1 /*
    2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
    3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
    4  *
    5  * This code is free software; you can redistribute it and/or modify it
    6  * under the terms of the GNU General Public License version 2 only, as
    7  * published by the Free Software Foundation.
    8  *
    9  * This code is distributed in the hope that it will be useful, but WITHOUT
   10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
   11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
   12  * version 2 for more details (a copy is included in the LICENSE file that
   13  * accompanied this code).
   14  *
   15  * You should have received a copy of the GNU General Public License version
   16  * 2 along with this work; if not, write to the Free Software Foundation,
   17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
   18  *
   19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
   20  * or visit www.oracle.com if you need additional information or have any
   21  * questions.
   22  *
   23  */
   24 
   25 #include "asm/assembler.hpp"
   26 #include "asm/assembler.inline.hpp"
   27 #include "code/aotCodeCache.hpp"
   28 #include "code/compiledIC.hpp"
   29 #include "compiler/compiler_globals.hpp"
   30 #include "compiler/disassembler.hpp"

   31 #include "crc32c.h"
   32 #include "gc/shared/barrierSet.hpp"
   33 #include "gc/shared/barrierSetAssembler.hpp"
   34 #include "gc/shared/collectedHeap.inline.hpp"
   35 #include "gc/shared/tlab_globals.hpp"
   36 #include "interpreter/bytecodeHistogram.hpp"
   37 #include "interpreter/interpreter.hpp"
   38 #include "interpreter/interpreterRuntime.hpp"
   39 #include "jvm.h"
   40 #include "memory/resourceArea.hpp"
   41 #include "memory/universe.hpp"
   42 #include "oops/accessDecorators.hpp"
   43 #include "oops/compressedKlass.inline.hpp"
   44 #include "oops/compressedOops.inline.hpp"
   45 #include "oops/klass.inline.hpp"

   46 #include "prims/methodHandles.hpp"

   47 #include "runtime/continuation.hpp"
   48 #include "runtime/interfaceSupport.inline.hpp"
   49 #include "runtime/javaThread.hpp"
   50 #include "runtime/jniHandles.hpp"
   51 #include "runtime/objectMonitor.hpp"
   52 #include "runtime/os.hpp"
   53 #include "runtime/safepoint.hpp"
   54 #include "runtime/safepointMechanism.hpp"
   55 #include "runtime/sharedRuntime.hpp"

   56 #include "runtime/stubRoutines.hpp"
   57 #include "utilities/checkedCast.hpp"
   58 #include "utilities/macros.hpp"




   59 
   60 #ifdef PRODUCT
   61 #define BLOCK_COMMENT(str) /* nothing */
   62 #define STOP(error) stop(error)
   63 #else
   64 #define BLOCK_COMMENT(str) block_comment(str)
   65 #define STOP(error) block_comment(error); stop(error)
   66 #endif
   67 
   68 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   69 
   70 #ifdef ASSERT
   71 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   72 #endif
   73 
   74 static const Assembler::Condition reverse[] = {
   75     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   76     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   77     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   78     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
   79     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
   80     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
   81     Assembler::above          /* belowEqual    = 0x6 */ ,
   82     Assembler::belowEqual     /* above         = 0x7 */ ,
   83     Assembler::positive       /* negative      = 0x8 */ ,
   84     Assembler::negative       /* positive      = 0x9 */ ,
   85     Assembler::noParity       /* parity        = 0xa */ ,
   86     Assembler::parity         /* noParity      = 0xb */ ,
   87     Assembler::greaterEqual   /* less          = 0xc */ ,
   88     Assembler::less           /* greaterEqual  = 0xd */ ,
   89     Assembler::greater        /* lessEqual     = 0xe */ ,
   90     Assembler::lessEqual      /* greater       = 0xf, */
   91 
   92 };
   93 
   94 
   95 // Implementation of MacroAssembler
   96 
   97 Address MacroAssembler::as_Address(AddressLiteral adr) {
   98   // amd64 always does this as a pc-rel
   99   // we can be absolute or disp based on the instruction type
  100   // jmp/call are displacements others are absolute
  101   assert(!adr.is_lval(), "must be rval");
  102   assert(reachable(adr), "must be");
  103   return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
  104 
  105 }
  106 
  107 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  108   AddressLiteral base = adr.base();
  109   lea(rscratch, base);
  110   Address index = adr.index();
  111   assert(index._disp == 0, "must not have disp"); // maybe it can?
  112   Address array(rscratch, index._index, index._scale, index._disp);
  113   return array;
  114 }
  115 
  116 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
  117   Label L, E;
  118 
  119 #ifdef _WIN64
  120   // Windows always allocates space for it's register args
  121   assert(num_args <= 4, "only register arguments supported");
  122   subq(rsp,  frame::arg_reg_save_area_bytes);
  123 #endif
  124 
  125   // Align stack if necessary
  126   testl(rsp, 15);
  127   jcc(Assembler::zero, L);
  128 
  129   subq(rsp, 8);
  130   call(RuntimeAddress(entry_point));
  131   addq(rsp, 8);
  132   jmp(E);
  133 
  134   bind(L);
  135   call(RuntimeAddress(entry_point));
  136 
  137   bind(E);
  138 
  139 #ifdef _WIN64
  140   // restore stack pointer
  141   addq(rsp, frame::arg_reg_save_area_bytes);
  142 #endif
  143 }
  144 
  145 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
  146   assert(!src2.is_lval(), "should use cmpptr");
  147   assert(rscratch != noreg || always_reachable(src2), "missing");
  148 
  149   if (reachable(src2)) {
  150     cmpq(src1, as_Address(src2));
  151   } else {
  152     lea(rscratch, src2);
  153     Assembler::cmpq(src1, Address(rscratch, 0));
  154   }
  155 }
  156 
  157 int MacroAssembler::corrected_idivq(Register reg) {
  158   // Full implementation of Java ldiv and lrem; checks for special
  159   // case as described in JVM spec., p.243 & p.271.  The function
  160   // returns the (pc) offset of the idivl instruction - may be needed
  161   // for implicit exceptions.
  162   //
  163   //         normal case                           special case
  164   //
  165   // input : rax: dividend                         min_long
  166   //         reg: divisor   (may not be eax/edx)   -1
  167   //
  168   // output: rax: quotient  (= rax idiv reg)       min_long
  169   //         rdx: remainder (= rax irem reg)       0
  170   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
  171   static const int64_t min_long = 0x8000000000000000;
  172   Label normal_case, special_case;
  173 
  174   // check for special case
  175   cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
  176   jcc(Assembler::notEqual, normal_case);
  177   xorl(rdx, rdx); // prepare rdx for possible special case (where
  178                   // remainder = 0)
  179   cmpq(reg, -1);
  180   jcc(Assembler::equal, special_case);
  181 
  182   // handle normal case
  183   bind(normal_case);
  184   cdqq();
  185   int idivq_offset = offset();
  186   idivq(reg);
  187 
  188   // normal and special case exit
  189   bind(special_case);
  190 
  191   return idivq_offset;
  192 }
  193 
  194 void MacroAssembler::decrementq(Register reg, int value) {
  195   if (value == min_jint) { subq(reg, value); return; }
  196   if (value <  0) { incrementq(reg, -value); return; }
  197   if (value == 0) {                        ; return; }
  198   if (value == 1 && UseIncDec) { decq(reg) ; return; }
  199   /* else */      { subq(reg, value)       ; return; }
  200 }
  201 
  202 void MacroAssembler::decrementq(Address dst, int value) {
  203   if (value == min_jint) { subq(dst, value); return; }
  204   if (value <  0) { incrementq(dst, -value); return; }
  205   if (value == 0) {                        ; return; }
  206   if (value == 1 && UseIncDec) { decq(dst) ; return; }
  207   /* else */      { subq(dst, value)       ; return; }
  208 }
  209 
  210 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
  211   assert(rscratch != noreg || always_reachable(dst), "missing");
  212 
  213   if (reachable(dst)) {
  214     incrementq(as_Address(dst));
  215   } else {
  216     lea(rscratch, dst);
  217     incrementq(Address(rscratch, 0));
  218   }
  219 }
  220 
  221 void MacroAssembler::incrementq(Register reg, int value) {
  222   if (value == min_jint) { addq(reg, value); return; }
  223   if (value <  0) { decrementq(reg, -value); return; }
  224   if (value == 0) {                        ; return; }
  225   if (value == 1 && UseIncDec) { incq(reg) ; return; }
  226   /* else */      { addq(reg, value)       ; return; }
  227 }
  228 
  229 void MacroAssembler::incrementq(Address dst, int value) {
  230   if (value == min_jint) { addq(dst, value); return; }
  231   if (value <  0) { decrementq(dst, -value); return; }
  232   if (value == 0) {                        ; return; }
  233   if (value == 1 && UseIncDec) { incq(dst) ; return; }
  234   /* else */      { addq(dst, value)       ; return; }
  235 }
  236 
  237 // 32bit can do a case table jump in one instruction but we no longer allow the base
  238 // to be installed in the Address class
  239 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  240   lea(rscratch, entry.base());
  241   Address dispatch = entry.index();
  242   assert(dispatch._base == noreg, "must be");
  243   dispatch._base = rscratch;
  244   jmp(dispatch);
  245 }
  246 
  247 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  248   ShouldNotReachHere(); // 64bit doesn't use two regs
  249   cmpq(x_lo, y_lo);
  250 }
  251 
  252 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  253   mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  254 }
  255 
  256 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  257   lea(rscratch, adr);
  258   movptr(dst, rscratch);
  259 }
  260 
  261 void MacroAssembler::leave() {
  262   // %%% is this really better? Why not on 32bit too?
  263   emit_int8((unsigned char)0xC9); // LEAVE
  264 }
  265 
  266 void MacroAssembler::lneg(Register hi, Register lo) {
  267   ShouldNotReachHere(); // 64bit doesn't use two regs
  268   negq(lo);
  269 }
  270 
  271 void MacroAssembler::movoop(Register dst, jobject obj) {
  272   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  273 }
  274 
  275 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  276   mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  277   movq(dst, rscratch);
  278 }
  279 
  280 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  281   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  282 }
  283 
  284 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  285   mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  286   movq(dst, rscratch);
  287 }
  288 
  289 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  290   if (src.is_lval()) {
  291     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  292   } else {
  293     if (reachable(src)) {
  294       movq(dst, as_Address(src));
  295     } else {
  296       lea(dst, src);
  297       movq(dst, Address(dst, 0));
  298     }
  299   }
  300 }
  301 
  302 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  303   movq(as_Address(dst, rscratch), src);
  304 }
  305 
  306 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  307   movq(dst, as_Address(src, dst /*rscratch*/));
  308 }
  309 
  310 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
  311 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  312   if (is_simm32(src)) {
  313     movptr(dst, checked_cast<int32_t>(src));
  314   } else {
  315     mov64(rscratch, src);
  316     movq(dst, rscratch);
  317   }
  318 }
  319 
  320 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  321   movoop(rscratch, obj);
  322   push(rscratch);
  323 }
  324 
  325 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  326   mov_metadata(rscratch, obj);
  327   push(rscratch);
  328 }
  329 
  330 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  331   lea(rscratch, src);
  332   if (src.is_lval()) {
  333     push(rscratch);
  334   } else {
  335     pushq(Address(rscratch, 0));
  336   }
  337 }
  338 
  339 static void pass_arg0(MacroAssembler* masm, Register arg) {
  340   if (c_rarg0 != arg ) {
  341     masm->mov(c_rarg0, arg);
  342   }
  343 }
  344 
  345 static void pass_arg1(MacroAssembler* masm, Register arg) {
  346   if (c_rarg1 != arg ) {
  347     masm->mov(c_rarg1, arg);
  348   }
  349 }
  350 
  351 static void pass_arg2(MacroAssembler* masm, Register arg) {
  352   if (c_rarg2 != arg ) {
  353     masm->mov(c_rarg2, arg);
  354   }
  355 }
  356 
  357 static void pass_arg3(MacroAssembler* masm, Register arg) {
  358   if (c_rarg3 != arg ) {
  359     masm->mov(c_rarg3, arg);
  360   }
  361 }
  362 
  363 void MacroAssembler::stop(const char* msg) {
  364   if (ShowMessageBoxOnError) {
  365     address rip = pc();
  366     pusha(); // get regs on stack
  367     lea(c_rarg1, InternalAddress(rip));
  368     movq(c_rarg2, rsp); // pass pointer to regs array
  369   }
  370   // Skip AOT caching C strings in scratch buffer.
  371   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  372   lea(c_rarg0, ExternalAddress((address) str));
  373   andq(rsp, -16); // align stack as required by ABI
  374   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
  375   hlt();
  376 }
  377 
  378 void MacroAssembler::warn(const char* msg) {
  379   push(rbp);
  380   movq(rbp, rsp);
  381   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  382   push_CPU_state();   // keeps alignment at 16 bytes
  383 
  384 #ifdef _WIN64
  385   // Windows always allocates space for its register args
  386   subq(rsp,  frame::arg_reg_save_area_bytes);
  387 #endif
  388   lea(c_rarg0, ExternalAddress((address) msg));
  389   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  390 
  391 #ifdef _WIN64
  392   // restore stack pointer
  393   addq(rsp, frame::arg_reg_save_area_bytes);
  394 #endif
  395   pop_CPU_state();
  396   mov(rsp, rbp);
  397   pop(rbp);
  398 }
  399 
  400 void MacroAssembler::print_state() {
  401   address rip = pc();
  402   pusha();            // get regs on stack
  403   push(rbp);
  404   movq(rbp, rsp);
  405   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  406   push_CPU_state();   // keeps alignment at 16 bytes
  407 
  408   lea(c_rarg0, InternalAddress(rip));
  409   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  410   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  411 
  412   pop_CPU_state();
  413   mov(rsp, rbp);
  414   pop(rbp);
  415   popa();
  416 }
  417 
  418 #ifndef PRODUCT
  419 extern "C" void findpc(intptr_t x);
  420 #endif
  421 
  422 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  423   // In order to get locks to work, we need to fake a in_VM state
  424   if (ShowMessageBoxOnError) {
  425     JavaThread* thread = JavaThread::current();
  426     JavaThreadState saved_state = thread->thread_state();
  427     thread->set_thread_state(_thread_in_vm);
  428 #ifndef PRODUCT
  429     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  430       ttyLocker ttyl;
  431       BytecodeCounter::print();
  432     }
  433 #endif
  434     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  435     // XXX correct this offset for amd64
  436     // This is the value of eip which points to where verify_oop will return.
  437     if (os::message_box(msg, "Execution stopped, print registers?")) {
  438       print_state64(pc, regs);
  439       BREAKPOINT;
  440     }
  441   }
  442   fatal("DEBUG MESSAGE: %s", msg);
  443 }
  444 
  445 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  446   ttyLocker ttyl;
  447   DebuggingContext debugging{};
  448   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  449 #ifndef PRODUCT
  450   tty->cr();
  451   findpc(pc);
  452   tty->cr();
  453 #endif
  454 #define PRINT_REG(rax, value) \
  455   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  456   PRINT_REG(rax, regs[15]);
  457   PRINT_REG(rbx, regs[12]);
  458   PRINT_REG(rcx, regs[14]);
  459   PRINT_REG(rdx, regs[13]);
  460   PRINT_REG(rdi, regs[8]);
  461   PRINT_REG(rsi, regs[9]);
  462   PRINT_REG(rbp, regs[10]);
  463   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  464   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  465   PRINT_REG(r8 , regs[7]);
  466   PRINT_REG(r9 , regs[6]);
  467   PRINT_REG(r10, regs[5]);
  468   PRINT_REG(r11, regs[4]);
  469   PRINT_REG(r12, regs[3]);
  470   PRINT_REG(r13, regs[2]);
  471   PRINT_REG(r14, regs[1]);
  472   PRINT_REG(r15, regs[0]);
  473 #undef PRINT_REG
  474   // Print some words near the top of the stack.
  475   int64_t* rsp = &regs[16];
  476   int64_t* dump_sp = rsp;
  477   for (int col1 = 0; col1 < 8; col1++) {
  478     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  479     os::print_location(tty, *dump_sp++);
  480   }
  481   for (int row = 0; row < 25; row++) {
  482     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  483     for (int col = 0; col < 4; col++) {
  484       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  485     }
  486     tty->cr();
  487   }
  488   // Print some instructions around pc:
  489   Disassembler::decode((address)pc-64, (address)pc);
  490   tty->print_cr("--------");
  491   Disassembler::decode((address)pc, (address)pc+32);
  492 }
  493 
  494 // The java_calling_convention describes stack locations as ideal slots on
  495 // a frame with no abi restrictions. Since we must observe abi restrictions
  496 // (like the placement of the register window) the slots must be biased by
  497 // the following value.
  498 static int reg2offset_in(VMReg r) {
  499   // Account for saved rbp and return address
  500   // This should really be in_preserve_stack_slots
  501   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  502 }
  503 
  504 static int reg2offset_out(VMReg r) {
  505   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  506 }
  507 
  508 // A long move
  509 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  510 
  511   // The calling conventions assures us that each VMregpair is either
  512   // all really one physical register or adjacent stack slots.
  513 
  514   if (src.is_single_phys_reg() ) {
  515     if (dst.is_single_phys_reg()) {
  516       if (dst.first() != src.first()) {
  517         mov(dst.first()->as_Register(), src.first()->as_Register());
  518       }
  519     } else {
  520       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  521              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  522       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  523     }
  524   } else if (dst.is_single_phys_reg()) {
  525     assert(src.is_single_reg(),  "not a stack pair");
  526     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  527   } else {
  528     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  529     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  530     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  531   }
  532 }
  533 
  534 // A double move
  535 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  536 
  537   // The calling conventions assures us that each VMregpair is either
  538   // all really one physical register or adjacent stack slots.
  539 
  540   if (src.is_single_phys_reg() ) {
  541     if (dst.is_single_phys_reg()) {
  542       // In theory these overlap but the ordering is such that this is likely a nop
  543       if ( src.first() != dst.first()) {
  544         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  545       }
  546     } else {
  547       assert(dst.is_single_reg(), "not a stack pair");
  548       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  549     }
  550   } else if (dst.is_single_phys_reg()) {
  551     assert(src.is_single_reg(),  "not a stack pair");
  552     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  553   } else {
  554     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  555     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  556     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  557   }
  558 }
  559 
  560 
  561 // A float arg may have to do float reg int reg conversion
  562 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  563   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  564 
  565   // The calling conventions assures us that each VMregpair is either
  566   // all really one physical register or adjacent stack slots.
  567 
  568   if (src.first()->is_stack()) {
  569     if (dst.first()->is_stack()) {
  570       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  571       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  572     } else {
  573       // stack to reg
  574       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  575       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  576     }
  577   } else if (dst.first()->is_stack()) {
  578     // reg to stack
  579     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  580     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  581   } else {
  582     // reg to reg
  583     // In theory these overlap but the ordering is such that this is likely a nop
  584     if ( src.first() != dst.first()) {
  585       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  586     }
  587   }
  588 }
  589 
  590 // On 64 bit we will store integer like items to the stack as
  591 // 64 bits items (x86_32/64 abi) even though java would only store
  592 // 32bits for a parameter. On 32bit it will simply be 32 bits
  593 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  594 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  595   if (src.first()->is_stack()) {
  596     if (dst.first()->is_stack()) {
  597       // stack to stack
  598       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  599       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  600     } else {
  601       // stack to reg
  602       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  603     }
  604   } else if (dst.first()->is_stack()) {
  605     // reg to stack
  606     // Do we really have to sign extend???
  607     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  608     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  609   } else {
  610     // Do we really have to sign extend???
  611     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
  612     if (dst.first() != src.first()) {
  613       movq(dst.first()->as_Register(), src.first()->as_Register());
  614     }
  615   }
  616 }
  617 
  618 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
  619   if (src.first()->is_stack()) {
  620     if (dst.first()->is_stack()) {
  621       // stack to stack
  622       movq(rax, Address(rbp, reg2offset_in(src.first())));
  623       movq(Address(rsp, reg2offset_out(dst.first())), rax);
  624     } else {
  625       // stack to reg
  626       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
  627     }
  628   } else if (dst.first()->is_stack()) {
  629     // reg to stack
  630     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
  631   } else {
  632     if (dst.first() != src.first()) {
  633       movq(dst.first()->as_Register(), src.first()->as_Register());
  634     }
  635   }
  636 }
  637 
  638 // An oop arg. Must pass a handle not the oop itself
  639 void MacroAssembler::object_move(OopMap* map,
  640                         int oop_handle_offset,
  641                         int framesize_in_slots,
  642                         VMRegPair src,
  643                         VMRegPair dst,
  644                         bool is_receiver,
  645                         int* receiver_offset) {
  646 
  647   // must pass a handle. First figure out the location we use as a handle
  648 
  649   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
  650 
  651   // See if oop is null if it is we need no handle
  652 
  653   if (src.first()->is_stack()) {
  654 
  655     // Oop is already on the stack as an argument
  656     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
  657     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
  658     if (is_receiver) {
  659       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
  660     }
  661 
  662     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
  663     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
  664     // conditionally move a null
  665     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
  666   } else {
  667 
  668     // Oop is in a register we must store it to the space we reserve
  669     // on the stack for oop_handles and pass a handle if oop is non-null
  670 
  671     const Register rOop = src.first()->as_Register();
  672     int oop_slot;
  673     if (rOop == j_rarg0)
  674       oop_slot = 0;
  675     else if (rOop == j_rarg1)
  676       oop_slot = 1;
  677     else if (rOop == j_rarg2)
  678       oop_slot = 2;
  679     else if (rOop == j_rarg3)
  680       oop_slot = 3;
  681     else if (rOop == j_rarg4)
  682       oop_slot = 4;
  683     else {
  684       assert(rOop == j_rarg5, "wrong register");
  685       oop_slot = 5;
  686     }
  687 
  688     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
  689     int offset = oop_slot*VMRegImpl::stack_slot_size;
  690 
  691     map->set_oop(VMRegImpl::stack2reg(oop_slot));
  692     // Store oop in handle area, may be null
  693     movptr(Address(rsp, offset), rOop);
  694     if (is_receiver) {
  695       *receiver_offset = offset;
  696     }
  697 
  698     cmpptr(rOop, NULL_WORD);
  699     lea(rHandle, Address(rsp, offset));
  700     // conditionally move a null from the handle area where it was just stored
  701     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
  702   }
  703 
  704   // If arg is on the stack then place it otherwise it is already in correct reg.
  705   if (dst.first()->is_stack()) {
  706     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
  707   }
  708 }
  709 
  710 void MacroAssembler::addptr(Register dst, int32_t imm32) {
  711   addq(dst, imm32);
  712 }
  713 
  714 void MacroAssembler::addptr(Register dst, Register src) {
  715   addq(dst, src);
  716 }
  717 
  718 void MacroAssembler::addptr(Address dst, Register src) {
  719   addq(dst, src);
  720 }
  721 
  722 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  723   assert(rscratch != noreg || always_reachable(src), "missing");
  724 
  725   if (reachable(src)) {
  726     Assembler::addsd(dst, as_Address(src));
  727   } else {
  728     lea(rscratch, src);
  729     Assembler::addsd(dst, Address(rscratch, 0));
  730   }
  731 }
  732 
  733 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
  734   assert(rscratch != noreg || always_reachable(src), "missing");
  735 
  736   if (reachable(src)) {
  737     addss(dst, as_Address(src));
  738   } else {
  739     lea(rscratch, src);
  740     addss(dst, Address(rscratch, 0));
  741   }
  742 }
  743 
  744 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  745   assert(rscratch != noreg || always_reachable(src), "missing");
  746 
  747   if (reachable(src)) {
  748     Assembler::addpd(dst, as_Address(src));
  749   } else {
  750     lea(rscratch, src);
  751     Assembler::addpd(dst, Address(rscratch, 0));
  752   }
  753 }
  754 
  755 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
  756 // Stub code is generated once and never copied.
  757 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
  758 void MacroAssembler::align64() {
  759   align(64, (uint)(uintptr_t)pc());
  760 }
  761 
  762 void MacroAssembler::align32() {
  763   align(32, (uint)(uintptr_t)pc());
  764 }
  765 
  766 void MacroAssembler::align(uint modulus) {
  767   // 8273459: Ensure alignment is possible with current segment alignment
  768   assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
  769   align(modulus, offset());
  770 }
  771 
  772 void MacroAssembler::align(uint modulus, uint target) {
  773   if (target % modulus != 0) {
  774     nop(modulus - (target % modulus));
  775   }
  776 }
  777 
  778 void MacroAssembler::push_f(XMMRegister r) {
  779   subptr(rsp, wordSize);
  780   movflt(Address(rsp, 0), r);
  781 }
  782 
  783 void MacroAssembler::pop_f(XMMRegister r) {
  784   movflt(r, Address(rsp, 0));
  785   addptr(rsp, wordSize);
  786 }
  787 
  788 void MacroAssembler::push_d(XMMRegister r) {
  789   subptr(rsp, 2 * wordSize);
  790   movdbl(Address(rsp, 0), r);
  791 }
  792 
  793 void MacroAssembler::pop_d(XMMRegister r) {
  794   movdbl(r, Address(rsp, 0));
  795   addptr(rsp, 2 * Interpreter::stackElementSize);
  796 }
  797 
  798 void MacroAssembler::push_ppx(Register src) {
  799   if (VM_Version::supports_apx_f()) {
  800     pushp(src);
  801   } else {
  802     Assembler::push(src);
  803   }
  804 }
  805 
  806 void MacroAssembler::pop_ppx(Register dst) {
  807   if (VM_Version::supports_apx_f()) {
  808     popp(dst);
  809   } else {
  810     Assembler::pop(dst);
  811   }
  812 }
  813 
  814 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  815   // Used in sign-masking with aligned address.
  816   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  817   assert(rscratch != noreg || always_reachable(src), "missing");
  818 
  819   if (UseAVX > 2 &&
  820       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
  821       (dst->encoding() >= 16)) {
  822     vpand(dst, dst, src, AVX_512bit, rscratch);
  823   } else if (reachable(src)) {
  824     Assembler::andpd(dst, as_Address(src));
  825   } else {
  826     lea(rscratch, src);
  827     Assembler::andpd(dst, Address(rscratch, 0));
  828   }
  829 }
  830 
  831 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
  832   // Used in sign-masking with aligned address.
  833   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  834   assert(rscratch != noreg || always_reachable(src), "missing");
  835 
  836   if (reachable(src)) {
  837     Assembler::andps(dst, as_Address(src));
  838   } else {
  839     lea(rscratch, src);
  840     Assembler::andps(dst, Address(rscratch, 0));
  841   }
  842 }
  843 
  844 void MacroAssembler::andptr(Register dst, int32_t imm32) {
  845   andq(dst, imm32);
  846 }
  847 
  848 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
  849   assert(rscratch != noreg || always_reachable(src), "missing");
  850 
  851   if (reachable(src)) {
  852     andq(dst, as_Address(src));
  853   } else {
  854     lea(rscratch, src);
  855     andq(dst, Address(rscratch, 0));
  856   }
  857 }
  858 
  859 void MacroAssembler::atomic_incl(Address counter_addr) {
  860   lock();
  861   incrementl(counter_addr);
  862 }
  863 
  864 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
  865   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  866 
  867   if (reachable(counter_addr)) {
  868     atomic_incl(as_Address(counter_addr));
  869   } else {
  870     lea(rscratch, counter_addr);
  871     atomic_incl(Address(rscratch, 0));
  872   }
  873 }
  874 
  875 void MacroAssembler::atomic_incq(Address counter_addr) {
  876   lock();
  877   incrementq(counter_addr);
  878 }
  879 
  880 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
  881   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  882 
  883   if (reachable(counter_addr)) {
  884     atomic_incq(as_Address(counter_addr));
  885   } else {
  886     lea(rscratch, counter_addr);
  887     atomic_incq(Address(rscratch, 0));
  888   }
  889 }
  890 
  891 // Writes to stack successive pages until offset reached to check for
  892 // stack overflow + shadow pages.  This clobbers tmp.
  893 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
  894   movptr(tmp, rsp);
  895   // Bang stack for total size given plus shadow page size.
  896   // Bang one page at a time because large size can bang beyond yellow and
  897   // red zones.
  898   Label loop;
  899   bind(loop);
  900   movl(Address(tmp, (-(int)os::vm_page_size())), size );
  901   subptr(tmp, (int)os::vm_page_size());
  902   subl(size, (int)os::vm_page_size());
  903   jcc(Assembler::greater, loop);
  904 
  905   // Bang down shadow pages too.
  906   // At this point, (tmp-0) is the last address touched, so don't
  907   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
  908   // was post-decremented.)  Skip this address by starting at i=1, and
  909   // touch a few more pages below.  N.B.  It is important to touch all
  910   // the way down including all pages in the shadow zone.
  911   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
  912     // this could be any sized move but this is can be a debugging crumb
  913     // so the bigger the better.
  914     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
  915   }
  916 }
  917 
  918 void MacroAssembler::reserved_stack_check() {
  919   // testing if reserved zone needs to be enabled
  920   Label no_reserved_zone_enabling;
  921 
  922   cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
  923   jcc(Assembler::below, no_reserved_zone_enabling);
  924 
  925   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
  926   jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
  927   should_not_reach_here();
  928 
  929   bind(no_reserved_zone_enabling);
  930 }
  931 
  932 void MacroAssembler::c2bool(Register x) {
  933   // implements x == 0 ? 0 : 1
  934   // note: must only look at least-significant byte of x
  935   //       since C-style booleans are stored in one byte
  936   //       only! (was bug)
  937   andl(x, 0xFF);
  938   setb(Assembler::notZero, x);
  939 }
  940 
  941 // Wouldn't need if AddressLiteral version had new name
  942 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
  943   Assembler::call(L, rtype);
  944 }
  945 
  946 void MacroAssembler::call(Register entry) {
  947   Assembler::call(entry);
  948 }
  949 
  950 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
  951   assert(rscratch != noreg || always_reachable(entry), "missing");
  952 
  953   if (reachable(entry)) {
  954     Assembler::call_literal(entry.target(), entry.rspec());
  955   } else {
  956     lea(rscratch, entry);
  957     Assembler::call(rscratch);
  958   }
  959 }
  960 
  961 void MacroAssembler::ic_call(address entry, jint method_index) {
  962   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
  963   // Needs full 64-bit immediate for later patching.
  964   Assembler::mov64(rax, (int64_t)Universe::non_oop_word());
  965   call(AddressLiteral(entry, rh));
  966 }
  967 
  968 int MacroAssembler::ic_check_size() {
  969   return UseCompactObjectHeaders ? 17 : 14;
  970 }
  971 
  972 int MacroAssembler::ic_check(int end_alignment) {
  973   Register receiver = j_rarg0;
  974   Register data = rax;
  975   Register temp = rscratch1;
  976 
  977   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
  978   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
  979   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
  980   // before the inline cache check here, and not after
  981   align(end_alignment, offset() + ic_check_size());
  982 
  983   int uep_offset = offset();
  984 
  985   if (UseCompactObjectHeaders) {
  986     load_narrow_klass_compact(temp, receiver);
  987     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  988   } else 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::mov64(Register dst, int64_t imm64) {
 1965   if (is_uimm32(imm64)) {
 1966     movl(dst, checked_cast<uint32_t>(imm64));
 1967   } else if (is_simm32(imm64)) {
 1968     movq(dst, checked_cast<int32_t>(imm64));
 1969   } else {
 1970     Assembler::mov64(dst, imm64);
 1971   }
 1972 }
 1973 
 1974 void MacroAssembler::mov64(Register dst, int64_t imm64, relocInfo::relocType rtype, int format) {
 1975   Assembler::mov64(dst, imm64, rtype, format);
 1976 }
 1977 
 1978 void MacroAssembler::movptr(Register dst, Register src) {
 1979   movq(dst, src);
 1980 }
 1981 
 1982 void MacroAssembler::movptr(Register dst, Address src) {
 1983   movq(dst, src);
 1984 }
 1985 
 1986 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 1987 void MacroAssembler::movptr(Register dst, intptr_t src) {
 1988   mov64(dst, src);
 1989 }
 1990 
 1991 void MacroAssembler::movptr(Address dst, Register src) {
 1992   movq(dst, src);
 1993 }
 1994 
 1995 void MacroAssembler::movptr(Address dst, int32_t src) {
 1996   movslq(dst, src);
 1997 }
 1998 
 1999 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 2000   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2001   Assembler::movdqu(dst, src);
 2002 }
 2003 
 2004 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 2005   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2006   Assembler::movdqu(dst, src);
 2007 }
 2008 
 2009 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2010   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2011   Assembler::movdqu(dst, src);
 2012 }
 2013 
 2014 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2015   assert(rscratch != noreg || always_reachable(src), "missing");
 2016 
 2017   if (reachable(src)) {
 2018     movdqu(dst, as_Address(src));
 2019   } else {
 2020     lea(rscratch, src);
 2021     movdqu(dst, Address(rscratch, 0));
 2022   }
 2023 }
 2024 
 2025 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2026   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2027   Assembler::vmovdqu(dst, src);
 2028 }
 2029 
 2030 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2031   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2032   Assembler::vmovdqu(dst, src);
 2033 }
 2034 
 2035 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2036   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2037   Assembler::vmovdqu(dst, src);
 2038 }
 2039 
 2040 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2041   assert(rscratch != noreg || always_reachable(src), "missing");
 2042 
 2043   if (reachable(src)) {
 2044     vmovdqu(dst, as_Address(src));
 2045   }
 2046   else {
 2047     lea(rscratch, src);
 2048     vmovdqu(dst, Address(rscratch, 0));
 2049   }
 2050 }
 2051 
 2052 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2053   assert(rscratch != noreg || always_reachable(src), "missing");
 2054 
 2055   if (vector_len == AVX_512bit) {
 2056     evmovdquq(dst, src, AVX_512bit, rscratch);
 2057   } else if (vector_len == AVX_256bit) {
 2058     vmovdqu(dst, src, rscratch);
 2059   } else {
 2060     movdqu(dst, src, rscratch);
 2061   }
 2062 }
 2063 
 2064 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
 2065   if (vector_len == AVX_512bit) {
 2066     evmovdquq(dst, src, AVX_512bit);
 2067   } else if (vector_len == AVX_256bit) {
 2068     vmovdqu(dst, src);
 2069   } else {
 2070     movdqu(dst, src);
 2071   }
 2072 }
 2073 
 2074 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
 2075   if (vector_len == AVX_512bit) {
 2076     evmovdquq(dst, src, AVX_512bit);
 2077   } else if (vector_len == AVX_256bit) {
 2078     vmovdqu(dst, src);
 2079   } else {
 2080     movdqu(dst, src);
 2081   }
 2082 }
 2083 
 2084 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
 2085   if (vector_len == AVX_512bit) {
 2086     evmovdquq(dst, src, AVX_512bit);
 2087   } else if (vector_len == AVX_256bit) {
 2088     vmovdqu(dst, src);
 2089   } else {
 2090     movdqu(dst, src);
 2091   }
 2092 }
 2093 
 2094 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2095   assert(rscratch != noreg || always_reachable(src), "missing");
 2096 
 2097   if (reachable(src)) {
 2098     vmovdqa(dst, as_Address(src));
 2099   }
 2100   else {
 2101     lea(rscratch, src);
 2102     vmovdqa(dst, Address(rscratch, 0));
 2103   }
 2104 }
 2105 
 2106 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2107   assert(rscratch != noreg || always_reachable(src), "missing");
 2108 
 2109   if (vector_len == AVX_512bit) {
 2110     evmovdqaq(dst, src, AVX_512bit, rscratch);
 2111   } else if (vector_len == AVX_256bit) {
 2112     vmovdqa(dst, src, rscratch);
 2113   } else {
 2114     movdqa(dst, src, rscratch);
 2115   }
 2116 }
 2117 
 2118 void MacroAssembler::kmov(KRegister dst, Address src) {
 2119   if (VM_Version::supports_avx512bw()) {
 2120     kmovql(dst, src);
 2121   } else {
 2122     assert(VM_Version::supports_evex(), "");
 2123     kmovwl(dst, src);
 2124   }
 2125 }
 2126 
 2127 void MacroAssembler::kmov(Address dst, KRegister src) {
 2128   if (VM_Version::supports_avx512bw()) {
 2129     kmovql(dst, src);
 2130   } else {
 2131     assert(VM_Version::supports_evex(), "");
 2132     kmovwl(dst, src);
 2133   }
 2134 }
 2135 
 2136 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2137   if (VM_Version::supports_avx512bw()) {
 2138     kmovql(dst, src);
 2139   } else {
 2140     assert(VM_Version::supports_evex(), "");
 2141     kmovwl(dst, src);
 2142   }
 2143 }
 2144 
 2145 void MacroAssembler::kmov(Register dst, KRegister src) {
 2146   if (VM_Version::supports_avx512bw()) {
 2147     kmovql(dst, src);
 2148   } else {
 2149     assert(VM_Version::supports_evex(), "");
 2150     kmovwl(dst, src);
 2151   }
 2152 }
 2153 
 2154 void MacroAssembler::kmov(KRegister dst, Register src) {
 2155   if (VM_Version::supports_avx512bw()) {
 2156     kmovql(dst, src);
 2157   } else {
 2158     assert(VM_Version::supports_evex(), "");
 2159     kmovwl(dst, src);
 2160   }
 2161 }
 2162 
 2163 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2164   assert(rscratch != noreg || always_reachable(src), "missing");
 2165 
 2166   if (reachable(src)) {
 2167     kmovql(dst, as_Address(src));
 2168   } else {
 2169     lea(rscratch, src);
 2170     kmovql(dst, Address(rscratch, 0));
 2171   }
 2172 }
 2173 
 2174 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2175   assert(rscratch != noreg || always_reachable(src), "missing");
 2176 
 2177   if (reachable(src)) {
 2178     kmovwl(dst, as_Address(src));
 2179   } else {
 2180     lea(rscratch, src);
 2181     kmovwl(dst, Address(rscratch, 0));
 2182   }
 2183 }
 2184 
 2185 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2186                                int vector_len, Register rscratch) {
 2187   assert(rscratch != noreg || always_reachable(src), "missing");
 2188 
 2189   if (reachable(src)) {
 2190     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2191   } else {
 2192     lea(rscratch, src);
 2193     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2194   }
 2195 }
 2196 
 2197 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2198                                int vector_len, Register rscratch) {
 2199   assert(rscratch != noreg || always_reachable(src), "missing");
 2200 
 2201   if (reachable(src)) {
 2202     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2203   } else {
 2204     lea(rscratch, src);
 2205     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2206   }
 2207 }
 2208 
 2209 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2210   assert(rscratch != noreg || always_reachable(src), "missing");
 2211 
 2212   if (reachable(src)) {
 2213     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2214   } else {
 2215     lea(rscratch, src);
 2216     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2217   }
 2218 }
 2219 
 2220 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2221   assert(rscratch != noreg || always_reachable(src), "missing");
 2222 
 2223   if (reachable(src)) {
 2224     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2225   } else {
 2226     lea(rscratch, src);
 2227     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2228   }
 2229 }
 2230 
 2231 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2232   assert(rscratch != noreg || always_reachable(src), "missing");
 2233 
 2234   if (reachable(src)) {
 2235     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2236   } else {
 2237     lea(rscratch, src);
 2238     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2239   }
 2240 }
 2241 
 2242 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2243   assert(rscratch != noreg || always_reachable(src), "missing");
 2244 
 2245   if (reachable(src)) {
 2246     Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
 2247   } else {
 2248     lea(rscratch, src);
 2249     Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2250   }
 2251 }
 2252 
 2253 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2254   assert(rscratch != noreg || always_reachable(src), "missing");
 2255 
 2256   if (reachable(src)) {
 2257     Assembler::evmovdqaq(dst, as_Address(src), vector_len);
 2258   } else {
 2259     lea(rscratch, src);
 2260     Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
 2261   }
 2262 }
 2263 
 2264 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2265   assert(rscratch != noreg || always_reachable(src), "missing");
 2266 
 2267   if (reachable(src)) {
 2268     Assembler::movapd(dst, as_Address(src));
 2269   } else {
 2270     lea(rscratch, src);
 2271     Assembler::movapd(dst, Address(rscratch, 0));
 2272   }
 2273 }
 2274 
 2275 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2276   assert(rscratch != noreg || always_reachable(src), "missing");
 2277 
 2278   if (reachable(src)) {
 2279     Assembler::movdqa(dst, as_Address(src));
 2280   } else {
 2281     lea(rscratch, src);
 2282     Assembler::movdqa(dst, Address(rscratch, 0));
 2283   }
 2284 }
 2285 
 2286 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2287   assert(rscratch != noreg || always_reachable(src), "missing");
 2288 
 2289   if (reachable(src)) {
 2290     Assembler::movsd(dst, as_Address(src));
 2291   } else {
 2292     lea(rscratch, src);
 2293     Assembler::movsd(dst, Address(rscratch, 0));
 2294   }
 2295 }
 2296 
 2297 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2298   assert(rscratch != noreg || always_reachable(src), "missing");
 2299 
 2300   if (reachable(src)) {
 2301     Assembler::movss(dst, as_Address(src));
 2302   } else {
 2303     lea(rscratch, src);
 2304     Assembler::movss(dst, Address(rscratch, 0));
 2305   }
 2306 }
 2307 
 2308 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2309   assert(rscratch != noreg || always_reachable(src), "missing");
 2310 
 2311   if (reachable(src)) {
 2312     Assembler::movddup(dst, as_Address(src));
 2313   } else {
 2314     lea(rscratch, src);
 2315     Assembler::movddup(dst, Address(rscratch, 0));
 2316   }
 2317 }
 2318 
 2319 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2320   assert(rscratch != noreg || always_reachable(src), "missing");
 2321 
 2322   if (reachable(src)) {
 2323     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2324   } else {
 2325     lea(rscratch, src);
 2326     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2327   }
 2328 }
 2329 
 2330 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2331   assert(rscratch != noreg || always_reachable(src), "missing");
 2332 
 2333   if (reachable(src)) {
 2334     Assembler::mulsd(dst, as_Address(src));
 2335   } else {
 2336     lea(rscratch, src);
 2337     Assembler::mulsd(dst, Address(rscratch, 0));
 2338   }
 2339 }
 2340 
 2341 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2342   assert(rscratch != noreg || always_reachable(src), "missing");
 2343 
 2344   if (reachable(src)) {
 2345     Assembler::mulss(dst, as_Address(src));
 2346   } else {
 2347     lea(rscratch, src);
 2348     Assembler::mulss(dst, Address(rscratch, 0));
 2349   }
 2350 }
 2351 
 2352 void MacroAssembler::null_check(Register reg, int offset) {
 2353   if (needs_explicit_null_check(offset)) {
 2354     // provoke OS null exception if reg is null by
 2355     // accessing M[reg] w/o changing any (non-CC) registers
 2356     // NOTE: cmpl is plenty here to provoke a segv
 2357     cmpptr(rax, Address(reg, 0));
 2358     // Note: should probably use testl(rax, Address(reg, 0));
 2359     //       may be shorter code (however, this version of
 2360     //       testl needs to be implemented first)
 2361   } else {
 2362     // nothing to do, (later) access of M[reg + offset]
 2363     // will provoke OS null exception if reg is null
 2364   }
 2365 }
 2366 











































































































 2367 void MacroAssembler::os_breakpoint() {
 2368   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2369   // (e.g., MSVC can't call ps() otherwise)
 2370   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2371 }
 2372 
 2373 void MacroAssembler::unimplemented(const char* what) {
 2374   const char* buf = nullptr;
 2375   {
 2376     ResourceMark rm;
 2377     stringStream ss;
 2378     ss.print("unimplemented: %s", what);
 2379     buf = code_string(ss.as_string());
 2380   }
 2381   stop(buf);
 2382 }
 2383 
 2384 #define XSTATE_BV 0x200
 2385 
 2386 void MacroAssembler::pop_CPU_state() {
 2387   pop_FPU_state();
 2388   pop_IU_state();
 2389 }
 2390 
 2391 void MacroAssembler::pop_FPU_state() {
 2392   fxrstor(Address(rsp, 0));
 2393   addptr(rsp, FPUStateSizeInWords * wordSize);
 2394 }
 2395 
 2396 void MacroAssembler::pop_IU_state() {
 2397   popa();
 2398   addq(rsp, 8);
 2399   popf();
 2400 }
 2401 
 2402 // Save Integer and Float state
 2403 // Warning: Stack must be 16 byte aligned (64bit)
 2404 void MacroAssembler::push_CPU_state() {
 2405   push_IU_state();
 2406   push_FPU_state();
 2407 }
 2408 
 2409 void MacroAssembler::push_FPU_state() {
 2410   subptr(rsp, FPUStateSizeInWords * wordSize);
 2411   fxsave(Address(rsp, 0));
 2412 }
 2413 
 2414 void MacroAssembler::push_IU_state() {
 2415   // Push flags first because pusha kills them
 2416   pushf();
 2417   // Make sure rsp stays 16-byte aligned
 2418   subq(rsp, 8);
 2419   pusha();
 2420 }
 2421 
 2422 void MacroAssembler::push_cont_fastpath() {
 2423   if (!Continuations::enabled()) return;
 2424 
 2425   Label L_done;
 2426   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2427   jccb(Assembler::belowEqual, L_done);
 2428   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2429   bind(L_done);
 2430 }
 2431 
 2432 void MacroAssembler::pop_cont_fastpath() {
 2433   if (!Continuations::enabled()) return;
 2434 
 2435   Label L_done;
 2436   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2437   jccb(Assembler::below, L_done);
 2438   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2439   bind(L_done);
 2440 }
 2441 
 2442 #ifdef ASSERT
 2443 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2444   Label no_cont;
 2445   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2446   testl(cont, cont);
 2447   jcc(Assembler::zero, no_cont);
 2448   stop(name);
 2449   bind(no_cont);
 2450 }
 2451 #endif
 2452 
 2453 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2454   // we must set sp to zero to clear frame
 2455   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2456   // must clear fp, so that compiled frames are not confused; it is
 2457   // possible that we need it only for debugging
 2458   if (clear_fp) {
 2459     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2460   }
 2461   // Always clear the pc because it could have been set by make_walkable()
 2462   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2463   vzeroupper();
 2464 }
 2465 
 2466 void MacroAssembler::round_to(Register reg, int modulus) {
 2467   addptr(reg, modulus - 1);
 2468   andptr(reg, -modulus);
 2469 }
 2470 
 2471 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2472   if (at_return) {
 2473     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2474     // we may safely use rsp instead to perform the stack watermark check.
 2475     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2476     jcc(Assembler::above, slow_path);
 2477     return;
 2478   }
 2479   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2480   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2481 }
 2482 
 2483 // Calls to C land
 2484 //
 2485 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2486 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2487 // has to be reset to 0. This is required to allow proper stack traversal.
 2488 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2489                                          Register last_java_fp,
 2490                                          address  last_java_pc,
 2491                                          Register rscratch) {
 2492   vzeroupper();
 2493   // determine last_java_sp register
 2494   if (!last_java_sp->is_valid()) {
 2495     last_java_sp = rsp;
 2496   }
 2497   // last_java_fp is optional
 2498   if (last_java_fp->is_valid()) {
 2499     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2500   }
 2501   // last_java_pc is optional
 2502   if (last_java_pc != nullptr) {
 2503     Address java_pc(r15_thread,
 2504                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2505     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2506   }
 2507   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2508 }
 2509 
 2510 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2511                                          Register last_java_fp,
 2512                                          Label &L,
 2513                                          Register scratch) {
 2514   lea(scratch, L);
 2515   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2516   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2517 }
 2518 
 2519 void MacroAssembler::shlptr(Register dst, int imm8) {
 2520   shlq(dst, imm8);
 2521 }
 2522 
 2523 void MacroAssembler::shrptr(Register dst, int imm8) {
 2524   shrq(dst, imm8);
 2525 }
 2526 
 2527 void MacroAssembler::sign_extend_byte(Register reg) {
 2528   movsbl(reg, reg); // movsxb
 2529 }
 2530 
 2531 void MacroAssembler::sign_extend_short(Register reg) {
 2532   movswl(reg, reg); // movsxw
 2533 }
 2534 
 2535 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2536   if (imm32 >= 0 && is8bit(imm32)) {
 2537     testb(dst, imm32);
 2538   } else {
 2539     Assembler::testl(dst, imm32);
 2540   }
 2541 }
 2542 
 2543 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2544   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2545     testb(dst, imm32);
 2546   } else {
 2547     Assembler::testl(dst, imm32);
 2548   }
 2549 }
 2550 
 2551 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2552   assert(always_reachable(src), "Address should be reachable");
 2553   testl(dst, as_Address(src));
 2554 }
 2555 
 2556 void MacroAssembler::testq(Address 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::testq(Register dst, int32_t imm32) {
 2565   if (imm32 >= 0) {
 2566     testl(dst, imm32);
 2567   } else {
 2568     Assembler::testq(dst, imm32);
 2569   }
 2570 }
 2571 
 2572 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2573   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2574   Assembler::pcmpeqb(dst, src);
 2575 }
 2576 
 2577 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2578   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2579   Assembler::pcmpeqw(dst, src);
 2580 }
 2581 
 2582 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2583   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2584   Assembler::pcmpestri(dst, src, imm8);
 2585 }
 2586 
 2587 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2588   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2589   Assembler::pcmpestri(dst, src, imm8);
 2590 }
 2591 
 2592 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2593   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2594   Assembler::pmovzxbw(dst, src);
 2595 }
 2596 
 2597 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2598   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2599   Assembler::pmovzxbw(dst, src);
 2600 }
 2601 
 2602 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2603   assert((src->encoding() < 16),"XMM register should be 0-15");
 2604   Assembler::pmovmskb(dst, src);
 2605 }
 2606 
 2607 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2608   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2609   Assembler::ptest(dst, src);
 2610 }
 2611 
 2612 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2613   assert(rscratch != noreg || always_reachable(src), "missing");
 2614 
 2615   if (reachable(src)) {
 2616     Assembler::sqrtss(dst, as_Address(src));
 2617   } else {
 2618     lea(rscratch, src);
 2619     Assembler::sqrtss(dst, Address(rscratch, 0));
 2620   }
 2621 }
 2622 
 2623 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2624   assert(rscratch != noreg || always_reachable(src), "missing");
 2625 
 2626   if (reachable(src)) {
 2627     Assembler::subsd(dst, as_Address(src));
 2628   } else {
 2629     lea(rscratch, src);
 2630     Assembler::subsd(dst, Address(rscratch, 0));
 2631   }
 2632 }
 2633 
 2634 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2635   assert(rscratch != noreg || always_reachable(src), "missing");
 2636 
 2637   if (reachable(src)) {
 2638     Assembler::roundsd(dst, as_Address(src), rmode);
 2639   } else {
 2640     lea(rscratch, src);
 2641     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2642   }
 2643 }
 2644 
 2645 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2646   assert(rscratch != noreg || always_reachable(src), "missing");
 2647 
 2648   if (reachable(src)) {
 2649     Assembler::subss(dst, as_Address(src));
 2650   } else {
 2651     lea(rscratch, src);
 2652     Assembler::subss(dst, Address(rscratch, 0));
 2653   }
 2654 }
 2655 
 2656 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2657   assert(rscratch != noreg || always_reachable(src), "missing");
 2658 
 2659   if (reachable(src)) {
 2660     Assembler::ucomisd(dst, as_Address(src));
 2661   } else {
 2662     lea(rscratch, src);
 2663     Assembler::ucomisd(dst, Address(rscratch, 0));
 2664   }
 2665 }
 2666 
 2667 void MacroAssembler::vucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2668   assert(rscratch != noreg || always_reachable(src), "missing");
 2669 
 2670   if (reachable(src)) {
 2671     Assembler::vucomxsd(dst, as_Address(src));
 2672   } else {
 2673     lea(rscratch, src);
 2674     Assembler::vucomxsd(dst, Address(rscratch, 0));
 2675   }
 2676 }
 2677 
 2678 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2679   assert(rscratch != noreg || always_reachable(src), "missing");
 2680 
 2681   if (reachable(src)) {
 2682     Assembler::ucomiss(dst, as_Address(src));
 2683   } else {
 2684     lea(rscratch, src);
 2685     Assembler::ucomiss(dst, Address(rscratch, 0));
 2686   }
 2687 }
 2688 
 2689 void MacroAssembler::vucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2690   assert(rscratch != noreg || always_reachable(src), "missing");
 2691 
 2692   if (reachable(src)) {
 2693     Assembler::vucomxss(dst, as_Address(src));
 2694   } else {
 2695     lea(rscratch, src);
 2696     Assembler::vucomxss(dst, Address(rscratch, 0));
 2697   }
 2698 }
 2699 
 2700 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2701   assert(rscratch != noreg || always_reachable(src), "missing");
 2702 
 2703   // Used in sign-bit flipping with aligned address.
 2704   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2705 
 2706   if (UseAVX > 2 &&
 2707       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2708       (dst->encoding() >= 16)) {
 2709     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2710   } else if (reachable(src)) {
 2711     Assembler::xorpd(dst, as_Address(src));
 2712   } else {
 2713     lea(rscratch, src);
 2714     Assembler::xorpd(dst, Address(rscratch, 0));
 2715   }
 2716 }
 2717 
 2718 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2719   if (UseAVX > 2 &&
 2720       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2721       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2722     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2723   } else {
 2724     Assembler::xorpd(dst, src);
 2725   }
 2726 }
 2727 
 2728 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2729   if (UseAVX > 2 &&
 2730       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2731       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2732     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2733   } else {
 2734     Assembler::xorps(dst, src);
 2735   }
 2736 }
 2737 
 2738 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2739   assert(rscratch != noreg || always_reachable(src), "missing");
 2740 
 2741   // Used in sign-bit flipping with aligned address.
 2742   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2743 
 2744   if (UseAVX > 2 &&
 2745       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2746       (dst->encoding() >= 16)) {
 2747     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2748   } else if (reachable(src)) {
 2749     Assembler::xorps(dst, as_Address(src));
 2750   } else {
 2751     lea(rscratch, src);
 2752     Assembler::xorps(dst, Address(rscratch, 0));
 2753   }
 2754 }
 2755 
 2756 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2757   assert(rscratch != noreg || always_reachable(src), "missing");
 2758 
 2759   // Used in sign-bit flipping with aligned address.
 2760   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2761   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2762   if (reachable(src)) {
 2763     Assembler::pshufb(dst, as_Address(src));
 2764   } else {
 2765     lea(rscratch, src);
 2766     Assembler::pshufb(dst, Address(rscratch, 0));
 2767   }
 2768 }
 2769 
 2770 // AVX 3-operands instructions
 2771 
 2772 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2773   assert(rscratch != noreg || always_reachable(src), "missing");
 2774 
 2775   if (reachable(src)) {
 2776     vaddsd(dst, nds, as_Address(src));
 2777   } else {
 2778     lea(rscratch, src);
 2779     vaddsd(dst, nds, Address(rscratch, 0));
 2780   }
 2781 }
 2782 
 2783 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2784   assert(rscratch != noreg || always_reachable(src), "missing");
 2785 
 2786   if (reachable(src)) {
 2787     vaddss(dst, nds, as_Address(src));
 2788   } else {
 2789     lea(rscratch, src);
 2790     vaddss(dst, nds, Address(rscratch, 0));
 2791   }
 2792 }
 2793 
 2794 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2795   assert(UseAVX > 0, "requires some form of AVX");
 2796   assert(rscratch != noreg || always_reachable(src), "missing");
 2797 
 2798   if (reachable(src)) {
 2799     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2800   } else {
 2801     lea(rscratch, src);
 2802     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2803   }
 2804 }
 2805 
 2806 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2807   assert(UseAVX > 0, "requires some form of AVX");
 2808   assert(rscratch != noreg || always_reachable(src), "missing");
 2809 
 2810   if (reachable(src)) {
 2811     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2812   } else {
 2813     lea(rscratch, src);
 2814     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2815   }
 2816 }
 2817 
 2818 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2819   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2820   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2821 
 2822   vandps(dst, nds, negate_field, vector_len, rscratch);
 2823 }
 2824 
 2825 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2826   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2827   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2828 
 2829   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2830 }
 2831 
 2832 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2833   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2834   Assembler::vpaddb(dst, nds, src, vector_len);
 2835 }
 2836 
 2837 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2838   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2839   Assembler::vpaddb(dst, nds, src, vector_len);
 2840 }
 2841 
 2842 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2843   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2844   Assembler::vpaddw(dst, nds, src, vector_len);
 2845 }
 2846 
 2847 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2848   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2849   Assembler::vpaddw(dst, nds, src, vector_len);
 2850 }
 2851 
 2852 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2853   assert(rscratch != noreg || always_reachable(src), "missing");
 2854 
 2855   if (reachable(src)) {
 2856     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 2857   } else {
 2858     lea(rscratch, src);
 2859     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 2860   }
 2861 }
 2862 
 2863 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2864   assert(rscratch != noreg || always_reachable(src), "missing");
 2865 
 2866   if (reachable(src)) {
 2867     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 2868   } else {
 2869     lea(rscratch, src);
 2870     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 2871   }
 2872 }
 2873 
 2874 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2875   assert(rscratch != noreg || always_reachable(src), "missing");
 2876 
 2877   if (reachable(src)) {
 2878     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 2879   } else {
 2880     lea(rscratch, src);
 2881     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 2882   }
 2883 }
 2884 
 2885 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2886   assert(rscratch != noreg || always_reachable(src), "missing");
 2887 
 2888   if (reachable(src)) {
 2889     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 2890   } else {
 2891     lea(rscratch, src);
 2892     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 2893   }
 2894 }
 2895 
 2896 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2897   assert(rscratch != noreg || always_reachable(src), "missing");
 2898 
 2899   if (reachable(src)) {
 2900     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 2901   } else {
 2902     lea(rscratch, src);
 2903     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 2904   }
 2905 }
 2906 
 2907 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2908   assert(rscratch != noreg || always_reachable(src), "missing");
 2909 
 2910   if (reachable(src)) {
 2911     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 2912   } else {
 2913     lea(rscratch, src);
 2914     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 2915   }
 2916 }
 2917 
 2918 // Vector float blend
 2919 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2920 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2921   // WARN: Allow dst == (src1|src2), mask == scratch
 2922   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2923                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2924   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 2925   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2926   if (blend_emulation && scratch_available && dst_available) {
 2927     if (compute_mask) {
 2928       vpsrad(scratch, mask, 32, vector_len);
 2929       mask = scratch;
 2930     }
 2931     if (dst == src1) {
 2932       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 2933       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 2934     } else {
 2935       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 2936       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 2937     }
 2938     vpor(dst, dst, scratch, vector_len);
 2939   } else {
 2940     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 2941   }
 2942 }
 2943 
 2944 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2945 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2946   // WARN: Allow dst == (src1|src2), mask == scratch
 2947   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2948                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2949   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 2950   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2951   if (blend_emulation && scratch_available && dst_available) {
 2952     if (compute_mask) {
 2953       vpxor(scratch, scratch, scratch, vector_len);
 2954       vpcmpgtq(scratch, scratch, mask, vector_len);
 2955       mask = scratch;
 2956     }
 2957     if (dst == src1) {
 2958       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 2959       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 2960     } else {
 2961       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 2962       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 2963     }
 2964     vpor(dst, dst, scratch, vector_len);
 2965   } else {
 2966     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 2967   }
 2968 }
 2969 
 2970 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2971   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2972   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 2973 }
 2974 
 2975 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 2976   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2977   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 2978 }
 2979 
 2980 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2981   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2982   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 2983 }
 2984 
 2985 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2986   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2987   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 2988 }
 2989 
 2990 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2991   assert(rscratch != noreg || always_reachable(src), "missing");
 2992 
 2993   if (reachable(src)) {
 2994     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 2995   } else {
 2996     lea(rscratch, src);
 2997     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 2998   }
 2999 }
 3000 
 3001 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3002                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3003   assert(rscratch != noreg || always_reachable(src), "missing");
 3004 
 3005   if (reachable(src)) {
 3006     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3007   } else {
 3008     lea(rscratch, src);
 3009     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3010   }
 3011 }
 3012 
 3013 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3014                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3015   assert(rscratch != noreg || always_reachable(src), "missing");
 3016 
 3017   if (reachable(src)) {
 3018     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3019   } else {
 3020     lea(rscratch, src);
 3021     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3022   }
 3023 }
 3024 
 3025 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3026                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3027   assert(rscratch != noreg || always_reachable(src), "missing");
 3028 
 3029   if (reachable(src)) {
 3030     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3031   } else {
 3032     lea(rscratch, src);
 3033     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3034   }
 3035 }
 3036 
 3037 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3038                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3039   assert(rscratch != noreg || always_reachable(src), "missing");
 3040 
 3041   if (reachable(src)) {
 3042     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3043   } else {
 3044     lea(rscratch, src);
 3045     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3046   }
 3047 }
 3048 
 3049 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3050   if (width == Assembler::Q) {
 3051     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3052   } else {
 3053     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3054   }
 3055 }
 3056 
 3057 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3058   int eq_cond_enc = 0x29;
 3059   int gt_cond_enc = 0x37;
 3060   if (width != Assembler::Q) {
 3061     eq_cond_enc = 0x74 + width;
 3062     gt_cond_enc = 0x64 + width;
 3063   }
 3064   switch (cond) {
 3065   case eq:
 3066     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3067     break;
 3068   case neq:
 3069     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3070     vallones(xtmp, vector_len);
 3071     vpxor(dst, xtmp, dst, vector_len);
 3072     break;
 3073   case le:
 3074     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3075     vallones(xtmp, vector_len);
 3076     vpxor(dst, xtmp, dst, vector_len);
 3077     break;
 3078   case nlt:
 3079     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3080     vallones(xtmp, vector_len);
 3081     vpxor(dst, xtmp, dst, vector_len);
 3082     break;
 3083   case lt:
 3084     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3085     break;
 3086   case nle:
 3087     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3088     break;
 3089   default:
 3090     assert(false, "Should not reach here");
 3091   }
 3092 }
 3093 
 3094 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3095   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3096   Assembler::vpmovzxbw(dst, src, vector_len);
 3097 }
 3098 
 3099 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3100   assert((src->encoding() < 16),"XMM register should be 0-15");
 3101   Assembler::vpmovmskb(dst, src, vector_len);
 3102 }
 3103 
 3104 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3105   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3106   Assembler::vpmullw(dst, nds, src, vector_len);
 3107 }
 3108 
 3109 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3110   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3111   Assembler::vpmullw(dst, nds, src, vector_len);
 3112 }
 3113 
 3114 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3115   assert((UseAVX > 0), "AVX support is needed");
 3116   assert(rscratch != noreg || always_reachable(src), "missing");
 3117 
 3118   if (reachable(src)) {
 3119     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3120   } else {
 3121     lea(rscratch, src);
 3122     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3123   }
 3124 }
 3125 
 3126 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3127   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3128   Assembler::vpsubb(dst, nds, src, vector_len);
 3129 }
 3130 
 3131 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3132   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3133   Assembler::vpsubb(dst, nds, src, vector_len);
 3134 }
 3135 
 3136 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3137   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3138   Assembler::vpsubw(dst, nds, src, vector_len);
 3139 }
 3140 
 3141 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3142   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3143   Assembler::vpsubw(dst, nds, src, vector_len);
 3144 }
 3145 
 3146 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3147   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3148   Assembler::vpsraw(dst, nds, shift, vector_len);
 3149 }
 3150 
 3151 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3152   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3153   Assembler::vpsraw(dst, nds, shift, vector_len);
 3154 }
 3155 
 3156 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3157   assert(UseAVX > 2,"");
 3158   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3159      vector_len = 2;
 3160   }
 3161   Assembler::evpsraq(dst, nds, shift, vector_len);
 3162 }
 3163 
 3164 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3165   assert(UseAVX > 2,"");
 3166   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3167      vector_len = 2;
 3168   }
 3169   Assembler::evpsraq(dst, nds, shift, vector_len);
 3170 }
 3171 
 3172 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3173   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3174   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3175 }
 3176 
 3177 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3178   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3179   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3180 }
 3181 
 3182 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3183   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3184   Assembler::vpsllw(dst, nds, shift, vector_len);
 3185 }
 3186 
 3187 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3188   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3189   Assembler::vpsllw(dst, nds, shift, vector_len);
 3190 }
 3191 
 3192 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3193   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3194   Assembler::vptest(dst, src);
 3195 }
 3196 
 3197 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3198   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3199   Assembler::punpcklbw(dst, src);
 3200 }
 3201 
 3202 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3203   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3204   Assembler::pshufd(dst, src, mode);
 3205 }
 3206 
 3207 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3208   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3209   Assembler::pshuflw(dst, src, mode);
 3210 }
 3211 
 3212 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3213   assert(rscratch != noreg || always_reachable(src), "missing");
 3214 
 3215   if (reachable(src)) {
 3216     vandpd(dst, nds, as_Address(src), vector_len);
 3217   } else {
 3218     lea(rscratch, src);
 3219     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3220   }
 3221 }
 3222 
 3223 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3224   assert(rscratch != noreg || always_reachable(src), "missing");
 3225 
 3226   if (reachable(src)) {
 3227     vandps(dst, nds, as_Address(src), vector_len);
 3228   } else {
 3229     lea(rscratch, src);
 3230     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3231   }
 3232 }
 3233 
 3234 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3235                             bool merge, int vector_len, Register rscratch) {
 3236   assert(rscratch != noreg || always_reachable(src), "missing");
 3237 
 3238   if (reachable(src)) {
 3239     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3240   } else {
 3241     lea(rscratch, src);
 3242     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3243   }
 3244 }
 3245 
 3246 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3247   assert(rscratch != noreg || always_reachable(src), "missing");
 3248 
 3249   if (reachable(src)) {
 3250     vdivsd(dst, nds, as_Address(src));
 3251   } else {
 3252     lea(rscratch, src);
 3253     vdivsd(dst, nds, Address(rscratch, 0));
 3254   }
 3255 }
 3256 
 3257 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3258   assert(rscratch != noreg || always_reachable(src), "missing");
 3259 
 3260   if (reachable(src)) {
 3261     vdivss(dst, nds, as_Address(src));
 3262   } else {
 3263     lea(rscratch, src);
 3264     vdivss(dst, nds, Address(rscratch, 0));
 3265   }
 3266 }
 3267 
 3268 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3269   assert(rscratch != noreg || always_reachable(src), "missing");
 3270 
 3271   if (reachable(src)) {
 3272     vmulsd(dst, nds, as_Address(src));
 3273   } else {
 3274     lea(rscratch, src);
 3275     vmulsd(dst, nds, Address(rscratch, 0));
 3276   }
 3277 }
 3278 
 3279 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3280   assert(rscratch != noreg || always_reachable(src), "missing");
 3281 
 3282   if (reachable(src)) {
 3283     vmulss(dst, nds, as_Address(src));
 3284   } else {
 3285     lea(rscratch, src);
 3286     vmulss(dst, nds, Address(rscratch, 0));
 3287   }
 3288 }
 3289 
 3290 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3291   assert(rscratch != noreg || always_reachable(src), "missing");
 3292 
 3293   if (reachable(src)) {
 3294     vsubsd(dst, nds, as_Address(src));
 3295   } else {
 3296     lea(rscratch, src);
 3297     vsubsd(dst, nds, Address(rscratch, 0));
 3298   }
 3299 }
 3300 
 3301 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3302   assert(rscratch != noreg || always_reachable(src), "missing");
 3303 
 3304   if (reachable(src)) {
 3305     vsubss(dst, nds, as_Address(src));
 3306   } else {
 3307     lea(rscratch, src);
 3308     vsubss(dst, nds, Address(rscratch, 0));
 3309   }
 3310 }
 3311 
 3312 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3313   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3314   assert(rscratch != noreg || always_reachable(src), "missing");
 3315 
 3316   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3317 }
 3318 
 3319 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3320   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3321   assert(rscratch != noreg || always_reachable(src), "missing");
 3322 
 3323   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3324 }
 3325 
 3326 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3327   assert(rscratch != noreg || always_reachable(src), "missing");
 3328 
 3329   if (reachable(src)) {
 3330     vxorpd(dst, nds, as_Address(src), vector_len);
 3331   } else {
 3332     lea(rscratch, src);
 3333     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3334   }
 3335 }
 3336 
 3337 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3338   assert(rscratch != noreg || always_reachable(src), "missing");
 3339 
 3340   if (reachable(src)) {
 3341     vxorps(dst, nds, as_Address(src), vector_len);
 3342   } else {
 3343     lea(rscratch, src);
 3344     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3345   }
 3346 }
 3347 
 3348 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3349   assert(rscratch != noreg || always_reachable(src), "missing");
 3350 
 3351   if (UseAVX > 1 || (vector_len < 1)) {
 3352     if (reachable(src)) {
 3353       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3354     } else {
 3355       lea(rscratch, src);
 3356       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3357     }
 3358   } else {
 3359     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3360   }
 3361 }
 3362 
 3363 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3364   assert(rscratch != noreg || always_reachable(src), "missing");
 3365 
 3366   if (reachable(src)) {
 3367     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3368   } else {
 3369     lea(rscratch, src);
 3370     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3371   }
 3372 }
 3373 
 3374 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3375   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3376   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3377   // The inverted mask is sign-extended
 3378   andptr(possibly_non_local, inverted_mask);
 3379 }
 3380 
 3381 void MacroAssembler::resolve_jobject(Register value,
 3382                                      Register tmp) {
 3383   Register thread = r15_thread;
 3384   assert_different_registers(value, thread, tmp);
 3385   Label done, tagged, weak_tagged;
 3386   testptr(value, value);
 3387   jcc(Assembler::zero, done);           // Use null as-is.
 3388   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3389   jcc(Assembler::notZero, tagged);
 3390 
 3391   // Resolve local handle
 3392   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3393   verify_oop(value);
 3394   jmp(done);
 3395 
 3396   bind(tagged);
 3397   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3398   jcc(Assembler::notZero, weak_tagged);
 3399 
 3400   // Resolve global handle
 3401   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3402   verify_oop(value);
 3403   jmp(done);
 3404 
 3405   bind(weak_tagged);
 3406   // Resolve jweak.
 3407   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3408                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3409   verify_oop(value);
 3410 
 3411   bind(done);
 3412 }
 3413 
 3414 void MacroAssembler::resolve_global_jobject(Register value,
 3415                                             Register tmp) {
 3416   Register thread = r15_thread;
 3417   assert_different_registers(value, thread, tmp);
 3418   Label done;
 3419 
 3420   testptr(value, value);
 3421   jcc(Assembler::zero, done);           // Use null as-is.
 3422 
 3423 #ifdef ASSERT
 3424   {
 3425     Label valid_global_tag;
 3426     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3427     jcc(Assembler::notZero, valid_global_tag);
 3428     stop("non global jobject using resolve_global_jobject");
 3429     bind(valid_global_tag);
 3430   }
 3431 #endif
 3432 
 3433   // Resolve global handle
 3434   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3435   verify_oop(value);
 3436 
 3437   bind(done);
 3438 }
 3439 
 3440 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3441   subq(dst, imm32);
 3442 }
 3443 
 3444 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3445 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3446   subq_imm32(dst, imm32);
 3447 }
 3448 
 3449 void MacroAssembler::subptr(Register dst, Register src) {
 3450   subq(dst, src);
 3451 }
 3452 
 3453 // C++ bool manipulation
 3454 void MacroAssembler::testbool(Register dst) {
 3455   if(sizeof(bool) == 1)
 3456     testb(dst, 0xff);
 3457   else if(sizeof(bool) == 2) {
 3458     // testw implementation needed for two byte bools
 3459     ShouldNotReachHere();
 3460   } else if(sizeof(bool) == 4)
 3461     testl(dst, dst);
 3462   else
 3463     // unsupported
 3464     ShouldNotReachHere();
 3465 }
 3466 
 3467 void MacroAssembler::testptr(Register dst, Register src) {
 3468   testq(dst, src);
 3469 }
 3470 






















































































































 3471 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3472 void MacroAssembler::tlab_allocate(Register obj,
 3473                                    Register var_size_in_bytes,
 3474                                    int con_size_in_bytes,
 3475                                    Register t1,
 3476                                    Register t2,
 3477                                    Label& slow_case) {
 3478   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3479   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3480 }
 3481 
 3482 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3483   RegSet regs;
 3484   regs += RegSet::of(rax, rcx, rdx);
 3485 #ifndef _WINDOWS
 3486   regs += RegSet::of(rsi, rdi);
 3487 #endif
 3488   regs += RegSet::range(r8, r11);
 3489   if (UseAPX) {
 3490     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3491   }
 3492   return regs;
 3493 }
 3494 
 3495 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3496   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3497 #if defined(_WINDOWS)
 3498   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3499   if (num_xmm_registers > 16) {
 3500      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3501   }
 3502   return result;
 3503 #else
 3504   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3505 #endif
 3506 }
 3507 
 3508 // C1 only ever uses the first double/float of the XMM register.
 3509 static int xmm_save_size() { return sizeof(double); }
 3510 
 3511 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3512   masm->movdbl(Address(rsp, offset), reg);
 3513 }
 3514 
 3515 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3516   masm->movdbl(reg, Address(rsp, offset));
 3517 }
 3518 
 3519 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3520                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3521 
 3522   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3523                          StackAlignmentInBytes);
 3524   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3525 
 3526   return gp_area_size + xmm_area_size;
 3527 }
 3528 
 3529 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3530   block_comment("push_call_clobbered_registers start");
 3531   // Regular registers
 3532   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3533 
 3534   int gp_area_size;
 3535   int xmm_area_size;
 3536   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3537                                                gp_area_size, xmm_area_size);
 3538   subptr(rsp, total_save_size);
 3539 
 3540   push_set(gp_registers_to_push, 0);
 3541 
 3542   if (save_fpu) {
 3543     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3544   }
 3545 
 3546   block_comment("push_call_clobbered_registers end");
 3547 }
 3548 
 3549 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3550   block_comment("pop_call_clobbered_registers start");
 3551 
 3552   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 3553 
 3554   int gp_area_size;
 3555   int xmm_area_size;
 3556   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 3557                                                gp_area_size, xmm_area_size);
 3558 
 3559   if (restore_fpu) {
 3560     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3561   }
 3562 
 3563   pop_set(gp_registers_to_pop, 0);
 3564 
 3565   addptr(rsp, total_save_size);
 3566 
 3567   vzeroupper();
 3568 
 3569   block_comment("pop_call_clobbered_registers end");
 3570 }
 3571 
 3572 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3573   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3574   int spill_offset = offset;
 3575 
 3576   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3577     save_xmm_register(this, spill_offset, *it);
 3578     spill_offset += xmm_save_size();
 3579   }
 3580 }
 3581 
 3582 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3583   int restore_size = set.size() * xmm_save_size();
 3584   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3585 
 3586   int restore_offset = offset + restore_size - xmm_save_size();
 3587 
 3588   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3589     restore_xmm_register(this, restore_offset, *it);
 3590     restore_offset -= xmm_save_size();
 3591   }
 3592 }
 3593 
 3594 void MacroAssembler::push_set(RegSet set, int offset) {
 3595   int spill_offset;
 3596   if (offset == -1) {
 3597     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3598     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3599     subptr(rsp, aligned_size);
 3600     spill_offset = 0;
 3601   } else {
 3602     spill_offset = offset;
 3603   }
 3604 
 3605   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3606     movptr(Address(rsp, spill_offset), *it);
 3607     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3608   }
 3609 }
 3610 
 3611 void MacroAssembler::pop_set(RegSet set, int offset) {
 3612 
 3613   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3614   int restore_size = set.size() * gp_reg_size;
 3615   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3616 
 3617   int restore_offset;
 3618   if (offset == -1) {
 3619     restore_offset = restore_size - gp_reg_size;
 3620   } else {
 3621     restore_offset = offset + restore_size - gp_reg_size;
 3622   }
 3623   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3624     movptr(*it, Address(rsp, restore_offset));
 3625     restore_offset -= gp_reg_size;
 3626   }
 3627 
 3628   if (offset == -1) {
 3629     addptr(rsp, aligned_size);
 3630   }
 3631 }
 3632 
 3633 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3634 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3635   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3636   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3637   Label done;
 3638 
 3639   testptr(length_in_bytes, length_in_bytes);
 3640   jcc(Assembler::zero, done);
 3641 
 3642   // initialize topmost word, divide index by 2, check if odd and test if zero
 3643   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3644 #ifdef ASSERT
 3645   {
 3646     Label L;
 3647     testptr(length_in_bytes, BytesPerWord - 1);
 3648     jcc(Assembler::zero, L);
 3649     stop("length must be a multiple of BytesPerWord");
 3650     bind(L);
 3651   }
 3652 #endif
 3653   Register index = length_in_bytes;
 3654   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3655   if (UseIncDec) {
 3656     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3657   } else {
 3658     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3659     shrptr(index, 1);
 3660   }
 3661 
 3662   // initialize remaining object fields: index is a multiple of 2 now
 3663   {
 3664     Label loop;
 3665     bind(loop);
 3666     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3667     decrement(index);
 3668     jcc(Assembler::notZero, loop);
 3669   }
 3670 
 3671   bind(done);
 3672 }
 3673 






















 3674 // Look up the method for a megamorphic invokeinterface call.
 3675 // The target method is determined by <intf_klass, itable_index>.
 3676 // The receiver klass is in recv_klass.
 3677 // On success, the result will be in method_result, and execution falls through.
 3678 // On failure, execution transfers to the given label.
 3679 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3680                                              Register intf_klass,
 3681                                              RegisterOrConstant itable_index,
 3682                                              Register method_result,
 3683                                              Register scan_temp,
 3684                                              Label& L_no_such_interface,
 3685                                              bool return_method) {
 3686   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3687   assert_different_registers(method_result, intf_klass, scan_temp);
 3688   assert(recv_klass != method_result || !return_method,
 3689          "recv_klass can be destroyed when method isn't needed");
 3690 
 3691   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3692          "caller must use same register for non-constant itable index as for method");
 3693 
 3694   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3695   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3696   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3697   int scan_step   = itableOffsetEntry::size() * wordSize;
 3698   int vte_size    = vtableEntry::size_in_bytes();
 3699   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3700   assert(vte_size == wordSize, "else adjust times_vte_scale");
 3701 
 3702   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3703 
 3704   // Could store the aligned, prescaled offset in the klass.
 3705   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 3706 
 3707   if (return_method) {
 3708     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 3709     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3710     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 3711   }
 3712 
 3713   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 3714   //   if (scan->interface() == intf) {
 3715   //     result = (klass + scan->offset() + itable_index);
 3716   //   }
 3717   // }
 3718   Label search, found_method;
 3719 
 3720   for (int peel = 1; peel >= 0; peel--) {
 3721     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 3722     cmpptr(intf_klass, method_result);
 3723 
 3724     if (peel) {
 3725       jccb(Assembler::equal, found_method);
 3726     } else {
 3727       jccb(Assembler::notEqual, search);
 3728       // (invert the test to fall through to found_method...)
 3729     }
 3730 
 3731     if (!peel)  break;
 3732 
 3733     bind(search);
 3734 
 3735     // Check that the previous entry is non-null.  A null entry means that
 3736     // the receiver class doesn't implement the interface, and wasn't the
 3737     // same as when the caller was compiled.
 3738     testptr(method_result, method_result);
 3739     jcc(Assembler::zero, L_no_such_interface);
 3740     addptr(scan_temp, scan_step);
 3741   }
 3742 
 3743   bind(found_method);
 3744 
 3745   if (return_method) {
 3746     // Got a hit.
 3747     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 3748     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 3749   }
 3750 }
 3751 
 3752 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 3753 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 3754 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 3755 // The target method is determined by <holder_klass, itable_index>.
 3756 // The receiver klass is in recv_klass.
 3757 // On success, the result will be in method_result, and execution falls through.
 3758 // On failure, execution transfers to the given label.
 3759 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 3760                                                   Register holder_klass,
 3761                                                   Register resolved_klass,
 3762                                                   Register method_result,
 3763                                                   Register scan_temp,
 3764                                                   Register temp_reg2,
 3765                                                   Register receiver,
 3766                                                   int itable_index,
 3767                                                   Label& L_no_such_interface) {
 3768   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 3769   Register temp_itbl_klass = method_result;
 3770   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 3771 
 3772   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3773   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3774   int scan_step = itableOffsetEntry::size() * wordSize;
 3775   int vte_size = vtableEntry::size_in_bytes();
 3776   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 3777   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 3778   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3779   assert(vte_size == wordSize, "adjust times_vte_scale");
 3780 
 3781   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 3782 
 3783   // temp_itbl_klass = recv_klass.itable[0]
 3784   // scan_temp = &recv_klass.itable[0] + step
 3785   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3786   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 3787   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 3788   xorptr(temp_reg, temp_reg);
 3789 
 3790   // Initial checks:
 3791   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 3792   //   - if (itable[0] == 0), no such interface
 3793   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 3794   cmpptr(holder_klass, resolved_klass);
 3795   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 3796   testptr(temp_itbl_klass, temp_itbl_klass);
 3797   jccb(Assembler::zero, L_no_such_interface);
 3798   cmpptr(holder_klass, temp_itbl_klass);
 3799   jccb(Assembler::equal, L_holder_found);
 3800 
 3801   // Loop: Look for holder_klass record in itable
 3802   //   do {
 3803   //     tmp = itable[index];
 3804   //     index += step;
 3805   //     if (tmp == holder_klass) {
 3806   //       goto L_holder_found; // Found!
 3807   //     }
 3808   //   } while (tmp != 0);
 3809   //   goto L_no_such_interface // Not found.
 3810   Label L_scan_holder;
 3811   bind(L_scan_holder);
 3812     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3813     addptr(scan_temp, scan_step);
 3814     cmpptr(holder_klass, temp_itbl_klass);
 3815     jccb(Assembler::equal, L_holder_found);
 3816     testptr(temp_itbl_klass, temp_itbl_klass);
 3817     jccb(Assembler::notZero, L_scan_holder);
 3818 
 3819   jmpb(L_no_such_interface);
 3820 
 3821   // Loop: Look for resolved_class record in itable
 3822   //   do {
 3823   //     tmp = itable[index];
 3824   //     index += step;
 3825   //     if (tmp == holder_klass) {
 3826   //        // Also check if we have met a holder klass
 3827   //        holder_tmp = itable[index-step-ioffset];
 3828   //     }
 3829   //     if (tmp == resolved_klass) {
 3830   //        goto L_resolved_found;  // Found!
 3831   //     }
 3832   //   } while (tmp != 0);
 3833   //   goto L_no_such_interface // Not found.
 3834   //
 3835   Label L_loop_scan_resolved;
 3836   bind(L_loop_scan_resolved);
 3837     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3838     addptr(scan_temp, scan_step);
 3839     bind(L_loop_scan_resolved_entry);
 3840     cmpptr(holder_klass, temp_itbl_klass);
 3841     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3842     cmpptr(resolved_klass, temp_itbl_klass);
 3843     jccb(Assembler::equal, L_resolved_found);
 3844     testptr(temp_itbl_klass, temp_itbl_klass);
 3845     jccb(Assembler::notZero, L_loop_scan_resolved);
 3846 
 3847   jmpb(L_no_such_interface);
 3848 
 3849   Label L_ready;
 3850 
 3851   // See if we already have a holder klass. If not, go and scan for it.
 3852   bind(L_resolved_found);
 3853   testptr(temp_reg, temp_reg);
 3854   jccb(Assembler::zero, L_scan_holder);
 3855   jmpb(L_ready);
 3856 
 3857   bind(L_holder_found);
 3858   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3859 
 3860   // Finally, temp_reg contains holder_klass vtable offset
 3861   bind(L_ready);
 3862   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3863   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 3864     load_klass(scan_temp, receiver, noreg);
 3865     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3866   } else {
 3867     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3868   }
 3869 }
 3870 
 3871 
 3872 // virtual method calling
 3873 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 3874                                            RegisterOrConstant vtable_index,
 3875                                            Register method_result) {
 3876   const ByteSize base = Klass::vtable_start_offset();
 3877   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 3878   Address vtable_entry_addr(recv_klass,
 3879                             vtable_index, Address::times_ptr,
 3880                             base + vtableEntry::method_offset());
 3881   movptr(method_result, vtable_entry_addr);
 3882 }
 3883 
 3884 
 3885 void MacroAssembler::check_klass_subtype(Register sub_klass,
 3886                            Register super_klass,
 3887                            Register temp_reg,
 3888                            Label& L_success) {
 3889   Label L_failure;
 3890   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 3891   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 3892   bind(L_failure);
 3893 }
 3894 
 3895 
 3896 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 3897                                                    Register super_klass,
 3898                                                    Register temp_reg,
 3899                                                    Label* L_success,
 3900                                                    Label* L_failure,
 3901                                                    Label* L_slow_path,
 3902                                         RegisterOrConstant super_check_offset) {
 3903   assert_different_registers(sub_klass, super_klass, temp_reg);
 3904   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 3905   if (super_check_offset.is_register()) {
 3906     assert_different_registers(sub_klass, super_klass,
 3907                                super_check_offset.as_register());
 3908   } else if (must_load_sco) {
 3909     assert(temp_reg != noreg, "supply either a temp or a register offset");
 3910   }
 3911 
 3912   Label L_fallthrough;
 3913   int label_nulls = 0;
 3914   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 3915   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 3916   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 3917   assert(label_nulls <= 1, "at most one null in the batch");
 3918 
 3919   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 3920   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 3921   Address super_check_offset_addr(super_klass, sco_offset);
 3922 
 3923   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 3924   // range of a jccb.  If this routine grows larger, reconsider at
 3925   // least some of these.
 3926 #define local_jcc(assembler_cond, label)                                \
 3927   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 3928   else                             jcc( assembler_cond, label) /*omit semi*/
 3929 
 3930   // Hacked jmp, which may only be used just before L_fallthrough.
 3931 #define final_jmp(label)                                                \
 3932   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 3933   else                            jmp(label)                /*omit semi*/
 3934 
 3935   // If the pointers are equal, we are done (e.g., String[] elements).
 3936   // This self-check enables sharing of secondary supertype arrays among
 3937   // non-primary types such as array-of-interface.  Otherwise, each such
 3938   // type would need its own customized SSA.
 3939   // We move this check to the front of the fast path because many
 3940   // type checks are in fact trivially successful in this manner,
 3941   // so we get a nicely predicted branch right at the start of the check.
 3942   cmpptr(sub_klass, super_klass);
 3943   local_jcc(Assembler::equal, *L_success);
 3944 
 3945   // Check the supertype display:
 3946   if (must_load_sco) {
 3947     // Positive movl does right thing on LP64.
 3948     movl(temp_reg, super_check_offset_addr);
 3949     super_check_offset = RegisterOrConstant(temp_reg);
 3950   }
 3951   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 3952   cmpptr(super_klass, super_check_addr); // load displayed supertype
 3953 
 3954   // This check has worked decisively for primary supers.
 3955   // Secondary supers are sought in the super_cache ('super_cache_addr').
 3956   // (Secondary supers are interfaces and very deeply nested subtypes.)
 3957   // This works in the same check above because of a tricky aliasing
 3958   // between the super_cache and the primary super display elements.
 3959   // (The 'super_check_addr' can address either, as the case requires.)
 3960   // Note that the cache is updated below if it does not help us find
 3961   // what we need immediately.
 3962   // So if it was a primary super, we can just fail immediately.
 3963   // Otherwise, it's the slow path for us (no success at this point).
 3964 
 3965   if (super_check_offset.is_register()) {
 3966     local_jcc(Assembler::equal, *L_success);
 3967     cmpl(super_check_offset.as_register(), sc_offset);
 3968     if (L_failure == &L_fallthrough) {
 3969       local_jcc(Assembler::equal, *L_slow_path);
 3970     } else {
 3971       local_jcc(Assembler::notEqual, *L_failure);
 3972       final_jmp(*L_slow_path);
 3973     }
 3974   } else if (super_check_offset.as_constant() == sc_offset) {
 3975     // Need a slow path; fast failure is impossible.
 3976     if (L_slow_path == &L_fallthrough) {
 3977       local_jcc(Assembler::equal, *L_success);
 3978     } else {
 3979       local_jcc(Assembler::notEqual, *L_slow_path);
 3980       final_jmp(*L_success);
 3981     }
 3982   } else {
 3983     // No slow path; it's a fast decision.
 3984     if (L_failure == &L_fallthrough) {
 3985       local_jcc(Assembler::equal, *L_success);
 3986     } else {
 3987       local_jcc(Assembler::notEqual, *L_failure);
 3988       final_jmp(*L_success);
 3989     }
 3990   }
 3991 
 3992   bind(L_fallthrough);
 3993 
 3994 #undef local_jcc
 3995 #undef final_jmp
 3996 }
 3997 
 3998 
 3999 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 4000                                                           Register super_klass,
 4001                                                           Register temp_reg,
 4002                                                           Register temp2_reg,
 4003                                                           Label* L_success,
 4004                                                           Label* L_failure,
 4005                                                           bool set_cond_codes) {
 4006   assert_different_registers(sub_klass, super_klass, temp_reg);
 4007   if (temp2_reg != noreg)
 4008     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4009 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4010 
 4011   Label L_fallthrough;
 4012   int label_nulls = 0;
 4013   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4014   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4015   assert(label_nulls <= 1, "at most one null in the batch");
 4016 
 4017   // a couple of useful fields in sub_klass:
 4018   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4019   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4020   Address secondary_supers_addr(sub_klass, ss_offset);
 4021   Address super_cache_addr(     sub_klass, sc_offset);
 4022 
 4023   // Do a linear scan of the secondary super-klass chain.
 4024   // This code is rarely used, so simplicity is a virtue here.
 4025   // The repne_scan instruction uses fixed registers, which we must spill.
 4026   // Don't worry too much about pre-existing connections with the input regs.
 4027 
 4028   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4029   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4030 
 4031   // Get super_klass value into rax (even if it was in rdi or rcx).
 4032   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4033   if (super_klass != rax) {
 4034     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4035     mov(rax, super_klass);
 4036   }
 4037   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4038   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4039 
 4040 #ifndef PRODUCT
 4041   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4042   ExternalAddress pst_counter_addr((address) pst_counter);
 4043   lea(rcx, pst_counter_addr);
 4044   incrementl(Address(rcx, 0));
 4045 #endif //PRODUCT
 4046 
 4047   // We will consult the secondary-super array.
 4048   movptr(rdi, secondary_supers_addr);
 4049   // Load the array length.  (Positive movl does right thing on LP64.)
 4050   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4051   // Skip to start of data.
 4052   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4053 
 4054   // Scan RCX words at [RDI] for an occurrence of RAX.
 4055   // Set NZ/Z based on last compare.
 4056   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4057   // not change flags (only scas instruction which is repeated sets flags).
 4058   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4059 
 4060     testptr(rax,rax); // Set Z = 0
 4061     repne_scan();
 4062 
 4063   // Unspill the temp. registers:
 4064   if (pushed_rdi)  pop(rdi);
 4065   if (pushed_rcx)  pop(rcx);
 4066   if (pushed_rax)  pop(rax);
 4067 
 4068   if (set_cond_codes) {
 4069     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4070     assert(!pushed_rdi, "rdi must be left non-null");
 4071     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4072   }
 4073 
 4074   if (L_failure == &L_fallthrough)
 4075         jccb(Assembler::notEqual, *L_failure);
 4076   else  jcc(Assembler::notEqual, *L_failure);
 4077 
 4078   // Success.  Cache the super we found and proceed in triumph.
 4079   movptr(super_cache_addr, super_klass);
 4080 
 4081   if (L_success != &L_fallthrough) {
 4082     jmp(*L_success);
 4083   }
 4084 
 4085 #undef IS_A_TEMP
 4086 
 4087   bind(L_fallthrough);
 4088 }
 4089 
 4090 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4091                                                    Register super_klass,
 4092                                                    Register temp_reg,
 4093                                                    Register temp2_reg,
 4094                                                    Label* L_success,
 4095                                                    Label* L_failure,
 4096                                                    bool set_cond_codes) {
 4097   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4098   check_klass_subtype_slow_path
 4099     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4100      L_success, L_failure);
 4101 }
 4102 
 4103 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4104                                                    Register super_klass,
 4105                                                    Register temp_reg,
 4106                                                    Register temp2_reg,
 4107                                                    Register temp3_reg,
 4108                                                    Register temp4_reg,
 4109                                                    Label* L_success,
 4110                                                    Label* L_failure) {
 4111   if (UseSecondarySupersTable) {
 4112     check_klass_subtype_slow_path_table
 4113       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4114        L_success, L_failure);
 4115   } else {
 4116     check_klass_subtype_slow_path_linear
 4117       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4118   }
 4119 }
 4120 
 4121 Register MacroAssembler::allocate_if_noreg(Register r,
 4122                                   RegSetIterator<Register> &available_regs,
 4123                                   RegSet &regs_to_push) {
 4124   if (!r->is_valid()) {
 4125     r = *available_regs++;
 4126     regs_to_push += r;
 4127   }
 4128   return r;
 4129 }
 4130 
 4131 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4132                                                          Register super_klass,
 4133                                                          Register temp_reg,
 4134                                                          Register temp2_reg,
 4135                                                          Register temp3_reg,
 4136                                                          Register result_reg,
 4137                                                          Label* L_success,
 4138                                                          Label* L_failure) {
 4139   // NB! Callers may assume that, when temp2_reg is a valid register,
 4140   // this code sets it to a nonzero value.
 4141   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4142 
 4143   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4144 
 4145   Label L_fallthrough;
 4146   int label_nulls = 0;
 4147   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4148   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4149   assert(label_nulls <= 1, "at most one null in the batch");
 4150 
 4151   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4152 
 4153   RegSetIterator<Register> available_regs
 4154     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4155 
 4156   RegSet pushed_regs;
 4157 
 4158   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4159   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4160   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4161   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4162   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4163 
 4164   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4165 
 4166   {
 4167 
 4168     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4169     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4170     subptr(rsp, aligned_size);
 4171     push_set(pushed_regs, 0);
 4172 
 4173     lookup_secondary_supers_table_var(sub_klass,
 4174                                       super_klass,
 4175                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4176     cmpq(result_reg, 0);
 4177 
 4178     // Unspill the temp. registers:
 4179     pop_set(pushed_regs, 0);
 4180     // Increment SP but do not clobber flags.
 4181     lea(rsp, Address(rsp, aligned_size));
 4182   }
 4183 
 4184   if (temp2_reg_was_valid) {
 4185     movq(temp2_reg, 1);
 4186   }
 4187 
 4188   jcc(Assembler::notEqual, *L_failure);
 4189 
 4190   if (L_success != &L_fallthrough) {
 4191     jmp(*L_success);
 4192   }
 4193 
 4194   bind(L_fallthrough);
 4195 }
 4196 
 4197 // population_count variant for running without the POPCNT
 4198 // instruction, which was introduced with SSE4.2 in 2008.
 4199 void MacroAssembler::population_count(Register dst, Register src,
 4200                                       Register scratch1, Register scratch2) {
 4201   assert_different_registers(src, scratch1, scratch2);
 4202   if (UsePopCountInstruction) {
 4203     Assembler::popcntq(dst, src);
 4204   } else {
 4205     assert_different_registers(src, scratch1, scratch2);
 4206     assert_different_registers(dst, scratch1, scratch2);
 4207     Label loop, done;
 4208 
 4209     mov(scratch1, src);
 4210     // dst = 0;
 4211     // while(scratch1 != 0) {
 4212     //   dst++;
 4213     //   scratch1 &= (scratch1 - 1);
 4214     // }
 4215     xorl(dst, dst);
 4216     testq(scratch1, scratch1);
 4217     jccb(Assembler::equal, done);
 4218     {
 4219       bind(loop);
 4220       incq(dst);
 4221       movq(scratch2, scratch1);
 4222       decq(scratch2);
 4223       andq(scratch1, scratch2);
 4224       jccb(Assembler::notEqual, loop);
 4225     }
 4226     bind(done);
 4227   }
 4228 #ifdef ASSERT
 4229   mov64(scratch1, 0xCafeBabeDeadBeef);
 4230   movq(scratch2, scratch1);
 4231 #endif
 4232 }
 4233 
 4234 // Ensure that the inline code and the stub are using the same registers.
 4235 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4236 do {                                                                 \
 4237   assert(r_super_klass  == rax, "mismatch");                         \
 4238   assert(r_array_base   == rbx, "mismatch");                         \
 4239   assert(r_array_length == rcx, "mismatch");                         \
 4240   assert(r_array_index  == rdx, "mismatch");                         \
 4241   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4242   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4243   assert(result         == rdi || result      == noreg, "mismatch"); \
 4244 } while(0)
 4245 
 4246 // Versions of salq and rorq that don't need count to be in rcx
 4247 
 4248 void MacroAssembler::salq(Register dest, Register count) {
 4249   if (count == rcx) {
 4250     Assembler::salq(dest);
 4251   } else {
 4252     assert_different_registers(rcx, dest);
 4253     xchgq(rcx, count);
 4254     Assembler::salq(dest);
 4255     xchgq(rcx, count);
 4256   }
 4257 }
 4258 
 4259 void MacroAssembler::rorq(Register dest, Register count) {
 4260   if (count == rcx) {
 4261     Assembler::rorq(dest);
 4262   } else {
 4263     assert_different_registers(rcx, dest);
 4264     xchgq(rcx, count);
 4265     Assembler::rorq(dest);
 4266     xchgq(rcx, count);
 4267   }
 4268 }
 4269 
 4270 // Return true: we succeeded in generating this code
 4271 //
 4272 // At runtime, return 0 in result if r_super_klass is a superclass of
 4273 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4274 // super_klass_slot of the class you're looking for. This is always
 4275 // the case for instanceof and checkcast.
 4276 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4277                                                          Register r_super_klass,
 4278                                                          Register temp1,
 4279                                                          Register temp2,
 4280                                                          Register temp3,
 4281                                                          Register temp4,
 4282                                                          Register result,
 4283                                                          u1 super_klass_slot) {
 4284   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4285 
 4286   Label L_fallthrough, L_success, L_failure;
 4287 
 4288   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4289 
 4290   const Register
 4291     r_array_index  = temp1,
 4292     r_array_length = temp2,
 4293     r_array_base   = temp3,
 4294     r_bitmap       = temp4;
 4295 
 4296   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4297 
 4298   xorq(result, result); // = 0
 4299 
 4300   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4301   movq(r_array_index, r_bitmap);
 4302 
 4303   // First check the bitmap to see if super_klass might be present. If
 4304   // the bit is zero, we are certain that super_klass is not one of
 4305   // the secondary supers.
 4306   u1 bit = super_klass_slot;
 4307   {
 4308     // NB: If the count in a x86 shift instruction is 0, the flags are
 4309     // not affected, so we do a testq instead.
 4310     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4311     if (shift_count != 0) {
 4312       salq(r_array_index, shift_count);
 4313     } else {
 4314       testq(r_array_index, r_array_index);
 4315     }
 4316   }
 4317   // We test the MSB of r_array_index, i.e. its sign bit
 4318   jcc(Assembler::positive, L_failure);
 4319 
 4320   // Get the first array index that can contain super_klass into r_array_index.
 4321   if (bit != 0) {
 4322     population_count(r_array_index, r_array_index, temp2, temp3);
 4323   } else {
 4324     movl(r_array_index, 1);
 4325   }
 4326   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4327 
 4328   // We will consult the secondary-super array.
 4329   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4330 
 4331   // We're asserting that the first word in an Array<Klass*> is the
 4332   // length, and the second word is the first word of the data. If
 4333   // that ever changes, r_array_base will have to be adjusted here.
 4334   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4335   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4336 
 4337   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4338   jccb(Assembler::equal, L_success);
 4339 
 4340   // Is there another entry to check? Consult the bitmap.
 4341   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4342   jccb(Assembler::carryClear, L_failure);
 4343 
 4344   // Linear probe. Rotate the bitmap so that the next bit to test is
 4345   // in Bit 1.
 4346   if (bit != 0) {
 4347     rorq(r_bitmap, bit);
 4348   }
 4349 
 4350   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4351   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4352   // Kills: r_array_length.
 4353   // Returns: result.
 4354   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4355   // Result (0/1) is in rdi
 4356   jmpb(L_fallthrough);
 4357 
 4358   bind(L_failure);
 4359   incq(result); // 0 => 1
 4360 
 4361   bind(L_success);
 4362   // result = 0;
 4363 
 4364   bind(L_fallthrough);
 4365   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4366 
 4367   if (VerifySecondarySupers) {
 4368     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4369                                   temp1, temp2, temp3);
 4370   }
 4371 }
 4372 
 4373 // At runtime, return 0 in result if r_super_klass is a superclass of
 4374 // r_sub_klass, otherwise return nonzero. Use this version of
 4375 // lookup_secondary_supers_table() if you don't know ahead of time
 4376 // which superclass will be searched for. Used by interpreter and
 4377 // runtime stubs. It is larger and has somewhat greater latency than
 4378 // the version above, which takes a constant super_klass_slot.
 4379 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4380                                                        Register r_super_klass,
 4381                                                        Register temp1,
 4382                                                        Register temp2,
 4383                                                        Register temp3,
 4384                                                        Register temp4,
 4385                                                        Register result) {
 4386   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4387   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4388   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4389 
 4390   Label L_fallthrough, L_success, L_failure;
 4391 
 4392   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4393 
 4394   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4395 
 4396   // FIXME. Once we are sure that all paths reaching this point really
 4397   // do pass rcx as one of our temps we can get rid of the following
 4398   // workaround.
 4399   assert(temps.contains(rcx), "fix this code");
 4400 
 4401   // We prefer to have our shift count in rcx. If rcx is one of our
 4402   // temps, use it for slot. If not, pick any of our temps.
 4403   Register slot;
 4404   if (!temps.contains(rcx)) {
 4405     slot = *available_regs++;
 4406   } else {
 4407     slot = rcx;
 4408   }
 4409 
 4410   const Register r_array_index = *available_regs++;
 4411   const Register r_bitmap      = *available_regs++;
 4412 
 4413   // The logic above guarantees this property, but we state it here.
 4414   assert_different_registers(r_array_index, r_bitmap, rcx);
 4415 
 4416   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4417   movq(r_array_index, r_bitmap);
 4418 
 4419   // First check the bitmap to see if super_klass might be present. If
 4420   // the bit is zero, we are certain that super_klass is not one of
 4421   // the secondary supers.
 4422   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4423   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4424   salq(r_array_index, slot);
 4425 
 4426   testq(r_array_index, r_array_index);
 4427   // We test the MSB of r_array_index, i.e. its sign bit
 4428   jcc(Assembler::positive, L_failure);
 4429 
 4430   const Register r_array_base = *available_regs++;
 4431 
 4432   // Get the first array index that can contain super_klass into r_array_index.
 4433   // Note: Clobbers r_array_base and slot.
 4434   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4435 
 4436   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4437 
 4438   // We will consult the secondary-super array.
 4439   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4440 
 4441   // We're asserting that the first word in an Array<Klass*> is the
 4442   // length, and the second word is the first word of the data. If
 4443   // that ever changes, r_array_base will have to be adjusted here.
 4444   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4445   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4446 
 4447   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4448   jccb(Assembler::equal, L_success);
 4449 
 4450   // Restore slot to its true value
 4451   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4452 
 4453   // Linear probe. Rotate the bitmap so that the next bit to test is
 4454   // in Bit 1.
 4455   rorq(r_bitmap, slot);
 4456 
 4457   // Is there another entry to check? Consult the bitmap.
 4458   btq(r_bitmap, 1);
 4459   jccb(Assembler::carryClear, L_failure);
 4460 
 4461   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4462   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4463   // Kills: r_array_length.
 4464   // Returns: result.
 4465   lookup_secondary_supers_table_slow_path(r_super_klass,
 4466                                           r_array_base,
 4467                                           r_array_index,
 4468                                           r_bitmap,
 4469                                           /*temp1*/result,
 4470                                           /*temp2*/slot,
 4471                                           &L_success,
 4472                                           nullptr);
 4473 
 4474   bind(L_failure);
 4475   movq(result, 1);
 4476   jmpb(L_fallthrough);
 4477 
 4478   bind(L_success);
 4479   xorq(result, result); // = 0
 4480 
 4481   bind(L_fallthrough);
 4482   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4483 
 4484   if (VerifySecondarySupers) {
 4485     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4486                                   temp1, temp2, temp3);
 4487   }
 4488 }
 4489 
 4490 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4491                                  Label* L_success, Label* L_failure) {
 4492   Label L_loop, L_fallthrough;
 4493   {
 4494     int label_nulls = 0;
 4495     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4496     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4497     assert(label_nulls <= 1, "at most one null in the batch");
 4498   }
 4499   bind(L_loop);
 4500   cmpq(value, Address(addr, count, Address::times_8));
 4501   jcc(Assembler::equal, *L_success);
 4502   addl(count, 1);
 4503   cmpl(count, limit);
 4504   jcc(Assembler::less, L_loop);
 4505 
 4506   if (&L_fallthrough != L_failure) {
 4507     jmp(*L_failure);
 4508   }
 4509   bind(L_fallthrough);
 4510 }
 4511 
 4512 // Called by code generated by check_klass_subtype_slow_path
 4513 // above. This is called when there is a collision in the hashed
 4514 // lookup in the secondary supers array.
 4515 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4516                                                              Register r_array_base,
 4517                                                              Register r_array_index,
 4518                                                              Register r_bitmap,
 4519                                                              Register temp1,
 4520                                                              Register temp2,
 4521                                                              Label* L_success,
 4522                                                              Label* L_failure) {
 4523   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4524 
 4525   const Register
 4526     r_array_length = temp1,
 4527     r_sub_klass    = noreg,
 4528     result         = noreg;
 4529 
 4530   Label L_fallthrough;
 4531   int label_nulls = 0;
 4532   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4533   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4534   assert(label_nulls <= 1, "at most one null in the batch");
 4535 
 4536   // Load the array length.
 4537   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4538   // And adjust the array base to point to the data.
 4539   // NB! Effectively increments current slot index by 1.
 4540   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4541   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4542 
 4543   // Linear probe
 4544   Label L_huge;
 4545 
 4546   // The bitmap is full to bursting.
 4547   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4548   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4549   jcc(Assembler::greater, L_huge);
 4550 
 4551   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4552   // current slot (at secondary_supers[r_array_index]) has not yet
 4553   // been inspected, and r_array_index may be out of bounds if we
 4554   // wrapped around the end of the array.
 4555 
 4556   { // This is conventional linear probing, but instead of terminating
 4557     // when a null entry is found in the table, we maintain a bitmap
 4558     // in which a 0 indicates missing entries.
 4559     // The check above guarantees there are 0s in the bitmap, so the loop
 4560     // eventually terminates.
 4561 
 4562     xorl(temp2, temp2); // = 0;
 4563 
 4564     Label L_again;
 4565     bind(L_again);
 4566 
 4567     // Check for array wraparound.
 4568     cmpl(r_array_index, r_array_length);
 4569     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4570 
 4571     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4572     jcc(Assembler::equal, *L_success);
 4573 
 4574     // If the next bit in bitmap is zero, we're done.
 4575     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4576     jcc(Assembler::carryClear, *L_failure);
 4577 
 4578     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4579     addl(r_array_index, 1);
 4580 
 4581     jmp(L_again);
 4582   }
 4583 
 4584   { // Degenerate case: more than 64 secondary supers.
 4585     // FIXME: We could do something smarter here, maybe a vectorized
 4586     // comparison or a binary search, but is that worth any added
 4587     // complexity?
 4588     bind(L_huge);
 4589     xorl(r_array_index, r_array_index); // = 0
 4590     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4591                 L_success,
 4592                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4593 
 4594     bind(L_fallthrough);
 4595   }
 4596 }
 4597 
 4598 struct VerifyHelperArguments {
 4599   Klass* _super;
 4600   Klass* _sub;
 4601   intptr_t _linear_result;
 4602   intptr_t _table_result;
 4603 };
 4604 
 4605 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4606   Klass::on_secondary_supers_verification_failure(args->_super,
 4607                                                   args->_sub,
 4608                                                   args->_linear_result,
 4609                                                   args->_table_result,
 4610                                                   msg);
 4611 }
 4612 
 4613 // Make sure that the hashed lookup and a linear scan agree.
 4614 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4615                                                    Register r_super_klass,
 4616                                                    Register result,
 4617                                                    Register temp1,
 4618                                                    Register temp2,
 4619                                                    Register temp3) {
 4620   const Register
 4621       r_array_index  = temp1,
 4622       r_array_length = temp2,
 4623       r_array_base   = temp3,
 4624       r_bitmap       = noreg;
 4625 
 4626   BLOCK_COMMENT("verify_secondary_supers_table {");
 4627 
 4628   Label L_success, L_failure, L_check, L_done;
 4629 
 4630   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4631   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4632   // And adjust the array base to point to the data.
 4633   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4634 
 4635   testl(r_array_length, r_array_length); // array_length == 0?
 4636   jcc(Assembler::zero, L_failure);
 4637 
 4638   movl(r_array_index, 0);
 4639   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4640   // fall through to L_failure
 4641 
 4642   const Register linear_result = r_array_index; // reuse temp1
 4643 
 4644   bind(L_failure); // not present
 4645   movl(linear_result, 1);
 4646   jmp(L_check);
 4647 
 4648   bind(L_success); // present
 4649   movl(linear_result, 0);
 4650 
 4651   bind(L_check);
 4652   cmpl(linear_result, result);
 4653   jcc(Assembler::equal, L_done);
 4654 
 4655   { // To avoid calling convention issues, build a record on the stack
 4656     // and pass the pointer to that instead.
 4657     push(result);
 4658     push(linear_result);
 4659     push(r_sub_klass);
 4660     push(r_super_klass);
 4661     movptr(c_rarg1, rsp);
 4662     movptr(c_rarg0, (uintptr_t) "mismatch");
 4663     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4664     should_not_reach_here();
 4665   }
 4666   bind(L_done);
 4667 
 4668   BLOCK_COMMENT("} verify_secondary_supers_table");
 4669 }
 4670 
 4671 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4672 
 4673 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4674   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4675 
 4676   Label L_fallthrough;
 4677   if (L_fast_path == nullptr) {
 4678     L_fast_path = &L_fallthrough;
 4679   } else if (L_slow_path == nullptr) {
 4680     L_slow_path = &L_fallthrough;
 4681   }
 4682 
 4683   // Fast path check: class is fully initialized.
 4684   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4685   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4686   jcc(Assembler::equal, *L_fast_path);
 4687 
 4688   // Fast path check: current thread is initializer thread
 4689   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4690   if (L_slow_path == &L_fallthrough) {
 4691     jcc(Assembler::equal, *L_fast_path);
 4692     bind(*L_slow_path);
 4693   } else if (L_fast_path == &L_fallthrough) {
 4694     jcc(Assembler::notEqual, *L_slow_path);
 4695     bind(*L_fast_path);
 4696   } else {
 4697     Unimplemented();
 4698   }
 4699 }
 4700 
 4701 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4702   if (VM_Version::supports_cmov()) {
 4703     cmovl(cc, dst, src);
 4704   } else {
 4705     Label L;
 4706     jccb(negate_condition(cc), L);
 4707     movl(dst, src);
 4708     bind(L);
 4709   }
 4710 }
 4711 
 4712 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4713   if (VM_Version::supports_cmov()) {
 4714     cmovl(cc, dst, src);
 4715   } else {
 4716     Label L;
 4717     jccb(negate_condition(cc), L);
 4718     movl(dst, src);
 4719     bind(L);
 4720   }
 4721 }
 4722 
 4723 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4724   if (!VerifyOops) return;




 4725 
 4726   BLOCK_COMMENT("verify_oop {");
 4727   push(rscratch1);
 4728   push(rax);                          // save rax
 4729   push(reg);                          // pass register argument
 4730 
 4731   // Pass register number to verify_oop_subroutine
 4732   const char* b = nullptr;
 4733   {
 4734     ResourceMark rm;
 4735     stringStream ss;
 4736     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4737     b = code_string(ss.as_string());
 4738   }
 4739   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4740   pushptr(buffer.addr(), rscratch1);
 4741 
 4742   // call indirectly to solve generation ordering problem
 4743   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4744   call(rax);
 4745   // Caller pops the arguments (oop, message) and restores rax, r10
 4746   BLOCK_COMMENT("} verify_oop");
 4747 }
 4748 
 4749 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 4750   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 4751     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 4752     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 4753     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 4754   } else if (VM_Version::supports_avx()) {
 4755     vpcmpeqd(dst, dst, dst, vector_len);
 4756   } else {
 4757     pcmpeqd(dst, dst);
 4758   }
 4759 }
 4760 
 4761 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 4762                                          int extra_slot_offset) {
 4763   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 4764   int stackElementSize = Interpreter::stackElementSize;
 4765   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 4766 #ifdef ASSERT
 4767   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 4768   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 4769 #endif
 4770   Register             scale_reg    = noreg;
 4771   Address::ScaleFactor scale_factor = Address::no_scale;
 4772   if (arg_slot.is_constant()) {
 4773     offset += arg_slot.as_constant() * stackElementSize;
 4774   } else {
 4775     scale_reg    = arg_slot.as_register();
 4776     scale_factor = Address::times(stackElementSize);
 4777   }
 4778   offset += wordSize;           // return PC is on stack
 4779   return Address(rsp, scale_reg, scale_factor, offset);
 4780 }
 4781 
 4782 // Handle the receiver type profile update given the "recv" klass.
 4783 //
 4784 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 4785 // If there are no matching or claimable receiver entries in RD, updates
 4786 // the polymorphic counter.
 4787 //
 4788 // This code expected to run by either the interpreter or JIT-ed code, without
 4789 // extra synchronization. For safety, receiver cells are claimed atomically, which
 4790 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 4791 // counter updates are not atomic.
 4792 //
 4793 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 4794   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 4795   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 4796   int poly_count_offset      = in_bytes(CounterData::count_offset());
 4797   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 4798   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 4799 
 4800   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 4801   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 4802   base_receiver_offset += mdp_offset;
 4803   end_receiver_offset  += mdp_offset;
 4804   poly_count_offset    += mdp_offset;
 4805 
 4806   // Scale down to optimize encoding. Slots are pointer-sized.
 4807   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 4808   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 4809   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 4810   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 4811   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 4812   base_receiver_offset   >>= LogBytesPerWord;
 4813   end_receiver_offset    >>= LogBytesPerWord;
 4814   poly_count_offset      >>= LogBytesPerWord;
 4815   receiver_step          >>= LogBytesPerWord;
 4816   receiver_to_count_step >>= LogBytesPerWord;
 4817 
 4818 #ifdef ASSERT
 4819   // We are about to walk the MDO slots without asking for offsets.
 4820   // Check that our math hits all the right spots.
 4821   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 4822     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 4823     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 4824     int offset = base_receiver_offset + receiver_step*c;
 4825     int count_offset = offset + receiver_to_count_step;
 4826     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 4827     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 4828   }
 4829   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 4830   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 4831 #endif
 4832 
 4833   // Corner case: no profile table. Increment poly counter and exit.
 4834   if (ReceiverTypeData::row_limit() == 0) {
 4835     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 4836     return;
 4837   }
 4838 
 4839   Register offset = rscratch1;
 4840 
 4841   Label L_loop_search_receiver, L_loop_search_empty;
 4842   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 4843 
 4844   // The code here recognizes three major cases:
 4845   //   A. Fastest: receiver found in the table
 4846   //   B. Fast: no receiver in the table, and the table is full
 4847   //   C. Slow: no receiver in the table, free slots in the table
 4848   //
 4849   // The case A performance is most important, as perfectly-behaved code would end up
 4850   // there, especially with larger TypeProfileWidth. The case B performance is
 4851   // important as well, this is where bulk of code would land for normally megamorphic
 4852   // cases. The case C performance is not essential, its job is to deal with installation
 4853   // races, we optimize for code density instead. Case C needs to make sure that receiver
 4854   // rows are only claimed once. This makes sure we never overwrite a row for another
 4855   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 4856   //
 4857   // It is very tempting to handle these cases in a single loop, and claim the first slot
 4858   // without checking the rest of the table. But, profiling code should tolerate free slots
 4859   // in the table, as class unloading can clear them. After such cleanup, the receiver
 4860   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 4861   // to complete, before trying to install new slots. Splitting the code in several tight
 4862   // loops also helpfully optimizes for cases A and B.
 4863   //
 4864   // This code is effectively:
 4865   //
 4866   // restart:
 4867   //   // Fastest: receiver is already installed
 4868   //   for (i = 0; i < receiver_count(); i++) {
 4869   //     if (receiver(i) == recv) goto found_recv(i);
 4870   //   }
 4871   //
 4872   //   // Fast: no receiver, but profile is full
 4873   //   for (i = 0; i < receiver_count(); i++) {
 4874   //     if (receiver(i) == null) goto found_null(i);
 4875   //   }
 4876   //   goto polymorphic
 4877   //
 4878   //   // Slow: try to install receiver
 4879   // found_null(i):
 4880   //   CAS(&receiver(i), null, recv);
 4881   //   goto restart
 4882   //
 4883   // polymorphic:
 4884   //   count++;
 4885   //   return
 4886   //
 4887   // found_recv(i):
 4888   //   *receiver_count(i)++
 4889   //
 4890 
 4891   bind(L_restart);
 4892 
 4893   // Fastest: receiver is already installed
 4894   movptr(offset, base_receiver_offset);
 4895   bind(L_loop_search_receiver);
 4896     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 4897     jccb(Assembler::equal, L_found_recv);
 4898   addptr(offset, receiver_step);
 4899   cmpptr(offset, end_receiver_offset);
 4900   jccb(Assembler::notEqual, L_loop_search_receiver);
 4901 
 4902   // Fast: no receiver, but profile is full
 4903   movptr(offset, base_receiver_offset);
 4904   bind(L_loop_search_empty);
 4905     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 4906     jccb(Assembler::equal, L_found_empty);
 4907   addptr(offset, receiver_step);
 4908   cmpptr(offset, end_receiver_offset);
 4909   jccb(Assembler::notEqual, L_loop_search_empty);
 4910   jmpb(L_polymorphic);
 4911 
 4912   // Slow: try to install receiver
 4913   bind(L_found_empty);
 4914 
 4915   // Atomically swing receiver slot: null -> recv.
 4916   //
 4917   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 4918   // other important registers untouched, as they form the address. Therefore, we need
 4919   // to shift any important registers from RAX into some other spare register. If we
 4920   // have a spare register, we are forced to save it on stack here.
 4921 
 4922   Register spare_reg = noreg;
 4923   Register shifted_mdp = mdp;
 4924   Register shifted_recv = recv;
 4925   if (recv == rax || mdp == rax) {
 4926     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 4927                 (recv != rcx && mdp != rcx) ? rcx :
 4928                 rdx;
 4929     assert_different_registers(mdp, recv, offset, spare_reg);
 4930 
 4931     push(spare_reg);
 4932     if (recv == rax) {
 4933       movptr(spare_reg, recv);
 4934       shifted_recv = spare_reg;
 4935     } else {
 4936       assert(mdp == rax, "Remaining case");
 4937       movptr(spare_reg, mdp);
 4938       shifted_mdp = spare_reg;
 4939     }
 4940   } else {
 4941     push(rax);
 4942   }
 4943 
 4944   // None of the important registers are in RAX after this shuffle.
 4945   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 4946 
 4947   xorptr(rax, rax);
 4948   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 4949 
 4950   // Unshift registers.
 4951   if (recv == rax || mdp == rax) {
 4952     movptr(rax, spare_reg);
 4953     pop(spare_reg);
 4954   } else {
 4955     pop(rax);
 4956   }
 4957 
 4958   // CAS success means the slot now has the receiver we want. CAS failure means
 4959   // something had claimed the slot concurrently: it can be the same receiver we want,
 4960   // or something else. Since this is a slow path, we can optimize for code density,
 4961   // and just restart the search from the beginning.
 4962   jmpb(L_restart);
 4963 
 4964   // Counter updates:
 4965 
 4966   // Increment polymorphic counter instead of receiver slot.
 4967   bind(L_polymorphic);
 4968   movptr(offset, poly_count_offset);
 4969   jmpb(L_count_update);
 4970 
 4971   // Found a receiver, convert its slot offset to corresponding count offset.
 4972   bind(L_found_recv);
 4973   addptr(offset, receiver_to_count_step);
 4974 
 4975   bind(L_count_update);
 4976   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 4977 }
 4978 
 4979 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 4980   if (!VerifyOops) return;




 4981 
 4982   push(rscratch1);
 4983   push(rax); // save rax,
 4984   // addr may contain rsp so we will have to adjust it based on the push
 4985   // we just did (and on 64 bit we do two pushes)
 4986   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 4987   // stores rax into addr which is backwards of what was intended.
 4988   if (addr.uses(rsp)) {
 4989     lea(rax, addr);
 4990     pushptr(Address(rax, 2 * BytesPerWord));
 4991   } else {
 4992     pushptr(addr);
 4993   }
 4994 
 4995   // Pass register number to verify_oop_subroutine
 4996   const char* b = nullptr;
 4997   {
 4998     ResourceMark rm;
 4999     stringStream ss;
 5000     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 5001     b = code_string(ss.as_string());
 5002   }
 5003   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5004   pushptr(buffer.addr(), rscratch1);
 5005 
 5006   // call indirectly to solve generation ordering problem
 5007   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5008   call(rax);
 5009   // Caller pops the arguments (addr, message) and restores rax, r10.
 5010 }
 5011 
 5012 void MacroAssembler::verify_tlab() {
 5013 #ifdef ASSERT
 5014   if (UseTLAB && VerifyOops) {
 5015     Label next, ok;
 5016     Register t1 = rsi;
 5017 
 5018     push(t1);
 5019 
 5020     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5021     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 5022     jcc(Assembler::aboveEqual, next);
 5023     STOP("assert(top >= start)");
 5024     should_not_reach_here();
 5025 
 5026     bind(next);
 5027     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 5028     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5029     jcc(Assembler::aboveEqual, ok);
 5030     STOP("assert(top <= end)");
 5031     should_not_reach_here();
 5032 
 5033     bind(ok);
 5034     pop(t1);
 5035   }
 5036 #endif
 5037 }
 5038 
 5039 class ControlWord {
 5040  public:
 5041   int32_t _value;
 5042 
 5043   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5044   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5045   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5046   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5047   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5048   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5049   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5050   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5051 
 5052   void print() const {
 5053     // rounding control
 5054     const char* rc;
 5055     switch (rounding_control()) {
 5056       case 0: rc = "round near"; break;
 5057       case 1: rc = "round down"; break;
 5058       case 2: rc = "round up  "; break;
 5059       case 3: rc = "chop      "; break;
 5060       default:
 5061         rc = nullptr; // silence compiler warnings
 5062         fatal("Unknown rounding control: %d", rounding_control());
 5063     };
 5064     // precision control
 5065     const char* pc;
 5066     switch (precision_control()) {
 5067       case 0: pc = "24 bits "; break;
 5068       case 1: pc = "reserved"; break;
 5069       case 2: pc = "53 bits "; break;
 5070       case 3: pc = "64 bits "; break;
 5071       default:
 5072         pc = nullptr; // silence compiler warnings
 5073         fatal("Unknown precision control: %d", precision_control());
 5074     };
 5075     // flags
 5076     char f[9];
 5077     f[0] = ' ';
 5078     f[1] = ' ';
 5079     f[2] = (precision   ()) ? 'P' : 'p';
 5080     f[3] = (underflow   ()) ? 'U' : 'u';
 5081     f[4] = (overflow    ()) ? 'O' : 'o';
 5082     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5083     f[6] = (denormalized()) ? 'D' : 'd';
 5084     f[7] = (invalid     ()) ? 'I' : 'i';
 5085     f[8] = '\x0';
 5086     // output
 5087     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5088   }
 5089 
 5090 };
 5091 
 5092 class StatusWord {
 5093  public:
 5094   int32_t _value;
 5095 
 5096   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5097   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5098   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5099   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5100   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5101   int  top() const                     { return  (_value >> 11) & 7      ; }
 5102   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5103   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5104   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5105   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5106   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5107   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5108   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5109   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5110 
 5111   void print() const {
 5112     // condition codes
 5113     char c[5];
 5114     c[0] = (C3()) ? '3' : '-';
 5115     c[1] = (C2()) ? '2' : '-';
 5116     c[2] = (C1()) ? '1' : '-';
 5117     c[3] = (C0()) ? '0' : '-';
 5118     c[4] = '\x0';
 5119     // flags
 5120     char f[9];
 5121     f[0] = (error_status()) ? 'E' : '-';
 5122     f[1] = (stack_fault ()) ? 'S' : '-';
 5123     f[2] = (precision   ()) ? 'P' : '-';
 5124     f[3] = (underflow   ()) ? 'U' : '-';
 5125     f[4] = (overflow    ()) ? 'O' : '-';
 5126     f[5] = (zero_divide ()) ? 'Z' : '-';
 5127     f[6] = (denormalized()) ? 'D' : '-';
 5128     f[7] = (invalid     ()) ? 'I' : '-';
 5129     f[8] = '\x0';
 5130     // output
 5131     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5132   }
 5133 
 5134 };
 5135 
 5136 class TagWord {
 5137  public:
 5138   int32_t _value;
 5139 
 5140   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5141 
 5142   void print() const {
 5143     printf("%04x", _value & 0xFFFF);
 5144   }
 5145 
 5146 };
 5147 
 5148 class FPU_Register {
 5149  public:
 5150   int32_t _m0;
 5151   int32_t _m1;
 5152   int16_t _ex;
 5153 
 5154   bool is_indefinite() const           {
 5155     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5156   }
 5157 
 5158   void print() const {
 5159     char  sign = (_ex < 0) ? '-' : '+';
 5160     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5161     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5162   };
 5163 
 5164 };
 5165 
 5166 class FPU_State {
 5167  public:
 5168   enum {
 5169     register_size       = 10,
 5170     number_of_registers =  8,
 5171     register_mask       =  7
 5172   };
 5173 
 5174   ControlWord  _control_word;
 5175   StatusWord   _status_word;
 5176   TagWord      _tag_word;
 5177   int32_t      _error_offset;
 5178   int32_t      _error_selector;
 5179   int32_t      _data_offset;
 5180   int32_t      _data_selector;
 5181   int8_t       _register[register_size * number_of_registers];
 5182 
 5183   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5184   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5185 
 5186   const char* tag_as_string(int tag) const {
 5187     switch (tag) {
 5188       case 0: return "valid";
 5189       case 1: return "zero";
 5190       case 2: return "special";
 5191       case 3: return "empty";
 5192     }
 5193     ShouldNotReachHere();
 5194     return nullptr;
 5195   }
 5196 
 5197   void print() const {
 5198     // print computation registers
 5199     { int t = _status_word.top();
 5200       for (int i = 0; i < number_of_registers; i++) {
 5201         int j = (i - t) & register_mask;
 5202         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5203         st(j)->print();
 5204         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5205       }
 5206     }
 5207     printf("\n");
 5208     // print control registers
 5209     printf("ctrl = "); _control_word.print(); printf("\n");
 5210     printf("stat = "); _status_word .print(); printf("\n");
 5211     printf("tags = "); _tag_word    .print(); printf("\n");
 5212   }
 5213 
 5214 };
 5215 
 5216 class Flag_Register {
 5217  public:
 5218   int32_t _value;
 5219 
 5220   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5221   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5222   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5223   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5224   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5225   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5226   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5227 
 5228   void print() const {
 5229     // flags
 5230     char f[8];
 5231     f[0] = (overflow       ()) ? 'O' : '-';
 5232     f[1] = (direction      ()) ? 'D' : '-';
 5233     f[2] = (sign           ()) ? 'S' : '-';
 5234     f[3] = (zero           ()) ? 'Z' : '-';
 5235     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5236     f[5] = (parity         ()) ? 'P' : '-';
 5237     f[6] = (carry          ()) ? 'C' : '-';
 5238     f[7] = '\x0';
 5239     // output
 5240     printf("%08x  flags = %s", _value, f);
 5241   }
 5242 
 5243 };
 5244 
 5245 class IU_Register {
 5246  public:
 5247   int32_t _value;
 5248 
 5249   void print() const {
 5250     printf("%08x  %11d", _value, _value);
 5251   }
 5252 
 5253 };
 5254 
 5255 class IU_State {
 5256  public:
 5257   Flag_Register _eflags;
 5258   IU_Register   _rdi;
 5259   IU_Register   _rsi;
 5260   IU_Register   _rbp;
 5261   IU_Register   _rsp;
 5262   IU_Register   _rbx;
 5263   IU_Register   _rdx;
 5264   IU_Register   _rcx;
 5265   IU_Register   _rax;
 5266 
 5267   void print() const {
 5268     // computation registers
 5269     printf("rax,  = "); _rax.print(); printf("\n");
 5270     printf("rbx,  = "); _rbx.print(); printf("\n");
 5271     printf("rcx  = "); _rcx.print(); printf("\n");
 5272     printf("rdx  = "); _rdx.print(); printf("\n");
 5273     printf("rdi  = "); _rdi.print(); printf("\n");
 5274     printf("rsi  = "); _rsi.print(); printf("\n");
 5275     printf("rbp,  = "); _rbp.print(); printf("\n");
 5276     printf("rsp  = "); _rsp.print(); printf("\n");
 5277     printf("\n");
 5278     // control registers
 5279     printf("flgs = "); _eflags.print(); printf("\n");
 5280   }
 5281 };
 5282 
 5283 
 5284 class CPU_State {
 5285  public:
 5286   FPU_State _fpu_state;
 5287   IU_State  _iu_state;
 5288 
 5289   void print() const {
 5290     printf("--------------------------------------------------\n");
 5291     _iu_state .print();
 5292     printf("\n");
 5293     _fpu_state.print();
 5294     printf("--------------------------------------------------\n");
 5295   }
 5296 
 5297 };
 5298 
 5299 
 5300 static void _print_CPU_state(CPU_State* state) {
 5301   state->print();
 5302 };
 5303 
 5304 
 5305 void MacroAssembler::print_CPU_state() {
 5306   push_CPU_state();
 5307   push(rsp);                // pass CPU state
 5308   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5309   addptr(rsp, wordSize);       // discard argument
 5310   pop_CPU_state();
 5311 }
 5312 
 5313 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5314   // Either restore the MXCSR register after returning from the JNI Call
 5315   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5316   if (VM_Version::supports_sse()) {
 5317     if (RestoreMXCSROnJNICalls) {
 5318       ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5319     } else if (CheckJNICalls) {
 5320       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5321     }
 5322   }
 5323   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5324   vzeroupper();
 5325 }
 5326 
 5327 // ((OopHandle)result).resolve();
 5328 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5329   assert_different_registers(result, tmp);
 5330 
 5331   // Only 64 bit platforms support GCs that require a tmp register
 5332   // Only IN_HEAP loads require a thread_tmp register
 5333   // OopHandle::resolve is an indirection like jobject.
 5334   access_load_at(T_OBJECT, IN_NATIVE,
 5335                  result, Address(result, 0), tmp);
 5336 }
 5337 
 5338 // ((WeakHandle)result).resolve();
 5339 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5340   assert_different_registers(rresult, rtmp);
 5341   Label resolved;
 5342 
 5343   // A null weak handle resolves to null.
 5344   cmpptr(rresult, 0);
 5345   jcc(Assembler::equal, resolved);
 5346 
 5347   // Only 64 bit platforms support GCs that require a tmp register
 5348   // Only IN_HEAP loads require a thread_tmp register
 5349   // WeakHandle::resolve is an indirection like jweak.
 5350   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5351                  rresult, Address(rresult, 0), rtmp);
 5352   bind(resolved);
 5353 }
 5354 
 5355 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5356   // get mirror
 5357   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5358   load_method_holder(mirror, method);
 5359   movptr(mirror, Address(mirror, mirror_offset));
 5360   resolve_oop_handle(mirror, tmp);
 5361 }
 5362 
 5363 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5364   load_method_holder(rresult, rmethod);
 5365   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5366 }
 5367 
 5368 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5369   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5370   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5371   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5372 }
 5373 










 5374 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5375   assert(UseCompactObjectHeaders, "expect compact object headers");
 5376   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5377   shrq(dst, markWord::klass_shift);
 5378 }
 5379 
 5380 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5381   assert_different_registers(src, tmp);
 5382   assert_different_registers(dst, tmp);
 5383 
 5384   if (UseCompactObjectHeaders) {
 5385     load_narrow_klass_compact(dst, src);
 5386     decode_klass_not_null(dst, tmp);
 5387   } else if (UseCompressedClassPointers) {
 5388     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5389     decode_klass_not_null(dst, tmp);
 5390   } else {
 5391     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5392   }
 5393 }
 5394 





 5395 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5396   assert(!UseCompactObjectHeaders, "not with compact headers");
 5397   assert_different_registers(src, tmp);
 5398   assert_different_registers(dst, tmp);
 5399   if (UseCompressedClassPointers) {
 5400     encode_klass_not_null(src, tmp);
 5401     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5402   } else {
 5403     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5404   }
 5405 }
 5406 
 5407 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5408   if (UseCompactObjectHeaders) {
 5409     assert(tmp != noreg, "need tmp");
 5410     assert_different_registers(klass, obj, tmp);
 5411     load_narrow_klass_compact(tmp, obj);
 5412     cmpl(klass, tmp);
 5413   } else if (UseCompressedClassPointers) {
 5414     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5415   } else {
 5416     cmpptr(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5417   }
 5418 }
 5419 
 5420 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5421   if (UseCompactObjectHeaders) {
 5422     assert(tmp2 != noreg, "need tmp2");
 5423     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5424     load_narrow_klass_compact(tmp1, obj1);
 5425     load_narrow_klass_compact(tmp2, obj2);
 5426     cmpl(tmp1, tmp2);
 5427   } else if (UseCompressedClassPointers) {
 5428     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5429     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5430   } else {
 5431     movptr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5432     cmpptr(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5433   }
 5434 }
 5435 
 5436 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5437                                     Register tmp1) {
 5438   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5439   decorators = AccessInternal::decorator_fixup(decorators, type);
 5440   bool as_raw = (decorators & AS_RAW) != 0;
 5441   if (as_raw) {
 5442     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5443   } else {
 5444     bs->load_at(this, decorators, type, dst, src, tmp1);
 5445   }
 5446 }
 5447 
 5448 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5449                                      Register tmp1, Register tmp2, Register tmp3) {
 5450   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5451   decorators = AccessInternal::decorator_fixup(decorators, type);
 5452   bool as_raw = (decorators & AS_RAW) != 0;
 5453   if (as_raw) {
 5454     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5455   } else {
 5456     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5457   }
 5458 }
 5459 








































 5460 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5461   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5462 }
 5463 
 5464 // Doesn't do verification, generates fixed size code
 5465 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5466   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5467 }
 5468 
 5469 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5470                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5471   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5472 }
 5473 
 5474 // Used for storing nulls.
 5475 void MacroAssembler::store_heap_oop_null(Address dst) {
 5476   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5477 }
 5478 
 5479 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5480   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5481   if (UseCompressedClassPointers) {
 5482     // Store to klass gap in destination
 5483     movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5484   }
 5485 }
 5486 
 5487 #ifdef ASSERT
 5488 void MacroAssembler::verify_heapbase(const char* msg) {
 5489   assert (UseCompressedOops, "should be compressed");
 5490   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5491   if (CheckCompressedOops) {
 5492     Label ok;
 5493     ExternalAddress src2(CompressedOops::base_addr());
 5494     const bool is_src2_reachable = reachable(src2);
 5495     if (!is_src2_reachable) {
 5496       push(rscratch1);  // cmpptr trashes rscratch1
 5497     }
 5498     cmpptr(r12_heapbase, src2, rscratch1);
 5499     jcc(Assembler::equal, ok);
 5500     STOP(msg);
 5501     bind(ok);
 5502     if (!is_src2_reachable) {
 5503       pop(rscratch1);
 5504     }
 5505   }
 5506 }
 5507 #endif
 5508 
 5509 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5510 void MacroAssembler::encode_heap_oop(Register r) {
 5511 #ifdef ASSERT
 5512   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5513 #endif
 5514   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5515   if (CompressedOops::base() == nullptr) {
 5516     if (CompressedOops::shift() != 0) {
 5517       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5518       shrq(r, LogMinObjAlignmentInBytes);
 5519     }
 5520     return;
 5521   }
 5522   testq(r, r);
 5523   cmovq(Assembler::equal, r, r12_heapbase);
 5524   subq(r, r12_heapbase);
 5525   shrq(r, LogMinObjAlignmentInBytes);
 5526 }
 5527 
 5528 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5529 #ifdef ASSERT
 5530   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5531   if (CheckCompressedOops) {
 5532     Label ok;
 5533     testq(r, r);
 5534     jcc(Assembler::notEqual, ok);
 5535     STOP("null oop passed to encode_heap_oop_not_null");
 5536     bind(ok);
 5537   }
 5538 #endif
 5539   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5540   if (CompressedOops::base() != nullptr) {
 5541     subq(r, r12_heapbase);
 5542   }
 5543   if (CompressedOops::shift() != 0) {
 5544     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5545     shrq(r, LogMinObjAlignmentInBytes);
 5546   }
 5547 }
 5548 
 5549 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5550 #ifdef ASSERT
 5551   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5552   if (CheckCompressedOops) {
 5553     Label ok;
 5554     testq(src, src);
 5555     jcc(Assembler::notEqual, ok);
 5556     STOP("null oop passed to encode_heap_oop_not_null2");
 5557     bind(ok);
 5558   }
 5559 #endif
 5560   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5561   if (dst != src) {
 5562     movq(dst, src);
 5563   }
 5564   if (CompressedOops::base() != nullptr) {
 5565     subq(dst, r12_heapbase);
 5566   }
 5567   if (CompressedOops::shift() != 0) {
 5568     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5569     shrq(dst, LogMinObjAlignmentInBytes);
 5570   }
 5571 }
 5572 
 5573 void  MacroAssembler::decode_heap_oop(Register r) {
 5574 #ifdef ASSERT
 5575   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5576 #endif
 5577   if (CompressedOops::base() == nullptr) {
 5578     if (CompressedOops::shift() != 0) {
 5579       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5580       shlq(r, LogMinObjAlignmentInBytes);
 5581     }
 5582   } else {
 5583     Label done;
 5584     shlq(r, LogMinObjAlignmentInBytes);
 5585     jccb(Assembler::equal, done);
 5586     addq(r, r12_heapbase);
 5587     bind(done);
 5588   }
 5589   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5590 }
 5591 
 5592 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5593   // Note: it will change flags
 5594   assert (UseCompressedOops, "should only be used for compressed headers");
 5595   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5596   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5597   // vtableStubs also counts instructions in pd_code_size_limit.
 5598   // Also do not verify_oop as this is called by verify_oop.
 5599   if (CompressedOops::shift() != 0) {
 5600     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5601     shlq(r, LogMinObjAlignmentInBytes);
 5602     if (CompressedOops::base() != nullptr) {
 5603       addq(r, r12_heapbase);
 5604     }
 5605   } else {
 5606     assert (CompressedOops::base() == nullptr, "sanity");
 5607   }
 5608 }
 5609 
 5610 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5611   // Note: it will change flags
 5612   assert (UseCompressedOops, "should only be used for compressed headers");
 5613   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5614   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5615   // vtableStubs also counts instructions in pd_code_size_limit.
 5616   // Also do not verify_oop as this is called by verify_oop.
 5617   if (CompressedOops::shift() != 0) {
 5618     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5619     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5620       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5621     } else {
 5622       if (dst != src) {
 5623         movq(dst, src);
 5624       }
 5625       shlq(dst, LogMinObjAlignmentInBytes);
 5626       if (CompressedOops::base() != nullptr) {
 5627         addq(dst, r12_heapbase);
 5628       }
 5629     }
 5630   } else {
 5631     assert (CompressedOops::base() == nullptr, "sanity");
 5632     if (dst != src) {
 5633       movq(dst, src);
 5634     }
 5635   }
 5636 }
 5637 
 5638 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5639   BLOCK_COMMENT("encode_klass_not_null {");
 5640   assert_different_registers(r, tmp);
 5641   if (CompressedKlassPointers::base() != nullptr) {
 5642     if (AOTCodeCache::is_on_for_dump()) {
 5643       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5644     } else {
 5645       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5646     }
 5647     subq(r, tmp);
 5648   }
 5649   if (CompressedKlassPointers::shift() != 0) {
 5650     shrq(r, CompressedKlassPointers::shift());
 5651   }
 5652   BLOCK_COMMENT("} encode_klass_not_null");
 5653 }
 5654 
 5655 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5656   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5657   assert_different_registers(src, dst);
 5658   if (CompressedKlassPointers::base() != nullptr) {
 5659     movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5660     addq(dst, src);
 5661   } else {
 5662     movptr(dst, src);
 5663   }
 5664   if (CompressedKlassPointers::shift() != 0) {
 5665     shrq(dst, CompressedKlassPointers::shift());
 5666   }
 5667   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5668 }
 5669 
 5670 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5671   BLOCK_COMMENT("decode_klass_not_null {");
 5672   assert_different_registers(r, tmp);
 5673   // Note: it will change flags
 5674   assert(UseCompressedClassPointers, "should only be used for compressed headers");
 5675   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5676   // vtableStubs also counts instructions in pd_code_size_limit.
 5677   // Also do not verify_oop as this is called by verify_oop.
 5678   if (CompressedKlassPointers::shift() != 0) {
 5679     shlq(r, CompressedKlassPointers::shift());
 5680   }
 5681   if (CompressedKlassPointers::base() != nullptr) {
 5682     if (AOTCodeCache::is_on_for_dump()) {
 5683       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5684     } else {
 5685       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5686     }
 5687     addq(r, tmp);
 5688   }
 5689   BLOCK_COMMENT("} decode_klass_not_null");
 5690 }
 5691 
 5692 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5693   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 5694   assert_different_registers(src, dst);
 5695   // Note: it will change flags
 5696   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5697   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5698   // vtableStubs also counts instructions in pd_code_size_limit.
 5699   // Also do not verify_oop as this is called by verify_oop.
 5700 
 5701   if (CompressedKlassPointers::base() == nullptr &&
 5702       CompressedKlassPointers::shift() == 0) {
 5703     // The best case scenario is that there is no base or shift. Then it is already
 5704     // a pointer that needs nothing but a register rename.
 5705     movl(dst, src);
 5706   } else {
 5707     if (CompressedKlassPointers::shift() <= Address::times_8) {
 5708       if (CompressedKlassPointers::base() != nullptr) {
 5709         movptr(dst, (intptr_t)CompressedKlassPointers::base());
 5710       } else {
 5711         xorq(dst, dst);
 5712       }
 5713       if (CompressedKlassPointers::shift() != 0) {
 5714         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 5715         leaq(dst, Address(dst, src, Address::times_8, 0));
 5716       } else {
 5717         addq(dst, src);
 5718       }
 5719     } else {
 5720       if (CompressedKlassPointers::base() != nullptr) {
 5721         const intptr_t base_right_shifted =
 5722             (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 5723         movptr(dst, base_right_shifted);
 5724       } else {
 5725         xorq(dst, dst);
 5726       }
 5727       addq(dst, src);
 5728       shlq(dst, CompressedKlassPointers::shift());
 5729     }
 5730   }
 5731   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 5732 }
 5733 
 5734 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5735   assert (UseCompressedOops, "should only be used for compressed headers");
 5736   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5737   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5738   int oop_index = oop_recorder()->find_index(obj);
 5739   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5740   mov_narrow_oop(dst, oop_index, rspec);
 5741 }
 5742 
 5743 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5744   assert (UseCompressedOops, "should only be used for compressed headers");
 5745   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5746   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5747   int oop_index = oop_recorder()->find_index(obj);
 5748   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5749   mov_narrow_oop(dst, oop_index, rspec);
 5750 }
 5751 
 5752 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5753   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5754   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5755   int klass_index = oop_recorder()->find_index(k);
 5756   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5757   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5758 }
 5759 
 5760 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 5761   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5762   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5763   int klass_index = oop_recorder()->find_index(k);
 5764   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5765   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5766 }
 5767 
 5768 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 5769   assert (UseCompressedOops, "should only be used for compressed headers");
 5770   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5771   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5772   int oop_index = oop_recorder()->find_index(obj);
 5773   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5774   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5775 }
 5776 
 5777 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 5778   assert (UseCompressedOops, "should only be used for compressed headers");
 5779   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5780   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5781   int oop_index = oop_recorder()->find_index(obj);
 5782   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5783   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5784 }
 5785 
 5786 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 5787   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5788   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5789   int klass_index = oop_recorder()->find_index(k);
 5790   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5791   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5792 }
 5793 
 5794 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 5795   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 5796   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5797   int klass_index = oop_recorder()->find_index(k);
 5798   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5799   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5800 }
 5801 
 5802 void MacroAssembler::reinit_heapbase() {
 5803   if (UseCompressedOops) {
 5804     if (Universe::heap() != nullptr) {
 5805       if (CompressedOops::base() == nullptr) {
 5806         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 5807       } else {
 5808         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 5809       }
 5810     } else {
 5811       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 5812     }
 5813   }
 5814 }
 5815 
















































































































































































































































































































































































































































































 5816 #if COMPILER2_OR_JVMCI
 5817 
 5818 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 5819 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5820   // cnt - number of qwords (8-byte words).
 5821   // base - start address, qword aligned.
 5822   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 5823   bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
 5824   if (use64byteVector) {
 5825     vpxor(xtmp, xtmp, xtmp, AVX_512bit);
 5826   } else if (MaxVectorSize >= 32) {
 5827     vpxor(xtmp, xtmp, xtmp, AVX_256bit);


 5828   } else {
 5829     pxor(xtmp, xtmp);

 5830   }
 5831   jmp(L_zero_64_bytes);
 5832 
 5833   BIND(L_loop);
 5834   if (MaxVectorSize >= 32) {
 5835     fill64(base, 0, xtmp, use64byteVector);
 5836   } else {
 5837     movdqu(Address(base,  0), xtmp);
 5838     movdqu(Address(base, 16), xtmp);
 5839     movdqu(Address(base, 32), xtmp);
 5840     movdqu(Address(base, 48), xtmp);
 5841   }
 5842   addptr(base, 64);
 5843 
 5844   BIND(L_zero_64_bytes);
 5845   subptr(cnt, 8);
 5846   jccb(Assembler::greaterEqual, L_loop);
 5847 
 5848   // Copy trailing 64 bytes
 5849   if (use64byteVector) {
 5850     addptr(cnt, 8);
 5851     jccb(Assembler::equal, L_end);
 5852     fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
 5853     jmp(L_end);
 5854   } else {
 5855     addptr(cnt, 4);
 5856     jccb(Assembler::less, L_tail);
 5857     if (MaxVectorSize >= 32) {
 5858       vmovdqu(Address(base, 0), xtmp);
 5859     } else {
 5860       movdqu(Address(base,  0), xtmp);
 5861       movdqu(Address(base, 16), xtmp);
 5862     }
 5863   }
 5864   addptr(base, 32);
 5865   subptr(cnt, 4);
 5866 
 5867   BIND(L_tail);
 5868   addptr(cnt, 4);
 5869   jccb(Assembler::lessEqual, L_end);
 5870   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 5871     fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
 5872   } else {
 5873     decrement(cnt);
 5874 
 5875     BIND(L_sloop);
 5876     movq(Address(base, 0), xtmp);
 5877     addptr(base, 8);
 5878     decrement(cnt);
 5879     jccb(Assembler::greaterEqual, L_sloop);
 5880   }
 5881   BIND(L_end);
 5882 }
 5883 
 5884 // Clearing constant sized memory using YMM/ZMM registers.
 5885 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5886   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 5887   bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
 5888 
 5889   int vector64_count = (cnt & (~0x7)) >> 3;
 5890   cnt = cnt & 0x7;
 5891   const int fill64_per_loop = 4;
 5892   const int max_unrolled_fill64 = 8;
 5893 
 5894   // 64 byte initialization loop.
 5895   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 5896   int start64 = 0;
 5897   if (vector64_count > max_unrolled_fill64) {
 5898     Label LOOP;
 5899     Register index = rtmp;
 5900 
 5901     start64 = vector64_count - (vector64_count % fill64_per_loop);
 5902 
 5903     movl(index, 0);
 5904     BIND(LOOP);
 5905     for (int i = 0; i < fill64_per_loop; i++) {
 5906       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 5907     }
 5908     addl(index, fill64_per_loop * 64);
 5909     cmpl(index, start64 * 64);
 5910     jccb(Assembler::less, LOOP);
 5911   }
 5912   for (int i = start64; i < vector64_count; i++) {
 5913     fill64(base, i * 64, xtmp, use64byteVector);
 5914   }
 5915 
 5916   // Clear remaining 64 byte tail.
 5917   int disp = vector64_count * 64;
 5918   if (cnt) {
 5919     switch (cnt) {
 5920       case 1:
 5921         movq(Address(base, disp), xtmp);
 5922         break;
 5923       case 2:
 5924         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 5925         break;
 5926       case 3:
 5927         movl(rtmp, 0x7);
 5928         kmovwl(mask, rtmp);
 5929         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 5930         break;
 5931       case 4:
 5932         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5933         break;
 5934       case 5:
 5935         if (use64byteVector) {
 5936           movl(rtmp, 0x1F);
 5937           kmovwl(mask, rtmp);
 5938           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5939         } else {
 5940           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5941           movq(Address(base, disp + 32), xtmp);
 5942         }
 5943         break;
 5944       case 6:
 5945         if (use64byteVector) {
 5946           movl(rtmp, 0x3F);
 5947           kmovwl(mask, rtmp);
 5948           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5949         } else {
 5950           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5951           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 5952         }
 5953         break;
 5954       case 7:
 5955         if (use64byteVector) {
 5956           movl(rtmp, 0x7F);
 5957           kmovwl(mask, rtmp);
 5958           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5959         } else {
 5960           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5961           movl(rtmp, 0x7);
 5962           kmovwl(mask, rtmp);
 5963           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 5964         }
 5965         break;
 5966       default:
 5967         fatal("Unexpected length : %d\n",cnt);
 5968         break;
 5969     }
 5970   }
 5971 }
 5972 
 5973 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
 5974                                bool is_large, KRegister mask) {
 5975   // cnt      - number of qwords (8-byte words).
 5976   // base     - start address, qword aligned.
 5977   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 5978   assert(base==rdi, "base register must be edi for rep stos");
 5979   assert(tmp==rax,   "tmp register must be eax for rep stos");
 5980   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 5981   assert(InitArrayShortSize % BytesPerLong == 0,
 5982     "InitArrayShortSize should be the multiple of BytesPerLong");
 5983 
 5984   Label DONE;
 5985   if (!is_large || !UseXMMForObjInit) {
 5986     xorptr(tmp, tmp);
 5987   }
 5988 
 5989   if (!is_large) {
 5990     Label LOOP, LONG;
 5991     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 5992     jccb(Assembler::greater, LONG);
 5993 
 5994     decrement(cnt);
 5995     jccb(Assembler::negative, DONE); // Zero length
 5996 
 5997     // Use individual pointer-sized stores for small counts:
 5998     BIND(LOOP);
 5999     movptr(Address(base, cnt, Address::times_ptr), tmp);
 6000     decrement(cnt);
 6001     jccb(Assembler::greaterEqual, LOOP);
 6002     jmpb(DONE);
 6003 
 6004     BIND(LONG);
 6005   }
 6006 
 6007   // Use longer rep-prefixed ops for non-small counts:
 6008   if (UseFastStosb) {
 6009     shlptr(cnt, 3); // convert to number of bytes
 6010     rep_stosb();
 6011   } else if (UseXMMForObjInit) {
 6012     xmm_clear_mem(base, cnt, tmp, xtmp, mask);
 6013   } else {
 6014     rep_stos();
 6015   }
 6016 
 6017   BIND(DONE);
 6018 }
 6019 
 6020 #endif //COMPILER2_OR_JVMCI
 6021 
 6022 
 6023 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6024                                    Register to, Register value, Register count,
 6025                                    Register rtmp, XMMRegister xtmp) {
 6026   ShortBranchVerifier sbv(this);
 6027   assert_different_registers(to, value, count, rtmp);
 6028   Label L_exit;
 6029   Label L_fill_2_bytes, L_fill_4_bytes;
 6030 
 6031 #if defined(COMPILER2)
 6032   if(MaxVectorSize >=32 &&
 6033      VM_Version::supports_avx512vlbw() &&
 6034      VM_Version::supports_bmi2()) {
 6035     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6036     return;
 6037   }
 6038 #endif
 6039 
 6040   int shift = -1;
 6041   switch (t) {
 6042     case T_BYTE:
 6043       shift = 2;
 6044       break;
 6045     case T_SHORT:
 6046       shift = 1;
 6047       break;
 6048     case T_INT:
 6049       shift = 0;
 6050       break;
 6051     default: ShouldNotReachHere();
 6052   }
 6053 
 6054   if (t == T_BYTE) {
 6055     andl(value, 0xff);
 6056     movl(rtmp, value);
 6057     shll(rtmp, 8);
 6058     orl(value, rtmp);
 6059   }
 6060   if (t == T_SHORT) {
 6061     andl(value, 0xffff);
 6062   }
 6063   if (t == T_BYTE || t == T_SHORT) {
 6064     movl(rtmp, value);
 6065     shll(rtmp, 16);
 6066     orl(value, rtmp);
 6067   }
 6068 
 6069   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6070   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6071   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6072     Label L_skip_align2;
 6073     // align source address at 4 bytes address boundary
 6074     if (t == T_BYTE) {
 6075       Label L_skip_align1;
 6076       // One byte misalignment happens only for byte arrays
 6077       testptr(to, 1);
 6078       jccb(Assembler::zero, L_skip_align1);
 6079       movb(Address(to, 0), value);
 6080       increment(to);
 6081       decrement(count);
 6082       BIND(L_skip_align1);
 6083     }
 6084     // Two bytes misalignment happens only for byte and short (char) arrays
 6085     testptr(to, 2);
 6086     jccb(Assembler::zero, L_skip_align2);
 6087     movw(Address(to, 0), value);
 6088     addptr(to, 2);
 6089     subptr(count, 1<<(shift-1));
 6090     BIND(L_skip_align2);
 6091   }
 6092   {
 6093     Label L_fill_32_bytes;
 6094     if (!UseUnalignedLoadStores) {
 6095       // align to 8 bytes, we know we are 4 byte aligned to start
 6096       testptr(to, 4);
 6097       jccb(Assembler::zero, L_fill_32_bytes);
 6098       movl(Address(to, 0), value);
 6099       addptr(to, 4);
 6100       subptr(count, 1<<shift);
 6101     }
 6102     BIND(L_fill_32_bytes);
 6103     {
 6104       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6105       movdl(xtmp, value);
 6106       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6107         Label L_check_fill_32_bytes;
 6108         if (UseAVX > 2) {
 6109           // Fill 64-byte chunks
 6110           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6111 
 6112           // If number of bytes to fill < VM_Version::avx3_threshold(), perform fill using AVX2
 6113           cmpptr(count, VM_Version::avx3_threshold());
 6114           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6115 
 6116           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6117 
 6118           subptr(count, 16 << shift);
 6119           jcc(Assembler::less, L_check_fill_32_bytes);
 6120           align(16);
 6121 
 6122           BIND(L_fill_64_bytes_loop_avx3);
 6123           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6124           addptr(to, 64);
 6125           subptr(count, 16 << shift);
 6126           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6127           jmpb(L_check_fill_32_bytes);
 6128 
 6129           BIND(L_check_fill_64_bytes_avx2);
 6130         }
 6131         // Fill 64-byte chunks
 6132         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6133 
 6134         subptr(count, 16 << shift);
 6135         jcc(Assembler::less, L_check_fill_32_bytes);
 6136 
 6137         // align data for 64-byte chunks
 6138         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6139         if (EnableX86ECoreOpts) {
 6140             // align 'big' arrays to cache lines to minimize split_stores
 6141             cmpptr(count, 96 << shift);
 6142             jcc(Assembler::below, L_fill_64_bytes_loop);
 6143 
 6144             // Find the bytes needed for alignment
 6145             movptr(rtmp, to);
 6146             andptr(rtmp, 0x1c);
 6147             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6148             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6149             addptr(rtmp, 32);
 6150             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6151             subptr(count, rtmp);    // adjust count by number of elements
 6152 
 6153             align(16);
 6154             BIND(L_align_64_bytes_loop);
 6155             movdl(Address(to, 0), xtmp);
 6156             addptr(to, 4);
 6157             subptr(rtmp, 1 << shift);
 6158             jcc(Assembler::greater, L_align_64_bytes_loop);
 6159         }
 6160 
 6161         align(16);
 6162         BIND(L_fill_64_bytes_loop);
 6163         vmovdqu(Address(to, 0), xtmp);
 6164         vmovdqu(Address(to, 32), xtmp);
 6165         addptr(to, 64);
 6166         subptr(count, 16 << shift);
 6167         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6168 
 6169         align(16);
 6170         BIND(L_check_fill_32_bytes);
 6171         addptr(count, 8 << shift);
 6172         jccb(Assembler::less, L_check_fill_8_bytes);
 6173         vmovdqu(Address(to, 0), xtmp);
 6174         addptr(to, 32);
 6175         subptr(count, 8 << shift);
 6176 
 6177         BIND(L_check_fill_8_bytes);
 6178         // clean upper bits of YMM registers
 6179         movdl(xtmp, value);
 6180         pshufd(xtmp, xtmp, 0);
 6181       } else {
 6182         // Fill 32-byte chunks
 6183         pshufd(xtmp, xtmp, 0);
 6184 
 6185         subptr(count, 8 << shift);
 6186         jcc(Assembler::less, L_check_fill_8_bytes);
 6187         align(16);
 6188 
 6189         BIND(L_fill_32_bytes_loop);
 6190 
 6191         if (UseUnalignedLoadStores) {
 6192           movdqu(Address(to, 0), xtmp);
 6193           movdqu(Address(to, 16), xtmp);
 6194         } else {
 6195           movq(Address(to, 0), xtmp);
 6196           movq(Address(to, 8), xtmp);
 6197           movq(Address(to, 16), xtmp);
 6198           movq(Address(to, 24), xtmp);
 6199         }
 6200 
 6201         addptr(to, 32);
 6202         subptr(count, 8 << shift);
 6203         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6204 
 6205         BIND(L_check_fill_8_bytes);
 6206       }
 6207       addptr(count, 8 << shift);
 6208       jccb(Assembler::zero, L_exit);
 6209       jmpb(L_fill_8_bytes);
 6210 
 6211       //
 6212       // length is too short, just fill qwords
 6213       //
 6214       align(16);
 6215       BIND(L_fill_8_bytes_loop);
 6216       movq(Address(to, 0), xtmp);
 6217       addptr(to, 8);
 6218       BIND(L_fill_8_bytes);
 6219       subptr(count, 1 << (shift + 1));
 6220       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6221     }
 6222   }
 6223 
 6224   Label L_fill_4_bytes_loop;
 6225   testl(count, 1 << shift);
 6226   jccb(Assembler::zero, L_fill_2_bytes);
 6227 
 6228   align(16);
 6229   BIND(L_fill_4_bytes_loop);
 6230   movl(Address(to, 0), value);
 6231   addptr(to, 4);
 6232 
 6233   BIND(L_fill_4_bytes);
 6234   subptr(count, 1 << shift);
 6235   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 6236 
 6237   if (t == T_BYTE || t == T_SHORT) {
 6238     Label L_fill_byte;
 6239     BIND(L_fill_2_bytes);
 6240     // fill trailing 2 bytes
 6241     testl(count, 1<<(shift-1));
 6242     jccb(Assembler::zero, L_fill_byte);
 6243     movw(Address(to, 0), value);
 6244     if (t == T_BYTE) {
 6245       addptr(to, 2);
 6246       BIND(L_fill_byte);
 6247       // fill trailing byte
 6248       testl(count, 1);
 6249       jccb(Assembler::zero, L_exit);
 6250       movb(Address(to, 0), value);
 6251     } else {
 6252       BIND(L_fill_byte);
 6253     }
 6254   } else {
 6255     BIND(L_fill_2_bytes);
 6256   }
 6257   BIND(L_exit);
 6258 }
 6259 
 6260 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6261   switch(type) {
 6262     case T_BYTE:
 6263     case T_BOOLEAN:
 6264       evpbroadcastb(dst, src, vector_len);
 6265       break;
 6266     case T_SHORT:
 6267     case T_CHAR:
 6268       evpbroadcastw(dst, src, vector_len);
 6269       break;
 6270     case T_INT:
 6271     case T_FLOAT:
 6272       evpbroadcastd(dst, src, vector_len);
 6273       break;
 6274     case T_LONG:
 6275     case T_DOUBLE:
 6276       evpbroadcastq(dst, src, vector_len);
 6277       break;
 6278     default:
 6279       fatal("Unhandled type : %s", type2name(type));
 6280       break;
 6281   }
 6282 }
 6283 
 6284 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 6285 //
 6286 // @IntrinsicCandidate
 6287 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 6288 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6289 //     int i = 0;
 6290 //     for (; i < len; i++) {
 6291 //         char c = sa[sp++];
 6292 //         if (c > '\u00FF')
 6293 //             break;
 6294 //         da[dp++] = (byte) c;
 6295 //     }
 6296 //     return i;
 6297 // }
 6298 //
 6299 // @IntrinsicCandidate
 6300 // int java.lang.StringCoding.encodeISOArray0(
 6301 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 6302 //   int i = 0;
 6303 //   for (; i < len; i++) {
 6304 //     char c = StringUTF16.getChar(sa, sp++);
 6305 //     if (c > '\u00FF')
 6306 //       break;
 6307 //     da[dp++] = (byte) c;
 6308 //   }
 6309 //   return i;
 6310 // }
 6311 //
 6312 // @IntrinsicCandidate
 6313 // int java.lang.StringCoding.encodeAsciiArray0(
 6314 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6315 //   int i = 0;
 6316 //   for (; i < len; i++) {
 6317 //     char c = sa[sp++];
 6318 //     if (c >= '\u0080')
 6319 //       break;
 6320 //     da[dp++] = (byte) c;
 6321 //   }
 6322 //   return i;
 6323 // }
 6324 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 6325   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 6326   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 6327   Register tmp5, Register result, bool ascii) {
 6328 
 6329   // rsi: src
 6330   // rdi: dst
 6331   // rdx: len
 6332   // rcx: tmp5
 6333   // rax: result
 6334   ShortBranchVerifier sbv(this);
 6335   assert_different_registers(src, dst, len, tmp5, result);
 6336   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 6337 
 6338   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 6339   int short_mask = ascii ? 0xff80 : 0xff00;
 6340 
 6341   // set result
 6342   xorl(result, result);
 6343   // check for zero length
 6344   testl(len, len);
 6345   jcc(Assembler::zero, L_done);
 6346 
 6347   movl(result, len);
 6348 
 6349   // Setup pointers
 6350   lea(src, Address(src, len, Address::times_2)); // char[]
 6351   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 6352   negptr(len);
 6353 
 6354   if (UseSSE42Intrinsics || UseAVX >= 2) {
 6355     Label L_copy_8_chars, L_copy_8_chars_exit;
 6356     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 6357 
 6358     if (UseAVX >= 2) {
 6359       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 6360       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6361       movdl(tmp1Reg, tmp5);
 6362       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 6363       jmp(L_chars_32_check);
 6364 
 6365       bind(L_copy_32_chars);
 6366       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 6367       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 6368       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6369       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6370       jccb(Assembler::notZero, L_copy_32_chars_exit);
 6371       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6372       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 6373       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 6374 
 6375       bind(L_chars_32_check);
 6376       addptr(len, 32);
 6377       jcc(Assembler::lessEqual, L_copy_32_chars);
 6378 
 6379       bind(L_copy_32_chars_exit);
 6380       subptr(len, 16);
 6381       jccb(Assembler::greater, L_copy_16_chars_exit);
 6382 
 6383     } else if (UseSSE42Intrinsics) {
 6384       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6385       movdl(tmp1Reg, tmp5);
 6386       pshufd(tmp1Reg, tmp1Reg, 0);
 6387       jmpb(L_chars_16_check);
 6388     }
 6389 
 6390     bind(L_copy_16_chars);
 6391     if (UseAVX >= 2) {
 6392       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 6393       vptest(tmp2Reg, tmp1Reg);
 6394       jcc(Assembler::notZero, L_copy_16_chars_exit);
 6395       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 6396       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 6397     } else {
 6398       if (UseAVX > 0) {
 6399         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6400         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6401         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 6402       } else {
 6403         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6404         por(tmp2Reg, tmp3Reg);
 6405         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6406         por(tmp2Reg, tmp4Reg);
 6407       }
 6408       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6409       jccb(Assembler::notZero, L_copy_16_chars_exit);
 6410       packuswb(tmp3Reg, tmp4Reg);
 6411     }
 6412     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 6413 
 6414     bind(L_chars_16_check);
 6415     addptr(len, 16);
 6416     jcc(Assembler::lessEqual, L_copy_16_chars);
 6417 
 6418     bind(L_copy_16_chars_exit);
 6419     if (UseAVX >= 2) {
 6420       // clean upper bits of YMM registers
 6421       vpxor(tmp2Reg, tmp2Reg);
 6422       vpxor(tmp3Reg, tmp3Reg);
 6423       vpxor(tmp4Reg, tmp4Reg);
 6424       movdl(tmp1Reg, tmp5);
 6425       pshufd(tmp1Reg, tmp1Reg, 0);
 6426     }
 6427     subptr(len, 8);
 6428     jccb(Assembler::greater, L_copy_8_chars_exit);
 6429 
 6430     bind(L_copy_8_chars);
 6431     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 6432     ptest(tmp3Reg, tmp1Reg);
 6433     jccb(Assembler::notZero, L_copy_8_chars_exit);
 6434     packuswb(tmp3Reg, tmp1Reg);
 6435     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 6436     addptr(len, 8);
 6437     jccb(Assembler::lessEqual, L_copy_8_chars);
 6438 
 6439     bind(L_copy_8_chars_exit);
 6440     subptr(len, 8);
 6441     jccb(Assembler::zero, L_done);
 6442   }
 6443 
 6444   bind(L_copy_1_char);
 6445   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 6446   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 6447   jccb(Assembler::notZero, L_copy_1_char_exit);
 6448   movb(Address(dst, len, Address::times_1, 0), tmp5);
 6449   addptr(len, 1);
 6450   jccb(Assembler::less, L_copy_1_char);
 6451 
 6452   bind(L_copy_1_char_exit);
 6453   addptr(result, len); // len is negative count of not processed elements
 6454 
 6455   bind(L_done);
 6456 }
 6457 
 6458 /**
 6459  * Helper for multiply_to_len().
 6460  */
 6461 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 6462   addq(dest_lo, src1);
 6463   adcq(dest_hi, 0);
 6464   addq(dest_lo, src2);
 6465   adcq(dest_hi, 0);
 6466 }
 6467 
 6468 /**
 6469  * Multiply 64 bit by 64 bit first loop.
 6470  */
 6471 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 6472                                            Register y, Register y_idx, Register z,
 6473                                            Register carry, Register product,
 6474                                            Register idx, Register kdx) {
 6475   //
 6476   //  jlong carry, x[], y[], z[];
 6477   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6478   //    huge_128 product = y[idx] * x[xstart] + carry;
 6479   //    z[kdx] = (jlong)product;
 6480   //    carry  = (jlong)(product >>> 64);
 6481   //  }
 6482   //  z[xstart] = carry;
 6483   //
 6484 
 6485   Label L_first_loop, L_first_loop_exit;
 6486   Label L_one_x, L_one_y, L_multiply;
 6487 
 6488   decrementl(xstart);
 6489   jcc(Assembler::negative, L_one_x);
 6490 
 6491   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6492   rorq(x_xstart, 32); // convert big-endian to little-endian
 6493 
 6494   bind(L_first_loop);
 6495   decrementl(idx);
 6496   jcc(Assembler::negative, L_first_loop_exit);
 6497   decrementl(idx);
 6498   jcc(Assembler::negative, L_one_y);
 6499   movq(y_idx, Address(y, idx, Address::times_4,  0));
 6500   rorq(y_idx, 32); // convert big-endian to little-endian
 6501   bind(L_multiply);
 6502   movq(product, x_xstart);
 6503   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 6504   addq(product, carry);
 6505   adcq(rdx, 0);
 6506   subl(kdx, 2);
 6507   movl(Address(z, kdx, Address::times_4,  4), product);
 6508   shrq(product, 32);
 6509   movl(Address(z, kdx, Address::times_4,  0), product);
 6510   movq(carry, rdx);
 6511   jmp(L_first_loop);
 6512 
 6513   bind(L_one_y);
 6514   movl(y_idx, Address(y,  0));
 6515   jmp(L_multiply);
 6516 
 6517   bind(L_one_x);
 6518   movl(x_xstart, Address(x,  0));
 6519   jmp(L_first_loop);
 6520 
 6521   bind(L_first_loop_exit);
 6522 }
 6523 
 6524 /**
 6525  * Multiply 64 bit by 64 bit and add 128 bit.
 6526  */
 6527 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 6528                                             Register yz_idx, Register idx,
 6529                                             Register carry, Register product, int offset) {
 6530   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 6531   //     z[kdx] = (jlong)product;
 6532 
 6533   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 6534   rorq(yz_idx, 32); // convert big-endian to little-endian
 6535   movq(product, x_xstart);
 6536   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 6537   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 6538   rorq(yz_idx, 32); // convert big-endian to little-endian
 6539 
 6540   add2_with_carry(rdx, product, carry, yz_idx);
 6541 
 6542   movl(Address(z, idx, Address::times_4,  offset+4), product);
 6543   shrq(product, 32);
 6544   movl(Address(z, idx, Address::times_4,  offset), product);
 6545 
 6546 }
 6547 
 6548 /**
 6549  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 6550  */
 6551 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 6552                                              Register yz_idx, Register idx, Register jdx,
 6553                                              Register carry, Register product,
 6554                                              Register carry2) {
 6555   //   jlong carry, x[], y[], z[];
 6556   //   int kdx = ystart+1;
 6557   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6558   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 6559   //     z[kdx+idx+1] = (jlong)product;
 6560   //     jlong carry2  = (jlong)(product >>> 64);
 6561   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 6562   //     z[kdx+idx] = (jlong)product;
 6563   //     carry  = (jlong)(product >>> 64);
 6564   //   }
 6565   //   idx += 2;
 6566   //   if (idx > 0) {
 6567   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 6568   //     z[kdx+idx] = (jlong)product;
 6569   //     carry  = (jlong)(product >>> 64);
 6570   //   }
 6571   //
 6572 
 6573   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6574 
 6575   movl(jdx, idx);
 6576   andl(jdx, 0xFFFFFFFC);
 6577   shrl(jdx, 2);
 6578 
 6579   bind(L_third_loop);
 6580   subl(jdx, 1);
 6581   jcc(Assembler::negative, L_third_loop_exit);
 6582   subl(idx, 4);
 6583 
 6584   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 6585   movq(carry2, rdx);
 6586 
 6587   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 6588   movq(carry, rdx);
 6589   jmp(L_third_loop);
 6590 
 6591   bind (L_third_loop_exit);
 6592 
 6593   andl (idx, 0x3);
 6594   jcc(Assembler::zero, L_post_third_loop_done);
 6595 
 6596   Label L_check_1;
 6597   subl(idx, 2);
 6598   jcc(Assembler::negative, L_check_1);
 6599 
 6600   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 6601   movq(carry, rdx);
 6602 
 6603   bind (L_check_1);
 6604   addl (idx, 0x2);
 6605   andl (idx, 0x1);
 6606   subl(idx, 1);
 6607   jcc(Assembler::negative, L_post_third_loop_done);
 6608 
 6609   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 6610   movq(product, x_xstart);
 6611   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 6612   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 6613 
 6614   add2_with_carry(rdx, product, yz_idx, carry);
 6615 
 6616   movl(Address(z, idx, Address::times_4,  0), product);
 6617   shrq(product, 32);
 6618 
 6619   shlq(rdx, 32);
 6620   orq(product, rdx);
 6621   movq(carry, product);
 6622 
 6623   bind(L_post_third_loop_done);
 6624 }
 6625 
 6626 /**
 6627  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 6628  *
 6629  */
 6630 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 6631                                                   Register carry, Register carry2,
 6632                                                   Register idx, Register jdx,
 6633                                                   Register yz_idx1, Register yz_idx2,
 6634                                                   Register tmp, Register tmp3, Register tmp4) {
 6635   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 6636 
 6637   //   jlong carry, x[], y[], z[];
 6638   //   int kdx = ystart+1;
 6639   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6640   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 6641   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 6642   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 6643   //     carry  = (jlong)(tmp4 >>> 64);
 6644   //     z[kdx+idx+1] = (jlong)tmp3;
 6645   //     z[kdx+idx] = (jlong)tmp4;
 6646   //   }
 6647   //   idx += 2;
 6648   //   if (idx > 0) {
 6649   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 6650   //     z[kdx+idx] = (jlong)yz_idx1;
 6651   //     carry  = (jlong)(yz_idx1 >>> 64);
 6652   //   }
 6653   //
 6654 
 6655   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6656 
 6657   movl(jdx, idx);
 6658   andl(jdx, 0xFFFFFFFC);
 6659   shrl(jdx, 2);
 6660 
 6661   bind(L_third_loop);
 6662   subl(jdx, 1);
 6663   jcc(Assembler::negative, L_third_loop_exit);
 6664   subl(idx, 4);
 6665 
 6666   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 6667   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 6668   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 6669   rorxq(yz_idx2, yz_idx2, 32);
 6670 
 6671   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 6672   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 6673 
 6674   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 6675   rorxq(yz_idx1, yz_idx1, 32);
 6676   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6677   rorxq(yz_idx2, yz_idx2, 32);
 6678 
 6679   if (VM_Version::supports_adx()) {
 6680     adcxq(tmp3, carry);
 6681     adoxq(tmp3, yz_idx1);
 6682 
 6683     adcxq(tmp4, tmp);
 6684     adoxq(tmp4, yz_idx2);
 6685 
 6686     movl(carry, 0); // does not affect flags
 6687     adcxq(carry2, carry);
 6688     adoxq(carry2, carry);
 6689   } else {
 6690     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 6691     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 6692   }
 6693   movq(carry, carry2);
 6694 
 6695   movl(Address(z, idx, Address::times_4, 12), tmp3);
 6696   shrq(tmp3, 32);
 6697   movl(Address(z, idx, Address::times_4,  8), tmp3);
 6698 
 6699   movl(Address(z, idx, Address::times_4,  4), tmp4);
 6700   shrq(tmp4, 32);
 6701   movl(Address(z, idx, Address::times_4,  0), tmp4);
 6702 
 6703   jmp(L_third_loop);
 6704 
 6705   bind (L_third_loop_exit);
 6706 
 6707   andl (idx, 0x3);
 6708   jcc(Assembler::zero, L_post_third_loop_done);
 6709 
 6710   Label L_check_1;
 6711   subl(idx, 2);
 6712   jcc(Assembler::negative, L_check_1);
 6713 
 6714   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 6715   rorxq(yz_idx1, yz_idx1, 32);
 6716   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 6717   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6718   rorxq(yz_idx2, yz_idx2, 32);
 6719 
 6720   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 6721 
 6722   movl(Address(z, idx, Address::times_4,  4), tmp3);
 6723   shrq(tmp3, 32);
 6724   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6725   movq(carry, tmp4);
 6726 
 6727   bind (L_check_1);
 6728   addl (idx, 0x2);
 6729   andl (idx, 0x1);
 6730   subl(idx, 1);
 6731   jcc(Assembler::negative, L_post_third_loop_done);
 6732   movl(tmp4, Address(y, idx, Address::times_4,  0));
 6733   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 6734   movl(tmp4, Address(z, idx, Address::times_4,  0));
 6735 
 6736   add2_with_carry(carry2, tmp3, tmp4, carry);
 6737 
 6738   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6739   shrq(tmp3, 32);
 6740 
 6741   shlq(carry2, 32);
 6742   orq(tmp3, carry2);
 6743   movq(carry, tmp3);
 6744 
 6745   bind(L_post_third_loop_done);
 6746 }
 6747 
 6748 /**
 6749  * Code for BigInteger::multiplyToLen() intrinsic.
 6750  *
 6751  * rdi: x
 6752  * rax: xlen
 6753  * rsi: y
 6754  * rcx: ylen
 6755  * r8:  z
 6756  * r11: tmp0
 6757  * r12: tmp1
 6758  * r13: tmp2
 6759  * r14: tmp3
 6760  * r15: tmp4
 6761  * rbx: tmp5
 6762  *
 6763  */
 6764 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 6765                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 6766   ShortBranchVerifier sbv(this);
 6767   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 6768 
 6769   push(tmp0);
 6770   push(tmp1);
 6771   push(tmp2);
 6772   push(tmp3);
 6773   push(tmp4);
 6774   push(tmp5);
 6775 
 6776   push(xlen);
 6777 
 6778   const Register idx = tmp1;
 6779   const Register kdx = tmp2;
 6780   const Register xstart = tmp3;
 6781 
 6782   const Register y_idx = tmp4;
 6783   const Register carry = tmp5;
 6784   const Register product  = xlen;
 6785   const Register x_xstart = tmp0;
 6786 
 6787   // First Loop.
 6788   //
 6789   //  final static long LONG_MASK = 0xffffffffL;
 6790   //  int xstart = xlen - 1;
 6791   //  int ystart = ylen - 1;
 6792   //  long carry = 0;
 6793   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6794   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 6795   //    z[kdx] = (int)product;
 6796   //    carry = product >>> 32;
 6797   //  }
 6798   //  z[xstart] = (int)carry;
 6799   //
 6800 
 6801   movl(idx, ylen);               // idx = ylen;
 6802   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 6803   xorq(carry, carry);            // carry = 0;
 6804 
 6805   Label L_done;
 6806 
 6807   movl(xstart, xlen);
 6808   decrementl(xstart);
 6809   jcc(Assembler::negative, L_done);
 6810 
 6811   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 6812 
 6813   Label L_second_loop;
 6814   testl(kdx, kdx);
 6815   jcc(Assembler::zero, L_second_loop);
 6816 
 6817   Label L_carry;
 6818   subl(kdx, 1);
 6819   jcc(Assembler::zero, L_carry);
 6820 
 6821   movl(Address(z, kdx, Address::times_4,  0), carry);
 6822   shrq(carry, 32);
 6823   subl(kdx, 1);
 6824 
 6825   bind(L_carry);
 6826   movl(Address(z, kdx, Address::times_4,  0), carry);
 6827 
 6828   // Second and third (nested) loops.
 6829   //
 6830   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 6831   //   carry = 0;
 6832   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 6833   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 6834   //                    (z[k] & LONG_MASK) + carry;
 6835   //     z[k] = (int)product;
 6836   //     carry = product >>> 32;
 6837   //   }
 6838   //   z[i] = (int)carry;
 6839   // }
 6840   //
 6841   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 6842 
 6843   const Register jdx = tmp1;
 6844 
 6845   bind(L_second_loop);
 6846   xorl(carry, carry);    // carry = 0;
 6847   movl(jdx, ylen);       // j = ystart+1
 6848 
 6849   subl(xstart, 1);       // i = xstart-1;
 6850   jcc(Assembler::negative, L_done);
 6851 
 6852   push (z);
 6853 
 6854   Label L_last_x;
 6855   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 6856   subl(xstart, 1);       // i = xstart-1;
 6857   jcc(Assembler::negative, L_last_x);
 6858 
 6859   if (UseBMI2Instructions) {
 6860     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 6861     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 6862   } else {
 6863     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6864     rorq(x_xstart, 32);  // convert big-endian to little-endian
 6865   }
 6866 
 6867   Label L_third_loop_prologue;
 6868   bind(L_third_loop_prologue);
 6869 
 6870   push (x);
 6871   push (xstart);
 6872   push (ylen);
 6873 
 6874 
 6875   if (UseBMI2Instructions) {
 6876     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 6877   } else { // !UseBMI2Instructions
 6878     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 6879   }
 6880 
 6881   pop(ylen);
 6882   pop(xlen);
 6883   pop(x);
 6884   pop(z);
 6885 
 6886   movl(tmp3, xlen);
 6887   addl(tmp3, 1);
 6888   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6889   subl(tmp3, 1);
 6890   jccb(Assembler::negative, L_done);
 6891 
 6892   shrq(carry, 32);
 6893   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6894   jmp(L_second_loop);
 6895 
 6896   // Next infrequent code is moved outside loops.
 6897   bind(L_last_x);
 6898   if (UseBMI2Instructions) {
 6899     movl(rdx, Address(x,  0));
 6900   } else {
 6901     movl(x_xstart, Address(x,  0));
 6902   }
 6903   jmp(L_third_loop_prologue);
 6904 
 6905   bind(L_done);
 6906 
 6907   pop(xlen);
 6908 
 6909   pop(tmp5);
 6910   pop(tmp4);
 6911   pop(tmp3);
 6912   pop(tmp2);
 6913   pop(tmp1);
 6914   pop(tmp0);
 6915 }
 6916 
 6917 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 6918   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 6919   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 6920   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 6921   Label VECTOR8_TAIL, VECTOR4_TAIL;
 6922   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 6923   Label SAME_TILL_END, DONE;
 6924   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 6925 
 6926   //scale is in rcx in both Win64 and Unix
 6927   ShortBranchVerifier sbv(this);
 6928 
 6929   shlq(length);
 6930   xorq(result, result);
 6931 
 6932   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 6933       VM_Version::supports_avx512vlbw()) {
 6934     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 6935 
 6936     cmpq(length, 64);
 6937     jcc(Assembler::less, VECTOR32_TAIL);
 6938 
 6939     movq(tmp1, length);
 6940     andq(tmp1, 0x3F);      // tail count
 6941     andq(length, ~(0x3F)); //vector count
 6942 
 6943     bind(VECTOR64_LOOP);
 6944     // AVX512 code to compare 64 byte vectors.
 6945     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 6946     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6947     kortestql(k7, k7);
 6948     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 6949     addq(result, 64);
 6950     subq(length, 64);
 6951     jccb(Assembler::notZero, VECTOR64_LOOP);
 6952 
 6953     //bind(VECTOR64_TAIL);
 6954     testq(tmp1, tmp1);
 6955     jcc(Assembler::zero, SAME_TILL_END);
 6956 
 6957     //bind(VECTOR64_TAIL);
 6958     // AVX512 code to compare up to 63 byte vectors.
 6959     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 6960     shlxq(tmp2, tmp2, tmp1);
 6961     notq(tmp2);
 6962     kmovql(k3, tmp2);
 6963 
 6964     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 6965     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6966 
 6967     ktestql(k7, k3);
 6968     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 6969 
 6970     bind(VECTOR64_NOT_EQUAL);
 6971     kmovql(tmp1, k7);
 6972     notq(tmp1);
 6973     tzcntq(tmp1, tmp1);
 6974     addq(result, tmp1);
 6975     shrq(result);
 6976     jmp(DONE);
 6977     bind(VECTOR32_TAIL);
 6978   }
 6979 
 6980   cmpq(length, 8);
 6981   jcc(Assembler::equal, VECTOR8_LOOP);
 6982   jcc(Assembler::less, VECTOR4_TAIL);
 6983 
 6984   if (UseAVX >= 2) {
 6985     Label VECTOR16_TAIL, VECTOR32_LOOP;
 6986 
 6987     cmpq(length, 16);
 6988     jcc(Assembler::equal, VECTOR16_LOOP);
 6989     jcc(Assembler::less, VECTOR8_LOOP);
 6990 
 6991     cmpq(length, 32);
 6992     jccb(Assembler::less, VECTOR16_TAIL);
 6993 
 6994     subq(length, 32);
 6995     bind(VECTOR32_LOOP);
 6996     vmovdqu(rymm0, Address(obja, result));
 6997     vmovdqu(rymm1, Address(objb, result));
 6998     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 6999     vptest(rymm2, rymm2);
 7000     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7001     addq(result, 32);
 7002     subq(length, 32);
 7003     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7004     addq(length, 32);
 7005     jcc(Assembler::equal, SAME_TILL_END);
 7006     //falling through if less than 32 bytes left //close the branch here.
 7007 
 7008     bind(VECTOR16_TAIL);
 7009     cmpq(length, 16);
 7010     jccb(Assembler::less, VECTOR8_TAIL);
 7011     bind(VECTOR16_LOOP);
 7012     movdqu(rymm0, Address(obja, result));
 7013     movdqu(rymm1, Address(objb, result));
 7014     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7015     ptest(rymm2, rymm2);
 7016     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7017     addq(result, 16);
 7018     subq(length, 16);
 7019     jcc(Assembler::equal, SAME_TILL_END);
 7020     //falling through if less than 16 bytes left
 7021   } else {//regular intrinsics
 7022 
 7023     cmpq(length, 16);
 7024     jccb(Assembler::less, VECTOR8_TAIL);
 7025 
 7026     subq(length, 16);
 7027     bind(VECTOR16_LOOP);
 7028     movdqu(rymm0, Address(obja, result));
 7029     movdqu(rymm1, Address(objb, result));
 7030     pxor(rymm0, rymm1);
 7031     ptest(rymm0, rymm0);
 7032     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7033     addq(result, 16);
 7034     subq(length, 16);
 7035     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7036     addq(length, 16);
 7037     jcc(Assembler::equal, SAME_TILL_END);
 7038     //falling through if less than 16 bytes left
 7039   }
 7040 
 7041   bind(VECTOR8_TAIL);
 7042   cmpq(length, 8);
 7043   jccb(Assembler::less, VECTOR4_TAIL);
 7044   bind(VECTOR8_LOOP);
 7045   movq(tmp1, Address(obja, result));
 7046   movq(tmp2, Address(objb, result));
 7047   xorq(tmp1, tmp2);
 7048   testq(tmp1, tmp1);
 7049   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7050   addq(result, 8);
 7051   subq(length, 8);
 7052   jcc(Assembler::equal, SAME_TILL_END);
 7053   //falling through if less than 8 bytes left
 7054 
 7055   bind(VECTOR4_TAIL);
 7056   cmpq(length, 4);
 7057   jccb(Assembler::less, BYTES_TAIL);
 7058   bind(VECTOR4_LOOP);
 7059   movl(tmp1, Address(obja, result));
 7060   xorl(tmp1, Address(objb, result));
 7061   testl(tmp1, tmp1);
 7062   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7063   addq(result, 4);
 7064   subq(length, 4);
 7065   jcc(Assembler::equal, SAME_TILL_END);
 7066   //falling through if less than 4 bytes left
 7067 
 7068   bind(BYTES_TAIL);
 7069   bind(BYTES_LOOP);
 7070   load_unsigned_byte(tmp1, Address(obja, result));
 7071   load_unsigned_byte(tmp2, Address(objb, result));
 7072   xorl(tmp1, tmp2);
 7073   testl(tmp1, tmp1);
 7074   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7075   decq(length);
 7076   jcc(Assembler::zero, SAME_TILL_END);
 7077   incq(result);
 7078   load_unsigned_byte(tmp1, Address(obja, result));
 7079   load_unsigned_byte(tmp2, Address(objb, result));
 7080   xorl(tmp1, tmp2);
 7081   testl(tmp1, tmp1);
 7082   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7083   decq(length);
 7084   jcc(Assembler::zero, SAME_TILL_END);
 7085   incq(result);
 7086   load_unsigned_byte(tmp1, Address(obja, result));
 7087   load_unsigned_byte(tmp2, Address(objb, result));
 7088   xorl(tmp1, tmp2);
 7089   testl(tmp1, tmp1);
 7090   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7091   jmp(SAME_TILL_END);
 7092 
 7093   if (UseAVX >= 2) {
 7094     bind(VECTOR32_NOT_EQUAL);
 7095     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7096     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7097     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7098     vpmovmskb(tmp1, rymm0);
 7099     bsfq(tmp1, tmp1);
 7100     addq(result, tmp1);
 7101     shrq(result);
 7102     jmp(DONE);
 7103   }
 7104 
 7105   bind(VECTOR16_NOT_EQUAL);
 7106   if (UseAVX >= 2) {
 7107     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7108     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7109     pxor(rymm0, rymm2);
 7110   } else {
 7111     pcmpeqb(rymm2, rymm2);
 7112     pxor(rymm0, rymm1);
 7113     pcmpeqb(rymm0, rymm1);
 7114     pxor(rymm0, rymm2);
 7115   }
 7116   pmovmskb(tmp1, rymm0);
 7117   bsfq(tmp1, tmp1);
 7118   addq(result, tmp1);
 7119   shrq(result);
 7120   jmpb(DONE);
 7121 
 7122   bind(VECTOR8_NOT_EQUAL);
 7123   bind(VECTOR4_NOT_EQUAL);
 7124   bsfq(tmp1, tmp1);
 7125   shrq(tmp1, 3);
 7126   addq(result, tmp1);
 7127   bind(BYTES_NOT_EQUAL);
 7128   shrq(result);
 7129   jmpb(DONE);
 7130 
 7131   bind(SAME_TILL_END);
 7132   mov64(result, -1);
 7133 
 7134   bind(DONE);
 7135 }
 7136 
 7137 //Helper functions for square_to_len()
 7138 
 7139 /**
 7140  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7141  * Preserves x and z and modifies rest of the registers.
 7142  */
 7143 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7144   // Perform square and right shift by 1
 7145   // Handle odd xlen case first, then for even xlen do the following
 7146   // jlong carry = 0;
 7147   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7148   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7149   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7150   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7151   //     carry = (jlong)product;
 7152   // }
 7153 
 7154   xorq(tmp5, tmp5);     // carry
 7155   xorq(rdxReg, rdxReg);
 7156   xorl(tmp1, tmp1);     // index for x
 7157   xorl(tmp4, tmp4);     // index for z
 7158 
 7159   Label L_first_loop, L_first_loop_exit;
 7160 
 7161   testl(xlen, 1);
 7162   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7163 
 7164   // Square and right shift by 1 the odd element using 32 bit multiply
 7165   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7166   imulq(raxReg, raxReg);
 7167   shrq(raxReg, 1);
 7168   adcq(tmp5, 0);
 7169   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7170   incrementl(tmp1);
 7171   addl(tmp4, 2);
 7172 
 7173   // Square and  right shift by 1 the rest using 64 bit multiply
 7174   bind(L_first_loop);
 7175   cmpptr(tmp1, xlen);
 7176   jccb(Assembler::equal, L_first_loop_exit);
 7177 
 7178   // Square
 7179   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7180   rorq(raxReg, 32);    // convert big-endian to little-endian
 7181   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7182 
 7183   // Right shift by 1 and save carry
 7184   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7185   rcrq(rdxReg, 1);
 7186   rcrq(raxReg, 1);
 7187   adcq(tmp5, 0);
 7188 
 7189   // Store result in z
 7190   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7191   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7192 
 7193   // Update indices for x and z
 7194   addl(tmp1, 2);
 7195   addl(tmp4, 4);
 7196   jmp(L_first_loop);
 7197 
 7198   bind(L_first_loop_exit);
 7199 }
 7200 
 7201 
 7202 /**
 7203  * Perform the following multiply add operation using BMI2 instructions
 7204  * carry:sum = sum + op1*op2 + carry
 7205  * op2 should be in rdx
 7206  * op2 is preserved, all other registers are modified
 7207  */
 7208 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7209   // assert op2 is rdx
 7210   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7211   addq(sum, carry);
 7212   adcq(tmp2, 0);
 7213   addq(sum, op1);
 7214   adcq(tmp2, 0);
 7215   movq(carry, tmp2);
 7216 }
 7217 
 7218 /**
 7219  * Perform the following multiply add operation:
 7220  * carry:sum = sum + op1*op2 + carry
 7221  * Preserves op1, op2 and modifies rest of registers
 7222  */
 7223 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7224   // rdx:rax = op1 * op2
 7225   movq(raxReg, op2);
 7226   mulq(op1);
 7227 
 7228   //  rdx:rax = sum + carry + rdx:rax
 7229   addq(sum, carry);
 7230   adcq(rdxReg, 0);
 7231   addq(sum, raxReg);
 7232   adcq(rdxReg, 0);
 7233 
 7234   // carry:sum = rdx:sum
 7235   movq(carry, rdxReg);
 7236 }
 7237 
 7238 /**
 7239  * Add 64 bit long carry into z[] with carry propagation.
 7240  * Preserves z and carry register values and modifies rest of registers.
 7241  *
 7242  */
 7243 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7244   Label L_fourth_loop, L_fourth_loop_exit;
 7245 
 7246   movl(tmp1, 1);
 7247   subl(zlen, 2);
 7248   addq(Address(z, zlen, Address::times_4, 0), carry);
 7249 
 7250   bind(L_fourth_loop);
 7251   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7252   subl(zlen, 2);
 7253   jccb(Assembler::negative, L_fourth_loop_exit);
 7254   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7255   jmp(L_fourth_loop);
 7256   bind(L_fourth_loop_exit);
 7257 }
 7258 
 7259 /**
 7260  * Shift z[] left by 1 bit.
 7261  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7262  *
 7263  */
 7264 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7265 
 7266   Label L_fifth_loop, L_fifth_loop_exit;
 7267 
 7268   // Fifth loop
 7269   // Perform primitiveLeftShift(z, zlen, 1)
 7270 
 7271   const Register prev_carry = tmp1;
 7272   const Register new_carry = tmp4;
 7273   const Register value = tmp2;
 7274   const Register zidx = tmp3;
 7275 
 7276   // int zidx, carry;
 7277   // long value;
 7278   // carry = 0;
 7279   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 7280   //    (carry:value)  = (z[i] << 1) | carry ;
 7281   //    z[i] = value;
 7282   // }
 7283 
 7284   movl(zidx, zlen);
 7285   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 7286 
 7287   bind(L_fifth_loop);
 7288   decl(zidx);  // Use decl to preserve carry flag
 7289   decl(zidx);
 7290   jccb(Assembler::negative, L_fifth_loop_exit);
 7291 
 7292   if (UseBMI2Instructions) {
 7293      movq(value, Address(z, zidx, Address::times_4, 0));
 7294      rclq(value, 1);
 7295      rorxq(value, value, 32);
 7296      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7297   }
 7298   else {
 7299     // clear new_carry
 7300     xorl(new_carry, new_carry);
 7301 
 7302     // Shift z[i] by 1, or in previous carry and save new carry
 7303     movq(value, Address(z, zidx, Address::times_4, 0));
 7304     shlq(value, 1);
 7305     adcl(new_carry, 0);
 7306 
 7307     orq(value, prev_carry);
 7308     rorq(value, 0x20);
 7309     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7310 
 7311     // Set previous carry = new carry
 7312     movl(prev_carry, new_carry);
 7313   }
 7314   jmp(L_fifth_loop);
 7315 
 7316   bind(L_fifth_loop_exit);
 7317 }
 7318 
 7319 
 7320 /**
 7321  * Code for BigInteger::squareToLen() intrinsic
 7322  *
 7323  * rdi: x
 7324  * rsi: len
 7325  * r8:  z
 7326  * rcx: zlen
 7327  * r12: tmp1
 7328  * r13: tmp2
 7329  * r14: tmp3
 7330  * r15: tmp4
 7331  * rbx: tmp5
 7332  *
 7333  */
 7334 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) {
 7335 
 7336   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 7337   push(tmp1);
 7338   push(tmp2);
 7339   push(tmp3);
 7340   push(tmp4);
 7341   push(tmp5);
 7342 
 7343   // First loop
 7344   // Store the squares, right shifted one bit (i.e., divided by 2).
 7345   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7346 
 7347   // Add in off-diagonal sums.
 7348   //
 7349   // Second, third (nested) and fourth loops.
 7350   // zlen +=2;
 7351   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 7352   //    carry = 0;
 7353   //    long op2 = x[xidx:xidx+1];
 7354   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 7355   //       k -= 2;
 7356   //       long op1 = x[j:j+1];
 7357   //       long sum = z[k:k+1];
 7358   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 7359   //       z[k:k+1] = sum;
 7360   //    }
 7361   //    add_one_64(z, k, carry, tmp_regs);
 7362   // }
 7363 
 7364   const Register carry = tmp5;
 7365   const Register sum = tmp3;
 7366   const Register op1 = tmp4;
 7367   Register op2 = tmp2;
 7368 
 7369   push(zlen);
 7370   push(len);
 7371   addl(zlen,2);
 7372   bind(L_second_loop);
 7373   xorq(carry, carry);
 7374   subl(zlen, 4);
 7375   subl(len, 2);
 7376   push(zlen);
 7377   push(len);
 7378   cmpl(len, 0);
 7379   jccb(Assembler::lessEqual, L_second_loop_exit);
 7380 
 7381   // Multiply an array by one 64 bit long.
 7382   if (UseBMI2Instructions) {
 7383     op2 = rdxReg;
 7384     movq(op2, Address(x, len, Address::times_4,  0));
 7385     rorxq(op2, op2, 32);
 7386   }
 7387   else {
 7388     movq(op2, Address(x, len, Address::times_4,  0));
 7389     rorq(op2, 32);
 7390   }
 7391 
 7392   bind(L_third_loop);
 7393   decrementl(len);
 7394   jccb(Assembler::negative, L_third_loop_exit);
 7395   decrementl(len);
 7396   jccb(Assembler::negative, L_last_x);
 7397 
 7398   movq(op1, Address(x, len, Address::times_4,  0));
 7399   rorq(op1, 32);
 7400 
 7401   bind(L_multiply);
 7402   subl(zlen, 2);
 7403   movq(sum, Address(z, zlen, Address::times_4,  0));
 7404 
 7405   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 7406   if (UseBMI2Instructions) {
 7407     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 7408   }
 7409   else {
 7410     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7411   }
 7412 
 7413   movq(Address(z, zlen, Address::times_4, 0), sum);
 7414 
 7415   jmp(L_third_loop);
 7416   bind(L_third_loop_exit);
 7417 
 7418   // Fourth loop
 7419   // Add 64 bit long carry into z with carry propagation.
 7420   // Uses offsetted zlen.
 7421   add_one_64(z, zlen, carry, tmp1);
 7422 
 7423   pop(len);
 7424   pop(zlen);
 7425   jmp(L_second_loop);
 7426 
 7427   // Next infrequent code is moved outside loops.
 7428   bind(L_last_x);
 7429   movl(op1, Address(x, 0));
 7430   jmp(L_multiply);
 7431 
 7432   bind(L_second_loop_exit);
 7433   pop(len);
 7434   pop(zlen);
 7435   pop(len);
 7436   pop(zlen);
 7437 
 7438   // Fifth loop
 7439   // Shift z left 1 bit.
 7440   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 7441 
 7442   // z[zlen-1] |= x[len-1] & 1;
 7443   movl(tmp3, Address(x, len, Address::times_4, -4));
 7444   andl(tmp3, 1);
 7445   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 7446 
 7447   pop(tmp5);
 7448   pop(tmp4);
 7449   pop(tmp3);
 7450   pop(tmp2);
 7451   pop(tmp1);
 7452 }
 7453 
 7454 /**
 7455  * Helper function for mul_add()
 7456  * Multiply the in[] by int k and add to out[] starting at offset offs using
 7457  * 128 bit by 32 bit multiply and return the carry in tmp5.
 7458  * Only quad int aligned length of in[] is operated on in this function.
 7459  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 7460  * This function preserves out, in and k registers.
 7461  * len and offset point to the appropriate index in "in" & "out" correspondingly
 7462  * tmp5 has the carry.
 7463  * other registers are temporary and are modified.
 7464  *
 7465  */
 7466 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 7467   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 7468   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7469 
 7470   Label L_first_loop, L_first_loop_exit;
 7471 
 7472   movl(tmp1, len);
 7473   shrl(tmp1, 2);
 7474 
 7475   bind(L_first_loop);
 7476   subl(tmp1, 1);
 7477   jccb(Assembler::negative, L_first_loop_exit);
 7478 
 7479   subl(len, 4);
 7480   subl(offset, 4);
 7481 
 7482   Register op2 = tmp2;
 7483   const Register sum = tmp3;
 7484   const Register op1 = tmp4;
 7485   const Register carry = tmp5;
 7486 
 7487   if (UseBMI2Instructions) {
 7488     op2 = rdxReg;
 7489   }
 7490 
 7491   movq(op1, Address(in, len, Address::times_4,  8));
 7492   rorq(op1, 32);
 7493   movq(sum, Address(out, offset, Address::times_4,  8));
 7494   rorq(sum, 32);
 7495   if (UseBMI2Instructions) {
 7496     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7497   }
 7498   else {
 7499     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7500   }
 7501   // Store back in big endian from little endian
 7502   rorq(sum, 0x20);
 7503   movq(Address(out, offset, Address::times_4,  8), sum);
 7504 
 7505   movq(op1, Address(in, len, Address::times_4,  0));
 7506   rorq(op1, 32);
 7507   movq(sum, Address(out, offset, Address::times_4,  0));
 7508   rorq(sum, 32);
 7509   if (UseBMI2Instructions) {
 7510     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7511   }
 7512   else {
 7513     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7514   }
 7515   // Store back in big endian from little endian
 7516   rorq(sum, 0x20);
 7517   movq(Address(out, offset, Address::times_4,  0), sum);
 7518 
 7519   jmp(L_first_loop);
 7520   bind(L_first_loop_exit);
 7521 }
 7522 
 7523 /**
 7524  * Code for BigInteger::mulAdd() intrinsic
 7525  *
 7526  * rdi: out
 7527  * rsi: in
 7528  * r11: offs (out.length - offset)
 7529  * rcx: len
 7530  * r8:  k
 7531  * r12: tmp1
 7532  * r13: tmp2
 7533  * r14: tmp3
 7534  * r15: tmp4
 7535  * rbx: tmp5
 7536  * Multiply the in[] by word k and add to out[], return the carry in rax
 7537  */
 7538 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 7539    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 7540    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7541 
 7542   Label L_carry, L_last_in, L_done;
 7543 
 7544 // carry = 0;
 7545 // for (int j=len-1; j >= 0; j--) {
 7546 //    long product = (in[j] & LONG_MASK) * kLong +
 7547 //                   (out[offs] & LONG_MASK) + carry;
 7548 //    out[offs--] = (int)product;
 7549 //    carry = product >>> 32;
 7550 // }
 7551 //
 7552   push(tmp1);
 7553   push(tmp2);
 7554   push(tmp3);
 7555   push(tmp4);
 7556   push(tmp5);
 7557 
 7558   Register op2 = tmp2;
 7559   const Register sum = tmp3;
 7560   const Register op1 = tmp4;
 7561   const Register carry =  tmp5;
 7562 
 7563   if (UseBMI2Instructions) {
 7564     op2 = rdxReg;
 7565     movl(op2, k);
 7566   }
 7567   else {
 7568     movl(op2, k);
 7569   }
 7570 
 7571   xorq(carry, carry);
 7572 
 7573   //First loop
 7574 
 7575   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 7576   //The carry is in tmp5
 7577   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7578 
 7579   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 7580   decrementl(len);
 7581   jccb(Assembler::negative, L_carry);
 7582   decrementl(len);
 7583   jccb(Assembler::negative, L_last_in);
 7584 
 7585   movq(op1, Address(in, len, Address::times_4,  0));
 7586   rorq(op1, 32);
 7587 
 7588   subl(offs, 2);
 7589   movq(sum, Address(out, offs, Address::times_4,  0));
 7590   rorq(sum, 32);
 7591 
 7592   if (UseBMI2Instructions) {
 7593     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7594   }
 7595   else {
 7596     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7597   }
 7598 
 7599   // Store back in big endian from little endian
 7600   rorq(sum, 0x20);
 7601   movq(Address(out, offs, Address::times_4,  0), sum);
 7602 
 7603   testl(len, len);
 7604   jccb(Assembler::zero, L_carry);
 7605 
 7606   //Multiply the last in[] entry, if any
 7607   bind(L_last_in);
 7608   movl(op1, Address(in, 0));
 7609   movl(sum, Address(out, offs, Address::times_4,  -4));
 7610 
 7611   movl(raxReg, k);
 7612   mull(op1); //tmp4 * eax -> edx:eax
 7613   addl(sum, carry);
 7614   adcl(rdxReg, 0);
 7615   addl(sum, raxReg);
 7616   adcl(rdxReg, 0);
 7617   movl(carry, rdxReg);
 7618 
 7619   movl(Address(out, offs, Address::times_4,  -4), sum);
 7620 
 7621   bind(L_carry);
 7622   //return tmp5/carry as carry in rax
 7623   movl(rax, carry);
 7624 
 7625   bind(L_done);
 7626   pop(tmp5);
 7627   pop(tmp4);
 7628   pop(tmp3);
 7629   pop(tmp2);
 7630   pop(tmp1);
 7631 }
 7632 
 7633 /**
 7634  * Emits code to update CRC-32 with a byte value according to constants in table
 7635  *
 7636  * @param [in,out]crc   Register containing the crc.
 7637  * @param [in]val       Register containing the byte to fold into the CRC.
 7638  * @param [in]table     Register containing the table of crc constants.
 7639  *
 7640  * uint32_t crc;
 7641  * val = crc_table[(val ^ crc) & 0xFF];
 7642  * crc = val ^ (crc >> 8);
 7643  *
 7644  */
 7645 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 7646   xorl(val, crc);
 7647   andl(val, 0xFF);
 7648   shrl(crc, 8); // unsigned shift
 7649   xorl(crc, Address(table, val, Address::times_4, 0));
 7650 }
 7651 
 7652 /**
 7653  * Fold 128-bit data chunk
 7654  */
 7655 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 7656   if (UseAVX > 0) {
 7657     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 7658     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 7659     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 7660     pxor(xcrc, xtmp);
 7661   } else {
 7662     movdqa(xtmp, xcrc);
 7663     pclmulhdq(xtmp, xK);   // [123:64]
 7664     pclmulldq(xcrc, xK);   // [63:0]
 7665     pxor(xcrc, xtmp);
 7666     movdqu(xtmp, Address(buf, offset));
 7667     pxor(xcrc, xtmp);
 7668   }
 7669 }
 7670 
 7671 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 7672   if (UseAVX > 0) {
 7673     vpclmulhdq(xtmp, xK, xcrc);
 7674     vpclmulldq(xcrc, xK, xcrc);
 7675     pxor(xcrc, xbuf);
 7676     pxor(xcrc, xtmp);
 7677   } else {
 7678     movdqa(xtmp, xcrc);
 7679     pclmulhdq(xtmp, xK);
 7680     pclmulldq(xcrc, xK);
 7681     pxor(xcrc, xbuf);
 7682     pxor(xcrc, xtmp);
 7683   }
 7684 }
 7685 
 7686 /**
 7687  * 8-bit folds to compute 32-bit CRC
 7688  *
 7689  * uint64_t xcrc;
 7690  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 7691  */
 7692 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 7693   movdl(tmp, xcrc);
 7694   andl(tmp, 0xFF);
 7695   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 7696   psrldq(xcrc, 1); // unsigned shift one byte
 7697   pxor(xcrc, xtmp);
 7698 }
 7699 
 7700 /**
 7701  * uint32_t crc;
 7702  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 7703  */
 7704 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 7705   movl(tmp, crc);
 7706   andl(tmp, 0xFF);
 7707   shrl(crc, 8);
 7708   xorl(crc, Address(table, tmp, Address::times_4, 0));
 7709 }
 7710 
 7711 /**
 7712  * @param crc   register containing existing CRC (32-bit)
 7713  * @param buf   register pointing to input byte buffer (byte*)
 7714  * @param len   register containing number of bytes
 7715  * @param table register that will contain address of CRC table
 7716  * @param tmp   scratch register
 7717  */
 7718 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 7719   assert_different_registers(crc, buf, len, table, tmp, rax);
 7720 
 7721   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 7722   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 7723 
 7724   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 7725   // context for the registers used, where all instructions below are using 128-bit mode
 7726   // On EVEX without VL and BW, these instructions will all be AVX.
 7727   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 7728   notl(crc); // ~crc
 7729   cmpl(len, 16);
 7730   jcc(Assembler::less, L_tail);
 7731 
 7732   // Align buffer to 16 bytes
 7733   movl(tmp, buf);
 7734   andl(tmp, 0xF);
 7735   jccb(Assembler::zero, L_aligned);
 7736   subl(tmp,  16);
 7737   addl(len, tmp);
 7738 
 7739   align(4);
 7740   BIND(L_align_loop);
 7741   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7742   update_byte_crc32(crc, rax, table);
 7743   increment(buf);
 7744   incrementl(tmp);
 7745   jccb(Assembler::less, L_align_loop);
 7746 
 7747   BIND(L_aligned);
 7748   movl(tmp, len); // save
 7749   shrl(len, 4);
 7750   jcc(Assembler::zero, L_tail_restore);
 7751 
 7752   // Fold crc into first bytes of vector
 7753   movdqa(xmm1, Address(buf, 0));
 7754   movdl(rax, xmm1);
 7755   xorl(crc, rax);
 7756   if (VM_Version::supports_sse4_1()) {
 7757     pinsrd(xmm1, crc, 0);
 7758   } else {
 7759     pinsrw(xmm1, crc, 0);
 7760     shrl(crc, 16);
 7761     pinsrw(xmm1, crc, 1);
 7762   }
 7763   addptr(buf, 16);
 7764   subl(len, 4); // len > 0
 7765   jcc(Assembler::less, L_fold_tail);
 7766 
 7767   movdqa(xmm2, Address(buf,  0));
 7768   movdqa(xmm3, Address(buf, 16));
 7769   movdqa(xmm4, Address(buf, 32));
 7770   addptr(buf, 48);
 7771   subl(len, 3);
 7772   jcc(Assembler::lessEqual, L_fold_512b);
 7773 
 7774   // Fold total 512 bits of polynomial on each iteration,
 7775   // 128 bits per each of 4 parallel streams.
 7776   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 7777 
 7778   align32();
 7779   BIND(L_fold_512b_loop);
 7780   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7781   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 7782   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 7783   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 7784   addptr(buf, 64);
 7785   subl(len, 4);
 7786   jcc(Assembler::greater, L_fold_512b_loop);
 7787 
 7788   // Fold 512 bits to 128 bits.
 7789   BIND(L_fold_512b);
 7790   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7791   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 7792   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 7793   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 7794 
 7795   // Fold the rest of 128 bits data chunks
 7796   BIND(L_fold_tail);
 7797   addl(len, 3);
 7798   jccb(Assembler::lessEqual, L_fold_128b);
 7799   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7800 
 7801   BIND(L_fold_tail_loop);
 7802   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7803   addptr(buf, 16);
 7804   decrementl(len);
 7805   jccb(Assembler::greater, L_fold_tail_loop);
 7806 
 7807   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 7808   BIND(L_fold_128b);
 7809   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 7810   if (UseAVX > 0) {
 7811     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 7812     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 7813     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 7814   } else {
 7815     movdqa(xmm2, xmm0);
 7816     pclmulqdq(xmm2, xmm1, 0x1);
 7817     movdqa(xmm3, xmm0);
 7818     pand(xmm3, xmm2);
 7819     pclmulqdq(xmm0, xmm3, 0x1);
 7820   }
 7821   psrldq(xmm1, 8);
 7822   psrldq(xmm2, 4);
 7823   pxor(xmm0, xmm1);
 7824   pxor(xmm0, xmm2);
 7825 
 7826   // 8 8-bit folds to compute 32-bit CRC.
 7827   for (int j = 0; j < 4; j++) {
 7828     fold_8bit_crc32(xmm0, table, xmm1, rax);
 7829   }
 7830   movdl(crc, xmm0); // mov 32 bits to general register
 7831   for (int j = 0; j < 4; j++) {
 7832     fold_8bit_crc32(crc, table, rax);
 7833   }
 7834 
 7835   BIND(L_tail_restore);
 7836   movl(len, tmp); // restore
 7837   BIND(L_tail);
 7838   andl(len, 0xf);
 7839   jccb(Assembler::zero, L_exit);
 7840 
 7841   // Fold the rest of bytes
 7842   align(4);
 7843   BIND(L_tail_loop);
 7844   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7845   update_byte_crc32(crc, rax, table);
 7846   increment(buf);
 7847   decrementl(len);
 7848   jccb(Assembler::greater, L_tail_loop);
 7849 
 7850   BIND(L_exit);
 7851   notl(crc); // ~c
 7852 }
 7853 
 7854 // Helper function for AVX 512 CRC32
 7855 // Fold 512-bit data chunks
 7856 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 7857                                              Register pos, int offset) {
 7858   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 7859   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 7860   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 7861   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 7862   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 7863 }
 7864 
 7865 // Helper function for AVX 512 CRC32
 7866 // Compute CRC32 for < 256B buffers
 7867 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 7868                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 7869                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 7870 
 7871   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 7872   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 7873   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 7874 
 7875   // check if there is enough buffer to be able to fold 16B at a time
 7876   cmpl(len, 32);
 7877   jcc(Assembler::less, L_less_than_32);
 7878 
 7879   // if there is, load the constants
 7880   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 7881   movdl(xmm0, crc);                        // get the initial crc value
 7882   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7883   pxor(xmm7, xmm0);
 7884 
 7885   // update the buffer pointer
 7886   addl(pos, 16);
 7887   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 7888   subl(len, 32);
 7889   jmp(L_16B_reduction_loop);
 7890 
 7891   bind(L_less_than_32);
 7892   //mov initial crc to the return value. this is necessary for zero - length buffers.
 7893   movl(rax, crc);
 7894   testl(len, len);
 7895   jcc(Assembler::equal, L_cleanup);
 7896 
 7897   movdl(xmm0, crc);                        //get the initial crc value
 7898 
 7899   cmpl(len, 16);
 7900   jcc(Assembler::equal, L_exact_16_left);
 7901   jcc(Assembler::less, L_less_than_16_left);
 7902 
 7903   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7904   pxor(xmm7, xmm0);                       //xor the initial crc value
 7905   addl(pos, 16);
 7906   subl(len, 16);
 7907   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 7908   jmp(L_get_last_two_xmms);
 7909 
 7910   bind(L_less_than_16_left);
 7911   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 7912   pxor(xmm1, xmm1);
 7913   movptr(tmp1, rsp);
 7914   movdqu(Address(tmp1, 0 * 16), xmm1);
 7915 
 7916   cmpl(len, 4);
 7917   jcc(Assembler::less, L_only_less_than_4);
 7918 
 7919   //backup the counter value
 7920   movl(tmp2, len);
 7921   cmpl(len, 8);
 7922   jcc(Assembler::less, L_less_than_8_left);
 7923 
 7924   //load 8 Bytes
 7925   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 7926   movq(Address(tmp1, 0 * 16), rax);
 7927   addptr(tmp1, 8);
 7928   subl(len, 8);
 7929   addl(pos, 8);
 7930 
 7931   bind(L_less_than_8_left);
 7932   cmpl(len, 4);
 7933   jcc(Assembler::less, L_less_than_4_left);
 7934 
 7935   //load 4 Bytes
 7936   movl(rax, Address(buf, pos, Address::times_1, 0));
 7937   movl(Address(tmp1, 0 * 16), rax);
 7938   addptr(tmp1, 4);
 7939   subl(len, 4);
 7940   addl(pos, 4);
 7941 
 7942   bind(L_less_than_4_left);
 7943   cmpl(len, 2);
 7944   jcc(Assembler::less, L_less_than_2_left);
 7945 
 7946   // load 2 Bytes
 7947   movw(rax, Address(buf, pos, Address::times_1, 0));
 7948   movl(Address(tmp1, 0 * 16), rax);
 7949   addptr(tmp1, 2);
 7950   subl(len, 2);
 7951   addl(pos, 2);
 7952 
 7953   bind(L_less_than_2_left);
 7954   cmpl(len, 1);
 7955   jcc(Assembler::less, L_zero_left);
 7956 
 7957   // load 1 Byte
 7958   movb(rax, Address(buf, pos, Address::times_1, 0));
 7959   movb(Address(tmp1, 0 * 16), rax);
 7960 
 7961   bind(L_zero_left);
 7962   movdqu(xmm7, Address(rsp, 0));
 7963   pxor(xmm7, xmm0);                       //xor the initial crc value
 7964 
 7965   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 7966   movdqu(xmm0, Address(rax, tmp2));
 7967   pshufb(xmm7, xmm0);
 7968   jmp(L_128_done);
 7969 
 7970   bind(L_exact_16_left);
 7971   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 7972   pxor(xmm7, xmm0);                       //xor the initial crc value
 7973   jmp(L_128_done);
 7974 
 7975   bind(L_only_less_than_4);
 7976   cmpl(len, 3);
 7977   jcc(Assembler::less, L_only_less_than_3);
 7978 
 7979   // load 3 Bytes
 7980   movb(rax, Address(buf, pos, Address::times_1, 0));
 7981   movb(Address(tmp1, 0), rax);
 7982 
 7983   movb(rax, Address(buf, pos, Address::times_1, 1));
 7984   movb(Address(tmp1, 1), rax);
 7985 
 7986   movb(rax, Address(buf, pos, Address::times_1, 2));
 7987   movb(Address(tmp1, 2), rax);
 7988 
 7989   movdqu(xmm7, Address(rsp, 0));
 7990   pxor(xmm7, xmm0);                     //xor the initial crc value
 7991 
 7992   pslldq(xmm7, 0x5);
 7993   jmp(L_barrett);
 7994   bind(L_only_less_than_3);
 7995   cmpl(len, 2);
 7996   jcc(Assembler::less, L_only_less_than_2);
 7997 
 7998   // load 2 Bytes
 7999   movb(rax, Address(buf, pos, Address::times_1, 0));
 8000   movb(Address(tmp1, 0), rax);
 8001 
 8002   movb(rax, Address(buf, pos, Address::times_1, 1));
 8003   movb(Address(tmp1, 1), rax);
 8004 
 8005   movdqu(xmm7, Address(rsp, 0));
 8006   pxor(xmm7, xmm0);                     //xor the initial crc value
 8007 
 8008   pslldq(xmm7, 0x6);
 8009   jmp(L_barrett);
 8010 
 8011   bind(L_only_less_than_2);
 8012   //load 1 Byte
 8013   movb(rax, Address(buf, pos, Address::times_1, 0));
 8014   movb(Address(tmp1, 0), rax);
 8015 
 8016   movdqu(xmm7, Address(rsp, 0));
 8017   pxor(xmm7, xmm0);                     //xor the initial crc value
 8018 
 8019   pslldq(xmm7, 0x7);
 8020 }
 8021 
 8022 /**
 8023 * Compute CRC32 using AVX512 instructions
 8024 * param crc   register containing existing CRC (32-bit)
 8025 * param buf   register pointing to input byte buffer (byte*)
 8026 * param len   register containing number of bytes
 8027 * param table address of crc or crc32c table
 8028 * param tmp1  scratch register
 8029 * param tmp2  scratch register
 8030 * return rax  result register
 8031 *
 8032 * This routine is identical for crc32c with the exception of the precomputed constant
 8033 * table which will be passed as the table argument.  The calculation steps are
 8034 * the same for both variants.
 8035 */
 8036 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8037   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8038 
 8039   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8040   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8041   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8042   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8043   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8044 
 8045   const Register pos = r12;
 8046   push(r12);
 8047   subptr(rsp, 16 * 2 + 8);
 8048 
 8049   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8050   // context for the registers used, where all instructions below are using 128-bit mode
 8051   // On EVEX without VL and BW, these instructions will all be AVX.
 8052   movl(pos, 0);
 8053 
 8054   // check if smaller than 256B
 8055   cmpl(len, 256);
 8056   jcc(Assembler::less, L_less_than_256);
 8057 
 8058   // load the initial crc value
 8059   movdl(xmm10, crc);
 8060 
 8061   // receive the initial 64B data, xor the initial crc value
 8062   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8063   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8064   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8065   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8066 
 8067   subl(len, 256);
 8068   cmpl(len, 256);
 8069   jcc(Assembler::less, L_fold_128_B_loop);
 8070 
 8071   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8072   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8073   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8074   subl(len, 256);
 8075 
 8076   bind(L_fold_256_B_loop);
 8077   addl(pos, 256);
 8078   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8079   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8080   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8081   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8082 
 8083   subl(len, 256);
 8084   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8085 
 8086   // Fold 256 into 128
 8087   addl(pos, 256);
 8088   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8089   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8090   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8091 
 8092   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8093   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8094   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8095 
 8096   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8097   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8098 
 8099   addl(len, 128);
 8100   jmp(L_fold_128_B_register);
 8101 
 8102   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8103   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8104 
 8105   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8106   bind(L_fold_128_B_loop);
 8107   addl(pos, 128);
 8108   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8109   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8110 
 8111   subl(len, 128);
 8112   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8113 
 8114   addl(pos, 128);
 8115 
 8116   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8117   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8118   bind(L_fold_128_B_register);
 8119   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8120   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8121   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8122   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8123   // save last that has no multiplicand
 8124   vextracti64x2(xmm7, xmm4, 3);
 8125 
 8126   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8127   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8128   // Needed later in reduction loop
 8129   movdqu(xmm10, Address(table, 1 * 16));
 8130   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8131   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8132 
 8133   // Swap 1,0,3,2 - 01 00 11 10
 8134   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8135   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8136   vextracti128(xmm5, xmm8, 1);
 8137   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8138 
 8139   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8140   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8141   addl(len, 128 - 16);
 8142   jcc(Assembler::less, L_final_reduction_for_128);
 8143 
 8144   bind(L_16B_reduction_loop);
 8145   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8146   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8147   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8148   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8149   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8150   addl(pos, 16);
 8151   subl(len, 16);
 8152   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8153 
 8154   bind(L_final_reduction_for_128);
 8155   addl(len, 16);
 8156   jcc(Assembler::equal, L_128_done);
 8157 
 8158   bind(L_get_last_two_xmms);
 8159   movdqu(xmm2, xmm7);
 8160   addl(pos, len);
 8161   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8162   subl(pos, len);
 8163 
 8164   // get rid of the extra data that was loaded before
 8165   // load the shift constant
 8166   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8167   movdqu(xmm0, Address(rax, len));
 8168   addl(rax, len);
 8169 
 8170   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8171   //Change mask to 512
 8172   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8173   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8174 
 8175   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8176   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8177   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8178   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8179   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8180 
 8181   bind(L_128_done);
 8182   // compute crc of a 128-bit value
 8183   movdqu(xmm10, Address(table, 3 * 16));
 8184   movdqu(xmm0, xmm7);
 8185 
 8186   // 64b fold
 8187   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8188   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8189   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8190 
 8191   // 32b fold
 8192   movdqu(xmm0, xmm7);
 8193   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8194   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8195   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8196   jmp(L_barrett);
 8197 
 8198   bind(L_less_than_256);
 8199   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);
 8200 
 8201   //barrett reduction
 8202   bind(L_barrett);
 8203   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8204   movdqu(xmm1, xmm7);
 8205   movdqu(xmm2, xmm7);
 8206   movdqu(xmm10, Address(table, 4 * 16));
 8207 
 8208   pclmulqdq(xmm7, xmm10, 0x0);
 8209   pxor(xmm7, xmm2);
 8210   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8211   movdqu(xmm2, xmm7);
 8212   pclmulqdq(xmm7, xmm10, 0x10);
 8213   pxor(xmm7, xmm2);
 8214   pxor(xmm7, xmm1);
 8215   pextrd(crc, xmm7, 2);
 8216 
 8217   bind(L_cleanup);
 8218   addptr(rsp, 16 * 2 + 8);
 8219   pop(r12);
 8220 }
 8221 
 8222 // S. Gueron / Information Processing Letters 112 (2012) 184
 8223 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8224 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8225 // Output: the 64-bit carry-less product of B * CONST
 8226 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8227                                      Register tmp1, Register tmp2, Register tmp3) {
 8228   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8229   if (n > 0) {
 8230     addq(tmp3, n * 256 * 8);
 8231   }
 8232   //    Q1 = TABLEExt[n][B & 0xFF];
 8233   movl(tmp1, in);
 8234   andl(tmp1, 0x000000FF);
 8235   shll(tmp1, 3);
 8236   addq(tmp1, tmp3);
 8237   movq(tmp1, Address(tmp1, 0));
 8238 
 8239   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8240   movl(tmp2, in);
 8241   shrl(tmp2, 8);
 8242   andl(tmp2, 0x000000FF);
 8243   shll(tmp2, 3);
 8244   addq(tmp2, tmp3);
 8245   movq(tmp2, Address(tmp2, 0));
 8246 
 8247   shlq(tmp2, 8);
 8248   xorq(tmp1, tmp2);
 8249 
 8250   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8251   movl(tmp2, in);
 8252   shrl(tmp2, 16);
 8253   andl(tmp2, 0x000000FF);
 8254   shll(tmp2, 3);
 8255   addq(tmp2, tmp3);
 8256   movq(tmp2, Address(tmp2, 0));
 8257 
 8258   shlq(tmp2, 16);
 8259   xorq(tmp1, tmp2);
 8260 
 8261   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8262   shrl(in, 24);
 8263   andl(in, 0x000000FF);
 8264   shll(in, 3);
 8265   addq(in, tmp3);
 8266   movq(in, Address(in, 0));
 8267 
 8268   shlq(in, 24);
 8269   xorq(in, tmp1);
 8270   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8271 }
 8272 
 8273 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8274                                       Register in_out,
 8275                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8276                                       XMMRegister w_xtmp2,
 8277                                       Register tmp1,
 8278                                       Register n_tmp2, Register n_tmp3) {
 8279   if (is_pclmulqdq_supported) {
 8280     movdl(w_xtmp1, in_out); // modified blindly
 8281 
 8282     movl(tmp1, const_or_pre_comp_const_index);
 8283     movdl(w_xtmp2, tmp1);
 8284     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8285 
 8286     movdq(in_out, w_xtmp1);
 8287   } else {
 8288     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 8289   }
 8290 }
 8291 
 8292 // Recombination Alternative 2: No bit-reflections
 8293 // T1 = (CRC_A * U1) << 1
 8294 // T2 = (CRC_B * U2) << 1
 8295 // C1 = T1 >> 32
 8296 // C2 = T2 >> 32
 8297 // T1 = T1 & 0xFFFFFFFF
 8298 // T2 = T2 & 0xFFFFFFFF
 8299 // T1 = CRC32(0, T1)
 8300 // T2 = CRC32(0, T2)
 8301 // C1 = C1 ^ T1
 8302 // C2 = C2 ^ T2
 8303 // CRC = C1 ^ C2 ^ CRC_C
 8304 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,
 8305                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8306                                      Register tmp1, Register tmp2,
 8307                                      Register n_tmp3) {
 8308   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8309   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8310   shlq(in_out, 1);
 8311   movl(tmp1, in_out);
 8312   shrq(in_out, 32);
 8313   xorl(tmp2, tmp2);
 8314   crc32(tmp2, tmp1, 4);
 8315   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 8316   shlq(in1, 1);
 8317   movl(tmp1, in1);
 8318   shrq(in1, 32);
 8319   xorl(tmp2, tmp2);
 8320   crc32(tmp2, tmp1, 4);
 8321   xorl(in1, tmp2);
 8322   xorl(in_out, in1);
 8323   xorl(in_out, in2);
 8324 }
 8325 
 8326 // Set N to predefined value
 8327 // Subtract from a length of a buffer
 8328 // execute in a loop:
 8329 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 8330 // for i = 1 to N do
 8331 //  CRC_A = CRC32(CRC_A, A[i])
 8332 //  CRC_B = CRC32(CRC_B, B[i])
 8333 //  CRC_C = CRC32(CRC_C, C[i])
 8334 // end for
 8335 // Recombine
 8336 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,
 8337                                        Register in_out1, Register in_out2, Register in_out3,
 8338                                        Register tmp1, Register tmp2, Register tmp3,
 8339                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8340                                        Register tmp4, Register tmp5,
 8341                                        Register n_tmp6) {
 8342   Label L_processPartitions;
 8343   Label L_processPartition;
 8344   Label L_exit;
 8345 
 8346   bind(L_processPartitions);
 8347   cmpl(in_out1, 3 * size);
 8348   jcc(Assembler::less, L_exit);
 8349     xorl(tmp1, tmp1);
 8350     xorl(tmp2, tmp2);
 8351     movq(tmp3, in_out2);
 8352     addq(tmp3, size);
 8353 
 8354     bind(L_processPartition);
 8355       crc32(in_out3, Address(in_out2, 0), 8);
 8356       crc32(tmp1, Address(in_out2, size), 8);
 8357       crc32(tmp2, Address(in_out2, size * 2), 8);
 8358       addq(in_out2, 8);
 8359       cmpq(in_out2, tmp3);
 8360       jcc(Assembler::less, L_processPartition);
 8361     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 8362             w_xtmp1, w_xtmp2, w_xtmp3,
 8363             tmp4, tmp5,
 8364             n_tmp6);
 8365     addq(in_out2, 2 * size);
 8366     subl(in_out1, 3 * size);
 8367     jmp(L_processPartitions);
 8368 
 8369   bind(L_exit);
 8370 }
 8371 
 8372 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 8373 // Input: A buffer I of L bytes.
 8374 // Output: the CRC32C value of the buffer.
 8375 // Notations:
 8376 // Write L = 24N + r, with N = floor (L/24).
 8377 // r = L mod 24 (0 <= r < 24).
 8378 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 8379 // N quadwords, and R consists of r bytes.
 8380 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 8381 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 8382 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 8383 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 8384 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 8385                                           Register tmp1, Register tmp2, Register tmp3,
 8386                                           Register tmp4, Register tmp5, Register tmp6,
 8387                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8388                                           bool is_pclmulqdq_supported) {
 8389   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 8390   Label L_wordByWord;
 8391   Label L_byteByByteProlog;
 8392   Label L_byteByByte;
 8393   Label L_exit;
 8394 
 8395   if (is_pclmulqdq_supported ) {
 8396     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 8397     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 8398 
 8399     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 8400     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 8401 
 8402     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 8403     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 8404     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 8405   } else {
 8406     const_or_pre_comp_const_index[0] = 1;
 8407     const_or_pre_comp_const_index[1] = 0;
 8408 
 8409     const_or_pre_comp_const_index[2] = 3;
 8410     const_or_pre_comp_const_index[3] = 2;
 8411 
 8412     const_or_pre_comp_const_index[4] = 5;
 8413     const_or_pre_comp_const_index[5] = 4;
 8414    }
 8415   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 8416                     in2, in1, in_out,
 8417                     tmp1, tmp2, tmp3,
 8418                     w_xtmp1, w_xtmp2, w_xtmp3,
 8419                     tmp4, tmp5,
 8420                     tmp6);
 8421   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 8422                     in2, in1, in_out,
 8423                     tmp1, tmp2, tmp3,
 8424                     w_xtmp1, w_xtmp2, w_xtmp3,
 8425                     tmp4, tmp5,
 8426                     tmp6);
 8427   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 8428                     in2, in1, in_out,
 8429                     tmp1, tmp2, tmp3,
 8430                     w_xtmp1, w_xtmp2, w_xtmp3,
 8431                     tmp4, tmp5,
 8432                     tmp6);
 8433   movl(tmp1, in2);
 8434   andl(tmp1, 0x00000007);
 8435   negl(tmp1);
 8436   addl(tmp1, in2);
 8437   addq(tmp1, in1);
 8438 
 8439   cmpq(in1, tmp1);
 8440   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 8441   align(16);
 8442   BIND(L_wordByWord);
 8443     crc32(in_out, Address(in1, 0), 8);
 8444     addq(in1, 8);
 8445     cmpq(in1, tmp1);
 8446     jcc(Assembler::less, L_wordByWord);
 8447 
 8448   BIND(L_byteByByteProlog);
 8449   andl(in2, 0x00000007);
 8450   movl(tmp2, 1);
 8451 
 8452   cmpl(tmp2, in2);
 8453   jccb(Assembler::greater, L_exit);
 8454   BIND(L_byteByByte);
 8455     crc32(in_out, Address(in1, 0), 1);
 8456     incq(in1);
 8457     incl(tmp2);
 8458     cmpl(tmp2, in2);
 8459     jcc(Assembler::lessEqual, L_byteByByte);
 8460 
 8461   BIND(L_exit);
 8462 }
 8463 #undef BIND
 8464 #undef BLOCK_COMMENT
 8465 
 8466 // Compress char[] array to byte[].
 8467 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 8468 // Return the array length if every element in array can be encoded,
 8469 // otherwise, the index of first non-latin1 (> 0xff) character.
 8470 //   @IntrinsicCandidate
 8471 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 8472 //     for (int i = 0; i < len; i++) {
 8473 //       char c = src[srcOff];
 8474 //       if (c > 0xff) {
 8475 //           return i;  // return index of non-latin1 char
 8476 //       }
 8477 //       dst[dstOff] = (byte)c;
 8478 //       srcOff++;
 8479 //       dstOff++;
 8480 //     }
 8481 //     return len;
 8482 //   }
 8483 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 8484   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 8485   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 8486   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 8487   Label copy_chars_loop, done, reset_sp, copy_tail;
 8488 
 8489   // rsi: src
 8490   // rdi: dst
 8491   // rdx: len
 8492   // rcx: tmp5
 8493   // rax: result
 8494 
 8495   // rsi holds start addr of source char[] to be compressed
 8496   // rdi holds start addr of destination byte[]
 8497   // rdx holds length
 8498 
 8499   assert(len != result, "");
 8500 
 8501   // save length for return
 8502   movl(result, len);
 8503 
 8504   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 8505     VM_Version::supports_avx512vlbw() &&
 8506     VM_Version::supports_bmi2()) {
 8507 
 8508     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 8509 
 8510     // alignment
 8511     Label post_alignment;
 8512 
 8513     // if length of the string is less than 32, handle it the old fashioned way
 8514     testl(len, -32);
 8515     jcc(Assembler::zero, below_threshold);
 8516 
 8517     // First check whether a character is compressible ( <= 0xFF).
 8518     // Create mask to test for Unicode chars inside zmm vector
 8519     movl(tmp5, 0x00FF);
 8520     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 8521 
 8522     testl(len, -64);
 8523     jccb(Assembler::zero, post_alignment);
 8524 
 8525     movl(tmp5, dst);
 8526     andl(tmp5, (32 - 1));
 8527     negl(tmp5);
 8528     andl(tmp5, (32 - 1));
 8529 
 8530     // bail out when there is nothing to be done
 8531     testl(tmp5, 0xFFFFFFFF);
 8532     jccb(Assembler::zero, post_alignment);
 8533 
 8534     // ~(~0 << len), where len is the # of remaining elements to process
 8535     movl(len, 0xFFFFFFFF);
 8536     shlxl(len, len, tmp5);
 8537     notl(len);
 8538     kmovdl(mask2, len);
 8539     movl(len, result);
 8540 
 8541     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8542     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8543     ktestd(mask1, mask2);
 8544     jcc(Assembler::carryClear, copy_tail);
 8545 
 8546     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8547 
 8548     addptr(src, tmp5);
 8549     addptr(src, tmp5);
 8550     addptr(dst, tmp5);
 8551     subl(len, tmp5);
 8552 
 8553     bind(post_alignment);
 8554     // end of alignment
 8555 
 8556     movl(tmp5, len);
 8557     andl(tmp5, (32 - 1));    // tail count (in chars)
 8558     andl(len, ~(32 - 1));    // vector count (in chars)
 8559     jccb(Assembler::zero, copy_loop_tail);
 8560 
 8561     lea(src, Address(src, len, Address::times_2));
 8562     lea(dst, Address(dst, len, Address::times_1));
 8563     negptr(len);
 8564 
 8565     bind(copy_32_loop);
 8566     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 8567     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 8568     kortestdl(mask1, mask1);
 8569     jccb(Assembler::carryClear, reset_for_copy_tail);
 8570 
 8571     // All elements in current processed chunk are valid candidates for
 8572     // compression. Write a truncated byte elements to the memory.
 8573     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 8574     addptr(len, 32);
 8575     jccb(Assembler::notZero, copy_32_loop);
 8576 
 8577     bind(copy_loop_tail);
 8578     // bail out when there is nothing to be done
 8579     testl(tmp5, 0xFFFFFFFF);
 8580     jcc(Assembler::zero, done);
 8581 
 8582     movl(len, tmp5);
 8583 
 8584     // ~(~0 << len), where len is the # of remaining elements to process
 8585     movl(tmp5, 0xFFFFFFFF);
 8586     shlxl(tmp5, tmp5, len);
 8587     notl(tmp5);
 8588 
 8589     kmovdl(mask2, tmp5);
 8590 
 8591     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8592     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8593     ktestd(mask1, mask2);
 8594     jcc(Assembler::carryClear, copy_tail);
 8595 
 8596     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8597     jmp(done);
 8598 
 8599     bind(reset_for_copy_tail);
 8600     lea(src, Address(src, tmp5, Address::times_2));
 8601     lea(dst, Address(dst, tmp5, Address::times_1));
 8602     subptr(len, tmp5);
 8603     jmp(copy_chars_loop);
 8604 
 8605     bind(below_threshold);
 8606   }
 8607 
 8608   if (UseSSE42Intrinsics) {
 8609     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 8610 
 8611     // vectored compression
 8612     testl(len, 0xfffffff8);
 8613     jcc(Assembler::zero, copy_tail);
 8614 
 8615     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 8616     movdl(tmp1Reg, tmp5);
 8617     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 8618 
 8619     andl(len, 0xfffffff0);
 8620     jccb(Assembler::zero, copy_16);
 8621 
 8622     // compress 16 chars per iter
 8623     pxor(tmp4Reg, tmp4Reg);
 8624 
 8625     lea(src, Address(src, len, Address::times_2));
 8626     lea(dst, Address(dst, len, Address::times_1));
 8627     negptr(len);
 8628 
 8629     bind(copy_32_loop);
 8630     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 8631     por(tmp4Reg, tmp2Reg);
 8632     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 8633     por(tmp4Reg, tmp3Reg);
 8634     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 8635     jccb(Assembler::notZero, reset_for_copy_tail);
 8636     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 8637     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 8638     addptr(len, 16);
 8639     jccb(Assembler::notZero, copy_32_loop);
 8640 
 8641     // compress next vector of 8 chars (if any)
 8642     bind(copy_16);
 8643     // len = 0
 8644     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 8645     jccb(Assembler::zero, copy_tail_sse);
 8646 
 8647     pxor(tmp3Reg, tmp3Reg);
 8648 
 8649     movdqu(tmp2Reg, Address(src, 0));
 8650     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 8651     jccb(Assembler::notZero, reset_for_copy_tail);
 8652     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 8653     movq(Address(dst, 0), tmp2Reg);
 8654     addptr(src, 16);
 8655     addptr(dst, 8);
 8656     jmpb(copy_tail_sse);
 8657 
 8658     bind(reset_for_copy_tail);
 8659     movl(tmp5, result);
 8660     andl(tmp5, 0x0000000f);
 8661     lea(src, Address(src, tmp5, Address::times_2));
 8662     lea(dst, Address(dst, tmp5, Address::times_1));
 8663     subptr(len, tmp5);
 8664     jmpb(copy_chars_loop);
 8665 
 8666     bind(copy_tail_sse);
 8667     movl(len, result);
 8668     andl(len, 0x00000007);    // tail count (in chars)
 8669   }
 8670   // compress 1 char per iter
 8671   bind(copy_tail);
 8672   testl(len, len);
 8673   jccb(Assembler::zero, done);
 8674   lea(src, Address(src, len, Address::times_2));
 8675   lea(dst, Address(dst, len, Address::times_1));
 8676   negptr(len);
 8677 
 8678   bind(copy_chars_loop);
 8679   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 8680   testl(tmp5, 0xff00);      // check if Unicode char
 8681   jccb(Assembler::notZero, reset_sp);
 8682   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 8683   increment(len);
 8684   jccb(Assembler::notZero, copy_chars_loop);
 8685 
 8686   // add len then return (len will be zero if compress succeeded, otherwise negative)
 8687   bind(reset_sp);
 8688   addl(result, len);
 8689 
 8690   bind(done);
 8691 }
 8692 
 8693 // Inflate byte[] array to char[].
 8694 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 8695 //   @IntrinsicCandidate
 8696 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 8697 //     for (int i = 0; i < len; i++) {
 8698 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 8699 //     }
 8700 //   }
 8701 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 8702   XMMRegister tmp1, Register tmp2, KRegister mask) {
 8703   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 8704   // rsi: src
 8705   // rdi: dst
 8706   // rdx: len
 8707   // rcx: tmp2
 8708 
 8709   // rsi holds start addr of source byte[] to be inflated
 8710   // rdi holds start addr of destination char[]
 8711   // rdx holds length
 8712   assert_different_registers(src, dst, len, tmp2);
 8713   movl(tmp2, len);
 8714   if ((UseAVX > 2) && // AVX512
 8715     VM_Version::supports_avx512vlbw() &&
 8716     VM_Version::supports_bmi2()) {
 8717 
 8718     Label copy_32_loop, copy_tail;
 8719     Register tmp3_aliased = len;
 8720 
 8721     // if length of the string is less than 16, handle it in an old fashioned way
 8722     testl(len, -16);
 8723     jcc(Assembler::zero, below_threshold);
 8724 
 8725     testl(len, -1 * AVX3Threshold);
 8726     jcc(Assembler::zero, avx3_threshold);
 8727 
 8728     // In order to use only one arithmetic operation for the main loop we use
 8729     // this pre-calculation
 8730     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 8731     andl(len, -32);     // vector count
 8732     jccb(Assembler::zero, copy_tail);
 8733 
 8734     lea(src, Address(src, len, Address::times_1));
 8735     lea(dst, Address(dst, len, Address::times_2));
 8736     negptr(len);
 8737 
 8738 
 8739     // inflate 32 chars per iter
 8740     bind(copy_32_loop);
 8741     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 8742     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 8743     addptr(len, 32);
 8744     jcc(Assembler::notZero, copy_32_loop);
 8745 
 8746     bind(copy_tail);
 8747     // bail out when there is nothing to be done
 8748     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 8749     jcc(Assembler::zero, done);
 8750 
 8751     // ~(~0 << length), where length is the # of remaining elements to process
 8752     movl(tmp3_aliased, -1);
 8753     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 8754     notl(tmp3_aliased);
 8755     kmovdl(mask, tmp3_aliased);
 8756     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 8757     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 8758 
 8759     jmp(done);
 8760     bind(avx3_threshold);
 8761   }
 8762   if (UseSSE42Intrinsics) {
 8763     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 8764 
 8765     if (UseAVX > 1) {
 8766       andl(tmp2, (16 - 1));
 8767       andl(len, -16);
 8768       jccb(Assembler::zero, copy_new_tail);
 8769     } else {
 8770       andl(tmp2, 0x00000007);   // tail count (in chars)
 8771       andl(len, 0xfffffff8);    // vector count (in chars)
 8772       jccb(Assembler::zero, copy_tail);
 8773     }
 8774 
 8775     // vectored inflation
 8776     lea(src, Address(src, len, Address::times_1));
 8777     lea(dst, Address(dst, len, Address::times_2));
 8778     negptr(len);
 8779 
 8780     if (UseAVX > 1) {
 8781       bind(copy_16_loop);
 8782       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 8783       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 8784       addptr(len, 16);
 8785       jcc(Assembler::notZero, copy_16_loop);
 8786 
 8787       bind(below_threshold);
 8788       bind(copy_new_tail);
 8789       movl(len, tmp2);
 8790       andl(tmp2, 0x00000007);
 8791       andl(len, 0xFFFFFFF8);
 8792       jccb(Assembler::zero, copy_tail);
 8793 
 8794       pmovzxbw(tmp1, Address(src, 0));
 8795       movdqu(Address(dst, 0), tmp1);
 8796       addptr(src, 8);
 8797       addptr(dst, 2 * 8);
 8798 
 8799       jmp(copy_tail, true);
 8800     }
 8801 
 8802     // inflate 8 chars per iter
 8803     bind(copy_8_loop);
 8804     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 8805     movdqu(Address(dst, len, Address::times_2), tmp1);
 8806     addptr(len, 8);
 8807     jcc(Assembler::notZero, copy_8_loop);
 8808 
 8809     bind(copy_tail);
 8810     movl(len, tmp2);
 8811 
 8812     cmpl(len, 4);
 8813     jccb(Assembler::less, copy_bytes);
 8814 
 8815     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 8816     pmovzxbw(tmp1, tmp1);
 8817     movq(Address(dst, 0), tmp1);
 8818     subptr(len, 4);
 8819     addptr(src, 4);
 8820     addptr(dst, 8);
 8821 
 8822     bind(copy_bytes);
 8823   } else {
 8824     bind(below_threshold);
 8825   }
 8826 
 8827   testl(len, len);
 8828   jccb(Assembler::zero, done);
 8829   lea(src, Address(src, len, Address::times_1));
 8830   lea(dst, Address(dst, len, Address::times_2));
 8831   negptr(len);
 8832 
 8833   // inflate 1 char per iter
 8834   bind(copy_chars_loop);
 8835   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 8836   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 8837   increment(len);
 8838   jcc(Assembler::notZero, copy_chars_loop);
 8839 
 8840   bind(done);
 8841 }
 8842 
 8843 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 8844   switch(type) {
 8845     case T_BYTE:
 8846     case T_BOOLEAN:
 8847       evmovdqub(dst, kmask, src, merge, vector_len);
 8848       break;
 8849     case T_CHAR:
 8850     case T_SHORT:
 8851       evmovdquw(dst, kmask, src, merge, vector_len);
 8852       break;
 8853     case T_INT:
 8854     case T_FLOAT:
 8855       evmovdqul(dst, kmask, src, merge, vector_len);
 8856       break;
 8857     case T_LONG:
 8858     case T_DOUBLE:
 8859       evmovdquq(dst, kmask, src, merge, vector_len);
 8860       break;
 8861     default:
 8862       fatal("Unexpected type argument %s", type2name(type));
 8863       break;
 8864   }
 8865 }
 8866 
 8867 
 8868 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 8869   switch(type) {
 8870     case T_BYTE:
 8871     case T_BOOLEAN:
 8872       evmovdqub(dst, kmask, src, merge, vector_len);
 8873       break;
 8874     case T_CHAR:
 8875     case T_SHORT:
 8876       evmovdquw(dst, kmask, src, merge, vector_len);
 8877       break;
 8878     case T_INT:
 8879     case T_FLOAT:
 8880       evmovdqul(dst, kmask, src, merge, vector_len);
 8881       break;
 8882     case T_LONG:
 8883     case T_DOUBLE:
 8884       evmovdquq(dst, kmask, src, merge, vector_len);
 8885       break;
 8886     default:
 8887       fatal("Unexpected type argument %s", type2name(type));
 8888       break;
 8889   }
 8890 }
 8891 
 8892 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 8893   switch(type) {
 8894     case T_BYTE:
 8895     case T_BOOLEAN:
 8896       evmovdqub(dst, kmask, src, merge, vector_len);
 8897       break;
 8898     case T_CHAR:
 8899     case T_SHORT:
 8900       evmovdquw(dst, kmask, src, merge, vector_len);
 8901       break;
 8902     case T_INT:
 8903     case T_FLOAT:
 8904       evmovdqul(dst, kmask, src, merge, vector_len);
 8905       break;
 8906     case T_LONG:
 8907     case T_DOUBLE:
 8908       evmovdquq(dst, kmask, src, merge, vector_len);
 8909       break;
 8910     default:
 8911       fatal("Unexpected type argument %s", type2name(type));
 8912       break;
 8913   }
 8914 }
 8915 
 8916 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 8917   switch(masklen) {
 8918     case 2:
 8919        knotbl(dst, src);
 8920        movl(rtmp, 3);
 8921        kmovbl(ktmp, rtmp);
 8922        kandbl(dst, ktmp, dst);
 8923        break;
 8924     case 4:
 8925        knotbl(dst, src);
 8926        movl(rtmp, 15);
 8927        kmovbl(ktmp, rtmp);
 8928        kandbl(dst, ktmp, dst);
 8929        break;
 8930     case 8:
 8931        knotbl(dst, src);
 8932        break;
 8933     case 16:
 8934        knotwl(dst, src);
 8935        break;
 8936     case 32:
 8937        knotdl(dst, src);
 8938        break;
 8939     case 64:
 8940        knotql(dst, src);
 8941        break;
 8942     default:
 8943       fatal("Unexpected vector length %d", masklen);
 8944       break;
 8945   }
 8946 }
 8947 
 8948 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8949   switch(type) {
 8950     case T_BOOLEAN:
 8951     case T_BYTE:
 8952        kandbl(dst, src1, src2);
 8953        break;
 8954     case T_CHAR:
 8955     case T_SHORT:
 8956        kandwl(dst, src1, src2);
 8957        break;
 8958     case T_INT:
 8959     case T_FLOAT:
 8960        kanddl(dst, src1, src2);
 8961        break;
 8962     case T_LONG:
 8963     case T_DOUBLE:
 8964        kandql(dst, src1, src2);
 8965        break;
 8966     default:
 8967       fatal("Unexpected type argument %s", type2name(type));
 8968       break;
 8969   }
 8970 }
 8971 
 8972 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8973   switch(type) {
 8974     case T_BOOLEAN:
 8975     case T_BYTE:
 8976        korbl(dst, src1, src2);
 8977        break;
 8978     case T_CHAR:
 8979     case T_SHORT:
 8980        korwl(dst, src1, src2);
 8981        break;
 8982     case T_INT:
 8983     case T_FLOAT:
 8984        kordl(dst, src1, src2);
 8985        break;
 8986     case T_LONG:
 8987     case T_DOUBLE:
 8988        korql(dst, src1, src2);
 8989        break;
 8990     default:
 8991       fatal("Unexpected type argument %s", type2name(type));
 8992       break;
 8993   }
 8994 }
 8995 
 8996 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8997   switch(type) {
 8998     case T_BOOLEAN:
 8999     case T_BYTE:
 9000        kxorbl(dst, src1, src2);
 9001        break;
 9002     case T_CHAR:
 9003     case T_SHORT:
 9004        kxorwl(dst, src1, src2);
 9005        break;
 9006     case T_INT:
 9007     case T_FLOAT:
 9008        kxordl(dst, src1, src2);
 9009        break;
 9010     case T_LONG:
 9011     case T_DOUBLE:
 9012        kxorql(dst, src1, src2);
 9013        break;
 9014     default:
 9015       fatal("Unexpected type argument %s", type2name(type));
 9016       break;
 9017   }
 9018 }
 9019 
 9020 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9021   switch(type) {
 9022     case T_BOOLEAN:
 9023     case T_BYTE:
 9024       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9025     case T_CHAR:
 9026     case T_SHORT:
 9027       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9028     case T_INT:
 9029     case T_FLOAT:
 9030       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9031     case T_LONG:
 9032     case T_DOUBLE:
 9033       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9034     default:
 9035       fatal("Unexpected type argument %s", type2name(type)); break;
 9036   }
 9037 }
 9038 
 9039 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9040   switch(type) {
 9041     case T_BOOLEAN:
 9042     case T_BYTE:
 9043       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9044     case T_CHAR:
 9045     case T_SHORT:
 9046       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9047     case T_INT:
 9048     case T_FLOAT:
 9049       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9050     case T_LONG:
 9051     case T_DOUBLE:
 9052       evpermq(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, Address 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, Address 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::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9089   switch(type) {
 9090     case T_BYTE:
 9091       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9092     case T_SHORT:
 9093       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9094     case T_INT:
 9095       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9096     case T_LONG:
 9097       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9098     default:
 9099       fatal("Unexpected type argument %s", type2name(type)); break;
 9100   }
 9101 }
 9102 
 9103 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9104   switch(type) {
 9105     case T_BYTE:
 9106       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9107     case T_SHORT:
 9108       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9109     case T_INT:
 9110       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9111     case T_LONG:
 9112       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9113     default:
 9114       fatal("Unexpected type argument %s", type2name(type)); break;
 9115   }
 9116 }
 9117 
 9118 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9119   switch(type) {
 9120     case T_BYTE:
 9121       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9122     case T_SHORT:
 9123       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9124     case T_INT:
 9125       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9126     case T_LONG:
 9127       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9128     case T_FLOAT:
 9129       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9130     case T_DOUBLE:
 9131       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9132     default:
 9133       fatal("Unexpected type argument %s", type2name(type)); break;
 9134   }
 9135 }
 9136 
 9137 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9138   switch(type) {
 9139     case T_BYTE:
 9140       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9141     case T_SHORT:
 9142       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9143     case T_INT:
 9144       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9145     case T_LONG:
 9146       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9147     case T_FLOAT:
 9148       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9149     case T_DOUBLE:
 9150       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9151     default:
 9152       fatal("Unexpected type argument %s", type2name(type)); break;
 9153   }
 9154 }
 9155 
 9156 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9157   switch(type) {
 9158     case T_BYTE:
 9159       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9160     case T_SHORT:
 9161       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9162     case T_INT:
 9163       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9164     case T_LONG:
 9165       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9166     case T_FLOAT:
 9167       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9168     case T_DOUBLE:
 9169       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9170     default:
 9171       fatal("Unexpected type argument %s", type2name(type)); break;
 9172   }
 9173 }
 9174 
 9175 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9176   switch(type) {
 9177     case T_BYTE:
 9178       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9179     case T_SHORT:
 9180       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9181     case T_INT:
 9182       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9183     case T_LONG:
 9184       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9185     case T_FLOAT:
 9186       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9187     case T_DOUBLE:
 9188       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9189     default:
 9190       fatal("Unexpected type argument %s", type2name(type)); break;
 9191   }
 9192 }
 9193 
 9194 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9195   switch(type) {
 9196     case T_INT:
 9197       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9198     case T_LONG:
 9199       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9200     default:
 9201       fatal("Unexpected type argument %s", type2name(type)); break;
 9202   }
 9203 }
 9204 
 9205 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9206   switch(type) {
 9207     case T_INT:
 9208       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9209     case T_LONG:
 9210       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9211     default:
 9212       fatal("Unexpected type argument %s", type2name(type)); break;
 9213   }
 9214 }
 9215 
 9216 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9217   switch(type) {
 9218     case T_INT:
 9219       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9220     case T_LONG:
 9221       evporq(dst, mask, nds, src, merge, vector_len); break;
 9222     default:
 9223       fatal("Unexpected type argument %s", type2name(type)); break;
 9224   }
 9225 }
 9226 
 9227 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9228   switch(type) {
 9229     case T_INT:
 9230       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9231     case T_LONG:
 9232       evporq(dst, mask, nds, src, merge, vector_len); break;
 9233     default:
 9234       fatal("Unexpected type argument %s", type2name(type)); break;
 9235   }
 9236 }
 9237 
 9238 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9239   switch(type) {
 9240     case T_INT:
 9241       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9242     case T_LONG:
 9243       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9244     default:
 9245       fatal("Unexpected type argument %s", type2name(type)); break;
 9246   }
 9247 }
 9248 
 9249 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9250   switch(type) {
 9251     case T_INT:
 9252       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9253     case T_LONG:
 9254       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9255     default:
 9256       fatal("Unexpected type argument %s", type2name(type)); break;
 9257   }
 9258 }
 9259 
 9260 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9261   switch(masklen) {
 9262     case 8:
 9263        kortestbl(src1, src2);
 9264        break;
 9265     case 16:
 9266        kortestwl(src1, src2);
 9267        break;
 9268     case 32:
 9269        kortestdl(src1, src2);
 9270        break;
 9271     case 64:
 9272        kortestql(src1, src2);
 9273        break;
 9274     default:
 9275       fatal("Unexpected mask length %d", masklen);
 9276       break;
 9277   }
 9278 }
 9279 
 9280 
 9281 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
 9282   switch(masklen)  {
 9283     case 8:
 9284        ktestbl(src1, src2);
 9285        break;
 9286     case 16:
 9287        ktestwl(src1, src2);
 9288        break;
 9289     case 32:
 9290        ktestdl(src1, src2);
 9291        break;
 9292     case 64:
 9293        ktestql(src1, src2);
 9294        break;
 9295     default:
 9296       fatal("Unexpected mask length %d", masklen);
 9297       break;
 9298   }
 9299 }
 9300 
 9301 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9302   switch(type) {
 9303     case T_INT:
 9304       evprold(dst, mask, src, shift, merge, vlen_enc); break;
 9305     case T_LONG:
 9306       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
 9307     default:
 9308       fatal("Unexpected type argument %s", type2name(type)); break;
 9309       break;
 9310   }
 9311 }
 9312 
 9313 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9314   switch(type) {
 9315     case T_INT:
 9316       evprord(dst, mask, src, shift, merge, vlen_enc); break;
 9317     case T_LONG:
 9318       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
 9319     default:
 9320       fatal("Unexpected type argument %s", type2name(type)); break;
 9321   }
 9322 }
 9323 
 9324 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9325   switch(type) {
 9326     case T_INT:
 9327       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9328     case T_LONG:
 9329       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9330     default:
 9331       fatal("Unexpected type argument %s", type2name(type)); break;
 9332   }
 9333 }
 9334 
 9335 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9336   switch(type) {
 9337     case T_INT:
 9338       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9339     case T_LONG:
 9340       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9341     default:
 9342       fatal("Unexpected type argument %s", type2name(type)); break;
 9343   }
 9344 }
 9345 
 9346 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9347   assert(rscratch != noreg || always_reachable(src), "missing");
 9348 
 9349   if (reachable(src)) {
 9350     evpandq(dst, nds, as_Address(src), vector_len);
 9351   } else {
 9352     lea(rscratch, src);
 9353     evpandq(dst, nds, Address(rscratch, 0), vector_len);
 9354   }
 9355 }
 9356 
 9357 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 9358   assert(rscratch != noreg || always_reachable(src), "missing");
 9359 
 9360   if (reachable(src)) {
 9361     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
 9362   } else {
 9363     lea(rscratch, src);
 9364     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 9365   }
 9366 }
 9367 
 9368 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9369   assert(rscratch != noreg || always_reachable(src), "missing");
 9370 
 9371   if (reachable(src)) {
 9372     evporq(dst, nds, as_Address(src), vector_len);
 9373   } else {
 9374     lea(rscratch, src);
 9375     evporq(dst, nds, Address(rscratch, 0), vector_len);
 9376   }
 9377 }
 9378 
 9379 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9380   assert(rscratch != noreg || always_reachable(src), "missing");
 9381 
 9382   if (reachable(src)) {
 9383     vpshufb(dst, nds, as_Address(src), vector_len);
 9384   } else {
 9385     lea(rscratch, src);
 9386     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
 9387   }
 9388 }
 9389 
 9390 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9391   assert(rscratch != noreg || always_reachable(src), "missing");
 9392 
 9393   if (reachable(src)) {
 9394     Assembler::vpor(dst, nds, as_Address(src), vector_len);
 9395   } else {
 9396     lea(rscratch, src);
 9397     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
 9398   }
 9399 }
 9400 
 9401 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
 9402   assert(rscratch != noreg || always_reachable(src3), "missing");
 9403 
 9404   if (reachable(src3)) {
 9405     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
 9406   } else {
 9407     lea(rscratch, src3);
 9408     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
 9409   }
 9410 }
 9411 
 9412 #if COMPILER2_OR_JVMCI
 9413 
 9414 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
 9415                                  Register length, Register temp, int vec_enc) {
 9416   // Computing mask for predicated vector store.
 9417   movptr(temp, -1);
 9418   bzhiq(temp, temp, length);
 9419   kmov(mask, temp);
 9420   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
 9421 }
 9422 
 9423 // Set memory operation for length "less than" 64 bytes.
 9424 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
 9425                                        XMMRegister xmm, KRegister mask, Register length,
 9426                                        Register temp, bool use64byteVector) {
 9427   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9428   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9429   if (!use64byteVector) {
 9430     fill32(dst, disp, xmm);
 9431     subptr(length, 32 >> shift);
 9432     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
 9433   } else {
 9434     assert(MaxVectorSize == 64, "vector length != 64");
 9435     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
 9436   }
 9437 }
 9438 
 9439 
 9440 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
 9441                                        XMMRegister xmm, KRegister mask, Register length,
 9442                                        Register temp) {
 9443   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9444   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9445   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
 9446 }
 9447 
 9448 
 9449 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
 9450   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9451   vmovdqu(dst, xmm);
 9452 }
 9453 
 9454 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
 9455   fill32(Address(dst, disp), xmm);
 9456 }
 9457 
 9458 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
 9459   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9460   if (!use64byteVector) {
 9461     fill32(dst, xmm);
 9462     fill32(dst.plus_disp(32), xmm);
 9463   } else {
 9464     evmovdquq(dst, xmm, Assembler::AVX_512bit);
 9465   }
 9466 }
 9467 
 9468 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
 9469   fill64(Address(dst, disp), xmm, use64byteVector);
 9470 }
 9471 
 9472 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
 9473                                         Register count, Register rtmp, XMMRegister xtmp) {
 9474   Label L_exit;
 9475   Label L_fill_start;
 9476   Label L_fill_64_bytes;
 9477   Label L_fill_96_bytes;
 9478   Label L_fill_128_bytes;
 9479   Label L_fill_128_bytes_loop;
 9480   Label L_fill_128_loop_header;
 9481   Label L_fill_128_bytes_loop_header;
 9482   Label L_fill_128_bytes_loop_pre_header;
 9483   Label L_fill_zmm_sequence;
 9484 
 9485   int shift = -1;
 9486   int avx3threshold = VM_Version::avx3_threshold();
 9487   switch(type) {
 9488     case T_BYTE:  shift = 0;
 9489       break;
 9490     case T_SHORT: shift = 1;
 9491       break;
 9492     case T_INT:   shift = 2;
 9493       break;
 9494     /* Uncomment when LONG fill stubs are supported.
 9495     case T_LONG:  shift = 3;
 9496       break;
 9497     */
 9498     default:
 9499       fatal("Unhandled type: %s\n", type2name(type));
 9500   }
 9501 
 9502   if ((avx3threshold != 0)  || (MaxVectorSize == 32)) {
 9503 
 9504     if (MaxVectorSize == 64) {
 9505       cmpq(count, avx3threshold >> shift);
 9506       jcc(Assembler::greater, L_fill_zmm_sequence);
 9507     }
 9508 
 9509     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
 9510 
 9511     bind(L_fill_start);
 9512 
 9513     cmpq(count, 32 >> shift);
 9514     jccb(Assembler::greater, L_fill_64_bytes);
 9515     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9516     jmp(L_exit);
 9517 
 9518     bind(L_fill_64_bytes);
 9519     cmpq(count, 64 >> shift);
 9520     jccb(Assembler::greater, L_fill_96_bytes);
 9521     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9522     jmp(L_exit);
 9523 
 9524     bind(L_fill_96_bytes);
 9525     cmpq(count, 96 >> shift);
 9526     jccb(Assembler::greater, L_fill_128_bytes);
 9527     fill64(to, 0, xtmp);
 9528     subq(count, 64 >> shift);
 9529     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
 9530     jmp(L_exit);
 9531 
 9532     bind(L_fill_128_bytes);
 9533     cmpq(count, 128 >> shift);
 9534     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
 9535     fill64(to, 0, xtmp);
 9536     fill32(to, 64, xtmp);
 9537     subq(count, 96 >> shift);
 9538     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
 9539     jmp(L_exit);
 9540 
 9541     bind(L_fill_128_bytes_loop_pre_header);
 9542     {
 9543       mov(rtmp, to);
 9544       andq(rtmp, 31);
 9545       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
 9546       negq(rtmp);
 9547       addq(rtmp, 32);
 9548       mov64(r8, -1L);
 9549       bzhiq(r8, r8, rtmp);
 9550       kmovql(k2, r8);
 9551       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
 9552       addq(to, rtmp);
 9553       shrq(rtmp, shift);
 9554       subq(count, rtmp);
 9555     }
 9556 
 9557     cmpq(count, 128 >> shift);
 9558     jcc(Assembler::less, L_fill_start);
 9559 
 9560     bind(L_fill_128_bytes_loop_header);
 9561     subq(count, 128 >> shift);
 9562 
 9563     align32();
 9564     bind(L_fill_128_bytes_loop);
 9565       fill64(to, 0, xtmp);
 9566       fill64(to, 64, xtmp);
 9567       addq(to, 128);
 9568       subq(count, 128 >> shift);
 9569       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
 9570 
 9571     addq(count, 128 >> shift);
 9572     jcc(Assembler::zero, L_exit);
 9573     jmp(L_fill_start);
 9574   }
 9575 
 9576   if (MaxVectorSize == 64) {
 9577     // Sequence using 64 byte ZMM register.
 9578     Label L_fill_128_bytes_zmm;
 9579     Label L_fill_192_bytes_zmm;
 9580     Label L_fill_192_bytes_loop_zmm;
 9581     Label L_fill_192_bytes_loop_header_zmm;
 9582     Label L_fill_192_bytes_loop_pre_header_zmm;
 9583     Label L_fill_start_zmm_sequence;
 9584 
 9585     bind(L_fill_zmm_sequence);
 9586     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
 9587 
 9588     bind(L_fill_start_zmm_sequence);
 9589     cmpq(count, 64 >> shift);
 9590     jccb(Assembler::greater, L_fill_128_bytes_zmm);
 9591     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
 9592     jmp(L_exit);
 9593 
 9594     bind(L_fill_128_bytes_zmm);
 9595     cmpq(count, 128 >> shift);
 9596     jccb(Assembler::greater, L_fill_192_bytes_zmm);
 9597     fill64(to, 0, xtmp, true);
 9598     subq(count, 64 >> shift);
 9599     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
 9600     jmp(L_exit);
 9601 
 9602     bind(L_fill_192_bytes_zmm);
 9603     cmpq(count, 192 >> shift);
 9604     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
 9605     fill64(to, 0, xtmp, true);
 9606     fill64(to, 64, xtmp, true);
 9607     subq(count, 128 >> shift);
 9608     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
 9609     jmp(L_exit);
 9610 
 9611     bind(L_fill_192_bytes_loop_pre_header_zmm);
 9612     {
 9613       movq(rtmp, to);
 9614       andq(rtmp, 63);
 9615       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
 9616       negq(rtmp);
 9617       addq(rtmp, 64);
 9618       mov64(r8, -1L);
 9619       bzhiq(r8, r8, rtmp);
 9620       kmovql(k2, r8);
 9621       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
 9622       addq(to, rtmp);
 9623       shrq(rtmp, shift);
 9624       subq(count, rtmp);
 9625     }
 9626 
 9627     cmpq(count, 192 >> shift);
 9628     jcc(Assembler::less, L_fill_start_zmm_sequence);
 9629 
 9630     bind(L_fill_192_bytes_loop_header_zmm);
 9631     subq(count, 192 >> shift);
 9632 
 9633     align32();
 9634     bind(L_fill_192_bytes_loop_zmm);
 9635       fill64(to, 0, xtmp, true);
 9636       fill64(to, 64, xtmp, true);
 9637       fill64(to, 128, xtmp, true);
 9638       addq(to, 192);
 9639       subq(count, 192 >> shift);
 9640       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
 9641 
 9642     addq(count, 192 >> shift);
 9643     jcc(Assembler::zero, L_exit);
 9644     jmp(L_fill_start_zmm_sequence);
 9645   }
 9646   bind(L_exit);
 9647 }
 9648 #endif //COMPILER2_OR_JVMCI
 9649 
 9650 
 9651 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
 9652   Label done;
 9653   cvttss2sil(dst, src);
 9654   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9655   cmpl(dst, 0x80000000); // float_sign_flip
 9656   jccb(Assembler::notEqual, done);
 9657   subptr(rsp, 8);
 9658   movflt(Address(rsp, 0), src);
 9659   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
 9660   pop(dst);
 9661   bind(done);
 9662 }
 9663 
 9664 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
 9665   Label done;
 9666   cvttsd2sil(dst, src);
 9667   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9668   cmpl(dst, 0x80000000); // float_sign_flip
 9669   jccb(Assembler::notEqual, done);
 9670   subptr(rsp, 8);
 9671   movdbl(Address(rsp, 0), src);
 9672   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
 9673   pop(dst);
 9674   bind(done);
 9675 }
 9676 
 9677 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
 9678   Label done;
 9679   cvttss2siq(dst, src);
 9680   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9681   jccb(Assembler::notEqual, done);
 9682   subptr(rsp, 8);
 9683   movflt(Address(rsp, 0), src);
 9684   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
 9685   pop(dst);
 9686   bind(done);
 9687 }
 9688 
 9689 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9690   // Following code is line by line assembly translation rounding algorithm.
 9691   // Please refer to java.lang.Math.round(float) algorithm for details.
 9692   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
 9693   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
 9694   const int32_t FloatConsts_EXP_BIAS = 127;
 9695   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
 9696   const int32_t MINUS_32 = 0xFFFFFFE0;
 9697   Label L_special_case, L_block1, L_exit;
 9698   movl(rtmp, FloatConsts_EXP_BIT_MASK);
 9699   movdl(dst, src);
 9700   andl(dst, rtmp);
 9701   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
 9702   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
 9703   subl(rtmp, dst);
 9704   movl(rcx, rtmp);
 9705   movl(dst, MINUS_32);
 9706   testl(rtmp, dst);
 9707   jccb(Assembler::notEqual, L_special_case);
 9708   movdl(dst, src);
 9709   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
 9710   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
 9711   movdl(rtmp, src);
 9712   testl(rtmp, rtmp);
 9713   jccb(Assembler::greaterEqual, L_block1);
 9714   negl(dst);
 9715   bind(L_block1);
 9716   sarl(dst);
 9717   addl(dst, 0x1);
 9718   sarl(dst, 0x1);
 9719   jmp(L_exit);
 9720   bind(L_special_case);
 9721   convert_f2i(dst, src);
 9722   bind(L_exit);
 9723 }
 9724 
 9725 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9726   // Following code is line by line assembly translation rounding algorithm.
 9727   // Please refer to java.lang.Math.round(double) algorithm for details.
 9728   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
 9729   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
 9730   const int64_t DoubleConsts_EXP_BIAS = 1023;
 9731   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
 9732   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
 9733   Label L_special_case, L_block1, L_exit;
 9734   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
 9735   movq(dst, src);
 9736   andq(dst, rtmp);
 9737   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
 9738   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
 9739   subq(rtmp, dst);
 9740   movq(rcx, rtmp);
 9741   mov64(dst, MINUS_64);
 9742   testq(rtmp, dst);
 9743   jccb(Assembler::notEqual, L_special_case);
 9744   movq(dst, src);
 9745   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
 9746   andq(dst, rtmp);
 9747   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
 9748   orq(dst, rtmp);
 9749   movq(rtmp, src);
 9750   testq(rtmp, rtmp);
 9751   jccb(Assembler::greaterEqual, L_block1);
 9752   negq(dst);
 9753   bind(L_block1);
 9754   sarq(dst);
 9755   addq(dst, 0x1);
 9756   sarq(dst, 0x1);
 9757   jmp(L_exit);
 9758   bind(L_special_case);
 9759   convert_d2l(dst, src);
 9760   bind(L_exit);
 9761 }
 9762 
 9763 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
 9764   Label done;
 9765   cvttsd2siq(dst, src);
 9766   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9767   jccb(Assembler::notEqual, done);
 9768   subptr(rsp, 8);
 9769   movdbl(Address(rsp, 0), src);
 9770   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
 9771   pop(dst);
 9772   bind(done);
 9773 }
 9774 
 9775 void MacroAssembler::cache_wb(Address line)
 9776 {
 9777   // 64 bit cpus always support clflush
 9778   assert(VM_Version::supports_clflush(), "clflush should be available");
 9779   bool optimized = VM_Version::supports_clflushopt();
 9780   bool no_evict = VM_Version::supports_clwb();
 9781 
 9782   // prefer clwb (writeback without evict) otherwise
 9783   // prefer clflushopt (potentially parallel writeback with evict)
 9784   // otherwise fallback on clflush (serial writeback with evict)
 9785 
 9786   if (optimized) {
 9787     if (no_evict) {
 9788       clwb(line);
 9789     } else {
 9790       clflushopt(line);
 9791     }
 9792   } else {
 9793     // no need for fence when using CLFLUSH
 9794     clflush(line);
 9795   }
 9796 }
 9797 
 9798 void MacroAssembler::cache_wbsync(bool is_pre)
 9799 {
 9800   assert(VM_Version::supports_clflush(), "clflush should be available");
 9801   bool optimized = VM_Version::supports_clflushopt();
 9802   bool no_evict = VM_Version::supports_clwb();
 9803 
 9804   // pick the correct implementation
 9805 
 9806   if (!is_pre && (optimized || no_evict)) {
 9807     // need an sfence for post flush when using clflushopt or clwb
 9808     // otherwise no no need for any synchroniaztion
 9809 
 9810     sfence();
 9811   }
 9812 }
 9813 
 9814 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
 9815   switch (cond) {
 9816     // Note some conditions are synonyms for others
 9817     case Assembler::zero:         return Assembler::notZero;
 9818     case Assembler::notZero:      return Assembler::zero;
 9819     case Assembler::less:         return Assembler::greaterEqual;
 9820     case Assembler::lessEqual:    return Assembler::greater;
 9821     case Assembler::greater:      return Assembler::lessEqual;
 9822     case Assembler::greaterEqual: return Assembler::less;
 9823     case Assembler::below:        return Assembler::aboveEqual;
 9824     case Assembler::belowEqual:   return Assembler::above;
 9825     case Assembler::above:        return Assembler::belowEqual;
 9826     case Assembler::aboveEqual:   return Assembler::below;
 9827     case Assembler::overflow:     return Assembler::noOverflow;
 9828     case Assembler::noOverflow:   return Assembler::overflow;
 9829     case Assembler::negative:     return Assembler::positive;
 9830     case Assembler::positive:     return Assembler::negative;
 9831     case Assembler::parity:       return Assembler::noParity;
 9832     case Assembler::noParity:     return Assembler::parity;
 9833   }
 9834   ShouldNotReachHere(); return Assembler::overflow;
 9835 }
 9836 
 9837 // This is simply a call to Thread::current()
 9838 void MacroAssembler::get_thread_slow(Register thread) {
 9839   if (thread != rax) {
 9840     push(rax);
 9841   }
 9842   push(rdi);
 9843   push(rsi);
 9844   push(rdx);
 9845   push(rcx);
 9846   push(r8);
 9847   push(r9);
 9848   push(r10);
 9849   push(r11);
 9850 
 9851   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
 9852 
 9853   pop(r11);
 9854   pop(r10);
 9855   pop(r9);
 9856   pop(r8);
 9857   pop(rcx);
 9858   pop(rdx);
 9859   pop(rsi);
 9860   pop(rdi);
 9861   if (thread != rax) {
 9862     mov(thread, rax);
 9863     pop(rax);
 9864   }
 9865 }
 9866 
 9867 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
 9868   Label L_stack_ok;
 9869   if (bias == 0) {
 9870     testptr(sp, 2 * wordSize - 1);
 9871   } else {
 9872     // lea(tmp, Address(rsp, bias);
 9873     mov(tmp, sp);
 9874     addptr(tmp, bias);
 9875     testptr(tmp, 2 * wordSize - 1);
 9876   }
 9877   jcc(Assembler::equal, L_stack_ok);
 9878   block_comment(msg);
 9879   stop(msg);
 9880   bind(L_stack_ok);
 9881 }
 9882 
 9883 // Implements fast-locking.
 9884 //
 9885 // obj: the object to be locked
 9886 // reg_rax: rax
 9887 // thread: the thread which attempts to lock obj
 9888 // tmp: a temporary register
 9889 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
 9890   Register thread = r15_thread;
 9891 
 9892   assert(reg_rax == rax, "");
 9893   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
 9894 
 9895   Label push;
 9896   const Register top = tmp;
 9897 
 9898   // Preload the markWord. It is important that this is the first
 9899   // instruction emitted as it is part of C1's null check semantics.
 9900   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9901 
 9902   if (UseObjectMonitorTable) {
 9903     // Clear cache in case fast locking succeeds or we need to take the slow-path.
 9904     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
 9905   }
 9906 
 9907   if (DiagnoseSyncOnValueBasedClasses != 0) {
 9908     load_klass(tmp, obj, rscratch1);
 9909     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
 9910     jcc(Assembler::notZero, slow);
 9911   }
 9912 
 9913   // Load top.
 9914   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9915 
 9916   // Check if the lock-stack is full.
 9917   cmpl(top, LockStack::end_offset());
 9918   jcc(Assembler::greaterEqual, slow);
 9919 
 9920   // Check for recursion.
 9921   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9922   jcc(Assembler::equal, push);
 9923 
 9924   // Check header for monitor (0b10).
 9925   testptr(reg_rax, markWord::monitor_value);
 9926   jcc(Assembler::notZero, slow);
 9927 
 9928   // Try to lock. Transition lock bits 0b01 => 0b00
 9929   movptr(tmp, reg_rax);
 9930   andptr(tmp, ~(int32_t)markWord::unlocked_value);
 9931   orptr(reg_rax, markWord::unlocked_value);



 9932   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9933   jcc(Assembler::notEqual, slow);
 9934 
 9935   // Restore top, CAS clobbers register.
 9936   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9937 
 9938   bind(push);
 9939   // After successful lock, push object on lock-stack.
 9940   movptr(Address(thread, top), obj);
 9941   incrementl(top, oopSize);
 9942   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
 9943 }
 9944 
 9945 // Implements fast-unlocking.
 9946 //
 9947 // obj: the object to be unlocked
 9948 // reg_rax: rax
 9949 // thread: the thread
 9950 // tmp: a temporary register
 9951 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
 9952   Register thread = r15_thread;
 9953 
 9954   assert(reg_rax == rax, "");
 9955   assert_different_registers(obj, reg_rax, thread, tmp);
 9956 
 9957   Label unlocked, push_and_slow;
 9958   const Register top = tmp;
 9959 
 9960   // Check if obj is top of lock-stack.
 9961   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9962   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9963   jcc(Assembler::notEqual, slow);
 9964 
 9965   // Pop lock-stack.
 9966   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
 9967   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
 9968 
 9969   // Check if recursive.
 9970   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
 9971   jcc(Assembler::equal, unlocked);
 9972 
 9973   // Not recursive. Check header for monitor (0b10).
 9974   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9975   testptr(reg_rax, markWord::monitor_value);
 9976   jcc(Assembler::notZero, push_and_slow);
 9977 
 9978 #ifdef ASSERT
 9979   // Check header not unlocked (0b01).
 9980   Label not_unlocked;
 9981   testptr(reg_rax, markWord::unlocked_value);
 9982   jcc(Assembler::zero, not_unlocked);
 9983   stop("fast_unlock already unlocked");
 9984   bind(not_unlocked);
 9985 #endif
 9986 
 9987   // Try to unlock. Transition lock bits 0b00 => 0b01
 9988   movptr(tmp, reg_rax);
 9989   orptr(tmp, markWord::unlocked_value);
 9990   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9991   jcc(Assembler::equal, unlocked);
 9992 
 9993   bind(push_and_slow);
 9994   // Restore lock-stack and handle the unlock in runtime.
 9995 #ifdef ASSERT
 9996   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9997   movptr(Address(thread, top), obj);
 9998 #endif
 9999   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10000   jmp(slow);
10001 
10002   bind(unlocked);
10003 }
10004 
10005 // Saves legacy GPRs state on stack.
10006 void MacroAssembler::save_legacy_gprs() {
10007   subq(rsp, 16 * wordSize);
10008   movq(Address(rsp, 15 * wordSize), rax);
10009   movq(Address(rsp, 14 * wordSize), rcx);
10010   movq(Address(rsp, 13 * wordSize), rdx);
10011   movq(Address(rsp, 12 * wordSize), rbx);
10012   movq(Address(rsp, 10 * wordSize), rbp);
10013   movq(Address(rsp, 9 * wordSize), rsi);
10014   movq(Address(rsp, 8 * wordSize), rdi);
10015   movq(Address(rsp, 7 * wordSize), r8);
10016   movq(Address(rsp, 6 * wordSize), r9);
10017   movq(Address(rsp, 5 * wordSize), r10);
10018   movq(Address(rsp, 4 * wordSize), r11);
10019   movq(Address(rsp, 3 * wordSize), r12);
10020   movq(Address(rsp, 2 * wordSize), r13);
10021   movq(Address(rsp, wordSize), r14);
10022   movq(Address(rsp, 0), r15);
10023 }
10024 
10025 // Resotres back legacy GPRs state from stack.
10026 void MacroAssembler::restore_legacy_gprs() {
10027   movq(r15, Address(rsp, 0));
10028   movq(r14, Address(rsp, wordSize));
10029   movq(r13, Address(rsp, 2 * wordSize));
10030   movq(r12, Address(rsp, 3 * wordSize));
10031   movq(r11, Address(rsp, 4 * wordSize));
10032   movq(r10, Address(rsp, 5 * wordSize));
10033   movq(r9,  Address(rsp, 6 * wordSize));
10034   movq(r8,  Address(rsp, 7 * wordSize));
10035   movq(rdi, Address(rsp, 8 * wordSize));
10036   movq(rsi, Address(rsp, 9 * wordSize));
10037   movq(rbp, Address(rsp, 10 * wordSize));
10038   movq(rbx, Address(rsp, 12 * wordSize));
10039   movq(rdx, Address(rsp, 13 * wordSize));
10040   movq(rcx, Address(rsp, 14 * wordSize));
10041   movq(rax, Address(rsp, 15 * wordSize));
10042   addq(rsp, 16 * wordSize);
10043 }
10044 
10045 void MacroAssembler::load_aotrc_address(Register reg, address a) {
10046 #if INCLUDE_CDS
10047   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
10048   if (AOTCodeCache::is_on_for_dump()) {
10049     // all aotrc field addresses should be registered in the AOTCodeCache address table
10050     lea(reg, ExternalAddress(a));
10051   } else {
10052     mov64(reg, (uint64_t)a);
10053   }
10054 #else
10055   ShouldNotReachHere();
10056 #endif
10057 }
10058 
10059 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10060   if (VM_Version::supports_apx_f()) {
10061     esetzucc(comparison, dst);
10062   } else {
10063     setb(comparison, dst);
10064     movzbl(dst, dst);
10065   }
10066 }
--- EOF ---