratch, src);
  341   if (src.is_lval()) {
  342     push(rscratch);
  343   } else {
  344     pushq(Address(rscratch, 0));
  345   }
  346 }
  347 
  348 static void pass_arg0(MacroAssembler* masm, Register arg) {
  349   if (c_rarg0 != arg ) {
  350     masm->mov(c_rarg0, arg);
  351   }
  352 }
  353 
  354 static void pass_arg1(MacroAssembler* masm, Register arg) {
  355   if (c_rarg1 != arg ) {
  356     masm->mov(c_rarg1, arg);
  357   }
  358 }
  359 
  360 static void pass_arg2(MacroAssembler* masm, Register arg) {
  361   if (c_rarg2 != arg ) {
  362     masm->mov(c_rarg2, arg);
  363   }
  364 }
  365 
  366 static void pass_arg3(MacroAssembler* masm, Register arg) {
  367   if (c_rarg3 != arg ) {
  368     masm->mov(c_rarg3, arg);
  369   }
  370 }
  371 
  372 void MacroAssembler::stop(const char* msg) {
  373   if (ShowMessageBoxOnError) {
  374     address rip = pc();
  375     pusha(); // get regs on stack
  376     lea(c_rarg1, InternalAddress(rip));
  377     movq(c_rarg2, rsp); // pass pointer to regs array
  378   }
  379   // Skip AOT caching C strings in scratch buffer.
  380   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  381   lea(c_rarg0, ExternalAddress((address) str));
  382   andq(rsp, -16); // align stack as required by ABI
  383   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
  384   hlt();
  385 }
  386 
  387 void MacroAssembler::warn(const char* msg) {
  388   push(rbp);
  389   movq(rbp, rsp);
  390   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  391   push_CPU_state();   // keeps alignment at 16 bytes
  392 
  393 #ifdef _WIN64
  394   // Windows always allocates space for its register args
  395   subq(rsp,  frame::arg_reg_save_area_bytes);
  396 #endif
  397   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  398   lea(c_rarg0, ExternalAddress((address) str));
  399   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  400 
  401 #ifdef _WIN64
  402   // restore stack pointer
  403   addq(rsp, frame::arg_reg_save_area_bytes);
  404 #endif
  405   pop_CPU_state();
  406   mov(rsp, rbp);
  407   pop(rbp);
  408 }
  409 
  410 void MacroAssembler::print_state() {
  411   address rip = pc();
  412   pusha();            // get regs on stack
  413   push(rbp);
  414   movq(rbp, rsp);
  415   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  416   push_CPU_state();   // keeps alignment at 16 bytes
  417 
  418   lea(c_rarg0, InternalAddress(rip));
  419   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  420   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  421 
  422   pop_CPU_state();
  423   mov(rsp, rbp);
  424   pop(rbp);
  425   popa();
  426 }
  427 
  428 #ifndef PRODUCT
  429 extern "C" void findpc(intptr_t x);
  430 #endif
  431 
  432 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  433   // In order to get locks to work, we need to fake a in_VM state
  434   if (ShowMessageBoxOnError) {
  435     JavaThread* thread = JavaThread::current();
  436     JavaThreadState saved_state = thread->thread_state();
  437     thread->set_thread_state(_thread_in_vm);
  438 #ifndef PRODUCT
  439     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  440       ttyLocker ttyl;
  441       BytecodeCounter::print();
  442     }
  443 #endif
  444     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  445     // XXX correct this offset for amd64
  446     // This is the value of eip which points to where verify_oop will return.
  447     if (os::message_box(msg, "Execution stopped, print registers?")) {
  448       print_state64(pc, regs);
  449       BREAKPOINT;
  450     }
  451   }
  452   fatal("DEBUG MESSAGE: %s", msg);
  453 }
  454 
  455 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  456   ttyLocker ttyl;
  457   DebuggingContext debugging{};
  458   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  459 #ifndef PRODUCT
  460   tty->cr();
  461   findpc(pc);
  462   tty->cr();
  463 #endif
  464 #define PRINT_REG(rax, value) \
  465   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  466   PRINT_REG(rax, regs[15]);
  467   PRINT_REG(rbx, regs[12]);
  468   PRINT_REG(rcx, regs[14]);
  469   PRINT_REG(rdx, regs[13]);
  470   PRINT_REG(rdi, regs[8]);
  471   PRINT_REG(rsi, regs[9]);
  472   PRINT_REG(rbp, regs[10]);
  473   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  474   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  475   PRINT_REG(r8 , regs[7]);
  476   PRINT_REG(r9 , regs[6]);
  477   PRINT_REG(r10, regs[5]);
  478   PRINT_REG(r11, regs[4]);
  479   PRINT_REG(r12, regs[3]);
  480   PRINT_REG(r13, regs[2]);
  481   PRINT_REG(r14, regs[1]);
  482   PRINT_REG(r15, regs[0]);
  483 #undef PRINT_REG
  484   // Print some words near the top of the stack.
  485   int64_t* rsp = &regs[16];
  486   int64_t* dump_sp = rsp;
  487   for (int col1 = 0; col1 < 8; col1++) {
  488     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  489     os::print_location(tty, *dump_sp++);
  490   }
  491   for (int row = 0; row < 25; row++) {
  492     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  493     for (int col = 0; col < 4; col++) {
  494       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  495     }
  496     tty->cr();
  497   }
  498   // Print some instructions around pc:
  499   Disassembler::decode((address)pc-64, (address)pc);
  500   tty->print_cr("--------");
  501   Disassembler::decode((address)pc, (address)pc+32);
  502 }
  503 
  504 // The java_calling_convention describes stack locations as ideal slots on
  505 // a frame with no abi restrictions. Since we must observe abi restrictions
  506 // (like the placement of the register window) the slots must be biased by
  507 // the following value.
  508 static int reg2offset_in(VMReg r) {
  509   // Account for saved rbp and return address
  510   // This should really be in_preserve_stack_slots
  511   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  512 }
  513 
  514 static int reg2offset_out(VMReg r) {
  515   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  516 }
  517 
  518 // A long move
  519 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  520 
  521   // The calling conventions assures us that each VMregpair is either
  522   // all really one physical register or adjacent stack slots.
  523 
  524   if (src.is_single_phys_reg() ) {
  525     if (dst.is_single_phys_reg()) {
  526       if (dst.first() != src.first()) {
  527         mov(dst.first()->as_Register(), src.first()->as_Register());
  528       }
  529     } else {
  530       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  531              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  532       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  533     }
  534   } else if (dst.is_single_phys_reg()) {
  535     assert(src.is_single_reg(),  "not a stack pair");
  536     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  537   } else {
  538     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  539     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  540     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  541   }
  542 }
  543 
  544 // A double move
  545 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  546 
  547   // The calling conventions assures us that each VMregpair is either
  548   // all really one physical register or adjacent stack slots.
  549 
  550   if (src.is_single_phys_reg() ) {
  551     if (dst.is_single_phys_reg()) {
  552       // In theory these overlap but the ordering is such that this is likely a nop
  553       if ( src.first() != dst.first()) {
  554         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  555       }
  556     } else {
  557       assert(dst.is_single_reg(), "not a stack pair");
  558       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  559     }
  560   } else if (dst.is_single_phys_reg()) {
  561     assert(src.is_single_reg(),  "not a stack pair");
  562     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  563   } else {
  564     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  565     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  566     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  567   }
  568 }
  569 
  570 
  571 // A float arg may have to do float reg int reg conversion
  572 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  573   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  574 
  575   // The calling conventions assures us that each VMregpair is either
  576   // all really one physical register or adjacent stack slots.
  577 
  578   if (src.first()->is_stack()) {
  579     if (dst.first()->is_stack()) {
  580       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  581       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  582     } else {
  583       // stack to reg
  584       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  585       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  586     }
  587   } else if (dst.first()->is_stack()) {
  588     // reg to stack
  589     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  590     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  591   } else {
  592     // reg to reg
  593     // In theory these overlap but the ordering is such that this is likely a nop
  594     if ( src.first() != dst.first()) {
  595       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  596     }
  597   }
  598 }
  599 
  600 // On 64 bit we will store integer like items to the stack as
  601 // 64 bits items (x86_32/64 abi) even though java would only store
  602 // 32bits for a parameter. On 32bit it will simply be 32 bits
  603 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  604 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  605   if (src.first()->is_stack()) {
  606     if (dst.first()->is_stack()) {
  607       // stack to stack
  608       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  609       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  610     } else {
  611       // stack to reg
  612       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  613     }
  614   } else if (dst.first()->is_stack()) {
  615     // reg to stack
  616     // Do we really have to sign extend???
  617     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  618     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  619   } else {
  620     // Do we really have to sign extend???
  621     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
  622     if (dst.first() != src.first()) {
  623       movq(dst.first()->as_Register(), src.first()->as_Register());
  624     }
  625   }
  626 }
  627 
  628 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
  629   if (src.first()->is_stack()) {
  630     if (dst.first()->is_stack()) {
  631       // stack to stack
  632       movq(rax, Address(rbp, reg2offset_in(src.first())));
  633       movq(Address(rsp, reg2offset_out(dst.first())), rax);
  634     } else {
  635       // stack to reg
  636       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
  637     }
  638   } else if (dst.first()->is_stack()) {
  639     // reg to stack
  640     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
  641   } else {
  642     if (dst.first() != src.first()) {
  643       movq(dst.first()->as_Register(), src.first()->as_Register());
  644     }
  645   }
  646 }
  647 
  648 // An oop arg. Must pass a handle not the oop itself
  649 void MacroAssembler::object_move(OopMap* map,
  650                         int oop_handle_offset,
  651                         int framesize_in_slots,
  652                         VMRegPair src,
  653                         VMRegPair dst,
  654                         bool is_receiver,
  655                         int* receiver_offset) {
  656 
  657   // must pass a handle. First figure out the location we use as a handle
  658 
  659   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
  660 
  661   // See if oop is null if it is we need no handle
  662 
  663   if (src.first()->is_stack()) {
  664 
  665     // Oop is already on the stack as an argument
  666     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
  667     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
  668     if (is_receiver) {
  669       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
  670     }
  671 
  672     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
  673     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
  674     // conditionally move a null
  675     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
  676   } else {
  677 
  678     // Oop is in a register we must store it to the space we reserve
  679     // on the stack for oop_handles and pass a handle if oop is non-null
  680 
  681     const Register rOop = src.first()->as_Register();
  682     int oop_slot;
  683     if (rOop == j_rarg0)
  684       oop_slot = 0;
  685     else if (rOop == j_rarg1)
  686       oop_slot = 1;
  687     else if (rOop == j_rarg2)
  688       oop_slot = 2;
  689     else if (rOop == j_rarg3)
  690       oop_slot = 3;
  691     else if (rOop == j_rarg4)
  692       oop_slot = 4;
  693     else {
  694       assert(rOop == j_rarg5, "wrong register");
  695       oop_slot = 5;
  696     }
  697 
  698     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
  699     int offset = oop_slot*VMRegImpl::stack_slot_size;
  700 
  701     map->set_oop(VMRegImpl::stack2reg(oop_slot));
  702     // Store oop in handle area, may be null
  703     movptr(Address(rsp, offset), rOop);
  704     if (is_receiver) {
  705       *receiver_offset = offset;
  706     }
  707 
  708     cmpptr(rOop, NULL_WORD);
  709     lea(rHandle, Address(rsp, offset));
  710     // conditionally move a null from the handle area where it was just stored
  711     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
  712   }
  713 
  714   // If arg is on the stack then place it otherwise it is already in correct reg.
  715   if (dst.first()->is_stack()) {
  716     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
  717   }
  718 }
  719 
  720 void MacroAssembler::addptr(Register dst, int32_t imm32) {
  721   addq(dst, imm32);
  722 }
  723 
  724 void MacroAssembler::addptr(Register dst, Register src) {
  725   addq(dst, src);
  726 }
  727 
  728 void MacroAssembler::addptr(Address dst, Register src) {
  729   addq(dst, src);
  730 }
  731 
  732 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  733   assert(rscratch != noreg || always_reachable(src), "missing");
  734 
  735   if (reachable(src)) {
  736     Assembler::addsd(dst, as_Address(src));
  737   } else {
  738     lea(rscratch, src);
  739     Assembler::addsd(dst, Address(rscratch, 0));
  740   }
  741 }
  742 
  743 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
  744   assert(rscratch != noreg || always_reachable(src), "missing");
  745 
  746   if (reachable(src)) {
  747     addss(dst, as_Address(src));
  748   } else {
  749     lea(rscratch, src);
  750     addss(dst, Address(rscratch, 0));
  751   }
  752 }
  753 
  754 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  755   assert(rscratch != noreg || always_reachable(src), "missing");
  756 
  757   if (reachable(src)) {
  758     Assembler::addpd(dst, as_Address(src));
  759   } else {
  760     lea(rscratch, src);
  761     Assembler::addpd(dst, Address(rscratch, 0));
  762   }
  763 }
  764 
  765 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
  766 // Stub code is generated once and never copied.
  767 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
  768 void MacroAssembler::align64() {
  769   align(64, (uint)(uintptr_t)pc());
  770 }
  771 
  772 void MacroAssembler::align32() {
  773   align(32, (uint)(uintptr_t)pc());
  774 }
  775 
  776 void MacroAssembler::align(uint modulus) {
  777   // 8273459: Ensure alignment is possible with current segment alignment
  778   assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
  779   align(modulus, offset());
  780 }
  781 
  782 void MacroAssembler::align(uint modulus, uint target) {
  783   if (target % modulus != 0) {
  784     nop(modulus - (target % modulus));
  785   }
  786 }
  787 
  788 void MacroAssembler::push_f(XMMRegister r) {
  789   subptr(rsp, wordSize);
  790   movflt(Address(rsp, 0), r);
  791 }
  792 
  793 void MacroAssembler::pop_f(XMMRegister r) {
  794   movflt(r, Address(rsp, 0));
  795   addptr(rsp, wordSize);
  796 }
  797 
  798 void MacroAssembler::push_d(XMMRegister r) {
  799   subptr(rsp, 2 * wordSize);
  800   movdbl(Address(rsp, 0), r);
  801 }
  802 
  803 void MacroAssembler::pop_d(XMMRegister r) {
  804   movdbl(r, Address(rsp, 0));
  805   addptr(rsp, 2 * Interpreter::stackElementSize);
  806 }
  807 
  808 void MacroAssembler::push_ppx(Register src) {
  809   if (VM_Version::supports_apx_f()) {
  810     pushp(src);
  811   } else {
  812     Assembler::push(src);
  813   }
  814 }
  815 
  816 void MacroAssembler::pop_ppx(Register dst) {
  817   if (VM_Version::supports_apx_f()) {
  818     popp(dst);
  819   } else {
  820     Assembler::pop(dst);
  821   }
  822 }
  823 
  824 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  825   // Used in sign-masking with aligned address.
  826   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  827   assert(rscratch != noreg || always_reachable(src), "missing");
  828 
  829   if (UseAVX > 2 &&
  830       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
  831       (dst->encoding() >= 16)) {
  832     vpand(dst, dst, src, AVX_512bit, rscratch);
  833   } else if (reachable(src)) {
  834     Assembler::andpd(dst, as_Address(src));
  835   } else {
  836     lea(rscratch, src);
  837     Assembler::andpd(dst, Address(rscratch, 0));
  838   }
  839 }
  840 
  841 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
  842   // Used in sign-masking with aligned address.
  843   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  844   assert(rscratch != noreg || always_reachable(src), "missing");
  845 
  846   if (reachable(src)) {
  847     Assembler::andps(dst, as_Address(src));
  848   } else {
  849     lea(rscratch, src);
  850     Assembler::andps(dst, Address(rscratch, 0));
  851   }
  852 }
  853 
  854 void MacroAssembler::andptr(Register dst, int32_t imm32) {
  855   andq(dst, imm32);
  856 }
  857 
  858 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
  859   assert(rscratch != noreg || always_reachable(src), "missing");
  860 
  861   if (reachable(src)) {
  862     andq(dst, as_Address(src));
  863   } else {
  864     lea(rscratch, src);
  865     andq(dst, Address(rscratch, 0));
  866   }
  867 }
  868 
  869 void MacroAssembler::atomic_incl(Address counter_addr) {
  870   lock();
  871   incrementl(counter_addr);
  872 }
  873 
  874 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
  875   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  876 
  877   if (reachable(counter_addr)) {
  878     atomic_incl(as_Address(counter_addr));
  879   } else {
  880     lea(rscratch, counter_addr);
  881     atomic_incl(Address(rscratch, 0));
  882   }
  883 }
  884 
  885 void MacroAssembler::atomic_incq(Address counter_addr) {
  886   lock();
  887   incrementq(counter_addr);
  888 }
  889 
  890 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
  891   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  892 
  893   if (reachable(counter_addr)) {
  894     atomic_incq(as_Address(counter_addr));
  895   } else {
  896     lea(rscratch, counter_addr);
  897     atomic_incq(Address(rscratch, 0));
  898   }
  899 }
  900 
  901 // Writes to stack successive pages until offset reached to check for
  902 // stack overflow + shadow pages.  This clobbers tmp.
  903 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
  904   movptr(tmp, rsp);
  905   // Bang stack for total size given plus shadow page size.
  906   // Bang one page at a time because large size can bang beyond yellow and
  907   // red zones.
  908   Label loop;
  909   bind(loop);
  910   movl(Address(tmp, (-(int)os::vm_page_size())), size );
  911   subptr(tmp, (int)os::vm_page_size());
  912   subl(size, (int)os::vm_page_size());
  913   jcc(Assembler::greater, loop);
  914 
  915   // Bang down shadow pages too.
  916   // At this point, (tmp-0) is the last address touched, so don't
  917   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
  918   // was post-decremented.)  Skip this address by starting at i=1, and
  919   // touch a few more pages below.  N.B.  It is important to touch all
  920   // the way down including all pages in the shadow zone.
  921   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
  922     // this could be any sized move but this is can be a debugging crumb
  923     // so the bigger the better.
  924     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
  925   }
  926 }
  927 
  928 void MacroAssembler::reserved_stack_check() {
  929   // testing if reserved zone needs to be enabled
  930   Label no_reserved_zone_enabling;
  931 
  932   cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
  933   jcc(Assembler::below, no_reserved_zone_enabling);
  934 
  935   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
  936   jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
  937   should_not_reach_here();
  938 
  939   bind(no_reserved_zone_enabling);
  940 }
  941 
  942 void MacroAssembler::c2bool(Register x) {
  943   // implements x == 0 ? 0 : 1
  944   // note: must only look at least-significant byte of x
  945   //       since C-style booleans are stored in one byte
  946   //       only! (was bug)
  947   andl(x, 0xFF);
  948   setb(Assembler::notZero, x);
  949 }
  950 
  951 // Wouldn't need if AddressLiteral version had new name
  952 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
  953   Assembler::call(L, rtype);
  954 }
  955 
  956 void MacroAssembler::call(Register entry) {
  957   Assembler::call(entry);
  958 }
  959 
  960 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
  961   assert(rscratch != noreg || always_reachable(entry), "missing");
  962 
  963   if (reachable(entry)) {
  964     Assembler::call_literal(entry.target(), entry.rspec());
  965   } else {
  966     lea(rscratch, entry);
  967     Assembler::call(rscratch);
  968   }
  969 }
  970 
  971 void MacroAssembler::ic_call(address entry, jint method_index) {
  972   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
  973   // Needs full 64-bit immediate for later patching.
  974   Assembler::mov64(rax, (int64_t)Universe::non_oop_word());
  975   call(AddressLiteral(entry, rh));
  976 }
  977 
  978 int MacroAssembler::ic_check_size() {
  979   return UseCompactObjectHeaders ? 17 : 14;
  980 }
  981 
  982 int MacroAssembler::ic_check(int end_alignment) {
  983   Register receiver = j_rarg0;
  984   Register data = rax;
  985   Register temp = rscratch1;
  986 
  987   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
  988   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
  989   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
  990   // before the inline cache check here, and not after
  991   align(end_alignment, offset() + ic_check_size());
  992 
  993   int uep_offset = offset();
  994 
  995   if (UseCompactObjectHeaders) {
  996     load_narrow_klass_compact(temp, receiver);
  997     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  998   } else {
  999     movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
 1000     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
 1001   }
 1002 
 1003   // if inline cache check fails, then jump to runtime routine
 1004   jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 1005   assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
 1006 
 1007   return uep_offset;
 1008 }
 1009 
 1010 void MacroAssembler::emit_static_call_stub() {
 1011   // Static stub relocation also tags the Method* in the code-stream.
 1012   mov_metadata(rbx, (Metadata*) nullptr);  // Method is zapped till fixup time.
 1013   // This is recognized as unresolved by relocs/nativeinst/ic code.
 1014   jump(RuntimeAddress(pc()));
 1015 }
 1016 
 1017 // Implementation of call_VM versions
 1018 
 1019 void MacroAssembler::call_VM(Register oop_result,
 1020                              address entry_point,
 1021                              bool check_exceptions) {
 1022   Label C, E;
 1023   call(C, relocInfo::none);
 1024   jmp(E);
 1025 
 1026   bind(C);
 1027   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
 1028   ret(0);
 1029 
 1030   bind(E);
 1031 }
 1032 
 1033 void MacroAssembler::call_VM(Register oop_result,
 1034                              address entry_point,
 1035                              Register arg_1,
 1036                              bool check_exceptions) {
 1037   Label C, E;
 1038   call(C, relocInfo::none);
 1039   jmp(E);
 1040 
 1041   bind(C);
 1042   pass_arg1(this, arg_1);
 1043   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
 1044   ret(0);
 1045 
 1046   bind(E);
 1047 }
 1048 
 1049 void MacroAssembler::call_VM(Register oop_result,
 1050                              address entry_point,
 1051                              Register arg_1,
 1052                              Register arg_2,
 1053                              bool check_exceptions) {
 1054   Label C, E;
 1055   call(C, relocInfo::none);
 1056   jmp(E);
 1057 
 1058   bind(C);
 1059 
 1060   assert_different_registers(arg_1, c_rarg2);
 1061 
 1062   pass_arg2(this, arg_2);
 1063   pass_arg1(this, arg_1);
 1064   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
 1065   ret(0);
 1066 
 1067   bind(E);
 1068 }
 1069 
 1070 void MacroAssembler::call_VM(Register oop_result,
 1071                              address entry_point,
 1072                              Register arg_1,
 1073                              Register arg_2,
 1074                              Register arg_3,
 1075                              bool check_exceptions) {
 1076   Label C, E;
 1077   call(C, relocInfo::none);
 1078   jmp(E);
 1079 
 1080   bind(C);
 1081 
 1082   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1083   assert_different_registers(arg_2, c_rarg3);
 1084   pass_arg3(this, arg_3);
 1085   pass_arg2(this, arg_2);
 1086   pass_arg1(this, arg_1);
 1087   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
 1088   ret(0);
 1089 
 1090   bind(E);
 1091 }
 1092 
 1093 void MacroAssembler::call_VM(Register oop_result,
 1094                              Register last_java_sp,
 1095                              address entry_point,
 1096                              int number_of_arguments,
 1097                              bool check_exceptions) {
 1098   call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1099 }
 1100 
 1101 void MacroAssembler::call_VM(Register oop_result,
 1102                              Register last_java_sp,
 1103                              address entry_point,
 1104                              Register arg_1,
 1105                              bool check_exceptions) {
 1106   pass_arg1(this, arg_1);
 1107   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1108 }
 1109 
 1110 void MacroAssembler::call_VM(Register oop_result,
 1111                              Register last_java_sp,
 1112                              address entry_point,
 1113                              Register arg_1,
 1114                              Register arg_2,
 1115                              bool check_exceptions) {
 1116 
 1117   assert_different_registers(arg_1, c_rarg2);
 1118   pass_arg2(this, arg_2);
 1119   pass_arg1(this, arg_1);
 1120   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1121 }
 1122 
 1123 void MacroAssembler::call_VM(Register oop_result,
 1124                              Register last_java_sp,
 1125                              address entry_point,
 1126                              Register arg_1,
 1127                              Register arg_2,
 1128                              Register arg_3,
 1129                              bool check_exceptions) {
 1130   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1131   assert_different_registers(arg_2, c_rarg3);
 1132   pass_arg3(this, arg_3);
 1133   pass_arg2(this, arg_2);
 1134   pass_arg1(this, arg_1);
 1135   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1136 }
 1137 
 1138 void MacroAssembler::super_call_VM(Register oop_result,
 1139                                    Register last_java_sp,
 1140                                    address entry_point,
 1141                                    int number_of_arguments,
 1142                                    bool check_exceptions) {
 1143   MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1144 }
 1145 
 1146 void MacroAssembler::super_call_VM(Register oop_result,
 1147                                    Register last_java_sp,
 1148                                    address entry_point,
 1149                                    Register arg_1,
 1150                                    bool check_exceptions) {
 1151   pass_arg1(this, arg_1);
 1152   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1153 }
 1154 
 1155 void MacroAssembler::super_call_VM(Register oop_result,
 1156                                    Register last_java_sp,
 1157                                    address entry_point,
 1158                                    Register arg_1,
 1159                                    Register arg_2,
 1160                                    bool check_exceptions) {
 1161 
 1162   assert_different_registers(arg_1, c_rarg2);
 1163   pass_arg2(this, arg_2);
 1164   pass_arg1(this, arg_1);
 1165   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1166 }
 1167 
 1168 void MacroAssembler::super_call_VM(Register oop_result,
 1169                                    Register last_java_sp,
 1170                                    address entry_point,
 1171                                    Register arg_1,
 1172                                    Register arg_2,
 1173                                    Register arg_3,
 1174                                    bool check_exceptions) {
 1175   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1176   assert_different_registers(arg_2, c_rarg3);
 1177   pass_arg3(this, arg_3);
 1178   pass_arg2(this, arg_2);
 1179   pass_arg1(this, arg_1);
 1180   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1181 }
 1182 
 1183 void MacroAssembler::call_VM_base(Register oop_result,
 1184                                   Register last_java_sp,
 1185                                   address  entry_point,
 1186                                   int      number_of_arguments,
 1187                                   bool     check_exceptions) {
 1188   Register java_thread = r15_thread;
 1189 
 1190   // determine last_java_sp register
 1191   if (!last_java_sp->is_valid()) {
 1192     last_java_sp = rsp;
 1193   }
 1194   // debugging support
 1195   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 1196 #ifdef ASSERT
 1197   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 1198   // r12 is the heapbase.
 1199   if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 1200 #endif // ASSERT
 1201 
 1202   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 1203   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 1204 
 1205   // push java thread (becomes first argument of C function)
 1206 
 1207   mov(c_rarg0, r15_thread);
 1208 
 1209   // set last Java frame before call
 1210   assert(last_java_sp != rbp, "can't use ebp/rbp");
 1211 
 1212   // Only interpreter should have to set fp
 1213   set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
 1214 
 1215   // do the call, remove parameters
 1216   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 1217 
 1218 #ifdef ASSERT
 1219   // Check that thread register is not clobbered.
 1220   guarantee(java_thread != rax, "change this code");
 1221   push(rax);
 1222   { Label L;
 1223     get_thread_slow(rax);
 1224     cmpptr(java_thread, rax);
 1225     jcc(Assembler::equal, L);
 1226     STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
 1227     bind(L);
 1228   }
 1229   pop(rax);
 1230 #endif
 1231 
 1232   // reset last Java frame
 1233   // Only interpreter should have to clear fp
 1234   reset_last_Java_frame(true);
 1235 
 1236    // C++ interp handles this in the interpreter
 1237   check_and_handle_popframe();
 1238   check_and_handle_earlyret();
 1239 
 1240   if (check_exceptions) {
 1241     // check for pending exceptions (java_thread is set upon return)
 1242     cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 1243     // This used to conditionally jump to forward_exception however it is
 1244     // possible if we relocate that the branch will not reach. So we must jump
 1245     // around so we can always reach
 1246 
 1247     Label ok;
 1248     jcc(Assembler::equal, ok);
 1249     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 1250     bind(ok);
 1251   }
 1252 
 1253   // get oop result if there is one and reset the value in the thread
 1254   if (oop_result->is_valid()) {
 1255     get_vm_result_oop(oop_result);
 1256   }
 1257 }
 1258 
 1259 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 1260   // Calculate the value for last_Java_sp somewhat subtle.
 1261   // call_VM does an intermediate call which places a return address on
 1262   // the stack just under the stack pointer as the user finished with it.
 1263   // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
 1264 
 1265   // We've pushed one address, correct last_Java_sp
 1266   lea(rax, Address(rsp, wordSize));
 1267 
 1268   call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
 1269 }
 1270 
 1271 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
 1272 void MacroAssembler::call_VM_leaf0(address entry_point) {
 1273   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 1274 }
 1275 
 1276 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
 1277   call_VM_leaf_base(entry_point, number_of_arguments);
 1278 }
 1279 
 1280 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
 1281   pass_arg0(this, arg_0);
 1282   call_VM_leaf(entry_point, 1);
 1283 }
 1284 
 1285 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1286 
 1287   assert_different_registers(arg_0, c_rarg1);
 1288   pass_arg1(this, arg_1);
 1289   pass_arg0(this, arg_0);
 1290   call_VM_leaf(entry_point, 2);
 1291 }
 1292 
 1293 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1294   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1295   assert_different_registers(arg_1, c_rarg2);
 1296   pass_arg2(this, arg_2);
 1297   pass_arg1(this, arg_1);
 1298   pass_arg0(this, arg_0);
 1299   call_VM_leaf(entry_point, 3);
 1300 }
 1301 
 1302 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1303   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1304   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1305   assert_different_registers(arg_2, c_rarg3);
 1306   pass_arg3(this, arg_3);
 1307   pass_arg2(this, arg_2);
 1308   pass_arg1(this, arg_1);
 1309   pass_arg0(this, arg_0);
 1310   call_VM_leaf(entry_point, 3);
 1311 }
 1312 
 1313 void MacroAssembler::super_call_VM_leaf(address entry_point) {
 1314   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1315 }
 1316 
 1317 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1318   pass_arg0(this, arg_0);
 1319   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1320 }
 1321 
 1322 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1323   assert_different_registers(arg_0, c_rarg1);
 1324   pass_arg1(this, arg_1);
 1325   pass_arg0(this, arg_0);
 1326   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1327 }
 1328 
 1329 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1330   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1331   assert_different_registers(arg_1, c_rarg2);
 1332   pass_arg2(this, arg_2);
 1333   pass_arg1(this, arg_1);
 1334   pass_arg0(this, arg_0);
 1335   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1336 }
 1337 
 1338 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1339   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1340   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1341   assert_different_registers(arg_2, c_rarg3);
 1342   pass_arg3(this, arg_3);
 1343   pass_arg2(this, arg_2);
 1344   pass_arg1(this, arg_1);
 1345   pass_arg0(this, arg_0);
 1346   MacroAssembler::call_VM_leaf_base(entry_point, 4);
 1347 }
 1348 
 1349 void MacroAssembler::get_vm_result_oop(Register oop_result) {
 1350   movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
 1351   movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
 1352   verify_oop_msg(oop_result, "broken oop in call_VM_base");
 1353 }
 1354 
 1355 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
 1356   movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
 1357   movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
 1358 }
 1359 
 1360 void MacroAssembler::check_and_handle_earlyret() {
 1361 }
 1362 
 1363 void MacroAssembler::check_and_handle_popframe() {
 1364 }
 1365 
 1366 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
 1367   assert(rscratch != noreg || always_reachable(src1), "missing");
 1368 
 1369   if (reachable(src1)) {
 1370     cmpl(as_Address(src1), imm);
 1371   } else {
 1372     lea(rscratch, src1);
 1373     cmpl(Address(rscratch, 0), imm);
 1374   }
 1375 }
 1376 
 1377 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
 1378   assert(!src2.is_lval(), "use cmpptr");
 1379   assert(rscratch != noreg || always_reachable(src2), "missing");
 1380 
 1381   if (reachable(src2)) {
 1382     cmpl(src1, as_Address(src2));
 1383   } else {
 1384     lea(rscratch, src2);
 1385     cmpl(src1, Address(rscratch, 0));
 1386   }
 1387 }
 1388 
 1389 void MacroAssembler::cmp32(Register src1, int32_t imm) {
 1390   Assembler::cmpl(src1, imm);
 1391 }
 1392 
 1393 void MacroAssembler::cmp32(Register src1, Address src2) {
 1394   Assembler::cmpl(src1, src2);
 1395 }
 1396 
 1397 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1398   ucomisd(opr1, opr2);
 1399 
 1400   Label L;
 1401   if (unordered_is_less) {
 1402     movl(dst, -1);
 1403     jcc(Assembler::parity, L);
 1404     jcc(Assembler::below , L);
 1405     movl(dst, 0);
 1406     jcc(Assembler::equal , L);
 1407     increment(dst);
 1408   } else { // unordered is greater
 1409     movl(dst, 1);
 1410     jcc(Assembler::parity, L);
 1411     jcc(Assembler::above , L);
 1412     movl(dst, 0);
 1413     jcc(Assembler::equal , L);
 1414     decrementl(dst);
 1415   }
 1416   bind(L);
 1417 }
 1418 
 1419 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1420   ucomiss(opr1, opr2);
 1421 
 1422   Label L;
 1423   if (unordered_is_less) {
 1424     movl(dst, -1);
 1425     jcc(Assembler::parity, L);
 1426     jcc(Assembler::below , L);
 1427     movl(dst, 0);
 1428     jcc(Assembler::equal , L);
 1429     increment(dst);
 1430   } else { // unordered is greater
 1431     movl(dst, 1);
 1432     jcc(Assembler::parity, L);
 1433     jcc(Assembler::above , L);
 1434     movl(dst, 0);
 1435     jcc(Assembler::equal , L);
 1436     decrementl(dst);
 1437   }
 1438   bind(L);
 1439 }
 1440 
 1441 
 1442 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
 1443   assert(rscratch != noreg || always_reachable(src1), "missing");
 1444 
 1445   if (reachable(src1)) {
 1446     cmpb(as_Address(src1), imm);
 1447   } else {
 1448     lea(rscratch, src1);
 1449     cmpb(Address(rscratch, 0), imm);
 1450   }
 1451 }
 1452 
 1453 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
 1454   assert(rscratch != noreg || always_reachable(src2), "missing");
 1455 
 1456   if (src2.is_lval()) {
 1457     movptr(rscratch, src2);
 1458     Assembler::cmpq(src1, rscratch);
 1459   } else if (reachable(src2)) {
 1460     cmpq(src1, as_Address(src2));
 1461   } else {
 1462     lea(rscratch, src2);
 1463     Assembler::cmpq(src1, Address(rscratch, 0));
 1464   }
 1465 }
 1466 
 1467 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
 1468   assert(src2.is_lval(), "not a mem-mem compare");
 1469   // moves src2's literal address
 1470   movptr(rscratch, src2);
 1471   Assembler::cmpq(src1, rscratch);
 1472 }
 1473 
 1474 void MacroAssembler::cmpoop(Register src1, Register src2) {
 1475   cmpptr(src1, src2);
 1476 }
 1477 
 1478 void MacroAssembler::cmpoop(Register src1, Address src2) {
 1479   cmpptr(src1, src2);
 1480 }
 1481 
 1482 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
 1483   movoop(rscratch, src2);
 1484   cmpptr(src1, rscratch);
 1485 }
 1486 
 1487 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
 1488   assert(rscratch != noreg || always_reachable(adr), "missing");
 1489 
 1490   if (reachable(adr)) {
 1491     lock();
 1492     cmpxchgptr(reg, as_Address(adr));
 1493   } else {
 1494     lea(rscratch, adr);
 1495     lock();
 1496     cmpxchgptr(reg, Address(rscratch, 0));
 1497   }
 1498 }
 1499 
 1500 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
 1501   cmpxchgq(reg, adr);
 1502 }
 1503 
 1504 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1505   assert(rscratch != noreg || always_reachable(src), "missing");
 1506 
 1507   if (reachable(src)) {
 1508     Assembler::comisd(dst, as_Address(src));
 1509   } else {
 1510     lea(rscratch, src);
 1511     Assembler::comisd(dst, Address(rscratch, 0));
 1512   }
 1513 }
 1514 
 1515 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1516   assert(rscratch != noreg || always_reachable(src), "missing");
 1517 
 1518   if (reachable(src)) {
 1519     Assembler::comiss(dst, as_Address(src));
 1520   } else {
 1521     lea(rscratch, src);
 1522     Assembler::comiss(dst, Address(rscratch, 0));
 1523   }
 1524 }
 1525 
 1526 
 1527 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
 1528   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1529 
 1530   Condition negated_cond = negate_condition(cond);
 1531   Label L;
 1532   jcc(negated_cond, L);
 1533   pushf(); // Preserve flags
 1534   atomic_incl(counter_addr, rscratch);
 1535   popf();
 1536   bind(L);
 1537 }
 1538 
 1539 int MacroAssembler::corrected_idivl(Register reg) {
 1540   // Full implementation of Java idiv and irem; checks for
 1541   // special case as described in JVM spec., p.243 & p.271.
 1542   // The function returns the (pc) offset of the idivl
 1543   // instruction - may be needed for implicit exceptions.
 1544   //
 1545   //         normal case                           special case
 1546   //
 1547   // input : rax,: dividend                         min_int
 1548   //         reg: divisor   (may not be rax,/rdx)   -1
 1549   //
 1550   // output: rax,: quotient  (= rax, idiv reg)       min_int
 1551   //         rdx: remainder (= rax, irem reg)       0
 1552   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
 1553   const int min_int = 0x80000000;
 1554   Label normal_case, special_case;
 1555 
 1556   // check for special case
 1557   cmpl(rax, min_int);
 1558   jcc(Assembler::notEqual, normal_case);
 1559   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
 1560   cmpl(reg, -1);
 1561   jcc(Assembler::equal, special_case);
 1562 
 1563   // handle normal case
 1564   bind(normal_case);
 1565   cdql();
 1566   int idivl_offset = offset();
 1567   idivl(reg);
 1568 
 1569   // normal and special case exit
 1570   bind(special_case);
 1571 
 1572   return idivl_offset;
 1573 }
 1574 
 1575 
 1576 
 1577 void MacroAssembler::decrementl(Register reg, int value) {
 1578   if (value == min_jint) {subl(reg, value) ; return; }
 1579   if (value <  0) { incrementl(reg, -value); return; }
 1580   if (value == 0) {                        ; return; }
 1581   if (value == 1 && UseIncDec) { decl(reg) ; return; }
 1582   /* else */      { subl(reg, value)       ; return; }
 1583 }
 1584 
 1585 void MacroAssembler::decrementl(Address dst, int value) {
 1586   if (value == min_jint) {subl(dst, value) ; return; }
 1587   if (value <  0) { incrementl(dst, -value); return; }
 1588   if (value == 0) {                        ; return; }
 1589   if (value == 1 && UseIncDec) { decl(dst) ; return; }
 1590   /* else */      { subl(dst, value)       ; return; }
 1591 }
 1592 
 1593 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
 1594   assert(shift_value > 0, "illegal shift value");
 1595   Label _is_positive;
 1596   testl (reg, reg);
 1597   jcc (Assembler::positive, _is_positive);
 1598   int offset = (1 << shift_value) - 1 ;
 1599 
 1600   if (offset == 1) {
 1601     incrementl(reg);
 1602   } else {
 1603     addl(reg, offset);
 1604   }
 1605 
 1606   bind (_is_positive);
 1607   sarl(reg, shift_value);
 1608 }
 1609 
 1610 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1611   assert(rscratch != noreg || always_reachable(src), "missing");
 1612 
 1613   if (reachable(src)) {
 1614     Assembler::divsd(dst, as_Address(src));
 1615   } else {
 1616     lea(rscratch, src);
 1617     Assembler::divsd(dst, Address(rscratch, 0));
 1618   }
 1619 }
 1620 
 1621 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1622   assert(rscratch != noreg || always_reachable(src), "missing");
 1623 
 1624   if (reachable(src)) {
 1625     Assembler::divss(dst, as_Address(src));
 1626   } else {
 1627     lea(rscratch, src);
 1628     Assembler::divss(dst, Address(rscratch, 0));
 1629   }
 1630 }
 1631 
 1632 void MacroAssembler::enter() {
 1633   push(rbp);
 1634   mov(rbp, rsp);
 1635 }
 1636 
 1637 void MacroAssembler::post_call_nop() {
 1638   if (!Continuations::enabled()) {
 1639     return;
 1640   }
 1641   InstructionMark im(this);
 1642   relocate(post_call_nop_Relocation::spec());
 1643   InlineSkippedInstructionsCounter skipCounter(this);
 1644   emit_int8((uint8_t)0x0f);
 1645   emit_int8((uint8_t)0x1f);
 1646   emit_int8((uint8_t)0x84);
 1647   emit_int8((uint8_t)0x00);
 1648   emit_int32(0x00);
 1649 }
 1650 
 1651 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1652   assert(rscratch != noreg || always_reachable(src), "missing");
 1653   if (reachable(src)) {
 1654     Assembler::mulpd(dst, as_Address(src));
 1655   } else {
 1656     lea(rscratch, src);
 1657     Assembler::mulpd(dst, Address(rscratch, 0));
 1658   }
 1659 }
 1660 
 1661 // dst = c = a * b + c
 1662 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1663   Assembler::vfmadd231sd(c, a, b);
 1664   if (dst != c) {
 1665     movdbl(dst, c);
 1666   }
 1667 }
 1668 
 1669 // dst = c = a * b + c
 1670 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1671   Assembler::vfmadd231ss(c, a, b);
 1672   if (dst != c) {
 1673     movflt(dst, c);
 1674   }
 1675 }
 1676 
 1677 // dst = c = a * b + c
 1678 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1679   Assembler::vfmadd231pd(c, a, b, vector_len);
 1680   if (dst != c) {
 1681     vmovdqu(dst, c);
 1682   }
 1683 }
 1684 
 1685 // dst = c = a * b + c
 1686 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1687   Assembler::vfmadd231ps(c, a, b, vector_len);
 1688   if (dst != c) {
 1689     vmovdqu(dst, c);
 1690   }
 1691 }
 1692 
 1693 // dst = c = a * b + c
 1694 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1695   Assembler::vfmadd231pd(c, a, b, vector_len);
 1696   if (dst != c) {
 1697     vmovdqu(dst, c);
 1698   }
 1699 }
 1700 
 1701 // dst = c = a * b + c
 1702 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1703   Assembler::vfmadd231ps(c, a, b, vector_len);
 1704   if (dst != c) {
 1705     vmovdqu(dst, c);
 1706   }
 1707 }
 1708 
 1709 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
 1710   assert(rscratch != noreg || always_reachable(dst), "missing");
 1711 
 1712   if (reachable(dst)) {
 1713     incrementl(as_Address(dst));
 1714   } else {
 1715     lea(rscratch, dst);
 1716     incrementl(Address(rscratch, 0));
 1717   }
 1718 }
 1719 
 1720 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
 1721   incrementl(as_Address(dst, rscratch));
 1722 }
 1723 
 1724 void MacroAssembler::incrementl(Register reg, int value) {
 1725   if (value == min_jint) {addl(reg, value) ; return; }
 1726   if (value <  0) { decrementl(reg, -value); return; }
 1727   if (value == 0) {                        ; return; }
 1728   if (value == 1 && UseIncDec) { incl(reg) ; return; }
 1729   /* else */      { addl(reg, value)       ; return; }
 1730 }
 1731 
 1732 void MacroAssembler::incrementl(Address dst, int value) {
 1733   if (value == min_jint) {addl(dst, value) ; return; }
 1734   if (value <  0) { decrementl(dst, -value); return; }
 1735   if (value == 0) {                        ; return; }
 1736   if (value == 1 && UseIncDec) { incl(dst) ; return; }
 1737   /* else */      { addl(dst, value)       ; return; }
 1738 }
 1739 
 1740 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
 1741   assert(rscratch != noreg || always_reachable(dst), "missing");
 1742   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
 1743   if (reachable(dst)) {
 1744     jmp_literal(dst.target(), dst.rspec());
 1745   } else {
 1746     lea(rscratch, dst);
 1747     jmp(rscratch);
 1748   }
 1749 }
 1750 
 1751 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
 1752   assert(rscratch != noreg || always_reachable(dst), "missing");
 1753   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump_cc");
 1754   if (reachable(dst)) {
 1755     InstructionMark im(this);
 1756     relocate(dst.reloc());
 1757     const int short_size = 2;
 1758     const int long_size = 6;
 1759     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
 1760     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
 1761       // 0111 tttn #8-bit disp
 1762       emit_int8(0x70 | cc);
 1763       emit_int8((offs - short_size) & 0xFF);
 1764     } else {
 1765       // 0000 1111 1000 tttn #32-bit disp
 1766       emit_int8(0x0F);
 1767       emit_int8((unsigned char)(0x80 | cc));
 1768       emit_int32(offs - long_size);
 1769     }
 1770   } else {
 1771 #ifdef ASSERT
 1772     warning("reversing conditional branch");
 1773 #endif /* ASSERT */
 1774     Label skip;
 1775     jccb(reverse[cc], skip);
 1776     lea(rscratch, dst);
 1777     Assembler::jmp(rscratch);
 1778     bind(skip);
 1779   }
 1780 }
 1781 
 1782 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
 1783   ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 1784   assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
 1785 
 1786   stmxcsr(mxcsr_save);
 1787   movl(tmp, mxcsr_save);
 1788   if (EnableX86ECoreOpts) {
 1789     // The mxcsr_std has status bits set for performance on ECore
 1790     orl(tmp, 0x003f);
 1791   } else {
 1792     // Mask out status bits (only check control and mask bits)
 1793     andl(tmp, 0xFFC0);
 1794   }
 1795   cmp32(tmp, mxcsr_std, rscratch);
 1796 }
 1797 
 1798 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
 1799   assert(rscratch != noreg || always_reachable(src), "missing");
 1800 
 1801   if (reachable(src)) {
 1802     Assembler::ldmxcsr(as_Address(src));
 1803   } else {
 1804     lea(rscratch, src);
 1805     Assembler::ldmxcsr(Address(rscratch, 0));
 1806   }
 1807 }
 1808 
 1809 int MacroAssembler::load_signed_byte(Register dst, Address src) {
 1810   int off = offset();
 1811   movsbl(dst, src); // movsxb
 1812   return off;
 1813 }
 1814 
 1815 // Note: load_signed_short used to be called load_signed_word.
 1816 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
 1817 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
 1818 // The term "word" in HotSpot means a 32- or 64-bit machine word.
 1819 int MacroAssembler::load_signed_short(Register dst, Address src) {
 1820   // This is dubious to me since it seems safe to do a signed 16 => 64 bit
 1821   // version but this is what 64bit has always done. This seems to imply
 1822   // that users are only using 32bits worth.
 1823   int off = offset();
 1824   movswl(dst, src); // movsxw
 1825   return off;
 1826 }
 1827 
 1828 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
 1829   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1830   // and "3.9 Partial Register Penalties", p. 22).
 1831   int off = offset();
 1832   movzbl(dst, src); // movzxb
 1833   return off;
 1834 }
 1835 
 1836 // Note: load_unsigned_short used to be called load_unsigned_word.
 1837 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
 1838   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1839   // and "3.9 Partial Register Penalties", p. 22).
 1840   int off = offset();
 1841   movzwl(dst, src); // movzxw
 1842   return off;
 1843 }
 1844 
 1845 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
 1846   switch (size_in_bytes) {
 1847   case  8:  movq(dst, src); break;
 1848   case  4:  movl(dst, src); break;
 1849   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
 1850   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
 1851   default:  ShouldNotReachHere();
 1852   }
 1853 }
 1854 
 1855 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
 1856   switch (size_in_bytes) {
 1857   case  8:  movq(dst, src); break;
 1858   case  4:  movl(dst, src); break;
 1859   case  2:  movw(dst, src); break;
 1860   case  1:  movb(dst, src); break;
 1861   default:  ShouldNotReachHere();
 1862   }
 1863 }
 1864 
 1865 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
 1866   assert(rscratch != noreg || always_reachable(dst), "missing");
 1867 
 1868   if (reachable(dst)) {
 1869     movl(as_Address(dst), src);
 1870   } else {
 1871     lea(rscratch, dst);
 1872     movl(Address(rscratch, 0), src);
 1873   }
 1874 }
 1875 
 1876 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
 1877   if (reachable(src)) {
 1878     movl(dst, as_Address(src));
 1879   } else {
 1880     lea(dst, src);
 1881     movl(dst, Address(dst, 0));
 1882   }
 1883 }
 1884 
 1885 // C++ bool manipulation
 1886 
 1887 void MacroAssembler::movbool(Register dst, Address src) {
 1888   if(sizeof(bool) == 1)
 1889     movb(dst, src);
 1890   else if(sizeof(bool) == 2)
 1891     movw(dst, src);
 1892   else if(sizeof(bool) == 4)
 1893     movl(dst, src);
 1894   else
 1895     // unsupported
 1896     ShouldNotReachHere();
 1897 }
 1898 
 1899 void MacroAssembler::movbool(Address dst, bool boolconst) {
 1900   if(sizeof(bool) == 1)
 1901     movb(dst, (int) boolconst);
 1902   else if(sizeof(bool) == 2)
 1903     movw(dst, (int) boolconst);
 1904   else if(sizeof(bool) == 4)
 1905     movl(dst, (int) boolconst);
 1906   else
 1907     // unsupported
 1908     ShouldNotReachHere();
 1909 }
 1910 
 1911 void MacroAssembler::movbool(Address dst, Register src) {
 1912   if(sizeof(bool) == 1)
 1913     movb(dst, src);
 1914   else if(sizeof(bool) == 2)
 1915     movw(dst, src);
 1916   else if(sizeof(bool) == 4)
 1917     movl(dst, src);
 1918   else
 1919     // unsupported
 1920     ShouldNotReachHere();
 1921 }
 1922 
 1923 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1924   assert(rscratch != noreg || always_reachable(src), "missing");
 1925 
 1926   if (reachable(src)) {
 1927     movdl(dst, as_Address(src));
 1928   } else {
 1929     lea(rscratch, src);
 1930     movdl(dst, Address(rscratch, 0));
 1931   }
 1932 }
 1933 
 1934 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1935   assert(rscratch != noreg || always_reachable(src), "missing");
 1936 
 1937   if (reachable(src)) {
 1938     movq(dst, as_Address(src));
 1939   } else {
 1940     lea(rscratch, src);
 1941     movq(dst, Address(rscratch, 0));
 1942   }
 1943 }
 1944 
 1945 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1946   assert(rscratch != noreg || always_reachable(src), "missing");
 1947 
 1948   if (reachable(src)) {
 1949     if (UseXmmLoadAndClearUpper) {
 1950       movsd (dst, as_Address(src));
 1951     } else {
 1952       movlpd(dst, as_Address(src));
 1953     }
 1954   } else {
 1955     lea(rscratch, src);
 1956     if (UseXmmLoadAndClearUpper) {
 1957       movsd (dst, Address(rscratch, 0));
 1958     } else {
 1959       movlpd(dst, Address(rscratch, 0));
 1960     }
 1961   }
 1962 }
 1963 
 1964 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1965   assert(rscratch != noreg || always_reachable(src), "missing");
 1966 
 1967   if (reachable(src)) {
 1968     movss(dst, as_Address(src));
 1969   } else {
 1970     lea(rscratch, src);
 1971     movss(dst, Address(rscratch, 0));
 1972   }
 1973 }
 1974 
 1975 void MacroAssembler::movhlf(XMMRegister dst, XMMRegister src, Register rscratch) {
 1976   if (VM_Version::supports_avx10_2()) {
 1977     evmovw(dst, src);
 1978   } else {
 1979     assert(rscratch != noreg, "missing");
 1980     evmovw(rscratch, src);
 1981     evmovw(dst, rscratch);
 1982   }
 1983 }
 1984 
 1985 void MacroAssembler::mov64(Register dst, int64_t imm64) {
 1986   if (is_uimm32(imm64)) {
 1987     movl(dst, checked_cast<uint32_t>(imm64));
 1988   } else if (is_simm32(imm64)) {
 1989     movq(dst, checked_cast<int32_t>(imm64));
 1990   } else {
 1991     Assembler::mov64(dst, imm64);
 1992   }
 1993 }
 1994 
 1995 void MacroAssembler::mov64(Register dst, int64_t imm64, relocInfo::relocType rtype, int format) {
 1996   Assembler::mov64(dst, imm64, rtype, format);
 1997 }
 1998 
 1999 void MacroAssembler::movptr(Register dst, Register src) {
 2000   movq(dst, src);
 2001 }
 2002 
 2003 void MacroAssembler::movptr(Register dst, Address src) {
 2004   movq(dst, src);
 2005 }
 2006 
 2007 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 2008 void MacroAssembler::movptr(Register dst, intptr_t src) {
 2009   mov64(dst, src);
 2010 }
 2011 
 2012 void MacroAssembler::movptr(Address dst, Register src) {
 2013   movq(dst, src);
 2014 }
 2015 
 2016 void MacroAssembler::movptr(Address dst, int32_t src) {
 2017   movslq(dst, src);
 2018 }
 2019 
 2020 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 2021   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2022   Assembler::movdqu(dst, src);
 2023 }
 2024 
 2025 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 2026   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2027   Assembler::movdqu(dst, src);
 2028 }
 2029 
 2030 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2031   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2032   Assembler::movdqu(dst, src);
 2033 }
 2034 
 2035 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2036   assert(rscratch != noreg || always_reachable(src), "missing");
 2037 
 2038   if (reachable(src)) {
 2039     movdqu(dst, as_Address(src));
 2040   } else {
 2041     lea(rscratch, src);
 2042     movdqu(dst, Address(rscratch, 0));
 2043   }
 2044 }
 2045 
 2046 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2047   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2048   Assembler::vmovdqu(dst, src);
 2049 }
 2050 
 2051 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2052   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2053   Assembler::vmovdqu(dst, src);
 2054 }
 2055 
 2056 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2057   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2058   Assembler::vmovdqu(dst, src);
 2059 }
 2060 
 2061 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2062   assert(rscratch != noreg || always_reachable(src), "missing");
 2063 
 2064   if (reachable(src)) {
 2065     vmovdqu(dst, as_Address(src));
 2066   }
 2067   else {
 2068     lea(rscratch, src);
 2069     vmovdqu(dst, Address(rscratch, 0));
 2070   }
 2071 }
 2072 
 2073 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2074   assert(rscratch != noreg || always_reachable(src), "missing");
 2075 
 2076   if (vector_len == AVX_512bit) {
 2077     evmovdquq(dst, src, AVX_512bit, rscratch);
 2078   } else if (vector_len == AVX_256bit) {
 2079     vmovdqu(dst, src, rscratch);
 2080   } else {
 2081     movdqu(dst, src, rscratch);
 2082   }
 2083 }
 2084 
 2085 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
 2086   if (vector_len == AVX_512bit) {
 2087     evmovdquq(dst, src, AVX_512bit);
 2088   } else if (vector_len == AVX_256bit) {
 2089     vmovdqu(dst, src);
 2090   } else {
 2091     movdqu(dst, src);
 2092   }
 2093 }
 2094 
 2095 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
 2096   if (vector_len == AVX_512bit) {
 2097     evmovdquq(dst, src, AVX_512bit);
 2098   } else if (vector_len == AVX_256bit) {
 2099     vmovdqu(dst, src);
 2100   } else {
 2101     movdqu(dst, src);
 2102   }
 2103 }
 2104 
 2105 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
 2106   if (vector_len == AVX_512bit) {
 2107     evmovdquq(dst, src, AVX_512bit);
 2108   } else if (vector_len == AVX_256bit) {
 2109     vmovdqu(dst, src);
 2110   } else {
 2111     movdqu(dst, src);
 2112   }
 2113 }
 2114 
 2115 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2116   assert(rscratch != noreg || always_reachable(src), "missing");
 2117 
 2118   if (reachable(src)) {
 2119     vmovdqa(dst, as_Address(src));
 2120   }
 2121   else {
 2122     lea(rscratch, src);
 2123     vmovdqa(dst, Address(rscratch, 0));
 2124   }
 2125 }
 2126 
 2127 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2128   assert(rscratch != noreg || always_reachable(src), "missing");
 2129 
 2130   if (vector_len == AVX_512bit) {
 2131     evmovdqaq(dst, src, AVX_512bit, rscratch);
 2132   } else if (vector_len == AVX_256bit) {
 2133     vmovdqa(dst, src, rscratch);
 2134   } else {
 2135     movdqa(dst, src, rscratch);
 2136   }
 2137 }
 2138 
 2139 void MacroAssembler::kmov(KRegister dst, Address src) {
 2140   if (VM_Version::supports_avx512bw()) {
 2141     kmovql(dst, src);
 2142   } else {
 2143     assert(VM_Version::supports_evex(), "");
 2144     kmovwl(dst, src);
 2145   }
 2146 }
 2147 
 2148 void MacroAssembler::kmov(Address dst, KRegister src) {
 2149   if (VM_Version::supports_avx512bw()) {
 2150     kmovql(dst, src);
 2151   } else {
 2152     assert(VM_Version::supports_evex(), "");
 2153     kmovwl(dst, src);
 2154   }
 2155 }
 2156 
 2157 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2158   if (VM_Version::supports_avx512bw()) {
 2159     kmovql(dst, src);
 2160   } else {
 2161     assert(VM_Version::supports_evex(), "");
 2162     kmovwl(dst, src);
 2163   }
 2164 }
 2165 
 2166 void MacroAssembler::kmov(Register dst, KRegister src) {
 2167   if (VM_Version::supports_avx512bw()) {
 2168     kmovql(dst, src);
 2169   } else {
 2170     assert(VM_Version::supports_evex(), "");
 2171     kmovwl(dst, src);
 2172   }
 2173 }
 2174 
 2175 void MacroAssembler::kmov(KRegister dst, Register src) {
 2176   if (VM_Version::supports_avx512bw()) {
 2177     kmovql(dst, src);
 2178   } else {
 2179     assert(VM_Version::supports_evex(), "");
 2180     kmovwl(dst, src);
 2181   }
 2182 }
 2183 
 2184 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2185   assert(rscratch != noreg || always_reachable(src), "missing");
 2186 
 2187   if (reachable(src)) {
 2188     kmovql(dst, as_Address(src));
 2189   } else {
 2190     lea(rscratch, src);
 2191     kmovql(dst, Address(rscratch, 0));
 2192   }
 2193 }
 2194 
 2195 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2196   assert(rscratch != noreg || always_reachable(src), "missing");
 2197 
 2198   if (reachable(src)) {
 2199     kmovwl(dst, as_Address(src));
 2200   } else {
 2201     lea(rscratch, src);
 2202     kmovwl(dst, Address(rscratch, 0));
 2203   }
 2204 }
 2205 
 2206 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2207                                int vector_len, Register rscratch) {
 2208   assert(rscratch != noreg || always_reachable(src), "missing");
 2209 
 2210   if (reachable(src)) {
 2211     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2212   } else {
 2213     lea(rscratch, src);
 2214     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2215   }
 2216 }
 2217 
 2218 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2219                                int vector_len, Register rscratch) {
 2220   assert(rscratch != noreg || always_reachable(src), "missing");
 2221 
 2222   if (reachable(src)) {
 2223     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2224   } else {
 2225     lea(rscratch, src);
 2226     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2227   }
 2228 }
 2229 
 2230 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2231   assert(rscratch != noreg || always_reachable(src), "missing");
 2232 
 2233   if (reachable(src)) {
 2234     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2235   } else {
 2236     lea(rscratch, src);
 2237     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2238   }
 2239 }
 2240 
 2241 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2242   assert(rscratch != noreg || always_reachable(src), "missing");
 2243 
 2244   if (reachable(src)) {
 2245     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2246   } else {
 2247     lea(rscratch, src);
 2248     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2249   }
 2250 }
 2251 
 2252 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2253   assert(rscratch != noreg || always_reachable(src), "missing");
 2254 
 2255   if (reachable(src)) {
 2256     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2257   } else {
 2258     lea(rscratch, src);
 2259     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2260   }
 2261 }
 2262 
 2263 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2264   assert(rscratch != noreg || always_reachable(src), "missing");
 2265 
 2266   if (reachable(src)) {
 2267     Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
 2268   } else {
 2269     lea(rscratch, src);
 2270     Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2271   }
 2272 }
 2273 
 2274 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2275   assert(rscratch != noreg || always_reachable(src), "missing");
 2276 
 2277   if (reachable(src)) {
 2278     Assembler::evmovdqaq(dst, as_Address(src), vector_len);
 2279   } else {
 2280     lea(rscratch, src);
 2281     Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
 2282   }
 2283 }
 2284 
 2285 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2286   assert(rscratch != noreg || always_reachable(src), "missing");
 2287 
 2288   if (reachable(src)) {
 2289     Assembler::movapd(dst, as_Address(src));
 2290   } else {
 2291     lea(rscratch, src);
 2292     Assembler::movapd(dst, Address(rscratch, 0));
 2293   }
 2294 }
 2295 
 2296 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2297   assert(rscratch != noreg || always_reachable(src), "missing");
 2298 
 2299   if (reachable(src)) {
 2300     Assembler::movdqa(dst, as_Address(src));
 2301   } else {
 2302     lea(rscratch, src);
 2303     Assembler::movdqa(dst, Address(rscratch, 0));
 2304   }
 2305 }
 2306 
 2307 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2308   assert(rscratch != noreg || always_reachable(src), "missing");
 2309 
 2310   if (reachable(src)) {
 2311     Assembler::movsd(dst, as_Address(src));
 2312   } else {
 2313     lea(rscratch, src);
 2314     Assembler::movsd(dst, Address(rscratch, 0));
 2315   }
 2316 }
 2317 
 2318 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2319   assert(rscratch != noreg || always_reachable(src), "missing");
 2320 
 2321   if (reachable(src)) {
 2322     Assembler::movss(dst, as_Address(src));
 2323   } else {
 2324     lea(rscratch, src);
 2325     Assembler::movss(dst, Address(rscratch, 0));
 2326   }
 2327 }
 2328 
 2329 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2330   assert(rscratch != noreg || always_reachable(src), "missing");
 2331 
 2332   if (reachable(src)) {
 2333     Assembler::movddup(dst, as_Address(src));
 2334   } else {
 2335     lea(rscratch, src);
 2336     Assembler::movddup(dst, Address(rscratch, 0));
 2337   }
 2338 }
 2339 
 2340 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2341   assert(rscratch != noreg || always_reachable(src), "missing");
 2342 
 2343   if (reachable(src)) {
 2344     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2345   } else {
 2346     lea(rscratch, src);
 2347     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2348   }
 2349 }
 2350 
 2351 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2352   assert(rscratch != noreg || always_reachable(src), "missing");
 2353 
 2354   if (reachable(src)) {
 2355     Assembler::mulsd(dst, as_Address(src));
 2356   } else {
 2357     lea(rscratch, src);
 2358     Assembler::mulsd(dst, Address(rscratch, 0));
 2359   }
 2360 }
 2361 
 2362 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2363   assert(rscratch != noreg || always_reachable(src), "missing");
 2364 
 2365   if (reachable(src)) {
 2366     Assembler::mulss(dst, as_Address(src));
 2367   } else {
 2368     lea(rscratch, src);
 2369     Assembler::mulss(dst, Address(rscratch, 0));
 2370   }
 2371 }
 2372 
 2373 void MacroAssembler::null_check(Register reg, int offset) {
 2374   if (needs_explicit_null_check(offset)) {
 2375     // provoke OS null exception if reg is null by
 2376     // accessing M[reg] w/o changing any (non-CC) registers
 2377     // NOTE: cmpl is plenty here to provoke a segv
 2378     cmpptr(rax, Address(reg, 0));
 2379     // Note: should probably use testl(rax, Address(reg, 0));
 2380     //       may be shorter code (however, this version of
 2381     //       testl needs to be implemented first)
 2382   } else {
 2383     // nothing to do, (later) access of M[reg + offset]
 2384     // will provoke OS null exception if reg is null
 2385   }
 2386 }
 2387 
 2388 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
 2389   andptr(markword, markWord::inline_type_pattern_mask);
 2390   cmpptr(markword, markWord::inline_type_pattern);
 2391   jcc(Assembler::equal, is_inline_type);
 2392 }
 2393 
 2394 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
 2395   if (can_be_null) {
 2396     testptr(object, object);
 2397     jcc(Assembler::zero, not_inline_type);
 2398   }
 2399   const int is_inline_type_mask = markWord::inline_type_pattern;
 2400   movptr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
 2401   andptr(tmp, is_inline_type_mask);
 2402   cmpptr(tmp, is_inline_type_mask);
 2403   jcc(Assembler::notEqual, not_inline_type);
 2404 }
 2405 
 2406 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
 2407   movl(temp_reg, flags);
 2408   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2409   jcc(Assembler::notEqual, is_null_free_inline_type);
 2410 }
 2411 
 2412 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
 2413   movl(temp_reg, flags);
 2414   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2415   jcc(Assembler::equal, not_null_free_inline_type);
 2416 }
 2417 
 2418 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
 2419   movl(temp_reg, flags);
 2420   testl(temp_reg, 1 << ResolvedFieldEntry::is_flat_shift);
 2421   jcc(Assembler::notEqual, is_flat);
 2422 }
 2423 
 2424 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
 2425   movl(temp_reg, flags);
 2426   testl(temp_reg, 1 << ResolvedFieldEntry::has_null_marker_shift);
 2427   jcc(Assembler::notEqual, has_null_marker);
 2428 }
 2429 
 2430 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
 2431   Label test_mark_word;
 2432   // load mark word
 2433   movptr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
 2434   // check displaced
 2435   testl(temp_reg, markWord::unlocked_value);
 2436   jccb(Assembler::notZero, test_mark_word);
 2437   // slow path use klass prototype
 2438   push(rscratch1);
 2439   load_prototype_header(temp_reg, oop, rscratch1);
 2440   pop(rscratch1);
 2441 
 2442   bind(test_mark_word);
 2443   testl(temp_reg, test_bit);
 2444   jcc((jmp_set) ? Assembler::notZero : Assembler::zero, jmp_label);
 2445 }
 2446 
 2447 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg,
 2448                                          Label& is_flat_array) {
 2449 #ifdef _LP64
 2450   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
 2451 #else
 2452   load_klass(temp_reg, oop, noreg);
 2453   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 2454   test_flat_array_layout(temp_reg, is_flat_array);
 2455 #endif
 2456 }
 2457 
 2458 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
 2459                                              Label& is_non_flat_array) {
 2460 #ifdef _LP64
 2461   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
 2462 #else
 2463   load_klass(temp_reg, oop, noreg);
 2464   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 2465   test_non_flat_array_layout(temp_reg, is_non_flat_array);
 2466 #endif
 2467 }
 2468 
 2469 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label&is_null_free_array) {
 2470 #ifdef _LP64
 2471   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
 2472 #else
 2473   Unimplemented();
 2474 #endif
 2475 }
 2476 
 2477 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
 2478 #ifdef _LP64
 2479   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
 2480 #else
 2481   Unimplemented();
 2482 #endif
 2483 }
 2484 
 2485 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
 2486   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2487   jcc(Assembler::notZero, is_flat_array);
 2488 }
 2489 
 2490 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
 2491   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2492   jcc(Assembler::zero, is_non_flat_array);
 2493 }
 2494 
 2495 void MacroAssembler::os_breakpoint() {
 2496   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2497   // (e.g., MSVC can't call ps() otherwise)
 2498   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2499 }
 2500 
 2501 void MacroAssembler::unimplemented(const char* what) {
 2502   const char* buf = nullptr;
 2503   {
 2504     ResourceMark rm;
 2505     stringStream ss;
 2506     ss.print("unimplemented: %s", what);
 2507     buf = code_string(ss.as_string());
 2508   }
 2509   stop(buf);
 2510 }
 2511 
 2512 #define XSTATE_BV 0x200
 2513 
 2514 void MacroAssembler::pop_CPU_state() {
 2515   pop_FPU_state();
 2516   pop_IU_state();
 2517 }
 2518 
 2519 void MacroAssembler::pop_FPU_state() {
 2520   fxrstor(Address(rsp, 0));
 2521   addptr(rsp, FPUStateSizeInWords * wordSize);
 2522 }
 2523 
 2524 void MacroAssembler::pop_IU_state() {
 2525   popa();
 2526   addq(rsp, 8);
 2527   popf();
 2528 }
 2529 
 2530 // Save Integer and Float state
 2531 // Warning: Stack must be 16 byte aligned (64bit)
 2532 void MacroAssembler::push_CPU_state() {
 2533   push_IU_state();
 2534   push_FPU_state();
 2535 }
 2536 
 2537 void MacroAssembler::push_FPU_state() {
 2538   subptr(rsp, FPUStateSizeInWords * wordSize);
 2539   fxsave(Address(rsp, 0));
 2540 }
 2541 
 2542 void MacroAssembler::push_IU_state() {
 2543   // Push flags first because pusha kills them
 2544   pushf();
 2545   // Make sure rsp stays 16-byte aligned
 2546   subq(rsp, 8);
 2547   pusha();
 2548 }
 2549 
 2550 void MacroAssembler::push_cont_fastpath() {
 2551   if (!Continuations::enabled()) return;
 2552 
 2553   Label L_done;
 2554   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2555   jccb(Assembler::belowEqual, L_done);
 2556   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2557   bind(L_done);
 2558 }
 2559 
 2560 void MacroAssembler::pop_cont_fastpath() {
 2561   if (!Continuations::enabled()) return;
 2562 
 2563   Label L_done;
 2564   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2565   jccb(Assembler::below, L_done);
 2566   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2567   bind(L_done);
 2568 }
 2569 
 2570 #ifdef ASSERT
 2571 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2572   Label no_cont;
 2573   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2574   testl(cont, cont);
 2575   jcc(Assembler::zero, no_cont);
 2576   stop(name);
 2577   bind(no_cont);
 2578 }
 2579 #endif
 2580 
 2581 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2582   // we must set sp to zero to clear frame
 2583   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2584   // must clear fp, so that compiled frames are not confused; it is
 2585   // possible that we need it only for debugging
 2586   if (clear_fp) {
 2587     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2588   }
 2589   // Always clear the pc because it could have been set by make_walkable()
 2590   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2591   vzeroupper();
 2592 }
 2593 
 2594 void MacroAssembler::round_to(Register reg, int modulus) {
 2595   addptr(reg, modulus - 1);
 2596   andptr(reg, -modulus);
 2597 }
 2598 
 2599 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2600   if (at_return) {
 2601     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2602     // we may safely use rsp instead to perform the stack watermark check.
 2603     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2604     jcc(Assembler::above, slow_path);
 2605     return;
 2606   }
 2607   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2608   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2609 }
 2610 
 2611 // Calls to C land
 2612 //
 2613 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2614 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2615 // has to be reset to 0. This is required to allow proper stack traversal.
 2616 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2617                                          Register last_java_fp,
 2618                                          address  last_java_pc,
 2619                                          Register rscratch) {
 2620   vzeroupper();
 2621   // determine last_java_sp register
 2622   if (!last_java_sp->is_valid()) {
 2623     last_java_sp = rsp;
 2624   }
 2625   // last_java_fp is optional
 2626   if (last_java_fp->is_valid()) {
 2627     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2628   }
 2629   // last_java_pc is optional
 2630   if (last_java_pc != nullptr) {
 2631     Address java_pc(r15_thread,
 2632                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2633     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2634   }
 2635   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2636 }
 2637 
 2638 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2639                                          Register last_java_fp,
 2640                                          Label &L,
 2641                                          Register scratch) {
 2642   lea(scratch, L);
 2643   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2644   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2645 }
 2646 
 2647 void MacroAssembler::shlptr(Register dst, int imm8) {
 2648   shlq(dst, imm8);
 2649 }
 2650 
 2651 void MacroAssembler::shrptr(Register dst, int imm8) {
 2652   shrq(dst, imm8);
 2653 }
 2654 
 2655 void MacroAssembler::sign_extend_byte(Register reg) {
 2656   movsbl(reg, reg); // movsxb
 2657 }
 2658 
 2659 void MacroAssembler::sign_extend_short(Register reg) {
 2660   movswl(reg, reg); // movsxw
 2661 }
 2662 
 2663 void MacroAssembler::narrow_subword_type(Register reg, BasicType bt) {
 2664   assert(is_subword_type(bt), "required");
 2665   switch (bt) {
 2666   case T_BOOLEAN: andl(reg, 1); break;
 2667   case T_BYTE:    movsbl(reg, reg); break;
 2668   case T_CHAR:    movzwl(reg, reg); break;
 2669   case T_SHORT:   movswl(reg, reg); break;
 2670   default:        ShouldNotReachHere();
 2671   }
 2672 }
 2673 
 2674 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2675   if (imm32 >= 0 && is8bit(imm32)) {
 2676     testb(dst, imm32);
 2677   } else {
 2678     Assembler::testl(dst, imm32);
 2679   }
 2680 }
 2681 
 2682 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2683   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2684     testb(dst, imm32);
 2685   } else {
 2686     Assembler::testl(dst, imm32);
 2687   }
 2688 }
 2689 
 2690 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2691   assert(always_reachable(src), "Address should be reachable");
 2692   testl(dst, as_Address(src));
 2693 }
 2694 
 2695 void MacroAssembler::testq(Address dst, int32_t imm32) {
 2696   if (imm32 >= 0) {
 2697     testl(dst, imm32);
 2698   } else {
 2699     Assembler::testq(dst, imm32);
 2700   }
 2701 }
 2702 
 2703 void MacroAssembler::testq(Register dst, int32_t imm32) {
 2704   if (imm32 >= 0) {
 2705     testl(dst, imm32);
 2706   } else {
 2707     Assembler::testq(dst, imm32);
 2708   }
 2709 }
 2710 
 2711 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2712   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2713   Assembler::pcmpeqb(dst, src);
 2714 }
 2715 
 2716 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2717   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2718   Assembler::pcmpeqw(dst, src);
 2719 }
 2720 
 2721 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2722   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2723   Assembler::pcmpestri(dst, src, imm8);
 2724 }
 2725 
 2726 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2727   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2728   Assembler::pcmpestri(dst, src, imm8);
 2729 }
 2730 
 2731 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2732   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2733   Assembler::pmovzxbw(dst, src);
 2734 }
 2735 
 2736 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2737   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2738   Assembler::pmovzxbw(dst, src);
 2739 }
 2740 
 2741 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2742   assert((src->encoding() < 16),"XMM register should be 0-15");
 2743   Assembler::pmovmskb(dst, src);
 2744 }
 2745 
 2746 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2747   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2748   Assembler::ptest(dst, src);
 2749 }
 2750 
 2751 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2752   assert(rscratch != noreg || always_reachable(src), "missing");
 2753 
 2754   if (reachable(src)) {
 2755     Assembler::sqrtss(dst, as_Address(src));
 2756   } else {
 2757     lea(rscratch, src);
 2758     Assembler::sqrtss(dst, Address(rscratch, 0));
 2759   }
 2760 }
 2761 
 2762 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2763   assert(rscratch != noreg || always_reachable(src), "missing");
 2764 
 2765   if (reachable(src)) {
 2766     Assembler::subsd(dst, as_Address(src));
 2767   } else {
 2768     lea(rscratch, src);
 2769     Assembler::subsd(dst, Address(rscratch, 0));
 2770   }
 2771 }
 2772 
 2773 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2774   assert(rscratch != noreg || always_reachable(src), "missing");
 2775 
 2776   if (reachable(src)) {
 2777     Assembler::roundsd(dst, as_Address(src), rmode);
 2778   } else {
 2779     lea(rscratch, src);
 2780     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2781   }
 2782 }
 2783 
 2784 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2785   assert(rscratch != noreg || always_reachable(src), "missing");
 2786 
 2787   if (reachable(src)) {
 2788     Assembler::subss(dst, as_Address(src));
 2789   } else {
 2790     lea(rscratch, src);
 2791     Assembler::subss(dst, Address(rscratch, 0));
 2792   }
 2793 }
 2794 
 2795 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2796   assert(rscratch != noreg || always_reachable(src), "missing");
 2797 
 2798   if (reachable(src)) {
 2799     Assembler::ucomisd(dst, as_Address(src));
 2800   } else {
 2801     lea(rscratch, src);
 2802     Assembler::ucomisd(dst, Address(rscratch, 0));
 2803   }
 2804 }
 2805 
 2806 void MacroAssembler::evucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2807   assert(rscratch != noreg || always_reachable(src), "missing");
 2808 
 2809   if (reachable(src)) {
 2810     Assembler::evucomxsd(dst, as_Address(src));
 2811   } else {
 2812     lea(rscratch, src);
 2813     Assembler::evucomxsd(dst, Address(rscratch, 0));
 2814   }
 2815 }
 2816 
 2817 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2818   assert(rscratch != noreg || always_reachable(src), "missing");
 2819 
 2820   if (reachable(src)) {
 2821     Assembler::ucomiss(dst, as_Address(src));
 2822   } else {
 2823     lea(rscratch, src);
 2824     Assembler::ucomiss(dst, Address(rscratch, 0));
 2825   }
 2826 }
 2827 
 2828 void MacroAssembler::evucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2829   assert(rscratch != noreg || always_reachable(src), "missing");
 2830 
 2831   if (reachable(src)) {
 2832     Assembler::evucomxss(dst, as_Address(src));
 2833   } else {
 2834     lea(rscratch, src);
 2835     Assembler::evucomxss(dst, Address(rscratch, 0));
 2836   }
 2837 }
 2838 
 2839 void MacroAssembler::evucomish(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2840   assert(rscratch != noreg || always_reachable(src), "missing");
 2841 
 2842   if (reachable(src)) {
 2843     Assembler::evucomish(dst, as_Address(src));
 2844   } else {
 2845     lea(rscratch, src);
 2846     Assembler::evucomish(dst, Address(rscratch, 0));
 2847   }
 2848 }
 2849 
 2850 void MacroAssembler::evucomxsh(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2851   assert(rscratch != noreg || always_reachable(src), "missing");
 2852 
 2853   if (reachable(src)) {
 2854     Assembler::evucomxsh(dst, as_Address(src));
 2855   } else {
 2856     lea(rscratch, src);
 2857     Assembler::evucomxsh(dst, Address(rscratch, 0));
 2858   }
 2859 }
 2860 
 2861 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2862   assert(rscratch != noreg || always_reachable(src), "missing");
 2863 
 2864   // Used in sign-bit flipping with aligned address.
 2865   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2866 
 2867   if (UseAVX > 2 &&
 2868       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2869       (dst->encoding() >= 16)) {
 2870     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2871   } else if (reachable(src)) {
 2872     Assembler::xorpd(dst, as_Address(src));
 2873   } else {
 2874     lea(rscratch, src);
 2875     Assembler::xorpd(dst, Address(rscratch, 0));
 2876   }
 2877 }
 2878 
 2879 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2880   if (UseAVX > 2 &&
 2881       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2882       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2883     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2884   } else {
 2885     Assembler::xorpd(dst, src);
 2886   }
 2887 }
 2888 
 2889 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2890   if (UseAVX > 2 &&
 2891       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2892       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2893     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2894   } else {
 2895     Assembler::xorps(dst, src);
 2896   }
 2897 }
 2898 
 2899 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2900   assert(rscratch != noreg || always_reachable(src), "missing");
 2901 
 2902   // Used in sign-bit flipping with aligned address.
 2903   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2904 
 2905   if (UseAVX > 2 &&
 2906       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2907       (dst->encoding() >= 16)) {
 2908     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2909   } else if (reachable(src)) {
 2910     Assembler::xorps(dst, as_Address(src));
 2911   } else {
 2912     lea(rscratch, src);
 2913     Assembler::xorps(dst, Address(rscratch, 0));
 2914   }
 2915 }
 2916 
 2917 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2918   assert(rscratch != noreg || always_reachable(src), "missing");
 2919 
 2920   // Used in sign-bit flipping with aligned address.
 2921   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2922   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2923   if (reachable(src)) {
 2924     Assembler::pshufb(dst, as_Address(src));
 2925   } else {
 2926     lea(rscratch, src);
 2927     Assembler::pshufb(dst, Address(rscratch, 0));
 2928   }
 2929 }
 2930 
 2931 // AVX 3-operands instructions
 2932 
 2933 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2934   assert(rscratch != noreg || always_reachable(src), "missing");
 2935 
 2936   if (reachable(src)) {
 2937     vaddsd(dst, nds, as_Address(src));
 2938   } else {
 2939     lea(rscratch, src);
 2940     vaddsd(dst, nds, Address(rscratch, 0));
 2941   }
 2942 }
 2943 
 2944 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2945   assert(rscratch != noreg || always_reachable(src), "missing");
 2946 
 2947   if (reachable(src)) {
 2948     vaddss(dst, nds, as_Address(src));
 2949   } else {
 2950     lea(rscratch, src);
 2951     vaddss(dst, nds, Address(rscratch, 0));
 2952   }
 2953 }
 2954 
 2955 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2956   assert(UseAVX > 0, "requires some form of AVX");
 2957   assert(rscratch != noreg || always_reachable(src), "missing");
 2958 
 2959   if (reachable(src)) {
 2960     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2961   } else {
 2962     lea(rscratch, src);
 2963     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2964   }
 2965 }
 2966 
 2967 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2968   assert(UseAVX > 0, "requires some form of AVX");
 2969   assert(rscratch != noreg || always_reachable(src), "missing");
 2970 
 2971   if (reachable(src)) {
 2972     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2973   } else {
 2974     lea(rscratch, src);
 2975     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2976   }
 2977 }
 2978 
 2979 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2980   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2981   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2982 
 2983   vandps(dst, nds, negate_field, vector_len, rscratch);
 2984 }
 2985 
 2986 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2987   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2988   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2989 
 2990   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2991 }
 2992 
 2993 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2994   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2995   Assembler::vpaddb(dst, nds, src, vector_len);
 2996 }
 2997 
 2998 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2999   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3000   Assembler::vpaddb(dst, nds, src, vector_len);
 3001 }
 3002 
 3003 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3004   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3005   Assembler::vpaddw(dst, nds, src, vector_len);
 3006 }
 3007 
 3008 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3009   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3010   Assembler::vpaddw(dst, nds, src, vector_len);
 3011 }
 3012 
 3013 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3014   assert(rscratch != noreg || always_reachable(src), "missing");
 3015 
 3016   if (reachable(src)) {
 3017     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 3018   } else {
 3019     lea(rscratch, src);
 3020     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 3021   }
 3022 }
 3023 
 3024 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3025   assert(rscratch != noreg || always_reachable(src), "missing");
 3026 
 3027   if (reachable(src)) {
 3028     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 3029   } else {
 3030     lea(rscratch, src);
 3031     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 3032   }
 3033 }
 3034 
 3035 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3036   assert(rscratch != noreg || always_reachable(src), "missing");
 3037 
 3038   if (reachable(src)) {
 3039     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 3040   } else {
 3041     lea(rscratch, src);
 3042     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 3043   }
 3044 }
 3045 
 3046 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3047   assert(rscratch != noreg || always_reachable(src), "missing");
 3048 
 3049   if (reachable(src)) {
 3050     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 3051   } else {
 3052     lea(rscratch, src);
 3053     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 3054   }
 3055 }
 3056 
 3057 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3058   assert(rscratch != noreg || always_reachable(src), "missing");
 3059 
 3060   if (reachable(src)) {
 3061     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 3062   } else {
 3063     lea(rscratch, src);
 3064     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 3065   }
 3066 }
 3067 
 3068 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3069   assert(rscratch != noreg || always_reachable(src), "missing");
 3070 
 3071   if (reachable(src)) {
 3072     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 3073   } else {
 3074     lea(rscratch, src);
 3075     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 3076   }
 3077 }
 3078 
 3079 // Vector float blend
 3080 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3081 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3082   // WARN: Allow dst == (src1|src2), mask == scratch
 3083   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3084                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3085   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 3086   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3087   if (blend_emulation && scratch_available && dst_available) {
 3088     if (compute_mask) {
 3089       vpsrad(scratch, mask, 32, vector_len);
 3090       mask = scratch;
 3091     }
 3092     if (dst == src1) {
 3093       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 3094       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3095     } else {
 3096       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3097       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 3098     }
 3099     vpor(dst, dst, scratch, vector_len);
 3100   } else {
 3101     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 3102   }
 3103 }
 3104 
 3105 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3106 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3107   // WARN: Allow dst == (src1|src2), mask == scratch
 3108   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3109                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3110   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 3111   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3112   if (blend_emulation && scratch_available && dst_available) {
 3113     if (compute_mask) {
 3114       vpxor(scratch, scratch, scratch, vector_len);
 3115       vpcmpgtq(scratch, scratch, mask, vector_len);
 3116       mask = scratch;
 3117     }
 3118     if (dst == src1) {
 3119       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 3120       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3121     } else {
 3122       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3123       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 3124     }
 3125     vpor(dst, dst, scratch, vector_len);
 3126   } else {
 3127     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 3128   }
 3129 }
 3130 
 3131 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3132   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3133   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 3134 }
 3135 
 3136 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 3137   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3138   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 3139 }
 3140 
 3141 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3142   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3143   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3144 }
 3145 
 3146 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3147   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3148   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3149 }
 3150 
 3151 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3152   assert(rscratch != noreg || always_reachable(src), "missing");
 3153 
 3154   if (reachable(src)) {
 3155     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 3156   } else {
 3157     lea(rscratch, src);
 3158     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 3159   }
 3160 }
 3161 
 3162 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3163                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3164   assert(rscratch != noreg || always_reachable(src), "missing");
 3165 
 3166   if (reachable(src)) {
 3167     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3168   } else {
 3169     lea(rscratch, src);
 3170     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3171   }
 3172 }
 3173 
 3174 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3175                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3176   assert(rscratch != noreg || always_reachable(src), "missing");
 3177 
 3178   if (reachable(src)) {
 3179     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3180   } else {
 3181     lea(rscratch, src);
 3182     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3183   }
 3184 }
 3185 
 3186 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3187                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3188   assert(rscratch != noreg || always_reachable(src), "missing");
 3189 
 3190   if (reachable(src)) {
 3191     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3192   } else {
 3193     lea(rscratch, src);
 3194     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3195   }
 3196 }
 3197 
 3198 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3199                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3200   assert(rscratch != noreg || always_reachable(src), "missing");
 3201 
 3202   if (reachable(src)) {
 3203     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3204   } else {
 3205     lea(rscratch, src);
 3206     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3207   }
 3208 }
 3209 
 3210 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3211   if (width == Assembler::Q) {
 3212     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3213   } else {
 3214     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3215   }
 3216 }
 3217 
 3218 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3219   int eq_cond_enc = 0x29;
 3220   int gt_cond_enc = 0x37;
 3221   if (width != Assembler::Q) {
 3222     eq_cond_enc = 0x74 + width;
 3223     gt_cond_enc = 0x64 + width;
 3224   }
 3225   switch (cond) {
 3226   case eq:
 3227     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3228     break;
 3229   case neq:
 3230     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3231     vallones(xtmp, vector_len);
 3232     vpxor(dst, xtmp, dst, vector_len);
 3233     break;
 3234   case le:
 3235     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3236     vallones(xtmp, vector_len);
 3237     vpxor(dst, xtmp, dst, vector_len);
 3238     break;
 3239   case nlt:
 3240     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3241     vallones(xtmp, vector_len);
 3242     vpxor(dst, xtmp, dst, vector_len);
 3243     break;
 3244   case lt:
 3245     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3246     break;
 3247   case nle:
 3248     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3249     break;
 3250   default:
 3251     assert(false, "Should not reach here");
 3252   }
 3253 }
 3254 
 3255 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3256   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3257   Assembler::vpmovzxbw(dst, src, vector_len);
 3258 }
 3259 
 3260 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3261   assert((src->encoding() < 16),"XMM register should be 0-15");
 3262   Assembler::vpmovmskb(dst, src, vector_len);
 3263 }
 3264 
 3265 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3266   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3267   Assembler::vpmullw(dst, nds, src, vector_len);
 3268 }
 3269 
 3270 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3271   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3272   Assembler::vpmullw(dst, nds, src, vector_len);
 3273 }
 3274 
 3275 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3276   assert((UseAVX > 0), "AVX support is needed");
 3277   assert(rscratch != noreg || always_reachable(src), "missing");
 3278 
 3279   if (reachable(src)) {
 3280     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3281   } else {
 3282     lea(rscratch, src);
 3283     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3284   }
 3285 }
 3286 
 3287 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3288   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3289   Assembler::vpsubb(dst, nds, src, vector_len);
 3290 }
 3291 
 3292 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3293   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3294   Assembler::vpsubb(dst, nds, src, vector_len);
 3295 }
 3296 
 3297 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3298   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3299   Assembler::vpsubw(dst, nds, src, vector_len);
 3300 }
 3301 
 3302 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3303   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3304   Assembler::vpsubw(dst, nds, src, vector_len);
 3305 }
 3306 
 3307 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3308   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3309   Assembler::vpsraw(dst, nds, shift, vector_len);
 3310 }
 3311 
 3312 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3313   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3314   Assembler::vpsraw(dst, nds, shift, vector_len);
 3315 }
 3316 
 3317 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3318   assert(UseAVX > 2,"");
 3319   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3320      vector_len = 2;
 3321   }
 3322   Assembler::evpsraq(dst, nds, shift, vector_len);
 3323 }
 3324 
 3325 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3326   assert(UseAVX > 2,"");
 3327   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3328      vector_len = 2;
 3329   }
 3330   Assembler::evpsraq(dst, nds, shift, vector_len);
 3331 }
 3332 
 3333 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3334   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3335   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3336 }
 3337 
 3338 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3339   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3340   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3341 }
 3342 
 3343 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3344   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3345   Assembler::vpsllw(dst, nds, shift, vector_len);
 3346 }
 3347 
 3348 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3349   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3350   Assembler::vpsllw(dst, nds, shift, vector_len);
 3351 }
 3352 
 3353 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3354   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3355   Assembler::vptest(dst, src);
 3356 }
 3357 
 3358 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3359   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3360   Assembler::punpcklbw(dst, src);
 3361 }
 3362 
 3363 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3364   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3365   Assembler::pshufd(dst, src, mode);
 3366 }
 3367 
 3368 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3369   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3370   Assembler::pshuflw(dst, src, mode);
 3371 }
 3372 
 3373 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3374   assert(rscratch != noreg || always_reachable(src), "missing");
 3375 
 3376   if (reachable(src)) {
 3377     vandpd(dst, nds, as_Address(src), vector_len);
 3378   } else {
 3379     lea(rscratch, src);
 3380     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3381   }
 3382 }
 3383 
 3384 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3385   assert(rscratch != noreg || always_reachable(src), "missing");
 3386 
 3387   if (reachable(src)) {
 3388     vandps(dst, nds, as_Address(src), vector_len);
 3389   } else {
 3390     lea(rscratch, src);
 3391     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3392   }
 3393 }
 3394 
 3395 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3396                             bool merge, int vector_len, Register rscratch) {
 3397   assert(rscratch != noreg || always_reachable(src), "missing");
 3398 
 3399   if (reachable(src)) {
 3400     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3401   } else {
 3402     lea(rscratch, src);
 3403     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3404   }
 3405 }
 3406 
 3407 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3408   assert(rscratch != noreg || always_reachable(src), "missing");
 3409 
 3410   if (reachable(src)) {
 3411     vdivsd(dst, nds, as_Address(src));
 3412   } else {
 3413     lea(rscratch, src);
 3414     vdivsd(dst, nds, Address(rscratch, 0));
 3415   }
 3416 }
 3417 
 3418 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3419   assert(rscratch != noreg || always_reachable(src), "missing");
 3420 
 3421   if (reachable(src)) {
 3422     vdivss(dst, nds, as_Address(src));
 3423   } else {
 3424     lea(rscratch, src);
 3425     vdivss(dst, nds, Address(rscratch, 0));
 3426   }
 3427 }
 3428 
 3429 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3430   assert(rscratch != noreg || always_reachable(src), "missing");
 3431 
 3432   if (reachable(src)) {
 3433     vmulsd(dst, nds, as_Address(src));
 3434   } else {
 3435     lea(rscratch, src);
 3436     vmulsd(dst, nds, Address(rscratch, 0));
 3437   }
 3438 }
 3439 
 3440 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3441   assert(rscratch != noreg || always_reachable(src), "missing");
 3442 
 3443   if (reachable(src)) {
 3444     vmulss(dst, nds, as_Address(src));
 3445   } else {
 3446     lea(rscratch, src);
 3447     vmulss(dst, nds, Address(rscratch, 0));
 3448   }
 3449 }
 3450 
 3451 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3452   assert(rscratch != noreg || always_reachable(src), "missing");
 3453 
 3454   if (reachable(src)) {
 3455     vsubsd(dst, nds, as_Address(src));
 3456   } else {
 3457     lea(rscratch, src);
 3458     vsubsd(dst, nds, Address(rscratch, 0));
 3459   }
 3460 }
 3461 
 3462 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3463   assert(rscratch != noreg || always_reachable(src), "missing");
 3464 
 3465   if (reachable(src)) {
 3466     vsubss(dst, nds, as_Address(src));
 3467   } else {
 3468     lea(rscratch, src);
 3469     vsubss(dst, nds, Address(rscratch, 0));
 3470   }
 3471 }
 3472 
 3473 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3474   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3475   assert(rscratch != noreg || always_reachable(src), "missing");
 3476 
 3477   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3478 }
 3479 
 3480 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3481   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3482   assert(rscratch != noreg || always_reachable(src), "missing");
 3483 
 3484   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3485 }
 3486 
 3487 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3488   assert(rscratch != noreg || always_reachable(src), "missing");
 3489 
 3490   if (reachable(src)) {
 3491     vxorpd(dst, nds, as_Address(src), vector_len);
 3492   } else {
 3493     lea(rscratch, src);
 3494     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3495   }
 3496 }
 3497 
 3498 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3499   assert(rscratch != noreg || always_reachable(src), "missing");
 3500 
 3501   if (reachable(src)) {
 3502     vxorps(dst, nds, as_Address(src), vector_len);
 3503   } else {
 3504     lea(rscratch, src);
 3505     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3506   }
 3507 }
 3508 
 3509 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3510   assert(rscratch != noreg || always_reachable(src), "missing");
 3511 
 3512   if (UseAVX > 1 || (vector_len < 1)) {
 3513     if (reachable(src)) {
 3514       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3515     } else {
 3516       lea(rscratch, src);
 3517       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3518     }
 3519   } else {
 3520     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3521   }
 3522 }
 3523 
 3524 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3525   assert(rscratch != noreg || always_reachable(src), "missing");
 3526 
 3527   if (reachable(src)) {
 3528     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3529   } else {
 3530     lea(rscratch, src);
 3531     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3532   }
 3533 }
 3534 
 3535 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3536   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3537   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3538   // The inverted mask is sign-extended
 3539   andptr(possibly_non_local, inverted_mask);
 3540 }
 3541 
 3542 void MacroAssembler::resolve_jobject(Register value,
 3543                                      Register tmp) {
 3544   Register thread = r15_thread;
 3545   assert_different_registers(value, thread, tmp);
 3546   Label done, tagged, weak_tagged;
 3547   testptr(value, value);
 3548   jcc(Assembler::zero, done);           // Use null as-is.
 3549   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3550   jcc(Assembler::notZero, tagged);
 3551 
 3552   // Resolve local handle
 3553   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3554   verify_oop(value);
 3555   jmp(done);
 3556 
 3557   bind(tagged);
 3558   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3559   jcc(Assembler::notZero, weak_tagged);
 3560 
 3561   // Resolve global handle
 3562   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3563   verify_oop(value);
 3564   jmp(done);
 3565 
 3566   bind(weak_tagged);
 3567   // Resolve jweak.
 3568   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3569                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3570   verify_oop(value);
 3571 
 3572   bind(done);
 3573 }
 3574 
 3575 void MacroAssembler::resolve_global_jobject(Register value,
 3576                                             Register tmp) {
 3577   Register thread = r15_thread;
 3578   assert_different_registers(value, thread, tmp);
 3579   Label done;
 3580 
 3581   testptr(value, value);
 3582   jcc(Assembler::zero, done);           // Use null as-is.
 3583 
 3584 #ifdef ASSERT
 3585   {
 3586     Label valid_global_tag;
 3587     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3588     jcc(Assembler::notZero, valid_global_tag);
 3589     stop("non global jobject using resolve_global_jobject");
 3590     bind(valid_global_tag);
 3591   }
 3592 #endif
 3593 
 3594   // Resolve global handle
 3595   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3596   verify_oop(value);
 3597 
 3598   bind(done);
 3599 }
 3600 
 3601 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3602   subq(dst, imm32);
 3603 }
 3604 
 3605 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3606 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3607   subq_imm32(dst, imm32);
 3608 }
 3609 
 3610 void MacroAssembler::subptr(Register dst, Register src) {
 3611   subq(dst, src);
 3612 }
 3613 
 3614 // C++ bool manipulation
 3615 void MacroAssembler::testbool(Register dst) {
 3616   if(sizeof(bool) == 1)
 3617     testb(dst, 0xff);
 3618   else if(sizeof(bool) == 2) {
 3619     // testw implementation needed for two byte bools
 3620     ShouldNotReachHere();
 3621   } else if(sizeof(bool) == 4)
 3622     testl(dst, dst);
 3623   else
 3624     // unsupported
 3625     ShouldNotReachHere();
 3626 }
 3627 
 3628 void MacroAssembler::testptr(Register dst, Register src) {
 3629   testq(dst, src);
 3630 }
 3631 
 3632 // Object / value buffer allocation...
 3633 //
 3634 // Kills klass and rsi on LP64
 3635 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
 3636                                        Register t1, Register t2,
 3637                                        bool clear_fields, Label& alloc_failed)
 3638 {
 3639   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
 3640   Register layout_size = t1;
 3641   assert(new_obj == rax, "needs to be rax");
 3642   assert_different_registers(klass, new_obj, t1, t2);
 3643 
 3644   // get instance_size in InstanceKlass (scaled to a count of bytes)
 3645   movl(layout_size, Address(klass, Klass::layout_helper_offset()));
 3646   // test to see if it is malformed in some way
 3647   testl(layout_size, Klass::_lh_instance_slow_path_bit);
 3648   jcc(Assembler::notZero, slow_case_no_pop);
 3649 
 3650   // Allocate the instance:
 3651   //  If TLAB is enabled:
 3652   //    Try to allocate in the TLAB.
 3653   //    If fails, go to the slow path.
 3654   //  Else If inline contiguous allocations are enabled:
 3655   //    Try to allocate in eden.
 3656   //    If fails due to heap end, go to slow path.
 3657   //
 3658   //  If TLAB is enabled OR inline contiguous is enabled:
 3659   //    Initialize the allocation.
 3660   //    Exit.
 3661   //
 3662   //  Go to slow path.
 3663 
 3664   push(klass);
 3665   if (UseTLAB) {
 3666     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
 3667     if (ZeroTLAB || (!clear_fields)) {
 3668       // the fields have been already cleared
 3669       jmp(initialize_header);
 3670     } else {
 3671       // initialize both the header and fields
 3672       jmp(initialize_object);
 3673     }
 3674   } else {
 3675     jmp(slow_case);
 3676   }
 3677 
 3678   // If UseTLAB is true, the object is created above and there is an initialize need.
 3679   // Otherwise, skip and go to the slow path.
 3680   if (UseTLAB) {
 3681     if (clear_fields) {
 3682       // The object is initialized before the header.  If the object size is
 3683       // zero, go directly to the header initialization.
 3684       bind(initialize_object);
 3685       if (UseCompactObjectHeaders) {
 3686         assert(is_aligned(oopDesc::base_offset_in_bytes(), BytesPerLong), "oop base offset must be 8-byte-aligned");
 3687         decrement(layout_size, oopDesc::base_offset_in_bytes());
 3688       } else {
 3689         decrement(layout_size, sizeof(oopDesc));
 3690       }
 3691       jcc(Assembler::zero, initialize_header);
 3692 
 3693       // Initialize topmost object field, divide size by 8, check if odd and
 3694       // test if zero.
 3695       Register zero = klass;
 3696       xorl(zero, zero);    // use zero reg to clear memory (shorter code)
 3697       shrl(layout_size, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
 3698 
 3699   #ifdef ASSERT
 3700       // make sure instance_size was multiple of 8
 3701       Label L;
 3702       // Ignore partial flag stall after shrl() since it is debug VM
 3703       jcc(Assembler::carryClear, L);
 3704       stop("object size is not multiple of 2 - adjust this code");
 3705       bind(L);
 3706       // must be > 0, no extra check needed here
 3707   #endif
 3708 
 3709       // initialize remaining object fields: instance_size was a multiple of 8
 3710       {
 3711         Label loop;
 3712         bind(loop);
 3713         int header_size_bytes = oopDesc::header_size() * HeapWordSize;
 3714         assert(is_aligned(header_size_bytes, BytesPerLong), "oop header size must be 8-byte-aligned");
 3715         movptr(Address(new_obj, layout_size, Address::times_8, header_size_bytes - 1*oopSize), zero);
 3716         decrement(layout_size);
 3717         jcc(Assembler::notZero, loop);
 3718       }
 3719     } // clear_fields
 3720 
 3721     // initialize object header only.
 3722     bind(initialize_header);
 3723     if (UseCompactObjectHeaders || Arguments::is_valhalla_enabled()) {
 3724       pop(klass);
 3725       Register mark_word = t2;
 3726       movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 3727       movptr(Address(new_obj, oopDesc::mark_offset_in_bytes ()), mark_word);
 3728     } else {
 3729      movptr(Address(new_obj, oopDesc::mark_offset_in_bytes()),
 3730             (intptr_t)markWord::prototype().value()); // header
 3731      pop(klass);   // get saved klass back in the register.
 3732     }
 3733     if (!UseCompactObjectHeaders) {
 3734       xorl(rsi, rsi);                 // use zero reg to clear memory (shorter code)
 3735       store_klass_gap(new_obj, rsi);  // zero klass gap for compressed oops
 3736       movptr(t2, klass);         // preserve klass
 3737       store_klass(new_obj, t2, rscratch1);  // src klass reg is potentially compressed
 3738     }
 3739     jmp(done);
 3740   }
 3741 
 3742   bind(slow_case);
 3743   pop(klass);
 3744   bind(slow_case_no_pop);
 3745   jmp(alloc_failed);
 3746 
 3747   bind(done);
 3748 }
 3749 
 3750 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3751 void MacroAssembler::tlab_allocate(Register obj,
 3752                                    Register var_size_in_bytes,
 3753                                    int con_size_in_bytes,
 3754                                    Register t1,
 3755                                    Register t2,
 3756                                    Label& slow_case) {
 3757   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3758   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3759 }
 3760 
 3761 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3762   RegSet regs;
 3763   regs += RegSet::of(rax, rcx, rdx);
 3764 #ifndef _WINDOWS
 3765   regs += RegSet::of(rsi, rdi);
 3766 #endif
 3767   regs += RegSet::range(r8, r11);
 3768   if (UseAPX) {
 3769     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3770   }
 3771   return regs;
 3772 }
 3773 
 3774 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3775   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3776 #if defined(_WINDOWS)
 3777   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3778   if (num_xmm_registers > 16) {
 3779      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3780   }
 3781   return result;
 3782 #else
 3783   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3784 #endif
 3785 }
 3786 
 3787 // C1 only ever uses the first double/float of the XMM register.
 3788 static int xmm_save_size() { return sizeof(double); }
 3789 
 3790 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3791   masm->movdbl(Address(rsp, offset), reg);
 3792 }
 3793 
 3794 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3795   masm->movdbl(reg, Address(rsp, offset));
 3796 }
 3797 
 3798 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3799                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3800 
 3801   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3802                          StackAlignmentInBytes);
 3803   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3804 
 3805   return gp_area_size + xmm_area_size;
 3806 }
 3807 
 3808 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3809   block_comment("push_call_clobbered_registers start");
 3810   // Regular registers
 3811   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3812 
 3813   int gp_area_size;
 3814   int xmm_area_size;
 3815   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3816                                                gp_area_size, xmm_area_size);
 3817   subptr(rsp, total_save_size);
 3818 
 3819   push_set(gp_registers_to_push, 0);
 3820 
 3821   if (save_fpu) {
 3822     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3823   }
 3824 
 3825   block_comment("push_call_clobbered_registers end");
 3826 }
 3827 
 3828 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3829   block_comment("pop_call_clobbered_registers start");
 3830 
 3831   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 3832 
 3833   int gp_area_size;
 3834   int xmm_area_size;
 3835   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 3836                                                gp_area_size, xmm_area_size);
 3837 
 3838   if (restore_fpu) {
 3839     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3840   }
 3841 
 3842   pop_set(gp_registers_to_pop, 0);
 3843 
 3844   addptr(rsp, total_save_size);
 3845 
 3846   vzeroupper();
 3847 
 3848   block_comment("pop_call_clobbered_registers end");
 3849 }
 3850 
 3851 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3852   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3853   int spill_offset = offset;
 3854 
 3855   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3856     save_xmm_register(this, spill_offset, *it);
 3857     spill_offset += xmm_save_size();
 3858   }
 3859 }
 3860 
 3861 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3862   int restore_size = set.size() * xmm_save_size();
 3863   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3864 
 3865   int restore_offset = offset + restore_size - xmm_save_size();
 3866 
 3867   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3868     restore_xmm_register(this, restore_offset, *it);
 3869     restore_offset -= xmm_save_size();
 3870   }
 3871 }
 3872 
 3873 void MacroAssembler::push_set(RegSet set, int offset) {
 3874   int spill_offset;
 3875   if (offset == -1) {
 3876     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3877     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3878     subptr(rsp, aligned_size);
 3879     spill_offset = 0;
 3880   } else {
 3881     spill_offset = offset;
 3882   }
 3883 
 3884   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3885     movptr(Address(rsp, spill_offset), *it);
 3886     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3887   }
 3888 }
 3889 
 3890 void MacroAssembler::pop_set(RegSet set, int offset) {
 3891 
 3892   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3893   int restore_size = set.size() * gp_reg_size;
 3894   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3895 
 3896   int restore_offset;
 3897   if (offset == -1) {
 3898     restore_offset = restore_size - gp_reg_size;
 3899   } else {
 3900     restore_offset = offset + restore_size - gp_reg_size;
 3901   }
 3902   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3903     movptr(*it, Address(rsp, restore_offset));
 3904     restore_offset -= gp_reg_size;
 3905   }
 3906 
 3907   if (offset == -1) {
 3908     addptr(rsp, aligned_size);
 3909   }
 3910 }
 3911 
 3912 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3913 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3914   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3915   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3916   Label done;
 3917 
 3918   testptr(length_in_bytes, length_in_bytes);
 3919   jcc(Assembler::zero, done);
 3920 
 3921   // initialize topmost word, divide index by 2, check if odd and test if zero
 3922   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3923 #ifdef ASSERT
 3924   {
 3925     Label L;
 3926     testptr(length_in_bytes, BytesPerWord - 1);
 3927     jcc(Assembler::zero, L);
 3928     stop("length must be a multiple of BytesPerWord");
 3929     bind(L);
 3930   }
 3931 #endif
 3932   Register index = length_in_bytes;
 3933   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3934   if (UseIncDec) {
 3935     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3936   } else {
 3937     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3938     shrptr(index, 1);
 3939   }
 3940 
 3941   // initialize remaining object fields: index is a multiple of 2 now
 3942   {
 3943     Label loop;
 3944     bind(loop);
 3945     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3946     decrement(index);
 3947     jcc(Assembler::notZero, loop);
 3948   }
 3949 
 3950   bind(done);
 3951 }
 3952 
 3953 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
 3954   movptr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
 3955 #ifdef ASSERT
 3956   {
 3957     Label done;
 3958     cmpptr(layout_info, 0);
 3959     jcc(Assembler::notEqual, done);
 3960     stop("inline_layout_info_array is null");
 3961     bind(done);
 3962   }
 3963 #endif
 3964 
 3965   InlineLayoutInfo array[2];
 3966   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
 3967   if (is_power_of_2(size)) {
 3968     shll(index, log2i_exact(size)); // Scale index by power of 2
 3969   } else {
 3970     imull(index, index, size); // Scale the index to be the entry index * array_element_size
 3971   }
 3972   lea(layout_info, Address(layout_info, index, Address::times_1, Array<InlineLayoutInfo>::base_offset_in_bytes()));
 3973 }
 3974 
 3975 // Look up the method for a megamorphic invokeinterface call.
 3976 // The target method is determined by <intf_klass, itable_index>.
 3977 // The receiver klass is in recv_klass.
 3978 // On success, the result will be in method_result, and execution falls through.
 3979 // On failure, execution transfers to the given label.
 3980 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3981                                              Register intf_klass,
 3982                                              RegisterOrConstant itable_index,
 3983                                              Register method_result,
 3984                                              Register scan_temp,
 3985                                              Label& L_no_such_interface,
 3986                                              bool return_method) {
 3987   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3988   assert_different_registers(method_result, intf_klass, scan_temp);
 3989   assert(recv_klass != method_result || !return_method,
 3990          "recv_klass can be destroyed when method isn't needed");
 3991 
 3992   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3993          "caller must use same register for non-constant itable index as for method");
 3994 
 3995   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3996   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3997   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3998   int scan_step   = itableOffsetEntry::size() * wordSize;
 3999   int vte_size    = vtableEntry::size_in_bytes();
 4000   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 4001   assert(vte_size == wordSize, "else adjust times_vte_scale");
 4002 
 4003   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 4004 
 4005   // Could store the aligned, prescaled offset in the klass.
 4006   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 4007 
 4008   if (return_method) {
 4009     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 4010     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 4011     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 4012   }
 4013 
 4014   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 4015   //   if (scan->interface() == intf) {
 4016   //     result = (klass + scan->offset() + itable_index);
 4017   //   }
 4018   // }
 4019   Label search, found_method;
 4020 
 4021   for (int peel = 1; peel >= 0; peel--) {
 4022     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 4023     cmpptr(intf_klass, method_result);
 4024 
 4025     if (peel) {
 4026       jccb(Assembler::equal, found_method);
 4027     } else {
 4028       jccb(Assembler::notEqual, search);
 4029       // (invert the test to fall through to found_method...)
 4030     }
 4031 
 4032     if (!peel)  break;
 4033 
 4034     bind(search);
 4035 
 4036     // Check that the previous entry is non-null.  A null entry means that
 4037     // the receiver class doesn't implement the interface, and wasn't the
 4038     // same as when the caller was compiled.
 4039     testptr(method_result, method_result);
 4040     jcc(Assembler::zero, L_no_such_interface);
 4041     addptr(scan_temp, scan_step);
 4042   }
 4043 
 4044   bind(found_method);
 4045 
 4046   if (return_method) {
 4047     // Got a hit.
 4048     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 4049     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 4050   }
 4051 }
 4052 
 4053 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 4054 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 4055 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 4056 // The target method is determined by <holder_klass, itable_index>.
 4057 // The receiver klass is in recv_klass.
 4058 // On success, the result will be in method_result, and execution falls through.
 4059 // On failure, execution transfers to the given label.
 4060 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 4061                                                   Register holder_klass,
 4062                                                   Register resolved_klass,
 4063                                                   Register method_result,
 4064                                                   Register scan_temp,
 4065                                                   Register temp_reg2,
 4066                                                   Register receiver,
 4067                                                   int itable_index,
 4068                                                   Label& L_no_such_interface) {
 4069   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 4070   Register temp_itbl_klass = method_result;
 4071   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 4072 
 4073   int vtable_base = in_bytes(Klass::vtable_start_offset());
 4074   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 4075   int scan_step = itableOffsetEntry::size() * wordSize;
 4076   int vte_size = vtableEntry::size_in_bytes();
 4077   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 4078   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 4079   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 4080   assert(vte_size == wordSize, "adjust times_vte_scale");
 4081 
 4082   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 4083 
 4084   // temp_itbl_klass = recv_klass.itable[0]
 4085   // scan_temp = &recv_klass.itable[0] + step
 4086   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 4087   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 4088   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 4089   xorptr(temp_reg, temp_reg);
 4090 
 4091   // Initial checks:
 4092   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 4093   //   - if (itable[0] == 0), no such interface
 4094   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 4095   cmpptr(holder_klass, resolved_klass);
 4096   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 4097   testptr(temp_itbl_klass, temp_itbl_klass);
 4098   jccb(Assembler::zero, L_no_such_interface);
 4099   cmpptr(holder_klass, temp_itbl_klass);
 4100   jccb(Assembler::equal, L_holder_found);
 4101 
 4102   // Loop: Look for holder_klass record in itable
 4103   //   do {
 4104   //     tmp = itable[index];
 4105   //     index += step;
 4106   //     if (tmp == holder_klass) {
 4107   //       goto L_holder_found; // Found!
 4108   //     }
 4109   //   } while (tmp != 0);
 4110   //   goto L_no_such_interface // Not found.
 4111   Label L_scan_holder;
 4112   bind(L_scan_holder);
 4113     movptr(temp_itbl_klass, Address(scan_temp, 0));
 4114     addptr(scan_temp, scan_step);
 4115     cmpptr(holder_klass, temp_itbl_klass);
 4116     jccb(Assembler::equal, L_holder_found);
 4117     testptr(temp_itbl_klass, temp_itbl_klass);
 4118     jccb(Assembler::notZero, L_scan_holder);
 4119 
 4120   jmpb(L_no_such_interface);
 4121 
 4122   // Loop: Look for resolved_class record in itable
 4123   //   do {
 4124   //     tmp = itable[index];
 4125   //     index += step;
 4126   //     if (tmp == holder_klass) {
 4127   //        // Also check if we have met a holder klass
 4128   //        holder_tmp = itable[index-step-ioffset];
 4129   //     }
 4130   //     if (tmp == resolved_klass) {
 4131   //        goto L_resolved_found;  // Found!
 4132   //     }
 4133   //   } while (tmp != 0);
 4134   //   goto L_no_such_interface // Not found.
 4135   //
 4136   Label L_loop_scan_resolved;
 4137   bind(L_loop_scan_resolved);
 4138     movptr(temp_itbl_klass, Address(scan_temp, 0));
 4139     addptr(scan_temp, scan_step);
 4140     bind(L_loop_scan_resolved_entry);
 4141     cmpptr(holder_klass, temp_itbl_klass);
 4142     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4143     cmpptr(resolved_klass, temp_itbl_klass);
 4144     jccb(Assembler::equal, L_resolved_found);
 4145     testptr(temp_itbl_klass, temp_itbl_klass);
 4146     jccb(Assembler::notZero, L_loop_scan_resolved);
 4147 
 4148   jmpb(L_no_such_interface);
 4149 
 4150   Label L_ready;
 4151 
 4152   // See if we already have a holder klass. If not, go and scan for it.
 4153   bind(L_resolved_found);
 4154   testptr(temp_reg, temp_reg);
 4155   jccb(Assembler::zero, L_scan_holder);
 4156   jmpb(L_ready);
 4157 
 4158   bind(L_holder_found);
 4159   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4160 
 4161   // Finally, temp_reg contains holder_klass vtable offset
 4162   bind(L_ready);
 4163   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 4164   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 4165     load_klass(scan_temp, receiver, noreg);
 4166     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4167   } else {
 4168     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4169   }
 4170 }
 4171 
 4172 
 4173 // virtual method calling
 4174 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 4175                                            RegisterOrConstant vtable_index,
 4176                                            Register method_result) {
 4177   const ByteSize base = Klass::vtable_start_offset();
 4178   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 4179   Address vtable_entry_addr(recv_klass,
 4180                             vtable_index, Address::times_ptr,
 4181                             base + vtableEntry::method_offset());
 4182   movptr(method_result, vtable_entry_addr);
 4183 }
 4184 
 4185 
 4186 void MacroAssembler::check_klass_subtype(Register sub_klass,
 4187                            Register super_klass,
 4188                            Register temp_reg,
 4189                            Label& L_success) {
 4190   Label L_failure;
 4191   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 4192   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 4193   bind(L_failure);
 4194 }
 4195 
 4196 
 4197 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 4198                                                    Register super_klass,
 4199                                                    Register temp_reg,
 4200                                                    Label* L_success,
 4201                                                    Label* L_failure,
 4202                                                    Label* L_slow_path,
 4203                                         RegisterOrConstant super_check_offset) {
 4204   assert_different_registers(sub_klass, super_klass, temp_reg);
 4205   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 4206   if (super_check_offset.is_register()) {
 4207     assert_different_registers(sub_klass, super_klass,
 4208                                super_check_offset.as_register());
 4209   } else if (must_load_sco) {
 4210     assert(temp_reg != noreg, "supply either a temp or a register offset");
 4211   }
 4212 
 4213   Label L_fallthrough;
 4214   int label_nulls = 0;
 4215   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4216   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4217   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 4218   assert(label_nulls <= 1, "at most one null in the batch");
 4219 
 4220   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4221   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 4222   Address super_check_offset_addr(super_klass, sco_offset);
 4223 
 4224   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 4225   // range of a jccb.  If this routine grows larger, reconsider at
 4226   // least some of these.
 4227 #define local_jcc(assembler_cond, label)                                \
 4228   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 4229   else                             jcc( assembler_cond, label) /*omit semi*/
 4230 
 4231   // Hacked jmp, which may only be used just before L_fallthrough.
 4232 #define final_jmp(label)                                                \
 4233   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 4234   else                            jmp(label)                /*omit semi*/
 4235 
 4236   // If the pointers are equal, we are done (e.g., String[] elements).
 4237   // This self-check enables sharing of secondary supertype arrays among
 4238   // non-primary types such as array-of-interface.  Otherwise, each such
 4239   // type would need its own customized SSA.
 4240   // We move this check to the front of the fast path because many
 4241   // type checks are in fact trivially successful in this manner,
 4242   // so we get a nicely predicted branch right at the start of the check.
 4243   cmpptr(sub_klass, super_klass);
 4244   local_jcc(Assembler::equal, *L_success);
 4245 
 4246   // Check the supertype display:
 4247   if (must_load_sco) {
 4248     // Positive movl does right thing on LP64.
 4249     movl(temp_reg, super_check_offset_addr);
 4250     super_check_offset = RegisterOrConstant(temp_reg);
 4251   }
 4252   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 4253   cmpptr(super_klass, super_check_addr); // load displayed supertype
 4254 
 4255   // This check has worked decisively for primary supers.
 4256   // Secondary supers are sought in the super_cache ('super_cache_addr').
 4257   // (Secondary supers are interfaces and very deeply nested subtypes.)
 4258   // This works in the same check above because of a tricky aliasing
 4259   // between the super_cache and the primary super display elements.
 4260   // (The 'super_check_addr' can address either, as the case requires.)
 4261   // Note that the cache is updated below if it does not help us find
 4262   // what we need immediately.
 4263   // So if it was a primary super, we can just fail immediately.
 4264   // Otherwise, it's the slow path for us (no success at this point).
 4265 
 4266   if (super_check_offset.is_register()) {
 4267     local_jcc(Assembler::equal, *L_success);
 4268     cmpl(super_check_offset.as_register(), sc_offset);
 4269     if (L_failure == &L_fallthrough) {
 4270       local_jcc(Assembler::equal, *L_slow_path);
 4271     } else {
 4272       local_jcc(Assembler::notEqual, *L_failure);
 4273       final_jmp(*L_slow_path);
 4274     }
 4275   } else if (super_check_offset.as_constant() == sc_offset) {
 4276     // Need a slow path; fast failure is impossible.
 4277     if (L_slow_path == &L_fallthrough) {
 4278       local_jcc(Assembler::equal, *L_success);
 4279     } else {
 4280       local_jcc(Assembler::notEqual, *L_slow_path);
 4281       final_jmp(*L_success);
 4282     }
 4283   } else {
 4284     // No slow path; it's a fast decision.
 4285     if (L_failure == &L_fallthrough) {
 4286       local_jcc(Assembler::equal, *L_success);
 4287     } else {
 4288       local_jcc(Assembler::notEqual, *L_failure);
 4289       final_jmp(*L_success);
 4290     }
 4291   }
 4292 
 4293   bind(L_fallthrough);
 4294 
 4295 #undef local_jcc
 4296 #undef final_jmp
 4297 }
 4298 
 4299 
 4300 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 4301                                                           Register super_klass,
 4302                                                           Register temp_reg,
 4303                                                           Register temp2_reg,
 4304                                                           Label* L_success,
 4305                                                           Label* L_failure,
 4306                                                           bool set_cond_codes) {
 4307   assert_different_registers(sub_klass, super_klass, temp_reg);
 4308   if (temp2_reg != noreg)
 4309     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4310 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4311 
 4312   Label L_fallthrough;
 4313   int label_nulls = 0;
 4314   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4315   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4316   assert(label_nulls <= 1, "at most one null in the batch");
 4317 
 4318   // a couple of useful fields in sub_klass:
 4319   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4320   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4321   Address secondary_supers_addr(sub_klass, ss_offset);
 4322   Address super_cache_addr(     sub_klass, sc_offset);
 4323 
 4324   // Do a linear scan of the secondary super-klass chain.
 4325   // This code is rarely used, so simplicity is a virtue here.
 4326   // The repne_scan instruction uses fixed registers, which we must spill.
 4327   // Don't worry too much about pre-existing connections with the input regs.
 4328 
 4329   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4330   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4331 
 4332   // Get super_klass value into rax (even if it was in rdi or rcx).
 4333   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4334   if (super_klass != rax) {
 4335     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4336     mov(rax, super_klass);
 4337   }
 4338   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4339   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4340 
 4341 #ifndef PRODUCT
 4342   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4343   ExternalAddress pst_counter_addr((address) pst_counter);
 4344   lea(rcx, pst_counter_addr);
 4345   incrementl(Address(rcx, 0));
 4346 #endif //PRODUCT
 4347 
 4348   // We will consult the secondary-super array.
 4349   movptr(rdi, secondary_supers_addr);
 4350   // Load the array length.  (Positive movl does right thing on LP64.)
 4351   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4352   // Skip to start of data.
 4353   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4354 
 4355   // Scan RCX words at [RDI] for an occurrence of RAX.
 4356   // Set NZ/Z based on last compare.
 4357   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4358   // not change flags (only scas instruction which is repeated sets flags).
 4359   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4360 
 4361     testptr(rax,rax); // Set Z = 0
 4362     repne_scan();
 4363 
 4364   // Unspill the temp. registers:
 4365   if (pushed_rdi)  pop(rdi);
 4366   if (pushed_rcx)  pop(rcx);
 4367   if (pushed_rax)  pop(rax);
 4368 
 4369   if (set_cond_codes) {
 4370     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4371     assert(!pushed_rdi, "rdi must be left non-null");
 4372     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4373   }
 4374 
 4375   if (L_failure == &L_fallthrough)
 4376         jccb(Assembler::notEqual, *L_failure);
 4377   else  jcc(Assembler::notEqual, *L_failure);
 4378 
 4379   // Success.  Cache the super we found and proceed in triumph.
 4380   movptr(super_cache_addr, super_klass);
 4381 
 4382   if (L_success != &L_fallthrough) {
 4383     jmp(*L_success);
 4384   }
 4385 
 4386 #undef IS_A_TEMP
 4387 
 4388   bind(L_fallthrough);
 4389 }
 4390 
 4391 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4392                                                    Register super_klass,
 4393                                                    Register temp_reg,
 4394                                                    Register temp2_reg,
 4395                                                    Label* L_success,
 4396                                                    Label* L_failure,
 4397                                                    bool set_cond_codes) {
 4398   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4399   check_klass_subtype_slow_path
 4400     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4401      L_success, L_failure);
 4402 }
 4403 
 4404 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4405                                                    Register super_klass,
 4406                                                    Register temp_reg,
 4407                                                    Register temp2_reg,
 4408                                                    Register temp3_reg,
 4409                                                    Register temp4_reg,
 4410                                                    Label* L_success,
 4411                                                    Label* L_failure) {
 4412   if (UseSecondarySupersTable) {
 4413     check_klass_subtype_slow_path_table
 4414       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4415        L_success, L_failure);
 4416   } else {
 4417     check_klass_subtype_slow_path_linear
 4418       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4419   }
 4420 }
 4421 
 4422 Register MacroAssembler::allocate_if_noreg(Register r,
 4423                                   RegSetIterator<Register> &available_regs,
 4424                                   RegSet &regs_to_push) {
 4425   if (!r->is_valid()) {
 4426     r = *available_regs++;
 4427     regs_to_push += r;
 4428   }
 4429   return r;
 4430 }
 4431 
 4432 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4433                                                          Register super_klass,
 4434                                                          Register temp_reg,
 4435                                                          Register temp2_reg,
 4436                                                          Register temp3_reg,
 4437                                                          Register result_reg,
 4438                                                          Label* L_success,
 4439                                                          Label* L_failure) {
 4440   // NB! Callers may assume that, when temp2_reg is a valid register,
 4441   // this code sets it to a nonzero value.
 4442   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4443 
 4444   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4445 
 4446   Label L_fallthrough;
 4447   int label_nulls = 0;
 4448   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4449   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4450   assert(label_nulls <= 1, "at most one null in the batch");
 4451 
 4452   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4453 
 4454   RegSetIterator<Register> available_regs
 4455     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4456 
 4457   RegSet pushed_regs;
 4458 
 4459   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4460   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4461   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4462   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4463   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4464 
 4465   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4466 
 4467   {
 4468 
 4469     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4470     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4471     subptr(rsp, aligned_size);
 4472     push_set(pushed_regs, 0);
 4473 
 4474     lookup_secondary_supers_table_var(sub_klass,
 4475                                       super_klass,
 4476                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4477     cmpq(result_reg, 0);
 4478 
 4479     // Unspill the temp. registers:
 4480     pop_set(pushed_regs, 0);
 4481     // Increment SP but do not clobber flags.
 4482     lea(rsp, Address(rsp, aligned_size));
 4483   }
 4484 
 4485   if (temp2_reg_was_valid) {
 4486     movq(temp2_reg, 1);
 4487   }
 4488 
 4489   jcc(Assembler::notEqual, *L_failure);
 4490 
 4491   if (L_success != &L_fallthrough) {
 4492     jmp(*L_success);
 4493   }
 4494 
 4495   bind(L_fallthrough);
 4496 }
 4497 
 4498 // population_count variant for running without the POPCNT
 4499 // instruction, which was introduced with SSE4.2 in 2008.
 4500 void MacroAssembler::population_count(Register dst, Register src,
 4501                                       Register scratch1, Register scratch2) {
 4502   assert_different_registers(src, scratch1, scratch2);
 4503   if (UsePopCountInstruction) {
 4504     Assembler::popcntq(dst, src);
 4505   } else {
 4506     assert_different_registers(src, scratch1, scratch2);
 4507     assert_different_registers(dst, scratch1, scratch2);
 4508     Label loop, done;
 4509 
 4510     mov(scratch1, src);
 4511     // dst = 0;
 4512     // while(scratch1 != 0) {
 4513     //   dst++;
 4514     //   scratch1 &= (scratch1 - 1);
 4515     // }
 4516     xorl(dst, dst);
 4517     testq(scratch1, scratch1);
 4518     jccb(Assembler::equal, done);
 4519     {
 4520       bind(loop);
 4521       incq(dst);
 4522       movq(scratch2, scratch1);
 4523       decq(scratch2);
 4524       andq(scratch1, scratch2);
 4525       jccb(Assembler::notEqual, loop);
 4526     }
 4527     bind(done);
 4528   }
 4529 #ifdef ASSERT
 4530   mov64(scratch1, 0xCafeBabeDeadBeef);
 4531   movq(scratch2, scratch1);
 4532 #endif
 4533 }
 4534 
 4535 // Ensure that the inline code and the stub are using the same registers.
 4536 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4537 do {                                                                 \
 4538   assert(r_super_klass  == rax, "mismatch");                         \
 4539   assert(r_array_base   == rbx, "mismatch");                         \
 4540   assert(r_array_length == rcx, "mismatch");                         \
 4541   assert(r_array_index  == rdx, "mismatch");                         \
 4542   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4543   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4544   assert(result         == rdi || result      == noreg, "mismatch"); \
 4545 } while(0)
 4546 
 4547 // Versions of salq and rorq that don't need count to be in rcx
 4548 
 4549 void MacroAssembler::salq(Register dest, Register count) {
 4550   if (count == rcx) {
 4551     Assembler::salq(dest);
 4552   } else {
 4553     assert_different_registers(rcx, dest);
 4554     xchgq(rcx, count);
 4555     Assembler::salq(dest);
 4556     xchgq(rcx, count);
 4557   }
 4558 }
 4559 
 4560 void MacroAssembler::rorq(Register dest, Register count) {
 4561   if (count == rcx) {
 4562     Assembler::rorq(dest);
 4563   } else {
 4564     assert_different_registers(rcx, dest);
 4565     xchgq(rcx, count);
 4566     Assembler::rorq(dest);
 4567     xchgq(rcx, count);
 4568   }
 4569 }
 4570 
 4571 // Return true: we succeeded in generating this code
 4572 //
 4573 // At runtime, return 0 in result if r_super_klass is a superclass of
 4574 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4575 // super_klass_slot of the class you're looking for. This is always
 4576 // the case for instanceof and checkcast.
 4577 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4578                                                          Register r_super_klass,
 4579                                                          Register temp1,
 4580                                                          Register temp2,
 4581                                                          Register temp3,
 4582                                                          Register temp4,
 4583                                                          Register result,
 4584                                                          u1 super_klass_slot) {
 4585   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4586 
 4587   Label L_fallthrough, L_success, L_failure;
 4588 
 4589   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4590 
 4591   const Register
 4592     r_array_index  = temp1,
 4593     r_array_length = temp2,
 4594     r_array_base   = temp3,
 4595     r_bitmap       = temp4;
 4596 
 4597   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4598 
 4599   xorq(result, result); // = 0
 4600 
 4601   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4602   movq(r_array_index, r_bitmap);
 4603 
 4604   // First check the bitmap to see if super_klass might be present. If
 4605   // the bit is zero, we are certain that super_klass is not one of
 4606   // the secondary supers.
 4607   u1 bit = super_klass_slot;
 4608   {
 4609     // NB: If the count in a x86 shift instruction is 0, the flags are
 4610     // not affected, so we do a testq instead.
 4611     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4612     if (shift_count != 0) {
 4613       salq(r_array_index, shift_count);
 4614     } else {
 4615       testq(r_array_index, r_array_index);
 4616     }
 4617   }
 4618   // We test the MSB of r_array_index, i.e. its sign bit
 4619   jcc(Assembler::positive, L_failure);
 4620 
 4621   // Get the first array index that can contain super_klass into r_array_index.
 4622   if (bit != 0) {
 4623     population_count(r_array_index, r_array_index, temp2, temp3);
 4624   } else {
 4625     movl(r_array_index, 1);
 4626   }
 4627   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4628 
 4629   // We will consult the secondary-super array.
 4630   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4631 
 4632   // We're asserting that the first word in an Array<Klass*> is the
 4633   // length, and the second word is the first word of the data. If
 4634   // that ever changes, r_array_base will have to be adjusted here.
 4635   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4636   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4637 
 4638   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4639   jccb(Assembler::equal, L_success);
 4640 
 4641   // Is there another entry to check? Consult the bitmap.
 4642   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4643   jccb(Assembler::carryClear, L_failure);
 4644 
 4645   // Linear probe. Rotate the bitmap so that the next bit to test is
 4646   // in Bit 1.
 4647   if (bit != 0) {
 4648     rorq(r_bitmap, bit);
 4649   }
 4650 
 4651   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4652   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4653   // Kills: r_array_length.
 4654   // Returns: result.
 4655   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4656   // Result (0/1) is in rdi
 4657   jmpb(L_fallthrough);
 4658 
 4659   bind(L_failure);
 4660   incq(result); // 0 => 1
 4661 
 4662   bind(L_success);
 4663   // result = 0;
 4664 
 4665   bind(L_fallthrough);
 4666   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4667 
 4668   if (VerifySecondarySupers) {
 4669     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4670                                   temp1, temp2, temp3);
 4671   }
 4672 }
 4673 
 4674 // At runtime, return 0 in result if r_super_klass is a superclass of
 4675 // r_sub_klass, otherwise return nonzero. Use this version of
 4676 // lookup_secondary_supers_table() if you don't know ahead of time
 4677 // which superclass will be searched for. Used by interpreter and
 4678 // runtime stubs. It is larger and has somewhat greater latency than
 4679 // the version above, which takes a constant super_klass_slot.
 4680 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4681                                                        Register r_super_klass,
 4682                                                        Register temp1,
 4683                                                        Register temp2,
 4684                                                        Register temp3,
 4685                                                        Register temp4,
 4686                                                        Register result) {
 4687   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4688   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4689   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4690 
 4691   Label L_fallthrough, L_success, L_failure;
 4692 
 4693   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4694 
 4695   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4696 
 4697   // FIXME. Once we are sure that all paths reaching this point really
 4698   // do pass rcx as one of our temps we can get rid of the following
 4699   // workaround.
 4700   assert(temps.contains(rcx), "fix this code");
 4701 
 4702   // We prefer to have our shift count in rcx. If rcx is one of our
 4703   // temps, use it for slot. If not, pick any of our temps.
 4704   Register slot;
 4705   if (!temps.contains(rcx)) {
 4706     slot = *available_regs++;
 4707   } else {
 4708     slot = rcx;
 4709   }
 4710 
 4711   const Register r_array_index = *available_regs++;
 4712   const Register r_bitmap      = *available_regs++;
 4713 
 4714   // The logic above guarantees this property, but we state it here.
 4715   assert_different_registers(r_array_index, r_bitmap, rcx);
 4716 
 4717   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4718   movq(r_array_index, r_bitmap);
 4719 
 4720   // First check the bitmap to see if super_klass might be present. If
 4721   // the bit is zero, we are certain that super_klass is not one of
 4722   // the secondary supers.
 4723   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4724   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4725   salq(r_array_index, slot);
 4726 
 4727   testq(r_array_index, r_array_index);
 4728   // We test the MSB of r_array_index, i.e. its sign bit
 4729   jcc(Assembler::positive, L_failure);
 4730 
 4731   const Register r_array_base = *available_regs++;
 4732 
 4733   // Get the first array index that can contain super_klass into r_array_index.
 4734   // Note: Clobbers r_array_base and slot.
 4735   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4736 
 4737   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4738 
 4739   // We will consult the secondary-super array.
 4740   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4741 
 4742   // We're asserting that the first word in an Array<Klass*> is the
 4743   // length, and the second word is the first word of the data. If
 4744   // that ever changes, r_array_base will have to be adjusted here.
 4745   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4746   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4747 
 4748   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4749   jccb(Assembler::equal, L_success);
 4750 
 4751   // Restore slot to its true value
 4752   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4753 
 4754   // Linear probe. Rotate the bitmap so that the next bit to test is
 4755   // in Bit 1.
 4756   rorq(r_bitmap, slot);
 4757 
 4758   // Is there another entry to check? Consult the bitmap.
 4759   btq(r_bitmap, 1);
 4760   jccb(Assembler::carryClear, L_failure);
 4761 
 4762   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4763   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4764   // Kills: r_array_length.
 4765   // Returns: result.
 4766   lookup_secondary_supers_table_slow_path(r_super_klass,
 4767                                           r_array_base,
 4768                                           r_array_index,
 4769                                           r_bitmap,
 4770                                           /*temp1*/result,
 4771                                           /*temp2*/slot,
 4772                                           &L_success,
 4773                                           nullptr);
 4774 
 4775   bind(L_failure);
 4776   movq(result, 1);
 4777   jmpb(L_fallthrough);
 4778 
 4779   bind(L_success);
 4780   xorq(result, result); // = 0
 4781 
 4782   bind(L_fallthrough);
 4783   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4784 
 4785   if (VerifySecondarySupers) {
 4786     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4787                                   temp1, temp2, temp3);
 4788   }
 4789 }
 4790 
 4791 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4792                                  Label* L_success, Label* L_failure) {
 4793   Label L_loop, L_fallthrough;
 4794   {
 4795     int label_nulls = 0;
 4796     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4797     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4798     assert(label_nulls <= 1, "at most one null in the batch");
 4799   }
 4800   bind(L_loop);
 4801   cmpq(value, Address(addr, count, Address::times_8));
 4802   jcc(Assembler::equal, *L_success);
 4803   addl(count, 1);
 4804   cmpl(count, limit);
 4805   jcc(Assembler::less, L_loop);
 4806 
 4807   if (&L_fallthrough != L_failure) {
 4808     jmp(*L_failure);
 4809   }
 4810   bind(L_fallthrough);
 4811 }
 4812 
 4813 // Called by code generated by check_klass_subtype_slow_path
 4814 // above. This is called when there is a collision in the hashed
 4815 // lookup in the secondary supers array.
 4816 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4817                                                              Register r_array_base,
 4818                                                              Register r_array_index,
 4819                                                              Register r_bitmap,
 4820                                                              Register temp1,
 4821                                                              Register temp2,
 4822                                                              Label* L_success,
 4823                                                              Label* L_failure) {
 4824   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4825 
 4826   const Register
 4827     r_array_length = temp1,
 4828     r_sub_klass    = noreg,
 4829     result         = noreg;
 4830 
 4831   Label L_fallthrough;
 4832   int label_nulls = 0;
 4833   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4834   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4835   assert(label_nulls <= 1, "at most one null in the batch");
 4836 
 4837   // Load the array length.
 4838   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4839   // And adjust the array base to point to the data.
 4840   // NB! Effectively increments current slot index by 1.
 4841   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4842   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4843 
 4844   // Linear probe
 4845   Label L_huge;
 4846 
 4847   // The bitmap is full to bursting.
 4848   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4849   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4850   jcc(Assembler::greater, L_huge);
 4851 
 4852   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4853   // current slot (at secondary_supers[r_array_index]) has not yet
 4854   // been inspected, and r_array_index may be out of bounds if we
 4855   // wrapped around the end of the array.
 4856 
 4857   { // This is conventional linear probing, but instead of terminating
 4858     // when a null entry is found in the table, we maintain a bitmap
 4859     // in which a 0 indicates missing entries.
 4860     // The check above guarantees there are 0s in the bitmap, so the loop
 4861     // eventually terminates.
 4862 
 4863     xorl(temp2, temp2); // = 0;
 4864 
 4865     Label L_again;
 4866     bind(L_again);
 4867 
 4868     // Check for array wraparound.
 4869     cmpl(r_array_index, r_array_length);
 4870     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4871 
 4872     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4873     jcc(Assembler::equal, *L_success);
 4874 
 4875     // If the next bit in bitmap is zero, we're done.
 4876     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4877     jcc(Assembler::carryClear, *L_failure);
 4878 
 4879     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4880     addl(r_array_index, 1);
 4881 
 4882     jmp(L_again);
 4883   }
 4884 
 4885   { // Degenerate case: more than 64 secondary supers.
 4886     // FIXME: We could do something smarter here, maybe a vectorized
 4887     // comparison or a binary search, but is that worth any added
 4888     // complexity?
 4889     bind(L_huge);
 4890     xorl(r_array_index, r_array_index); // = 0
 4891     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4892                 L_success,
 4893                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4894 
 4895     bind(L_fallthrough);
 4896   }
 4897 }
 4898 
 4899 struct VerifyHelperArguments {
 4900   Klass* _super;
 4901   Klass* _sub;
 4902   intptr_t _linear_result;
 4903   intptr_t _table_result;
 4904 };
 4905 
 4906 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4907   Klass::on_secondary_supers_verification_failure(args->_super,
 4908                                                   args->_sub,
 4909                                                   args->_linear_result,
 4910                                                   args->_table_result,
 4911                                                   msg);
 4912 }
 4913 
 4914 // Make sure that the hashed lookup and a linear scan agree.
 4915 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4916                                                    Register r_super_klass,
 4917                                                    Register result,
 4918                                                    Register temp1,
 4919                                                    Register temp2,
 4920                                                    Register temp3) {
 4921   const Register
 4922       r_array_index  = temp1,
 4923       r_array_length = temp2,
 4924       r_array_base   = temp3,
 4925       r_bitmap       = noreg;
 4926 
 4927   BLOCK_COMMENT("verify_secondary_supers_table {");
 4928 
 4929   Label L_success, L_failure, L_check, L_done;
 4930 
 4931   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4932   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4933   // And adjust the array base to point to the data.
 4934   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4935 
 4936   testl(r_array_length, r_array_length); // array_length == 0?
 4937   jcc(Assembler::zero, L_failure);
 4938 
 4939   movl(r_array_index, 0);
 4940   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4941   // fall through to L_failure
 4942 
 4943   const Register linear_result = r_array_index; // reuse temp1
 4944 
 4945   bind(L_failure); // not present
 4946   movl(linear_result, 1);
 4947   jmp(L_check);
 4948 
 4949   bind(L_success); // present
 4950   movl(linear_result, 0);
 4951 
 4952   bind(L_check);
 4953   cmpl(linear_result, result);
 4954   jcc(Assembler::equal, L_done);
 4955 
 4956   { // To avoid calling convention issues, build a record on the stack
 4957     // and pass the pointer to that instead.
 4958     push(result);
 4959     push(linear_result);
 4960     push(r_sub_klass);
 4961     push(r_super_klass);
 4962     movptr(c_rarg1, rsp);
 4963     movptr(c_rarg0, (uintptr_t) "mismatch");
 4964     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4965     should_not_reach_here();
 4966   }
 4967   bind(L_done);
 4968 
 4969   BLOCK_COMMENT("} verify_secondary_supers_table");
 4970 }
 4971 
 4972 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4973 
 4974 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4975   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4976 
 4977   Label L_fallthrough;
 4978   if (L_fast_path == nullptr) {
 4979     L_fast_path = &L_fallthrough;
 4980   } else if (L_slow_path == nullptr) {
 4981     L_slow_path = &L_fallthrough;
 4982   }
 4983 
 4984   // Fast path check: class is fully initialized.
 4985   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4986   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4987   jcc(Assembler::equal, *L_fast_path);
 4988 
 4989   // Fast path check: current thread is initializer thread
 4990   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4991   if (L_slow_path == &L_fallthrough) {
 4992     jcc(Assembler::equal, *L_fast_path);
 4993     bind(*L_slow_path);
 4994   } else if (L_fast_path == &L_fallthrough) {
 4995     jcc(Assembler::notEqual, *L_slow_path);
 4996     bind(*L_fast_path);
 4997   } else {
 4998     Unimplemented();
 4999   }
 5000 }
 5001 
 5002 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 5003   if (VM_Version::supports_cmov()) {
 5004     cmovl(cc, dst, src);
 5005   } else {
 5006     Label L;
 5007     jccb(negate_condition(cc), L);
 5008     movl(dst, src);
 5009     bind(L);
 5010   }
 5011 }
 5012 
 5013 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 5014   if (VM_Version::supports_cmov()) {
 5015     cmovl(cc, dst, src);
 5016   } else {
 5017     Label L;
 5018     jccb(negate_condition(cc), L);
 5019     movl(dst, src);
 5020     bind(L);
 5021   }
 5022 }
 5023 
 5024 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 5025   if (!VerifyOops || VerifyAdapterSharing) {
 5026     // Below address of the code string confuses VerifyAdapterSharing
 5027     // because it may differ between otherwise equivalent adapters.
 5028     return;
 5029   }
 5030 
 5031   BLOCK_COMMENT("verify_oop {");
 5032   push(rscratch1);
 5033   push(rax);                          // save rax
 5034   push(reg);                          // pass register argument
 5035 
 5036   // Pass register number to verify_oop_subroutine
 5037   const char* b = nullptr;
 5038   {
 5039     ResourceMark rm;
 5040     stringStream ss;
 5041     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 5042     b = code_string(ss.as_string());
 5043   }
 5044   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5045   pushptr(buffer.addr(), rscratch1);
 5046 
 5047   // call indirectly to solve generation ordering problem
 5048   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5049   call(rax);
 5050   // Caller pops the arguments (oop, message) and restores rax, r10
 5051   BLOCK_COMMENT("} verify_oop");
 5052 }
 5053 
 5054 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 5055   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 5056     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 5057     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 5058     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 5059   } else if (VM_Version::supports_avx()) {
 5060     vpcmpeqd(dst, dst, dst, vector_len);
 5061   } else {
 5062     pcmpeqd(dst, dst);
 5063   }
 5064 }
 5065 
 5066 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 5067                                          int extra_slot_offset) {
 5068   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 5069   int stackElementSize = Interpreter::stackElementSize;
 5070   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 5071 #ifdef ASSERT
 5072   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 5073   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 5074 #endif
 5075   Register             scale_reg    = noreg;
 5076   Address::ScaleFactor scale_factor = Address::no_scale;
 5077   if (arg_slot.is_constant()) {
 5078     offset += arg_slot.as_constant() * stackElementSize;
 5079   } else {
 5080     scale_reg    = arg_slot.as_register();
 5081     scale_factor = Address::times(stackElementSize);
 5082   }
 5083   offset += wordSize;           // return PC is on stack
 5084   return Address(rsp, scale_reg, scale_factor, offset);
 5085 }
 5086 
 5087 // Handle the receiver type profile update given the "recv" klass.
 5088 //
 5089 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 5090 // If there are no matching or claimable receiver entries in RD, updates
 5091 // the polymorphic counter.
 5092 //
 5093 // This code expected to run by either the interpreter or JIT-ed code, without
 5094 // extra synchronization. For safety, receiver cells are claimed atomically, which
 5095 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 5096 // counter updates are not atomic.
 5097 //
 5098 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 5099   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 5100   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 5101   int poly_count_offset      = in_bytes(CounterData::count_offset());
 5102   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 5103   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 5104 
 5105   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 5106   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 5107   base_receiver_offset += mdp_offset;
 5108   end_receiver_offset  += mdp_offset;
 5109   poly_count_offset    += mdp_offset;
 5110 
 5111   // Scale down to optimize encoding. Slots are pointer-sized.
 5112   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 5113   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 5114   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 5115   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 5116   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 5117   base_receiver_offset   >>= LogBytesPerWord;
 5118   end_receiver_offset    >>= LogBytesPerWord;
 5119   poly_count_offset      >>= LogBytesPerWord;
 5120   receiver_step          >>= LogBytesPerWord;
 5121   receiver_to_count_step >>= LogBytesPerWord;
 5122 
 5123 #ifdef ASSERT
 5124   // We are about to walk the MDO slots without asking for offsets.
 5125   // Check that our math hits all the right spots.
 5126   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 5127     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 5128     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 5129     int offset = base_receiver_offset + receiver_step*c;
 5130     int count_offset = offset + receiver_to_count_step;
 5131     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 5132     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 5133   }
 5134   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 5135   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 5136 #endif
 5137 
 5138   // Corner case: no profile table. Increment poly counter and exit.
 5139   if (ReceiverTypeData::row_limit() == 0) {
 5140     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 5141     return;
 5142   }
 5143 
 5144   Register offset = rscratch1;
 5145 
 5146   Label L_loop_search_receiver, L_loop_search_empty;
 5147   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 5148 
 5149   // The code here recognizes three major cases:
 5150   //   A. Fastest: receiver found in the table
 5151   //   B. Fast: no receiver in the table, and the table is full
 5152   //   C. Slow: no receiver in the table, free slots in the table
 5153   //
 5154   // The case A performance is most important, as perfectly-behaved code would end up
 5155   // there, especially with larger TypeProfileWidth. The case B performance is
 5156   // important as well, this is where bulk of code would land for normally megamorphic
 5157   // cases. The case C performance is not essential, its job is to deal with installation
 5158   // races, we optimize for code density instead. Case C needs to make sure that receiver
 5159   // rows are only claimed once. This makes sure we never overwrite a row for another
 5160   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 5161   //
 5162   // It is very tempting to handle these cases in a single loop, and claim the first slot
 5163   // without checking the rest of the table. But, profiling code should tolerate free slots
 5164   // in the table, as class unloading can clear them. After such cleanup, the receiver
 5165   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 5166   // to complete, before trying to install new slots. Splitting the code in several tight
 5167   // loops also helpfully optimizes for cases A and B.
 5168   //
 5169   // This code is effectively:
 5170   //
 5171   // restart:
 5172   //   // Fastest: receiver is already installed
 5173   //   for (i = 0; i < receiver_count(); i++) {
 5174   //     if (receiver(i) == recv) goto found_recv(i);
 5175   //   }
 5176   //
 5177   //   // Fast: no receiver, but profile is full
 5178   //   for (i = 0; i < receiver_count(); i++) {
 5179   //     if (receiver(i) == null) goto found_null(i);
 5180   //   }
 5181   //   goto polymorphic
 5182   //
 5183   //   // Slow: try to install receiver
 5184   // found_null(i):
 5185   //   CAS(&receiver(i), null, recv);
 5186   //   goto restart
 5187   //
 5188   // polymorphic:
 5189   //   count++;
 5190   //   return
 5191   //
 5192   // found_recv(i):
 5193   //   *receiver_count(i)++
 5194   //
 5195 
 5196   bind(L_restart);
 5197 
 5198   // Fastest: receiver is already installed
 5199   movptr(offset, base_receiver_offset);
 5200   bind(L_loop_search_receiver);
 5201     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 5202     jccb(Assembler::equal, L_found_recv);
 5203   addptr(offset, receiver_step);
 5204   cmpptr(offset, end_receiver_offset);
 5205   jccb(Assembler::notEqual, L_loop_search_receiver);
 5206 
 5207   // Fast: no receiver, but profile is full
 5208   movptr(offset, base_receiver_offset);
 5209   bind(L_loop_search_empty);
 5210     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 5211     jccb(Assembler::equal, L_found_empty);
 5212   addptr(offset, receiver_step);
 5213   cmpptr(offset, end_receiver_offset);
 5214   jccb(Assembler::notEqual, L_loop_search_empty);
 5215   jmpb(L_polymorphic);
 5216 
 5217   // Slow: try to install receiver
 5218   bind(L_found_empty);
 5219 
 5220   // Atomically swing receiver slot: null -> recv.
 5221   //
 5222   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 5223   // other important registers untouched, as they form the address. Therefore, we need
 5224   // to shift any important registers from RAX into some other spare register. If we
 5225   // have a spare register, we are forced to save it on stack here.
 5226 
 5227   Register spare_reg = noreg;
 5228   Register shifted_mdp = mdp;
 5229   Register shifted_recv = recv;
 5230   if (recv == rax || mdp == rax) {
 5231     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 5232                 (recv != rcx && mdp != rcx) ? rcx :
 5233                 rdx;
 5234     assert_different_registers(mdp, recv, offset, spare_reg);
 5235 
 5236     push(spare_reg);
 5237     if (recv == rax) {
 5238       movptr(spare_reg, recv);
 5239       shifted_recv = spare_reg;
 5240     } else {
 5241       assert(mdp == rax, "Remaining case");
 5242       movptr(spare_reg, mdp);
 5243       shifted_mdp = spare_reg;
 5244     }
 5245   } else {
 5246     push(rax);
 5247   }
 5248 
 5249   // None of the important registers are in RAX after this shuffle.
 5250   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 5251 
 5252   xorptr(rax, rax);
 5253   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 5254 
 5255   // Unshift registers.
 5256   if (recv == rax || mdp == rax) {
 5257     movptr(rax, spare_reg);
 5258     pop(spare_reg);
 5259   } else {
 5260     pop(rax);
 5261   }
 5262 
 5263   // CAS success means the slot now has the receiver we want. CAS failure means
 5264   // something had claimed the slot concurrently: it can be the same receiver we want,
 5265   // or something else. Since this is a slow path, we can optimize for code density,
 5266   // and just restart the search from the beginning.
 5267   jmpb(L_restart);
 5268 
 5269   // Counter updates:
 5270 
 5271   // Increment polymorphic counter instead of receiver slot.
 5272   bind(L_polymorphic);
 5273   movptr(offset, poly_count_offset);
 5274   jmpb(L_count_update);
 5275 
 5276   // Found a receiver, convert its slot offset to corresponding count offset.
 5277   bind(L_found_recv);
 5278   addptr(offset, receiver_to_count_step);
 5279 
 5280   bind(L_count_update);
 5281   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 5282 }
 5283 
 5284 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5285   if (!VerifyOops || VerifyAdapterSharing) {
 5286     // Below address of the code string confuses VerifyAdapterSharing
 5287     // because it may differ between otherwise equivalent adapters.
 5288     return;
 5289   }
 5290 
 5291   push(rscratch1);
 5292   push(rax); // save rax,
 5293   // addr may contain rsp so we will have to adjust it based on the push
 5294   // we just did (and on 64 bit we do two pushes)
 5295   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5296   // stores rax into addr which is backwards of what was intended.
 5297   if (addr.uses(rsp)) {
 5298     lea(rax, addr);
 5299     pushptr(Address(rax, 2 * BytesPerWord));
 5300   } else {
 5301     pushptr(addr);
 5302   }
 5303 
 5304   // Pass register number to verify_oop_subroutine
 5305   const char* b = nullptr;
 5306   {
 5307     ResourceMark rm;
 5308     stringStream ss;
 5309     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 5310     b = code_string(ss.as_string());
 5311   }
 5312   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5313   pushptr(buffer.addr(), rscratch1);
 5314 
 5315   // call indirectly to solve generation ordering problem
 5316   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5317   call(rax);
 5318   // Caller pops the arguments (addr, message) and restores rax, r10.
 5319 }
 5320 
 5321 void MacroAssembler::verify_tlab() {
 5322 #ifdef ASSERT
 5323   if (UseTLAB && VerifyOops) {
 5324     Label next, ok;
 5325     Register t1 = rsi;
 5326 
 5327     push(t1);
 5328 
 5329     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5330     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 5331     jcc(Assembler::aboveEqual, next);
 5332     STOP("assert(top >= start)");
 5333     should_not_reach_here();
 5334 
 5335     bind(next);
 5336     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 5337     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5338     jcc(Assembler::aboveEqual, ok);
 5339     STOP("assert(top <= end)");
 5340     should_not_reach_here();
 5341 
 5342     bind(ok);
 5343     pop(t1);
 5344   }
 5345 #endif
 5346 }
 5347 
 5348 class ControlWord {
 5349  public:
 5350   int32_t _value;
 5351 
 5352   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5353   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5354   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5355   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5356   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5357   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5358   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5359   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5360 
 5361   void print() const {
 5362     // rounding control
 5363     const char* rc;
 5364     switch (rounding_control()) {
 5365       case 0: rc = "round near"; break;
 5366       case 1: rc = "round down"; break;
 5367       case 2: rc = "round up  "; break;
 5368       case 3: rc = "chop      "; break;
 5369       default:
 5370         rc = nullptr; // silence compiler warnings
 5371         fatal("Unknown rounding control: %d", rounding_control());
 5372     };
 5373     // precision control
 5374     const char* pc;
 5375     switch (precision_control()) {
 5376       case 0: pc = "24 bits "; break;
 5377       case 1: pc = "reserved"; break;
 5378       case 2: pc = "53 bits "; break;
 5379       case 3: pc = "64 bits "; break;
 5380       default:
 5381         pc = nullptr; // silence compiler warnings
 5382         fatal("Unknown precision control: %d", precision_control());
 5383     };
 5384     // flags
 5385     char f[9];
 5386     f[0] = ' ';
 5387     f[1] = ' ';
 5388     f[2] = (precision   ()) ? 'P' : 'p';
 5389     f[3] = (underflow   ()) ? 'U' : 'u';
 5390     f[4] = (overflow    ()) ? 'O' : 'o';
 5391     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5392     f[6] = (denormalized()) ? 'D' : 'd';
 5393     f[7] = (invalid     ()) ? 'I' : 'i';
 5394     f[8] = '\x0';
 5395     // output
 5396     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5397   }
 5398 
 5399 };
 5400 
 5401 class StatusWord {
 5402  public:
 5403   int32_t _value;
 5404 
 5405   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5406   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5407   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5408   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5409   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5410   int  top() const                     { return  (_value >> 11) & 7      ; }
 5411   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5412   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5413   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5414   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5415   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5416   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5417   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5418   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5419 
 5420   void print() const {
 5421     // condition codes
 5422     char c[5];
 5423     c[0] = (C3()) ? '3' : '-';
 5424     c[1] = (C2()) ? '2' : '-';
 5425     c[2] = (C1()) ? '1' : '-';
 5426     c[3] = (C0()) ? '0' : '-';
 5427     c[4] = '\x0';
 5428     // flags
 5429     char f[9];
 5430     f[0] = (error_status()) ? 'E' : '-';
 5431     f[1] = (stack_fault ()) ? 'S' : '-';
 5432     f[2] = (precision   ()) ? 'P' : '-';
 5433     f[3] = (underflow   ()) ? 'U' : '-';
 5434     f[4] = (overflow    ()) ? 'O' : '-';
 5435     f[5] = (zero_divide ()) ? 'Z' : '-';
 5436     f[6] = (denormalized()) ? 'D' : '-';
 5437     f[7] = (invalid     ()) ? 'I' : '-';
 5438     f[8] = '\x0';
 5439     // output
 5440     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5441   }
 5442 
 5443 };
 5444 
 5445 class TagWord {
 5446  public:
 5447   int32_t _value;
 5448 
 5449   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5450 
 5451   void print() const {
 5452     printf("%04x", _value & 0xFFFF);
 5453   }
 5454 
 5455 };
 5456 
 5457 class FPU_Register {
 5458  public:
 5459   int32_t _m0;
 5460   int32_t _m1;
 5461   int16_t _ex;
 5462 
 5463   bool is_indefinite() const           {
 5464     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5465   }
 5466 
 5467   void print() const {
 5468     char  sign = (_ex < 0) ? '-' : '+';
 5469     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5470     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5471   };
 5472 
 5473 };
 5474 
 5475 class FPU_State {
 5476  public:
 5477   enum {
 5478     register_size       = 10,
 5479     number_of_registers =  8,
 5480     register_mask       =  7
 5481   };
 5482 
 5483   ControlWord  _control_word;
 5484   StatusWord   _status_word;
 5485   TagWord      _tag_word;
 5486   int32_t      _error_offset;
 5487   int32_t      _error_selector;
 5488   int32_t      _data_offset;
 5489   int32_t      _data_selector;
 5490   int8_t       _register[register_size * number_of_registers];
 5491 
 5492   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5493   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5494 
 5495   const char* tag_as_string(int tag) const {
 5496     switch (tag) {
 5497       case 0: return "valid";
 5498       case 1: return "zero";
 5499       case 2: return "special";
 5500       case 3: return "empty";
 5501     }
 5502     ShouldNotReachHere();
 5503     return nullptr;
 5504   }
 5505 
 5506   void print() const {
 5507     // print computation registers
 5508     { int t = _status_word.top();
 5509       for (int i = 0; i < number_of_registers; i++) {
 5510         int j = (i - t) & register_mask;
 5511         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5512         st(j)->print();
 5513         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5514       }
 5515     }
 5516     printf("\n");
 5517     // print control registers
 5518     printf("ctrl = "); _control_word.print(); printf("\n");
 5519     printf("stat = "); _status_word .print(); printf("\n");
 5520     printf("tags = "); _tag_word    .print(); printf("\n");
 5521   }
 5522 
 5523 };
 5524 
 5525 class Flag_Register {
 5526  public:
 5527   int32_t _value;
 5528 
 5529   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5530   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5531   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5532   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5533   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5534   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5535   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5536 
 5537   void print() const {
 5538     // flags
 5539     char f[8];
 5540     f[0] = (overflow       ()) ? 'O' : '-';
 5541     f[1] = (direction      ()) ? 'D' : '-';
 5542     f[2] = (sign           ()) ? 'S' : '-';
 5543     f[3] = (zero           ()) ? 'Z' : '-';
 5544     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5545     f[5] = (parity         ()) ? 'P' : '-';
 5546     f[6] = (carry          ()) ? 'C' : '-';
 5547     f[7] = '\x0';
 5548     // output
 5549     printf("%08x  flags = %s", _value, f);
 5550   }
 5551 
 5552 };
 5553 
 5554 class IU_Register {
 5555  public:
 5556   int32_t _value;
 5557 
 5558   void print() const {
 5559     printf("%08x  %11d", _value, _value);
 5560   }
 5561 
 5562 };
 5563 
 5564 class IU_State {
 5565  public:
 5566   Flag_Register _eflags;
 5567   IU_Register   _rdi;
 5568   IU_Register   _rsi;
 5569   IU_Register   _rbp;
 5570   IU_Register   _rsp;
 5571   IU_Register   _rbx;
 5572   IU_Register   _rdx;
 5573   IU_Register   _rcx;
 5574   IU_Register   _rax;
 5575 
 5576   void print() const {
 5577     // computation registers
 5578     printf("rax,  = "); _rax.print(); printf("\n");
 5579     printf("rbx,  = "); _rbx.print(); printf("\n");
 5580     printf("rcx  = "); _rcx.print(); printf("\n");
 5581     printf("rdx  = "); _rdx.print(); printf("\n");
 5582     printf("rdi  = "); _rdi.print(); printf("\n");
 5583     printf("rsi  = "); _rsi.print(); printf("\n");
 5584     printf("rbp,  = "); _rbp.print(); printf("\n");
 5585     printf("rsp  = "); _rsp.print(); printf("\n");
 5586     printf("\n");
 5587     // control registers
 5588     printf("flgs = "); _eflags.print(); printf("\n");
 5589   }
 5590 };
 5591 
 5592 
 5593 class CPU_State {
 5594  public:
 5595   FPU_State _fpu_state;
 5596   IU_State  _iu_state;
 5597 
 5598   void print() const {
 5599     printf("--------------------------------------------------\n");
 5600     _iu_state .print();
 5601     printf("\n");
 5602     _fpu_state.print();
 5603     printf("--------------------------------------------------\n");
 5604   }
 5605 
 5606 };
 5607 
 5608 
 5609 static void _print_CPU_state(CPU_State* state) {
 5610   state->print();
 5611 };
 5612 
 5613 
 5614 void MacroAssembler::print_CPU_state() {
 5615   push_CPU_state();
 5616   push(rsp);                // pass CPU state
 5617   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5618   addptr(rsp, wordSize);       // discard argument
 5619   pop_CPU_state();
 5620 }
 5621 
 5622 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5623   // Either restore the MXCSR register after returning from the JNI Call
 5624   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5625   if (VM_Version::supports_sse()) {
 5626     if (RestoreMXCSROnJNICalls) {
 5627       ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5628     } else if (CheckJNICalls) {
 5629       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5630     }
 5631   }
 5632   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5633   vzeroupper();
 5634 }
 5635 
 5636 // ((OopHandle)result).resolve();
 5637 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5638   assert_different_registers(result, tmp);
 5639 
 5640   // Only 64 bit platforms support GCs that require a tmp register
 5641   // Only IN_HEAP loads require a thread_tmp register
 5642   // OopHandle::resolve is an indirection like jobject.
 5643   access_load_at(T_OBJECT, IN_NATIVE,
 5644                  result, Address(result, 0), tmp);
 5645 }
 5646 
 5647 // ((WeakHandle)result).resolve();
 5648 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5649   assert_different_registers(rresult, rtmp);
 5650   Label resolved;
 5651 
 5652   // A null weak handle resolves to null.
 5653   cmpptr(rresult, 0);
 5654   jcc(Assembler::equal, resolved);
 5655 
 5656   // Only 64 bit platforms support GCs that require a tmp register
 5657   // Only IN_HEAP loads require a thread_tmp register
 5658   // WeakHandle::resolve is an indirection like jweak.
 5659   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5660                  rresult, Address(rresult, 0), rtmp);
 5661   bind(resolved);
 5662 }
 5663 
 5664 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5665   // get mirror
 5666   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5667   load_method_holder(mirror, method);
 5668   movptr(mirror, Address(mirror, mirror_offset));
 5669   resolve_oop_handle(mirror, tmp);
 5670 }
 5671 
 5672 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5673   load_method_holder(rresult, rmethod);
 5674   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5675 }
 5676 
 5677 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5678   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5679   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5680   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5681 }
 5682 
 5683 void MacroAssembler::load_metadata(Register dst, Register src) {
 5684   if (UseCompactObjectHeaders) {
 5685     load_narrow_klass_compact(dst, src);
 5686   } else {
 5687     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5688   }
 5689 }
 5690 
 5691 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5692   assert(UseCompactObjectHeaders, "expect compact object headers");
 5693   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5694   shrq(dst, markWord::klass_shift);
 5695 }
 5696 
 5697 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5698   assert_different_registers(src, tmp);
 5699   assert_different_registers(dst, tmp);
 5700 
 5701   if (UseCompactObjectHeaders) {
 5702     load_narrow_klass_compact(dst, src);
 5703     decode_klass_not_null(dst, tmp);
 5704   } else {
 5705     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5706     decode_klass_not_null(dst, tmp);
 5707   }
 5708 }
 5709 
 5710 void MacroAssembler::load_prototype_header(Register dst, Register src, Register tmp) {
 5711   load_klass(dst, src, tmp);
 5712   movptr(dst, Address(dst, Klass::prototype_header_offset()));
 5713 }
 5714 
 5715 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5716   assert(!UseCompactObjectHeaders, "not with compact headers");
 5717   assert_different_registers(src, tmp);
 5718   assert_different_registers(dst, tmp);
 5719   encode_klass_not_null(src, tmp);
 5720   movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5721 }
 5722 
 5723 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5724   if (UseCompactObjectHeaders) {
 5725     assert(tmp != noreg, "need tmp");
 5726     assert_different_registers(klass, obj, tmp);
 5727     load_narrow_klass_compact(tmp, obj);
 5728     cmpl(klass, tmp);
 5729   } else {
 5730     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5731   }
 5732 }
 5733 
 5734 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5735   if (UseCompactObjectHeaders) {
 5736     assert(tmp2 != noreg, "need tmp2");
 5737     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5738     load_narrow_klass_compact(tmp1, obj1);
 5739     load_narrow_klass_compact(tmp2, obj2);
 5740     cmpl(tmp1, tmp2);
 5741   } else {
 5742     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5743     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5744   }
 5745 }
 5746 
 5747 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5748                                     Register tmp1) {
 5749   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5750   decorators = AccessInternal::decorator_fixup(decorators, type);
 5751   bool as_raw = (decorators & AS_RAW) != 0;
 5752   if (as_raw) {
 5753     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5754   } else {
 5755     bs->load_at(this, decorators, type, dst, src, tmp1);
 5756   }
 5757 }
 5758 
 5759 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5760                                      Register tmp1, Register tmp2, Register tmp3) {
 5761   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5762   decorators = AccessInternal::decorator_fixup(decorators, type);
 5763   bool as_raw = (decorators & AS_RAW) != 0;
 5764   if (as_raw) {
 5765     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5766   } else {
 5767     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5768   }
 5769 }
 5770 
 5771 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
 5772                                      Register inline_layout_info) {
 5773   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5774   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
 5775 }
 5776 
 5777 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
 5778   movptr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
 5779   movl(offset, Address(offset, InlineKlass::payload_offset_offset()));
 5780 }
 5781 
 5782 void MacroAssembler::payload_addr(Register oop, Register data, Register inline_klass) {
 5783   // ((address) (void*) o) + vk->payload_offset();
 5784   Register offset = (data == oop) ? rscratch1 : data;
 5785   payload_offset(inline_klass, offset);
 5786   if (data == oop) {
 5787     addptr(data, offset);
 5788   } else {
 5789     lea(data, Address(oop, offset));
 5790   }
 5791 }
 5792 
 5793 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
 5794                                                 Register index, Register data) {
 5795   assert(index != rcx, "index needs to shift by rcx");
 5796   assert_different_registers(array, array_klass, index);
 5797   assert_different_registers(rcx, array, index);
 5798 
 5799   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
 5800   movl(rcx, Address(array_klass, Klass::layout_helper_offset()));
 5801 
 5802   // Klass::layout_helper_log2_element_size(lh)
 5803   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
 5804   shrl(rcx, Klass::_lh_log2_element_size_shift);
 5805   andl(rcx, Klass::_lh_log2_element_size_mask);
 5806   shlptr(index); // index << rcx
 5807 
 5808   lea(data, Address(array, index, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT)));
 5809 }
 5810 
 5811 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5812   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5813 }
 5814 
 5815 // Doesn't do verification, generates fixed size code
 5816 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5817   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5818 }
 5819 
 5820 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5821                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5822   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5823 }
 5824 
 5825 // Used for storing nulls.
 5826 void MacroAssembler::store_heap_oop_null(Address dst) {
 5827   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5828 }
 5829 
 5830 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5831   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5832   // Store to klass gap in destination
 5833   movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5834 }
 5835 
 5836 #ifdef ASSERT
 5837 void MacroAssembler::verify_heapbase(const char* msg) {
 5838   assert (UseCompressedOops, "should be compressed");
 5839   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5840   if (CheckCompressedOops) {
 5841     Label ok;
 5842     ExternalAddress src2(CompressedOops::base_addr());
 5843     const bool is_src2_reachable = reachable(src2);
 5844     if (!is_src2_reachable) {
 5845       push(rscratch1);  // cmpptr trashes rscratch1
 5846     }
 5847     cmpptr(r12_heapbase, src2, rscratch1);
 5848     jcc(Assembler::equal, ok);
 5849     STOP(msg);
 5850     bind(ok);
 5851     if (!is_src2_reachable) {
 5852       pop(rscratch1);
 5853     }
 5854   }
 5855 }
 5856 #endif
 5857 
 5858 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5859 void MacroAssembler::encode_heap_oop(Register r) {
 5860 #ifdef ASSERT
 5861   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5862 #endif
 5863   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5864   if (CompressedOops::base() == nullptr) {
 5865     if (CompressedOops::shift() != 0) {
 5866       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5867       shrq(r, LogMinObjAlignmentInBytes);
 5868     }
 5869     return;
 5870   }
 5871   testq(r, r);
 5872   cmovq(Assembler::equal, r, r12_heapbase);
 5873   subq(r, r12_heapbase);
 5874   shrq(r, LogMinObjAlignmentInBytes);
 5875 }
 5876 
 5877 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5878 #ifdef ASSERT
 5879   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5880   if (CheckCompressedOops) {
 5881     Label ok;
 5882     testq(r, r);
 5883     jcc(Assembler::notEqual, ok);
 5884     STOP("null oop passed to encode_heap_oop_not_null");
 5885     bind(ok);
 5886   }
 5887 #endif
 5888   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5889   if (CompressedOops::base() != nullptr) {
 5890     subq(r, r12_heapbase);
 5891   }
 5892   if (CompressedOops::shift() != 0) {
 5893     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5894     shrq(r, LogMinObjAlignmentInBytes);
 5895   }
 5896 }
 5897 
 5898 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5899 #ifdef ASSERT
 5900   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5901   if (CheckCompressedOops) {
 5902     Label ok;
 5903     testq(src, src);
 5904     jcc(Assembler::notEqual, ok);
 5905     STOP("null oop passed to encode_heap_oop_not_null2");
 5906     bind(ok);
 5907   }
 5908 #endif
 5909   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5910   if (dst != src) {
 5911     movq(dst, src);
 5912   }
 5913   if (CompressedOops::base() != nullptr) {
 5914     subq(dst, r12_heapbase);
 5915   }
 5916   if (CompressedOops::shift() != 0) {
 5917     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5918     shrq(dst, LogMinObjAlignmentInBytes);
 5919   }
 5920 }
 5921 
 5922 void  MacroAssembler::decode_heap_oop(Register r) {
 5923 #ifdef ASSERT
 5924   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5925 #endif
 5926   if (CompressedOops::base() == nullptr) {
 5927     if (CompressedOops::shift() != 0) {
 5928       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5929       shlq(r, LogMinObjAlignmentInBytes);
 5930     }
 5931   } else {
 5932     Label done;
 5933     shlq(r, LogMinObjAlignmentInBytes);
 5934     jccb(Assembler::equal, done);
 5935     addq(r, r12_heapbase);
 5936     bind(done);
 5937   }
 5938   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5939 }
 5940 
 5941 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5942   // Note: it will change flags
 5943   assert (UseCompressedOops, "should only be used for compressed headers");
 5944   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5945   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5946   // vtableStubs also counts instructions in pd_code_size_limit.
 5947   // Also do not verify_oop as this is called by verify_oop.
 5948   if (CompressedOops::shift() != 0) {
 5949     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5950     shlq(r, LogMinObjAlignmentInBytes);
 5951     if (CompressedOops::base() != nullptr) {
 5952       addq(r, r12_heapbase);
 5953     }
 5954   } else {
 5955     assert (CompressedOops::base() == nullptr, "sanity");
 5956   }
 5957 }
 5958 
 5959 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5960   // Note: it will change flags
 5961   assert (UseCompressedOops, "should only be used for compressed headers");
 5962   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5963   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5964   // vtableStubs also counts instructions in pd_code_size_limit.
 5965   // Also do not verify_oop as this is called by verify_oop.
 5966   if (CompressedOops::shift() != 0) {
 5967     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5968     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5969       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5970     } else {
 5971       if (dst != src) {
 5972         movq(dst, src);
 5973       }
 5974       shlq(dst, LogMinObjAlignmentInBytes);
 5975       if (CompressedOops::base() != nullptr) {
 5976         addq(dst, r12_heapbase);
 5977       }
 5978     }
 5979   } else {
 5980     assert (CompressedOops::base() == nullptr, "sanity");
 5981     if (dst != src) {
 5982       movq(dst, src);
 5983     }
 5984   }
 5985 }
 5986 
 5987 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5988   BLOCK_COMMENT("encode_klass_not_null {");
 5989   assert_different_registers(r, tmp);
 5990   if (CompressedKlassPointers::base() != nullptr) {
 5991     if (AOTCodeCache::is_on_for_dump()) {
 5992       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5993     } else {
 5994       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5995     }
 5996     subq(r, tmp);
 5997   }
 5998   if (CompressedKlassPointers::shift() != 0) {
 5999     shrq(r, CompressedKlassPointers::shift());
 6000   }
 6001   BLOCK_COMMENT("} encode_klass_not_null");
 6002 }
 6003 
 6004 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 6005   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 6006   assert_different_registers(src, dst);
 6007   if (CompressedKlassPointers::base() != nullptr) {
 6008     if (AOTCodeCache::is_on_for_dump()) {
 6009       movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 6010       negq(dst);
 6011     } else {
 6012       movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 6013     }
 6014     addq(dst, src);
 6015   } else {
 6016     movptr(dst, src);
 6017   }
 6018   if (CompressedKlassPointers::shift() != 0) {
 6019     shrq(dst, CompressedKlassPointers::shift());
 6020   }
 6021   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 6022 }
 6023 
 6024 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 6025   BLOCK_COMMENT("decode_klass_not_null {");
 6026   assert_different_registers(r, tmp);
 6027   // Note: it will change flags
 6028   // Cannot assert, unverified entry point counts instructions (see .ad file)
 6029   // vtableStubs also counts instructions in pd_code_size_limit.
 6030   // Also do not verify_oop as this is called by verify_oop.
 6031   if (CompressedKlassPointers::shift() != 0) {
 6032     shlq(r, CompressedKlassPointers::shift());
 6033   }
 6034   if (CompressedKlassPointers::base() != nullptr) {
 6035     if (AOTCodeCache::is_on_for_dump()) {
 6036       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 6037     } else {
 6038       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 6039     }
 6040     addq(r, tmp);
 6041   }
 6042   BLOCK_COMMENT("} decode_klass_not_null");
 6043 }
 6044 
 6045 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 6046   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 6047   assert_different_registers(src, dst);
 6048   // Note: it will change flags
 6049   // Cannot assert, unverified entry point counts instructions (see .ad file)
 6050   // vtableStubs also counts instructions in pd_code_size_limit.
 6051   // Also do not verify_oop as this is called by verify_oop.
 6052 
 6053   if (CompressedKlassPointers::base() == nullptr &&
 6054       CompressedKlassPointers::shift() == 0) {
 6055     // The best case scenario is that there is no base or shift. Then it is already
 6056     // a pointer that needs nothing but a register rename.
 6057     movl(dst, src);
 6058   } else {
 6059     if (CompressedKlassPointers::shift() <= Address::times_8) {
 6060       if (CompressedKlassPointers::base() != nullptr) {
 6061         if (AOTCodeCache::is_on_for_dump()) {
 6062           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 6063         } else {
 6064           movptr(dst, (intptr_t)CompressedKlassPointers::base());
 6065         }
 6066       } else {
 6067         xorq(dst, dst);
 6068       }
 6069       if (CompressedKlassPointers::shift() != 0) {
 6070         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 6071         leaq(dst, Address(dst, src, Address::times_8, 0));
 6072       } else {
 6073         addq(dst, src);
 6074       }
 6075     } else {
 6076       if (CompressedKlassPointers::base() != nullptr) {
 6077         if (AOTCodeCache::is_on_for_dump()) {
 6078           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 6079           shrq(dst, CompressedKlassPointers::shift());
 6080         } else {
 6081           const intptr_t base_right_shifted =
 6082                (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 6083           movptr(dst, base_right_shifted);
 6084         }
 6085       } else {
 6086         xorq(dst, dst);
 6087       }
 6088       addq(dst, src);
 6089       shlq(dst, CompressedKlassPointers::shift());
 6090     }
 6091   }
 6092   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 6093 }
 6094 
 6095 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 6096   assert (UseCompressedOops, "should only be used for compressed headers");
 6097   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6098   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6099   int oop_index = oop_recorder()->find_index(obj);
 6100   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6101   mov_narrow_oop(dst, oop_index, rspec);
 6102 }
 6103 
 6104 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 6105   assert (UseCompressedOops, "should only be used for compressed headers");
 6106   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6107   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6108   int oop_index = oop_recorder()->find_index(obj);
 6109   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6110   mov_narrow_oop(dst, oop_index, rspec);
 6111 }
 6112 
 6113 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 6114   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6115   int klass_index = oop_recorder()->find_index(k);
 6116   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6117   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6118 }
 6119 
 6120 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 6121   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6122   int klass_index = oop_recorder()->find_index(k);
 6123   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6124   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6125 }
 6126 
 6127 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 6128   assert (UseCompressedOops, "should only be used for compressed headers");
 6129   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6130   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6131   int oop_index = oop_recorder()->find_index(obj);
 6132   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6133   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 6134 }
 6135 
 6136 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 6137   assert (UseCompressedOops, "should only be used for compressed headers");
 6138   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6139   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6140   int oop_index = oop_recorder()->find_index(obj);
 6141   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6142   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 6143 }
 6144 
 6145 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 6146   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6147   int klass_index = oop_recorder()->find_index(k);
 6148   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6149   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6150 }
 6151 
 6152 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 6153   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6154   int klass_index = oop_recorder()->find_index(k);
 6155   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6156   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6157 }
 6158 
 6159 void MacroAssembler::reinit_heapbase() {
 6160   if (UseCompressedOops) {
 6161     if (Universe::heap() != nullptr && !AOTCodeCache::is_on_for_dump()) {
 6162       if (CompressedOops::base() == nullptr) {
 6163         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 6164       } else {
 6165         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 6166       }
 6167     } else {
 6168       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 6169     }
 6170   }
 6171 }
 6172 
 6173 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
 6174   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
 6175   // An inline type might be returned. If fields are in registers we
 6176   // need to allocate an inline type instance and initialize it with
 6177   // the value of the fields.
 6178   Label skip;
 6179   // We only need a new buffered inline type if a new one is not returned
 6180   testptr(rax, 1);
 6181   jcc(Assembler::zero, skip);
 6182   int call_offset = -1;
 6183 
 6184 #ifdef _LP64
 6185   // The following code is similar to allocate_instance but has some slight differences,
 6186   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
 6187   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
 6188   Label slow_case;
 6189   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
 6190   mov(rscratch1, rax); // save rax for slow_case since *_allocate may corrupt it when allocation failed
 6191   if (vk != nullptr) {
 6192     // Called from C1, where the return type is statically known.
 6193     movptr(rbx, (intptr_t)vk->get_InlineKlass());
 6194     jint lh = vk->layout_helper();
 6195     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
 6196     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
 6197       tlab_allocate(rax, noreg, lh, r13, r14, slow_case);
 6198     } else {
 6199       jmp(slow_case);
 6200     }
 6201   } else {
 6202     // Call from interpreter. RAX contains ((the InlineKlass* of the return type) | 0x01)
 6203     mov(rbx, rax);
 6204     andptr(rbx, -2);
 6205     if (UseTLAB) {
 6206       movl(r14, Address(rbx, Klass::layout_helper_offset()));
 6207       testl(r14, Klass::_lh_instance_slow_path_bit);
 6208       jcc(Assembler::notZero, slow_case);
 6209       tlab_allocate(rax, r14, 0, r13, r14, slow_case);
 6210     } else {
 6211       jmp(slow_case);
 6212     }
 6213   }
 6214   if (UseTLAB) {
 6215     // 2. Initialize buffered inline instance header
 6216     Register buffer_obj = rax;
 6217     Register klass = rbx;
 6218     if (UseCompactObjectHeaders) {
 6219       Register mark_word = r13;
 6220       movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 6221       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), mark_word);
 6222     } else {
 6223       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), (intptr_t)markWord::inline_type_prototype().value());
 6224       xorl(r13, r13);
 6225       store_klass_gap(buffer_obj, r13);
 6226       if (vk == nullptr) {
 6227         // store_klass corrupts rbx(klass), so save it in r13 for later use (interpreter case only).
 6228         mov(r13, klass);
 6229       }
 6230       store_klass(buffer_obj, klass, rscratch1);
 6231       klass = r13;
 6232     }
 6233     // 3. Initialize its fields with an inline class specific handler
 6234     if (vk != nullptr) {
 6235       call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
 6236     } else {
 6237       movptr(rbx, Address(klass, InlineKlass::adr_members_offset()));
 6238       movptr(rbx, Address(rbx, InlineKlass::pack_handler_offset()));
 6239       call(rbx);
 6240     }
 6241     jmp(skip);
 6242   }
 6243   bind(slow_case);
 6244   // We failed to allocate a new inline type, fall back to a runtime
 6245   // call. Some oop field may be live in some registers but we can't
 6246   // tell. That runtime call will take care of preserving them
 6247   // across a GC if there's one.
 6248   mov(rax, rscratch1);
 6249 #endif
 6250 
 6251   if (from_interpreter) {
 6252     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
 6253   } else {
 6254     call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
 6255     call_offset = offset();
 6256   }
 6257 
 6258   bind(skip);
 6259   return call_offset;
 6260 }
 6261 
 6262 // Move a value between registers/stack slots and update the reg_state
 6263 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
 6264   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
 6265   if (reg_state[to->value()] == reg_written) {
 6266     return true; // Already written
 6267   }
 6268   if (from != to && bt != T_VOID) {
 6269     if (reg_state[to->value()] == reg_readonly) {
 6270       return false; // Not yet writable
 6271     }
 6272     if (from->is_reg()) {
 6273       if (to->is_reg()) {
 6274         if (from->is_XMMRegister()) {
 6275           if (bt == T_DOUBLE) {
 6276             movdbl(to->as_XMMRegister(), from->as_XMMRegister());
 6277           } else {
 6278             assert(bt == T_FLOAT, "must be float");
 6279             movflt(to->as_XMMRegister(), from->as_XMMRegister());
 6280           }
 6281         } else {
 6282           movq(to->as_Register(), from->as_Register());
 6283         }
 6284       } else {
 6285         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6286         Address to_addr = Address(rsp, st_off);
 6287         if (from->is_XMMRegister()) {
 6288           if (bt == T_DOUBLE) {
 6289             movdbl(to_addr, from->as_XMMRegister());
 6290           } else {
 6291             assert(bt == T_FLOAT, "must be float");
 6292             movflt(to_addr, from->as_XMMRegister());
 6293           }
 6294         } else {
 6295           movq(to_addr, from->as_Register());
 6296         }
 6297       }
 6298     } else {
 6299       Address from_addr = Address(rsp, from->reg2stack() * VMRegImpl::stack_slot_size + wordSize);
 6300       if (to->is_reg()) {
 6301         if (to->is_XMMRegister()) {
 6302           if (bt == T_DOUBLE) {
 6303             movdbl(to->as_XMMRegister(), from_addr);
 6304           } else {
 6305             assert(bt == T_FLOAT, "must be float");
 6306             movflt(to->as_XMMRegister(), from_addr);
 6307           }
 6308         } else {
 6309           movq(to->as_Register(), from_addr);
 6310         }
 6311       } else {
 6312         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6313         movq(r13, from_addr);
 6314         movq(Address(rsp, st_off), r13);
 6315       }
 6316     }
 6317   }
 6318   // Update register states
 6319   reg_state[from->value()] = reg_writable;
 6320   reg_state[to->value()] = reg_written;
 6321   return true;
 6322 }
 6323 
 6324 // Calculate the extra stack space required for packing or unpacking inline
 6325 // args and adjust the stack pointer (see MacroAssembler::remove_frame).
 6326 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
 6327   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
 6328   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
 6329   assert(sp_inc > 0, "sanity");
 6330   // Two additional slots to account for return address
 6331   sp_inc +=  2 * VMRegImpl::stack_slot_size;
 6332 
 6333   push(rbp);
 6334   subptr(rsp, sp_inc);
 6335 #ifdef ASSERT
 6336   movl(Address(rsp, 0), badRegWordVal);
 6337   movl(Address(rsp, VMRegImpl::stack_slot_size), badRegWordVal);
 6338 #endif
 6339   return sp_inc + wordSize; // account for rbp space
 6340 }
 6341 
 6342 // Read all fields from an inline type buffer and store the field values in registers/stack slots.
 6343 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
 6344                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
 6345                                           RegState reg_state[]) {
 6346   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
 6347   assert(from->is_valid(), "source must be valid");
 6348   bool progress = false;
 6349 #ifdef ASSERT
 6350   const int start_offset = offset();
 6351 #endif
 6352 
 6353   Label L_null, L_notNull;
 6354   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
 6355   Register tmp1 = r10;
 6356   Register tmp2 = r13;
 6357   Register fromReg = noreg;
 6358   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
 6359   bool done = true;
 6360   bool mark_done = true;
 6361   VMReg toReg;
 6362   BasicType bt;
 6363   // Check if argument requires a null check
 6364   bool null_check = false;
 6365   VMReg nullCheckReg;
 6366   while (stream.next(nullCheckReg, bt)) {
 6367     if (sig->at(stream.sig_index())._offset == -1) {
 6368       null_check = true;
 6369       break;
 6370     }
 6371   }
 6372   stream.reset(sig_index, to_index);
 6373   while (stream.next(toReg, bt)) {
 6374     assert(toReg->is_valid(), "destination must be valid");
 6375     int idx = (int)toReg->value();
 6376     if (reg_state[idx] == reg_readonly) {
 6377       if (idx != from->value()) {
 6378         mark_done = false;
 6379       }
 6380       done = false;
 6381       continue;
 6382     } else if (reg_state[idx] == reg_written) {
 6383       continue;
 6384     }
 6385     assert(reg_state[idx] == reg_writable, "must be writable");
 6386     reg_state[idx] = reg_written;
 6387     progress = true;
 6388 
 6389     if (fromReg == noreg) {
 6390       if (from->is_reg()) {
 6391         fromReg = from->as_Register();
 6392       } else {
 6393         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6394         movq(tmp1, Address(rsp, st_off));
 6395         fromReg = tmp1;
 6396       }
 6397       if (null_check) {
 6398         // Nullable inline type argument, emit null check
 6399         testptr(fromReg, fromReg);
 6400         jcc(Assembler::zero, L_null);
 6401       }
 6402     }
 6403     int off = sig->at(stream.sig_index())._offset;
 6404     if (off == -1) {
 6405       assert(null_check, "Missing null check at");
 6406       if (toReg->is_stack()) {
 6407         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6408         movq(Address(rsp, st_off), 1);
 6409       } else {
 6410         movq(toReg->as_Register(), 1);
 6411       }
 6412       continue;
 6413     }
 6414     if (sig->at(stream.sig_index())._vt_oop) {
 6415       if (toReg->is_stack()) {
 6416         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6417         movq(Address(rsp, st_off), fromReg);
 6418       } else {
 6419         movq(toReg->as_Register(), fromReg);
 6420       }
 6421       continue;
 6422     }
 6423     assert(off > 0, "offset in object should be positive");
 6424     Address fromAddr = Address(fromReg, off);
 6425     if (!toReg->is_XMMRegister()) {
 6426       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
 6427       if (is_reference_type(bt)) {
 6428         load_heap_oop(dst, fromAddr);
 6429       } else {
 6430         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
 6431         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
 6432       }
 6433       if (toReg->is_stack()) {
 6434         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6435         movq(Address(rsp, st_off), dst);
 6436       }
 6437     } else if (bt == T_DOUBLE) {
 6438       movdbl(toReg->as_XMMRegister(), fromAddr);
 6439     } else {
 6440       assert(bt == T_FLOAT, "must be float");
 6441       movflt(toReg->as_XMMRegister(), fromAddr);
 6442     }
 6443   }
 6444   if (progress && null_check) {
 6445     if (done) {
 6446       jmp(L_notNull);
 6447       bind(L_null);
 6448       // Set null marker to zero to signal that the argument is null.
 6449       // Also set all fields to zero since the runtime requires a canonical
 6450       // representation of a flat null.
 6451       stream.reset(sig_index, to_index);
 6452       while (stream.next(toReg, bt)) {
 6453         if (toReg->is_stack()) {
 6454           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6455           movq(Address(rsp, st_off), 0);
 6456         } else if (toReg->is_XMMRegister()) {
 6457           xorps(toReg->as_XMMRegister(), toReg->as_XMMRegister());
 6458         } else {
 6459           xorl(toReg->as_Register(), toReg->as_Register());
 6460         }
 6461       }
 6462       bind(L_notNull);
 6463     } else {
 6464       bind(L_null);
 6465     }
 6466   }
 6467 
 6468   sig_index = stream.sig_index();
 6469   to_index = stream.regs_index();
 6470 
 6471   if (mark_done && reg_state[from->value()] != reg_written) {
 6472     // This is okay because no one else will write to that slot
 6473     reg_state[from->value()] = reg_writable;
 6474   }
 6475   from_index--;
 6476   assert(progress || (start_offset == offset()), "should not emit code");
 6477   return done;
 6478 }
 6479 
 6480 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
 6481                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
 6482                                         RegState reg_state[], Register val_array) {
 6483   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
 6484   assert(to->is_valid(), "destination must be valid");
 6485 
 6486   if (reg_state[to->value()] == reg_written) {
 6487     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6488     return true; // Already written
 6489   }
 6490 
 6491   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
 6492   Register val_obj_tmp = r11;
 6493   Register from_reg_tmp = r14;
 6494   Register tmp1 = r10;
 6495   Register tmp2 = r13;
 6496   Register tmp3 = rbx;
 6497   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
 6498 
 6499   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
 6500 
 6501   if (reg_state[to->value()] == reg_readonly) {
 6502     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
 6503       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6504       return false; // Not yet writable
 6505     }
 6506     val_obj = val_obj_tmp;
 6507   }
 6508 
 6509   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
 6510   VMReg fromReg;
 6511   BasicType bt;
 6512   Label L_null;
 6513   while (stream.next(fromReg, bt)) {
 6514     assert(fromReg->is_valid(), "source must be valid");
 6515     reg_state[fromReg->value()] = reg_writable;
 6516 
 6517     int off = sig->at(stream.sig_index())._offset;
 6518     if (off == -1) {
 6519       // Nullable inline type argument, emit null check
 6520       Label L_notNull;
 6521       if (fromReg->is_stack()) {
 6522         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6523         testb(Address(rsp, ld_off), 1);
 6524       } else {
 6525         testb(fromReg->as_Register(), 1);
 6526       }
 6527       jcc(Assembler::notZero, L_notNull);
 6528       movptr(val_obj, 0);
 6529       jmp(L_null);
 6530       bind(L_notNull);
 6531       continue;
 6532     }
 6533     if (sig->at(stream.sig_index())._vt_oop) {
 6534       // buffer argument: use if non null
 6535       if (fromReg->is_stack()) {
 6536         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6537         movptr(val_obj, Address(rsp, ld_off));
 6538       } else {
 6539         movptr(val_obj, fromReg->as_Register());
 6540       }
 6541       testptr(val_obj, val_obj);
 6542       jcc(Assembler::notEqual, L_null);
 6543       // otherwise get the buffer from the just allocated pool of buffers
 6544       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
 6545       load_heap_oop(val_obj, Address(val_array, index));
 6546       continue;
 6547     }
 6548 
 6549     assert(off > 0, "offset in object should be positive");
 6550     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
 6551 
 6552     // Pack the scalarized field into the value object.
 6553     Address dst(val_obj, off);
 6554     if (!fromReg->is_XMMRegister()) {
 6555       Register src;
 6556       if (fromReg->is_stack()) {
 6557         src = from_reg_tmp;
 6558         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6559         load_sized_value(src, Address(rsp, ld_off), size_in_bytes, /* is_signed */ false);
 6560       } else {
 6561         src = fromReg->as_Register();
 6562       }
 6563       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
 6564       if (is_reference_type(bt)) {
 6565         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
 6566         mov(tmp3, val_obj);
 6567         Address dst_with_tmp3(tmp3, off);
 6568         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
 6569       } else {
 6570         store_sized_value(dst, src, size_in_bytes);
 6571       }
 6572     } else if (bt == T_DOUBLE) {
 6573       movdbl(dst, fromReg->as_XMMRegister());
 6574     } else {
 6575       assert(bt == T_FLOAT, "must be float");
 6576       movflt(dst, fromReg->as_XMMRegister());
 6577     }
 6578   }
 6579   bind(L_null);
 6580   sig_index = stream.sig_index();
 6581   from_index = stream.regs_index();
 6582 
 6583   assert(reg_state[to->value()] == reg_writable, "must have already been read");
 6584   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
 6585   assert(success, "to register must be writeable");
 6586   return true;
 6587 }
 6588 
 6589 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
 6590   return reg->is_XMMRegister() ? xmm8->as_VMReg() : r14->as_VMReg();
 6591 }
 6592 
 6593 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
 6594   assert((initial_framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 6595   if (needs_stack_repair) {
 6596     // The method has a scalarized entry point (where fields of value object arguments
 6597     // are passed through registers and stack), and a non-scalarized entry point (where
 6598     // value object arguments are given as oops). The non-scalarized entry point will
 6599     // first load each field of value object arguments and store them in registers and on
 6600     // the stack in a way compatible with the scalarized entry point. To do so, some extra
 6601     // stack space might be reserved (if argument registers are not enough). On leaving the
 6602     // method, this space must be freed.
 6603     //
 6604     // In case we used the non-scalarized entry point the stack looks like this:
 6605     //
 6606     // | Arguments from caller     |
 6607     // |---------------------------|  <-- caller's SP
 6608     // | Return address #1         |
 6609     // | Saved RBP #1              |
 6610     // |---------------------------|
 6611     // | Extension space for       |
 6612     // |   inline arg (un)packing  |
 6613     // |---------------------------|  <-- start of this method's frame
 6614     // | Return address #2         |
 6615     // | Saved RBP #2              |
 6616     // |---------------------------|  <-- RBP (with -XX:+PreserveFramePointer)
 6617     // | sp_inc                    |
 6618     // | method locals             |
 6619     // |---------------------------|  <-- SP
 6620     //
 6621     // Space for the return pc and saved rbp is reserved twice. But only the #1 copies
 6622     // contain the real values of return pc and saved rbp. The #2 copies are not reliable
 6623     // and should not be used. They are mostly needed to add space between the extension
 6624     // space and the locals, as there would be between the real arguments and the locals
 6625     // if we don't need to do unpacking (from the scalarized entry point).
 6626     //
 6627     // When leaving, one must load RBP #1 into RBP, and use the copy #1 of the return address,
 6628     // while keeping in mind that from the scalarized entry point, there will be only one
 6629     // copy. Indeed, in the case we used the scalarized calling convention, the stack looks like this:
 6630     //
 6631     // | Arguments from caller     |
 6632     // |---------------------------|  <-- caller's SP
 6633     // | Return address            |
 6634     // | Saved RBP                 |
 6635     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
 6636     // | sp_inc                    |
 6637     // | method locals             |
 6638     // |---------------------------|  <-- SP
 6639     //
 6640     // The sp_inc stack slot holds the total size of the frame, including the extension
 6641     // space and copies #2 of the return address and the saved RBP (but never the copies
 6642     // #1 of the return address and saved RBP). That is how to find the copies #1 of the
 6643     // return address and saved rbp. This size is expressed in bytes. Be careful when using
 6644     // it from C++ in pointer arithmetic you might need to divide it by wordSize.
 6645 
 6646     // The stack increment resides just below the saved rbp
 6647     addq(rsp, Address(rsp, initial_framesize - wordSize));
 6648     pop(rbp);
 6649   } else {
 6650     if (initial_framesize > 0) {
 6651       addq(rsp, initial_framesize);
 6652     }
 6653     pop(rbp);
 6654   }
 6655 }
 6656 
 6657 #if COMPILER2_OR_JVMCI
 6658 
 6659 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 6660 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp, KRegister mask) {
 6661   // cnt - number of qwords (8-byte words).
 6662   // base - start address, qword aligned.
 6663   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 6664   bool use64byteVector = (MaxVectorSize == 64) && (CopyAVX3Threshold == 0);
 6665   if (use64byteVector) {
 6666     evpbroadcastq(xtmp, val, AVX_512bit);
 6667   } else if (MaxVectorSize >= 32) {
 6668     movdq(xtmp, val);
 6669     punpcklqdq(xtmp, xtmp);
 6670     vinserti128_high(xtmp, xtmp);
 6671   } else {
 6672     movdq(xtmp, val);
 6673     punpcklqdq(xtmp, xtmp);
 6674   }
 6675   jmp(L_zero_64_bytes);
 6676 
 6677   BIND(L_loop);
 6678   if (MaxVectorSize >= 32) {
 6679     fill64(base, 0, xtmp, use64byteVector);
 6680   } else {
 6681     movdqu(Address(base,  0), xtmp);
 6682     movdqu(Address(base, 16), xtmp);
 6683     movdqu(Address(base, 32), xtmp);
 6684     movdqu(Address(base, 48), xtmp);
 6685   }
 6686   addptr(base, 64);
 6687 
 6688   BIND(L_zero_64_bytes);
 6689   subptr(cnt, 8);
 6690   jccb(Assembler::greaterEqual, L_loop);
 6691 
 6692   // Copy trailing 64 bytes
 6693   if (use64byteVector) {
 6694     addptr(cnt, 8);
 6695     jccb(Assembler::equal, L_end);
 6696     fill64_masked(3, base, 0, xtmp, mask, cnt, val, true);
 6697     jmp(L_end);
 6698   } else {
 6699     addptr(cnt, 4);
 6700     jccb(Assembler::less, L_tail);
 6701     if (MaxVectorSize >= 32) {
 6702       vmovdqu(Address(base, 0), xtmp);
 6703     } else {
 6704       movdqu(Address(base,  0), xtmp);
 6705       movdqu(Address(base, 16), xtmp);
 6706     }
 6707   }
 6708   addptr(base, 32);
 6709   subptr(cnt, 4);
 6710 
 6711   BIND(L_tail);
 6712   addptr(cnt, 4);
 6713   jccb(Assembler::lessEqual, L_end);
 6714   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 6715     fill32_masked(3, base, 0, xtmp, mask, cnt, val);
 6716   } else {
 6717     decrement(cnt);
 6718 
 6719     BIND(L_sloop);
 6720     movq(Address(base, 0), xtmp);
 6721     addptr(base, 8);
 6722     decrement(cnt);
 6723     jccb(Assembler::greaterEqual, L_sloop);
 6724   }
 6725   BIND(L_end);
 6726 }
 6727 
 6728 // Clearing constant sized memory using YMM/ZMM registers.
 6729 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 6730   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 6731   bool use64byteVector = (MaxVectorSize > 32) && (CopyAVX3Threshold == 0);
 6732 
 6733   int vector64_count = (cnt & (~0x7)) >> 3;
 6734   cnt = cnt & 0x7;
 6735   const int fill64_per_loop = 4;
 6736   const int max_unrolled_fill64 = 8;
 6737 
 6738   // 64 byte initialization loop.
 6739   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 6740   int start64 = 0;
 6741   if (vector64_count > max_unrolled_fill64) {
 6742     Label LOOP;
 6743     Register index = rtmp;
 6744 
 6745     start64 = vector64_count - (vector64_count % fill64_per_loop);
 6746 
 6747     movl(index, 0);
 6748     BIND(LOOP);
 6749     for (int i = 0; i < fill64_per_loop; i++) {
 6750       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 6751     }
 6752     addl(index, fill64_per_loop * 64);
 6753     cmpl(index, start64 * 64);
 6754     jccb(Assembler::less, LOOP);
 6755   }
 6756   for (int i = start64; i < vector64_count; i++) {
 6757     fill64(base, i * 64, xtmp, use64byteVector);
 6758   }
 6759 
 6760   // Clear remaining 64 byte tail.
 6761   int disp = vector64_count * 64;
 6762   if (cnt) {
 6763     switch (cnt) {
 6764       case 1:
 6765         movq(Address(base, disp), xtmp);
 6766         break;
 6767       case 2:
 6768         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 6769         break;
 6770       case 3:
 6771         movl(rtmp, 0x7);
 6772         kmovwl(mask, rtmp);
 6773         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 6774         break;
 6775       case 4:
 6776         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6777         break;
 6778       case 5:
 6779         if (use64byteVector) {
 6780           movl(rtmp, 0x1F);
 6781           kmovwl(mask, rtmp);
 6782           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6783         } else {
 6784           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6785           movq(Address(base, disp + 32), xtmp);
 6786         }
 6787         break;
 6788       case 6:
 6789         if (use64byteVector) {
 6790           movl(rtmp, 0x3F);
 6791           kmovwl(mask, rtmp);
 6792           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6793         } else {
 6794           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6795           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 6796         }
 6797         break;
 6798       case 7:
 6799         if (use64byteVector) {
 6800           movl(rtmp, 0x7F);
 6801           kmovwl(mask, rtmp);
 6802           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6803         } else {
 6804           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6805           movl(rtmp, 0x7);
 6806           kmovwl(mask, rtmp);
 6807           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 6808         }
 6809         break;
 6810       default:
 6811         fatal("Unexpected length : %d\n",cnt);
 6812         break;
 6813     }
 6814   }
 6815 }
 6816 
 6817 void MacroAssembler::clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp,
 6818                                bool is_large, bool word_copy_only, KRegister mask) {
 6819   // cnt      - number of qwords (8-byte words).
 6820   // base     - start address, qword aligned.
 6821   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6822   assert(base==rdi, "base register must be edi for rep stos");
 6823   assert(val==rax,   "val register must be eax for rep stos");
 6824   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6825   assert(InitArrayShortSize % BytesPerLong == 0,
 6826     "InitArrayShortSize should be the multiple of BytesPerLong");
 6827 
 6828   Label DONE;
 6829 
 6830   if (!is_large) {
 6831     Label LOOP, LONG;
 6832     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6833     jccb(Assembler::greater, LONG);
 6834 
 6835     decrement(cnt);
 6836     jccb(Assembler::negative, DONE); // Zero length
 6837 
 6838     // Use individual pointer-sized stores for small counts:
 6839     BIND(LOOP);
 6840     movptr(Address(base, cnt, Address::times_ptr), val);
 6841     decrement(cnt);
 6842     jccb(Assembler::greaterEqual, LOOP);
 6843     jmpb(DONE);
 6844 
 6845     BIND(LONG);
 6846   }
 6847 
 6848   // Use longer rep-prefixed ops for non-small counts:
 6849   if (UseFastStosb && !word_copy_only) {
 6850     shlptr(cnt, 3); // convert to number of bytes
 6851     rep_stosb();
 6852   } else if (UseXMMForObjInit) {
 6853     xmm_clear_mem(base, cnt, val, xtmp, mask);
 6854   } else {
 6855     rep_stos();
 6856   }
 6857 
 6858   BIND(DONE);
 6859 }
 6860 
 6861 #endif //COMPILER2_OR_JVMCI
 6862 
 6863 
 6864 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6865                                    Register to, Register value, Register count,
 6866                                    Register rtmp, XMMRegister xtmp) {
 6867   ShortBranchVerifier sbv(this);
 6868   assert_different_registers(to, value, count, rtmp);
 6869   Label L_exit;
 6870   Label L_fill_2_bytes, L_fill_4_bytes;
 6871 
 6872 #if defined(COMPILER2)
 6873   if(MaxVectorSize >=32 &&
 6874      VM_Version::supports_avx512vlbw() &&
 6875      VM_Version::supports_bmi2()) {
 6876     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6877     return;
 6878   }
 6879 #endif
 6880 
 6881   int shift = -1;
 6882   switch (t) {
 6883     case T_BYTE:
 6884       shift = 2;
 6885       break;
 6886     case T_SHORT:
 6887       shift = 1;
 6888       break;
 6889     case T_INT:
 6890       shift = 0;
 6891       break;
 6892     default: ShouldNotReachHere();
 6893   }
 6894 
 6895   if (t == T_BYTE) {
 6896     andl(value, 0xff);
 6897     movl(rtmp, value);
 6898     shll(rtmp, 8);
 6899     orl(value, rtmp);
 6900   }
 6901   if (t == T_SHORT) {
 6902     andl(value, 0xffff);
 6903   }
 6904   if (t == T_BYTE || t == T_SHORT) {
 6905     movl(rtmp, value);
 6906     shll(rtmp, 16);
 6907     orl(value, rtmp);
 6908   }
 6909 
 6910   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6911   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6912   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6913     Label L_skip_align2;
 6914     // align source address at 4 bytes address boundary
 6915     if (t == T_BYTE) {
 6916       Label L_skip_align1;
 6917       // One byte misalignment happens only for byte arrays
 6918       testptr(to, 1);
 6919       jccb(Assembler::zero, L_skip_align1);
 6920       movb(Address(to, 0), value);
 6921       increment(to);
 6922       decrement(count);
 6923       BIND(L_skip_align1);
 6924     }
 6925     // Two bytes misalignment happens only for byte and short (char) arrays
 6926     testptr(to, 2);
 6927     jccb(Assembler::zero, L_skip_align2);
 6928     movw(Address(to, 0), value);
 6929     addptr(to, 2);
 6930     subptr(count, 1<<(shift-1));
 6931     BIND(L_skip_align2);
 6932   }
 6933   {
 6934     Label L_fill_32_bytes;
 6935     if (!UseUnalignedLoadStores) {
 6936       // align to 8 bytes, we know we are 4 byte aligned to start
 6937       testptr(to, 4);
 6938       jccb(Assembler::zero, L_fill_32_bytes);
 6939       movl(Address(to, 0), value);
 6940       addptr(to, 4);
 6941       subptr(count, 1<<shift);
 6942     }
 6943     BIND(L_fill_32_bytes);
 6944     {
 6945       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6946       movdl(xtmp, value);
 6947       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6948         Label L_check_fill_32_bytes;
 6949         if (UseAVX > 2) {
 6950           // Fill 64-byte chunks
 6951           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6952 
 6953           // If number of bytes to fill < CopyAVX3Threshold, perform fill using AVX2
 6954           cmpptr(count, CopyAVX3Threshold);
 6955           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6956 
 6957           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6958 
 6959           subptr(count, 16 << shift);
 6960           jcc(Assembler::less, L_check_fill_32_bytes);
 6961           align(16);
 6962 
 6963           BIND(L_fill_64_bytes_loop_avx3);
 6964           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6965           addptr(to, 64);
 6966           subptr(count, 16 << shift);
 6967           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6968           jmpb(L_check_fill_32_bytes);
 6969 
 6970           BIND(L_check_fill_64_bytes_avx2);
 6971         }
 6972         // Fill 64-byte chunks
 6973         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6974 
 6975         subptr(count, 16 << shift);
 6976         jcc(Assembler::less, L_check_fill_32_bytes);
 6977 
 6978         // align data for 64-byte chunks
 6979         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6980         if (EnableX86ECoreOpts) {
 6981             // align 'big' arrays to cache lines to minimize split_stores
 6982             cmpptr(count, 96 << shift);
 6983             jcc(Assembler::below, L_fill_64_bytes_loop);
 6984 
 6985             // Find the bytes needed for alignment
 6986             movptr(rtmp, to);
 6987             andptr(rtmp, 0x1c);
 6988             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6989             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6990             addptr(rtmp, 32);
 6991             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6992             subptr(count, rtmp);    // adjust count by number of elements
 6993 
 6994             align(16);
 6995             BIND(L_align_64_bytes_loop);
 6996             movdl(Address(to, 0), xtmp);
 6997             addptr(to, 4);
 6998             subptr(rtmp, 1 << shift);
 6999             jcc(Assembler::greater, L_align_64_bytes_loop);
 7000         }
 7001 
 7002         align(16);
 7003         BIND(L_fill_64_bytes_loop);
 7004         vmovdqu(Address(to, 0), xtmp);
 7005         vmovdqu(Address(to, 32), xtmp);
 7006         addptr(to, 64);
 7007         subptr(count, 16 << shift);
 7008         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 7009 
 7010         align(16);
 7011         BIND(L_check_fill_32_bytes);
 7012         addptr(count, 8 << shift);
 7013         jccb(Assembler::less, L_check_fill_8_bytes);
 7014         vmovdqu(Address(to, 0), xtmp);
 7015         addptr(to, 32);
 7016         subptr(count, 8 << shift);
 7017 
 7018         BIND(L_check_fill_8_bytes);
 7019         // clean upper bits of YMM registers
 7020         movdl(xtmp, value);
 7021         pshufd(xtmp, xtmp, 0);
 7022       } else {
 7023         // Fill 32-byte chunks
 7024         pshufd(xtmp, xtmp, 0);
 7025 
 7026         subptr(count, 8 << shift);
 7027         jcc(Assembler::less, L_check_fill_8_bytes);
 7028         align(16);
 7029 
 7030         BIND(L_fill_32_bytes_loop);
 7031 
 7032         if (UseUnalignedLoadStores) {
 7033           movdqu(Address(to, 0), xtmp);
 7034           movdqu(Address(to, 16), xtmp);
 7035         } else {
 7036           movq(Address(to, 0), xtmp);
 7037           movq(Address(to, 8), xtmp);
 7038           movq(Address(to, 16), xtmp);
 7039           movq(Address(to, 24), xtmp);
 7040         }
 7041 
 7042         addptr(to, 32);
 7043         subptr(count, 8 << shift);
 7044         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 7045 
 7046         BIND(L_check_fill_8_bytes);
 7047       }
 7048       addptr(count, 8 << shift);
 7049       jccb(Assembler::zero, L_exit);
 7050       jmpb(L_fill_8_bytes);
 7051 
 7052       //
 7053       // length is too short, just fill qwords
 7054       //
 7055       align(16);
 7056       BIND(L_fill_8_bytes_loop);
 7057       movq(Address(to, 0), xtmp);
 7058       addptr(to, 8);
 7059       BIND(L_fill_8_bytes);
 7060       subptr(count, 1 << (shift + 1));
 7061       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 7062     }
 7063   }
 7064 
 7065   Label L_fill_4_bytes_loop;
 7066   testl(count, 1 << shift);
 7067   jccb(Assembler::zero, L_fill_2_bytes);
 7068 
 7069   align(16);
 7070   BIND(L_fill_4_bytes_loop);
 7071   movl(Address(to, 0), value);
 7072   addptr(to, 4);
 7073 
 7074   BIND(L_fill_4_bytes);
 7075   subptr(count, 1 << shift);
 7076   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 7077 
 7078   if (t == T_BYTE || t == T_SHORT) {
 7079     Label L_fill_byte;
 7080     BIND(L_fill_2_bytes);
 7081     // fill trailing 2 bytes
 7082     testl(count, 1<<(shift-1));
 7083     jccb(Assembler::zero, L_fill_byte);
 7084     movw(Address(to, 0), value);
 7085     if (t == T_BYTE) {
 7086       addptr(to, 2);
 7087       BIND(L_fill_byte);
 7088       // fill trailing byte
 7089       testl(count, 1);
 7090       jccb(Assembler::zero, L_exit);
 7091       movb(Address(to, 0), value);
 7092     } else {
 7093       BIND(L_fill_byte);
 7094     }
 7095   } else {
 7096     BIND(L_fill_2_bytes);
 7097   }
 7098   BIND(L_exit);
 7099 }
 7100 
 7101 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 7102   switch(type) {
 7103     case T_BYTE:
 7104     case T_BOOLEAN:
 7105       evpbroadcastb(dst, src, vector_len);
 7106       break;
 7107     case T_SHORT:
 7108     case T_CHAR:
 7109       evpbroadcastw(dst, src, vector_len);
 7110       break;
 7111     case T_INT:
 7112     case T_FLOAT:
 7113       evpbroadcastd(dst, src, vector_len);
 7114       break;
 7115     case T_LONG:
 7116     case T_DOUBLE:
 7117       evpbroadcastq(dst, src, vector_len);
 7118       break;
 7119     default:
 7120       fatal("Unhandled type : %s", type2name(type));
 7121       break;
 7122   }
 7123 }
 7124 
 7125 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 7126 //
 7127 // @IntrinsicCandidate
 7128 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 7129 //         char[] sa, int sp, byte[] da, int dp, int len) {
 7130 //     int i = 0;
 7131 //     for (; i < len; i++) {
 7132 //         char c = sa[sp++];
 7133 //         if (c > '\u00FF')
 7134 //             break;
 7135 //         da[dp++] = (byte) c;
 7136 //     }
 7137 //     return i;
 7138 // }
 7139 //
 7140 // @IntrinsicCandidate
 7141 // int java.lang.StringCoding.encodeISOArray0(
 7142 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 7143 //   int i = 0;
 7144 //   for (; i < len; i++) {
 7145 //     char c = StringUTF16.getChar(sa, sp++);
 7146 //     if (c > '\u00FF')
 7147 //       break;
 7148 //     da[dp++] = (byte) c;
 7149 //   }
 7150 //   return i;
 7151 // }
 7152 //
 7153 // @IntrinsicCandidate
 7154 // int java.lang.StringCoding.encodeAsciiArray0(
 7155 //         char[] sa, int sp, byte[] da, int dp, int len) {
 7156 //   int i = 0;
 7157 //   for (; i < len; i++) {
 7158 //     char c = sa[sp++];
 7159 //     if (c >= '\u0080')
 7160 //       break;
 7161 //     da[dp++] = (byte) c;
 7162 //   }
 7163 //   return i;
 7164 // }
 7165 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 7166   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 7167   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 7168   Register tmp5, Register result, bool ascii) {
 7169 
 7170   // rsi: src
 7171   // rdi: dst
 7172   // rdx: len
 7173   // rcx: tmp5
 7174   // rax: result
 7175   ShortBranchVerifier sbv(this);
 7176   assert_different_registers(src, dst, len, tmp5, result);
 7177   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 7178 
 7179   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 7180   int short_mask = ascii ? 0xff80 : 0xff00;
 7181 
 7182   // set result
 7183   xorl(result, result);
 7184   // check for zero length
 7185   testl(len, len);
 7186   jcc(Assembler::zero, L_done);
 7187 
 7188   movl(result, len);
 7189 
 7190   // Setup pointers
 7191   lea(src, Address(src, len, Address::times_2)); // char[]
 7192   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 7193   negptr(len);
 7194 
 7195   if (UseSSE42Intrinsics || UseAVX >= 2) {
 7196     Label L_copy_8_chars, L_copy_8_chars_exit;
 7197     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 7198 
 7199     if (UseAVX >= 2) {
 7200       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 7201       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7202       movdl(tmp1Reg, tmp5);
 7203       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 7204       jmp(L_chars_32_check);
 7205 
 7206       bind(L_copy_32_chars);
 7207       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 7208       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 7209       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7210       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7211       jccb(Assembler::notZero, L_copy_32_chars_exit);
 7212       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7213       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 7214       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 7215 
 7216       bind(L_chars_32_check);
 7217       addptr(len, 32);
 7218       jcc(Assembler::lessEqual, L_copy_32_chars);
 7219 
 7220       bind(L_copy_32_chars_exit);
 7221       subptr(len, 16);
 7222       jccb(Assembler::greater, L_copy_16_chars_exit);
 7223 
 7224     } else if (UseSSE42Intrinsics) {
 7225       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7226       movdl(tmp1Reg, tmp5);
 7227       pshufd(tmp1Reg, tmp1Reg, 0);
 7228       jmpb(L_chars_16_check);
 7229     }
 7230 
 7231     bind(L_copy_16_chars);
 7232     if (UseAVX >= 2) {
 7233       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 7234       vptest(tmp2Reg, tmp1Reg);
 7235       jcc(Assembler::notZero, L_copy_16_chars_exit);
 7236       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 7237       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 7238     } else {
 7239       if (UseAVX > 0) {
 7240         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7241         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7242         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 7243       } else {
 7244         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7245         por(tmp2Reg, tmp3Reg);
 7246         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7247         por(tmp2Reg, tmp4Reg);
 7248       }
 7249       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7250       jccb(Assembler::notZero, L_copy_16_chars_exit);
 7251       packuswb(tmp3Reg, tmp4Reg);
 7252     }
 7253     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 7254 
 7255     bind(L_chars_16_check);
 7256     addptr(len, 16);
 7257     jcc(Assembler::lessEqual, L_copy_16_chars);
 7258 
 7259     bind(L_copy_16_chars_exit);
 7260     if (UseAVX >= 2) {
 7261       // clean upper bits of YMM registers
 7262       vpxor(tmp2Reg, tmp2Reg);
 7263       vpxor(tmp3Reg, tmp3Reg);
 7264       vpxor(tmp4Reg, tmp4Reg);
 7265       movdl(tmp1Reg, tmp5);
 7266       pshufd(tmp1Reg, tmp1Reg, 0);
 7267     }
 7268     subptr(len, 8);
 7269     jccb(Assembler::greater, L_copy_8_chars_exit);
 7270 
 7271     bind(L_copy_8_chars);
 7272     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 7273     ptest(tmp3Reg, tmp1Reg);
 7274     jccb(Assembler::notZero, L_copy_8_chars_exit);
 7275     packuswb(tmp3Reg, tmp1Reg);
 7276     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 7277     addptr(len, 8);
 7278     jccb(Assembler::lessEqual, L_copy_8_chars);
 7279 
 7280     bind(L_copy_8_chars_exit);
 7281     subptr(len, 8);
 7282     jccb(Assembler::zero, L_done);
 7283   }
 7284 
 7285   bind(L_copy_1_char);
 7286   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 7287   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 7288   jccb(Assembler::notZero, L_copy_1_char_exit);
 7289   movb(Address(dst, len, Address::times_1, 0), tmp5);
 7290   addptr(len, 1);
 7291   jccb(Assembler::less, L_copy_1_char);
 7292 
 7293   bind(L_copy_1_char_exit);
 7294   addptr(result, len); // len is negative count of not processed elements
 7295 
 7296   bind(L_done);
 7297 }
 7298 
 7299 /**
 7300  * Helper for multiply_to_len().
 7301  */
 7302 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 7303   addq(dest_lo, src1);
 7304   adcq(dest_hi, 0);
 7305   addq(dest_lo, src2);
 7306   adcq(dest_hi, 0);
 7307 }
 7308 
 7309 /**
 7310  * Multiply 64 bit by 64 bit first loop.
 7311  */
 7312 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 7313                                            Register y, Register y_idx, Register z,
 7314                                            Register carry, Register product,
 7315                                            Register idx, Register kdx) {
 7316   //
 7317   //  jlong carry, x[], y[], z[];
 7318   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7319   //    huge_128 product = y[idx] * x[xstart] + carry;
 7320   //    z[kdx] = (jlong)product;
 7321   //    carry  = (jlong)(product >>> 64);
 7322   //  }
 7323   //  z[xstart] = carry;
 7324   //
 7325 
 7326   Label L_first_loop, L_first_loop_exit;
 7327   Label L_one_x, L_one_y, L_multiply;
 7328 
 7329   decrementl(xstart);
 7330   jcc(Assembler::negative, L_one_x);
 7331 
 7332   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7333   rorq(x_xstart, 32); // convert big-endian to little-endian
 7334 
 7335   bind(L_first_loop);
 7336   decrementl(idx);
 7337   jcc(Assembler::negative, L_first_loop_exit);
 7338   decrementl(idx);
 7339   jcc(Assembler::negative, L_one_y);
 7340   movq(y_idx, Address(y, idx, Address::times_4,  0));
 7341   rorq(y_idx, 32); // convert big-endian to little-endian
 7342   bind(L_multiply);
 7343   movq(product, x_xstart);
 7344   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 7345   addq(product, carry);
 7346   adcq(rdx, 0);
 7347   subl(kdx, 2);
 7348   movl(Address(z, kdx, Address::times_4,  4), product);
 7349   shrq(product, 32);
 7350   movl(Address(z, kdx, Address::times_4,  0), product);
 7351   movq(carry, rdx);
 7352   jmp(L_first_loop);
 7353 
 7354   bind(L_one_y);
 7355   movl(y_idx, Address(y,  0));
 7356   jmp(L_multiply);
 7357 
 7358   bind(L_one_x);
 7359   movl(x_xstart, Address(x,  0));
 7360   jmp(L_first_loop);
 7361 
 7362   bind(L_first_loop_exit);
 7363 }
 7364 
 7365 /**
 7366  * Multiply 64 bit by 64 bit and add 128 bit.
 7367  */
 7368 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 7369                                             Register yz_idx, Register idx,
 7370                                             Register carry, Register product, int offset) {
 7371   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 7372   //     z[kdx] = (jlong)product;
 7373 
 7374   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 7375   rorq(yz_idx, 32); // convert big-endian to little-endian
 7376   movq(product, x_xstart);
 7377   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 7378   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 7379   rorq(yz_idx, 32); // convert big-endian to little-endian
 7380 
 7381   add2_with_carry(rdx, product, carry, yz_idx);
 7382 
 7383   movl(Address(z, idx, Address::times_4,  offset+4), product);
 7384   shrq(product, 32);
 7385   movl(Address(z, idx, Address::times_4,  offset), product);
 7386 
 7387 }
 7388 
 7389 /**
 7390  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 7391  */
 7392 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 7393                                              Register yz_idx, Register idx, Register jdx,
 7394                                              Register carry, Register product,
 7395                                              Register carry2) {
 7396   //   jlong carry, x[], y[], z[];
 7397   //   int kdx = ystart+1;
 7398   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7399   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 7400   //     z[kdx+idx+1] = (jlong)product;
 7401   //     jlong carry2  = (jlong)(product >>> 64);
 7402   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 7403   //     z[kdx+idx] = (jlong)product;
 7404   //     carry  = (jlong)(product >>> 64);
 7405   //   }
 7406   //   idx += 2;
 7407   //   if (idx > 0) {
 7408   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 7409   //     z[kdx+idx] = (jlong)product;
 7410   //     carry  = (jlong)(product >>> 64);
 7411   //   }
 7412   //
 7413 
 7414   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7415 
 7416   movl(jdx, idx);
 7417   andl(jdx, 0xFFFFFFFC);
 7418   shrl(jdx, 2);
 7419 
 7420   bind(L_third_loop);
 7421   subl(jdx, 1);
 7422   jcc(Assembler::negative, L_third_loop_exit);
 7423   subl(idx, 4);
 7424 
 7425   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 7426   movq(carry2, rdx);
 7427 
 7428   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 7429   movq(carry, rdx);
 7430   jmp(L_third_loop);
 7431 
 7432   bind (L_third_loop_exit);
 7433 
 7434   andl (idx, 0x3);
 7435   jcc(Assembler::zero, L_post_third_loop_done);
 7436 
 7437   Label L_check_1;
 7438   subl(idx, 2);
 7439   jcc(Assembler::negative, L_check_1);
 7440 
 7441   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 7442   movq(carry, rdx);
 7443 
 7444   bind (L_check_1);
 7445   addl (idx, 0x2);
 7446   andl (idx, 0x1);
 7447   subl(idx, 1);
 7448   jcc(Assembler::negative, L_post_third_loop_done);
 7449 
 7450   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 7451   movq(product, x_xstart);
 7452   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 7453   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 7454 
 7455   add2_with_carry(rdx, product, yz_idx, carry);
 7456 
 7457   movl(Address(z, idx, Address::times_4,  0), product);
 7458   shrq(product, 32);
 7459 
 7460   shlq(rdx, 32);
 7461   orq(product, rdx);
 7462   movq(carry, product);
 7463 
 7464   bind(L_post_third_loop_done);
 7465 }
 7466 
 7467 /**
 7468  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 7469  *
 7470  */
 7471 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 7472                                                   Register carry, Register carry2,
 7473                                                   Register idx, Register jdx,
 7474                                                   Register yz_idx1, Register yz_idx2,
 7475                                                   Register tmp, Register tmp3, Register tmp4) {
 7476   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 7477 
 7478   //   jlong carry, x[], y[], z[];
 7479   //   int kdx = ystart+1;
 7480   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7481   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 7482   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 7483   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 7484   //     carry  = (jlong)(tmp4 >>> 64);
 7485   //     z[kdx+idx+1] = (jlong)tmp3;
 7486   //     z[kdx+idx] = (jlong)tmp4;
 7487   //   }
 7488   //   idx += 2;
 7489   //   if (idx > 0) {
 7490   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 7491   //     z[kdx+idx] = (jlong)yz_idx1;
 7492   //     carry  = (jlong)(yz_idx1 >>> 64);
 7493   //   }
 7494   //
 7495 
 7496   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7497 
 7498   movl(jdx, idx);
 7499   andl(jdx, 0xFFFFFFFC);
 7500   shrl(jdx, 2);
 7501 
 7502   bind(L_third_loop);
 7503   subl(jdx, 1);
 7504   jcc(Assembler::negative, L_third_loop_exit);
 7505   subl(idx, 4);
 7506 
 7507   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 7508   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 7509   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 7510   rorxq(yz_idx2, yz_idx2, 32);
 7511 
 7512   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 7513   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 7514 
 7515   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 7516   rorxq(yz_idx1, yz_idx1, 32);
 7517   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7518   rorxq(yz_idx2, yz_idx2, 32);
 7519 
 7520   if (VM_Version::supports_adx()) {
 7521     adcxq(tmp3, carry);
 7522     adoxq(tmp3, yz_idx1);
 7523 
 7524     adcxq(tmp4, tmp);
 7525     adoxq(tmp4, yz_idx2);
 7526 
 7527     movl(carry, 0); // does not affect flags
 7528     adcxq(carry2, carry);
 7529     adoxq(carry2, carry);
 7530   } else {
 7531     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 7532     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 7533   }
 7534   movq(carry, carry2);
 7535 
 7536   movl(Address(z, idx, Address::times_4, 12), tmp3);
 7537   shrq(tmp3, 32);
 7538   movl(Address(z, idx, Address::times_4,  8), tmp3);
 7539 
 7540   movl(Address(z, idx, Address::times_4,  4), tmp4);
 7541   shrq(tmp4, 32);
 7542   movl(Address(z, idx, Address::times_4,  0), tmp4);
 7543 
 7544   jmp(L_third_loop);
 7545 
 7546   bind (L_third_loop_exit);
 7547 
 7548   andl (idx, 0x3);
 7549   jcc(Assembler::zero, L_post_third_loop_done);
 7550 
 7551   Label L_check_1;
 7552   subl(idx, 2);
 7553   jcc(Assembler::negative, L_check_1);
 7554 
 7555   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 7556   rorxq(yz_idx1, yz_idx1, 32);
 7557   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 7558   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7559   rorxq(yz_idx2, yz_idx2, 32);
 7560 
 7561   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 7562 
 7563   movl(Address(z, idx, Address::times_4,  4), tmp3);
 7564   shrq(tmp3, 32);
 7565   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7566   movq(carry, tmp4);
 7567 
 7568   bind (L_check_1);
 7569   addl (idx, 0x2);
 7570   andl (idx, 0x1);
 7571   subl(idx, 1);
 7572   jcc(Assembler::negative, L_post_third_loop_done);
 7573   movl(tmp4, Address(y, idx, Address::times_4,  0));
 7574   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 7575   movl(tmp4, Address(z, idx, Address::times_4,  0));
 7576 
 7577   add2_with_carry(carry2, tmp3, tmp4, carry);
 7578 
 7579   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7580   shrq(tmp3, 32);
 7581 
 7582   shlq(carry2, 32);
 7583   orq(tmp3, carry2);
 7584   movq(carry, tmp3);
 7585 
 7586   bind(L_post_third_loop_done);
 7587 }
 7588 
 7589 /**
 7590  * Code for BigInteger::multiplyToLen() intrinsic.
 7591  *
 7592  * rdi: x
 7593  * rax: xlen
 7594  * rsi: y
 7595  * rcx: ylen
 7596  * r8:  z
 7597  * r11: tmp0
 7598  * r12: tmp1
 7599  * r13: tmp2
 7600  * r14: tmp3
 7601  * r15: tmp4
 7602  * rbx: tmp5
 7603  *
 7604  */
 7605 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 7606                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 7607   ShortBranchVerifier sbv(this);
 7608   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 7609 
 7610   push(tmp0);
 7611   push(tmp1);
 7612   push(tmp2);
 7613   push(tmp3);
 7614   push(tmp4);
 7615   push(tmp5);
 7616 
 7617   push(xlen);
 7618 
 7619   const Register idx = tmp1;
 7620   const Register kdx = tmp2;
 7621   const Register xstart = tmp3;
 7622 
 7623   const Register y_idx = tmp4;
 7624   const Register carry = tmp5;
 7625   const Register product  = xlen;
 7626   const Register x_xstart = tmp0;
 7627 
 7628   // First Loop.
 7629   //
 7630   //  final static long LONG_MASK = 0xffffffffL;
 7631   //  int xstart = xlen - 1;
 7632   //  int ystart = ylen - 1;
 7633   //  long carry = 0;
 7634   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7635   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 7636   //    z[kdx] = (int)product;
 7637   //    carry = product >>> 32;
 7638   //  }
 7639   //  z[xstart] = (int)carry;
 7640   //
 7641 
 7642   movl(idx, ylen);               // idx = ylen;
 7643   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 7644   xorq(carry, carry);            // carry = 0;
 7645 
 7646   Label L_done;
 7647 
 7648   movl(xstart, xlen);
 7649   decrementl(xstart);
 7650   jcc(Assembler::negative, L_done);
 7651 
 7652   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 7653 
 7654   Label L_second_loop;
 7655   testl(kdx, kdx);
 7656   jcc(Assembler::zero, L_second_loop);
 7657 
 7658   Label L_carry;
 7659   subl(kdx, 1);
 7660   jcc(Assembler::zero, L_carry);
 7661 
 7662   movl(Address(z, kdx, Address::times_4,  0), carry);
 7663   shrq(carry, 32);
 7664   subl(kdx, 1);
 7665 
 7666   bind(L_carry);
 7667   movl(Address(z, kdx, Address::times_4,  0), carry);
 7668 
 7669   // Second and third (nested) loops.
 7670   //
 7671   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 7672   //   carry = 0;
 7673   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 7674   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 7675   //                    (z[k] & LONG_MASK) + carry;
 7676   //     z[k] = (int)product;
 7677   //     carry = product >>> 32;
 7678   //   }
 7679   //   z[i] = (int)carry;
 7680   // }
 7681   //
 7682   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 7683 
 7684   const Register jdx = tmp1;
 7685 
 7686   bind(L_second_loop);
 7687   xorl(carry, carry);    // carry = 0;
 7688   movl(jdx, ylen);       // j = ystart+1
 7689 
 7690   subl(xstart, 1);       // i = xstart-1;
 7691   jcc(Assembler::negative, L_done);
 7692 
 7693   push (z);
 7694 
 7695   Label L_last_x;
 7696   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 7697   subl(xstart, 1);       // i = xstart-1;
 7698   jcc(Assembler::negative, L_last_x);
 7699 
 7700   if (UseBMI2Instructions) {
 7701     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 7702     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 7703   } else {
 7704     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7705     rorq(x_xstart, 32);  // convert big-endian to little-endian
 7706   }
 7707 
 7708   Label L_third_loop_prologue;
 7709   bind(L_third_loop_prologue);
 7710 
 7711   push (x);
 7712   push (xstart);
 7713   push (ylen);
 7714 
 7715 
 7716   if (UseBMI2Instructions) {
 7717     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 7718   } else { // !UseBMI2Instructions
 7719     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 7720   }
 7721 
 7722   pop(ylen);
 7723   pop(xlen);
 7724   pop(x);
 7725   pop(z);
 7726 
 7727   movl(tmp3, xlen);
 7728   addl(tmp3, 1);
 7729   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7730   subl(tmp3, 1);
 7731   jccb(Assembler::negative, L_done);
 7732 
 7733   shrq(carry, 32);
 7734   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7735   jmp(L_second_loop);
 7736 
 7737   // Next infrequent code is moved outside loops.
 7738   bind(L_last_x);
 7739   if (UseBMI2Instructions) {
 7740     movl(rdx, Address(x,  0));
 7741   } else {
 7742     movl(x_xstart, Address(x,  0));
 7743   }
 7744   jmp(L_third_loop_prologue);
 7745 
 7746   bind(L_done);
 7747 
 7748   pop(xlen);
 7749 
 7750   pop(tmp5);
 7751   pop(tmp4);
 7752   pop(tmp3);
 7753   pop(tmp2);
 7754   pop(tmp1);
 7755   pop(tmp0);
 7756 }
 7757 
 7758 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 7759   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 7760   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 7761   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 7762   Label VECTOR8_TAIL, VECTOR4_TAIL;
 7763   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 7764   Label SAME_TILL_END, DONE;
 7765   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 7766 
 7767   //scale is in rcx in both Win64 and Unix
 7768   ShortBranchVerifier sbv(this);
 7769 
 7770   shlq(length);
 7771   xorq(result, result);
 7772 
 7773   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 7774       VM_Version::supports_avx512vlbw()) {
 7775     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 7776 
 7777     cmpq(length, 64);
 7778     jcc(Assembler::less, VECTOR32_TAIL);
 7779 
 7780     movq(tmp1, length);
 7781     andq(tmp1, 0x3F);      // tail count
 7782     andq(length, ~(0x3F)); //vector count
 7783 
 7784     bind(VECTOR64_LOOP);
 7785     // AVX512 code to compare 64 byte vectors.
 7786     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 7787     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7788     kortestql(k7, k7);
 7789     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 7790     addq(result, 64);
 7791     subq(length, 64);
 7792     jccb(Assembler::notZero, VECTOR64_LOOP);
 7793 
 7794     //bind(VECTOR64_TAIL);
 7795     testq(tmp1, tmp1);
 7796     jcc(Assembler::zero, SAME_TILL_END);
 7797 
 7798     //bind(VECTOR64_TAIL);
 7799     // AVX512 code to compare up to 63 byte vectors.
 7800     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 7801     shlxq(tmp2, tmp2, tmp1);
 7802     notq(tmp2);
 7803     kmovql(k3, tmp2);
 7804 
 7805     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 7806     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7807 
 7808     ktestql(k7, k3);
 7809     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 7810 
 7811     bind(VECTOR64_NOT_EQUAL);
 7812     kmovql(tmp1, k7);
 7813     notq(tmp1);
 7814     tzcntq(tmp1, tmp1);
 7815     addq(result, tmp1);
 7816     shrq(result);
 7817     jmp(DONE);
 7818     bind(VECTOR32_TAIL);
 7819   }
 7820 
 7821   cmpq(length, 8);
 7822   jcc(Assembler::equal, VECTOR8_LOOP);
 7823   jcc(Assembler::less, VECTOR4_TAIL);
 7824 
 7825   if (UseAVX >= 2) {
 7826     Label VECTOR16_TAIL, VECTOR32_LOOP;
 7827 
 7828     cmpq(length, 16);
 7829     jcc(Assembler::equal, VECTOR16_LOOP);
 7830     jcc(Assembler::less, VECTOR8_LOOP);
 7831 
 7832     cmpq(length, 32);
 7833     jccb(Assembler::less, VECTOR16_TAIL);
 7834 
 7835     subq(length, 32);
 7836     bind(VECTOR32_LOOP);
 7837     vmovdqu(rymm0, Address(obja, result));
 7838     vmovdqu(rymm1, Address(objb, result));
 7839     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 7840     vptest(rymm2, rymm2);
 7841     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7842     addq(result, 32);
 7843     subq(length, 32);
 7844     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7845     addq(length, 32);
 7846     jcc(Assembler::equal, SAME_TILL_END);
 7847     //falling through if less than 32 bytes left //close the branch here.
 7848 
 7849     bind(VECTOR16_TAIL);
 7850     cmpq(length, 16);
 7851     jccb(Assembler::less, VECTOR8_TAIL);
 7852     bind(VECTOR16_LOOP);
 7853     movdqu(rymm0, Address(obja, result));
 7854     movdqu(rymm1, Address(objb, result));
 7855     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7856     ptest(rymm2, rymm2);
 7857     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7858     addq(result, 16);
 7859     subq(length, 16);
 7860     jcc(Assembler::equal, SAME_TILL_END);
 7861     //falling through if less than 16 bytes left
 7862   } else {//regular intrinsics
 7863 
 7864     cmpq(length, 16);
 7865     jccb(Assembler::less, VECTOR8_TAIL);
 7866 
 7867     subq(length, 16);
 7868     bind(VECTOR16_LOOP);
 7869     movdqu(rymm0, Address(obja, result));
 7870     movdqu(rymm1, Address(objb, result));
 7871     pxor(rymm0, rymm1);
 7872     ptest(rymm0, rymm0);
 7873     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7874     addq(result, 16);
 7875     subq(length, 16);
 7876     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7877     addq(length, 16);
 7878     jcc(Assembler::equal, SAME_TILL_END);
 7879     //falling through if less than 16 bytes left
 7880   }
 7881 
 7882   bind(VECTOR8_TAIL);
 7883   cmpq(length, 8);
 7884   jccb(Assembler::less, VECTOR4_TAIL);
 7885   bind(VECTOR8_LOOP);
 7886   movq(tmp1, Address(obja, result));
 7887   movq(tmp2, Address(objb, result));
 7888   xorq(tmp1, tmp2);
 7889   testq(tmp1, tmp1);
 7890   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7891   addq(result, 8);
 7892   subq(length, 8);
 7893   jcc(Assembler::equal, SAME_TILL_END);
 7894   //falling through if less than 8 bytes left
 7895 
 7896   bind(VECTOR4_TAIL);
 7897   cmpq(length, 4);
 7898   jccb(Assembler::less, BYTES_TAIL);
 7899   bind(VECTOR4_LOOP);
 7900   movl(tmp1, Address(obja, result));
 7901   xorl(tmp1, Address(objb, result));
 7902   testl(tmp1, tmp1);
 7903   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7904   addq(result, 4);
 7905   subq(length, 4);
 7906   jcc(Assembler::equal, SAME_TILL_END);
 7907   //falling through if less than 4 bytes left
 7908 
 7909   bind(BYTES_TAIL);
 7910   bind(BYTES_LOOP);
 7911   load_unsigned_byte(tmp1, Address(obja, result));
 7912   load_unsigned_byte(tmp2, Address(objb, result));
 7913   xorl(tmp1, tmp2);
 7914   testl(tmp1, tmp1);
 7915   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7916   decq(length);
 7917   jcc(Assembler::zero, SAME_TILL_END);
 7918   incq(result);
 7919   load_unsigned_byte(tmp1, Address(obja, result));
 7920   load_unsigned_byte(tmp2, Address(objb, result));
 7921   xorl(tmp1, tmp2);
 7922   testl(tmp1, tmp1);
 7923   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7924   decq(length);
 7925   jcc(Assembler::zero, SAME_TILL_END);
 7926   incq(result);
 7927   load_unsigned_byte(tmp1, Address(obja, result));
 7928   load_unsigned_byte(tmp2, Address(objb, result));
 7929   xorl(tmp1, tmp2);
 7930   testl(tmp1, tmp1);
 7931   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7932   jmp(SAME_TILL_END);
 7933 
 7934   if (UseAVX >= 2) {
 7935     bind(VECTOR32_NOT_EQUAL);
 7936     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7937     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7938     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7939     vpmovmskb(tmp1, rymm0);
 7940     bsfq(tmp1, tmp1);
 7941     addq(result, tmp1);
 7942     shrq(result);
 7943     jmp(DONE);
 7944   }
 7945 
 7946   bind(VECTOR16_NOT_EQUAL);
 7947   if (UseAVX >= 2) {
 7948     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7949     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7950     pxor(rymm0, rymm2);
 7951   } else {
 7952     pcmpeqb(rymm2, rymm2);
 7953     pxor(rymm0, rymm1);
 7954     pcmpeqb(rymm0, rymm1);
 7955     pxor(rymm0, rymm2);
 7956   }
 7957   pmovmskb(tmp1, rymm0);
 7958   bsfq(tmp1, tmp1);
 7959   addq(result, tmp1);
 7960   shrq(result);
 7961   jmpb(DONE);
 7962 
 7963   bind(VECTOR8_NOT_EQUAL);
 7964   bind(VECTOR4_NOT_EQUAL);
 7965   bsfq(tmp1, tmp1);
 7966   shrq(tmp1, 3);
 7967   addq(result, tmp1);
 7968   bind(BYTES_NOT_EQUAL);
 7969   shrq(result);
 7970   jmpb(DONE);
 7971 
 7972   bind(SAME_TILL_END);
 7973   mov64(result, -1);
 7974 
 7975   bind(DONE);
 7976 }
 7977 
 7978 //Helper functions for square_to_len()
 7979 
 7980 /**
 7981  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7982  * Preserves x and z and modifies rest of the registers.
 7983  */
 7984 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7985   // Perform square and right shift by 1
 7986   // Handle odd xlen case first, then for even xlen do the following
 7987   // jlong carry = 0;
 7988   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7989   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7990   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7991   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7992   //     carry = (jlong)product;
 7993   // }
 7994 
 7995   xorq(tmp5, tmp5);     // carry
 7996   xorq(rdxReg, rdxReg);
 7997   xorl(tmp1, tmp1);     // index for x
 7998   xorl(tmp4, tmp4);     // index for z
 7999 
 8000   Label L_first_loop, L_first_loop_exit;
 8001 
 8002   testl(xlen, 1);
 8003   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 8004 
 8005   // Square and right shift by 1 the odd element using 32 bit multiply
 8006   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 8007   imulq(raxReg, raxReg);
 8008   shrq(raxReg, 1);
 8009   adcq(tmp5, 0);
 8010   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 8011   incrementl(tmp1);
 8012   addl(tmp4, 2);
 8013 
 8014   // Square and  right shift by 1 the rest using 64 bit multiply
 8015   bind(L_first_loop);
 8016   cmpptr(tmp1, xlen);
 8017   jccb(Assembler::equal, L_first_loop_exit);
 8018 
 8019   // Square
 8020   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 8021   rorq(raxReg, 32);    // convert big-endian to little-endian
 8022   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 8023 
 8024   // Right shift by 1 and save carry
 8025   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 8026   rcrq(rdxReg, 1);
 8027   rcrq(raxReg, 1);
 8028   adcq(tmp5, 0);
 8029 
 8030   // Store result in z
 8031   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 8032   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 8033 
 8034   // Update indices for x and z
 8035   addl(tmp1, 2);
 8036   addl(tmp4, 4);
 8037   jmp(L_first_loop);
 8038 
 8039   bind(L_first_loop_exit);
 8040 }
 8041 
 8042 
 8043 /**
 8044  * Perform the following multiply add operation using BMI2 instructions
 8045  * carry:sum = sum + op1*op2 + carry
 8046  * op2 should be in rdx
 8047  * op2 is preserved, all other registers are modified
 8048  */
 8049 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 8050   // assert op2 is rdx
 8051   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 8052   addq(sum, carry);
 8053   adcq(tmp2, 0);
 8054   addq(sum, op1);
 8055   adcq(tmp2, 0);
 8056   movq(carry, tmp2);
 8057 }
 8058 
 8059 /**
 8060  * Perform the following multiply add operation:
 8061  * carry:sum = sum + op1*op2 + carry
 8062  * Preserves op1, op2 and modifies rest of registers
 8063  */
 8064 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 8065   // rdx:rax = op1 * op2
 8066   movq(raxReg, op2);
 8067   mulq(op1);
 8068 
 8069   //  rdx:rax = sum + carry + rdx:rax
 8070   addq(sum, carry);
 8071   adcq(rdxReg, 0);
 8072   addq(sum, raxReg);
 8073   adcq(rdxReg, 0);
 8074 
 8075   // carry:sum = rdx:sum
 8076   movq(carry, rdxReg);
 8077 }
 8078 
 8079 /**
 8080  * Add 64 bit long carry into z[] with carry propagation.
 8081  * Preserves z and carry register values and modifies rest of registers.
 8082  *
 8083  */
 8084 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 8085   Label L_fourth_loop, L_fourth_loop_exit;
 8086 
 8087   movl(tmp1, 1);
 8088   subl(zlen, 2);
 8089   addq(Address(z, zlen, Address::times_4, 0), carry);
 8090 
 8091   bind(L_fourth_loop);
 8092   jccb(Assembler::carryClear, L_fourth_loop_exit);
 8093   subl(zlen, 2);
 8094   jccb(Assembler::negative, L_fourth_loop_exit);
 8095   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 8096   jmp(L_fourth_loop);
 8097   bind(L_fourth_loop_exit);
 8098 }
 8099 
 8100 /**
 8101  * Shift z[] left by 1 bit.
 8102  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 8103  *
 8104  */
 8105 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 8106 
 8107   Label L_fifth_loop, L_fifth_loop_exit;
 8108 
 8109   // Fifth loop
 8110   // Perform primitiveLeftShift(z, zlen, 1)
 8111 
 8112   const Register prev_carry = tmp1;
 8113   const Register new_carry = tmp4;
 8114   const Register value = tmp2;
 8115   const Register zidx = tmp3;
 8116 
 8117   // int zidx, carry;
 8118   // long value;
 8119   // carry = 0;
 8120   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 8121   //    (carry:value)  = (z[i] << 1) | carry ;
 8122   //    z[i] = value;
 8123   // }
 8124 
 8125   movl(zidx, zlen);
 8126   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 8127 
 8128   bind(L_fifth_loop);
 8129   decl(zidx);  // Use decl to preserve carry flag
 8130   decl(zidx);
 8131   jccb(Assembler::negative, L_fifth_loop_exit);
 8132 
 8133   if (UseBMI2Instructions) {
 8134      movq(value, Address(z, zidx, Address::times_4, 0));
 8135      rclq(value, 1);
 8136      rorxq(value, value, 32);
 8137      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 8138   }
 8139   else {
 8140     // clear new_carry
 8141     xorl(new_carry, new_carry);
 8142 
 8143     // Shift z[i] by 1, or in previous carry and save new carry
 8144     movq(value, Address(z, zidx, Address::times_4, 0));
 8145     shlq(value, 1);
 8146     adcl(new_carry, 0);
 8147 
 8148     orq(value, prev_carry);
 8149     rorq(value, 0x20);
 8150     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 8151 
 8152     // Set previous carry = new carry
 8153     movl(prev_carry, new_carry);
 8154   }
 8155   jmp(L_fifth_loop);
 8156 
 8157   bind(L_fifth_loop_exit);
 8158 }
 8159 
 8160 
 8161 /**
 8162  * Code for BigInteger::squareToLen() intrinsic
 8163  *
 8164  * rdi: x
 8165  * rsi: len
 8166  * r8:  z
 8167  * rcx: zlen
 8168  * r12: tmp1
 8169  * r13: tmp2
 8170  * r14: tmp3
 8171  * r15: tmp4
 8172  * rbx: tmp5
 8173  *
 8174  */
 8175 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) {
 8176 
 8177   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 8178   push(tmp1);
 8179   push(tmp2);
 8180   push(tmp3);
 8181   push(tmp4);
 8182   push(tmp5);
 8183 
 8184   // First loop
 8185   // Store the squares, right shifted one bit (i.e., divided by 2).
 8186   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8187 
 8188   // Add in off-diagonal sums.
 8189   //
 8190   // Second, third (nested) and fourth loops.
 8191   // zlen +=2;
 8192   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 8193   //    carry = 0;
 8194   //    long op2 = x[xidx:xidx+1];
 8195   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 8196   //       k -= 2;
 8197   //       long op1 = x[j:j+1];
 8198   //       long sum = z[k:k+1];
 8199   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 8200   //       z[k:k+1] = sum;
 8201   //    }
 8202   //    add_one_64(z, k, carry, tmp_regs);
 8203   // }
 8204 
 8205   const Register carry = tmp5;
 8206   const Register sum = tmp3;
 8207   const Register op1 = tmp4;
 8208   Register op2 = tmp2;
 8209 
 8210   push(zlen);
 8211   push(len);
 8212   addl(zlen,2);
 8213   bind(L_second_loop);
 8214   xorq(carry, carry);
 8215   subl(zlen, 4);
 8216   subl(len, 2);
 8217   push(zlen);
 8218   push(len);
 8219   cmpl(len, 0);
 8220   jccb(Assembler::lessEqual, L_second_loop_exit);
 8221 
 8222   // Multiply an array by one 64 bit long.
 8223   if (UseBMI2Instructions) {
 8224     op2 = rdxReg;
 8225     movq(op2, Address(x, len, Address::times_4,  0));
 8226     rorxq(op2, op2, 32);
 8227   }
 8228   else {
 8229     movq(op2, Address(x, len, Address::times_4,  0));
 8230     rorq(op2, 32);
 8231   }
 8232 
 8233   bind(L_third_loop);
 8234   decrementl(len);
 8235   jccb(Assembler::negative, L_third_loop_exit);
 8236   decrementl(len);
 8237   jccb(Assembler::negative, L_last_x);
 8238 
 8239   movq(op1, Address(x, len, Address::times_4,  0));
 8240   rorq(op1, 32);
 8241 
 8242   bind(L_multiply);
 8243   subl(zlen, 2);
 8244   movq(sum, Address(z, zlen, Address::times_4,  0));
 8245 
 8246   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 8247   if (UseBMI2Instructions) {
 8248     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 8249   }
 8250   else {
 8251     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8252   }
 8253 
 8254   movq(Address(z, zlen, Address::times_4, 0), sum);
 8255 
 8256   jmp(L_third_loop);
 8257   bind(L_third_loop_exit);
 8258 
 8259   // Fourth loop
 8260   // Add 64 bit long carry into z with carry propagation.
 8261   // Uses offsetted zlen.
 8262   add_one_64(z, zlen, carry, tmp1);
 8263 
 8264   pop(len);
 8265   pop(zlen);
 8266   jmp(L_second_loop);
 8267 
 8268   // Next infrequent code is moved outside loops.
 8269   bind(L_last_x);
 8270   movl(op1, Address(x, 0));
 8271   jmp(L_multiply);
 8272 
 8273   bind(L_second_loop_exit);
 8274   pop(len);
 8275   pop(zlen);
 8276   pop(len);
 8277   pop(zlen);
 8278 
 8279   // Fifth loop
 8280   // Shift z left 1 bit.
 8281   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 8282 
 8283   // z[zlen-1] |= x[len-1] & 1;
 8284   movl(tmp3, Address(x, len, Address::times_4, -4));
 8285   andl(tmp3, 1);
 8286   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 8287 
 8288   pop(tmp5);
 8289   pop(tmp4);
 8290   pop(tmp3);
 8291   pop(tmp2);
 8292   pop(tmp1);
 8293 }
 8294 
 8295 /**
 8296  * Helper function for mul_add()
 8297  * Multiply the in[] by int k and add to out[] starting at offset offs using
 8298  * 128 bit by 32 bit multiply and return the carry in tmp5.
 8299  * Only quad int aligned length of in[] is operated on in this function.
 8300  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 8301  * This function preserves out, in and k registers.
 8302  * len and offset point to the appropriate index in "in" & "out" correspondingly
 8303  * tmp5 has the carry.
 8304  * other registers are temporary and are modified.
 8305  *
 8306  */
 8307 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 8308   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 8309   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8310 
 8311   Label L_first_loop, L_first_loop_exit;
 8312 
 8313   movl(tmp1, len);
 8314   shrl(tmp1, 2);
 8315 
 8316   bind(L_first_loop);
 8317   subl(tmp1, 1);
 8318   jccb(Assembler::negative, L_first_loop_exit);
 8319 
 8320   subl(len, 4);
 8321   subl(offset, 4);
 8322 
 8323   Register op2 = tmp2;
 8324   const Register sum = tmp3;
 8325   const Register op1 = tmp4;
 8326   const Register carry = tmp5;
 8327 
 8328   if (UseBMI2Instructions) {
 8329     op2 = rdxReg;
 8330   }
 8331 
 8332   movq(op1, Address(in, len, Address::times_4,  8));
 8333   rorq(op1, 32);
 8334   movq(sum, Address(out, offset, Address::times_4,  8));
 8335   rorq(sum, 32);
 8336   if (UseBMI2Instructions) {
 8337     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8338   }
 8339   else {
 8340     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8341   }
 8342   // Store back in big endian from little endian
 8343   rorq(sum, 0x20);
 8344   movq(Address(out, offset, Address::times_4,  8), sum);
 8345 
 8346   movq(op1, Address(in, len, Address::times_4,  0));
 8347   rorq(op1, 32);
 8348   movq(sum, Address(out, offset, Address::times_4,  0));
 8349   rorq(sum, 32);
 8350   if (UseBMI2Instructions) {
 8351     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8352   }
 8353   else {
 8354     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8355   }
 8356   // Store back in big endian from little endian
 8357   rorq(sum, 0x20);
 8358   movq(Address(out, offset, Address::times_4,  0), sum);
 8359 
 8360   jmp(L_first_loop);
 8361   bind(L_first_loop_exit);
 8362 }
 8363 
 8364 /**
 8365  * Code for BigInteger::mulAdd() intrinsic
 8366  *
 8367  * rdi: out
 8368  * rsi: in
 8369  * r11: offs (out.length - offset)
 8370  * rcx: len
 8371  * r8:  k
 8372  * r12: tmp1
 8373  * r13: tmp2
 8374  * r14: tmp3
 8375  * r15: tmp4
 8376  * rbx: tmp5
 8377  * Multiply the in[] by word k and add to out[], return the carry in rax
 8378  */
 8379 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 8380    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 8381    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8382 
 8383   Label L_carry, L_last_in, L_done;
 8384 
 8385 // carry = 0;
 8386 // for (int j=len-1; j >= 0; j--) {
 8387 //    long product = (in[j] & LONG_MASK) * kLong +
 8388 //                   (out[offs] & LONG_MASK) + carry;
 8389 //    out[offs--] = (int)product;
 8390 //    carry = product >>> 32;
 8391 // }
 8392 //
 8393   push(tmp1);
 8394   push(tmp2);
 8395   push(tmp3);
 8396   push(tmp4);
 8397   push(tmp5);
 8398 
 8399   Register op2 = tmp2;
 8400   const Register sum = tmp3;
 8401   const Register op1 = tmp4;
 8402   const Register carry =  tmp5;
 8403 
 8404   if (UseBMI2Instructions) {
 8405     op2 = rdxReg;
 8406     movl(op2, k);
 8407   }
 8408   else {
 8409     movl(op2, k);
 8410   }
 8411 
 8412   xorq(carry, carry);
 8413 
 8414   //First loop
 8415 
 8416   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 8417   //The carry is in tmp5
 8418   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8419 
 8420   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 8421   decrementl(len);
 8422   jccb(Assembler::negative, L_carry);
 8423   decrementl(len);
 8424   jccb(Assembler::negative, L_last_in);
 8425 
 8426   movq(op1, Address(in, len, Address::times_4,  0));
 8427   rorq(op1, 32);
 8428 
 8429   subl(offs, 2);
 8430   movq(sum, Address(out, offs, Address::times_4,  0));
 8431   rorq(sum, 32);
 8432 
 8433   if (UseBMI2Instructions) {
 8434     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8435   }
 8436   else {
 8437     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8438   }
 8439 
 8440   // Store back in big endian from little endian
 8441   rorq(sum, 0x20);
 8442   movq(Address(out, offs, Address::times_4,  0), sum);
 8443 
 8444   testl(len, len);
 8445   jccb(Assembler::zero, L_carry);
 8446 
 8447   //Multiply the last in[] entry, if any
 8448   bind(L_last_in);
 8449   movl(op1, Address(in, 0));
 8450   movl(sum, Address(out, offs, Address::times_4,  -4));
 8451 
 8452   movl(raxReg, k);
 8453   mull(op1); //tmp4 * eax -> edx:eax
 8454   addl(sum, carry);
 8455   adcl(rdxReg, 0);
 8456   addl(sum, raxReg);
 8457   adcl(rdxReg, 0);
 8458   movl(carry, rdxReg);
 8459 
 8460   movl(Address(out, offs, Address::times_4,  -4), sum);
 8461 
 8462   bind(L_carry);
 8463   //return tmp5/carry as carry in rax
 8464   movl(rax, carry);
 8465 
 8466   bind(L_done);
 8467   pop(tmp5);
 8468   pop(tmp4);
 8469   pop(tmp3);
 8470   pop(tmp2);
 8471   pop(tmp1);
 8472 }
 8473 
 8474 /**
 8475  * Emits code to update CRC-32 with a byte value according to constants in table
 8476  *
 8477  * @param [in,out]crc   Register containing the crc.
 8478  * @param [in]val       Register containing the byte to fold into the CRC.
 8479  * @param [in]table     Register containing the table of crc constants.
 8480  *
 8481  * uint32_t crc;
 8482  * val = crc_table[(val ^ crc) & 0xFF];
 8483  * crc = val ^ (crc >> 8);
 8484  *
 8485  */
 8486 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 8487   xorl(val, crc);
 8488   andl(val, 0xFF);
 8489   shrl(crc, 8); // unsigned shift
 8490   xorl(crc, Address(table, val, Address::times_4, 0));
 8491 }
 8492 
 8493 /**
 8494  * Fold 128-bit data chunk
 8495  */
 8496 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 8497   if (UseAVX > 0) {
 8498     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 8499     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 8500     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 8501     pxor(xcrc, xtmp);
 8502   } else {
 8503     movdqa(xtmp, xcrc);
 8504     pclmulhdq(xtmp, xK);   // [123:64]
 8505     pclmulldq(xcrc, xK);   // [63:0]
 8506     pxor(xcrc, xtmp);
 8507     movdqu(xtmp, Address(buf, offset));
 8508     pxor(xcrc, xtmp);
 8509   }
 8510 }
 8511 
 8512 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 8513   if (UseAVX > 0) {
 8514     vpclmulhdq(xtmp, xK, xcrc);
 8515     vpclmulldq(xcrc, xK, xcrc);
 8516     pxor(xcrc, xbuf);
 8517     pxor(xcrc, xtmp);
 8518   } else {
 8519     movdqa(xtmp, xcrc);
 8520     pclmulhdq(xtmp, xK);
 8521     pclmulldq(xcrc, xK);
 8522     pxor(xcrc, xbuf);
 8523     pxor(xcrc, xtmp);
 8524   }
 8525 }
 8526 
 8527 /**
 8528  * 8-bit folds to compute 32-bit CRC
 8529  *
 8530  * uint64_t xcrc;
 8531  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 8532  */
 8533 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 8534   movdl(tmp, xcrc);
 8535   andl(tmp, 0xFF);
 8536   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 8537   psrldq(xcrc, 1); // unsigned shift one byte
 8538   pxor(xcrc, xtmp);
 8539 }
 8540 
 8541 /**
 8542  * uint32_t crc;
 8543  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 8544  */
 8545 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 8546   movl(tmp, crc);
 8547   andl(tmp, 0xFF);
 8548   shrl(crc, 8);
 8549   xorl(crc, Address(table, tmp, Address::times_4, 0));
 8550 }
 8551 
 8552 /**
 8553  * @param crc   register containing existing CRC (32-bit)
 8554  * @param buf   register pointing to input byte buffer (byte*)
 8555  * @param len   register containing number of bytes
 8556  * @param table register that will contain address of CRC table
 8557  * @param tmp   scratch register
 8558  */
 8559 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 8560   assert_different_registers(crc, buf, len, table, tmp, rax);
 8561 
 8562   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8563   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8564 
 8565   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8566   // context for the registers used, where all instructions below are using 128-bit mode
 8567   // On EVEX without VL and BW, these instructions will all be AVX.
 8568   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 8569   notl(crc); // ~crc
 8570   cmpl(len, 16);
 8571   jcc(Assembler::less, L_tail);
 8572 
 8573   // Align buffer to 16 bytes
 8574   movl(tmp, buf);
 8575   andl(tmp, 0xF);
 8576   jccb(Assembler::zero, L_aligned);
 8577   subl(tmp,  16);
 8578   addl(len, tmp);
 8579 
 8580   align(4);
 8581   BIND(L_align_loop);
 8582   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8583   update_byte_crc32(crc, rax, table);
 8584   increment(buf);
 8585   incrementl(tmp);
 8586   jccb(Assembler::less, L_align_loop);
 8587 
 8588   BIND(L_aligned);
 8589   movl(tmp, len); // save
 8590   shrl(len, 4);
 8591   jcc(Assembler::zero, L_tail_restore);
 8592 
 8593   // Fold crc into first bytes of vector
 8594   movdqa(xmm1, Address(buf, 0));
 8595   movdl(rax, xmm1);
 8596   xorl(crc, rax);
 8597   if (VM_Version::supports_sse4_1()) {
 8598     pinsrd(xmm1, crc, 0);
 8599   } else {
 8600     pinsrw(xmm1, crc, 0);
 8601     shrl(crc, 16);
 8602     pinsrw(xmm1, crc, 1);
 8603   }
 8604   addptr(buf, 16);
 8605   subl(len, 4); // len > 0
 8606   jcc(Assembler::less, L_fold_tail);
 8607 
 8608   movdqa(xmm2, Address(buf,  0));
 8609   movdqa(xmm3, Address(buf, 16));
 8610   movdqa(xmm4, Address(buf, 32));
 8611   addptr(buf, 48);
 8612   subl(len, 3);
 8613   jcc(Assembler::lessEqual, L_fold_512b);
 8614 
 8615   // Fold total 512 bits of polynomial on each iteration,
 8616   // 128 bits per each of 4 parallel streams.
 8617   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 8618 
 8619   align32();
 8620   BIND(L_fold_512b_loop);
 8621   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8622   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 8623   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 8624   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 8625   addptr(buf, 64);
 8626   subl(len, 4);
 8627   jcc(Assembler::greater, L_fold_512b_loop);
 8628 
 8629   // Fold 512 bits to 128 bits.
 8630   BIND(L_fold_512b);
 8631   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8632   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 8633   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 8634   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 8635 
 8636   // Fold the rest of 128 bits data chunks
 8637   BIND(L_fold_tail);
 8638   addl(len, 3);
 8639   jccb(Assembler::lessEqual, L_fold_128b);
 8640   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8641 
 8642   BIND(L_fold_tail_loop);
 8643   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8644   addptr(buf, 16);
 8645   decrementl(len);
 8646   jccb(Assembler::greater, L_fold_tail_loop);
 8647 
 8648   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 8649   BIND(L_fold_128b);
 8650   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 8651   if (UseAVX > 0) {
 8652     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 8653     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 8654     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 8655   } else {
 8656     movdqa(xmm2, xmm0);
 8657     pclmulqdq(xmm2, xmm1, 0x1);
 8658     movdqa(xmm3, xmm0);
 8659     pand(xmm3, xmm2);
 8660     pclmulqdq(xmm0, xmm3, 0x1);
 8661   }
 8662   psrldq(xmm1, 8);
 8663   psrldq(xmm2, 4);
 8664   pxor(xmm0, xmm1);
 8665   pxor(xmm0, xmm2);
 8666 
 8667   // 8 8-bit folds to compute 32-bit CRC.
 8668   for (int j = 0; j < 4; j++) {
 8669     fold_8bit_crc32(xmm0, table, xmm1, rax);
 8670   }
 8671   movdl(crc, xmm0); // mov 32 bits to general register
 8672   for (int j = 0; j < 4; j++) {
 8673     fold_8bit_crc32(crc, table, rax);
 8674   }
 8675 
 8676   BIND(L_tail_restore);
 8677   movl(len, tmp); // restore
 8678   BIND(L_tail);
 8679   andl(len, 0xf);
 8680   jccb(Assembler::zero, L_exit);
 8681 
 8682   // Fold the rest of bytes
 8683   align(4);
 8684   BIND(L_tail_loop);
 8685   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8686   update_byte_crc32(crc, rax, table);
 8687   increment(buf);
 8688   decrementl(len);
 8689   jccb(Assembler::greater, L_tail_loop);
 8690 
 8691   BIND(L_exit);
 8692   notl(crc); // ~c
 8693 }
 8694 
 8695 // Helper function for AVX 512 CRC32
 8696 // Fold 512-bit data chunks
 8697 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 8698                                              Register pos, int offset) {
 8699   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 8700   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 8701   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 8702   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 8703   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 8704 }
 8705 
 8706 // Helper function for AVX 512 CRC32
 8707 // Compute CRC32 for < 256B buffers
 8708 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 8709                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 8710                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 8711 
 8712   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 8713   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 8714   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 8715 
 8716   // check if there is enough buffer to be able to fold 16B at a time
 8717   cmpl(len, 32);
 8718   jcc(Assembler::less, L_less_than_32);
 8719 
 8720   // if there is, load the constants
 8721   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 8722   movdl(xmm0, crc);                        // get the initial crc value
 8723   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8724   pxor(xmm7, xmm0);
 8725 
 8726   // update the buffer pointer
 8727   addl(pos, 16);
 8728   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 8729   subl(len, 32);
 8730   jmp(L_16B_reduction_loop);
 8731 
 8732   bind(L_less_than_32);
 8733   //mov initial crc to the return value. this is necessary for zero - length buffers.
 8734   movl(rax, crc);
 8735   testl(len, len);
 8736   jcc(Assembler::equal, L_cleanup);
 8737 
 8738   movdl(xmm0, crc);                        //get the initial crc value
 8739 
 8740   cmpl(len, 16);
 8741   jcc(Assembler::equal, L_exact_16_left);
 8742   jcc(Assembler::less, L_less_than_16_left);
 8743 
 8744   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8745   pxor(xmm7, xmm0);                       //xor the initial crc value
 8746   addl(pos, 16);
 8747   subl(len, 16);
 8748   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 8749   jmp(L_get_last_two_xmms);
 8750 
 8751   bind(L_less_than_16_left);
 8752   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 8753   pxor(xmm1, xmm1);
 8754   movptr(tmp1, rsp);
 8755   movdqu(Address(tmp1, 0 * 16), xmm1);
 8756 
 8757   cmpl(len, 4);
 8758   jcc(Assembler::less, L_only_less_than_4);
 8759 
 8760   //backup the counter value
 8761   movl(tmp2, len);
 8762   cmpl(len, 8);
 8763   jcc(Assembler::less, L_less_than_8_left);
 8764 
 8765   //load 8 Bytes
 8766   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 8767   movq(Address(tmp1, 0 * 16), rax);
 8768   addptr(tmp1, 8);
 8769   subl(len, 8);
 8770   addl(pos, 8);
 8771 
 8772   bind(L_less_than_8_left);
 8773   cmpl(len, 4);
 8774   jcc(Assembler::less, L_less_than_4_left);
 8775 
 8776   //load 4 Bytes
 8777   movl(rax, Address(buf, pos, Address::times_1, 0));
 8778   movl(Address(tmp1, 0 * 16), rax);
 8779   addptr(tmp1, 4);
 8780   subl(len, 4);
 8781   addl(pos, 4);
 8782 
 8783   bind(L_less_than_4_left);
 8784   cmpl(len, 2);
 8785   jcc(Assembler::less, L_less_than_2_left);
 8786 
 8787   // load 2 Bytes
 8788   movw(rax, Address(buf, pos, Address::times_1, 0));
 8789   movl(Address(tmp1, 0 * 16), rax);
 8790   addptr(tmp1, 2);
 8791   subl(len, 2);
 8792   addl(pos, 2);
 8793 
 8794   bind(L_less_than_2_left);
 8795   cmpl(len, 1);
 8796   jcc(Assembler::less, L_zero_left);
 8797 
 8798   // load 1 Byte
 8799   movb(rax, Address(buf, pos, Address::times_1, 0));
 8800   movb(Address(tmp1, 0 * 16), rax);
 8801 
 8802   bind(L_zero_left);
 8803   movdqu(xmm7, Address(rsp, 0));
 8804   pxor(xmm7, xmm0);                       //xor the initial crc value
 8805 
 8806   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8807   movdqu(xmm0, Address(rax, tmp2));
 8808   pshufb(xmm7, xmm0);
 8809   jmp(L_128_done);
 8810 
 8811   bind(L_exact_16_left);
 8812   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 8813   pxor(xmm7, xmm0);                       //xor the initial crc value
 8814   jmp(L_128_done);
 8815 
 8816   bind(L_only_less_than_4);
 8817   cmpl(len, 3);
 8818   jcc(Assembler::less, L_only_less_than_3);
 8819 
 8820   // load 3 Bytes
 8821   movb(rax, Address(buf, pos, Address::times_1, 0));
 8822   movb(Address(tmp1, 0), rax);
 8823 
 8824   movb(rax, Address(buf, pos, Address::times_1, 1));
 8825   movb(Address(tmp1, 1), rax);
 8826 
 8827   movb(rax, Address(buf, pos, Address::times_1, 2));
 8828   movb(Address(tmp1, 2), rax);
 8829 
 8830   movdqu(xmm7, Address(rsp, 0));
 8831   pxor(xmm7, xmm0);                     //xor the initial crc value
 8832 
 8833   pslldq(xmm7, 0x5);
 8834   jmp(L_barrett);
 8835   bind(L_only_less_than_3);
 8836   cmpl(len, 2);
 8837   jcc(Assembler::less, L_only_less_than_2);
 8838 
 8839   // load 2 Bytes
 8840   movb(rax, Address(buf, pos, Address::times_1, 0));
 8841   movb(Address(tmp1, 0), rax);
 8842 
 8843   movb(rax, Address(buf, pos, Address::times_1, 1));
 8844   movb(Address(tmp1, 1), rax);
 8845 
 8846   movdqu(xmm7, Address(rsp, 0));
 8847   pxor(xmm7, xmm0);                     //xor the initial crc value
 8848 
 8849   pslldq(xmm7, 0x6);
 8850   jmp(L_barrett);
 8851 
 8852   bind(L_only_less_than_2);
 8853   //load 1 Byte
 8854   movb(rax, Address(buf, pos, Address::times_1, 0));
 8855   movb(Address(tmp1, 0), rax);
 8856 
 8857   movdqu(xmm7, Address(rsp, 0));
 8858   pxor(xmm7, xmm0);                     //xor the initial crc value
 8859 
 8860   pslldq(xmm7, 0x7);
 8861 }
 8862 
 8863 /**
 8864 * Compute CRC32 using AVX512 instructions
 8865 * param crc   register containing existing CRC (32-bit)
 8866 * param buf   register pointing to input byte buffer (byte*)
 8867 * param len   register containing number of bytes
 8868 * param table address of crc or crc32c table
 8869 * param tmp1  scratch register
 8870 * param tmp2  scratch register
 8871 * return rax  result register
 8872 *
 8873 * This routine is identical for crc32c with the exception of the precomputed constant
 8874 * table which will be passed as the table argument.  The calculation steps are
 8875 * the same for both variants.
 8876 */
 8877 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8878   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8879 
 8880   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8881   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8882   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8883   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8884   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8885 
 8886   const Register pos = r12;
 8887   push(r12);
 8888   subptr(rsp, 16 * 2 + 8);
 8889 
 8890   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8891   // context for the registers used, where all instructions below are using 128-bit mode
 8892   // On EVEX without VL and BW, these instructions will all be AVX.
 8893   movl(pos, 0);
 8894 
 8895   // check if smaller than 256B
 8896   cmpl(len, 256);
 8897   jcc(Assembler::less, L_less_than_256);
 8898 
 8899   // load the initial crc value
 8900   movdl(xmm10, crc);
 8901 
 8902   // receive the initial 64B data, xor the initial crc value
 8903   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8904   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8905   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8906   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8907 
 8908   subl(len, 256);
 8909   cmpl(len, 256);
 8910   jcc(Assembler::less, L_fold_128_B_loop);
 8911 
 8912   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8913   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8914   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8915   subl(len, 256);
 8916 
 8917   bind(L_fold_256_B_loop);
 8918   addl(pos, 256);
 8919   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8920   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8921   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8922   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8923 
 8924   subl(len, 256);
 8925   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8926 
 8927   // Fold 256 into 128
 8928   addl(pos, 256);
 8929   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8930   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8931   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8932 
 8933   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8934   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8935   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8936 
 8937   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8938   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8939 
 8940   addl(len, 128);
 8941   jmp(L_fold_128_B_register);
 8942 
 8943   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8944   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8945 
 8946   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8947   bind(L_fold_128_B_loop);
 8948   addl(pos, 128);
 8949   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8950   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8951 
 8952   subl(len, 128);
 8953   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8954 
 8955   addl(pos, 128);
 8956 
 8957   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8958   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8959   bind(L_fold_128_B_register);
 8960   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8961   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8962   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8963   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8964   // save last that has no multiplicand
 8965   vextracti64x2(xmm7, xmm4, 3);
 8966 
 8967   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8968   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8969   // Needed later in reduction loop
 8970   movdqu(xmm10, Address(table, 1 * 16));
 8971   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8972   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8973 
 8974   // Swap 1,0,3,2 - 01 00 11 10
 8975   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8976   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8977   vextracti128(xmm5, xmm8, 1);
 8978   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8979 
 8980   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8981   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8982   addl(len, 128 - 16);
 8983   jcc(Assembler::less, L_final_reduction_for_128);
 8984 
 8985   bind(L_16B_reduction_loop);
 8986   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8987   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8988   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8989   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8990   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8991   addl(pos, 16);
 8992   subl(len, 16);
 8993   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8994 
 8995   bind(L_final_reduction_for_128);
 8996   addl(len, 16);
 8997   jcc(Assembler::equal, L_128_done);
 8998 
 8999   bind(L_get_last_two_xmms);
 9000   movdqu(xmm2, xmm7);
 9001   addl(pos, len);
 9002   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 9003   subl(pos, len);
 9004 
 9005   // get rid of the extra data that was loaded before
 9006   // load the shift constant
 9007   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 9008   movdqu(xmm0, Address(rax, len));
 9009   addl(rax, len);
 9010 
 9011   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 9012   //Change mask to 512
 9013   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 9014   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 9015 
 9016   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 9017   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 9018   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 9019   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 9020   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 9021 
 9022   bind(L_128_done);
 9023   // compute crc of a 128-bit value
 9024   movdqu(xmm10, Address(table, 3 * 16));
 9025   movdqu(xmm0, xmm7);
 9026 
 9027   // 64b fold
 9028   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 9029   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 9030   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 9031 
 9032   // 32b fold
 9033   movdqu(xmm0, xmm7);
 9034   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 9035   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 9036   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 9037   jmp(L_barrett);
 9038 
 9039   bind(L_less_than_256);
 9040   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);
 9041 
 9042   //barrett reduction
 9043   bind(L_barrett);
 9044   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 9045   movdqu(xmm1, xmm7);
 9046   movdqu(xmm2, xmm7);
 9047   movdqu(xmm10, Address(table, 4 * 16));
 9048 
 9049   pclmulqdq(xmm7, xmm10, 0x0);
 9050   pxor(xmm7, xmm2);
 9051   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 9052   movdqu(xmm2, xmm7);
 9053   pclmulqdq(xmm7, xmm10, 0x10);
 9054   pxor(xmm7, xmm2);
 9055   pxor(xmm7, xmm1);
 9056   pextrd(crc, xmm7, 2);
 9057 
 9058   bind(L_cleanup);
 9059   addptr(rsp, 16 * 2 + 8);
 9060   pop(r12);
 9061 }
 9062 
 9063 // S. Gueron / Information Processing Letters 112 (2012) 184
 9064 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 9065 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 9066 // Output: the 64-bit carry-less product of B * CONST
 9067 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 9068                                      Register tmp1, Register tmp2, Register tmp3) {
 9069   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 9070   if (n > 0) {
 9071     addq(tmp3, n * 256 * 8);
 9072   }
 9073   //    Q1 = TABLEExt[n][B & 0xFF];
 9074   movl(tmp1, in);
 9075   andl(tmp1, 0x000000FF);
 9076   shll(tmp1, 3);
 9077   addq(tmp1, tmp3);
 9078   movq(tmp1, Address(tmp1, 0));
 9079 
 9080   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 9081   movl(tmp2, in);
 9082   shrl(tmp2, 8);
 9083   andl(tmp2, 0x000000FF);
 9084   shll(tmp2, 3);
 9085   addq(tmp2, tmp3);
 9086   movq(tmp2, Address(tmp2, 0));
 9087 
 9088   shlq(tmp2, 8);
 9089   xorq(tmp1, tmp2);
 9090 
 9091   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 9092   movl(tmp2, in);
 9093   shrl(tmp2, 16);
 9094   andl(tmp2, 0x000000FF);
 9095   shll(tmp2, 3);
 9096   addq(tmp2, tmp3);
 9097   movq(tmp2, Address(tmp2, 0));
 9098 
 9099   shlq(tmp2, 16);
 9100   xorq(tmp1, tmp2);
 9101 
 9102   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 9103   shrl(in, 24);
 9104   andl(in, 0x000000FF);
 9105   shll(in, 3);
 9106   addq(in, tmp3);
 9107   movq(in, Address(in, 0));
 9108 
 9109   shlq(in, 24);
 9110   xorq(in, tmp1);
 9111   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 9112 }
 9113 
 9114 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 9115                                       Register in_out,
 9116                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 9117                                       XMMRegister w_xtmp2,
 9118                                       Register tmp1,
 9119                                       Register n_tmp2, Register n_tmp3) {
 9120   if (is_pclmulqdq_supported) {
 9121     movdl(w_xtmp1, in_out); // modified blindly
 9122 
 9123     movl(tmp1, const_or_pre_comp_const_index);
 9124     movdl(w_xtmp2, tmp1);
 9125     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 9126 
 9127     movdq(in_out, w_xtmp1);
 9128   } else {
 9129     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 9130   }
 9131 }
 9132 
 9133 // Recombination Alternative 2: No bit-reflections
 9134 // T1 = (CRC_A * U1) << 1
 9135 // T2 = (CRC_B * U2) << 1
 9136 // C1 = T1 >> 32
 9137 // C2 = T2 >> 32
 9138 // T1 = T1 & 0xFFFFFFFF
 9139 // T2 = T2 & 0xFFFFFFFF
 9140 // T1 = CRC32(0, T1)
 9141 // T2 = CRC32(0, T2)
 9142 // C1 = C1 ^ T1
 9143 // C2 = C2 ^ T2
 9144 // CRC = C1 ^ C2 ^ CRC_C
 9145 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,
 9146                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9147                                      Register tmp1, Register tmp2,
 9148                                      Register n_tmp3) {
 9149   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 9150   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 9151   shlq(in_out, 1);
 9152   movl(tmp1, in_out);
 9153   shrq(in_out, 32);
 9154   xorl(tmp2, tmp2);
 9155   crc32(tmp2, tmp1, 4);
 9156   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 9157   shlq(in1, 1);
 9158   movl(tmp1, in1);
 9159   shrq(in1, 32);
 9160   xorl(tmp2, tmp2);
 9161   crc32(tmp2, tmp1, 4);
 9162   xorl(in1, tmp2);
 9163   xorl(in_out, in1);
 9164   xorl(in_out, in2);
 9165 }
 9166 
 9167 // Set N to predefined value
 9168 // Subtract from a length of a buffer
 9169 // execute in a loop:
 9170 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 9171 // for i = 1 to N do
 9172 //  CRC_A = CRC32(CRC_A, A[i])
 9173 //  CRC_B = CRC32(CRC_B, B[i])
 9174 //  CRC_C = CRC32(CRC_C, C[i])
 9175 // end for
 9176 // Recombine
 9177 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,
 9178                                        Register in_out1, Register in_out2, Register in_out3,
 9179                                        Register tmp1, Register tmp2, Register tmp3,
 9180                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9181                                        Register tmp4, Register tmp5,
 9182                                        Register n_tmp6) {
 9183   Label L_processPartitions;
 9184   Label L_processPartition;
 9185   Label L_exit;
 9186 
 9187   bind(L_processPartitions);
 9188   cmpl(in_out1, 3 * size);
 9189   jcc(Assembler::less, L_exit);
 9190     xorl(tmp1, tmp1);
 9191     xorl(tmp2, tmp2);
 9192     movq(tmp3, in_out2);
 9193     addq(tmp3, size);
 9194 
 9195     bind(L_processPartition);
 9196       crc32(in_out3, Address(in_out2, 0), 8);
 9197       crc32(tmp1, Address(in_out2, size), 8);
 9198       crc32(tmp2, Address(in_out2, size * 2), 8);
 9199       addq(in_out2, 8);
 9200       cmpq(in_out2, tmp3);
 9201       jcc(Assembler::less, L_processPartition);
 9202     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 9203             w_xtmp1, w_xtmp2, w_xtmp3,
 9204             tmp4, tmp5,
 9205             n_tmp6);
 9206     addq(in_out2, 2 * size);
 9207     subl(in_out1, 3 * size);
 9208     jmp(L_processPartitions);
 9209 
 9210   bind(L_exit);
 9211 }
 9212 
 9213 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 9214 // Input: A buffer I of L bytes.
 9215 // Output: the CRC32C value of the buffer.
 9216 // Notations:
 9217 // Write L = 24N + r, with N = floor (L/24).
 9218 // r = L mod 24 (0 <= r < 24).
 9219 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 9220 // N quadwords, and R consists of r bytes.
 9221 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 9222 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 9223 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 9224 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 9225 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 9226                                           Register tmp1, Register tmp2, Register tmp3,
 9227                                           Register tmp4, Register tmp5, Register tmp6,
 9228                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9229                                           bool is_pclmulqdq_supported) {
 9230   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 9231   Label L_wordByWord;
 9232   Label L_byteByByteProlog;
 9233   Label L_byteByByte;
 9234   Label L_exit;
 9235 
 9236   if (is_pclmulqdq_supported ) {
 9237     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 9238     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 9239 
 9240     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 9241     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 9242 
 9243     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 9244     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 9245     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 9246   } else {
 9247     const_or_pre_comp_const_index[0] = 1;
 9248     const_or_pre_comp_const_index[1] = 0;
 9249 
 9250     const_or_pre_comp_const_index[2] = 3;
 9251     const_or_pre_comp_const_index[3] = 2;
 9252 
 9253     const_or_pre_comp_const_index[4] = 5;
 9254     const_or_pre_comp_const_index[5] = 4;
 9255    }
 9256   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 9257                     in2, in1, in_out,
 9258                     tmp1, tmp2, tmp3,
 9259                     w_xtmp1, w_xtmp2, w_xtmp3,
 9260                     tmp4, tmp5,
 9261                     tmp6);
 9262   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 9263                     in2, in1, in_out,
 9264                     tmp1, tmp2, tmp3,
 9265                     w_xtmp1, w_xtmp2, w_xtmp3,
 9266                     tmp4, tmp5,
 9267                     tmp6);
 9268   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 9269                     in2, in1, in_out,
 9270                     tmp1, tmp2, tmp3,
 9271                     w_xtmp1, w_xtmp2, w_xtmp3,
 9272                     tmp4, tmp5,
 9273                     tmp6);
 9274   movl(tmp1, in2);
 9275   andl(tmp1, 0x00000007);
 9276   negl(tmp1);
 9277   addl(tmp1, in2);
 9278   addq(tmp1, in1);
 9279 
 9280   cmpq(in1, tmp1);
 9281   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 9282   align(16);
 9283   BIND(L_wordByWord);
 9284     crc32(in_out, Address(in1, 0), 8);
 9285     addq(in1, 8);
 9286     cmpq(in1, tmp1);
 9287     jcc(Assembler::less, L_wordByWord);
 9288 
 9289   BIND(L_byteByByteProlog);
 9290   andl(in2, 0x00000007);
 9291   movl(tmp2, 1);
 9292 
 9293   cmpl(tmp2, in2);
 9294   jccb(Assembler::greater, L_exit);
 9295   BIND(L_byteByByte);
 9296     crc32(in_out, Address(in1, 0), 1);
 9297     incq(in1);
 9298     incl(tmp2);
 9299     cmpl(tmp2, in2);
 9300     jcc(Assembler::lessEqual, L_byteByByte);
 9301 
 9302   BIND(L_exit);
 9303 }
 9304 #undef BIND
 9305 #undef BLOCK_COMMENT
 9306 
 9307 // Compress char[] array to byte[].
 9308 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 9309 // Return the array length if every element in array can be encoded,
 9310 // otherwise, the index of first non-latin1 (> 0xff) character.
 9311 //   @IntrinsicCandidate
 9312 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 9313 //     for (int i = 0; i < len; i++) {
 9314 //       char c = src[srcOff];
 9315 //       if (c > 0xff) {
 9316 //           return i;  // return index of non-latin1 char
 9317 //       }
 9318 //       dst[dstOff] = (byte)c;
 9319 //       srcOff++;
 9320 //       dstOff++;
 9321 //     }
 9322 //     return len;
 9323 //   }
 9324 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 9325   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 9326   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 9327   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 9328   Label copy_chars_loop, done, reset_sp, copy_tail;
 9329 
 9330   // rsi: src
 9331   // rdi: dst
 9332   // rdx: len
 9333   // rcx: tmp5
 9334   // rax: result
 9335 
 9336   // rsi holds start addr of source char[] to be compressed
 9337   // rdi holds start addr of destination byte[]
 9338   // rdx holds length
 9339 
 9340   assert(len != result, "");
 9341 
 9342   // save length for return
 9343   movl(result, len);
 9344 
 9345   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 9346     VM_Version::supports_avx512vlbw() &&
 9347     VM_Version::supports_bmi2()) {
 9348 
 9349     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 9350 
 9351     // alignment
 9352     Label post_alignment;
 9353 
 9354     // if length of the string is less than 32, handle it the old fashioned way
 9355     testl(len, -32);
 9356     jcc(Assembler::zero, below_threshold);
 9357 
 9358     // First check whether a character is compressible ( <= 0xFF).
 9359     // Create mask to test for Unicode chars inside zmm vector
 9360     movl(tmp5, 0x00FF);
 9361     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 9362 
 9363     testl(len, -64);
 9364     jccb(Assembler::zero, post_alignment);
 9365 
 9366     movl(tmp5, dst);
 9367     andl(tmp5, (32 - 1));
 9368     negl(tmp5);
 9369     andl(tmp5, (32 - 1));
 9370 
 9371     // bail out when there is nothing to be done
 9372     testl(tmp5, 0xFFFFFFFF);
 9373     jccb(Assembler::zero, post_alignment);
 9374 
 9375     // ~(~0 << len), where len is the # of remaining elements to process
 9376     movl(len, 0xFFFFFFFF);
 9377     shlxl(len, len, tmp5);
 9378     notl(len);
 9379     kmovdl(mask2, len);
 9380     movl(len, result);
 9381 
 9382     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9383     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9384     ktestd(mask1, mask2);
 9385     jcc(Assembler::carryClear, copy_tail);
 9386 
 9387     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9388 
 9389     addptr(src, tmp5);
 9390     addptr(src, tmp5);
 9391     addptr(dst, tmp5);
 9392     subl(len, tmp5);
 9393 
 9394     bind(post_alignment);
 9395     // end of alignment
 9396 
 9397     movl(tmp5, len);
 9398     andl(tmp5, (32 - 1));    // tail count (in chars)
 9399     andl(len, ~(32 - 1));    // vector count (in chars)
 9400     jccb(Assembler::zero, copy_loop_tail);
 9401 
 9402     lea(src, Address(src, len, Address::times_2));
 9403     lea(dst, Address(dst, len, Address::times_1));
 9404     negptr(len);
 9405 
 9406     bind(copy_32_loop);
 9407     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 9408     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 9409     kortestdl(mask1, mask1);
 9410     jccb(Assembler::carryClear, reset_for_copy_tail);
 9411 
 9412     // All elements in current processed chunk are valid candidates for
 9413     // compression. Write a truncated byte elements to the memory.
 9414     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 9415     addptr(len, 32);
 9416     jccb(Assembler::notZero, copy_32_loop);
 9417 
 9418     bind(copy_loop_tail);
 9419     // bail out when there is nothing to be done
 9420     testl(tmp5, 0xFFFFFFFF);
 9421     jcc(Assembler::zero, done);
 9422 
 9423     movl(len, tmp5);
 9424 
 9425     // ~(~0 << len), where len is the # of remaining elements to process
 9426     movl(tmp5, 0xFFFFFFFF);
 9427     shlxl(tmp5, tmp5, len);
 9428     notl(tmp5);
 9429 
 9430     kmovdl(mask2, tmp5);
 9431 
 9432     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9433     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9434     ktestd(mask1, mask2);
 9435     jcc(Assembler::carryClear, copy_tail);
 9436 
 9437     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9438     jmp(done);
 9439 
 9440     bind(reset_for_copy_tail);
 9441     lea(src, Address(src, tmp5, Address::times_2));
 9442     lea(dst, Address(dst, tmp5, Address::times_1));
 9443     subptr(len, tmp5);
 9444     jmp(copy_chars_loop);
 9445 
 9446     bind(below_threshold);
 9447   }
 9448 
 9449   if (UseSSE42Intrinsics) {
 9450     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 9451 
 9452     // vectored compression
 9453     testl(len, 0xfffffff8);
 9454     jcc(Assembler::zero, copy_tail);
 9455 
 9456     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 9457     movdl(tmp1Reg, tmp5);
 9458     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 9459 
 9460     andl(len, 0xfffffff0);
 9461     jccb(Assembler::zero, copy_16);
 9462 
 9463     // compress 16 chars per iter
 9464     pxor(tmp4Reg, tmp4Reg);
 9465 
 9466     lea(src, Address(src, len, Address::times_2));
 9467     lea(dst, Address(dst, len, Address::times_1));
 9468     negptr(len);
 9469 
 9470     bind(copy_32_loop);
 9471     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 9472     por(tmp4Reg, tmp2Reg);
 9473     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 9474     por(tmp4Reg, tmp3Reg);
 9475     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 9476     jccb(Assembler::notZero, reset_for_copy_tail);
 9477     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 9478     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 9479     addptr(len, 16);
 9480     jccb(Assembler::notZero, copy_32_loop);
 9481 
 9482     // compress next vector of 8 chars (if any)
 9483     bind(copy_16);
 9484     // len = 0
 9485     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 9486     jccb(Assembler::zero, copy_tail_sse);
 9487 
 9488     pxor(tmp3Reg, tmp3Reg);
 9489 
 9490     movdqu(tmp2Reg, Address(src, 0));
 9491     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 9492     jccb(Assembler::notZero, reset_for_copy_tail);
 9493     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 9494     movq(Address(dst, 0), tmp2Reg);
 9495     addptr(src, 16);
 9496     addptr(dst, 8);
 9497     jmpb(copy_tail_sse);
 9498 
 9499     bind(reset_for_copy_tail);
 9500     movl(tmp5, result);
 9501     andl(tmp5, 0x0000000f);
 9502     lea(src, Address(src, tmp5, Address::times_2));
 9503     lea(dst, Address(dst, tmp5, Address::times_1));
 9504     subptr(len, tmp5);
 9505     jmpb(copy_chars_loop);
 9506 
 9507     bind(copy_tail_sse);
 9508     movl(len, result);
 9509     andl(len, 0x00000007);    // tail count (in chars)
 9510   }
 9511   // compress 1 char per iter
 9512   bind(copy_tail);
 9513   testl(len, len);
 9514   jccb(Assembler::zero, done);
 9515   lea(src, Address(src, len, Address::times_2));
 9516   lea(dst, Address(dst, len, Address::times_1));
 9517   negptr(len);
 9518 
 9519   bind(copy_chars_loop);
 9520   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 9521   testl(tmp5, 0xff00);      // check if Unicode char
 9522   jccb(Assembler::notZero, reset_sp);
 9523   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 9524   increment(len);
 9525   jccb(Assembler::notZero, copy_chars_loop);
 9526 
 9527   // add len then return (len will be zero if compress succeeded, otherwise negative)
 9528   bind(reset_sp);
 9529   addl(result, len);
 9530 
 9531   bind(done);
 9532 }
 9533 
 9534 // Inflate byte[] array to char[].
 9535 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 9536 //   @IntrinsicCandidate
 9537 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 9538 //     for (int i = 0; i < len; i++) {
 9539 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 9540 //     }
 9541 //   }
 9542 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 9543   XMMRegister tmp1, Register tmp2, KRegister mask) {
 9544   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 9545   // rsi: src
 9546   // rdi: dst
 9547   // rdx: len
 9548   // rcx: tmp2
 9549 
 9550   // rsi holds start addr of source byte[] to be inflated
 9551   // rdi holds start addr of destination char[]
 9552   // rdx holds length
 9553   assert_different_registers(src, dst, len, tmp2);
 9554   movl(tmp2, len);
 9555   if ((UseAVX > 2) && // AVX512
 9556     VM_Version::supports_avx512vlbw() &&
 9557     VM_Version::supports_bmi2()) {
 9558 
 9559     Label copy_32_loop, copy_tail;
 9560     Register tmp3_aliased = len;
 9561 
 9562     // if length of the string is less than 16, handle it in an old fashioned way
 9563     testl(len, -16);
 9564     jcc(Assembler::zero, below_threshold);
 9565 
 9566     testl(len, -1 * AVX3Threshold);
 9567     jcc(Assembler::zero, avx3_threshold);
 9568 
 9569     // In order to use only one arithmetic operation for the main loop we use
 9570     // this pre-calculation
 9571     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 9572     andl(len, -32);     // vector count
 9573     jccb(Assembler::zero, copy_tail);
 9574 
 9575     lea(src, Address(src, len, Address::times_1));
 9576     lea(dst, Address(dst, len, Address::times_2));
 9577     negptr(len);
 9578 
 9579 
 9580     // inflate 32 chars per iter
 9581     bind(copy_32_loop);
 9582     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 9583     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 9584     addptr(len, 32);
 9585     jcc(Assembler::notZero, copy_32_loop);
 9586 
 9587     bind(copy_tail);
 9588     // bail out when there is nothing to be done
 9589     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 9590     jcc(Assembler::zero, done);
 9591 
 9592     // ~(~0 << length), where length is the # of remaining elements to process
 9593     movl(tmp3_aliased, -1);
 9594     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 9595     notl(tmp3_aliased);
 9596     kmovdl(mask, tmp3_aliased);
 9597     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 9598     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 9599 
 9600     jmp(done);
 9601     bind(avx3_threshold);
 9602   }
 9603   if (UseSSE42Intrinsics) {
 9604     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 9605 
 9606     if (UseAVX > 1) {
 9607       andl(tmp2, (16 - 1));
 9608       andl(len, -16);
 9609       jccb(Assembler::zero, copy_new_tail);
 9610     } else {
 9611       andl(tmp2, 0x00000007);   // tail count (in chars)
 9612       andl(len, 0xfffffff8);    // vector count (in chars)
 9613       jccb(Assembler::zero, copy_tail);
 9614     }
 9615 
 9616     // vectored inflation
 9617     lea(src, Address(src, len, Address::times_1));
 9618     lea(dst, Address(dst, len, Address::times_2));
 9619     negptr(len);
 9620 
 9621     if (UseAVX > 1) {
 9622       bind(copy_16_loop);
 9623       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 9624       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 9625       addptr(len, 16);
 9626       jcc(Assembler::notZero, copy_16_loop);
 9627 
 9628       bind(below_threshold);
 9629       bind(copy_new_tail);
 9630       movl(len, tmp2);
 9631       andl(tmp2, 0x00000007);
 9632       andl(len, 0xFFFFFFF8);
 9633       jccb(Assembler::zero, copy_tail);
 9634 
 9635       pmovzxbw(tmp1, Address(src, 0));
 9636       movdqu(Address(dst, 0), tmp1);
 9637       addptr(src, 8);
 9638       addptr(dst, 2 * 8);
 9639 
 9640       jmp(copy_tail, true);
 9641     }
 9642 
 9643     // inflate 8 chars per iter
 9644     bind(copy_8_loop);
 9645     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 9646     movdqu(Address(dst, len, Address::times_2), tmp1);
 9647     addptr(len, 8);
 9648     jcc(Assembler::notZero, copy_8_loop);
 9649 
 9650     bind(copy_tail);
 9651     movl(len, tmp2);
 9652 
 9653     cmpl(len, 4);
 9654     jccb(Assembler::less, copy_bytes);
 9655 
 9656     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 9657     pmovzxbw(tmp1, tmp1);
 9658     movq(Address(dst, 0), tmp1);
 9659     subptr(len, 4);
 9660     addptr(src, 4);
 9661     addptr(dst, 8);
 9662 
 9663     bind(copy_bytes);
 9664   } else {
 9665     bind(below_threshold);
 9666   }
 9667 
 9668   testl(len, len);
 9669   jccb(Assembler::zero, done);
 9670   lea(src, Address(src, len, Address::times_1));
 9671   lea(dst, Address(dst, len, Address::times_2));
 9672   negptr(len);
 9673 
 9674   // inflate 1 char per iter
 9675   bind(copy_chars_loop);
 9676   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 9677   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 9678   increment(len);
 9679   jcc(Assembler::notZero, copy_chars_loop);
 9680 
 9681   bind(done);
 9682 }
 9683 
 9684 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 9685   switch(type) {
 9686     case T_BYTE:
 9687     case T_BOOLEAN:
 9688       evmovdqub(dst, kmask, src, merge, vector_len);
 9689       break;
 9690     case T_CHAR:
 9691     case T_SHORT:
 9692       evmovdquw(dst, kmask, src, merge, vector_len);
 9693       break;
 9694     case T_INT:
 9695     case T_FLOAT:
 9696       evmovdqul(dst, kmask, src, merge, vector_len);
 9697       break;
 9698     case T_LONG:
 9699     case T_DOUBLE:
 9700       evmovdquq(dst, kmask, src, merge, vector_len);
 9701       break;
 9702     default:
 9703       fatal("Unexpected type argument %s", type2name(type));
 9704       break;
 9705   }
 9706 }
 9707 
 9708 
 9709 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 9710   switch(type) {
 9711     case T_BYTE:
 9712     case T_BOOLEAN:
 9713       evmovdqub(dst, kmask, src, merge, vector_len);
 9714       break;
 9715     case T_CHAR:
 9716     case T_SHORT:
 9717       evmovdquw(dst, kmask, src, merge, vector_len);
 9718       break;
 9719     case T_INT:
 9720     case T_FLOAT:
 9721       evmovdqul(dst, kmask, src, merge, vector_len);
 9722       break;
 9723     case T_LONG:
 9724     case T_DOUBLE:
 9725       evmovdquq(dst, kmask, src, merge, vector_len);
 9726       break;
 9727     default:
 9728       fatal("Unexpected type argument %s", type2name(type));
 9729       break;
 9730   }
 9731 }
 9732 
 9733 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 9734   switch(type) {
 9735     case T_BYTE:
 9736     case T_BOOLEAN:
 9737       evmovdqub(dst, kmask, src, merge, vector_len);
 9738       break;
 9739     case T_CHAR:
 9740     case T_SHORT:
 9741       evmovdquw(dst, kmask, src, merge, vector_len);
 9742       break;
 9743     case T_INT:
 9744     case T_FLOAT:
 9745       evmovdqul(dst, kmask, src, merge, vector_len);
 9746       break;
 9747     case T_LONG:
 9748     case T_DOUBLE:
 9749       evmovdquq(dst, kmask, src, merge, vector_len);
 9750       break;
 9751     default:
 9752       fatal("Unexpected type argument %s", type2name(type));
 9753       break;
 9754   }
 9755 }
 9756 
 9757 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 9758   switch(masklen) {
 9759     case 2:
 9760        knotbl(dst, src);
 9761        movl(rtmp, 3);
 9762        kmovbl(ktmp, rtmp);
 9763        kandbl(dst, ktmp, dst);
 9764        break;
 9765     case 4:
 9766        knotbl(dst, src);
 9767        movl(rtmp, 15);
 9768        kmovbl(ktmp, rtmp);
 9769        kandbl(dst, ktmp, dst);
 9770        break;
 9771     case 8:
 9772        knotbl(dst, src);
 9773        break;
 9774     case 16:
 9775        knotwl(dst, src);
 9776        break;
 9777     case 32:
 9778        knotdl(dst, src);
 9779        break;
 9780     case 64:
 9781        knotql(dst, src);
 9782        break;
 9783     default:
 9784       fatal("Unexpected vector length %d", masklen);
 9785       break;
 9786   }
 9787 }
 9788 
 9789 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9790   switch(type) {
 9791     case T_BOOLEAN:
 9792     case T_BYTE:
 9793        kandbl(dst, src1, src2);
 9794        break;
 9795     case T_CHAR:
 9796     case T_SHORT:
 9797        kandwl(dst, src1, src2);
 9798        break;
 9799     case T_INT:
 9800     case T_FLOAT:
 9801        kanddl(dst, src1, src2);
 9802        break;
 9803     case T_LONG:
 9804     case T_DOUBLE:
 9805        kandql(dst, src1, src2);
 9806        break;
 9807     default:
 9808       fatal("Unexpected type argument %s", type2name(type));
 9809       break;
 9810   }
 9811 }
 9812 
 9813 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9814   switch(type) {
 9815     case T_BOOLEAN:
 9816     case T_BYTE:
 9817        korbl(dst, src1, src2);
 9818        break;
 9819     case T_CHAR:
 9820     case T_SHORT:
 9821        korwl(dst, src1, src2);
 9822        break;
 9823     case T_INT:
 9824     case T_FLOAT:
 9825        kordl(dst, src1, src2);
 9826        break;
 9827     case T_LONG:
 9828     case T_DOUBLE:
 9829        korql(dst, src1, src2);
 9830        break;
 9831     default:
 9832       fatal("Unexpected type argument %s", type2name(type));
 9833       break;
 9834   }
 9835 }
 9836 
 9837 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9838   switch(type) {
 9839     case T_BOOLEAN:
 9840     case T_BYTE:
 9841        kxorbl(dst, src1, src2);
 9842        break;
 9843     case T_CHAR:
 9844     case T_SHORT:
 9845        kxorwl(dst, src1, src2);
 9846        break;
 9847     case T_INT:
 9848     case T_FLOAT:
 9849        kxordl(dst, src1, src2);
 9850        break;
 9851     case T_LONG:
 9852     case T_DOUBLE:
 9853        kxorql(dst, src1, src2);
 9854        break;
 9855     default:
 9856       fatal("Unexpected type argument %s", type2name(type));
 9857       break;
 9858   }
 9859 }
 9860 
 9861 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9862   switch(type) {
 9863     case T_BOOLEAN:
 9864     case T_BYTE:
 9865       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9866     case T_CHAR:
 9867     case T_SHORT:
 9868       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9869     case T_INT:
 9870     case T_FLOAT:
 9871       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9872     case T_LONG:
 9873     case T_DOUBLE:
 9874       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9875     default:
 9876       fatal("Unexpected type argument %s", type2name(type)); break;
 9877   }
 9878 }
 9879 
 9880 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9881   switch(type) {
 9882     case T_BOOLEAN:
 9883     case T_BYTE:
 9884       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9885     case T_CHAR:
 9886     case T_SHORT:
 9887       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9888     case T_INT:
 9889     case T_FLOAT:
 9890       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9891     case T_LONG:
 9892     case T_DOUBLE:
 9893       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9894     default:
 9895       fatal("Unexpected type argument %s", type2name(type)); break;
 9896   }
 9897 }
 9898 
 9899 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9900   switch(type) {
 9901     case T_BYTE:
 9902       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9903     case T_SHORT:
 9904       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9905     case T_INT:
 9906       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9907     case T_LONG:
 9908       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9909     default:
 9910       fatal("Unexpected type argument %s", type2name(type)); break;
 9911   }
 9912 }
 9913 
 9914 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9915   switch(type) {
 9916     case T_BYTE:
 9917       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9918     case T_SHORT:
 9919       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9920     case T_INT:
 9921       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9922     case T_LONG:
 9923       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9924     default:
 9925       fatal("Unexpected type argument %s", type2name(type)); break;
 9926   }
 9927 }
 9928 
 9929 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9930   switch(type) {
 9931     case T_BYTE:
 9932       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9933     case T_SHORT:
 9934       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9935     case T_INT:
 9936       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9937     case T_LONG:
 9938       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9939     default:
 9940       fatal("Unexpected type argument %s", type2name(type)); break;
 9941   }
 9942 }
 9943 
 9944 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9945   switch(type) {
 9946     case T_BYTE:
 9947       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9948     case T_SHORT:
 9949       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9950     case T_INT:
 9951       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9952     case T_LONG:
 9953       evpmaxuq(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::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9960   switch(type) {
 9961     case T_BYTE:
 9962       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9963     case T_SHORT:
 9964       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9965     case T_INT:
 9966       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9967     case T_LONG:
 9968       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9969     case T_FLOAT:
 9970       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9971     case T_DOUBLE:
 9972       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9973     default:
 9974       fatal("Unexpected type argument %s", type2name(type)); break;
 9975   }
 9976 }
 9977 
 9978 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9979   switch(type) {
 9980     case T_BYTE:
 9981       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9982     case T_SHORT:
 9983       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9984     case T_INT:
 9985       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9986     case T_LONG:
 9987       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9988     case T_FLOAT:
 9989       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9990     case T_DOUBLE:
 9991       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9992     default:
 9993       fatal("Unexpected type argument %s", type2name(type)); break;
 9994   }
 9995 }
 9996 
 9997 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9998   switch(type) {
 9999     case T_BYTE:
10000       evpminsb(dst, mask, nds, src, merge, vector_len); break;
10001     case T_SHORT:
10002       evpminsw(dst, mask, nds, src, merge, vector_len); break;
10003     case T_INT:
10004       evpminsd(dst, mask, nds, src, merge, vector_len); break;
10005     case T_LONG:
10006       evpminsq(dst, mask, nds, src, merge, vector_len); break;
10007     case T_FLOAT:
10008       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
10009     case T_DOUBLE:
10010       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
10011     default:
10012       fatal("Unexpected type argument %s", type2name(type)); break;
10013   }
10014 }
10015 
10016 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
10017   switch(type) {
10018     case T_BYTE:
10019       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
10020     case T_SHORT:
10021       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
10022     case T_INT:
10023       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
10024     case T_LONG:
10025       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
10026     case T_FLOAT:
10027       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
10028     case T_DOUBLE:
10029       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
10030     default:
10031       fatal("Unexpected type argument %s", type2name(type)); break;
10032   }
10033 }
10034 
10035 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
10036   switch(type) {
10037     case T_INT:
10038       evpxord(dst, mask, nds, src, merge, vector_len); break;
10039     case T_LONG:
10040       evpxorq(dst, mask, nds, src, merge, vector_len); break;
10041     default:
10042       fatal("Unexpected type argument %s", type2name(type)); break;
10043   }
10044 }
10045 
10046 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
10047   switch(type) {
10048     case T_INT:
10049       evpxord(dst, mask, nds, src, merge, vector_len); break;
10050     case T_LONG:
10051       evpxorq(dst, mask, nds, src, merge, vector_len); break;
10052     default:
10053       fatal("Unexpected type argument %s", type2name(type)); break;
10054   }
10055 }
10056 
10057 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
10058   switch(type) {
10059     case T_INT:
10060       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
10061     case T_LONG:
10062       evporq(dst, mask, nds, src, merge, vector_len); break;
10063     default:
10064       fatal("Unexpected type argument %s", type2name(type)); break;
10065   }
10066 }
10067 
10068 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
10069   switch(type) {
10070     case T_INT:
10071       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
10072     case T_LONG:
10073       evporq(dst, mask, nds, src, merge, vector_len); break;
10074     default:
10075       fatal("Unexpected type argument %s", type2name(type)); break;
10076   }
10077 }
10078 
10079 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
10080   switch(type) {
10081     case T_INT:
10082       evpandd(dst, mask, nds, src, merge, vector_len); break;
10083     case T_LONG:
10084       evpandq(dst, mask, nds, src, merge, vector_len); break;
10085     default:
10086       fatal("Unexpected type argument %s", type2name(type)); break;
10087   }
10088 }
10089 
10090 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
10091   switch(type) {
10092     case T_INT:
10093       evpandd(dst, mask, nds, src, merge, vector_len); break;
10094     case T_LONG:
10095       evpandq(dst, mask, nds, src, merge, vector_len); break;
10096     default:
10097       fatal("Unexpected type argument %s", type2name(type)); break;
10098   }
10099 }
10100 
10101 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
10102   switch(masklen) {
10103     case 8:
10104        kortestbl(src1, src2);
10105        break;
10106     case 16:
10107        kortestwl(src1, src2);
10108        break;
10109     case 32:
10110        kortestdl(src1, src2);
10111        break;
10112     case 64:
10113        kortestql(src1, src2);
10114        break;
10115     default:
10116       fatal("Unexpected mask length %d", masklen);
10117       break;
10118   }
10119 }
10120 
10121 
10122 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
10123   switch(masklen)  {
10124     case 8:
10125        ktestbl(src1, src2);
10126        break;
10127     case 16:
10128        ktestwl(src1, src2);
10129        break;
10130     case 32:
10131        ktestdl(src1, src2);
10132        break;
10133     case 64:
10134        ktestql(src1, src2);
10135        break;
10136     default:
10137       fatal("Unexpected mask length %d", masklen);
10138       break;
10139   }
10140 }
10141 
10142 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
10143   switch(type) {
10144     case T_INT:
10145       evprold(dst, mask, src, shift, merge, vlen_enc); break;
10146     case T_LONG:
10147       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
10148     default:
10149       fatal("Unexpected type argument %s", type2name(type)); break;
10150       break;
10151   }
10152 }
10153 
10154 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
10155   switch(type) {
10156     case T_INT:
10157       evprord(dst, mask, src, shift, merge, vlen_enc); break;
10158     case T_LONG:
10159       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
10160     default:
10161       fatal("Unexpected type argument %s", type2name(type)); break;
10162   }
10163 }
10164 
10165 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
10166   switch(type) {
10167     case T_INT:
10168       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
10169     case T_LONG:
10170       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
10171     default:
10172       fatal("Unexpected type argument %s", type2name(type)); break;
10173   }
10174 }
10175 
10176 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
10177   switch(type) {
10178     case T_INT:
10179       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
10180     case T_LONG:
10181       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
10182     default:
10183       fatal("Unexpected type argument %s", type2name(type)); break;
10184   }
10185 }
10186 
10187 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10188   assert(rscratch != noreg || always_reachable(src), "missing");
10189 
10190   if (reachable(src)) {
10191     evpandq(dst, nds, as_Address(src), vector_len);
10192   } else {
10193     lea(rscratch, src);
10194     evpandq(dst, nds, Address(rscratch, 0), vector_len);
10195   }
10196 }
10197 
10198 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
10199   assert(rscratch != noreg || always_reachable(src), "missing");
10200 
10201   if (reachable(src)) {
10202     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
10203   } else {
10204     lea(rscratch, src);
10205     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
10206   }
10207 }
10208 
10209 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10210   assert(rscratch != noreg || always_reachable(src), "missing");
10211 
10212   if (reachable(src)) {
10213     evporq(dst, nds, as_Address(src), vector_len);
10214   } else {
10215     lea(rscratch, src);
10216     evporq(dst, nds, Address(rscratch, 0), vector_len);
10217   }
10218 }
10219 
10220 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10221   assert(rscratch != noreg || always_reachable(src), "missing");
10222 
10223   if (reachable(src)) {
10224     vpshufb(dst, nds, as_Address(src), vector_len);
10225   } else {
10226     lea(rscratch, src);
10227     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
10228   }
10229 }
10230 
10231 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10232   assert(rscratch != noreg || always_reachable(src), "missing");
10233 
10234   if (reachable(src)) {
10235     Assembler::vpor(dst, nds, as_Address(src), vector_len);
10236   } else {
10237     lea(rscratch, src);
10238     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
10239   }
10240 }
10241 
10242 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
10243   assert(rscratch != noreg || always_reachable(src3), "missing");
10244 
10245   if (reachable(src3)) {
10246     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
10247   } else {
10248     lea(rscratch, src3);
10249     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
10250   }
10251 }
10252 
10253 #if COMPILER2_OR_JVMCI
10254 
10255 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
10256                                  Register length, Register temp, int vec_enc) {
10257   // Computing mask for predicated vector store.
10258   movptr(temp, -1);
10259   bzhiq(temp, temp, length);
10260   kmov(mask, temp);
10261   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
10262 }
10263 
10264 // Set memory operation for length "less than" 64 bytes.
10265 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
10266                                        XMMRegister xmm, KRegister mask, Register length,
10267                                        Register temp, bool use64byteVector) {
10268   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10269   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10270   if (!use64byteVector) {
10271     fill32(dst, disp, xmm);
10272     subptr(length, 32 >> shift);
10273     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
10274   } else {
10275     assert(MaxVectorSize == 64, "vector length != 64");
10276     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
10277   }
10278 }
10279 
10280 
10281 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
10282                                        XMMRegister xmm, KRegister mask, Register length,
10283                                        Register temp) {
10284   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10285   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10286   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
10287 }
10288 
10289 
10290 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
10291   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10292   vmovdqu(dst, xmm);
10293 }
10294 
10295 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
10296   fill32(Address(dst, disp), xmm);
10297 }
10298 
10299 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
10300   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10301   if (!use64byteVector) {
10302     fill32(dst, xmm);
10303     fill32(dst.plus_disp(32), xmm);
10304   } else {
10305     evmovdquq(dst, xmm, Assembler::AVX_512bit);
10306   }
10307 }
10308 
10309 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
10310   fill64(Address(dst, disp), xmm, use64byteVector);
10311 }
10312 
10313 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
10314                                         Register count, Register rtmp, XMMRegister xtmp) {
10315   Label L_exit;
10316   Label L_fill_start;
10317   Label L_fill_64_bytes;
10318   Label L_fill_96_bytes;
10319   Label L_fill_128_bytes;
10320   Label L_fill_128_bytes_loop;
10321   Label L_fill_128_loop_header;
10322   Label L_fill_128_bytes_loop_header;
10323   Label L_fill_128_bytes_loop_pre_header;
10324   Label L_fill_zmm_sequence;
10325 
10326   int shift = -1;
10327   switch(type) {
10328     case T_BYTE:  shift = 0;
10329       break;
10330     case T_SHORT: shift = 1;
10331       break;
10332     case T_INT:   shift = 2;
10333       break;
10334     /* Uncomment when LONG fill stubs are supported.
10335     case T_LONG:  shift = 3;
10336       break;
10337     */
10338     default:
10339       fatal("Unhandled type: %s\n", type2name(type));
10340   }
10341 
10342   if ((CopyAVX3Threshold != 0)  || (MaxVectorSize == 32)) {
10343 
10344     if (MaxVectorSize == 64) {
10345       cmpq(count, CopyAVX3Threshold >> shift);
10346       jcc(Assembler::greater, L_fill_zmm_sequence);
10347     }
10348 
10349     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
10350 
10351     bind(L_fill_start);
10352 
10353     cmpq(count, 32 >> shift);
10354     jccb(Assembler::greater, L_fill_64_bytes);
10355     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
10356     jmp(L_exit);
10357 
10358     bind(L_fill_64_bytes);
10359     cmpq(count, 64 >> shift);
10360     jccb(Assembler::greater, L_fill_96_bytes);
10361     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
10362     jmp(L_exit);
10363 
10364     bind(L_fill_96_bytes);
10365     cmpq(count, 96 >> shift);
10366     jccb(Assembler::greater, L_fill_128_bytes);
10367     fill64(to, 0, xtmp);
10368     subq(count, 64 >> shift);
10369     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
10370     jmp(L_exit);
10371 
10372     bind(L_fill_128_bytes);
10373     cmpq(count, 128 >> shift);
10374     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
10375     fill64(to, 0, xtmp);
10376     fill32(to, 64, xtmp);
10377     subq(count, 96 >> shift);
10378     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
10379     jmp(L_exit);
10380 
10381     bind(L_fill_128_bytes_loop_pre_header);
10382     {
10383       mov(rtmp, to);
10384       andq(rtmp, 31);
10385       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
10386       negq(rtmp);
10387       addq(rtmp, 32);
10388       mov64(r8, -1L);
10389       bzhiq(r8, r8, rtmp);
10390       kmovql(k2, r8);
10391       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
10392       addq(to, rtmp);
10393       shrq(rtmp, shift);
10394       subq(count, rtmp);
10395     }
10396 
10397     cmpq(count, 128 >> shift);
10398     jcc(Assembler::less, L_fill_start);
10399 
10400     bind(L_fill_128_bytes_loop_header);
10401     subq(count, 128 >> shift);
10402 
10403     align32();
10404     bind(L_fill_128_bytes_loop);
10405       fill64(to, 0, xtmp);
10406       fill64(to, 64, xtmp);
10407       addq(to, 128);
10408       subq(count, 128 >> shift);
10409       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
10410 
10411     addq(count, 128 >> shift);
10412     jcc(Assembler::zero, L_exit);
10413     jmp(L_fill_start);
10414   }
10415 
10416   if (MaxVectorSize == 64) {
10417     // Sequence using 64 byte ZMM register.
10418     Label L_fill_128_bytes_zmm;
10419     Label L_fill_192_bytes_zmm;
10420     Label L_fill_192_bytes_loop_zmm;
10421     Label L_fill_192_bytes_loop_header_zmm;
10422     Label L_fill_192_bytes_loop_pre_header_zmm;
10423     Label L_fill_start_zmm_sequence;
10424 
10425     bind(L_fill_zmm_sequence);
10426     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
10427 
10428     bind(L_fill_start_zmm_sequence);
10429     cmpq(count, 64 >> shift);
10430     jccb(Assembler::greater, L_fill_128_bytes_zmm);
10431     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
10432     jmp(L_exit);
10433 
10434     bind(L_fill_128_bytes_zmm);
10435     cmpq(count, 128 >> shift);
10436     jccb(Assembler::greater, L_fill_192_bytes_zmm);
10437     fill64(to, 0, xtmp, true);
10438     subq(count, 64 >> shift);
10439     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
10440     jmp(L_exit);
10441 
10442     bind(L_fill_192_bytes_zmm);
10443     cmpq(count, 192 >> shift);
10444     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
10445     fill64(to, 0, xtmp, true);
10446     fill64(to, 64, xtmp, true);
10447     subq(count, 128 >> shift);
10448     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
10449     jmp(L_exit);
10450 
10451     bind(L_fill_192_bytes_loop_pre_header_zmm);
10452     {
10453       movq(rtmp, to);
10454       andq(rtmp, 63);
10455       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
10456       negq(rtmp);
10457       addq(rtmp, 64);
10458       mov64(r8, -1L);
10459       bzhiq(r8, r8, rtmp);
10460       kmovql(k2, r8);
10461       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
10462       addq(to, rtmp);
10463       shrq(rtmp, shift);
10464       subq(count, rtmp);
10465     }
10466 
10467     cmpq(count, 192 >> shift);
10468     jcc(Assembler::less, L_fill_start_zmm_sequence);
10469 
10470     bind(L_fill_192_bytes_loop_header_zmm);
10471     subq(count, 192 >> shift);
10472 
10473     align32();
10474     bind(L_fill_192_bytes_loop_zmm);
10475       fill64(to, 0, xtmp, true);
10476       fill64(to, 64, xtmp, true);
10477       fill64(to, 128, xtmp, true);
10478       addq(to, 192);
10479       subq(count, 192 >> shift);
10480       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
10481 
10482     addq(count, 192 >> shift);
10483     jcc(Assembler::zero, L_exit);
10484     jmp(L_fill_start_zmm_sequence);
10485   }
10486   bind(L_exit);
10487 }
10488 #endif //COMPILER2_OR_JVMCI
10489 
10490 
10491 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
10492   Label done;
10493   cvttss2sil(dst, src);
10494   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10495   cmpl(dst, 0x80000000); // float_sign_flip
10496   jccb(Assembler::notEqual, done);
10497   subptr(rsp, 8);
10498   movflt(Address(rsp, 0), src);
10499   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
10500   pop(dst);
10501   bind(done);
10502 }
10503 
10504 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
10505   Label done;
10506   cvttsd2sil(dst, src);
10507   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10508   cmpl(dst, 0x80000000); // float_sign_flip
10509   jccb(Assembler::notEqual, done);
10510   subptr(rsp, 8);
10511   movdbl(Address(rsp, 0), src);
10512   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
10513   pop(dst);
10514   bind(done);
10515 }
10516 
10517 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
10518   Label done;
10519   cvttss2siq(dst, src);
10520   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10521   jccb(Assembler::notEqual, done);
10522   subptr(rsp, 8);
10523   movflt(Address(rsp, 0), src);
10524   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
10525   pop(dst);
10526   bind(done);
10527 }
10528 
10529 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10530   // Following code is line by line assembly translation rounding algorithm.
10531   // Please refer to java.lang.Math.round(float) algorithm for details.
10532   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
10533   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
10534   const int32_t FloatConsts_EXP_BIAS = 127;
10535   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
10536   const int32_t MINUS_32 = 0xFFFFFFE0;
10537   Label L_special_case, L_block1, L_exit;
10538   movl(rtmp, FloatConsts_EXP_BIT_MASK);
10539   movdl(dst, src);
10540   andl(dst, rtmp);
10541   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
10542   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
10543   subl(rtmp, dst);
10544   movl(rcx, rtmp);
10545   movl(dst, MINUS_32);
10546   testl(rtmp, dst);
10547   jccb(Assembler::notEqual, L_special_case);
10548   movdl(dst, src);
10549   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
10550   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
10551   movdl(rtmp, src);
10552   testl(rtmp, rtmp);
10553   jccb(Assembler::greaterEqual, L_block1);
10554   negl(dst);
10555   bind(L_block1);
10556   sarl(dst);
10557   addl(dst, 0x1);
10558   sarl(dst, 0x1);
10559   jmp(L_exit);
10560   bind(L_special_case);
10561   convert_f2i(dst, src);
10562   bind(L_exit);
10563 }
10564 
10565 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10566   // Following code is line by line assembly translation rounding algorithm.
10567   // Please refer to java.lang.Math.round(double) algorithm for details.
10568   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
10569   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
10570   const int64_t DoubleConsts_EXP_BIAS = 1023;
10571   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
10572   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
10573   Label L_special_case, L_block1, L_exit;
10574   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
10575   movq(dst, src);
10576   andq(dst, rtmp);
10577   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
10578   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
10579   subq(rtmp, dst);
10580   movq(rcx, rtmp);
10581   mov64(dst, MINUS_64);
10582   testq(rtmp, dst);
10583   jccb(Assembler::notEqual, L_special_case);
10584   movq(dst, src);
10585   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
10586   andq(dst, rtmp);
10587   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
10588   orq(dst, rtmp);
10589   movq(rtmp, src);
10590   testq(rtmp, rtmp);
10591   jccb(Assembler::greaterEqual, L_block1);
10592   negq(dst);
10593   bind(L_block1);
10594   sarq(dst);
10595   addq(dst, 0x1);
10596   sarq(dst, 0x1);
10597   jmp(L_exit);
10598   bind(L_special_case);
10599   convert_d2l(dst, src);
10600   bind(L_exit);
10601 }
10602 
10603 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
10604   Label done;
10605   cvttsd2siq(dst, src);
10606   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10607   jccb(Assembler::notEqual, done);
10608   subptr(rsp, 8);
10609   movdbl(Address(rsp, 0), src);
10610   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
10611   pop(dst);
10612   bind(done);
10613 }
10614 
10615 void MacroAssembler::cache_wb(Address line)
10616 {
10617   // 64 bit cpus always support clflush
10618   assert(VM_Version::supports_clflush(), "clflush should be available");
10619   bool optimized = VM_Version::supports_clflushopt();
10620   bool no_evict = VM_Version::supports_clwb();
10621 
10622   // prefer clwb (writeback without evict) otherwise
10623   // prefer clflushopt (potentially parallel writeback with evict)
10624   // otherwise fallback on clflush (serial writeback with evict)
10625 
10626   if (optimized) {
10627     if (no_evict) {
10628       clwb(line);
10629     } else {
10630       clflushopt(line);
10631     }
10632   } else {
10633     // no need for fence when using CLFLUSH
10634     clflush(line);
10635   }
10636 }
10637 
10638 void MacroAssembler::cache_wbsync(bool is_pre)
10639 {
10640   assert(VM_Version::supports_clflush(), "clflush should be available");
10641   bool optimized = VM_Version::supports_clflushopt();
10642   bool no_evict = VM_Version::supports_clwb();
10643 
10644   // pick the correct implementation
10645 
10646   if (!is_pre && (optimized || no_evict)) {
10647     // need an sfence for post flush when using clflushopt or clwb
10648     // otherwise no no need for any synchroniaztion
10649 
10650     sfence();
10651   }
10652 }
10653 
10654 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
10655   switch (cond) {
10656     // Note some conditions are synonyms for others
10657     case Assembler::zero:         return Assembler::notZero;
10658     case Assembler::notZero:      return Assembler::zero;
10659     case Assembler::less:         return Assembler::greaterEqual;
10660     case Assembler::lessEqual:    return Assembler::greater;
10661     case Assembler::greater:      return Assembler::lessEqual;
10662     case Assembler::greaterEqual: return Assembler::less;
10663     case Assembler::below:        return Assembler::aboveEqual;
10664     case Assembler::belowEqual:   return Assembler::above;
10665     case Assembler::above:        return Assembler::belowEqual;
10666     case Assembler::aboveEqual:   return Assembler::below;
10667     case Assembler::overflow:     return Assembler::noOverflow;
10668     case Assembler::noOverflow:   return Assembler::overflow;
10669     case Assembler::negative:     return Assembler::positive;
10670     case Assembler::positive:     return Assembler::negative;
10671     case Assembler::parity:       return Assembler::noParity;
10672     case Assembler::noParity:     return Assembler::parity;
10673   }
10674   ShouldNotReachHere(); return Assembler::overflow;
10675 }
10676 
10677 // This is simply a call to Thread::current()
10678 void MacroAssembler::get_thread_slow(Register thread) {
10679   if (thread != rax) {
10680     push(rax);
10681   }
10682   push(rdi);
10683   push(rsi);
10684   push(rdx);
10685   push(rcx);
10686   push(r8);
10687   push(r9);
10688   push(r10);
10689   push(r11);
10690 
10691   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
10692 
10693   pop(r11);
10694   pop(r10);
10695   pop(r9);
10696   pop(r8);
10697   pop(rcx);
10698   pop(rdx);
10699   pop(rsi);
10700   pop(rdi);
10701   if (thread != rax) {
10702     mov(thread, rax);
10703     pop(rax);
10704   }
10705 }
10706 
10707 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
10708   Label L_stack_ok;
10709   if (bias == 0) {
10710     testptr(sp, 2 * wordSize - 1);
10711   } else {
10712     // lea(tmp, Address(rsp, bias);
10713     mov(tmp, sp);
10714     addptr(tmp, bias);
10715     testptr(tmp, 2 * wordSize - 1);
10716   }
10717   jcc(Assembler::equal, L_stack_ok);
10718   block_comment(msg);
10719   stop(msg);
10720   bind(L_stack_ok);
10721 }
10722 
10723 // Implements fast-locking.
10724 //
10725 // obj: the object to be locked
10726 // reg_rax: rax
10727 // thread: the thread which attempts to lock obj
10728 // tmp: a temporary register
10729 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
10730   Register thread = r15_thread;
10731 
10732   assert(reg_rax == rax, "");
10733   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
10734 
10735   Label push;
10736   const Register top = tmp;
10737 
10738   // Preload the markWord. It is important that this is the first
10739   // instruction emitted as it is part of C1's null check semantics.
10740   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10741 
10742   if (UseObjectMonitorTable) {
10743     // Clear cache in case fast locking succeeds or we need to take the slow-path.
10744     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
10745   }
10746 
10747   if (DiagnoseSyncOnValueBasedClasses != 0) {
10748     load_klass(tmp, obj, rscratch1);
10749     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
10750     jcc(Assembler::notZero, slow);
10751   }
10752 
10753   // Load top.
10754   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10755 
10756   // Check if the lock-stack is full.
10757   cmpl(top, LockStack::end_offset());
10758   jcc(Assembler::greaterEqual, slow);
10759 
10760   // Check for recursion.
10761   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10762   jcc(Assembler::equal, push);
10763 
10764   // Check header for monitor (0b10).
10765   testptr(reg_rax, markWord::monitor_value);
10766   jcc(Assembler::notZero, slow);
10767 
10768   // Try to lock. Transition lock bits 0b01 => 0b00
10769   movptr(tmp, reg_rax);
10770   andptr(tmp, ~(int32_t)markWord::unlocked_value);
10771   orptr(reg_rax, markWord::unlocked_value);
10772   // Mask inline_type bit such that we go to the slow path if object is an inline type
10773   andptr(reg_rax, ~((int) markWord::inline_type_bit_in_place));
10774 
10775   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10776   jcc(Assembler::notEqual, slow);
10777 
10778   // Restore top, CAS clobbers register.
10779   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10780 
10781   bind(push);
10782   // After successful lock, push object on lock-stack.
10783   movptr(Address(thread, top), obj);
10784   incrementl(top, oopSize);
10785   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
10786 }
10787 
10788 // Implements fast-unlocking.
10789 //
10790 // obj: the object to be unlocked
10791 // reg_rax: rax
10792 // thread: the thread
10793 // tmp: a temporary register
10794 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
10795   Register thread = r15_thread;
10796 
10797   assert(reg_rax == rax, "");
10798   assert_different_registers(obj, reg_rax, thread, tmp);
10799 
10800   Label unlocked, push_and_slow;
10801   const Register top = tmp;
10802 
10803   // Check if obj is top of lock-stack.
10804   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10805   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10806   jcc(Assembler::notEqual, slow);
10807 
10808   // Pop lock-stack.
10809   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
10810   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10811 
10812   // Check if recursive.
10813   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
10814   jcc(Assembler::equal, unlocked);
10815 
10816   // Not recursive. Check header for monitor (0b10).
10817   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10818   testptr(reg_rax, markWord::monitor_value);
10819   jcc(Assembler::notZero, push_and_slow);
10820 
10821 #ifdef ASSERT
10822   // Check header not unlocked (0b01).
10823   Label not_unlocked;
10824   testptr(reg_rax, markWord::unlocked_value);
10825   jcc(Assembler::zero, not_unlocked);
10826   stop("fast_unlock already unlocked");
10827   bind(not_unlocked);
10828 #endif
10829 
10830   // Try to unlock. Transition lock bits 0b00 => 0b01
10831   movptr(tmp, reg_rax);
10832   orptr(tmp, markWord::unlocked_value);
10833   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10834   jcc(Assembler::equal, unlocked);
10835 
10836   bind(push_and_slow);
10837   // Restore lock-stack and handle the unlock in runtime.
10838 #ifdef ASSERT
10839   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10840   movptr(Address(thread, top), obj);
10841 #endif
10842   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10843   jmp(slow);
10844 
10845   bind(unlocked);
10846 }
10847 
10848 // Saves legacy GPRs state on stack.
10849 void MacroAssembler::save_legacy_gprs() {
10850   subq(rsp, 16 * wordSize);
10851   movq(Address(rsp, 15 * wordSize), rax);
10852   movq(Address(rsp, 14 * wordSize), rcx);
10853   movq(Address(rsp, 13 * wordSize), rdx);
10854   movq(Address(rsp, 12 * wordSize), rbx);
10855   movq(Address(rsp, 10 * wordSize), rbp);
10856   movq(Address(rsp, 9 * wordSize), rsi);
10857   movq(Address(rsp, 8 * wordSize), rdi);
10858   movq(Address(rsp, 7 * wordSize), r8);
10859   movq(Address(rsp, 6 * wordSize), r9);
10860   movq(Address(rsp, 5 * wordSize), r10);
10861   movq(Address(rsp, 4 * wordSize), r11);
10862   movq(Address(rsp, 3 * wordSize), r12);
10863   movq(Address(rsp, 2 * wordSize), r13);
10864   movq(Address(rsp, wordSize), r14);
10865   movq(Address(rsp, 0), r15);
10866 }
10867 
10868 // Resotres back legacy GPRs state from stack.
10869 void MacroAssembler::restore_legacy_gprs() {
10870   movq(r15, Address(rsp, 0));
10871   movq(r14, Address(rsp, wordSize));
10872   movq(r13, Address(rsp, 2 * wordSize));
10873   movq(r12, Address(rsp, 3 * wordSize));
10874   movq(r11, Address(rsp, 4 * wordSize));
10875   movq(r10, Address(rsp, 5 * wordSize));
10876   movq(r9,  Address(rsp, 6 * wordSize));
10877   movq(r8,  Address(rsp, 7 * wordSize));
10878   movq(rdi, Address(rsp, 8 * wordSize));
10879   movq(rsi, Address(rsp, 9 * wordSize));
10880   movq(rbp, Address(rsp, 10 * wordSize));
10881   movq(rbx, Address(rsp, 12 * wordSize));
10882   movq(rdx, Address(rsp, 13 * wordSize));
10883   movq(rcx, Address(rsp, 14 * wordSize));
10884   movq(rax, Address(rsp, 15 * wordSize));
10885   addq(rsp, 16 * wordSize);
10886 }
10887 
10888 void MacroAssembler::load_aotrc_address(Register reg, address a) {
10889 #if INCLUDE_CDS
10890   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
10891   if (AOTCodeCache::is_on_for_dump()) {
10892     // all aotrc field addresses should be registered in the AOTCodeCache address table
10893     lea(reg, ExternalAddress(a));
10894   } else {
10895     mov64(reg, (uint64_t)a);
10896   }
10897 #else
10898   ShouldNotReachHere();
10899 #endif
10900 }
10901 
10902 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10903   if (VM_Version::supports_apx_f()) {
10904     esetzucc(comparison, dst);
10905   } else {
10906     setb(comparison, dst);
10907     movzbl(dst, dst);
10908   }
10909 }