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   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  389   lea(c_rarg0, ExternalAddress((address) str));
  390   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  391 
  392 #ifdef _WIN64
  393   // restore stack pointer
  394   addq(rsp, frame::arg_reg_save_area_bytes);
  395 #endif
  396   pop_CPU_state();
  397   mov(rsp, rbp);
  398   pop(rbp);
  399 }
  400 
  401 void MacroAssembler::print_state() {
  402   address rip = pc();
  403   pusha();            // get regs on stack
  404   push(rbp);
  405   movq(rbp, rsp);
  406   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  407   push_CPU_state();   // keeps alignment at 16 bytes
  408 
  409   lea(c_rarg0, InternalAddress(rip));
  410   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  411   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  412 
  413   pop_CPU_state();
  414   mov(rsp, rbp);
  415   pop(rbp);
  416   popa();
  417 }
  418 
  419 #ifndef PRODUCT
  420 extern "C" void findpc(intptr_t x);
  421 #endif
  422 
  423 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  424   // In order to get locks to work, we need to fake a in_VM state
  425   if (ShowMessageBoxOnError) {
  426     JavaThread* thread = JavaThread::current();
  427     JavaThreadState saved_state = thread->thread_state();
  428     thread->set_thread_state(_thread_in_vm);
  429 #ifndef PRODUCT
  430     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  431       ttyLocker ttyl;
  432       BytecodeCounter::print();
  433     }
  434 #endif
  435     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  436     // XXX correct this offset for amd64
  437     // This is the value of eip which points to where verify_oop will return.
  438     if (os::message_box(msg, "Execution stopped, print registers?")) {
  439       print_state64(pc, regs);
  440       BREAKPOINT;
  441     }
  442   }
  443   fatal("DEBUG MESSAGE: %s", msg);
  444 }
  445 
  446 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  447   ttyLocker ttyl;
  448   DebuggingContext debugging{};
  449   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  450 #ifndef PRODUCT
  451   tty->cr();
  452   findpc(pc);
  453   tty->cr();
  454 #endif
  455 #define PRINT_REG(rax, value) \
  456   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  457   PRINT_REG(rax, regs[15]);
  458   PRINT_REG(rbx, regs[12]);
  459   PRINT_REG(rcx, regs[14]);
  460   PRINT_REG(rdx, regs[13]);
  461   PRINT_REG(rdi, regs[8]);
  462   PRINT_REG(rsi, regs[9]);
  463   PRINT_REG(rbp, regs[10]);
  464   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  465   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  466   PRINT_REG(r8 , regs[7]);
  467   PRINT_REG(r9 , regs[6]);
  468   PRINT_REG(r10, regs[5]);
  469   PRINT_REG(r11, regs[4]);
  470   PRINT_REG(r12, regs[3]);
  471   PRINT_REG(r13, regs[2]);
  472   PRINT_REG(r14, regs[1]);
  473   PRINT_REG(r15, regs[0]);
  474 #undef PRINT_REG
  475   // Print some words near the top of the stack.
  476   int64_t* rsp = &regs[16];
  477   int64_t* dump_sp = rsp;
  478   for (int col1 = 0; col1 < 8; col1++) {
  479     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  480     os::print_location(tty, *dump_sp++);
  481   }
  482   for (int row = 0; row < 25; row++) {
  483     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  484     for (int col = 0; col < 4; col++) {
  485       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  486     }
  487     tty->cr();
  488   }
  489   // Print some instructions around pc:
  490   Disassembler::decode((address)pc-64, (address)pc);
  491   tty->print_cr("--------");
  492   Disassembler::decode((address)pc, (address)pc+32);
  493 }
  494 
  495 // The java_calling_convention describes stack locations as ideal slots on
  496 // a frame with no abi restrictions. Since we must observe abi restrictions
  497 // (like the placement of the register window) the slots must be biased by
  498 // the following value.
  499 static int reg2offset_in(VMReg r) {
  500   // Account for saved rbp and return address
  501   // This should really be in_preserve_stack_slots
  502   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  503 }
  504 
  505 static int reg2offset_out(VMReg r) {
  506   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  507 }
  508 
  509 // A long move
  510 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  511 
  512   // The calling conventions assures us that each VMregpair is either
  513   // all really one physical register or adjacent stack slots.
  514 
  515   if (src.is_single_phys_reg() ) {
  516     if (dst.is_single_phys_reg()) {
  517       if (dst.first() != src.first()) {
  518         mov(dst.first()->as_Register(), src.first()->as_Register());
  519       }
  520     } else {
  521       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  522              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  523       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  524     }
  525   } else if (dst.is_single_phys_reg()) {
  526     assert(src.is_single_reg(),  "not a stack pair");
  527     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  528   } else {
  529     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  530     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  531     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  532   }
  533 }
  534 
  535 // A double move
  536 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  537 
  538   // The calling conventions assures us that each VMregpair is either
  539   // all really one physical register or adjacent stack slots.
  540 
  541   if (src.is_single_phys_reg() ) {
  542     if (dst.is_single_phys_reg()) {
  543       // In theory these overlap but the ordering is such that this is likely a nop
  544       if ( src.first() != dst.first()) {
  545         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  546       }
  547     } else {
  548       assert(dst.is_single_reg(), "not a stack pair");
  549       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  550     }
  551   } else if (dst.is_single_phys_reg()) {
  552     assert(src.is_single_reg(),  "not a stack pair");
  553     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  554   } else {
  555     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  556     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  557     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  558   }
  559 }
  560 
  561 
  562 // A float arg may have to do float reg int reg conversion
  563 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  564   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  565 
  566   // The calling conventions assures us that each VMregpair is either
  567   // all really one physical register or adjacent stack slots.
  568 
  569   if (src.first()->is_stack()) {
  570     if (dst.first()->is_stack()) {
  571       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  572       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  573     } else {
  574       // stack to reg
  575       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  576       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  577     }
  578   } else if (dst.first()->is_stack()) {
  579     // reg to stack
  580     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  581     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  582   } else {
  583     // reg to reg
  584     // In theory these overlap but the ordering is such that this is likely a nop
  585     if ( src.first() != dst.first()) {
  586       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  587     }
  588   }
  589 }
  590 
  591 // On 64 bit we will store integer like items to the stack as
  592 // 64 bits items (x86_32/64 abi) even though java would only store
  593 // 32bits for a parameter. On 32bit it will simply be 32 bits
  594 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  595 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  596   if (src.first()->is_stack()) {
  597     if (dst.first()->is_stack()) {
  598       // stack to stack
  599       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  600       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  601     } else {
  602       // stack to reg
  603       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  604     }
  605   } else if (dst.first()->is_stack()) {
  606     // reg to stack
  607     // Do we really have to sign extend???
  608     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  609     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  610   } else {
  611     // Do we really have to sign extend???
  612     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
  613     if (dst.first() != src.first()) {
  614       movq(dst.first()->as_Register(), src.first()->as_Register());
  615     }
  616   }
  617 }
  618 
  619 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
  620   if (src.first()->is_stack()) {
  621     if (dst.first()->is_stack()) {
  622       // stack to stack
  623       movq(rax, Address(rbp, reg2offset_in(src.first())));
  624       movq(Address(rsp, reg2offset_out(dst.first())), rax);
  625     } else {
  626       // stack to reg
  627       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
  628     }
  629   } else if (dst.first()->is_stack()) {
  630     // reg to stack
  631     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
  632   } else {
  633     if (dst.first() != src.first()) {
  634       movq(dst.first()->as_Register(), src.first()->as_Register());
  635     }
  636   }
  637 }
  638 
  639 // An oop arg. Must pass a handle not the oop itself
  640 void MacroAssembler::object_move(OopMap* map,
  641                         int oop_handle_offset,
  642                         int framesize_in_slots,
  643                         VMRegPair src,
  644                         VMRegPair dst,
  645                         bool is_receiver,
  646                         int* receiver_offset) {
  647 
  648   // must pass a handle. First figure out the location we use as a handle
  649 
  650   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
  651 
  652   // See if oop is null if it is we need no handle
  653 
  654   if (src.first()->is_stack()) {
  655 
  656     // Oop is already on the stack as an argument
  657     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
  658     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
  659     if (is_receiver) {
  660       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
  661     }
  662 
  663     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
  664     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
  665     // conditionally move a null
  666     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
  667   } else {
  668 
  669     // Oop is in a register we must store it to the space we reserve
  670     // on the stack for oop_handles and pass a handle if oop is non-null
  671 
  672     const Register rOop = src.first()->as_Register();
  673     int oop_slot;
  674     if (rOop == j_rarg0)
  675       oop_slot = 0;
  676     else if (rOop == j_rarg1)
  677       oop_slot = 1;
  678     else if (rOop == j_rarg2)
  679       oop_slot = 2;
  680     else if (rOop == j_rarg3)
  681       oop_slot = 3;
  682     else if (rOop == j_rarg4)
  683       oop_slot = 4;
  684     else {
  685       assert(rOop == j_rarg5, "wrong register");
  686       oop_slot = 5;
  687     }
  688 
  689     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
  690     int offset = oop_slot*VMRegImpl::stack_slot_size;
  691 
  692     map->set_oop(VMRegImpl::stack2reg(oop_slot));
  693     // Store oop in handle area, may be null
  694     movptr(Address(rsp, offset), rOop);
  695     if (is_receiver) {
  696       *receiver_offset = offset;
  697     }
  698 
  699     cmpptr(rOop, NULL_WORD);
  700     lea(rHandle, Address(rsp, offset));
  701     // conditionally move a null from the handle area where it was just stored
  702     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
  703   }
  704 
  705   // If arg is on the stack then place it otherwise it is already in correct reg.
  706   if (dst.first()->is_stack()) {
  707     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
  708   }
  709 }
  710 
  711 void MacroAssembler::addptr(Register dst, int32_t imm32) {
  712   addq(dst, imm32);
  713 }
  714 
  715 void MacroAssembler::addptr(Register dst, Register src) {
  716   addq(dst, src);
  717 }
  718 
  719 void MacroAssembler::addptr(Address dst, Register src) {
  720   addq(dst, src);
  721 }
  722 
  723 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  724   assert(rscratch != noreg || always_reachable(src), "missing");
  725 
  726   if (reachable(src)) {
  727     Assembler::addsd(dst, as_Address(src));
  728   } else {
  729     lea(rscratch, src);
  730     Assembler::addsd(dst, Address(rscratch, 0));
  731   }
  732 }
  733 
  734 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
  735   assert(rscratch != noreg || always_reachable(src), "missing");
  736 
  737   if (reachable(src)) {
  738     addss(dst, as_Address(src));
  739   } else {
  740     lea(rscratch, src);
  741     addss(dst, Address(rscratch, 0));
  742   }
  743 }
  744 
  745 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  746   assert(rscratch != noreg || always_reachable(src), "missing");
  747 
  748   if (reachable(src)) {
  749     Assembler::addpd(dst, as_Address(src));
  750   } else {
  751     lea(rscratch, src);
  752     Assembler::addpd(dst, Address(rscratch, 0));
  753   }
  754 }
  755 
  756 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
  757 // Stub code is generated once and never copied.
  758 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
  759 void MacroAssembler::align64() {
  760   align(64, (uint)(uintptr_t)pc());
  761 }
  762 
  763 void MacroAssembler::align32() {
  764   align(32, (uint)(uintptr_t)pc());
  765 }
  766 
  767 void MacroAssembler::align(uint modulus) {
  768   // 8273459: Ensure alignment is possible with current segment alignment
  769   assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
  770   align(modulus, offset());
  771 }
  772 
  773 void MacroAssembler::align(uint modulus, uint target) {
  774   if (target % modulus != 0) {
  775     nop(modulus - (target % modulus));
  776   }
  777 }
  778 
  779 void MacroAssembler::push_f(XMMRegister r) {
  780   subptr(rsp, wordSize);
  781   movflt(Address(rsp, 0), r);
  782 }
  783 
  784 void MacroAssembler::pop_f(XMMRegister r) {
  785   movflt(r, Address(rsp, 0));
  786   addptr(rsp, wordSize);
  787 }
  788 
  789 void MacroAssembler::push_d(XMMRegister r) {
  790   subptr(rsp, 2 * wordSize);
  791   movdbl(Address(rsp, 0), r);
  792 }
  793 
  794 void MacroAssembler::pop_d(XMMRegister r) {
  795   movdbl(r, Address(rsp, 0));
  796   addptr(rsp, 2 * Interpreter::stackElementSize);
  797 }
  798 
  799 void MacroAssembler::push_ppx(Register src) {
  800   if (VM_Version::supports_apx_f()) {
  801     pushp(src);
  802   } else {
  803     Assembler::push(src);
  804   }
  805 }
  806 
  807 void MacroAssembler::pop_ppx(Register dst) {
  808   if (VM_Version::supports_apx_f()) {
  809     popp(dst);
  810   } else {
  811     Assembler::pop(dst);
  812   }
  813 }
  814 
  815 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  816   // Used in sign-masking with aligned address.
  817   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  818   assert(rscratch != noreg || always_reachable(src), "missing");
  819 
  820   if (UseAVX > 2 &&
  821       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
  822       (dst->encoding() >= 16)) {
  823     vpand(dst, dst, src, AVX_512bit, rscratch);
  824   } else if (reachable(src)) {
  825     Assembler::andpd(dst, as_Address(src));
  826   } else {
  827     lea(rscratch, src);
  828     Assembler::andpd(dst, Address(rscratch, 0));
  829   }
  830 }
  831 
  832 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
  833   // Used in sign-masking with aligned address.
  834   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  835   assert(rscratch != noreg || always_reachable(src), "missing");
  836 
  837   if (reachable(src)) {
  838     Assembler::andps(dst, as_Address(src));
  839   } else {
  840     lea(rscratch, src);
  841     Assembler::andps(dst, Address(rscratch, 0));
  842   }
  843 }
  844 
  845 void MacroAssembler::andptr(Register dst, int32_t imm32) {
  846   andq(dst, imm32);
  847 }
  848 
  849 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
  850   assert(rscratch != noreg || always_reachable(src), "missing");
  851 
  852   if (reachable(src)) {
  853     andq(dst, as_Address(src));
  854   } else {
  855     lea(rscratch, src);
  856     andq(dst, Address(rscratch, 0));
  857   }
  858 }
  859 
  860 void MacroAssembler::atomic_incl(Address counter_addr) {
  861   lock();
  862   incrementl(counter_addr);
  863 }
  864 
  865 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
  866   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  867 
  868   if (reachable(counter_addr)) {
  869     atomic_incl(as_Address(counter_addr));
  870   } else {
  871     lea(rscratch, counter_addr);
  872     atomic_incl(Address(rscratch, 0));
  873   }
  874 }
  875 
  876 void MacroAssembler::atomic_incq(Address counter_addr) {
  877   lock();
  878   incrementq(counter_addr);
  879 }
  880 
  881 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
  882   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  883 
  884   if (reachable(counter_addr)) {
  885     atomic_incq(as_Address(counter_addr));
  886   } else {
  887     lea(rscratch, counter_addr);
  888     atomic_incq(Address(rscratch, 0));
  889   }
  890 }
  891 
  892 // Writes to stack successive pages until offset reached to check for
  893 // stack overflow + shadow pages.  This clobbers tmp.
  894 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
  895   movptr(tmp, rsp);
  896   // Bang stack for total size given plus shadow page size.
  897   // Bang one page at a time because large size can bang beyond yellow and
  898   // red zones.
  899   Label loop;
  900   bind(loop);
  901   movl(Address(tmp, (-(int)os::vm_page_size())), size );
  902   subptr(tmp, (int)os::vm_page_size());
  903   subl(size, (int)os::vm_page_size());
  904   jcc(Assembler::greater, loop);
  905 
  906   // Bang down shadow pages too.
  907   // At this point, (tmp-0) is the last address touched, so don't
  908   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
  909   // was post-decremented.)  Skip this address by starting at i=1, and
  910   // touch a few more pages below.  N.B.  It is important to touch all
  911   // the way down including all pages in the shadow zone.
  912   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
  913     // this could be any sized move but this is can be a debugging crumb
  914     // so the bigger the better.
  915     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
  916   }
  917 }
  918 
  919 void MacroAssembler::reserved_stack_check() {
  920   // testing if reserved zone needs to be enabled
  921   Label no_reserved_zone_enabling;
  922 
  923   cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
  924   jcc(Assembler::below, no_reserved_zone_enabling);
  925 
  926   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
  927   jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
  928   should_not_reach_here();
  929 
  930   bind(no_reserved_zone_enabling);
  931 }
  932 
  933 void MacroAssembler::c2bool(Register x) {
  934   // implements x == 0 ? 0 : 1
  935   // note: must only look at least-significant byte of x
  936   //       since C-style booleans are stored in one byte
  937   //       only! (was bug)
  938   andl(x, 0xFF);
  939   setb(Assembler::notZero, x);
  940 }
  941 
  942 // Wouldn't need if AddressLiteral version had new name
  943 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
  944   Assembler::call(L, rtype);
  945 }
  946 
  947 void MacroAssembler::call(Register entry) {
  948   Assembler::call(entry);
  949 }
  950 
  951 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
  952   assert(rscratch != noreg || always_reachable(entry), "missing");
  953 
  954   if (reachable(entry)) {
  955     Assembler::call_literal(entry.target(), entry.rspec());
  956   } else {
  957     lea(rscratch, entry);
  958     Assembler::call(rscratch);
  959   }
  960 }
  961 
  962 void MacroAssembler::ic_call(address entry, jint method_index) {
  963   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
  964   // Needs full 64-bit immediate for later patching.
  965   Assembler::mov64(rax, (int64_t)Universe::non_oop_word());
  966   call(AddressLiteral(entry, rh));
  967 }
  968 
  969 int MacroAssembler::ic_check_size() {
  970   return UseCompactObjectHeaders ? 17 : 14;
  971 }
  972 
  973 int MacroAssembler::ic_check(int end_alignment) {
  974   Register receiver = j_rarg0;
  975   Register data = rax;
  976   Register temp = rscratch1;
  977 
  978   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
  979   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
  980   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
  981   // before the inline cache check here, and not after
  982   align(end_alignment, offset() + ic_check_size());
  983 
  984   int uep_offset = offset();
  985 
  986   if (UseCompactObjectHeaders) {
  987     load_narrow_klass_compact(temp, receiver);
  988     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  989   } else {
  990     movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
  991     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  992   }
  993 
  994   // if inline cache check fails, then jump to runtime routine
  995   jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
  996   assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
  997 
  998   return uep_offset;
  999 }
 1000 
 1001 void MacroAssembler::emit_static_call_stub() {
 1002   // Static stub relocation also tags the Method* in the code-stream.
 1003   mov_metadata(rbx, (Metadata*) nullptr);  // Method is zapped till fixup time.
 1004   // This is recognized as unresolved by relocs/nativeinst/ic code.
 1005   jump(RuntimeAddress(pc()));
 1006 }
 1007 
 1008 // Implementation of call_VM versions
 1009 
 1010 void MacroAssembler::call_VM(Register oop_result,
 1011                              address entry_point,
 1012                              bool check_exceptions) {
 1013   Label C, E;
 1014   call(C, relocInfo::none);
 1015   jmp(E);
 1016 
 1017   bind(C);
 1018   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
 1019   ret(0);
 1020 
 1021   bind(E);
 1022 }
 1023 
 1024 void MacroAssembler::call_VM(Register oop_result,
 1025                              address entry_point,
 1026                              Register arg_1,
 1027                              bool check_exceptions) {
 1028   Label C, E;
 1029   call(C, relocInfo::none);
 1030   jmp(E);
 1031 
 1032   bind(C);
 1033   pass_arg1(this, arg_1);
 1034   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
 1035   ret(0);
 1036 
 1037   bind(E);
 1038 }
 1039 
 1040 void MacroAssembler::call_VM(Register oop_result,
 1041                              address entry_point,
 1042                              Register arg_1,
 1043                              Register arg_2,
 1044                              bool check_exceptions) {
 1045   Label C, E;
 1046   call(C, relocInfo::none);
 1047   jmp(E);
 1048 
 1049   bind(C);
 1050 
 1051   assert_different_registers(arg_1, c_rarg2);
 1052 
 1053   pass_arg2(this, arg_2);
 1054   pass_arg1(this, arg_1);
 1055   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
 1056   ret(0);
 1057 
 1058   bind(E);
 1059 }
 1060 
 1061 void MacroAssembler::call_VM(Register oop_result,
 1062                              address entry_point,
 1063                              Register arg_1,
 1064                              Register arg_2,
 1065                              Register arg_3,
 1066                              bool check_exceptions) {
 1067   Label C, E;
 1068   call(C, relocInfo::none);
 1069   jmp(E);
 1070 
 1071   bind(C);
 1072 
 1073   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1074   assert_different_registers(arg_2, c_rarg3);
 1075   pass_arg3(this, arg_3);
 1076   pass_arg2(this, arg_2);
 1077   pass_arg1(this, arg_1);
 1078   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
 1079   ret(0);
 1080 
 1081   bind(E);
 1082 }
 1083 
 1084 void MacroAssembler::call_VM(Register oop_result,
 1085                              Register last_java_sp,
 1086                              address entry_point,
 1087                              int number_of_arguments,
 1088                              bool check_exceptions) {
 1089   call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1090 }
 1091 
 1092 void MacroAssembler::call_VM(Register oop_result,
 1093                              Register last_java_sp,
 1094                              address entry_point,
 1095                              Register arg_1,
 1096                              bool check_exceptions) {
 1097   pass_arg1(this, arg_1);
 1098   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1099 }
 1100 
 1101 void MacroAssembler::call_VM(Register oop_result,
 1102                              Register last_java_sp,
 1103                              address entry_point,
 1104                              Register arg_1,
 1105                              Register arg_2,
 1106                              bool check_exceptions) {
 1107 
 1108   assert_different_registers(arg_1, c_rarg2);
 1109   pass_arg2(this, arg_2);
 1110   pass_arg1(this, arg_1);
 1111   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1112 }
 1113 
 1114 void MacroAssembler::call_VM(Register oop_result,
 1115                              Register last_java_sp,
 1116                              address entry_point,
 1117                              Register arg_1,
 1118                              Register arg_2,
 1119                              Register arg_3,
 1120                              bool check_exceptions) {
 1121   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1122   assert_different_registers(arg_2, c_rarg3);
 1123   pass_arg3(this, arg_3);
 1124   pass_arg2(this, arg_2);
 1125   pass_arg1(this, arg_1);
 1126   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1127 }
 1128 
 1129 void MacroAssembler::super_call_VM(Register oop_result,
 1130                                    Register last_java_sp,
 1131                                    address entry_point,
 1132                                    int number_of_arguments,
 1133                                    bool check_exceptions) {
 1134   MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1135 }
 1136 
 1137 void MacroAssembler::super_call_VM(Register oop_result,
 1138                                    Register last_java_sp,
 1139                                    address entry_point,
 1140                                    Register arg_1,
 1141                                    bool check_exceptions) {
 1142   pass_arg1(this, arg_1);
 1143   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1144 }
 1145 
 1146 void MacroAssembler::super_call_VM(Register oop_result,
 1147                                    Register last_java_sp,
 1148                                    address entry_point,
 1149                                    Register arg_1,
 1150                                    Register arg_2,
 1151                                    bool check_exceptions) {
 1152 
 1153   assert_different_registers(arg_1, c_rarg2);
 1154   pass_arg2(this, arg_2);
 1155   pass_arg1(this, arg_1);
 1156   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1157 }
 1158 
 1159 void MacroAssembler::super_call_VM(Register oop_result,
 1160                                    Register last_java_sp,
 1161                                    address entry_point,
 1162                                    Register arg_1,
 1163                                    Register arg_2,
 1164                                    Register arg_3,
 1165                                    bool check_exceptions) {
 1166   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1167   assert_different_registers(arg_2, c_rarg3);
 1168   pass_arg3(this, arg_3);
 1169   pass_arg2(this, arg_2);
 1170   pass_arg1(this, arg_1);
 1171   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1172 }
 1173 
 1174 void MacroAssembler::call_VM_base(Register oop_result,
 1175                                   Register last_java_sp,
 1176                                   address  entry_point,
 1177                                   int      number_of_arguments,
 1178                                   bool     check_exceptions) {
 1179   Register java_thread = r15_thread;
 1180 
 1181   // determine last_java_sp register
 1182   if (!last_java_sp->is_valid()) {
 1183     last_java_sp = rsp;
 1184   }
 1185   // debugging support
 1186   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 1187 #ifdef ASSERT
 1188   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 1189   // r12 is the heapbase.
 1190   if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 1191 #endif // ASSERT
 1192 
 1193   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 1194   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 1195 
 1196   // push java thread (becomes first argument of C function)
 1197 
 1198   mov(c_rarg0, r15_thread);
 1199 
 1200   // set last Java frame before call
 1201   assert(last_java_sp != rbp, "can't use ebp/rbp");
 1202 
 1203   // Only interpreter should have to set fp
 1204   set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
 1205 
 1206   // do the call, remove parameters
 1207   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 1208 
 1209 #ifdef ASSERT
 1210   // Check that thread register is not clobbered.
 1211   guarantee(java_thread != rax, "change this code");
 1212   push(rax);
 1213   { Label L;
 1214     get_thread_slow(rax);
 1215     cmpptr(java_thread, rax);
 1216     jcc(Assembler::equal, L);
 1217     STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
 1218     bind(L);
 1219   }
 1220   pop(rax);
 1221 #endif
 1222 
 1223   // reset last Java frame
 1224   // Only interpreter should have to clear fp
 1225   reset_last_Java_frame(true);
 1226 
 1227    // C++ interp handles this in the interpreter
 1228   check_and_handle_popframe();
 1229   check_and_handle_earlyret();
 1230 
 1231   if (check_exceptions) {
 1232     // check for pending exceptions (java_thread is set upon return)
 1233     cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 1234     // This used to conditionally jump to forward_exception however it is
 1235     // possible if we relocate that the branch will not reach. So we must jump
 1236     // around so we can always reach
 1237 
 1238     Label ok;
 1239     jcc(Assembler::equal, ok);
 1240     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 1241     bind(ok);
 1242   }
 1243 
 1244   // get oop result if there is one and reset the value in the thread
 1245   if (oop_result->is_valid()) {
 1246     get_vm_result_oop(oop_result);
 1247   }
 1248 }
 1249 
 1250 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 1251   // Calculate the value for last_Java_sp somewhat subtle.
 1252   // call_VM does an intermediate call which places a return address on
 1253   // the stack just under the stack pointer as the user finished with it.
 1254   // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
 1255 
 1256   // We've pushed one address, correct last_Java_sp
 1257   lea(rax, Address(rsp, wordSize));
 1258 
 1259   call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
 1260 }
 1261 
 1262 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
 1263 void MacroAssembler::call_VM_leaf0(address entry_point) {
 1264   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 1265 }
 1266 
 1267 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
 1268   call_VM_leaf_base(entry_point, number_of_arguments);
 1269 }
 1270 
 1271 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
 1272   pass_arg0(this, arg_0);
 1273   call_VM_leaf(entry_point, 1);
 1274 }
 1275 
 1276 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1277 
 1278   assert_different_registers(arg_0, c_rarg1);
 1279   pass_arg1(this, arg_1);
 1280   pass_arg0(this, arg_0);
 1281   call_VM_leaf(entry_point, 2);
 1282 }
 1283 
 1284 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1285   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1286   assert_different_registers(arg_1, c_rarg2);
 1287   pass_arg2(this, arg_2);
 1288   pass_arg1(this, arg_1);
 1289   pass_arg0(this, arg_0);
 1290   call_VM_leaf(entry_point, 3);
 1291 }
 1292 
 1293 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1294   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1295   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1296   assert_different_registers(arg_2, c_rarg3);
 1297   pass_arg3(this, arg_3);
 1298   pass_arg2(this, arg_2);
 1299   pass_arg1(this, arg_1);
 1300   pass_arg0(this, arg_0);
 1301   call_VM_leaf(entry_point, 3);
 1302 }
 1303 




 1304 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1305   pass_arg0(this, arg_0);
 1306   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1307 }
 1308 
 1309 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1310   assert_different_registers(arg_0, c_rarg1);
 1311   pass_arg1(this, arg_1);
 1312   pass_arg0(this, arg_0);
 1313   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1314 }
 1315 
 1316 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1317   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1318   assert_different_registers(arg_1, c_rarg2);
 1319   pass_arg2(this, arg_2);
 1320   pass_arg1(this, arg_1);
 1321   pass_arg0(this, arg_0);
 1322   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1323 }
 1324 
 1325 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1326   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1327   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1328   assert_different_registers(arg_2, c_rarg3);
 1329   pass_arg3(this, arg_3);
 1330   pass_arg2(this, arg_2);
 1331   pass_arg1(this, arg_1);
 1332   pass_arg0(this, arg_0);
 1333   MacroAssembler::call_VM_leaf_base(entry_point, 4);
 1334 }
 1335 
 1336 void MacroAssembler::get_vm_result_oop(Register oop_result) {
 1337   movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
 1338   movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
 1339   verify_oop_msg(oop_result, "broken oop in call_VM_base");
 1340 }
 1341 
 1342 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
 1343   movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
 1344   movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
 1345 }
 1346 
 1347 void MacroAssembler::check_and_handle_earlyret() {
 1348 }
 1349 
 1350 void MacroAssembler::check_and_handle_popframe() {
 1351 }
 1352 
 1353 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
 1354   assert(rscratch != noreg || always_reachable(src1), "missing");
 1355 
 1356   if (reachable(src1)) {
 1357     cmpl(as_Address(src1), imm);
 1358   } else {
 1359     lea(rscratch, src1);
 1360     cmpl(Address(rscratch, 0), imm);
 1361   }
 1362 }
 1363 
 1364 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
 1365   assert(!src2.is_lval(), "use cmpptr");
 1366   assert(rscratch != noreg || always_reachable(src2), "missing");
 1367 
 1368   if (reachable(src2)) {
 1369     cmpl(src1, as_Address(src2));
 1370   } else {
 1371     lea(rscratch, src2);
 1372     cmpl(src1, Address(rscratch, 0));
 1373   }
 1374 }
 1375 
 1376 void MacroAssembler::cmp32(Register src1, int32_t imm) {
 1377   Assembler::cmpl(src1, imm);
 1378 }
 1379 
 1380 void MacroAssembler::cmp32(Register src1, Address src2) {
 1381   Assembler::cmpl(src1, src2);
 1382 }
 1383 
 1384 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1385   ucomisd(opr1, opr2);
 1386 
 1387   Label L;
 1388   if (unordered_is_less) {
 1389     movl(dst, -1);
 1390     jcc(Assembler::parity, L);
 1391     jcc(Assembler::below , L);
 1392     movl(dst, 0);
 1393     jcc(Assembler::equal , L);
 1394     increment(dst);
 1395   } else { // unordered is greater
 1396     movl(dst, 1);
 1397     jcc(Assembler::parity, L);
 1398     jcc(Assembler::above , L);
 1399     movl(dst, 0);
 1400     jcc(Assembler::equal , L);
 1401     decrementl(dst);
 1402   }
 1403   bind(L);
 1404 }
 1405 
 1406 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1407   ucomiss(opr1, opr2);
 1408 
 1409   Label L;
 1410   if (unordered_is_less) {
 1411     movl(dst, -1);
 1412     jcc(Assembler::parity, L);
 1413     jcc(Assembler::below , L);
 1414     movl(dst, 0);
 1415     jcc(Assembler::equal , L);
 1416     increment(dst);
 1417   } else { // unordered is greater
 1418     movl(dst, 1);
 1419     jcc(Assembler::parity, L);
 1420     jcc(Assembler::above , L);
 1421     movl(dst, 0);
 1422     jcc(Assembler::equal , L);
 1423     decrementl(dst);
 1424   }
 1425   bind(L);
 1426 }
 1427 
 1428 
 1429 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
 1430   assert(rscratch != noreg || always_reachable(src1), "missing");
 1431 
 1432   if (reachable(src1)) {
 1433     cmpb(as_Address(src1), imm);
 1434   } else {
 1435     lea(rscratch, src1);
 1436     cmpb(Address(rscratch, 0), imm);
 1437   }
 1438 }
 1439 
 1440 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
 1441   assert(rscratch != noreg || always_reachable(src2), "missing");
 1442 
 1443   if (src2.is_lval()) {
 1444     movptr(rscratch, src2);
 1445     Assembler::cmpq(src1, rscratch);
 1446   } else if (reachable(src2)) {
 1447     cmpq(src1, as_Address(src2));
 1448   } else {
 1449     lea(rscratch, src2);
 1450     Assembler::cmpq(src1, Address(rscratch, 0));
 1451   }
 1452 }
 1453 
 1454 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
 1455   assert(src2.is_lval(), "not a mem-mem compare");
 1456   // moves src2's literal address
 1457   movptr(rscratch, src2);
 1458   Assembler::cmpq(src1, rscratch);
 1459 }
 1460 
 1461 void MacroAssembler::cmpoop(Register src1, Register src2) {
 1462   cmpptr(src1, src2);
 1463 }
 1464 
 1465 void MacroAssembler::cmpoop(Register src1, Address src2) {
 1466   cmpptr(src1, src2);
 1467 }
 1468 
 1469 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
 1470   movoop(rscratch, src2);
 1471   cmpptr(src1, rscratch);
 1472 }
 1473 
 1474 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
 1475   assert(rscratch != noreg || always_reachable(adr), "missing");
 1476 
 1477   if (reachable(adr)) {
 1478     lock();
 1479     cmpxchgptr(reg, as_Address(adr));
 1480   } else {
 1481     lea(rscratch, adr);
 1482     lock();
 1483     cmpxchgptr(reg, Address(rscratch, 0));
 1484   }
 1485 }
 1486 
 1487 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
 1488   cmpxchgq(reg, adr);
 1489 }
 1490 
 1491 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1492   assert(rscratch != noreg || always_reachable(src), "missing");
 1493 
 1494   if (reachable(src)) {
 1495     Assembler::comisd(dst, as_Address(src));
 1496   } else {
 1497     lea(rscratch, src);
 1498     Assembler::comisd(dst, Address(rscratch, 0));
 1499   }
 1500 }
 1501 
 1502 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1503   assert(rscratch != noreg || always_reachable(src), "missing");
 1504 
 1505   if (reachable(src)) {
 1506     Assembler::comiss(dst, as_Address(src));
 1507   } else {
 1508     lea(rscratch, src);
 1509     Assembler::comiss(dst, Address(rscratch, 0));
 1510   }
 1511 }
 1512 
 1513 
 1514 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
 1515   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1516 
 1517   Condition negated_cond = negate_condition(cond);
 1518   Label L;
 1519   jcc(negated_cond, L);
 1520   pushf(); // Preserve flags
 1521   atomic_incl(counter_addr, rscratch);
 1522   popf();
 1523   bind(L);
 1524 }
 1525 
 1526 int MacroAssembler::corrected_idivl(Register reg) {
 1527   // Full implementation of Java idiv and irem; checks for
 1528   // special case as described in JVM spec., p.243 & p.271.
 1529   // The function returns the (pc) offset of the idivl
 1530   // instruction - may be needed for implicit exceptions.
 1531   //
 1532   //         normal case                           special case
 1533   //
 1534   // input : rax,: dividend                         min_int
 1535   //         reg: divisor   (may not be rax,/rdx)   -1
 1536   //
 1537   // output: rax,: quotient  (= rax, idiv reg)       min_int
 1538   //         rdx: remainder (= rax, irem reg)       0
 1539   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
 1540   const int min_int = 0x80000000;
 1541   Label normal_case, special_case;
 1542 
 1543   // check for special case
 1544   cmpl(rax, min_int);
 1545   jcc(Assembler::notEqual, normal_case);
 1546   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
 1547   cmpl(reg, -1);
 1548   jcc(Assembler::equal, special_case);
 1549 
 1550   // handle normal case
 1551   bind(normal_case);
 1552   cdql();
 1553   int idivl_offset = offset();
 1554   idivl(reg);
 1555 
 1556   // normal and special case exit
 1557   bind(special_case);
 1558 
 1559   return idivl_offset;
 1560 }
 1561 
 1562 
 1563 
 1564 void MacroAssembler::decrementl(Register reg, int value) {
 1565   if (value == min_jint) {subl(reg, value) ; return; }
 1566   if (value <  0) { incrementl(reg, -value); return; }
 1567   if (value == 0) {                        ; return; }
 1568   if (value == 1 && UseIncDec) { decl(reg) ; return; }
 1569   /* else */      { subl(reg, value)       ; return; }
 1570 }
 1571 
 1572 void MacroAssembler::decrementl(Address dst, int value) {
 1573   if (value == min_jint) {subl(dst, value) ; return; }
 1574   if (value <  0) { incrementl(dst, -value); return; }
 1575   if (value == 0) {                        ; return; }
 1576   if (value == 1 && UseIncDec) { decl(dst) ; return; }
 1577   /* else */      { subl(dst, value)       ; return; }
 1578 }
 1579 
 1580 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
 1581   assert(shift_value > 0, "illegal shift value");
 1582   Label _is_positive;
 1583   testl (reg, reg);
 1584   jcc (Assembler::positive, _is_positive);
 1585   int offset = (1 << shift_value) - 1 ;
 1586 
 1587   if (offset == 1) {
 1588     incrementl(reg);
 1589   } else {
 1590     addl(reg, offset);
 1591   }
 1592 
 1593   bind (_is_positive);
 1594   sarl(reg, shift_value);
 1595 }
 1596 
 1597 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1598   assert(rscratch != noreg || always_reachable(src), "missing");
 1599 
 1600   if (reachable(src)) {
 1601     Assembler::divsd(dst, as_Address(src));
 1602   } else {
 1603     lea(rscratch, src);
 1604     Assembler::divsd(dst, Address(rscratch, 0));
 1605   }
 1606 }
 1607 
 1608 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1609   assert(rscratch != noreg || always_reachable(src), "missing");
 1610 
 1611   if (reachable(src)) {
 1612     Assembler::divss(dst, as_Address(src));
 1613   } else {
 1614     lea(rscratch, src);
 1615     Assembler::divss(dst, Address(rscratch, 0));
 1616   }
 1617 }
 1618 
 1619 void MacroAssembler::enter() {
 1620   push(rbp);
 1621   mov(rbp, rsp);
 1622 }
 1623 
 1624 void MacroAssembler::post_call_nop() {
 1625   if (!Continuations::enabled()) {
 1626     return;
 1627   }
 1628   InstructionMark im(this);
 1629   relocate(post_call_nop_Relocation::spec());
 1630   InlineSkippedInstructionsCounter skipCounter(this);
 1631   emit_int8((uint8_t)0x0f);
 1632   emit_int8((uint8_t)0x1f);
 1633   emit_int8((uint8_t)0x84);
 1634   emit_int8((uint8_t)0x00);
 1635   emit_int32(0x00);
 1636 }
 1637 
 1638 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1639   assert(rscratch != noreg || always_reachable(src), "missing");
 1640   if (reachable(src)) {
 1641     Assembler::mulpd(dst, as_Address(src));
 1642   } else {
 1643     lea(rscratch, src);
 1644     Assembler::mulpd(dst, Address(rscratch, 0));
 1645   }
 1646 }
 1647 
 1648 // dst = c = a * b + c
 1649 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1650   Assembler::vfmadd231sd(c, a, b);
 1651   if (dst != c) {
 1652     movdbl(dst, c);
 1653   }
 1654 }
 1655 
 1656 // dst = c = a * b + c
 1657 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1658   Assembler::vfmadd231ss(c, a, b);
 1659   if (dst != c) {
 1660     movflt(dst, c);
 1661   }
 1662 }
 1663 
 1664 // dst = c = a * b + c
 1665 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1666   Assembler::vfmadd231pd(c, a, b, vector_len);
 1667   if (dst != c) {
 1668     vmovdqu(dst, c);
 1669   }
 1670 }
 1671 
 1672 // dst = c = a * b + c
 1673 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1674   Assembler::vfmadd231ps(c, a, b, vector_len);
 1675   if (dst != c) {
 1676     vmovdqu(dst, c);
 1677   }
 1678 }
 1679 
 1680 // dst = c = a * b + c
 1681 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1682   Assembler::vfmadd231pd(c, a, b, vector_len);
 1683   if (dst != c) {
 1684     vmovdqu(dst, c);
 1685   }
 1686 }
 1687 
 1688 // dst = c = a * b + c
 1689 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1690   Assembler::vfmadd231ps(c, a, b, vector_len);
 1691   if (dst != c) {
 1692     vmovdqu(dst, c);
 1693   }
 1694 }
 1695 
 1696 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
 1697   assert(rscratch != noreg || always_reachable(dst), "missing");
 1698 
 1699   if (reachable(dst)) {
 1700     incrementl(as_Address(dst));
 1701   } else {
 1702     lea(rscratch, dst);
 1703     incrementl(Address(rscratch, 0));
 1704   }
 1705 }
 1706 
 1707 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
 1708   incrementl(as_Address(dst, rscratch));
 1709 }
 1710 
 1711 void MacroAssembler::incrementl(Register reg, int value) {
 1712   if (value == min_jint) {addl(reg, value) ; return; }
 1713   if (value <  0) { decrementl(reg, -value); return; }
 1714   if (value == 0) {                        ; return; }
 1715   if (value == 1 && UseIncDec) { incl(reg) ; return; }
 1716   /* else */      { addl(reg, value)       ; return; }
 1717 }
 1718 
 1719 void MacroAssembler::incrementl(Address dst, int value) {
 1720   if (value == min_jint) {addl(dst, value) ; return; }
 1721   if (value <  0) { decrementl(dst, -value); return; }
 1722   if (value == 0) {                        ; return; }
 1723   if (value == 1 && UseIncDec) { incl(dst) ; return; }
 1724   /* else */      { addl(dst, value)       ; return; }
 1725 }
 1726 
 1727 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
 1728   assert(rscratch != noreg || always_reachable(dst), "missing");
 1729   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
 1730   if (reachable(dst)) {
 1731     jmp_literal(dst.target(), dst.rspec());
 1732   } else {
 1733     lea(rscratch, dst);
 1734     jmp(rscratch);
 1735   }
 1736 }
 1737 
 1738 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
 1739   assert(rscratch != noreg || always_reachable(dst), "missing");
 1740   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump_cc");
 1741   if (reachable(dst)) {
 1742     InstructionMark im(this);
 1743     relocate(dst.reloc());
 1744     const int short_size = 2;
 1745     const int long_size = 6;
 1746     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
 1747     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
 1748       // 0111 tttn #8-bit disp
 1749       emit_int8(0x70 | cc);
 1750       emit_int8((offs - short_size) & 0xFF);
 1751     } else {
 1752       // 0000 1111 1000 tttn #32-bit disp
 1753       emit_int8(0x0F);
 1754       emit_int8((unsigned char)(0x80 | cc));
 1755       emit_int32(offs - long_size);
 1756     }
 1757   } else {
 1758 #ifdef ASSERT
 1759     warning("reversing conditional branch");
 1760 #endif /* ASSERT */
 1761     Label skip;
 1762     jccb(reverse[cc], skip);
 1763     lea(rscratch, dst);
 1764     Assembler::jmp(rscratch);
 1765     bind(skip);
 1766   }
 1767 }
 1768 
 1769 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
 1770   ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 1771   assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
 1772 
 1773   stmxcsr(mxcsr_save);
 1774   movl(tmp, mxcsr_save);
 1775   if (EnableX86ECoreOpts) {
 1776     // The mxcsr_std has status bits set for performance on ECore
 1777     orl(tmp, 0x003f);
 1778   } else {
 1779     // Mask out status bits (only check control and mask bits)
 1780     andl(tmp, 0xFFC0);
 1781   }
 1782   cmp32(tmp, mxcsr_std, rscratch);
 1783 }
 1784 
 1785 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
 1786   assert(rscratch != noreg || always_reachable(src), "missing");
 1787 
 1788   if (reachable(src)) {
 1789     Assembler::ldmxcsr(as_Address(src));
 1790   } else {
 1791     lea(rscratch, src);
 1792     Assembler::ldmxcsr(Address(rscratch, 0));
 1793   }
 1794 }
 1795 
 1796 int MacroAssembler::load_signed_byte(Register dst, Address src) {
 1797   int off = offset();
 1798   movsbl(dst, src); // movsxb
 1799   return off;
 1800 }
 1801 
 1802 // Note: load_signed_short used to be called load_signed_word.
 1803 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
 1804 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
 1805 // The term "word" in HotSpot means a 32- or 64-bit machine word.
 1806 int MacroAssembler::load_signed_short(Register dst, Address src) {
 1807   // This is dubious to me since it seems safe to do a signed 16 => 64 bit
 1808   // version but this is what 64bit has always done. This seems to imply
 1809   // that users are only using 32bits worth.
 1810   int off = offset();
 1811   movswl(dst, src); // movsxw
 1812   return off;
 1813 }
 1814 
 1815 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
 1816   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1817   // and "3.9 Partial Register Penalties", p. 22).
 1818   int off = offset();
 1819   movzbl(dst, src); // movzxb
 1820   return off;
 1821 }
 1822 
 1823 // Note: load_unsigned_short used to be called load_unsigned_word.
 1824 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
 1825   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1826   // and "3.9 Partial Register Penalties", p. 22).
 1827   int off = offset();
 1828   movzwl(dst, src); // movzxw
 1829   return off;
 1830 }
 1831 
 1832 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
 1833   switch (size_in_bytes) {
 1834   case  8:  movq(dst, src); break;
 1835   case  4:  movl(dst, src); break;
 1836   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
 1837   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
 1838   default:  ShouldNotReachHere();
 1839   }
 1840 }
 1841 
 1842 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
 1843   switch (size_in_bytes) {
 1844   case  8:  movq(dst, src); break;
 1845   case  4:  movl(dst, src); break;
 1846   case  2:  movw(dst, src); break;
 1847   case  1:  movb(dst, src); break;
 1848   default:  ShouldNotReachHere();
 1849   }
 1850 }
 1851 
 1852 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
 1853   assert(rscratch != noreg || always_reachable(dst), "missing");
 1854 
 1855   if (reachable(dst)) {
 1856     movl(as_Address(dst), src);
 1857   } else {
 1858     lea(rscratch, dst);
 1859     movl(Address(rscratch, 0), src);
 1860   }
 1861 }
 1862 
 1863 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
 1864   if (reachable(src)) {
 1865     movl(dst, as_Address(src));
 1866   } else {
 1867     lea(dst, src);
 1868     movl(dst, Address(dst, 0));
 1869   }
 1870 }
 1871 
 1872 // C++ bool manipulation
 1873 
 1874 void MacroAssembler::movbool(Register dst, Address src) {
 1875   if(sizeof(bool) == 1)
 1876     movb(dst, src);
 1877   else if(sizeof(bool) == 2)
 1878     movw(dst, src);
 1879   else if(sizeof(bool) == 4)
 1880     movl(dst, src);
 1881   else
 1882     // unsupported
 1883     ShouldNotReachHere();
 1884 }
 1885 
 1886 void MacroAssembler::movbool(Address dst, bool boolconst) {
 1887   if(sizeof(bool) == 1)
 1888     movb(dst, (int) boolconst);
 1889   else if(sizeof(bool) == 2)
 1890     movw(dst, (int) boolconst);
 1891   else if(sizeof(bool) == 4)
 1892     movl(dst, (int) boolconst);
 1893   else
 1894     // unsupported
 1895     ShouldNotReachHere();
 1896 }
 1897 
 1898 void MacroAssembler::movbool(Address dst, Register src) {
 1899   if(sizeof(bool) == 1)
 1900     movb(dst, src);
 1901   else if(sizeof(bool) == 2)
 1902     movw(dst, src);
 1903   else if(sizeof(bool) == 4)
 1904     movl(dst, src);
 1905   else
 1906     // unsupported
 1907     ShouldNotReachHere();
 1908 }
 1909 
 1910 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1911   assert(rscratch != noreg || always_reachable(src), "missing");
 1912 
 1913   if (reachable(src)) {
 1914     movdl(dst, as_Address(src));
 1915   } else {
 1916     lea(rscratch, src);
 1917     movdl(dst, Address(rscratch, 0));
 1918   }
 1919 }
 1920 
 1921 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1922   assert(rscratch != noreg || always_reachable(src), "missing");
 1923 
 1924   if (reachable(src)) {
 1925     movq(dst, as_Address(src));
 1926   } else {
 1927     lea(rscratch, src);
 1928     movq(dst, Address(rscratch, 0));
 1929   }
 1930 }
 1931 
 1932 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1933   assert(rscratch != noreg || always_reachable(src), "missing");
 1934 
 1935   if (reachable(src)) {
 1936     if (UseXmmLoadAndClearUpper) {
 1937       movsd (dst, as_Address(src));
 1938     } else {
 1939       movlpd(dst, as_Address(src));
 1940     }
 1941   } else {
 1942     lea(rscratch, src);
 1943     if (UseXmmLoadAndClearUpper) {
 1944       movsd (dst, Address(rscratch, 0));
 1945     } else {
 1946       movlpd(dst, Address(rscratch, 0));
 1947     }
 1948   }
 1949 }
 1950 
 1951 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1952   assert(rscratch != noreg || always_reachable(src), "missing");
 1953 
 1954   if (reachable(src)) {
 1955     movss(dst, as_Address(src));
 1956   } else {
 1957     lea(rscratch, src);
 1958     movss(dst, Address(rscratch, 0));
 1959   }
 1960 }
 1961 
 1962 void MacroAssembler::movhlf(XMMRegister dst, XMMRegister src, Register rscratch) {
 1963   if (VM_Version::supports_avx10_2()) {
 1964     evmovw(dst, src);
 1965   } else {
 1966     assert(rscratch != noreg, "missing");
 1967     evmovw(rscratch, src);
 1968     evmovw(dst, rscratch);
 1969   }
 1970 }
 1971 
 1972 void MacroAssembler::mov64(Register dst, int64_t imm64) {
 1973   if (is_uimm32(imm64)) {
 1974     movl(dst, checked_cast<uint32_t>(imm64));
 1975   } else if (is_simm32(imm64)) {
 1976     movq(dst, checked_cast<int32_t>(imm64));
 1977   } else {
 1978     Assembler::mov64(dst, imm64);
 1979   }
 1980 }
 1981 
 1982 void MacroAssembler::mov64(Register dst, int64_t imm64, relocInfo::relocType rtype, int format) {
 1983   Assembler::mov64(dst, imm64, rtype, format);
 1984 }
 1985 
 1986 void MacroAssembler::movptr(Register dst, Register src) {
 1987   movq(dst, src);
 1988 }
 1989 
 1990 void MacroAssembler::movptr(Register dst, Address src) {
 1991   movq(dst, src);
 1992 }
 1993 
 1994 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 1995 void MacroAssembler::movptr(Register dst, intptr_t src) {
 1996   mov64(dst, src);
 1997 }
 1998 
 1999 void MacroAssembler::movptr(Address dst, Register src) {
 2000   movq(dst, src);
 2001 }
 2002 
 2003 void MacroAssembler::movptr(Address dst, int32_t src) {
 2004   movslq(dst, src);
 2005 }
 2006 
 2007 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 2008   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2009   Assembler::movdqu(dst, src);
 2010 }
 2011 
 2012 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 2013   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2014   Assembler::movdqu(dst, src);
 2015 }
 2016 
 2017 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2018   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2019   Assembler::movdqu(dst, src);
 2020 }
 2021 
 2022 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2023   assert(rscratch != noreg || always_reachable(src), "missing");
 2024 
 2025   if (reachable(src)) {
 2026     movdqu(dst, as_Address(src));
 2027   } else {
 2028     lea(rscratch, src);
 2029     movdqu(dst, Address(rscratch, 0));
 2030   }
 2031 }
 2032 
 2033 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2034   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2035   Assembler::vmovdqu(dst, src);
 2036 }
 2037 
 2038 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2039   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2040   Assembler::vmovdqu(dst, src);
 2041 }
 2042 
 2043 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2044   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2045   Assembler::vmovdqu(dst, src);
 2046 }
 2047 
 2048 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2049   assert(rscratch != noreg || always_reachable(src), "missing");
 2050 
 2051   if (reachable(src)) {
 2052     vmovdqu(dst, as_Address(src));
 2053   }
 2054   else {
 2055     lea(rscratch, src);
 2056     vmovdqu(dst, Address(rscratch, 0));
 2057   }
 2058 }
 2059 
 2060 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2061   assert(rscratch != noreg || always_reachable(src), "missing");
 2062 
 2063   if (vector_len == AVX_512bit) {
 2064     evmovdquq(dst, src, AVX_512bit, rscratch);
 2065   } else if (vector_len == AVX_256bit) {
 2066     vmovdqu(dst, src, rscratch);
 2067   } else {
 2068     movdqu(dst, src, rscratch);
 2069   }
 2070 }
 2071 
 2072 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
 2073   if (vector_len == AVX_512bit) {
 2074     evmovdquq(dst, src, AVX_512bit);
 2075   } else if (vector_len == AVX_256bit) {
 2076     vmovdqu(dst, src);
 2077   } else {
 2078     movdqu(dst, src);
 2079   }
 2080 }
 2081 
 2082 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
 2083   if (vector_len == AVX_512bit) {
 2084     evmovdquq(dst, src, AVX_512bit);
 2085   } else if (vector_len == AVX_256bit) {
 2086     vmovdqu(dst, src);
 2087   } else {
 2088     movdqu(dst, src);
 2089   }
 2090 }
 2091 
 2092 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
 2093   if (vector_len == AVX_512bit) {
 2094     evmovdquq(dst, src, AVX_512bit);
 2095   } else if (vector_len == AVX_256bit) {
 2096     vmovdqu(dst, src);
 2097   } else {
 2098     movdqu(dst, src);
 2099   }
 2100 }
 2101 
 2102 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2103   assert(rscratch != noreg || always_reachable(src), "missing");
 2104 
 2105   if (reachable(src)) {
 2106     vmovdqa(dst, as_Address(src));
 2107   }
 2108   else {
 2109     lea(rscratch, src);
 2110     vmovdqa(dst, Address(rscratch, 0));
 2111   }
 2112 }
 2113 
 2114 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2115   assert(rscratch != noreg || always_reachable(src), "missing");
 2116 
 2117   if (vector_len == AVX_512bit) {
 2118     evmovdqaq(dst, src, AVX_512bit, rscratch);
 2119   } else if (vector_len == AVX_256bit) {
 2120     vmovdqa(dst, src, rscratch);
 2121   } else {
 2122     movdqa(dst, src, rscratch);
 2123   }
 2124 }
 2125 
 2126 void MacroAssembler::kmov(KRegister dst, Address src) {
 2127   if (VM_Version::supports_avx512bw()) {
 2128     kmovql(dst, src);
 2129   } else {
 2130     assert(VM_Version::supports_evex(), "");
 2131     kmovwl(dst, src);
 2132   }
 2133 }
 2134 
 2135 void MacroAssembler::kmov(Address dst, KRegister src) {
 2136   if (VM_Version::supports_avx512bw()) {
 2137     kmovql(dst, src);
 2138   } else {
 2139     assert(VM_Version::supports_evex(), "");
 2140     kmovwl(dst, src);
 2141   }
 2142 }
 2143 
 2144 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2145   if (VM_Version::supports_avx512bw()) {
 2146     kmovql(dst, src);
 2147   } else {
 2148     assert(VM_Version::supports_evex(), "");
 2149     kmovwl(dst, src);
 2150   }
 2151 }
 2152 
 2153 void MacroAssembler::kmov(Register dst, KRegister src) {
 2154   if (VM_Version::supports_avx512bw()) {
 2155     kmovql(dst, src);
 2156   } else {
 2157     assert(VM_Version::supports_evex(), "");
 2158     kmovwl(dst, src);
 2159   }
 2160 }
 2161 
 2162 void MacroAssembler::kmov(KRegister dst, Register src) {
 2163   if (VM_Version::supports_avx512bw()) {
 2164     kmovql(dst, src);
 2165   } else {
 2166     assert(VM_Version::supports_evex(), "");
 2167     kmovwl(dst, src);
 2168   }
 2169 }
 2170 
 2171 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2172   assert(rscratch != noreg || always_reachable(src), "missing");
 2173 
 2174   if (reachable(src)) {
 2175     kmovql(dst, as_Address(src));
 2176   } else {
 2177     lea(rscratch, src);
 2178     kmovql(dst, Address(rscratch, 0));
 2179   }
 2180 }
 2181 
 2182 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2183   assert(rscratch != noreg || always_reachable(src), "missing");
 2184 
 2185   if (reachable(src)) {
 2186     kmovwl(dst, as_Address(src));
 2187   } else {
 2188     lea(rscratch, src);
 2189     kmovwl(dst, Address(rscratch, 0));
 2190   }
 2191 }
 2192 
 2193 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2194                                int vector_len, Register rscratch) {
 2195   assert(rscratch != noreg || always_reachable(src), "missing");
 2196 
 2197   if (reachable(src)) {
 2198     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2199   } else {
 2200     lea(rscratch, src);
 2201     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2202   }
 2203 }
 2204 
 2205 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2206                                int vector_len, Register rscratch) {
 2207   assert(rscratch != noreg || always_reachable(src), "missing");
 2208 
 2209   if (reachable(src)) {
 2210     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2211   } else {
 2212     lea(rscratch, src);
 2213     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2214   }
 2215 }
 2216 
 2217 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2218   assert(rscratch != noreg || always_reachable(src), "missing");
 2219 
 2220   if (reachable(src)) {
 2221     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2222   } else {
 2223     lea(rscratch, src);
 2224     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2225   }
 2226 }
 2227 
 2228 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2229   assert(rscratch != noreg || always_reachable(src), "missing");
 2230 
 2231   if (reachable(src)) {
 2232     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2233   } else {
 2234     lea(rscratch, src);
 2235     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2236   }
 2237 }
 2238 
 2239 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2240   assert(rscratch != noreg || always_reachable(src), "missing");
 2241 
 2242   if (reachable(src)) {
 2243     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2244   } else {
 2245     lea(rscratch, src);
 2246     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2247   }
 2248 }
 2249 
 2250 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2251   assert(rscratch != noreg || always_reachable(src), "missing");
 2252 
 2253   if (reachable(src)) {
 2254     Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
 2255   } else {
 2256     lea(rscratch, src);
 2257     Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2258   }
 2259 }
 2260 
 2261 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2262   assert(rscratch != noreg || always_reachable(src), "missing");
 2263 
 2264   if (reachable(src)) {
 2265     Assembler::evmovdqaq(dst, as_Address(src), vector_len);
 2266   } else {
 2267     lea(rscratch, src);
 2268     Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
 2269   }
 2270 }
 2271 
 2272 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2273   assert(rscratch != noreg || always_reachable(src), "missing");
 2274 
 2275   if (reachable(src)) {
 2276     Assembler::movapd(dst, as_Address(src));
 2277   } else {
 2278     lea(rscratch, src);
 2279     Assembler::movapd(dst, Address(rscratch, 0));
 2280   }
 2281 }
 2282 
 2283 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2284   assert(rscratch != noreg || always_reachable(src), "missing");
 2285 
 2286   if (reachable(src)) {
 2287     Assembler::movdqa(dst, as_Address(src));
 2288   } else {
 2289     lea(rscratch, src);
 2290     Assembler::movdqa(dst, Address(rscratch, 0));
 2291   }
 2292 }
 2293 
 2294 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2295   assert(rscratch != noreg || always_reachable(src), "missing");
 2296 
 2297   if (reachable(src)) {
 2298     Assembler::movsd(dst, as_Address(src));
 2299   } else {
 2300     lea(rscratch, src);
 2301     Assembler::movsd(dst, Address(rscratch, 0));
 2302   }
 2303 }
 2304 
 2305 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2306   assert(rscratch != noreg || always_reachable(src), "missing");
 2307 
 2308   if (reachable(src)) {
 2309     Assembler::movss(dst, as_Address(src));
 2310   } else {
 2311     lea(rscratch, src);
 2312     Assembler::movss(dst, Address(rscratch, 0));
 2313   }
 2314 }
 2315 
 2316 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2317   assert(rscratch != noreg || always_reachable(src), "missing");
 2318 
 2319   if (reachable(src)) {
 2320     Assembler::movddup(dst, as_Address(src));
 2321   } else {
 2322     lea(rscratch, src);
 2323     Assembler::movddup(dst, Address(rscratch, 0));
 2324   }
 2325 }
 2326 
 2327 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2328   assert(rscratch != noreg || always_reachable(src), "missing");
 2329 
 2330   if (reachable(src)) {
 2331     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2332   } else {
 2333     lea(rscratch, src);
 2334     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2335   }
 2336 }
 2337 
 2338 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2339   assert(rscratch != noreg || always_reachable(src), "missing");
 2340 
 2341   if (reachable(src)) {
 2342     Assembler::mulsd(dst, as_Address(src));
 2343   } else {
 2344     lea(rscratch, src);
 2345     Assembler::mulsd(dst, Address(rscratch, 0));
 2346   }
 2347 }
 2348 
 2349 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2350   assert(rscratch != noreg || always_reachable(src), "missing");
 2351 
 2352   if (reachable(src)) {
 2353     Assembler::mulss(dst, as_Address(src));
 2354   } else {
 2355     lea(rscratch, src);
 2356     Assembler::mulss(dst, Address(rscratch, 0));
 2357   }
 2358 }
 2359 
 2360 void MacroAssembler::null_check(Register reg, int offset) {
 2361   if (needs_explicit_null_check(offset)) {
 2362     // provoke OS null exception if reg is null by
 2363     // accessing M[reg] w/o changing any (non-CC) registers
 2364     // NOTE: cmpl is plenty here to provoke a segv
 2365     cmpptr(rax, Address(reg, 0));
 2366     // Note: should probably use testl(rax, Address(reg, 0));
 2367     //       may be shorter code (however, this version of
 2368     //       testl needs to be implemented first)
 2369   } else {
 2370     // nothing to do, (later) access of M[reg + offset]
 2371     // will provoke OS null exception if reg is null
 2372   }
 2373 }
 2374 











































































































 2375 void MacroAssembler::os_breakpoint() {
 2376   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2377   // (e.g., MSVC can't call ps() otherwise)
 2378   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2379 }
 2380 
 2381 void MacroAssembler::unimplemented(const char* what) {
 2382   const char* buf = nullptr;
 2383   {
 2384     ResourceMark rm;
 2385     stringStream ss;
 2386     ss.print("unimplemented: %s", what);
 2387     buf = code_string(ss.as_string());
 2388   }
 2389   stop(buf);
 2390 }
 2391 
 2392 #define XSTATE_BV 0x200
 2393 
 2394 void MacroAssembler::pop_CPU_state() {
 2395   pop_FPU_state();
 2396   pop_IU_state();
 2397 }
 2398 
 2399 void MacroAssembler::pop_FPU_state() {
 2400   fxrstor(Address(rsp, 0));
 2401   addptr(rsp, FPUStateSizeInWords * wordSize);
 2402 }
 2403 
 2404 void MacroAssembler::pop_IU_state() {
 2405   popa();
 2406   addq(rsp, 8);
 2407   popf();
 2408 }
 2409 
 2410 // Save Integer and Float state
 2411 // Warning: Stack must be 16 byte aligned (64bit)
 2412 void MacroAssembler::push_CPU_state() {
 2413   push_IU_state();
 2414   push_FPU_state();
 2415 }
 2416 
 2417 void MacroAssembler::push_FPU_state() {
 2418   subptr(rsp, FPUStateSizeInWords * wordSize);
 2419   fxsave(Address(rsp, 0));
 2420 }
 2421 
 2422 void MacroAssembler::push_IU_state() {
 2423   // Push flags first because pusha kills them
 2424   pushf();
 2425   // Make sure rsp stays 16-byte aligned
 2426   subq(rsp, 8);
 2427   pusha();
 2428 }
 2429 
 2430 void MacroAssembler::push_cont_fastpath() {
 2431   if (!Continuations::enabled()) return;
 2432 
 2433   Label L_done;
 2434   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2435   jccb(Assembler::belowEqual, L_done);
 2436   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2437   bind(L_done);
 2438 }
 2439 
 2440 void MacroAssembler::pop_cont_fastpath() {
 2441   if (!Continuations::enabled()) return;
 2442 
 2443   Label L_done;
 2444   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2445   jccb(Assembler::below, L_done);
 2446   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2447   bind(L_done);
 2448 }
 2449 
 2450 #ifdef ASSERT
 2451 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2452   Label no_cont;
 2453   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2454   testl(cont, cont);
 2455   jcc(Assembler::zero, no_cont);
 2456   stop(name);
 2457   bind(no_cont);
 2458 }
 2459 #endif
 2460 
 2461 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2462   // we must set sp to zero to clear frame
 2463   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2464   // must clear fp, so that compiled frames are not confused; it is
 2465   // possible that we need it only for debugging
 2466   if (clear_fp) {
 2467     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2468   }
 2469   // Always clear the pc because it could have been set by make_walkable()
 2470   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2471   vzeroupper();
 2472 }
 2473 
 2474 void MacroAssembler::round_to(Register reg, int modulus) {
 2475   addptr(reg, modulus - 1);
 2476   andptr(reg, -modulus);
 2477 }
 2478 
 2479 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2480   if (at_return) {
 2481     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2482     // we may safely use rsp instead to perform the stack watermark check.
 2483     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2484     jcc(Assembler::above, slow_path);
 2485     return;
 2486   }
 2487   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2488   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2489 }
 2490 
 2491 // Calls to C land
 2492 //
 2493 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2494 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2495 // has to be reset to 0. This is required to allow proper stack traversal.
 2496 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2497                                          Register last_java_fp,
 2498                                          address  last_java_pc,
 2499                                          Register rscratch) {
 2500   vzeroupper();
 2501   // determine last_java_sp register
 2502   if (!last_java_sp->is_valid()) {
 2503     last_java_sp = rsp;
 2504   }
 2505   // last_java_fp is optional
 2506   if (last_java_fp->is_valid()) {
 2507     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2508   }
 2509   // last_java_pc is optional
 2510   if (last_java_pc != nullptr) {
 2511     Address java_pc(r15_thread,
 2512                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2513     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2514   }
 2515   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2516 }
 2517 
 2518 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2519                                          Register last_java_fp,
 2520                                          Label &L,
 2521                                          Register scratch) {
 2522   lea(scratch, L);
 2523   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2524   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2525 }
 2526 
 2527 void MacroAssembler::shlptr(Register dst, int imm8) {
 2528   shlq(dst, imm8);
 2529 }
 2530 
 2531 void MacroAssembler::shrptr(Register dst, int imm8) {
 2532   shrq(dst, imm8);
 2533 }
 2534 
 2535 void MacroAssembler::sign_extend_byte(Register reg) {
 2536   movsbl(reg, reg); // movsxb
 2537 }
 2538 
 2539 void MacroAssembler::sign_extend_short(Register reg) {
 2540   movswl(reg, reg); // movsxw
 2541 }
 2542 
 2543 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2544   if (imm32 >= 0 && is8bit(imm32)) {
 2545     testb(dst, imm32);
 2546   } else {
 2547     Assembler::testl(dst, imm32);
 2548   }
 2549 }
 2550 
 2551 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2552   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2553     testb(dst, imm32);
 2554   } else {
 2555     Assembler::testl(dst, imm32);
 2556   }
 2557 }
 2558 
 2559 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2560   assert(always_reachable(src), "Address should be reachable");
 2561   testl(dst, as_Address(src));
 2562 }
 2563 
 2564 void MacroAssembler::testq(Address 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::testq(Register dst, int32_t imm32) {
 2573   if (imm32 >= 0) {
 2574     testl(dst, imm32);
 2575   } else {
 2576     Assembler::testq(dst, imm32);
 2577   }
 2578 }
 2579 
 2580 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2581   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2582   Assembler::pcmpeqb(dst, src);
 2583 }
 2584 
 2585 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2586   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2587   Assembler::pcmpeqw(dst, src);
 2588 }
 2589 
 2590 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2591   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2592   Assembler::pcmpestri(dst, src, imm8);
 2593 }
 2594 
 2595 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2596   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2597   Assembler::pcmpestri(dst, src, imm8);
 2598 }
 2599 
 2600 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2601   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2602   Assembler::pmovzxbw(dst, src);
 2603 }
 2604 
 2605 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2606   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2607   Assembler::pmovzxbw(dst, src);
 2608 }
 2609 
 2610 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2611   assert((src->encoding() < 16),"XMM register should be 0-15");
 2612   Assembler::pmovmskb(dst, src);
 2613 }
 2614 
 2615 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2616   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2617   Assembler::ptest(dst, src);
 2618 }
 2619 
 2620 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2621   assert(rscratch != noreg || always_reachable(src), "missing");
 2622 
 2623   if (reachable(src)) {
 2624     Assembler::sqrtss(dst, as_Address(src));
 2625   } else {
 2626     lea(rscratch, src);
 2627     Assembler::sqrtss(dst, Address(rscratch, 0));
 2628   }
 2629 }
 2630 
 2631 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2632   assert(rscratch != noreg || always_reachable(src), "missing");
 2633 
 2634   if (reachable(src)) {
 2635     Assembler::subsd(dst, as_Address(src));
 2636   } else {
 2637     lea(rscratch, src);
 2638     Assembler::subsd(dst, Address(rscratch, 0));
 2639   }
 2640 }
 2641 
 2642 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2643   assert(rscratch != noreg || always_reachable(src), "missing");
 2644 
 2645   if (reachable(src)) {
 2646     Assembler::roundsd(dst, as_Address(src), rmode);
 2647   } else {
 2648     lea(rscratch, src);
 2649     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2650   }
 2651 }
 2652 
 2653 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2654   assert(rscratch != noreg || always_reachable(src), "missing");
 2655 
 2656   if (reachable(src)) {
 2657     Assembler::subss(dst, as_Address(src));
 2658   } else {
 2659     lea(rscratch, src);
 2660     Assembler::subss(dst, Address(rscratch, 0));
 2661   }
 2662 }
 2663 
 2664 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2665   assert(rscratch != noreg || always_reachable(src), "missing");
 2666 
 2667   if (reachable(src)) {
 2668     Assembler::ucomisd(dst, as_Address(src));
 2669   } else {
 2670     lea(rscratch, src);
 2671     Assembler::ucomisd(dst, Address(rscratch, 0));
 2672   }
 2673 }
 2674 
 2675 void MacroAssembler::evucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2676   assert(rscratch != noreg || always_reachable(src), "missing");
 2677 
 2678   if (reachable(src)) {
 2679     Assembler::evucomxsd(dst, as_Address(src));
 2680   } else {
 2681     lea(rscratch, src);
 2682     Assembler::evucomxsd(dst, Address(rscratch, 0));
 2683   }
 2684 }
 2685 
 2686 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2687   assert(rscratch != noreg || always_reachable(src), "missing");
 2688 
 2689   if (reachable(src)) {
 2690     Assembler::ucomiss(dst, as_Address(src));
 2691   } else {
 2692     lea(rscratch, src);
 2693     Assembler::ucomiss(dst, Address(rscratch, 0));
 2694   }
 2695 }
 2696 
 2697 void MacroAssembler::evucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2698   assert(rscratch != noreg || always_reachable(src), "missing");
 2699 
 2700   if (reachable(src)) {
 2701     Assembler::evucomxss(dst, as_Address(src));
 2702   } else {
 2703     lea(rscratch, src);
 2704     Assembler::evucomxss(dst, Address(rscratch, 0));
 2705   }
 2706 }
 2707 
 2708 void MacroAssembler::evucomish(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2709   assert(rscratch != noreg || always_reachable(src), "missing");
 2710 
 2711   if (reachable(src)) {
 2712     Assembler::evucomish(dst, as_Address(src));
 2713   } else {
 2714     lea(rscratch, src);
 2715     Assembler::evucomish(dst, Address(rscratch, 0));
 2716   }
 2717 }
 2718 
 2719 void MacroAssembler::evucomxsh(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2720   assert(rscratch != noreg || always_reachable(src), "missing");
 2721 
 2722   if (reachable(src)) {
 2723     Assembler::evucomxsh(dst, as_Address(src));
 2724   } else {
 2725     lea(rscratch, src);
 2726     Assembler::evucomxsh(dst, Address(rscratch, 0));
 2727   }
 2728 }
 2729 
 2730 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2731   assert(rscratch != noreg || always_reachable(src), "missing");
 2732 
 2733   // Used in sign-bit flipping with aligned address.
 2734   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2735 
 2736   if (UseAVX > 2 &&
 2737       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2738       (dst->encoding() >= 16)) {
 2739     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2740   } else if (reachable(src)) {
 2741     Assembler::xorpd(dst, as_Address(src));
 2742   } else {
 2743     lea(rscratch, src);
 2744     Assembler::xorpd(dst, Address(rscratch, 0));
 2745   }
 2746 }
 2747 
 2748 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2749   if (UseAVX > 2 &&
 2750       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2751       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2752     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2753   } else {
 2754     Assembler::xorpd(dst, src);
 2755   }
 2756 }
 2757 
 2758 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2759   if (UseAVX > 2 &&
 2760       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2761       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2762     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2763   } else {
 2764     Assembler::xorps(dst, src);
 2765   }
 2766 }
 2767 
 2768 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2769   assert(rscratch != noreg || always_reachable(src), "missing");
 2770 
 2771   // Used in sign-bit flipping with aligned address.
 2772   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2773 
 2774   if (UseAVX > 2 &&
 2775       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2776       (dst->encoding() >= 16)) {
 2777     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2778   } else if (reachable(src)) {
 2779     Assembler::xorps(dst, as_Address(src));
 2780   } else {
 2781     lea(rscratch, src);
 2782     Assembler::xorps(dst, Address(rscratch, 0));
 2783   }
 2784 }
 2785 
 2786 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2787   assert(rscratch != noreg || always_reachable(src), "missing");
 2788 
 2789   // Used in sign-bit flipping with aligned address.
 2790   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2791   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2792   if (reachable(src)) {
 2793     Assembler::pshufb(dst, as_Address(src));
 2794   } else {
 2795     lea(rscratch, src);
 2796     Assembler::pshufb(dst, Address(rscratch, 0));
 2797   }
 2798 }
 2799 
 2800 // AVX 3-operands instructions
 2801 
 2802 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2803   assert(rscratch != noreg || always_reachable(src), "missing");
 2804 
 2805   if (reachable(src)) {
 2806     vaddsd(dst, nds, as_Address(src));
 2807   } else {
 2808     lea(rscratch, src);
 2809     vaddsd(dst, nds, Address(rscratch, 0));
 2810   }
 2811 }
 2812 
 2813 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2814   assert(rscratch != noreg || always_reachable(src), "missing");
 2815 
 2816   if (reachable(src)) {
 2817     vaddss(dst, nds, as_Address(src));
 2818   } else {
 2819     lea(rscratch, src);
 2820     vaddss(dst, nds, Address(rscratch, 0));
 2821   }
 2822 }
 2823 
 2824 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2825   assert(UseAVX > 0, "requires some form of AVX");
 2826   assert(rscratch != noreg || always_reachable(src), "missing");
 2827 
 2828   if (reachable(src)) {
 2829     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2830   } else {
 2831     lea(rscratch, src);
 2832     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2833   }
 2834 }
 2835 
 2836 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2837   assert(UseAVX > 0, "requires some form of AVX");
 2838   assert(rscratch != noreg || always_reachable(src), "missing");
 2839 
 2840   if (reachable(src)) {
 2841     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2842   } else {
 2843     lea(rscratch, src);
 2844     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2845   }
 2846 }
 2847 
 2848 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2849   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2850   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2851 
 2852   vandps(dst, nds, negate_field, vector_len, rscratch);
 2853 }
 2854 
 2855 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2856   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2857   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2858 
 2859   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2860 }
 2861 
 2862 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2863   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2864   Assembler::vpaddb(dst, nds, src, vector_len);
 2865 }
 2866 
 2867 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2868   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2869   Assembler::vpaddb(dst, nds, src, vector_len);
 2870 }
 2871 
 2872 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2873   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2874   Assembler::vpaddw(dst, nds, src, vector_len);
 2875 }
 2876 
 2877 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2878   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2879   Assembler::vpaddw(dst, nds, src, vector_len);
 2880 }
 2881 
 2882 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2883   assert(rscratch != noreg || always_reachable(src), "missing");
 2884 
 2885   if (reachable(src)) {
 2886     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 2887   } else {
 2888     lea(rscratch, src);
 2889     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 2890   }
 2891 }
 2892 
 2893 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2894   assert(rscratch != noreg || always_reachable(src), "missing");
 2895 
 2896   if (reachable(src)) {
 2897     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 2898   } else {
 2899     lea(rscratch, src);
 2900     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 2901   }
 2902 }
 2903 
 2904 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2905   assert(rscratch != noreg || always_reachable(src), "missing");
 2906 
 2907   if (reachable(src)) {
 2908     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 2909   } else {
 2910     lea(rscratch, src);
 2911     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 2912   }
 2913 }
 2914 
 2915 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2916   assert(rscratch != noreg || always_reachable(src), "missing");
 2917 
 2918   if (reachable(src)) {
 2919     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 2920   } else {
 2921     lea(rscratch, src);
 2922     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 2923   }
 2924 }
 2925 
 2926 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2927   assert(rscratch != noreg || always_reachable(src), "missing");
 2928 
 2929   if (reachable(src)) {
 2930     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 2931   } else {
 2932     lea(rscratch, src);
 2933     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 2934   }
 2935 }
 2936 
 2937 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2938   assert(rscratch != noreg || always_reachable(src), "missing");
 2939 
 2940   if (reachable(src)) {
 2941     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 2942   } else {
 2943     lea(rscratch, src);
 2944     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 2945   }
 2946 }
 2947 
 2948 // Vector float blend
 2949 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2950 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2951   // WARN: Allow dst == (src1|src2), mask == scratch
 2952   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2953                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2954   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 2955   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2956   if (blend_emulation && scratch_available && dst_available) {
 2957     if (compute_mask) {
 2958       vpsrad(scratch, mask, 32, vector_len);
 2959       mask = scratch;
 2960     }
 2961     if (dst == src1) {
 2962       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 2963       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 2964     } else {
 2965       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 2966       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 2967     }
 2968     vpor(dst, dst, scratch, vector_len);
 2969   } else {
 2970     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 2971   }
 2972 }
 2973 
 2974 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 2975 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 2976   // WARN: Allow dst == (src1|src2), mask == scratch
 2977   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 2978                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 2979   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 2980   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 2981   if (blend_emulation && scratch_available && dst_available) {
 2982     if (compute_mask) {
 2983       vpxor(scratch, scratch, scratch, vector_len);
 2984       vpcmpgtq(scratch, scratch, mask, vector_len);
 2985       mask = scratch;
 2986     }
 2987     if (dst == src1) {
 2988       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 2989       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 2990     } else {
 2991       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 2992       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 2993     }
 2994     vpor(dst, dst, scratch, vector_len);
 2995   } else {
 2996     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 2997   }
 2998 }
 2999 
 3000 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3001   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3002   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 3003 }
 3004 
 3005 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 3006   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3007   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 3008 }
 3009 
 3010 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3011   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3012   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3013 }
 3014 
 3015 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3016   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3017   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3018 }
 3019 
 3020 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3021   assert(rscratch != noreg || always_reachable(src), "missing");
 3022 
 3023   if (reachable(src)) {
 3024     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 3025   } else {
 3026     lea(rscratch, src);
 3027     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 3028   }
 3029 }
 3030 
 3031 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3032                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3033   assert(rscratch != noreg || always_reachable(src), "missing");
 3034 
 3035   if (reachable(src)) {
 3036     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3037   } else {
 3038     lea(rscratch, src);
 3039     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3040   }
 3041 }
 3042 
 3043 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3044                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3045   assert(rscratch != noreg || always_reachable(src), "missing");
 3046 
 3047   if (reachable(src)) {
 3048     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3049   } else {
 3050     lea(rscratch, src);
 3051     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3052   }
 3053 }
 3054 
 3055 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3056                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3057   assert(rscratch != noreg || always_reachable(src), "missing");
 3058 
 3059   if (reachable(src)) {
 3060     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3061   } else {
 3062     lea(rscratch, src);
 3063     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3064   }
 3065 }
 3066 
 3067 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3068                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3069   assert(rscratch != noreg || always_reachable(src), "missing");
 3070 
 3071   if (reachable(src)) {
 3072     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3073   } else {
 3074     lea(rscratch, src);
 3075     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3076   }
 3077 }
 3078 
 3079 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3080   if (width == Assembler::Q) {
 3081     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3082   } else {
 3083     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3084   }
 3085 }
 3086 
 3087 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3088   int eq_cond_enc = 0x29;
 3089   int gt_cond_enc = 0x37;
 3090   if (width != Assembler::Q) {
 3091     eq_cond_enc = 0x74 + width;
 3092     gt_cond_enc = 0x64 + width;
 3093   }
 3094   switch (cond) {
 3095   case eq:
 3096     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3097     break;
 3098   case neq:
 3099     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3100     vallones(xtmp, vector_len);
 3101     vpxor(dst, xtmp, dst, vector_len);
 3102     break;
 3103   case le:
 3104     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3105     vallones(xtmp, vector_len);
 3106     vpxor(dst, xtmp, dst, vector_len);
 3107     break;
 3108   case nlt:
 3109     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3110     vallones(xtmp, vector_len);
 3111     vpxor(dst, xtmp, dst, vector_len);
 3112     break;
 3113   case lt:
 3114     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3115     break;
 3116   case nle:
 3117     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3118     break;
 3119   default:
 3120     assert(false, "Should not reach here");
 3121   }
 3122 }
 3123 
 3124 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3125   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3126   Assembler::vpmovzxbw(dst, src, vector_len);
 3127 }
 3128 
 3129 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3130   assert((src->encoding() < 16),"XMM register should be 0-15");
 3131   Assembler::vpmovmskb(dst, src, vector_len);
 3132 }
 3133 
 3134 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3135   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3136   Assembler::vpmullw(dst, nds, src, vector_len);
 3137 }
 3138 
 3139 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3140   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3141   Assembler::vpmullw(dst, nds, src, vector_len);
 3142 }
 3143 
 3144 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3145   assert((UseAVX > 0), "AVX support is needed");
 3146   assert(rscratch != noreg || always_reachable(src), "missing");
 3147 
 3148   if (reachable(src)) {
 3149     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3150   } else {
 3151     lea(rscratch, src);
 3152     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3153   }
 3154 }
 3155 
 3156 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3157   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3158   Assembler::vpsubb(dst, nds, src, vector_len);
 3159 }
 3160 
 3161 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3162   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3163   Assembler::vpsubb(dst, nds, src, vector_len);
 3164 }
 3165 
 3166 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3167   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3168   Assembler::vpsubw(dst, nds, src, vector_len);
 3169 }
 3170 
 3171 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3172   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3173   Assembler::vpsubw(dst, nds, src, vector_len);
 3174 }
 3175 
 3176 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3177   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3178   Assembler::vpsraw(dst, nds, shift, vector_len);
 3179 }
 3180 
 3181 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3182   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3183   Assembler::vpsraw(dst, nds, shift, vector_len);
 3184 }
 3185 
 3186 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3187   assert(UseAVX > 2,"");
 3188   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3189      vector_len = 2;
 3190   }
 3191   Assembler::evpsraq(dst, nds, shift, vector_len);
 3192 }
 3193 
 3194 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3195   assert(UseAVX > 2,"");
 3196   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3197      vector_len = 2;
 3198   }
 3199   Assembler::evpsraq(dst, nds, shift, vector_len);
 3200 }
 3201 
 3202 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3203   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3204   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3205 }
 3206 
 3207 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3208   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3209   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3210 }
 3211 
 3212 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3213   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3214   Assembler::vpsllw(dst, nds, shift, vector_len);
 3215 }
 3216 
 3217 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3218   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3219   Assembler::vpsllw(dst, nds, shift, vector_len);
 3220 }
 3221 
 3222 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3223   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3224   Assembler::vptest(dst, src);
 3225 }
 3226 
 3227 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3228   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3229   Assembler::punpcklbw(dst, src);
 3230 }
 3231 
 3232 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3233   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3234   Assembler::pshufd(dst, src, mode);
 3235 }
 3236 
 3237 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3238   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3239   Assembler::pshuflw(dst, src, mode);
 3240 }
 3241 
 3242 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3243   assert(rscratch != noreg || always_reachable(src), "missing");
 3244 
 3245   if (reachable(src)) {
 3246     vandpd(dst, nds, as_Address(src), vector_len);
 3247   } else {
 3248     lea(rscratch, src);
 3249     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3250   }
 3251 }
 3252 
 3253 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3254   assert(rscratch != noreg || always_reachable(src), "missing");
 3255 
 3256   if (reachable(src)) {
 3257     vandps(dst, nds, as_Address(src), vector_len);
 3258   } else {
 3259     lea(rscratch, src);
 3260     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3261   }
 3262 }
 3263 
 3264 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3265                             bool merge, int vector_len, Register rscratch) {
 3266   assert(rscratch != noreg || always_reachable(src), "missing");
 3267 
 3268   if (reachable(src)) {
 3269     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3270   } else {
 3271     lea(rscratch, src);
 3272     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3273   }
 3274 }
 3275 
 3276 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3277   assert(rscratch != noreg || always_reachable(src), "missing");
 3278 
 3279   if (reachable(src)) {
 3280     vdivsd(dst, nds, as_Address(src));
 3281   } else {
 3282     lea(rscratch, src);
 3283     vdivsd(dst, nds, Address(rscratch, 0));
 3284   }
 3285 }
 3286 
 3287 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3288   assert(rscratch != noreg || always_reachable(src), "missing");
 3289 
 3290   if (reachable(src)) {
 3291     vdivss(dst, nds, as_Address(src));
 3292   } else {
 3293     lea(rscratch, src);
 3294     vdivss(dst, nds, Address(rscratch, 0));
 3295   }
 3296 }
 3297 
 3298 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3299   assert(rscratch != noreg || always_reachable(src), "missing");
 3300 
 3301   if (reachable(src)) {
 3302     vmulsd(dst, nds, as_Address(src));
 3303   } else {
 3304     lea(rscratch, src);
 3305     vmulsd(dst, nds, Address(rscratch, 0));
 3306   }
 3307 }
 3308 
 3309 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3310   assert(rscratch != noreg || always_reachable(src), "missing");
 3311 
 3312   if (reachable(src)) {
 3313     vmulss(dst, nds, as_Address(src));
 3314   } else {
 3315     lea(rscratch, src);
 3316     vmulss(dst, nds, Address(rscratch, 0));
 3317   }
 3318 }
 3319 
 3320 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3321   assert(rscratch != noreg || always_reachable(src), "missing");
 3322 
 3323   if (reachable(src)) {
 3324     vsubsd(dst, nds, as_Address(src));
 3325   } else {
 3326     lea(rscratch, src);
 3327     vsubsd(dst, nds, Address(rscratch, 0));
 3328   }
 3329 }
 3330 
 3331 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3332   assert(rscratch != noreg || always_reachable(src), "missing");
 3333 
 3334   if (reachable(src)) {
 3335     vsubss(dst, nds, as_Address(src));
 3336   } else {
 3337     lea(rscratch, src);
 3338     vsubss(dst, nds, Address(rscratch, 0));
 3339   }
 3340 }
 3341 
 3342 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3343   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3344   assert(rscratch != noreg || always_reachable(src), "missing");
 3345 
 3346   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3347 }
 3348 
 3349 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3350   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3351   assert(rscratch != noreg || always_reachable(src), "missing");
 3352 
 3353   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3354 }
 3355 
 3356 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3357   assert(rscratch != noreg || always_reachable(src), "missing");
 3358 
 3359   if (reachable(src)) {
 3360     vxorpd(dst, nds, as_Address(src), vector_len);
 3361   } else {
 3362     lea(rscratch, src);
 3363     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3364   }
 3365 }
 3366 
 3367 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3368   assert(rscratch != noreg || always_reachable(src), "missing");
 3369 
 3370   if (reachable(src)) {
 3371     vxorps(dst, nds, as_Address(src), vector_len);
 3372   } else {
 3373     lea(rscratch, src);
 3374     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3375   }
 3376 }
 3377 
 3378 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3379   assert(rscratch != noreg || always_reachable(src), "missing");
 3380 
 3381   if (UseAVX > 1 || (vector_len < 1)) {
 3382     if (reachable(src)) {
 3383       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3384     } else {
 3385       lea(rscratch, src);
 3386       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3387     }
 3388   } else {
 3389     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3390   }
 3391 }
 3392 
 3393 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3394   assert(rscratch != noreg || always_reachable(src), "missing");
 3395 
 3396   if (reachable(src)) {
 3397     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3398   } else {
 3399     lea(rscratch, src);
 3400     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3401   }
 3402 }
 3403 
 3404 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3405   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3406   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3407   // The inverted mask is sign-extended
 3408   andptr(possibly_non_local, inverted_mask);
 3409 }
 3410 
 3411 void MacroAssembler::resolve_jobject(Register value,
 3412                                      Register tmp) {
 3413   Register thread = r15_thread;
 3414   assert_different_registers(value, thread, tmp);
 3415   Label done, tagged, weak_tagged;
 3416   testptr(value, value);
 3417   jcc(Assembler::zero, done);           // Use null as-is.
 3418   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3419   jcc(Assembler::notZero, tagged);
 3420 
 3421   // Resolve local handle
 3422   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3423   verify_oop(value);
 3424   jmp(done);
 3425 
 3426   bind(tagged);
 3427   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3428   jcc(Assembler::notZero, weak_tagged);
 3429 
 3430   // Resolve global handle
 3431   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3432   verify_oop(value);
 3433   jmp(done);
 3434 
 3435   bind(weak_tagged);
 3436   // Resolve jweak.
 3437   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3438                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3439   verify_oop(value);
 3440 
 3441   bind(done);
 3442 }
 3443 
 3444 void MacroAssembler::resolve_global_jobject(Register value,
 3445                                             Register tmp) {
 3446   Register thread = r15_thread;
 3447   assert_different_registers(value, thread, tmp);
 3448   Label done;
 3449 
 3450   testptr(value, value);
 3451   jcc(Assembler::zero, done);           // Use null as-is.
 3452 
 3453 #ifdef ASSERT
 3454   {
 3455     Label valid_global_tag;
 3456     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3457     jcc(Assembler::notZero, valid_global_tag);
 3458     stop("non global jobject using resolve_global_jobject");
 3459     bind(valid_global_tag);
 3460   }
 3461 #endif
 3462 
 3463   // Resolve global handle
 3464   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3465   verify_oop(value);
 3466 
 3467   bind(done);
 3468 }
 3469 
 3470 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3471   subq(dst, imm32);
 3472 }
 3473 
 3474 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3475 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3476   subq_imm32(dst, imm32);
 3477 }
 3478 
 3479 void MacroAssembler::subptr(Register dst, Register src) {
 3480   subq(dst, src);
 3481 }
 3482 
 3483 // C++ bool manipulation
 3484 void MacroAssembler::testbool(Register dst) {
 3485   if(sizeof(bool) == 1)
 3486     testb(dst, 0xff);
 3487   else if(sizeof(bool) == 2) {
 3488     // testw implementation needed for two byte bools
 3489     ShouldNotReachHere();
 3490   } else if(sizeof(bool) == 4)
 3491     testl(dst, dst);
 3492   else
 3493     // unsupported
 3494     ShouldNotReachHere();
 3495 }
 3496 
 3497 void MacroAssembler::testptr(Register dst, Register src) {
 3498   testq(dst, src);
 3499 }
 3500 






















































































































 3501 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3502 void MacroAssembler::tlab_allocate(Register obj,
 3503                                    Register var_size_in_bytes,
 3504                                    int con_size_in_bytes,
 3505                                    Register t1,
 3506                                    Register t2,
 3507                                    Label& slow_case) {
 3508   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3509   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3510 }
 3511 
 3512 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3513   RegSet regs;
 3514   regs += RegSet::of(rax, rcx, rdx);
 3515 #ifndef _WINDOWS
 3516   regs += RegSet::of(rsi, rdi);
 3517 #endif
 3518   regs += RegSet::range(r8, r11);
 3519   if (UseAPX) {
 3520     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3521   }
 3522   return regs;
 3523 }
 3524 
 3525 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3526   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3527 #if defined(_WINDOWS)
 3528   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3529   if (num_xmm_registers > 16) {
 3530      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3531   }
 3532   return result;
 3533 #else
 3534   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3535 #endif
 3536 }
 3537 
 3538 // C1 only ever uses the first double/float of the XMM register.
 3539 static int xmm_save_size() { return sizeof(double); }
 3540 
 3541 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3542   masm->movdbl(Address(rsp, offset), reg);
 3543 }
 3544 
 3545 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3546   masm->movdbl(reg, Address(rsp, offset));
 3547 }
 3548 
 3549 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3550                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3551 
 3552   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3553                          StackAlignmentInBytes);
 3554   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3555 
 3556   return gp_area_size + xmm_area_size;
 3557 }
 3558 
 3559 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3560   block_comment("push_call_clobbered_registers start");
 3561   // Regular registers
 3562   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3563 
 3564   int gp_area_size;
 3565   int xmm_area_size;
 3566   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3567                                                gp_area_size, xmm_area_size);
 3568   subptr(rsp, total_save_size);
 3569 
 3570   push_set(gp_registers_to_push, 0);
 3571 
 3572   if (save_fpu) {
 3573     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3574   }
 3575 
 3576   block_comment("push_call_clobbered_registers end");
 3577 }
 3578 
 3579 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3580   block_comment("pop_call_clobbered_registers start");
 3581 
 3582   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 3583 
 3584   int gp_area_size;
 3585   int xmm_area_size;
 3586   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 3587                                                gp_area_size, xmm_area_size);
 3588 
 3589   if (restore_fpu) {
 3590     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3591   }
 3592 
 3593   pop_set(gp_registers_to_pop, 0);
 3594 
 3595   addptr(rsp, total_save_size);
 3596 
 3597   vzeroupper();
 3598 
 3599   block_comment("pop_call_clobbered_registers end");
 3600 }
 3601 
 3602 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3603   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3604   int spill_offset = offset;
 3605 
 3606   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3607     save_xmm_register(this, spill_offset, *it);
 3608     spill_offset += xmm_save_size();
 3609   }
 3610 }
 3611 
 3612 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3613   int restore_size = set.size() * xmm_save_size();
 3614   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3615 
 3616   int restore_offset = offset + restore_size - xmm_save_size();
 3617 
 3618   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3619     restore_xmm_register(this, restore_offset, *it);
 3620     restore_offset -= xmm_save_size();
 3621   }
 3622 }
 3623 
 3624 void MacroAssembler::push_set(RegSet set, int offset) {
 3625   int spill_offset;
 3626   if (offset == -1) {
 3627     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3628     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3629     subptr(rsp, aligned_size);
 3630     spill_offset = 0;
 3631   } else {
 3632     spill_offset = offset;
 3633   }
 3634 
 3635   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3636     movptr(Address(rsp, spill_offset), *it);
 3637     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3638   }
 3639 }
 3640 
 3641 void MacroAssembler::pop_set(RegSet set, int offset) {
 3642 
 3643   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3644   int restore_size = set.size() * gp_reg_size;
 3645   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3646 
 3647   int restore_offset;
 3648   if (offset == -1) {
 3649     restore_offset = restore_size - gp_reg_size;
 3650   } else {
 3651     restore_offset = offset + restore_size - gp_reg_size;
 3652   }
 3653   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3654     movptr(*it, Address(rsp, restore_offset));
 3655     restore_offset -= gp_reg_size;
 3656   }
 3657 
 3658   if (offset == -1) {
 3659     addptr(rsp, aligned_size);
 3660   }
 3661 }
 3662 
 3663 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3664 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3665   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3666   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3667   Label done;
 3668 
 3669   testptr(length_in_bytes, length_in_bytes);
 3670   jcc(Assembler::zero, done);
 3671 
 3672   // initialize topmost word, divide index by 2, check if odd and test if zero
 3673   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3674 #ifdef ASSERT
 3675   {
 3676     Label L;
 3677     testptr(length_in_bytes, BytesPerWord - 1);
 3678     jcc(Assembler::zero, L);
 3679     stop("length must be a multiple of BytesPerWord");
 3680     bind(L);
 3681   }
 3682 #endif
 3683   Register index = length_in_bytes;
 3684   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3685   if (UseIncDec) {
 3686     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3687   } else {
 3688     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3689     shrptr(index, 1);
 3690   }
 3691 
 3692   // initialize remaining object fields: index is a multiple of 2 now
 3693   {
 3694     Label loop;
 3695     bind(loop);
 3696     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3697     decrement(index);
 3698     jcc(Assembler::notZero, loop);
 3699   }
 3700 
 3701   bind(done);
 3702 }
 3703 






















 3704 // Look up the method for a megamorphic invokeinterface call.
 3705 // The target method is determined by <intf_klass, itable_index>.
 3706 // The receiver klass is in recv_klass.
 3707 // On success, the result will be in method_result, and execution falls through.
 3708 // On failure, execution transfers to the given label.
 3709 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3710                                              Register intf_klass,
 3711                                              RegisterOrConstant itable_index,
 3712                                              Register method_result,
 3713                                              Register scan_temp,
 3714                                              Label& L_no_such_interface,
 3715                                              bool return_method) {
 3716   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3717   assert_different_registers(method_result, intf_klass, scan_temp);
 3718   assert(recv_klass != method_result || !return_method,
 3719          "recv_klass can be destroyed when method isn't needed");
 3720 
 3721   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3722          "caller must use same register for non-constant itable index as for method");
 3723 
 3724   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3725   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3726   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3727   int scan_step   = itableOffsetEntry::size() * wordSize;
 3728   int vte_size    = vtableEntry::size_in_bytes();
 3729   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3730   assert(vte_size == wordSize, "else adjust times_vte_scale");
 3731 
 3732   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3733 
 3734   // Could store the aligned, prescaled offset in the klass.
 3735   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 3736 
 3737   if (return_method) {
 3738     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 3739     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3740     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 3741   }
 3742 
 3743   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 3744   //   if (scan->interface() == intf) {
 3745   //     result = (klass + scan->offset() + itable_index);
 3746   //   }
 3747   // }
 3748   Label search, found_method;
 3749 
 3750   for (int peel = 1; peel >= 0; peel--) {
 3751     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 3752     cmpptr(intf_klass, method_result);
 3753 
 3754     if (peel) {
 3755       jccb(Assembler::equal, found_method);
 3756     } else {
 3757       jccb(Assembler::notEqual, search);
 3758       // (invert the test to fall through to found_method...)
 3759     }
 3760 
 3761     if (!peel)  break;
 3762 
 3763     bind(search);
 3764 
 3765     // Check that the previous entry is non-null.  A null entry means that
 3766     // the receiver class doesn't implement the interface, and wasn't the
 3767     // same as when the caller was compiled.
 3768     testptr(method_result, method_result);
 3769     jcc(Assembler::zero, L_no_such_interface);
 3770     addptr(scan_temp, scan_step);
 3771   }
 3772 
 3773   bind(found_method);
 3774 
 3775   if (return_method) {
 3776     // Got a hit.
 3777     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 3778     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 3779   }
 3780 }
 3781 
 3782 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 3783 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 3784 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 3785 // The target method is determined by <holder_klass, itable_index>.
 3786 // The receiver klass is in recv_klass.
 3787 // On success, the result will be in method_result, and execution falls through.
 3788 // On failure, execution transfers to the given label.
 3789 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 3790                                                   Register holder_klass,
 3791                                                   Register resolved_klass,
 3792                                                   Register method_result,
 3793                                                   Register scan_temp,
 3794                                                   Register temp_reg2,
 3795                                                   Register receiver,
 3796                                                   int itable_index,
 3797                                                   Label& L_no_such_interface) {
 3798   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 3799   Register temp_itbl_klass = method_result;
 3800   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 3801 
 3802   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3803   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3804   int scan_step = itableOffsetEntry::size() * wordSize;
 3805   int vte_size = vtableEntry::size_in_bytes();
 3806   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 3807   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 3808   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3809   assert(vte_size == wordSize, "adjust times_vte_scale");
 3810 
 3811   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 3812 
 3813   // temp_itbl_klass = recv_klass.itable[0]
 3814   // scan_temp = &recv_klass.itable[0] + step
 3815   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3816   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 3817   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 3818   xorptr(temp_reg, temp_reg);
 3819 
 3820   // Initial checks:
 3821   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 3822   //   - if (itable[0] == 0), no such interface
 3823   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 3824   cmpptr(holder_klass, resolved_klass);
 3825   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 3826   testptr(temp_itbl_klass, temp_itbl_klass);
 3827   jccb(Assembler::zero, L_no_such_interface);
 3828   cmpptr(holder_klass, temp_itbl_klass);
 3829   jccb(Assembler::equal, L_holder_found);
 3830 
 3831   // Loop: Look for holder_klass record in itable
 3832   //   do {
 3833   //     tmp = itable[index];
 3834   //     index += step;
 3835   //     if (tmp == holder_klass) {
 3836   //       goto L_holder_found; // Found!
 3837   //     }
 3838   //   } while (tmp != 0);
 3839   //   goto L_no_such_interface // Not found.
 3840   Label L_scan_holder;
 3841   bind(L_scan_holder);
 3842     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3843     addptr(scan_temp, scan_step);
 3844     cmpptr(holder_klass, temp_itbl_klass);
 3845     jccb(Assembler::equal, L_holder_found);
 3846     testptr(temp_itbl_klass, temp_itbl_klass);
 3847     jccb(Assembler::notZero, L_scan_holder);
 3848 
 3849   jmpb(L_no_such_interface);
 3850 
 3851   // Loop: Look for resolved_class record in itable
 3852   //   do {
 3853   //     tmp = itable[index];
 3854   //     index += step;
 3855   //     if (tmp == holder_klass) {
 3856   //        // Also check if we have met a holder klass
 3857   //        holder_tmp = itable[index-step-ioffset];
 3858   //     }
 3859   //     if (tmp == resolved_klass) {
 3860   //        goto L_resolved_found;  // Found!
 3861   //     }
 3862   //   } while (tmp != 0);
 3863   //   goto L_no_such_interface // Not found.
 3864   //
 3865   Label L_loop_scan_resolved;
 3866   bind(L_loop_scan_resolved);
 3867     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3868     addptr(scan_temp, scan_step);
 3869     bind(L_loop_scan_resolved_entry);
 3870     cmpptr(holder_klass, temp_itbl_klass);
 3871     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3872     cmpptr(resolved_klass, temp_itbl_klass);
 3873     jccb(Assembler::equal, L_resolved_found);
 3874     testptr(temp_itbl_klass, temp_itbl_klass);
 3875     jccb(Assembler::notZero, L_loop_scan_resolved);
 3876 
 3877   jmpb(L_no_such_interface);
 3878 
 3879   Label L_ready;
 3880 
 3881   // See if we already have a holder klass. If not, go and scan for it.
 3882   bind(L_resolved_found);
 3883   testptr(temp_reg, temp_reg);
 3884   jccb(Assembler::zero, L_scan_holder);
 3885   jmpb(L_ready);
 3886 
 3887   bind(L_holder_found);
 3888   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3889 
 3890   // Finally, temp_reg contains holder_klass vtable offset
 3891   bind(L_ready);
 3892   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3893   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 3894     load_klass(scan_temp, receiver, noreg);
 3895     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3896   } else {
 3897     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 3898   }
 3899 }
 3900 
 3901 
 3902 // virtual method calling
 3903 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 3904                                            RegisterOrConstant vtable_index,
 3905                                            Register method_result) {
 3906   const ByteSize base = Klass::vtable_start_offset();
 3907   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 3908   Address vtable_entry_addr(recv_klass,
 3909                             vtable_index, Address::times_ptr,
 3910                             base + vtableEntry::method_offset());
 3911   movptr(method_result, vtable_entry_addr);
 3912 }
 3913 
 3914 
 3915 void MacroAssembler::check_klass_subtype(Register sub_klass,
 3916                            Register super_klass,
 3917                            Register temp_reg,
 3918                            Label& L_success) {
 3919   Label L_failure;
 3920   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 3921   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 3922   bind(L_failure);
 3923 }
 3924 
 3925 
 3926 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 3927                                                    Register super_klass,
 3928                                                    Register temp_reg,
 3929                                                    Label* L_success,
 3930                                                    Label* L_failure,
 3931                                                    Label* L_slow_path,
 3932                                         RegisterOrConstant super_check_offset) {
 3933   assert_different_registers(sub_klass, super_klass, temp_reg);
 3934   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 3935   if (super_check_offset.is_register()) {
 3936     assert_different_registers(sub_klass, super_klass,
 3937                                super_check_offset.as_register());
 3938   } else if (must_load_sco) {
 3939     assert(temp_reg != noreg, "supply either a temp or a register offset");
 3940   }
 3941 
 3942   Label L_fallthrough;
 3943   int label_nulls = 0;
 3944   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 3945   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 3946   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 3947   assert(label_nulls <= 1, "at most one null in the batch");
 3948 
 3949   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 3950   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 3951   Address super_check_offset_addr(super_klass, sco_offset);
 3952 
 3953   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 3954   // range of a jccb.  If this routine grows larger, reconsider at
 3955   // least some of these.
 3956 #define local_jcc(assembler_cond, label)                                \
 3957   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 3958   else                             jcc( assembler_cond, label) /*omit semi*/
 3959 
 3960   // Hacked jmp, which may only be used just before L_fallthrough.
 3961 #define final_jmp(label)                                                \
 3962   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 3963   else                            jmp(label)                /*omit semi*/
 3964 
 3965   // If the pointers are equal, we are done (e.g., String[] elements).
 3966   // This self-check enables sharing of secondary supertype arrays among
 3967   // non-primary types such as array-of-interface.  Otherwise, each such
 3968   // type would need its own customized SSA.
 3969   // We move this check to the front of the fast path because many
 3970   // type checks are in fact trivially successful in this manner,
 3971   // so we get a nicely predicted branch right at the start of the check.
 3972   cmpptr(sub_klass, super_klass);
 3973   local_jcc(Assembler::equal, *L_success);
 3974 
 3975   // Check the supertype display:
 3976   if (must_load_sco) {
 3977     // Positive movl does right thing on LP64.
 3978     movl(temp_reg, super_check_offset_addr);
 3979     super_check_offset = RegisterOrConstant(temp_reg);
 3980   }
 3981   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 3982   cmpptr(super_klass, super_check_addr); // load displayed supertype
 3983 
 3984   // This check has worked decisively for primary supers.
 3985   // Secondary supers are sought in the super_cache ('super_cache_addr').
 3986   // (Secondary supers are interfaces and very deeply nested subtypes.)
 3987   // This works in the same check above because of a tricky aliasing
 3988   // between the super_cache and the primary super display elements.
 3989   // (The 'super_check_addr' can address either, as the case requires.)
 3990   // Note that the cache is updated below if it does not help us find
 3991   // what we need immediately.
 3992   // So if it was a primary super, we can just fail immediately.
 3993   // Otherwise, it's the slow path for us (no success at this point).
 3994 
 3995   if (super_check_offset.is_register()) {
 3996     local_jcc(Assembler::equal, *L_success);
 3997     cmpl(super_check_offset.as_register(), sc_offset);
 3998     if (L_failure == &L_fallthrough) {
 3999       local_jcc(Assembler::equal, *L_slow_path);
 4000     } else {
 4001       local_jcc(Assembler::notEqual, *L_failure);
 4002       final_jmp(*L_slow_path);
 4003     }
 4004   } else if (super_check_offset.as_constant() == sc_offset) {
 4005     // Need a slow path; fast failure is impossible.
 4006     if (L_slow_path == &L_fallthrough) {
 4007       local_jcc(Assembler::equal, *L_success);
 4008     } else {
 4009       local_jcc(Assembler::notEqual, *L_slow_path);
 4010       final_jmp(*L_success);
 4011     }
 4012   } else {
 4013     // No slow path; it's a fast decision.
 4014     if (L_failure == &L_fallthrough) {
 4015       local_jcc(Assembler::equal, *L_success);
 4016     } else {
 4017       local_jcc(Assembler::notEqual, *L_failure);
 4018       final_jmp(*L_success);
 4019     }
 4020   }
 4021 
 4022   bind(L_fallthrough);
 4023 
 4024 #undef local_jcc
 4025 #undef final_jmp
 4026 }
 4027 
 4028 
 4029 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 4030                                                           Register super_klass,
 4031                                                           Register temp_reg,
 4032                                                           Register temp2_reg,
 4033                                                           Label* L_success,
 4034                                                           Label* L_failure,
 4035                                                           bool set_cond_codes) {
 4036   assert_different_registers(sub_klass, super_klass, temp_reg);
 4037   if (temp2_reg != noreg)
 4038     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4039 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4040 
 4041   Label L_fallthrough;
 4042   int label_nulls = 0;
 4043   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4044   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4045   assert(label_nulls <= 1, "at most one null in the batch");
 4046 
 4047   // a couple of useful fields in sub_klass:
 4048   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4049   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4050   Address secondary_supers_addr(sub_klass, ss_offset);
 4051   Address super_cache_addr(     sub_klass, sc_offset);
 4052 
 4053   // Do a linear scan of the secondary super-klass chain.
 4054   // This code is rarely used, so simplicity is a virtue here.
 4055   // The repne_scan instruction uses fixed registers, which we must spill.
 4056   // Don't worry too much about pre-existing connections with the input regs.
 4057 
 4058   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4059   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4060 
 4061   // Get super_klass value into rax (even if it was in rdi or rcx).
 4062   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4063   if (super_klass != rax) {
 4064     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4065     mov(rax, super_klass);
 4066   }
 4067   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4068   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4069 
 4070 #ifndef PRODUCT
 4071   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4072   ExternalAddress pst_counter_addr((address) pst_counter);
 4073   lea(rcx, pst_counter_addr);
 4074   incrementl(Address(rcx, 0));
 4075 #endif //PRODUCT
 4076 
 4077   // We will consult the secondary-super array.
 4078   movptr(rdi, secondary_supers_addr);
 4079   // Load the array length.  (Positive movl does right thing on LP64.)
 4080   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4081   // Skip to start of data.
 4082   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4083 
 4084   // Scan RCX words at [RDI] for an occurrence of RAX.
 4085   // Set NZ/Z based on last compare.
 4086   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4087   // not change flags (only scas instruction which is repeated sets flags).
 4088   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4089 
 4090     testptr(rax,rax); // Set Z = 0
 4091     repne_scan();
 4092 
 4093   // Unspill the temp. registers:
 4094   if (pushed_rdi)  pop(rdi);
 4095   if (pushed_rcx)  pop(rcx);
 4096   if (pushed_rax)  pop(rax);
 4097 
 4098   if (set_cond_codes) {
 4099     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4100     assert(!pushed_rdi, "rdi must be left non-null");
 4101     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4102   }
 4103 
 4104   if (L_failure == &L_fallthrough)
 4105         jccb(Assembler::notEqual, *L_failure);
 4106   else  jcc(Assembler::notEqual, *L_failure);
 4107 
 4108   // Success.  Cache the super we found and proceed in triumph.
 4109   movptr(super_cache_addr, super_klass);
 4110 
 4111   if (L_success != &L_fallthrough) {
 4112     jmp(*L_success);
 4113   }
 4114 
 4115 #undef IS_A_TEMP
 4116 
 4117   bind(L_fallthrough);
 4118 }
 4119 
 4120 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4121                                                    Register super_klass,
 4122                                                    Register temp_reg,
 4123                                                    Register temp2_reg,
 4124                                                    Label* L_success,
 4125                                                    Label* L_failure,
 4126                                                    bool set_cond_codes) {
 4127   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4128   check_klass_subtype_slow_path
 4129     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4130      L_success, L_failure);
 4131 }
 4132 
 4133 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4134                                                    Register super_klass,
 4135                                                    Register temp_reg,
 4136                                                    Register temp2_reg,
 4137                                                    Register temp3_reg,
 4138                                                    Register temp4_reg,
 4139                                                    Label* L_success,
 4140                                                    Label* L_failure) {
 4141   if (UseSecondarySupersTable) {
 4142     check_klass_subtype_slow_path_table
 4143       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4144        L_success, L_failure);
 4145   } else {
 4146     check_klass_subtype_slow_path_linear
 4147       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4148   }
 4149 }
 4150 
 4151 Register MacroAssembler::allocate_if_noreg(Register r,
 4152                                   RegSetIterator<Register> &available_regs,
 4153                                   RegSet &regs_to_push) {
 4154   if (!r->is_valid()) {
 4155     r = *available_regs++;
 4156     regs_to_push += r;
 4157   }
 4158   return r;
 4159 }
 4160 
 4161 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4162                                                          Register super_klass,
 4163                                                          Register temp_reg,
 4164                                                          Register temp2_reg,
 4165                                                          Register temp3_reg,
 4166                                                          Register result_reg,
 4167                                                          Label* L_success,
 4168                                                          Label* L_failure) {
 4169   // NB! Callers may assume that, when temp2_reg is a valid register,
 4170   // this code sets it to a nonzero value.
 4171   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4172 
 4173   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4174 
 4175   Label L_fallthrough;
 4176   int label_nulls = 0;
 4177   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4178   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4179   assert(label_nulls <= 1, "at most one null in the batch");
 4180 
 4181   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4182 
 4183   RegSetIterator<Register> available_regs
 4184     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4185 
 4186   RegSet pushed_regs;
 4187 
 4188   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4189   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4190   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4191   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4192   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4193 
 4194   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4195 
 4196   {
 4197 
 4198     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4199     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4200     subptr(rsp, aligned_size);
 4201     push_set(pushed_regs, 0);
 4202 
 4203     lookup_secondary_supers_table_var(sub_klass,
 4204                                       super_klass,
 4205                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4206     cmpq(result_reg, 0);
 4207 
 4208     // Unspill the temp. registers:
 4209     pop_set(pushed_regs, 0);
 4210     // Increment SP but do not clobber flags.
 4211     lea(rsp, Address(rsp, aligned_size));
 4212   }
 4213 
 4214   if (temp2_reg_was_valid) {
 4215     movq(temp2_reg, 1);
 4216   }
 4217 
 4218   jcc(Assembler::notEqual, *L_failure);
 4219 
 4220   if (L_success != &L_fallthrough) {
 4221     jmp(*L_success);
 4222   }
 4223 
 4224   bind(L_fallthrough);
 4225 }
 4226 
 4227 // population_count variant for running without the POPCNT
 4228 // instruction, which was introduced with SSE4.2 in 2008.
 4229 void MacroAssembler::population_count(Register dst, Register src,
 4230                                       Register scratch1, Register scratch2) {
 4231   assert_different_registers(src, scratch1, scratch2);
 4232   if (UsePopCountInstruction) {
 4233     Assembler::popcntq(dst, src);
 4234   } else {
 4235     assert_different_registers(src, scratch1, scratch2);
 4236     assert_different_registers(dst, scratch1, scratch2);
 4237     Label loop, done;
 4238 
 4239     mov(scratch1, src);
 4240     // dst = 0;
 4241     // while(scratch1 != 0) {
 4242     //   dst++;
 4243     //   scratch1 &= (scratch1 - 1);
 4244     // }
 4245     xorl(dst, dst);
 4246     testq(scratch1, scratch1);
 4247     jccb(Assembler::equal, done);
 4248     {
 4249       bind(loop);
 4250       incq(dst);
 4251       movq(scratch2, scratch1);
 4252       decq(scratch2);
 4253       andq(scratch1, scratch2);
 4254       jccb(Assembler::notEqual, loop);
 4255     }
 4256     bind(done);
 4257   }
 4258 #ifdef ASSERT
 4259   mov64(scratch1, 0xCafeBabeDeadBeef);
 4260   movq(scratch2, scratch1);
 4261 #endif
 4262 }
 4263 
 4264 // Ensure that the inline code and the stub are using the same registers.
 4265 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4266 do {                                                                 \
 4267   assert(r_super_klass  == rax, "mismatch");                         \
 4268   assert(r_array_base   == rbx, "mismatch");                         \
 4269   assert(r_array_length == rcx, "mismatch");                         \
 4270   assert(r_array_index  == rdx, "mismatch");                         \
 4271   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4272   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4273   assert(result         == rdi || result      == noreg, "mismatch"); \
 4274 } while(0)
 4275 
 4276 // Versions of salq and rorq that don't need count to be in rcx
 4277 
 4278 void MacroAssembler::salq(Register dest, Register count) {
 4279   if (count == rcx) {
 4280     Assembler::salq(dest);
 4281   } else {
 4282     assert_different_registers(rcx, dest);
 4283     xchgq(rcx, count);
 4284     Assembler::salq(dest);
 4285     xchgq(rcx, count);
 4286   }
 4287 }
 4288 
 4289 void MacroAssembler::rorq(Register dest, Register count) {
 4290   if (count == rcx) {
 4291     Assembler::rorq(dest);
 4292   } else {
 4293     assert_different_registers(rcx, dest);
 4294     xchgq(rcx, count);
 4295     Assembler::rorq(dest);
 4296     xchgq(rcx, count);
 4297   }
 4298 }
 4299 
 4300 // Return true: we succeeded in generating this code
 4301 //
 4302 // At runtime, return 0 in result if r_super_klass is a superclass of
 4303 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4304 // super_klass_slot of the class you're looking for. This is always
 4305 // the case for instanceof and checkcast.
 4306 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4307                                                          Register r_super_klass,
 4308                                                          Register temp1,
 4309                                                          Register temp2,
 4310                                                          Register temp3,
 4311                                                          Register temp4,
 4312                                                          Register result,
 4313                                                          u1 super_klass_slot) {
 4314   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4315 
 4316   Label L_fallthrough, L_success, L_failure;
 4317 
 4318   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4319 
 4320   const Register
 4321     r_array_index  = temp1,
 4322     r_array_length = temp2,
 4323     r_array_base   = temp3,
 4324     r_bitmap       = temp4;
 4325 
 4326   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4327 
 4328   xorq(result, result); // = 0
 4329 
 4330   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4331   movq(r_array_index, r_bitmap);
 4332 
 4333   // First check the bitmap to see if super_klass might be present. If
 4334   // the bit is zero, we are certain that super_klass is not one of
 4335   // the secondary supers.
 4336   u1 bit = super_klass_slot;
 4337   {
 4338     // NB: If the count in a x86 shift instruction is 0, the flags are
 4339     // not affected, so we do a testq instead.
 4340     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4341     if (shift_count != 0) {
 4342       salq(r_array_index, shift_count);
 4343     } else {
 4344       testq(r_array_index, r_array_index);
 4345     }
 4346   }
 4347   // We test the MSB of r_array_index, i.e. its sign bit
 4348   jcc(Assembler::positive, L_failure);
 4349 
 4350   // Get the first array index that can contain super_klass into r_array_index.
 4351   if (bit != 0) {
 4352     population_count(r_array_index, r_array_index, temp2, temp3);
 4353   } else {
 4354     movl(r_array_index, 1);
 4355   }
 4356   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4357 
 4358   // We will consult the secondary-super array.
 4359   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4360 
 4361   // We're asserting that the first word in an Array<Klass*> is the
 4362   // length, and the second word is the first word of the data. If
 4363   // that ever changes, r_array_base will have to be adjusted here.
 4364   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4365   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4366 
 4367   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4368   jccb(Assembler::equal, L_success);
 4369 
 4370   // Is there another entry to check? Consult the bitmap.
 4371   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4372   jccb(Assembler::carryClear, L_failure);
 4373 
 4374   // Linear probe. Rotate the bitmap so that the next bit to test is
 4375   // in Bit 1.
 4376   if (bit != 0) {
 4377     rorq(r_bitmap, bit);
 4378   }
 4379 
 4380   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4381   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4382   // Kills: r_array_length.
 4383   // Returns: result.
 4384   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4385   // Result (0/1) is in rdi
 4386   jmpb(L_fallthrough);
 4387 
 4388   bind(L_failure);
 4389   incq(result); // 0 => 1
 4390 
 4391   bind(L_success);
 4392   // result = 0;
 4393 
 4394   bind(L_fallthrough);
 4395   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4396 
 4397   if (VerifySecondarySupers) {
 4398     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4399                                   temp1, temp2, temp3);
 4400   }
 4401 }
 4402 
 4403 // At runtime, return 0 in result if r_super_klass is a superclass of
 4404 // r_sub_klass, otherwise return nonzero. Use this version of
 4405 // lookup_secondary_supers_table() if you don't know ahead of time
 4406 // which superclass will be searched for. Used by interpreter and
 4407 // runtime stubs. It is larger and has somewhat greater latency than
 4408 // the version above, which takes a constant super_klass_slot.
 4409 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4410                                                        Register r_super_klass,
 4411                                                        Register temp1,
 4412                                                        Register temp2,
 4413                                                        Register temp3,
 4414                                                        Register temp4,
 4415                                                        Register result) {
 4416   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4417   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4418   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4419 
 4420   Label L_fallthrough, L_success, L_failure;
 4421 
 4422   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4423 
 4424   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4425 
 4426   // FIXME. Once we are sure that all paths reaching this point really
 4427   // do pass rcx as one of our temps we can get rid of the following
 4428   // workaround.
 4429   assert(temps.contains(rcx), "fix this code");
 4430 
 4431   // We prefer to have our shift count in rcx. If rcx is one of our
 4432   // temps, use it for slot. If not, pick any of our temps.
 4433   Register slot;
 4434   if (!temps.contains(rcx)) {
 4435     slot = *available_regs++;
 4436   } else {
 4437     slot = rcx;
 4438   }
 4439 
 4440   const Register r_array_index = *available_regs++;
 4441   const Register r_bitmap      = *available_regs++;
 4442 
 4443   // The logic above guarantees this property, but we state it here.
 4444   assert_different_registers(r_array_index, r_bitmap, rcx);
 4445 
 4446   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4447   movq(r_array_index, r_bitmap);
 4448 
 4449   // First check the bitmap to see if super_klass might be present. If
 4450   // the bit is zero, we are certain that super_klass is not one of
 4451   // the secondary supers.
 4452   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4453   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4454   salq(r_array_index, slot);
 4455 
 4456   testq(r_array_index, r_array_index);
 4457   // We test the MSB of r_array_index, i.e. its sign bit
 4458   jcc(Assembler::positive, L_failure);
 4459 
 4460   const Register r_array_base = *available_regs++;
 4461 
 4462   // Get the first array index that can contain super_klass into r_array_index.
 4463   // Note: Clobbers r_array_base and slot.
 4464   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4465 
 4466   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4467 
 4468   // We will consult the secondary-super array.
 4469   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4470 
 4471   // We're asserting that the first word in an Array<Klass*> is the
 4472   // length, and the second word is the first word of the data. If
 4473   // that ever changes, r_array_base will have to be adjusted here.
 4474   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4475   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4476 
 4477   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4478   jccb(Assembler::equal, L_success);
 4479 
 4480   // Restore slot to its true value
 4481   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4482 
 4483   // Linear probe. Rotate the bitmap so that the next bit to test is
 4484   // in Bit 1.
 4485   rorq(r_bitmap, slot);
 4486 
 4487   // Is there another entry to check? Consult the bitmap.
 4488   btq(r_bitmap, 1);
 4489   jccb(Assembler::carryClear, L_failure);
 4490 
 4491   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4492   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4493   // Kills: r_array_length.
 4494   // Returns: result.
 4495   lookup_secondary_supers_table_slow_path(r_super_klass,
 4496                                           r_array_base,
 4497                                           r_array_index,
 4498                                           r_bitmap,
 4499                                           /*temp1*/result,
 4500                                           /*temp2*/slot,
 4501                                           &L_success,
 4502                                           nullptr);
 4503 
 4504   bind(L_failure);
 4505   movq(result, 1);
 4506   jmpb(L_fallthrough);
 4507 
 4508   bind(L_success);
 4509   xorq(result, result); // = 0
 4510 
 4511   bind(L_fallthrough);
 4512   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4513 
 4514   if (VerifySecondarySupers) {
 4515     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4516                                   temp1, temp2, temp3);
 4517   }
 4518 }
 4519 
 4520 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4521                                  Label* L_success, Label* L_failure) {
 4522   Label L_loop, L_fallthrough;
 4523   {
 4524     int label_nulls = 0;
 4525     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4526     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4527     assert(label_nulls <= 1, "at most one null in the batch");
 4528   }
 4529   bind(L_loop);
 4530   cmpq(value, Address(addr, count, Address::times_8));
 4531   jcc(Assembler::equal, *L_success);
 4532   addl(count, 1);
 4533   cmpl(count, limit);
 4534   jcc(Assembler::less, L_loop);
 4535 
 4536   if (&L_fallthrough != L_failure) {
 4537     jmp(*L_failure);
 4538   }
 4539   bind(L_fallthrough);
 4540 }
 4541 
 4542 // Called by code generated by check_klass_subtype_slow_path
 4543 // above. This is called when there is a collision in the hashed
 4544 // lookup in the secondary supers array.
 4545 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4546                                                              Register r_array_base,
 4547                                                              Register r_array_index,
 4548                                                              Register r_bitmap,
 4549                                                              Register temp1,
 4550                                                              Register temp2,
 4551                                                              Label* L_success,
 4552                                                              Label* L_failure) {
 4553   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4554 
 4555   const Register
 4556     r_array_length = temp1,
 4557     r_sub_klass    = noreg,
 4558     result         = noreg;
 4559 
 4560   Label L_fallthrough;
 4561   int label_nulls = 0;
 4562   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4563   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4564   assert(label_nulls <= 1, "at most one null in the batch");
 4565 
 4566   // Load the array length.
 4567   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4568   // And adjust the array base to point to the data.
 4569   // NB! Effectively increments current slot index by 1.
 4570   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4571   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4572 
 4573   // Linear probe
 4574   Label L_huge;
 4575 
 4576   // The bitmap is full to bursting.
 4577   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4578   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4579   jcc(Assembler::greater, L_huge);
 4580 
 4581   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4582   // current slot (at secondary_supers[r_array_index]) has not yet
 4583   // been inspected, and r_array_index may be out of bounds if we
 4584   // wrapped around the end of the array.
 4585 
 4586   { // This is conventional linear probing, but instead of terminating
 4587     // when a null entry is found in the table, we maintain a bitmap
 4588     // in which a 0 indicates missing entries.
 4589     // The check above guarantees there are 0s in the bitmap, so the loop
 4590     // eventually terminates.
 4591 
 4592     xorl(temp2, temp2); // = 0;
 4593 
 4594     Label L_again;
 4595     bind(L_again);
 4596 
 4597     // Check for array wraparound.
 4598     cmpl(r_array_index, r_array_length);
 4599     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4600 
 4601     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4602     jcc(Assembler::equal, *L_success);
 4603 
 4604     // If the next bit in bitmap is zero, we're done.
 4605     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4606     jcc(Assembler::carryClear, *L_failure);
 4607 
 4608     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4609     addl(r_array_index, 1);
 4610 
 4611     jmp(L_again);
 4612   }
 4613 
 4614   { // Degenerate case: more than 64 secondary supers.
 4615     // FIXME: We could do something smarter here, maybe a vectorized
 4616     // comparison or a binary search, but is that worth any added
 4617     // complexity?
 4618     bind(L_huge);
 4619     xorl(r_array_index, r_array_index); // = 0
 4620     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4621                 L_success,
 4622                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4623 
 4624     bind(L_fallthrough);
 4625   }
 4626 }
 4627 
 4628 struct VerifyHelperArguments {
 4629   Klass* _super;
 4630   Klass* _sub;
 4631   intptr_t _linear_result;
 4632   intptr_t _table_result;
 4633 };
 4634 
 4635 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4636   Klass::on_secondary_supers_verification_failure(args->_super,
 4637                                                   args->_sub,
 4638                                                   args->_linear_result,
 4639                                                   args->_table_result,
 4640                                                   msg);
 4641 }
 4642 
 4643 // Make sure that the hashed lookup and a linear scan agree.
 4644 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4645                                                    Register r_super_klass,
 4646                                                    Register result,
 4647                                                    Register temp1,
 4648                                                    Register temp2,
 4649                                                    Register temp3) {
 4650   const Register
 4651       r_array_index  = temp1,
 4652       r_array_length = temp2,
 4653       r_array_base   = temp3,
 4654       r_bitmap       = noreg;
 4655 
 4656   BLOCK_COMMENT("verify_secondary_supers_table {");
 4657 
 4658   Label L_success, L_failure, L_check, L_done;
 4659 
 4660   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4661   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4662   // And adjust the array base to point to the data.
 4663   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4664 
 4665   testl(r_array_length, r_array_length); // array_length == 0?
 4666   jcc(Assembler::zero, L_failure);
 4667 
 4668   movl(r_array_index, 0);
 4669   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4670   // fall through to L_failure
 4671 
 4672   const Register linear_result = r_array_index; // reuse temp1
 4673 
 4674   bind(L_failure); // not present
 4675   movl(linear_result, 1);
 4676   jmp(L_check);
 4677 
 4678   bind(L_success); // present
 4679   movl(linear_result, 0);
 4680 
 4681   bind(L_check);
 4682   cmpl(linear_result, result);
 4683   jcc(Assembler::equal, L_done);
 4684 
 4685   { // To avoid calling convention issues, build a record on the stack
 4686     // and pass the pointer to that instead.
 4687     push(result);
 4688     push(linear_result);
 4689     push(r_sub_klass);
 4690     push(r_super_klass);
 4691     movptr(c_rarg1, rsp);
 4692     movptr(c_rarg0, (uintptr_t) "mismatch");
 4693     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4694     should_not_reach_here();
 4695   }
 4696   bind(L_done);
 4697 
 4698   BLOCK_COMMENT("} verify_secondary_supers_table");
 4699 }
 4700 
 4701 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4702 
 4703 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4704   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4705 
 4706   Label L_fallthrough;
 4707   if (L_fast_path == nullptr) {
 4708     L_fast_path = &L_fallthrough;
 4709   } else if (L_slow_path == nullptr) {
 4710     L_slow_path = &L_fallthrough;
 4711   }
 4712 
 4713   // Fast path check: class is fully initialized.
 4714   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4715   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4716   jcc(Assembler::equal, *L_fast_path);
 4717 
 4718   // Fast path check: current thread is initializer thread
 4719   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4720   if (L_slow_path == &L_fallthrough) {
 4721     jcc(Assembler::equal, *L_fast_path);
 4722     bind(*L_slow_path);
 4723   } else if (L_fast_path == &L_fallthrough) {
 4724     jcc(Assembler::notEqual, *L_slow_path);
 4725     bind(*L_fast_path);
 4726   } else {
 4727     Unimplemented();
 4728   }
 4729 }
 4730 
 4731 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4732   if (VM_Version::supports_cmov()) {
 4733     cmovl(cc, dst, src);
 4734   } else {
 4735     Label L;
 4736     jccb(negate_condition(cc), L);
 4737     movl(dst, src);
 4738     bind(L);
 4739   }
 4740 }
 4741 
 4742 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4743   if (VM_Version::supports_cmov()) {
 4744     cmovl(cc, dst, src);
 4745   } else {
 4746     Label L;
 4747     jccb(negate_condition(cc), L);
 4748     movl(dst, src);
 4749     bind(L);
 4750   }
 4751 }
 4752 
 4753 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4754   if (!VerifyOops) return;




 4755 
 4756   BLOCK_COMMENT("verify_oop {");
 4757   push(rscratch1);
 4758   push(rax);                          // save rax
 4759   push(reg);                          // pass register argument
 4760 
 4761   // Pass register number to verify_oop_subroutine
 4762   const char* b = nullptr;
 4763   {
 4764     ResourceMark rm;
 4765     stringStream ss;
 4766     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4767     b = code_string(ss.as_string());
 4768   }
 4769   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4770   pushptr(buffer.addr(), rscratch1);
 4771 
 4772   // call indirectly to solve generation ordering problem
 4773   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4774   call(rax);
 4775   // Caller pops the arguments (oop, message) and restores rax, r10
 4776   BLOCK_COMMENT("} verify_oop");
 4777 }
 4778 
 4779 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 4780   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 4781     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 4782     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 4783     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 4784   } else if (VM_Version::supports_avx()) {
 4785     vpcmpeqd(dst, dst, dst, vector_len);
 4786   } else {
 4787     pcmpeqd(dst, dst);
 4788   }
 4789 }
 4790 
 4791 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 4792                                          int extra_slot_offset) {
 4793   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 4794   int stackElementSize = Interpreter::stackElementSize;
 4795   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 4796 #ifdef ASSERT
 4797   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 4798   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 4799 #endif
 4800   Register             scale_reg    = noreg;
 4801   Address::ScaleFactor scale_factor = Address::no_scale;
 4802   if (arg_slot.is_constant()) {
 4803     offset += arg_slot.as_constant() * stackElementSize;
 4804   } else {
 4805     scale_reg    = arg_slot.as_register();
 4806     scale_factor = Address::times(stackElementSize);
 4807   }
 4808   offset += wordSize;           // return PC is on stack
 4809   return Address(rsp, scale_reg, scale_factor, offset);
 4810 }
 4811 
 4812 // Handle the receiver type profile update given the "recv" klass.
 4813 //
 4814 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 4815 // If there are no matching or claimable receiver entries in RD, updates
 4816 // the polymorphic counter.
 4817 //
 4818 // This code expected to run by either the interpreter or JIT-ed code, without
 4819 // extra synchronization. For safety, receiver cells are claimed atomically, which
 4820 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 4821 // counter updates are not atomic.
 4822 //
 4823 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 4824   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 4825   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 4826   int poly_count_offset      = in_bytes(CounterData::count_offset());
 4827   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 4828   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 4829 
 4830   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 4831   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 4832   base_receiver_offset += mdp_offset;
 4833   end_receiver_offset  += mdp_offset;
 4834   poly_count_offset    += mdp_offset;
 4835 
 4836   // Scale down to optimize encoding. Slots are pointer-sized.
 4837   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 4838   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 4839   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 4840   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 4841   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 4842   base_receiver_offset   >>= LogBytesPerWord;
 4843   end_receiver_offset    >>= LogBytesPerWord;
 4844   poly_count_offset      >>= LogBytesPerWord;
 4845   receiver_step          >>= LogBytesPerWord;
 4846   receiver_to_count_step >>= LogBytesPerWord;
 4847 
 4848 #ifdef ASSERT
 4849   // We are about to walk the MDO slots without asking for offsets.
 4850   // Check that our math hits all the right spots.
 4851   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 4852     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 4853     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 4854     int offset = base_receiver_offset + receiver_step*c;
 4855     int count_offset = offset + receiver_to_count_step;
 4856     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 4857     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 4858   }
 4859   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 4860   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 4861 #endif
 4862 
 4863   // Corner case: no profile table. Increment poly counter and exit.
 4864   if (ReceiverTypeData::row_limit() == 0) {
 4865     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 4866     return;
 4867   }
 4868 
 4869   Register offset = rscratch1;
 4870 
 4871   Label L_loop_search_receiver, L_loop_search_empty;
 4872   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 4873 
 4874   // The code here recognizes three major cases:
 4875   //   A. Fastest: receiver found in the table
 4876   //   B. Fast: no receiver in the table, and the table is full
 4877   //   C. Slow: no receiver in the table, free slots in the table
 4878   //
 4879   // The case A performance is most important, as perfectly-behaved code would end up
 4880   // there, especially with larger TypeProfileWidth. The case B performance is
 4881   // important as well, this is where bulk of code would land for normally megamorphic
 4882   // cases. The case C performance is not essential, its job is to deal with installation
 4883   // races, we optimize for code density instead. Case C needs to make sure that receiver
 4884   // rows are only claimed once. This makes sure we never overwrite a row for another
 4885   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 4886   //
 4887   // It is very tempting to handle these cases in a single loop, and claim the first slot
 4888   // without checking the rest of the table. But, profiling code should tolerate free slots
 4889   // in the table, as class unloading can clear them. After such cleanup, the receiver
 4890   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 4891   // to complete, before trying to install new slots. Splitting the code in several tight
 4892   // loops also helpfully optimizes for cases A and B.
 4893   //
 4894   // This code is effectively:
 4895   //
 4896   // restart:
 4897   //   // Fastest: receiver is already installed
 4898   //   for (i = 0; i < receiver_count(); i++) {
 4899   //     if (receiver(i) == recv) goto found_recv(i);
 4900   //   }
 4901   //
 4902   //   // Fast: no receiver, but profile is full
 4903   //   for (i = 0; i < receiver_count(); i++) {
 4904   //     if (receiver(i) == null) goto found_null(i);
 4905   //   }
 4906   //   goto polymorphic
 4907   //
 4908   //   // Slow: try to install receiver
 4909   // found_null(i):
 4910   //   CAS(&receiver(i), null, recv);
 4911   //   goto restart
 4912   //
 4913   // polymorphic:
 4914   //   count++;
 4915   //   return
 4916   //
 4917   // found_recv(i):
 4918   //   *receiver_count(i)++
 4919   //
 4920 
 4921   bind(L_restart);
 4922 
 4923   // Fastest: receiver is already installed
 4924   movptr(offset, base_receiver_offset);
 4925   bind(L_loop_search_receiver);
 4926     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 4927     jccb(Assembler::equal, L_found_recv);
 4928   addptr(offset, receiver_step);
 4929   cmpptr(offset, end_receiver_offset);
 4930   jccb(Assembler::notEqual, L_loop_search_receiver);
 4931 
 4932   // Fast: no receiver, but profile is full
 4933   movptr(offset, base_receiver_offset);
 4934   bind(L_loop_search_empty);
 4935     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 4936     jccb(Assembler::equal, L_found_empty);
 4937   addptr(offset, receiver_step);
 4938   cmpptr(offset, end_receiver_offset);
 4939   jccb(Assembler::notEqual, L_loop_search_empty);
 4940   jmpb(L_polymorphic);
 4941 
 4942   // Slow: try to install receiver
 4943   bind(L_found_empty);
 4944 
 4945   // Atomically swing receiver slot: null -> recv.
 4946   //
 4947   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 4948   // other important registers untouched, as they form the address. Therefore, we need
 4949   // to shift any important registers from RAX into some other spare register. If we
 4950   // have a spare register, we are forced to save it on stack here.
 4951 
 4952   Register spare_reg = noreg;
 4953   Register shifted_mdp = mdp;
 4954   Register shifted_recv = recv;
 4955   if (recv == rax || mdp == rax) {
 4956     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 4957                 (recv != rcx && mdp != rcx) ? rcx :
 4958                 rdx;
 4959     assert_different_registers(mdp, recv, offset, spare_reg);
 4960 
 4961     push(spare_reg);
 4962     if (recv == rax) {
 4963       movptr(spare_reg, recv);
 4964       shifted_recv = spare_reg;
 4965     } else {
 4966       assert(mdp == rax, "Remaining case");
 4967       movptr(spare_reg, mdp);
 4968       shifted_mdp = spare_reg;
 4969     }
 4970   } else {
 4971     push(rax);
 4972   }
 4973 
 4974   // None of the important registers are in RAX after this shuffle.
 4975   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 4976 
 4977   xorptr(rax, rax);
 4978   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 4979 
 4980   // Unshift registers.
 4981   if (recv == rax || mdp == rax) {
 4982     movptr(rax, spare_reg);
 4983     pop(spare_reg);
 4984   } else {
 4985     pop(rax);
 4986   }
 4987 
 4988   // CAS success means the slot now has the receiver we want. CAS failure means
 4989   // something had claimed the slot concurrently: it can be the same receiver we want,
 4990   // or something else. Since this is a slow path, we can optimize for code density,
 4991   // and just restart the search from the beginning.
 4992   jmpb(L_restart);
 4993 
 4994   // Counter updates:
 4995 
 4996   // Increment polymorphic counter instead of receiver slot.
 4997   bind(L_polymorphic);
 4998   movptr(offset, poly_count_offset);
 4999   jmpb(L_count_update);
 5000 
 5001   // Found a receiver, convert its slot offset to corresponding count offset.
 5002   bind(L_found_recv);
 5003   addptr(offset, receiver_to_count_step);
 5004 
 5005   bind(L_count_update);
 5006   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 5007 }
 5008 
 5009 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5010   if (!VerifyOops) return;




 5011 
 5012   push(rscratch1);
 5013   push(rax); // save rax,
 5014   // addr may contain rsp so we will have to adjust it based on the push
 5015   // we just did (and on 64 bit we do two pushes)
 5016   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5017   // stores rax into addr which is backwards of what was intended.
 5018   if (addr.uses(rsp)) {
 5019     lea(rax, addr);
 5020     pushptr(Address(rax, 2 * BytesPerWord));
 5021   } else {
 5022     pushptr(addr);
 5023   }
 5024 
 5025   // Pass register number to verify_oop_subroutine
 5026   const char* b = nullptr;
 5027   {
 5028     ResourceMark rm;
 5029     stringStream ss;
 5030     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 5031     b = code_string(ss.as_string());
 5032   }
 5033   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5034   pushptr(buffer.addr(), rscratch1);
 5035 
 5036   // call indirectly to solve generation ordering problem
 5037   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5038   call(rax);
 5039   // Caller pops the arguments (addr, message) and restores rax, r10.
 5040 }
 5041 
 5042 void MacroAssembler::verify_tlab() {
 5043 #ifdef ASSERT
 5044   if (UseTLAB && VerifyOops) {
 5045     Label next, ok;
 5046     Register t1 = rsi;
 5047 
 5048     push(t1);
 5049 
 5050     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5051     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 5052     jcc(Assembler::aboveEqual, next);
 5053     STOP("assert(top >= start)");
 5054     should_not_reach_here();
 5055 
 5056     bind(next);
 5057     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 5058     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5059     jcc(Assembler::aboveEqual, ok);
 5060     STOP("assert(top <= end)");
 5061     should_not_reach_here();
 5062 
 5063     bind(ok);
 5064     pop(t1);
 5065   }
 5066 #endif
 5067 }
 5068 
 5069 class ControlWord {
 5070  public:
 5071   int32_t _value;
 5072 
 5073   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5074   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5075   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5076   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5077   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5078   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5079   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5080   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5081 
 5082   void print() const {
 5083     // rounding control
 5084     const char* rc;
 5085     switch (rounding_control()) {
 5086       case 0: rc = "round near"; break;
 5087       case 1: rc = "round down"; break;
 5088       case 2: rc = "round up  "; break;
 5089       case 3: rc = "chop      "; break;
 5090       default:
 5091         rc = nullptr; // silence compiler warnings
 5092         fatal("Unknown rounding control: %d", rounding_control());
 5093     };
 5094     // precision control
 5095     const char* pc;
 5096     switch (precision_control()) {
 5097       case 0: pc = "24 bits "; break;
 5098       case 1: pc = "reserved"; break;
 5099       case 2: pc = "53 bits "; break;
 5100       case 3: pc = "64 bits "; break;
 5101       default:
 5102         pc = nullptr; // silence compiler warnings
 5103         fatal("Unknown precision control: %d", precision_control());
 5104     };
 5105     // flags
 5106     char f[9];
 5107     f[0] = ' ';
 5108     f[1] = ' ';
 5109     f[2] = (precision   ()) ? 'P' : 'p';
 5110     f[3] = (underflow   ()) ? 'U' : 'u';
 5111     f[4] = (overflow    ()) ? 'O' : 'o';
 5112     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5113     f[6] = (denormalized()) ? 'D' : 'd';
 5114     f[7] = (invalid     ()) ? 'I' : 'i';
 5115     f[8] = '\x0';
 5116     // output
 5117     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5118   }
 5119 
 5120 };
 5121 
 5122 class StatusWord {
 5123  public:
 5124   int32_t _value;
 5125 
 5126   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5127   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5128   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5129   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5130   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5131   int  top() const                     { return  (_value >> 11) & 7      ; }
 5132   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5133   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5134   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5135   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5136   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5137   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5138   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5139   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5140 
 5141   void print() const {
 5142     // condition codes
 5143     char c[5];
 5144     c[0] = (C3()) ? '3' : '-';
 5145     c[1] = (C2()) ? '2' : '-';
 5146     c[2] = (C1()) ? '1' : '-';
 5147     c[3] = (C0()) ? '0' : '-';
 5148     c[4] = '\x0';
 5149     // flags
 5150     char f[9];
 5151     f[0] = (error_status()) ? 'E' : '-';
 5152     f[1] = (stack_fault ()) ? 'S' : '-';
 5153     f[2] = (precision   ()) ? 'P' : '-';
 5154     f[3] = (underflow   ()) ? 'U' : '-';
 5155     f[4] = (overflow    ()) ? 'O' : '-';
 5156     f[5] = (zero_divide ()) ? 'Z' : '-';
 5157     f[6] = (denormalized()) ? 'D' : '-';
 5158     f[7] = (invalid     ()) ? 'I' : '-';
 5159     f[8] = '\x0';
 5160     // output
 5161     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5162   }
 5163 
 5164 };
 5165 
 5166 class TagWord {
 5167  public:
 5168   int32_t _value;
 5169 
 5170   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5171 
 5172   void print() const {
 5173     printf("%04x", _value & 0xFFFF);
 5174   }
 5175 
 5176 };
 5177 
 5178 class FPU_Register {
 5179  public:
 5180   int32_t _m0;
 5181   int32_t _m1;
 5182   int16_t _ex;
 5183 
 5184   bool is_indefinite() const           {
 5185     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5186   }
 5187 
 5188   void print() const {
 5189     char  sign = (_ex < 0) ? '-' : '+';
 5190     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5191     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5192   };
 5193 
 5194 };
 5195 
 5196 class FPU_State {
 5197  public:
 5198   enum {
 5199     register_size       = 10,
 5200     number_of_registers =  8,
 5201     register_mask       =  7
 5202   };
 5203 
 5204   ControlWord  _control_word;
 5205   StatusWord   _status_word;
 5206   TagWord      _tag_word;
 5207   int32_t      _error_offset;
 5208   int32_t      _error_selector;
 5209   int32_t      _data_offset;
 5210   int32_t      _data_selector;
 5211   int8_t       _register[register_size * number_of_registers];
 5212 
 5213   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5214   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5215 
 5216   const char* tag_as_string(int tag) const {
 5217     switch (tag) {
 5218       case 0: return "valid";
 5219       case 1: return "zero";
 5220       case 2: return "special";
 5221       case 3: return "empty";
 5222     }
 5223     ShouldNotReachHere();
 5224     return nullptr;
 5225   }
 5226 
 5227   void print() const {
 5228     // print computation registers
 5229     { int t = _status_word.top();
 5230       for (int i = 0; i < number_of_registers; i++) {
 5231         int j = (i - t) & register_mask;
 5232         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5233         st(j)->print();
 5234         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5235       }
 5236     }
 5237     printf("\n");
 5238     // print control registers
 5239     printf("ctrl = "); _control_word.print(); printf("\n");
 5240     printf("stat = "); _status_word .print(); printf("\n");
 5241     printf("tags = "); _tag_word    .print(); printf("\n");
 5242   }
 5243 
 5244 };
 5245 
 5246 class Flag_Register {
 5247  public:
 5248   int32_t _value;
 5249 
 5250   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5251   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5252   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5253   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5254   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5255   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5256   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5257 
 5258   void print() const {
 5259     // flags
 5260     char f[8];
 5261     f[0] = (overflow       ()) ? 'O' : '-';
 5262     f[1] = (direction      ()) ? 'D' : '-';
 5263     f[2] = (sign           ()) ? 'S' : '-';
 5264     f[3] = (zero           ()) ? 'Z' : '-';
 5265     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5266     f[5] = (parity         ()) ? 'P' : '-';
 5267     f[6] = (carry          ()) ? 'C' : '-';
 5268     f[7] = '\x0';
 5269     // output
 5270     printf("%08x  flags = %s", _value, f);
 5271   }
 5272 
 5273 };
 5274 
 5275 class IU_Register {
 5276  public:
 5277   int32_t _value;
 5278 
 5279   void print() const {
 5280     printf("%08x  %11d", _value, _value);
 5281   }
 5282 
 5283 };
 5284 
 5285 class IU_State {
 5286  public:
 5287   Flag_Register _eflags;
 5288   IU_Register   _rdi;
 5289   IU_Register   _rsi;
 5290   IU_Register   _rbp;
 5291   IU_Register   _rsp;
 5292   IU_Register   _rbx;
 5293   IU_Register   _rdx;
 5294   IU_Register   _rcx;
 5295   IU_Register   _rax;
 5296 
 5297   void print() const {
 5298     // computation registers
 5299     printf("rax,  = "); _rax.print(); printf("\n");
 5300     printf("rbx,  = "); _rbx.print(); printf("\n");
 5301     printf("rcx  = "); _rcx.print(); printf("\n");
 5302     printf("rdx  = "); _rdx.print(); printf("\n");
 5303     printf("rdi  = "); _rdi.print(); printf("\n");
 5304     printf("rsi  = "); _rsi.print(); printf("\n");
 5305     printf("rbp,  = "); _rbp.print(); printf("\n");
 5306     printf("rsp  = "); _rsp.print(); printf("\n");
 5307     printf("\n");
 5308     // control registers
 5309     printf("flgs = "); _eflags.print(); printf("\n");
 5310   }
 5311 };
 5312 
 5313 
 5314 class CPU_State {
 5315  public:
 5316   FPU_State _fpu_state;
 5317   IU_State  _iu_state;
 5318 
 5319   void print() const {
 5320     printf("--------------------------------------------------\n");
 5321     _iu_state .print();
 5322     printf("\n");
 5323     _fpu_state.print();
 5324     printf("--------------------------------------------------\n");
 5325   }
 5326 
 5327 };
 5328 
 5329 
 5330 static void _print_CPU_state(CPU_State* state) {
 5331   state->print();
 5332 };
 5333 
 5334 
 5335 void MacroAssembler::print_CPU_state() {
 5336   push_CPU_state();
 5337   push(rsp);                // pass CPU state
 5338   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5339   addptr(rsp, wordSize);       // discard argument
 5340   pop_CPU_state();
 5341 }
 5342 
 5343 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5344   // Either restore the MXCSR register after returning from the JNI Call
 5345   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5346   if (VM_Version::supports_sse()) {
 5347     if (RestoreMXCSROnJNICalls) {
 5348       ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5349     } else if (CheckJNICalls) {
 5350       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5351     }
 5352   }
 5353   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5354   vzeroupper();
 5355 }
 5356 
 5357 // ((OopHandle)result).resolve();
 5358 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5359   assert_different_registers(result, tmp);
 5360 
 5361   // Only 64 bit platforms support GCs that require a tmp register
 5362   // Only IN_HEAP loads require a thread_tmp register
 5363   // OopHandle::resolve is an indirection like jobject.
 5364   access_load_at(T_OBJECT, IN_NATIVE,
 5365                  result, Address(result, 0), tmp);
 5366 }
 5367 
 5368 // ((WeakHandle)result).resolve();
 5369 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5370   assert_different_registers(rresult, rtmp);
 5371   Label resolved;
 5372 
 5373   // A null weak handle resolves to null.
 5374   cmpptr(rresult, 0);
 5375   jcc(Assembler::equal, resolved);
 5376 
 5377   // Only 64 bit platforms support GCs that require a tmp register
 5378   // Only IN_HEAP loads require a thread_tmp register
 5379   // WeakHandle::resolve is an indirection like jweak.
 5380   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5381                  rresult, Address(rresult, 0), rtmp);
 5382   bind(resolved);
 5383 }
 5384 
 5385 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5386   // get mirror
 5387   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5388   load_method_holder(mirror, method);
 5389   movptr(mirror, Address(mirror, mirror_offset));
 5390   resolve_oop_handle(mirror, tmp);
 5391 }
 5392 
 5393 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5394   load_method_holder(rresult, rmethod);
 5395   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5396 }
 5397 
 5398 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5399   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5400   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5401   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5402 }
 5403 








 5404 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5405   assert(UseCompactObjectHeaders, "expect compact object headers");
 5406   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5407   shrq(dst, markWord::klass_shift);
 5408 }
 5409 
 5410 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5411   assert_different_registers(src, tmp);
 5412   assert_different_registers(dst, tmp);
 5413 
 5414   if (UseCompactObjectHeaders) {
 5415     load_narrow_klass_compact(dst, src);
 5416     decode_klass_not_null(dst, tmp);
 5417   } else {
 5418     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5419     decode_klass_not_null(dst, tmp);
 5420   }
 5421 }
 5422 





 5423 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5424   assert(!UseCompactObjectHeaders, "not with compact headers");
 5425   assert_different_registers(src, tmp);
 5426   assert_different_registers(dst, tmp);
 5427   encode_klass_not_null(src, tmp);
 5428   movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5429 }
 5430 
 5431 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5432   if (UseCompactObjectHeaders) {
 5433     assert(tmp != noreg, "need tmp");
 5434     assert_different_registers(klass, obj, tmp);
 5435     load_narrow_klass_compact(tmp, obj);
 5436     cmpl(klass, tmp);
 5437   } else {
 5438     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5439   }
 5440 }
 5441 
 5442 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5443   if (UseCompactObjectHeaders) {
 5444     assert(tmp2 != noreg, "need tmp2");
 5445     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5446     load_narrow_klass_compact(tmp1, obj1);
 5447     load_narrow_klass_compact(tmp2, obj2);
 5448     cmpl(tmp1, tmp2);
 5449   } else {
 5450     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5451     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5452   }
 5453 }
 5454 
 5455 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5456                                     Register tmp1) {
 5457   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5458   decorators = AccessInternal::decorator_fixup(decorators, type);
 5459   bool as_raw = (decorators & AS_RAW) != 0;
 5460   if (as_raw) {
 5461     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5462   } else {
 5463     bs->load_at(this, decorators, type, dst, src, tmp1);
 5464   }
 5465 }
 5466 
 5467 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5468                                      Register tmp1, Register tmp2, Register tmp3) {
 5469   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5470   decorators = AccessInternal::decorator_fixup(decorators, type);
 5471   bool as_raw = (decorators & AS_RAW) != 0;
 5472   if (as_raw) {
 5473     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5474   } else {
 5475     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5476   }
 5477 }
 5478 








































 5479 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5480   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5481 }
 5482 
 5483 // Doesn't do verification, generates fixed size code
 5484 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5485   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5486 }
 5487 
 5488 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5489                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5490   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5491 }
 5492 
 5493 // Used for storing nulls.
 5494 void MacroAssembler::store_heap_oop_null(Address dst) {
 5495   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5496 }
 5497 
 5498 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5499   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5500   // Store to klass gap in destination
 5501   movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5502 }
 5503 
 5504 #ifdef ASSERT
 5505 void MacroAssembler::verify_heapbase(const char* msg) {
 5506   assert (UseCompressedOops, "should be compressed");
 5507   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5508   if (CheckCompressedOops) {
 5509     Label ok;
 5510     ExternalAddress src2(CompressedOops::base_addr());
 5511     const bool is_src2_reachable = reachable(src2);
 5512     if (!is_src2_reachable) {
 5513       push(rscratch1);  // cmpptr trashes rscratch1
 5514     }
 5515     cmpptr(r12_heapbase, src2, rscratch1);
 5516     jcc(Assembler::equal, ok);
 5517     STOP(msg);
 5518     bind(ok);
 5519     if (!is_src2_reachable) {
 5520       pop(rscratch1);
 5521     }
 5522   }
 5523 }
 5524 #endif
 5525 
 5526 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5527 void MacroAssembler::encode_heap_oop(Register r) {
 5528 #ifdef ASSERT
 5529   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5530 #endif
 5531   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5532   if (CompressedOops::base() == nullptr) {
 5533     if (CompressedOops::shift() != 0) {
 5534       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5535       shrq(r, LogMinObjAlignmentInBytes);
 5536     }
 5537     return;
 5538   }
 5539   testq(r, r);
 5540   cmovq(Assembler::equal, r, r12_heapbase);
 5541   subq(r, r12_heapbase);
 5542   shrq(r, LogMinObjAlignmentInBytes);
 5543 }
 5544 
 5545 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5546 #ifdef ASSERT
 5547   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5548   if (CheckCompressedOops) {
 5549     Label ok;
 5550     testq(r, r);
 5551     jcc(Assembler::notEqual, ok);
 5552     STOP("null oop passed to encode_heap_oop_not_null");
 5553     bind(ok);
 5554   }
 5555 #endif
 5556   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5557   if (CompressedOops::base() != nullptr) {
 5558     subq(r, r12_heapbase);
 5559   }
 5560   if (CompressedOops::shift() != 0) {
 5561     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5562     shrq(r, LogMinObjAlignmentInBytes);
 5563   }
 5564 }
 5565 
 5566 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5567 #ifdef ASSERT
 5568   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5569   if (CheckCompressedOops) {
 5570     Label ok;
 5571     testq(src, src);
 5572     jcc(Assembler::notEqual, ok);
 5573     STOP("null oop passed to encode_heap_oop_not_null2");
 5574     bind(ok);
 5575   }
 5576 #endif
 5577   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5578   if (dst != src) {
 5579     movq(dst, src);
 5580   }
 5581   if (CompressedOops::base() != nullptr) {
 5582     subq(dst, r12_heapbase);
 5583   }
 5584   if (CompressedOops::shift() != 0) {
 5585     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5586     shrq(dst, LogMinObjAlignmentInBytes);
 5587   }
 5588 }
 5589 
 5590 void  MacroAssembler::decode_heap_oop(Register r) {
 5591 #ifdef ASSERT
 5592   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5593 #endif
 5594   if (CompressedOops::base() == nullptr) {
 5595     if (CompressedOops::shift() != 0) {
 5596       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5597       shlq(r, LogMinObjAlignmentInBytes);
 5598     }
 5599   } else {
 5600     Label done;
 5601     shlq(r, LogMinObjAlignmentInBytes);
 5602     jccb(Assembler::equal, done);
 5603     addq(r, r12_heapbase);
 5604     bind(done);
 5605   }
 5606   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5607 }
 5608 
 5609 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5610   // Note: it will change flags
 5611   assert (UseCompressedOops, "should only be used for compressed headers");
 5612   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5613   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5614   // vtableStubs also counts instructions in pd_code_size_limit.
 5615   // Also do not verify_oop as this is called by verify_oop.
 5616   if (CompressedOops::shift() != 0) {
 5617     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5618     shlq(r, LogMinObjAlignmentInBytes);
 5619     if (CompressedOops::base() != nullptr) {
 5620       addq(r, r12_heapbase);
 5621     }
 5622   } else {
 5623     assert (CompressedOops::base() == nullptr, "sanity");
 5624   }
 5625 }
 5626 
 5627 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5628   // Note: it will change flags
 5629   assert (UseCompressedOops, "should only be used for compressed headers");
 5630   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5631   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5632   // vtableStubs also counts instructions in pd_code_size_limit.
 5633   // Also do not verify_oop as this is called by verify_oop.
 5634   if (CompressedOops::shift() != 0) {
 5635     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5636     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5637       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5638     } else {
 5639       if (dst != src) {
 5640         movq(dst, src);
 5641       }
 5642       shlq(dst, LogMinObjAlignmentInBytes);
 5643       if (CompressedOops::base() != nullptr) {
 5644         addq(dst, r12_heapbase);
 5645       }
 5646     }
 5647   } else {
 5648     assert (CompressedOops::base() == nullptr, "sanity");
 5649     if (dst != src) {
 5650       movq(dst, src);
 5651     }
 5652   }
 5653 }
 5654 
 5655 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5656   BLOCK_COMMENT("encode_klass_not_null {");
 5657   assert_different_registers(r, tmp);
 5658   if (CompressedKlassPointers::base() != nullptr) {
 5659     if (AOTCodeCache::is_on_for_dump()) {
 5660       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5661     } else {
 5662       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5663     }
 5664     subq(r, tmp);
 5665   }
 5666   if (CompressedKlassPointers::shift() != 0) {
 5667     shrq(r, CompressedKlassPointers::shift());
 5668   }
 5669   BLOCK_COMMENT("} encode_klass_not_null");
 5670 }
 5671 
 5672 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5673   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5674   assert_different_registers(src, dst);
 5675   if (CompressedKlassPointers::base() != nullptr) {
 5676     if (AOTCodeCache::is_on_for_dump()) {
 5677       movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5678       negq(dst);
 5679     } else {
 5680       movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5681     }
 5682     addq(dst, src);
 5683   } else {
 5684     movptr(dst, src);
 5685   }
 5686   if (CompressedKlassPointers::shift() != 0) {
 5687     shrq(dst, CompressedKlassPointers::shift());
 5688   }
 5689   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5690 }
 5691 
 5692 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5693   BLOCK_COMMENT("decode_klass_not_null {");
 5694   assert_different_registers(r, tmp);
 5695   // Note: it will change flags
 5696   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5697   // vtableStubs also counts instructions in pd_code_size_limit.
 5698   // Also do not verify_oop as this is called by verify_oop.
 5699   if (CompressedKlassPointers::shift() != 0) {
 5700     shlq(r, CompressedKlassPointers::shift());
 5701   }
 5702   if (CompressedKlassPointers::base() != nullptr) {
 5703     if (AOTCodeCache::is_on_for_dump()) {
 5704       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5705     } else {
 5706       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5707     }
 5708     addq(r, tmp);
 5709   }
 5710   BLOCK_COMMENT("} decode_klass_not_null");
 5711 }
 5712 
 5713 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5714   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 5715   assert_different_registers(src, dst);
 5716   // Note: it will change flags
 5717   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5718   // vtableStubs also counts instructions in pd_code_size_limit.
 5719   // Also do not verify_oop as this is called by verify_oop.
 5720 
 5721   if (CompressedKlassPointers::base() == nullptr &&
 5722       CompressedKlassPointers::shift() == 0) {
 5723     // The best case scenario is that there is no base or shift. Then it is already
 5724     // a pointer that needs nothing but a register rename.
 5725     movl(dst, src);
 5726   } else {
 5727     if (CompressedKlassPointers::shift() <= Address::times_8) {
 5728       if (CompressedKlassPointers::base() != nullptr) {
 5729         if (AOTCodeCache::is_on_for_dump()) {
 5730           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5731         } else {
 5732           movptr(dst, (intptr_t)CompressedKlassPointers::base());
 5733         }
 5734       } else {
 5735         xorq(dst, dst);
 5736       }
 5737       if (CompressedKlassPointers::shift() != 0) {
 5738         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 5739         leaq(dst, Address(dst, src, Address::times_8, 0));
 5740       } else {
 5741         addq(dst, src);
 5742       }
 5743     } else {
 5744       if (CompressedKlassPointers::base() != nullptr) {
 5745         if (AOTCodeCache::is_on_for_dump()) {
 5746           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5747           shrq(dst, CompressedKlassPointers::shift());
 5748         } else {
 5749           const intptr_t base_right_shifted =
 5750                (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 5751           movptr(dst, base_right_shifted);
 5752         }
 5753       } else {
 5754         xorq(dst, dst);
 5755       }
 5756       addq(dst, src);
 5757       shlq(dst, CompressedKlassPointers::shift());
 5758     }
 5759   }
 5760   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 5761 }
 5762 
 5763 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5764   assert (UseCompressedOops, "should only be used for compressed headers");
 5765   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5766   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5767   int oop_index = oop_recorder()->find_index(obj);
 5768   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5769   mov_narrow_oop(dst, oop_index, rspec);
 5770 }
 5771 
 5772 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5773   assert (UseCompressedOops, "should only be used for compressed headers");
 5774   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5775   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5776   int oop_index = oop_recorder()->find_index(obj);
 5777   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5778   mov_narrow_oop(dst, oop_index, rspec);
 5779 }
 5780 
 5781 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5782   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5783   int klass_index = oop_recorder()->find_index(k);
 5784   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5785   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5786 }
 5787 
 5788 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 5789   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5790   int klass_index = oop_recorder()->find_index(k);
 5791   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5792   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5793 }
 5794 
 5795 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 5796   assert (UseCompressedOops, "should only be used for compressed headers");
 5797   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5798   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5799   int oop_index = oop_recorder()->find_index(obj);
 5800   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5801   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5802 }
 5803 
 5804 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 5805   assert (UseCompressedOops, "should only be used for compressed headers");
 5806   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5807   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5808   int oop_index = oop_recorder()->find_index(obj);
 5809   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5810   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5811 }
 5812 
 5813 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 5814   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5815   int klass_index = oop_recorder()->find_index(k);
 5816   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5817   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5818 }
 5819 
 5820 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 5821   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5822   int klass_index = oop_recorder()->find_index(k);
 5823   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5824   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5825 }
 5826 
 5827 void MacroAssembler::reinit_heapbase() {
 5828   if (UseCompressedOops) {
 5829     if (Universe::heap() != nullptr && !AOTCodeCache::is_on_for_dump()) {
 5830       if (CompressedOops::base() == nullptr) {
 5831         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 5832       } else {
 5833         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 5834       }
 5835     } else {
 5836       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 5837     }
 5838   }
 5839 }
 5840 





































































































































































































































































































































































































































































































 5841 #if COMPILER2_OR_JVMCI
 5842 
 5843 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 5844 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5845   // cnt - number of qwords (8-byte words).
 5846   // base - start address, qword aligned.
 5847   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 5848   bool use64byteVector = (MaxVectorSize == 64) && (CopyAVX3Threshold == 0);
 5849   if (use64byteVector) {
 5850     vpxor(xtmp, xtmp, xtmp, AVX_512bit);
 5851   } else if (MaxVectorSize >= 32) {
 5852     vpxor(xtmp, xtmp, xtmp, AVX_256bit);


 5853   } else {
 5854     pxor(xtmp, xtmp);

 5855   }
 5856   jmp(L_zero_64_bytes);
 5857 
 5858   BIND(L_loop);
 5859   if (MaxVectorSize >= 32) {
 5860     fill64(base, 0, xtmp, use64byteVector);
 5861   } else {
 5862     movdqu(Address(base,  0), xtmp);
 5863     movdqu(Address(base, 16), xtmp);
 5864     movdqu(Address(base, 32), xtmp);
 5865     movdqu(Address(base, 48), xtmp);
 5866   }
 5867   addptr(base, 64);
 5868 
 5869   BIND(L_zero_64_bytes);
 5870   subptr(cnt, 8);
 5871   jccb(Assembler::greaterEqual, L_loop);
 5872 
 5873   // Copy trailing 64 bytes
 5874   if (use64byteVector) {
 5875     addptr(cnt, 8);
 5876     jccb(Assembler::equal, L_end);
 5877     fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
 5878     jmp(L_end);
 5879   } else {
 5880     addptr(cnt, 4);
 5881     jccb(Assembler::less, L_tail);
 5882     if (MaxVectorSize >= 32) {
 5883       vmovdqu(Address(base, 0), xtmp);
 5884     } else {
 5885       movdqu(Address(base,  0), xtmp);
 5886       movdqu(Address(base, 16), xtmp);
 5887     }
 5888   }
 5889   addptr(base, 32);
 5890   subptr(cnt, 4);
 5891 
 5892   BIND(L_tail);
 5893   addptr(cnt, 4);
 5894   jccb(Assembler::lessEqual, L_end);
 5895   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 5896     fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
 5897   } else {
 5898     decrement(cnt);
 5899 
 5900     BIND(L_sloop);
 5901     movq(Address(base, 0), xtmp);
 5902     addptr(base, 8);
 5903     decrement(cnt);
 5904     jccb(Assembler::greaterEqual, L_sloop);
 5905   }
 5906   BIND(L_end);
 5907 }
 5908 
 5909 // Clearing constant sized memory using YMM/ZMM registers.
 5910 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5911   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 5912   bool use64byteVector = (MaxVectorSize > 32) && (CopyAVX3Threshold == 0);
 5913 
 5914   int vector64_count = (cnt & (~0x7)) >> 3;
 5915   cnt = cnt & 0x7;
 5916   const int fill64_per_loop = 4;
 5917   const int max_unrolled_fill64 = 8;
 5918 
 5919   // 64 byte initialization loop.
 5920   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 5921   int start64 = 0;
 5922   if (vector64_count > max_unrolled_fill64) {
 5923     Label LOOP;
 5924     Register index = rtmp;
 5925 
 5926     start64 = vector64_count - (vector64_count % fill64_per_loop);
 5927 
 5928     movl(index, 0);
 5929     BIND(LOOP);
 5930     for (int i = 0; i < fill64_per_loop; i++) {
 5931       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 5932     }
 5933     addl(index, fill64_per_loop * 64);
 5934     cmpl(index, start64 * 64);
 5935     jccb(Assembler::less, LOOP);
 5936   }
 5937   for (int i = start64; i < vector64_count; i++) {
 5938     fill64(base, i * 64, xtmp, use64byteVector);
 5939   }
 5940 
 5941   // Clear remaining 64 byte tail.
 5942   int disp = vector64_count * 64;
 5943   if (cnt) {
 5944     switch (cnt) {
 5945       case 1:
 5946         movq(Address(base, disp), xtmp);
 5947         break;
 5948       case 2:
 5949         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 5950         break;
 5951       case 3:
 5952         movl(rtmp, 0x7);
 5953         kmovwl(mask, rtmp);
 5954         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 5955         break;
 5956       case 4:
 5957         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5958         break;
 5959       case 5:
 5960         if (use64byteVector) {
 5961           movl(rtmp, 0x1F);
 5962           kmovwl(mask, rtmp);
 5963           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5964         } else {
 5965           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5966           movq(Address(base, disp + 32), xtmp);
 5967         }
 5968         break;
 5969       case 6:
 5970         if (use64byteVector) {
 5971           movl(rtmp, 0x3F);
 5972           kmovwl(mask, rtmp);
 5973           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5974         } else {
 5975           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5976           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 5977         }
 5978         break;
 5979       case 7:
 5980         if (use64byteVector) {
 5981           movl(rtmp, 0x7F);
 5982           kmovwl(mask, rtmp);
 5983           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5984         } else {
 5985           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5986           movl(rtmp, 0x7);
 5987           kmovwl(mask, rtmp);
 5988           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 5989         }
 5990         break;
 5991       default:
 5992         fatal("Unexpected length : %d\n",cnt);
 5993         break;
 5994     }
 5995   }
 5996 }
 5997 
 5998 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
 5999                                bool is_large, KRegister mask) {
 6000   // cnt      - number of qwords (8-byte words).
 6001   // base     - start address, qword aligned.
 6002   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6003   assert(base==rdi, "base register must be edi for rep stos");
 6004   assert(tmp==rax,   "tmp register must be eax for rep stos");
 6005   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6006   assert(InitArrayShortSize % BytesPerLong == 0,
 6007     "InitArrayShortSize should be the multiple of BytesPerLong");
 6008 
 6009   Label DONE;
 6010   if (!is_large || !UseXMMForObjInit) {
 6011     xorptr(tmp, tmp);
 6012   }
 6013 
 6014   if (!is_large) {
 6015     Label LOOP, LONG;
 6016     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6017     jccb(Assembler::greater, LONG);
 6018 
 6019     decrement(cnt);
 6020     jccb(Assembler::negative, DONE); // Zero length
 6021 
 6022     // Use individual pointer-sized stores for small counts:
 6023     BIND(LOOP);
 6024     movptr(Address(base, cnt, Address::times_ptr), tmp);
 6025     decrement(cnt);
 6026     jccb(Assembler::greaterEqual, LOOP);
 6027     jmpb(DONE);
 6028 
 6029     BIND(LONG);
 6030   }
 6031 
 6032   // Use longer rep-prefixed ops for non-small counts:
 6033   if (UseFastStosb) {
 6034     shlptr(cnt, 3); // convert to number of bytes
 6035     rep_stosb();
 6036   } else if (UseXMMForObjInit) {
 6037     xmm_clear_mem(base, cnt, tmp, xtmp, mask);
 6038   } else {
 6039     rep_stos();
 6040   }
 6041 
 6042   BIND(DONE);
 6043 }
 6044 
 6045 #endif //COMPILER2_OR_JVMCI
 6046 
 6047 
 6048 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6049                                    Register to, Register value, Register count,
 6050                                    Register rtmp, XMMRegister xtmp) {
 6051   ShortBranchVerifier sbv(this);
 6052   assert_different_registers(to, value, count, rtmp);
 6053   Label L_exit;
 6054   Label L_fill_2_bytes, L_fill_4_bytes;
 6055 
 6056 #if defined(COMPILER2)
 6057   if(MaxVectorSize >=32 &&
 6058      VM_Version::supports_avx512vlbw() &&
 6059      VM_Version::supports_bmi2()) {
 6060     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6061     return;
 6062   }
 6063 #endif
 6064 
 6065   int shift = -1;
 6066   switch (t) {
 6067     case T_BYTE:
 6068       shift = 2;
 6069       break;
 6070     case T_SHORT:
 6071       shift = 1;
 6072       break;
 6073     case T_INT:
 6074       shift = 0;
 6075       break;
 6076     default: ShouldNotReachHere();
 6077   }
 6078 
 6079   if (t == T_BYTE) {
 6080     andl(value, 0xff);
 6081     movl(rtmp, value);
 6082     shll(rtmp, 8);
 6083     orl(value, rtmp);
 6084   }
 6085   if (t == T_SHORT) {
 6086     andl(value, 0xffff);
 6087   }
 6088   if (t == T_BYTE || t == T_SHORT) {
 6089     movl(rtmp, value);
 6090     shll(rtmp, 16);
 6091     orl(value, rtmp);
 6092   }
 6093 
 6094   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6095   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6096   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6097     Label L_skip_align2;
 6098     // align source address at 4 bytes address boundary
 6099     if (t == T_BYTE) {
 6100       Label L_skip_align1;
 6101       // One byte misalignment happens only for byte arrays
 6102       testptr(to, 1);
 6103       jccb(Assembler::zero, L_skip_align1);
 6104       movb(Address(to, 0), value);
 6105       increment(to);
 6106       decrement(count);
 6107       BIND(L_skip_align1);
 6108     }
 6109     // Two bytes misalignment happens only for byte and short (char) arrays
 6110     testptr(to, 2);
 6111     jccb(Assembler::zero, L_skip_align2);
 6112     movw(Address(to, 0), value);
 6113     addptr(to, 2);
 6114     subptr(count, 1<<(shift-1));
 6115     BIND(L_skip_align2);
 6116   }
 6117   {
 6118     Label L_fill_32_bytes;
 6119     if (!UseUnalignedLoadStores) {
 6120       // align to 8 bytes, we know we are 4 byte aligned to start
 6121       testptr(to, 4);
 6122       jccb(Assembler::zero, L_fill_32_bytes);
 6123       movl(Address(to, 0), value);
 6124       addptr(to, 4);
 6125       subptr(count, 1<<shift);
 6126     }
 6127     BIND(L_fill_32_bytes);
 6128     {
 6129       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6130       movdl(xtmp, value);
 6131       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6132         Label L_check_fill_32_bytes;
 6133         if (UseAVX > 2) {
 6134           // Fill 64-byte chunks
 6135           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6136 
 6137           // If number of bytes to fill < CopyAVX3Threshold, perform fill using AVX2
 6138           cmpptr(count, CopyAVX3Threshold);
 6139           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6140 
 6141           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6142 
 6143           subptr(count, 16 << shift);
 6144           jcc(Assembler::less, L_check_fill_32_bytes);
 6145           align(16);
 6146 
 6147           BIND(L_fill_64_bytes_loop_avx3);
 6148           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6149           addptr(to, 64);
 6150           subptr(count, 16 << shift);
 6151           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6152           jmpb(L_check_fill_32_bytes);
 6153 
 6154           BIND(L_check_fill_64_bytes_avx2);
 6155         }
 6156         // Fill 64-byte chunks
 6157         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6158 
 6159         subptr(count, 16 << shift);
 6160         jcc(Assembler::less, L_check_fill_32_bytes);
 6161 
 6162         // align data for 64-byte chunks
 6163         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6164         if (EnableX86ECoreOpts) {
 6165             // align 'big' arrays to cache lines to minimize split_stores
 6166             cmpptr(count, 96 << shift);
 6167             jcc(Assembler::below, L_fill_64_bytes_loop);
 6168 
 6169             // Find the bytes needed for alignment
 6170             movptr(rtmp, to);
 6171             andptr(rtmp, 0x1c);
 6172             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6173             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6174             addptr(rtmp, 32);
 6175             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6176             subptr(count, rtmp);    // adjust count by number of elements
 6177 
 6178             align(16);
 6179             BIND(L_align_64_bytes_loop);
 6180             movdl(Address(to, 0), xtmp);
 6181             addptr(to, 4);
 6182             subptr(rtmp, 1 << shift);
 6183             jcc(Assembler::greater, L_align_64_bytes_loop);
 6184         }
 6185 
 6186         align(16);
 6187         BIND(L_fill_64_bytes_loop);
 6188         vmovdqu(Address(to, 0), xtmp);
 6189         vmovdqu(Address(to, 32), xtmp);
 6190         addptr(to, 64);
 6191         subptr(count, 16 << shift);
 6192         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6193 
 6194         align(16);
 6195         BIND(L_check_fill_32_bytes);
 6196         addptr(count, 8 << shift);
 6197         jccb(Assembler::less, L_check_fill_8_bytes);
 6198         vmovdqu(Address(to, 0), xtmp);
 6199         addptr(to, 32);
 6200         subptr(count, 8 << shift);
 6201 
 6202         BIND(L_check_fill_8_bytes);
 6203         // clean upper bits of YMM registers
 6204         movdl(xtmp, value);
 6205         pshufd(xtmp, xtmp, 0);
 6206       } else {
 6207         // Fill 32-byte chunks
 6208         pshufd(xtmp, xtmp, 0);
 6209 
 6210         subptr(count, 8 << shift);
 6211         jcc(Assembler::less, L_check_fill_8_bytes);
 6212         align(16);
 6213 
 6214         BIND(L_fill_32_bytes_loop);
 6215 
 6216         if (UseUnalignedLoadStores) {
 6217           movdqu(Address(to, 0), xtmp);
 6218           movdqu(Address(to, 16), xtmp);
 6219         } else {
 6220           movq(Address(to, 0), xtmp);
 6221           movq(Address(to, 8), xtmp);
 6222           movq(Address(to, 16), xtmp);
 6223           movq(Address(to, 24), xtmp);
 6224         }
 6225 
 6226         addptr(to, 32);
 6227         subptr(count, 8 << shift);
 6228         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6229 
 6230         BIND(L_check_fill_8_bytes);
 6231       }
 6232       addptr(count, 8 << shift);
 6233       jccb(Assembler::zero, L_exit);
 6234       jmpb(L_fill_8_bytes);
 6235 
 6236       //
 6237       // length is too short, just fill qwords
 6238       //
 6239       align(16);
 6240       BIND(L_fill_8_bytes_loop);
 6241       movq(Address(to, 0), xtmp);
 6242       addptr(to, 8);
 6243       BIND(L_fill_8_bytes);
 6244       subptr(count, 1 << (shift + 1));
 6245       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6246     }
 6247   }
 6248 
 6249   Label L_fill_4_bytes_loop;
 6250   testl(count, 1 << shift);
 6251   jccb(Assembler::zero, L_fill_2_bytes);
 6252 
 6253   align(16);
 6254   BIND(L_fill_4_bytes_loop);
 6255   movl(Address(to, 0), value);
 6256   addptr(to, 4);
 6257 
 6258   BIND(L_fill_4_bytes);
 6259   subptr(count, 1 << shift);
 6260   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 6261 
 6262   if (t == T_BYTE || t == T_SHORT) {
 6263     Label L_fill_byte;
 6264     BIND(L_fill_2_bytes);
 6265     // fill trailing 2 bytes
 6266     testl(count, 1<<(shift-1));
 6267     jccb(Assembler::zero, L_fill_byte);
 6268     movw(Address(to, 0), value);
 6269     if (t == T_BYTE) {
 6270       addptr(to, 2);
 6271       BIND(L_fill_byte);
 6272       // fill trailing byte
 6273       testl(count, 1);
 6274       jccb(Assembler::zero, L_exit);
 6275       movb(Address(to, 0), value);
 6276     } else {
 6277       BIND(L_fill_byte);
 6278     }
 6279   } else {
 6280     BIND(L_fill_2_bytes);
 6281   }
 6282   BIND(L_exit);
 6283 }
 6284 
 6285 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6286   switch(type) {
 6287     case T_BYTE:
 6288     case T_BOOLEAN:
 6289       evpbroadcastb(dst, src, vector_len);
 6290       break;
 6291     case T_SHORT:
 6292     case T_CHAR:
 6293       evpbroadcastw(dst, src, vector_len);
 6294       break;
 6295     case T_INT:
 6296     case T_FLOAT:
 6297       evpbroadcastd(dst, src, vector_len);
 6298       break;
 6299     case T_LONG:
 6300     case T_DOUBLE:
 6301       evpbroadcastq(dst, src, vector_len);
 6302       break;
 6303     default:
 6304       fatal("Unhandled type : %s", type2name(type));
 6305       break;
 6306   }
 6307 }
 6308 
 6309 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 6310 //
 6311 // @IntrinsicCandidate
 6312 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 6313 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6314 //     int i = 0;
 6315 //     for (; i < len; i++) {
 6316 //         char c = sa[sp++];
 6317 //         if (c > '\u00FF')
 6318 //             break;
 6319 //         da[dp++] = (byte) c;
 6320 //     }
 6321 //     return i;
 6322 // }
 6323 //
 6324 // @IntrinsicCandidate
 6325 // int java.lang.StringCoding.encodeISOArray0(
 6326 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 6327 //   int i = 0;
 6328 //   for (; i < len; i++) {
 6329 //     char c = StringUTF16.getChar(sa, sp++);
 6330 //     if (c > '\u00FF')
 6331 //       break;
 6332 //     da[dp++] = (byte) c;
 6333 //   }
 6334 //   return i;
 6335 // }
 6336 //
 6337 // @IntrinsicCandidate
 6338 // int java.lang.StringCoding.encodeAsciiArray0(
 6339 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6340 //   int i = 0;
 6341 //   for (; i < len; i++) {
 6342 //     char c = sa[sp++];
 6343 //     if (c >= '\u0080')
 6344 //       break;
 6345 //     da[dp++] = (byte) c;
 6346 //   }
 6347 //   return i;
 6348 // }
 6349 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 6350   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 6351   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 6352   Register tmp5, Register result, bool ascii) {
 6353 
 6354   // rsi: src
 6355   // rdi: dst
 6356   // rdx: len
 6357   // rcx: tmp5
 6358   // rax: result
 6359   ShortBranchVerifier sbv(this);
 6360   assert_different_registers(src, dst, len, tmp5, result);
 6361   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 6362 
 6363   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 6364   int short_mask = ascii ? 0xff80 : 0xff00;
 6365 
 6366   // set result
 6367   xorl(result, result);
 6368   // check for zero length
 6369   testl(len, len);
 6370   jcc(Assembler::zero, L_done);
 6371 
 6372   movl(result, len);
 6373 
 6374   // Setup pointers
 6375   lea(src, Address(src, len, Address::times_2)); // char[]
 6376   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 6377   negptr(len);
 6378 
 6379   if (UseSSE42Intrinsics || UseAVX >= 2) {
 6380     Label L_copy_8_chars, L_copy_8_chars_exit;
 6381     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 6382 
 6383     if (UseAVX >= 2) {
 6384       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 6385       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6386       movdl(tmp1Reg, tmp5);
 6387       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 6388       jmp(L_chars_32_check);
 6389 
 6390       bind(L_copy_32_chars);
 6391       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 6392       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 6393       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6394       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6395       jccb(Assembler::notZero, L_copy_32_chars_exit);
 6396       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 6397       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 6398       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 6399 
 6400       bind(L_chars_32_check);
 6401       addptr(len, 32);
 6402       jcc(Assembler::lessEqual, L_copy_32_chars);
 6403 
 6404       bind(L_copy_32_chars_exit);
 6405       subptr(len, 16);
 6406       jccb(Assembler::greater, L_copy_16_chars_exit);
 6407 
 6408     } else if (UseSSE42Intrinsics) {
 6409       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 6410       movdl(tmp1Reg, tmp5);
 6411       pshufd(tmp1Reg, tmp1Reg, 0);
 6412       jmpb(L_chars_16_check);
 6413     }
 6414 
 6415     bind(L_copy_16_chars);
 6416     if (UseAVX >= 2) {
 6417       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 6418       vptest(tmp2Reg, tmp1Reg);
 6419       jcc(Assembler::notZero, L_copy_16_chars_exit);
 6420       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 6421       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 6422     } else {
 6423       if (UseAVX > 0) {
 6424         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6425         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6426         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 6427       } else {
 6428         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 6429         por(tmp2Reg, tmp3Reg);
 6430         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 6431         por(tmp2Reg, tmp4Reg);
 6432       }
 6433       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 6434       jccb(Assembler::notZero, L_copy_16_chars_exit);
 6435       packuswb(tmp3Reg, tmp4Reg);
 6436     }
 6437     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 6438 
 6439     bind(L_chars_16_check);
 6440     addptr(len, 16);
 6441     jcc(Assembler::lessEqual, L_copy_16_chars);
 6442 
 6443     bind(L_copy_16_chars_exit);
 6444     if (UseAVX >= 2) {
 6445       // clean upper bits of YMM registers
 6446       vpxor(tmp2Reg, tmp2Reg);
 6447       vpxor(tmp3Reg, tmp3Reg);
 6448       vpxor(tmp4Reg, tmp4Reg);
 6449       movdl(tmp1Reg, tmp5);
 6450       pshufd(tmp1Reg, tmp1Reg, 0);
 6451     }
 6452     subptr(len, 8);
 6453     jccb(Assembler::greater, L_copy_8_chars_exit);
 6454 
 6455     bind(L_copy_8_chars);
 6456     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 6457     ptest(tmp3Reg, tmp1Reg);
 6458     jccb(Assembler::notZero, L_copy_8_chars_exit);
 6459     packuswb(tmp3Reg, tmp1Reg);
 6460     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 6461     addptr(len, 8);
 6462     jccb(Assembler::lessEqual, L_copy_8_chars);
 6463 
 6464     bind(L_copy_8_chars_exit);
 6465     subptr(len, 8);
 6466     jccb(Assembler::zero, L_done);
 6467   }
 6468 
 6469   bind(L_copy_1_char);
 6470   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 6471   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 6472   jccb(Assembler::notZero, L_copy_1_char_exit);
 6473   movb(Address(dst, len, Address::times_1, 0), tmp5);
 6474   addptr(len, 1);
 6475   jccb(Assembler::less, L_copy_1_char);
 6476 
 6477   bind(L_copy_1_char_exit);
 6478   addptr(result, len); // len is negative count of not processed elements
 6479 
 6480   bind(L_done);
 6481 }
 6482 
 6483 /**
 6484  * Helper for multiply_to_len().
 6485  */
 6486 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 6487   addq(dest_lo, src1);
 6488   adcq(dest_hi, 0);
 6489   addq(dest_lo, src2);
 6490   adcq(dest_hi, 0);
 6491 }
 6492 
 6493 /**
 6494  * Multiply 64 bit by 64 bit first loop.
 6495  */
 6496 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 6497                                            Register y, Register y_idx, Register z,
 6498                                            Register carry, Register product,
 6499                                            Register idx, Register kdx) {
 6500   //
 6501   //  jlong carry, x[], y[], z[];
 6502   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6503   //    huge_128 product = y[idx] * x[xstart] + carry;
 6504   //    z[kdx] = (jlong)product;
 6505   //    carry  = (jlong)(product >>> 64);
 6506   //  }
 6507   //  z[xstart] = carry;
 6508   //
 6509 
 6510   Label L_first_loop, L_first_loop_exit;
 6511   Label L_one_x, L_one_y, L_multiply;
 6512 
 6513   decrementl(xstart);
 6514   jcc(Assembler::negative, L_one_x);
 6515 
 6516   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6517   rorq(x_xstart, 32); // convert big-endian to little-endian
 6518 
 6519   bind(L_first_loop);
 6520   decrementl(idx);
 6521   jcc(Assembler::negative, L_first_loop_exit);
 6522   decrementl(idx);
 6523   jcc(Assembler::negative, L_one_y);
 6524   movq(y_idx, Address(y, idx, Address::times_4,  0));
 6525   rorq(y_idx, 32); // convert big-endian to little-endian
 6526   bind(L_multiply);
 6527   movq(product, x_xstart);
 6528   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 6529   addq(product, carry);
 6530   adcq(rdx, 0);
 6531   subl(kdx, 2);
 6532   movl(Address(z, kdx, Address::times_4,  4), product);
 6533   shrq(product, 32);
 6534   movl(Address(z, kdx, Address::times_4,  0), product);
 6535   movq(carry, rdx);
 6536   jmp(L_first_loop);
 6537 
 6538   bind(L_one_y);
 6539   movl(y_idx, Address(y,  0));
 6540   jmp(L_multiply);
 6541 
 6542   bind(L_one_x);
 6543   movl(x_xstart, Address(x,  0));
 6544   jmp(L_first_loop);
 6545 
 6546   bind(L_first_loop_exit);
 6547 }
 6548 
 6549 /**
 6550  * Multiply 64 bit by 64 bit and add 128 bit.
 6551  */
 6552 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 6553                                             Register yz_idx, Register idx,
 6554                                             Register carry, Register product, int offset) {
 6555   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 6556   //     z[kdx] = (jlong)product;
 6557 
 6558   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 6559   rorq(yz_idx, 32); // convert big-endian to little-endian
 6560   movq(product, x_xstart);
 6561   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 6562   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 6563   rorq(yz_idx, 32); // convert big-endian to little-endian
 6564 
 6565   add2_with_carry(rdx, product, carry, yz_idx);
 6566 
 6567   movl(Address(z, idx, Address::times_4,  offset+4), product);
 6568   shrq(product, 32);
 6569   movl(Address(z, idx, Address::times_4,  offset), product);
 6570 
 6571 }
 6572 
 6573 /**
 6574  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 6575  */
 6576 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 6577                                              Register yz_idx, Register idx, Register jdx,
 6578                                              Register carry, Register product,
 6579                                              Register carry2) {
 6580   //   jlong carry, x[], y[], z[];
 6581   //   int kdx = ystart+1;
 6582   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6583   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 6584   //     z[kdx+idx+1] = (jlong)product;
 6585   //     jlong carry2  = (jlong)(product >>> 64);
 6586   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 6587   //     z[kdx+idx] = (jlong)product;
 6588   //     carry  = (jlong)(product >>> 64);
 6589   //   }
 6590   //   idx += 2;
 6591   //   if (idx > 0) {
 6592   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 6593   //     z[kdx+idx] = (jlong)product;
 6594   //     carry  = (jlong)(product >>> 64);
 6595   //   }
 6596   //
 6597 
 6598   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6599 
 6600   movl(jdx, idx);
 6601   andl(jdx, 0xFFFFFFFC);
 6602   shrl(jdx, 2);
 6603 
 6604   bind(L_third_loop);
 6605   subl(jdx, 1);
 6606   jcc(Assembler::negative, L_third_loop_exit);
 6607   subl(idx, 4);
 6608 
 6609   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 6610   movq(carry2, rdx);
 6611 
 6612   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 6613   movq(carry, rdx);
 6614   jmp(L_third_loop);
 6615 
 6616   bind (L_third_loop_exit);
 6617 
 6618   andl (idx, 0x3);
 6619   jcc(Assembler::zero, L_post_third_loop_done);
 6620 
 6621   Label L_check_1;
 6622   subl(idx, 2);
 6623   jcc(Assembler::negative, L_check_1);
 6624 
 6625   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 6626   movq(carry, rdx);
 6627 
 6628   bind (L_check_1);
 6629   addl (idx, 0x2);
 6630   andl (idx, 0x1);
 6631   subl(idx, 1);
 6632   jcc(Assembler::negative, L_post_third_loop_done);
 6633 
 6634   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 6635   movq(product, x_xstart);
 6636   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 6637   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 6638 
 6639   add2_with_carry(rdx, product, yz_idx, carry);
 6640 
 6641   movl(Address(z, idx, Address::times_4,  0), product);
 6642   shrq(product, 32);
 6643 
 6644   shlq(rdx, 32);
 6645   orq(product, rdx);
 6646   movq(carry, product);
 6647 
 6648   bind(L_post_third_loop_done);
 6649 }
 6650 
 6651 /**
 6652  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 6653  *
 6654  */
 6655 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 6656                                                   Register carry, Register carry2,
 6657                                                   Register idx, Register jdx,
 6658                                                   Register yz_idx1, Register yz_idx2,
 6659                                                   Register tmp, Register tmp3, Register tmp4) {
 6660   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 6661 
 6662   //   jlong carry, x[], y[], z[];
 6663   //   int kdx = ystart+1;
 6664   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 6665   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 6666   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 6667   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 6668   //     carry  = (jlong)(tmp4 >>> 64);
 6669   //     z[kdx+idx+1] = (jlong)tmp3;
 6670   //     z[kdx+idx] = (jlong)tmp4;
 6671   //   }
 6672   //   idx += 2;
 6673   //   if (idx > 0) {
 6674   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 6675   //     z[kdx+idx] = (jlong)yz_idx1;
 6676   //     carry  = (jlong)(yz_idx1 >>> 64);
 6677   //   }
 6678   //
 6679 
 6680   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 6681 
 6682   movl(jdx, idx);
 6683   andl(jdx, 0xFFFFFFFC);
 6684   shrl(jdx, 2);
 6685 
 6686   bind(L_third_loop);
 6687   subl(jdx, 1);
 6688   jcc(Assembler::negative, L_third_loop_exit);
 6689   subl(idx, 4);
 6690 
 6691   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 6692   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 6693   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 6694   rorxq(yz_idx2, yz_idx2, 32);
 6695 
 6696   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 6697   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 6698 
 6699   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 6700   rorxq(yz_idx1, yz_idx1, 32);
 6701   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6702   rorxq(yz_idx2, yz_idx2, 32);
 6703 
 6704   if (VM_Version::supports_adx()) {
 6705     adcxq(tmp3, carry);
 6706     adoxq(tmp3, yz_idx1);
 6707 
 6708     adcxq(tmp4, tmp);
 6709     adoxq(tmp4, yz_idx2);
 6710 
 6711     movl(carry, 0); // does not affect flags
 6712     adcxq(carry2, carry);
 6713     adoxq(carry2, carry);
 6714   } else {
 6715     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 6716     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 6717   }
 6718   movq(carry, carry2);
 6719 
 6720   movl(Address(z, idx, Address::times_4, 12), tmp3);
 6721   shrq(tmp3, 32);
 6722   movl(Address(z, idx, Address::times_4,  8), tmp3);
 6723 
 6724   movl(Address(z, idx, Address::times_4,  4), tmp4);
 6725   shrq(tmp4, 32);
 6726   movl(Address(z, idx, Address::times_4,  0), tmp4);
 6727 
 6728   jmp(L_third_loop);
 6729 
 6730   bind (L_third_loop_exit);
 6731 
 6732   andl (idx, 0x3);
 6733   jcc(Assembler::zero, L_post_third_loop_done);
 6734 
 6735   Label L_check_1;
 6736   subl(idx, 2);
 6737   jcc(Assembler::negative, L_check_1);
 6738 
 6739   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 6740   rorxq(yz_idx1, yz_idx1, 32);
 6741   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 6742   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 6743   rorxq(yz_idx2, yz_idx2, 32);
 6744 
 6745   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 6746 
 6747   movl(Address(z, idx, Address::times_4,  4), tmp3);
 6748   shrq(tmp3, 32);
 6749   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6750   movq(carry, tmp4);
 6751 
 6752   bind (L_check_1);
 6753   addl (idx, 0x2);
 6754   andl (idx, 0x1);
 6755   subl(idx, 1);
 6756   jcc(Assembler::negative, L_post_third_loop_done);
 6757   movl(tmp4, Address(y, idx, Address::times_4,  0));
 6758   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 6759   movl(tmp4, Address(z, idx, Address::times_4,  0));
 6760 
 6761   add2_with_carry(carry2, tmp3, tmp4, carry);
 6762 
 6763   movl(Address(z, idx, Address::times_4,  0), tmp3);
 6764   shrq(tmp3, 32);
 6765 
 6766   shlq(carry2, 32);
 6767   orq(tmp3, carry2);
 6768   movq(carry, tmp3);
 6769 
 6770   bind(L_post_third_loop_done);
 6771 }
 6772 
 6773 /**
 6774  * Code for BigInteger::multiplyToLen() intrinsic.
 6775  *
 6776  * rdi: x
 6777  * rax: xlen
 6778  * rsi: y
 6779  * rcx: ylen
 6780  * r8:  z
 6781  * r11: tmp0
 6782  * r12: tmp1
 6783  * r13: tmp2
 6784  * r14: tmp3
 6785  * r15: tmp4
 6786  * rbx: tmp5
 6787  *
 6788  */
 6789 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 6790                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 6791   ShortBranchVerifier sbv(this);
 6792   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 6793 
 6794   push(tmp0);
 6795   push(tmp1);
 6796   push(tmp2);
 6797   push(tmp3);
 6798   push(tmp4);
 6799   push(tmp5);
 6800 
 6801   push(xlen);
 6802 
 6803   const Register idx = tmp1;
 6804   const Register kdx = tmp2;
 6805   const Register xstart = tmp3;
 6806 
 6807   const Register y_idx = tmp4;
 6808   const Register carry = tmp5;
 6809   const Register product  = xlen;
 6810   const Register x_xstart = tmp0;
 6811 
 6812   // First Loop.
 6813   //
 6814   //  final static long LONG_MASK = 0xffffffffL;
 6815   //  int xstart = xlen - 1;
 6816   //  int ystart = ylen - 1;
 6817   //  long carry = 0;
 6818   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 6819   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 6820   //    z[kdx] = (int)product;
 6821   //    carry = product >>> 32;
 6822   //  }
 6823   //  z[xstart] = (int)carry;
 6824   //
 6825 
 6826   movl(idx, ylen);               // idx = ylen;
 6827   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 6828   xorq(carry, carry);            // carry = 0;
 6829 
 6830   Label L_done;
 6831 
 6832   movl(xstart, xlen);
 6833   decrementl(xstart);
 6834   jcc(Assembler::negative, L_done);
 6835 
 6836   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 6837 
 6838   Label L_second_loop;
 6839   testl(kdx, kdx);
 6840   jcc(Assembler::zero, L_second_loop);
 6841 
 6842   Label L_carry;
 6843   subl(kdx, 1);
 6844   jcc(Assembler::zero, L_carry);
 6845 
 6846   movl(Address(z, kdx, Address::times_4,  0), carry);
 6847   shrq(carry, 32);
 6848   subl(kdx, 1);
 6849 
 6850   bind(L_carry);
 6851   movl(Address(z, kdx, Address::times_4,  0), carry);
 6852 
 6853   // Second and third (nested) loops.
 6854   //
 6855   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 6856   //   carry = 0;
 6857   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 6858   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 6859   //                    (z[k] & LONG_MASK) + carry;
 6860   //     z[k] = (int)product;
 6861   //     carry = product >>> 32;
 6862   //   }
 6863   //   z[i] = (int)carry;
 6864   // }
 6865   //
 6866   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 6867 
 6868   const Register jdx = tmp1;
 6869 
 6870   bind(L_second_loop);
 6871   xorl(carry, carry);    // carry = 0;
 6872   movl(jdx, ylen);       // j = ystart+1
 6873 
 6874   subl(xstart, 1);       // i = xstart-1;
 6875   jcc(Assembler::negative, L_done);
 6876 
 6877   push (z);
 6878 
 6879   Label L_last_x;
 6880   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 6881   subl(xstart, 1);       // i = xstart-1;
 6882   jcc(Assembler::negative, L_last_x);
 6883 
 6884   if (UseBMI2Instructions) {
 6885     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 6886     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 6887   } else {
 6888     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 6889     rorq(x_xstart, 32);  // convert big-endian to little-endian
 6890   }
 6891 
 6892   Label L_third_loop_prologue;
 6893   bind(L_third_loop_prologue);
 6894 
 6895   push (x);
 6896   push (xstart);
 6897   push (ylen);
 6898 
 6899 
 6900   if (UseBMI2Instructions) {
 6901     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 6902   } else { // !UseBMI2Instructions
 6903     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 6904   }
 6905 
 6906   pop(ylen);
 6907   pop(xlen);
 6908   pop(x);
 6909   pop(z);
 6910 
 6911   movl(tmp3, xlen);
 6912   addl(tmp3, 1);
 6913   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6914   subl(tmp3, 1);
 6915   jccb(Assembler::negative, L_done);
 6916 
 6917   shrq(carry, 32);
 6918   movl(Address(z, tmp3, Address::times_4,  0), carry);
 6919   jmp(L_second_loop);
 6920 
 6921   // Next infrequent code is moved outside loops.
 6922   bind(L_last_x);
 6923   if (UseBMI2Instructions) {
 6924     movl(rdx, Address(x,  0));
 6925   } else {
 6926     movl(x_xstart, Address(x,  0));
 6927   }
 6928   jmp(L_third_loop_prologue);
 6929 
 6930   bind(L_done);
 6931 
 6932   pop(xlen);
 6933 
 6934   pop(tmp5);
 6935   pop(tmp4);
 6936   pop(tmp3);
 6937   pop(tmp2);
 6938   pop(tmp1);
 6939   pop(tmp0);
 6940 }
 6941 
 6942 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 6943   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 6944   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 6945   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 6946   Label VECTOR8_TAIL, VECTOR4_TAIL;
 6947   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 6948   Label SAME_TILL_END, DONE;
 6949   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 6950 
 6951   //scale is in rcx in both Win64 and Unix
 6952   ShortBranchVerifier sbv(this);
 6953 
 6954   shlq(length);
 6955   xorq(result, result);
 6956 
 6957   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 6958       VM_Version::supports_avx512vlbw()) {
 6959     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 6960 
 6961     cmpq(length, 64);
 6962     jcc(Assembler::less, VECTOR32_TAIL);
 6963 
 6964     movq(tmp1, length);
 6965     andq(tmp1, 0x3F);      // tail count
 6966     andq(length, ~(0x3F)); //vector count
 6967 
 6968     bind(VECTOR64_LOOP);
 6969     // AVX512 code to compare 64 byte vectors.
 6970     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 6971     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6972     kortestql(k7, k7);
 6973     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 6974     addq(result, 64);
 6975     subq(length, 64);
 6976     jccb(Assembler::notZero, VECTOR64_LOOP);
 6977 
 6978     //bind(VECTOR64_TAIL);
 6979     testq(tmp1, tmp1);
 6980     jcc(Assembler::zero, SAME_TILL_END);
 6981 
 6982     //bind(VECTOR64_TAIL);
 6983     // AVX512 code to compare up to 63 byte vectors.
 6984     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 6985     shlxq(tmp2, tmp2, tmp1);
 6986     notq(tmp2);
 6987     kmovql(k3, tmp2);
 6988 
 6989     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 6990     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 6991 
 6992     ktestql(k7, k3);
 6993     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 6994 
 6995     bind(VECTOR64_NOT_EQUAL);
 6996     kmovql(tmp1, k7);
 6997     notq(tmp1);
 6998     tzcntq(tmp1, tmp1);
 6999     addq(result, tmp1);
 7000     shrq(result);
 7001     jmp(DONE);
 7002     bind(VECTOR32_TAIL);
 7003   }
 7004 
 7005   cmpq(length, 8);
 7006   jcc(Assembler::equal, VECTOR8_LOOP);
 7007   jcc(Assembler::less, VECTOR4_TAIL);
 7008 
 7009   if (UseAVX >= 2) {
 7010     Label VECTOR16_TAIL, VECTOR32_LOOP;
 7011 
 7012     cmpq(length, 16);
 7013     jcc(Assembler::equal, VECTOR16_LOOP);
 7014     jcc(Assembler::less, VECTOR8_LOOP);
 7015 
 7016     cmpq(length, 32);
 7017     jccb(Assembler::less, VECTOR16_TAIL);
 7018 
 7019     subq(length, 32);
 7020     bind(VECTOR32_LOOP);
 7021     vmovdqu(rymm0, Address(obja, result));
 7022     vmovdqu(rymm1, Address(objb, result));
 7023     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 7024     vptest(rymm2, rymm2);
 7025     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7026     addq(result, 32);
 7027     subq(length, 32);
 7028     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7029     addq(length, 32);
 7030     jcc(Assembler::equal, SAME_TILL_END);
 7031     //falling through if less than 32 bytes left //close the branch here.
 7032 
 7033     bind(VECTOR16_TAIL);
 7034     cmpq(length, 16);
 7035     jccb(Assembler::less, VECTOR8_TAIL);
 7036     bind(VECTOR16_LOOP);
 7037     movdqu(rymm0, Address(obja, result));
 7038     movdqu(rymm1, Address(objb, result));
 7039     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7040     ptest(rymm2, rymm2);
 7041     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7042     addq(result, 16);
 7043     subq(length, 16);
 7044     jcc(Assembler::equal, SAME_TILL_END);
 7045     //falling through if less than 16 bytes left
 7046   } else {//regular intrinsics
 7047 
 7048     cmpq(length, 16);
 7049     jccb(Assembler::less, VECTOR8_TAIL);
 7050 
 7051     subq(length, 16);
 7052     bind(VECTOR16_LOOP);
 7053     movdqu(rymm0, Address(obja, result));
 7054     movdqu(rymm1, Address(objb, result));
 7055     pxor(rymm0, rymm1);
 7056     ptest(rymm0, rymm0);
 7057     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7058     addq(result, 16);
 7059     subq(length, 16);
 7060     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7061     addq(length, 16);
 7062     jcc(Assembler::equal, SAME_TILL_END);
 7063     //falling through if less than 16 bytes left
 7064   }
 7065 
 7066   bind(VECTOR8_TAIL);
 7067   cmpq(length, 8);
 7068   jccb(Assembler::less, VECTOR4_TAIL);
 7069   bind(VECTOR8_LOOP);
 7070   movq(tmp1, Address(obja, result));
 7071   movq(tmp2, Address(objb, result));
 7072   xorq(tmp1, tmp2);
 7073   testq(tmp1, tmp1);
 7074   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7075   addq(result, 8);
 7076   subq(length, 8);
 7077   jcc(Assembler::equal, SAME_TILL_END);
 7078   //falling through if less than 8 bytes left
 7079 
 7080   bind(VECTOR4_TAIL);
 7081   cmpq(length, 4);
 7082   jccb(Assembler::less, BYTES_TAIL);
 7083   bind(VECTOR4_LOOP);
 7084   movl(tmp1, Address(obja, result));
 7085   xorl(tmp1, Address(objb, result));
 7086   testl(tmp1, tmp1);
 7087   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7088   addq(result, 4);
 7089   subq(length, 4);
 7090   jcc(Assembler::equal, SAME_TILL_END);
 7091   //falling through if less than 4 bytes left
 7092 
 7093   bind(BYTES_TAIL);
 7094   bind(BYTES_LOOP);
 7095   load_unsigned_byte(tmp1, Address(obja, result));
 7096   load_unsigned_byte(tmp2, Address(objb, result));
 7097   xorl(tmp1, tmp2);
 7098   testl(tmp1, tmp1);
 7099   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7100   decq(length);
 7101   jcc(Assembler::zero, SAME_TILL_END);
 7102   incq(result);
 7103   load_unsigned_byte(tmp1, Address(obja, result));
 7104   load_unsigned_byte(tmp2, Address(objb, result));
 7105   xorl(tmp1, tmp2);
 7106   testl(tmp1, tmp1);
 7107   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7108   decq(length);
 7109   jcc(Assembler::zero, SAME_TILL_END);
 7110   incq(result);
 7111   load_unsigned_byte(tmp1, Address(obja, result));
 7112   load_unsigned_byte(tmp2, Address(objb, result));
 7113   xorl(tmp1, tmp2);
 7114   testl(tmp1, tmp1);
 7115   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7116   jmp(SAME_TILL_END);
 7117 
 7118   if (UseAVX >= 2) {
 7119     bind(VECTOR32_NOT_EQUAL);
 7120     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7121     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7122     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7123     vpmovmskb(tmp1, rymm0);
 7124     bsfq(tmp1, tmp1);
 7125     addq(result, tmp1);
 7126     shrq(result);
 7127     jmp(DONE);
 7128   }
 7129 
 7130   bind(VECTOR16_NOT_EQUAL);
 7131   if (UseAVX >= 2) {
 7132     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7133     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7134     pxor(rymm0, rymm2);
 7135   } else {
 7136     pcmpeqb(rymm2, rymm2);
 7137     pxor(rymm0, rymm1);
 7138     pcmpeqb(rymm0, rymm1);
 7139     pxor(rymm0, rymm2);
 7140   }
 7141   pmovmskb(tmp1, rymm0);
 7142   bsfq(tmp1, tmp1);
 7143   addq(result, tmp1);
 7144   shrq(result);
 7145   jmpb(DONE);
 7146 
 7147   bind(VECTOR8_NOT_EQUAL);
 7148   bind(VECTOR4_NOT_EQUAL);
 7149   bsfq(tmp1, tmp1);
 7150   shrq(tmp1, 3);
 7151   addq(result, tmp1);
 7152   bind(BYTES_NOT_EQUAL);
 7153   shrq(result);
 7154   jmpb(DONE);
 7155 
 7156   bind(SAME_TILL_END);
 7157   mov64(result, -1);
 7158 
 7159   bind(DONE);
 7160 }
 7161 
 7162 //Helper functions for square_to_len()
 7163 
 7164 /**
 7165  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7166  * Preserves x and z and modifies rest of the registers.
 7167  */
 7168 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7169   // Perform square and right shift by 1
 7170   // Handle odd xlen case first, then for even xlen do the following
 7171   // jlong carry = 0;
 7172   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7173   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7174   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7175   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7176   //     carry = (jlong)product;
 7177   // }
 7178 
 7179   xorq(tmp5, tmp5);     // carry
 7180   xorq(rdxReg, rdxReg);
 7181   xorl(tmp1, tmp1);     // index for x
 7182   xorl(tmp4, tmp4);     // index for z
 7183 
 7184   Label L_first_loop, L_first_loop_exit;
 7185 
 7186   testl(xlen, 1);
 7187   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7188 
 7189   // Square and right shift by 1 the odd element using 32 bit multiply
 7190   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7191   imulq(raxReg, raxReg);
 7192   shrq(raxReg, 1);
 7193   adcq(tmp5, 0);
 7194   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7195   incrementl(tmp1);
 7196   addl(tmp4, 2);
 7197 
 7198   // Square and  right shift by 1 the rest using 64 bit multiply
 7199   bind(L_first_loop);
 7200   cmpptr(tmp1, xlen);
 7201   jccb(Assembler::equal, L_first_loop_exit);
 7202 
 7203   // Square
 7204   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7205   rorq(raxReg, 32);    // convert big-endian to little-endian
 7206   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7207 
 7208   // Right shift by 1 and save carry
 7209   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7210   rcrq(rdxReg, 1);
 7211   rcrq(raxReg, 1);
 7212   adcq(tmp5, 0);
 7213 
 7214   // Store result in z
 7215   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7216   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7217 
 7218   // Update indices for x and z
 7219   addl(tmp1, 2);
 7220   addl(tmp4, 4);
 7221   jmp(L_first_loop);
 7222 
 7223   bind(L_first_loop_exit);
 7224 }
 7225 
 7226 
 7227 /**
 7228  * Perform the following multiply add operation using BMI2 instructions
 7229  * carry:sum = sum + op1*op2 + carry
 7230  * op2 should be in rdx
 7231  * op2 is preserved, all other registers are modified
 7232  */
 7233 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7234   // assert op2 is rdx
 7235   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7236   addq(sum, carry);
 7237   adcq(tmp2, 0);
 7238   addq(sum, op1);
 7239   adcq(tmp2, 0);
 7240   movq(carry, tmp2);
 7241 }
 7242 
 7243 /**
 7244  * Perform the following multiply add operation:
 7245  * carry:sum = sum + op1*op2 + carry
 7246  * Preserves op1, op2 and modifies rest of registers
 7247  */
 7248 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7249   // rdx:rax = op1 * op2
 7250   movq(raxReg, op2);
 7251   mulq(op1);
 7252 
 7253   //  rdx:rax = sum + carry + rdx:rax
 7254   addq(sum, carry);
 7255   adcq(rdxReg, 0);
 7256   addq(sum, raxReg);
 7257   adcq(rdxReg, 0);
 7258 
 7259   // carry:sum = rdx:sum
 7260   movq(carry, rdxReg);
 7261 }
 7262 
 7263 /**
 7264  * Add 64 bit long carry into z[] with carry propagation.
 7265  * Preserves z and carry register values and modifies rest of registers.
 7266  *
 7267  */
 7268 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7269   Label L_fourth_loop, L_fourth_loop_exit;
 7270 
 7271   movl(tmp1, 1);
 7272   subl(zlen, 2);
 7273   addq(Address(z, zlen, Address::times_4, 0), carry);
 7274 
 7275   bind(L_fourth_loop);
 7276   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7277   subl(zlen, 2);
 7278   jccb(Assembler::negative, L_fourth_loop_exit);
 7279   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7280   jmp(L_fourth_loop);
 7281   bind(L_fourth_loop_exit);
 7282 }
 7283 
 7284 /**
 7285  * Shift z[] left by 1 bit.
 7286  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7287  *
 7288  */
 7289 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7290 
 7291   Label L_fifth_loop, L_fifth_loop_exit;
 7292 
 7293   // Fifth loop
 7294   // Perform primitiveLeftShift(z, zlen, 1)
 7295 
 7296   const Register prev_carry = tmp1;
 7297   const Register new_carry = tmp4;
 7298   const Register value = tmp2;
 7299   const Register zidx = tmp3;
 7300 
 7301   // int zidx, carry;
 7302   // long value;
 7303   // carry = 0;
 7304   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 7305   //    (carry:value)  = (z[i] << 1) | carry ;
 7306   //    z[i] = value;
 7307   // }
 7308 
 7309   movl(zidx, zlen);
 7310   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 7311 
 7312   bind(L_fifth_loop);
 7313   decl(zidx);  // Use decl to preserve carry flag
 7314   decl(zidx);
 7315   jccb(Assembler::negative, L_fifth_loop_exit);
 7316 
 7317   if (UseBMI2Instructions) {
 7318      movq(value, Address(z, zidx, Address::times_4, 0));
 7319      rclq(value, 1);
 7320      rorxq(value, value, 32);
 7321      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7322   }
 7323   else {
 7324     // clear new_carry
 7325     xorl(new_carry, new_carry);
 7326 
 7327     // Shift z[i] by 1, or in previous carry and save new carry
 7328     movq(value, Address(z, zidx, Address::times_4, 0));
 7329     shlq(value, 1);
 7330     adcl(new_carry, 0);
 7331 
 7332     orq(value, prev_carry);
 7333     rorq(value, 0x20);
 7334     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7335 
 7336     // Set previous carry = new carry
 7337     movl(prev_carry, new_carry);
 7338   }
 7339   jmp(L_fifth_loop);
 7340 
 7341   bind(L_fifth_loop_exit);
 7342 }
 7343 
 7344 
 7345 /**
 7346  * Code for BigInteger::squareToLen() intrinsic
 7347  *
 7348  * rdi: x
 7349  * rsi: len
 7350  * r8:  z
 7351  * rcx: zlen
 7352  * r12: tmp1
 7353  * r13: tmp2
 7354  * r14: tmp3
 7355  * r15: tmp4
 7356  * rbx: tmp5
 7357  *
 7358  */
 7359 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) {
 7360 
 7361   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 7362   push(tmp1);
 7363   push(tmp2);
 7364   push(tmp3);
 7365   push(tmp4);
 7366   push(tmp5);
 7367 
 7368   // First loop
 7369   // Store the squares, right shifted one bit (i.e., divided by 2).
 7370   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7371 
 7372   // Add in off-diagonal sums.
 7373   //
 7374   // Second, third (nested) and fourth loops.
 7375   // zlen +=2;
 7376   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 7377   //    carry = 0;
 7378   //    long op2 = x[xidx:xidx+1];
 7379   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 7380   //       k -= 2;
 7381   //       long op1 = x[j:j+1];
 7382   //       long sum = z[k:k+1];
 7383   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 7384   //       z[k:k+1] = sum;
 7385   //    }
 7386   //    add_one_64(z, k, carry, tmp_regs);
 7387   // }
 7388 
 7389   const Register carry = tmp5;
 7390   const Register sum = tmp3;
 7391   const Register op1 = tmp4;
 7392   Register op2 = tmp2;
 7393 
 7394   push(zlen);
 7395   push(len);
 7396   addl(zlen,2);
 7397   bind(L_second_loop);
 7398   xorq(carry, carry);
 7399   subl(zlen, 4);
 7400   subl(len, 2);
 7401   push(zlen);
 7402   push(len);
 7403   cmpl(len, 0);
 7404   jccb(Assembler::lessEqual, L_second_loop_exit);
 7405 
 7406   // Multiply an array by one 64 bit long.
 7407   if (UseBMI2Instructions) {
 7408     op2 = rdxReg;
 7409     movq(op2, Address(x, len, Address::times_4,  0));
 7410     rorxq(op2, op2, 32);
 7411   }
 7412   else {
 7413     movq(op2, Address(x, len, Address::times_4,  0));
 7414     rorq(op2, 32);
 7415   }
 7416 
 7417   bind(L_third_loop);
 7418   decrementl(len);
 7419   jccb(Assembler::negative, L_third_loop_exit);
 7420   decrementl(len);
 7421   jccb(Assembler::negative, L_last_x);
 7422 
 7423   movq(op1, Address(x, len, Address::times_4,  0));
 7424   rorq(op1, 32);
 7425 
 7426   bind(L_multiply);
 7427   subl(zlen, 2);
 7428   movq(sum, Address(z, zlen, Address::times_4,  0));
 7429 
 7430   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 7431   if (UseBMI2Instructions) {
 7432     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 7433   }
 7434   else {
 7435     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7436   }
 7437 
 7438   movq(Address(z, zlen, Address::times_4, 0), sum);
 7439 
 7440   jmp(L_third_loop);
 7441   bind(L_third_loop_exit);
 7442 
 7443   // Fourth loop
 7444   // Add 64 bit long carry into z with carry propagation.
 7445   // Uses offsetted zlen.
 7446   add_one_64(z, zlen, carry, tmp1);
 7447 
 7448   pop(len);
 7449   pop(zlen);
 7450   jmp(L_second_loop);
 7451 
 7452   // Next infrequent code is moved outside loops.
 7453   bind(L_last_x);
 7454   movl(op1, Address(x, 0));
 7455   jmp(L_multiply);
 7456 
 7457   bind(L_second_loop_exit);
 7458   pop(len);
 7459   pop(zlen);
 7460   pop(len);
 7461   pop(zlen);
 7462 
 7463   // Fifth loop
 7464   // Shift z left 1 bit.
 7465   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 7466 
 7467   // z[zlen-1] |= x[len-1] & 1;
 7468   movl(tmp3, Address(x, len, Address::times_4, -4));
 7469   andl(tmp3, 1);
 7470   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 7471 
 7472   pop(tmp5);
 7473   pop(tmp4);
 7474   pop(tmp3);
 7475   pop(tmp2);
 7476   pop(tmp1);
 7477 }
 7478 
 7479 /**
 7480  * Helper function for mul_add()
 7481  * Multiply the in[] by int k and add to out[] starting at offset offs using
 7482  * 128 bit by 32 bit multiply and return the carry in tmp5.
 7483  * Only quad int aligned length of in[] is operated on in this function.
 7484  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 7485  * This function preserves out, in and k registers.
 7486  * len and offset point to the appropriate index in "in" & "out" correspondingly
 7487  * tmp5 has the carry.
 7488  * other registers are temporary and are modified.
 7489  *
 7490  */
 7491 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 7492   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 7493   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7494 
 7495   Label L_first_loop, L_first_loop_exit;
 7496 
 7497   movl(tmp1, len);
 7498   shrl(tmp1, 2);
 7499 
 7500   bind(L_first_loop);
 7501   subl(tmp1, 1);
 7502   jccb(Assembler::negative, L_first_loop_exit);
 7503 
 7504   subl(len, 4);
 7505   subl(offset, 4);
 7506 
 7507   Register op2 = tmp2;
 7508   const Register sum = tmp3;
 7509   const Register op1 = tmp4;
 7510   const Register carry = tmp5;
 7511 
 7512   if (UseBMI2Instructions) {
 7513     op2 = rdxReg;
 7514   }
 7515 
 7516   movq(op1, Address(in, len, Address::times_4,  8));
 7517   rorq(op1, 32);
 7518   movq(sum, Address(out, offset, Address::times_4,  8));
 7519   rorq(sum, 32);
 7520   if (UseBMI2Instructions) {
 7521     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7522   }
 7523   else {
 7524     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7525   }
 7526   // Store back in big endian from little endian
 7527   rorq(sum, 0x20);
 7528   movq(Address(out, offset, Address::times_4,  8), sum);
 7529 
 7530   movq(op1, Address(in, len, Address::times_4,  0));
 7531   rorq(op1, 32);
 7532   movq(sum, Address(out, offset, Address::times_4,  0));
 7533   rorq(sum, 32);
 7534   if (UseBMI2Instructions) {
 7535     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7536   }
 7537   else {
 7538     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7539   }
 7540   // Store back in big endian from little endian
 7541   rorq(sum, 0x20);
 7542   movq(Address(out, offset, Address::times_4,  0), sum);
 7543 
 7544   jmp(L_first_loop);
 7545   bind(L_first_loop_exit);
 7546 }
 7547 
 7548 /**
 7549  * Code for BigInteger::mulAdd() intrinsic
 7550  *
 7551  * rdi: out
 7552  * rsi: in
 7553  * r11: offs (out.length - offset)
 7554  * rcx: len
 7555  * r8:  k
 7556  * r12: tmp1
 7557  * r13: tmp2
 7558  * r14: tmp3
 7559  * r15: tmp4
 7560  * rbx: tmp5
 7561  * Multiply the in[] by word k and add to out[], return the carry in rax
 7562  */
 7563 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 7564    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 7565    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7566 
 7567   Label L_carry, L_last_in, L_done;
 7568 
 7569 // carry = 0;
 7570 // for (int j=len-1; j >= 0; j--) {
 7571 //    long product = (in[j] & LONG_MASK) * kLong +
 7572 //                   (out[offs] & LONG_MASK) + carry;
 7573 //    out[offs--] = (int)product;
 7574 //    carry = product >>> 32;
 7575 // }
 7576 //
 7577   push(tmp1);
 7578   push(tmp2);
 7579   push(tmp3);
 7580   push(tmp4);
 7581   push(tmp5);
 7582 
 7583   Register op2 = tmp2;
 7584   const Register sum = tmp3;
 7585   const Register op1 = tmp4;
 7586   const Register carry =  tmp5;
 7587 
 7588   if (UseBMI2Instructions) {
 7589     op2 = rdxReg;
 7590     movl(op2, k);
 7591   }
 7592   else {
 7593     movl(op2, k);
 7594   }
 7595 
 7596   xorq(carry, carry);
 7597 
 7598   //First loop
 7599 
 7600   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 7601   //The carry is in tmp5
 7602   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 7603 
 7604   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 7605   decrementl(len);
 7606   jccb(Assembler::negative, L_carry);
 7607   decrementl(len);
 7608   jccb(Assembler::negative, L_last_in);
 7609 
 7610   movq(op1, Address(in, len, Address::times_4,  0));
 7611   rorq(op1, 32);
 7612 
 7613   subl(offs, 2);
 7614   movq(sum, Address(out, offs, Address::times_4,  0));
 7615   rorq(sum, 32);
 7616 
 7617   if (UseBMI2Instructions) {
 7618     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 7619   }
 7620   else {
 7621     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 7622   }
 7623 
 7624   // Store back in big endian from little endian
 7625   rorq(sum, 0x20);
 7626   movq(Address(out, offs, Address::times_4,  0), sum);
 7627 
 7628   testl(len, len);
 7629   jccb(Assembler::zero, L_carry);
 7630 
 7631   //Multiply the last in[] entry, if any
 7632   bind(L_last_in);
 7633   movl(op1, Address(in, 0));
 7634   movl(sum, Address(out, offs, Address::times_4,  -4));
 7635 
 7636   movl(raxReg, k);
 7637   mull(op1); //tmp4 * eax -> edx:eax
 7638   addl(sum, carry);
 7639   adcl(rdxReg, 0);
 7640   addl(sum, raxReg);
 7641   adcl(rdxReg, 0);
 7642   movl(carry, rdxReg);
 7643 
 7644   movl(Address(out, offs, Address::times_4,  -4), sum);
 7645 
 7646   bind(L_carry);
 7647   //return tmp5/carry as carry in rax
 7648   movl(rax, carry);
 7649 
 7650   bind(L_done);
 7651   pop(tmp5);
 7652   pop(tmp4);
 7653   pop(tmp3);
 7654   pop(tmp2);
 7655   pop(tmp1);
 7656 }
 7657 
 7658 /**
 7659  * Emits code to update CRC-32 with a byte value according to constants in table
 7660  *
 7661  * @param [in,out]crc   Register containing the crc.
 7662  * @param [in]val       Register containing the byte to fold into the CRC.
 7663  * @param [in]table     Register containing the table of crc constants.
 7664  *
 7665  * uint32_t crc;
 7666  * val = crc_table[(val ^ crc) & 0xFF];
 7667  * crc = val ^ (crc >> 8);
 7668  *
 7669  */
 7670 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 7671   xorl(val, crc);
 7672   andl(val, 0xFF);
 7673   shrl(crc, 8); // unsigned shift
 7674   xorl(crc, Address(table, val, Address::times_4, 0));
 7675 }
 7676 
 7677 /**
 7678  * Fold 128-bit data chunk
 7679  */
 7680 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 7681   if (UseAVX > 0) {
 7682     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 7683     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 7684     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 7685     pxor(xcrc, xtmp);
 7686   } else {
 7687     movdqa(xtmp, xcrc);
 7688     pclmulhdq(xtmp, xK);   // [123:64]
 7689     pclmulldq(xcrc, xK);   // [63:0]
 7690     pxor(xcrc, xtmp);
 7691     movdqu(xtmp, Address(buf, offset));
 7692     pxor(xcrc, xtmp);
 7693   }
 7694 }
 7695 
 7696 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 7697   if (UseAVX > 0) {
 7698     vpclmulhdq(xtmp, xK, xcrc);
 7699     vpclmulldq(xcrc, xK, xcrc);
 7700     pxor(xcrc, xbuf);
 7701     pxor(xcrc, xtmp);
 7702   } else {
 7703     movdqa(xtmp, xcrc);
 7704     pclmulhdq(xtmp, xK);
 7705     pclmulldq(xcrc, xK);
 7706     pxor(xcrc, xbuf);
 7707     pxor(xcrc, xtmp);
 7708   }
 7709 }
 7710 
 7711 /**
 7712  * 8-bit folds to compute 32-bit CRC
 7713  *
 7714  * uint64_t xcrc;
 7715  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 7716  */
 7717 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 7718   movdl(tmp, xcrc);
 7719   andl(tmp, 0xFF);
 7720   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 7721   psrldq(xcrc, 1); // unsigned shift one byte
 7722   pxor(xcrc, xtmp);
 7723 }
 7724 
 7725 /**
 7726  * uint32_t crc;
 7727  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 7728  */
 7729 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 7730   movl(tmp, crc);
 7731   andl(tmp, 0xFF);
 7732   shrl(crc, 8);
 7733   xorl(crc, Address(table, tmp, Address::times_4, 0));
 7734 }
 7735 
 7736 /**
 7737  * @param crc   register containing existing CRC (32-bit)
 7738  * @param buf   register pointing to input byte buffer (byte*)
 7739  * @param len   register containing number of bytes
 7740  * @param table register that will contain address of CRC table
 7741  * @param tmp   scratch register
 7742  */
 7743 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 7744   assert_different_registers(crc, buf, len, table, tmp, rax);
 7745 
 7746   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 7747   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 7748 
 7749   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 7750   // context for the registers used, where all instructions below are using 128-bit mode
 7751   // On EVEX without VL and BW, these instructions will all be AVX.
 7752   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 7753   notl(crc); // ~crc
 7754   cmpl(len, 16);
 7755   jcc(Assembler::less, L_tail);
 7756 
 7757   // Align buffer to 16 bytes
 7758   movl(tmp, buf);
 7759   andl(tmp, 0xF);
 7760   jccb(Assembler::zero, L_aligned);
 7761   subl(tmp,  16);
 7762   addl(len, tmp);
 7763 
 7764   align(4);
 7765   BIND(L_align_loop);
 7766   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7767   update_byte_crc32(crc, rax, table);
 7768   increment(buf);
 7769   incrementl(tmp);
 7770   jccb(Assembler::less, L_align_loop);
 7771 
 7772   BIND(L_aligned);
 7773   movl(tmp, len); // save
 7774   shrl(len, 4);
 7775   jcc(Assembler::zero, L_tail_restore);
 7776 
 7777   // Fold crc into first bytes of vector
 7778   movdqa(xmm1, Address(buf, 0));
 7779   movdl(rax, xmm1);
 7780   xorl(crc, rax);
 7781   if (VM_Version::supports_sse4_1()) {
 7782     pinsrd(xmm1, crc, 0);
 7783   } else {
 7784     pinsrw(xmm1, crc, 0);
 7785     shrl(crc, 16);
 7786     pinsrw(xmm1, crc, 1);
 7787   }
 7788   addptr(buf, 16);
 7789   subl(len, 4); // len > 0
 7790   jcc(Assembler::less, L_fold_tail);
 7791 
 7792   movdqa(xmm2, Address(buf,  0));
 7793   movdqa(xmm3, Address(buf, 16));
 7794   movdqa(xmm4, Address(buf, 32));
 7795   addptr(buf, 48);
 7796   subl(len, 3);
 7797   jcc(Assembler::lessEqual, L_fold_512b);
 7798 
 7799   // Fold total 512 bits of polynomial on each iteration,
 7800   // 128 bits per each of 4 parallel streams.
 7801   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 7802 
 7803   align32();
 7804   BIND(L_fold_512b_loop);
 7805   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7806   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 7807   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 7808   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 7809   addptr(buf, 64);
 7810   subl(len, 4);
 7811   jcc(Assembler::greater, L_fold_512b_loop);
 7812 
 7813   // Fold 512 bits to 128 bits.
 7814   BIND(L_fold_512b);
 7815   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7816   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 7817   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 7818   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 7819 
 7820   // Fold the rest of 128 bits data chunks
 7821   BIND(L_fold_tail);
 7822   addl(len, 3);
 7823   jccb(Assembler::lessEqual, L_fold_128b);
 7824   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 7825 
 7826   BIND(L_fold_tail_loop);
 7827   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 7828   addptr(buf, 16);
 7829   decrementl(len);
 7830   jccb(Assembler::greater, L_fold_tail_loop);
 7831 
 7832   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 7833   BIND(L_fold_128b);
 7834   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 7835   if (UseAVX > 0) {
 7836     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 7837     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 7838     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 7839   } else {
 7840     movdqa(xmm2, xmm0);
 7841     pclmulqdq(xmm2, xmm1, 0x1);
 7842     movdqa(xmm3, xmm0);
 7843     pand(xmm3, xmm2);
 7844     pclmulqdq(xmm0, xmm3, 0x1);
 7845   }
 7846   psrldq(xmm1, 8);
 7847   psrldq(xmm2, 4);
 7848   pxor(xmm0, xmm1);
 7849   pxor(xmm0, xmm2);
 7850 
 7851   // 8 8-bit folds to compute 32-bit CRC.
 7852   for (int j = 0; j < 4; j++) {
 7853     fold_8bit_crc32(xmm0, table, xmm1, rax);
 7854   }
 7855   movdl(crc, xmm0); // mov 32 bits to general register
 7856   for (int j = 0; j < 4; j++) {
 7857     fold_8bit_crc32(crc, table, rax);
 7858   }
 7859 
 7860   BIND(L_tail_restore);
 7861   movl(len, tmp); // restore
 7862   BIND(L_tail);
 7863   andl(len, 0xf);
 7864   jccb(Assembler::zero, L_exit);
 7865 
 7866   // Fold the rest of bytes
 7867   align(4);
 7868   BIND(L_tail_loop);
 7869   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 7870   update_byte_crc32(crc, rax, table);
 7871   increment(buf);
 7872   decrementl(len);
 7873   jccb(Assembler::greater, L_tail_loop);
 7874 
 7875   BIND(L_exit);
 7876   notl(crc); // ~c
 7877 }
 7878 
 7879 // Helper function for AVX 512 CRC32
 7880 // Fold 512-bit data chunks
 7881 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 7882                                              Register pos, int offset) {
 7883   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 7884   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 7885   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 7886   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 7887   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 7888 }
 7889 
 7890 // Helper function for AVX 512 CRC32
 7891 // Compute CRC32 for < 256B buffers
 7892 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 7893                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 7894                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 7895 
 7896   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 7897   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 7898   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 7899 
 7900   // check if there is enough buffer to be able to fold 16B at a time
 7901   cmpl(len, 32);
 7902   jcc(Assembler::less, L_less_than_32);
 7903 
 7904   // if there is, load the constants
 7905   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 7906   movdl(xmm0, crc);                        // get the initial crc value
 7907   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7908   pxor(xmm7, xmm0);
 7909 
 7910   // update the buffer pointer
 7911   addl(pos, 16);
 7912   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 7913   subl(len, 32);
 7914   jmp(L_16B_reduction_loop);
 7915 
 7916   bind(L_less_than_32);
 7917   //mov initial crc to the return value. this is necessary for zero - length buffers.
 7918   movl(rax, crc);
 7919   testl(len, len);
 7920   jcc(Assembler::equal, L_cleanup);
 7921 
 7922   movdl(xmm0, crc);                        //get the initial crc value
 7923 
 7924   cmpl(len, 16);
 7925   jcc(Assembler::equal, L_exact_16_left);
 7926   jcc(Assembler::less, L_less_than_16_left);
 7927 
 7928   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 7929   pxor(xmm7, xmm0);                       //xor the initial crc value
 7930   addl(pos, 16);
 7931   subl(len, 16);
 7932   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 7933   jmp(L_get_last_two_xmms);
 7934 
 7935   bind(L_less_than_16_left);
 7936   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 7937   pxor(xmm1, xmm1);
 7938   movptr(tmp1, rsp);
 7939   movdqu(Address(tmp1, 0 * 16), xmm1);
 7940 
 7941   cmpl(len, 4);
 7942   jcc(Assembler::less, L_only_less_than_4);
 7943 
 7944   //backup the counter value
 7945   movl(tmp2, len);
 7946   cmpl(len, 8);
 7947   jcc(Assembler::less, L_less_than_8_left);
 7948 
 7949   //load 8 Bytes
 7950   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 7951   movq(Address(tmp1, 0 * 16), rax);
 7952   addptr(tmp1, 8);
 7953   subl(len, 8);
 7954   addl(pos, 8);
 7955 
 7956   bind(L_less_than_8_left);
 7957   cmpl(len, 4);
 7958   jcc(Assembler::less, L_less_than_4_left);
 7959 
 7960   //load 4 Bytes
 7961   movl(rax, Address(buf, pos, Address::times_1, 0));
 7962   movl(Address(tmp1, 0 * 16), rax);
 7963   addptr(tmp1, 4);
 7964   subl(len, 4);
 7965   addl(pos, 4);
 7966 
 7967   bind(L_less_than_4_left);
 7968   cmpl(len, 2);
 7969   jcc(Assembler::less, L_less_than_2_left);
 7970 
 7971   // load 2 Bytes
 7972   movw(rax, Address(buf, pos, Address::times_1, 0));
 7973   movl(Address(tmp1, 0 * 16), rax);
 7974   addptr(tmp1, 2);
 7975   subl(len, 2);
 7976   addl(pos, 2);
 7977 
 7978   bind(L_less_than_2_left);
 7979   cmpl(len, 1);
 7980   jcc(Assembler::less, L_zero_left);
 7981 
 7982   // load 1 Byte
 7983   movb(rax, Address(buf, pos, Address::times_1, 0));
 7984   movb(Address(tmp1, 0 * 16), rax);
 7985 
 7986   bind(L_zero_left);
 7987   movdqu(xmm7, Address(rsp, 0));
 7988   pxor(xmm7, xmm0);                       //xor the initial crc value
 7989 
 7990   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 7991   movdqu(xmm0, Address(rax, tmp2));
 7992   pshufb(xmm7, xmm0);
 7993   jmp(L_128_done);
 7994 
 7995   bind(L_exact_16_left);
 7996   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 7997   pxor(xmm7, xmm0);                       //xor the initial crc value
 7998   jmp(L_128_done);
 7999 
 8000   bind(L_only_less_than_4);
 8001   cmpl(len, 3);
 8002   jcc(Assembler::less, L_only_less_than_3);
 8003 
 8004   // load 3 Bytes
 8005   movb(rax, Address(buf, pos, Address::times_1, 0));
 8006   movb(Address(tmp1, 0), rax);
 8007 
 8008   movb(rax, Address(buf, pos, Address::times_1, 1));
 8009   movb(Address(tmp1, 1), rax);
 8010 
 8011   movb(rax, Address(buf, pos, Address::times_1, 2));
 8012   movb(Address(tmp1, 2), rax);
 8013 
 8014   movdqu(xmm7, Address(rsp, 0));
 8015   pxor(xmm7, xmm0);                     //xor the initial crc value
 8016 
 8017   pslldq(xmm7, 0x5);
 8018   jmp(L_barrett);
 8019   bind(L_only_less_than_3);
 8020   cmpl(len, 2);
 8021   jcc(Assembler::less, L_only_less_than_2);
 8022 
 8023   // load 2 Bytes
 8024   movb(rax, Address(buf, pos, Address::times_1, 0));
 8025   movb(Address(tmp1, 0), rax);
 8026 
 8027   movb(rax, Address(buf, pos, Address::times_1, 1));
 8028   movb(Address(tmp1, 1), rax);
 8029 
 8030   movdqu(xmm7, Address(rsp, 0));
 8031   pxor(xmm7, xmm0);                     //xor the initial crc value
 8032 
 8033   pslldq(xmm7, 0x6);
 8034   jmp(L_barrett);
 8035 
 8036   bind(L_only_less_than_2);
 8037   //load 1 Byte
 8038   movb(rax, Address(buf, pos, Address::times_1, 0));
 8039   movb(Address(tmp1, 0), rax);
 8040 
 8041   movdqu(xmm7, Address(rsp, 0));
 8042   pxor(xmm7, xmm0);                     //xor the initial crc value
 8043 
 8044   pslldq(xmm7, 0x7);
 8045 }
 8046 
 8047 /**
 8048 * Compute CRC32 using AVX512 instructions
 8049 * param crc   register containing existing CRC (32-bit)
 8050 * param buf   register pointing to input byte buffer (byte*)
 8051 * param len   register containing number of bytes
 8052 * param table address of crc or crc32c table
 8053 * param tmp1  scratch register
 8054 * param tmp2  scratch register
 8055 * return rax  result register
 8056 *
 8057 * This routine is identical for crc32c with the exception of the precomputed constant
 8058 * table which will be passed as the table argument.  The calculation steps are
 8059 * the same for both variants.
 8060 */
 8061 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8062   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8063 
 8064   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8065   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8066   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8067   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8068   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8069 
 8070   const Register pos = r12;
 8071   push(r12);
 8072   subptr(rsp, 16 * 2 + 8);
 8073 
 8074   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8075   // context for the registers used, where all instructions below are using 128-bit mode
 8076   // On EVEX without VL and BW, these instructions will all be AVX.
 8077   movl(pos, 0);
 8078 
 8079   // check if smaller than 256B
 8080   cmpl(len, 256);
 8081   jcc(Assembler::less, L_less_than_256);
 8082 
 8083   // load the initial crc value
 8084   movdl(xmm10, crc);
 8085 
 8086   // receive the initial 64B data, xor the initial crc value
 8087   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8088   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8089   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8090   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8091 
 8092   subl(len, 256);
 8093   cmpl(len, 256);
 8094   jcc(Assembler::less, L_fold_128_B_loop);
 8095 
 8096   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8097   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8098   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8099   subl(len, 256);
 8100 
 8101   bind(L_fold_256_B_loop);
 8102   addl(pos, 256);
 8103   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8104   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8105   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8106   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8107 
 8108   subl(len, 256);
 8109   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8110 
 8111   // Fold 256 into 128
 8112   addl(pos, 256);
 8113   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8114   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8115   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8116 
 8117   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8118   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8119   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8120 
 8121   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8122   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8123 
 8124   addl(len, 128);
 8125   jmp(L_fold_128_B_register);
 8126 
 8127   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8128   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8129 
 8130   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8131   bind(L_fold_128_B_loop);
 8132   addl(pos, 128);
 8133   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8134   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8135 
 8136   subl(len, 128);
 8137   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8138 
 8139   addl(pos, 128);
 8140 
 8141   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8142   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8143   bind(L_fold_128_B_register);
 8144   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8145   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8146   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8147   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8148   // save last that has no multiplicand
 8149   vextracti64x2(xmm7, xmm4, 3);
 8150 
 8151   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8152   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8153   // Needed later in reduction loop
 8154   movdqu(xmm10, Address(table, 1 * 16));
 8155   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8156   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8157 
 8158   // Swap 1,0,3,2 - 01 00 11 10
 8159   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8160   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8161   vextracti128(xmm5, xmm8, 1);
 8162   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8163 
 8164   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8165   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8166   addl(len, 128 - 16);
 8167   jcc(Assembler::less, L_final_reduction_for_128);
 8168 
 8169   bind(L_16B_reduction_loop);
 8170   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8171   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8172   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8173   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8174   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8175   addl(pos, 16);
 8176   subl(len, 16);
 8177   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8178 
 8179   bind(L_final_reduction_for_128);
 8180   addl(len, 16);
 8181   jcc(Assembler::equal, L_128_done);
 8182 
 8183   bind(L_get_last_two_xmms);
 8184   movdqu(xmm2, xmm7);
 8185   addl(pos, len);
 8186   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8187   subl(pos, len);
 8188 
 8189   // get rid of the extra data that was loaded before
 8190   // load the shift constant
 8191   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8192   movdqu(xmm0, Address(rax, len));
 8193   addl(rax, len);
 8194 
 8195   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8196   //Change mask to 512
 8197   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8198   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8199 
 8200   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8201   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8202   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8203   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8204   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8205 
 8206   bind(L_128_done);
 8207   // compute crc of a 128-bit value
 8208   movdqu(xmm10, Address(table, 3 * 16));
 8209   movdqu(xmm0, xmm7);
 8210 
 8211   // 64b fold
 8212   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8213   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8214   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8215 
 8216   // 32b fold
 8217   movdqu(xmm0, xmm7);
 8218   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8219   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8220   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8221   jmp(L_barrett);
 8222 
 8223   bind(L_less_than_256);
 8224   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);
 8225 
 8226   //barrett reduction
 8227   bind(L_barrett);
 8228   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8229   movdqu(xmm1, xmm7);
 8230   movdqu(xmm2, xmm7);
 8231   movdqu(xmm10, Address(table, 4 * 16));
 8232 
 8233   pclmulqdq(xmm7, xmm10, 0x0);
 8234   pxor(xmm7, xmm2);
 8235   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8236   movdqu(xmm2, xmm7);
 8237   pclmulqdq(xmm7, xmm10, 0x10);
 8238   pxor(xmm7, xmm2);
 8239   pxor(xmm7, xmm1);
 8240   pextrd(crc, xmm7, 2);
 8241 
 8242   bind(L_cleanup);
 8243   addptr(rsp, 16 * 2 + 8);
 8244   pop(r12);
 8245 }
 8246 
 8247 // S. Gueron / Information Processing Letters 112 (2012) 184
 8248 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8249 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8250 // Output: the 64-bit carry-less product of B * CONST
 8251 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8252                                      Register tmp1, Register tmp2, Register tmp3) {
 8253   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8254   if (n > 0) {
 8255     addq(tmp3, n * 256 * 8);
 8256   }
 8257   //    Q1 = TABLEExt[n][B & 0xFF];
 8258   movl(tmp1, in);
 8259   andl(tmp1, 0x000000FF);
 8260   shll(tmp1, 3);
 8261   addq(tmp1, tmp3);
 8262   movq(tmp1, Address(tmp1, 0));
 8263 
 8264   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8265   movl(tmp2, in);
 8266   shrl(tmp2, 8);
 8267   andl(tmp2, 0x000000FF);
 8268   shll(tmp2, 3);
 8269   addq(tmp2, tmp3);
 8270   movq(tmp2, Address(tmp2, 0));
 8271 
 8272   shlq(tmp2, 8);
 8273   xorq(tmp1, tmp2);
 8274 
 8275   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8276   movl(tmp2, in);
 8277   shrl(tmp2, 16);
 8278   andl(tmp2, 0x000000FF);
 8279   shll(tmp2, 3);
 8280   addq(tmp2, tmp3);
 8281   movq(tmp2, Address(tmp2, 0));
 8282 
 8283   shlq(tmp2, 16);
 8284   xorq(tmp1, tmp2);
 8285 
 8286   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8287   shrl(in, 24);
 8288   andl(in, 0x000000FF);
 8289   shll(in, 3);
 8290   addq(in, tmp3);
 8291   movq(in, Address(in, 0));
 8292 
 8293   shlq(in, 24);
 8294   xorq(in, tmp1);
 8295   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8296 }
 8297 
 8298 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8299                                       Register in_out,
 8300                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8301                                       XMMRegister w_xtmp2,
 8302                                       Register tmp1,
 8303                                       Register n_tmp2, Register n_tmp3) {
 8304   if (is_pclmulqdq_supported) {
 8305     movdl(w_xtmp1, in_out); // modified blindly
 8306 
 8307     movl(tmp1, const_or_pre_comp_const_index);
 8308     movdl(w_xtmp2, tmp1);
 8309     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8310 
 8311     movdq(in_out, w_xtmp1);
 8312   } else {
 8313     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 8314   }
 8315 }
 8316 
 8317 // Recombination Alternative 2: No bit-reflections
 8318 // T1 = (CRC_A * U1) << 1
 8319 // T2 = (CRC_B * U2) << 1
 8320 // C1 = T1 >> 32
 8321 // C2 = T2 >> 32
 8322 // T1 = T1 & 0xFFFFFFFF
 8323 // T2 = T2 & 0xFFFFFFFF
 8324 // T1 = CRC32(0, T1)
 8325 // T2 = CRC32(0, T2)
 8326 // C1 = C1 ^ T1
 8327 // C2 = C2 ^ T2
 8328 // CRC = C1 ^ C2 ^ CRC_C
 8329 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,
 8330                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8331                                      Register tmp1, Register tmp2,
 8332                                      Register n_tmp3) {
 8333   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8334   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8335   shlq(in_out, 1);
 8336   movl(tmp1, in_out);
 8337   shrq(in_out, 32);
 8338   xorl(tmp2, tmp2);
 8339   crc32(tmp2, tmp1, 4);
 8340   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 8341   shlq(in1, 1);
 8342   movl(tmp1, in1);
 8343   shrq(in1, 32);
 8344   xorl(tmp2, tmp2);
 8345   crc32(tmp2, tmp1, 4);
 8346   xorl(in1, tmp2);
 8347   xorl(in_out, in1);
 8348   xorl(in_out, in2);
 8349 }
 8350 
 8351 // Set N to predefined value
 8352 // Subtract from a length of a buffer
 8353 // execute in a loop:
 8354 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 8355 // for i = 1 to N do
 8356 //  CRC_A = CRC32(CRC_A, A[i])
 8357 //  CRC_B = CRC32(CRC_B, B[i])
 8358 //  CRC_C = CRC32(CRC_C, C[i])
 8359 // end for
 8360 // Recombine
 8361 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,
 8362                                        Register in_out1, Register in_out2, Register in_out3,
 8363                                        Register tmp1, Register tmp2, Register tmp3,
 8364                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8365                                        Register tmp4, Register tmp5,
 8366                                        Register n_tmp6) {
 8367   Label L_processPartitions;
 8368   Label L_processPartition;
 8369   Label L_exit;
 8370 
 8371   bind(L_processPartitions);
 8372   cmpl(in_out1, 3 * size);
 8373   jcc(Assembler::less, L_exit);
 8374     xorl(tmp1, tmp1);
 8375     xorl(tmp2, tmp2);
 8376     movq(tmp3, in_out2);
 8377     addq(tmp3, size);
 8378 
 8379     bind(L_processPartition);
 8380       crc32(in_out3, Address(in_out2, 0), 8);
 8381       crc32(tmp1, Address(in_out2, size), 8);
 8382       crc32(tmp2, Address(in_out2, size * 2), 8);
 8383       addq(in_out2, 8);
 8384       cmpq(in_out2, tmp3);
 8385       jcc(Assembler::less, L_processPartition);
 8386     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 8387             w_xtmp1, w_xtmp2, w_xtmp3,
 8388             tmp4, tmp5,
 8389             n_tmp6);
 8390     addq(in_out2, 2 * size);
 8391     subl(in_out1, 3 * size);
 8392     jmp(L_processPartitions);
 8393 
 8394   bind(L_exit);
 8395 }
 8396 
 8397 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 8398 // Input: A buffer I of L bytes.
 8399 // Output: the CRC32C value of the buffer.
 8400 // Notations:
 8401 // Write L = 24N + r, with N = floor (L/24).
 8402 // r = L mod 24 (0 <= r < 24).
 8403 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 8404 // N quadwords, and R consists of r bytes.
 8405 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 8406 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 8407 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 8408 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 8409 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 8410                                           Register tmp1, Register tmp2, Register tmp3,
 8411                                           Register tmp4, Register tmp5, Register tmp6,
 8412                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8413                                           bool is_pclmulqdq_supported) {
 8414   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 8415   Label L_wordByWord;
 8416   Label L_byteByByteProlog;
 8417   Label L_byteByByte;
 8418   Label L_exit;
 8419 
 8420   if (is_pclmulqdq_supported ) {
 8421     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 8422     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 8423 
 8424     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 8425     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 8426 
 8427     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 8428     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 8429     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 8430   } else {
 8431     const_or_pre_comp_const_index[0] = 1;
 8432     const_or_pre_comp_const_index[1] = 0;
 8433 
 8434     const_or_pre_comp_const_index[2] = 3;
 8435     const_or_pre_comp_const_index[3] = 2;
 8436 
 8437     const_or_pre_comp_const_index[4] = 5;
 8438     const_or_pre_comp_const_index[5] = 4;
 8439    }
 8440   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 8441                     in2, in1, in_out,
 8442                     tmp1, tmp2, tmp3,
 8443                     w_xtmp1, w_xtmp2, w_xtmp3,
 8444                     tmp4, tmp5,
 8445                     tmp6);
 8446   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 8447                     in2, in1, in_out,
 8448                     tmp1, tmp2, tmp3,
 8449                     w_xtmp1, w_xtmp2, w_xtmp3,
 8450                     tmp4, tmp5,
 8451                     tmp6);
 8452   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 8453                     in2, in1, in_out,
 8454                     tmp1, tmp2, tmp3,
 8455                     w_xtmp1, w_xtmp2, w_xtmp3,
 8456                     tmp4, tmp5,
 8457                     tmp6);
 8458   movl(tmp1, in2);
 8459   andl(tmp1, 0x00000007);
 8460   negl(tmp1);
 8461   addl(tmp1, in2);
 8462   addq(tmp1, in1);
 8463 
 8464   cmpq(in1, tmp1);
 8465   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 8466   align(16);
 8467   BIND(L_wordByWord);
 8468     crc32(in_out, Address(in1, 0), 8);
 8469     addq(in1, 8);
 8470     cmpq(in1, tmp1);
 8471     jcc(Assembler::less, L_wordByWord);
 8472 
 8473   BIND(L_byteByByteProlog);
 8474   andl(in2, 0x00000007);
 8475   movl(tmp2, 1);
 8476 
 8477   cmpl(tmp2, in2);
 8478   jccb(Assembler::greater, L_exit);
 8479   BIND(L_byteByByte);
 8480     crc32(in_out, Address(in1, 0), 1);
 8481     incq(in1);
 8482     incl(tmp2);
 8483     cmpl(tmp2, in2);
 8484     jcc(Assembler::lessEqual, L_byteByByte);
 8485 
 8486   BIND(L_exit);
 8487 }
 8488 #undef BIND
 8489 #undef BLOCK_COMMENT
 8490 
 8491 // Compress char[] array to byte[].
 8492 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 8493 // Return the array length if every element in array can be encoded,
 8494 // otherwise, the index of first non-latin1 (> 0xff) character.
 8495 //   @IntrinsicCandidate
 8496 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 8497 //     for (int i = 0; i < len; i++) {
 8498 //       char c = src[srcOff];
 8499 //       if (c > 0xff) {
 8500 //           return i;  // return index of non-latin1 char
 8501 //       }
 8502 //       dst[dstOff] = (byte)c;
 8503 //       srcOff++;
 8504 //       dstOff++;
 8505 //     }
 8506 //     return len;
 8507 //   }
 8508 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 8509   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 8510   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 8511   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 8512   Label copy_chars_loop, done, reset_sp, copy_tail;
 8513 
 8514   // rsi: src
 8515   // rdi: dst
 8516   // rdx: len
 8517   // rcx: tmp5
 8518   // rax: result
 8519 
 8520   // rsi holds start addr of source char[] to be compressed
 8521   // rdi holds start addr of destination byte[]
 8522   // rdx holds length
 8523 
 8524   assert(len != result, "");
 8525 
 8526   // save length for return
 8527   movl(result, len);
 8528 
 8529   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 8530     VM_Version::supports_avx512vlbw() &&
 8531     VM_Version::supports_bmi2()) {
 8532 
 8533     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 8534 
 8535     // alignment
 8536     Label post_alignment;
 8537 
 8538     // if length of the string is less than 32, handle it the old fashioned way
 8539     testl(len, -32);
 8540     jcc(Assembler::zero, below_threshold);
 8541 
 8542     // First check whether a character is compressible ( <= 0xFF).
 8543     // Create mask to test for Unicode chars inside zmm vector
 8544     movl(tmp5, 0x00FF);
 8545     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 8546 
 8547     testl(len, -64);
 8548     jccb(Assembler::zero, post_alignment);
 8549 
 8550     movl(tmp5, dst);
 8551     andl(tmp5, (32 - 1));
 8552     negl(tmp5);
 8553     andl(tmp5, (32 - 1));
 8554 
 8555     // bail out when there is nothing to be done
 8556     testl(tmp5, 0xFFFFFFFF);
 8557     jccb(Assembler::zero, post_alignment);
 8558 
 8559     // ~(~0 << len), where len is the # of remaining elements to process
 8560     movl(len, 0xFFFFFFFF);
 8561     shlxl(len, len, tmp5);
 8562     notl(len);
 8563     kmovdl(mask2, len);
 8564     movl(len, result);
 8565 
 8566     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8567     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8568     ktestd(mask1, mask2);
 8569     jcc(Assembler::carryClear, copy_tail);
 8570 
 8571     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8572 
 8573     addptr(src, tmp5);
 8574     addptr(src, tmp5);
 8575     addptr(dst, tmp5);
 8576     subl(len, tmp5);
 8577 
 8578     bind(post_alignment);
 8579     // end of alignment
 8580 
 8581     movl(tmp5, len);
 8582     andl(tmp5, (32 - 1));    // tail count (in chars)
 8583     andl(len, ~(32 - 1));    // vector count (in chars)
 8584     jccb(Assembler::zero, copy_loop_tail);
 8585 
 8586     lea(src, Address(src, len, Address::times_2));
 8587     lea(dst, Address(dst, len, Address::times_1));
 8588     negptr(len);
 8589 
 8590     bind(copy_32_loop);
 8591     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 8592     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 8593     kortestdl(mask1, mask1);
 8594     jccb(Assembler::carryClear, reset_for_copy_tail);
 8595 
 8596     // All elements in current processed chunk are valid candidates for
 8597     // compression. Write a truncated byte elements to the memory.
 8598     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 8599     addptr(len, 32);
 8600     jccb(Assembler::notZero, copy_32_loop);
 8601 
 8602     bind(copy_loop_tail);
 8603     // bail out when there is nothing to be done
 8604     testl(tmp5, 0xFFFFFFFF);
 8605     jcc(Assembler::zero, done);
 8606 
 8607     movl(len, tmp5);
 8608 
 8609     // ~(~0 << len), where len is the # of remaining elements to process
 8610     movl(tmp5, 0xFFFFFFFF);
 8611     shlxl(tmp5, tmp5, len);
 8612     notl(tmp5);
 8613 
 8614     kmovdl(mask2, tmp5);
 8615 
 8616     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 8617     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 8618     ktestd(mask1, mask2);
 8619     jcc(Assembler::carryClear, copy_tail);
 8620 
 8621     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 8622     jmp(done);
 8623 
 8624     bind(reset_for_copy_tail);
 8625     lea(src, Address(src, tmp5, Address::times_2));
 8626     lea(dst, Address(dst, tmp5, Address::times_1));
 8627     subptr(len, tmp5);
 8628     jmp(copy_chars_loop);
 8629 
 8630     bind(below_threshold);
 8631   }
 8632 
 8633   if (UseSSE42Intrinsics) {
 8634     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 8635 
 8636     // vectored compression
 8637     testl(len, 0xfffffff8);
 8638     jcc(Assembler::zero, copy_tail);
 8639 
 8640     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 8641     movdl(tmp1Reg, tmp5);
 8642     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 8643 
 8644     andl(len, 0xfffffff0);
 8645     jccb(Assembler::zero, copy_16);
 8646 
 8647     // compress 16 chars per iter
 8648     pxor(tmp4Reg, tmp4Reg);
 8649 
 8650     lea(src, Address(src, len, Address::times_2));
 8651     lea(dst, Address(dst, len, Address::times_1));
 8652     negptr(len);
 8653 
 8654     bind(copy_32_loop);
 8655     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 8656     por(tmp4Reg, tmp2Reg);
 8657     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 8658     por(tmp4Reg, tmp3Reg);
 8659     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 8660     jccb(Assembler::notZero, reset_for_copy_tail);
 8661     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 8662     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 8663     addptr(len, 16);
 8664     jccb(Assembler::notZero, copy_32_loop);
 8665 
 8666     // compress next vector of 8 chars (if any)
 8667     bind(copy_16);
 8668     // len = 0
 8669     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 8670     jccb(Assembler::zero, copy_tail_sse);
 8671 
 8672     pxor(tmp3Reg, tmp3Reg);
 8673 
 8674     movdqu(tmp2Reg, Address(src, 0));
 8675     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 8676     jccb(Assembler::notZero, reset_for_copy_tail);
 8677     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 8678     movq(Address(dst, 0), tmp2Reg);
 8679     addptr(src, 16);
 8680     addptr(dst, 8);
 8681     jmpb(copy_tail_sse);
 8682 
 8683     bind(reset_for_copy_tail);
 8684     movl(tmp5, result);
 8685     andl(tmp5, 0x0000000f);
 8686     lea(src, Address(src, tmp5, Address::times_2));
 8687     lea(dst, Address(dst, tmp5, Address::times_1));
 8688     subptr(len, tmp5);
 8689     jmpb(copy_chars_loop);
 8690 
 8691     bind(copy_tail_sse);
 8692     movl(len, result);
 8693     andl(len, 0x00000007);    // tail count (in chars)
 8694   }
 8695   // compress 1 char per iter
 8696   bind(copy_tail);
 8697   testl(len, len);
 8698   jccb(Assembler::zero, done);
 8699   lea(src, Address(src, len, Address::times_2));
 8700   lea(dst, Address(dst, len, Address::times_1));
 8701   negptr(len);
 8702 
 8703   bind(copy_chars_loop);
 8704   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 8705   testl(tmp5, 0xff00);      // check if Unicode char
 8706   jccb(Assembler::notZero, reset_sp);
 8707   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 8708   increment(len);
 8709   jccb(Assembler::notZero, copy_chars_loop);
 8710 
 8711   // add len then return (len will be zero if compress succeeded, otherwise negative)
 8712   bind(reset_sp);
 8713   addl(result, len);
 8714 
 8715   bind(done);
 8716 }
 8717 
 8718 // Inflate byte[] array to char[].
 8719 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 8720 //   @IntrinsicCandidate
 8721 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 8722 //     for (int i = 0; i < len; i++) {
 8723 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 8724 //     }
 8725 //   }
 8726 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 8727   XMMRegister tmp1, Register tmp2, KRegister mask) {
 8728   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 8729   // rsi: src
 8730   // rdi: dst
 8731   // rdx: len
 8732   // rcx: tmp2
 8733 
 8734   // rsi holds start addr of source byte[] to be inflated
 8735   // rdi holds start addr of destination char[]
 8736   // rdx holds length
 8737   assert_different_registers(src, dst, len, tmp2);
 8738   movl(tmp2, len);
 8739   if ((UseAVX > 2) && // AVX512
 8740     VM_Version::supports_avx512vlbw() &&
 8741     VM_Version::supports_bmi2()) {
 8742 
 8743     Label copy_32_loop, copy_tail;
 8744     Register tmp3_aliased = len;
 8745 
 8746     // if length of the string is less than 16, handle it in an old fashioned way
 8747     testl(len, -16);
 8748     jcc(Assembler::zero, below_threshold);
 8749 
 8750     testl(len, -1 * AVX3Threshold);
 8751     jcc(Assembler::zero, avx3_threshold);
 8752 
 8753     // In order to use only one arithmetic operation for the main loop we use
 8754     // this pre-calculation
 8755     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 8756     andl(len, -32);     // vector count
 8757     jccb(Assembler::zero, copy_tail);
 8758 
 8759     lea(src, Address(src, len, Address::times_1));
 8760     lea(dst, Address(dst, len, Address::times_2));
 8761     negptr(len);
 8762 
 8763 
 8764     // inflate 32 chars per iter
 8765     bind(copy_32_loop);
 8766     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 8767     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 8768     addptr(len, 32);
 8769     jcc(Assembler::notZero, copy_32_loop);
 8770 
 8771     bind(copy_tail);
 8772     // bail out when there is nothing to be done
 8773     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 8774     jcc(Assembler::zero, done);
 8775 
 8776     // ~(~0 << length), where length is the # of remaining elements to process
 8777     movl(tmp3_aliased, -1);
 8778     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 8779     notl(tmp3_aliased);
 8780     kmovdl(mask, tmp3_aliased);
 8781     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 8782     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 8783 
 8784     jmp(done);
 8785     bind(avx3_threshold);
 8786   }
 8787   if (UseSSE42Intrinsics) {
 8788     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 8789 
 8790     if (UseAVX > 1) {
 8791       andl(tmp2, (16 - 1));
 8792       andl(len, -16);
 8793       jccb(Assembler::zero, copy_new_tail);
 8794     } else {
 8795       andl(tmp2, 0x00000007);   // tail count (in chars)
 8796       andl(len, 0xfffffff8);    // vector count (in chars)
 8797       jccb(Assembler::zero, copy_tail);
 8798     }
 8799 
 8800     // vectored inflation
 8801     lea(src, Address(src, len, Address::times_1));
 8802     lea(dst, Address(dst, len, Address::times_2));
 8803     negptr(len);
 8804 
 8805     if (UseAVX > 1) {
 8806       bind(copy_16_loop);
 8807       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 8808       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 8809       addptr(len, 16);
 8810       jcc(Assembler::notZero, copy_16_loop);
 8811 
 8812       bind(below_threshold);
 8813       bind(copy_new_tail);
 8814       movl(len, tmp2);
 8815       andl(tmp2, 0x00000007);
 8816       andl(len, 0xFFFFFFF8);
 8817       jccb(Assembler::zero, copy_tail);
 8818 
 8819       pmovzxbw(tmp1, Address(src, 0));
 8820       movdqu(Address(dst, 0), tmp1);
 8821       addptr(src, 8);
 8822       addptr(dst, 2 * 8);
 8823 
 8824       jmp(copy_tail, true);
 8825     }
 8826 
 8827     // inflate 8 chars per iter
 8828     bind(copy_8_loop);
 8829     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 8830     movdqu(Address(dst, len, Address::times_2), tmp1);
 8831     addptr(len, 8);
 8832     jcc(Assembler::notZero, copy_8_loop);
 8833 
 8834     bind(copy_tail);
 8835     movl(len, tmp2);
 8836 
 8837     cmpl(len, 4);
 8838     jccb(Assembler::less, copy_bytes);
 8839 
 8840     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 8841     pmovzxbw(tmp1, tmp1);
 8842     movq(Address(dst, 0), tmp1);
 8843     subptr(len, 4);
 8844     addptr(src, 4);
 8845     addptr(dst, 8);
 8846 
 8847     bind(copy_bytes);
 8848   } else {
 8849     bind(below_threshold);
 8850   }
 8851 
 8852   testl(len, len);
 8853   jccb(Assembler::zero, done);
 8854   lea(src, Address(src, len, Address::times_1));
 8855   lea(dst, Address(dst, len, Address::times_2));
 8856   negptr(len);
 8857 
 8858   // inflate 1 char per iter
 8859   bind(copy_chars_loop);
 8860   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 8861   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 8862   increment(len);
 8863   jcc(Assembler::notZero, copy_chars_loop);
 8864 
 8865   bind(done);
 8866 }
 8867 
 8868 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister 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 
 8893 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 8894   switch(type) {
 8895     case T_BYTE:
 8896     case T_BOOLEAN:
 8897       evmovdqub(dst, kmask, src, merge, vector_len);
 8898       break;
 8899     case T_CHAR:
 8900     case T_SHORT:
 8901       evmovdquw(dst, kmask, src, merge, vector_len);
 8902       break;
 8903     case T_INT:
 8904     case T_FLOAT:
 8905       evmovdqul(dst, kmask, src, merge, vector_len);
 8906       break;
 8907     case T_LONG:
 8908     case T_DOUBLE:
 8909       evmovdquq(dst, kmask, src, merge, vector_len);
 8910       break;
 8911     default:
 8912       fatal("Unexpected type argument %s", type2name(type));
 8913       break;
 8914   }
 8915 }
 8916 
 8917 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 8918   switch(type) {
 8919     case T_BYTE:
 8920     case T_BOOLEAN:
 8921       evmovdqub(dst, kmask, src, merge, vector_len);
 8922       break;
 8923     case T_CHAR:
 8924     case T_SHORT:
 8925       evmovdquw(dst, kmask, src, merge, vector_len);
 8926       break;
 8927     case T_INT:
 8928     case T_FLOAT:
 8929       evmovdqul(dst, kmask, src, merge, vector_len);
 8930       break;
 8931     case T_LONG:
 8932     case T_DOUBLE:
 8933       evmovdquq(dst, kmask, src, merge, vector_len);
 8934       break;
 8935     default:
 8936       fatal("Unexpected type argument %s", type2name(type));
 8937       break;
 8938   }
 8939 }
 8940 
 8941 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 8942   switch(masklen) {
 8943     case 2:
 8944        knotbl(dst, src);
 8945        movl(rtmp, 3);
 8946        kmovbl(ktmp, rtmp);
 8947        kandbl(dst, ktmp, dst);
 8948        break;
 8949     case 4:
 8950        knotbl(dst, src);
 8951        movl(rtmp, 15);
 8952        kmovbl(ktmp, rtmp);
 8953        kandbl(dst, ktmp, dst);
 8954        break;
 8955     case 8:
 8956        knotbl(dst, src);
 8957        break;
 8958     case 16:
 8959        knotwl(dst, src);
 8960        break;
 8961     case 32:
 8962        knotdl(dst, src);
 8963        break;
 8964     case 64:
 8965        knotql(dst, src);
 8966        break;
 8967     default:
 8968       fatal("Unexpected vector length %d", masklen);
 8969       break;
 8970   }
 8971 }
 8972 
 8973 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8974   switch(type) {
 8975     case T_BOOLEAN:
 8976     case T_BYTE:
 8977        kandbl(dst, src1, src2);
 8978        break;
 8979     case T_CHAR:
 8980     case T_SHORT:
 8981        kandwl(dst, src1, src2);
 8982        break;
 8983     case T_INT:
 8984     case T_FLOAT:
 8985        kanddl(dst, src1, src2);
 8986        break;
 8987     case T_LONG:
 8988     case T_DOUBLE:
 8989        kandql(dst, src1, src2);
 8990        break;
 8991     default:
 8992       fatal("Unexpected type argument %s", type2name(type));
 8993       break;
 8994   }
 8995 }
 8996 
 8997 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 8998   switch(type) {
 8999     case T_BOOLEAN:
 9000     case T_BYTE:
 9001        korbl(dst, src1, src2);
 9002        break;
 9003     case T_CHAR:
 9004     case T_SHORT:
 9005        korwl(dst, src1, src2);
 9006        break;
 9007     case T_INT:
 9008     case T_FLOAT:
 9009        kordl(dst, src1, src2);
 9010        break;
 9011     case T_LONG:
 9012     case T_DOUBLE:
 9013        korql(dst, src1, src2);
 9014        break;
 9015     default:
 9016       fatal("Unexpected type argument %s", type2name(type));
 9017       break;
 9018   }
 9019 }
 9020 
 9021 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9022   switch(type) {
 9023     case T_BOOLEAN:
 9024     case T_BYTE:
 9025        kxorbl(dst, src1, src2);
 9026        break;
 9027     case T_CHAR:
 9028     case T_SHORT:
 9029        kxorwl(dst, src1, src2);
 9030        break;
 9031     case T_INT:
 9032     case T_FLOAT:
 9033        kxordl(dst, src1, src2);
 9034        break;
 9035     case T_LONG:
 9036     case T_DOUBLE:
 9037        kxorql(dst, src1, src2);
 9038        break;
 9039     default:
 9040       fatal("Unexpected type argument %s", type2name(type));
 9041       break;
 9042   }
 9043 }
 9044 
 9045 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9046   switch(type) {
 9047     case T_BOOLEAN:
 9048     case T_BYTE:
 9049       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9050     case T_CHAR:
 9051     case T_SHORT:
 9052       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9053     case T_INT:
 9054     case T_FLOAT:
 9055       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9056     case T_LONG:
 9057     case T_DOUBLE:
 9058       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9059     default:
 9060       fatal("Unexpected type argument %s", type2name(type)); break;
 9061   }
 9062 }
 9063 
 9064 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9065   switch(type) {
 9066     case T_BOOLEAN:
 9067     case T_BYTE:
 9068       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9069     case T_CHAR:
 9070     case T_SHORT:
 9071       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9072     case T_INT:
 9073     case T_FLOAT:
 9074       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9075     case T_LONG:
 9076     case T_DOUBLE:
 9077       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9078     default:
 9079       fatal("Unexpected type argument %s", type2name(type)); break;
 9080   }
 9081 }
 9082 
 9083 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9084   switch(type) {
 9085     case T_BYTE:
 9086       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9087     case T_SHORT:
 9088       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9089     case T_INT:
 9090       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9091     case T_LONG:
 9092       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9093     default:
 9094       fatal("Unexpected type argument %s", type2name(type)); break;
 9095   }
 9096 }
 9097 
 9098 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9099   switch(type) {
 9100     case T_BYTE:
 9101       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9102     case T_SHORT:
 9103       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9104     case T_INT:
 9105       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9106     case T_LONG:
 9107       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9108     default:
 9109       fatal("Unexpected type argument %s", type2name(type)); break;
 9110   }
 9111 }
 9112 
 9113 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9114   switch(type) {
 9115     case T_BYTE:
 9116       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9117     case T_SHORT:
 9118       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9119     case T_INT:
 9120       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9121     case T_LONG:
 9122       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9123     default:
 9124       fatal("Unexpected type argument %s", type2name(type)); break;
 9125   }
 9126 }
 9127 
 9128 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9129   switch(type) {
 9130     case T_BYTE:
 9131       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9132     case T_SHORT:
 9133       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9134     case T_INT:
 9135       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9136     case T_LONG:
 9137       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9138     default:
 9139       fatal("Unexpected type argument %s", type2name(type)); break;
 9140   }
 9141 }
 9142 
 9143 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9144   switch(type) {
 9145     case T_BYTE:
 9146       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9147     case T_SHORT:
 9148       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9149     case T_INT:
 9150       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9151     case T_LONG:
 9152       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9153     case T_FLOAT:
 9154       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9155     case T_DOUBLE:
 9156       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9157     default:
 9158       fatal("Unexpected type argument %s", type2name(type)); break;
 9159   }
 9160 }
 9161 
 9162 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9163   switch(type) {
 9164     case T_BYTE:
 9165       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9166     case T_SHORT:
 9167       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9168     case T_INT:
 9169       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9170     case T_LONG:
 9171       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9172     case T_FLOAT:
 9173       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9174     case T_DOUBLE:
 9175       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9176     default:
 9177       fatal("Unexpected type argument %s", type2name(type)); break;
 9178   }
 9179 }
 9180 
 9181 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9182   switch(type) {
 9183     case T_BYTE:
 9184       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9185     case T_SHORT:
 9186       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9187     case T_INT:
 9188       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9189     case T_LONG:
 9190       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9191     case T_FLOAT:
 9192       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9193     case T_DOUBLE:
 9194       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9195     default:
 9196       fatal("Unexpected type argument %s", type2name(type)); break;
 9197   }
 9198 }
 9199 
 9200 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9201   switch(type) {
 9202     case T_BYTE:
 9203       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9204     case T_SHORT:
 9205       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9206     case T_INT:
 9207       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9208     case T_LONG:
 9209       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9210     case T_FLOAT:
 9211       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9212     case T_DOUBLE:
 9213       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9214     default:
 9215       fatal("Unexpected type argument %s", type2name(type)); break;
 9216   }
 9217 }
 9218 
 9219 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9220   switch(type) {
 9221     case T_INT:
 9222       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9223     case T_LONG:
 9224       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9225     default:
 9226       fatal("Unexpected type argument %s", type2name(type)); break;
 9227   }
 9228 }
 9229 
 9230 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9231   switch(type) {
 9232     case T_INT:
 9233       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9234     case T_LONG:
 9235       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9236     default:
 9237       fatal("Unexpected type argument %s", type2name(type)); break;
 9238   }
 9239 }
 9240 
 9241 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9242   switch(type) {
 9243     case T_INT:
 9244       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9245     case T_LONG:
 9246       evporq(dst, mask, nds, src, merge, vector_len); break;
 9247     default:
 9248       fatal("Unexpected type argument %s", type2name(type)); break;
 9249   }
 9250 }
 9251 
 9252 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9253   switch(type) {
 9254     case T_INT:
 9255       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9256     case T_LONG:
 9257       evporq(dst, mask, nds, src, merge, vector_len); break;
 9258     default:
 9259       fatal("Unexpected type argument %s", type2name(type)); break;
 9260   }
 9261 }
 9262 
 9263 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9264   switch(type) {
 9265     case T_INT:
 9266       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9267     case T_LONG:
 9268       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9269     default:
 9270       fatal("Unexpected type argument %s", type2name(type)); break;
 9271   }
 9272 }
 9273 
 9274 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9275   switch(type) {
 9276     case T_INT:
 9277       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9278     case T_LONG:
 9279       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9280     default:
 9281       fatal("Unexpected type argument %s", type2name(type)); break;
 9282   }
 9283 }
 9284 
 9285 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9286   switch(masklen) {
 9287     case 8:
 9288        kortestbl(src1, src2);
 9289        break;
 9290     case 16:
 9291        kortestwl(src1, src2);
 9292        break;
 9293     case 32:
 9294        kortestdl(src1, src2);
 9295        break;
 9296     case 64:
 9297        kortestql(src1, src2);
 9298        break;
 9299     default:
 9300       fatal("Unexpected mask length %d", masklen);
 9301       break;
 9302   }
 9303 }
 9304 
 9305 
 9306 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
 9307   switch(masklen)  {
 9308     case 8:
 9309        ktestbl(src1, src2);
 9310        break;
 9311     case 16:
 9312        ktestwl(src1, src2);
 9313        break;
 9314     case 32:
 9315        ktestdl(src1, src2);
 9316        break;
 9317     case 64:
 9318        ktestql(src1, src2);
 9319        break;
 9320     default:
 9321       fatal("Unexpected mask length %d", masklen);
 9322       break;
 9323   }
 9324 }
 9325 
 9326 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9327   switch(type) {
 9328     case T_INT:
 9329       evprold(dst, mask, src, shift, merge, vlen_enc); break;
 9330     case T_LONG:
 9331       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
 9332     default:
 9333       fatal("Unexpected type argument %s", type2name(type)); break;
 9334       break;
 9335   }
 9336 }
 9337 
 9338 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9339   switch(type) {
 9340     case T_INT:
 9341       evprord(dst, mask, src, shift, merge, vlen_enc); break;
 9342     case T_LONG:
 9343       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
 9344     default:
 9345       fatal("Unexpected type argument %s", type2name(type)); break;
 9346   }
 9347 }
 9348 
 9349 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9350   switch(type) {
 9351     case T_INT:
 9352       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9353     case T_LONG:
 9354       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9355     default:
 9356       fatal("Unexpected type argument %s", type2name(type)); break;
 9357   }
 9358 }
 9359 
 9360 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9361   switch(type) {
 9362     case T_INT:
 9363       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9364     case T_LONG:
 9365       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
 9366     default:
 9367       fatal("Unexpected type argument %s", type2name(type)); break;
 9368   }
 9369 }
 9370 
 9371 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9372   assert(rscratch != noreg || always_reachable(src), "missing");
 9373 
 9374   if (reachable(src)) {
 9375     evpandq(dst, nds, as_Address(src), vector_len);
 9376   } else {
 9377     lea(rscratch, src);
 9378     evpandq(dst, nds, Address(rscratch, 0), vector_len);
 9379   }
 9380 }
 9381 
 9382 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 9383   assert(rscratch != noreg || always_reachable(src), "missing");
 9384 
 9385   if (reachable(src)) {
 9386     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
 9387   } else {
 9388     lea(rscratch, src);
 9389     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 9390   }
 9391 }
 9392 
 9393 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9394   assert(rscratch != noreg || always_reachable(src), "missing");
 9395 
 9396   if (reachable(src)) {
 9397     evporq(dst, nds, as_Address(src), vector_len);
 9398   } else {
 9399     lea(rscratch, src);
 9400     evporq(dst, nds, Address(rscratch, 0), vector_len);
 9401   }
 9402 }
 9403 
 9404 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9405   assert(rscratch != noreg || always_reachable(src), "missing");
 9406 
 9407   if (reachable(src)) {
 9408     vpshufb(dst, nds, as_Address(src), vector_len);
 9409   } else {
 9410     lea(rscratch, src);
 9411     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
 9412   }
 9413 }
 9414 
 9415 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 9416   assert(rscratch != noreg || always_reachable(src), "missing");
 9417 
 9418   if (reachable(src)) {
 9419     Assembler::vpor(dst, nds, as_Address(src), vector_len);
 9420   } else {
 9421     lea(rscratch, src);
 9422     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
 9423   }
 9424 }
 9425 
 9426 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
 9427   assert(rscratch != noreg || always_reachable(src3), "missing");
 9428 
 9429   if (reachable(src3)) {
 9430     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
 9431   } else {
 9432     lea(rscratch, src3);
 9433     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
 9434   }
 9435 }
 9436 
 9437 #if COMPILER2_OR_JVMCI
 9438 
 9439 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
 9440                                  Register length, Register temp, int vec_enc) {
 9441   // Computing mask for predicated vector store.
 9442   movptr(temp, -1);
 9443   bzhiq(temp, temp, length);
 9444   kmov(mask, temp);
 9445   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
 9446 }
 9447 
 9448 // Set memory operation for length "less than" 64 bytes.
 9449 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
 9450                                        XMMRegister xmm, KRegister mask, Register length,
 9451                                        Register temp, bool use64byteVector) {
 9452   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9453   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9454   if (!use64byteVector) {
 9455     fill32(dst, disp, xmm);
 9456     subptr(length, 32 >> shift);
 9457     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
 9458   } else {
 9459     assert(MaxVectorSize == 64, "vector length != 64");
 9460     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
 9461   }
 9462 }
 9463 
 9464 
 9465 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
 9466                                        XMMRegister xmm, KRegister mask, Register length,
 9467                                        Register temp) {
 9468   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9469   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
 9470   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
 9471 }
 9472 
 9473 
 9474 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
 9475   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9476   vmovdqu(dst, xmm);
 9477 }
 9478 
 9479 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
 9480   fill32(Address(dst, disp), xmm);
 9481 }
 9482 
 9483 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
 9484   assert(MaxVectorSize >= 32, "vector length should be >= 32");
 9485   if (!use64byteVector) {
 9486     fill32(dst, xmm);
 9487     fill32(dst.plus_disp(32), xmm);
 9488   } else {
 9489     evmovdquq(dst, xmm, Assembler::AVX_512bit);
 9490   }
 9491 }
 9492 
 9493 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
 9494   fill64(Address(dst, disp), xmm, use64byteVector);
 9495 }
 9496 
 9497 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
 9498                                         Register count, Register rtmp, XMMRegister xtmp) {
 9499   Label L_exit;
 9500   Label L_fill_start;
 9501   Label L_fill_64_bytes;
 9502   Label L_fill_96_bytes;
 9503   Label L_fill_128_bytes;
 9504   Label L_fill_128_bytes_loop;
 9505   Label L_fill_128_loop_header;
 9506   Label L_fill_128_bytes_loop_header;
 9507   Label L_fill_128_bytes_loop_pre_header;
 9508   Label L_fill_zmm_sequence;
 9509 
 9510   int shift = -1;
 9511   switch(type) {
 9512     case T_BYTE:  shift = 0;
 9513       break;
 9514     case T_SHORT: shift = 1;
 9515       break;
 9516     case T_INT:   shift = 2;
 9517       break;
 9518     /* Uncomment when LONG fill stubs are supported.
 9519     case T_LONG:  shift = 3;
 9520       break;
 9521     */
 9522     default:
 9523       fatal("Unhandled type: %s\n", type2name(type));
 9524   }
 9525 
 9526   if ((CopyAVX3Threshold != 0)  || (MaxVectorSize == 32)) {
 9527 
 9528     if (MaxVectorSize == 64) {
 9529       cmpq(count, CopyAVX3Threshold >> shift);
 9530       jcc(Assembler::greater, L_fill_zmm_sequence);
 9531     }
 9532 
 9533     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
 9534 
 9535     bind(L_fill_start);
 9536 
 9537     cmpq(count, 32 >> shift);
 9538     jccb(Assembler::greater, L_fill_64_bytes);
 9539     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9540     jmp(L_exit);
 9541 
 9542     bind(L_fill_64_bytes);
 9543     cmpq(count, 64 >> shift);
 9544     jccb(Assembler::greater, L_fill_96_bytes);
 9545     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
 9546     jmp(L_exit);
 9547 
 9548     bind(L_fill_96_bytes);
 9549     cmpq(count, 96 >> shift);
 9550     jccb(Assembler::greater, L_fill_128_bytes);
 9551     fill64(to, 0, xtmp);
 9552     subq(count, 64 >> shift);
 9553     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
 9554     jmp(L_exit);
 9555 
 9556     bind(L_fill_128_bytes);
 9557     cmpq(count, 128 >> shift);
 9558     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
 9559     fill64(to, 0, xtmp);
 9560     fill32(to, 64, xtmp);
 9561     subq(count, 96 >> shift);
 9562     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
 9563     jmp(L_exit);
 9564 
 9565     bind(L_fill_128_bytes_loop_pre_header);
 9566     {
 9567       mov(rtmp, to);
 9568       andq(rtmp, 31);
 9569       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
 9570       negq(rtmp);
 9571       addq(rtmp, 32);
 9572       mov64(r8, -1L);
 9573       bzhiq(r8, r8, rtmp);
 9574       kmovql(k2, r8);
 9575       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
 9576       addq(to, rtmp);
 9577       shrq(rtmp, shift);
 9578       subq(count, rtmp);
 9579     }
 9580 
 9581     cmpq(count, 128 >> shift);
 9582     jcc(Assembler::less, L_fill_start);
 9583 
 9584     bind(L_fill_128_bytes_loop_header);
 9585     subq(count, 128 >> shift);
 9586 
 9587     align32();
 9588     bind(L_fill_128_bytes_loop);
 9589       fill64(to, 0, xtmp);
 9590       fill64(to, 64, xtmp);
 9591       addq(to, 128);
 9592       subq(count, 128 >> shift);
 9593       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
 9594 
 9595     addq(count, 128 >> shift);
 9596     jcc(Assembler::zero, L_exit);
 9597     jmp(L_fill_start);
 9598   }
 9599 
 9600   if (MaxVectorSize == 64) {
 9601     // Sequence using 64 byte ZMM register.
 9602     Label L_fill_128_bytes_zmm;
 9603     Label L_fill_192_bytes_zmm;
 9604     Label L_fill_192_bytes_loop_zmm;
 9605     Label L_fill_192_bytes_loop_header_zmm;
 9606     Label L_fill_192_bytes_loop_pre_header_zmm;
 9607     Label L_fill_start_zmm_sequence;
 9608 
 9609     bind(L_fill_zmm_sequence);
 9610     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
 9611 
 9612     bind(L_fill_start_zmm_sequence);
 9613     cmpq(count, 64 >> shift);
 9614     jccb(Assembler::greater, L_fill_128_bytes_zmm);
 9615     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
 9616     jmp(L_exit);
 9617 
 9618     bind(L_fill_128_bytes_zmm);
 9619     cmpq(count, 128 >> shift);
 9620     jccb(Assembler::greater, L_fill_192_bytes_zmm);
 9621     fill64(to, 0, xtmp, true);
 9622     subq(count, 64 >> shift);
 9623     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
 9624     jmp(L_exit);
 9625 
 9626     bind(L_fill_192_bytes_zmm);
 9627     cmpq(count, 192 >> shift);
 9628     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
 9629     fill64(to, 0, xtmp, true);
 9630     fill64(to, 64, xtmp, true);
 9631     subq(count, 128 >> shift);
 9632     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
 9633     jmp(L_exit);
 9634 
 9635     bind(L_fill_192_bytes_loop_pre_header_zmm);
 9636     {
 9637       movq(rtmp, to);
 9638       andq(rtmp, 63);
 9639       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
 9640       negq(rtmp);
 9641       addq(rtmp, 64);
 9642       mov64(r8, -1L);
 9643       bzhiq(r8, r8, rtmp);
 9644       kmovql(k2, r8);
 9645       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
 9646       addq(to, rtmp);
 9647       shrq(rtmp, shift);
 9648       subq(count, rtmp);
 9649     }
 9650 
 9651     cmpq(count, 192 >> shift);
 9652     jcc(Assembler::less, L_fill_start_zmm_sequence);
 9653 
 9654     bind(L_fill_192_bytes_loop_header_zmm);
 9655     subq(count, 192 >> shift);
 9656 
 9657     align32();
 9658     bind(L_fill_192_bytes_loop_zmm);
 9659       fill64(to, 0, xtmp, true);
 9660       fill64(to, 64, xtmp, true);
 9661       fill64(to, 128, xtmp, true);
 9662       addq(to, 192);
 9663       subq(count, 192 >> shift);
 9664       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
 9665 
 9666     addq(count, 192 >> shift);
 9667     jcc(Assembler::zero, L_exit);
 9668     jmp(L_fill_start_zmm_sequence);
 9669   }
 9670   bind(L_exit);
 9671 }
 9672 #endif //COMPILER2_OR_JVMCI
 9673 
 9674 
 9675 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
 9676   Label done;
 9677   cvttss2sil(dst, src);
 9678   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9679   cmpl(dst, 0x80000000); // float_sign_flip
 9680   jccb(Assembler::notEqual, done);
 9681   subptr(rsp, 8);
 9682   movflt(Address(rsp, 0), src);
 9683   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
 9684   pop(dst);
 9685   bind(done);
 9686 }
 9687 
 9688 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
 9689   Label done;
 9690   cvttsd2sil(dst, src);
 9691   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
 9692   cmpl(dst, 0x80000000); // float_sign_flip
 9693   jccb(Assembler::notEqual, done);
 9694   subptr(rsp, 8);
 9695   movdbl(Address(rsp, 0), src);
 9696   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
 9697   pop(dst);
 9698   bind(done);
 9699 }
 9700 
 9701 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
 9702   Label done;
 9703   cvttss2siq(dst, src);
 9704   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9705   jccb(Assembler::notEqual, done);
 9706   subptr(rsp, 8);
 9707   movflt(Address(rsp, 0), src);
 9708   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
 9709   pop(dst);
 9710   bind(done);
 9711 }
 9712 
 9713 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9714   // Following code is line by line assembly translation rounding algorithm.
 9715   // Please refer to java.lang.Math.round(float) algorithm for details.
 9716   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
 9717   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
 9718   const int32_t FloatConsts_EXP_BIAS = 127;
 9719   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
 9720   const int32_t MINUS_32 = 0xFFFFFFE0;
 9721   Label L_special_case, L_block1, L_exit;
 9722   movl(rtmp, FloatConsts_EXP_BIT_MASK);
 9723   movdl(dst, src);
 9724   andl(dst, rtmp);
 9725   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
 9726   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
 9727   subl(rtmp, dst);
 9728   movl(rcx, rtmp);
 9729   movl(dst, MINUS_32);
 9730   testl(rtmp, dst);
 9731   jccb(Assembler::notEqual, L_special_case);
 9732   movdl(dst, src);
 9733   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
 9734   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
 9735   movdl(rtmp, src);
 9736   testl(rtmp, rtmp);
 9737   jccb(Assembler::greaterEqual, L_block1);
 9738   negl(dst);
 9739   bind(L_block1);
 9740   sarl(dst);
 9741   addl(dst, 0x1);
 9742   sarl(dst, 0x1);
 9743   jmp(L_exit);
 9744   bind(L_special_case);
 9745   convert_f2i(dst, src);
 9746   bind(L_exit);
 9747 }
 9748 
 9749 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
 9750   // Following code is line by line assembly translation rounding algorithm.
 9751   // Please refer to java.lang.Math.round(double) algorithm for details.
 9752   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
 9753   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
 9754   const int64_t DoubleConsts_EXP_BIAS = 1023;
 9755   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
 9756   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
 9757   Label L_special_case, L_block1, L_exit;
 9758   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
 9759   movq(dst, src);
 9760   andq(dst, rtmp);
 9761   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
 9762   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
 9763   subq(rtmp, dst);
 9764   movq(rcx, rtmp);
 9765   mov64(dst, MINUS_64);
 9766   testq(rtmp, dst);
 9767   jccb(Assembler::notEqual, L_special_case);
 9768   movq(dst, src);
 9769   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
 9770   andq(dst, rtmp);
 9771   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
 9772   orq(dst, rtmp);
 9773   movq(rtmp, src);
 9774   testq(rtmp, rtmp);
 9775   jccb(Assembler::greaterEqual, L_block1);
 9776   negq(dst);
 9777   bind(L_block1);
 9778   sarq(dst);
 9779   addq(dst, 0x1);
 9780   sarq(dst, 0x1);
 9781   jmp(L_exit);
 9782   bind(L_special_case);
 9783   convert_d2l(dst, src);
 9784   bind(L_exit);
 9785 }
 9786 
 9787 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
 9788   Label done;
 9789   cvttsd2siq(dst, src);
 9790   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
 9791   jccb(Assembler::notEqual, done);
 9792   subptr(rsp, 8);
 9793   movdbl(Address(rsp, 0), src);
 9794   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
 9795   pop(dst);
 9796   bind(done);
 9797 }
 9798 
 9799 void MacroAssembler::cache_wb(Address line)
 9800 {
 9801   // 64 bit cpus always support clflush
 9802   assert(VM_Version::supports_clflush(), "clflush should be available");
 9803   bool optimized = VM_Version::supports_clflushopt();
 9804   bool no_evict = VM_Version::supports_clwb();
 9805 
 9806   // prefer clwb (writeback without evict) otherwise
 9807   // prefer clflushopt (potentially parallel writeback with evict)
 9808   // otherwise fallback on clflush (serial writeback with evict)
 9809 
 9810   if (optimized) {
 9811     if (no_evict) {
 9812       clwb(line);
 9813     } else {
 9814       clflushopt(line);
 9815     }
 9816   } else {
 9817     // no need for fence when using CLFLUSH
 9818     clflush(line);
 9819   }
 9820 }
 9821 
 9822 void MacroAssembler::cache_wbsync(bool is_pre)
 9823 {
 9824   assert(VM_Version::supports_clflush(), "clflush should be available");
 9825   bool optimized = VM_Version::supports_clflushopt();
 9826   bool no_evict = VM_Version::supports_clwb();
 9827 
 9828   // pick the correct implementation
 9829 
 9830   if (!is_pre && (optimized || no_evict)) {
 9831     // need an sfence for post flush when using clflushopt or clwb
 9832     // otherwise no no need for any synchroniaztion
 9833 
 9834     sfence();
 9835   }
 9836 }
 9837 
 9838 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
 9839   switch (cond) {
 9840     // Note some conditions are synonyms for others
 9841     case Assembler::zero:         return Assembler::notZero;
 9842     case Assembler::notZero:      return Assembler::zero;
 9843     case Assembler::less:         return Assembler::greaterEqual;
 9844     case Assembler::lessEqual:    return Assembler::greater;
 9845     case Assembler::greater:      return Assembler::lessEqual;
 9846     case Assembler::greaterEqual: return Assembler::less;
 9847     case Assembler::below:        return Assembler::aboveEqual;
 9848     case Assembler::belowEqual:   return Assembler::above;
 9849     case Assembler::above:        return Assembler::belowEqual;
 9850     case Assembler::aboveEqual:   return Assembler::below;
 9851     case Assembler::overflow:     return Assembler::noOverflow;
 9852     case Assembler::noOverflow:   return Assembler::overflow;
 9853     case Assembler::negative:     return Assembler::positive;
 9854     case Assembler::positive:     return Assembler::negative;
 9855     case Assembler::parity:       return Assembler::noParity;
 9856     case Assembler::noParity:     return Assembler::parity;
 9857   }
 9858   ShouldNotReachHere(); return Assembler::overflow;
 9859 }
 9860 
 9861 // This is simply a call to Thread::current()
 9862 void MacroAssembler::get_thread_slow(Register thread) {
 9863   if (thread != rax) {
 9864     push(rax);
 9865   }
 9866   push(rdi);
 9867   push(rsi);
 9868   push(rdx);
 9869   push(rcx);
 9870   push(r8);
 9871   push(r9);
 9872   push(r10);
 9873   push(r11);
 9874 
 9875   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
 9876 
 9877   pop(r11);
 9878   pop(r10);
 9879   pop(r9);
 9880   pop(r8);
 9881   pop(rcx);
 9882   pop(rdx);
 9883   pop(rsi);
 9884   pop(rdi);
 9885   if (thread != rax) {
 9886     mov(thread, rax);
 9887     pop(rax);
 9888   }
 9889 }
 9890 
 9891 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
 9892   Label L_stack_ok;
 9893   if (bias == 0) {
 9894     testptr(sp, 2 * wordSize - 1);
 9895   } else {
 9896     // lea(tmp, Address(rsp, bias);
 9897     mov(tmp, sp);
 9898     addptr(tmp, bias);
 9899     testptr(tmp, 2 * wordSize - 1);
 9900   }
 9901   jcc(Assembler::equal, L_stack_ok);
 9902   block_comment(msg);
 9903   stop(msg);
 9904   bind(L_stack_ok);
 9905 }
 9906 
 9907 // Implements fast-locking.
 9908 //
 9909 // obj: the object to be locked
 9910 // reg_rax: rax
 9911 // thread: the thread which attempts to lock obj
 9912 // tmp: a temporary register
 9913 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
 9914   Register thread = r15_thread;
 9915 
 9916   assert(reg_rax == rax, "");
 9917   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
 9918 
 9919   Label push;
 9920   const Register top = tmp;
 9921 
 9922   // Preload the markWord. It is important that this is the first
 9923   // instruction emitted as it is part of C1's null check semantics.
 9924   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9925 
 9926   if (UseObjectMonitorTable) {
 9927     // Clear cache in case fast locking succeeds or we need to take the slow-path.
 9928     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
 9929   }
 9930 
 9931   if (DiagnoseSyncOnValueBasedClasses != 0) {
 9932     load_klass(tmp, obj, rscratch1);
 9933     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
 9934     jcc(Assembler::notZero, slow);
 9935   }
 9936 
 9937   // Load top.
 9938   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9939 
 9940   // Check if the lock-stack is full.
 9941   cmpl(top, LockStack::end_offset());
 9942   jcc(Assembler::greaterEqual, slow);
 9943 
 9944   // Check for recursion.
 9945   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9946   jcc(Assembler::equal, push);
 9947 
 9948   // Check header for monitor (0b10).
 9949   testptr(reg_rax, markWord::monitor_value);
 9950   jcc(Assembler::notZero, slow);
 9951 
 9952   // Try to lock. Transition lock bits 0b01 => 0b00
 9953   movptr(tmp, reg_rax);
 9954   andptr(tmp, ~(int32_t)markWord::unlocked_value);
 9955   orptr(reg_rax, markWord::unlocked_value);



 9956   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9957   jcc(Assembler::notEqual, slow);
 9958 
 9959   // Restore top, CAS clobbers register.
 9960   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9961 
 9962   bind(push);
 9963   // After successful lock, push object on lock-stack.
 9964   movptr(Address(thread, top), obj);
 9965   incrementl(top, oopSize);
 9966   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
 9967 }
 9968 
 9969 // Implements fast-unlocking.
 9970 //
 9971 // obj: the object to be unlocked
 9972 // reg_rax: rax
 9973 // thread: the thread
 9974 // tmp: a temporary register
 9975 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
 9976   Register thread = r15_thread;
 9977 
 9978   assert(reg_rax == rax, "");
 9979   assert_different_registers(obj, reg_rax, thread, tmp);
 9980 
 9981   Label unlocked, push_and_slow;
 9982   const Register top = tmp;
 9983 
 9984   // Check if obj is top of lock-stack.
 9985   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9986   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9987   jcc(Assembler::notEqual, slow);
 9988 
 9989   // Pop lock-stack.
 9990   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
 9991   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
 9992 
 9993   // Check if recursive.
 9994   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
 9995   jcc(Assembler::equal, unlocked);
 9996 
 9997   // Not recursive. Check header for monitor (0b10).
 9998   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
 9999   testptr(reg_rax, markWord::monitor_value);
10000   jcc(Assembler::notZero, push_and_slow);
10001 
10002 #ifdef ASSERT
10003   // Check header not unlocked (0b01).
10004   Label not_unlocked;
10005   testptr(reg_rax, markWord::unlocked_value);
10006   jcc(Assembler::zero, not_unlocked);
10007   stop("fast_unlock already unlocked");
10008   bind(not_unlocked);
10009 #endif
10010 
10011   // Try to unlock. Transition lock bits 0b00 => 0b01
10012   movptr(tmp, reg_rax);
10013   orptr(tmp, markWord::unlocked_value);
10014   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10015   jcc(Assembler::equal, unlocked);
10016 
10017   bind(push_and_slow);
10018   // Restore lock-stack and handle the unlock in runtime.
10019 #ifdef ASSERT
10020   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10021   movptr(Address(thread, top), obj);
10022 #endif
10023   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10024   jmp(slow);
10025 
10026   bind(unlocked);
10027 }
10028 
10029 // Saves legacy GPRs state on stack.
10030 void MacroAssembler::save_legacy_gprs() {
10031   subq(rsp, 16 * wordSize);
10032   movq(Address(rsp, 15 * wordSize), rax);
10033   movq(Address(rsp, 14 * wordSize), rcx);
10034   movq(Address(rsp, 13 * wordSize), rdx);
10035   movq(Address(rsp, 12 * wordSize), rbx);
10036   movq(Address(rsp, 10 * wordSize), rbp);
10037   movq(Address(rsp, 9 * wordSize), rsi);
10038   movq(Address(rsp, 8 * wordSize), rdi);
10039   movq(Address(rsp, 7 * wordSize), r8);
10040   movq(Address(rsp, 6 * wordSize), r9);
10041   movq(Address(rsp, 5 * wordSize), r10);
10042   movq(Address(rsp, 4 * wordSize), r11);
10043   movq(Address(rsp, 3 * wordSize), r12);
10044   movq(Address(rsp, 2 * wordSize), r13);
10045   movq(Address(rsp, wordSize), r14);
10046   movq(Address(rsp, 0), r15);
10047 }
10048 
10049 // Resotres back legacy GPRs state from stack.
10050 void MacroAssembler::restore_legacy_gprs() {
10051   movq(r15, Address(rsp, 0));
10052   movq(r14, Address(rsp, wordSize));
10053   movq(r13, Address(rsp, 2 * wordSize));
10054   movq(r12, Address(rsp, 3 * wordSize));
10055   movq(r11, Address(rsp, 4 * wordSize));
10056   movq(r10, Address(rsp, 5 * wordSize));
10057   movq(r9,  Address(rsp, 6 * wordSize));
10058   movq(r8,  Address(rsp, 7 * wordSize));
10059   movq(rdi, Address(rsp, 8 * wordSize));
10060   movq(rsi, Address(rsp, 9 * wordSize));
10061   movq(rbp, Address(rsp, 10 * wordSize));
10062   movq(rbx, Address(rsp, 12 * wordSize));
10063   movq(rdx, Address(rsp, 13 * wordSize));
10064   movq(rcx, Address(rsp, 14 * wordSize));
10065   movq(rax, Address(rsp, 15 * wordSize));
10066   addq(rsp, 16 * wordSize);
10067 }
10068 
10069 void MacroAssembler::load_aotrc_address(Register reg, address a) {
10070 #if INCLUDE_CDS
10071   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
10072   if (AOTCodeCache::is_on_for_dump()) {
10073     // all aotrc field addresses should be registered in the AOTCodeCache address table
10074     lea(reg, ExternalAddress(a));
10075   } else {
10076     mov64(reg, (uint64_t)a);
10077   }
10078 #else
10079   ShouldNotReachHere();
10080 #endif
10081 }
10082 
10083 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10084   if (VM_Version::supports_apx_f()) {
10085     esetzucc(comparison, dst);
10086   } else {
10087     setb(comparison, dst);
10088     movzbl(dst, dst);
10089   }
10090 }
--- EOF ---