1 /*
    2  * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
    3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
    4  *
    5  * This code is free software; you can redistribute it and/or modify it
    6  * under the terms of the GNU General Public License version 2 only, as
    7  * published by the Free Software Foundation.
    8  *
    9  * This code is distributed in the hope that it will be useful, but WITHOUT
   10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
   11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
   12  * version 2 for more details (a copy is included in the LICENSE file that
   13  * accompanied this code).
   14  *
   15  * You should have received a copy of the GNU General Public License version
   16  * 2 along with this work; if not, write to the Free Software Foundation,
   17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
   18  *
   19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
   20  * or visit www.oracle.com if you need additional information or have any
   21  * questions.
   22  *
   23  */
   24 
   25 #include "asm/assembler.hpp"
   26 #include "asm/assembler.inline.hpp"
   27 #include "code/aotCodeCache.hpp"
   28 #include "code/compiledIC.hpp"
   29 #include "compiler/compiler_globals.hpp"
   30 #include "compiler/disassembler.hpp"
   31 #include "crc32c.h"
   32 #include "gc/shared/barrierSet.hpp"
   33 #include "gc/shared/barrierSetAssembler.hpp"
   34 #include "gc/shared/collectedHeap.inline.hpp"
   35 #include "gc/shared/tlab_globals.hpp"
   36 #include "interpreter/bytecodeHistogram.hpp"
   37 #include "interpreter/interpreter.hpp"
   38 #include "interpreter/interpreterRuntime.hpp"
   39 #include "jvm.h"
   40 #include "memory/resourceArea.hpp"
   41 #include "memory/universe.hpp"
   42 #include "oops/accessDecorators.hpp"
   43 #include "oops/compressedKlass.inline.hpp"
   44 #include "oops/compressedOops.inline.hpp"
   45 #include "oops/klass.inline.hpp"
   46 #include "prims/methodHandles.hpp"
   47 #include "runtime/continuation.hpp"
   48 #include "runtime/interfaceSupport.inline.hpp"
   49 #include "runtime/javaThread.hpp"
   50 #include "runtime/jniHandles.hpp"
   51 #include "runtime/objectMonitor.hpp"
   52 #include "runtime/os.hpp"
   53 #include "runtime/safepoint.hpp"
   54 #include "runtime/safepointMechanism.hpp"
   55 #include "runtime/sharedRuntime.hpp"
   56 #include "runtime/stubRoutines.hpp"
   57 #include "utilities/checkedCast.hpp"
   58 #include "utilities/macros.hpp"
   59 
   60 #ifdef PRODUCT
   61 #define BLOCK_COMMENT(str) /* nothing */
   62 #define STOP(error) stop(error)
   63 #else
   64 #define BLOCK_COMMENT(str) block_comment(str)
   65 #define STOP(error) block_comment(error); stop(error)
   66 #endif
   67 
   68 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   69 
   70 #ifdef ASSERT
   71 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   72 #endif
   73 
   74 static const Assembler::Condition reverse[] = {
   75     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   76     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   77     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   78     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
   79     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
   80     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
   81     Assembler::above          /* belowEqual    = 0x6 */ ,
   82     Assembler::belowEqual     /* above         = 0x7 */ ,
   83     Assembler::positive       /* negative      = 0x8 */ ,
   84     Assembler::negative       /* positive      = 0x9 */ ,
   85     Assembler::noParity       /* parity        = 0xa */ ,
   86     Assembler::parity         /* noParity      = 0xb */ ,
   87     Assembler::greaterEqual   /* less          = 0xc */ ,
   88     Assembler::less           /* greaterEqual  = 0xd */ ,
   89     Assembler::greater        /* lessEqual     = 0xe */ ,
   90     Assembler::lessEqual      /* greater       = 0xf, */
   91 
   92 };
   93 
   94 
   95 // Implementation of MacroAssembler
   96 
   97 Address MacroAssembler::as_Address(AddressLiteral adr) {
   98   // amd64 always does this as a pc-rel
   99   // we can be absolute or disp based on the instruction type
  100   // jmp/call are displacements others are absolute
  101   assert(!adr.is_lval(), "must be rval");
  102   assert(reachable(adr), "must be");
  103   return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
  104 
  105 }
  106 
  107 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  108   AddressLiteral base = adr.base();
  109   lea(rscratch, base);
  110   Address index = adr.index();
  111   assert(index._disp == 0, "must not have disp"); // maybe it can?
  112   Address array(rscratch, index._index, index._scale, index._disp);
  113   return array;
  114 }
  115 
  116 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
  117   Label L, E;
  118 
  119 #ifdef _WIN64
  120   // Windows always allocates space for it's register args
  121   assert(num_args <= 4, "only register arguments supported");
  122   subq(rsp,  frame::arg_reg_save_area_bytes);
  123 #endif
  124 
  125   // Align stack if necessary
  126   testl(rsp, 15);
  127   jcc(Assembler::zero, L);
  128 
  129   subq(rsp, 8);
  130   call(RuntimeAddress(entry_point));
  131   addq(rsp, 8);
  132   jmp(E);
  133 
  134   bind(L);
  135   call(RuntimeAddress(entry_point));
  136 
  137   bind(E);
  138 
  139 #ifdef _WIN64
  140   // restore stack pointer
  141   addq(rsp, frame::arg_reg_save_area_bytes);
  142 #endif
  143 }
  144 
  145 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
  146   assert(!src2.is_lval(), "should use cmpptr");
  147   assert(rscratch != noreg || always_reachable(src2), "missing");
  148 
  149   if (reachable(src2)) {
  150     cmpq(src1, as_Address(src2));
  151   } else {
  152     lea(rscratch, src2);
  153     Assembler::cmpq(src1, Address(rscratch, 0));
  154   }
  155 }
  156 
  157 int MacroAssembler::corrected_idivq(Register reg) {
  158   // Full implementation of Java ldiv and lrem; checks for special
  159   // case as described in JVM spec., p.243 & p.271.  The function
  160   // returns the (pc) offset of the idivl instruction - may be needed
  161   // for implicit exceptions.
  162   //
  163   //         normal case                           special case
  164   //
  165   // input : rax: dividend                         min_long
  166   //         reg: divisor   (may not be eax/edx)   -1
  167   //
  168   // output: rax: quotient  (= rax idiv reg)       min_long
  169   //         rdx: remainder (= rax irem reg)       0
  170   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
  171   static const int64_t min_long = 0x8000000000000000;
  172   Label normal_case, special_case;
  173 
  174   // check for special case
  175   cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
  176   jcc(Assembler::notEqual, normal_case);
  177   xorl(rdx, rdx); // prepare rdx for possible special case (where
  178                   // remainder = 0)
  179   cmpq(reg, -1);
  180   jcc(Assembler::equal, special_case);
  181 
  182   // handle normal case
  183   bind(normal_case);
  184   cdqq();
  185   int idivq_offset = offset();
  186   idivq(reg);
  187 
  188   // normal and special case exit
  189   bind(special_case);
  190 
  191   return idivq_offset;
  192 }
  193 
  194 void MacroAssembler::decrementq(Register reg, int value) {
  195   if (value == min_jint) { subq(reg, value); return; }
  196   if (value <  0) { incrementq(reg, -value); return; }
  197   if (value == 0) {                        ; return; }
  198   if (value == 1 && UseIncDec) { decq(reg) ; return; }
  199   /* else */      { subq(reg, value)       ; return; }
  200 }
  201 
  202 void MacroAssembler::decrementq(Address dst, int value) {
  203   if (value == min_jint) { subq(dst, value); return; }
  204   if (value <  0) { incrementq(dst, -value); return; }
  205   if (value == 0) {                        ; return; }
  206   if (value == 1 && UseIncDec) { decq(dst) ; return; }
  207   /* else */      { subq(dst, value)       ; return; }
  208 }
  209 
  210 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
  211   assert(rscratch != noreg || always_reachable(dst), "missing");
  212 
  213   if (reachable(dst)) {
  214     incrementq(as_Address(dst));
  215   } else {
  216     lea(rscratch, dst);
  217     incrementq(Address(rscratch, 0));
  218   }
  219 }
  220 
  221 void MacroAssembler::incrementq(Register reg, int value) {
  222   if (value == min_jint) { addq(reg, value); return; }
  223   if (value <  0) { decrementq(reg, -value); return; }
  224   if (value == 0) {                        ; return; }
  225   if (value == 1 && UseIncDec) { incq(reg) ; return; }
  226   /* else */      { addq(reg, value)       ; return; }
  227 }
  228 
  229 void MacroAssembler::incrementq(Address dst, int value) {
  230   if (value == min_jint) { addq(dst, value); return; }
  231   if (value <  0) { decrementq(dst, -value); return; }
  232   if (value == 0) {                        ; return; }
  233   if (value == 1 && UseIncDec) { incq(dst) ; return; }
  234   /* else */      { addq(dst, value)       ; return; }
  235 }
  236 
  237 // 32bit can do a case table jump in one instruction but we no longer allow the base
  238 // to be installed in the Address class
  239 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  240   lea(rscratch, entry.base());
  241   Address dispatch = entry.index();
  242   assert(dispatch._base == noreg, "must be");
  243   dispatch._base = rscratch;
  244   jmp(dispatch);
  245 }
  246 
  247 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  248   ShouldNotReachHere(); // 64bit doesn't use two regs
  249   cmpq(x_lo, y_lo);
  250 }
  251 
  252 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  253   mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  254 }
  255 
  256 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  257   lea(rscratch, adr);
  258   movptr(dst, rscratch);
  259 }
  260 
  261 void MacroAssembler::leave() {
  262   // %%% is this really better? Why not on 32bit too?
  263   emit_int8((unsigned char)0xC9); // LEAVE
  264 }
  265 
  266 void MacroAssembler::lneg(Register hi, Register lo) {
  267   ShouldNotReachHere(); // 64bit doesn't use two regs
  268   negq(lo);
  269 }
  270 
  271 void MacroAssembler::movoop(Register dst, jobject obj) {
  272   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  273 }
  274 
  275 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  276   mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  277   movq(dst, rscratch);
  278 }
  279 
  280 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  281   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  282 }
  283 
  284 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  285   mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  286   movq(dst, rscratch);
  287 }
  288 
  289 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  290   if (src.is_lval()) {
  291     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  292   } else {
  293     if (reachable(src)) {
  294       movq(dst, as_Address(src));
  295     } else {
  296       lea(dst, src);
  297       movq(dst, Address(dst, 0));
  298     }
  299   }
  300 }
  301 
  302 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  303   movq(as_Address(dst, rscratch), src);
  304 }
  305 
  306 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  307   movq(dst, as_Address(src, dst /*rscratch*/));
  308 }
  309 
  310 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
  311 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  312   if (is_simm32(src)) {
  313     movptr(dst, checked_cast<int32_t>(src));
  314   } else {
  315     mov64(rscratch, src);
  316     movq(dst, rscratch);
  317   }
  318 }
  319 
  320 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  321   movoop(rscratch, obj);
  322   push(rscratch);
  323 }
  324 
  325 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  326   mov_metadata(rscratch, obj);
  327   push(rscratch);
  328 }
  329 
  330 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  331   lea(rscratch, src);
  332   if (src.is_lval()) {
  333     push(rscratch);
  334   } else {
  335     pushq(Address(rscratch, 0));
  336   }
  337 }
  338 
  339 static void pass_arg0(MacroAssembler* masm, Register arg) {
  340   if (c_rarg0 != arg ) {
  341     masm->mov(c_rarg0, arg);
  342   }
  343 }
  344 
  345 static void pass_arg1(MacroAssembler* masm, Register arg) {
  346   if (c_rarg1 != arg ) {
  347     masm->mov(c_rarg1, arg);
  348   }
  349 }
  350 
  351 static void pass_arg2(MacroAssembler* masm, Register arg) {
  352   if (c_rarg2 != arg ) {
  353     masm->mov(c_rarg2, arg);
  354   }
  355 }
  356 
  357 static void pass_arg3(MacroAssembler* masm, Register arg) {
  358   if (c_rarg3 != arg ) {
  359     masm->mov(c_rarg3, arg);
  360   }
  361 }
  362 
  363 void MacroAssembler::stop(const char* msg) {
  364   if (ShowMessageBoxOnError) {
  365     address rip = pc();
  366     pusha(); // get regs on stack
  367     lea(c_rarg1, InternalAddress(rip));
  368     movq(c_rarg2, rsp); // pass pointer to regs array
  369   }
  370   // Skip AOT caching C strings in scratch buffer.
  371   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  372   lea(c_rarg0, ExternalAddress((address) str));
  373   andq(rsp, -16); // align stack as required by ABI
  374   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
  375   hlt();
  376 }
  377 
  378 void MacroAssembler::warn(const char* msg) {
  379   push(rbp);
  380   movq(rbp, rsp);
  381   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  382   push_CPU_state();   // keeps alignment at 16 bytes
  383 
  384 #ifdef _WIN64
  385   // Windows always allocates space for its register args
  386   subq(rsp,  frame::arg_reg_save_area_bytes);
  387 #endif
  388   lea(c_rarg0, ExternalAddress((address) msg));
  389   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  390 
  391 #ifdef _WIN64
  392   // restore stack pointer
  393   addq(rsp, frame::arg_reg_save_area_bytes);
  394 #endif
  395   pop_CPU_state();
  396   mov(rsp, rbp);
  397   pop(rbp);
  398 }
  399 
  400 void MacroAssembler::print_state() {
  401   address rip = pc();
  402   pusha();            // get regs on stack
  403   push(rbp);
  404   movq(rbp, rsp);
  405   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  406   push_CPU_state();   // keeps alignment at 16 bytes
  407 
  408   lea(c_rarg0, InternalAddress(rip));
  409   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  410   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  411 
  412   pop_CPU_state();
  413   mov(rsp, rbp);
  414   pop(rbp);
  415   popa();
  416 }
  417 
  418 #ifndef PRODUCT
  419 extern "C" void findpc(intptr_t x);
  420 #endif
  421 
  422 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  423   // In order to get locks to work, we need to fake a in_VM state
  424   if (ShowMessageBoxOnError) {
  425     JavaThread* thread = JavaThread::current();
  426     JavaThreadState saved_state = thread->thread_state();
  427     thread->set_thread_state(_thread_in_vm);
  428 #ifndef PRODUCT
  429     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  430       ttyLocker ttyl;
  431       BytecodeCounter::print();
  432     }
  433 #endif
  434     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  435     // XXX correct this offset for amd64
  436     // This is the value of eip which points to where verify_oop will return.
  437     if (os::message_box(msg, "Execution stopped, print registers?")) {
  438       print_state64(pc, regs);
  439       BREAKPOINT;
  440     }
  441   }
  442   fatal("DEBUG MESSAGE: %s", msg);
  443 }
  444 
  445 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  446   ttyLocker ttyl;
  447   DebuggingContext debugging{};
  448   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  449 #ifndef PRODUCT
  450   tty->cr();
  451   findpc(pc);
  452   tty->cr();
  453 #endif
  454 #define PRINT_REG(rax, value) \
  455   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  456   PRINT_REG(rax, regs[15]);
  457   PRINT_REG(rbx, regs[12]);
  458   PRINT_REG(rcx, regs[14]);
  459   PRINT_REG(rdx, regs[13]);
  460   PRINT_REG(rdi, regs[8]);
  461   PRINT_REG(rsi, regs[9]);
  462   PRINT_REG(rbp, regs[10]);
  463   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  464   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  465   PRINT_REG(r8 , regs[7]);
  466   PRINT_REG(r9 , regs[6]);
  467   PRINT_REG(r10, regs[5]);
  468   PRINT_REG(r11, regs[4]);
  469   PRINT_REG(r12, regs[3]);
  470   PRINT_REG(r13, regs[2]);
  471   PRINT_REG(r14, regs[1]);
  472   PRINT_REG(r15, regs[0]);
  473 #undef PRINT_REG
  474   // Print some words near the top of the stack.
  475   int64_t* rsp = &regs[16];
  476   int64_t* dump_sp = rsp;
  477   for (int col1 = 0; col1 < 8; col1++) {
  478     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  479     os::print_location(tty, *dump_sp++);
  480   }
  481   for (int row = 0; row < 25; row++) {
  482     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  483     for (int col = 0; col < 4; col++) {
  484       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  485     }
  486     tty->cr();
  487   }
  488   // Print some instructions around pc:
  489   Disassembler::decode((address)pc-64, (address)pc);
  490   tty->print_cr("--------");
  491   Disassembler::decode((address)pc, (address)pc+32);
  492 }
  493 
  494 // The java_calling_convention describes stack locations as ideal slots on
  495 // a frame with no abi restrictions. Since we must observe abi restrictions
  496 // (like the placement of the register window) the slots must be biased by
  497 // the following value.
  498 static int reg2offset_in(VMReg r) {
  499   // Account for saved rbp and return address
  500   // This should really be in_preserve_stack_slots
  501   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  502 }
  503 
  504 static int reg2offset_out(VMReg r) {
  505   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  506 }
  507 
  508 // A long move
  509 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  510 
  511   // The calling conventions assures us that each VMregpair is either
  512   // all really one physical register or adjacent stack slots.
  513 
  514   if (src.is_single_phys_reg() ) {
  515     if (dst.is_single_phys_reg()) {
  516       if (dst.first() != src.first()) {
  517         mov(dst.first()->as_Register(), src.first()->as_Register());
  518       }
  519     } else {
  520       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  521              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  522       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  523     }
  524   } else if (dst.is_single_phys_reg()) {
  525     assert(src.is_single_reg(),  "not a stack pair");
  526     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  527   } else {
  528     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  529     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  530     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  531   }
  532 }
  533 
  534 // A double move
  535 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  536 
  537   // The calling conventions assures us that each VMregpair is either
  538   // all really one physical register or adjacent stack slots.
  539 
  540   if (src.is_single_phys_reg() ) {
  541     if (dst.is_single_phys_reg()) {
  542       // In theory these overlap but the ordering is such that this is likely a nop
  543       if ( src.first() != dst.first()) {
  544         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  545       }
  546     } else {
  547       assert(dst.is_single_reg(), "not a stack pair");
  548       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  549     }
  550   } else if (dst.is_single_phys_reg()) {
  551     assert(src.is_single_reg(),  "not a stack pair");
  552     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  553   } else {
  554     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  555     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  556     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  557   }
  558 }
  559 
  560 
  561 // A float arg may have to do float reg int reg conversion
  562 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  563   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  564 
  565   // The calling conventions assures us that each VMregpair is either
  566   // all really one physical register or adjacent stack slots.
  567 
  568   if (src.first()->is_stack()) {
  569     if (dst.first()->is_stack()) {
  570       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  571       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  572     } else {
  573       // stack to reg
  574       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  575       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  576     }
  577   } else if (dst.first()->is_stack()) {
  578     // reg to stack
  579     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  580     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  581   } else {
  582     // reg to reg
  583     // In theory these overlap but the ordering is such that this is likely a nop
  584     if ( src.first() != dst.first()) {
  585       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  586     }
  587   }
  588 }
  589 
  590 // On 64 bit we will store integer like items to the stack as
  591 // 64 bits items (x86_32/64 abi) even though java would only store
  592 // 32bits for a parameter. On 32bit it will simply be 32 bits
  593 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  594 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  595   if (src.first()->is_stack()) {
  596     if (dst.first()->is_stack()) {
  597       // stack to stack
  598       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  599       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  600     } else {
  601       // stack to reg
  602       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  603     }
  604   } else if (dst.first()->is_stack()) {
  605     // reg to stack
  606     // Do we really have to sign extend???
  607     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  608     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  609   } else {
  610     // Do we really have to sign extend???
  611     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
  612     if (dst.first() != src.first()) {
  613       movq(dst.first()->as_Register(), src.first()->as_Register());
  614     }
  615   }
  616 }
  617 
  618 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
  619   if (src.first()->is_stack()) {
  620     if (dst.first()->is_stack()) {
  621       // stack to stack
  622       movq(rax, Address(rbp, reg2offset_in(src.first())));
  623       movq(Address(rsp, reg2offset_out(dst.first())), rax);
  624     } else {
  625       // stack to reg
  626       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
  627     }
  628   } else if (dst.first()->is_stack()) {
  629     // reg to stack
  630     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
  631   } else {
  632     if (dst.first() != src.first()) {
  633       movq(dst.first()->as_Register(), src.first()->as_Register());
  634     }
  635   }
  636 }
  637 
  638 // An oop arg. Must pass a handle not the oop itself
  639 void MacroAssembler::object_move(OopMap* map,
  640                         int oop_handle_offset,
  641                         int framesize_in_slots,
  642                         VMRegPair src,
  643                         VMRegPair dst,
  644                         bool is_receiver,
  645                         int* receiver_offset) {
  646 
  647   // must pass a handle. First figure out the location we use as a handle
  648 
  649   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
  650 
  651   // See if oop is null if it is we need no handle
  652 
  653   if (src.first()->is_stack()) {
  654 
  655     // Oop is already on the stack as an argument
  656     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
  657     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
  658     if (is_receiver) {
  659       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
  660     }
  661 
  662     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
  663     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
  664     // conditionally move a null
  665     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
  666   } else {
  667 
  668     // Oop is in a register we must store it to the space we reserve
  669     // on the stack for oop_handles and pass a handle if oop is non-null
  670 
  671     const Register rOop = src.first()->as_Register();
  672     int oop_slot;
  673     if (rOop == j_rarg0)
  674       oop_slot = 0;
  675     else if (rOop == j_rarg1)
  676       oop_slot = 1;
  677     else if (rOop == j_rarg2)
  678       oop_slot = 2;
  679     else if (rOop == j_rarg3)
  680       oop_slot = 3;
  681     else if (rOop == j_rarg4)
  682       oop_slot = 4;
  683     else {
  684       assert(rOop == j_rarg5, "wrong register");
  685       oop_slot = 5;
  686     }
  687 
  688     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
  689     int offset = oop_slot*VMRegImpl::stack_slot_size;
  690 
  691     map->set_oop(VMRegImpl::stack2reg(oop_slot));
  692     // Store oop in handle area, may be null
  693     movptr(Address(rsp, offset), rOop);
  694     if (is_receiver) {
  695       *receiver_offset = offset;
  696     }
  697 
  698     cmpptr(rOop, NULL_WORD);
  699     lea(rHandle, Address(rsp, offset));
  700     // conditionally move a null from the handle area where it was just stored
  701     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
  702   }
  703 
  704   // If arg is on the stack then place it otherwise it is already in correct reg.
  705   if (dst.first()->is_stack()) {
  706     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
  707   }
  708 }
  709 
  710 void MacroAssembler::addptr(Register dst, int32_t imm32) {
  711   addq(dst, imm32);
  712 }
  713 
  714 void MacroAssembler::addptr(Register dst, Register src) {
  715   addq(dst, src);
  716 }
  717 
  718 void MacroAssembler::addptr(Address dst, Register src) {
  719   addq(dst, src);
  720 }
  721 
  722 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  723   assert(rscratch != noreg || always_reachable(src), "missing");
  724 
  725   if (reachable(src)) {
  726     Assembler::addsd(dst, as_Address(src));
  727   } else {
  728     lea(rscratch, src);
  729     Assembler::addsd(dst, Address(rscratch, 0));
  730   }
  731 }
  732 
  733 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
  734   assert(rscratch != noreg || always_reachable(src), "missing");
  735 
  736   if (reachable(src)) {
  737     addss(dst, as_Address(src));
  738   } else {
  739     lea(rscratch, src);
  740     addss(dst, Address(rscratch, 0));
  741   }
  742 }
  743 
  744 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  745   assert(rscratch != noreg || always_reachable(src), "missing");
  746 
  747   if (reachable(src)) {
  748     Assembler::addpd(dst, as_Address(src));
  749   } else {
  750     lea(rscratch, src);
  751     Assembler::addpd(dst, Address(rscratch, 0));
  752   }
  753 }
  754 
  755 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
  756 // Stub code is generated once and never copied.
  757 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
  758 void MacroAssembler::align64() {
  759   align(64, (uint)(uintptr_t)pc());
  760 }
  761 
  762 void MacroAssembler::align32() {
  763   align(32, (uint)(uintptr_t)pc());
  764 }
  765 
  766 void MacroAssembler::align(uint modulus) {
  767   // 8273459: Ensure alignment is possible with current segment alignment
  768   assert(modulus <= (uintx)CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
  769   align(modulus, offset());
  770 }
  771 
  772 void MacroAssembler::align(uint modulus, uint target) {
  773   if (target % modulus != 0) {
  774     nop(modulus - (target % modulus));
  775   }
  776 }
  777 
  778 void MacroAssembler::push_f(XMMRegister r) {
  779   subptr(rsp, wordSize);
  780   movflt(Address(rsp, 0), r);
  781 }
  782 
  783 void MacroAssembler::pop_f(XMMRegister r) {
  784   movflt(r, Address(rsp, 0));
  785   addptr(rsp, wordSize);
  786 }
  787 
  788 void MacroAssembler::push_d(XMMRegister r) {
  789   subptr(rsp, 2 * wordSize);
  790   movdbl(Address(rsp, 0), r);
  791 }
  792 
  793 void MacroAssembler::pop_d(XMMRegister r) {
  794   movdbl(r, Address(rsp, 0));
  795   addptr(rsp, 2 * Interpreter::stackElementSize);
  796 }
  797 
  798 void MacroAssembler::push_ppx(Register src) {
  799   if (VM_Version::supports_apx_f()) {
  800     pushp(src);
  801   } else {
  802     Assembler::push(src);
  803   }
  804 }
  805 
  806 void MacroAssembler::pop_ppx(Register dst) {
  807   if (VM_Version::supports_apx_f()) {
  808     popp(dst);
  809   } else {
  810     Assembler::pop(dst);
  811   }
  812 }
  813 
  814 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  815   // Used in sign-masking with aligned address.
  816   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  817   assert(rscratch != noreg || always_reachable(src), "missing");
  818 
  819   if (UseAVX > 2 &&
  820       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
  821       (dst->encoding() >= 16)) {
  822     vpand(dst, dst, src, AVX_512bit, rscratch);
  823   } else if (reachable(src)) {
  824     Assembler::andpd(dst, as_Address(src));
  825   } else {
  826     lea(rscratch, src);
  827     Assembler::andpd(dst, Address(rscratch, 0));
  828   }
  829 }
  830 
  831 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
  832   // Used in sign-masking with aligned address.
  833   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  834   assert(rscratch != noreg || always_reachable(src), "missing");
  835 
  836   if (reachable(src)) {
  837     Assembler::andps(dst, as_Address(src));
  838   } else {
  839     lea(rscratch, src);
  840     Assembler::andps(dst, Address(rscratch, 0));
  841   }
  842 }
  843 
  844 void MacroAssembler::andptr(Register dst, int32_t imm32) {
  845   andq(dst, imm32);
  846 }
  847 
  848 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
  849   assert(rscratch != noreg || always_reachable(src), "missing");
  850 
  851   if (reachable(src)) {
  852     andq(dst, as_Address(src));
  853   } else {
  854     lea(rscratch, src);
  855     andq(dst, Address(rscratch, 0));
  856   }
  857 }
  858 
  859 void MacroAssembler::atomic_incl(Address counter_addr) {
  860   lock();
  861   incrementl(counter_addr);
  862 }
  863 
  864 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
  865   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  866 
  867   if (reachable(counter_addr)) {
  868     atomic_incl(as_Address(counter_addr));
  869   } else {
  870     lea(rscratch, counter_addr);
  871     atomic_incl(Address(rscratch, 0));
  872   }
  873 }
  874 
  875 void MacroAssembler::atomic_incq(Address counter_addr) {
  876   lock();
  877   incrementq(counter_addr);
  878 }
  879 
  880 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
  881   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  882 
  883   if (reachable(counter_addr)) {
  884     atomic_incq(as_Address(counter_addr));
  885   } else {
  886     lea(rscratch, counter_addr);
  887     atomic_incq(Address(rscratch, 0));
  888   }
  889 }
  890 
  891 // Writes to stack successive pages until offset reached to check for
  892 // stack overflow + shadow pages.  This clobbers tmp.
  893 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
  894   movptr(tmp, rsp);
  895   // Bang stack for total size given plus shadow page size.
  896   // Bang one page at a time because large size can bang beyond yellow and
  897   // red zones.
  898   Label loop;
  899   bind(loop);
  900   movl(Address(tmp, (-(int)os::vm_page_size())), size );
  901   subptr(tmp, (int)os::vm_page_size());
  902   subl(size, (int)os::vm_page_size());
  903   jcc(Assembler::greater, loop);
  904 
  905   // Bang down shadow pages too.
  906   // At this point, (tmp-0) is the last address touched, so don't
  907   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
  908   // was post-decremented.)  Skip this address by starting at i=1, and
  909   // touch a few more pages below.  N.B.  It is important to touch all
  910   // the way down including all pages in the shadow zone.
  911   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
  912     // this could be any sized move but this is can be a debugging crumb
  913     // so the bigger the better.
  914     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
  915   }
  916 }
  917 
  918 void MacroAssembler::reserved_stack_check() {
  919   // testing if reserved zone needs to be enabled
  920   Label no_reserved_zone_enabling;
  921 
  922   cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
  923   jcc(Assembler::below, no_reserved_zone_enabling);
  924 
  925   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
  926   jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
  927   should_not_reach_here();
  928 
  929   bind(no_reserved_zone_enabling);
  930 }
  931 
  932 void MacroAssembler::c2bool(Register x) {
  933   // implements x == 0 ? 0 : 1
  934   // note: must only look at least-significant byte of x
  935   //       since C-style booleans are stored in one byte
  936   //       only! (was bug)
  937   andl(x, 0xFF);
  938   setb(Assembler::notZero, x);
  939 }
  940 
  941 // Wouldn't need if AddressLiteral version had new name
  942 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
  943   Assembler::call(L, rtype);
  944 }
  945 
  946 void MacroAssembler::call(Register entry) {
  947   Assembler::call(entry);
  948 }
  949 
  950 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
  951   assert(rscratch != noreg || always_reachable(entry), "missing");
  952 
  953   if (reachable(entry)) {
  954     Assembler::call_literal(entry.target(), entry.rspec());
  955   } else {
  956     lea(rscratch, entry);
  957     Assembler::call(rscratch);
  958   }
  959 }
  960 
  961 void MacroAssembler::ic_call(address entry, jint method_index) {
  962   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
  963   // Needs full 64-bit immediate for later patching.
  964   mov64(rax, (int64_t)Universe::non_oop_word());
  965   call(AddressLiteral(entry, rh));
  966 }
  967 
  968 int MacroAssembler::ic_check_size() {
  969   return UseCompactObjectHeaders ? 17 : 14;
  970 }
  971 
  972 int MacroAssembler::ic_check(int end_alignment) {
  973   Register receiver = j_rarg0;
  974   Register data = rax;
  975   Register temp = rscratch1;
  976 
  977   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
  978   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
  979   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
  980   // before the inline cache check here, and not after
  981   align(end_alignment, offset() + ic_check_size());
  982 
  983   int uep_offset = offset();
  984 
  985   if (UseCompactObjectHeaders) {
  986     load_narrow_klass_compact(temp, receiver);
  987     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  988   } else if (UseCompressedClassPointers) {
  989     movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
  990     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  991   } else {
  992     movptr(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
  993     cmpptr(temp, Address(data, CompiledICData::speculated_klass_offset()));
  994   }
  995 
  996   // if inline cache check fails, then jump to runtime routine
  997   jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
  998   assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
  999 
 1000   return uep_offset;
 1001 }
 1002 
 1003 void MacroAssembler::emit_static_call_stub() {
 1004   // Static stub relocation also tags the Method* in the code-stream.
 1005   mov_metadata(rbx, (Metadata*) nullptr);  // Method is zapped till fixup time.
 1006   // This is recognized as unresolved by relocs/nativeinst/ic code.
 1007   jump(RuntimeAddress(pc()));
 1008 }
 1009 
 1010 // Implementation of call_VM versions
 1011 
 1012 void MacroAssembler::call_VM(Register oop_result,
 1013                              address entry_point,
 1014                              bool check_exceptions) {
 1015   Label C, E;
 1016   call(C, relocInfo::none);
 1017   jmp(E);
 1018 
 1019   bind(C);
 1020   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
 1021   ret(0);
 1022 
 1023   bind(E);
 1024 }
 1025 
 1026 void MacroAssembler::call_VM(Register oop_result,
 1027                              address entry_point,
 1028                              Register arg_1,
 1029                              bool check_exceptions) {
 1030   Label C, E;
 1031   call(C, relocInfo::none);
 1032   jmp(E);
 1033 
 1034   bind(C);
 1035   pass_arg1(this, arg_1);
 1036   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
 1037   ret(0);
 1038 
 1039   bind(E);
 1040 }
 1041 
 1042 void MacroAssembler::call_VM(Register oop_result,
 1043                              address entry_point,
 1044                              Register arg_1,
 1045                              Register arg_2,
 1046                              bool check_exceptions) {
 1047   Label C, E;
 1048   call(C, relocInfo::none);
 1049   jmp(E);
 1050 
 1051   bind(C);
 1052 
 1053   assert_different_registers(arg_1, c_rarg2);
 1054 
 1055   pass_arg2(this, arg_2);
 1056   pass_arg1(this, arg_1);
 1057   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
 1058   ret(0);
 1059 
 1060   bind(E);
 1061 }
 1062 
 1063 void MacroAssembler::call_VM(Register oop_result,
 1064                              address entry_point,
 1065                              Register arg_1,
 1066                              Register arg_2,
 1067                              Register arg_3,
 1068                              bool check_exceptions) {
 1069   Label C, E;
 1070   call(C, relocInfo::none);
 1071   jmp(E);
 1072 
 1073   bind(C);
 1074 
 1075   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1076   assert_different_registers(arg_2, c_rarg3);
 1077   pass_arg3(this, arg_3);
 1078   pass_arg2(this, arg_2);
 1079   pass_arg1(this, arg_1);
 1080   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
 1081   ret(0);
 1082 
 1083   bind(E);
 1084 }
 1085 
 1086 void MacroAssembler::call_VM(Register oop_result,
 1087                              Register last_java_sp,
 1088                              address entry_point,
 1089                              int number_of_arguments,
 1090                              bool check_exceptions) {
 1091   call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1092 }
 1093 
 1094 void MacroAssembler::call_VM(Register oop_result,
 1095                              Register last_java_sp,
 1096                              address entry_point,
 1097                              Register arg_1,
 1098                              bool check_exceptions) {
 1099   pass_arg1(this, arg_1);
 1100   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1101 }
 1102 
 1103 void MacroAssembler::call_VM(Register oop_result,
 1104                              Register last_java_sp,
 1105                              address entry_point,
 1106                              Register arg_1,
 1107                              Register arg_2,
 1108                              bool check_exceptions) {
 1109 
 1110   assert_different_registers(arg_1, c_rarg2);
 1111   pass_arg2(this, arg_2);
 1112   pass_arg1(this, arg_1);
 1113   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1114 }
 1115 
 1116 void MacroAssembler::call_VM(Register oop_result,
 1117                              Register last_java_sp,
 1118                              address entry_point,
 1119                              Register arg_1,
 1120                              Register arg_2,
 1121                              Register arg_3,
 1122                              bool check_exceptions) {
 1123   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1124   assert_different_registers(arg_2, c_rarg3);
 1125   pass_arg3(this, arg_3);
 1126   pass_arg2(this, arg_2);
 1127   pass_arg1(this, arg_1);
 1128   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1129 }
 1130 
 1131 void MacroAssembler::super_call_VM(Register oop_result,
 1132                                    Register last_java_sp,
 1133                                    address entry_point,
 1134                                    int number_of_arguments,
 1135                                    bool check_exceptions) {
 1136   MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1137 }
 1138 
 1139 void MacroAssembler::super_call_VM(Register oop_result,
 1140                                    Register last_java_sp,
 1141                                    address entry_point,
 1142                                    Register arg_1,
 1143                                    bool check_exceptions) {
 1144   pass_arg1(this, arg_1);
 1145   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1146 }
 1147 
 1148 void MacroAssembler::super_call_VM(Register oop_result,
 1149                                    Register last_java_sp,
 1150                                    address entry_point,
 1151                                    Register arg_1,
 1152                                    Register arg_2,
 1153                                    bool check_exceptions) {
 1154 
 1155   assert_different_registers(arg_1, c_rarg2);
 1156   pass_arg2(this, arg_2);
 1157   pass_arg1(this, arg_1);
 1158   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1159 }
 1160 
 1161 void MacroAssembler::super_call_VM(Register oop_result,
 1162                                    Register last_java_sp,
 1163                                    address entry_point,
 1164                                    Register arg_1,
 1165                                    Register arg_2,
 1166                                    Register arg_3,
 1167                                    bool check_exceptions) {
 1168   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1169   assert_different_registers(arg_2, c_rarg3);
 1170   pass_arg3(this, arg_3);
 1171   pass_arg2(this, arg_2);
 1172   pass_arg1(this, arg_1);
 1173   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1174 }
 1175 
 1176 void MacroAssembler::call_VM_base(Register oop_result,
 1177                                   Register last_java_sp,
 1178                                   address  entry_point,
 1179                                   int      number_of_arguments,
 1180                                   bool     check_exceptions) {
 1181   Register java_thread = r15_thread;
 1182 
 1183   // determine last_java_sp register
 1184   if (!last_java_sp->is_valid()) {
 1185     last_java_sp = rsp;
 1186   }
 1187   // debugging support
 1188   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 1189 #ifdef ASSERT
 1190   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 1191   // r12 is the heapbase.
 1192   if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 1193 #endif // ASSERT
 1194 
 1195   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 1196   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 1197 
 1198   // push java thread (becomes first argument of C function)
 1199 
 1200   mov(c_rarg0, r15_thread);
 1201 
 1202   // set last Java frame before call
 1203   assert(last_java_sp != rbp, "can't use ebp/rbp");
 1204 
 1205   // Only interpreter should have to set fp
 1206   set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
 1207 
 1208   // do the call, remove parameters
 1209   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 1210 
 1211 #ifdef ASSERT
 1212   // Check that thread register is not clobbered.
 1213   guarantee(java_thread != rax, "change this code");
 1214   push(rax);
 1215   { Label L;
 1216     get_thread_slow(rax);
 1217     cmpptr(java_thread, rax);
 1218     jcc(Assembler::equal, L);
 1219     STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
 1220     bind(L);
 1221   }
 1222   pop(rax);
 1223 #endif
 1224 
 1225   // reset last Java frame
 1226   // Only interpreter should have to clear fp
 1227   reset_last_Java_frame(true);
 1228 
 1229    // C++ interp handles this in the interpreter
 1230   check_and_handle_popframe();
 1231   check_and_handle_earlyret();
 1232 
 1233   if (check_exceptions) {
 1234     // check for pending exceptions (java_thread is set upon return)
 1235     cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 1236     // This used to conditionally jump to forward_exception however it is
 1237     // possible if we relocate that the branch will not reach. So we must jump
 1238     // around so we can always reach
 1239 
 1240     Label ok;
 1241     jcc(Assembler::equal, ok);
 1242     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 1243     bind(ok);
 1244   }
 1245 
 1246   // get oop result if there is one and reset the value in the thread
 1247   if (oop_result->is_valid()) {
 1248     get_vm_result_oop(oop_result);
 1249   }
 1250 }
 1251 
 1252 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 1253   // Calculate the value for last_Java_sp somewhat subtle.
 1254   // call_VM does an intermediate call which places a return address on
 1255   // the stack just under the stack pointer as the user finished with it.
 1256   // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
 1257 
 1258   // We've pushed one address, correct last_Java_sp
 1259   lea(rax, Address(rsp, wordSize));
 1260 
 1261   call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
 1262 }
 1263 
 1264 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
 1265 void MacroAssembler::call_VM_leaf0(address entry_point) {
 1266   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 1267 }
 1268 
 1269 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
 1270   call_VM_leaf_base(entry_point, number_of_arguments);
 1271 }
 1272 
 1273 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
 1274   pass_arg0(this, arg_0);
 1275   call_VM_leaf(entry_point, 1);
 1276 }
 1277 
 1278 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1279 
 1280   assert_different_registers(arg_0, c_rarg1);
 1281   pass_arg1(this, arg_1);
 1282   pass_arg0(this, arg_0);
 1283   call_VM_leaf(entry_point, 2);
 1284 }
 1285 
 1286 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1287   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1288   assert_different_registers(arg_1, c_rarg2);
 1289   pass_arg2(this, arg_2);
 1290   pass_arg1(this, arg_1);
 1291   pass_arg0(this, arg_0);
 1292   call_VM_leaf(entry_point, 3);
 1293 }
 1294 
 1295 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1296   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1297   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1298   assert_different_registers(arg_2, c_rarg3);
 1299   pass_arg3(this, arg_3);
 1300   pass_arg2(this, arg_2);
 1301   pass_arg1(this, arg_1);
 1302   pass_arg0(this, arg_0);
 1303   call_VM_leaf(entry_point, 3);
 1304 }
 1305 
 1306 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1307   pass_arg0(this, arg_0);
 1308   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1309 }
 1310 
 1311 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1312   assert_different_registers(arg_0, c_rarg1);
 1313   pass_arg1(this, arg_1);
 1314   pass_arg0(this, arg_0);
 1315   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1316 }
 1317 
 1318 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1319   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1320   assert_different_registers(arg_1, c_rarg2);
 1321   pass_arg2(this, arg_2);
 1322   pass_arg1(this, arg_1);
 1323   pass_arg0(this, arg_0);
 1324   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1325 }
 1326 
 1327 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1328   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1329   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1330   assert_different_registers(arg_2, c_rarg3);
 1331   pass_arg3(this, arg_3);
 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, 4);
 1336 }
 1337 
 1338 void MacroAssembler::get_vm_result_oop(Register oop_result) {
 1339   movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
 1340   movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
 1341   verify_oop_msg(oop_result, "broken oop in call_VM_base");
 1342 }
 1343 
 1344 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
 1345   movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
 1346   movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
 1347 }
 1348 
 1349 void MacroAssembler::check_and_handle_earlyret() {
 1350 }
 1351 
 1352 void MacroAssembler::check_and_handle_popframe() {
 1353 }
 1354 
 1355 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
 1356   assert(rscratch != noreg || always_reachable(src1), "missing");
 1357 
 1358   if (reachable(src1)) {
 1359     cmpl(as_Address(src1), imm);
 1360   } else {
 1361     lea(rscratch, src1);
 1362     cmpl(Address(rscratch, 0), imm);
 1363   }
 1364 }
 1365 
 1366 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
 1367   assert(!src2.is_lval(), "use cmpptr");
 1368   assert(rscratch != noreg || always_reachable(src2), "missing");
 1369 
 1370   if (reachable(src2)) {
 1371     cmpl(src1, as_Address(src2));
 1372   } else {
 1373     lea(rscratch, src2);
 1374     cmpl(src1, Address(rscratch, 0));
 1375   }
 1376 }
 1377 
 1378 void MacroAssembler::cmp32(Register src1, int32_t imm) {
 1379   Assembler::cmpl(src1, imm);
 1380 }
 1381 
 1382 void MacroAssembler::cmp32(Register src1, Address src2) {
 1383   Assembler::cmpl(src1, src2);
 1384 }
 1385 
 1386 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1387   ucomisd(opr1, opr2);
 1388 
 1389   Label L;
 1390   if (unordered_is_less) {
 1391     movl(dst, -1);
 1392     jcc(Assembler::parity, L);
 1393     jcc(Assembler::below , L);
 1394     movl(dst, 0);
 1395     jcc(Assembler::equal , L);
 1396     increment(dst);
 1397   } else { // unordered is greater
 1398     movl(dst, 1);
 1399     jcc(Assembler::parity, L);
 1400     jcc(Assembler::above , L);
 1401     movl(dst, 0);
 1402     jcc(Assembler::equal , L);
 1403     decrementl(dst);
 1404   }
 1405   bind(L);
 1406 }
 1407 
 1408 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1409   ucomiss(opr1, opr2);
 1410 
 1411   Label L;
 1412   if (unordered_is_less) {
 1413     movl(dst, -1);
 1414     jcc(Assembler::parity, L);
 1415     jcc(Assembler::below , L);
 1416     movl(dst, 0);
 1417     jcc(Assembler::equal , L);
 1418     increment(dst);
 1419   } else { // unordered is greater
 1420     movl(dst, 1);
 1421     jcc(Assembler::parity, L);
 1422     jcc(Assembler::above , L);
 1423     movl(dst, 0);
 1424     jcc(Assembler::equal , L);
 1425     decrementl(dst);
 1426   }
 1427   bind(L);
 1428 }
 1429 
 1430 
 1431 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
 1432   assert(rscratch != noreg || always_reachable(src1), "missing");
 1433 
 1434   if (reachable(src1)) {
 1435     cmpb(as_Address(src1), imm);
 1436   } else {
 1437     lea(rscratch, src1);
 1438     cmpb(Address(rscratch, 0), imm);
 1439   }
 1440 }
 1441 
 1442 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
 1443   assert(rscratch != noreg || always_reachable(src2), "missing");
 1444 
 1445   if (src2.is_lval()) {
 1446     movptr(rscratch, src2);
 1447     Assembler::cmpq(src1, rscratch);
 1448   } else if (reachable(src2)) {
 1449     cmpq(src1, as_Address(src2));
 1450   } else {
 1451     lea(rscratch, src2);
 1452     Assembler::cmpq(src1, Address(rscratch, 0));
 1453   }
 1454 }
 1455 
 1456 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
 1457   assert(src2.is_lval(), "not a mem-mem compare");
 1458   // moves src2's literal address
 1459   movptr(rscratch, src2);
 1460   Assembler::cmpq(src1, rscratch);
 1461 }
 1462 
 1463 void MacroAssembler::cmpoop(Register src1, Register src2) {
 1464   cmpptr(src1, src2);
 1465 }
 1466 
 1467 void MacroAssembler::cmpoop(Register src1, Address src2) {
 1468   cmpptr(src1, src2);
 1469 }
 1470 
 1471 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
 1472   movoop(rscratch, src2);
 1473   cmpptr(src1, rscratch);
 1474 }
 1475 
 1476 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
 1477   assert(rscratch != noreg || always_reachable(adr), "missing");
 1478 
 1479   if (reachable(adr)) {
 1480     lock();
 1481     cmpxchgptr(reg, as_Address(adr));
 1482   } else {
 1483     lea(rscratch, adr);
 1484     lock();
 1485     cmpxchgptr(reg, Address(rscratch, 0));
 1486   }
 1487 }
 1488 
 1489 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
 1490   cmpxchgq(reg, adr);
 1491 }
 1492 
 1493 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1494   assert(rscratch != noreg || always_reachable(src), "missing");
 1495 
 1496   if (reachable(src)) {
 1497     Assembler::comisd(dst, as_Address(src));
 1498   } else {
 1499     lea(rscratch, src);
 1500     Assembler::comisd(dst, Address(rscratch, 0));
 1501   }
 1502 }
 1503 
 1504 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1505   assert(rscratch != noreg || always_reachable(src), "missing");
 1506 
 1507   if (reachable(src)) {
 1508     Assembler::comiss(dst, as_Address(src));
 1509   } else {
 1510     lea(rscratch, src);
 1511     Assembler::comiss(dst, Address(rscratch, 0));
 1512   }
 1513 }
 1514 
 1515 
 1516 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
 1517   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1518 
 1519   Condition negated_cond = negate_condition(cond);
 1520   Label L;
 1521   jcc(negated_cond, L);
 1522   pushf(); // Preserve flags
 1523   atomic_incl(counter_addr, rscratch);
 1524   popf();
 1525   bind(L);
 1526 }
 1527 
 1528 int MacroAssembler::corrected_idivl(Register reg) {
 1529   // Full implementation of Java idiv and irem; checks for
 1530   // special case as described in JVM spec., p.243 & p.271.
 1531   // The function returns the (pc) offset of the idivl
 1532   // instruction - may be needed for implicit exceptions.
 1533   //
 1534   //         normal case                           special case
 1535   //
 1536   // input : rax,: dividend                         min_int
 1537   //         reg: divisor   (may not be rax,/rdx)   -1
 1538   //
 1539   // output: rax,: quotient  (= rax, idiv reg)       min_int
 1540   //         rdx: remainder (= rax, irem reg)       0
 1541   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
 1542   const int min_int = 0x80000000;
 1543   Label normal_case, special_case;
 1544 
 1545   // check for special case
 1546   cmpl(rax, min_int);
 1547   jcc(Assembler::notEqual, normal_case);
 1548   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
 1549   cmpl(reg, -1);
 1550   jcc(Assembler::equal, special_case);
 1551 
 1552   // handle normal case
 1553   bind(normal_case);
 1554   cdql();
 1555   int idivl_offset = offset();
 1556   idivl(reg);
 1557 
 1558   // normal and special case exit
 1559   bind(special_case);
 1560 
 1561   return idivl_offset;
 1562 }
 1563 
 1564 
 1565 
 1566 void MacroAssembler::decrementl(Register reg, int value) {
 1567   if (value == min_jint) {subl(reg, value) ; return; }
 1568   if (value <  0) { incrementl(reg, -value); return; }
 1569   if (value == 0) {                        ; return; }
 1570   if (value == 1 && UseIncDec) { decl(reg) ; return; }
 1571   /* else */      { subl(reg, value)       ; return; }
 1572 }
 1573 
 1574 void MacroAssembler::decrementl(Address dst, int value) {
 1575   if (value == min_jint) {subl(dst, value) ; return; }
 1576   if (value <  0) { incrementl(dst, -value); return; }
 1577   if (value == 0) {                        ; return; }
 1578   if (value == 1 && UseIncDec) { decl(dst) ; return; }
 1579   /* else */      { subl(dst, value)       ; return; }
 1580 }
 1581 
 1582 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
 1583   assert(shift_value > 0, "illegal shift value");
 1584   Label _is_positive;
 1585   testl (reg, reg);
 1586   jcc (Assembler::positive, _is_positive);
 1587   int offset = (1 << shift_value) - 1 ;
 1588 
 1589   if (offset == 1) {
 1590     incrementl(reg);
 1591   } else {
 1592     addl(reg, offset);
 1593   }
 1594 
 1595   bind (_is_positive);
 1596   sarl(reg, shift_value);
 1597 }
 1598 
 1599 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1600   assert(rscratch != noreg || always_reachable(src), "missing");
 1601 
 1602   if (reachable(src)) {
 1603     Assembler::divsd(dst, as_Address(src));
 1604   } else {
 1605     lea(rscratch, src);
 1606     Assembler::divsd(dst, Address(rscratch, 0));
 1607   }
 1608 }
 1609 
 1610 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1611   assert(rscratch != noreg || always_reachable(src), "missing");
 1612 
 1613   if (reachable(src)) {
 1614     Assembler::divss(dst, as_Address(src));
 1615   } else {
 1616     lea(rscratch, src);
 1617     Assembler::divss(dst, Address(rscratch, 0));
 1618   }
 1619 }
 1620 
 1621 void MacroAssembler::enter() {
 1622   push(rbp);
 1623   mov(rbp, rsp);
 1624 }
 1625 
 1626 void MacroAssembler::post_call_nop() {
 1627   if (!Continuations::enabled()) {
 1628     return;
 1629   }
 1630   InstructionMark im(this);
 1631   relocate(post_call_nop_Relocation::spec());
 1632   InlineSkippedInstructionsCounter skipCounter(this);
 1633   emit_int8((uint8_t)0x0f);
 1634   emit_int8((uint8_t)0x1f);
 1635   emit_int8((uint8_t)0x84);
 1636   emit_int8((uint8_t)0x00);
 1637   emit_int32(0x00);
 1638 }
 1639 
 1640 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1641   assert(rscratch != noreg || always_reachable(src), "missing");
 1642   if (reachable(src)) {
 1643     Assembler::mulpd(dst, as_Address(src));
 1644   } else {
 1645     lea(rscratch, src);
 1646     Assembler::mulpd(dst, Address(rscratch, 0));
 1647   }
 1648 }
 1649 
 1650 // dst = c = a * b + c
 1651 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1652   Assembler::vfmadd231sd(c, a, b);
 1653   if (dst != c) {
 1654     movdbl(dst, c);
 1655   }
 1656 }
 1657 
 1658 // dst = c = a * b + c
 1659 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1660   Assembler::vfmadd231ss(c, a, b);
 1661   if (dst != c) {
 1662     movflt(dst, c);
 1663   }
 1664 }
 1665 
 1666 // dst = c = a * b + c
 1667 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1668   Assembler::vfmadd231pd(c, a, b, vector_len);
 1669   if (dst != c) {
 1670     vmovdqu(dst, c);
 1671   }
 1672 }
 1673 
 1674 // dst = c = a * b + c
 1675 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1676   Assembler::vfmadd231ps(c, a, b, vector_len);
 1677   if (dst != c) {
 1678     vmovdqu(dst, c);
 1679   }
 1680 }
 1681 
 1682 // dst = c = a * b + c
 1683 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1684   Assembler::vfmadd231pd(c, a, b, vector_len);
 1685   if (dst != c) {
 1686     vmovdqu(dst, c);
 1687   }
 1688 }
 1689 
 1690 // dst = c = a * b + c
 1691 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1692   Assembler::vfmadd231ps(c, a, b, vector_len);
 1693   if (dst != c) {
 1694     vmovdqu(dst, c);
 1695   }
 1696 }
 1697 
 1698 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
 1699   assert(rscratch != noreg || always_reachable(dst), "missing");
 1700 
 1701   if (reachable(dst)) {
 1702     incrementl(as_Address(dst));
 1703   } else {
 1704     lea(rscratch, dst);
 1705     incrementl(Address(rscratch, 0));
 1706   }
 1707 }
 1708 
 1709 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
 1710   incrementl(as_Address(dst, rscratch));
 1711 }
 1712 
 1713 void MacroAssembler::incrementl(Register reg, int value) {
 1714   if (value == min_jint) {addl(reg, value) ; return; }
 1715   if (value <  0) { decrementl(reg, -value); return; }
 1716   if (value == 0) {                        ; return; }
 1717   if (value == 1 && UseIncDec) { incl(reg) ; return; }
 1718   /* else */      { addl(reg, value)       ; return; }
 1719 }
 1720 
 1721 void MacroAssembler::incrementl(Address dst, int value) {
 1722   if (value == min_jint) {addl(dst, value) ; return; }
 1723   if (value <  0) { decrementl(dst, -value); return; }
 1724   if (value == 0) {                        ; return; }
 1725   if (value == 1 && UseIncDec) { incl(dst) ; return; }
 1726   /* else */      { addl(dst, value)       ; return; }
 1727 }
 1728 
 1729 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
 1730   assert(rscratch != noreg || always_reachable(dst), "missing");
 1731   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
 1732   if (reachable(dst)) {
 1733     jmp_literal(dst.target(), dst.rspec());
 1734   } else {
 1735     lea(rscratch, dst);
 1736     jmp(rscratch);
 1737   }
 1738 }
 1739 
 1740 void MacroAssembler::jump_cc(Condition cc, 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_cc");
 1743   if (reachable(dst)) {
 1744     InstructionMark im(this);
 1745     relocate(dst.reloc());
 1746     const int short_size = 2;
 1747     const int long_size = 6;
 1748     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
 1749     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
 1750       // 0111 tttn #8-bit disp
 1751       emit_int8(0x70 | cc);
 1752       emit_int8((offs - short_size) & 0xFF);
 1753     } else {
 1754       // 0000 1111 1000 tttn #32-bit disp
 1755       emit_int8(0x0F);
 1756       emit_int8((unsigned char)(0x80 | cc));
 1757       emit_int32(offs - long_size);
 1758     }
 1759   } else {
 1760 #ifdef ASSERT
 1761     warning("reversing conditional branch");
 1762 #endif /* ASSERT */
 1763     Label skip;
 1764     jccb(reverse[cc], skip);
 1765     lea(rscratch, dst);
 1766     Assembler::jmp(rscratch);
 1767     bind(skip);
 1768   }
 1769 }
 1770 
 1771 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
 1772   ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 1773   assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
 1774 
 1775   stmxcsr(mxcsr_save);
 1776   movl(tmp, mxcsr_save);
 1777   if (EnableX86ECoreOpts) {
 1778     // The mxcsr_std has status bits set for performance on ECore
 1779     orl(tmp, 0x003f);
 1780   } else {
 1781     // Mask out status bits (only check control and mask bits)
 1782     andl(tmp, 0xFFC0);
 1783   }
 1784   cmp32(tmp, mxcsr_std, rscratch);
 1785 }
 1786 
 1787 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
 1788   assert(rscratch != noreg || always_reachable(src), "missing");
 1789 
 1790   if (reachable(src)) {
 1791     Assembler::ldmxcsr(as_Address(src));
 1792   } else {
 1793     lea(rscratch, src);
 1794     Assembler::ldmxcsr(Address(rscratch, 0));
 1795   }
 1796 }
 1797 
 1798 int MacroAssembler::load_signed_byte(Register dst, Address src) {
 1799   int off = offset();
 1800   movsbl(dst, src); // movsxb
 1801   return off;
 1802 }
 1803 
 1804 // Note: load_signed_short used to be called load_signed_word.
 1805 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
 1806 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
 1807 // The term "word" in HotSpot means a 32- or 64-bit machine word.
 1808 int MacroAssembler::load_signed_short(Register dst, Address src) {
 1809   // This is dubious to me since it seems safe to do a signed 16 => 64 bit
 1810   // version but this is what 64bit has always done. This seems to imply
 1811   // that users are only using 32bits worth.
 1812   int off = offset();
 1813   movswl(dst, src); // movsxw
 1814   return off;
 1815 }
 1816 
 1817 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
 1818   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1819   // and "3.9 Partial Register Penalties", p. 22).
 1820   int off = offset();
 1821   movzbl(dst, src); // movzxb
 1822   return off;
 1823 }
 1824 
 1825 // Note: load_unsigned_short used to be called load_unsigned_word.
 1826 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
 1827   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1828   // and "3.9 Partial Register Penalties", p. 22).
 1829   int off = offset();
 1830   movzwl(dst, src); // movzxw
 1831   return off;
 1832 }
 1833 
 1834 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
 1835   switch (size_in_bytes) {
 1836   case  8:  movq(dst, src); break;
 1837   case  4:  movl(dst, src); break;
 1838   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
 1839   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
 1840   default:  ShouldNotReachHere();
 1841   }
 1842 }
 1843 
 1844 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
 1845   switch (size_in_bytes) {
 1846   case  8:  movq(dst, src); break;
 1847   case  4:  movl(dst, src); break;
 1848   case  2:  movw(dst, src); break;
 1849   case  1:  movb(dst, src); break;
 1850   default:  ShouldNotReachHere();
 1851   }
 1852 }
 1853 
 1854 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
 1855   assert(rscratch != noreg || always_reachable(dst), "missing");
 1856 
 1857   if (reachable(dst)) {
 1858     movl(as_Address(dst), src);
 1859   } else {
 1860     lea(rscratch, dst);
 1861     movl(Address(rscratch, 0), src);
 1862   }
 1863 }
 1864 
 1865 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
 1866   if (reachable(src)) {
 1867     movl(dst, as_Address(src));
 1868   } else {
 1869     lea(dst, src);
 1870     movl(dst, Address(dst, 0));
 1871   }
 1872 }
 1873 
 1874 // C++ bool manipulation
 1875 
 1876 void MacroAssembler::movbool(Register dst, Address src) {
 1877   if(sizeof(bool) == 1)
 1878     movb(dst, src);
 1879   else if(sizeof(bool) == 2)
 1880     movw(dst, src);
 1881   else if(sizeof(bool) == 4)
 1882     movl(dst, src);
 1883   else
 1884     // unsupported
 1885     ShouldNotReachHere();
 1886 }
 1887 
 1888 void MacroAssembler::movbool(Address dst, bool boolconst) {
 1889   if(sizeof(bool) == 1)
 1890     movb(dst, (int) boolconst);
 1891   else if(sizeof(bool) == 2)
 1892     movw(dst, (int) boolconst);
 1893   else if(sizeof(bool) == 4)
 1894     movl(dst, (int) boolconst);
 1895   else
 1896     // unsupported
 1897     ShouldNotReachHere();
 1898 }
 1899 
 1900 void MacroAssembler::movbool(Address dst, Register src) {
 1901   if(sizeof(bool) == 1)
 1902     movb(dst, src);
 1903   else if(sizeof(bool) == 2)
 1904     movw(dst, src);
 1905   else if(sizeof(bool) == 4)
 1906     movl(dst, src);
 1907   else
 1908     // unsupported
 1909     ShouldNotReachHere();
 1910 }
 1911 
 1912 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1913   assert(rscratch != noreg || always_reachable(src), "missing");
 1914 
 1915   if (reachable(src)) {
 1916     movdl(dst, as_Address(src));
 1917   } else {
 1918     lea(rscratch, src);
 1919     movdl(dst, Address(rscratch, 0));
 1920   }
 1921 }
 1922 
 1923 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1924   assert(rscratch != noreg || always_reachable(src), "missing");
 1925 
 1926   if (reachable(src)) {
 1927     movq(dst, as_Address(src));
 1928   } else {
 1929     lea(rscratch, src);
 1930     movq(dst, Address(rscratch, 0));
 1931   }
 1932 }
 1933 
 1934 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1935   assert(rscratch != noreg || always_reachable(src), "missing");
 1936 
 1937   if (reachable(src)) {
 1938     if (UseXmmLoadAndClearUpper) {
 1939       movsd (dst, as_Address(src));
 1940     } else {
 1941       movlpd(dst, as_Address(src));
 1942     }
 1943   } else {
 1944     lea(rscratch, src);
 1945     if (UseXmmLoadAndClearUpper) {
 1946       movsd (dst, Address(rscratch, 0));
 1947     } else {
 1948       movlpd(dst, Address(rscratch, 0));
 1949     }
 1950   }
 1951 }
 1952 
 1953 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1954   assert(rscratch != noreg || always_reachable(src), "missing");
 1955 
 1956   if (reachable(src)) {
 1957     movss(dst, as_Address(src));
 1958   } else {
 1959     lea(rscratch, src);
 1960     movss(dst, Address(rscratch, 0));
 1961   }
 1962 }
 1963 
 1964 void MacroAssembler::movptr(Register dst, Register src) {
 1965   movq(dst, src);
 1966 }
 1967 
 1968 void MacroAssembler::movptr(Register dst, Address src) {
 1969   movq(dst, src);
 1970 }
 1971 
 1972 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 1973 void MacroAssembler::movptr(Register dst, intptr_t src) {
 1974   if (is_uimm32(src)) {
 1975     movl(dst, checked_cast<uint32_t>(src));
 1976   } else if (is_simm32(src)) {
 1977     movq(dst, checked_cast<int32_t>(src));
 1978   } else {
 1979     mov64(dst, src);
 1980   }
 1981 }
 1982 
 1983 void MacroAssembler::movptr(Address dst, Register src) {
 1984   movq(dst, src);
 1985 }
 1986 
 1987 void MacroAssembler::movptr(Address dst, int32_t src) {
 1988   movslq(dst, src);
 1989 }
 1990 
 1991 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 1992   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 1993   Assembler::movdqu(dst, src);
 1994 }
 1995 
 1996 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 1997   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 1998   Assembler::movdqu(dst, src);
 1999 }
 2000 
 2001 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2002   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2003   Assembler::movdqu(dst, src);
 2004 }
 2005 
 2006 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2007   assert(rscratch != noreg || always_reachable(src), "missing");
 2008 
 2009   if (reachable(src)) {
 2010     movdqu(dst, as_Address(src));
 2011   } else {
 2012     lea(rscratch, src);
 2013     movdqu(dst, Address(rscratch, 0));
 2014   }
 2015 }
 2016 
 2017 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2018   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2019   Assembler::vmovdqu(dst, src);
 2020 }
 2021 
 2022 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2023   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2024   Assembler::vmovdqu(dst, src);
 2025 }
 2026 
 2027 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2028   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2029   Assembler::vmovdqu(dst, src);
 2030 }
 2031 
 2032 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2033   assert(rscratch != noreg || always_reachable(src), "missing");
 2034 
 2035   if (reachable(src)) {
 2036     vmovdqu(dst, as_Address(src));
 2037   }
 2038   else {
 2039     lea(rscratch, src);
 2040     vmovdqu(dst, Address(rscratch, 0));
 2041   }
 2042 }
 2043 
 2044 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2045   assert(rscratch != noreg || always_reachable(src), "missing");
 2046 
 2047   if (vector_len == AVX_512bit) {
 2048     evmovdquq(dst, src, AVX_512bit, rscratch);
 2049   } else if (vector_len == AVX_256bit) {
 2050     vmovdqu(dst, src, rscratch);
 2051   } else {
 2052     movdqu(dst, src, rscratch);
 2053   }
 2054 }
 2055 
 2056 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
 2057   if (vector_len == AVX_512bit) {
 2058     evmovdquq(dst, src, AVX_512bit);
 2059   } else if (vector_len == AVX_256bit) {
 2060     vmovdqu(dst, src);
 2061   } else {
 2062     movdqu(dst, src);
 2063   }
 2064 }
 2065 
 2066 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
 2067   if (vector_len == AVX_512bit) {
 2068     evmovdquq(dst, src, AVX_512bit);
 2069   } else if (vector_len == AVX_256bit) {
 2070     vmovdqu(dst, src);
 2071   } else {
 2072     movdqu(dst, src);
 2073   }
 2074 }
 2075 
 2076 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
 2077   if (vector_len == AVX_512bit) {
 2078     evmovdquq(dst, src, AVX_512bit);
 2079   } else if (vector_len == AVX_256bit) {
 2080     vmovdqu(dst, src);
 2081   } else {
 2082     movdqu(dst, src);
 2083   }
 2084 }
 2085 
 2086 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2087   assert(rscratch != noreg || always_reachable(src), "missing");
 2088 
 2089   if (reachable(src)) {
 2090     vmovdqa(dst, as_Address(src));
 2091   }
 2092   else {
 2093     lea(rscratch, src);
 2094     vmovdqa(dst, Address(rscratch, 0));
 2095   }
 2096 }
 2097 
 2098 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2099   assert(rscratch != noreg || always_reachable(src), "missing");
 2100 
 2101   if (vector_len == AVX_512bit) {
 2102     evmovdqaq(dst, src, AVX_512bit, rscratch);
 2103   } else if (vector_len == AVX_256bit) {
 2104     vmovdqa(dst, src, rscratch);
 2105   } else {
 2106     movdqa(dst, src, rscratch);
 2107   }
 2108 }
 2109 
 2110 void MacroAssembler::kmov(KRegister dst, Address src) {
 2111   if (VM_Version::supports_avx512bw()) {
 2112     kmovql(dst, src);
 2113   } else {
 2114     assert(VM_Version::supports_evex(), "");
 2115     kmovwl(dst, src);
 2116   }
 2117 }
 2118 
 2119 void MacroAssembler::kmov(Address dst, KRegister src) {
 2120   if (VM_Version::supports_avx512bw()) {
 2121     kmovql(dst, src);
 2122   } else {
 2123     assert(VM_Version::supports_evex(), "");
 2124     kmovwl(dst, src);
 2125   }
 2126 }
 2127 
 2128 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2129   if (VM_Version::supports_avx512bw()) {
 2130     kmovql(dst, src);
 2131   } else {
 2132     assert(VM_Version::supports_evex(), "");
 2133     kmovwl(dst, src);
 2134   }
 2135 }
 2136 
 2137 void MacroAssembler::kmov(Register dst, KRegister src) {
 2138   if (VM_Version::supports_avx512bw()) {
 2139     kmovql(dst, src);
 2140   } else {
 2141     assert(VM_Version::supports_evex(), "");
 2142     kmovwl(dst, src);
 2143   }
 2144 }
 2145 
 2146 void MacroAssembler::kmov(KRegister dst, Register src) {
 2147   if (VM_Version::supports_avx512bw()) {
 2148     kmovql(dst, src);
 2149   } else {
 2150     assert(VM_Version::supports_evex(), "");
 2151     kmovwl(dst, src);
 2152   }
 2153 }
 2154 
 2155 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2156   assert(rscratch != noreg || always_reachable(src), "missing");
 2157 
 2158   if (reachable(src)) {
 2159     kmovql(dst, as_Address(src));
 2160   } else {
 2161     lea(rscratch, src);
 2162     kmovql(dst, Address(rscratch, 0));
 2163   }
 2164 }
 2165 
 2166 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2167   assert(rscratch != noreg || always_reachable(src), "missing");
 2168 
 2169   if (reachable(src)) {
 2170     kmovwl(dst, as_Address(src));
 2171   } else {
 2172     lea(rscratch, src);
 2173     kmovwl(dst, Address(rscratch, 0));
 2174   }
 2175 }
 2176 
 2177 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2178                                int vector_len, Register rscratch) {
 2179   assert(rscratch != noreg || always_reachable(src), "missing");
 2180 
 2181   if (reachable(src)) {
 2182     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2183   } else {
 2184     lea(rscratch, src);
 2185     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2186   }
 2187 }
 2188 
 2189 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2190                                int vector_len, Register rscratch) {
 2191   assert(rscratch != noreg || always_reachable(src), "missing");
 2192 
 2193   if (reachable(src)) {
 2194     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2195   } else {
 2196     lea(rscratch, src);
 2197     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2198   }
 2199 }
 2200 
 2201 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2202   assert(rscratch != noreg || always_reachable(src), "missing");
 2203 
 2204   if (reachable(src)) {
 2205     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2206   } else {
 2207     lea(rscratch, src);
 2208     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2209   }
 2210 }
 2211 
 2212 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2213   assert(rscratch != noreg || always_reachable(src), "missing");
 2214 
 2215   if (reachable(src)) {
 2216     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2217   } else {
 2218     lea(rscratch, src);
 2219     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2220   }
 2221 }
 2222 
 2223 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2224   assert(rscratch != noreg || always_reachable(src), "missing");
 2225 
 2226   if (reachable(src)) {
 2227     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2228   } else {
 2229     lea(rscratch, src);
 2230     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2231   }
 2232 }
 2233 
 2234 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2235   assert(rscratch != noreg || always_reachable(src), "missing");
 2236 
 2237   if (reachable(src)) {
 2238     Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
 2239   } else {
 2240     lea(rscratch, src);
 2241     Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2242   }
 2243 }
 2244 
 2245 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2246   assert(rscratch != noreg || always_reachable(src), "missing");
 2247 
 2248   if (reachable(src)) {
 2249     Assembler::evmovdqaq(dst, as_Address(src), vector_len);
 2250   } else {
 2251     lea(rscratch, src);
 2252     Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
 2253   }
 2254 }
 2255 
 2256 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2257   assert(rscratch != noreg || always_reachable(src), "missing");
 2258 
 2259   if (reachable(src)) {
 2260     Assembler::movapd(dst, as_Address(src));
 2261   } else {
 2262     lea(rscratch, src);
 2263     Assembler::movapd(dst, Address(rscratch, 0));
 2264   }
 2265 }
 2266 
 2267 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2268   assert(rscratch != noreg || always_reachable(src), "missing");
 2269 
 2270   if (reachable(src)) {
 2271     Assembler::movdqa(dst, as_Address(src));
 2272   } else {
 2273     lea(rscratch, src);
 2274     Assembler::movdqa(dst, Address(rscratch, 0));
 2275   }
 2276 }
 2277 
 2278 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2279   assert(rscratch != noreg || always_reachable(src), "missing");
 2280 
 2281   if (reachable(src)) {
 2282     Assembler::movsd(dst, as_Address(src));
 2283   } else {
 2284     lea(rscratch, src);
 2285     Assembler::movsd(dst, Address(rscratch, 0));
 2286   }
 2287 }
 2288 
 2289 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2290   assert(rscratch != noreg || always_reachable(src), "missing");
 2291 
 2292   if (reachable(src)) {
 2293     Assembler::movss(dst, as_Address(src));
 2294   } else {
 2295     lea(rscratch, src);
 2296     Assembler::movss(dst, Address(rscratch, 0));
 2297   }
 2298 }
 2299 
 2300 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2301   assert(rscratch != noreg || always_reachable(src), "missing");
 2302 
 2303   if (reachable(src)) {
 2304     Assembler::movddup(dst, as_Address(src));
 2305   } else {
 2306     lea(rscratch, src);
 2307     Assembler::movddup(dst, Address(rscratch, 0));
 2308   }
 2309 }
 2310 
 2311 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2312   assert(rscratch != noreg || always_reachable(src), "missing");
 2313 
 2314   if (reachable(src)) {
 2315     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2316   } else {
 2317     lea(rscratch, src);
 2318     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2319   }
 2320 }
 2321 
 2322 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2323   assert(rscratch != noreg || always_reachable(src), "missing");
 2324 
 2325   if (reachable(src)) {
 2326     Assembler::mulsd(dst, as_Address(src));
 2327   } else {
 2328     lea(rscratch, src);
 2329     Assembler::mulsd(dst, Address(rscratch, 0));
 2330   }
 2331 }
 2332 
 2333 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2334   assert(rscratch != noreg || always_reachable(src), "missing");
 2335 
 2336   if (reachable(src)) {
 2337     Assembler::mulss(dst, as_Address(src));
 2338   } else {
 2339     lea(rscratch, src);
 2340     Assembler::mulss(dst, Address(rscratch, 0));
 2341   }
 2342 }
 2343 
 2344 void MacroAssembler::null_check(Register reg, int offset) {
 2345   if (needs_explicit_null_check(offset)) {
 2346     // provoke OS null exception if reg is null by
 2347     // accessing M[reg] w/o changing any (non-CC) registers
 2348     // NOTE: cmpl is plenty here to provoke a segv
 2349     cmpptr(rax, Address(reg, 0));
 2350     // Note: should probably use testl(rax, Address(reg, 0));
 2351     //       may be shorter code (however, this version of
 2352     //       testl needs to be implemented first)
 2353   } else {
 2354     // nothing to do, (later) access of M[reg + offset]
 2355     // will provoke OS null exception if reg is null
 2356   }
 2357 }
 2358 
 2359 void MacroAssembler::os_breakpoint() {
 2360   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2361   // (e.g., MSVC can't call ps() otherwise)
 2362   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2363 }
 2364 
 2365 void MacroAssembler::unimplemented(const char* what) {
 2366   const char* buf = nullptr;
 2367   {
 2368     ResourceMark rm;
 2369     stringStream ss;
 2370     ss.print("unimplemented: %s", what);
 2371     buf = code_string(ss.as_string());
 2372   }
 2373   stop(buf);
 2374 }
 2375 
 2376 #define XSTATE_BV 0x200
 2377 
 2378 void MacroAssembler::pop_CPU_state() {
 2379   pop_FPU_state();
 2380   pop_IU_state();
 2381 }
 2382 
 2383 void MacroAssembler::pop_FPU_state() {
 2384   fxrstor(Address(rsp, 0));
 2385   addptr(rsp, FPUStateSizeInWords * wordSize);
 2386 }
 2387 
 2388 void MacroAssembler::pop_IU_state() {
 2389   popa();
 2390   addq(rsp, 8);
 2391   popf();
 2392 }
 2393 
 2394 // Save Integer and Float state
 2395 // Warning: Stack must be 16 byte aligned (64bit)
 2396 void MacroAssembler::push_CPU_state() {
 2397   push_IU_state();
 2398   push_FPU_state();
 2399 }
 2400 
 2401 void MacroAssembler::push_FPU_state() {
 2402   subptr(rsp, FPUStateSizeInWords * wordSize);
 2403   fxsave(Address(rsp, 0));
 2404 }
 2405 
 2406 void MacroAssembler::push_IU_state() {
 2407   // Push flags first because pusha kills them
 2408   pushf();
 2409   // Make sure rsp stays 16-byte aligned
 2410   subq(rsp, 8);
 2411   pusha();
 2412 }
 2413 
 2414 void MacroAssembler::push_cont_fastpath() {
 2415   if (!Continuations::enabled()) return;
 2416 
 2417   Label L_done;
 2418   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2419   jccb(Assembler::belowEqual, L_done);
 2420   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2421   bind(L_done);
 2422 }
 2423 
 2424 void MacroAssembler::pop_cont_fastpath() {
 2425   if (!Continuations::enabled()) return;
 2426 
 2427   Label L_done;
 2428   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2429   jccb(Assembler::below, L_done);
 2430   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2431   bind(L_done);
 2432 }
 2433 
 2434 #ifdef ASSERT
 2435 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2436   Label no_cont;
 2437   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2438   testl(cont, cont);
 2439   jcc(Assembler::zero, no_cont);
 2440   stop(name);
 2441   bind(no_cont);
 2442 }
 2443 #endif
 2444 
 2445 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2446   // we must set sp to zero to clear frame
 2447   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2448   // must clear fp, so that compiled frames are not confused; it is
 2449   // possible that we need it only for debugging
 2450   if (clear_fp) {
 2451     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2452   }
 2453   // Always clear the pc because it could have been set by make_walkable()
 2454   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2455   vzeroupper();
 2456 }
 2457 
 2458 void MacroAssembler::round_to(Register reg, int modulus) {
 2459   addptr(reg, modulus - 1);
 2460   andptr(reg, -modulus);
 2461 }
 2462 
 2463 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2464   if (at_return) {
 2465     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2466     // we may safely use rsp instead to perform the stack watermark check.
 2467     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2468     jcc(Assembler::above, slow_path);
 2469     return;
 2470   }
 2471   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2472   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2473 }
 2474 
 2475 // Calls to C land
 2476 //
 2477 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2478 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2479 // has to be reset to 0. This is required to allow proper stack traversal.
 2480 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2481                                          Register last_java_fp,
 2482                                          address  last_java_pc,
 2483                                          Register rscratch) {
 2484   vzeroupper();
 2485   // determine last_java_sp register
 2486   if (!last_java_sp->is_valid()) {
 2487     last_java_sp = rsp;
 2488   }
 2489   // last_java_fp is optional
 2490   if (last_java_fp->is_valid()) {
 2491     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2492   }
 2493   // last_java_pc is optional
 2494   if (last_java_pc != nullptr) {
 2495     Address java_pc(r15_thread,
 2496                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2497     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2498   }
 2499   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2500 }
 2501 
 2502 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2503                                          Register last_java_fp,
 2504                                          Label &L,
 2505                                          Register scratch) {
 2506   lea(scratch, L);
 2507   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2508   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2509 }
 2510 
 2511 void MacroAssembler::shlptr(Register dst, int imm8) {
 2512   shlq(dst, imm8);
 2513 }
 2514 
 2515 void MacroAssembler::shrptr(Register dst, int imm8) {
 2516   shrq(dst, imm8);
 2517 }
 2518 
 2519 void MacroAssembler::sign_extend_byte(Register reg) {
 2520   movsbl(reg, reg); // movsxb
 2521 }
 2522 
 2523 void MacroAssembler::sign_extend_short(Register reg) {
 2524   movswl(reg, reg); // movsxw
 2525 }
 2526 
 2527 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2528   if (imm32 >= 0 && is8bit(imm32)) {
 2529     testb(dst, imm32);
 2530   } else {
 2531     Assembler::testl(dst, imm32);
 2532   }
 2533 }
 2534 
 2535 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2536   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2537     testb(dst, imm32);
 2538   } else {
 2539     Assembler::testl(dst, imm32);
 2540   }
 2541 }
 2542 
 2543 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2544   assert(always_reachable(src), "Address should be reachable");
 2545   testl(dst, as_Address(src));
 2546 }
 2547 
 2548 void MacroAssembler::testq(Address dst, int32_t imm32) {
 2549   if (imm32 >= 0) {
 2550     testl(dst, imm32);
 2551   } else {
 2552     Assembler::testq(dst, imm32);
 2553   }
 2554 }
 2555 
 2556 void MacroAssembler::testq(Register dst, int32_t imm32) {
 2557   if (imm32 >= 0) {
 2558     testl(dst, imm32);
 2559   } else {
 2560     Assembler::testq(dst, imm32);
 2561   }
 2562 }
 2563 
 2564 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2565   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2566   Assembler::pcmpeqb(dst, src);
 2567 }
 2568 
 2569 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2570   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2571   Assembler::pcmpeqw(dst, src);
 2572 }
 2573 
 2574 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2575   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2576   Assembler::pcmpestri(dst, src, imm8);
 2577 }
 2578 
 2579 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2580   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2581   Assembler::pcmpestri(dst, src, imm8);
 2582 }
 2583 
 2584 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2585   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2586   Assembler::pmovzxbw(dst, src);
 2587 }
 2588 
 2589 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2590   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2591   Assembler::pmovzxbw(dst, src);
 2592 }
 2593 
 2594 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2595   assert((src->encoding() < 16),"XMM register should be 0-15");
 2596   Assembler::pmovmskb(dst, src);
 2597 }
 2598 
 2599 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2600   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2601   Assembler::ptest(dst, src);
 2602 }
 2603 
 2604 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2605   assert(rscratch != noreg || always_reachable(src), "missing");
 2606 
 2607   if (reachable(src)) {
 2608     Assembler::sqrtss(dst, as_Address(src));
 2609   } else {
 2610     lea(rscratch, src);
 2611     Assembler::sqrtss(dst, Address(rscratch, 0));
 2612   }
 2613 }
 2614 
 2615 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2616   assert(rscratch != noreg || always_reachable(src), "missing");
 2617 
 2618   if (reachable(src)) {
 2619     Assembler::subsd(dst, as_Address(src));
 2620   } else {
 2621     lea(rscratch, src);
 2622     Assembler::subsd(dst, Address(rscratch, 0));
 2623   }
 2624 }
 2625 
 2626 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2627   assert(rscratch != noreg || always_reachable(src), "missing");
 2628 
 2629   if (reachable(src)) {
 2630     Assembler::roundsd(dst, as_Address(src), rmode);
 2631   } else {
 2632     lea(rscratch, src);
 2633     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2634   }
 2635 }
 2636 
 2637 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2638   assert(rscratch != noreg || always_reachable(src), "missing");
 2639 
 2640   if (reachable(src)) {
 2641     Assembler::subss(dst, as_Address(src));
 2642   } else {
 2643     lea(rscratch, src);
 2644     Assembler::subss(dst, Address(rscratch, 0));
 2645   }
 2646 }
 2647 
 2648 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2649   assert(rscratch != noreg || always_reachable(src), "missing");
 2650 
 2651   if (reachable(src)) {
 2652     Assembler::ucomisd(dst, as_Address(src));
 2653   } else {
 2654     lea(rscratch, src);
 2655     Assembler::ucomisd(dst, Address(rscratch, 0));
 2656   }
 2657 }
 2658 
 2659 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2660   assert(rscratch != noreg || always_reachable(src), "missing");
 2661 
 2662   if (reachable(src)) {
 2663     Assembler::ucomiss(dst, as_Address(src));
 2664   } else {
 2665     lea(rscratch, src);
 2666     Assembler::ucomiss(dst, Address(rscratch, 0));
 2667   }
 2668 }
 2669 
 2670 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2671   assert(rscratch != noreg || always_reachable(src), "missing");
 2672 
 2673   // Used in sign-bit flipping with aligned address.
 2674   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2675 
 2676   if (UseAVX > 2 &&
 2677       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2678       (dst->encoding() >= 16)) {
 2679     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2680   } else if (reachable(src)) {
 2681     Assembler::xorpd(dst, as_Address(src));
 2682   } else {
 2683     lea(rscratch, src);
 2684     Assembler::xorpd(dst, Address(rscratch, 0));
 2685   }
 2686 }
 2687 
 2688 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2689   if (UseAVX > 2 &&
 2690       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2691       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2692     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2693   } else {
 2694     Assembler::xorpd(dst, src);
 2695   }
 2696 }
 2697 
 2698 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2699   if (UseAVX > 2 &&
 2700       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2701       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2702     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2703   } else {
 2704     Assembler::xorps(dst, src);
 2705   }
 2706 }
 2707 
 2708 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2709   assert(rscratch != noreg || always_reachable(src), "missing");
 2710 
 2711   // Used in sign-bit flipping with aligned address.
 2712   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2713 
 2714   if (UseAVX > 2 &&
 2715       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2716       (dst->encoding() >= 16)) {
 2717     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2718   } else if (reachable(src)) {
 2719     Assembler::xorps(dst, as_Address(src));
 2720   } else {
 2721     lea(rscratch, src);
 2722     Assembler::xorps(dst, Address(rscratch, 0));
 2723   }
 2724 }
 2725 
 2726 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2727   assert(rscratch != noreg || always_reachable(src), "missing");
 2728 
 2729   // Used in sign-bit flipping with aligned address.
 2730   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2731   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2732   if (reachable(src)) {
 2733     Assembler::pshufb(dst, as_Address(src));
 2734   } else {
 2735     lea(rscratch, src);
 2736     Assembler::pshufb(dst, Address(rscratch, 0));
 2737   }
 2738 }
 2739 
 2740 // AVX 3-operands instructions
 2741 
 2742 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2743   assert(rscratch != noreg || always_reachable(src), "missing");
 2744 
 2745   if (reachable(src)) {
 2746     vaddsd(dst, nds, as_Address(src));
 2747   } else {
 2748     lea(rscratch, src);
 2749     vaddsd(dst, nds, Address(rscratch, 0));
 2750   }
 2751 }
 2752 
 2753 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2754   assert(rscratch != noreg || always_reachable(src), "missing");
 2755 
 2756   if (reachable(src)) {
 2757     vaddss(dst, nds, as_Address(src));
 2758   } else {
 2759     lea(rscratch, src);
 2760     vaddss(dst, nds, Address(rscratch, 0));
 2761   }
 2762 }
 2763 
 2764 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2765   assert(UseAVX > 0, "requires some form of AVX");
 2766   assert(rscratch != noreg || always_reachable(src), "missing");
 2767 
 2768   if (reachable(src)) {
 2769     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2770   } else {
 2771     lea(rscratch, src);
 2772     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2773   }
 2774 }
 2775 
 2776 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2777   assert(UseAVX > 0, "requires some form of AVX");
 2778   assert(rscratch != noreg || always_reachable(src), "missing");
 2779 
 2780   if (reachable(src)) {
 2781     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2782   } else {
 2783     lea(rscratch, src);
 2784     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2785   }
 2786 }
 2787 
 2788 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2789   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2790   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2791 
 2792   vandps(dst, nds, negate_field, vector_len, rscratch);
 2793 }
 2794 
 2795 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2796   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2797   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2798 
 2799   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2800 }
 2801 
 2802 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2803   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2804   Assembler::vpaddb(dst, nds, src, vector_len);
 2805 }
 2806 
 2807 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2808   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2809   Assembler::vpaddb(dst, nds, src, vector_len);
 2810 }
 2811 
 2812 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2813   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2814   Assembler::vpaddw(dst, nds, src, vector_len);
 2815 }
 2816 
 2817 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2818   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2819   Assembler::vpaddw(dst, nds, src, vector_len);
 2820 }
 2821 
 2822 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2823   assert(rscratch != noreg || always_reachable(src), "missing");
 2824 
 2825   if (reachable(src)) {
 2826     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 2827   } else {
 2828     lea(rscratch, src);
 2829     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 2830   }
 2831 }
 2832 
 2833 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2834   assert(rscratch != noreg || always_reachable(src), "missing");
 2835 
 2836   if (reachable(src)) {
 2837     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 2838   } else {
 2839     lea(rscratch, src);
 2840     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 2841   }
 2842 }
 2843 
 2844 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2845   assert(rscratch != noreg || always_reachable(src), "missing");
 2846 
 2847   if (reachable(src)) {
 2848     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 2849   } else {
 2850     lea(rscratch, src);
 2851     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 2852   }
 2853 }
 2854 
 2855 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2856   assert(rscratch != noreg || always_reachable(src), "missing");
 2857 
 2858   if (reachable(src)) {
 2859     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 2860   } else {
 2861     lea(rscratch, src);
 2862     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 2863   }
 2864 }
 2865 
 2866 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2867   assert(rscratch != noreg || always_reachable(src), "missing");
 2868 
 2869   if (reachable(src)) {
 2870     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 2871   } else {
 2872     lea(rscratch, src);
 2873     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 2874   }
 2875 }
 2876 
 2877 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2878   assert(rscratch != noreg || always_reachable(src), "missing");
 2879 
 2880   if (reachable(src)) {
 2881     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 2882   } else {
 2883     lea(rscratch, src);
 2884     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 2885   }
 2886 }
 2887 
 2888 // Vector float blend
 2889 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2890 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2891   // WARN: Allow dst == (src1|src2), mask == scratch
 2892   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2893                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2894   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 2895   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2896   if (blend_emulation && scratch_available && dst_available) {
 2897     if (compute_mask) {
 2898       vpsrad(scratch, mask, 32, vector_len);
 2899       mask = scratch;
 2900     }
 2901     if (dst == src1) {
 2902       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 2903       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 2904     } else {
 2905       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 2906       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 2907     }
 2908     vpor(dst, dst, scratch, vector_len);
 2909   } else {
 2910     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 2911   }
 2912 }
 2913 
 2914 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2915 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2916   // WARN: Allow dst == (src1|src2), mask == scratch
 2917   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2918                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2919   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 2920   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2921   if (blend_emulation && scratch_available && dst_available) {
 2922     if (compute_mask) {
 2923       vpxor(scratch, scratch, scratch, vector_len);
 2924       vpcmpgtq(scratch, scratch, mask, vector_len);
 2925       mask = scratch;
 2926     }
 2927     if (dst == src1) {
 2928       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 2929       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 2930     } else {
 2931       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 2932       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 2933     }
 2934     vpor(dst, dst, scratch, vector_len);
 2935   } else {
 2936     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 2937   }
 2938 }
 2939 
 2940 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2941   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2942   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 2943 }
 2944 
 2945 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 2946   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2947   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 2948 }
 2949 
 2950 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2951   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2952   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 2953 }
 2954 
 2955 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2956   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2957   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 2958 }
 2959 
 2960 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2961   assert(rscratch != noreg || always_reachable(src), "missing");
 2962 
 2963   if (reachable(src)) {
 2964     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 2965   } else {
 2966     lea(rscratch, src);
 2967     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 2968   }
 2969 }
 2970 
 2971 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 2972                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 2973   assert(rscratch != noreg || always_reachable(src), "missing");
 2974 
 2975   if (reachable(src)) {
 2976     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 2977   } else {
 2978     lea(rscratch, src);
 2979     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 2980   }
 2981 }
 2982 
 2983 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 2984                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 2985   assert(rscratch != noreg || always_reachable(src), "missing");
 2986 
 2987   if (reachable(src)) {
 2988     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 2989   } else {
 2990     lea(rscratch, src);
 2991     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 2992   }
 2993 }
 2994 
 2995 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 2996                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 2997   assert(rscratch != noreg || always_reachable(src), "missing");
 2998 
 2999   if (reachable(src)) {
 3000     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3001   } else {
 3002     lea(rscratch, src);
 3003     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3004   }
 3005 }
 3006 
 3007 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3008                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3009   assert(rscratch != noreg || always_reachable(src), "missing");
 3010 
 3011   if (reachable(src)) {
 3012     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3013   } else {
 3014     lea(rscratch, src);
 3015     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3016   }
 3017 }
 3018 
 3019 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3020   if (width == Assembler::Q) {
 3021     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3022   } else {
 3023     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3024   }
 3025 }
 3026 
 3027 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3028   int eq_cond_enc = 0x29;
 3029   int gt_cond_enc = 0x37;
 3030   if (width != Assembler::Q) {
 3031     eq_cond_enc = 0x74 + width;
 3032     gt_cond_enc = 0x64 + width;
 3033   }
 3034   switch (cond) {
 3035   case eq:
 3036     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3037     break;
 3038   case neq:
 3039     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3040     vallones(xtmp, vector_len);
 3041     vpxor(dst, xtmp, dst, vector_len);
 3042     break;
 3043   case le:
 3044     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3045     vallones(xtmp, vector_len);
 3046     vpxor(dst, xtmp, dst, vector_len);
 3047     break;
 3048   case nlt:
 3049     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3050     vallones(xtmp, vector_len);
 3051     vpxor(dst, xtmp, dst, vector_len);
 3052     break;
 3053   case lt:
 3054     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3055     break;
 3056   case nle:
 3057     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3058     break;
 3059   default:
 3060     assert(false, "Should not reach here");
 3061   }
 3062 }
 3063 
 3064 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3065   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3066   Assembler::vpmovzxbw(dst, src, vector_len);
 3067 }
 3068 
 3069 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3070   assert((src->encoding() < 16),"XMM register should be 0-15");
 3071   Assembler::vpmovmskb(dst, src, vector_len);
 3072 }
 3073 
 3074 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3075   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3076   Assembler::vpmullw(dst, nds, src, vector_len);
 3077 }
 3078 
 3079 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3080   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3081   Assembler::vpmullw(dst, nds, src, vector_len);
 3082 }
 3083 
 3084 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3085   assert((UseAVX > 0), "AVX support is needed");
 3086   assert(rscratch != noreg || always_reachable(src), "missing");
 3087 
 3088   if (reachable(src)) {
 3089     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3090   } else {
 3091     lea(rscratch, src);
 3092     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3093   }
 3094 }
 3095 
 3096 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3097   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3098   Assembler::vpsubb(dst, nds, src, vector_len);
 3099 }
 3100 
 3101 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3102   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3103   Assembler::vpsubb(dst, nds, src, vector_len);
 3104 }
 3105 
 3106 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3107   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3108   Assembler::vpsubw(dst, nds, src, vector_len);
 3109 }
 3110 
 3111 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3112   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3113   Assembler::vpsubw(dst, nds, src, vector_len);
 3114 }
 3115 
 3116 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3117   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3118   Assembler::vpsraw(dst, nds, shift, vector_len);
 3119 }
 3120 
 3121 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3122   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3123   Assembler::vpsraw(dst, nds, shift, vector_len);
 3124 }
 3125 
 3126 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3127   assert(UseAVX > 2,"");
 3128   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3129      vector_len = 2;
 3130   }
 3131   Assembler::evpsraq(dst, nds, shift, vector_len);
 3132 }
 3133 
 3134 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3135   assert(UseAVX > 2,"");
 3136   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3137      vector_len = 2;
 3138   }
 3139   Assembler::evpsraq(dst, nds, shift, vector_len);
 3140 }
 3141 
 3142 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3143   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3144   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3145 }
 3146 
 3147 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3148   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3149   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3150 }
 3151 
 3152 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3153   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3154   Assembler::vpsllw(dst, nds, shift, vector_len);
 3155 }
 3156 
 3157 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3158   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3159   Assembler::vpsllw(dst, nds, shift, vector_len);
 3160 }
 3161 
 3162 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3163   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3164   Assembler::vptest(dst, src);
 3165 }
 3166 
 3167 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3168   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3169   Assembler::punpcklbw(dst, src);
 3170 }
 3171 
 3172 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3173   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3174   Assembler::pshufd(dst, src, mode);
 3175 }
 3176 
 3177 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3178   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3179   Assembler::pshuflw(dst, src, mode);
 3180 }
 3181 
 3182 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3183   assert(rscratch != noreg || always_reachable(src), "missing");
 3184 
 3185   if (reachable(src)) {
 3186     vandpd(dst, nds, as_Address(src), vector_len);
 3187   } else {
 3188     lea(rscratch, src);
 3189     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3190   }
 3191 }
 3192 
 3193 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3194   assert(rscratch != noreg || always_reachable(src), "missing");
 3195 
 3196   if (reachable(src)) {
 3197     vandps(dst, nds, as_Address(src), vector_len);
 3198   } else {
 3199     lea(rscratch, src);
 3200     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3201   }
 3202 }
 3203 
 3204 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3205                             bool merge, int vector_len, Register rscratch) {
 3206   assert(rscratch != noreg || always_reachable(src), "missing");
 3207 
 3208   if (reachable(src)) {
 3209     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3210   } else {
 3211     lea(rscratch, src);
 3212     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3213   }
 3214 }
 3215 
 3216 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3217   assert(rscratch != noreg || always_reachable(src), "missing");
 3218 
 3219   if (reachable(src)) {
 3220     vdivsd(dst, nds, as_Address(src));
 3221   } else {
 3222     lea(rscratch, src);
 3223     vdivsd(dst, nds, Address(rscratch, 0));
 3224   }
 3225 }
 3226 
 3227 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3228   assert(rscratch != noreg || always_reachable(src), "missing");
 3229 
 3230   if (reachable(src)) {
 3231     vdivss(dst, nds, as_Address(src));
 3232   } else {
 3233     lea(rscratch, src);
 3234     vdivss(dst, nds, Address(rscratch, 0));
 3235   }
 3236 }
 3237 
 3238 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3239   assert(rscratch != noreg || always_reachable(src), "missing");
 3240 
 3241   if (reachable(src)) {
 3242     vmulsd(dst, nds, as_Address(src));
 3243   } else {
 3244     lea(rscratch, src);
 3245     vmulsd(dst, nds, Address(rscratch, 0));
 3246   }
 3247 }
 3248 
 3249 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3250   assert(rscratch != noreg || always_reachable(src), "missing");
 3251 
 3252   if (reachable(src)) {
 3253     vmulss(dst, nds, as_Address(src));
 3254   } else {
 3255     lea(rscratch, src);
 3256     vmulss(dst, nds, Address(rscratch, 0));
 3257   }
 3258 }
 3259 
 3260 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3261   assert(rscratch != noreg || always_reachable(src), "missing");
 3262 
 3263   if (reachable(src)) {
 3264     vsubsd(dst, nds, as_Address(src));
 3265   } else {
 3266     lea(rscratch, src);
 3267     vsubsd(dst, nds, Address(rscratch, 0));
 3268   }
 3269 }
 3270 
 3271 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3272   assert(rscratch != noreg || always_reachable(src), "missing");
 3273 
 3274   if (reachable(src)) {
 3275     vsubss(dst, nds, as_Address(src));
 3276   } else {
 3277     lea(rscratch, src);
 3278     vsubss(dst, nds, Address(rscratch, 0));
 3279   }
 3280 }
 3281 
 3282 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3283   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3284   assert(rscratch != noreg || always_reachable(src), "missing");
 3285 
 3286   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3287 }
 3288 
 3289 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3290   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3291   assert(rscratch != noreg || always_reachable(src), "missing");
 3292 
 3293   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3294 }
 3295 
 3296 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3297   assert(rscratch != noreg || always_reachable(src), "missing");
 3298 
 3299   if (reachable(src)) {
 3300     vxorpd(dst, nds, as_Address(src), vector_len);
 3301   } else {
 3302     lea(rscratch, src);
 3303     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3304   }
 3305 }
 3306 
 3307 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3308   assert(rscratch != noreg || always_reachable(src), "missing");
 3309 
 3310   if (reachable(src)) {
 3311     vxorps(dst, nds, as_Address(src), vector_len);
 3312   } else {
 3313     lea(rscratch, src);
 3314     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3315   }
 3316 }
 3317 
 3318 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3319   assert(rscratch != noreg || always_reachable(src), "missing");
 3320 
 3321   if (UseAVX > 1 || (vector_len < 1)) {
 3322     if (reachable(src)) {
 3323       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3324     } else {
 3325       lea(rscratch, src);
 3326       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3327     }
 3328   } else {
 3329     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3330   }
 3331 }
 3332 
 3333 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3334   assert(rscratch != noreg || always_reachable(src), "missing");
 3335 
 3336   if (reachable(src)) {
 3337     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3338   } else {
 3339     lea(rscratch, src);
 3340     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3341   }
 3342 }
 3343 
 3344 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3345   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3346   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3347   // The inverted mask is sign-extended
 3348   andptr(possibly_non_local, inverted_mask);
 3349 }
 3350 
 3351 void MacroAssembler::resolve_jobject(Register value,
 3352                                      Register tmp) {
 3353   Register thread = r15_thread;
 3354   assert_different_registers(value, thread, tmp);
 3355   Label done, tagged, weak_tagged;
 3356   testptr(value, value);
 3357   jcc(Assembler::zero, done);           // Use null as-is.
 3358   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3359   jcc(Assembler::notZero, tagged);
 3360 
 3361   // Resolve local handle
 3362   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3363   verify_oop(value);
 3364   jmp(done);
 3365 
 3366   bind(tagged);
 3367   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3368   jcc(Assembler::notZero, weak_tagged);
 3369 
 3370   // Resolve global handle
 3371   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3372   verify_oop(value);
 3373   jmp(done);
 3374 
 3375   bind(weak_tagged);
 3376   // Resolve jweak.
 3377   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3378                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3379   verify_oop(value);
 3380 
 3381   bind(done);
 3382 }
 3383 
 3384 void MacroAssembler::resolve_global_jobject(Register value,
 3385                                             Register tmp) {
 3386   Register thread = r15_thread;
 3387   assert_different_registers(value, thread, tmp);
 3388   Label done;
 3389 
 3390   testptr(value, value);
 3391   jcc(Assembler::zero, done);           // Use null as-is.
 3392 
 3393 #ifdef ASSERT
 3394   {
 3395     Label valid_global_tag;
 3396     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3397     jcc(Assembler::notZero, valid_global_tag);
 3398     stop("non global jobject using resolve_global_jobject");
 3399     bind(valid_global_tag);
 3400   }
 3401 #endif
 3402 
 3403   // Resolve global handle
 3404   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3405   verify_oop(value);
 3406 
 3407   bind(done);
 3408 }
 3409 
 3410 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3411   subq(dst, imm32);
 3412 }
 3413 
 3414 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3415 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3416   subq_imm32(dst, imm32);
 3417 }
 3418 
 3419 void MacroAssembler::subptr(Register dst, Register src) {
 3420   subq(dst, src);
 3421 }
 3422 
 3423 // C++ bool manipulation
 3424 void MacroAssembler::testbool(Register dst) {
 3425   if(sizeof(bool) == 1)
 3426     testb(dst, 0xff);
 3427   else if(sizeof(bool) == 2) {
 3428     // testw implementation needed for two byte bools
 3429     ShouldNotReachHere();
 3430   } else if(sizeof(bool) == 4)
 3431     testl(dst, dst);
 3432   else
 3433     // unsupported
 3434     ShouldNotReachHere();
 3435 }
 3436 
 3437 void MacroAssembler::testptr(Register dst, Register src) {
 3438   testq(dst, src);
 3439 }
 3440 
 3441 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3442 void MacroAssembler::tlab_allocate(Register obj,
 3443                                    Register var_size_in_bytes,
 3444                                    int con_size_in_bytes,
 3445                                    Register t1,
 3446                                    Register t2,
 3447                                    Label& slow_case) {
 3448   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3449   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3450 }
 3451 
 3452 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3453   RegSet regs;
 3454   regs += RegSet::of(rax, rcx, rdx);
 3455 #ifndef _WINDOWS
 3456   regs += RegSet::of(rsi, rdi);
 3457 #endif
 3458   regs += RegSet::range(r8, r11);
 3459   if (UseAPX) {
 3460     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3461   }
 3462   return regs;
 3463 }
 3464 
 3465 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3466   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3467 #if defined(_WINDOWS)
 3468   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3469   if (num_xmm_registers > 16) {
 3470      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3471   }
 3472   return result;
 3473 #else
 3474   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3475 #endif
 3476 }
 3477 
 3478 // C1 only ever uses the first double/float of the XMM register.
 3479 static int xmm_save_size() { return sizeof(double); }
 3480 
 3481 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3482   masm->movdbl(Address(rsp, offset), reg);
 3483 }
 3484 
 3485 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3486   masm->movdbl(reg, Address(rsp, offset));
 3487 }
 3488 
 3489 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3490                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3491 
 3492   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3493                          StackAlignmentInBytes);
 3494   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3495 
 3496   return gp_area_size + xmm_area_size;
 3497 }
 3498 
 3499 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3500   block_comment("push_call_clobbered_registers start");
 3501   // Regular registers
 3502   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3503 
 3504   int gp_area_size;
 3505   int xmm_area_size;
 3506   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3507                                                gp_area_size, xmm_area_size);
 3508   subptr(rsp, total_save_size);
 3509 
 3510   push_set(gp_registers_to_push, 0);
 3511 
 3512   if (save_fpu) {
 3513     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3514   }
 3515 
 3516   block_comment("push_call_clobbered_registers end");
 3517 }
 3518 
 3519 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3520   block_comment("pop_call_clobbered_registers start");
 3521 
 3522   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 3523 
 3524   int gp_area_size;
 3525   int xmm_area_size;
 3526   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 3527                                                gp_area_size, xmm_area_size);
 3528 
 3529   if (restore_fpu) {
 3530     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3531   }
 3532 
 3533   pop_set(gp_registers_to_pop, 0);
 3534 
 3535   addptr(rsp, total_save_size);
 3536 
 3537   vzeroupper();
 3538 
 3539   block_comment("pop_call_clobbered_registers end");
 3540 }
 3541 
 3542 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3543   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3544   int spill_offset = offset;
 3545 
 3546   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3547     save_xmm_register(this, spill_offset, *it);
 3548     spill_offset += xmm_save_size();
 3549   }
 3550 }
 3551 
 3552 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3553   int restore_size = set.size() * xmm_save_size();
 3554   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3555 
 3556   int restore_offset = offset + restore_size - xmm_save_size();
 3557 
 3558   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3559     restore_xmm_register(this, restore_offset, *it);
 3560     restore_offset -= xmm_save_size();
 3561   }
 3562 }
 3563 
 3564 void MacroAssembler::push_set(RegSet set, int offset) {
 3565   int spill_offset;
 3566   if (offset == -1) {
 3567     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3568     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3569     subptr(rsp, aligned_size);
 3570     spill_offset = 0;
 3571   } else {
 3572     spill_offset = offset;
 3573   }
 3574 
 3575   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3576     movptr(Address(rsp, spill_offset), *it);
 3577     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3578   }
 3579 }
 3580 
 3581 void MacroAssembler::pop_set(RegSet set, int offset) {
 3582 
 3583   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3584   int restore_size = set.size() * gp_reg_size;
 3585   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3586 
 3587   int restore_offset;
 3588   if (offset == -1) {
 3589     restore_offset = restore_size - gp_reg_size;
 3590   } else {
 3591     restore_offset = offset + restore_size - gp_reg_size;
 3592   }
 3593   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3594     movptr(*it, Address(rsp, restore_offset));
 3595     restore_offset -= gp_reg_size;
 3596   }
 3597 
 3598   if (offset == -1) {
 3599     addptr(rsp, aligned_size);
 3600   }
 3601 }
 3602 
 3603 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3604 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3605   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3606   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3607   Label done;
 3608 
 3609   testptr(length_in_bytes, length_in_bytes);
 3610   jcc(Assembler::zero, done);
 3611 
 3612   // initialize topmost word, divide index by 2, check if odd and test if zero
 3613   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3614 #ifdef ASSERT
 3615   {
 3616     Label L;
 3617     testptr(length_in_bytes, BytesPerWord - 1);
 3618     jcc(Assembler::zero, L);
 3619     stop("length must be a multiple of BytesPerWord");
 3620     bind(L);
 3621   }
 3622 #endif
 3623   Register index = length_in_bytes;
 3624   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3625   if (UseIncDec) {
 3626     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3627   } else {
 3628     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3629     shrptr(index, 1);
 3630   }
 3631 
 3632   // initialize remaining object fields: index is a multiple of 2 now
 3633   {
 3634     Label loop;
 3635     bind(loop);
 3636     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3637     decrement(index);
 3638     jcc(Assembler::notZero, loop);
 3639   }
 3640 
 3641   bind(done);
 3642 }
 3643 
 3644 // Look up the method for a megamorphic invokeinterface call.
 3645 // The target method is determined by <intf_klass, itable_index>.
 3646 // The receiver klass is in recv_klass.
 3647 // On success, the result will be in method_result, and execution falls through.
 3648 // On failure, execution transfers to the given label.
 3649 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3650                                              Register intf_klass,
 3651                                              RegisterOrConstant itable_index,
 3652                                              Register method_result,
 3653                                              Register scan_temp,
 3654                                              Label& L_no_such_interface,
 3655                                              bool return_method) {
 3656   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3657   assert_different_registers(method_result, intf_klass, scan_temp);
 3658   assert(recv_klass != method_result || !return_method,
 3659          "recv_klass can be destroyed when method isn't needed");
 3660 
 3661   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3662          "caller must use same register for non-constant itable index as for method");
 3663 
 3664   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3665   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3666   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3667   int scan_step   = itableOffsetEntry::size() * wordSize;
 3668   int vte_size    = vtableEntry::size_in_bytes();
 3669   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3670   assert(vte_size == wordSize, "else adjust times_vte_scale");
 3671 
 3672   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3673 
 3674   // Could store the aligned, prescaled offset in the klass.
 3675   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 3676 
 3677   if (return_method) {
 3678     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 3679     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3680     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 3681   }
 3682 
 3683   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 3684   //   if (scan->interface() == intf) {
 3685   //     result = (klass + scan->offset() + itable_index);
 3686   //   }
 3687   // }
 3688   Label search, found_method;
 3689 
 3690   for (int peel = 1; peel >= 0; peel--) {
 3691     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 3692     cmpptr(intf_klass, method_result);
 3693 
 3694     if (peel) {
 3695       jccb(Assembler::equal, found_method);
 3696     } else {
 3697       jccb(Assembler::notEqual, search);
 3698       // (invert the test to fall through to found_method...)
 3699     }
 3700 
 3701     if (!peel)  break;
 3702 
 3703     bind(search);
 3704 
 3705     // Check that the previous entry is non-null.  A null entry means that
 3706     // the receiver class doesn't implement the interface, and wasn't the
 3707     // same as when the caller was compiled.
 3708     testptr(method_result, method_result);
 3709     jcc(Assembler::zero, L_no_such_interface);
 3710     addptr(scan_temp, scan_step);
 3711   }
 3712 
 3713   bind(found_method);
 3714 
 3715   if (return_method) {
 3716     // Got a hit.
 3717     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 3718     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 3719   }
 3720 }
 3721 
 3722 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 3723 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 3724 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 3725 // The target method is determined by <holder_klass, itable_index>.
 3726 // The receiver klass is in recv_klass.
 3727 // On success, the result will be in method_result, and execution falls through.
 3728 // On failure, execution transfers to the given label.
 3729 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 3730                                                   Register holder_klass,
 3731                                                   Register resolved_klass,
 3732                                                   Register method_result,
 3733                                                   Register scan_temp,
 3734                                                   Register temp_reg2,
 3735                                                   Register receiver,
 3736                                                   int itable_index,
 3737                                                   Label& L_no_such_interface) {
 3738   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 3739   Register temp_itbl_klass = method_result;
 3740   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 3741 
 3742   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3743   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3744   int scan_step = itableOffsetEntry::size() * wordSize;
 3745   int vte_size = vtableEntry::size_in_bytes();
 3746   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 3747   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 3748   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3749   assert(vte_size == wordSize, "adjust times_vte_scale");
 3750 
 3751   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 3752 
 3753   // temp_itbl_klass = recv_klass.itable[0]
 3754   // scan_temp = &recv_klass.itable[0] + step
 3755   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3756   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 3757   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 3758   xorptr(temp_reg, temp_reg);
 3759 
 3760   // Initial checks:
 3761   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 3762   //   - if (itable[0] == 0), no such interface
 3763   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 3764   cmpptr(holder_klass, resolved_klass);
 3765   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 3766   testptr(temp_itbl_klass, temp_itbl_klass);
 3767   jccb(Assembler::zero, L_no_such_interface);
 3768   cmpptr(holder_klass, temp_itbl_klass);
 3769   jccb(Assembler::equal, L_holder_found);
 3770 
 3771   // Loop: Look for holder_klass record in itable
 3772   //   do {
 3773   //     tmp = itable[index];
 3774   //     index += step;
 3775   //     if (tmp == holder_klass) {
 3776   //       goto L_holder_found; // Found!
 3777   //     }
 3778   //   } while (tmp != 0);
 3779   //   goto L_no_such_interface // Not found.
 3780   Label L_scan_holder;
 3781   bind(L_scan_holder);
 3782     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3783     addptr(scan_temp, scan_step);
 3784     cmpptr(holder_klass, temp_itbl_klass);
 3785     jccb(Assembler::equal, L_holder_found);
 3786     testptr(temp_itbl_klass, temp_itbl_klass);
 3787     jccb(Assembler::notZero, L_scan_holder);
 3788 
 3789   jmpb(L_no_such_interface);
 3790 
 3791   // Loop: Look for resolved_class record in itable
 3792   //   do {
 3793   //     tmp = itable[index];
 3794   //     index += step;
 3795   //     if (tmp == holder_klass) {
 3796   //        // Also check if we have met a holder klass
 3797   //        holder_tmp = itable[index-step-ioffset];
 3798   //     }
 3799   //     if (tmp == resolved_klass) {
 3800   //        goto L_resolved_found;  // Found!
 3801   //     }
 3802   //   } while (tmp != 0);
 3803   //   goto L_no_such_interface // Not found.
 3804   //
 3805   Label L_loop_scan_resolved;
 3806   bind(L_loop_scan_resolved);
 3807     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3808     addptr(scan_temp, scan_step);
 3809     bind(L_loop_scan_resolved_entry);
 3810     cmpptr(holder_klass, temp_itbl_klass);
 3811     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3812     cmpptr(resolved_klass, temp_itbl_klass);
 3813     jccb(Assembler::equal, L_resolved_found);
 3814     testptr(temp_itbl_klass, temp_itbl_klass);
 3815     jccb(Assembler::notZero, L_loop_scan_resolved);
 3816 
 3817   jmpb(L_no_such_interface);
 3818 
 3819   Label L_ready;
 3820 
 3821   // See if we already have a holder klass. If not, go and scan for it.
 3822   bind(L_resolved_found);
 3823   testptr(temp_reg, temp_reg);
 3824   jccb(Assembler::zero, L_scan_holder);
 3825   jmpb(L_ready);
 3826 
 3827   bind(L_holder_found);
 3828   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3829 
 3830   // Finally, temp_reg contains holder_klass vtable offset
 3831   bind(L_ready);
 3832   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3833   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 3834     load_klass(scan_temp, receiver, noreg);
 3835     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3836   } else {
 3837     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3838   }
 3839 }
 3840 
 3841 
 3842 // virtual method calling
 3843 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 3844                                            RegisterOrConstant vtable_index,
 3845                                            Register method_result) {
 3846   const ByteSize base = Klass::vtable_start_offset();
 3847   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 3848   Address vtable_entry_addr(recv_klass,
 3849                             vtable_index, Address::times_ptr,
 3850                             base + vtableEntry::method_offset());
 3851   movptr(method_result, vtable_entry_addr);
 3852 }
 3853 
 3854 
 3855 void MacroAssembler::check_klass_subtype(Register sub_klass,
 3856                            Register super_klass,
 3857                            Register temp_reg,
 3858                            Label& L_success) {
 3859   Label L_failure;
 3860   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 3861   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 3862   bind(L_failure);
 3863 }
 3864 
 3865 
 3866 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 3867                                                    Register super_klass,
 3868                                                    Register temp_reg,
 3869                                                    Label* L_success,
 3870                                                    Label* L_failure,
 3871                                                    Label* L_slow_path,
 3872                                         RegisterOrConstant super_check_offset) {
 3873   assert_different_registers(sub_klass, super_klass, temp_reg);
 3874   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 3875   if (super_check_offset.is_register()) {
 3876     assert_different_registers(sub_klass, super_klass,
 3877                                super_check_offset.as_register());
 3878   } else if (must_load_sco) {
 3879     assert(temp_reg != noreg, "supply either a temp or a register offset");
 3880   }
 3881 
 3882   Label L_fallthrough;
 3883   int label_nulls = 0;
 3884   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 3885   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 3886   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 3887   assert(label_nulls <= 1, "at most one null in the batch");
 3888 
 3889   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 3890   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 3891   Address super_check_offset_addr(super_klass, sco_offset);
 3892 
 3893   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 3894   // range of a jccb.  If this routine grows larger, reconsider at
 3895   // least some of these.
 3896 #define local_jcc(assembler_cond, label)                                \
 3897   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 3898   else                             jcc( assembler_cond, label) /*omit semi*/
 3899 
 3900   // Hacked jmp, which may only be used just before L_fallthrough.
 3901 #define final_jmp(label)                                                \
 3902   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 3903   else                            jmp(label)                /*omit semi*/
 3904 
 3905   // If the pointers are equal, we are done (e.g., String[] elements).
 3906   // This self-check enables sharing of secondary supertype arrays among
 3907   // non-primary types such as array-of-interface.  Otherwise, each such
 3908   // type would need its own customized SSA.
 3909   // We move this check to the front of the fast path because many
 3910   // type checks are in fact trivially successful in this manner,
 3911   // so we get a nicely predicted branch right at the start of the check.
 3912   cmpptr(sub_klass, super_klass);
 3913   local_jcc(Assembler::equal, *L_success);
 3914 
 3915   // Check the supertype display:
 3916   if (must_load_sco) {
 3917     // Positive movl does right thing on LP64.
 3918     movl(temp_reg, super_check_offset_addr);
 3919     super_check_offset = RegisterOrConstant(temp_reg);
 3920   }
 3921   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 3922   cmpptr(super_klass, super_check_addr); // load displayed supertype
 3923 
 3924   // This check has worked decisively for primary supers.
 3925   // Secondary supers are sought in the super_cache ('super_cache_addr').
 3926   // (Secondary supers are interfaces and very deeply nested subtypes.)
 3927   // This works in the same check above because of a tricky aliasing
 3928   // between the super_cache and the primary super display elements.
 3929   // (The 'super_check_addr' can address either, as the case requires.)
 3930   // Note that the cache is updated below if it does not help us find
 3931   // what we need immediately.
 3932   // So if it was a primary super, we can just fail immediately.
 3933   // Otherwise, it's the slow path for us (no success at this point).
 3934 
 3935   if (super_check_offset.is_register()) {
 3936     local_jcc(Assembler::equal, *L_success);
 3937     cmpl(super_check_offset.as_register(), sc_offset);
 3938     if (L_failure == &L_fallthrough) {
 3939       local_jcc(Assembler::equal, *L_slow_path);
 3940     } else {
 3941       local_jcc(Assembler::notEqual, *L_failure);
 3942       final_jmp(*L_slow_path);
 3943     }
 3944   } else if (super_check_offset.as_constant() == sc_offset) {
 3945     // Need a slow path; fast failure is impossible.
 3946     if (L_slow_path == &L_fallthrough) {
 3947       local_jcc(Assembler::equal, *L_success);
 3948     } else {
 3949       local_jcc(Assembler::notEqual, *L_slow_path);
 3950       final_jmp(*L_success);
 3951     }
 3952   } else {
 3953     // No slow path; it's a fast decision.
 3954     if (L_failure == &L_fallthrough) {
 3955       local_jcc(Assembler::equal, *L_success);
 3956     } else {
 3957       local_jcc(Assembler::notEqual, *L_failure);
 3958       final_jmp(*L_success);
 3959     }
 3960   }
 3961 
 3962   bind(L_fallthrough);
 3963 
 3964 #undef local_jcc
 3965 #undef final_jmp
 3966 }
 3967 
 3968 
 3969 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 3970                                                           Register super_klass,
 3971                                                           Register temp_reg,
 3972                                                           Register temp2_reg,
 3973                                                           Label* L_success,
 3974                                                           Label* L_failure,
 3975                                                           bool set_cond_codes) {
 3976   assert_different_registers(sub_klass, super_klass, temp_reg);
 3977   if (temp2_reg != noreg)
 3978     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 3979 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 3980 
 3981   Label L_fallthrough;
 3982   int label_nulls = 0;
 3983   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 3984   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 3985   assert(label_nulls <= 1, "at most one null in the batch");
 3986 
 3987   // a couple of useful fields in sub_klass:
 3988   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 3989   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 3990   Address secondary_supers_addr(sub_klass, ss_offset);
 3991   Address super_cache_addr(     sub_klass, sc_offset);
 3992 
 3993   // Do a linear scan of the secondary super-klass chain.
 3994   // This code is rarely used, so simplicity is a virtue here.
 3995   // The repne_scan instruction uses fixed registers, which we must spill.
 3996   // Don't worry too much about pre-existing connections with the input regs.
 3997 
 3998   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 3999   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4000 
 4001   // Get super_klass value into rax (even if it was in rdi or rcx).
 4002   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4003   if (super_klass != rax) {
 4004     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4005     mov(rax, super_klass);
 4006   }
 4007   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4008   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4009 
 4010 #ifndef PRODUCT
 4011   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4012   ExternalAddress pst_counter_addr((address) pst_counter);
 4013   lea(rcx, pst_counter_addr);
 4014   incrementl(Address(rcx, 0));
 4015 #endif //PRODUCT
 4016 
 4017   // We will consult the secondary-super array.
 4018   movptr(rdi, secondary_supers_addr);
 4019   // Load the array length.  (Positive movl does right thing on LP64.)
 4020   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4021   // Skip to start of data.
 4022   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4023 
 4024   // Scan RCX words at [RDI] for an occurrence of RAX.
 4025   // Set NZ/Z based on last compare.
 4026   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4027   // not change flags (only scas instruction which is repeated sets flags).
 4028   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4029 
 4030     testptr(rax,rax); // Set Z = 0
 4031     repne_scan();
 4032 
 4033   // Unspill the temp. registers:
 4034   if (pushed_rdi)  pop(rdi);
 4035   if (pushed_rcx)  pop(rcx);
 4036   if (pushed_rax)  pop(rax);
 4037 
 4038   if (set_cond_codes) {
 4039     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4040     assert(!pushed_rdi, "rdi must be left non-null");
 4041     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4042   }
 4043 
 4044   if (L_failure == &L_fallthrough)
 4045         jccb(Assembler::notEqual, *L_failure);
 4046   else  jcc(Assembler::notEqual, *L_failure);
 4047 
 4048   // Success.  Cache the super we found and proceed in triumph.
 4049   movptr(super_cache_addr, super_klass);
 4050 
 4051   if (L_success != &L_fallthrough) {
 4052     jmp(*L_success);
 4053   }
 4054 
 4055 #undef IS_A_TEMP
 4056 
 4057   bind(L_fallthrough);
 4058 }
 4059 
 4060 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4061                                                    Register super_klass,
 4062                                                    Register temp_reg,
 4063                                                    Register temp2_reg,
 4064                                                    Label* L_success,
 4065                                                    Label* L_failure,
 4066                                                    bool set_cond_codes) {
 4067   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4068   check_klass_subtype_slow_path
 4069     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4070      L_success, L_failure);
 4071 }
 4072 
 4073 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4074                                                    Register super_klass,
 4075                                                    Register temp_reg,
 4076                                                    Register temp2_reg,
 4077                                                    Register temp3_reg,
 4078                                                    Register temp4_reg,
 4079                                                    Label* L_success,
 4080                                                    Label* L_failure) {
 4081   if (UseSecondarySupersTable) {
 4082     check_klass_subtype_slow_path_table
 4083       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4084        L_success, L_failure);
 4085   } else {
 4086     check_klass_subtype_slow_path_linear
 4087       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4088   }
 4089 }
 4090 
 4091 Register MacroAssembler::allocate_if_noreg(Register r,
 4092                                   RegSetIterator<Register> &available_regs,
 4093                                   RegSet &regs_to_push) {
 4094   if (!r->is_valid()) {
 4095     r = *available_regs++;
 4096     regs_to_push += r;
 4097   }
 4098   return r;
 4099 }
 4100 
 4101 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4102                                                          Register super_klass,
 4103                                                          Register temp_reg,
 4104                                                          Register temp2_reg,
 4105                                                          Register temp3_reg,
 4106                                                          Register result_reg,
 4107                                                          Label* L_success,
 4108                                                          Label* L_failure) {
 4109   // NB! Callers may assume that, when temp2_reg is a valid register,
 4110   // this code sets it to a nonzero value.
 4111   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4112 
 4113   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4114 
 4115   Label L_fallthrough;
 4116   int label_nulls = 0;
 4117   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4118   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4119   assert(label_nulls <= 1, "at most one null in the batch");
 4120 
 4121   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4122 
 4123   RegSetIterator<Register> available_regs
 4124     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4125 
 4126   RegSet pushed_regs;
 4127 
 4128   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4129   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4130   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4131   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4132   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4133 
 4134   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4135 
 4136   {
 4137 
 4138     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4139     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4140     subptr(rsp, aligned_size);
 4141     push_set(pushed_regs, 0);
 4142 
 4143     lookup_secondary_supers_table_var(sub_klass,
 4144                                       super_klass,
 4145                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4146     cmpq(result_reg, 0);
 4147 
 4148     // Unspill the temp. registers:
 4149     pop_set(pushed_regs, 0);
 4150     // Increment SP but do not clobber flags.
 4151     lea(rsp, Address(rsp, aligned_size));
 4152   }
 4153 
 4154   if (temp2_reg_was_valid) {
 4155     movq(temp2_reg, 1);
 4156   }
 4157 
 4158   jcc(Assembler::notEqual, *L_failure);
 4159 
 4160   if (L_success != &L_fallthrough) {
 4161     jmp(*L_success);
 4162   }
 4163 
 4164   bind(L_fallthrough);
 4165 }
 4166 
 4167 // population_count variant for running without the POPCNT
 4168 // instruction, which was introduced with SSE4.2 in 2008.
 4169 void MacroAssembler::population_count(Register dst, Register src,
 4170                                       Register scratch1, Register scratch2) {
 4171   assert_different_registers(src, scratch1, scratch2);
 4172   if (UsePopCountInstruction) {
 4173     Assembler::popcntq(dst, src);
 4174   } else {
 4175     assert_different_registers(src, scratch1, scratch2);
 4176     assert_different_registers(dst, scratch1, scratch2);
 4177     Label loop, done;
 4178 
 4179     mov(scratch1, src);
 4180     // dst = 0;
 4181     // while(scratch1 != 0) {
 4182     //   dst++;
 4183     //   scratch1 &= (scratch1 - 1);
 4184     // }
 4185     xorl(dst, dst);
 4186     testq(scratch1, scratch1);
 4187     jccb(Assembler::equal, done);
 4188     {
 4189       bind(loop);
 4190       incq(dst);
 4191       movq(scratch2, scratch1);
 4192       decq(scratch2);
 4193       andq(scratch1, scratch2);
 4194       jccb(Assembler::notEqual, loop);
 4195     }
 4196     bind(done);
 4197   }
 4198 #ifdef ASSERT
 4199   mov64(scratch1, 0xCafeBabeDeadBeef);
 4200   movq(scratch2, scratch1);
 4201 #endif
 4202 }
 4203 
 4204 // Ensure that the inline code and the stub are using the same registers.
 4205 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4206 do {                                                                 \
 4207   assert(r_super_klass  == rax, "mismatch");                         \
 4208   assert(r_array_base   == rbx, "mismatch");                         \
 4209   assert(r_array_length == rcx, "mismatch");                         \
 4210   assert(r_array_index  == rdx, "mismatch");                         \
 4211   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4212   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4213   assert(result         == rdi || result      == noreg, "mismatch"); \
 4214 } while(0)
 4215 
 4216 // Versions of salq and rorq that don't need count to be in rcx
 4217 
 4218 void MacroAssembler::salq(Register dest, Register count) {
 4219   if (count == rcx) {
 4220     Assembler::salq(dest);
 4221   } else {
 4222     assert_different_registers(rcx, dest);
 4223     xchgq(rcx, count);
 4224     Assembler::salq(dest);
 4225     xchgq(rcx, count);
 4226   }
 4227 }
 4228 
 4229 void MacroAssembler::rorq(Register dest, Register count) {
 4230   if (count == rcx) {
 4231     Assembler::rorq(dest);
 4232   } else {
 4233     assert_different_registers(rcx, dest);
 4234     xchgq(rcx, count);
 4235     Assembler::rorq(dest);
 4236     xchgq(rcx, count);
 4237   }
 4238 }
 4239 
 4240 // Return true: we succeeded in generating this code
 4241 //
 4242 // At runtime, return 0 in result if r_super_klass is a superclass of
 4243 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4244 // super_klass_slot of the class you're looking for. This is always
 4245 // the case for instanceof and checkcast.
 4246 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4247                                                          Register r_super_klass,
 4248                                                          Register temp1,
 4249                                                          Register temp2,
 4250                                                          Register temp3,
 4251                                                          Register temp4,
 4252                                                          Register result,
 4253                                                          u1 super_klass_slot) {
 4254   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4255 
 4256   Label L_fallthrough, L_success, L_failure;
 4257 
 4258   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4259 
 4260   const Register
 4261     r_array_index  = temp1,
 4262     r_array_length = temp2,
 4263     r_array_base   = temp3,
 4264     r_bitmap       = temp4;
 4265 
 4266   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4267 
 4268   xorq(result, result); // = 0
 4269 
 4270   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4271   movq(r_array_index, r_bitmap);
 4272 
 4273   // First check the bitmap to see if super_klass might be present. If
 4274   // the bit is zero, we are certain that super_klass is not one of
 4275   // the secondary supers.
 4276   u1 bit = super_klass_slot;
 4277   {
 4278     // NB: If the count in a x86 shift instruction is 0, the flags are
 4279     // not affected, so we do a testq instead.
 4280     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4281     if (shift_count != 0) {
 4282       salq(r_array_index, shift_count);
 4283     } else {
 4284       testq(r_array_index, r_array_index);
 4285     }
 4286   }
 4287   // We test the MSB of r_array_index, i.e. its sign bit
 4288   jcc(Assembler::positive, L_failure);
 4289 
 4290   // Get the first array index that can contain super_klass into r_array_index.
 4291   if (bit != 0) {
 4292     population_count(r_array_index, r_array_index, temp2, temp3);
 4293   } else {
 4294     movl(r_array_index, 1);
 4295   }
 4296   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4297 
 4298   // We will consult the secondary-super array.
 4299   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4300 
 4301   // We're asserting that the first word in an Array<Klass*> is the
 4302   // length, and the second word is the first word of the data. If
 4303   // that ever changes, r_array_base will have to be adjusted here.
 4304   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4305   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4306 
 4307   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4308   jccb(Assembler::equal, L_success);
 4309 
 4310   // Is there another entry to check? Consult the bitmap.
 4311   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4312   jccb(Assembler::carryClear, L_failure);
 4313 
 4314   // Linear probe. Rotate the bitmap so that the next bit to test is
 4315   // in Bit 1.
 4316   if (bit != 0) {
 4317     rorq(r_bitmap, bit);
 4318   }
 4319 
 4320   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4321   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4322   // Kills: r_array_length.
 4323   // Returns: result.
 4324   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4325   // Result (0/1) is in rdi
 4326   jmpb(L_fallthrough);
 4327 
 4328   bind(L_failure);
 4329   incq(result); // 0 => 1
 4330 
 4331   bind(L_success);
 4332   // result = 0;
 4333 
 4334   bind(L_fallthrough);
 4335   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4336 
 4337   if (VerifySecondarySupers) {
 4338     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4339                                   temp1, temp2, temp3);
 4340   }
 4341 }
 4342 
 4343 // At runtime, return 0 in result if r_super_klass is a superclass of
 4344 // r_sub_klass, otherwise return nonzero. Use this version of
 4345 // lookup_secondary_supers_table() if you don't know ahead of time
 4346 // which superclass will be searched for. Used by interpreter and
 4347 // runtime stubs. It is larger and has somewhat greater latency than
 4348 // the version above, which takes a constant super_klass_slot.
 4349 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4350                                                        Register r_super_klass,
 4351                                                        Register temp1,
 4352                                                        Register temp2,
 4353                                                        Register temp3,
 4354                                                        Register temp4,
 4355                                                        Register result) {
 4356   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4357   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4358   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4359 
 4360   Label L_fallthrough, L_success, L_failure;
 4361 
 4362   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4363 
 4364   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4365 
 4366   // FIXME. Once we are sure that all paths reaching this point really
 4367   // do pass rcx as one of our temps we can get rid of the following
 4368   // workaround.
 4369   assert(temps.contains(rcx), "fix this code");
 4370 
 4371   // We prefer to have our shift count in rcx. If rcx is one of our
 4372   // temps, use it for slot. If not, pick any of our temps.
 4373   Register slot;
 4374   if (!temps.contains(rcx)) {
 4375     slot = *available_regs++;
 4376   } else {
 4377     slot = rcx;
 4378   }
 4379 
 4380   const Register r_array_index = *available_regs++;
 4381   const Register r_bitmap      = *available_regs++;
 4382 
 4383   // The logic above guarantees this property, but we state it here.
 4384   assert_different_registers(r_array_index, r_bitmap, rcx);
 4385 
 4386   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4387   movq(r_array_index, r_bitmap);
 4388 
 4389   // First check the bitmap to see if super_klass might be present. If
 4390   // the bit is zero, we are certain that super_klass is not one of
 4391   // the secondary supers.
 4392   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4393   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4394   salq(r_array_index, slot);
 4395 
 4396   testq(r_array_index, r_array_index);
 4397   // We test the MSB of r_array_index, i.e. its sign bit
 4398   jcc(Assembler::positive, L_failure);
 4399 
 4400   const Register r_array_base = *available_regs++;
 4401 
 4402   // Get the first array index that can contain super_klass into r_array_index.
 4403   // Note: Clobbers r_array_base and slot.
 4404   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4405 
 4406   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4407 
 4408   // We will consult the secondary-super array.
 4409   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4410 
 4411   // We're asserting that the first word in an Array<Klass*> is the
 4412   // length, and the second word is the first word of the data. If
 4413   // that ever changes, r_array_base will have to be adjusted here.
 4414   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4415   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4416 
 4417   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4418   jccb(Assembler::equal, L_success);
 4419 
 4420   // Restore slot to its true value
 4421   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4422 
 4423   // Linear probe. Rotate the bitmap so that the next bit to test is
 4424   // in Bit 1.
 4425   rorq(r_bitmap, slot);
 4426 
 4427   // Is there another entry to check? Consult the bitmap.
 4428   btq(r_bitmap, 1);
 4429   jccb(Assembler::carryClear, L_failure);
 4430 
 4431   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4432   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4433   // Kills: r_array_length.
 4434   // Returns: result.
 4435   lookup_secondary_supers_table_slow_path(r_super_klass,
 4436                                           r_array_base,
 4437                                           r_array_index,
 4438                                           r_bitmap,
 4439                                           /*temp1*/result,
 4440                                           /*temp2*/slot,
 4441                                           &L_success,
 4442                                           nullptr);
 4443 
 4444   bind(L_failure);
 4445   movq(result, 1);
 4446   jmpb(L_fallthrough);
 4447 
 4448   bind(L_success);
 4449   xorq(result, result); // = 0
 4450 
 4451   bind(L_fallthrough);
 4452   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4453 
 4454   if (VerifySecondarySupers) {
 4455     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4456                                   temp1, temp2, temp3);
 4457   }
 4458 }
 4459 
 4460 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4461                                  Label* L_success, Label* L_failure) {
 4462   Label L_loop, L_fallthrough;
 4463   {
 4464     int label_nulls = 0;
 4465     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4466     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4467     assert(label_nulls <= 1, "at most one null in the batch");
 4468   }
 4469   bind(L_loop);
 4470   cmpq(value, Address(addr, count, Address::times_8));
 4471   jcc(Assembler::equal, *L_success);
 4472   addl(count, 1);
 4473   cmpl(count, limit);
 4474   jcc(Assembler::less, L_loop);
 4475 
 4476   if (&L_fallthrough != L_failure) {
 4477     jmp(*L_failure);
 4478   }
 4479   bind(L_fallthrough);
 4480 }
 4481 
 4482 // Called by code generated by check_klass_subtype_slow_path
 4483 // above. This is called when there is a collision in the hashed
 4484 // lookup in the secondary supers array.
 4485 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4486                                                              Register r_array_base,
 4487                                                              Register r_array_index,
 4488                                                              Register r_bitmap,
 4489                                                              Register temp1,
 4490                                                              Register temp2,
 4491                                                              Label* L_success,
 4492                                                              Label* L_failure) {
 4493   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4494 
 4495   const Register
 4496     r_array_length = temp1,
 4497     r_sub_klass    = noreg,
 4498     result         = noreg;
 4499 
 4500   Label L_fallthrough;
 4501   int label_nulls = 0;
 4502   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4503   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4504   assert(label_nulls <= 1, "at most one null in the batch");
 4505 
 4506   // Load the array length.
 4507   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4508   // And adjust the array base to point to the data.
 4509   // NB! Effectively increments current slot index by 1.
 4510   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4511   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4512 
 4513   // Linear probe
 4514   Label L_huge;
 4515 
 4516   // The bitmap is full to bursting.
 4517   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4518   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4519   jcc(Assembler::greater, L_huge);
 4520 
 4521   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4522   // current slot (at secondary_supers[r_array_index]) has not yet
 4523   // been inspected, and r_array_index may be out of bounds if we
 4524   // wrapped around the end of the array.
 4525 
 4526   { // This is conventional linear probing, but instead of terminating
 4527     // when a null entry is found in the table, we maintain a bitmap
 4528     // in which a 0 indicates missing entries.
 4529     // The check above guarantees there are 0s in the bitmap, so the loop
 4530     // eventually terminates.
 4531 
 4532     xorl(temp2, temp2); // = 0;
 4533 
 4534     Label L_again;
 4535     bind(L_again);
 4536 
 4537     // Check for array wraparound.
 4538     cmpl(r_array_index, r_array_length);
 4539     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4540 
 4541     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4542     jcc(Assembler::equal, *L_success);
 4543 
 4544     // If the next bit in bitmap is zero, we're done.
 4545     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4546     jcc(Assembler::carryClear, *L_failure);
 4547 
 4548     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4549     addl(r_array_index, 1);
 4550 
 4551     jmp(L_again);
 4552   }
 4553 
 4554   { // Degenerate case: more than 64 secondary supers.
 4555     // FIXME: We could do something smarter here, maybe a vectorized
 4556     // comparison or a binary search, but is that worth any added
 4557     // complexity?
 4558     bind(L_huge);
 4559     xorl(r_array_index, r_array_index); // = 0
 4560     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4561                 L_success,
 4562                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4563 
 4564     bind(L_fallthrough);
 4565   }
 4566 }
 4567 
 4568 struct VerifyHelperArguments {
 4569   Klass* _super;
 4570   Klass* _sub;
 4571   intptr_t _linear_result;
 4572   intptr_t _table_result;
 4573 };
 4574 
 4575 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4576   Klass::on_secondary_supers_verification_failure(args->_super,
 4577                                                   args->_sub,
 4578                                                   args->_linear_result,
 4579                                                   args->_table_result,
 4580                                                   msg);
 4581 }
 4582 
 4583 // Make sure that the hashed lookup and a linear scan agree.
 4584 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4585                                                    Register r_super_klass,
 4586                                                    Register result,
 4587                                                    Register temp1,
 4588                                                    Register temp2,
 4589                                                    Register temp3) {
 4590   const Register
 4591       r_array_index  = temp1,
 4592       r_array_length = temp2,
 4593       r_array_base   = temp3,
 4594       r_bitmap       = noreg;
 4595 
 4596   BLOCK_COMMENT("verify_secondary_supers_table {");
 4597 
 4598   Label L_success, L_failure, L_check, L_done;
 4599 
 4600   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4601   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4602   // And adjust the array base to point to the data.
 4603   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4604 
 4605   testl(r_array_length, r_array_length); // array_length == 0?
 4606   jcc(Assembler::zero, L_failure);
 4607 
 4608   movl(r_array_index, 0);
 4609   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4610   // fall through to L_failure
 4611 
 4612   const Register linear_result = r_array_index; // reuse temp1
 4613 
 4614   bind(L_failure); // not present
 4615   movl(linear_result, 1);
 4616   jmp(L_check);
 4617 
 4618   bind(L_success); // present
 4619   movl(linear_result, 0);
 4620 
 4621   bind(L_check);
 4622   cmpl(linear_result, result);
 4623   jcc(Assembler::equal, L_done);
 4624 
 4625   { // To avoid calling convention issues, build a record on the stack
 4626     // and pass the pointer to that instead.
 4627     push(result);
 4628     push(linear_result);
 4629     push(r_sub_klass);
 4630     push(r_super_klass);
 4631     movptr(c_rarg1, rsp);
 4632     movptr(c_rarg0, (uintptr_t) "mismatch");
 4633     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4634     should_not_reach_here();
 4635   }
 4636   bind(L_done);
 4637 
 4638   BLOCK_COMMENT("} verify_secondary_supers_table");
 4639 }
 4640 
 4641 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4642 
 4643 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4644   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4645 
 4646   Label L_fallthrough;
 4647   if (L_fast_path == nullptr) {
 4648     L_fast_path = &L_fallthrough;
 4649   } else if (L_slow_path == nullptr) {
 4650     L_slow_path = &L_fallthrough;
 4651   }
 4652 
 4653   // Fast path check: class is fully initialized.
 4654   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4655   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4656   jcc(Assembler::equal, *L_fast_path);
 4657 
 4658   // Fast path check: current thread is initializer thread
 4659   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4660   if (L_slow_path == &L_fallthrough) {
 4661     jcc(Assembler::equal, *L_fast_path);
 4662     bind(*L_slow_path);
 4663   } else if (L_fast_path == &L_fallthrough) {
 4664     jcc(Assembler::notEqual, *L_slow_path);
 4665     bind(*L_fast_path);
 4666   } else {
 4667     Unimplemented();
 4668   }
 4669 }
 4670 
 4671 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4672   if (VM_Version::supports_cmov()) {
 4673     cmovl(cc, dst, src);
 4674   } else {
 4675     Label L;
 4676     jccb(negate_condition(cc), L);
 4677     movl(dst, src);
 4678     bind(L);
 4679   }
 4680 }
 4681 
 4682 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4683   if (VM_Version::supports_cmov()) {
 4684     cmovl(cc, dst, src);
 4685   } else {
 4686     Label L;
 4687     jccb(negate_condition(cc), L);
 4688     movl(dst, src);
 4689     bind(L);
 4690   }
 4691 }
 4692 
 4693 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4694   if (!VerifyOops) return;
 4695 
 4696   BLOCK_COMMENT("verify_oop {");
 4697   push(rscratch1);
 4698   push(rax);                          // save rax
 4699   push(reg);                          // pass register argument
 4700 
 4701   // Pass register number to verify_oop_subroutine
 4702   const char* b = nullptr;
 4703   {
 4704     ResourceMark rm;
 4705     stringStream ss;
 4706     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4707     b = code_string(ss.as_string());
 4708   }
 4709   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4710   pushptr(buffer.addr(), rscratch1);
 4711 
 4712   // call indirectly to solve generation ordering problem
 4713   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4714   call(rax);
 4715   // Caller pops the arguments (oop, message) and restores rax, r10
 4716   BLOCK_COMMENT("} verify_oop");
 4717 }
 4718 
 4719 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 4720   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 4721     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 4722     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 4723     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 4724   } else if (VM_Version::supports_avx()) {
 4725     vpcmpeqd(dst, dst, dst, vector_len);
 4726   } else {
 4727     pcmpeqd(dst, dst);
 4728   }
 4729 }
 4730 
 4731 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 4732                                          int extra_slot_offset) {
 4733   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 4734   int stackElementSize = Interpreter::stackElementSize;
 4735   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 4736 #ifdef ASSERT
 4737   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 4738   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 4739 #endif
 4740   Register             scale_reg    = noreg;
 4741   Address::ScaleFactor scale_factor = Address::no_scale;
 4742   if (arg_slot.is_constant()) {
 4743     offset += arg_slot.as_constant() * stackElementSize;
 4744   } else {
 4745     scale_reg    = arg_slot.as_register();
 4746     scale_factor = Address::times(stackElementSize);
 4747   }
 4748   offset += wordSize;           // return PC is on stack
 4749   return Address(rsp, scale_reg, scale_factor, offset);
 4750 }
 4751 
 4752 // Handle the receiver type profile update given the "recv" klass.
 4753 //
 4754 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 4755 // If there are no matching or claimable receiver entries in RD, updates
 4756 // the polymorphic counter.
 4757 //
 4758 // This code expected to run by either the interpreter or JIT-ed code, without
 4759 // extra synchronization. For safety, receiver cells are claimed atomically, which
 4760 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 4761 // counter updates are not atomic.
 4762 //
 4763 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 4764   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 4765   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 4766   int poly_count_offset      = in_bytes(CounterData::count_offset());
 4767   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 4768   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 4769 
 4770   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 4771   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 4772   base_receiver_offset += mdp_offset;
 4773   end_receiver_offset  += mdp_offset;
 4774   poly_count_offset    += mdp_offset;
 4775 
 4776   // Scale down to optimize encoding. Slots are pointer-sized.
 4777   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 4778   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 4779   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 4780   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 4781   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 4782   base_receiver_offset   >>= LogBytesPerWord;
 4783   end_receiver_offset    >>= LogBytesPerWord;
 4784   poly_count_offset      >>= LogBytesPerWord;
 4785   receiver_step          >>= LogBytesPerWord;
 4786   receiver_to_count_step >>= LogBytesPerWord;
 4787 
 4788 #ifdef ASSERT
 4789   // We are about to walk the MDO slots without asking for offsets.
 4790   // Check that our math hits all the right spots.
 4791   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 4792     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 4793     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 4794     int offset = base_receiver_offset + receiver_step*c;
 4795     int count_offset = offset + receiver_to_count_step;
 4796     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 4797     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 4798   }
 4799   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 4800   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 4801 #endif
 4802 
 4803   // Corner case: no profile table. Increment poly counter and exit.
 4804   if (ReceiverTypeData::row_limit() == 0) {
 4805     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 4806     return;
 4807   }
 4808 
 4809   Register offset = rscratch1;
 4810 
 4811   Label L_loop_search_receiver, L_loop_search_empty;
 4812   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 4813 
 4814   // The code here recognizes three major cases:
 4815   //   A. Fastest: receiver found in the table
 4816   //   B. Fast: no receiver in the table, and the table is full
 4817   //   C. Slow: no receiver in the table, free slots in the table
 4818   //
 4819   // The case A performance is most important, as perfectly-behaved code would end up
 4820   // there, especially with larger TypeProfileWidth. The case B performance is
 4821   // important as well, this is where bulk of code would land for normally megamorphic
 4822   // cases. The case C performance is not essential, its job is to deal with installation
 4823   // races, we optimize for code density instead. Case C needs to make sure that receiver
 4824   // rows are only claimed once. This makes sure we never overwrite a row for another
 4825   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 4826   //
 4827   // It is very tempting to handle these cases in a single loop, and claim the first slot
 4828   // without checking the rest of the table. But, profiling code should tolerate free slots
 4829   // in the table, as class unloading can clear them. After such cleanup, the receiver
 4830   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 4831   // to complete, before trying to install new slots. Splitting the code in several tight
 4832   // loops also helpfully optimizes for cases A and B.
 4833   //
 4834   // This code is effectively:
 4835   //
 4836   // restart:
 4837   //   // Fastest: receiver is already installed
 4838   //   for (i = 0; i < receiver_count(); i++) {
 4839   //     if (receiver(i) == recv) goto found_recv(i);
 4840   //   }
 4841   //
 4842   //   // Fast: no receiver, but profile is full
 4843   //   for (i = 0; i < receiver_count(); i++) {
 4844   //     if (receiver(i) == null) goto found_null(i);
 4845   //   }
 4846   //   goto polymorphic
 4847   //
 4848   //   // Slow: try to install receiver
 4849   // found_null(i):
 4850   //   CAS(&receiver(i), null, recv);
 4851   //   goto restart
 4852   //
 4853   // polymorphic:
 4854   //   count++;
 4855   //   return
 4856   //
 4857   // found_recv(i):
 4858   //   *receiver_count(i)++
 4859   //
 4860 
 4861   bind(L_restart);
 4862 
 4863   // Fastest: receiver is already installed
 4864   movptr(offset, base_receiver_offset);
 4865   bind(L_loop_search_receiver);
 4866     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 4867     jccb(Assembler::equal, L_found_recv);
 4868   addptr(offset, receiver_step);
 4869   cmpptr(offset, end_receiver_offset);
 4870   jccb(Assembler::notEqual, L_loop_search_receiver);
 4871 
 4872   // Fast: no receiver, but profile is full
 4873   movptr(offset, base_receiver_offset);
 4874   bind(L_loop_search_empty);
 4875     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 4876     jccb(Assembler::equal, L_found_empty);
 4877   addptr(offset, receiver_step);
 4878   cmpptr(offset, end_receiver_offset);
 4879   jccb(Assembler::notEqual, L_loop_search_empty);
 4880   jmpb(L_polymorphic);
 4881 
 4882   // Slow: try to install receiver
 4883   bind(L_found_empty);
 4884 
 4885   // Atomically swing receiver slot: null -> recv.
 4886   //
 4887   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 4888   // other important registers untouched, as they form the address. Therefore, we need
 4889   // to shift any important registers from RAX into some other spare register. If we
 4890   // have a spare register, we are forced to save it on stack here.
 4891 
 4892   Register spare_reg = noreg;
 4893   Register shifted_mdp = mdp;
 4894   Register shifted_recv = recv;
 4895   if (recv == rax || mdp == rax) {
 4896     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 4897                 (recv != rcx && mdp != rcx) ? rcx :
 4898                 rdx;
 4899     assert_different_registers(mdp, recv, offset, spare_reg);
 4900 
 4901     push(spare_reg);
 4902     if (recv == rax) {
 4903       movptr(spare_reg, recv);
 4904       shifted_recv = spare_reg;
 4905     } else {
 4906       assert(mdp == rax, "Remaining case");
 4907       movptr(spare_reg, mdp);
 4908       shifted_mdp = spare_reg;
 4909     }
 4910   } else {
 4911     push(rax);
 4912   }
 4913 
 4914   // None of the important registers are in RAX after this shuffle.
 4915   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 4916 
 4917   xorptr(rax, rax);
 4918   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 4919 
 4920   // Unshift registers.
 4921   if (recv == rax || mdp == rax) {
 4922     movptr(rax, spare_reg);
 4923     pop(spare_reg);
 4924   } else {
 4925     pop(rax);
 4926   }
 4927 
 4928   // CAS success means the slot now has the receiver we want. CAS failure means
 4929   // something had claimed the slot concurrently: it can be the same receiver we want,
 4930   // or something else. Since this is a slow path, we can optimize for code density,
 4931   // and just restart the search from the beginning.
 4932   jmpb(L_restart);
 4933 
 4934   // Counter updates:
 4935 
 4936   // Increment polymorphic counter instead of receiver slot.
 4937   bind(L_polymorphic);
 4938   movptr(offset, poly_count_offset);
 4939   jmpb(L_count_update);
 4940 
 4941   // Found a receiver, convert its slot offset to corresponding count offset.
 4942   bind(L_found_recv);
 4943   addptr(offset, receiver_to_count_step);
 4944 
 4945   bind(L_count_update);
 4946   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 4947 }
 4948 
 4949 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 4950   if (!VerifyOops) return;
 4951 
 4952   push(rscratch1);
 4953   push(rax); // save rax,
 4954   // addr may contain rsp so we will have to adjust it based on the push
 4955   // we just did (and on 64 bit we do two pushes)
 4956   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 4957   // stores rax into addr which is backwards of what was intended.
 4958   if (addr.uses(rsp)) {
 4959     lea(rax, addr);
 4960     pushptr(Address(rax, 2 * BytesPerWord));
 4961   } else {
 4962     pushptr(addr);
 4963   }
 4964 
 4965   // Pass register number to verify_oop_subroutine
 4966   const char* b = nullptr;
 4967   {
 4968     ResourceMark rm;
 4969     stringStream ss;
 4970     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 4971     b = code_string(ss.as_string());
 4972   }
 4973   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4974   pushptr(buffer.addr(), rscratch1);
 4975 
 4976   // call indirectly to solve generation ordering problem
 4977   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4978   call(rax);
 4979   // Caller pops the arguments (addr, message) and restores rax, r10.
 4980 }
 4981 
 4982 void MacroAssembler::verify_tlab() {
 4983 #ifdef ASSERT
 4984   if (UseTLAB && VerifyOops) {
 4985     Label next, ok;
 4986     Register t1 = rsi;
 4987 
 4988     push(t1);
 4989 
 4990     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 4991     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 4992     jcc(Assembler::aboveEqual, next);
 4993     STOP("assert(top >= start)");
 4994     should_not_reach_here();
 4995 
 4996     bind(next);
 4997     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 4998     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 4999     jcc(Assembler::aboveEqual, ok);
 5000     STOP("assert(top <= end)");
 5001     should_not_reach_here();
 5002 
 5003     bind(ok);
 5004     pop(t1);
 5005   }
 5006 #endif
 5007 }
 5008 
 5009 class ControlWord {
 5010  public:
 5011   int32_t _value;
 5012 
 5013   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5014   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5015   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5016   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5017   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5018   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5019   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5020   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5021 
 5022   void print() const {
 5023     // rounding control
 5024     const char* rc;
 5025     switch (rounding_control()) {
 5026       case 0: rc = "round near"; break;
 5027       case 1: rc = "round down"; break;
 5028       case 2: rc = "round up  "; break;
 5029       case 3: rc = "chop      "; break;
 5030       default:
 5031         rc = nullptr; // silence compiler warnings
 5032         fatal("Unknown rounding control: %d", rounding_control());
 5033     };
 5034     // precision control
 5035     const char* pc;
 5036     switch (precision_control()) {
 5037       case 0: pc = "24 bits "; break;
 5038       case 1: pc = "reserved"; break;
 5039       case 2: pc = "53 bits "; break;
 5040       case 3: pc = "64 bits "; break;
 5041       default:
 5042         pc = nullptr; // silence compiler warnings
 5043         fatal("Unknown precision control: %d", precision_control());
 5044     };
 5045     // flags
 5046     char f[9];
 5047     f[0] = ' ';
 5048     f[1] = ' ';
 5049     f[2] = (precision   ()) ? 'P' : 'p';
 5050     f[3] = (underflow   ()) ? 'U' : 'u';
 5051     f[4] = (overflow    ()) ? 'O' : 'o';
 5052     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5053     f[6] = (denormalized()) ? 'D' : 'd';
 5054     f[7] = (invalid     ()) ? 'I' : 'i';
 5055     f[8] = '\x0';
 5056     // output
 5057     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5058   }
 5059 
 5060 };
 5061 
 5062 class StatusWord {
 5063  public:
 5064   int32_t _value;
 5065 
 5066   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5067   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5068   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5069   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5070   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5071   int  top() const                     { return  (_value >> 11) & 7      ; }
 5072   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5073   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5074   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5075   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5076   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5077   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5078   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5079   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5080 
 5081   void print() const {
 5082     // condition codes
 5083     char c[5];
 5084     c[0] = (C3()) ? '3' : '-';
 5085     c[1] = (C2()) ? '2' : '-';
 5086     c[2] = (C1()) ? '1' : '-';
 5087     c[3] = (C0()) ? '0' : '-';
 5088     c[4] = '\x0';
 5089     // flags
 5090     char f[9];
 5091     f[0] = (error_status()) ? 'E' : '-';
 5092     f[1] = (stack_fault ()) ? 'S' : '-';
 5093     f[2] = (precision   ()) ? 'P' : '-';
 5094     f[3] = (underflow   ()) ? 'U' : '-';
 5095     f[4] = (overflow    ()) ? 'O' : '-';
 5096     f[5] = (zero_divide ()) ? 'Z' : '-';
 5097     f[6] = (denormalized()) ? 'D' : '-';
 5098     f[7] = (invalid     ()) ? 'I' : '-';
 5099     f[8] = '\x0';
 5100     // output
 5101     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5102   }
 5103 
 5104 };
 5105 
 5106 class TagWord {
 5107  public:
 5108   int32_t _value;
 5109 
 5110   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5111 
 5112   void print() const {
 5113     printf("%04x", _value & 0xFFFF);
 5114   }
 5115 
 5116 };
 5117 
 5118 class FPU_Register {
 5119  public:
 5120   int32_t _m0;
 5121   int32_t _m1;
 5122   int16_t _ex;
 5123 
 5124   bool is_indefinite() const           {
 5125     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5126   }
 5127 
 5128   void print() const {
 5129     char  sign = (_ex < 0) ? '-' : '+';
 5130     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5131     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5132   };
 5133 
 5134 };
 5135 
 5136 class FPU_State {
 5137  public:
 5138   enum {
 5139     register_size       = 10,
 5140     number_of_registers =  8,
 5141     register_mask       =  7
 5142   };
 5143 
 5144   ControlWord  _control_word;
 5145   StatusWord   _status_word;
 5146   TagWord      _tag_word;
 5147   int32_t      _error_offset;
 5148   int32_t      _error_selector;
 5149   int32_t      _data_offset;
 5150   int32_t      _data_selector;
 5151   int8_t       _register[register_size * number_of_registers];
 5152 
 5153   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5154   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5155 
 5156   const char* tag_as_string(int tag) const {
 5157     switch (tag) {
 5158       case 0: return "valid";
 5159       case 1: return "zero";
 5160       case 2: return "special";
 5161       case 3: return "empty";
 5162     }
 5163     ShouldNotReachHere();
 5164     return nullptr;
 5165   }
 5166 
 5167   void print() const {
 5168     // print computation registers
 5169     { int t = _status_word.top();
 5170       for (int i = 0; i < number_of_registers; i++) {
 5171         int j = (i - t) & register_mask;
 5172         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5173         st(j)->print();
 5174         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5175       }
 5176     }
 5177     printf("\n");
 5178     // print control registers
 5179     printf("ctrl = "); _control_word.print(); printf("\n");
 5180     printf("stat = "); _status_word .print(); printf("\n");
 5181     printf("tags = "); _tag_word    .print(); printf("\n");
 5182   }
 5183 
 5184 };
 5185 
 5186 class Flag_Register {
 5187  public:
 5188   int32_t _value;
 5189 
 5190   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5191   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5192   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5193   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5194   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5195   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5196   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5197 
 5198   void print() const {
 5199     // flags
 5200     char f[8];
 5201     f[0] = (overflow       ()) ? 'O' : '-';
 5202     f[1] = (direction      ()) ? 'D' : '-';
 5203     f[2] = (sign           ()) ? 'S' : '-';
 5204     f[3] = (zero           ()) ? 'Z' : '-';
 5205     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5206     f[5] = (parity         ()) ? 'P' : '-';
 5207     f[6] = (carry          ()) ? 'C' : '-';
 5208     f[7] = '\x0';
 5209     // output
 5210     printf("%08x  flags = %s", _value, f);
 5211   }
 5212 
 5213 };
 5214 
 5215 class IU_Register {
 5216  public:
 5217   int32_t _value;
 5218 
 5219   void print() const {
 5220     printf("%08x  %11d", _value, _value);
 5221   }
 5222 
 5223 };
 5224 
 5225 class IU_State {
 5226  public:
 5227   Flag_Register _eflags;
 5228   IU_Register   _rdi;
 5229   IU_Register   _rsi;
 5230   IU_Register   _rbp;
 5231   IU_Register   _rsp;
 5232   IU_Register   _rbx;
 5233   IU_Register   _rdx;
 5234   IU_Register   _rcx;
 5235   IU_Register   _rax;
 5236 
 5237   void print() const {
 5238     // computation registers
 5239     printf("rax,  = "); _rax.print(); printf("\n");
 5240     printf("rbx,  = "); _rbx.print(); printf("\n");
 5241     printf("rcx  = "); _rcx.print(); printf("\n");
 5242     printf("rdx  = "); _rdx.print(); printf("\n");
 5243     printf("rdi  = "); _rdi.print(); printf("\n");
 5244     printf("rsi  = "); _rsi.print(); printf("\n");
 5245     printf("rbp,  = "); _rbp.print(); printf("\n");
 5246     printf("rsp  = "); _rsp.print(); printf("\n");
 5247     printf("\n");
 5248     // control registers
 5249     printf("flgs = "); _eflags.print(); printf("\n");
 5250   }
 5251 };
 5252 
 5253 
 5254 class CPU_State {
 5255  public:
 5256   FPU_State _fpu_state;
 5257   IU_State  _iu_state;
 5258 
 5259   void print() const {
 5260     printf("--------------------------------------------------\n");
 5261     _iu_state .print();
 5262     printf("\n");
 5263     _fpu_state.print();
 5264     printf("--------------------------------------------------\n");
 5265   }
 5266 
 5267 };
 5268 
 5269 
 5270 static void _print_CPU_state(CPU_State* state) {
 5271   state->print();
 5272 };
 5273 
 5274 
 5275 void MacroAssembler::print_CPU_state() {
 5276   push_CPU_state();
 5277   push(rsp);                // pass CPU state
 5278   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5279   addptr(rsp, wordSize);       // discard argument
 5280   pop_CPU_state();
 5281 }
 5282 
 5283 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5284   // Either restore the MXCSR register after returning from the JNI Call
 5285   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5286   if (VM_Version::supports_sse()) {
 5287     if (RestoreMXCSROnJNICalls) {
 5288       ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5289     } else if (CheckJNICalls) {
 5290       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5291     }
 5292   }
 5293   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5294   vzeroupper();
 5295 }
 5296 
 5297 // ((OopHandle)result).resolve();
 5298 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5299   assert_different_registers(result, tmp);
 5300 
 5301   // Only 64 bit platforms support GCs that require a tmp register
 5302   // Only IN_HEAP loads require a thread_tmp register
 5303   // OopHandle::resolve is an indirection like jobject.
 5304   access_load_at(T_OBJECT, IN_NATIVE,
 5305                  result, Address(result, 0), tmp);
 5306 }
 5307 
 5308 // ((WeakHandle)result).resolve();
 5309 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5310   assert_different_registers(rresult, rtmp);
 5311   Label resolved;
 5312 
 5313   // A null weak handle resolves to null.
 5314   cmpptr(rresult, 0);
 5315   jcc(Assembler::equal, resolved);
 5316 
 5317   // Only 64 bit platforms support GCs that require a tmp register
 5318   // Only IN_HEAP loads require a thread_tmp register
 5319   // WeakHandle::resolve is an indirection like jweak.
 5320   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5321                  rresult, Address(rresult, 0), rtmp);
 5322   bind(resolved);
 5323 }
 5324 
 5325 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5326   // get mirror
 5327   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5328   load_method_holder(mirror, method);
 5329   movptr(mirror, Address(mirror, mirror_offset));
 5330   resolve_oop_handle(mirror, tmp);
 5331 }
 5332 
 5333 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5334   load_method_holder(rresult, rmethod);
 5335   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5336 }
 5337 
 5338 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5339   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5340   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5341   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5342 }
 5343 
 5344 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5345   assert(UseCompactObjectHeaders, "expect compact object headers");
 5346   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5347   shrq(dst, markWord::klass_shift);
 5348 }
 5349 
 5350 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5351   assert_different_registers(src, tmp);
 5352   assert_different_registers(dst, tmp);
 5353 
 5354   if (UseCompactObjectHeaders) {
 5355     load_narrow_klass_compact(dst, src);
 5356     decode_klass_not_null(dst, tmp);
 5357   } else if (UseCompressedClassPointers) {
 5358     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5359     decode_klass_not_null(dst, tmp);
 5360   } else {
 5361     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5362   }
 5363 }
 5364 
 5365 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5366   assert(!UseCompactObjectHeaders, "not with compact headers");
 5367   assert_different_registers(src, tmp);
 5368   assert_different_registers(dst, tmp);
 5369   if (UseCompressedClassPointers) {
 5370     encode_klass_not_null(src, tmp);
 5371     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5372   } else {
 5373     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5374   }
 5375 }
 5376 
 5377 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5378   if (UseCompactObjectHeaders) {
 5379     assert(tmp != noreg, "need tmp");
 5380     assert_different_registers(klass, obj, tmp);
 5381     load_narrow_klass_compact(tmp, obj);
 5382     cmpl(klass, tmp);
 5383   } else if (UseCompressedClassPointers) {
 5384     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5385   } else {
 5386     cmpptr(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5387   }
 5388 }
 5389 
 5390 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5391   if (UseCompactObjectHeaders) {
 5392     assert(tmp2 != noreg, "need tmp2");
 5393     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5394     load_narrow_klass_compact(tmp1, obj1);
 5395     load_narrow_klass_compact(tmp2, obj2);
 5396     cmpl(tmp1, tmp2);
 5397   } else if (UseCompressedClassPointers) {
 5398     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5399     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5400   } else {
 5401     movptr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5402     cmpptr(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5403   }
 5404 }
 5405 
 5406 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5407                                     Register tmp1) {
 5408   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5409   decorators = AccessInternal::decorator_fixup(decorators, type);
 5410   bool as_raw = (decorators & AS_RAW) != 0;
 5411   if (as_raw) {
 5412     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5413   } else {
 5414     bs->load_at(this, decorators, type, dst, src, tmp1);
 5415   }
 5416 }
 5417 
 5418 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5419                                      Register tmp1, Register tmp2, Register tmp3) {
 5420   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5421   decorators = AccessInternal::decorator_fixup(decorators, type);
 5422   bool as_raw = (decorators & AS_RAW) != 0;
 5423   if (as_raw) {
 5424     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5425   } else {
 5426     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5427   }
 5428 }
 5429 
 5430 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5431   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5432 }
 5433 
 5434 // Doesn't do verification, generates fixed size code
 5435 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5436   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5437 }
 5438 
 5439 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5440                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5441   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5442 }
 5443 
 5444 // Used for storing nulls.
 5445 void MacroAssembler::store_heap_oop_null(Address dst) {
 5446   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5447 }
 5448 
 5449 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5450   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5451   if (UseCompressedClassPointers) {
 5452     // Store to klass gap in destination
 5453     movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5454   }
 5455 }
 5456 
 5457 #ifdef ASSERT
 5458 void MacroAssembler::verify_heapbase(const char* msg) {
 5459   assert (UseCompressedOops, "should be compressed");
 5460   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5461   if (CheckCompressedOops) {
 5462     Label ok;
 5463     ExternalAddress src2(CompressedOops::base_addr());
 5464     const bool is_src2_reachable = reachable(src2);
 5465     if (!is_src2_reachable) {
 5466       push(rscratch1);  // cmpptr trashes rscratch1
 5467     }
 5468     cmpptr(r12_heapbase, src2, rscratch1);
 5469     jcc(Assembler::equal, ok);
 5470     STOP(msg);
 5471     bind(ok);
 5472     if (!is_src2_reachable) {
 5473       pop(rscratch1);
 5474     }
 5475   }
 5476 }
 5477 #endif
 5478 
 5479 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5480 void MacroAssembler::encode_heap_oop(Register r) {
 5481 #ifdef ASSERT
 5482   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5483 #endif
 5484   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5485   if (CompressedOops::base() == nullptr) {
 5486     if (CompressedOops::shift() != 0) {
 5487       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5488       shrq(r, LogMinObjAlignmentInBytes);
 5489     }
 5490     return;
 5491   }
 5492   testq(r, r);
 5493   cmovq(Assembler::equal, r, r12_heapbase);
 5494   subq(r, r12_heapbase);
 5495   shrq(r, LogMinObjAlignmentInBytes);
 5496 }
 5497 
 5498 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5499 #ifdef ASSERT
 5500   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5501   if (CheckCompressedOops) {
 5502     Label ok;
 5503     testq(r, r);
 5504     jcc(Assembler::notEqual, ok);
 5505     STOP("null oop passed to encode_heap_oop_not_null");
 5506     bind(ok);
 5507   }
 5508 #endif
 5509   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5510   if (CompressedOops::base() != nullptr) {
 5511     subq(r, r12_heapbase);
 5512   }
 5513   if (CompressedOops::shift() != 0) {
 5514     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5515     shrq(r, LogMinObjAlignmentInBytes);
 5516   }
 5517 }
 5518 
 5519 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5520 #ifdef ASSERT
 5521   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5522   if (CheckCompressedOops) {
 5523     Label ok;
 5524     testq(src, src);
 5525     jcc(Assembler::notEqual, ok);
 5526     STOP("null oop passed to encode_heap_oop_not_null2");
 5527     bind(ok);
 5528   }
 5529 #endif
 5530   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5531   if (dst != src) {
 5532     movq(dst, src);
 5533   }
 5534   if (CompressedOops::base() != nullptr) {
 5535     subq(dst, r12_heapbase);
 5536   }
 5537   if (CompressedOops::shift() != 0) {
 5538     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5539     shrq(dst, LogMinObjAlignmentInBytes);
 5540   }
 5541 }
 5542 
 5543 void  MacroAssembler::decode_heap_oop(Register r) {
 5544 #ifdef ASSERT
 5545   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5546 #endif
 5547   if (CompressedOops::base() == nullptr) {
 5548     if (CompressedOops::shift() != 0) {
 5549       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5550       shlq(r, LogMinObjAlignmentInBytes);
 5551     }
 5552   } else {
 5553     Label done;
 5554     shlq(r, LogMinObjAlignmentInBytes);
 5555     jccb(Assembler::equal, done);
 5556     addq(r, r12_heapbase);
 5557     bind(done);
 5558   }
 5559   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5560 }
 5561 
 5562 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5563   // Note: it will change flags
 5564   assert (UseCompressedOops, "should only be used for compressed headers");
 5565   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5566   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5567   // vtableStubs also counts instructions in pd_code_size_limit.
 5568   // Also do not verify_oop as this is called by verify_oop.
 5569   if (CompressedOops::shift() != 0) {
 5570     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5571     shlq(r, LogMinObjAlignmentInBytes);
 5572     if (CompressedOops::base() != nullptr) {
 5573       addq(r, r12_heapbase);
 5574     }
 5575   } else {
 5576     assert (CompressedOops::base() == nullptr, "sanity");
 5577   }
 5578 }
 5579 
 5580 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5581   // Note: it will change flags
 5582   assert (UseCompressedOops, "should only be used for compressed headers");
 5583   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5584   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5585   // vtableStubs also counts instructions in pd_code_size_limit.
 5586   // Also do not verify_oop as this is called by verify_oop.
 5587   if (CompressedOops::shift() != 0) {
 5588     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5589     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5590       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5591     } else {
 5592       if (dst != src) {
 5593         movq(dst, src);
 5594       }
 5595       shlq(dst, LogMinObjAlignmentInBytes);
 5596       if (CompressedOops::base() != nullptr) {
 5597         addq(dst, r12_heapbase);
 5598       }
 5599     }
 5600   } else {
 5601     assert (CompressedOops::base() == nullptr, "sanity");
 5602     if (dst != src) {
 5603       movq(dst, src);
 5604     }
 5605   }
 5606 }
 5607 
 5608 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5609   BLOCK_COMMENT("encode_klass_not_null {");
 5610   assert_different_registers(r, tmp);
 5611   if (CompressedKlassPointers::base() != nullptr) {
 5612     if (AOTCodeCache::is_on_for_dump()) {
 5613       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5614     } else {
 5615       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5616     }
 5617     subq(r, tmp);
 5618   }
 5619   if (CompressedKlassPointers::shift() != 0) {
 5620     shrq(r, CompressedKlassPointers::shift());
 5621   }
 5622   BLOCK_COMMENT("} encode_klass_not_null");
 5623 }
 5624 
 5625 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5626   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5627   assert_different_registers(src, dst);
 5628   if (CompressedKlassPointers::base() != nullptr) {
 5629     movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5630     addq(dst, src);
 5631   } else {
 5632     movptr(dst, src);
 5633   }
 5634   if (CompressedKlassPointers::shift() != 0) {
 5635     shrq(dst, CompressedKlassPointers::shift());
 5636   }
 5637   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5638 }
 5639 
 5640 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5641   BLOCK_COMMENT("decode_klass_not_null {");
 5642   assert_different_registers(r, tmp);
 5643   // Note: it will change flags
 5644   assert(UseCompressedClassPointers, "should only be used for compressed headers");
 5645   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5646   // vtableStubs also counts instructions in pd_code_size_limit.
 5647   // Also do not verify_oop as this is called by verify_oop.
 5648   if (CompressedKlassPointers::shift() != 0) {
 5649     shlq(r, CompressedKlassPointers::shift());
 5650   }
 5651   if (CompressedKlassPointers::base() != nullptr) {
 5652     if (AOTCodeCache::is_on_for_dump()) {
 5653       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5654     } else {
 5655       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5656     }
 5657     addq(r, tmp);
 5658   }
 5659   BLOCK_COMMENT("} decode_klass_not_null");
 5660 }
 5661 
 5662 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5663   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 5664   assert_different_registers(src, dst);
 5665   // Note: it will change flags
 5666   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5667   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5668   // vtableStubs also counts instructions in pd_code_size_limit.
 5669   // Also do not verify_oop as this is called by verify_oop.
 5670 
 5671   if (CompressedKlassPointers::base() == nullptr &&
 5672       CompressedKlassPointers::shift() == 0) {
 5673     // The best case scenario is that there is no base or shift. Then it is already
 5674     // a pointer that needs nothing but a register rename.
 5675     movl(dst, src);
 5676   } else {
 5677     if (CompressedKlassPointers::shift() <= Address::times_8) {
 5678       if (CompressedKlassPointers::base() != nullptr) {
 5679         movptr(dst, (intptr_t)CompressedKlassPointers::base());
 5680       } else {
 5681         xorq(dst, dst);
 5682       }
 5683       if (CompressedKlassPointers::shift() != 0) {
 5684         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 5685         leaq(dst, Address(dst, src, Address::times_8, 0));
 5686       } else {
 5687         addq(dst, src);
 5688       }
 5689     } else {
 5690       if (CompressedKlassPointers::base() != nullptr) {
 5691         const intptr_t base_right_shifted =
 5692             (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 5693         movptr(dst, base_right_shifted);
 5694       } else {
 5695         xorq(dst, dst);
 5696       }
 5697       addq(dst, src);
 5698       shlq(dst, CompressedKlassPointers::shift());
 5699     }
 5700   }
 5701   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 5702 }
 5703 
 5704 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5705   assert (UseCompressedOops, "should only be used for compressed headers");
 5706   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5707   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5708   int oop_index = oop_recorder()->find_index(obj);
 5709   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5710   mov_narrow_oop(dst, oop_index, rspec);
 5711 }
 5712 
 5713 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5714   assert (UseCompressedOops, "should only be used for compressed headers");
 5715   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5716   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5717   int oop_index = oop_recorder()->find_index(obj);
 5718   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5719   mov_narrow_oop(dst, oop_index, rspec);
 5720 }
 5721 
 5722 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5723   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5724   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5725   int klass_index = oop_recorder()->find_index(k);
 5726   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5727   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5728 }
 5729 
 5730 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 5731   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5732   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5733   int klass_index = oop_recorder()->find_index(k);
 5734   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5735   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5736 }
 5737 
 5738 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 5739   assert (UseCompressedOops, "should only be used for compressed headers");
 5740   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5741   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5742   int oop_index = oop_recorder()->find_index(obj);
 5743   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5744   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5745 }
 5746 
 5747 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 5748   assert (UseCompressedOops, "should only be used for compressed headers");
 5749   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5750   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5751   int oop_index = oop_recorder()->find_index(obj);
 5752   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5753   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5754 }
 5755 
 5756 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 5757   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5758   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5759   int klass_index = oop_recorder()->find_index(k);
 5760   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5761   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5762 }
 5763 
 5764 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 5765   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5766   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5767   int klass_index = oop_recorder()->find_index(k);
 5768   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5769   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5770 }
 5771 
 5772 void MacroAssembler::reinit_heapbase() {
 5773   if (UseCompressedOops) {
 5774     if (Universe::heap() != nullptr) {
 5775       if (CompressedOops::base() == nullptr) {
 5776         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 5777       } else {
 5778         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 5779       }
 5780     } else {
 5781       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 5782     }
 5783   }
 5784 }
 5785 
 5786 #if COMPILER2_OR_JVMCI
 5787 
 5788 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 5789 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5790   // cnt - number of qwords (8-byte words).
 5791   // base - start address, qword aligned.
 5792   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 5793   bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
 5794   if (use64byteVector) {
 5795     vpxor(xtmp, xtmp, xtmp, AVX_512bit);
 5796   } else if (MaxVectorSize >= 32) {
 5797     vpxor(xtmp, xtmp, xtmp, AVX_256bit);
 5798   } else {
 5799     pxor(xtmp, xtmp);
 5800   }
 5801   jmp(L_zero_64_bytes);
 5802 
 5803   BIND(L_loop);
 5804   if (MaxVectorSize >= 32) {
 5805     fill64(base, 0, xtmp, use64byteVector);
 5806   } else {
 5807     movdqu(Address(base,  0), xtmp);
 5808     movdqu(Address(base, 16), xtmp);
 5809     movdqu(Address(base, 32), xtmp);
 5810     movdqu(Address(base, 48), xtmp);
 5811   }
 5812   addptr(base, 64);
 5813 
 5814   BIND(L_zero_64_bytes);
 5815   subptr(cnt, 8);
 5816   jccb(Assembler::greaterEqual, L_loop);
 5817 
 5818   // Copy trailing 64 bytes
 5819   if (use64byteVector) {
 5820     addptr(cnt, 8);
 5821     jccb(Assembler::equal, L_end);
 5822     fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
 5823     jmp(L_end);
 5824   } else {
 5825     addptr(cnt, 4);
 5826     jccb(Assembler::less, L_tail);
 5827     if (MaxVectorSize >= 32) {
 5828       vmovdqu(Address(base, 0), xtmp);
 5829     } else {
 5830       movdqu(Address(base,  0), xtmp);
 5831       movdqu(Address(base, 16), xtmp);
 5832     }
 5833   }
 5834   addptr(base, 32);
 5835   subptr(cnt, 4);
 5836 
 5837   BIND(L_tail);
 5838   addptr(cnt, 4);
 5839   jccb(Assembler::lessEqual, L_end);
 5840   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 5841     fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
 5842   } else {
 5843     decrement(cnt);
 5844 
 5845     BIND(L_sloop);
 5846     movq(Address(base, 0), xtmp);
 5847     addptr(base, 8);
 5848     decrement(cnt);
 5849     jccb(Assembler::greaterEqual, L_sloop);
 5850   }
 5851   BIND(L_end);
 5852 }
 5853 
 5854 // Clearing constant sized memory using YMM/ZMM registers.
 5855 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5856   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 5857   bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
 5858 
 5859   int vector64_count = (cnt & (~0x7)) >> 3;
 5860   cnt = cnt & 0x7;
 5861   const int fill64_per_loop = 4;
 5862   const int max_unrolled_fill64 = 8;
 5863 
 5864   // 64 byte initialization loop.
 5865   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 5866   int start64 = 0;
 5867   if (vector64_count > max_unrolled_fill64) {
 5868     Label LOOP;
 5869     Register index = rtmp;
 5870 
 5871     start64 = vector64_count - (vector64_count % fill64_per_loop);
 5872 
 5873     movl(index, 0);
 5874     BIND(LOOP);
 5875     for (int i = 0; i < fill64_per_loop; i++) {
 5876       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 5877     }
 5878     addl(index, fill64_per_loop * 64);
 5879     cmpl(index, start64 * 64);
 5880     jccb(Assembler::less, LOOP);
 5881   }
 5882   for (int i = start64; i < vector64_count; i++) {
 5883     fill64(base, i * 64, xtmp, use64byteVector);
 5884   }
 5885 
 5886   // Clear remaining 64 byte tail.
 5887   int disp = vector64_count * 64;
 5888   if (cnt) {
 5889     switch (cnt) {
 5890       case 1:
 5891         movq(Address(base, disp), xtmp);
 5892         break;
 5893       case 2:
 5894         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 5895         break;
 5896       case 3:
 5897         movl(rtmp, 0x7);
 5898         kmovwl(mask, rtmp);
 5899         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 5900         break;
 5901       case 4:
 5902         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5903         break;
 5904       case 5:
 5905         if (use64byteVector) {
 5906           movl(rtmp, 0x1F);
 5907           kmovwl(mask, rtmp);
 5908           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5909         } else {
 5910           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5911           movq(Address(base, disp + 32), xtmp);
 5912         }
 5913         break;
 5914       case 6:
 5915         if (use64byteVector) {
 5916           movl(rtmp, 0x3F);
 5917           kmovwl(mask, rtmp);
 5918           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5919         } else {
 5920           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5921           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 5922         }
 5923         break;
 5924       case 7:
 5925         if (use64byteVector) {
 5926           movl(rtmp, 0x7F);
 5927           kmovwl(mask, rtmp);
 5928           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5929         } else {
 5930           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5931           movl(rtmp, 0x7);
 5932           kmovwl(mask, rtmp);
 5933           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 5934         }
 5935         break;
 5936       default:
 5937         fatal("Unexpected length : %d\n",cnt);
 5938         break;
 5939     }
 5940   }
 5941 }
 5942 
 5943 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
 5944                                bool is_large, KRegister mask) {
 5945   // cnt      - number of qwords (8-byte words).
 5946   // base     - start address, qword aligned.
 5947   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 5948   assert(base==rdi, "base register must be edi for rep stos");
 5949   assert(tmp==rax,   "tmp register must be eax for rep stos");
 5950   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 5951   assert(InitArrayShortSize % BytesPerLong == 0,
 5952     "InitArrayShortSize should be the multiple of BytesPerLong");
 5953 
 5954   Label DONE;
 5955   if (!is_large || !UseXMMForObjInit) {
 5956     xorptr(tmp, tmp);
 5957   }
 5958 
 5959   if (!is_large) {
 5960     Label LOOP, LONG;
 5961     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 5962     jccb(Assembler::greater, LONG);
 5963 
 5964     decrement(cnt);
 5965     jccb(Assembler::negative, DONE); // Zero length
 5966 
 5967     // Use individual pointer-sized stores for small counts:
 5968     BIND(LOOP);
 5969     movptr(Address(base, cnt, Address::times_ptr), tmp);
 5970     decrement(cnt);
 5971     jccb(Assembler::greaterEqual, LOOP);
 5972     jmpb(DONE);
 5973 
 5974     BIND(LONG);
 5975   }
 5976 
 5977   // Use longer rep-prefixed ops for non-small counts:
 5978   if (UseFastStosb) {
 5979     shlptr(cnt, 3); // convert to number of bytes
 5980     rep_stosb();
 5981   } else if (UseXMMForObjInit) {
 5982     xmm_clear_mem(base, cnt, tmp, xtmp, mask);
 5983   } else {
 5984     rep_stos();
 5985   }
 5986 
 5987   BIND(DONE);
 5988 }
 5989 
 5990 #endif //COMPILER2_OR_JVMCI
 5991 
 5992 
 5993 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 5994                                    Register to, Register value, Register count,
 5995                                    Register rtmp, XMMRegister xtmp) {
 5996   ShortBranchVerifier sbv(this);
 5997   assert_different_registers(to, value, count, rtmp);
 5998   Label L_exit;
 5999   Label L_fill_2_bytes, L_fill_4_bytes;
 6000 
 6001 #if defined(COMPILER2)
 6002   if(MaxVectorSize >=32 &&
 6003      VM_Version::supports_avx512vlbw() &&
 6004      VM_Version::supports_bmi2()) {
 6005     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6006     return;
 6007   }
 6008 #endif
 6009 
 6010   int shift = -1;
 6011   switch (t) {
 6012     case T_BYTE:
 6013       shift = 2;
 6014       break;
 6015     case T_SHORT:
 6016       shift = 1;
 6017       break;
 6018     case T_INT:
 6019       shift = 0;
 6020       break;
 6021     default: ShouldNotReachHere();
 6022   }
 6023 
 6024   if (t == T_BYTE) {
 6025     andl(value, 0xff);
 6026     movl(rtmp, value);
 6027     shll(rtmp, 8);
 6028     orl(value, rtmp);
 6029   }
 6030   if (t == T_SHORT) {
 6031     andl(value, 0xffff);
 6032   }
 6033   if (t == T_BYTE || t == T_SHORT) {
 6034     movl(rtmp, value);
 6035     shll(rtmp, 16);
 6036     orl(value, rtmp);
 6037   }
 6038 
 6039   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6040   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6041   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6042     Label L_skip_align2;
 6043     // align source address at 4 bytes address boundary
 6044     if (t == T_BYTE) {
 6045       Label L_skip_align1;
 6046       // One byte misalignment happens only for byte arrays
 6047       testptr(to, 1);
 6048       jccb(Assembler::zero, L_skip_align1);
 6049       movb(Address(to, 0), value);
 6050       increment(to);
 6051       decrement(count);
 6052       BIND(L_skip_align1);
 6053     }
 6054     // Two bytes misalignment happens only for byte and short (char) arrays
 6055     testptr(to, 2);
 6056     jccb(Assembler::zero, L_skip_align2);
 6057     movw(Address(to, 0), value);
 6058     addptr(to, 2);
 6059     subptr(count, 1<<(shift-1));
 6060     BIND(L_skip_align2);
 6061   }
 6062   {
 6063     Label L_fill_32_bytes;
 6064     if (!UseUnalignedLoadStores) {
 6065       // align to 8 bytes, we know we are 4 byte aligned to start
 6066       testptr(to, 4);
 6067       jccb(Assembler::zero, L_fill_32_bytes);
 6068       movl(Address(to, 0), value);
 6069       addptr(to, 4);
 6070       subptr(count, 1<<shift);
 6071     }
 6072     BIND(L_fill_32_bytes);
 6073     {
 6074       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6075       movdl(xtmp, value);
 6076       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6077         Label L_check_fill_32_bytes;
 6078         if (UseAVX > 2) {
 6079           // Fill 64-byte chunks
 6080           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6081 
 6082           // If number of bytes to fill < VM_Version::avx3_threshold(), perform fill using AVX2
 6083           cmpptr(count, VM_Version::avx3_threshold());
 6084           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6085 
 6086           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6087 
 6088           subptr(count, 16 << shift);
 6089           jccb(Assembler::less, L_check_fill_32_bytes);
 6090           align(16);
 6091 
 6092           BIND(L_fill_64_bytes_loop_avx3);
 6093           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6094           addptr(to, 64);
 6095           subptr(count, 16 << shift);
 6096           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6097           jmpb(L_check_fill_32_bytes);
 6098 
 6099           BIND(L_check_fill_64_bytes_avx2);
 6100         }
 6101         // Fill 64-byte chunks
 6102         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6103 
 6104         subptr(count, 16 << shift);
 6105         jcc(Assembler::less, L_check_fill_32_bytes);
 6106 
 6107         // align data for 64-byte chunks
 6108         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6109         if (EnableX86ECoreOpts) {
 6110             // align 'big' arrays to cache lines to minimize split_stores
 6111             cmpptr(count, 96 << shift);
 6112             jcc(Assembler::below, L_fill_64_bytes_loop);
 6113 
 6114             // Find the bytes needed for alignment
 6115             movptr(rtmp, to);
 6116             andptr(rtmp, 0x1c);
 6117             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6118             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6119             addptr(rtmp, 32);
 6120             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6121             subptr(count, rtmp);    // adjust count by number of elements
 6122 
 6123             align(16);
 6124             BIND(L_align_64_bytes_loop);
 6125             movdl(Address(to, 0), xtmp);
 6126             addptr(to, 4);
 6127             subptr(rtmp, 1 << shift);
 6128             jcc(Assembler::greater, L_align_64_bytes_loop);
 6129         }
 6130 
 6131         align(16);
 6132         BIND(L_fill_64_bytes_loop);
 6133         vmovdqu(Address(to, 0), xtmp);
 6134         vmovdqu(Address(to, 32), xtmp);
 6135         addptr(to, 64);
 6136         subptr(count, 16 << shift);
 6137         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6138 
 6139         align(16);
 6140         BIND(L_check_fill_32_bytes);
 6141         addptr(count, 8 << shift);
 6142         jccb(Assembler::less, L_check_fill_8_bytes);
 6143         vmovdqu(Address(to, 0), xtmp);
 6144         addptr(to, 32);
 6145         subptr(count, 8 << shift);
 6146 
 6147         BIND(L_check_fill_8_bytes);
 6148         // clean upper bits of YMM registers
 6149         movdl(xtmp, value);
 6150         pshufd(xtmp, xtmp, 0);
 6151       } else {
 6152         // Fill 32-byte chunks
 6153         pshufd(xtmp, xtmp, 0);
 6154 
 6155         subptr(count, 8 << shift);
 6156         jcc(Assembler::less, L_check_fill_8_bytes);
 6157         align(16);
 6158 
 6159         BIND(L_fill_32_bytes_loop);
 6160 
 6161         if (UseUnalignedLoadStores) {
 6162           movdqu(Address(to, 0), xtmp);
 6163           movdqu(Address(to, 16), xtmp);
 6164         } else {
 6165           movq(Address(to, 0), xtmp);
 6166           movq(Address(to, 8), xtmp);
 6167           movq(Address(to, 16), xtmp);
 6168           movq(Address(to, 24), xtmp);
 6169         }
 6170 
 6171         addptr(to, 32);
 6172         subptr(count, 8 << shift);
 6173         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6174 
 6175         BIND(L_check_fill_8_bytes);
 6176       }
 6177       addptr(count, 8 << shift);
 6178       jccb(Assembler::zero, L_exit);
 6179       jmpb(L_fill_8_bytes);
 6180 
 6181       //
 6182       // length is too short, just fill qwords
 6183       //
 6184       align(16);
 6185       BIND(L_fill_8_bytes_loop);
 6186       movq(Address(to, 0), xtmp);
 6187       addptr(to, 8);
 6188       BIND(L_fill_8_bytes);
 6189       subptr(count, 1 << (shift + 1));
 6190       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6191     }
 6192   }
 6193 
 6194   Label L_fill_4_bytes_loop;
 6195   testl(count, 1 << shift);
 6196   jccb(Assembler::zero, L_fill_2_bytes);
 6197 
 6198   align(16);
 6199   BIND(L_fill_4_bytes_loop);
 6200   movl(Address(to, 0), value);
 6201   addptr(to, 4);
 6202 
 6203   BIND(L_fill_4_bytes);
 6204   subptr(count, 1 << shift);
 6205   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 6206 
 6207   if (t == T_BYTE || t == T_SHORT) {
 6208     Label L_fill_byte;
 6209     BIND(L_fill_2_bytes);
 6210     // fill trailing 2 bytes
 6211     testl(count, 1<<(shift-1));
 6212     jccb(Assembler::zero, L_fill_byte);
 6213     movw(Address(to, 0), value);
 6214     if (t == T_BYTE) {
 6215       addptr(to, 2);
 6216       BIND(L_fill_byte);
 6217       // fill trailing byte
 6218       testl(count, 1);
 6219       jccb(Assembler::zero, L_exit);
 6220       movb(Address(to, 0), value);
 6221     } else {
 6222       BIND(L_fill_byte);
 6223     }
 6224   } else {
 6225     BIND(L_fill_2_bytes);
 6226   }
 6227   BIND(L_exit);
 6228 }
 6229 
 6230 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6231   switch(type) {
 6232     case T_BYTE:
 6233     case T_BOOLEAN:
 6234       evpbroadcastb(dst, src, vector_len);
 6235       break;
 6236     case T_SHORT:
 6237     case T_CHAR:
 6238       evpbroadcastw(dst, src, vector_len);
 6239       break;
 6240     case T_INT:
 6241     case T_FLOAT:
 6242       evpbroadcastd(dst, src, vector_len);
 6243       break;
 6244     case T_LONG:
 6245     case T_DOUBLE:
 6246       evpbroadcastq(dst, src, vector_len);
 6247       break;
 6248     default:
 6249       fatal("Unhandled type : %s", type2name(type));
 6250       break;
 6251   }
 6252 }
 6253 
 6254 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 6255 //
 6256 // @IntrinsicCandidate
 6257 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 6258 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6259 //     int i = 0;
 6260 //     for (; i < len; i++) {
 6261 //         char c = sa[sp++];
 6262 //         if (c > '\u00FF')
 6263 //             break;
 6264 //         da[dp++] = (byte) c;
 6265 //     }
 6266 //     return i;
 6267 // }
 6268 //
 6269 // @IntrinsicCandidate
 6270 // int java.lang.StringCoding.encodeISOArray0(
 6271 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 6272 //   int i = 0;
 6273 //   for (; i < len; i++) {
 6274 //     char c = StringUTF16.getChar(sa, sp++);
 6275 //     if (c > '\u00FF')
 6276 //       break;
 6277 //     da[dp++] = (byte) c;
 6278 //   }
 6279 //   return i;
 6280 // }
 6281 //
 6282 // @IntrinsicCandidate
 6283 // int java.lang.StringCoding.encodeAsciiArray0(
 6284 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6285 //   int i = 0;
 6286 //   for (; i < len; i++) {
 6287 //     char c = sa[sp++];
 6288 //     if (c >= '\u0080')
 6289 //       break;
 6290 //     da[dp++] = (byte) c;
 6291 //   }
 6292 //   return i;
 6293 // }
 6294 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 6295   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 6296   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 6297   Register tmp5, Register result, bool ascii) {
 6298 
 6299   // rsi: src
 6300   // rdi: dst
 6301   // rdx: len
 6302   // rcx: tmp5
 6303   // rax: result
 6304   ShortBranchVerifier sbv(this);
 6305   assert_different_registers(src, dst, len, tmp5, result);
 6306   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 6307 
 6308   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 6309   int short_mask = ascii ? 0xff80 : 0xff00;
 6310 
 6311   // set result
 6312   xorl(result, result);
 6313   // check for zero length
 6314   testl(len, len);
 6315   jcc(Assembler::zero, L_done);
 6316 
 6317   movl(result, len);
 6318 
 6319   // Setup pointers
 6320   lea(src, Address(src, len, Address::times_2)); // char[]
 6321   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 6322   negptr(len);
 6323 
 6324   if (UseSSE42Intrinsics || UseAVX >= 2) {
 6325     Label L_copy_8_chars, L_copy_8_chars_exit;
 6326     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 6327 
 6328     if (UseAVX >= 2) {
 6329       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 6330       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6331       movdl(tmp1Reg, tmp5);
 6332       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 6333       jmp(L_chars_32_check);
 6334 
 6335       bind(L_copy_32_chars);
 6336       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 6337       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 6338       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6339       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6340       jccb(Assembler::notZero, L_copy_32_chars_exit);
 6341       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6342       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 6343       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 6344 
 6345       bind(L_chars_32_check);
 6346       addptr(len, 32);
 6347       jcc(Assembler::lessEqual, L_copy_32_chars);
 6348 
 6349       bind(L_copy_32_chars_exit);
 6350       subptr(len, 16);
 6351       jccb(Assembler::greater, L_copy_16_chars_exit);
 6352 
 6353     } else if (UseSSE42Intrinsics) {
 6354       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6355       movdl(tmp1Reg, tmp5);
 6356       pshufd(tmp1Reg, tmp1Reg, 0);
 6357       jmpb(L_chars_16_check);
 6358     }
 6359 
 6360     bind(L_copy_16_chars);
 6361     if (UseAVX >= 2) {
 6362       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 6363       vptest(tmp2Reg, tmp1Reg);
 6364       jcc(Assembler::notZero, L_copy_16_chars_exit);
 6365       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 6366       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 6367     } else {
 6368       if (UseAVX > 0) {
 6369         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6370         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6371         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 6372       } else {
 6373         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6374         por(tmp2Reg, tmp3Reg);
 6375         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6376         por(tmp2Reg, tmp4Reg);
 6377       }
 6378       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6379       jccb(Assembler::notZero, L_copy_16_chars_exit);
 6380       packuswb(tmp3Reg, tmp4Reg);
 6381     }
 6382     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 6383 
 6384     bind(L_chars_16_check);
 6385     addptr(len, 16);
 6386     jcc(Assembler::lessEqual, L_copy_16_chars);
 6387 
 6388     bind(L_copy_16_chars_exit);
 6389     if (UseAVX >= 2) {
 6390       // clean upper bits of YMM registers
 6391       vpxor(tmp2Reg, tmp2Reg);
 6392       vpxor(tmp3Reg, tmp3Reg);
 6393       vpxor(tmp4Reg, tmp4Reg);
 6394       movdl(tmp1Reg, tmp5);
 6395       pshufd(tmp1Reg, tmp1Reg, 0);
 6396     }
 6397     subptr(len, 8);
 6398     jccb(Assembler::greater, L_copy_8_chars_exit);
 6399 
 6400     bind(L_copy_8_chars);
 6401     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 6402     ptest(tmp3Reg, tmp1Reg);
 6403     jccb(Assembler::notZero, L_copy_8_chars_exit);
 6404     packuswb(tmp3Reg, tmp1Reg);
 6405     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 6406     addptr(len, 8);
 6407     jccb(Assembler::lessEqual, L_copy_8_chars);
 6408 
 6409     bind(L_copy_8_chars_exit);
 6410     subptr(len, 8);
 6411     jccb(Assembler::zero, L_done);
 6412   }
 6413 
 6414   bind(L_copy_1_char);
 6415   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 6416   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 6417   jccb(Assembler::notZero, L_copy_1_char_exit);
 6418   movb(Address(dst, len, Address::times_1, 0), tmp5);
 6419   addptr(len, 1);
 6420   jccb(Assembler::less, L_copy_1_char);
 6421 
 6422   bind(L_copy_1_char_exit);
 6423   addptr(result, len); // len is negative count of not processed elements
 6424 
 6425   bind(L_done);
 6426 }
 6427 
 6428 /**
 6429  * Helper for multiply_to_len().
 6430  */
 6431 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 6432   addq(dest_lo, src1);
 6433   adcq(dest_hi, 0);
 6434   addq(dest_lo, src2);
 6435   adcq(dest_hi, 0);
 6436 }
 6437 
 6438 /**
 6439  * Multiply 64 bit by 64 bit first loop.
 6440  */
 6441 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 6442                                            Register y, Register y_idx, Register z,
 6443                                            Register carry, Register product,
 6444                                            Register idx, Register kdx) {
 6445   //
 6446   //  jlong carry, x[], y[], z[];
 6447   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6448   //    huge_128 product = y[idx] * x[xstart] + carry;
 6449   //    z[kdx] = (jlong)product;
 6450   //    carry  = (jlong)(product >>> 64);
 6451   //  }
 6452   //  z[xstart] = carry;
 6453   //
 6454 
 6455   Label L_first_loop, L_first_loop_exit;
 6456   Label L_one_x, L_one_y, L_multiply;
 6457 
 6458   decrementl(xstart);
 6459   jcc(Assembler::negative, L_one_x);
 6460 
 6461   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6462   rorq(x_xstart, 32); // convert big-endian to little-endian
 6463 
 6464   bind(L_first_loop);
 6465   decrementl(idx);
 6466   jcc(Assembler::negative, L_first_loop_exit);
 6467   decrementl(idx);
 6468   jcc(Assembler::negative, L_one_y);
 6469   movq(y_idx, Address(y, idx, Address::times_4,  0));
 6470   rorq(y_idx, 32); // convert big-endian to little-endian
 6471   bind(L_multiply);
 6472   movq(product, x_xstart);
 6473   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 6474   addq(product, carry);
 6475   adcq(rdx, 0);
 6476   subl(kdx, 2);
 6477   movl(Address(z, kdx, Address::times_4,  4), product);
 6478   shrq(product, 32);
 6479   movl(Address(z, kdx, Address::times_4,  0), product);
 6480   movq(carry, rdx);
 6481   jmp(L_first_loop);
 6482 
 6483   bind(L_one_y);
 6484   movl(y_idx, Address(y,  0));
 6485   jmp(L_multiply);
 6486 
 6487   bind(L_one_x);
 6488   movl(x_xstart, Address(x,  0));
 6489   jmp(L_first_loop);
 6490 
 6491   bind(L_first_loop_exit);
 6492 }
 6493 
 6494 /**
 6495  * Multiply 64 bit by 64 bit and add 128 bit.
 6496  */
 6497 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 6498                                             Register yz_idx, Register idx,
 6499                                             Register carry, Register product, int offset) {
 6500   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 6501   //     z[kdx] = (jlong)product;
 6502 
 6503   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 6504   rorq(yz_idx, 32); // convert big-endian to little-endian
 6505   movq(product, x_xstart);
 6506   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 6507   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 6508   rorq(yz_idx, 32); // convert big-endian to little-endian
 6509 
 6510   add2_with_carry(rdx, product, carry, yz_idx);
 6511 
 6512   movl(Address(z, idx, Address::times_4,  offset+4), product);
 6513   shrq(product, 32);
 6514   movl(Address(z, idx, Address::times_4,  offset), product);
 6515 
 6516 }
 6517 
 6518 /**
 6519  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 6520  */
 6521 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 6522                                              Register yz_idx, Register idx, Register jdx,
 6523                                              Register carry, Register product,
 6524                                              Register carry2) {
 6525   //   jlong carry, x[], y[], z[];
 6526   //   int kdx = ystart+1;
 6527   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6528   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 6529   //     z[kdx+idx+1] = (jlong)product;
 6530   //     jlong carry2  = (jlong)(product >>> 64);
 6531   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 6532   //     z[kdx+idx] = (jlong)product;
 6533   //     carry  = (jlong)(product >>> 64);
 6534   //   }
 6535   //   idx += 2;
 6536   //   if (idx > 0) {
 6537   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 6538   //     z[kdx+idx] = (jlong)product;
 6539   //     carry  = (jlong)(product >>> 64);
 6540   //   }
 6541   //
 6542 
 6543   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6544 
 6545   movl(jdx, idx);
 6546   andl(jdx, 0xFFFFFFFC);
 6547   shrl(jdx, 2);
 6548 
 6549   bind(L_third_loop);
 6550   subl(jdx, 1);
 6551   jcc(Assembler::negative, L_third_loop_exit);
 6552   subl(idx, 4);
 6553 
 6554   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 6555   movq(carry2, rdx);
 6556 
 6557   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 6558   movq(carry, rdx);
 6559   jmp(L_third_loop);
 6560 
 6561   bind (L_third_loop_exit);
 6562 
 6563   andl (idx, 0x3);
 6564   jcc(Assembler::zero, L_post_third_loop_done);
 6565 
 6566   Label L_check_1;
 6567   subl(idx, 2);
 6568   jcc(Assembler::negative, L_check_1);
 6569 
 6570   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 6571   movq(carry, rdx);
 6572 
 6573   bind (L_check_1);
 6574   addl (idx, 0x2);
 6575   andl (idx, 0x1);
 6576   subl(idx, 1);
 6577   jcc(Assembler::negative, L_post_third_loop_done);
 6578 
 6579   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 6580   movq(product, x_xstart);
 6581   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 6582   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 6583 
 6584   add2_with_carry(rdx, product, yz_idx, carry);
 6585 
 6586   movl(Address(z, idx, Address::times_4,  0), product);
 6587   shrq(product, 32);
 6588 
 6589   shlq(rdx, 32);
 6590   orq(product, rdx);
 6591   movq(carry, product);
 6592 
 6593   bind(L_post_third_loop_done);
 6594 }
 6595 
 6596 /**
 6597  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 6598  *
 6599  */
 6600 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 6601                                                   Register carry, Register carry2,
 6602                                                   Register idx, Register jdx,
 6603                                                   Register yz_idx1, Register yz_idx2,
 6604                                                   Register tmp, Register tmp3, Register tmp4) {
 6605   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 6606 
 6607   //   jlong carry, x[], y[], z[];
 6608   //   int kdx = ystart+1;
 6609   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6610   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 6611   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 6612   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 6613   //     carry  = (jlong)(tmp4 >>> 64);
 6614   //     z[kdx+idx+1] = (jlong)tmp3;
 6615   //     z[kdx+idx] = (jlong)tmp4;
 6616   //   }
 6617   //   idx += 2;
 6618   //   if (idx > 0) {
 6619   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 6620   //     z[kdx+idx] = (jlong)yz_idx1;
 6621   //     carry  = (jlong)(yz_idx1 >>> 64);
 6622   //   }
 6623   //
 6624 
 6625   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6626 
 6627   movl(jdx, idx);
 6628   andl(jdx, 0xFFFFFFFC);
 6629   shrl(jdx, 2);
 6630 
 6631   bind(L_third_loop);
 6632   subl(jdx, 1);
 6633   jcc(Assembler::negative, L_third_loop_exit);
 6634   subl(idx, 4);
 6635 
 6636   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 6637   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 6638   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 6639   rorxq(yz_idx2, yz_idx2, 32);
 6640 
 6641   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 6642   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 6643 
 6644   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 6645   rorxq(yz_idx1, yz_idx1, 32);
 6646   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6647   rorxq(yz_idx2, yz_idx2, 32);
 6648 
 6649   if (VM_Version::supports_adx()) {
 6650     adcxq(tmp3, carry);
 6651     adoxq(tmp3, yz_idx1);
 6652 
 6653     adcxq(tmp4, tmp);
 6654     adoxq(tmp4, yz_idx2);
 6655 
 6656     movl(carry, 0); // does not affect flags
 6657     adcxq(carry2, carry);
 6658     adoxq(carry2, carry);
 6659   } else {
 6660     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 6661     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 6662   }
 6663   movq(carry, carry2);
 6664 
 6665   movl(Address(z, idx, Address::times_4, 12), tmp3);
 6666   shrq(tmp3, 32);
 6667   movl(Address(z, idx, Address::times_4,  8), tmp3);
 6668 
 6669   movl(Address(z, idx, Address::times_4,  4), tmp4);
 6670   shrq(tmp4, 32);
 6671   movl(Address(z, idx, Address::times_4,  0), tmp4);
 6672 
 6673   jmp(L_third_loop);
 6674 
 6675   bind (L_third_loop_exit);
 6676 
 6677   andl (idx, 0x3);
 6678   jcc(Assembler::zero, L_post_third_loop_done);
 6679 
 6680   Label L_check_1;
 6681   subl(idx, 2);
 6682   jcc(Assembler::negative, L_check_1);
 6683 
 6684   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 6685   rorxq(yz_idx1, yz_idx1, 32);
 6686   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 6687   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6688   rorxq(yz_idx2, yz_idx2, 32);
 6689 
 6690   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 6691 
 6692   movl(Address(z, idx, Address::times_4,  4), tmp3);
 6693   shrq(tmp3, 32);
 6694   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6695   movq(carry, tmp4);
 6696 
 6697   bind (L_check_1);
 6698   addl (idx, 0x2);
 6699   andl (idx, 0x1);
 6700   subl(idx, 1);
 6701   jcc(Assembler::negative, L_post_third_loop_done);
 6702   movl(tmp4, Address(y, idx, Address::times_4,  0));
 6703   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 6704   movl(tmp4, Address(z, idx, Address::times_4,  0));
 6705 
 6706   add2_with_carry(carry2, tmp3, tmp4, carry);
 6707 
 6708   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6709   shrq(tmp3, 32);
 6710 
 6711   shlq(carry2, 32);
 6712   orq(tmp3, carry2);
 6713   movq(carry, tmp3);
 6714 
 6715   bind(L_post_third_loop_done);
 6716 }
 6717 
 6718 /**
 6719  * Code for BigInteger::multiplyToLen() intrinsic.
 6720  *
 6721  * rdi: x
 6722  * rax: xlen
 6723  * rsi: y
 6724  * rcx: ylen
 6725  * r8:  z
 6726  * r11: tmp0
 6727  * r12: tmp1
 6728  * r13: tmp2
 6729  * r14: tmp3
 6730  * r15: tmp4
 6731  * rbx: tmp5
 6732  *
 6733  */
 6734 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 6735                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 6736   ShortBranchVerifier sbv(this);
 6737   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 6738 
 6739   push(tmp0);
 6740   push(tmp1);
 6741   push(tmp2);
 6742   push(tmp3);
 6743   push(tmp4);
 6744   push(tmp5);
 6745 
 6746   push(xlen);
 6747 
 6748   const Register idx = tmp1;
 6749   const Register kdx = tmp2;
 6750   const Register xstart = tmp3;
 6751 
 6752   const Register y_idx = tmp4;
 6753   const Register carry = tmp5;
 6754   const Register product  = xlen;
 6755   const Register x_xstart = tmp0;
 6756 
 6757   // First Loop.
 6758   //
 6759   //  final static long LONG_MASK = 0xffffffffL;
 6760   //  int xstart = xlen - 1;
 6761   //  int ystart = ylen - 1;
 6762   //  long carry = 0;
 6763   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6764   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 6765   //    z[kdx] = (int)product;
 6766   //    carry = product >>> 32;
 6767   //  }
 6768   //  z[xstart] = (int)carry;
 6769   //
 6770 
 6771   movl(idx, ylen);               // idx = ylen;
 6772   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 6773   xorq(carry, carry);            // carry = 0;
 6774 
 6775   Label L_done;
 6776 
 6777   movl(xstart, xlen);
 6778   decrementl(xstart);
 6779   jcc(Assembler::negative, L_done);
 6780 
 6781   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 6782 
 6783   Label L_second_loop;
 6784   testl(kdx, kdx);
 6785   jcc(Assembler::zero, L_second_loop);
 6786 
 6787   Label L_carry;
 6788   subl(kdx, 1);
 6789   jcc(Assembler::zero, L_carry);
 6790 
 6791   movl(Address(z, kdx, Address::times_4,  0), carry);
 6792   shrq(carry, 32);
 6793   subl(kdx, 1);
 6794 
 6795   bind(L_carry);
 6796   movl(Address(z, kdx, Address::times_4,  0), carry);
 6797 
 6798   // Second and third (nested) loops.
 6799   //
 6800   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 6801   //   carry = 0;
 6802   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 6803   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 6804   //                    (z[k] & LONG_MASK) + carry;
 6805   //     z[k] = (int)product;
 6806   //     carry = product >>> 32;
 6807   //   }
 6808   //   z[i] = (int)carry;
 6809   // }
 6810   //
 6811   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 6812 
 6813   const Register jdx = tmp1;
 6814 
 6815   bind(L_second_loop);
 6816   xorl(carry, carry);    // carry = 0;
 6817   movl(jdx, ylen);       // j = ystart+1
 6818 
 6819   subl(xstart, 1);       // i = xstart-1;
 6820   jcc(Assembler::negative, L_done);
 6821 
 6822   push (z);
 6823 
 6824   Label L_last_x;
 6825   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 6826   subl(xstart, 1);       // i = xstart-1;
 6827   jcc(Assembler::negative, L_last_x);
 6828 
 6829   if (UseBMI2Instructions) {
 6830     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 6831     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 6832   } else {
 6833     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6834     rorq(x_xstart, 32);  // convert big-endian to little-endian
 6835   }
 6836 
 6837   Label L_third_loop_prologue;
 6838   bind(L_third_loop_prologue);
 6839 
 6840   push (x);
 6841   push (xstart);
 6842   push (ylen);
 6843 
 6844 
 6845   if (UseBMI2Instructions) {
 6846     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 6847   } else { // !UseBMI2Instructions
 6848     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 6849   }
 6850 
 6851   pop(ylen);
 6852   pop(xlen);
 6853   pop(x);
 6854   pop(z);
 6855 
 6856   movl(tmp3, xlen);
 6857   addl(tmp3, 1);
 6858   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6859   subl(tmp3, 1);
 6860   jccb(Assembler::negative, L_done);
 6861 
 6862   shrq(carry, 32);
 6863   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6864   jmp(L_second_loop);
 6865 
 6866   // Next infrequent code is moved outside loops.
 6867   bind(L_last_x);
 6868   if (UseBMI2Instructions) {
 6869     movl(rdx, Address(x,  0));
 6870   } else {
 6871     movl(x_xstart, Address(x,  0));
 6872   }
 6873   jmp(L_third_loop_prologue);
 6874 
 6875   bind(L_done);
 6876 
 6877   pop(xlen);
 6878 
 6879   pop(tmp5);
 6880   pop(tmp4);
 6881   pop(tmp3);
 6882   pop(tmp2);
 6883   pop(tmp1);
 6884   pop(tmp0);
 6885 }
 6886 
 6887 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 6888   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 6889   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 6890   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 6891   Label VECTOR8_TAIL, VECTOR4_TAIL;
 6892   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 6893   Label SAME_TILL_END, DONE;
 6894   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 6895 
 6896   //scale is in rcx in both Win64 and Unix
 6897   ShortBranchVerifier sbv(this);
 6898 
 6899   shlq(length);
 6900   xorq(result, result);
 6901 
 6902   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 6903       VM_Version::supports_avx512vlbw()) {
 6904     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 6905 
 6906     cmpq(length, 64);
 6907     jcc(Assembler::less, VECTOR32_TAIL);
 6908 
 6909     movq(tmp1, length);
 6910     andq(tmp1, 0x3F);      // tail count
 6911     andq(length, ~(0x3F)); //vector count
 6912 
 6913     bind(VECTOR64_LOOP);
 6914     // AVX512 code to compare 64 byte vectors.
 6915     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 6916     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6917     kortestql(k7, k7);
 6918     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 6919     addq(result, 64);
 6920     subq(length, 64);
 6921     jccb(Assembler::notZero, VECTOR64_LOOP);
 6922 
 6923     //bind(VECTOR64_TAIL);
 6924     testq(tmp1, tmp1);
 6925     jcc(Assembler::zero, SAME_TILL_END);
 6926 
 6927     //bind(VECTOR64_TAIL);
 6928     // AVX512 code to compare up to 63 byte vectors.
 6929     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 6930     shlxq(tmp2, tmp2, tmp1);
 6931     notq(tmp2);
 6932     kmovql(k3, tmp2);
 6933 
 6934     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 6935     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6936 
 6937     ktestql(k7, k3);
 6938     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 6939 
 6940     bind(VECTOR64_NOT_EQUAL);
 6941     kmovql(tmp1, k7);
 6942     notq(tmp1);
 6943     tzcntq(tmp1, tmp1);
 6944     addq(result, tmp1);
 6945     shrq(result);
 6946     jmp(DONE);
 6947     bind(VECTOR32_TAIL);
 6948   }
 6949 
 6950   cmpq(length, 8);
 6951   jcc(Assembler::equal, VECTOR8_LOOP);
 6952   jcc(Assembler::less, VECTOR4_TAIL);
 6953 
 6954   if (UseAVX >= 2) {
 6955     Label VECTOR16_TAIL, VECTOR32_LOOP;
 6956 
 6957     cmpq(length, 16);
 6958     jcc(Assembler::equal, VECTOR16_LOOP);
 6959     jcc(Assembler::less, VECTOR8_LOOP);
 6960 
 6961     cmpq(length, 32);
 6962     jccb(Assembler::less, VECTOR16_TAIL);
 6963 
 6964     subq(length, 32);
 6965     bind(VECTOR32_LOOP);
 6966     vmovdqu(rymm0, Address(obja, result));
 6967     vmovdqu(rymm1, Address(objb, result));
 6968     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 6969     vptest(rymm2, rymm2);
 6970     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 6971     addq(result, 32);
 6972     subq(length, 32);
 6973     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 6974     addq(length, 32);
 6975     jcc(Assembler::equal, SAME_TILL_END);
 6976     //falling through if less than 32 bytes left //close the branch here.
 6977 
 6978     bind(VECTOR16_TAIL);
 6979     cmpq(length, 16);
 6980     jccb(Assembler::less, VECTOR8_TAIL);
 6981     bind(VECTOR16_LOOP);
 6982     movdqu(rymm0, Address(obja, result));
 6983     movdqu(rymm1, Address(objb, result));
 6984     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 6985     ptest(rymm2, rymm2);
 6986     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 6987     addq(result, 16);
 6988     subq(length, 16);
 6989     jcc(Assembler::equal, SAME_TILL_END);
 6990     //falling through if less than 16 bytes left
 6991   } else {//regular intrinsics
 6992 
 6993     cmpq(length, 16);
 6994     jccb(Assembler::less, VECTOR8_TAIL);
 6995 
 6996     subq(length, 16);
 6997     bind(VECTOR16_LOOP);
 6998     movdqu(rymm0, Address(obja, result));
 6999     movdqu(rymm1, Address(objb, result));
 7000     pxor(rymm0, rymm1);
 7001     ptest(rymm0, rymm0);
 7002     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7003     addq(result, 16);
 7004     subq(length, 16);
 7005     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7006     addq(length, 16);
 7007     jcc(Assembler::equal, SAME_TILL_END);
 7008     //falling through if less than 16 bytes left
 7009   }
 7010 
 7011   bind(VECTOR8_TAIL);
 7012   cmpq(length, 8);
 7013   jccb(Assembler::less, VECTOR4_TAIL);
 7014   bind(VECTOR8_LOOP);
 7015   movq(tmp1, Address(obja, result));
 7016   movq(tmp2, Address(objb, result));
 7017   xorq(tmp1, tmp2);
 7018   testq(tmp1, tmp1);
 7019   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7020   addq(result, 8);
 7021   subq(length, 8);
 7022   jcc(Assembler::equal, SAME_TILL_END);
 7023   //falling through if less than 8 bytes left
 7024 
 7025   bind(VECTOR4_TAIL);
 7026   cmpq(length, 4);
 7027   jccb(Assembler::less, BYTES_TAIL);
 7028   bind(VECTOR4_LOOP);
 7029   movl(tmp1, Address(obja, result));
 7030   xorl(tmp1, Address(objb, result));
 7031   testl(tmp1, tmp1);
 7032   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7033   addq(result, 4);
 7034   subq(length, 4);
 7035   jcc(Assembler::equal, SAME_TILL_END);
 7036   //falling through if less than 4 bytes left
 7037 
 7038   bind(BYTES_TAIL);
 7039   bind(BYTES_LOOP);
 7040   load_unsigned_byte(tmp1, Address(obja, result));
 7041   load_unsigned_byte(tmp2, Address(objb, result));
 7042   xorl(tmp1, tmp2);
 7043   testl(tmp1, tmp1);
 7044   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7045   decq(length);
 7046   jcc(Assembler::zero, SAME_TILL_END);
 7047   incq(result);
 7048   load_unsigned_byte(tmp1, Address(obja, result));
 7049   load_unsigned_byte(tmp2, Address(objb, result));
 7050   xorl(tmp1, tmp2);
 7051   testl(tmp1, tmp1);
 7052   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7053   decq(length);
 7054   jcc(Assembler::zero, SAME_TILL_END);
 7055   incq(result);
 7056   load_unsigned_byte(tmp1, Address(obja, result));
 7057   load_unsigned_byte(tmp2, Address(objb, result));
 7058   xorl(tmp1, tmp2);
 7059   testl(tmp1, tmp1);
 7060   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7061   jmp(SAME_TILL_END);
 7062 
 7063   if (UseAVX >= 2) {
 7064     bind(VECTOR32_NOT_EQUAL);
 7065     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7066     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7067     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7068     vpmovmskb(tmp1, rymm0);
 7069     bsfq(tmp1, tmp1);
 7070     addq(result, tmp1);
 7071     shrq(result);
 7072     jmp(DONE);
 7073   }
 7074 
 7075   bind(VECTOR16_NOT_EQUAL);
 7076   if (UseAVX >= 2) {
 7077     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7078     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7079     pxor(rymm0, rymm2);
 7080   } else {
 7081     pcmpeqb(rymm2, rymm2);
 7082     pxor(rymm0, rymm1);
 7083     pcmpeqb(rymm0, rymm1);
 7084     pxor(rymm0, rymm2);
 7085   }
 7086   pmovmskb(tmp1, rymm0);
 7087   bsfq(tmp1, tmp1);
 7088   addq(result, tmp1);
 7089   shrq(result);
 7090   jmpb(DONE);
 7091 
 7092   bind(VECTOR8_NOT_EQUAL);
 7093   bind(VECTOR4_NOT_EQUAL);
 7094   bsfq(tmp1, tmp1);
 7095   shrq(tmp1, 3);
 7096   addq(result, tmp1);
 7097   bind(BYTES_NOT_EQUAL);
 7098   shrq(result);
 7099   jmpb(DONE);
 7100 
 7101   bind(SAME_TILL_END);
 7102   mov64(result, -1);
 7103 
 7104   bind(DONE);
 7105 }
 7106 
 7107 //Helper functions for square_to_len()
 7108 
 7109 /**
 7110  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7111  * Preserves x and z and modifies rest of the registers.
 7112  */
 7113 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7114   // Perform square and right shift by 1
 7115   // Handle odd xlen case first, then for even xlen do the following
 7116   // jlong carry = 0;
 7117   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7118   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7119   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7120   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7121   //     carry = (jlong)product;
 7122   // }
 7123 
 7124   xorq(tmp5, tmp5);     // carry
 7125   xorq(rdxReg, rdxReg);
 7126   xorl(tmp1, tmp1);     // index for x
 7127   xorl(tmp4, tmp4);     // index for z
 7128 
 7129   Label L_first_loop, L_first_loop_exit;
 7130 
 7131   testl(xlen, 1);
 7132   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7133 
 7134   // Square and right shift by 1 the odd element using 32 bit multiply
 7135   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7136   imulq(raxReg, raxReg);
 7137   shrq(raxReg, 1);
 7138   adcq(tmp5, 0);
 7139   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7140   incrementl(tmp1);
 7141   addl(tmp4, 2);
 7142 
 7143   // Square and  right shift by 1 the rest using 64 bit multiply
 7144   bind(L_first_loop);
 7145   cmpptr(tmp1, xlen);
 7146   jccb(Assembler::equal, L_first_loop_exit);
 7147 
 7148   // Square
 7149   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7150   rorq(raxReg, 32);    // convert big-endian to little-endian
 7151   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7152 
 7153   // Right shift by 1 and save carry
 7154   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7155   rcrq(rdxReg, 1);
 7156   rcrq(raxReg, 1);
 7157   adcq(tmp5, 0);
 7158 
 7159   // Store result in z
 7160   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7161   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7162 
 7163   // Update indices for x and z
 7164   addl(tmp1, 2);
 7165   addl(tmp4, 4);
 7166   jmp(L_first_loop);
 7167 
 7168   bind(L_first_loop_exit);
 7169 }
 7170 
 7171 
 7172 /**
 7173  * Perform the following multiply add operation using BMI2 instructions
 7174  * carry:sum = sum + op1*op2 + carry
 7175  * op2 should be in rdx
 7176  * op2 is preserved, all other registers are modified
 7177  */
 7178 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7179   // assert op2 is rdx
 7180   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7181   addq(sum, carry);
 7182   adcq(tmp2, 0);
 7183   addq(sum, op1);
 7184   adcq(tmp2, 0);
 7185   movq(carry, tmp2);
 7186 }
 7187 
 7188 /**
 7189  * Perform the following multiply add operation:
 7190  * carry:sum = sum + op1*op2 + carry
 7191  * Preserves op1, op2 and modifies rest of registers
 7192  */
 7193 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7194   // rdx:rax = op1 * op2
 7195   movq(raxReg, op2);
 7196   mulq(op1);
 7197 
 7198   //  rdx:rax = sum + carry + rdx:rax
 7199   addq(sum, carry);
 7200   adcq(rdxReg, 0);
 7201   addq(sum, raxReg);
 7202   adcq(rdxReg, 0);
 7203 
 7204   // carry:sum = rdx:sum
 7205   movq(carry, rdxReg);
 7206 }
 7207 
 7208 /**
 7209  * Add 64 bit long carry into z[] with carry propagation.
 7210  * Preserves z and carry register values and modifies rest of registers.
 7211  *
 7212  */
 7213 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7214   Label L_fourth_loop, L_fourth_loop_exit;
 7215 
 7216   movl(tmp1, 1);
 7217   subl(zlen, 2);
 7218   addq(Address(z, zlen, Address::times_4, 0), carry);
 7219 
 7220   bind(L_fourth_loop);
 7221   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7222   subl(zlen, 2);
 7223   jccb(Assembler::negative, L_fourth_loop_exit);
 7224   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7225   jmp(L_fourth_loop);
 7226   bind(L_fourth_loop_exit);
 7227 }
 7228 
 7229 /**
 7230  * Shift z[] left by 1 bit.
 7231  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7232  *
 7233  */
 7234 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7235 
 7236   Label L_fifth_loop, L_fifth_loop_exit;
 7237 
 7238   // Fifth loop
 7239   // Perform primitiveLeftShift(z, zlen, 1)
 7240 
 7241   const Register prev_carry = tmp1;
 7242   const Register new_carry = tmp4;
 7243   const Register value = tmp2;
 7244   const Register zidx = tmp3;
 7245 
 7246   // int zidx, carry;
 7247   // long value;
 7248   // carry = 0;
 7249   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 7250   //    (carry:value)  = (z[i] << 1) | carry ;
 7251   //    z[i] = value;
 7252   // }
 7253 
 7254   movl(zidx, zlen);
 7255   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 7256 
 7257   bind(L_fifth_loop);
 7258   decl(zidx);  // Use decl to preserve carry flag
 7259   decl(zidx);
 7260   jccb(Assembler::negative, L_fifth_loop_exit);
 7261 
 7262   if (UseBMI2Instructions) {
 7263      movq(value, Address(z, zidx, Address::times_4, 0));
 7264      rclq(value, 1);
 7265      rorxq(value, value, 32);
 7266      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7267   }
 7268   else {
 7269     // clear new_carry
 7270     xorl(new_carry, new_carry);
 7271 
 7272     // Shift z[i] by 1, or in previous carry and save new carry
 7273     movq(value, Address(z, zidx, Address::times_4, 0));
 7274     shlq(value, 1);
 7275     adcl(new_carry, 0);
 7276 
 7277     orq(value, prev_carry);
 7278     rorq(value, 0x20);
 7279     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7280 
 7281     // Set previous carry = new carry
 7282     movl(prev_carry, new_carry);
 7283   }
 7284   jmp(L_fifth_loop);
 7285 
 7286   bind(L_fifth_loop_exit);
 7287 }
 7288 
 7289 
 7290 /**
 7291  * Code for BigInteger::squareToLen() intrinsic
 7292  *
 7293  * rdi: x
 7294  * rsi: len
 7295  * r8:  z
 7296  * rcx: zlen
 7297  * r12: tmp1
 7298  * r13: tmp2
 7299  * r14: tmp3
 7300  * r15: tmp4
 7301  * rbx: tmp5
 7302  *
 7303  */
 7304 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) {
 7305 
 7306   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 7307   push(tmp1);
 7308   push(tmp2);
 7309   push(tmp3);
 7310   push(tmp4);
 7311   push(tmp5);
 7312 
 7313   // First loop
 7314   // Store the squares, right shifted one bit (i.e., divided by 2).
 7315   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7316 
 7317   // Add in off-diagonal sums.
 7318   //
 7319   // Second, third (nested) and fourth loops.
 7320   // zlen +=2;
 7321   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 7322   //    carry = 0;
 7323   //    long op2 = x[xidx:xidx+1];
 7324   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 7325   //       k -= 2;
 7326   //       long op1 = x[j:j+1];
 7327   //       long sum = z[k:k+1];
 7328   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 7329   //       z[k:k+1] = sum;
 7330   //    }
 7331   //    add_one_64(z, k, carry, tmp_regs);
 7332   // }
 7333 
 7334   const Register carry = tmp5;
 7335   const Register sum = tmp3;
 7336   const Register op1 = tmp4;
 7337   Register op2 = tmp2;
 7338 
 7339   push(zlen);
 7340   push(len);
 7341   addl(zlen,2);
 7342   bind(L_second_loop);
 7343   xorq(carry, carry);
 7344   subl(zlen, 4);
 7345   subl(len, 2);
 7346   push(zlen);
 7347   push(len);
 7348   cmpl(len, 0);
 7349   jccb(Assembler::lessEqual, L_second_loop_exit);
 7350 
 7351   // Multiply an array by one 64 bit long.
 7352   if (UseBMI2Instructions) {
 7353     op2 = rdxReg;
 7354     movq(op2, Address(x, len, Address::times_4,  0));
 7355     rorxq(op2, op2, 32);
 7356   }
 7357   else {
 7358     movq(op2, Address(x, len, Address::times_4,  0));
 7359     rorq(op2, 32);
 7360   }
 7361 
 7362   bind(L_third_loop);
 7363   decrementl(len);
 7364   jccb(Assembler::negative, L_third_loop_exit);
 7365   decrementl(len);
 7366   jccb(Assembler::negative, L_last_x);
 7367 
 7368   movq(op1, Address(x, len, Address::times_4,  0));
 7369   rorq(op1, 32);
 7370 
 7371   bind(L_multiply);
 7372   subl(zlen, 2);
 7373   movq(sum, Address(z, zlen, Address::times_4,  0));
 7374 
 7375   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 7376   if (UseBMI2Instructions) {
 7377     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 7378   }
 7379   else {
 7380     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7381   }
 7382 
 7383   movq(Address(z, zlen, Address::times_4, 0), sum);
 7384 
 7385   jmp(L_third_loop);
 7386   bind(L_third_loop_exit);
 7387 
 7388   // Fourth loop
 7389   // Add 64 bit long carry into z with carry propagation.
 7390   // Uses offsetted zlen.
 7391   add_one_64(z, zlen, carry, tmp1);
 7392 
 7393   pop(len);
 7394   pop(zlen);
 7395   jmp(L_second_loop);
 7396 
 7397   // Next infrequent code is moved outside loops.
 7398   bind(L_last_x);
 7399   movl(op1, Address(x, 0));
 7400   jmp(L_multiply);
 7401 
 7402   bind(L_second_loop_exit);
 7403   pop(len);
 7404   pop(zlen);
 7405   pop(len);
 7406   pop(zlen);
 7407 
 7408   // Fifth loop
 7409   // Shift z left 1 bit.
 7410   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 7411 
 7412   // z[zlen-1] |= x[len-1] & 1;
 7413   movl(tmp3, Address(x, len, Address::times_4, -4));
 7414   andl(tmp3, 1);
 7415   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 7416 
 7417   pop(tmp5);
 7418   pop(tmp4);
 7419   pop(tmp3);
 7420   pop(tmp2);
 7421   pop(tmp1);
 7422 }
 7423 
 7424 /**
 7425  * Helper function for mul_add()
 7426  * Multiply the in[] by int k and add to out[] starting at offset offs using
 7427  * 128 bit by 32 bit multiply and return the carry in tmp5.
 7428  * Only quad int aligned length of in[] is operated on in this function.
 7429  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 7430  * This function preserves out, in and k registers.
 7431  * len and offset point to the appropriate index in "in" & "out" correspondingly
 7432  * tmp5 has the carry.
 7433  * other registers are temporary and are modified.
 7434  *
 7435  */
 7436 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 7437   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 7438   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7439 
 7440   Label L_first_loop, L_first_loop_exit;
 7441 
 7442   movl(tmp1, len);
 7443   shrl(tmp1, 2);
 7444 
 7445   bind(L_first_loop);
 7446   subl(tmp1, 1);
 7447   jccb(Assembler::negative, L_first_loop_exit);
 7448 
 7449   subl(len, 4);
 7450   subl(offset, 4);
 7451 
 7452   Register op2 = tmp2;
 7453   const Register sum = tmp3;
 7454   const Register op1 = tmp4;
 7455   const Register carry = tmp5;
 7456 
 7457   if (UseBMI2Instructions) {
 7458     op2 = rdxReg;
 7459   }
 7460 
 7461   movq(op1, Address(in, len, Address::times_4,  8));
 7462   rorq(op1, 32);
 7463   movq(sum, Address(out, offset, Address::times_4,  8));
 7464   rorq(sum, 32);
 7465   if (UseBMI2Instructions) {
 7466     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7467   }
 7468   else {
 7469     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7470   }
 7471   // Store back in big endian from little endian
 7472   rorq(sum, 0x20);
 7473   movq(Address(out, offset, Address::times_4,  8), sum);
 7474 
 7475   movq(op1, Address(in, len, Address::times_4,  0));
 7476   rorq(op1, 32);
 7477   movq(sum, Address(out, offset, Address::times_4,  0));
 7478   rorq(sum, 32);
 7479   if (UseBMI2Instructions) {
 7480     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7481   }
 7482   else {
 7483     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7484   }
 7485   // Store back in big endian from little endian
 7486   rorq(sum, 0x20);
 7487   movq(Address(out, offset, Address::times_4,  0), sum);
 7488 
 7489   jmp(L_first_loop);
 7490   bind(L_first_loop_exit);
 7491 }
 7492 
 7493 /**
 7494  * Code for BigInteger::mulAdd() intrinsic
 7495  *
 7496  * rdi: out
 7497  * rsi: in
 7498  * r11: offs (out.length - offset)
 7499  * rcx: len
 7500  * r8:  k
 7501  * r12: tmp1
 7502  * r13: tmp2
 7503  * r14: tmp3
 7504  * r15: tmp4
 7505  * rbx: tmp5
 7506  * Multiply the in[] by word k and add to out[], return the carry in rax
 7507  */
 7508 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 7509    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 7510    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7511 
 7512   Label L_carry, L_last_in, L_done;
 7513 
 7514 // carry = 0;
 7515 // for (int j=len-1; j >= 0; j--) {
 7516 //    long product = (in[j] & LONG_MASK) * kLong +
 7517 //                   (out[offs] & LONG_MASK) + carry;
 7518 //    out[offs--] = (int)product;
 7519 //    carry = product >>> 32;
 7520 // }
 7521 //
 7522   push(tmp1);
 7523   push(tmp2);
 7524   push(tmp3);
 7525   push(tmp4);
 7526   push(tmp5);
 7527 
 7528   Register op2 = tmp2;
 7529   const Register sum = tmp3;
 7530   const Register op1 = tmp4;
 7531   const Register carry =  tmp5;
 7532 
 7533   if (UseBMI2Instructions) {
 7534     op2 = rdxReg;
 7535     movl(op2, k);
 7536   }
 7537   else {
 7538     movl(op2, k);
 7539   }
 7540 
 7541   xorq(carry, carry);
 7542 
 7543   //First loop
 7544 
 7545   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 7546   //The carry is in tmp5
 7547   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7548 
 7549   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 7550   decrementl(len);
 7551   jccb(Assembler::negative, L_carry);
 7552   decrementl(len);
 7553   jccb(Assembler::negative, L_last_in);
 7554 
 7555   movq(op1, Address(in, len, Address::times_4,  0));
 7556   rorq(op1, 32);
 7557 
 7558   subl(offs, 2);
 7559   movq(sum, Address(out, offs, Address::times_4,  0));
 7560   rorq(sum, 32);
 7561 
 7562   if (UseBMI2Instructions) {
 7563     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7564   }
 7565   else {
 7566     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7567   }
 7568 
 7569   // Store back in big endian from little endian
 7570   rorq(sum, 0x20);
 7571   movq(Address(out, offs, Address::times_4,  0), sum);
 7572 
 7573   testl(len, len);
 7574   jccb(Assembler::zero, L_carry);
 7575 
 7576   //Multiply the last in[] entry, if any
 7577   bind(L_last_in);
 7578   movl(op1, Address(in, 0));
 7579   movl(sum, Address(out, offs, Address::times_4,  -4));
 7580 
 7581   movl(raxReg, k);
 7582   mull(op1); //tmp4 * eax -> edx:eax
 7583   addl(sum, carry);
 7584   adcl(rdxReg, 0);
 7585   addl(sum, raxReg);
 7586   adcl(rdxReg, 0);
 7587   movl(carry, rdxReg);
 7588 
 7589   movl(Address(out, offs, Address::times_4,  -4), sum);
 7590 
 7591   bind(L_carry);
 7592   //return tmp5/carry as carry in rax
 7593   movl(rax, carry);
 7594 
 7595   bind(L_done);
 7596   pop(tmp5);
 7597   pop(tmp4);
 7598   pop(tmp3);
 7599   pop(tmp2);
 7600   pop(tmp1);
 7601 }
 7602 
 7603 /**
 7604  * Emits code to update CRC-32 with a byte value according to constants in table
 7605  *
 7606  * @param [in,out]crc   Register containing the crc.
 7607  * @param [in]val       Register containing the byte to fold into the CRC.
 7608  * @param [in]table     Register containing the table of crc constants.
 7609  *
 7610  * uint32_t crc;
 7611  * val = crc_table[(val ^ crc) & 0xFF];
 7612  * crc = val ^ (crc >> 8);
 7613  *
 7614  */
 7615 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 7616   xorl(val, crc);
 7617   andl(val, 0xFF);
 7618   shrl(crc, 8); // unsigned shift
 7619   xorl(crc, Address(table, val, Address::times_4, 0));
 7620 }
 7621 
 7622 /**
 7623  * Fold 128-bit data chunk
 7624  */
 7625 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 7626   if (UseAVX > 0) {
 7627     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 7628     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 7629     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 7630     pxor(xcrc, xtmp);
 7631   } else {
 7632     movdqa(xtmp, xcrc);
 7633     pclmulhdq(xtmp, xK);   // [123:64]
 7634     pclmulldq(xcrc, xK);   // [63:0]
 7635     pxor(xcrc, xtmp);
 7636     movdqu(xtmp, Address(buf, offset));
 7637     pxor(xcrc, xtmp);
 7638   }
 7639 }
 7640 
 7641 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 7642   if (UseAVX > 0) {
 7643     vpclmulhdq(xtmp, xK, xcrc);
 7644     vpclmulldq(xcrc, xK, xcrc);
 7645     pxor(xcrc, xbuf);
 7646     pxor(xcrc, xtmp);
 7647   } else {
 7648     movdqa(xtmp, xcrc);
 7649     pclmulhdq(xtmp, xK);
 7650     pclmulldq(xcrc, xK);
 7651     pxor(xcrc, xbuf);
 7652     pxor(xcrc, xtmp);
 7653   }
 7654 }
 7655 
 7656 /**
 7657  * 8-bit folds to compute 32-bit CRC
 7658  *
 7659  * uint64_t xcrc;
 7660  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 7661  */
 7662 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 7663   movdl(tmp, xcrc);
 7664   andl(tmp, 0xFF);
 7665   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 7666   psrldq(xcrc, 1); // unsigned shift one byte
 7667   pxor(xcrc, xtmp);
 7668 }
 7669 
 7670 /**
 7671  * uint32_t crc;
 7672  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 7673  */
 7674 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 7675   movl(tmp, crc);
 7676   andl(tmp, 0xFF);
 7677   shrl(crc, 8);
 7678   xorl(crc, Address(table, tmp, Address::times_4, 0));
 7679 }
 7680 
 7681 /**
 7682  * @param crc   register containing existing CRC (32-bit)
 7683  * @param buf   register pointing to input byte buffer (byte*)
 7684  * @param len   register containing number of bytes
 7685  * @param table register that will contain address of CRC table
 7686  * @param tmp   scratch register
 7687  */
 7688 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 7689   assert_different_registers(crc, buf, len, table, tmp, rax);
 7690 
 7691   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 7692   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 7693 
 7694   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 7695   // context for the registers used, where all instructions below are using 128-bit mode
 7696   // On EVEX without VL and BW, these instructions will all be AVX.
 7697   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 7698   notl(crc); // ~crc
 7699   cmpl(len, 16);
 7700   jcc(Assembler::less, L_tail);
 7701 
 7702   // Align buffer to 16 bytes
 7703   movl(tmp, buf);
 7704   andl(tmp, 0xF);
 7705   jccb(Assembler::zero, L_aligned);
 7706   subl(tmp,  16);
 7707   addl(len, tmp);
 7708 
 7709   align(4);
 7710   BIND(L_align_loop);
 7711   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7712   update_byte_crc32(crc, rax, table);
 7713   increment(buf);
 7714   incrementl(tmp);
 7715   jccb(Assembler::less, L_align_loop);
 7716 
 7717   BIND(L_aligned);
 7718   movl(tmp, len); // save
 7719   shrl(len, 4);
 7720   jcc(Assembler::zero, L_tail_restore);
 7721 
 7722   // Fold crc into first bytes of vector
 7723   movdqa(xmm1, Address(buf, 0));
 7724   movdl(rax, xmm1);
 7725   xorl(crc, rax);
 7726   if (VM_Version::supports_sse4_1()) {
 7727     pinsrd(xmm1, crc, 0);
 7728   } else {
 7729     pinsrw(xmm1, crc, 0);
 7730     shrl(crc, 16);
 7731     pinsrw(xmm1, crc, 1);
 7732   }
 7733   addptr(buf, 16);
 7734   subl(len, 4); // len > 0
 7735   jcc(Assembler::less, L_fold_tail);
 7736 
 7737   movdqa(xmm2, Address(buf,  0));
 7738   movdqa(xmm3, Address(buf, 16));
 7739   movdqa(xmm4, Address(buf, 32));
 7740   addptr(buf, 48);
 7741   subl(len, 3);
 7742   jcc(Assembler::lessEqual, L_fold_512b);
 7743 
 7744   // Fold total 512 bits of polynomial on each iteration,
 7745   // 128 bits per each of 4 parallel streams.
 7746   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 7747 
 7748   align32();
 7749   BIND(L_fold_512b_loop);
 7750   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7751   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 7752   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 7753   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 7754   addptr(buf, 64);
 7755   subl(len, 4);
 7756   jcc(Assembler::greater, L_fold_512b_loop);
 7757 
 7758   // Fold 512 bits to 128 bits.
 7759   BIND(L_fold_512b);
 7760   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7761   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 7762   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 7763   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 7764 
 7765   // Fold the rest of 128 bits data chunks
 7766   BIND(L_fold_tail);
 7767   addl(len, 3);
 7768   jccb(Assembler::lessEqual, L_fold_128b);
 7769   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7770 
 7771   BIND(L_fold_tail_loop);
 7772   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7773   addptr(buf, 16);
 7774   decrementl(len);
 7775   jccb(Assembler::greater, L_fold_tail_loop);
 7776 
 7777   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 7778   BIND(L_fold_128b);
 7779   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 7780   if (UseAVX > 0) {
 7781     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 7782     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 7783     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 7784   } else {
 7785     movdqa(xmm2, xmm0);
 7786     pclmulqdq(xmm2, xmm1, 0x1);
 7787     movdqa(xmm3, xmm0);
 7788     pand(xmm3, xmm2);
 7789     pclmulqdq(xmm0, xmm3, 0x1);
 7790   }
 7791   psrldq(xmm1, 8);
 7792   psrldq(xmm2, 4);
 7793   pxor(xmm0, xmm1);
 7794   pxor(xmm0, xmm2);
 7795 
 7796   // 8 8-bit folds to compute 32-bit CRC.
 7797   for (int j = 0; j < 4; j++) {
 7798     fold_8bit_crc32(xmm0, table, xmm1, rax);
 7799   }
 7800   movdl(crc, xmm0); // mov 32 bits to general register
 7801   for (int j = 0; j < 4; j++) {
 7802     fold_8bit_crc32(crc, table, rax);
 7803   }
 7804 
 7805   BIND(L_tail_restore);
 7806   movl(len, tmp); // restore
 7807   BIND(L_tail);
 7808   andl(len, 0xf);
 7809   jccb(Assembler::zero, L_exit);
 7810 
 7811   // Fold the rest of bytes
 7812   align(4);
 7813   BIND(L_tail_loop);
 7814   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7815   update_byte_crc32(crc, rax, table);
 7816   increment(buf);
 7817   decrementl(len);
 7818   jccb(Assembler::greater, L_tail_loop);
 7819 
 7820   BIND(L_exit);
 7821   notl(crc); // ~c
 7822 }
 7823 
 7824 // Helper function for AVX 512 CRC32
 7825 // Fold 512-bit data chunks
 7826 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 7827                                              Register pos, int offset) {
 7828   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 7829   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 7830   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 7831   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 7832   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 7833 }
 7834 
 7835 // Helper function for AVX 512 CRC32
 7836 // Compute CRC32 for < 256B buffers
 7837 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 7838                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 7839                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 7840 
 7841   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 7842   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 7843   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 7844 
 7845   // check if there is enough buffer to be able to fold 16B at a time
 7846   cmpl(len, 32);
 7847   jcc(Assembler::less, L_less_than_32);
 7848 
 7849   // if there is, load the constants
 7850   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 7851   movdl(xmm0, crc);                        // get the initial crc value
 7852   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7853   pxor(xmm7, xmm0);
 7854 
 7855   // update the buffer pointer
 7856   addl(pos, 16);
 7857   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 7858   subl(len, 32);
 7859   jmp(L_16B_reduction_loop);
 7860 
 7861   bind(L_less_than_32);
 7862   //mov initial crc to the return value. this is necessary for zero - length buffers.
 7863   movl(rax, crc);
 7864   testl(len, len);
 7865   jcc(Assembler::equal, L_cleanup);
 7866 
 7867   movdl(xmm0, crc);                        //get the initial crc value
 7868 
 7869   cmpl(len, 16);
 7870   jcc(Assembler::equal, L_exact_16_left);
 7871   jcc(Assembler::less, L_less_than_16_left);
 7872 
 7873   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7874   pxor(xmm7, xmm0);                       //xor the initial crc value
 7875   addl(pos, 16);
 7876   subl(len, 16);
 7877   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 7878   jmp(L_get_last_two_xmms);
 7879 
 7880   bind(L_less_than_16_left);
 7881   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 7882   pxor(xmm1, xmm1);
 7883   movptr(tmp1, rsp);
 7884   movdqu(Address(tmp1, 0 * 16), xmm1);
 7885 
 7886   cmpl(len, 4);
 7887   jcc(Assembler::less, L_only_less_than_4);
 7888 
 7889   //backup the counter value
 7890   movl(tmp2, len);
 7891   cmpl(len, 8);
 7892   jcc(Assembler::less, L_less_than_8_left);
 7893 
 7894   //load 8 Bytes
 7895   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 7896   movq(Address(tmp1, 0 * 16), rax);
 7897   addptr(tmp1, 8);
 7898   subl(len, 8);
 7899   addl(pos, 8);
 7900 
 7901   bind(L_less_than_8_left);
 7902   cmpl(len, 4);
 7903   jcc(Assembler::less, L_less_than_4_left);
 7904 
 7905   //load 4 Bytes
 7906   movl(rax, Address(buf, pos, Address::times_1, 0));
 7907   movl(Address(tmp1, 0 * 16), rax);
 7908   addptr(tmp1, 4);
 7909   subl(len, 4);
 7910   addl(pos, 4);
 7911 
 7912   bind(L_less_than_4_left);
 7913   cmpl(len, 2);
 7914   jcc(Assembler::less, L_less_than_2_left);
 7915 
 7916   // load 2 Bytes
 7917   movw(rax, Address(buf, pos, Address::times_1, 0));
 7918   movl(Address(tmp1, 0 * 16), rax);
 7919   addptr(tmp1, 2);
 7920   subl(len, 2);
 7921   addl(pos, 2);
 7922 
 7923   bind(L_less_than_2_left);
 7924   cmpl(len, 1);
 7925   jcc(Assembler::less, L_zero_left);
 7926 
 7927   // load 1 Byte
 7928   movb(rax, Address(buf, pos, Address::times_1, 0));
 7929   movb(Address(tmp1, 0 * 16), rax);
 7930 
 7931   bind(L_zero_left);
 7932   movdqu(xmm7, Address(rsp, 0));
 7933   pxor(xmm7, xmm0);                       //xor the initial crc value
 7934 
 7935   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 7936   movdqu(xmm0, Address(rax, tmp2));
 7937   pshufb(xmm7, xmm0);
 7938   jmp(L_128_done);
 7939 
 7940   bind(L_exact_16_left);
 7941   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 7942   pxor(xmm7, xmm0);                       //xor the initial crc value
 7943   jmp(L_128_done);
 7944 
 7945   bind(L_only_less_than_4);
 7946   cmpl(len, 3);
 7947   jcc(Assembler::less, L_only_less_than_3);
 7948 
 7949   // load 3 Bytes
 7950   movb(rax, Address(buf, pos, Address::times_1, 0));
 7951   movb(Address(tmp1, 0), rax);
 7952 
 7953   movb(rax, Address(buf, pos, Address::times_1, 1));
 7954   movb(Address(tmp1, 1), rax);
 7955 
 7956   movb(rax, Address(buf, pos, Address::times_1, 2));
 7957   movb(Address(tmp1, 2), rax);
 7958 
 7959   movdqu(xmm7, Address(rsp, 0));
 7960   pxor(xmm7, xmm0);                     //xor the initial crc value
 7961 
 7962   pslldq(xmm7, 0x5);
 7963   jmp(L_barrett);
 7964   bind(L_only_less_than_3);
 7965   cmpl(len, 2);
 7966   jcc(Assembler::less, L_only_less_than_2);
 7967 
 7968   // load 2 Bytes
 7969   movb(rax, Address(buf, pos, Address::times_1, 0));
 7970   movb(Address(tmp1, 0), rax);
 7971 
 7972   movb(rax, Address(buf, pos, Address::times_1, 1));
 7973   movb(Address(tmp1, 1), rax);
 7974 
 7975   movdqu(xmm7, Address(rsp, 0));
 7976   pxor(xmm7, xmm0);                     //xor the initial crc value
 7977 
 7978   pslldq(xmm7, 0x6);
 7979   jmp(L_barrett);
 7980 
 7981   bind(L_only_less_than_2);
 7982   //load 1 Byte
 7983   movb(rax, Address(buf, pos, Address::times_1, 0));
 7984   movb(Address(tmp1, 0), rax);
 7985 
 7986   movdqu(xmm7, Address(rsp, 0));
 7987   pxor(xmm7, xmm0);                     //xor the initial crc value
 7988 
 7989   pslldq(xmm7, 0x7);
 7990 }
 7991 
 7992 /**
 7993 * Compute CRC32 using AVX512 instructions
 7994 * param crc   register containing existing CRC (32-bit)
 7995 * param buf   register pointing to input byte buffer (byte*)
 7996 * param len   register containing number of bytes
 7997 * param table address of crc or crc32c table
 7998 * param tmp1  scratch register
 7999 * param tmp2  scratch register
 8000 * return rax  result register
 8001 *
 8002 * This routine is identical for crc32c with the exception of the precomputed constant
 8003 * table which will be passed as the table argument.  The calculation steps are
 8004 * the same for both variants.
 8005 */
 8006 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8007   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8008 
 8009   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8010   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8011   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8012   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8013   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8014 
 8015   const Register pos = r12;
 8016   push(r12);
 8017   subptr(rsp, 16 * 2 + 8);
 8018 
 8019   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8020   // context for the registers used, where all instructions below are using 128-bit mode
 8021   // On EVEX without VL and BW, these instructions will all be AVX.
 8022   movl(pos, 0);
 8023 
 8024   // check if smaller than 256B
 8025   cmpl(len, 256);
 8026   jcc(Assembler::less, L_less_than_256);
 8027 
 8028   // load the initial crc value
 8029   movdl(xmm10, crc);
 8030 
 8031   // receive the initial 64B data, xor the initial crc value
 8032   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8033   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8034   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8035   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8036 
 8037   subl(len, 256);
 8038   cmpl(len, 256);
 8039   jcc(Assembler::less, L_fold_128_B_loop);
 8040 
 8041   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8042   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8043   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8044   subl(len, 256);
 8045 
 8046   bind(L_fold_256_B_loop);
 8047   addl(pos, 256);
 8048   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8049   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8050   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8051   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8052 
 8053   subl(len, 256);
 8054   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8055 
 8056   // Fold 256 into 128
 8057   addl(pos, 256);
 8058   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8059   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8060   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8061 
 8062   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8063   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8064   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8065 
 8066   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8067   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8068 
 8069   addl(len, 128);
 8070   jmp(L_fold_128_B_register);
 8071 
 8072   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8073   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8074 
 8075   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8076   bind(L_fold_128_B_loop);
 8077   addl(pos, 128);
 8078   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8079   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8080 
 8081   subl(len, 128);
 8082   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8083 
 8084   addl(pos, 128);
 8085 
 8086   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8087   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8088   bind(L_fold_128_B_register);
 8089   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8090   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8091   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8092   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8093   // save last that has no multiplicand
 8094   vextracti64x2(xmm7, xmm4, 3);
 8095 
 8096   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8097   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8098   // Needed later in reduction loop
 8099   movdqu(xmm10, Address(table, 1 * 16));
 8100   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8101   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8102 
 8103   // Swap 1,0,3,2 - 01 00 11 10
 8104   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8105   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8106   vextracti128(xmm5, xmm8, 1);
 8107   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8108 
 8109   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8110   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8111   addl(len, 128 - 16);
 8112   jcc(Assembler::less, L_final_reduction_for_128);
 8113 
 8114   bind(L_16B_reduction_loop);
 8115   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8116   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8117   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8118   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8119   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8120   addl(pos, 16);
 8121   subl(len, 16);
 8122   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8123 
 8124   bind(L_final_reduction_for_128);
 8125   addl(len, 16);
 8126   jcc(Assembler::equal, L_128_done);
 8127 
 8128   bind(L_get_last_two_xmms);
 8129   movdqu(xmm2, xmm7);
 8130   addl(pos, len);
 8131   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8132   subl(pos, len);
 8133 
 8134   // get rid of the extra data that was loaded before
 8135   // load the shift constant
 8136   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8137   movdqu(xmm0, Address(rax, len));
 8138   addl(rax, len);
 8139 
 8140   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8141   //Change mask to 512
 8142   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8143   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8144 
 8145   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8146   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8147   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8148   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8149   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8150 
 8151   bind(L_128_done);
 8152   // compute crc of a 128-bit value
 8153   movdqu(xmm10, Address(table, 3 * 16));
 8154   movdqu(xmm0, xmm7);
 8155 
 8156   // 64b fold
 8157   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8158   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8159   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8160 
 8161   // 32b fold
 8162   movdqu(xmm0, xmm7);
 8163   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8164   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8165   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8166   jmp(L_barrett);
 8167 
 8168   bind(L_less_than_256);
 8169   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);
 8170 
 8171   //barrett reduction
 8172   bind(L_barrett);
 8173   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8174   movdqu(xmm1, xmm7);
 8175   movdqu(xmm2, xmm7);
 8176   movdqu(xmm10, Address(table, 4 * 16));
 8177 
 8178   pclmulqdq(xmm7, xmm10, 0x0);
 8179   pxor(xmm7, xmm2);
 8180   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8181   movdqu(xmm2, xmm7);
 8182   pclmulqdq(xmm7, xmm10, 0x10);
 8183   pxor(xmm7, xmm2);
 8184   pxor(xmm7, xmm1);
 8185   pextrd(crc, xmm7, 2);
 8186 
 8187   bind(L_cleanup);
 8188   addptr(rsp, 16 * 2 + 8);
 8189   pop(r12);
 8190 }
 8191 
 8192 // S. Gueron / Information Processing Letters 112 (2012) 184
 8193 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8194 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8195 // Output: the 64-bit carry-less product of B * CONST
 8196 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8197                                      Register tmp1, Register tmp2, Register tmp3) {
 8198   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8199   if (n > 0) {
 8200     addq(tmp3, n * 256 * 8);
 8201   }
 8202   //    Q1 = TABLEExt[n][B & 0xFF];
 8203   movl(tmp1, in);
 8204   andl(tmp1, 0x000000FF);
 8205   shll(tmp1, 3);
 8206   addq(tmp1, tmp3);
 8207   movq(tmp1, Address(tmp1, 0));
 8208 
 8209   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8210   movl(tmp2, in);
 8211   shrl(tmp2, 8);
 8212   andl(tmp2, 0x000000FF);
 8213   shll(tmp2, 3);
 8214   addq(tmp2, tmp3);
 8215   movq(tmp2, Address(tmp2, 0));
 8216 
 8217   shlq(tmp2, 8);
 8218   xorq(tmp1, tmp2);
 8219 
 8220   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8221   movl(tmp2, in);
 8222   shrl(tmp2, 16);
 8223   andl(tmp2, 0x000000FF);
 8224   shll(tmp2, 3);
 8225   addq(tmp2, tmp3);
 8226   movq(tmp2, Address(tmp2, 0));
 8227 
 8228   shlq(tmp2, 16);
 8229   xorq(tmp1, tmp2);
 8230 
 8231   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8232   shrl(in, 24);
 8233   andl(in, 0x000000FF);
 8234   shll(in, 3);
 8235   addq(in, tmp3);
 8236   movq(in, Address(in, 0));
 8237 
 8238   shlq(in, 24);
 8239   xorq(in, tmp1);
 8240   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8241 }
 8242 
 8243 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8244                                       Register in_out,
 8245                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8246                                       XMMRegister w_xtmp2,
 8247                                       Register tmp1,
 8248                                       Register n_tmp2, Register n_tmp3) {
 8249   if (is_pclmulqdq_supported) {
 8250     movdl(w_xtmp1, in_out); // modified blindly
 8251 
 8252     movl(tmp1, const_or_pre_comp_const_index);
 8253     movdl(w_xtmp2, tmp1);
 8254     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8255 
 8256     movdq(in_out, w_xtmp1);
 8257   } else {
 8258     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 8259   }
 8260 }
 8261 
 8262 // Recombination Alternative 2: No bit-reflections
 8263 // T1 = (CRC_A * U1) << 1
 8264 // T2 = (CRC_B * U2) << 1
 8265 // C1 = T1 >> 32
 8266 // C2 = T2 >> 32
 8267 // T1 = T1 & 0xFFFFFFFF
 8268 // T2 = T2 & 0xFFFFFFFF
 8269 // T1 = CRC32(0, T1)
 8270 // T2 = CRC32(0, T2)
 8271 // C1 = C1 ^ T1
 8272 // C2 = C2 ^ T2
 8273 // CRC = C1 ^ C2 ^ CRC_C
 8274 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,
 8275                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8276                                      Register tmp1, Register tmp2,
 8277                                      Register n_tmp3) {
 8278   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8279   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8280   shlq(in_out, 1);
 8281   movl(tmp1, in_out);
 8282   shrq(in_out, 32);
 8283   xorl(tmp2, tmp2);
 8284   crc32(tmp2, tmp1, 4);
 8285   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 8286   shlq(in1, 1);
 8287   movl(tmp1, in1);
 8288   shrq(in1, 32);
 8289   xorl(tmp2, tmp2);
 8290   crc32(tmp2, tmp1, 4);
 8291   xorl(in1, tmp2);
 8292   xorl(in_out, in1);
 8293   xorl(in_out, in2);
 8294 }
 8295 
 8296 // Set N to predefined value
 8297 // Subtract from a length of a buffer
 8298 // execute in a loop:
 8299 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 8300 // for i = 1 to N do
 8301 //  CRC_A = CRC32(CRC_A, A[i])
 8302 //  CRC_B = CRC32(CRC_B, B[i])
 8303 //  CRC_C = CRC32(CRC_C, C[i])
 8304 // end for
 8305 // Recombine
 8306 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,
 8307                                        Register in_out1, Register in_out2, Register in_out3,
 8308                                        Register tmp1, Register tmp2, Register tmp3,
 8309                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8310                                        Register tmp4, Register tmp5,
 8311                                        Register n_tmp6) {
 8312   Label L_processPartitions;
 8313   Label L_processPartition;
 8314   Label L_exit;
 8315 
 8316   bind(L_processPartitions);
 8317   cmpl(in_out1, 3 * size);
 8318   jcc(Assembler::less, L_exit);
 8319     xorl(tmp1, tmp1);
 8320     xorl(tmp2, tmp2);
 8321     movq(tmp3, in_out2);
 8322     addq(tmp3, size);
 8323 
 8324     bind(L_processPartition);
 8325       crc32(in_out3, Address(in_out2, 0), 8);
 8326       crc32(tmp1, Address(in_out2, size), 8);
 8327       crc32(tmp2, Address(in_out2, size * 2), 8);
 8328       addq(in_out2, 8);
 8329       cmpq(in_out2, tmp3);
 8330       jcc(Assembler::less, L_processPartition);
 8331     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 8332             w_xtmp1, w_xtmp2, w_xtmp3,
 8333             tmp4, tmp5,
 8334             n_tmp6);
 8335     addq(in_out2, 2 * size);
 8336     subl(in_out1, 3 * size);
 8337     jmp(L_processPartitions);
 8338 
 8339   bind(L_exit);
 8340 }
 8341 
 8342 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 8343 // Input: A buffer I of L bytes.
 8344 // Output: the CRC32C value of the buffer.
 8345 // Notations:
 8346 // Write L = 24N + r, with N = floor (L/24).
 8347 // r = L mod 24 (0 <= r < 24).
 8348 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 8349 // N quadwords, and R consists of r bytes.
 8350 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 8351 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 8352 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 8353 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 8354 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 8355                                           Register tmp1, Register tmp2, Register tmp3,
 8356                                           Register tmp4, Register tmp5, Register tmp6,
 8357                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8358                                           bool is_pclmulqdq_supported) {
 8359   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 8360   Label L_wordByWord;
 8361   Label L_byteByByteProlog;
 8362   Label L_byteByByte;
 8363   Label L_exit;
 8364 
 8365   if (is_pclmulqdq_supported ) {
 8366     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 8367     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 8368 
 8369     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 8370     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 8371 
 8372     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 8373     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 8374     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 8375   } else {
 8376     const_or_pre_comp_const_index[0] = 1;
 8377     const_or_pre_comp_const_index[1] = 0;
 8378 
 8379     const_or_pre_comp_const_index[2] = 3;
 8380     const_or_pre_comp_const_index[3] = 2;
 8381 
 8382     const_or_pre_comp_const_index[4] = 5;
 8383     const_or_pre_comp_const_index[5] = 4;
 8384    }
 8385   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 8386                     in2, in1, in_out,
 8387                     tmp1, tmp2, tmp3,
 8388                     w_xtmp1, w_xtmp2, w_xtmp3,
 8389                     tmp4, tmp5,
 8390                     tmp6);
 8391   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 8392                     in2, in1, in_out,
 8393                     tmp1, tmp2, tmp3,
 8394                     w_xtmp1, w_xtmp2, w_xtmp3,
 8395                     tmp4, tmp5,
 8396                     tmp6);
 8397   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 8398                     in2, in1, in_out,
 8399                     tmp1, tmp2, tmp3,
 8400                     w_xtmp1, w_xtmp2, w_xtmp3,
 8401                     tmp4, tmp5,
 8402                     tmp6);
 8403   movl(tmp1, in2);
 8404   andl(tmp1, 0x00000007);
 8405   negl(tmp1);
 8406   addl(tmp1, in2);
 8407   addq(tmp1, in1);
 8408 
 8409   cmpq(in1, tmp1);
 8410   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 8411   align(16);
 8412   BIND(L_wordByWord);
 8413     crc32(in_out, Address(in1, 0), 8);
 8414     addq(in1, 8);
 8415     cmpq(in1, tmp1);
 8416     jcc(Assembler::less, L_wordByWord);
 8417 
 8418   BIND(L_byteByByteProlog);
 8419   andl(in2, 0x00000007);
 8420   movl(tmp2, 1);
 8421 
 8422   cmpl(tmp2, in2);
 8423   jccb(Assembler::greater, L_exit);
 8424   BIND(L_byteByByte);
 8425     crc32(in_out, Address(in1, 0), 1);
 8426     incq(in1);
 8427     incl(tmp2);
 8428     cmpl(tmp2, in2);
 8429     jcc(Assembler::lessEqual, L_byteByByte);
 8430 
 8431   BIND(L_exit);
 8432 }
 8433 #undef BIND
 8434 #undef BLOCK_COMMENT
 8435 
 8436 // Compress char[] array to byte[].
 8437 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 8438 // Return the array length if every element in array can be encoded,
 8439 // otherwise, the index of first non-latin1 (> 0xff) character.
 8440 //   @IntrinsicCandidate
 8441 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 8442 //     for (int i = 0; i < len; i++) {
 8443 //       char c = src[srcOff];
 8444 //       if (c > 0xff) {
 8445 //           return i;  // return index of non-latin1 char
 8446 //       }
 8447 //       dst[dstOff] = (byte)c;
 8448 //       srcOff++;
 8449 //       dstOff++;
 8450 //     }
 8451 //     return len;
 8452 //   }
 8453 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 8454   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 8455   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 8456   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 8457   Label copy_chars_loop, done, reset_sp, copy_tail;
 8458 
 8459   // rsi: src
 8460   // rdi: dst
 8461   // rdx: len
 8462   // rcx: tmp5
 8463   // rax: result
 8464 
 8465   // rsi holds start addr of source char[] to be compressed
 8466   // rdi holds start addr of destination byte[]
 8467   // rdx holds length
 8468 
 8469   assert(len != result, "");
 8470 
 8471   // save length for return
 8472   movl(result, len);
 8473 
 8474   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 8475     VM_Version::supports_avx512vlbw() &&
 8476     VM_Version::supports_bmi2()) {
 8477 
 8478     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 8479 
 8480     // alignment
 8481     Label post_alignment;
 8482 
 8483     // if length of the string is less than 32, handle it the old fashioned way
 8484     testl(len, -32);
 8485     jcc(Assembler::zero, below_threshold);
 8486 
 8487     // First check whether a character is compressible ( <= 0xFF).
 8488     // Create mask to test for Unicode chars inside zmm vector
 8489     movl(tmp5, 0x00FF);
 8490     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 8491 
 8492     testl(len, -64);
 8493     jccb(Assembler::zero, post_alignment);
 8494 
 8495     movl(tmp5, dst);
 8496     andl(tmp5, (32 - 1));
 8497     negl(tmp5);
 8498     andl(tmp5, (32 - 1));
 8499 
 8500     // bail out when there is nothing to be done
 8501     testl(tmp5, 0xFFFFFFFF);
 8502     jccb(Assembler::zero, post_alignment);
 8503 
 8504     // ~(~0 << len), where len is the # of remaining elements to process
 8505     movl(len, 0xFFFFFFFF);
 8506     shlxl(len, len, tmp5);
 8507     notl(len);
 8508     kmovdl(mask2, len);
 8509     movl(len, result);
 8510 
 8511     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8512     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8513     ktestd(mask1, mask2);
 8514     jcc(Assembler::carryClear, copy_tail);
 8515 
 8516     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8517 
 8518     addptr(src, tmp5);
 8519     addptr(src, tmp5);
 8520     addptr(dst, tmp5);
 8521     subl(len, tmp5);
 8522 
 8523     bind(post_alignment);
 8524     // end of alignment
 8525 
 8526     movl(tmp5, len);
 8527     andl(tmp5, (32 - 1));    // tail count (in chars)
 8528     andl(len, ~(32 - 1));    // vector count (in chars)
 8529     jccb(Assembler::zero, copy_loop_tail);
 8530 
 8531     lea(src, Address(src, len, Address::times_2));
 8532     lea(dst, Address(dst, len, Address::times_1));
 8533     negptr(len);
 8534 
 8535     bind(copy_32_loop);
 8536     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 8537     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 8538     kortestdl(mask1, mask1);
 8539     jccb(Assembler::carryClear, reset_for_copy_tail);
 8540 
 8541     // All elements in current processed chunk are valid candidates for
 8542     // compression. Write a truncated byte elements to the memory.
 8543     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 8544     addptr(len, 32);
 8545     jccb(Assembler::notZero, copy_32_loop);
 8546 
 8547     bind(copy_loop_tail);
 8548     // bail out when there is nothing to be done
 8549     testl(tmp5, 0xFFFFFFFF);
 8550     jcc(Assembler::zero, done);
 8551 
 8552     movl(len, tmp5);
 8553 
 8554     // ~(~0 << len), where len is the # of remaining elements to process
 8555     movl(tmp5, 0xFFFFFFFF);
 8556     shlxl(tmp5, tmp5, len);
 8557     notl(tmp5);
 8558 
 8559     kmovdl(mask2, tmp5);
 8560 
 8561     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8562     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8563     ktestd(mask1, mask2);
 8564     jcc(Assembler::carryClear, copy_tail);
 8565 
 8566     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8567     jmp(done);
 8568 
 8569     bind(reset_for_copy_tail);
 8570     lea(src, Address(src, tmp5, Address::times_2));
 8571     lea(dst, Address(dst, tmp5, Address::times_1));
 8572     subptr(len, tmp5);
 8573     jmp(copy_chars_loop);
 8574 
 8575     bind(below_threshold);
 8576   }
 8577 
 8578   if (UseSSE42Intrinsics) {
 8579     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 8580 
 8581     // vectored compression
 8582     testl(len, 0xfffffff8);
 8583     jcc(Assembler::zero, copy_tail);
 8584 
 8585     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 8586     movdl(tmp1Reg, tmp5);
 8587     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 8588 
 8589     andl(len, 0xfffffff0);
 8590     jccb(Assembler::zero, copy_16);
 8591 
 8592     // compress 16 chars per iter
 8593     pxor(tmp4Reg, tmp4Reg);
 8594 
 8595     lea(src, Address(src, len, Address::times_2));
 8596     lea(dst, Address(dst, len, Address::times_1));
 8597     negptr(len);
 8598 
 8599     bind(copy_32_loop);
 8600     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 8601     por(tmp4Reg, tmp2Reg);
 8602     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 8603     por(tmp4Reg, tmp3Reg);
 8604     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 8605     jccb(Assembler::notZero, reset_for_copy_tail);
 8606     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 8607     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 8608     addptr(len, 16);
 8609     jccb(Assembler::notZero, copy_32_loop);
 8610 
 8611     // compress next vector of 8 chars (if any)
 8612     bind(copy_16);
 8613     // len = 0
 8614     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 8615     jccb(Assembler::zero, copy_tail_sse);
 8616 
 8617     pxor(tmp3Reg, tmp3Reg);
 8618 
 8619     movdqu(tmp2Reg, Address(src, 0));
 8620     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 8621     jccb(Assembler::notZero, reset_for_copy_tail);
 8622     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 8623     movq(Address(dst, 0), tmp2Reg);
 8624     addptr(src, 16);
 8625     addptr(dst, 8);
 8626     jmpb(copy_tail_sse);
 8627 
 8628     bind(reset_for_copy_tail);
 8629     movl(tmp5, result);
 8630     andl(tmp5, 0x0000000f);
 8631     lea(src, Address(src, tmp5, Address::times_2));
 8632     lea(dst, Address(dst, tmp5, Address::times_1));
 8633     subptr(len, tmp5);
 8634     jmpb(copy_chars_loop);
 8635 
 8636     bind(copy_tail_sse);
 8637     movl(len, result);
 8638     andl(len, 0x00000007);    // tail count (in chars)
 8639   }
 8640   // compress 1 char per iter
 8641   bind(copy_tail);
 8642   testl(len, len);
 8643   jccb(Assembler::zero, done);
 8644   lea(src, Address(src, len, Address::times_2));
 8645   lea(dst, Address(dst, len, Address::times_1));
 8646   negptr(len);
 8647 
 8648   bind(copy_chars_loop);
 8649   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 8650   testl(tmp5, 0xff00);      // check if Unicode char
 8651   jccb(Assembler::notZero, reset_sp);
 8652   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 8653   increment(len);
 8654   jccb(Assembler::notZero, copy_chars_loop);
 8655 
 8656   // add len then return (len will be zero if compress succeeded, otherwise negative)
 8657   bind(reset_sp);
 8658   addl(result, len);
 8659 
 8660   bind(done);
 8661 }
 8662 
 8663 // Inflate byte[] array to char[].
 8664 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 8665 //   @IntrinsicCandidate
 8666 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 8667 //     for (int i = 0; i < len; i++) {
 8668 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 8669 //     }
 8670 //   }
 8671 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 8672   XMMRegister tmp1, Register tmp2, KRegister mask) {
 8673   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 8674   // rsi: src
 8675   // rdi: dst
 8676   // rdx: len
 8677   // rcx: tmp2
 8678 
 8679   // rsi holds start addr of source byte[] to be inflated
 8680   // rdi holds start addr of destination char[]
 8681   // rdx holds length
 8682   assert_different_registers(src, dst, len, tmp2);
 8683   movl(tmp2, len);
 8684   if ((UseAVX > 2) && // AVX512
 8685     VM_Version::supports_avx512vlbw() &&
 8686     VM_Version::supports_bmi2()) {
 8687 
 8688     Label copy_32_loop, copy_tail;
 8689     Register tmp3_aliased = len;
 8690 
 8691     // if length of the string is less than 16, handle it in an old fashioned way
 8692     testl(len, -16);
 8693     jcc(Assembler::zero, below_threshold);
 8694 
 8695     testl(len, -1 * AVX3Threshold);
 8696     jcc(Assembler::zero, avx3_threshold);
 8697 
 8698     // In order to use only one arithmetic operation for the main loop we use
 8699     // this pre-calculation
 8700     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 8701     andl(len, -32);     // vector count
 8702     jccb(Assembler::zero, copy_tail);
 8703 
 8704     lea(src, Address(src, len, Address::times_1));
 8705     lea(dst, Address(dst, len, Address::times_2));
 8706     negptr(len);
 8707 
 8708 
 8709     // inflate 32 chars per iter
 8710     bind(copy_32_loop);
 8711     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 8712     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 8713     addptr(len, 32);
 8714     jcc(Assembler::notZero, copy_32_loop);
 8715 
 8716     bind(copy_tail);
 8717     // bail out when there is nothing to be done
 8718     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 8719     jcc(Assembler::zero, done);
 8720 
 8721     // ~(~0 << length), where length is the # of remaining elements to process
 8722     movl(tmp3_aliased, -1);
 8723     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 8724     notl(tmp3_aliased);
 8725     kmovdl(mask, tmp3_aliased);
 8726     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 8727     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 8728 
 8729     jmp(done);
 8730     bind(avx3_threshold);
 8731   }
 8732   if (UseSSE42Intrinsics) {
 8733     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 8734 
 8735     if (UseAVX > 1) {
 8736       andl(tmp2, (16 - 1));
 8737       andl(len, -16);
 8738       jccb(Assembler::zero, copy_new_tail);
 8739     } else {
 8740       andl(tmp2, 0x00000007);   // tail count (in chars)
 8741       andl(len, 0xfffffff8);    // vector count (in chars)
 8742       jccb(Assembler::zero, copy_tail);
 8743     }
 8744 
 8745     // vectored inflation
 8746     lea(src, Address(src, len, Address::times_1));
 8747     lea(dst, Address(dst, len, Address::times_2));
 8748     negptr(len);
 8749 
 8750     if (UseAVX > 1) {
 8751       bind(copy_16_loop);
 8752       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 8753       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 8754       addptr(len, 16);
 8755       jcc(Assembler::notZero, copy_16_loop);
 8756 
 8757       bind(below_threshold);
 8758       bind(copy_new_tail);
 8759       movl(len, tmp2);
 8760       andl(tmp2, 0x00000007);
 8761       andl(len, 0xFFFFFFF8);
 8762       jccb(Assembler::zero, copy_tail);
 8763 
 8764       pmovzxbw(tmp1, Address(src, 0));
 8765       movdqu(Address(dst, 0), tmp1);
 8766       addptr(src, 8);
 8767       addptr(dst, 2 * 8);
 8768 
 8769       jmp(copy_tail, true);
 8770     }
 8771 
 8772     // inflate 8 chars per iter
 8773     bind(copy_8_loop);
 8774     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 8775     movdqu(Address(dst, len, Address::times_2), tmp1);
 8776     addptr(len, 8);
 8777     jcc(Assembler::notZero, copy_8_loop);
 8778 
 8779     bind(copy_tail);
 8780     movl(len, tmp2);
 8781 
 8782     cmpl(len, 4);
 8783     jccb(Assembler::less, copy_bytes);
 8784 
 8785     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 8786     pmovzxbw(tmp1, tmp1);
 8787     movq(Address(dst, 0), tmp1);
 8788     subptr(len, 4);
 8789     addptr(src, 4);
 8790     addptr(dst, 8);
 8791 
 8792     bind(copy_bytes);
 8793   } else {
 8794     bind(below_threshold);
 8795   }
 8796 
 8797   testl(len, len);
 8798   jccb(Assembler::zero, done);
 8799   lea(src, Address(src, len, Address::times_1));
 8800   lea(dst, Address(dst, len, Address::times_2));
 8801   negptr(len);
 8802 
 8803   // inflate 1 char per iter
 8804   bind(copy_chars_loop);
 8805   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 8806   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 8807   increment(len);
 8808   jcc(Assembler::notZero, copy_chars_loop);
 8809 
 8810   bind(done);
 8811 }
 8812 
 8813 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 8814   switch(type) {
 8815     case T_BYTE:
 8816     case T_BOOLEAN:
 8817       evmovdqub(dst, kmask, src, merge, vector_len);
 8818       break;
 8819     case T_CHAR:
 8820     case T_SHORT:
 8821       evmovdquw(dst, kmask, src, merge, vector_len);
 8822       break;
 8823     case T_INT:
 8824     case T_FLOAT:
 8825       evmovdqul(dst, kmask, src, merge, vector_len);
 8826       break;
 8827     case T_LONG:
 8828     case T_DOUBLE:
 8829       evmovdquq(dst, kmask, src, merge, vector_len);
 8830       break;
 8831     default:
 8832       fatal("Unexpected type argument %s", type2name(type));
 8833       break;
 8834   }
 8835 }
 8836 
 8837 
 8838 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 8839   switch(type) {
 8840     case T_BYTE:
 8841     case T_BOOLEAN:
 8842       evmovdqub(dst, kmask, src, merge, vector_len);
 8843       break;
 8844     case T_CHAR:
 8845     case T_SHORT:
 8846       evmovdquw(dst, kmask, src, merge, vector_len);
 8847       break;
 8848     case T_INT:
 8849     case T_FLOAT:
 8850       evmovdqul(dst, kmask, src, merge, vector_len);
 8851       break;
 8852     case T_LONG:
 8853     case T_DOUBLE:
 8854       evmovdquq(dst, kmask, src, merge, vector_len);
 8855       break;
 8856     default:
 8857       fatal("Unexpected type argument %s", type2name(type));
 8858       break;
 8859   }
 8860 }
 8861 
 8862 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 8863   switch(type) {
 8864     case T_BYTE:
 8865     case T_BOOLEAN:
 8866       evmovdqub(dst, kmask, src, merge, vector_len);
 8867       break;
 8868     case T_CHAR:
 8869     case T_SHORT:
 8870       evmovdquw(dst, kmask, src, merge, vector_len);
 8871       break;
 8872     case T_INT:
 8873     case T_FLOAT:
 8874       evmovdqul(dst, kmask, src, merge, vector_len);
 8875       break;
 8876     case T_LONG:
 8877     case T_DOUBLE:
 8878       evmovdquq(dst, kmask, src, merge, vector_len);
 8879       break;
 8880     default:
 8881       fatal("Unexpected type argument %s", type2name(type));
 8882       break;
 8883   }
 8884 }
 8885 
 8886 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 8887   switch(masklen) {
 8888     case 2:
 8889        knotbl(dst, src);
 8890        movl(rtmp, 3);
 8891        kmovbl(ktmp, rtmp);
 8892        kandbl(dst, ktmp, dst);
 8893        break;
 8894     case 4:
 8895        knotbl(dst, src);
 8896        movl(rtmp, 15);
 8897        kmovbl(ktmp, rtmp);
 8898        kandbl(dst, ktmp, dst);
 8899        break;
 8900     case 8:
 8901        knotbl(dst, src);
 8902        break;
 8903     case 16:
 8904        knotwl(dst, src);
 8905        break;
 8906     case 32:
 8907        knotdl(dst, src);
 8908        break;
 8909     case 64:
 8910        knotql(dst, src);
 8911        break;
 8912     default:
 8913       fatal("Unexpected vector length %d", masklen);
 8914       break;
 8915   }
 8916 }
 8917 
 8918 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8919   switch(type) {
 8920     case T_BOOLEAN:
 8921     case T_BYTE:
 8922        kandbl(dst, src1, src2);
 8923        break;
 8924     case T_CHAR:
 8925     case T_SHORT:
 8926        kandwl(dst, src1, src2);
 8927        break;
 8928     case T_INT:
 8929     case T_FLOAT:
 8930        kanddl(dst, src1, src2);
 8931        break;
 8932     case T_LONG:
 8933     case T_DOUBLE:
 8934        kandql(dst, src1, src2);
 8935        break;
 8936     default:
 8937       fatal("Unexpected type argument %s", type2name(type));
 8938       break;
 8939   }
 8940 }
 8941 
 8942 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8943   switch(type) {
 8944     case T_BOOLEAN:
 8945     case T_BYTE:
 8946        korbl(dst, src1, src2);
 8947        break;
 8948     case T_CHAR:
 8949     case T_SHORT:
 8950        korwl(dst, src1, src2);
 8951        break;
 8952     case T_INT:
 8953     case T_FLOAT:
 8954        kordl(dst, src1, src2);
 8955        break;
 8956     case T_LONG:
 8957     case T_DOUBLE:
 8958        korql(dst, src1, src2);
 8959        break;
 8960     default:
 8961       fatal("Unexpected type argument %s", type2name(type));
 8962       break;
 8963   }
 8964 }
 8965 
 8966 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8967   switch(type) {
 8968     case T_BOOLEAN:
 8969     case T_BYTE:
 8970        kxorbl(dst, src1, src2);
 8971        break;
 8972     case T_CHAR:
 8973     case T_SHORT:
 8974        kxorwl(dst, src1, src2);
 8975        break;
 8976     case T_INT:
 8977     case T_FLOAT:
 8978        kxordl(dst, src1, src2);
 8979        break;
 8980     case T_LONG:
 8981     case T_DOUBLE:
 8982        kxorql(dst, src1, src2);
 8983        break;
 8984     default:
 8985       fatal("Unexpected type argument %s", type2name(type));
 8986       break;
 8987   }
 8988 }
 8989 
 8990 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 8991   switch(type) {
 8992     case T_BOOLEAN:
 8993     case T_BYTE:
 8994       evpermb(dst, mask, nds, src, merge, vector_len); break;
 8995     case T_CHAR:
 8996     case T_SHORT:
 8997       evpermw(dst, mask, nds, src, merge, vector_len); break;
 8998     case T_INT:
 8999     case T_FLOAT:
 9000       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9001     case T_LONG:
 9002     case T_DOUBLE:
 9003       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9004     default:
 9005       fatal("Unexpected type argument %s", type2name(type)); break;
 9006   }
 9007 }
 9008 
 9009 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9010   switch(type) {
 9011     case T_BOOLEAN:
 9012     case T_BYTE:
 9013       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9014     case T_CHAR:
 9015     case T_SHORT:
 9016       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9017     case T_INT:
 9018     case T_FLOAT:
 9019       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9020     case T_LONG:
 9021     case T_DOUBLE:
 9022       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9023     default:
 9024       fatal("Unexpected type argument %s", type2name(type)); break;
 9025   }
 9026 }
 9027 
 9028 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9029   switch(type) {
 9030     case T_BYTE:
 9031       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9032     case T_SHORT:
 9033       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9034     case T_INT:
 9035       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9036     case T_LONG:
 9037       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9038     default:
 9039       fatal("Unexpected type argument %s", type2name(type)); break;
 9040   }
 9041 }
 9042 
 9043 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9044   switch(type) {
 9045     case T_BYTE:
 9046       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9047     case T_SHORT:
 9048       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9049     case T_INT:
 9050       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9051     case T_LONG:
 9052       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9053     default:
 9054       fatal("Unexpected type argument %s", type2name(type)); break;
 9055   }
 9056 }
 9057 
 9058 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9059   switch(type) {
 9060     case T_BYTE:
 9061       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9062     case T_SHORT:
 9063       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9064     case T_INT:
 9065       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9066     case T_LONG:
 9067       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9068     default:
 9069       fatal("Unexpected type argument %s", type2name(type)); break;
 9070   }
 9071 }
 9072 
 9073 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9074   switch(type) {
 9075     case T_BYTE:
 9076       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9077     case T_SHORT:
 9078       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9079     case T_INT:
 9080       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9081     case T_LONG:
 9082       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9083     default:
 9084       fatal("Unexpected type argument %s", type2name(type)); break;
 9085   }
 9086 }
 9087 
 9088 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9089   switch(type) {
 9090     case T_BYTE:
 9091       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9092     case T_SHORT:
 9093       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9094     case T_INT:
 9095       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9096     case T_LONG:
 9097       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9098     case T_FLOAT:
 9099       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9100     case T_DOUBLE:
 9101       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9102     default:
 9103       fatal("Unexpected type argument %s", type2name(type)); break;
 9104   }
 9105 }
 9106 
 9107 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9108   switch(type) {
 9109     case T_BYTE:
 9110       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9111     case T_SHORT:
 9112       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9113     case T_INT:
 9114       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9115     case T_LONG:
 9116       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9117     case T_FLOAT:
 9118       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9119     case T_DOUBLE:
 9120       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9121     default:
 9122       fatal("Unexpected type argument %s", type2name(type)); break;
 9123   }
 9124 }
 9125 
 9126 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9127   switch(type) {
 9128     case T_BYTE:
 9129       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9130     case T_SHORT:
 9131       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9132     case T_INT:
 9133       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9134     case T_LONG:
 9135       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9136     case T_FLOAT:
 9137       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9138     case T_DOUBLE:
 9139       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9140     default:
 9141       fatal("Unexpected type argument %s", type2name(type)); break;
 9142   }
 9143 }
 9144 
 9145 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9146   switch(type) {
 9147     case T_BYTE:
 9148       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9149     case T_SHORT:
 9150       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9151     case T_INT:
 9152       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9153     case T_LONG:
 9154       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9155     case T_FLOAT:
 9156       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9157     case T_DOUBLE:
 9158       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9159     default:
 9160       fatal("Unexpected type argument %s", type2name(type)); break;
 9161   }
 9162 }
 9163 
 9164 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9165   switch(type) {
 9166     case T_INT:
 9167       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9168     case T_LONG:
 9169       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9170     default:
 9171       fatal("Unexpected type argument %s", type2name(type)); break;
 9172   }
 9173 }
 9174 
 9175 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9176   switch(type) {
 9177     case T_INT:
 9178       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9179     case T_LONG:
 9180       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9181     default:
 9182       fatal("Unexpected type argument %s", type2name(type)); break;
 9183   }
 9184 }
 9185 
 9186 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9187   switch(type) {
 9188     case T_INT:
 9189       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9190     case T_LONG:
 9191       evporq(dst, mask, nds, src, merge, vector_len); break;
 9192     default:
 9193       fatal("Unexpected type argument %s", type2name(type)); break;
 9194   }
 9195 }
 9196 
 9197 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9198   switch(type) {
 9199     case T_INT:
 9200       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9201     case T_LONG:
 9202       evporq(dst, mask, nds, src, merge, vector_len); break;
 9203     default:
 9204       fatal("Unexpected type argument %s", type2name(type)); break;
 9205   }
 9206 }
 9207 
 9208 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9209   switch(type) {
 9210     case T_INT:
 9211       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9212     case T_LONG:
 9213       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9214     default:
 9215       fatal("Unexpected type argument %s", type2name(type)); break;
 9216   }
 9217 }
 9218 
 9219 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9220   switch(type) {
 9221     case T_INT:
 9222       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9223     case T_LONG:
 9224       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9225     default:
 9226       fatal("Unexpected type argument %s", type2name(type)); break;
 9227   }
 9228 }
 9229 
 9230 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9231   switch(masklen) {
 9232     case 8:
 9233        kortestbl(src1, src2);
 9234        break;
 9235     case 16:
 9236        kortestwl(src1, src2);
 9237        break;
 9238     case 32:
 9239        kortestdl(src1, src2);
 9240        break;
 9241     case 64:
 9242        kortestql(src1, src2);
 9243        break;
 9244     default:
 9245       fatal("Unexpected mask length %d", masklen);
 9246       break;
 9247   }
 9248 }
 9249 
 9250 
 9251 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
 9252   switch(masklen)  {
 9253     case 8:
 9254        ktestbl(src1, src2);
 9255        break;
 9256     case 16:
 9257        ktestwl(src1, src2);
 9258        break;
 9259     case 32:
 9260        ktestdl(src1, src2);
 9261        break;
 9262     case 64:
 9263        ktestql(src1, src2);
 9264        break;
 9265     default:
 9266       fatal("Unexpected mask length %d", masklen);
 9267       break;
 9268   }
 9269 }
 9270 
 9271 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9272   switch(type) {
 9273     case T_INT:
 9274       evprold(dst, mask, src, shift, merge, vlen_enc); break;
 9275     case T_LONG:
 9276       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
 9277     default:
 9278       fatal("Unexpected type argument %s", type2name(type)); break;
 9279       break;
 9280   }
 9281 }
 9282 
 9283 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9284   switch(type) {
 9285     case T_INT:
 9286       evprord(dst, mask, src, shift, merge, vlen_enc); break;
 9287     case T_LONG:
 9288       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
 9289     default:
 9290       fatal("Unexpected type argument %s", type2name(type)); break;
 9291   }
 9292 }
 9293 
 9294 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9295   switch(type) {
 9296     case T_INT:
 9297       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9298     case T_LONG:
 9299       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9300     default:
 9301       fatal("Unexpected type argument %s", type2name(type)); break;
 9302   }
 9303 }
 9304 
 9305 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9306   switch(type) {
 9307     case T_INT:
 9308       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9309     case T_LONG:
 9310       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9311     default:
 9312       fatal("Unexpected type argument %s", type2name(type)); break;
 9313   }
 9314 }
 9315 
 9316 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9317   assert(rscratch != noreg || always_reachable(src), "missing");
 9318 
 9319   if (reachable(src)) {
 9320     evpandq(dst, nds, as_Address(src), vector_len);
 9321   } else {
 9322     lea(rscratch, src);
 9323     evpandq(dst, nds, Address(rscratch, 0), vector_len);
 9324   }
 9325 }
 9326 
 9327 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 9328   assert(rscratch != noreg || always_reachable(src), "missing");
 9329 
 9330   if (reachable(src)) {
 9331     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
 9332   } else {
 9333     lea(rscratch, src);
 9334     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 9335   }
 9336 }
 9337 
 9338 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9339   assert(rscratch != noreg || always_reachable(src), "missing");
 9340 
 9341   if (reachable(src)) {
 9342     evporq(dst, nds, as_Address(src), vector_len);
 9343   } else {
 9344     lea(rscratch, src);
 9345     evporq(dst, nds, Address(rscratch, 0), vector_len);
 9346   }
 9347 }
 9348 
 9349 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9350   assert(rscratch != noreg || always_reachable(src), "missing");
 9351 
 9352   if (reachable(src)) {
 9353     vpshufb(dst, nds, as_Address(src), vector_len);
 9354   } else {
 9355     lea(rscratch, src);
 9356     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
 9357   }
 9358 }
 9359 
 9360 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9361   assert(rscratch != noreg || always_reachable(src), "missing");
 9362 
 9363   if (reachable(src)) {
 9364     Assembler::vpor(dst, nds, as_Address(src), vector_len);
 9365   } else {
 9366     lea(rscratch, src);
 9367     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
 9368   }
 9369 }
 9370 
 9371 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
 9372   assert(rscratch != noreg || always_reachable(src3), "missing");
 9373 
 9374   if (reachable(src3)) {
 9375     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
 9376   } else {
 9377     lea(rscratch, src3);
 9378     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
 9379   }
 9380 }
 9381 
 9382 #if COMPILER2_OR_JVMCI
 9383 
 9384 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
 9385                                  Register length, Register temp, int vec_enc) {
 9386   // Computing mask for predicated vector store.
 9387   movptr(temp, -1);
 9388   bzhiq(temp, temp, length);
 9389   kmov(mask, temp);
 9390   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
 9391 }
 9392 
 9393 // Set memory operation for length "less than" 64 bytes.
 9394 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
 9395                                        XMMRegister xmm, KRegister mask, Register length,
 9396                                        Register temp, bool use64byteVector) {
 9397   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9398   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9399   if (!use64byteVector) {
 9400     fill32(dst, disp, xmm);
 9401     subptr(length, 32 >> shift);
 9402     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
 9403   } else {
 9404     assert(MaxVectorSize == 64, "vector length != 64");
 9405     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
 9406   }
 9407 }
 9408 
 9409 
 9410 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
 9411                                        XMMRegister xmm, KRegister mask, Register length,
 9412                                        Register temp) {
 9413   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9414   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9415   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
 9416 }
 9417 
 9418 
 9419 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
 9420   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9421   vmovdqu(dst, xmm);
 9422 }
 9423 
 9424 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
 9425   fill32(Address(dst, disp), xmm);
 9426 }
 9427 
 9428 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
 9429   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9430   if (!use64byteVector) {
 9431     fill32(dst, xmm);
 9432     fill32(dst.plus_disp(32), xmm);
 9433   } else {
 9434     evmovdquq(dst, xmm, Assembler::AVX_512bit);
 9435   }
 9436 }
 9437 
 9438 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
 9439   fill64(Address(dst, disp), xmm, use64byteVector);
 9440 }
 9441 
 9442 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
 9443                                         Register count, Register rtmp, XMMRegister xtmp) {
 9444   Label L_exit;
 9445   Label L_fill_start;
 9446   Label L_fill_64_bytes;
 9447   Label L_fill_96_bytes;
 9448   Label L_fill_128_bytes;
 9449   Label L_fill_128_bytes_loop;
 9450   Label L_fill_128_loop_header;
 9451   Label L_fill_128_bytes_loop_header;
 9452   Label L_fill_128_bytes_loop_pre_header;
 9453   Label L_fill_zmm_sequence;
 9454 
 9455   int shift = -1;
 9456   int avx3threshold = VM_Version::avx3_threshold();
 9457   switch(type) {
 9458     case T_BYTE:  shift = 0;
 9459       break;
 9460     case T_SHORT: shift = 1;
 9461       break;
 9462     case T_INT:   shift = 2;
 9463       break;
 9464     /* Uncomment when LONG fill stubs are supported.
 9465     case T_LONG:  shift = 3;
 9466       break;
 9467     */
 9468     default:
 9469       fatal("Unhandled type: %s\n", type2name(type));
 9470   }
 9471 
 9472   if ((avx3threshold != 0)  || (MaxVectorSize == 32)) {
 9473 
 9474     if (MaxVectorSize == 64) {
 9475       cmpq(count, avx3threshold >> shift);
 9476       jcc(Assembler::greater, L_fill_zmm_sequence);
 9477     }
 9478 
 9479     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
 9480 
 9481     bind(L_fill_start);
 9482 
 9483     cmpq(count, 32 >> shift);
 9484     jccb(Assembler::greater, L_fill_64_bytes);
 9485     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9486     jmp(L_exit);
 9487 
 9488     bind(L_fill_64_bytes);
 9489     cmpq(count, 64 >> shift);
 9490     jccb(Assembler::greater, L_fill_96_bytes);
 9491     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9492     jmp(L_exit);
 9493 
 9494     bind(L_fill_96_bytes);
 9495     cmpq(count, 96 >> shift);
 9496     jccb(Assembler::greater, L_fill_128_bytes);
 9497     fill64(to, 0, xtmp);
 9498     subq(count, 64 >> shift);
 9499     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
 9500     jmp(L_exit);
 9501 
 9502     bind(L_fill_128_bytes);
 9503     cmpq(count, 128 >> shift);
 9504     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
 9505     fill64(to, 0, xtmp);
 9506     fill32(to, 64, xtmp);
 9507     subq(count, 96 >> shift);
 9508     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
 9509     jmp(L_exit);
 9510 
 9511     bind(L_fill_128_bytes_loop_pre_header);
 9512     {
 9513       mov(rtmp, to);
 9514       andq(rtmp, 31);
 9515       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
 9516       negq(rtmp);
 9517       addq(rtmp, 32);
 9518       mov64(r8, -1L);
 9519       bzhiq(r8, r8, rtmp);
 9520       kmovql(k2, r8);
 9521       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
 9522       addq(to, rtmp);
 9523       shrq(rtmp, shift);
 9524       subq(count, rtmp);
 9525     }
 9526 
 9527     cmpq(count, 128 >> shift);
 9528     jcc(Assembler::less, L_fill_start);
 9529 
 9530     bind(L_fill_128_bytes_loop_header);
 9531     subq(count, 128 >> shift);
 9532 
 9533     align32();
 9534     bind(L_fill_128_bytes_loop);
 9535       fill64(to, 0, xtmp);
 9536       fill64(to, 64, xtmp);
 9537       addq(to, 128);
 9538       subq(count, 128 >> shift);
 9539       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
 9540 
 9541     addq(count, 128 >> shift);
 9542     jcc(Assembler::zero, L_exit);
 9543     jmp(L_fill_start);
 9544   }
 9545 
 9546   if (MaxVectorSize == 64) {
 9547     // Sequence using 64 byte ZMM register.
 9548     Label L_fill_128_bytes_zmm;
 9549     Label L_fill_192_bytes_zmm;
 9550     Label L_fill_192_bytes_loop_zmm;
 9551     Label L_fill_192_bytes_loop_header_zmm;
 9552     Label L_fill_192_bytes_loop_pre_header_zmm;
 9553     Label L_fill_start_zmm_sequence;
 9554 
 9555     bind(L_fill_zmm_sequence);
 9556     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
 9557 
 9558     bind(L_fill_start_zmm_sequence);
 9559     cmpq(count, 64 >> shift);
 9560     jccb(Assembler::greater, L_fill_128_bytes_zmm);
 9561     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
 9562     jmp(L_exit);
 9563 
 9564     bind(L_fill_128_bytes_zmm);
 9565     cmpq(count, 128 >> shift);
 9566     jccb(Assembler::greater, L_fill_192_bytes_zmm);
 9567     fill64(to, 0, xtmp, true);
 9568     subq(count, 64 >> shift);
 9569     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
 9570     jmp(L_exit);
 9571 
 9572     bind(L_fill_192_bytes_zmm);
 9573     cmpq(count, 192 >> shift);
 9574     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
 9575     fill64(to, 0, xtmp, true);
 9576     fill64(to, 64, xtmp, true);
 9577     subq(count, 128 >> shift);
 9578     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
 9579     jmp(L_exit);
 9580 
 9581     bind(L_fill_192_bytes_loop_pre_header_zmm);
 9582     {
 9583       movq(rtmp, to);
 9584       andq(rtmp, 63);
 9585       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
 9586       negq(rtmp);
 9587       addq(rtmp, 64);
 9588       mov64(r8, -1L);
 9589       bzhiq(r8, r8, rtmp);
 9590       kmovql(k2, r8);
 9591       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
 9592       addq(to, rtmp);
 9593       shrq(rtmp, shift);
 9594       subq(count, rtmp);
 9595     }
 9596 
 9597     cmpq(count, 192 >> shift);
 9598     jcc(Assembler::less, L_fill_start_zmm_sequence);
 9599 
 9600     bind(L_fill_192_bytes_loop_header_zmm);
 9601     subq(count, 192 >> shift);
 9602 
 9603     align32();
 9604     bind(L_fill_192_bytes_loop_zmm);
 9605       fill64(to, 0, xtmp, true);
 9606       fill64(to, 64, xtmp, true);
 9607       fill64(to, 128, xtmp, true);
 9608       addq(to, 192);
 9609       subq(count, 192 >> shift);
 9610       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
 9611 
 9612     addq(count, 192 >> shift);
 9613     jcc(Assembler::zero, L_exit);
 9614     jmp(L_fill_start_zmm_sequence);
 9615   }
 9616   bind(L_exit);
 9617 }
 9618 #endif //COMPILER2_OR_JVMCI
 9619 
 9620 
 9621 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
 9622   Label done;
 9623   cvttss2sil(dst, src);
 9624   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9625   cmpl(dst, 0x80000000); // float_sign_flip
 9626   jccb(Assembler::notEqual, done);
 9627   subptr(rsp, 8);
 9628   movflt(Address(rsp, 0), src);
 9629   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
 9630   pop(dst);
 9631   bind(done);
 9632 }
 9633 
 9634 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
 9635   Label done;
 9636   cvttsd2sil(dst, src);
 9637   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9638   cmpl(dst, 0x80000000); // float_sign_flip
 9639   jccb(Assembler::notEqual, done);
 9640   subptr(rsp, 8);
 9641   movdbl(Address(rsp, 0), src);
 9642   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
 9643   pop(dst);
 9644   bind(done);
 9645 }
 9646 
 9647 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
 9648   Label done;
 9649   cvttss2siq(dst, src);
 9650   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9651   jccb(Assembler::notEqual, done);
 9652   subptr(rsp, 8);
 9653   movflt(Address(rsp, 0), src);
 9654   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
 9655   pop(dst);
 9656   bind(done);
 9657 }
 9658 
 9659 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9660   // Following code is line by line assembly translation rounding algorithm.
 9661   // Please refer to java.lang.Math.round(float) algorithm for details.
 9662   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
 9663   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
 9664   const int32_t FloatConsts_EXP_BIAS = 127;
 9665   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
 9666   const int32_t MINUS_32 = 0xFFFFFFE0;
 9667   Label L_special_case, L_block1, L_exit;
 9668   movl(rtmp, FloatConsts_EXP_BIT_MASK);
 9669   movdl(dst, src);
 9670   andl(dst, rtmp);
 9671   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
 9672   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
 9673   subl(rtmp, dst);
 9674   movl(rcx, rtmp);
 9675   movl(dst, MINUS_32);
 9676   testl(rtmp, dst);
 9677   jccb(Assembler::notEqual, L_special_case);
 9678   movdl(dst, src);
 9679   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
 9680   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
 9681   movdl(rtmp, src);
 9682   testl(rtmp, rtmp);
 9683   jccb(Assembler::greaterEqual, L_block1);
 9684   negl(dst);
 9685   bind(L_block1);
 9686   sarl(dst);
 9687   addl(dst, 0x1);
 9688   sarl(dst, 0x1);
 9689   jmp(L_exit);
 9690   bind(L_special_case);
 9691   convert_f2i(dst, src);
 9692   bind(L_exit);
 9693 }
 9694 
 9695 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9696   // Following code is line by line assembly translation rounding algorithm.
 9697   // Please refer to java.lang.Math.round(double) algorithm for details.
 9698   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
 9699   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
 9700   const int64_t DoubleConsts_EXP_BIAS = 1023;
 9701   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
 9702   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
 9703   Label L_special_case, L_block1, L_exit;
 9704   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
 9705   movq(dst, src);
 9706   andq(dst, rtmp);
 9707   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
 9708   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
 9709   subq(rtmp, dst);
 9710   movq(rcx, rtmp);
 9711   mov64(dst, MINUS_64);
 9712   testq(rtmp, dst);
 9713   jccb(Assembler::notEqual, L_special_case);
 9714   movq(dst, src);
 9715   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
 9716   andq(dst, rtmp);
 9717   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
 9718   orq(dst, rtmp);
 9719   movq(rtmp, src);
 9720   testq(rtmp, rtmp);
 9721   jccb(Assembler::greaterEqual, L_block1);
 9722   negq(dst);
 9723   bind(L_block1);
 9724   sarq(dst);
 9725   addq(dst, 0x1);
 9726   sarq(dst, 0x1);
 9727   jmp(L_exit);
 9728   bind(L_special_case);
 9729   convert_d2l(dst, src);
 9730   bind(L_exit);
 9731 }
 9732 
 9733 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
 9734   Label done;
 9735   cvttsd2siq(dst, src);
 9736   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9737   jccb(Assembler::notEqual, done);
 9738   subptr(rsp, 8);
 9739   movdbl(Address(rsp, 0), src);
 9740   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
 9741   pop(dst);
 9742   bind(done);
 9743 }
 9744 
 9745 void MacroAssembler::cache_wb(Address line)
 9746 {
 9747   // 64 bit cpus always support clflush
 9748   assert(VM_Version::supports_clflush(), "clflush should be available");
 9749   bool optimized = VM_Version::supports_clflushopt();
 9750   bool no_evict = VM_Version::supports_clwb();
 9751 
 9752   // prefer clwb (writeback without evict) otherwise
 9753   // prefer clflushopt (potentially parallel writeback with evict)
 9754   // otherwise fallback on clflush (serial writeback with evict)
 9755 
 9756   if (optimized) {
 9757     if (no_evict) {
 9758       clwb(line);
 9759     } else {
 9760       clflushopt(line);
 9761     }
 9762   } else {
 9763     // no need for fence when using CLFLUSH
 9764     clflush(line);
 9765   }
 9766 }
 9767 
 9768 void MacroAssembler::cache_wbsync(bool is_pre)
 9769 {
 9770   assert(VM_Version::supports_clflush(), "clflush should be available");
 9771   bool optimized = VM_Version::supports_clflushopt();
 9772   bool no_evict = VM_Version::supports_clwb();
 9773 
 9774   // pick the correct implementation
 9775 
 9776   if (!is_pre && (optimized || no_evict)) {
 9777     // need an sfence for post flush when using clflushopt or clwb
 9778     // otherwise no no need for any synchroniaztion
 9779 
 9780     sfence();
 9781   }
 9782 }
 9783 
 9784 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
 9785   switch (cond) {
 9786     // Note some conditions are synonyms for others
 9787     case Assembler::zero:         return Assembler::notZero;
 9788     case Assembler::notZero:      return Assembler::zero;
 9789     case Assembler::less:         return Assembler::greaterEqual;
 9790     case Assembler::lessEqual:    return Assembler::greater;
 9791     case Assembler::greater:      return Assembler::lessEqual;
 9792     case Assembler::greaterEqual: return Assembler::less;
 9793     case Assembler::below:        return Assembler::aboveEqual;
 9794     case Assembler::belowEqual:   return Assembler::above;
 9795     case Assembler::above:        return Assembler::belowEqual;
 9796     case Assembler::aboveEqual:   return Assembler::below;
 9797     case Assembler::overflow:     return Assembler::noOverflow;
 9798     case Assembler::noOverflow:   return Assembler::overflow;
 9799     case Assembler::negative:     return Assembler::positive;
 9800     case Assembler::positive:     return Assembler::negative;
 9801     case Assembler::parity:       return Assembler::noParity;
 9802     case Assembler::noParity:     return Assembler::parity;
 9803   }
 9804   ShouldNotReachHere(); return Assembler::overflow;
 9805 }
 9806 
 9807 // This is simply a call to Thread::current()
 9808 void MacroAssembler::get_thread_slow(Register thread) {
 9809   if (thread != rax) {
 9810     push(rax);
 9811   }
 9812   push(rdi);
 9813   push(rsi);
 9814   push(rdx);
 9815   push(rcx);
 9816   push(r8);
 9817   push(r9);
 9818   push(r10);
 9819   push(r11);
 9820 
 9821   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
 9822 
 9823   pop(r11);
 9824   pop(r10);
 9825   pop(r9);
 9826   pop(r8);
 9827   pop(rcx);
 9828   pop(rdx);
 9829   pop(rsi);
 9830   pop(rdi);
 9831   if (thread != rax) {
 9832     mov(thread, rax);
 9833     pop(rax);
 9834   }
 9835 }
 9836 
 9837 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
 9838   Label L_stack_ok;
 9839   if (bias == 0) {
 9840     testptr(sp, 2 * wordSize - 1);
 9841   } else {
 9842     // lea(tmp, Address(rsp, bias);
 9843     mov(tmp, sp);
 9844     addptr(tmp, bias);
 9845     testptr(tmp, 2 * wordSize - 1);
 9846   }
 9847   jcc(Assembler::equal, L_stack_ok);
 9848   block_comment(msg);
 9849   stop(msg);
 9850   bind(L_stack_ok);
 9851 }
 9852 
 9853 // Implements fast-locking.
 9854 //
 9855 // obj: the object to be locked
 9856 // reg_rax: rax
 9857 // thread: the thread which attempts to lock obj
 9858 // tmp: a temporary register
 9859 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
 9860   Register thread = r15_thread;
 9861 
 9862   assert(reg_rax == rax, "");
 9863   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
 9864 
 9865   Label push;
 9866   const Register top = tmp;
 9867 
 9868   // Preload the markWord. It is important that this is the first
 9869   // instruction emitted as it is part of C1's null check semantics.
 9870   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9871 
 9872   if (UseObjectMonitorTable) {
 9873     // Clear cache in case fast locking succeeds or we need to take the slow-path.
 9874     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
 9875   }
 9876 
 9877   if (DiagnoseSyncOnValueBasedClasses != 0) {
 9878     load_klass(tmp, obj, rscratch1);
 9879     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
 9880     jcc(Assembler::notZero, slow);
 9881   }
 9882 
 9883   // Load top.
 9884   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9885 
 9886   // Check if the lock-stack is full.
 9887   cmpl(top, LockStack::end_offset());
 9888   jcc(Assembler::greaterEqual, slow);
 9889 
 9890   // Check for recursion.
 9891   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9892   jcc(Assembler::equal, push);
 9893 
 9894   // Check header for monitor (0b10).
 9895   testptr(reg_rax, markWord::monitor_value);
 9896   jcc(Assembler::notZero, slow);
 9897 
 9898   // Try to lock. Transition lock bits 0b01 => 0b00
 9899   movptr(tmp, reg_rax);
 9900   andptr(tmp, ~(int32_t)markWord::unlocked_value);
 9901   orptr(reg_rax, markWord::unlocked_value);
 9902   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9903   jcc(Assembler::notEqual, slow);
 9904 
 9905   // Restore top, CAS clobbers register.
 9906   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9907 
 9908   bind(push);
 9909   // After successful lock, push object on lock-stack.
 9910   movptr(Address(thread, top), obj);
 9911   incrementl(top, oopSize);
 9912   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
 9913 }
 9914 
 9915 // Implements fast-unlocking.
 9916 //
 9917 // obj: the object to be unlocked
 9918 // reg_rax: rax
 9919 // thread: the thread
 9920 // tmp: a temporary register
 9921 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
 9922   Register thread = r15_thread;
 9923 
 9924   assert(reg_rax == rax, "");
 9925   assert_different_registers(obj, reg_rax, thread, tmp);
 9926 
 9927   Label unlocked, push_and_slow;
 9928   const Register top = tmp;
 9929 
 9930   // Check if obj is top of lock-stack.
 9931   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9932   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9933   jcc(Assembler::notEqual, slow);
 9934 
 9935   // Pop lock-stack.
 9936   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
 9937   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
 9938 
 9939   // Check if recursive.
 9940   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
 9941   jcc(Assembler::equal, unlocked);
 9942 
 9943   // Not recursive. Check header for monitor (0b10).
 9944   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9945   testptr(reg_rax, markWord::monitor_value);
 9946   jcc(Assembler::notZero, push_and_slow);
 9947 
 9948 #ifdef ASSERT
 9949   // Check header not unlocked (0b01).
 9950   Label not_unlocked;
 9951   testptr(reg_rax, markWord::unlocked_value);
 9952   jcc(Assembler::zero, not_unlocked);
 9953   stop("fast_unlock already unlocked");
 9954   bind(not_unlocked);
 9955 #endif
 9956 
 9957   // Try to unlock. Transition lock bits 0b00 => 0b01
 9958   movptr(tmp, reg_rax);
 9959   orptr(tmp, markWord::unlocked_value);
 9960   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9961   jcc(Assembler::equal, unlocked);
 9962 
 9963   bind(push_and_slow);
 9964   // Restore lock-stack and handle the unlock in runtime.
 9965 #ifdef ASSERT
 9966   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9967   movptr(Address(thread, top), obj);
 9968 #endif
 9969   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
 9970   jmp(slow);
 9971 
 9972   bind(unlocked);
 9973 }
 9974 
 9975 // Saves legacy GPRs state on stack.
 9976 void MacroAssembler::save_legacy_gprs() {
 9977   subq(rsp, 16 * wordSize);
 9978   movq(Address(rsp, 15 * wordSize), rax);
 9979   movq(Address(rsp, 14 * wordSize), rcx);
 9980   movq(Address(rsp, 13 * wordSize), rdx);
 9981   movq(Address(rsp, 12 * wordSize), rbx);
 9982   movq(Address(rsp, 10 * wordSize), rbp);
 9983   movq(Address(rsp, 9 * wordSize), rsi);
 9984   movq(Address(rsp, 8 * wordSize), rdi);
 9985   movq(Address(rsp, 7 * wordSize), r8);
 9986   movq(Address(rsp, 6 * wordSize), r9);
 9987   movq(Address(rsp, 5 * wordSize), r10);
 9988   movq(Address(rsp, 4 * wordSize), r11);
 9989   movq(Address(rsp, 3 * wordSize), r12);
 9990   movq(Address(rsp, 2 * wordSize), r13);
 9991   movq(Address(rsp, wordSize), r14);
 9992   movq(Address(rsp, 0), r15);
 9993 }
 9994 
 9995 // Resotres back legacy GPRs state from stack.
 9996 void MacroAssembler::restore_legacy_gprs() {
 9997   movq(r15, Address(rsp, 0));
 9998   movq(r14, Address(rsp, wordSize));
 9999   movq(r13, Address(rsp, 2 * wordSize));
10000   movq(r12, Address(rsp, 3 * wordSize));
10001   movq(r11, Address(rsp, 4 * wordSize));
10002   movq(r10, Address(rsp, 5 * wordSize));
10003   movq(r9,  Address(rsp, 6 * wordSize));
10004   movq(r8,  Address(rsp, 7 * wordSize));
10005   movq(rdi, Address(rsp, 8 * wordSize));
10006   movq(rsi, Address(rsp, 9 * wordSize));
10007   movq(rbp, Address(rsp, 10 * wordSize));
10008   movq(rbx, Address(rsp, 12 * wordSize));
10009   movq(rdx, Address(rsp, 13 * wordSize));
10010   movq(rcx, Address(rsp, 14 * wordSize));
10011   movq(rax, Address(rsp, 15 * wordSize));
10012   addq(rsp, 16 * wordSize);
10013 }
10014 
10015 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10016   if (VM_Version::supports_apx_f()) {
10017     esetzucc(comparison, dst);
10018   } else {
10019     setb(comparison, dst);
10020     movzbl(dst, dst);
10021   }
10022 }