1 /*
    2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
    3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
    4  *
    5  * This code is free software; you can redistribute it and/or modify it
    6  * under the terms of the GNU General Public License version 2 only, as
    7  * published by the Free Software Foundation.
    8  *
    9  * This code is distributed in the hope that it will be useful, but WITHOUT
   10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
   11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
   12  * version 2 for more details (a copy is included in the LICENSE file that
   13  * accompanied this code).
   14  *
   15  * You should have received a copy of the GNU General Public License version
   16  * 2 along with this work; if not, write to the Free Software Foundation,
   17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
   18  *
   19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
   20  * or visit www.oracle.com if you need additional information or have any
   21  * questions.
   22  *
   23  */
   24 
   25 #include "asm/assembler.hpp"
   26 #include "asm/assembler.inline.hpp"
   27 #include "code/aotCodeCache.hpp"
   28 #include "code/compiledIC.hpp"
   29 #include "compiler/compiler_globals.hpp"
   30 #include "compiler/disassembler.hpp"
   31 #include "ci/ciInlineKlass.hpp"
   32 #include "crc32c.h"
   33 #include "gc/shared/barrierSet.hpp"
   34 #include "gc/shared/barrierSetAssembler.hpp"
   35 #include "gc/shared/collectedHeap.inline.hpp"
   36 #include "gc/shared/tlab_globals.hpp"
   37 #include "interpreter/bytecodeHistogram.hpp"
   38 #include "interpreter/interpreter.hpp"
   39 #include "interpreter/interpreterRuntime.hpp"
   40 #include "jvm.h"
   41 #include "memory/resourceArea.hpp"
   42 #include "memory/universe.hpp"
   43 #include "oops/accessDecorators.hpp"
   44 #include "oops/compressedKlass.inline.hpp"
   45 #include "oops/compressedOops.inline.hpp"
   46 #include "oops/klass.inline.hpp"
   47 #include "oops/resolvedFieldEntry.hpp"
   48 #include "prims/methodHandles.hpp"
   49 #include "runtime/arguments.hpp"
   50 #include "runtime/continuation.hpp"
   51 #include "runtime/interfaceSupport.inline.hpp"
   52 #include "runtime/javaThread.hpp"
   53 #include "runtime/jniHandles.hpp"
   54 #include "runtime/objectMonitor.hpp"
   55 #include "runtime/os.hpp"
   56 #include "runtime/safepoint.hpp"
   57 #include "runtime/safepointMechanism.hpp"
   58 #include "runtime/sharedRuntime.hpp"
   59 #include "runtime/signature_cc.hpp"
   60 #include "runtime/stubRoutines.hpp"
   61 #include "utilities/checkedCast.hpp"
   62 #include "utilities/globalDefinitions.hpp"
   63 #include "utilities/macros.hpp"
   64 #include "vmreg_x86.inline.hpp"
   65 #ifdef COMPILER2
   66 #include "opto/output.hpp"
   67 #endif
   68 
   69 #ifdef PRODUCT
   70 #define BLOCK_COMMENT(str) /* nothing */
   71 #define STOP(error) stop(error)
   72 #else
   73 #define BLOCK_COMMENT(str) block_comment(str)
   74 #define STOP(error) block_comment(error); stop(error)
   75 #endif
   76 
   77 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   78 
   79 #ifdef ASSERT
   80 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   81 #endif
   82 
   83 static const Assembler::Condition reverse[] = {
   84     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   85     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   86     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   87     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
   88     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
   89     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
   90     Assembler::above          /* belowEqual    = 0x6 */ ,
   91     Assembler::belowEqual     /* above         = 0x7 */ ,
   92     Assembler::positive       /* negative      = 0x8 */ ,
   93     Assembler::negative       /* positive      = 0x9 */ ,
   94     Assembler::noParity       /* parity        = 0xa */ ,
   95     Assembler::parity         /* noParity      = 0xb */ ,
   96     Assembler::greaterEqual   /* less          = 0xc */ ,
   97     Assembler::less           /* greaterEqual  = 0xd */ ,
   98     Assembler::greater        /* lessEqual     = 0xe */ ,
   99     Assembler::lessEqual      /* greater       = 0xf, */
  100 
  101 };
  102 
  103 
  104 // Implementation of MacroAssembler
  105 
  106 Address MacroAssembler::as_Address(AddressLiteral adr) {
  107   // amd64 always does this as a pc-rel
  108   // we can be absolute or disp based on the instruction type
  109   // jmp/call are displacements others are absolute
  110   assert(!adr.is_lval(), "must be rval");
  111   assert(reachable(adr), "must be");
  112   return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
  113 
  114 }
  115 
  116 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  117   AddressLiteral base = adr.base();
  118   lea(rscratch, base);
  119   Address index = adr.index();
  120   assert(index._disp == 0, "must not have disp"); // maybe it can?
  121   Address array(rscratch, index._index, index._scale, index._disp);
  122   return array;
  123 }
  124 
  125 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
  126   Label L, E;
  127 
  128 #ifdef _WIN64
  129   // Windows always allocates space for it's register args
  130   assert(num_args <= 4, "only register arguments supported");
  131   subq(rsp,  frame::arg_reg_save_area_bytes);
  132 #endif
  133 
  134   // Align stack if necessary
  135   testl(rsp, 15);
  136   jcc(Assembler::zero, L);
  137 
  138   subq(rsp, 8);
  139   call(RuntimeAddress(entry_point));
  140   addq(rsp, 8);
  141   jmp(E);
  142 
  143   bind(L);
  144   call(RuntimeAddress(entry_point));
  145 
  146   bind(E);
  147 
  148 #ifdef _WIN64
  149   // restore stack pointer
  150   addq(rsp, frame::arg_reg_save_area_bytes);
  151 #endif
  152 }
  153 
  154 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
  155   assert(!src2.is_lval(), "should use cmpptr");
  156   assert(rscratch != noreg || always_reachable(src2), "missing");
  157 
  158   if (reachable(src2)) {
  159     cmpq(src1, as_Address(src2));
  160   } else {
  161     lea(rscratch, src2);
  162     Assembler::cmpq(src1, Address(rscratch, 0));
  163   }
  164 }
  165 
  166 int MacroAssembler::corrected_idivq(Register reg) {
  167   // Full implementation of Java ldiv and lrem; checks for special
  168   // case as described in JVM spec., p.243 & p.271.  The function
  169   // returns the (pc) offset of the idivl instruction - may be needed
  170   // for implicit exceptions.
  171   //
  172   //         normal case                           special case
  173   //
  174   // input : rax: dividend                         min_long
  175   //         reg: divisor   (may not be eax/edx)   -1
  176   //
  177   // output: rax: quotient  (= rax idiv reg)       min_long
  178   //         rdx: remainder (= rax irem reg)       0
  179   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
  180   static const int64_t min_long = 0x8000000000000000;
  181   Label normal_case, special_case;
  182 
  183   // check for special case
  184   cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
  185   jcc(Assembler::notEqual, normal_case);
  186   xorl(rdx, rdx); // prepare rdx for possible special case (where
  187                   // remainder = 0)
  188   cmpq(reg, -1);
  189   jcc(Assembler::equal, special_case);
  190 
  191   // handle normal case
  192   bind(normal_case);
  193   cdqq();
  194   int idivq_offset = offset();
  195   idivq(reg);
  196 
  197   // normal and special case exit
  198   bind(special_case);
  199 
  200   return idivq_offset;
  201 }
  202 
  203 void MacroAssembler::decrementq(Register reg, int value) {
  204   if (value == min_jint) { subq(reg, value); return; }
  205   if (value <  0) { incrementq(reg, -value); return; }
  206   if (value == 0) {                        ; return; }
  207   if (value == 1 && UseIncDec) { decq(reg) ; return; }
  208   /* else */      { subq(reg, value)       ; return; }
  209 }
  210 
  211 void MacroAssembler::decrementq(Address dst, int value) {
  212   if (value == min_jint) { subq(dst, value); return; }
  213   if (value <  0) { incrementq(dst, -value); return; }
  214   if (value == 0) {                        ; return; }
  215   if (value == 1 && UseIncDec) { decq(dst) ; return; }
  216   /* else */      { subq(dst, value)       ; return; }
  217 }
  218 
  219 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
  220   assert(rscratch != noreg || always_reachable(dst), "missing");
  221 
  222   if (reachable(dst)) {
  223     incrementq(as_Address(dst));
  224   } else {
  225     lea(rscratch, dst);
  226     incrementq(Address(rscratch, 0));
  227   }
  228 }
  229 
  230 void MacroAssembler::incrementq(Register reg, int value) {
  231   if (value == min_jint) { addq(reg, value); return; }
  232   if (value <  0) { decrementq(reg, -value); return; }
  233   if (value == 0) {                        ; return; }
  234   if (value == 1 && UseIncDec) { incq(reg) ; return; }
  235   /* else */      { addq(reg, value)       ; return; }
  236 }
  237 
  238 void MacroAssembler::incrementq(Address dst, int value) {
  239   if (value == min_jint) { addq(dst, value); return; }
  240   if (value <  0) { decrementq(dst, -value); return; }
  241   if (value == 0) {                        ; return; }
  242   if (value == 1 && UseIncDec) { incq(dst) ; return; }
  243   /* else */      { addq(dst, value)       ; return; }
  244 }
  245 
  246 // 32bit can do a case table jump in one instruction but we no longer allow the base
  247 // to be installed in the Address class
  248 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  249   lea(rscratch, entry.base());
  250   Address dispatch = entry.index();
  251   assert(dispatch._base == noreg, "must be");
  252   dispatch._base = rscratch;
  253   jmp(dispatch);
  254 }
  255 
  256 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  257   ShouldNotReachHere(); // 64bit doesn't use two regs
  258   cmpq(x_lo, y_lo);
  259 }
  260 
  261 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  262   mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  263 }
  264 
  265 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  266   lea(rscratch, adr);
  267   movptr(dst, rscratch);
  268 }
  269 
  270 void MacroAssembler::leave() {
  271   // %%% is this really better? Why not on 32bit too?
  272   emit_int8((unsigned char)0xC9); // LEAVE
  273 }
  274 
  275 void MacroAssembler::lneg(Register hi, Register lo) {
  276   ShouldNotReachHere(); // 64bit doesn't use two regs
  277   negq(lo);
  278 }
  279 
  280 void MacroAssembler::movoop(Register dst, jobject obj) {
  281   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  282 }
  283 
  284 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  285   mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  286   movq(dst, rscratch);
  287 }
  288 
  289 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  290   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  291 }
  292 
  293 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  294   mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  295   movq(dst, rscratch);
  296 }
  297 
  298 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  299   if (src.is_lval()) {
  300     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  301   } else {
  302     if (reachable(src)) {
  303       movq(dst, as_Address(src));
  304     } else {
  305       lea(dst, src);
  306       movq(dst, Address(dst, 0));
  307     }
  308   }
  309 }
  310 
  311 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  312   movq(as_Address(dst, rscratch), src);
  313 }
  314 
  315 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  316   movq(dst, as_Address(src, dst /*rscratch*/));
  317 }
  318 
  319 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
  320 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  321   if (is_simm32(src)) {
  322     movptr(dst, checked_cast<int32_t>(src));
  323   } else {
  324     mov64(rscratch, src);
  325     movq(dst, rscratch);
  326   }
  327 }
  328 
  329 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  330   movoop(rscratch, obj);
  331   push(rscratch);
  332 }
  333 
  334 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  335   mov_metadata(rscratch, obj);
  336   push(rscratch);
  337 }
  338 
  339 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  340   lea(rscratch, src);
  341   if (src.is_lval()) {
  342     push(rscratch);
  343   } else {
  344     pushq(Address(rscratch, 0));
  345   }
  346 }
  347 
  348 static void pass_arg0(MacroAssembler* masm, Register arg) {
  349   if (c_rarg0 != arg ) {
  350     masm->mov(c_rarg0, arg);
  351   }
  352 }
  353 
  354 static void pass_arg1(MacroAssembler* masm, Register arg) {
  355   if (c_rarg1 != arg ) {
  356     masm->mov(c_rarg1, arg);
  357   }
  358 }
  359 
  360 static void pass_arg2(MacroAssembler* masm, Register arg) {
  361   if (c_rarg2 != arg ) {
  362     masm->mov(c_rarg2, arg);
  363   }
  364 }
  365 
  366 static void pass_arg3(MacroAssembler* masm, Register arg) {
  367   if (c_rarg3 != arg ) {
  368     masm->mov(c_rarg3, arg);
  369   }
  370 }
  371 
  372 void MacroAssembler::stop(const char* msg) {
  373   if (ShowMessageBoxOnError) {
  374     address rip = pc();
  375     pusha(); // get regs on stack
  376     lea(c_rarg1, InternalAddress(rip));
  377     movq(c_rarg2, rsp); // pass pointer to regs array
  378   }
  379   // Skip AOT caching C strings in scratch buffer.
  380   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  381   lea(c_rarg0, ExternalAddress((address) str));
  382   andq(rsp, -16); // align stack as required by ABI
  383   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
  384   hlt();
  385 }
  386 
  387 void MacroAssembler::warn(const char* msg) {
  388   push(rbp);
  389   movq(rbp, rsp);
  390   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  391   push_CPU_state();   // keeps alignment at 16 bytes
  392 
  393 #ifdef _WIN64
  394   // Windows always allocates space for its register args
  395   subq(rsp,  frame::arg_reg_save_area_bytes);
  396 #endif
  397   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  398   lea(c_rarg0, ExternalAddress((address) str));
  399   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  400 
  401 #ifdef _WIN64
  402   // restore stack pointer
  403   addq(rsp, frame::arg_reg_save_area_bytes);
  404 #endif
  405   pop_CPU_state();
  406   mov(rsp, rbp);
  407   pop(rbp);
  408 }
  409 
  410 void MacroAssembler::print_state() {
  411   address rip = pc();
  412   pusha();            // get regs on stack
  413   push(rbp);
  414   movq(rbp, rsp);
  415   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  416   push_CPU_state();   // keeps alignment at 16 bytes
  417 
  418   lea(c_rarg0, InternalAddress(rip));
  419   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  420   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  421 
  422   pop_CPU_state();
  423   mov(rsp, rbp);
  424   pop(rbp);
  425   popa();
  426 }
  427 
  428 #ifndef PRODUCT
  429 extern "C" void findpc(intptr_t x);
  430 #endif
  431 
  432 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  433   // In order to get locks to work, we need to fake a in_VM state
  434   if (ShowMessageBoxOnError) {
  435     JavaThread* thread = JavaThread::current();
  436     JavaThreadState saved_state = thread->thread_state();
  437     thread->set_thread_state(_thread_in_vm);
  438 #ifndef PRODUCT
  439     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  440       ttyLocker ttyl;
  441       BytecodeCounter::print();
  442     }
  443 #endif
  444     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  445     // XXX correct this offset for amd64
  446     // This is the value of eip which points to where verify_oop will return.
  447     if (os::message_box(msg, "Execution stopped, print registers?")) {
  448       print_state64(pc, regs);
  449       BREAKPOINT;
  450     }
  451   }
  452   fatal("DEBUG MESSAGE: %s", msg);
  453 }
  454 
  455 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  456   ttyLocker ttyl;
  457   DebuggingContext debugging{};
  458   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  459 #ifndef PRODUCT
  460   tty->cr();
  461   findpc(pc);
  462   tty->cr();
  463 #endif
  464 #define PRINT_REG(rax, value) \
  465   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  466   PRINT_REG(rax, regs[15]);
  467   PRINT_REG(rbx, regs[12]);
  468   PRINT_REG(rcx, regs[14]);
  469   PRINT_REG(rdx, regs[13]);
  470   PRINT_REG(rdi, regs[8]);
  471   PRINT_REG(rsi, regs[9]);
  472   PRINT_REG(rbp, regs[10]);
  473   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  474   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  475   PRINT_REG(r8 , regs[7]);
  476   PRINT_REG(r9 , regs[6]);
  477   PRINT_REG(r10, regs[5]);
  478   PRINT_REG(r11, regs[4]);
  479   PRINT_REG(r12, regs[3]);
  480   PRINT_REG(r13, regs[2]);
  481   PRINT_REG(r14, regs[1]);
  482   PRINT_REG(r15, regs[0]);
  483 #undef PRINT_REG
  484   // Print some words near the top of the stack.
  485   int64_t* rsp = &regs[16];
  486   int64_t* dump_sp = rsp;
  487   for (int col1 = 0; col1 < 8; col1++) {
  488     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  489     os::print_location(tty, *dump_sp++);
  490   }
  491   for (int row = 0; row < 25; row++) {
  492     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  493     for (int col = 0; col < 4; col++) {
  494       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  495     }
  496     tty->cr();
  497   }
  498   // Print some instructions around pc:
  499   Disassembler::decode((address)pc-64, (address)pc);
  500   tty->print_cr("--------");
  501   Disassembler::decode((address)pc, (address)pc+32);
  502 }
  503 
  504 // The java_calling_convention describes stack locations as ideal slots on
  505 // a frame with no abi restrictions. Since we must observe abi restrictions
  506 // (like the placement of the register window) the slots must be biased by
  507 // the following value.
  508 static int reg2offset_in(VMReg r) {
  509   // Account for saved rbp and return address
  510   // This should really be in_preserve_stack_slots
  511   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  512 }
  513 
  514 static int reg2offset_out(VMReg r) {
  515   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  516 }
  517 
  518 // A long move
  519 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  520 
  521   // The calling conventions assures us that each VMregpair is either
  522   // all really one physical register or adjacent stack slots.
  523 
  524   if (src.is_single_phys_reg() ) {
  525     if (dst.is_single_phys_reg()) {
  526       if (dst.first() != src.first()) {
  527         mov(dst.first()->as_Register(), src.first()->as_Register());
  528       }
  529     } else {
  530       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  531              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  532       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  533     }
  534   } else if (dst.is_single_phys_reg()) {
  535     assert(src.is_single_reg(),  "not a stack pair");
  536     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  537   } else {
  538     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  539     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  540     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  541   }
  542 }
  543 
  544 // A double move
  545 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  546 
  547   // The calling conventions assures us that each VMregpair is either
  548   // all really one physical register or adjacent stack slots.
  549 
  550   if (src.is_single_phys_reg() ) {
  551     if (dst.is_single_phys_reg()) {
  552       // In theory these overlap but the ordering is such that this is likely a nop
  553       if ( src.first() != dst.first()) {
  554         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  555       }
  556     } else {
  557       assert(dst.is_single_reg(), "not a stack pair");
  558       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  559     }
  560   } else if (dst.is_single_phys_reg()) {
  561     assert(src.is_single_reg(),  "not a stack pair");
  562     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  563   } else {
  564     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  565     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  566     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  567   }
  568 }
  569 
  570 
  571 // A float arg may have to do float reg int reg conversion
  572 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  573   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  574 
  575   // The calling conventions assures us that each VMregpair is either
  576   // all really one physical register or adjacent stack slots.
  577 
  578   if (src.first()->is_stack()) {
  579     if (dst.first()->is_stack()) {
  580       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  581       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  582     } else {
  583       // stack to reg
  584       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  585       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  586     }
  587   } else if (dst.first()->is_stack()) {
  588     // reg to stack
  589     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  590     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  591   } else {
  592     // reg to reg
  593     // In theory these overlap but the ordering is such that this is likely a nop
  594     if ( src.first() != dst.first()) {
  595       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  596     }
  597   }
  598 }
  599 
  600 // On 64 bit we will store integer like items to the stack as
  601 // 64 bits items (x86_32/64 abi) even though java would only store
  602 // 32bits for a parameter. On 32bit it will simply be 32 bits
  603 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  604 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  605   if (src.first()->is_stack()) {
  606     if (dst.first()->is_stack()) {
  607       // stack to stack
  608       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  609       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  610     } else {
  611       // stack to reg
  612       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  613     }
  614   } else if (dst.first()->is_stack()) {
  615     // reg to stack
  616     // Do we really have to sign extend???
  617     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  618     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  619   } else {
  620     // Do we really have to sign extend???
  621     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
  622     if (dst.first() != src.first()) {
  623       movq(dst.first()->as_Register(), src.first()->as_Register());
  624     }
  625   }
  626 }
  627 
  628 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
  629   if (src.first()->is_stack()) {
  630     if (dst.first()->is_stack()) {
  631       // stack to stack
  632       movq(rax, Address(rbp, reg2offset_in(src.first())));
  633       movq(Address(rsp, reg2offset_out(dst.first())), rax);
  634     } else {
  635       // stack to reg
  636       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
  637     }
  638   } else if (dst.first()->is_stack()) {
  639     // reg to stack
  640     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
  641   } else {
  642     if (dst.first() != src.first()) {
  643       movq(dst.first()->as_Register(), src.first()->as_Register());
  644     }
  645   }
  646 }
  647 
  648 // An oop arg. Must pass a handle not the oop itself
  649 void MacroAssembler::object_move(OopMap* map,
  650                         int oop_handle_offset,
  651                         int framesize_in_slots,
  652                         VMRegPair src,
  653                         VMRegPair dst,
  654                         bool is_receiver,
  655                         int* receiver_offset) {
  656 
  657   // must pass a handle. First figure out the location we use as a handle
  658 
  659   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
  660 
  661   // See if oop is null if it is we need no handle
  662 
  663   if (src.first()->is_stack()) {
  664 
  665     // Oop is already on the stack as an argument
  666     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
  667     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
  668     if (is_receiver) {
  669       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
  670     }
  671 
  672     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
  673     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
  674     // conditionally move a null
  675     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
  676   } else {
  677 
  678     // Oop is in a register we must store it to the space we reserve
  679     // on the stack for oop_handles and pass a handle if oop is non-null
  680 
  681     const Register rOop = src.first()->as_Register();
  682     int oop_slot;
  683     if (rOop == j_rarg0)
  684       oop_slot = 0;
  685     else if (rOop == j_rarg1)
  686       oop_slot = 1;
  687     else if (rOop == j_rarg2)
  688       oop_slot = 2;
  689     else if (rOop == j_rarg3)
  690       oop_slot = 3;
  691     else if (rOop == j_rarg4)
  692       oop_slot = 4;
  693     else {
  694       assert(rOop == j_rarg5, "wrong register");
  695       oop_slot = 5;
  696     }
  697 
  698     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
  699     int offset = oop_slot*VMRegImpl::stack_slot_size;
  700 
  701     map->set_oop(VMRegImpl::stack2reg(oop_slot));
  702     // Store oop in handle area, may be null
  703     movptr(Address(rsp, offset), rOop);
  704     if (is_receiver) {
  705       *receiver_offset = offset;
  706     }
  707 
  708     cmpptr(rOop, NULL_WORD);
  709     lea(rHandle, Address(rsp, offset));
  710     // conditionally move a null from the handle area where it was just stored
  711     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
  712   }
  713 
  714   // If arg is on the stack then place it otherwise it is already in correct reg.
  715   if (dst.first()->is_stack()) {
  716     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
  717   }
  718 }
  719 
  720 void MacroAssembler::addptr(Register dst, int32_t imm32) {
  721   addq(dst, imm32);
  722 }
  723 
  724 void MacroAssembler::addptr(Register dst, Register src) {
  725   addq(dst, src);
  726 }
  727 
  728 void MacroAssembler::addptr(Address dst, Register src) {
  729   addq(dst, src);
  730 }
  731 
  732 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  733   assert(rscratch != noreg || always_reachable(src), "missing");
  734 
  735   if (reachable(src)) {
  736     Assembler::addsd(dst, as_Address(src));
  737   } else {
  738     lea(rscratch, src);
  739     Assembler::addsd(dst, Address(rscratch, 0));
  740   }
  741 }
  742 
  743 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
  744   assert(rscratch != noreg || always_reachable(src), "missing");
  745 
  746   if (reachable(src)) {
  747     addss(dst, as_Address(src));
  748   } else {
  749     lea(rscratch, src);
  750     addss(dst, Address(rscratch, 0));
  751   }
  752 }
  753 
  754 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  755   assert(rscratch != noreg || always_reachable(src), "missing");
  756 
  757   if (reachable(src)) {
  758     Assembler::addpd(dst, as_Address(src));
  759   } else {
  760     lea(rscratch, src);
  761     Assembler::addpd(dst, Address(rscratch, 0));
  762   }
  763 }
  764 
  765 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
  766 // Stub code is generated once and never copied.
  767 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
  768 void MacroAssembler::align64() {
  769   align(64, (uint)(uintptr_t)pc());
  770 }
  771 
  772 void MacroAssembler::align32() {
  773   align(32, (uint)(uintptr_t)pc());
  774 }
  775 
  776 void MacroAssembler::align(uint modulus) {
  777   // 8273459: Ensure alignment is possible with current segment alignment
  778   assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
  779   align(modulus, offset());
  780 }
  781 
  782 void MacroAssembler::align(uint modulus, uint target) {
  783   if (target % modulus != 0) {
  784     nop(modulus - (target % modulus));
  785   }
  786 }
  787 
  788 void MacroAssembler::push_f(XMMRegister r) {
  789   subptr(rsp, wordSize);
  790   movflt(Address(rsp, 0), r);
  791 }
  792 
  793 void MacroAssembler::pop_f(XMMRegister r) {
  794   movflt(r, Address(rsp, 0));
  795   addptr(rsp, wordSize);
  796 }
  797 
  798 void MacroAssembler::push_d(XMMRegister r) {
  799   subptr(rsp, 2 * wordSize);
  800   movdbl(Address(rsp, 0), r);
  801 }
  802 
  803 void MacroAssembler::pop_d(XMMRegister r) {
  804   movdbl(r, Address(rsp, 0));
  805   addptr(rsp, 2 * Interpreter::stackElementSize);
  806 }
  807 
  808 void MacroAssembler::push_ppx(Register src) {
  809   if (VM_Version::supports_apx_f()) {
  810     pushp(src);
  811   } else {
  812     Assembler::push(src);
  813   }
  814 }
  815 
  816 void MacroAssembler::pop_ppx(Register dst) {
  817   if (VM_Version::supports_apx_f()) {
  818     popp(dst);
  819   } else {
  820     Assembler::pop(dst);
  821   }
  822 }
  823 
  824 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  825   // Used in sign-masking with aligned address.
  826   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  827   assert(rscratch != noreg || always_reachable(src), "missing");
  828 
  829   if (UseAVX > 2 &&
  830       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
  831       (dst->encoding() >= 16)) {
  832     vpand(dst, dst, src, AVX_512bit, rscratch);
  833   } else if (reachable(src)) {
  834     Assembler::andpd(dst, as_Address(src));
  835   } else {
  836     lea(rscratch, src);
  837     Assembler::andpd(dst, Address(rscratch, 0));
  838   }
  839 }
  840 
  841 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
  842   // Used in sign-masking with aligned address.
  843   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  844   assert(rscratch != noreg || always_reachable(src), "missing");
  845 
  846   if (reachable(src)) {
  847     Assembler::andps(dst, as_Address(src));
  848   } else {
  849     lea(rscratch, src);
  850     Assembler::andps(dst, Address(rscratch, 0));
  851   }
  852 }
  853 
  854 void MacroAssembler::andptr(Register dst, int32_t imm32) {
  855   andq(dst, imm32);
  856 }
  857 
  858 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
  859   assert(rscratch != noreg || always_reachable(src), "missing");
  860 
  861   if (reachable(src)) {
  862     andq(dst, as_Address(src));
  863   } else {
  864     lea(rscratch, src);
  865     andq(dst, Address(rscratch, 0));
  866   }
  867 }
  868 
  869 void MacroAssembler::atomic_incl(Address counter_addr) {
  870   lock();
  871   incrementl(counter_addr);
  872 }
  873 
  874 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
  875   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  876 
  877   if (reachable(counter_addr)) {
  878     atomic_incl(as_Address(counter_addr));
  879   } else {
  880     lea(rscratch, counter_addr);
  881     atomic_incl(Address(rscratch, 0));
  882   }
  883 }
  884 
  885 void MacroAssembler::atomic_incq(Address counter_addr) {
  886   lock();
  887   incrementq(counter_addr);
  888 }
  889 
  890 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
  891   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  892 
  893   if (reachable(counter_addr)) {
  894     atomic_incq(as_Address(counter_addr));
  895   } else {
  896     lea(rscratch, counter_addr);
  897     atomic_incq(Address(rscratch, 0));
  898   }
  899 }
  900 
  901 // Writes to stack successive pages until offset reached to check for
  902 // stack overflow + shadow pages.  This clobbers tmp.
  903 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
  904   movptr(tmp, rsp);
  905   // Bang stack for total size given plus shadow page size.
  906   // Bang one page at a time because large size can bang beyond yellow and
  907   // red zones.
  908   Label loop;
  909   bind(loop);
  910   movl(Address(tmp, (-(int)os::vm_page_size())), size );
  911   subptr(tmp, (int)os::vm_page_size());
  912   subl(size, (int)os::vm_page_size());
  913   jcc(Assembler::greater, loop);
  914 
  915   // Bang down shadow pages too.
  916   // At this point, (tmp-0) is the last address touched, so don't
  917   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
  918   // was post-decremented.)  Skip this address by starting at i=1, and
  919   // touch a few more pages below.  N.B.  It is important to touch all
  920   // the way down including all pages in the shadow zone.
  921   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
  922     // this could be any sized move but this is can be a debugging crumb
  923     // so the bigger the better.
  924     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
  925   }
  926 }
  927 
  928 void MacroAssembler::reserved_stack_check() {
  929   // testing if reserved zone needs to be enabled
  930   Label no_reserved_zone_enabling;
  931 
  932   cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
  933   jcc(Assembler::below, no_reserved_zone_enabling);
  934 
  935   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
  936   jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
  937   should_not_reach_here();
  938 
  939   bind(no_reserved_zone_enabling);
  940 }
  941 
  942 void MacroAssembler::c2bool(Register x) {
  943   // implements x == 0 ? 0 : 1
  944   // note: must only look at least-significant byte of x
  945   //       since C-style booleans are stored in one byte
  946   //       only! (was bug)
  947   andl(x, 0xFF);
  948   setb(Assembler::notZero, x);
  949 }
  950 
  951 // Wouldn't need if AddressLiteral version had new name
  952 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
  953   Assembler::call(L, rtype);
  954 }
  955 
  956 void MacroAssembler::call(Register entry) {
  957   Assembler::call(entry);
  958 }
  959 
  960 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
  961   assert(rscratch != noreg || always_reachable(entry), "missing");
  962 
  963   if (reachable(entry)) {
  964     Assembler::call_literal(entry.target(), entry.rspec());
  965   } else {
  966     lea(rscratch, entry);
  967     Assembler::call(rscratch);
  968   }
  969 }
  970 
  971 void MacroAssembler::ic_call(address entry, jint method_index) {
  972   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
  973   // Needs full 64-bit immediate for later patching.
  974   Assembler::mov64(rax, (int64_t)Universe::non_oop_word());
  975   call(AddressLiteral(entry, rh));
  976 }
  977 
  978 int MacroAssembler::ic_check_size() {
  979   return UseCompactObjectHeaders ? 17 : 14;
  980 }
  981 
  982 int MacroAssembler::ic_check(int end_alignment) {
  983   Register receiver = j_rarg0;
  984   Register data = rax;
  985   Register temp = rscratch1;
  986 
  987   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
  988   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
  989   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
  990   // before the inline cache check here, and not after
  991   align(end_alignment, offset() + ic_check_size());
  992 
  993   int uep_offset = offset();
  994 
  995   if (UseCompactObjectHeaders) {
  996     load_narrow_klass_compact(temp, receiver);
  997     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  998   } else {
  999     movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
 1000     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
 1001   }
 1002 
 1003   // if inline cache check fails, then jump to runtime routine
 1004   jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 1005   assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
 1006 
 1007   return uep_offset;
 1008 }
 1009 
 1010 void MacroAssembler::emit_static_call_stub() {
 1011   // Static stub relocation also tags the Method* in the code-stream.
 1012   mov_metadata(rbx, (Metadata*) nullptr);  // Method is zapped till fixup time.
 1013   // This is recognized as unresolved by relocs/nativeinst/ic code.
 1014   jump(RuntimeAddress(pc()));
 1015 }
 1016 
 1017 // Implementation of call_VM versions
 1018 
 1019 void MacroAssembler::call_VM(Register oop_result,
 1020                              address entry_point,
 1021                              bool check_exceptions) {
 1022   Label C, E;
 1023   call(C, relocInfo::none);
 1024   jmp(E);
 1025 
 1026   bind(C);
 1027   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
 1028   ret(0);
 1029 
 1030   bind(E);
 1031 }
 1032 
 1033 void MacroAssembler::call_VM(Register oop_result,
 1034                              address entry_point,
 1035                              Register arg_1,
 1036                              bool check_exceptions) {
 1037   Label C, E;
 1038   call(C, relocInfo::none);
 1039   jmp(E);
 1040 
 1041   bind(C);
 1042   pass_arg1(this, arg_1);
 1043   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
 1044   ret(0);
 1045 
 1046   bind(E);
 1047 }
 1048 
 1049 void MacroAssembler::call_VM(Register oop_result,
 1050                              address entry_point,
 1051                              Register arg_1,
 1052                              Register arg_2,
 1053                              bool check_exceptions) {
 1054   Label C, E;
 1055   call(C, relocInfo::none);
 1056   jmp(E);
 1057 
 1058   bind(C);
 1059 
 1060   assert_different_registers(arg_1, c_rarg2);
 1061 
 1062   pass_arg2(this, arg_2);
 1063   pass_arg1(this, arg_1);
 1064   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
 1065   ret(0);
 1066 
 1067   bind(E);
 1068 }
 1069 
 1070 void MacroAssembler::call_VM(Register oop_result,
 1071                              address entry_point,
 1072                              Register arg_1,
 1073                              Register arg_2,
 1074                              Register arg_3,
 1075                              bool check_exceptions) {
 1076   Label C, E;
 1077   call(C, relocInfo::none);
 1078   jmp(E);
 1079 
 1080   bind(C);
 1081 
 1082   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1083   assert_different_registers(arg_2, c_rarg3);
 1084   pass_arg3(this, arg_3);
 1085   pass_arg2(this, arg_2);
 1086   pass_arg1(this, arg_1);
 1087   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
 1088   ret(0);
 1089 
 1090   bind(E);
 1091 }
 1092 
 1093 void MacroAssembler::call_VM(Register oop_result,
 1094                              Register last_java_sp,
 1095                              address entry_point,
 1096                              int number_of_arguments,
 1097                              bool check_exceptions) {
 1098   call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1099 }
 1100 
 1101 void MacroAssembler::call_VM(Register oop_result,
 1102                              Register last_java_sp,
 1103                              address entry_point,
 1104                              Register arg_1,
 1105                              bool check_exceptions) {
 1106   pass_arg1(this, arg_1);
 1107   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1108 }
 1109 
 1110 void MacroAssembler::call_VM(Register oop_result,
 1111                              Register last_java_sp,
 1112                              address entry_point,
 1113                              Register arg_1,
 1114                              Register arg_2,
 1115                              bool check_exceptions) {
 1116 
 1117   assert_different_registers(arg_1, c_rarg2);
 1118   pass_arg2(this, arg_2);
 1119   pass_arg1(this, arg_1);
 1120   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1121 }
 1122 
 1123 void MacroAssembler::call_VM(Register oop_result,
 1124                              Register last_java_sp,
 1125                              address entry_point,
 1126                              Register arg_1,
 1127                              Register arg_2,
 1128                              Register arg_3,
 1129                              bool check_exceptions) {
 1130   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1131   assert_different_registers(arg_2, c_rarg3);
 1132   pass_arg3(this, arg_3);
 1133   pass_arg2(this, arg_2);
 1134   pass_arg1(this, arg_1);
 1135   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1136 }
 1137 
 1138 void MacroAssembler::super_call_VM(Register oop_result,
 1139                                    Register last_java_sp,
 1140                                    address entry_point,
 1141                                    int number_of_arguments,
 1142                                    bool check_exceptions) {
 1143   MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1144 }
 1145 
 1146 void MacroAssembler::super_call_VM(Register oop_result,
 1147                                    Register last_java_sp,
 1148                                    address entry_point,
 1149                                    Register arg_1,
 1150                                    bool check_exceptions) {
 1151   pass_arg1(this, arg_1);
 1152   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1153 }
 1154 
 1155 void MacroAssembler::super_call_VM(Register oop_result,
 1156                                    Register last_java_sp,
 1157                                    address entry_point,
 1158                                    Register arg_1,
 1159                                    Register arg_2,
 1160                                    bool check_exceptions) {
 1161 
 1162   assert_different_registers(arg_1, c_rarg2);
 1163   pass_arg2(this, arg_2);
 1164   pass_arg1(this, arg_1);
 1165   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1166 }
 1167 
 1168 void MacroAssembler::super_call_VM(Register oop_result,
 1169                                    Register last_java_sp,
 1170                                    address entry_point,
 1171                                    Register arg_1,
 1172                                    Register arg_2,
 1173                                    Register arg_3,
 1174                                    bool check_exceptions) {
 1175   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1176   assert_different_registers(arg_2, c_rarg3);
 1177   pass_arg3(this, arg_3);
 1178   pass_arg2(this, arg_2);
 1179   pass_arg1(this, arg_1);
 1180   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1181 }
 1182 
 1183 void MacroAssembler::call_VM_base(Register oop_result,
 1184                                   Register last_java_sp,
 1185                                   address  entry_point,
 1186                                   int      number_of_arguments,
 1187                                   bool     check_exceptions) {
 1188   Register java_thread = r15_thread;
 1189 
 1190   // determine last_java_sp register
 1191   if (!last_java_sp->is_valid()) {
 1192     last_java_sp = rsp;
 1193   }
 1194   // debugging support
 1195   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 1196 #ifdef ASSERT
 1197   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 1198   // r12 is the heapbase.
 1199   if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 1200 #endif // ASSERT
 1201 
 1202   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 1203   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 1204 
 1205   // push java thread (becomes first argument of C function)
 1206 
 1207   mov(c_rarg0, r15_thread);
 1208 
 1209   // set last Java frame before call
 1210   assert(last_java_sp != rbp, "can't use ebp/rbp");
 1211 
 1212   // Only interpreter should have to set fp
 1213   set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
 1214 
 1215   // do the call, remove parameters
 1216   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 1217 
 1218 #ifdef ASSERT
 1219   // Check that thread register is not clobbered.
 1220   guarantee(java_thread != rax, "change this code");
 1221   push(rax);
 1222   { Label L;
 1223     get_thread_slow(rax);
 1224     cmpptr(java_thread, rax);
 1225     jcc(Assembler::equal, L);
 1226     STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
 1227     bind(L);
 1228   }
 1229   pop(rax);
 1230 #endif
 1231 
 1232   // reset last Java frame
 1233   // Only interpreter should have to clear fp
 1234   reset_last_Java_frame(true);
 1235 
 1236    // C++ interp handles this in the interpreter
 1237   check_and_handle_popframe();
 1238   check_and_handle_earlyret();
 1239 
 1240   if (check_exceptions) {
 1241     // check for pending exceptions (java_thread is set upon return)
 1242     cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 1243     // This used to conditionally jump to forward_exception however it is
 1244     // possible if we relocate that the branch will not reach. So we must jump
 1245     // around so we can always reach
 1246 
 1247     Label ok;
 1248     jcc(Assembler::equal, ok);
 1249     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 1250     bind(ok);
 1251   }
 1252 
 1253   // get oop result if there is one and reset the value in the thread
 1254   if (oop_result->is_valid()) {
 1255     get_vm_result_oop(oop_result);
 1256   }
 1257 }
 1258 
 1259 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 1260   // Calculate the value for last_Java_sp somewhat subtle.
 1261   // call_VM does an intermediate call which places a return address on
 1262   // the stack just under the stack pointer as the user finished with it.
 1263   // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
 1264 
 1265   // We've pushed one address, correct last_Java_sp
 1266   lea(rax, Address(rsp, wordSize));
 1267 
 1268   call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
 1269 }
 1270 
 1271 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
 1272 void MacroAssembler::call_VM_leaf0(address entry_point) {
 1273   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 1274 }
 1275 
 1276 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
 1277   call_VM_leaf_base(entry_point, number_of_arguments);
 1278 }
 1279 
 1280 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
 1281   pass_arg0(this, arg_0);
 1282   call_VM_leaf(entry_point, 1);
 1283 }
 1284 
 1285 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1286 
 1287   assert_different_registers(arg_0, c_rarg1);
 1288   pass_arg1(this, arg_1);
 1289   pass_arg0(this, arg_0);
 1290   call_VM_leaf(entry_point, 2);
 1291 }
 1292 
 1293 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1294   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1295   assert_different_registers(arg_1, c_rarg2);
 1296   pass_arg2(this, arg_2);
 1297   pass_arg1(this, arg_1);
 1298   pass_arg0(this, arg_0);
 1299   call_VM_leaf(entry_point, 3);
 1300 }
 1301 
 1302 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1303   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1304   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1305   assert_different_registers(arg_2, c_rarg3);
 1306   pass_arg3(this, arg_3);
 1307   pass_arg2(this, arg_2);
 1308   pass_arg1(this, arg_1);
 1309   pass_arg0(this, arg_0);
 1310   call_VM_leaf(entry_point, 3);
 1311 }
 1312 
 1313 void MacroAssembler::super_call_VM_leaf(address entry_point) {
 1314   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1315 }
 1316 
 1317 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1318   pass_arg0(this, arg_0);
 1319   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1320 }
 1321 
 1322 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1323   assert_different_registers(arg_0, c_rarg1);
 1324   pass_arg1(this, arg_1);
 1325   pass_arg0(this, arg_0);
 1326   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1327 }
 1328 
 1329 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1330   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1331   assert_different_registers(arg_1, c_rarg2);
 1332   pass_arg2(this, arg_2);
 1333   pass_arg1(this, arg_1);
 1334   pass_arg0(this, arg_0);
 1335   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1336 }
 1337 
 1338 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1339   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1340   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1341   assert_different_registers(arg_2, c_rarg3);
 1342   pass_arg3(this, arg_3);
 1343   pass_arg2(this, arg_2);
 1344   pass_arg1(this, arg_1);
 1345   pass_arg0(this, arg_0);
 1346   MacroAssembler::call_VM_leaf_base(entry_point, 4);
 1347 }
 1348 
 1349 void MacroAssembler::get_vm_result_oop(Register oop_result) {
 1350   movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
 1351   movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
 1352   verify_oop_msg(oop_result, "broken oop in call_VM_base");
 1353 }
 1354 
 1355 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
 1356   movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
 1357   movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
 1358 }
 1359 
 1360 void MacroAssembler::check_and_handle_earlyret() {
 1361 }
 1362 
 1363 void MacroAssembler::check_and_handle_popframe() {
 1364 }
 1365 
 1366 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
 1367   assert(rscratch != noreg || always_reachable(src1), "missing");
 1368 
 1369   if (reachable(src1)) {
 1370     cmpl(as_Address(src1), imm);
 1371   } else {
 1372     lea(rscratch, src1);
 1373     cmpl(Address(rscratch, 0), imm);
 1374   }
 1375 }
 1376 
 1377 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
 1378   assert(!src2.is_lval(), "use cmpptr");
 1379   assert(rscratch != noreg || always_reachable(src2), "missing");
 1380 
 1381   if (reachable(src2)) {
 1382     cmpl(src1, as_Address(src2));
 1383   } else {
 1384     lea(rscratch, src2);
 1385     cmpl(src1, Address(rscratch, 0));
 1386   }
 1387 }
 1388 
 1389 void MacroAssembler::cmp32(Register src1, int32_t imm) {
 1390   Assembler::cmpl(src1, imm);
 1391 }
 1392 
 1393 void MacroAssembler::cmp32(Register src1, Address src2) {
 1394   Assembler::cmpl(src1, src2);
 1395 }
 1396 
 1397 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1398   ucomisd(opr1, opr2);
 1399 
 1400   Label L;
 1401   if (unordered_is_less) {
 1402     movl(dst, -1);
 1403     jcc(Assembler::parity, L);
 1404     jcc(Assembler::below , L);
 1405     movl(dst, 0);
 1406     jcc(Assembler::equal , L);
 1407     increment(dst);
 1408   } else { // unordered is greater
 1409     movl(dst, 1);
 1410     jcc(Assembler::parity, L);
 1411     jcc(Assembler::above , L);
 1412     movl(dst, 0);
 1413     jcc(Assembler::equal , L);
 1414     decrementl(dst);
 1415   }
 1416   bind(L);
 1417 }
 1418 
 1419 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1420   ucomiss(opr1, opr2);
 1421 
 1422   Label L;
 1423   if (unordered_is_less) {
 1424     movl(dst, -1);
 1425     jcc(Assembler::parity, L);
 1426     jcc(Assembler::below , L);
 1427     movl(dst, 0);
 1428     jcc(Assembler::equal , L);
 1429     increment(dst);
 1430   } else { // unordered is greater
 1431     movl(dst, 1);
 1432     jcc(Assembler::parity, L);
 1433     jcc(Assembler::above , L);
 1434     movl(dst, 0);
 1435     jcc(Assembler::equal , L);
 1436     decrementl(dst);
 1437   }
 1438   bind(L);
 1439 }
 1440 
 1441 
 1442 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
 1443   assert(rscratch != noreg || always_reachable(src1), "missing");
 1444 
 1445   if (reachable(src1)) {
 1446     cmpb(as_Address(src1), imm);
 1447   } else {
 1448     lea(rscratch, src1);
 1449     cmpb(Address(rscratch, 0), imm);
 1450   }
 1451 }
 1452 
 1453 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
 1454   assert(rscratch != noreg || always_reachable(src2), "missing");
 1455 
 1456   if (src2.is_lval()) {
 1457     movptr(rscratch, src2);
 1458     Assembler::cmpq(src1, rscratch);
 1459   } else if (reachable(src2)) {
 1460     cmpq(src1, as_Address(src2));
 1461   } else {
 1462     lea(rscratch, src2);
 1463     Assembler::cmpq(src1, Address(rscratch, 0));
 1464   }
 1465 }
 1466 
 1467 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
 1468   assert(src2.is_lval(), "not a mem-mem compare");
 1469   // moves src2's literal address
 1470   movptr(rscratch, src2);
 1471   Assembler::cmpq(src1, rscratch);
 1472 }
 1473 
 1474 void MacroAssembler::cmpoop(Register src1, Register src2) {
 1475   cmpptr(src1, src2);
 1476 }
 1477 
 1478 void MacroAssembler::cmpoop(Register src1, Address src2) {
 1479   cmpptr(src1, src2);
 1480 }
 1481 
 1482 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
 1483   movoop(rscratch, src2);
 1484   cmpptr(src1, rscratch);
 1485 }
 1486 
 1487 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
 1488   assert(rscratch != noreg || always_reachable(adr), "missing");
 1489 
 1490   if (reachable(adr)) {
 1491     lock();
 1492     cmpxchgptr(reg, as_Address(adr));
 1493   } else {
 1494     lea(rscratch, adr);
 1495     lock();
 1496     cmpxchgptr(reg, Address(rscratch, 0));
 1497   }
 1498 }
 1499 
 1500 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
 1501   cmpxchgq(reg, adr);
 1502 }
 1503 
 1504 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1505   assert(rscratch != noreg || always_reachable(src), "missing");
 1506 
 1507   if (reachable(src)) {
 1508     Assembler::comisd(dst, as_Address(src));
 1509   } else {
 1510     lea(rscratch, src);
 1511     Assembler::comisd(dst, Address(rscratch, 0));
 1512   }
 1513 }
 1514 
 1515 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1516   assert(rscratch != noreg || always_reachable(src), "missing");
 1517 
 1518   if (reachable(src)) {
 1519     Assembler::comiss(dst, as_Address(src));
 1520   } else {
 1521     lea(rscratch, src);
 1522     Assembler::comiss(dst, Address(rscratch, 0));
 1523   }
 1524 }
 1525 
 1526 
 1527 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
 1528   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1529 
 1530   Condition negated_cond = negate_condition(cond);
 1531   Label L;
 1532   jcc(negated_cond, L);
 1533   pushf(); // Preserve flags
 1534   atomic_incl(counter_addr, rscratch);
 1535   popf();
 1536   bind(L);
 1537 }
 1538 
 1539 int MacroAssembler::corrected_idivl(Register reg) {
 1540   // Full implementation of Java idiv and irem; checks for
 1541   // special case as described in JVM spec., p.243 & p.271.
 1542   // The function returns the (pc) offset of the idivl
 1543   // instruction - may be needed for implicit exceptions.
 1544   //
 1545   //         normal case                           special case
 1546   //
 1547   // input : rax,: dividend                         min_int
 1548   //         reg: divisor   (may not be rax,/rdx)   -1
 1549   //
 1550   // output: rax,: quotient  (= rax, idiv reg)       min_int
 1551   //         rdx: remainder (= rax, irem reg)       0
 1552   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
 1553   const int min_int = 0x80000000;
 1554   Label normal_case, special_case;
 1555 
 1556   // check for special case
 1557   cmpl(rax, min_int);
 1558   jcc(Assembler::notEqual, normal_case);
 1559   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
 1560   cmpl(reg, -1);
 1561   jcc(Assembler::equal, special_case);
 1562 
 1563   // handle normal case
 1564   bind(normal_case);
 1565   cdql();
 1566   int idivl_offset = offset();
 1567   idivl(reg);
 1568 
 1569   // normal and special case exit
 1570   bind(special_case);
 1571 
 1572   return idivl_offset;
 1573 }
 1574 
 1575 
 1576 
 1577 void MacroAssembler::decrementl(Register reg, int value) {
 1578   if (value == min_jint) {subl(reg, value) ; return; }
 1579   if (value <  0) { incrementl(reg, -value); return; }
 1580   if (value == 0) {                        ; return; }
 1581   if (value == 1 && UseIncDec) { decl(reg) ; return; }
 1582   /* else */      { subl(reg, value)       ; return; }
 1583 }
 1584 
 1585 void MacroAssembler::decrementl(Address dst, int value) {
 1586   if (value == min_jint) {subl(dst, value) ; return; }
 1587   if (value <  0) { incrementl(dst, -value); return; }
 1588   if (value == 0) {                        ; return; }
 1589   if (value == 1 && UseIncDec) { decl(dst) ; return; }
 1590   /* else */      { subl(dst, value)       ; return; }
 1591 }
 1592 
 1593 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
 1594   assert(shift_value > 0, "illegal shift value");
 1595   Label _is_positive;
 1596   testl (reg, reg);
 1597   jcc (Assembler::positive, _is_positive);
 1598   int offset = (1 << shift_value) - 1 ;
 1599 
 1600   if (offset == 1) {
 1601     incrementl(reg);
 1602   } else {
 1603     addl(reg, offset);
 1604   }
 1605 
 1606   bind (_is_positive);
 1607   sarl(reg, shift_value);
 1608 }
 1609 
 1610 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1611   assert(rscratch != noreg || always_reachable(src), "missing");
 1612 
 1613   if (reachable(src)) {
 1614     Assembler::divsd(dst, as_Address(src));
 1615   } else {
 1616     lea(rscratch, src);
 1617     Assembler::divsd(dst, Address(rscratch, 0));
 1618   }
 1619 }
 1620 
 1621 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1622   assert(rscratch != noreg || always_reachable(src), "missing");
 1623 
 1624   if (reachable(src)) {
 1625     Assembler::divss(dst, as_Address(src));
 1626   } else {
 1627     lea(rscratch, src);
 1628     Assembler::divss(dst, Address(rscratch, 0));
 1629   }
 1630 }
 1631 
 1632 void MacroAssembler::enter() {
 1633   push(rbp);
 1634   mov(rbp, rsp);
 1635 }
 1636 
 1637 void MacroAssembler::post_call_nop() {
 1638   if (!Continuations::enabled()) {
 1639     return;
 1640   }
 1641   InstructionMark im(this);
 1642   relocate(post_call_nop_Relocation::spec());
 1643   InlineSkippedInstructionsCounter skipCounter(this);
 1644   emit_int8((uint8_t)0x0f);
 1645   emit_int8((uint8_t)0x1f);
 1646   emit_int8((uint8_t)0x84);
 1647   emit_int8((uint8_t)0x00);
 1648   emit_int32(0x00);
 1649 }
 1650 
 1651 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1652   assert(rscratch != noreg || always_reachable(src), "missing");
 1653   if (reachable(src)) {
 1654     Assembler::mulpd(dst, as_Address(src));
 1655   } else {
 1656     lea(rscratch, src);
 1657     Assembler::mulpd(dst, Address(rscratch, 0));
 1658   }
 1659 }
 1660 
 1661 // dst = c = a * b + c
 1662 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1663   Assembler::vfmadd231sd(c, a, b);
 1664   if (dst != c) {
 1665     movdbl(dst, c);
 1666   }
 1667 }
 1668 
 1669 // dst = c = a * b + c
 1670 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1671   Assembler::vfmadd231ss(c, a, b);
 1672   if (dst != c) {
 1673     movflt(dst, c);
 1674   }
 1675 }
 1676 
 1677 // dst = c = a * b + c
 1678 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1679   Assembler::vfmadd231pd(c, a, b, vector_len);
 1680   if (dst != c) {
 1681     vmovdqu(dst, c);
 1682   }
 1683 }
 1684 
 1685 // dst = c = a * b + c
 1686 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1687   Assembler::vfmadd231ps(c, a, b, vector_len);
 1688   if (dst != c) {
 1689     vmovdqu(dst, c);
 1690   }
 1691 }
 1692 
 1693 // dst = c = a * b + c
 1694 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1695   Assembler::vfmadd231pd(c, a, b, vector_len);
 1696   if (dst != c) {
 1697     vmovdqu(dst, c);
 1698   }
 1699 }
 1700 
 1701 // dst = c = a * b + c
 1702 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1703   Assembler::vfmadd231ps(c, a, b, vector_len);
 1704   if (dst != c) {
 1705     vmovdqu(dst, c);
 1706   }
 1707 }
 1708 
 1709 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
 1710   assert(rscratch != noreg || always_reachable(dst), "missing");
 1711 
 1712   if (reachable(dst)) {
 1713     incrementl(as_Address(dst));
 1714   } else {
 1715     lea(rscratch, dst);
 1716     incrementl(Address(rscratch, 0));
 1717   }
 1718 }
 1719 
 1720 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
 1721   incrementl(as_Address(dst, rscratch));
 1722 }
 1723 
 1724 void MacroAssembler::incrementl(Register reg, int value) {
 1725   if (value == min_jint) {addl(reg, value) ; return; }
 1726   if (value <  0) { decrementl(reg, -value); return; }
 1727   if (value == 0) {                        ; return; }
 1728   if (value == 1 && UseIncDec) { incl(reg) ; return; }
 1729   /* else */      { addl(reg, value)       ; return; }
 1730 }
 1731 
 1732 void MacroAssembler::incrementl(Address dst, int value) {
 1733   if (value == min_jint) {addl(dst, value) ; return; }
 1734   if (value <  0) { decrementl(dst, -value); return; }
 1735   if (value == 0) {                        ; return; }
 1736   if (value == 1 && UseIncDec) { incl(dst) ; return; }
 1737   /* else */      { addl(dst, value)       ; return; }
 1738 }
 1739 
 1740 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
 1741   assert(rscratch != noreg || always_reachable(dst), "missing");
 1742   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
 1743   if (reachable(dst)) {
 1744     jmp_literal(dst.target(), dst.rspec());
 1745   } else {
 1746     lea(rscratch, dst);
 1747     jmp(rscratch);
 1748   }
 1749 }
 1750 
 1751 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
 1752   assert(rscratch != noreg || always_reachable(dst), "missing");
 1753   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump_cc");
 1754   if (reachable(dst)) {
 1755     InstructionMark im(this);
 1756     relocate(dst.reloc());
 1757     const int short_size = 2;
 1758     const int long_size = 6;
 1759     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
 1760     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
 1761       // 0111 tttn #8-bit disp
 1762       emit_int8(0x70 | cc);
 1763       emit_int8((offs - short_size) & 0xFF);
 1764     } else {
 1765       // 0000 1111 1000 tttn #32-bit disp
 1766       emit_int8(0x0F);
 1767       emit_int8((unsigned char)(0x80 | cc));
 1768       emit_int32(offs - long_size);
 1769     }
 1770   } else {
 1771 #ifdef ASSERT
 1772     warning("reversing conditional branch");
 1773 #endif /* ASSERT */
 1774     Label skip;
 1775     jccb(reverse[cc], skip);
 1776     lea(rscratch, dst);
 1777     Assembler::jmp(rscratch);
 1778     bind(skip);
 1779   }
 1780 }
 1781 
 1782 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
 1783   ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 1784   assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
 1785 
 1786   stmxcsr(mxcsr_save);
 1787   movl(tmp, mxcsr_save);
 1788   if (EnableX86ECoreOpts) {
 1789     // The mxcsr_std has status bits set for performance on ECore
 1790     orl(tmp, 0x003f);
 1791   } else {
 1792     // Mask out status bits (only check control and mask bits)
 1793     andl(tmp, 0xFFC0);
 1794   }
 1795   cmp32(tmp, mxcsr_std, rscratch);
 1796 }
 1797 
 1798 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
 1799   assert(rscratch != noreg || always_reachable(src), "missing");
 1800 
 1801   if (reachable(src)) {
 1802     Assembler::ldmxcsr(as_Address(src));
 1803   } else {
 1804     lea(rscratch, src);
 1805     Assembler::ldmxcsr(Address(rscratch, 0));
 1806   }
 1807 }
 1808 
 1809 int MacroAssembler::load_signed_byte(Register dst, Address src) {
 1810   int off = offset();
 1811   movsbl(dst, src); // movsxb
 1812   return off;
 1813 }
 1814 
 1815 // Note: load_signed_short used to be called load_signed_word.
 1816 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
 1817 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
 1818 // The term "word" in HotSpot means a 32- or 64-bit machine word.
 1819 int MacroAssembler::load_signed_short(Register dst, Address src) {
 1820   // This is dubious to me since it seems safe to do a signed 16 => 64 bit
 1821   // version but this is what 64bit has always done. This seems to imply
 1822   // that users are only using 32bits worth.
 1823   int off = offset();
 1824   movswl(dst, src); // movsxw
 1825   return off;
 1826 }
 1827 
 1828 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
 1829   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1830   // and "3.9 Partial Register Penalties", p. 22).
 1831   int off = offset();
 1832   movzbl(dst, src); // movzxb
 1833   return off;
 1834 }
 1835 
 1836 // Note: load_unsigned_short used to be called load_unsigned_word.
 1837 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
 1838   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1839   // and "3.9 Partial Register Penalties", p. 22).
 1840   int off = offset();
 1841   movzwl(dst, src); // movzxw
 1842   return off;
 1843 }
 1844 
 1845 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
 1846   switch (size_in_bytes) {
 1847   case  8:  movq(dst, src); break;
 1848   case  4:  movl(dst, src); break;
 1849   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
 1850   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
 1851   default:  ShouldNotReachHere();
 1852   }
 1853 }
 1854 
 1855 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
 1856   switch (size_in_bytes) {
 1857   case  8:  movq(dst, src); break;
 1858   case  4:  movl(dst, src); break;
 1859   case  2:  movw(dst, src); break;
 1860   case  1:  movb(dst, src); break;
 1861   default:  ShouldNotReachHere();
 1862   }
 1863 }
 1864 
 1865 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
 1866   assert(rscratch != noreg || always_reachable(dst), "missing");
 1867 
 1868   if (reachable(dst)) {
 1869     movl(as_Address(dst), src);
 1870   } else {
 1871     lea(rscratch, dst);
 1872     movl(Address(rscratch, 0), src);
 1873   }
 1874 }
 1875 
 1876 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
 1877   if (reachable(src)) {
 1878     movl(dst, as_Address(src));
 1879   } else {
 1880     lea(dst, src);
 1881     movl(dst, Address(dst, 0));
 1882   }
 1883 }
 1884 
 1885 // C++ bool manipulation
 1886 
 1887 void MacroAssembler::movbool(Register dst, Address src) {
 1888   if(sizeof(bool) == 1)
 1889     movb(dst, src);
 1890   else if(sizeof(bool) == 2)
 1891     movw(dst, src);
 1892   else if(sizeof(bool) == 4)
 1893     movl(dst, src);
 1894   else
 1895     // unsupported
 1896     ShouldNotReachHere();
 1897 }
 1898 
 1899 void MacroAssembler::movbool(Address dst, bool boolconst) {
 1900   if(sizeof(bool) == 1)
 1901     movb(dst, (int) boolconst);
 1902   else if(sizeof(bool) == 2)
 1903     movw(dst, (int) boolconst);
 1904   else if(sizeof(bool) == 4)
 1905     movl(dst, (int) boolconst);
 1906   else
 1907     // unsupported
 1908     ShouldNotReachHere();
 1909 }
 1910 
 1911 void MacroAssembler::movbool(Address dst, Register src) {
 1912   if(sizeof(bool) == 1)
 1913     movb(dst, src);
 1914   else if(sizeof(bool) == 2)
 1915     movw(dst, src);
 1916   else if(sizeof(bool) == 4)
 1917     movl(dst, src);
 1918   else
 1919     // unsupported
 1920     ShouldNotReachHere();
 1921 }
 1922 
 1923 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1924   assert(rscratch != noreg || always_reachable(src), "missing");
 1925 
 1926   if (reachable(src)) {
 1927     movdl(dst, as_Address(src));
 1928   } else {
 1929     lea(rscratch, src);
 1930     movdl(dst, Address(rscratch, 0));
 1931   }
 1932 }
 1933 
 1934 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1935   assert(rscratch != noreg || always_reachable(src), "missing");
 1936 
 1937   if (reachable(src)) {
 1938     movq(dst, as_Address(src));
 1939   } else {
 1940     lea(rscratch, src);
 1941     movq(dst, Address(rscratch, 0));
 1942   }
 1943 }
 1944 
 1945 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1946   assert(rscratch != noreg || always_reachable(src), "missing");
 1947 
 1948   if (reachable(src)) {
 1949     if (UseXmmLoadAndClearUpper) {
 1950       movsd (dst, as_Address(src));
 1951     } else {
 1952       movlpd(dst, as_Address(src));
 1953     }
 1954   } else {
 1955     lea(rscratch, src);
 1956     if (UseXmmLoadAndClearUpper) {
 1957       movsd (dst, Address(rscratch, 0));
 1958     } else {
 1959       movlpd(dst, Address(rscratch, 0));
 1960     }
 1961   }
 1962 }
 1963 
 1964 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1965   assert(rscratch != noreg || always_reachable(src), "missing");
 1966 
 1967   if (reachable(src)) {
 1968     movss(dst, as_Address(src));
 1969   } else {
 1970     lea(rscratch, src);
 1971     movss(dst, Address(rscratch, 0));
 1972   }
 1973 }
 1974 
 1975 void MacroAssembler::movhlf(XMMRegister dst, XMMRegister src, Register rscratch) {
 1976   if (VM_Version::supports_avx10_2()) {
 1977     evmovw(dst, src);
 1978   } else {
 1979     assert(rscratch != noreg, "missing");
 1980     evmovw(rscratch, src);
 1981     evmovw(dst, rscratch);
 1982   }
 1983 }
 1984 
 1985 void MacroAssembler::mov64(Register dst, int64_t imm64) {
 1986   if (is_uimm32(imm64)) {
 1987     movl(dst, checked_cast<uint32_t>(imm64));
 1988   } else if (is_simm32(imm64)) {
 1989     movq(dst, checked_cast<int32_t>(imm64));
 1990   } else {
 1991     Assembler::mov64(dst, imm64);
 1992   }
 1993 }
 1994 
 1995 void MacroAssembler::mov64(Register dst, int64_t imm64, relocInfo::relocType rtype, int format) {
 1996   Assembler::mov64(dst, imm64, rtype, format);
 1997 }
 1998 
 1999 void MacroAssembler::movptr(Register dst, Register src) {
 2000   movq(dst, src);
 2001 }
 2002 
 2003 void MacroAssembler::movptr(Register dst, Address src) {
 2004   movq(dst, src);
 2005 }
 2006 
 2007 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 2008 void MacroAssembler::movptr(Register dst, intptr_t src) {
 2009   mov64(dst, src);
 2010 }
 2011 
 2012 void MacroAssembler::movptr(Address dst, Register src) {
 2013   movq(dst, src);
 2014 }
 2015 
 2016 void MacroAssembler::movptr(Address dst, int32_t src) {
 2017   movslq(dst, src);
 2018 }
 2019 
 2020 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 2021   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2022   Assembler::movdqu(dst, src);
 2023 }
 2024 
 2025 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 2026   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2027   Assembler::movdqu(dst, src);
 2028 }
 2029 
 2030 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2031   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2032   Assembler::movdqu(dst, src);
 2033 }
 2034 
 2035 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2036   assert(rscratch != noreg || always_reachable(src), "missing");
 2037 
 2038   if (reachable(src)) {
 2039     movdqu(dst, as_Address(src));
 2040   } else {
 2041     lea(rscratch, src);
 2042     movdqu(dst, Address(rscratch, 0));
 2043   }
 2044 }
 2045 
 2046 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2047   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2048   Assembler::vmovdqu(dst, src);
 2049 }
 2050 
 2051 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2052   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2053   Assembler::vmovdqu(dst, src);
 2054 }
 2055 
 2056 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2057   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2058   Assembler::vmovdqu(dst, src);
 2059 }
 2060 
 2061 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2062   assert(rscratch != noreg || always_reachable(src), "missing");
 2063 
 2064   if (reachable(src)) {
 2065     vmovdqu(dst, as_Address(src));
 2066   }
 2067   else {
 2068     lea(rscratch, src);
 2069     vmovdqu(dst, Address(rscratch, 0));
 2070   }
 2071 }
 2072 
 2073 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2074   assert(rscratch != noreg || always_reachable(src), "missing");
 2075 
 2076   if (vector_len == AVX_512bit) {
 2077     evmovdquq(dst, src, AVX_512bit, rscratch);
 2078   } else if (vector_len == AVX_256bit) {
 2079     vmovdqu(dst, src, rscratch);
 2080   } else {
 2081     movdqu(dst, src, rscratch);
 2082   }
 2083 }
 2084 
 2085 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
 2086   if (vector_len == AVX_512bit) {
 2087     evmovdquq(dst, src, AVX_512bit);
 2088   } else if (vector_len == AVX_256bit) {
 2089     vmovdqu(dst, src);
 2090   } else {
 2091     movdqu(dst, src);
 2092   }
 2093 }
 2094 
 2095 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
 2096   if (vector_len == AVX_512bit) {
 2097     evmovdquq(dst, src, AVX_512bit);
 2098   } else if (vector_len == AVX_256bit) {
 2099     vmovdqu(dst, src);
 2100   } else {
 2101     movdqu(dst, src);
 2102   }
 2103 }
 2104 
 2105 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
 2106   if (vector_len == AVX_512bit) {
 2107     evmovdquq(dst, src, AVX_512bit);
 2108   } else if (vector_len == AVX_256bit) {
 2109     vmovdqu(dst, src);
 2110   } else {
 2111     movdqu(dst, src);
 2112   }
 2113 }
 2114 
 2115 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2116   assert(rscratch != noreg || always_reachable(src), "missing");
 2117 
 2118   if (reachable(src)) {
 2119     vmovdqa(dst, as_Address(src));
 2120   }
 2121   else {
 2122     lea(rscratch, src);
 2123     vmovdqa(dst, Address(rscratch, 0));
 2124   }
 2125 }
 2126 
 2127 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2128   assert(rscratch != noreg || always_reachable(src), "missing");
 2129 
 2130   if (vector_len == AVX_512bit) {
 2131     evmovdqaq(dst, src, AVX_512bit, rscratch);
 2132   } else if (vector_len == AVX_256bit) {
 2133     vmovdqa(dst, src, rscratch);
 2134   } else {
 2135     movdqa(dst, src, rscratch);
 2136   }
 2137 }
 2138 
 2139 void MacroAssembler::kmov(KRegister dst, Address src) {
 2140   if (VM_Version::supports_avx512bw()) {
 2141     kmovql(dst, src);
 2142   } else {
 2143     assert(VM_Version::supports_evex(), "");
 2144     kmovwl(dst, src);
 2145   }
 2146 }
 2147 
 2148 void MacroAssembler::kmov(Address dst, KRegister src) {
 2149   if (VM_Version::supports_avx512bw()) {
 2150     kmovql(dst, src);
 2151   } else {
 2152     assert(VM_Version::supports_evex(), "");
 2153     kmovwl(dst, src);
 2154   }
 2155 }
 2156 
 2157 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2158   if (VM_Version::supports_avx512bw()) {
 2159     kmovql(dst, src);
 2160   } else {
 2161     assert(VM_Version::supports_evex(), "");
 2162     kmovwl(dst, src);
 2163   }
 2164 }
 2165 
 2166 void MacroAssembler::kmov(Register dst, KRegister src) {
 2167   if (VM_Version::supports_avx512bw()) {
 2168     kmovql(dst, src);
 2169   } else {
 2170     assert(VM_Version::supports_evex(), "");
 2171     kmovwl(dst, src);
 2172   }
 2173 }
 2174 
 2175 void MacroAssembler::kmov(KRegister dst, Register src) {
 2176   if (VM_Version::supports_avx512bw()) {
 2177     kmovql(dst, src);
 2178   } else {
 2179     assert(VM_Version::supports_evex(), "");
 2180     kmovwl(dst, src);
 2181   }
 2182 }
 2183 
 2184 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2185   assert(rscratch != noreg || always_reachable(src), "missing");
 2186 
 2187   if (reachable(src)) {
 2188     kmovql(dst, as_Address(src));
 2189   } else {
 2190     lea(rscratch, src);
 2191     kmovql(dst, Address(rscratch, 0));
 2192   }
 2193 }
 2194 
 2195 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2196   assert(rscratch != noreg || always_reachable(src), "missing");
 2197 
 2198   if (reachable(src)) {
 2199     kmovwl(dst, as_Address(src));
 2200   } else {
 2201     lea(rscratch, src);
 2202     kmovwl(dst, Address(rscratch, 0));
 2203   }
 2204 }
 2205 
 2206 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2207                                int vector_len, Register rscratch) {
 2208   assert(rscratch != noreg || always_reachable(src), "missing");
 2209 
 2210   if (reachable(src)) {
 2211     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2212   } else {
 2213     lea(rscratch, src);
 2214     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2215   }
 2216 }
 2217 
 2218 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2219                                int vector_len, Register rscratch) {
 2220   assert(rscratch != noreg || always_reachable(src), "missing");
 2221 
 2222   if (reachable(src)) {
 2223     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2224   } else {
 2225     lea(rscratch, src);
 2226     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2227   }
 2228 }
 2229 
 2230 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2231   assert(rscratch != noreg || always_reachable(src), "missing");
 2232 
 2233   if (reachable(src)) {
 2234     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2235   } else {
 2236     lea(rscratch, src);
 2237     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2238   }
 2239 }
 2240 
 2241 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2242   assert(rscratch != noreg || always_reachable(src), "missing");
 2243 
 2244   if (reachable(src)) {
 2245     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2246   } else {
 2247     lea(rscratch, src);
 2248     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2249   }
 2250 }
 2251 
 2252 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2253   assert(rscratch != noreg || always_reachable(src), "missing");
 2254 
 2255   if (reachable(src)) {
 2256     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2257   } else {
 2258     lea(rscratch, src);
 2259     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2260   }
 2261 }
 2262 
 2263 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2264   assert(rscratch != noreg || always_reachable(src), "missing");
 2265 
 2266   if (reachable(src)) {
 2267     Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
 2268   } else {
 2269     lea(rscratch, src);
 2270     Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2271   }
 2272 }
 2273 
 2274 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2275   assert(rscratch != noreg || always_reachable(src), "missing");
 2276 
 2277   if (reachable(src)) {
 2278     Assembler::evmovdqaq(dst, as_Address(src), vector_len);
 2279   } else {
 2280     lea(rscratch, src);
 2281     Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
 2282   }
 2283 }
 2284 
 2285 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2286   assert(rscratch != noreg || always_reachable(src), "missing");
 2287 
 2288   if (reachable(src)) {
 2289     Assembler::movapd(dst, as_Address(src));
 2290   } else {
 2291     lea(rscratch, src);
 2292     Assembler::movapd(dst, Address(rscratch, 0));
 2293   }
 2294 }
 2295 
 2296 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2297   assert(rscratch != noreg || always_reachable(src), "missing");
 2298 
 2299   if (reachable(src)) {
 2300     Assembler::movdqa(dst, as_Address(src));
 2301   } else {
 2302     lea(rscratch, src);
 2303     Assembler::movdqa(dst, Address(rscratch, 0));
 2304   }
 2305 }
 2306 
 2307 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2308   assert(rscratch != noreg || always_reachable(src), "missing");
 2309 
 2310   if (reachable(src)) {
 2311     Assembler::movsd(dst, as_Address(src));
 2312   } else {
 2313     lea(rscratch, src);
 2314     Assembler::movsd(dst, Address(rscratch, 0));
 2315   }
 2316 }
 2317 
 2318 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2319   assert(rscratch != noreg || always_reachable(src), "missing");
 2320 
 2321   if (reachable(src)) {
 2322     Assembler::movss(dst, as_Address(src));
 2323   } else {
 2324     lea(rscratch, src);
 2325     Assembler::movss(dst, Address(rscratch, 0));
 2326   }
 2327 }
 2328 
 2329 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2330   assert(rscratch != noreg || always_reachable(src), "missing");
 2331 
 2332   if (reachable(src)) {
 2333     Assembler::movddup(dst, as_Address(src));
 2334   } else {
 2335     lea(rscratch, src);
 2336     Assembler::movddup(dst, Address(rscratch, 0));
 2337   }
 2338 }
 2339 
 2340 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2341   assert(rscratch != noreg || always_reachable(src), "missing");
 2342 
 2343   if (reachable(src)) {
 2344     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2345   } else {
 2346     lea(rscratch, src);
 2347     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2348   }
 2349 }
 2350 
 2351 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2352   assert(rscratch != noreg || always_reachable(src), "missing");
 2353 
 2354   if (reachable(src)) {
 2355     Assembler::mulsd(dst, as_Address(src));
 2356   } else {
 2357     lea(rscratch, src);
 2358     Assembler::mulsd(dst, Address(rscratch, 0));
 2359   }
 2360 }
 2361 
 2362 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2363   assert(rscratch != noreg || always_reachable(src), "missing");
 2364 
 2365   if (reachable(src)) {
 2366     Assembler::mulss(dst, as_Address(src));
 2367   } else {
 2368     lea(rscratch, src);
 2369     Assembler::mulss(dst, Address(rscratch, 0));
 2370   }
 2371 }
 2372 
 2373 void MacroAssembler::null_check(Register reg, int offset) {
 2374   if (needs_explicit_null_check(offset)) {
 2375     // provoke OS null exception if reg is null by
 2376     // accessing M[reg] w/o changing any (non-CC) registers
 2377     // NOTE: cmpl is plenty here to provoke a segv
 2378     cmpptr(rax, Address(reg, 0));
 2379     // Note: should probably use testl(rax, Address(reg, 0));
 2380     //       may be shorter code (however, this version of
 2381     //       testl needs to be implemented first)
 2382   } else {
 2383     // nothing to do, (later) access of M[reg + offset]
 2384     // will provoke OS null exception if reg is null
 2385   }
 2386 }
 2387 
 2388 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
 2389   andptr(markword, markWord::inline_type_pattern_mask);
 2390   cmpptr(markword, markWord::inline_type_pattern);
 2391   jcc(Assembler::equal, is_inline_type);
 2392 }
 2393 
 2394 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
 2395   if (can_be_null) {
 2396     testptr(object, object);
 2397     jcc(Assembler::zero, not_inline_type);
 2398   }
 2399   const int is_inline_type_mask = markWord::inline_type_pattern;
 2400   movptr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
 2401   andptr(tmp, is_inline_type_mask);
 2402   cmpptr(tmp, is_inline_type_mask);
 2403   jcc(Assembler::notEqual, not_inline_type);
 2404 }
 2405 
 2406 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
 2407   movl(temp_reg, flags);
 2408   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2409   jcc(Assembler::notEqual, is_null_free_inline_type);
 2410 }
 2411 
 2412 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
 2413   movl(temp_reg, flags);
 2414   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2415   jcc(Assembler::equal, not_null_free_inline_type);
 2416 }
 2417 
 2418 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
 2419   movl(temp_reg, flags);
 2420   testl(temp_reg, 1 << ResolvedFieldEntry::is_flat_shift);
 2421   jcc(Assembler::notEqual, is_flat);
 2422 }
 2423 
 2424 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
 2425   Label test_mark_word;
 2426   // load mark word
 2427   movptr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
 2428   // check displaced
 2429   testl(temp_reg, markWord::unlocked_value);
 2430   jccb(Assembler::notZero, test_mark_word);
 2431   // slow path use klass prototype
 2432   push(rscratch1);
 2433   load_prototype_header(temp_reg, oop, rscratch1);
 2434   pop(rscratch1);
 2435 
 2436   bind(test_mark_word);
 2437   testl(temp_reg, test_bit);
 2438   jcc((jmp_set) ? Assembler::notZero : Assembler::zero, jmp_label);
 2439 }
 2440 
 2441 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg,
 2442                                          Label& is_flat_array) {
 2443   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
 2444 }
 2445 
 2446 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
 2447                                              Label& is_non_flat_array) {
 2448   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
 2449 }
 2450 
 2451 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label&is_null_free_array) {
 2452   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
 2453 }
 2454 
 2455 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
 2456   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
 2457 }
 2458 
 2459 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
 2460   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2461   jcc(Assembler::notZero, is_flat_array);
 2462 }
 2463 
 2464 void MacroAssembler::os_breakpoint() {
 2465   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2466   // (e.g., MSVC can't call ps() otherwise)
 2467   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2468 }
 2469 
 2470 void MacroAssembler::unimplemented(const char* what) {
 2471   const char* buf = nullptr;
 2472   {
 2473     ResourceMark rm;
 2474     stringStream ss;
 2475     ss.print("unimplemented: %s", what);
 2476     buf = code_string(ss.as_string());
 2477   }
 2478   stop(buf);
 2479 }
 2480 
 2481 #define XSTATE_BV 0x200
 2482 
 2483 void MacroAssembler::pop_CPU_state() {
 2484   pop_FPU_state();
 2485   pop_IU_state();
 2486 }
 2487 
 2488 void MacroAssembler::pop_FPU_state() {
 2489   fxrstor(Address(rsp, 0));
 2490   addptr(rsp, FPUStateSizeInWords * wordSize);
 2491 }
 2492 
 2493 void MacroAssembler::pop_IU_state() {
 2494   popa();
 2495   addq(rsp, 8);
 2496   popf();
 2497 }
 2498 
 2499 // Save Integer and Float state
 2500 // Warning: Stack must be 16 byte aligned (64bit)
 2501 void MacroAssembler::push_CPU_state() {
 2502   push_IU_state();
 2503   push_FPU_state();
 2504 }
 2505 
 2506 void MacroAssembler::push_FPU_state() {
 2507   subptr(rsp, FPUStateSizeInWords * wordSize);
 2508   fxsave(Address(rsp, 0));
 2509 }
 2510 
 2511 void MacroAssembler::push_IU_state() {
 2512   // Push flags first because pusha kills them
 2513   pushf();
 2514   // Make sure rsp stays 16-byte aligned
 2515   subq(rsp, 8);
 2516   pusha();
 2517 }
 2518 
 2519 void MacroAssembler::push_cont_fastpath() {
 2520   if (!Continuations::enabled()) return;
 2521 
 2522   Label L_done;
 2523   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2524   jccb(Assembler::belowEqual, L_done);
 2525   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2526   bind(L_done);
 2527 }
 2528 
 2529 void MacroAssembler::pop_cont_fastpath() {
 2530   if (!Continuations::enabled()) return;
 2531 
 2532   Label L_done;
 2533   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2534   jccb(Assembler::below, L_done);
 2535   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2536   bind(L_done);
 2537 }
 2538 
 2539 #ifdef ASSERT
 2540 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2541   Label no_cont;
 2542   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2543   testl(cont, cont);
 2544   jcc(Assembler::zero, no_cont);
 2545   stop(name);
 2546   bind(no_cont);
 2547 }
 2548 #endif
 2549 
 2550 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2551   // we must set sp to zero to clear frame
 2552   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2553   // must clear fp, so that compiled frames are not confused; it is
 2554   // possible that we need it only for debugging
 2555   if (clear_fp) {
 2556     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2557   }
 2558   // Always clear the pc because it could have been set by make_walkable()
 2559   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2560   vzeroupper();
 2561 }
 2562 
 2563 void MacroAssembler::round_to(Register reg, int modulus) {
 2564   addptr(reg, modulus - 1);
 2565   andptr(reg, -modulus);
 2566 }
 2567 
 2568 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2569   if (at_return) {
 2570     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2571     // we may safely use rsp instead to perform the stack watermark check.
 2572     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2573     jcc(Assembler::above, slow_path);
 2574     return;
 2575   }
 2576   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2577   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2578 }
 2579 
 2580 // Calls to C land
 2581 //
 2582 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2583 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2584 // has to be reset to 0. This is required to allow proper stack traversal.
 2585 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2586                                          Register last_java_fp,
 2587                                          address  last_java_pc,
 2588                                          Register rscratch) {
 2589   vzeroupper();
 2590   // determine last_java_sp register
 2591   if (!last_java_sp->is_valid()) {
 2592     last_java_sp = rsp;
 2593   }
 2594   // last_java_fp is optional
 2595   if (last_java_fp->is_valid()) {
 2596     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2597   }
 2598   // last_java_pc is optional
 2599   if (last_java_pc != nullptr) {
 2600     Address java_pc(r15_thread,
 2601                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2602     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2603   }
 2604   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2605 }
 2606 
 2607 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2608                                          Register last_java_fp,
 2609                                          Label &L,
 2610                                          Register scratch) {
 2611   lea(scratch, L);
 2612   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2613   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2614 }
 2615 
 2616 void MacroAssembler::shlptr(Register dst, int imm8) {
 2617   shlq(dst, imm8);
 2618 }
 2619 
 2620 void MacroAssembler::shrptr(Register dst, int imm8) {
 2621   shrq(dst, imm8);
 2622 }
 2623 
 2624 void MacroAssembler::sign_extend_byte(Register reg) {
 2625   movsbl(reg, reg); // movsxb
 2626 }
 2627 
 2628 void MacroAssembler::sign_extend_short(Register reg) {
 2629   movswl(reg, reg); // movsxw
 2630 }
 2631 
 2632 void MacroAssembler::narrow_subword_type(Register reg, BasicType bt) {
 2633   assert(is_subword_type(bt), "required");
 2634   switch (bt) {
 2635   case T_BOOLEAN: andl(reg, 1); break;
 2636   case T_BYTE:    movsbl(reg, reg); break;
 2637   case T_CHAR:    movzwl(reg, reg); break;
 2638   case T_SHORT:   movswl(reg, reg); break;
 2639   default:        ShouldNotReachHere();
 2640   }
 2641 }
 2642 
 2643 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2644   if (imm32 >= 0 && is8bit(imm32)) {
 2645     testb(dst, imm32);
 2646   } else {
 2647     Assembler::testl(dst, imm32);
 2648   }
 2649 }
 2650 
 2651 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2652   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2653     testb(dst, imm32);
 2654   } else {
 2655     Assembler::testl(dst, imm32);
 2656   }
 2657 }
 2658 
 2659 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2660   assert(always_reachable(src), "Address should be reachable");
 2661   testl(dst, as_Address(src));
 2662 }
 2663 
 2664 void MacroAssembler::testq(Address dst, int32_t imm32) {
 2665   if (imm32 >= 0) {
 2666     testl(dst, imm32);
 2667   } else {
 2668     Assembler::testq(dst, imm32);
 2669   }
 2670 }
 2671 
 2672 void MacroAssembler::testq(Register dst, int32_t imm32) {
 2673   if (imm32 >= 0) {
 2674     testl(dst, imm32);
 2675   } else {
 2676     Assembler::testq(dst, imm32);
 2677   }
 2678 }
 2679 
 2680 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2681   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2682   Assembler::pcmpeqb(dst, src);
 2683 }
 2684 
 2685 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2686   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2687   Assembler::pcmpeqw(dst, src);
 2688 }
 2689 
 2690 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2691   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2692   Assembler::pcmpestri(dst, src, imm8);
 2693 }
 2694 
 2695 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2696   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2697   Assembler::pcmpestri(dst, src, imm8);
 2698 }
 2699 
 2700 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2701   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2702   Assembler::pmovzxbw(dst, src);
 2703 }
 2704 
 2705 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2706   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2707   Assembler::pmovzxbw(dst, src);
 2708 }
 2709 
 2710 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2711   assert((src->encoding() < 16),"XMM register should be 0-15");
 2712   Assembler::pmovmskb(dst, src);
 2713 }
 2714 
 2715 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2716   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2717   Assembler::ptest(dst, src);
 2718 }
 2719 
 2720 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2721   assert(rscratch != noreg || always_reachable(src), "missing");
 2722 
 2723   if (reachable(src)) {
 2724     Assembler::sqrtss(dst, as_Address(src));
 2725   } else {
 2726     lea(rscratch, src);
 2727     Assembler::sqrtss(dst, Address(rscratch, 0));
 2728   }
 2729 }
 2730 
 2731 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2732   assert(rscratch != noreg || always_reachable(src), "missing");
 2733 
 2734   if (reachable(src)) {
 2735     Assembler::subsd(dst, as_Address(src));
 2736   } else {
 2737     lea(rscratch, src);
 2738     Assembler::subsd(dst, Address(rscratch, 0));
 2739   }
 2740 }
 2741 
 2742 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2743   assert(rscratch != noreg || always_reachable(src), "missing");
 2744 
 2745   if (reachable(src)) {
 2746     Assembler::roundsd(dst, as_Address(src), rmode);
 2747   } else {
 2748     lea(rscratch, src);
 2749     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2750   }
 2751 }
 2752 
 2753 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2754   assert(rscratch != noreg || always_reachable(src), "missing");
 2755 
 2756   if (reachable(src)) {
 2757     Assembler::subss(dst, as_Address(src));
 2758   } else {
 2759     lea(rscratch, src);
 2760     Assembler::subss(dst, Address(rscratch, 0));
 2761   }
 2762 }
 2763 
 2764 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2765   assert(rscratch != noreg || always_reachable(src), "missing");
 2766 
 2767   if (reachable(src)) {
 2768     Assembler::ucomisd(dst, as_Address(src));
 2769   } else {
 2770     lea(rscratch, src);
 2771     Assembler::ucomisd(dst, Address(rscratch, 0));
 2772   }
 2773 }
 2774 
 2775 void MacroAssembler::evucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2776   assert(rscratch != noreg || always_reachable(src), "missing");
 2777 
 2778   if (reachable(src)) {
 2779     Assembler::evucomxsd(dst, as_Address(src));
 2780   } else {
 2781     lea(rscratch, src);
 2782     Assembler::evucomxsd(dst, Address(rscratch, 0));
 2783   }
 2784 }
 2785 
 2786 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2787   assert(rscratch != noreg || always_reachable(src), "missing");
 2788 
 2789   if (reachable(src)) {
 2790     Assembler::ucomiss(dst, as_Address(src));
 2791   } else {
 2792     lea(rscratch, src);
 2793     Assembler::ucomiss(dst, Address(rscratch, 0));
 2794   }
 2795 }
 2796 
 2797 void MacroAssembler::evucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2798   assert(rscratch != noreg || always_reachable(src), "missing");
 2799 
 2800   if (reachable(src)) {
 2801     Assembler::evucomxss(dst, as_Address(src));
 2802   } else {
 2803     lea(rscratch, src);
 2804     Assembler::evucomxss(dst, Address(rscratch, 0));
 2805   }
 2806 }
 2807 
 2808 void MacroAssembler::evucomish(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2809   assert(rscratch != noreg || always_reachable(src), "missing");
 2810 
 2811   if (reachable(src)) {
 2812     Assembler::evucomish(dst, as_Address(src));
 2813   } else {
 2814     lea(rscratch, src);
 2815     Assembler::evucomish(dst, Address(rscratch, 0));
 2816   }
 2817 }
 2818 
 2819 void MacroAssembler::evucomxsh(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2820   assert(rscratch != noreg || always_reachable(src), "missing");
 2821 
 2822   if (reachable(src)) {
 2823     Assembler::evucomxsh(dst, as_Address(src));
 2824   } else {
 2825     lea(rscratch, src);
 2826     Assembler::evucomxsh(dst, Address(rscratch, 0));
 2827   }
 2828 }
 2829 
 2830 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2831   assert(rscratch != noreg || always_reachable(src), "missing");
 2832 
 2833   // Used in sign-bit flipping with aligned address.
 2834   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2835 
 2836   if (UseAVX > 2 &&
 2837       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2838       (dst->encoding() >= 16)) {
 2839     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2840   } else if (reachable(src)) {
 2841     Assembler::xorpd(dst, as_Address(src));
 2842   } else {
 2843     lea(rscratch, src);
 2844     Assembler::xorpd(dst, Address(rscratch, 0));
 2845   }
 2846 }
 2847 
 2848 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2849   if (UseAVX > 2 &&
 2850       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2851       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2852     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2853   } else {
 2854     Assembler::xorpd(dst, src);
 2855   }
 2856 }
 2857 
 2858 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2859   if (UseAVX > 2 &&
 2860       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2861       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2862     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2863   } else {
 2864     Assembler::xorps(dst, src);
 2865   }
 2866 }
 2867 
 2868 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2869   assert(rscratch != noreg || always_reachable(src), "missing");
 2870 
 2871   // Used in sign-bit flipping with aligned address.
 2872   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2873 
 2874   if (UseAVX > 2 &&
 2875       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2876       (dst->encoding() >= 16)) {
 2877     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2878   } else if (reachable(src)) {
 2879     Assembler::xorps(dst, as_Address(src));
 2880   } else {
 2881     lea(rscratch, src);
 2882     Assembler::xorps(dst, Address(rscratch, 0));
 2883   }
 2884 }
 2885 
 2886 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2887   assert(rscratch != noreg || always_reachable(src), "missing");
 2888 
 2889   // Used in sign-bit flipping with aligned address.
 2890   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2891   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2892   if (reachable(src)) {
 2893     Assembler::pshufb(dst, as_Address(src));
 2894   } else {
 2895     lea(rscratch, src);
 2896     Assembler::pshufb(dst, Address(rscratch, 0));
 2897   }
 2898 }
 2899 
 2900 // AVX 3-operands instructions
 2901 
 2902 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2903   assert(rscratch != noreg || always_reachable(src), "missing");
 2904 
 2905   if (reachable(src)) {
 2906     vaddsd(dst, nds, as_Address(src));
 2907   } else {
 2908     lea(rscratch, src);
 2909     vaddsd(dst, nds, Address(rscratch, 0));
 2910   }
 2911 }
 2912 
 2913 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2914   assert(rscratch != noreg || always_reachable(src), "missing");
 2915 
 2916   if (reachable(src)) {
 2917     vaddss(dst, nds, as_Address(src));
 2918   } else {
 2919     lea(rscratch, src);
 2920     vaddss(dst, nds, Address(rscratch, 0));
 2921   }
 2922 }
 2923 
 2924 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2925   assert(UseAVX > 0, "requires some form of AVX");
 2926   assert(rscratch != noreg || always_reachable(src), "missing");
 2927 
 2928   if (reachable(src)) {
 2929     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2930   } else {
 2931     lea(rscratch, src);
 2932     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2933   }
 2934 }
 2935 
 2936 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2937   assert(UseAVX > 0, "requires some form of AVX");
 2938   assert(rscratch != noreg || always_reachable(src), "missing");
 2939 
 2940   if (reachable(src)) {
 2941     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2942   } else {
 2943     lea(rscratch, src);
 2944     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2945   }
 2946 }
 2947 
 2948 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2949   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2950   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2951 
 2952   vandps(dst, nds, negate_field, vector_len, rscratch);
 2953 }
 2954 
 2955 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2956   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2957   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2958 
 2959   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2960 }
 2961 
 2962 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2963   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2964   Assembler::vpaddb(dst, nds, src, vector_len);
 2965 }
 2966 
 2967 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2968   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2969   Assembler::vpaddb(dst, nds, src, vector_len);
 2970 }
 2971 
 2972 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2973   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2974   Assembler::vpaddw(dst, nds, src, vector_len);
 2975 }
 2976 
 2977 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2978   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2979   Assembler::vpaddw(dst, nds, src, vector_len);
 2980 }
 2981 
 2982 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2983   assert(rscratch != noreg || always_reachable(src), "missing");
 2984 
 2985   if (reachable(src)) {
 2986     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 2987   } else {
 2988     lea(rscratch, src);
 2989     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 2990   }
 2991 }
 2992 
 2993 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2994   assert(rscratch != noreg || always_reachable(src), "missing");
 2995 
 2996   if (reachable(src)) {
 2997     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 2998   } else {
 2999     lea(rscratch, src);
 3000     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 3001   }
 3002 }
 3003 
 3004 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3005   assert(rscratch != noreg || always_reachable(src), "missing");
 3006 
 3007   if (reachable(src)) {
 3008     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 3009   } else {
 3010     lea(rscratch, src);
 3011     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 3012   }
 3013 }
 3014 
 3015 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3016   assert(rscratch != noreg || always_reachable(src), "missing");
 3017 
 3018   if (reachable(src)) {
 3019     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 3020   } else {
 3021     lea(rscratch, src);
 3022     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 3023   }
 3024 }
 3025 
 3026 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3027   assert(rscratch != noreg || always_reachable(src), "missing");
 3028 
 3029   if (reachable(src)) {
 3030     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 3031   } else {
 3032     lea(rscratch, src);
 3033     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 3034   }
 3035 }
 3036 
 3037 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3038   assert(rscratch != noreg || always_reachable(src), "missing");
 3039 
 3040   if (reachable(src)) {
 3041     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 3042   } else {
 3043     lea(rscratch, src);
 3044     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 3045   }
 3046 }
 3047 
 3048 // Vector float blend
 3049 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3050 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3051   // WARN: Allow dst == (src1|src2), mask == scratch
 3052   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3053                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3054   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 3055   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3056   if (blend_emulation && scratch_available && dst_available) {
 3057     if (compute_mask) {
 3058       vpsrad(scratch, mask, 32, vector_len);
 3059       mask = scratch;
 3060     }
 3061     if (dst == src1) {
 3062       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 3063       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3064     } else {
 3065       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3066       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 3067     }
 3068     vpor(dst, dst, scratch, vector_len);
 3069   } else {
 3070     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 3071   }
 3072 }
 3073 
 3074 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3075 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3076   // WARN: Allow dst == (src1|src2), mask == scratch
 3077   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3078                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3079   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 3080   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3081   if (blend_emulation && scratch_available && dst_available) {
 3082     if (compute_mask) {
 3083       vpxor(scratch, scratch, scratch, vector_len);
 3084       vpcmpgtq(scratch, scratch, mask, vector_len);
 3085       mask = scratch;
 3086     }
 3087     if (dst == src1) {
 3088       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 3089       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3090     } else {
 3091       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3092       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 3093     }
 3094     vpor(dst, dst, scratch, vector_len);
 3095   } else {
 3096     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 3097   }
 3098 }
 3099 
 3100 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3101   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3102   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 3103 }
 3104 
 3105 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 3106   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3107   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 3108 }
 3109 
 3110 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3111   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3112   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3113 }
 3114 
 3115 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3116   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3117   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3118 }
 3119 
 3120 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3121   assert(rscratch != noreg || always_reachable(src), "missing");
 3122 
 3123   if (reachable(src)) {
 3124     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 3125   } else {
 3126     lea(rscratch, src);
 3127     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 3128   }
 3129 }
 3130 
 3131 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3132                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3133   assert(rscratch != noreg || always_reachable(src), "missing");
 3134 
 3135   if (reachable(src)) {
 3136     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3137   } else {
 3138     lea(rscratch, src);
 3139     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3140   }
 3141 }
 3142 
 3143 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3144                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3145   assert(rscratch != noreg || always_reachable(src), "missing");
 3146 
 3147   if (reachable(src)) {
 3148     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3149   } else {
 3150     lea(rscratch, src);
 3151     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3152   }
 3153 }
 3154 
 3155 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3156                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3157   assert(rscratch != noreg || always_reachable(src), "missing");
 3158 
 3159   if (reachable(src)) {
 3160     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3161   } else {
 3162     lea(rscratch, src);
 3163     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3164   }
 3165 }
 3166 
 3167 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3168                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3169   assert(rscratch != noreg || always_reachable(src), "missing");
 3170 
 3171   if (reachable(src)) {
 3172     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3173   } else {
 3174     lea(rscratch, src);
 3175     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3176   }
 3177 }
 3178 
 3179 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3180   if (width == Assembler::Q) {
 3181     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3182   } else {
 3183     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3184   }
 3185 }
 3186 
 3187 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3188   int eq_cond_enc = 0x29;
 3189   int gt_cond_enc = 0x37;
 3190   if (width != Assembler::Q) {
 3191     eq_cond_enc = 0x74 + width;
 3192     gt_cond_enc = 0x64 + width;
 3193   }
 3194   switch (cond) {
 3195   case eq:
 3196     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3197     break;
 3198   case neq:
 3199     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3200     vallones(xtmp, vector_len);
 3201     vpxor(dst, xtmp, dst, vector_len);
 3202     break;
 3203   case le:
 3204     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3205     vallones(xtmp, vector_len);
 3206     vpxor(dst, xtmp, dst, vector_len);
 3207     break;
 3208   case nlt:
 3209     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3210     vallones(xtmp, vector_len);
 3211     vpxor(dst, xtmp, dst, vector_len);
 3212     break;
 3213   case lt:
 3214     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3215     break;
 3216   case nle:
 3217     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3218     break;
 3219   default:
 3220     assert(false, "Should not reach here");
 3221   }
 3222 }
 3223 
 3224 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3225   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3226   Assembler::vpmovzxbw(dst, src, vector_len);
 3227 }
 3228 
 3229 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3230   assert((src->encoding() < 16),"XMM register should be 0-15");
 3231   Assembler::vpmovmskb(dst, src, vector_len);
 3232 }
 3233 
 3234 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3235   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3236   Assembler::vpmullw(dst, nds, src, vector_len);
 3237 }
 3238 
 3239 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3240   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3241   Assembler::vpmullw(dst, nds, src, vector_len);
 3242 }
 3243 
 3244 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3245   assert((UseAVX > 0), "AVX support is needed");
 3246   assert(rscratch != noreg || always_reachable(src), "missing");
 3247 
 3248   if (reachable(src)) {
 3249     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3250   } else {
 3251     lea(rscratch, src);
 3252     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3253   }
 3254 }
 3255 
 3256 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3257   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3258   Assembler::vpsubb(dst, nds, src, vector_len);
 3259 }
 3260 
 3261 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3262   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3263   Assembler::vpsubb(dst, nds, src, vector_len);
 3264 }
 3265 
 3266 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3267   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3268   Assembler::vpsubw(dst, nds, src, vector_len);
 3269 }
 3270 
 3271 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3272   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3273   Assembler::vpsubw(dst, nds, src, vector_len);
 3274 }
 3275 
 3276 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3277   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3278   Assembler::vpsraw(dst, nds, shift, vector_len);
 3279 }
 3280 
 3281 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3282   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3283   Assembler::vpsraw(dst, nds, shift, vector_len);
 3284 }
 3285 
 3286 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3287   assert(UseAVX > 2,"");
 3288   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3289      vector_len = 2;
 3290   }
 3291   Assembler::evpsraq(dst, nds, shift, vector_len);
 3292 }
 3293 
 3294 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3295   assert(UseAVX > 2,"");
 3296   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3297      vector_len = 2;
 3298   }
 3299   Assembler::evpsraq(dst, nds, shift, vector_len);
 3300 }
 3301 
 3302 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3303   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3304   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3305 }
 3306 
 3307 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3308   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3309   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3310 }
 3311 
 3312 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3313   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3314   Assembler::vpsllw(dst, nds, shift, vector_len);
 3315 }
 3316 
 3317 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3318   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3319   Assembler::vpsllw(dst, nds, shift, vector_len);
 3320 }
 3321 
 3322 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3323   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3324   Assembler::vptest(dst, src);
 3325 }
 3326 
 3327 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3328   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3329   Assembler::punpcklbw(dst, src);
 3330 }
 3331 
 3332 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3333   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3334   Assembler::pshufd(dst, src, mode);
 3335 }
 3336 
 3337 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3338   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3339   Assembler::pshuflw(dst, src, mode);
 3340 }
 3341 
 3342 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3343   assert(rscratch != noreg || always_reachable(src), "missing");
 3344 
 3345   if (reachable(src)) {
 3346     vandpd(dst, nds, as_Address(src), vector_len);
 3347   } else {
 3348     lea(rscratch, src);
 3349     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3350   }
 3351 }
 3352 
 3353 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3354   assert(rscratch != noreg || always_reachable(src), "missing");
 3355 
 3356   if (reachable(src)) {
 3357     vandps(dst, nds, as_Address(src), vector_len);
 3358   } else {
 3359     lea(rscratch, src);
 3360     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3361   }
 3362 }
 3363 
 3364 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3365                             bool merge, int vector_len, Register rscratch) {
 3366   assert(rscratch != noreg || always_reachable(src), "missing");
 3367 
 3368   if (reachable(src)) {
 3369     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3370   } else {
 3371     lea(rscratch, src);
 3372     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3373   }
 3374 }
 3375 
 3376 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3377   assert(rscratch != noreg || always_reachable(src), "missing");
 3378 
 3379   if (reachable(src)) {
 3380     vdivsd(dst, nds, as_Address(src));
 3381   } else {
 3382     lea(rscratch, src);
 3383     vdivsd(dst, nds, Address(rscratch, 0));
 3384   }
 3385 }
 3386 
 3387 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3388   assert(rscratch != noreg || always_reachable(src), "missing");
 3389 
 3390   if (reachable(src)) {
 3391     vdivss(dst, nds, as_Address(src));
 3392   } else {
 3393     lea(rscratch, src);
 3394     vdivss(dst, nds, Address(rscratch, 0));
 3395   }
 3396 }
 3397 
 3398 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3399   assert(rscratch != noreg || always_reachable(src), "missing");
 3400 
 3401   if (reachable(src)) {
 3402     vmulsd(dst, nds, as_Address(src));
 3403   } else {
 3404     lea(rscratch, src);
 3405     vmulsd(dst, nds, Address(rscratch, 0));
 3406   }
 3407 }
 3408 
 3409 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3410   assert(rscratch != noreg || always_reachable(src), "missing");
 3411 
 3412   if (reachable(src)) {
 3413     vmulss(dst, nds, as_Address(src));
 3414   } else {
 3415     lea(rscratch, src);
 3416     vmulss(dst, nds, Address(rscratch, 0));
 3417   }
 3418 }
 3419 
 3420 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3421   assert(rscratch != noreg || always_reachable(src), "missing");
 3422 
 3423   if (reachable(src)) {
 3424     vsubsd(dst, nds, as_Address(src));
 3425   } else {
 3426     lea(rscratch, src);
 3427     vsubsd(dst, nds, Address(rscratch, 0));
 3428   }
 3429 }
 3430 
 3431 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3432   assert(rscratch != noreg || always_reachable(src), "missing");
 3433 
 3434   if (reachable(src)) {
 3435     vsubss(dst, nds, as_Address(src));
 3436   } else {
 3437     lea(rscratch, src);
 3438     vsubss(dst, nds, Address(rscratch, 0));
 3439   }
 3440 }
 3441 
 3442 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3443   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3444   assert(rscratch != noreg || always_reachable(src), "missing");
 3445 
 3446   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3447 }
 3448 
 3449 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3450   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3451   assert(rscratch != noreg || always_reachable(src), "missing");
 3452 
 3453   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3454 }
 3455 
 3456 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3457   assert(rscratch != noreg || always_reachable(src), "missing");
 3458 
 3459   if (reachable(src)) {
 3460     vxorpd(dst, nds, as_Address(src), vector_len);
 3461   } else {
 3462     lea(rscratch, src);
 3463     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3464   }
 3465 }
 3466 
 3467 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3468   assert(rscratch != noreg || always_reachable(src), "missing");
 3469 
 3470   if (reachable(src)) {
 3471     vxorps(dst, nds, as_Address(src), vector_len);
 3472   } else {
 3473     lea(rscratch, src);
 3474     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3475   }
 3476 }
 3477 
 3478 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3479   assert(rscratch != noreg || always_reachable(src), "missing");
 3480 
 3481   if (UseAVX > 1 || (vector_len < 1)) {
 3482     if (reachable(src)) {
 3483       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3484     } else {
 3485       lea(rscratch, src);
 3486       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3487     }
 3488   } else {
 3489     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3490   }
 3491 }
 3492 
 3493 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3494   assert(rscratch != noreg || always_reachable(src), "missing");
 3495 
 3496   if (reachable(src)) {
 3497     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3498   } else {
 3499     lea(rscratch, src);
 3500     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3501   }
 3502 }
 3503 
 3504 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3505   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3506   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3507   // The inverted mask is sign-extended
 3508   andptr(possibly_non_local, inverted_mask);
 3509 }
 3510 
 3511 void MacroAssembler::resolve_jobject(Register value,
 3512                                      Register tmp) {
 3513   Register thread = r15_thread;
 3514   assert_different_registers(value, thread, tmp);
 3515   Label done, tagged, weak_tagged;
 3516   testptr(value, value);
 3517   jcc(Assembler::zero, done);           // Use null as-is.
 3518   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3519   jcc(Assembler::notZero, tagged);
 3520 
 3521   // Resolve local handle
 3522   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3523   verify_oop(value);
 3524   jmp(done);
 3525 
 3526   bind(tagged);
 3527   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3528   jcc(Assembler::notZero, weak_tagged);
 3529 
 3530   // Resolve global handle
 3531   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3532   verify_oop(value);
 3533   jmp(done);
 3534 
 3535   bind(weak_tagged);
 3536   // Resolve jweak.
 3537   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3538                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3539   verify_oop(value);
 3540 
 3541   bind(done);
 3542 }
 3543 
 3544 void MacroAssembler::resolve_global_jobject(Register value,
 3545                                             Register tmp) {
 3546   Register thread = r15_thread;
 3547   assert_different_registers(value, thread, tmp);
 3548   Label done;
 3549 
 3550   testptr(value, value);
 3551   jcc(Assembler::zero, done);           // Use null as-is.
 3552 
 3553 #ifdef ASSERT
 3554   {
 3555     Label valid_global_tag;
 3556     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3557     jcc(Assembler::notZero, valid_global_tag);
 3558     stop("non global jobject using resolve_global_jobject");
 3559     bind(valid_global_tag);
 3560   }
 3561 #endif
 3562 
 3563   // Resolve global handle
 3564   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3565   verify_oop(value);
 3566 
 3567   bind(done);
 3568 }
 3569 
 3570 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3571   subq(dst, imm32);
 3572 }
 3573 
 3574 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3575 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3576   subq_imm32(dst, imm32);
 3577 }
 3578 
 3579 void MacroAssembler::subptr(Register dst, Register src) {
 3580   subq(dst, src);
 3581 }
 3582 
 3583 // C++ bool manipulation
 3584 void MacroAssembler::testbool(Register dst) {
 3585   if(sizeof(bool) == 1)
 3586     testb(dst, 0xff);
 3587   else if(sizeof(bool) == 2) {
 3588     // testw implementation needed for two byte bools
 3589     ShouldNotReachHere();
 3590   } else if(sizeof(bool) == 4)
 3591     testl(dst, dst);
 3592   else
 3593     // unsupported
 3594     ShouldNotReachHere();
 3595 }
 3596 
 3597 void MacroAssembler::testptr(Register dst, Register src) {
 3598   testq(dst, src);
 3599 }
 3600 
 3601 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3602 void MacroAssembler::tlab_allocate(Register obj,
 3603                                    Register var_size_in_bytes,
 3604                                    int con_size_in_bytes,
 3605                                    Register t1,
 3606                                    Register t2,
 3607                                    Label& slow_case) {
 3608   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3609   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3610 }
 3611 
 3612 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3613   RegSet regs;
 3614   regs += RegSet::of(rax, rcx, rdx);
 3615 #ifndef _WINDOWS
 3616   regs += RegSet::of(rsi, rdi);
 3617 #endif
 3618   regs += RegSet::range(r8, r11);
 3619   if (UseAPX) {
 3620     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3621   }
 3622   return regs;
 3623 }
 3624 
 3625 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3626   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3627 #if defined(_WINDOWS)
 3628   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3629   if (num_xmm_registers > 16) {
 3630      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3631   }
 3632   return result;
 3633 #else
 3634   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3635 #endif
 3636 }
 3637 
 3638 // C1 only ever uses the first double/float of the XMM register.
 3639 static int xmm_save_size() { return sizeof(double); }
 3640 
 3641 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3642   masm->movdbl(Address(rsp, offset), reg);
 3643 }
 3644 
 3645 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3646   masm->movdbl(reg, Address(rsp, offset));
 3647 }
 3648 
 3649 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3650                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3651 
 3652   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3653                          StackAlignmentInBytes);
 3654   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3655 
 3656   return gp_area_size + xmm_area_size;
 3657 }
 3658 
 3659 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3660   block_comment("push_call_clobbered_registers start");
 3661   // Regular registers
 3662   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3663 
 3664   int gp_area_size;
 3665   int xmm_area_size;
 3666   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3667                                                gp_area_size, xmm_area_size);
 3668   subptr(rsp, total_save_size);
 3669 
 3670   push_set(gp_registers_to_push, 0);
 3671 
 3672   if (save_fpu) {
 3673     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3674   }
 3675 
 3676   block_comment("push_call_clobbered_registers end");
 3677 }
 3678 
 3679 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3680   block_comment("pop_call_clobbered_registers start");
 3681 
 3682   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 3683 
 3684   int gp_area_size;
 3685   int xmm_area_size;
 3686   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 3687                                                gp_area_size, xmm_area_size);
 3688 
 3689   if (restore_fpu) {
 3690     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3691   }
 3692 
 3693   pop_set(gp_registers_to_pop, 0);
 3694 
 3695   addptr(rsp, total_save_size);
 3696 
 3697   vzeroupper();
 3698 
 3699   block_comment("pop_call_clobbered_registers end");
 3700 }
 3701 
 3702 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3703   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3704   int spill_offset = offset;
 3705 
 3706   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3707     save_xmm_register(this, spill_offset, *it);
 3708     spill_offset += xmm_save_size();
 3709   }
 3710 }
 3711 
 3712 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3713   int restore_size = set.size() * xmm_save_size();
 3714   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3715 
 3716   int restore_offset = offset + restore_size - xmm_save_size();
 3717 
 3718   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3719     restore_xmm_register(this, restore_offset, *it);
 3720     restore_offset -= xmm_save_size();
 3721   }
 3722 }
 3723 
 3724 void MacroAssembler::push_set(RegSet set, int offset) {
 3725   int spill_offset;
 3726   if (offset == -1) {
 3727     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3728     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3729     subptr(rsp, aligned_size);
 3730     spill_offset = 0;
 3731   } else {
 3732     spill_offset = offset;
 3733   }
 3734 
 3735   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3736     movptr(Address(rsp, spill_offset), *it);
 3737     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3738   }
 3739 }
 3740 
 3741 void MacroAssembler::pop_set(RegSet set, int offset) {
 3742 
 3743   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3744   int restore_size = set.size() * gp_reg_size;
 3745   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3746 
 3747   int restore_offset;
 3748   if (offset == -1) {
 3749     restore_offset = restore_size - gp_reg_size;
 3750   } else {
 3751     restore_offset = offset + restore_size - gp_reg_size;
 3752   }
 3753   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3754     movptr(*it, Address(rsp, restore_offset));
 3755     restore_offset -= gp_reg_size;
 3756   }
 3757 
 3758   if (offset == -1) {
 3759     addptr(rsp, aligned_size);
 3760   }
 3761 }
 3762 
 3763 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3764 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3765   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3766   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3767   Label done;
 3768 
 3769   testptr(length_in_bytes, length_in_bytes);
 3770   jcc(Assembler::zero, done);
 3771 
 3772   // initialize topmost word, divide index by 2, check if odd and test if zero
 3773   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3774 #ifdef ASSERT
 3775   {
 3776     Label L;
 3777     testptr(length_in_bytes, BytesPerWord - 1);
 3778     jcc(Assembler::zero, L);
 3779     stop("length must be a multiple of BytesPerWord");
 3780     bind(L);
 3781   }
 3782 #endif
 3783   Register index = length_in_bytes;
 3784   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3785   if (UseIncDec) {
 3786     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3787   } else {
 3788     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3789     shrptr(index, 1);
 3790   }
 3791 
 3792   // initialize remaining object fields: index is a multiple of 2 now
 3793   {
 3794     Label loop;
 3795     bind(loop);
 3796     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3797     decrement(index);
 3798     jcc(Assembler::notZero, loop);
 3799   }
 3800 
 3801   bind(done);
 3802 }
 3803 
 3804 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
 3805   movptr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
 3806 #ifdef ASSERT
 3807   {
 3808     Label done;
 3809     cmpptr(layout_info, 0);
 3810     jcc(Assembler::notEqual, done);
 3811     stop("inline_layout_info_array is null");
 3812     bind(done);
 3813   }
 3814 #endif
 3815 
 3816   InlineLayoutInfo array[2];
 3817   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
 3818   if (is_power_of_2(size)) {
 3819     shll(index, log2i_exact(size)); // Scale index by power of 2
 3820   } else {
 3821     imull(index, index, size); // Scale the index to be the entry index * array_element_size
 3822   }
 3823   lea(layout_info, Address(layout_info, index, Address::times_1, Array<InlineLayoutInfo>::base_offset_in_bytes()));
 3824 }
 3825 
 3826 // Look up the method for a megamorphic invokeinterface call.
 3827 // The target method is determined by <intf_klass, itable_index>.
 3828 // The receiver klass is in recv_klass.
 3829 // On success, the result will be in method_result, and execution falls through.
 3830 // On failure, execution transfers to the given label.
 3831 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3832                                              Register intf_klass,
 3833                                              RegisterOrConstant itable_index,
 3834                                              Register method_result,
 3835                                              Register scan_temp,
 3836                                              Label& L_no_such_interface,
 3837                                              bool return_method) {
 3838   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3839   assert_different_registers(method_result, intf_klass, scan_temp);
 3840   assert(recv_klass != method_result || !return_method,
 3841          "recv_klass can be destroyed when method isn't needed");
 3842 
 3843   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3844          "caller must use same register for non-constant itable index as for method");
 3845 
 3846   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3847   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3848   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3849   int scan_step   = itableOffsetEntry::size() * wordSize;
 3850   int vte_size    = vtableEntry::size_in_bytes();
 3851   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3852   assert(vte_size == wordSize, "else adjust times_vte_scale");
 3853 
 3854   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3855 
 3856   // Could store the aligned, prescaled offset in the klass.
 3857   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 3858 
 3859   if (return_method) {
 3860     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 3861     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3862     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 3863   }
 3864 
 3865   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 3866   //   if (scan->interface() == intf) {
 3867   //     result = (klass + scan->offset() + itable_index);
 3868   //   }
 3869   // }
 3870   Label search, found_method;
 3871 
 3872   for (int peel = 1; peel >= 0; peel--) {
 3873     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 3874     cmpptr(intf_klass, method_result);
 3875 
 3876     if (peel) {
 3877       jccb(Assembler::equal, found_method);
 3878     } else {
 3879       jccb(Assembler::notEqual, search);
 3880       // (invert the test to fall through to found_method...)
 3881     }
 3882 
 3883     if (!peel)  break;
 3884 
 3885     bind(search);
 3886 
 3887     // Check that the previous entry is non-null.  A null entry means that
 3888     // the receiver class doesn't implement the interface, and wasn't the
 3889     // same as when the caller was compiled.
 3890     testptr(method_result, method_result);
 3891     jcc(Assembler::zero, L_no_such_interface);
 3892     addptr(scan_temp, scan_step);
 3893   }
 3894 
 3895   bind(found_method);
 3896 
 3897   if (return_method) {
 3898     // Got a hit.
 3899     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 3900     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 3901   }
 3902 }
 3903 
 3904 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 3905 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 3906 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 3907 // The target method is determined by <holder_klass, itable_index>.
 3908 // The receiver klass is in recv_klass.
 3909 // On success, the result will be in method_result, and execution falls through.
 3910 // On failure, execution transfers to the given label.
 3911 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 3912                                                   Register holder_klass,
 3913                                                   Register resolved_klass,
 3914                                                   Register method_result,
 3915                                                   Register scan_temp,
 3916                                                   Register temp_reg2,
 3917                                                   Register receiver,
 3918                                                   int itable_index,
 3919                                                   Label& L_no_such_interface) {
 3920   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 3921   Register temp_itbl_klass = method_result;
 3922   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 3923 
 3924   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3925   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3926   int scan_step = itableOffsetEntry::size() * wordSize;
 3927   int vte_size = vtableEntry::size_in_bytes();
 3928   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 3929   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 3930   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3931   assert(vte_size == wordSize, "adjust times_vte_scale");
 3932 
 3933   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 3934 
 3935   // temp_itbl_klass = recv_klass.itable[0]
 3936   // scan_temp = &recv_klass.itable[0] + step
 3937   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3938   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 3939   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 3940   xorptr(temp_reg, temp_reg);
 3941 
 3942   // Initial checks:
 3943   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 3944   //   - if (itable[0] == 0), no such interface
 3945   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 3946   cmpptr(holder_klass, resolved_klass);
 3947   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 3948   testptr(temp_itbl_klass, temp_itbl_klass);
 3949   jccb(Assembler::zero, L_no_such_interface);
 3950   cmpptr(holder_klass, temp_itbl_klass);
 3951   jccb(Assembler::equal, L_holder_found);
 3952 
 3953   // Loop: Look for holder_klass record in itable
 3954   //   do {
 3955   //     tmp = itable[index];
 3956   //     index += step;
 3957   //     if (tmp == holder_klass) {
 3958   //       goto L_holder_found; // Found!
 3959   //     }
 3960   //   } while (tmp != 0);
 3961   //   goto L_no_such_interface // Not found.
 3962   Label L_scan_holder;
 3963   bind(L_scan_holder);
 3964     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3965     addptr(scan_temp, scan_step);
 3966     cmpptr(holder_klass, temp_itbl_klass);
 3967     jccb(Assembler::equal, L_holder_found);
 3968     testptr(temp_itbl_klass, temp_itbl_klass);
 3969     jccb(Assembler::notZero, L_scan_holder);
 3970 
 3971   jmpb(L_no_such_interface);
 3972 
 3973   // Loop: Look for resolved_class record in itable
 3974   //   do {
 3975   //     tmp = itable[index];
 3976   //     index += step;
 3977   //     if (tmp == holder_klass) {
 3978   //        // Also check if we have met a holder klass
 3979   //        holder_tmp = itable[index-step-ioffset];
 3980   //     }
 3981   //     if (tmp == resolved_klass) {
 3982   //        goto L_resolved_found;  // Found!
 3983   //     }
 3984   //   } while (tmp != 0);
 3985   //   goto L_no_such_interface // Not found.
 3986   //
 3987   Label L_loop_scan_resolved;
 3988   bind(L_loop_scan_resolved);
 3989     movptr(temp_itbl_klass, Address(scan_temp, 0));
 3990     addptr(scan_temp, scan_step);
 3991     bind(L_loop_scan_resolved_entry);
 3992     cmpptr(holder_klass, temp_itbl_klass);
 3993     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 3994     cmpptr(resolved_klass, temp_itbl_klass);
 3995     jccb(Assembler::equal, L_resolved_found);
 3996     testptr(temp_itbl_klass, temp_itbl_klass);
 3997     jccb(Assembler::notZero, L_loop_scan_resolved);
 3998 
 3999   jmpb(L_no_such_interface);
 4000 
 4001   Label L_ready;
 4002 
 4003   // See if we already have a holder klass. If not, go and scan for it.
 4004   bind(L_resolved_found);
 4005   testptr(temp_reg, temp_reg);
 4006   jccb(Assembler::zero, L_scan_holder);
 4007   jmpb(L_ready);
 4008 
 4009   bind(L_holder_found);
 4010   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4011 
 4012   // Finally, temp_reg contains holder_klass vtable offset
 4013   bind(L_ready);
 4014   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 4015   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 4016     load_klass(scan_temp, receiver, noreg);
 4017     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4018   } else {
 4019     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4020   }
 4021 }
 4022 
 4023 
 4024 // virtual method calling
 4025 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 4026                                            RegisterOrConstant vtable_index,
 4027                                            Register method_result) {
 4028   const ByteSize base = Klass::vtable_start_offset();
 4029   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 4030   Address vtable_entry_addr(recv_klass,
 4031                             vtable_index, Address::times_ptr,
 4032                             base + vtableEntry::method_offset());
 4033   movptr(method_result, vtable_entry_addr);
 4034 }
 4035 
 4036 
 4037 void MacroAssembler::check_klass_subtype(Register sub_klass,
 4038                            Register super_klass,
 4039                            Register temp_reg,
 4040                            Label& L_success) {
 4041   Label L_failure;
 4042   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 4043   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 4044   bind(L_failure);
 4045 }
 4046 
 4047 
 4048 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 4049                                                    Register super_klass,
 4050                                                    Register temp_reg,
 4051                                                    Label* L_success,
 4052                                                    Label* L_failure,
 4053                                                    Label* L_slow_path,
 4054                                         RegisterOrConstant super_check_offset) {
 4055   assert_different_registers(sub_klass, super_klass, temp_reg);
 4056   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 4057   if (super_check_offset.is_register()) {
 4058     assert_different_registers(sub_klass, super_klass,
 4059                                super_check_offset.as_register());
 4060   } else if (must_load_sco) {
 4061     assert(temp_reg != noreg, "supply either a temp or a register offset");
 4062   }
 4063 
 4064   Label L_fallthrough;
 4065   int label_nulls = 0;
 4066   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4067   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4068   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 4069   assert(label_nulls <= 1, "at most one null in the batch");
 4070 
 4071   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4072   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 4073   Address super_check_offset_addr(super_klass, sco_offset);
 4074 
 4075   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 4076   // range of a jccb.  If this routine grows larger, reconsider at
 4077   // least some of these.
 4078 #define local_jcc(assembler_cond, label)                                \
 4079   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 4080   else                             jcc( assembler_cond, label) /*omit semi*/
 4081 
 4082   // Hacked jmp, which may only be used just before L_fallthrough.
 4083 #define final_jmp(label)                                                \
 4084   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 4085   else                            jmp(label)                /*omit semi*/
 4086 
 4087   // If the pointers are equal, we are done (e.g., String[] elements).
 4088   // This self-check enables sharing of secondary supertype arrays among
 4089   // non-primary types such as array-of-interface.  Otherwise, each such
 4090   // type would need its own customized SSA.
 4091   // We move this check to the front of the fast path because many
 4092   // type checks are in fact trivially successful in this manner,
 4093   // so we get a nicely predicted branch right at the start of the check.
 4094   cmpptr(sub_klass, super_klass);
 4095   local_jcc(Assembler::equal, *L_success);
 4096 
 4097   // Check the supertype display:
 4098   if (must_load_sco) {
 4099     // Positive movl does right thing on LP64.
 4100     movl(temp_reg, super_check_offset_addr);
 4101     super_check_offset = RegisterOrConstant(temp_reg);
 4102   }
 4103   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 4104   cmpptr(super_klass, super_check_addr); // load displayed supertype
 4105 
 4106   // This check has worked decisively for primary supers.
 4107   // Secondary supers are sought in the super_cache ('super_cache_addr').
 4108   // (Secondary supers are interfaces and very deeply nested subtypes.)
 4109   // This works in the same check above because of a tricky aliasing
 4110   // between the super_cache and the primary super display elements.
 4111   // (The 'super_check_addr' can address either, as the case requires.)
 4112   // Note that the cache is updated below if it does not help us find
 4113   // what we need immediately.
 4114   // So if it was a primary super, we can just fail immediately.
 4115   // Otherwise, it's the slow path for us (no success at this point).
 4116 
 4117   if (super_check_offset.is_register()) {
 4118     local_jcc(Assembler::equal, *L_success);
 4119     cmpl(super_check_offset.as_register(), sc_offset);
 4120     if (L_failure == &L_fallthrough) {
 4121       local_jcc(Assembler::equal, *L_slow_path);
 4122     } else {
 4123       local_jcc(Assembler::notEqual, *L_failure);
 4124       final_jmp(*L_slow_path);
 4125     }
 4126   } else if (super_check_offset.as_constant() == sc_offset) {
 4127     // Need a slow path; fast failure is impossible.
 4128     if (L_slow_path == &L_fallthrough) {
 4129       local_jcc(Assembler::equal, *L_success);
 4130     } else {
 4131       local_jcc(Assembler::notEqual, *L_slow_path);
 4132       final_jmp(*L_success);
 4133     }
 4134   } else {
 4135     // No slow path; it's a fast decision.
 4136     if (L_failure == &L_fallthrough) {
 4137       local_jcc(Assembler::equal, *L_success);
 4138     } else {
 4139       local_jcc(Assembler::notEqual, *L_failure);
 4140       final_jmp(*L_success);
 4141     }
 4142   }
 4143 
 4144   bind(L_fallthrough);
 4145 
 4146 #undef local_jcc
 4147 #undef final_jmp
 4148 }
 4149 
 4150 
 4151 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 4152                                                           Register super_klass,
 4153                                                           Register temp_reg,
 4154                                                           Register temp2_reg,
 4155                                                           Label* L_success,
 4156                                                           Label* L_failure,
 4157                                                           bool set_cond_codes) {
 4158   assert_different_registers(sub_klass, super_klass, temp_reg);
 4159   if (temp2_reg != noreg)
 4160     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4161 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4162 
 4163   Label L_fallthrough;
 4164   int label_nulls = 0;
 4165   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4166   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4167   assert(label_nulls <= 1, "at most one null in the batch");
 4168 
 4169   // a couple of useful fields in sub_klass:
 4170   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4171   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4172   Address secondary_supers_addr(sub_klass, ss_offset);
 4173   Address super_cache_addr(     sub_klass, sc_offset);
 4174 
 4175   // Do a linear scan of the secondary super-klass chain.
 4176   // This code is rarely used, so simplicity is a virtue here.
 4177   // The repne_scan instruction uses fixed registers, which we must spill.
 4178   // Don't worry too much about pre-existing connections with the input regs.
 4179 
 4180   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4181   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4182 
 4183   // Get super_klass value into rax (even if it was in rdi or rcx).
 4184   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4185   if (super_klass != rax) {
 4186     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4187     mov(rax, super_klass);
 4188   }
 4189   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4190   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4191 
 4192 #ifndef PRODUCT
 4193   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4194   ExternalAddress pst_counter_addr((address) pst_counter);
 4195   lea(rcx, pst_counter_addr);
 4196   incrementl(Address(rcx, 0));
 4197 #endif //PRODUCT
 4198 
 4199   // We will consult the secondary-super array.
 4200   movptr(rdi, secondary_supers_addr);
 4201   // Load the array length.  (Positive movl does right thing on LP64.)
 4202   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4203   // Skip to start of data.
 4204   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4205 
 4206   // Scan RCX words at [RDI] for an occurrence of RAX.
 4207   // Set NZ/Z based on last compare.
 4208   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4209   // not change flags (only scas instruction which is repeated sets flags).
 4210   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4211 
 4212     testptr(rax,rax); // Set Z = 0
 4213     repne_scan();
 4214 
 4215   // Unspill the temp. registers:
 4216   if (pushed_rdi)  pop(rdi);
 4217   if (pushed_rcx)  pop(rcx);
 4218   if (pushed_rax)  pop(rax);
 4219 
 4220   if (set_cond_codes) {
 4221     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4222     assert(!pushed_rdi, "rdi must be left non-null");
 4223     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4224   }
 4225 
 4226   if (L_failure == &L_fallthrough)
 4227         jccb(Assembler::notEqual, *L_failure);
 4228   else  jcc(Assembler::notEqual, *L_failure);
 4229 
 4230   // Success.  Cache the super we found and proceed in triumph.
 4231   movptr(super_cache_addr, super_klass);
 4232 
 4233   if (L_success != &L_fallthrough) {
 4234     jmp(*L_success);
 4235   }
 4236 
 4237 #undef IS_A_TEMP
 4238 
 4239   bind(L_fallthrough);
 4240 }
 4241 
 4242 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4243                                                    Register super_klass,
 4244                                                    Register temp_reg,
 4245                                                    Register temp2_reg,
 4246                                                    Label* L_success,
 4247                                                    Label* L_failure,
 4248                                                    bool set_cond_codes) {
 4249   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4250   check_klass_subtype_slow_path
 4251     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4252      L_success, L_failure);
 4253 }
 4254 
 4255 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4256                                                    Register super_klass,
 4257                                                    Register temp_reg,
 4258                                                    Register temp2_reg,
 4259                                                    Register temp3_reg,
 4260                                                    Register temp4_reg,
 4261                                                    Label* L_success,
 4262                                                    Label* L_failure) {
 4263   if (UseSecondarySupersTable) {
 4264     check_klass_subtype_slow_path_table
 4265       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4266        L_success, L_failure);
 4267   } else {
 4268     check_klass_subtype_slow_path_linear
 4269       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4270   }
 4271 }
 4272 
 4273 Register MacroAssembler::allocate_if_noreg(Register r,
 4274                                   RegSetIterator<Register> &available_regs,
 4275                                   RegSet &regs_to_push) {
 4276   if (!r->is_valid()) {
 4277     r = *available_regs++;
 4278     regs_to_push += r;
 4279   }
 4280   return r;
 4281 }
 4282 
 4283 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4284                                                          Register super_klass,
 4285                                                          Register temp_reg,
 4286                                                          Register temp2_reg,
 4287                                                          Register temp3_reg,
 4288                                                          Register result_reg,
 4289                                                          Label* L_success,
 4290                                                          Label* L_failure) {
 4291   // NB! Callers may assume that, when temp2_reg is a valid register,
 4292   // this code sets it to a nonzero value.
 4293   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4294 
 4295   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4296 
 4297   Label L_fallthrough;
 4298   int label_nulls = 0;
 4299   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4300   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4301   assert(label_nulls <= 1, "at most one null in the batch");
 4302 
 4303   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4304 
 4305   RegSetIterator<Register> available_regs
 4306     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4307 
 4308   RegSet pushed_regs;
 4309 
 4310   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4311   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4312   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4313   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4314   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4315 
 4316   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4317 
 4318   {
 4319 
 4320     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4321     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4322     subptr(rsp, aligned_size);
 4323     push_set(pushed_regs, 0);
 4324 
 4325     lookup_secondary_supers_table_var(sub_klass,
 4326                                       super_klass,
 4327                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4328     cmpq(result_reg, 0);
 4329 
 4330     // Unspill the temp. registers:
 4331     pop_set(pushed_regs, 0);
 4332     // Increment SP but do not clobber flags.
 4333     lea(rsp, Address(rsp, aligned_size));
 4334   }
 4335 
 4336   if (temp2_reg_was_valid) {
 4337     movq(temp2_reg, 1);
 4338   }
 4339 
 4340   jcc(Assembler::notEqual, *L_failure);
 4341 
 4342   if (L_success != &L_fallthrough) {
 4343     jmp(*L_success);
 4344   }
 4345 
 4346   bind(L_fallthrough);
 4347 }
 4348 
 4349 // population_count variant for running without the POPCNT
 4350 // instruction, which was introduced with SSE4.2 in 2008.
 4351 void MacroAssembler::population_count(Register dst, Register src,
 4352                                       Register scratch1, Register scratch2) {
 4353   assert_different_registers(src, scratch1, scratch2);
 4354   if (UsePopCountInstruction) {
 4355     Assembler::popcntq(dst, src);
 4356   } else {
 4357     assert_different_registers(src, scratch1, scratch2);
 4358     assert_different_registers(dst, scratch1, scratch2);
 4359     Label loop, done;
 4360 
 4361     mov(scratch1, src);
 4362     // dst = 0;
 4363     // while(scratch1 != 0) {
 4364     //   dst++;
 4365     //   scratch1 &= (scratch1 - 1);
 4366     // }
 4367     xorl(dst, dst);
 4368     testq(scratch1, scratch1);
 4369     jccb(Assembler::equal, done);
 4370     {
 4371       bind(loop);
 4372       incq(dst);
 4373       movq(scratch2, scratch1);
 4374       decq(scratch2);
 4375       andq(scratch1, scratch2);
 4376       jccb(Assembler::notEqual, loop);
 4377     }
 4378     bind(done);
 4379   }
 4380 #ifdef ASSERT
 4381   mov64(scratch1, 0xCafeBabeDeadBeef);
 4382   movq(scratch2, scratch1);
 4383 #endif
 4384 }
 4385 
 4386 // Ensure that the inline code and the stub are using the same registers.
 4387 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4388 do {                                                                 \
 4389   assert(r_super_klass  == rax, "mismatch");                         \
 4390   assert(r_array_base   == rbx, "mismatch");                         \
 4391   assert(r_array_length == rcx, "mismatch");                         \
 4392   assert(r_array_index  == rdx, "mismatch");                         \
 4393   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4394   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4395   assert(result         == rdi || result      == noreg, "mismatch"); \
 4396 } while(0)
 4397 
 4398 // Versions of salq and rorq that don't need count to be in rcx
 4399 
 4400 void MacroAssembler::salq(Register dest, Register count) {
 4401   if (count == rcx) {
 4402     Assembler::salq(dest);
 4403   } else {
 4404     assert_different_registers(rcx, dest);
 4405     xchgq(rcx, count);
 4406     Assembler::salq(dest);
 4407     xchgq(rcx, count);
 4408   }
 4409 }
 4410 
 4411 void MacroAssembler::rorq(Register dest, Register count) {
 4412   if (count == rcx) {
 4413     Assembler::rorq(dest);
 4414   } else {
 4415     assert_different_registers(rcx, dest);
 4416     xchgq(rcx, count);
 4417     Assembler::rorq(dest);
 4418     xchgq(rcx, count);
 4419   }
 4420 }
 4421 
 4422 // Return true: we succeeded in generating this code
 4423 //
 4424 // At runtime, return 0 in result if r_super_klass is a superclass of
 4425 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4426 // super_klass_slot of the class you're looking for. This is always
 4427 // the case for instanceof and checkcast.
 4428 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4429                                                          Register r_super_klass,
 4430                                                          Register temp1,
 4431                                                          Register temp2,
 4432                                                          Register temp3,
 4433                                                          Register temp4,
 4434                                                          Register result,
 4435                                                          u1 super_klass_slot) {
 4436   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4437 
 4438   Label L_fallthrough, L_success, L_failure;
 4439 
 4440   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4441 
 4442   const Register
 4443     r_array_index  = temp1,
 4444     r_array_length = temp2,
 4445     r_array_base   = temp3,
 4446     r_bitmap       = temp4;
 4447 
 4448   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4449 
 4450   xorq(result, result); // = 0
 4451 
 4452   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4453   movq(r_array_index, r_bitmap);
 4454 
 4455   // First check the bitmap to see if super_klass might be present. If
 4456   // the bit is zero, we are certain that super_klass is not one of
 4457   // the secondary supers.
 4458   u1 bit = super_klass_slot;
 4459   {
 4460     // NB: If the count in a x86 shift instruction is 0, the flags are
 4461     // not affected, so we do a testq instead.
 4462     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4463     if (shift_count != 0) {
 4464       salq(r_array_index, shift_count);
 4465     } else {
 4466       testq(r_array_index, r_array_index);
 4467     }
 4468   }
 4469   // We test the MSB of r_array_index, i.e. its sign bit
 4470   jcc(Assembler::positive, L_failure);
 4471 
 4472   // Get the first array index that can contain super_klass into r_array_index.
 4473   if (bit != 0) {
 4474     population_count(r_array_index, r_array_index, temp2, temp3);
 4475   } else {
 4476     movl(r_array_index, 1);
 4477   }
 4478   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4479 
 4480   // We will consult the secondary-super array.
 4481   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4482 
 4483   // We're asserting that the first word in an Array<Klass*> is the
 4484   // length, and the second word is the first word of the data. If
 4485   // that ever changes, r_array_base will have to be adjusted here.
 4486   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4487   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4488 
 4489   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4490   jccb(Assembler::equal, L_success);
 4491 
 4492   // Is there another entry to check? Consult the bitmap.
 4493   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4494   jccb(Assembler::carryClear, L_failure);
 4495 
 4496   // Linear probe. Rotate the bitmap so that the next bit to test is
 4497   // in Bit 1.
 4498   if (bit != 0) {
 4499     rorq(r_bitmap, bit);
 4500   }
 4501 
 4502   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4503   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4504   // Kills: r_array_length.
 4505   // Returns: result.
 4506   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4507   // Result (0/1) is in rdi
 4508   jmpb(L_fallthrough);
 4509 
 4510   bind(L_failure);
 4511   incq(result); // 0 => 1
 4512 
 4513   bind(L_success);
 4514   // result = 0;
 4515 
 4516   bind(L_fallthrough);
 4517   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4518 
 4519   if (VerifySecondarySupers) {
 4520     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4521                                   temp1, temp2, temp3);
 4522   }
 4523 }
 4524 
 4525 // At runtime, return 0 in result if r_super_klass is a superclass of
 4526 // r_sub_klass, otherwise return nonzero. Use this version of
 4527 // lookup_secondary_supers_table() if you don't know ahead of time
 4528 // which superclass will be searched for. Used by interpreter and
 4529 // runtime stubs. It is larger and has somewhat greater latency than
 4530 // the version above, which takes a constant super_klass_slot.
 4531 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4532                                                        Register r_super_klass,
 4533                                                        Register temp1,
 4534                                                        Register temp2,
 4535                                                        Register temp3,
 4536                                                        Register temp4,
 4537                                                        Register result) {
 4538   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4539   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4540   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4541 
 4542   Label L_fallthrough, L_success, L_failure;
 4543 
 4544   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4545 
 4546   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4547 
 4548   // FIXME. Once we are sure that all paths reaching this point really
 4549   // do pass rcx as one of our temps we can get rid of the following
 4550   // workaround.
 4551   assert(temps.contains(rcx), "fix this code");
 4552 
 4553   // We prefer to have our shift count in rcx. If rcx is one of our
 4554   // temps, use it for slot. If not, pick any of our temps.
 4555   Register slot;
 4556   if (!temps.contains(rcx)) {
 4557     slot = *available_regs++;
 4558   } else {
 4559     slot = rcx;
 4560   }
 4561 
 4562   const Register r_array_index = *available_regs++;
 4563   const Register r_bitmap      = *available_regs++;
 4564 
 4565   // The logic above guarantees this property, but we state it here.
 4566   assert_different_registers(r_array_index, r_bitmap, rcx);
 4567 
 4568   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4569   movq(r_array_index, r_bitmap);
 4570 
 4571   // First check the bitmap to see if super_klass might be present. If
 4572   // the bit is zero, we are certain that super_klass is not one of
 4573   // the secondary supers.
 4574   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4575   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4576   salq(r_array_index, slot);
 4577 
 4578   testq(r_array_index, r_array_index);
 4579   // We test the MSB of r_array_index, i.e. its sign bit
 4580   jcc(Assembler::positive, L_failure);
 4581 
 4582   const Register r_array_base = *available_regs++;
 4583 
 4584   // Get the first array index that can contain super_klass into r_array_index.
 4585   // Note: Clobbers r_array_base and slot.
 4586   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4587 
 4588   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4589 
 4590   // We will consult the secondary-super array.
 4591   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4592 
 4593   // We're asserting that the first word in an Array<Klass*> is the
 4594   // length, and the second word is the first word of the data. If
 4595   // that ever changes, r_array_base will have to be adjusted here.
 4596   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4597   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4598 
 4599   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4600   jccb(Assembler::equal, L_success);
 4601 
 4602   // Restore slot to its true value
 4603   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4604 
 4605   // Linear probe. Rotate the bitmap so that the next bit to test is
 4606   // in Bit 1.
 4607   rorq(r_bitmap, slot);
 4608 
 4609   // Is there another entry to check? Consult the bitmap.
 4610   btq(r_bitmap, 1);
 4611   jccb(Assembler::carryClear, L_failure);
 4612 
 4613   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4614   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4615   // Kills: r_array_length.
 4616   // Returns: result.
 4617   lookup_secondary_supers_table_slow_path(r_super_klass,
 4618                                           r_array_base,
 4619                                           r_array_index,
 4620                                           r_bitmap,
 4621                                           /*temp1*/result,
 4622                                           /*temp2*/slot,
 4623                                           &L_success,
 4624                                           nullptr);
 4625 
 4626   bind(L_failure);
 4627   movq(result, 1);
 4628   jmpb(L_fallthrough);
 4629 
 4630   bind(L_success);
 4631   xorq(result, result); // = 0
 4632 
 4633   bind(L_fallthrough);
 4634   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4635 
 4636   if (VerifySecondarySupers) {
 4637     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4638                                   temp1, temp2, temp3);
 4639   }
 4640 }
 4641 
 4642 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4643                                  Label* L_success, Label* L_failure) {
 4644   Label L_loop, L_fallthrough;
 4645   {
 4646     int label_nulls = 0;
 4647     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4648     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4649     assert(label_nulls <= 1, "at most one null in the batch");
 4650   }
 4651   bind(L_loop);
 4652   cmpq(value, Address(addr, count, Address::times_8));
 4653   jcc(Assembler::equal, *L_success);
 4654   addl(count, 1);
 4655   cmpl(count, limit);
 4656   jcc(Assembler::less, L_loop);
 4657 
 4658   if (&L_fallthrough != L_failure) {
 4659     jmp(*L_failure);
 4660   }
 4661   bind(L_fallthrough);
 4662 }
 4663 
 4664 // Called by code generated by check_klass_subtype_slow_path
 4665 // above. This is called when there is a collision in the hashed
 4666 // lookup in the secondary supers array.
 4667 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4668                                                              Register r_array_base,
 4669                                                              Register r_array_index,
 4670                                                              Register r_bitmap,
 4671                                                              Register temp1,
 4672                                                              Register temp2,
 4673                                                              Label* L_success,
 4674                                                              Label* L_failure) {
 4675   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4676 
 4677   const Register
 4678     r_array_length = temp1,
 4679     r_sub_klass    = noreg,
 4680     result         = noreg;
 4681 
 4682   Label L_fallthrough;
 4683   int label_nulls = 0;
 4684   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4685   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4686   assert(label_nulls <= 1, "at most one null in the batch");
 4687 
 4688   // Load the array length.
 4689   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4690   // And adjust the array base to point to the data.
 4691   // NB! Effectively increments current slot index by 1.
 4692   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4693   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4694 
 4695   // Linear probe
 4696   Label L_huge;
 4697 
 4698   // The bitmap is full to bursting.
 4699   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4700   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4701   jcc(Assembler::greater, L_huge);
 4702 
 4703   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4704   // current slot (at secondary_supers[r_array_index]) has not yet
 4705   // been inspected, and r_array_index may be out of bounds if we
 4706   // wrapped around the end of the array.
 4707 
 4708   { // This is conventional linear probing, but instead of terminating
 4709     // when a null entry is found in the table, we maintain a bitmap
 4710     // in which a 0 indicates missing entries.
 4711     // The check above guarantees there are 0s in the bitmap, so the loop
 4712     // eventually terminates.
 4713 
 4714     xorl(temp2, temp2); // = 0;
 4715 
 4716     Label L_again;
 4717     bind(L_again);
 4718 
 4719     // Check for array wraparound.
 4720     cmpl(r_array_index, r_array_length);
 4721     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4722 
 4723     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4724     jcc(Assembler::equal, *L_success);
 4725 
 4726     // If the next bit in bitmap is zero, we're done.
 4727     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4728     jcc(Assembler::carryClear, *L_failure);
 4729 
 4730     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4731     addl(r_array_index, 1);
 4732 
 4733     jmp(L_again);
 4734   }
 4735 
 4736   { // Degenerate case: more than 64 secondary supers.
 4737     // FIXME: We could do something smarter here, maybe a vectorized
 4738     // comparison or a binary search, but is that worth any added
 4739     // complexity?
 4740     bind(L_huge);
 4741     xorl(r_array_index, r_array_index); // = 0
 4742     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4743                 L_success,
 4744                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4745 
 4746     bind(L_fallthrough);
 4747   }
 4748 }
 4749 
 4750 struct VerifyHelperArguments {
 4751   Klass* _super;
 4752   Klass* _sub;
 4753   intptr_t _linear_result;
 4754   intptr_t _table_result;
 4755 };
 4756 
 4757 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4758   Klass::on_secondary_supers_verification_failure(args->_super,
 4759                                                   args->_sub,
 4760                                                   args->_linear_result,
 4761                                                   args->_table_result,
 4762                                                   msg);
 4763 }
 4764 
 4765 // Make sure that the hashed lookup and a linear scan agree.
 4766 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4767                                                    Register r_super_klass,
 4768                                                    Register result,
 4769                                                    Register temp1,
 4770                                                    Register temp2,
 4771                                                    Register temp3) {
 4772   const Register
 4773       r_array_index  = temp1,
 4774       r_array_length = temp2,
 4775       r_array_base   = temp3,
 4776       r_bitmap       = noreg;
 4777 
 4778   BLOCK_COMMENT("verify_secondary_supers_table {");
 4779 
 4780   Label L_success, L_failure, L_check, L_done;
 4781 
 4782   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4783   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4784   // And adjust the array base to point to the data.
 4785   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4786 
 4787   testl(r_array_length, r_array_length); // array_length == 0?
 4788   jcc(Assembler::zero, L_failure);
 4789 
 4790   movl(r_array_index, 0);
 4791   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4792   // fall through to L_failure
 4793 
 4794   const Register linear_result = r_array_index; // reuse temp1
 4795 
 4796   bind(L_failure); // not present
 4797   movl(linear_result, 1);
 4798   jmp(L_check);
 4799 
 4800   bind(L_success); // present
 4801   movl(linear_result, 0);
 4802 
 4803   bind(L_check);
 4804   cmpl(linear_result, result);
 4805   jcc(Assembler::equal, L_done);
 4806 
 4807   { // To avoid calling convention issues, build a record on the stack
 4808     // and pass the pointer to that instead.
 4809     push(result);
 4810     push(linear_result);
 4811     push(r_sub_klass);
 4812     push(r_super_klass);
 4813     movptr(c_rarg1, rsp);
 4814     movptr(c_rarg0, (uintptr_t) "mismatch");
 4815     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4816     should_not_reach_here();
 4817   }
 4818   bind(L_done);
 4819 
 4820   BLOCK_COMMENT("} verify_secondary_supers_table");
 4821 }
 4822 
 4823 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4824 
 4825 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4826   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4827 
 4828   Label L_fallthrough;
 4829   if (L_fast_path == nullptr) {
 4830     L_fast_path = &L_fallthrough;
 4831   } else if (L_slow_path == nullptr) {
 4832     L_slow_path = &L_fallthrough;
 4833   }
 4834 
 4835   // Fast path check: class is fully initialized.
 4836   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4837   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4838   jcc(Assembler::equal, *L_fast_path);
 4839 
 4840   // Fast path check: current thread is initializer thread
 4841   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4842   if (L_slow_path == &L_fallthrough) {
 4843     jcc(Assembler::equal, *L_fast_path);
 4844     bind(*L_slow_path);
 4845   } else if (L_fast_path == &L_fallthrough) {
 4846     jcc(Assembler::notEqual, *L_slow_path);
 4847     bind(*L_fast_path);
 4848   } else {
 4849     Unimplemented();
 4850   }
 4851 }
 4852 
 4853 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4854   if (VM_Version::supports_cmov()) {
 4855     cmovl(cc, dst, src);
 4856   } else {
 4857     Label L;
 4858     jccb(negate_condition(cc), L);
 4859     movl(dst, src);
 4860     bind(L);
 4861   }
 4862 }
 4863 
 4864 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4865   if (VM_Version::supports_cmov()) {
 4866     cmovl(cc, dst, src);
 4867   } else {
 4868     Label L;
 4869     jccb(negate_condition(cc), L);
 4870     movl(dst, src);
 4871     bind(L);
 4872   }
 4873 }
 4874 
 4875 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4876   if (!VerifyOops || VerifyAdapterSharing) {
 4877     // Below address of the code string confuses VerifyAdapterSharing
 4878     // because it may differ between otherwise equivalent adapters.
 4879     return;
 4880   }
 4881 
 4882   BLOCK_COMMENT("verify_oop {");
 4883   push(rscratch1);
 4884   push(rax);                          // save rax
 4885   push(reg);                          // pass register argument
 4886 
 4887   // Pass register number to verify_oop_subroutine
 4888   const char* b = nullptr;
 4889   {
 4890     ResourceMark rm;
 4891     stringStream ss;
 4892     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4893     b = code_string(ss.as_string());
 4894   }
 4895   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4896   pushptr(buffer.addr(), rscratch1);
 4897 
 4898   // call indirectly to solve generation ordering problem
 4899   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4900   call(rax);
 4901   // Caller pops the arguments (oop, message) and restores rax, r10
 4902   BLOCK_COMMENT("} verify_oop");
 4903 }
 4904 
 4905 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 4906   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 4907     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 4908     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 4909     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 4910   } else if (VM_Version::supports_avx()) {
 4911     vpcmpeqd(dst, dst, dst, vector_len);
 4912   } else {
 4913     pcmpeqd(dst, dst);
 4914   }
 4915 }
 4916 
 4917 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 4918                                          int extra_slot_offset) {
 4919   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 4920   int stackElementSize = Interpreter::stackElementSize;
 4921   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 4922 #ifdef ASSERT
 4923   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 4924   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 4925 #endif
 4926   Register             scale_reg    = noreg;
 4927   Address::ScaleFactor scale_factor = Address::no_scale;
 4928   if (arg_slot.is_constant()) {
 4929     offset += arg_slot.as_constant() * stackElementSize;
 4930   } else {
 4931     scale_reg    = arg_slot.as_register();
 4932     scale_factor = Address::times(stackElementSize);
 4933   }
 4934   offset += wordSize;           // return PC is on stack
 4935   return Address(rsp, scale_reg, scale_factor, offset);
 4936 }
 4937 
 4938 // Handle the receiver type profile update given the "recv" klass.
 4939 //
 4940 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 4941 // If there are no matching or claimable receiver entries in RD, updates
 4942 // the polymorphic counter.
 4943 //
 4944 // This code expected to run by either the interpreter or JIT-ed code, without
 4945 // extra synchronization. For safety, receiver cells are claimed atomically, which
 4946 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 4947 // counter updates are not atomic.
 4948 //
 4949 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 4950   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 4951   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 4952   int poly_count_offset      = in_bytes(CounterData::count_offset());
 4953   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 4954   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 4955 
 4956   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 4957   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 4958   base_receiver_offset += mdp_offset;
 4959   end_receiver_offset  += mdp_offset;
 4960   poly_count_offset    += mdp_offset;
 4961 
 4962   // Scale down to optimize encoding. Slots are pointer-sized.
 4963   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 4964   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 4965   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 4966   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 4967   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 4968   base_receiver_offset   >>= LogBytesPerWord;
 4969   end_receiver_offset    >>= LogBytesPerWord;
 4970   poly_count_offset      >>= LogBytesPerWord;
 4971   receiver_step          >>= LogBytesPerWord;
 4972   receiver_to_count_step >>= LogBytesPerWord;
 4973 
 4974 #ifdef ASSERT
 4975   // We are about to walk the MDO slots without asking for offsets.
 4976   // Check that our math hits all the right spots.
 4977   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 4978     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 4979     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 4980     int offset = base_receiver_offset + receiver_step*c;
 4981     int count_offset = offset + receiver_to_count_step;
 4982     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 4983     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 4984   }
 4985   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 4986   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 4987 #endif
 4988 
 4989   // Corner case: no profile table. Increment poly counter and exit.
 4990   if (ReceiverTypeData::row_limit() == 0) {
 4991     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 4992     return;
 4993   }
 4994 
 4995   Register offset = rscratch1;
 4996 
 4997   Label L_loop_search_receiver, L_loop_search_empty;
 4998   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 4999 
 5000   // The code here recognizes three major cases:
 5001   //   A. Fastest: receiver found in the table
 5002   //   B. Fast: no receiver in the table, and the table is full
 5003   //   C. Slow: no receiver in the table, free slots in the table
 5004   //
 5005   // The case A performance is most important, as perfectly-behaved code would end up
 5006   // there, especially with larger TypeProfileWidth. The case B performance is
 5007   // important as well, this is where bulk of code would land for normally megamorphic
 5008   // cases. The case C performance is not essential, its job is to deal with installation
 5009   // races, we optimize for code density instead. Case C needs to make sure that receiver
 5010   // rows are only claimed once. This makes sure we never overwrite a row for another
 5011   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 5012   //
 5013   // It is very tempting to handle these cases in a single loop, and claim the first slot
 5014   // without checking the rest of the table. But, profiling code should tolerate free slots
 5015   // in the table, as class unloading can clear them. After such cleanup, the receiver
 5016   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 5017   // to complete, before trying to install new slots. Splitting the code in several tight
 5018   // loops also helpfully optimizes for cases A and B.
 5019   //
 5020   // This code is effectively:
 5021   //
 5022   // restart:
 5023   //   // Fastest: receiver is already installed
 5024   //   for (i = 0; i < receiver_count(); i++) {
 5025   //     if (receiver(i) == recv) goto found_recv(i);
 5026   //   }
 5027   //
 5028   //   // Fast: no receiver, but profile is full
 5029   //   for (i = 0; i < receiver_count(); i++) {
 5030   //     if (receiver(i) == null) goto found_null(i);
 5031   //   }
 5032   //   goto polymorphic
 5033   //
 5034   //   // Slow: try to install receiver
 5035   // found_null(i):
 5036   //   CAS(&receiver(i), null, recv);
 5037   //   goto restart
 5038   //
 5039   // polymorphic:
 5040   //   count++;
 5041   //   return
 5042   //
 5043   // found_recv(i):
 5044   //   *receiver_count(i)++
 5045   //
 5046 
 5047   bind(L_restart);
 5048 
 5049   // Fastest: receiver is already installed
 5050   movptr(offset, base_receiver_offset);
 5051   bind(L_loop_search_receiver);
 5052     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 5053     jccb(Assembler::equal, L_found_recv);
 5054   addptr(offset, receiver_step);
 5055   cmpptr(offset, end_receiver_offset);
 5056   jccb(Assembler::notEqual, L_loop_search_receiver);
 5057 
 5058   // Fast: no receiver, but profile is full
 5059   movptr(offset, base_receiver_offset);
 5060   bind(L_loop_search_empty);
 5061     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 5062     jccb(Assembler::equal, L_found_empty);
 5063   addptr(offset, receiver_step);
 5064   cmpptr(offset, end_receiver_offset);
 5065   jccb(Assembler::notEqual, L_loop_search_empty);
 5066   jmpb(L_polymorphic);
 5067 
 5068   // Slow: try to install receiver
 5069   bind(L_found_empty);
 5070 
 5071   // Atomically swing receiver slot: null -> recv.
 5072   //
 5073   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 5074   // other important registers untouched, as they form the address. Therefore, we need
 5075   // to shift any important registers from RAX into some other spare register. If we
 5076   // have a spare register, we are forced to save it on stack here.
 5077 
 5078   Register spare_reg = noreg;
 5079   Register shifted_mdp = mdp;
 5080   Register shifted_recv = recv;
 5081   if (recv == rax || mdp == rax) {
 5082     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 5083                 (recv != rcx && mdp != rcx) ? rcx :
 5084                 rdx;
 5085     assert_different_registers(mdp, recv, offset, spare_reg);
 5086 
 5087     push(spare_reg);
 5088     if (recv == rax) {
 5089       movptr(spare_reg, recv);
 5090       shifted_recv = spare_reg;
 5091     } else {
 5092       assert(mdp == rax, "Remaining case");
 5093       movptr(spare_reg, mdp);
 5094       shifted_mdp = spare_reg;
 5095     }
 5096   } else {
 5097     push(rax);
 5098   }
 5099 
 5100   // None of the important registers are in RAX after this shuffle.
 5101   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 5102 
 5103   xorptr(rax, rax);
 5104   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 5105 
 5106   // Unshift registers.
 5107   if (recv == rax || mdp == rax) {
 5108     movptr(rax, spare_reg);
 5109     pop(spare_reg);
 5110   } else {
 5111     pop(rax);
 5112   }
 5113 
 5114   // CAS success means the slot now has the receiver we want. CAS failure means
 5115   // something had claimed the slot concurrently: it can be the same receiver we want,
 5116   // or something else. Since this is a slow path, we can optimize for code density,
 5117   // and just restart the search from the beginning.
 5118   jmpb(L_restart);
 5119 
 5120   // Counter updates:
 5121 
 5122   // Increment polymorphic counter instead of receiver slot.
 5123   bind(L_polymorphic);
 5124   movptr(offset, poly_count_offset);
 5125   jmpb(L_count_update);
 5126 
 5127   // Found a receiver, convert its slot offset to corresponding count offset.
 5128   bind(L_found_recv);
 5129   addptr(offset, receiver_to_count_step);
 5130 
 5131   bind(L_count_update);
 5132   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 5133 }
 5134 
 5135 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5136   if (!VerifyOops || VerifyAdapterSharing) {
 5137     // Below address of the code string confuses VerifyAdapterSharing
 5138     // because it may differ between otherwise equivalent adapters.
 5139     return;
 5140   }
 5141 
 5142   push(rscratch1);
 5143   push(rax); // save rax,
 5144   // addr may contain rsp so we will have to adjust it based on the push
 5145   // we just did (and on 64 bit we do two pushes)
 5146   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5147   // stores rax into addr which is backwards of what was intended.
 5148   if (addr.uses(rsp)) {
 5149     lea(rax, addr);
 5150     pushptr(Address(rax, 2 * BytesPerWord));
 5151   } else {
 5152     pushptr(addr);
 5153   }
 5154 
 5155   // Pass register number to verify_oop_subroutine
 5156   const char* b = nullptr;
 5157   {
 5158     ResourceMark rm;
 5159     stringStream ss;
 5160     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 5161     b = code_string(ss.as_string());
 5162   }
 5163   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5164   pushptr(buffer.addr(), rscratch1);
 5165 
 5166   // call indirectly to solve generation ordering problem
 5167   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5168   call(rax);
 5169   // Caller pops the arguments (addr, message) and restores rax, r10.
 5170 }
 5171 
 5172 void MacroAssembler::verify_tlab() {
 5173 #ifdef ASSERT
 5174   if (UseTLAB && VerifyOops) {
 5175     Label next, ok;
 5176     Register t1 = rsi;
 5177 
 5178     push(t1);
 5179 
 5180     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5181     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 5182     jcc(Assembler::aboveEqual, next);
 5183     STOP("assert(top >= start)");
 5184     should_not_reach_here();
 5185 
 5186     bind(next);
 5187     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 5188     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5189     jcc(Assembler::aboveEqual, ok);
 5190     STOP("assert(top <= end)");
 5191     should_not_reach_here();
 5192 
 5193     bind(ok);
 5194     pop(t1);
 5195   }
 5196 #endif
 5197 }
 5198 
 5199 class ControlWord {
 5200  public:
 5201   int32_t _value;
 5202 
 5203   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5204   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5205   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5206   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5207   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5208   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5209   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5210   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5211 
 5212   void print() const {
 5213     // rounding control
 5214     const char* rc;
 5215     switch (rounding_control()) {
 5216       case 0: rc = "round near"; break;
 5217       case 1: rc = "round down"; break;
 5218       case 2: rc = "round up  "; break;
 5219       case 3: rc = "chop      "; break;
 5220       default:
 5221         rc = nullptr; // silence compiler warnings
 5222         fatal("Unknown rounding control: %d", rounding_control());
 5223     };
 5224     // precision control
 5225     const char* pc;
 5226     switch (precision_control()) {
 5227       case 0: pc = "24 bits "; break;
 5228       case 1: pc = "reserved"; break;
 5229       case 2: pc = "53 bits "; break;
 5230       case 3: pc = "64 bits "; break;
 5231       default:
 5232         pc = nullptr; // silence compiler warnings
 5233         fatal("Unknown precision control: %d", precision_control());
 5234     };
 5235     // flags
 5236     char f[9];
 5237     f[0] = ' ';
 5238     f[1] = ' ';
 5239     f[2] = (precision   ()) ? 'P' : 'p';
 5240     f[3] = (underflow   ()) ? 'U' : 'u';
 5241     f[4] = (overflow    ()) ? 'O' : 'o';
 5242     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5243     f[6] = (denormalized()) ? 'D' : 'd';
 5244     f[7] = (invalid     ()) ? 'I' : 'i';
 5245     f[8] = '\x0';
 5246     // output
 5247     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5248   }
 5249 
 5250 };
 5251 
 5252 class StatusWord {
 5253  public:
 5254   int32_t _value;
 5255 
 5256   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5257   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5258   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5259   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5260   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5261   int  top() const                     { return  (_value >> 11) & 7      ; }
 5262   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5263   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5264   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5265   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5266   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5267   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5268   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5269   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5270 
 5271   void print() const {
 5272     // condition codes
 5273     char c[5];
 5274     c[0] = (C3()) ? '3' : '-';
 5275     c[1] = (C2()) ? '2' : '-';
 5276     c[2] = (C1()) ? '1' : '-';
 5277     c[3] = (C0()) ? '0' : '-';
 5278     c[4] = '\x0';
 5279     // flags
 5280     char f[9];
 5281     f[0] = (error_status()) ? 'E' : '-';
 5282     f[1] = (stack_fault ()) ? 'S' : '-';
 5283     f[2] = (precision   ()) ? 'P' : '-';
 5284     f[3] = (underflow   ()) ? 'U' : '-';
 5285     f[4] = (overflow    ()) ? 'O' : '-';
 5286     f[5] = (zero_divide ()) ? 'Z' : '-';
 5287     f[6] = (denormalized()) ? 'D' : '-';
 5288     f[7] = (invalid     ()) ? 'I' : '-';
 5289     f[8] = '\x0';
 5290     // output
 5291     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5292   }
 5293 
 5294 };
 5295 
 5296 class TagWord {
 5297  public:
 5298   int32_t _value;
 5299 
 5300   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5301 
 5302   void print() const {
 5303     printf("%04x", _value & 0xFFFF);
 5304   }
 5305 
 5306 };
 5307 
 5308 class FPU_Register {
 5309  public:
 5310   int32_t _m0;
 5311   int32_t _m1;
 5312   int16_t _ex;
 5313 
 5314   bool is_indefinite() const           {
 5315     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5316   }
 5317 
 5318   void print() const {
 5319     char  sign = (_ex < 0) ? '-' : '+';
 5320     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5321     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5322   };
 5323 
 5324 };
 5325 
 5326 class FPU_State {
 5327  public:
 5328   enum {
 5329     register_size       = 10,
 5330     number_of_registers =  8,
 5331     register_mask       =  7
 5332   };
 5333 
 5334   ControlWord  _control_word;
 5335   StatusWord   _status_word;
 5336   TagWord      _tag_word;
 5337   int32_t      _error_offset;
 5338   int32_t      _error_selector;
 5339   int32_t      _data_offset;
 5340   int32_t      _data_selector;
 5341   int8_t       _register[register_size * number_of_registers];
 5342 
 5343   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5344   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5345 
 5346   const char* tag_as_string(int tag) const {
 5347     switch (tag) {
 5348       case 0: return "valid";
 5349       case 1: return "zero";
 5350       case 2: return "special";
 5351       case 3: return "empty";
 5352     }
 5353     ShouldNotReachHere();
 5354     return nullptr;
 5355   }
 5356 
 5357   void print() const {
 5358     // print computation registers
 5359     { int t = _status_word.top();
 5360       for (int i = 0; i < number_of_registers; i++) {
 5361         int j = (i - t) & register_mask;
 5362         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5363         st(j)->print();
 5364         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5365       }
 5366     }
 5367     printf("\n");
 5368     // print control registers
 5369     printf("ctrl = "); _control_word.print(); printf("\n");
 5370     printf("stat = "); _status_word .print(); printf("\n");
 5371     printf("tags = "); _tag_word    .print(); printf("\n");
 5372   }
 5373 
 5374 };
 5375 
 5376 class Flag_Register {
 5377  public:
 5378   int32_t _value;
 5379 
 5380   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5381   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5382   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5383   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5384   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5385   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5386   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5387 
 5388   void print() const {
 5389     // flags
 5390     char f[8];
 5391     f[0] = (overflow       ()) ? 'O' : '-';
 5392     f[1] = (direction      ()) ? 'D' : '-';
 5393     f[2] = (sign           ()) ? 'S' : '-';
 5394     f[3] = (zero           ()) ? 'Z' : '-';
 5395     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5396     f[5] = (parity         ()) ? 'P' : '-';
 5397     f[6] = (carry          ()) ? 'C' : '-';
 5398     f[7] = '\x0';
 5399     // output
 5400     printf("%08x  flags = %s", _value, f);
 5401   }
 5402 
 5403 };
 5404 
 5405 class IU_Register {
 5406  public:
 5407   int32_t _value;
 5408 
 5409   void print() const {
 5410     printf("%08x  %11d", _value, _value);
 5411   }
 5412 
 5413 };
 5414 
 5415 class IU_State {
 5416  public:
 5417   Flag_Register _eflags;
 5418   IU_Register   _rdi;
 5419   IU_Register   _rsi;
 5420   IU_Register   _rbp;
 5421   IU_Register   _rsp;
 5422   IU_Register   _rbx;
 5423   IU_Register   _rdx;
 5424   IU_Register   _rcx;
 5425   IU_Register   _rax;
 5426 
 5427   void print() const {
 5428     // computation registers
 5429     printf("rax,  = "); _rax.print(); printf("\n");
 5430     printf("rbx,  = "); _rbx.print(); printf("\n");
 5431     printf("rcx  = "); _rcx.print(); printf("\n");
 5432     printf("rdx  = "); _rdx.print(); printf("\n");
 5433     printf("rdi  = "); _rdi.print(); printf("\n");
 5434     printf("rsi  = "); _rsi.print(); printf("\n");
 5435     printf("rbp,  = "); _rbp.print(); printf("\n");
 5436     printf("rsp  = "); _rsp.print(); printf("\n");
 5437     printf("\n");
 5438     // control registers
 5439     printf("flgs = "); _eflags.print(); printf("\n");
 5440   }
 5441 };
 5442 
 5443 
 5444 class CPU_State {
 5445  public:
 5446   FPU_State _fpu_state;
 5447   IU_State  _iu_state;
 5448 
 5449   void print() const {
 5450     printf("--------------------------------------------------\n");
 5451     _iu_state .print();
 5452     printf("\n");
 5453     _fpu_state.print();
 5454     printf("--------------------------------------------------\n");
 5455   }
 5456 
 5457 };
 5458 
 5459 
 5460 static void _print_CPU_state(CPU_State* state) {
 5461   state->print();
 5462 };
 5463 
 5464 
 5465 void MacroAssembler::print_CPU_state() {
 5466   push_CPU_state();
 5467   push(rsp);                // pass CPU state
 5468   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5469   addptr(rsp, wordSize);       // discard argument
 5470   pop_CPU_state();
 5471 }
 5472 
 5473 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5474   // Either restore the MXCSR register after returning from the JNI Call
 5475   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5476   if (RestoreMXCSROnJNICalls) {
 5477     ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5478   } else if (CheckJNICalls) {
 5479     call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5480   }
 5481   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5482   vzeroupper();
 5483 }
 5484 
 5485 // ((OopHandle)result).resolve();
 5486 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5487   assert_different_registers(result, tmp);
 5488 
 5489   // Only 64 bit platforms support GCs that require a tmp register
 5490   // Only IN_HEAP loads require a thread_tmp register
 5491   // OopHandle::resolve is an indirection like jobject.
 5492   access_load_at(T_OBJECT, IN_NATIVE,
 5493                  result, Address(result, 0), tmp);
 5494 }
 5495 
 5496 // ((WeakHandle)result).resolve();
 5497 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5498   assert_different_registers(rresult, rtmp);
 5499   Label resolved;
 5500 
 5501   // A null weak handle resolves to null.
 5502   cmpptr(rresult, 0);
 5503   jcc(Assembler::equal, resolved);
 5504 
 5505   // Only 64 bit platforms support GCs that require a tmp register
 5506   // Only IN_HEAP loads require a thread_tmp register
 5507   // WeakHandle::resolve is an indirection like jweak.
 5508   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5509                  rresult, Address(rresult, 0), rtmp);
 5510   bind(resolved);
 5511 }
 5512 
 5513 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5514   // get mirror
 5515   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5516   load_method_holder(mirror, method);
 5517   movptr(mirror, Address(mirror, mirror_offset));
 5518   resolve_oop_handle(mirror, tmp);
 5519 }
 5520 
 5521 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5522   load_method_holder(rresult, rmethod);
 5523   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5524 }
 5525 
 5526 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5527   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5528   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5529   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5530 }
 5531 
 5532 void MacroAssembler::load_metadata(Register dst, Register src) {
 5533   if (UseCompactObjectHeaders) {
 5534     load_narrow_klass_compact(dst, src);
 5535   } else {
 5536     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5537   }
 5538 }
 5539 
 5540 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5541   assert(UseCompactObjectHeaders, "expect compact object headers");
 5542   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5543   shrq(dst, markWord::klass_shift);
 5544 }
 5545 
 5546 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5547   assert_different_registers(src, tmp);
 5548   assert_different_registers(dst, tmp);
 5549 
 5550   if (UseCompactObjectHeaders) {
 5551     load_narrow_klass_compact(dst, src);
 5552     decode_klass_not_null(dst, tmp);
 5553   } else {
 5554     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5555     decode_klass_not_null(dst, tmp);
 5556   }
 5557 }
 5558 
 5559 void MacroAssembler::load_prototype_header(Register dst, Register src, Register tmp) {
 5560   load_klass(dst, src, tmp);
 5561   movptr(dst, Address(dst, Klass::prototype_header_offset()));
 5562 }
 5563 
 5564 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5565   assert(!UseCompactObjectHeaders, "not with compact headers");
 5566   assert_different_registers(src, tmp);
 5567   assert_different_registers(dst, tmp);
 5568   encode_klass_not_null(src, tmp);
 5569   movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5570 }
 5571 
 5572 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5573   if (UseCompactObjectHeaders) {
 5574     assert(tmp != noreg, "need tmp");
 5575     assert_different_registers(klass, obj, tmp);
 5576     load_narrow_klass_compact(tmp, obj);
 5577     cmpl(klass, tmp);
 5578   } else {
 5579     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5580   }
 5581 }
 5582 
 5583 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5584   if (UseCompactObjectHeaders) {
 5585     assert(tmp2 != noreg, "need tmp2");
 5586     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5587     load_narrow_klass_compact(tmp1, obj1);
 5588     load_narrow_klass_compact(tmp2, obj2);
 5589     cmpl(tmp1, tmp2);
 5590   } else {
 5591     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5592     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5593   }
 5594 }
 5595 
 5596 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5597                                     Register tmp1) {
 5598   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5599   decorators = AccessInternal::decorator_fixup(decorators, type);
 5600   bool as_raw = (decorators & AS_RAW) != 0;
 5601   if (as_raw) {
 5602     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5603   } else {
 5604     bs->load_at(this, decorators, type, dst, src, tmp1);
 5605   }
 5606 }
 5607 
 5608 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5609                                      Register tmp1, Register tmp2, Register tmp3) {
 5610   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5611   decorators = AccessInternal::decorator_fixup(decorators, type);
 5612   bool as_raw = (decorators & AS_RAW) != 0;
 5613   if (as_raw) {
 5614     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5615   } else {
 5616     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5617   }
 5618 }
 5619 
 5620 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
 5621                                      Register inline_layout_info) {
 5622   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5623   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
 5624 }
 5625 
 5626 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
 5627   movptr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
 5628   movl(offset, Address(offset, InlineKlass::payload_offset_offset()));
 5629 }
 5630 
 5631 void MacroAssembler::payload_addr(Register oop, Register data, Register inline_klass) {
 5632   // ((address) (void*) o) + vk->payload_offset();
 5633   Register offset = (data == oop) ? rscratch1 : data;
 5634   payload_offset(inline_klass, offset);
 5635   if (data == oop) {
 5636     addptr(data, offset);
 5637   } else {
 5638     lea(data, Address(oop, offset));
 5639   }
 5640 }
 5641 
 5642 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5643   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5644 }
 5645 
 5646 // Doesn't do verification, generates fixed size code
 5647 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5648   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5649 }
 5650 
 5651 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5652                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5653   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5654 }
 5655 
 5656 // Used for storing nulls.
 5657 void MacroAssembler::store_heap_oop_null(Address dst) {
 5658   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5659 }
 5660 
 5661 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5662   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5663   // Store to klass gap in destination
 5664   movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5665 }
 5666 
 5667 #ifdef ASSERT
 5668 void MacroAssembler::verify_heapbase(const char* msg) {
 5669   assert (UseCompressedOops, "should be compressed");
 5670   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5671   if (CheckCompressedOops) {
 5672     Label ok;
 5673     ExternalAddress src2(CompressedOops::base_addr());
 5674     const bool is_src2_reachable = reachable(src2);
 5675     if (!is_src2_reachable) {
 5676       push(rscratch1);  // cmpptr trashes rscratch1
 5677     }
 5678     cmpptr(r12_heapbase, src2, rscratch1);
 5679     jcc(Assembler::equal, ok);
 5680     STOP(msg);
 5681     bind(ok);
 5682     if (!is_src2_reachable) {
 5683       pop(rscratch1);
 5684     }
 5685   }
 5686 }
 5687 #endif
 5688 
 5689 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5690 void MacroAssembler::encode_heap_oop(Register r) {
 5691 #ifdef ASSERT
 5692   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5693 #endif
 5694   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5695   if (CompressedOops::base() == nullptr) {
 5696     if (CompressedOops::shift() != 0) {
 5697       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5698       shrq(r, LogMinObjAlignmentInBytes);
 5699     }
 5700     return;
 5701   }
 5702   testq(r, r);
 5703   cmovq(Assembler::equal, r, r12_heapbase);
 5704   subq(r, r12_heapbase);
 5705   shrq(r, LogMinObjAlignmentInBytes);
 5706 }
 5707 
 5708 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5709 #ifdef ASSERT
 5710   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5711   if (CheckCompressedOops) {
 5712     Label ok;
 5713     testq(r, r);
 5714     jcc(Assembler::notEqual, ok);
 5715     STOP("null oop passed to encode_heap_oop_not_null");
 5716     bind(ok);
 5717   }
 5718 #endif
 5719   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5720   if (CompressedOops::base() != nullptr) {
 5721     subq(r, r12_heapbase);
 5722   }
 5723   if (CompressedOops::shift() != 0) {
 5724     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5725     shrq(r, LogMinObjAlignmentInBytes);
 5726   }
 5727 }
 5728 
 5729 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5730 #ifdef ASSERT
 5731   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5732   if (CheckCompressedOops) {
 5733     Label ok;
 5734     testq(src, src);
 5735     jcc(Assembler::notEqual, ok);
 5736     STOP("null oop passed to encode_heap_oop_not_null2");
 5737     bind(ok);
 5738   }
 5739 #endif
 5740   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5741   if (dst != src) {
 5742     movq(dst, src);
 5743   }
 5744   if (CompressedOops::base() != nullptr) {
 5745     subq(dst, r12_heapbase);
 5746   }
 5747   if (CompressedOops::shift() != 0) {
 5748     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5749     shrq(dst, LogMinObjAlignmentInBytes);
 5750   }
 5751 }
 5752 
 5753 void  MacroAssembler::decode_heap_oop(Register r) {
 5754 #ifdef ASSERT
 5755   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5756 #endif
 5757   if (CompressedOops::base() == nullptr) {
 5758     if (CompressedOops::shift() != 0) {
 5759       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5760       shlq(r, LogMinObjAlignmentInBytes);
 5761     }
 5762   } else {
 5763     Label done;
 5764     shlq(r, LogMinObjAlignmentInBytes);
 5765     jccb(Assembler::equal, done);
 5766     addq(r, r12_heapbase);
 5767     bind(done);
 5768   }
 5769   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5770 }
 5771 
 5772 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5773   // Note: it will change flags
 5774   assert (UseCompressedOops, "should only be used for compressed headers");
 5775   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5776   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5777   // vtableStubs also counts instructions in pd_code_size_limit.
 5778   // Also do not verify_oop as this is called by verify_oop.
 5779   if (CompressedOops::shift() != 0) {
 5780     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5781     shlq(r, LogMinObjAlignmentInBytes);
 5782     if (CompressedOops::base() != nullptr) {
 5783       addq(r, r12_heapbase);
 5784     }
 5785   } else {
 5786     assert (CompressedOops::base() == nullptr, "sanity");
 5787   }
 5788 }
 5789 
 5790 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5791   // Note: it will change flags
 5792   assert (UseCompressedOops, "should only be used for compressed headers");
 5793   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5794   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5795   // vtableStubs also counts instructions in pd_code_size_limit.
 5796   // Also do not verify_oop as this is called by verify_oop.
 5797   if (CompressedOops::shift() != 0) {
 5798     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5799     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5800       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5801     } else {
 5802       if (dst != src) {
 5803         movq(dst, src);
 5804       }
 5805       shlq(dst, LogMinObjAlignmentInBytes);
 5806       if (CompressedOops::base() != nullptr) {
 5807         addq(dst, r12_heapbase);
 5808       }
 5809     }
 5810   } else {
 5811     assert (CompressedOops::base() == nullptr, "sanity");
 5812     if (dst != src) {
 5813       movq(dst, src);
 5814     }
 5815   }
 5816 }
 5817 
 5818 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5819   BLOCK_COMMENT("encode_klass_not_null {");
 5820   assert_different_registers(r, tmp);
 5821   if (CompressedKlassPointers::base() != nullptr) {
 5822     if (AOTCodeCache::is_on_for_dump()) {
 5823       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5824     } else {
 5825       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5826     }
 5827     subq(r, tmp);
 5828   }
 5829   if (CompressedKlassPointers::shift() != 0) {
 5830     shrq(r, CompressedKlassPointers::shift());
 5831   }
 5832   BLOCK_COMMENT("} encode_klass_not_null");
 5833 }
 5834 
 5835 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5836   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5837   assert_different_registers(src, dst);
 5838   if (CompressedKlassPointers::base() != nullptr) {
 5839     if (AOTCodeCache::is_on_for_dump()) {
 5840       movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5841       negq(dst);
 5842     } else {
 5843       movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5844     }
 5845     addq(dst, src);
 5846   } else {
 5847     movptr(dst, src);
 5848   }
 5849   if (CompressedKlassPointers::shift() != 0) {
 5850     shrq(dst, CompressedKlassPointers::shift());
 5851   }
 5852   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5853 }
 5854 
 5855 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5856   BLOCK_COMMENT("decode_klass_not_null {");
 5857   assert_different_registers(r, tmp);
 5858   // Note: it will change flags
 5859   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5860   // vtableStubs also counts instructions in pd_code_size_limit.
 5861   // Also do not verify_oop as this is called by verify_oop.
 5862   if (CompressedKlassPointers::shift() != 0) {
 5863     shlq(r, CompressedKlassPointers::shift());
 5864   }
 5865   if (CompressedKlassPointers::base() != nullptr) {
 5866     if (AOTCodeCache::is_on_for_dump()) {
 5867       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5868     } else {
 5869       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5870     }
 5871     addq(r, tmp);
 5872   }
 5873   BLOCK_COMMENT("} decode_klass_not_null");
 5874 }
 5875 
 5876 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 5877   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 5878   assert_different_registers(src, dst);
 5879   // Note: it will change flags
 5880   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5881   // vtableStubs also counts instructions in pd_code_size_limit.
 5882   // Also do not verify_oop as this is called by verify_oop.
 5883 
 5884   if (CompressedKlassPointers::base() == nullptr &&
 5885       CompressedKlassPointers::shift() == 0) {
 5886     // The best case scenario is that there is no base or shift. Then it is already
 5887     // a pointer that needs nothing but a register rename.
 5888     movl(dst, src);
 5889   } else {
 5890     if (CompressedKlassPointers::shift() <= Address::times_8) {
 5891       if (CompressedKlassPointers::base() != nullptr) {
 5892         if (AOTCodeCache::is_on_for_dump()) {
 5893           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5894         } else {
 5895           movptr(dst, (intptr_t)CompressedKlassPointers::base());
 5896         }
 5897       } else {
 5898         xorq(dst, dst);
 5899       }
 5900       if (CompressedKlassPointers::shift() != 0) {
 5901         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 5902         leaq(dst, Address(dst, src, Address::times_8, 0));
 5903       } else {
 5904         addq(dst, src);
 5905       }
 5906     } else {
 5907       if (CompressedKlassPointers::base() != nullptr) {
 5908         if (AOTCodeCache::is_on_for_dump()) {
 5909           movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
 5910           shrq(dst, CompressedKlassPointers::shift());
 5911         } else {
 5912           const intptr_t base_right_shifted =
 5913                (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 5914           movptr(dst, base_right_shifted);
 5915         }
 5916       } else {
 5917         xorq(dst, dst);
 5918       }
 5919       addq(dst, src);
 5920       shlq(dst, CompressedKlassPointers::shift());
 5921     }
 5922   }
 5923   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 5924 }
 5925 
 5926 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 5927   assert (UseCompressedOops, "should only be used for compressed headers");
 5928   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5929   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5930   int oop_index = oop_recorder()->find_index(obj);
 5931   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5932   mov_narrow_oop(dst, oop_index, rspec);
 5933 }
 5934 
 5935 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 5936   assert (UseCompressedOops, "should only be used for compressed headers");
 5937   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5938   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5939   int oop_index = oop_recorder()->find_index(obj);
 5940   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5941   mov_narrow_oop(dst, oop_index, rspec);
 5942 }
 5943 
 5944 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 5945   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5946   int klass_index = oop_recorder()->find_index(k);
 5947   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5948   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5949 }
 5950 
 5951 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 5952   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5953   int klass_index = oop_recorder()->find_index(k);
 5954   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5955   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5956 }
 5957 
 5958 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 5959   assert (UseCompressedOops, "should only be used for compressed headers");
 5960   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5961   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5962   int oop_index = oop_recorder()->find_index(obj);
 5963   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5964   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5965 }
 5966 
 5967 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 5968   assert (UseCompressedOops, "should only be used for compressed headers");
 5969   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5970   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5971   int oop_index = oop_recorder()->find_index(obj);
 5972   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 5973   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 5974 }
 5975 
 5976 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 5977   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5978   int klass_index = oop_recorder()->find_index(k);
 5979   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5980   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5981 }
 5982 
 5983 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 5984   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5985   int klass_index = oop_recorder()->find_index(k);
 5986   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5987   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5988 }
 5989 
 5990 void MacroAssembler::reinit_heapbase() {
 5991   if (UseCompressedOops) {
 5992     if (Universe::heap() != nullptr && !AOTCodeCache::is_on_for_dump()) {
 5993       if (CompressedOops::base() == nullptr) {
 5994         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 5995       } else {
 5996         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 5997       }
 5998     } else {
 5999       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 6000     }
 6001   }
 6002 }
 6003 
 6004 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
 6005   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
 6006   // An inline type might be returned. If fields are in registers we
 6007   // need to allocate an inline type instance and initialize it with
 6008   // the value of the fields.
 6009   Label skip;
 6010   // We only need a new buffered inline type if a new one is not returned
 6011   testptr(rax, 1);
 6012   jcc(Assembler::zero, skip);
 6013   int call_offset = -1;
 6014 
 6015   // The following code is similar to allocation code in TemplateTable::_new but has some slight differences,
 6016   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
 6017   // allocating is not necessary if vk != nullptr, etc.
 6018   Label slow_case;
 6019   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
 6020   mov(rscratch1, rax); // save rax for slow_case since *_allocate may corrupt it when allocation failed
 6021   if (vk != nullptr) {
 6022     // Called from C1, where the return type is statically known.
 6023     movptr(rbx, (intptr_t)vk->get_InlineKlass());
 6024     jint lh = vk->layout_helper();
 6025     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
 6026     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
 6027       tlab_allocate(rax, noreg, lh, r13, r14, slow_case);
 6028     } else {
 6029       jmp(slow_case);
 6030     }
 6031   } else {
 6032     // Call from interpreter. RAX contains ((the InlineKlass* of the return type) | 0x01)
 6033     mov(rbx, rax);
 6034     andptr(rbx, -2);
 6035     if (UseTLAB) {
 6036       movl(r14, Address(rbx, Klass::layout_helper_offset()));
 6037       testl(r14, Klass::_lh_instance_slow_path_bit);
 6038       jcc(Assembler::notZero, slow_case);
 6039       tlab_allocate(rax, r14, 0, r13, r14, slow_case);
 6040     } else {
 6041       jmp(slow_case);
 6042     }
 6043   }
 6044   if (UseTLAB) {
 6045     // 2. Initialize buffered inline instance header
 6046     Register buffer_obj = rax;
 6047     Register klass = rbx;
 6048     if (UseCompactObjectHeaders) {
 6049       Register mark_word = r13;
 6050       movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 6051       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), mark_word);
 6052     } else {
 6053       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), (intptr_t)markWord::inline_type_prototype().value());
 6054       xorl(r13, r13);
 6055       store_klass_gap(buffer_obj, r13);
 6056       if (vk == nullptr) {
 6057         // store_klass corrupts rbx(klass), so save it in r13 for later use (interpreter case only).
 6058         mov(r13, klass);
 6059       }
 6060       store_klass(buffer_obj, klass, rscratch1);
 6061       klass = r13;
 6062     }
 6063     // 3. Initialize its fields with an inline class specific handler
 6064     if (vk != nullptr) {
 6065       call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
 6066     } else {
 6067       movptr(rbx, Address(klass, InlineKlass::adr_members_offset()));
 6068       movptr(rbx, Address(rbx, InlineKlass::pack_handler_offset()));
 6069       call(rbx);
 6070     }
 6071     jmp(skip);
 6072   }
 6073   bind(slow_case);
 6074   // We failed to allocate a new inline type, fall back to a runtime
 6075   // call. Some oop field may be live in some registers but we can't
 6076   // tell. That runtime call will take care of preserving them
 6077   // across a GC if there's one.
 6078   mov(rax, rscratch1);
 6079 
 6080   if (from_interpreter) {
 6081     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
 6082   } else {
 6083     call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
 6084     call_offset = offset();
 6085   }
 6086 
 6087   bind(skip);
 6088   return call_offset;
 6089 }
 6090 
 6091 // Move a value between registers/stack slots and update the reg_state
 6092 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
 6093   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
 6094   if (reg_state[to->value()] == reg_written) {
 6095     return true; // Already written
 6096   }
 6097   if (from != to && bt != T_VOID) {
 6098     if (reg_state[to->value()] == reg_readonly) {
 6099       return false; // Not yet writable
 6100     }
 6101     if (from->is_reg()) {
 6102       if (to->is_reg()) {
 6103         if (from->is_XMMRegister()) {
 6104           if (bt == T_DOUBLE) {
 6105             movdbl(to->as_XMMRegister(), from->as_XMMRegister());
 6106           } else {
 6107             assert(bt == T_FLOAT, "must be float");
 6108             movflt(to->as_XMMRegister(), from->as_XMMRegister());
 6109           }
 6110         } else {
 6111           movq(to->as_Register(), from->as_Register());
 6112         }
 6113       } else {
 6114         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6115         Address to_addr = Address(rsp, st_off);
 6116         if (from->is_XMMRegister()) {
 6117           if (bt == T_DOUBLE) {
 6118             movdbl(to_addr, from->as_XMMRegister());
 6119           } else {
 6120             assert(bt == T_FLOAT, "must be float");
 6121             movflt(to_addr, from->as_XMMRegister());
 6122           }
 6123         } else {
 6124           movq(to_addr, from->as_Register());
 6125         }
 6126       }
 6127     } else {
 6128       Address from_addr = Address(rsp, from->reg2stack() * VMRegImpl::stack_slot_size + wordSize);
 6129       if (to->is_reg()) {
 6130         if (to->is_XMMRegister()) {
 6131           if (bt == T_DOUBLE) {
 6132             movdbl(to->as_XMMRegister(), from_addr);
 6133           } else {
 6134             assert(bt == T_FLOAT, "must be float");
 6135             movflt(to->as_XMMRegister(), from_addr);
 6136           }
 6137         } else {
 6138           movq(to->as_Register(), from_addr);
 6139         }
 6140       } else {
 6141         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6142         movq(r13, from_addr);
 6143         movq(Address(rsp, st_off), r13);
 6144       }
 6145     }
 6146   }
 6147   // Update register states
 6148   reg_state[from->value()] = reg_writable;
 6149   reg_state[to->value()] = reg_written;
 6150   return true;
 6151 }
 6152 
 6153 // Calculate the extra stack space required for packing or unpacking inline
 6154 // args and adjust the stack pointer (see MacroAssembler::remove_frame).
 6155 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
 6156   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
 6157   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
 6158   assert(sp_inc > 0, "sanity");
 6159   // Two additional slots to account for return address
 6160   sp_inc +=  2 * VMRegImpl::stack_slot_size;
 6161 
 6162   push(rbp);
 6163   subptr(rsp, sp_inc);
 6164 #ifdef ASSERT
 6165   movl(Address(rsp, 0), badRegWordVal);
 6166   movl(Address(rsp, VMRegImpl::stack_slot_size), badRegWordVal);
 6167 #endif
 6168   return sp_inc + wordSize; // account for rbp space
 6169 }
 6170 
 6171 // Read all fields from an inline type buffer and store the field values in registers/stack slots.
 6172 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
 6173                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
 6174                                           RegState reg_state[]) {
 6175   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
 6176   assert(from->is_valid(), "source must be valid");
 6177   bool progress = false;
 6178 #ifdef ASSERT
 6179   const int start_offset = offset();
 6180 #endif
 6181 
 6182   Label L_null, L_notNull;
 6183   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
 6184   Register tmp1 = r10;
 6185   Register tmp2 = r13;
 6186   Register fromReg = noreg;
 6187   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
 6188   bool done = true;
 6189   bool mark_done = true;
 6190   VMReg toReg;
 6191   BasicType bt;
 6192   // Check if argument requires a null check
 6193   bool null_check = false;
 6194   VMReg nullCheckReg;
 6195   while (stream.next(nullCheckReg, bt)) {
 6196     if (sig->at(stream.sig_index())._offset == -1) {
 6197       null_check = true;
 6198       break;
 6199     }
 6200   }
 6201   stream.reset(sig_index, to_index);
 6202   while (stream.next(toReg, bt)) {
 6203     assert(toReg->is_valid(), "destination must be valid");
 6204     int idx = (int)toReg->value();
 6205     if (reg_state[idx] == reg_readonly) {
 6206       if (idx != from->value()) {
 6207         mark_done = false;
 6208       }
 6209       done = false;
 6210       continue;
 6211     } else if (reg_state[idx] == reg_written) {
 6212       continue;
 6213     }
 6214     assert(reg_state[idx] == reg_writable, "must be writable");
 6215     reg_state[idx] = reg_written;
 6216     progress = true;
 6217 
 6218     if (fromReg == noreg) {
 6219       if (from->is_reg()) {
 6220         fromReg = from->as_Register();
 6221       } else {
 6222         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6223         movq(tmp1, Address(rsp, st_off));
 6224         fromReg = tmp1;
 6225       }
 6226       if (null_check) {
 6227         // Nullable inline type argument, emit null check
 6228         testptr(fromReg, fromReg);
 6229         jcc(Assembler::zero, L_null);
 6230       }
 6231     }
 6232     int off = sig->at(stream.sig_index())._offset;
 6233     if (off == -1) {
 6234       assert(null_check, "Missing null check at");
 6235       if (toReg->is_stack()) {
 6236         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6237         movq(Address(rsp, st_off), 1);
 6238       } else {
 6239         movq(toReg->as_Register(), 1);
 6240       }
 6241       continue;
 6242     }
 6243     if (sig->at(stream.sig_index())._vt_oop) {
 6244       if (toReg->is_stack()) {
 6245         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6246         movq(Address(rsp, st_off), fromReg);
 6247       } else {
 6248         movq(toReg->as_Register(), fromReg);
 6249       }
 6250       continue;
 6251     }
 6252     assert(off > 0, "offset in object should be positive");
 6253     Address fromAddr = Address(fromReg, off);
 6254     if (!toReg->is_XMMRegister()) {
 6255       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
 6256       if (is_reference_type(bt)) {
 6257         load_heap_oop(dst, fromAddr);
 6258       } else {
 6259         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
 6260         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
 6261       }
 6262       if (toReg->is_stack()) {
 6263         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6264         movq(Address(rsp, st_off), dst);
 6265       }
 6266     } else if (bt == T_DOUBLE) {
 6267       movdbl(toReg->as_XMMRegister(), fromAddr);
 6268     } else {
 6269       assert(bt == T_FLOAT, "must be float");
 6270       movflt(toReg->as_XMMRegister(), fromAddr);
 6271     }
 6272   }
 6273   if (progress && null_check) {
 6274     if (done) {
 6275       jmp(L_notNull);
 6276       bind(L_null);
 6277       // Set null marker to zero to signal that the argument is null.
 6278       // Also set all fields to zero since the runtime requires a canonical
 6279       // representation of a flat null.
 6280       stream.reset(sig_index, to_index);
 6281       while (stream.next(toReg, bt)) {
 6282         if (toReg->is_stack()) {
 6283           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6284           movq(Address(rsp, st_off), 0);
 6285         } else if (toReg->is_XMMRegister()) {
 6286           xorps(toReg->as_XMMRegister(), toReg->as_XMMRegister());
 6287         } else {
 6288           xorl(toReg->as_Register(), toReg->as_Register());
 6289         }
 6290       }
 6291       bind(L_notNull);
 6292     } else {
 6293       bind(L_null);
 6294     }
 6295   }
 6296 
 6297   sig_index = stream.sig_index();
 6298   to_index = stream.regs_index();
 6299 
 6300   if (mark_done && reg_state[from->value()] != reg_written) {
 6301     // This is okay because no one else will write to that slot
 6302     reg_state[from->value()] = reg_writable;
 6303   }
 6304   from_index--;
 6305   assert(progress || (start_offset == offset()), "should not emit code");
 6306   return done;
 6307 }
 6308 
 6309 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
 6310                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
 6311                                         RegState reg_state[], Register val_array) {
 6312   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
 6313   assert(to->is_valid(), "destination must be valid");
 6314 
 6315   if (reg_state[to->value()] == reg_written) {
 6316     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6317     return true; // Already written
 6318   }
 6319 
 6320   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
 6321   Register val_obj_tmp = r11;
 6322   Register from_reg_tmp = r14;
 6323   Register tmp1 = r10;
 6324   Register tmp2 = r13;
 6325   Register tmp3 = rbx;
 6326   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
 6327 
 6328   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
 6329 
 6330   if (reg_state[to->value()] == reg_readonly) {
 6331     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
 6332       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6333       return false; // Not yet writable
 6334     }
 6335     val_obj = val_obj_tmp;
 6336   }
 6337 
 6338   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
 6339   VMReg fromReg;
 6340   BasicType bt;
 6341   Label L_null;
 6342   while (stream.next(fromReg, bt)) {
 6343     assert(fromReg->is_valid(), "source must be valid");
 6344     reg_state[fromReg->value()] = reg_writable;
 6345 
 6346     int off = sig->at(stream.sig_index())._offset;
 6347     if (off == -1) {
 6348       // Nullable inline type argument, emit null check
 6349       Label L_notNull;
 6350       if (fromReg->is_stack()) {
 6351         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6352         testb(Address(rsp, ld_off), 1);
 6353       } else {
 6354         testb(fromReg->as_Register(), 1);
 6355       }
 6356       jcc(Assembler::notZero, L_notNull);
 6357       movptr(val_obj, 0);
 6358       jmp(L_null);
 6359       bind(L_notNull);
 6360       continue;
 6361     }
 6362     if (sig->at(stream.sig_index())._vt_oop) {
 6363       // buffer argument: use if non null
 6364       if (fromReg->is_stack()) {
 6365         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6366         movptr(val_obj, Address(rsp, ld_off));
 6367       } else {
 6368         movptr(val_obj, fromReg->as_Register());
 6369       }
 6370       testptr(val_obj, val_obj);
 6371       jcc(Assembler::notEqual, L_null);
 6372       // otherwise get the buffer from the just allocated pool of buffers
 6373       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
 6374       load_heap_oop(val_obj, Address(val_array, index));
 6375       continue;
 6376     }
 6377 
 6378     assert(off > 0, "offset in object should be positive");
 6379     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
 6380 
 6381     // Pack the scalarized field into the value object.
 6382     Address dst(val_obj, off);
 6383     if (!fromReg->is_XMMRegister()) {
 6384       Register src;
 6385       if (fromReg->is_stack()) {
 6386         src = from_reg_tmp;
 6387         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6388         load_sized_value(src, Address(rsp, ld_off), size_in_bytes, /* is_signed */ false);
 6389       } else {
 6390         src = fromReg->as_Register();
 6391       }
 6392       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
 6393       if (is_reference_type(bt)) {
 6394         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
 6395         mov(tmp3, val_obj);
 6396         Address dst_with_tmp3(tmp3, off);
 6397         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
 6398       } else {
 6399         store_sized_value(dst, src, size_in_bytes);
 6400       }
 6401     } else if (bt == T_DOUBLE) {
 6402       movdbl(dst, fromReg->as_XMMRegister());
 6403     } else {
 6404       assert(bt == T_FLOAT, "must be float");
 6405       movflt(dst, fromReg->as_XMMRegister());
 6406     }
 6407   }
 6408   bind(L_null);
 6409   sig_index = stream.sig_index();
 6410   from_index = stream.regs_index();
 6411 
 6412   assert(reg_state[to->value()] == reg_writable, "must have already been read");
 6413   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
 6414   assert(success, "to register must be writeable");
 6415   return true;
 6416 }
 6417 
 6418 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
 6419   return reg->is_XMMRegister() ? xmm8->as_VMReg() : r14->as_VMReg();
 6420 }
 6421 
 6422 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
 6423   assert((initial_framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 6424   if (needs_stack_repair) {
 6425     // The method has a scalarized entry point (where fields of value object arguments
 6426     // are passed through registers and stack), and a non-scalarized entry point (where
 6427     // value object arguments are given as oops). The non-scalarized entry point will
 6428     // first load each field of value object arguments and store them in registers and on
 6429     // the stack in a way compatible with the scalarized entry point. To do so, some extra
 6430     // stack space might be reserved (if argument registers are not enough). On leaving the
 6431     // method, this space must be freed.
 6432     //
 6433     // In case we used the non-scalarized entry point the stack looks like this:
 6434     //
 6435     // | Arguments from caller     |
 6436     // |---------------------------|  <-- caller's SP
 6437     // | Return address #1         |
 6438     // | Saved RBP #1              |
 6439     // |---------------------------|
 6440     // | Extension space for       |
 6441     // |   inline arg (un)packing  |
 6442     // |---------------------------|  <-- start of this method's frame
 6443     // | Return address #2         |
 6444     // | Saved RBP #2              |
 6445     // |---------------------------|  <-- RBP (with -XX:+PreserveFramePointer)
 6446     // | sp_inc                    |
 6447     // | method locals             |
 6448     // |---------------------------|  <-- SP
 6449     //
 6450     // Space for the return pc and saved rbp is reserved twice. But only the #1 copies
 6451     // contain the real values of return pc and saved rbp. The #2 copies are not reliable
 6452     // and should not be used. They are mostly needed to add space between the extension
 6453     // space and the locals, as there would be between the real arguments and the locals
 6454     // if we don't need to do unpacking (from the scalarized entry point).
 6455     //
 6456     // When leaving, one must load RBP #1 into RBP, and use the copy #1 of the return address,
 6457     // while keeping in mind that from the scalarized entry point, there will be only one
 6458     // copy. Indeed, in the case we used the scalarized calling convention, the stack looks like this:
 6459     //
 6460     // | Arguments from caller     |
 6461     // |---------------------------|  <-- caller's SP
 6462     // | Return address            |
 6463     // | Saved RBP                 |
 6464     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
 6465     // | sp_inc                    |
 6466     // | method locals             |
 6467     // |---------------------------|  <-- SP
 6468     //
 6469     // The sp_inc stack slot holds the total size of the frame, including the extension
 6470     // space and copies #2 of the return address and the saved RBP (but never the copies
 6471     // #1 of the return address and saved RBP). That is how to find the copies #1 of the
 6472     // return address and saved rbp. This size is expressed in bytes. Be careful when using
 6473     // it from C++ in pointer arithmetic you might need to divide it by wordSize.
 6474 
 6475     // The stack increment resides just below the saved rbp
 6476     addq(rsp, Address(rsp, initial_framesize - wordSize));
 6477     pop(rbp);
 6478   } else {
 6479     if (initial_framesize > 0) {
 6480       addq(rsp, initial_framesize);
 6481     }
 6482     pop(rbp);
 6483   }
 6484 }
 6485 
 6486 #if COMPILER2_OR_JVMCI
 6487 
 6488 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 6489 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp, KRegister mask) {
 6490   // cnt - number of qwords (8-byte words).
 6491   // base - start address, qword aligned.
 6492   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 6493   bool use64byteVector = (MaxVectorSize == 64) && (CopyAVX3Threshold == 0);
 6494   if (use64byteVector) {
 6495     evpbroadcastq(xtmp, val, AVX_512bit);
 6496   } else if (MaxVectorSize >= 32) {
 6497     movdq(xtmp, val);
 6498     punpcklqdq(xtmp, xtmp);
 6499     vinserti128_high(xtmp, xtmp);
 6500   } else {
 6501     movdq(xtmp, val);
 6502     punpcklqdq(xtmp, xtmp);
 6503   }
 6504   jmp(L_zero_64_bytes);
 6505 
 6506   BIND(L_loop);
 6507   if (MaxVectorSize >= 32) {
 6508     fill64(base, 0, xtmp, use64byteVector);
 6509   } else {
 6510     movdqu(Address(base,  0), xtmp);
 6511     movdqu(Address(base, 16), xtmp);
 6512     movdqu(Address(base, 32), xtmp);
 6513     movdqu(Address(base, 48), xtmp);
 6514   }
 6515   addptr(base, 64);
 6516 
 6517   BIND(L_zero_64_bytes);
 6518   subptr(cnt, 8);
 6519   jccb(Assembler::greaterEqual, L_loop);
 6520 
 6521   // Copy trailing 64 bytes
 6522   if (use64byteVector) {
 6523     addptr(cnt, 8);
 6524     jccb(Assembler::equal, L_end);
 6525     fill64_masked(3, base, 0, xtmp, mask, cnt, val, true);
 6526     jmp(L_end);
 6527   } else {
 6528     addptr(cnt, 4);
 6529     jccb(Assembler::less, L_tail);
 6530     if (MaxVectorSize >= 32) {
 6531       vmovdqu(Address(base, 0), xtmp);
 6532     } else {
 6533       movdqu(Address(base,  0), xtmp);
 6534       movdqu(Address(base, 16), xtmp);
 6535     }
 6536   }
 6537   addptr(base, 32);
 6538   subptr(cnt, 4);
 6539 
 6540   BIND(L_tail);
 6541   addptr(cnt, 4);
 6542   jccb(Assembler::lessEqual, L_end);
 6543   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 6544     fill32_masked(3, base, 0, xtmp, mask, cnt, val);
 6545   } else {
 6546     decrement(cnt);
 6547 
 6548     BIND(L_sloop);
 6549     movq(Address(base, 0), xtmp);
 6550     addptr(base, 8);
 6551     decrement(cnt);
 6552     jccb(Assembler::greaterEqual, L_sloop);
 6553   }
 6554   BIND(L_end);
 6555 }
 6556 
 6557 // Clearing constant sized memory using YMM/ZMM registers.
 6558 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 6559   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 6560   bool use64byteVector = (MaxVectorSize > 32) && (CopyAVX3Threshold == 0);
 6561 
 6562   int vector64_count = (cnt & (~0x7)) >> 3;
 6563   cnt = cnt & 0x7;
 6564   const int fill64_per_loop = 4;
 6565   const int max_unrolled_fill64 = 8;
 6566 
 6567   // 64 byte initialization loop.
 6568   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 6569   int start64 = 0;
 6570   if (vector64_count > max_unrolled_fill64) {
 6571     Label LOOP;
 6572     Register index = rtmp;
 6573 
 6574     start64 = vector64_count - (vector64_count % fill64_per_loop);
 6575 
 6576     movl(index, 0);
 6577     BIND(LOOP);
 6578     for (int i = 0; i < fill64_per_loop; i++) {
 6579       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 6580     }
 6581     addl(index, fill64_per_loop * 64);
 6582     cmpl(index, start64 * 64);
 6583     jccb(Assembler::less, LOOP);
 6584   }
 6585   for (int i = start64; i < vector64_count; i++) {
 6586     fill64(base, i * 64, xtmp, use64byteVector);
 6587   }
 6588 
 6589   // Clear remaining 64 byte tail.
 6590   int disp = vector64_count * 64;
 6591   if (cnt) {
 6592     switch (cnt) {
 6593       case 1:
 6594         movq(Address(base, disp), xtmp);
 6595         break;
 6596       case 2:
 6597         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 6598         break;
 6599       case 3:
 6600         movl(rtmp, 0x7);
 6601         kmovwl(mask, rtmp);
 6602         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 6603         break;
 6604       case 4:
 6605         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6606         break;
 6607       case 5:
 6608         if (use64byteVector) {
 6609           movl(rtmp, 0x1F);
 6610           kmovwl(mask, rtmp);
 6611           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6612         } else {
 6613           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6614           movq(Address(base, disp + 32), xtmp);
 6615         }
 6616         break;
 6617       case 6:
 6618         if (use64byteVector) {
 6619           movl(rtmp, 0x3F);
 6620           kmovwl(mask, rtmp);
 6621           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6622         } else {
 6623           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6624           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 6625         }
 6626         break;
 6627       case 7:
 6628         if (use64byteVector) {
 6629           movl(rtmp, 0x7F);
 6630           kmovwl(mask, rtmp);
 6631           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6632         } else {
 6633           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6634           movl(rtmp, 0x7);
 6635           kmovwl(mask, rtmp);
 6636           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 6637         }
 6638         break;
 6639       default:
 6640         fatal("Unexpected length : %d\n",cnt);
 6641         break;
 6642     }
 6643   }
 6644 }
 6645 
 6646 void MacroAssembler::clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp,
 6647                                bool is_large, bool word_copy_only, KRegister mask) {
 6648   // cnt      - number of qwords (8-byte words).
 6649   // base     - start address, qword aligned.
 6650   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6651   assert(base==rdi, "base register must be edi for rep stos");
 6652   assert(val==rax,   "val register must be eax for rep stos");
 6653   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6654   assert(InitArrayShortSize % BytesPerLong == 0,
 6655     "InitArrayShortSize should be the multiple of BytesPerLong");
 6656 
 6657   Label DONE;
 6658 
 6659   if (!is_large) {
 6660     Label LOOP, LONG;
 6661     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6662     jccb(Assembler::greater, LONG);
 6663 
 6664     decrement(cnt);
 6665     jccb(Assembler::negative, DONE); // Zero length
 6666 
 6667     // Use individual pointer-sized stores for small counts:
 6668     BIND(LOOP);
 6669     movptr(Address(base, cnt, Address::times_ptr), val);
 6670     decrement(cnt);
 6671     jccb(Assembler::greaterEqual, LOOP);
 6672     jmpb(DONE);
 6673 
 6674     BIND(LONG);
 6675   }
 6676 
 6677   // Use longer rep-prefixed ops for non-small counts:
 6678   if (UseFastStosb && !word_copy_only) {
 6679     shlptr(cnt, 3); // convert to number of bytes
 6680     rep_stosb();
 6681   } else if (UseXMMForObjInit) {
 6682     xmm_clear_mem(base, cnt, val, xtmp, mask);
 6683   } else {
 6684     rep_stos();
 6685   }
 6686 
 6687   BIND(DONE);
 6688 }
 6689 
 6690 #endif //COMPILER2_OR_JVMCI
 6691 
 6692 
 6693 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6694                                    Register to, Register value, Register count,
 6695                                    Register rtmp, XMMRegister xtmp) {
 6696   ShortBranchVerifier sbv(this);
 6697   assert_different_registers(to, value, count, rtmp);
 6698   Label L_exit;
 6699   Label L_fill_2_bytes, L_fill_4_bytes;
 6700 
 6701 #if defined(COMPILER2)
 6702   if(MaxVectorSize >=32 &&
 6703      VM_Version::supports_avx512vlbw() &&
 6704      VM_Version::supports_bmi2()) {
 6705     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6706     return;
 6707   }
 6708 #endif
 6709 
 6710   int shift = -1;
 6711   switch (t) {
 6712     case T_BYTE:
 6713       shift = 2;
 6714       break;
 6715     case T_SHORT:
 6716       shift = 1;
 6717       break;
 6718     case T_INT:
 6719       shift = 0;
 6720       break;
 6721     default: ShouldNotReachHere();
 6722   }
 6723 
 6724   if (t == T_BYTE) {
 6725     andl(value, 0xff);
 6726     movl(rtmp, value);
 6727     shll(rtmp, 8);
 6728     orl(value, rtmp);
 6729   }
 6730   if (t == T_SHORT) {
 6731     andl(value, 0xffff);
 6732   }
 6733   if (t == T_BYTE || t == T_SHORT) {
 6734     movl(rtmp, value);
 6735     shll(rtmp, 16);
 6736     orl(value, rtmp);
 6737   }
 6738 
 6739   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6740   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6741   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6742     Label L_skip_align2;
 6743     // align source address at 4 bytes address boundary
 6744     if (t == T_BYTE) {
 6745       Label L_skip_align1;
 6746       // One byte misalignment happens only for byte arrays
 6747       testptr(to, 1);
 6748       jccb(Assembler::zero, L_skip_align1);
 6749       movb(Address(to, 0), value);
 6750       increment(to);
 6751       decrement(count);
 6752       BIND(L_skip_align1);
 6753     }
 6754     // Two bytes misalignment happens only for byte and short (char) arrays
 6755     testptr(to, 2);
 6756     jccb(Assembler::zero, L_skip_align2);
 6757     movw(Address(to, 0), value);
 6758     addptr(to, 2);
 6759     subptr(count, 1<<(shift-1));
 6760     BIND(L_skip_align2);
 6761   }
 6762   {
 6763     Label L_fill_32_bytes;
 6764     if (!UseUnalignedLoadStores) {
 6765       // align to 8 bytes, we know we are 4 byte aligned to start
 6766       testptr(to, 4);
 6767       jccb(Assembler::zero, L_fill_32_bytes);
 6768       movl(Address(to, 0), value);
 6769       addptr(to, 4);
 6770       subptr(count, 1<<shift);
 6771     }
 6772     BIND(L_fill_32_bytes);
 6773     {
 6774       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6775       movdl(xtmp, value);
 6776       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6777         Label L_check_fill_32_bytes;
 6778         if (UseAVX > 2) {
 6779           // Fill 64-byte chunks
 6780           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6781 
 6782           // If number of bytes to fill < CopyAVX3Threshold, perform fill using AVX2
 6783           cmpptr(count, CopyAVX3Threshold);
 6784           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6785 
 6786           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6787 
 6788           subptr(count, 16 << shift);
 6789           jcc(Assembler::less, L_check_fill_32_bytes);
 6790           align(16);
 6791 
 6792           BIND(L_fill_64_bytes_loop_avx3);
 6793           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6794           addptr(to, 64);
 6795           subptr(count, 16 << shift);
 6796           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6797           jmpb(L_check_fill_32_bytes);
 6798 
 6799           BIND(L_check_fill_64_bytes_avx2);
 6800         }
 6801         // Fill 64-byte chunks
 6802         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6803 
 6804         subptr(count, 16 << shift);
 6805         jcc(Assembler::less, L_check_fill_32_bytes);
 6806 
 6807         // align data for 64-byte chunks
 6808         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6809         if (EnableX86ECoreOpts) {
 6810             // align 'big' arrays to cache lines to minimize split_stores
 6811             cmpptr(count, 96 << shift);
 6812             jcc(Assembler::below, L_fill_64_bytes_loop);
 6813 
 6814             // Find the bytes needed for alignment
 6815             movptr(rtmp, to);
 6816             andptr(rtmp, 0x1c);
 6817             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6818             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6819             addptr(rtmp, 32);
 6820             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6821             subptr(count, rtmp);    // adjust count by number of elements
 6822 
 6823             align(16);
 6824             BIND(L_align_64_bytes_loop);
 6825             movdl(Address(to, 0), xtmp);
 6826             addptr(to, 4);
 6827             subptr(rtmp, 1 << shift);
 6828             jcc(Assembler::greater, L_align_64_bytes_loop);
 6829         }
 6830 
 6831         align(16);
 6832         BIND(L_fill_64_bytes_loop);
 6833         vmovdqu(Address(to, 0), xtmp);
 6834         vmovdqu(Address(to, 32), xtmp);
 6835         addptr(to, 64);
 6836         subptr(count, 16 << shift);
 6837         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6838 
 6839         align(16);
 6840         BIND(L_check_fill_32_bytes);
 6841         addptr(count, 8 << shift);
 6842         jccb(Assembler::less, L_check_fill_8_bytes);
 6843         vmovdqu(Address(to, 0), xtmp);
 6844         addptr(to, 32);
 6845         subptr(count, 8 << shift);
 6846 
 6847         BIND(L_check_fill_8_bytes);
 6848         // clean upper bits of YMM registers
 6849         movdl(xtmp, value);
 6850         pshufd(xtmp, xtmp, 0);
 6851       } else {
 6852         // Fill 32-byte chunks
 6853         pshufd(xtmp, xtmp, 0);
 6854 
 6855         subptr(count, 8 << shift);
 6856         jcc(Assembler::less, L_check_fill_8_bytes);
 6857         align(16);
 6858 
 6859         BIND(L_fill_32_bytes_loop);
 6860 
 6861         if (UseUnalignedLoadStores) {
 6862           movdqu(Address(to, 0), xtmp);
 6863           movdqu(Address(to, 16), xtmp);
 6864         } else {
 6865           movq(Address(to, 0), xtmp);
 6866           movq(Address(to, 8), xtmp);
 6867           movq(Address(to, 16), xtmp);
 6868           movq(Address(to, 24), xtmp);
 6869         }
 6870 
 6871         addptr(to, 32);
 6872         subptr(count, 8 << shift);
 6873         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6874 
 6875         BIND(L_check_fill_8_bytes);
 6876       }
 6877       addptr(count, 8 << shift);
 6878       jccb(Assembler::zero, L_exit);
 6879       jmpb(L_fill_8_bytes);
 6880 
 6881       //
 6882       // length is too short, just fill qwords
 6883       //
 6884       align(16);
 6885       BIND(L_fill_8_bytes_loop);
 6886       movq(Address(to, 0), xtmp);
 6887       addptr(to, 8);
 6888       BIND(L_fill_8_bytes);
 6889       subptr(count, 1 << (shift + 1));
 6890       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6891     }
 6892   }
 6893 
 6894   Label L_fill_4_bytes_loop;
 6895   testl(count, 1 << shift);
 6896   jccb(Assembler::zero, L_fill_2_bytes);
 6897 
 6898   align(16);
 6899   BIND(L_fill_4_bytes_loop);
 6900   movl(Address(to, 0), value);
 6901   addptr(to, 4);
 6902 
 6903   BIND(L_fill_4_bytes);
 6904   subptr(count, 1 << shift);
 6905   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 6906 
 6907   if (t == T_BYTE || t == T_SHORT) {
 6908     Label L_fill_byte;
 6909     BIND(L_fill_2_bytes);
 6910     // fill trailing 2 bytes
 6911     testl(count, 1<<(shift-1));
 6912     jccb(Assembler::zero, L_fill_byte);
 6913     movw(Address(to, 0), value);
 6914     if (t == T_BYTE) {
 6915       addptr(to, 2);
 6916       BIND(L_fill_byte);
 6917       // fill trailing byte
 6918       testl(count, 1);
 6919       jccb(Assembler::zero, L_exit);
 6920       movb(Address(to, 0), value);
 6921     } else {
 6922       BIND(L_fill_byte);
 6923     }
 6924   } else {
 6925     BIND(L_fill_2_bytes);
 6926   }
 6927   BIND(L_exit);
 6928 }
 6929 
 6930 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 6931   switch(type) {
 6932     case T_BYTE:
 6933     case T_BOOLEAN:
 6934       evpbroadcastb(dst, src, vector_len);
 6935       break;
 6936     case T_SHORT:
 6937     case T_CHAR:
 6938       evpbroadcastw(dst, src, vector_len);
 6939       break;
 6940     case T_INT:
 6941     case T_FLOAT:
 6942       evpbroadcastd(dst, src, vector_len);
 6943       break;
 6944     case T_LONG:
 6945     case T_DOUBLE:
 6946       evpbroadcastq(dst, src, vector_len);
 6947       break;
 6948     default:
 6949       fatal("Unhandled type : %s", type2name(type));
 6950       break;
 6951   }
 6952 }
 6953 
 6954 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 6955 //
 6956 // @IntrinsicCandidate
 6957 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 6958 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6959 //     int i = 0;
 6960 //     for (; i < len; i++) {
 6961 //         char c = sa[sp++];
 6962 //         if (c > '\u00FF')
 6963 //             break;
 6964 //         da[dp++] = (byte) c;
 6965 //     }
 6966 //     return i;
 6967 // }
 6968 //
 6969 // @IntrinsicCandidate
 6970 // int java.lang.StringCoding.encodeISOArray0(
 6971 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 6972 //   int i = 0;
 6973 //   for (; i < len; i++) {
 6974 //     char c = StringUTF16.getChar(sa, sp++);
 6975 //     if (c > '\u00FF')
 6976 //       break;
 6977 //     da[dp++] = (byte) c;
 6978 //   }
 6979 //   return i;
 6980 // }
 6981 //
 6982 // @IntrinsicCandidate
 6983 // int java.lang.StringCoding.encodeAsciiArray0(
 6984 //         char[] sa, int sp, byte[] da, int dp, int len) {
 6985 //   int i = 0;
 6986 //   for (; i < len; i++) {
 6987 //     char c = sa[sp++];
 6988 //     if (c >= '\u0080')
 6989 //       break;
 6990 //     da[dp++] = (byte) c;
 6991 //   }
 6992 //   return i;
 6993 // }
 6994 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 6995   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 6996   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 6997   Register tmp5, Register result, bool ascii) {
 6998 
 6999   // rsi: src
 7000   // rdi: dst
 7001   // rdx: len
 7002   // rcx: tmp5
 7003   // rax: result
 7004   ShortBranchVerifier sbv(this);
 7005   assert_different_registers(src, dst, len, tmp5, result);
 7006   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 7007 
 7008   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 7009   int short_mask = ascii ? 0xff80 : 0xff00;
 7010 
 7011   // set result
 7012   xorl(result, result);
 7013   // check for zero length
 7014   testl(len, len);
 7015   jcc(Assembler::zero, L_done);
 7016 
 7017   movl(result, len);
 7018 
 7019   // Setup pointers
 7020   lea(src, Address(src, len, Address::times_2)); // char[]
 7021   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 7022   negptr(len);
 7023 
 7024   if (UseSSE42Intrinsics || UseAVX >= 2) {
 7025     Label L_copy_8_chars, L_copy_8_chars_exit;
 7026     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 7027 
 7028     if (UseAVX >= 2) {
 7029       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 7030       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7031       movdl(tmp1Reg, tmp5);
 7032       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 7033       jmp(L_chars_32_check);
 7034 
 7035       bind(L_copy_32_chars);
 7036       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 7037       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 7038       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7039       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7040       jccb(Assembler::notZero, L_copy_32_chars_exit);
 7041       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7042       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 7043       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 7044 
 7045       bind(L_chars_32_check);
 7046       addptr(len, 32);
 7047       jcc(Assembler::lessEqual, L_copy_32_chars);
 7048 
 7049       bind(L_copy_32_chars_exit);
 7050       subptr(len, 16);
 7051       jccb(Assembler::greater, L_copy_16_chars_exit);
 7052 
 7053     } else if (UseSSE42Intrinsics) {
 7054       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7055       movdl(tmp1Reg, tmp5);
 7056       pshufd(tmp1Reg, tmp1Reg, 0);
 7057       jmpb(L_chars_16_check);
 7058     }
 7059 
 7060     bind(L_copy_16_chars);
 7061     if (UseAVX >= 2) {
 7062       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 7063       vptest(tmp2Reg, tmp1Reg);
 7064       jcc(Assembler::notZero, L_copy_16_chars_exit);
 7065       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 7066       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 7067     } else {
 7068       if (UseAVX > 0) {
 7069         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7070         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7071         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 7072       } else {
 7073         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7074         por(tmp2Reg, tmp3Reg);
 7075         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7076         por(tmp2Reg, tmp4Reg);
 7077       }
 7078       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7079       jccb(Assembler::notZero, L_copy_16_chars_exit);
 7080       packuswb(tmp3Reg, tmp4Reg);
 7081     }
 7082     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 7083 
 7084     bind(L_chars_16_check);
 7085     addptr(len, 16);
 7086     jcc(Assembler::lessEqual, L_copy_16_chars);
 7087 
 7088     bind(L_copy_16_chars_exit);
 7089     if (UseAVX >= 2) {
 7090       // clean upper bits of YMM registers
 7091       vpxor(tmp2Reg, tmp2Reg);
 7092       vpxor(tmp3Reg, tmp3Reg);
 7093       vpxor(tmp4Reg, tmp4Reg);
 7094       movdl(tmp1Reg, tmp5);
 7095       pshufd(tmp1Reg, tmp1Reg, 0);
 7096     }
 7097     subptr(len, 8);
 7098     jccb(Assembler::greater, L_copy_8_chars_exit);
 7099 
 7100     bind(L_copy_8_chars);
 7101     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 7102     ptest(tmp3Reg, tmp1Reg);
 7103     jccb(Assembler::notZero, L_copy_8_chars_exit);
 7104     packuswb(tmp3Reg, tmp1Reg);
 7105     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 7106     addptr(len, 8);
 7107     jccb(Assembler::lessEqual, L_copy_8_chars);
 7108 
 7109     bind(L_copy_8_chars_exit);
 7110     subptr(len, 8);
 7111     jccb(Assembler::zero, L_done);
 7112   }
 7113 
 7114   bind(L_copy_1_char);
 7115   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 7116   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 7117   jccb(Assembler::notZero, L_copy_1_char_exit);
 7118   movb(Address(dst, len, Address::times_1, 0), tmp5);
 7119   addptr(len, 1);
 7120   jccb(Assembler::less, L_copy_1_char);
 7121 
 7122   bind(L_copy_1_char_exit);
 7123   addptr(result, len); // len is negative count of not processed elements
 7124 
 7125   bind(L_done);
 7126 }
 7127 
 7128 /**
 7129  * Helper for multiply_to_len().
 7130  */
 7131 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 7132   addq(dest_lo, src1);
 7133   adcq(dest_hi, 0);
 7134   addq(dest_lo, src2);
 7135   adcq(dest_hi, 0);
 7136 }
 7137 
 7138 /**
 7139  * Multiply 64 bit by 64 bit first loop.
 7140  */
 7141 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 7142                                            Register y, Register y_idx, Register z,
 7143                                            Register carry, Register product,
 7144                                            Register idx, Register kdx) {
 7145   //
 7146   //  jlong carry, x[], y[], z[];
 7147   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7148   //    huge_128 product = y[idx] * x[xstart] + carry;
 7149   //    z[kdx] = (jlong)product;
 7150   //    carry  = (jlong)(product >>> 64);
 7151   //  }
 7152   //  z[xstart] = carry;
 7153   //
 7154 
 7155   Label L_first_loop, L_first_loop_exit;
 7156   Label L_one_x, L_one_y, L_multiply;
 7157 
 7158   decrementl(xstart);
 7159   jcc(Assembler::negative, L_one_x);
 7160 
 7161   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7162   rorq(x_xstart, 32); // convert big-endian to little-endian
 7163 
 7164   bind(L_first_loop);
 7165   decrementl(idx);
 7166   jcc(Assembler::negative, L_first_loop_exit);
 7167   decrementl(idx);
 7168   jcc(Assembler::negative, L_one_y);
 7169   movq(y_idx, Address(y, idx, Address::times_4,  0));
 7170   rorq(y_idx, 32); // convert big-endian to little-endian
 7171   bind(L_multiply);
 7172   movq(product, x_xstart);
 7173   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 7174   addq(product, carry);
 7175   adcq(rdx, 0);
 7176   subl(kdx, 2);
 7177   movl(Address(z, kdx, Address::times_4,  4), product);
 7178   shrq(product, 32);
 7179   movl(Address(z, kdx, Address::times_4,  0), product);
 7180   movq(carry, rdx);
 7181   jmp(L_first_loop);
 7182 
 7183   bind(L_one_y);
 7184   movl(y_idx, Address(y,  0));
 7185   jmp(L_multiply);
 7186 
 7187   bind(L_one_x);
 7188   movl(x_xstart, Address(x,  0));
 7189   jmp(L_first_loop);
 7190 
 7191   bind(L_first_loop_exit);
 7192 }
 7193 
 7194 /**
 7195  * Multiply 64 bit by 64 bit and add 128 bit.
 7196  */
 7197 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 7198                                             Register yz_idx, Register idx,
 7199                                             Register carry, Register product, int offset) {
 7200   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 7201   //     z[kdx] = (jlong)product;
 7202 
 7203   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 7204   rorq(yz_idx, 32); // convert big-endian to little-endian
 7205   movq(product, x_xstart);
 7206   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 7207   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 7208   rorq(yz_idx, 32); // convert big-endian to little-endian
 7209 
 7210   add2_with_carry(rdx, product, carry, yz_idx);
 7211 
 7212   movl(Address(z, idx, Address::times_4,  offset+4), product);
 7213   shrq(product, 32);
 7214   movl(Address(z, idx, Address::times_4,  offset), product);
 7215 
 7216 }
 7217 
 7218 /**
 7219  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 7220  */
 7221 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 7222                                              Register yz_idx, Register idx, Register jdx,
 7223                                              Register carry, Register product,
 7224                                              Register carry2) {
 7225   //   jlong carry, x[], y[], z[];
 7226   //   int kdx = ystart+1;
 7227   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7228   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 7229   //     z[kdx+idx+1] = (jlong)product;
 7230   //     jlong carry2  = (jlong)(product >>> 64);
 7231   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 7232   //     z[kdx+idx] = (jlong)product;
 7233   //     carry  = (jlong)(product >>> 64);
 7234   //   }
 7235   //   idx += 2;
 7236   //   if (idx > 0) {
 7237   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 7238   //     z[kdx+idx] = (jlong)product;
 7239   //     carry  = (jlong)(product >>> 64);
 7240   //   }
 7241   //
 7242 
 7243   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7244 
 7245   movl(jdx, idx);
 7246   andl(jdx, 0xFFFFFFFC);
 7247   shrl(jdx, 2);
 7248 
 7249   bind(L_third_loop);
 7250   subl(jdx, 1);
 7251   jcc(Assembler::negative, L_third_loop_exit);
 7252   subl(idx, 4);
 7253 
 7254   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 7255   movq(carry2, rdx);
 7256 
 7257   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 7258   movq(carry, rdx);
 7259   jmp(L_third_loop);
 7260 
 7261   bind (L_third_loop_exit);
 7262 
 7263   andl (idx, 0x3);
 7264   jcc(Assembler::zero, L_post_third_loop_done);
 7265 
 7266   Label L_check_1;
 7267   subl(idx, 2);
 7268   jcc(Assembler::negative, L_check_1);
 7269 
 7270   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 7271   movq(carry, rdx);
 7272 
 7273   bind (L_check_1);
 7274   addl (idx, 0x2);
 7275   andl (idx, 0x1);
 7276   subl(idx, 1);
 7277   jcc(Assembler::negative, L_post_third_loop_done);
 7278 
 7279   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 7280   movq(product, x_xstart);
 7281   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 7282   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 7283 
 7284   add2_with_carry(rdx, product, yz_idx, carry);
 7285 
 7286   movl(Address(z, idx, Address::times_4,  0), product);
 7287   shrq(product, 32);
 7288 
 7289   shlq(rdx, 32);
 7290   orq(product, rdx);
 7291   movq(carry, product);
 7292 
 7293   bind(L_post_third_loop_done);
 7294 }
 7295 
 7296 /**
 7297  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 7298  *
 7299  */
 7300 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 7301                                                   Register carry, Register carry2,
 7302                                                   Register idx, Register jdx,
 7303                                                   Register yz_idx1, Register yz_idx2,
 7304                                                   Register tmp, Register tmp3, Register tmp4) {
 7305   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 7306 
 7307   //   jlong carry, x[], y[], z[];
 7308   //   int kdx = ystart+1;
 7309   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7310   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 7311   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 7312   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 7313   //     carry  = (jlong)(tmp4 >>> 64);
 7314   //     z[kdx+idx+1] = (jlong)tmp3;
 7315   //     z[kdx+idx] = (jlong)tmp4;
 7316   //   }
 7317   //   idx += 2;
 7318   //   if (idx > 0) {
 7319   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 7320   //     z[kdx+idx] = (jlong)yz_idx1;
 7321   //     carry  = (jlong)(yz_idx1 >>> 64);
 7322   //   }
 7323   //
 7324 
 7325   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7326 
 7327   movl(jdx, idx);
 7328   andl(jdx, 0xFFFFFFFC);
 7329   shrl(jdx, 2);
 7330 
 7331   bind(L_third_loop);
 7332   subl(jdx, 1);
 7333   jcc(Assembler::negative, L_third_loop_exit);
 7334   subl(idx, 4);
 7335 
 7336   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 7337   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 7338   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 7339   rorxq(yz_idx2, yz_idx2, 32);
 7340 
 7341   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 7342   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 7343 
 7344   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 7345   rorxq(yz_idx1, yz_idx1, 32);
 7346   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7347   rorxq(yz_idx2, yz_idx2, 32);
 7348 
 7349   if (VM_Version::supports_adx()) {
 7350     adcxq(tmp3, carry);
 7351     adoxq(tmp3, yz_idx1);
 7352 
 7353     adcxq(tmp4, tmp);
 7354     adoxq(tmp4, yz_idx2);
 7355 
 7356     movl(carry, 0); // does not affect flags
 7357     adcxq(carry2, carry);
 7358     adoxq(carry2, carry);
 7359   } else {
 7360     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 7361     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 7362   }
 7363   movq(carry, carry2);
 7364 
 7365   movl(Address(z, idx, Address::times_4, 12), tmp3);
 7366   shrq(tmp3, 32);
 7367   movl(Address(z, idx, Address::times_4,  8), tmp3);
 7368 
 7369   movl(Address(z, idx, Address::times_4,  4), tmp4);
 7370   shrq(tmp4, 32);
 7371   movl(Address(z, idx, Address::times_4,  0), tmp4);
 7372 
 7373   jmp(L_third_loop);
 7374 
 7375   bind (L_third_loop_exit);
 7376 
 7377   andl (idx, 0x3);
 7378   jcc(Assembler::zero, L_post_third_loop_done);
 7379 
 7380   Label L_check_1;
 7381   subl(idx, 2);
 7382   jcc(Assembler::negative, L_check_1);
 7383 
 7384   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 7385   rorxq(yz_idx1, yz_idx1, 32);
 7386   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 7387   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7388   rorxq(yz_idx2, yz_idx2, 32);
 7389 
 7390   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 7391 
 7392   movl(Address(z, idx, Address::times_4,  4), tmp3);
 7393   shrq(tmp3, 32);
 7394   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7395   movq(carry, tmp4);
 7396 
 7397   bind (L_check_1);
 7398   addl (idx, 0x2);
 7399   andl (idx, 0x1);
 7400   subl(idx, 1);
 7401   jcc(Assembler::negative, L_post_third_loop_done);
 7402   movl(tmp4, Address(y, idx, Address::times_4,  0));
 7403   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 7404   movl(tmp4, Address(z, idx, Address::times_4,  0));
 7405 
 7406   add2_with_carry(carry2, tmp3, tmp4, carry);
 7407 
 7408   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7409   shrq(tmp3, 32);
 7410 
 7411   shlq(carry2, 32);
 7412   orq(tmp3, carry2);
 7413   movq(carry, tmp3);
 7414 
 7415   bind(L_post_third_loop_done);
 7416 }
 7417 
 7418 /**
 7419  * Code for BigInteger::multiplyToLen() intrinsic.
 7420  *
 7421  * rdi: x
 7422  * rax: xlen
 7423  * rsi: y
 7424  * rcx: ylen
 7425  * r8:  z
 7426  * r11: tmp0
 7427  * r12: tmp1
 7428  * r13: tmp2
 7429  * r14: tmp3
 7430  * r15: tmp4
 7431  * rbx: tmp5
 7432  *
 7433  */
 7434 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 7435                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 7436   ShortBranchVerifier sbv(this);
 7437   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 7438 
 7439   push(tmp0);
 7440   push(tmp1);
 7441   push(tmp2);
 7442   push(tmp3);
 7443   push(tmp4);
 7444   push(tmp5);
 7445 
 7446   push(xlen);
 7447 
 7448   const Register idx = tmp1;
 7449   const Register kdx = tmp2;
 7450   const Register xstart = tmp3;
 7451 
 7452   const Register y_idx = tmp4;
 7453   const Register carry = tmp5;
 7454   const Register product  = xlen;
 7455   const Register x_xstart = tmp0;
 7456 
 7457   // First Loop.
 7458   //
 7459   //  final static long LONG_MASK = 0xffffffffL;
 7460   //  int xstart = xlen - 1;
 7461   //  int ystart = ylen - 1;
 7462   //  long carry = 0;
 7463   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7464   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 7465   //    z[kdx] = (int)product;
 7466   //    carry = product >>> 32;
 7467   //  }
 7468   //  z[xstart] = (int)carry;
 7469   //
 7470 
 7471   movl(idx, ylen);               // idx = ylen;
 7472   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 7473   xorq(carry, carry);            // carry = 0;
 7474 
 7475   Label L_done;
 7476 
 7477   movl(xstart, xlen);
 7478   decrementl(xstart);
 7479   jcc(Assembler::negative, L_done);
 7480 
 7481   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 7482 
 7483   Label L_second_loop;
 7484   testl(kdx, kdx);
 7485   jcc(Assembler::zero, L_second_loop);
 7486 
 7487   Label L_carry;
 7488   subl(kdx, 1);
 7489   jcc(Assembler::zero, L_carry);
 7490 
 7491   movl(Address(z, kdx, Address::times_4,  0), carry);
 7492   shrq(carry, 32);
 7493   subl(kdx, 1);
 7494 
 7495   bind(L_carry);
 7496   movl(Address(z, kdx, Address::times_4,  0), carry);
 7497 
 7498   // Second and third (nested) loops.
 7499   //
 7500   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 7501   //   carry = 0;
 7502   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 7503   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 7504   //                    (z[k] & LONG_MASK) + carry;
 7505   //     z[k] = (int)product;
 7506   //     carry = product >>> 32;
 7507   //   }
 7508   //   z[i] = (int)carry;
 7509   // }
 7510   //
 7511   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 7512 
 7513   const Register jdx = tmp1;
 7514 
 7515   bind(L_second_loop);
 7516   xorl(carry, carry);    // carry = 0;
 7517   movl(jdx, ylen);       // j = ystart+1
 7518 
 7519   subl(xstart, 1);       // i = xstart-1;
 7520   jcc(Assembler::negative, L_done);
 7521 
 7522   push (z);
 7523 
 7524   Label L_last_x;
 7525   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 7526   subl(xstart, 1);       // i = xstart-1;
 7527   jcc(Assembler::negative, L_last_x);
 7528 
 7529   if (UseBMI2Instructions) {
 7530     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 7531     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 7532   } else {
 7533     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7534     rorq(x_xstart, 32);  // convert big-endian to little-endian
 7535   }
 7536 
 7537   Label L_third_loop_prologue;
 7538   bind(L_third_loop_prologue);
 7539 
 7540   push (x);
 7541   push (xstart);
 7542   push (ylen);
 7543 
 7544 
 7545   if (UseBMI2Instructions) {
 7546     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 7547   } else { // !UseBMI2Instructions
 7548     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 7549   }
 7550 
 7551   pop(ylen);
 7552   pop(xlen);
 7553   pop(x);
 7554   pop(z);
 7555 
 7556   movl(tmp3, xlen);
 7557   addl(tmp3, 1);
 7558   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7559   subl(tmp3, 1);
 7560   jccb(Assembler::negative, L_done);
 7561 
 7562   shrq(carry, 32);
 7563   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7564   jmp(L_second_loop);
 7565 
 7566   // Next infrequent code is moved outside loops.
 7567   bind(L_last_x);
 7568   if (UseBMI2Instructions) {
 7569     movl(rdx, Address(x,  0));
 7570   } else {
 7571     movl(x_xstart, Address(x,  0));
 7572   }
 7573   jmp(L_third_loop_prologue);
 7574 
 7575   bind(L_done);
 7576 
 7577   pop(xlen);
 7578 
 7579   pop(tmp5);
 7580   pop(tmp4);
 7581   pop(tmp3);
 7582   pop(tmp2);
 7583   pop(tmp1);
 7584   pop(tmp0);
 7585 }
 7586 
 7587 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 7588   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 7589   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 7590   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 7591   Label VECTOR8_TAIL, VECTOR4_TAIL;
 7592   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 7593   Label SAME_TILL_END, DONE;
 7594   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 7595 
 7596   //scale is in rcx in both Win64 and Unix
 7597   ShortBranchVerifier sbv(this);
 7598 
 7599   shlq(length);
 7600   xorq(result, result);
 7601 
 7602   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 7603       VM_Version::supports_avx512vlbw() && UseCountTrailingZerosInstruction) {
 7604     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 7605 
 7606     cmpq(length, 64);
 7607     jcc(Assembler::less, VECTOR32_TAIL);
 7608 
 7609     movq(tmp1, length);
 7610     andq(tmp1, 0x3F);      // tail count
 7611     andq(length, ~(0x3F)); //vector count
 7612 
 7613     bind(VECTOR64_LOOP);
 7614     // AVX512 code to compare 64 byte vectors.
 7615     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 7616     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7617     kortestql(k7, k7);
 7618     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 7619     addq(result, 64);
 7620     subq(length, 64);
 7621     jccb(Assembler::notZero, VECTOR64_LOOP);
 7622 
 7623     //bind(VECTOR64_TAIL);
 7624     testq(tmp1, tmp1);
 7625     jcc(Assembler::zero, SAME_TILL_END);
 7626 
 7627     //bind(VECTOR64_TAIL);
 7628     // AVX512 code to compare up to 63 byte vectors.
 7629     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 7630     shlxq(tmp2, tmp2, tmp1);
 7631     notq(tmp2);
 7632     kmovql(k3, tmp2);
 7633 
 7634     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 7635     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7636 
 7637     ktestql(k7, k3);
 7638     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 7639 
 7640     bind(VECTOR64_NOT_EQUAL);
 7641     kmovql(tmp1, k7);
 7642     notq(tmp1);
 7643     tzcntq(tmp1, tmp1);
 7644     addq(result, tmp1);
 7645     shrq(result);
 7646     jmp(DONE);
 7647     bind(VECTOR32_TAIL);
 7648   }
 7649 
 7650   cmpq(length, 8);
 7651   jcc(Assembler::equal, VECTOR8_LOOP);
 7652   jcc(Assembler::less, VECTOR4_TAIL);
 7653 
 7654   if (UseAVX >= 2) {
 7655     Label VECTOR16_TAIL, VECTOR32_LOOP;
 7656 
 7657     cmpq(length, 16);
 7658     jcc(Assembler::equal, VECTOR16_LOOP);
 7659     jcc(Assembler::less, VECTOR8_LOOP);
 7660 
 7661     cmpq(length, 32);
 7662     jccb(Assembler::less, VECTOR16_TAIL);
 7663 
 7664     subq(length, 32);
 7665     bind(VECTOR32_LOOP);
 7666     vmovdqu(rymm0, Address(obja, result));
 7667     vmovdqu(rymm1, Address(objb, result));
 7668     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 7669     vptest(rymm2, rymm2);
 7670     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7671     addq(result, 32);
 7672     subq(length, 32);
 7673     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7674     addq(length, 32);
 7675     jcc(Assembler::equal, SAME_TILL_END);
 7676     //falling through if less than 32 bytes left //close the branch here.
 7677 
 7678     bind(VECTOR16_TAIL);
 7679     cmpq(length, 16);
 7680     jccb(Assembler::less, VECTOR8_TAIL);
 7681     bind(VECTOR16_LOOP);
 7682     movdqu(rymm0, Address(obja, result));
 7683     movdqu(rymm1, Address(objb, result));
 7684     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7685     ptest(rymm2, rymm2);
 7686     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7687     addq(result, 16);
 7688     subq(length, 16);
 7689     jcc(Assembler::equal, SAME_TILL_END);
 7690     //falling through if less than 16 bytes left
 7691   } else {//regular intrinsics
 7692 
 7693     cmpq(length, 16);
 7694     jccb(Assembler::less, VECTOR8_TAIL);
 7695 
 7696     subq(length, 16);
 7697     bind(VECTOR16_LOOP);
 7698     movdqu(rymm0, Address(obja, result));
 7699     movdqu(rymm1, Address(objb, result));
 7700     pxor(rymm0, rymm1);
 7701     ptest(rymm0, rymm0);
 7702     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7703     addq(result, 16);
 7704     subq(length, 16);
 7705     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7706     addq(length, 16);
 7707     jcc(Assembler::equal, SAME_TILL_END);
 7708     //falling through if less than 16 bytes left
 7709   }
 7710 
 7711   bind(VECTOR8_TAIL);
 7712   cmpq(length, 8);
 7713   jccb(Assembler::less, VECTOR4_TAIL);
 7714   bind(VECTOR8_LOOP);
 7715   movq(tmp1, Address(obja, result));
 7716   movq(tmp2, Address(objb, result));
 7717   xorq(tmp1, tmp2);
 7718   testq(tmp1, tmp1);
 7719   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7720   addq(result, 8);
 7721   subq(length, 8);
 7722   jcc(Assembler::equal, SAME_TILL_END);
 7723   //falling through if less than 8 bytes left
 7724 
 7725   bind(VECTOR4_TAIL);
 7726   cmpq(length, 4);
 7727   jccb(Assembler::less, BYTES_TAIL);
 7728   bind(VECTOR4_LOOP);
 7729   movl(tmp1, Address(obja, result));
 7730   xorl(tmp1, Address(objb, result));
 7731   testl(tmp1, tmp1);
 7732   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7733   addq(result, 4);
 7734   subq(length, 4);
 7735   jcc(Assembler::equal, SAME_TILL_END);
 7736   //falling through if less than 4 bytes left
 7737 
 7738   bind(BYTES_TAIL);
 7739   bind(BYTES_LOOP);
 7740   load_unsigned_byte(tmp1, Address(obja, result));
 7741   load_unsigned_byte(tmp2, Address(objb, result));
 7742   xorl(tmp1, tmp2);
 7743   testl(tmp1, tmp1);
 7744   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7745   decq(length);
 7746   jcc(Assembler::zero, SAME_TILL_END);
 7747   incq(result);
 7748   load_unsigned_byte(tmp1, Address(obja, result));
 7749   load_unsigned_byte(tmp2, Address(objb, result));
 7750   xorl(tmp1, tmp2);
 7751   testl(tmp1, tmp1);
 7752   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7753   decq(length);
 7754   jcc(Assembler::zero, SAME_TILL_END);
 7755   incq(result);
 7756   load_unsigned_byte(tmp1, Address(obja, result));
 7757   load_unsigned_byte(tmp2, Address(objb, result));
 7758   xorl(tmp1, tmp2);
 7759   testl(tmp1, tmp1);
 7760   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7761   jmp(SAME_TILL_END);
 7762 
 7763   if (UseAVX >= 2) {
 7764     bind(VECTOR32_NOT_EQUAL);
 7765     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7766     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7767     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7768     vpmovmskb(tmp1, rymm0);
 7769     bsfq(tmp1, tmp1);
 7770     addq(result, tmp1);
 7771     shrq(result);
 7772     jmp(DONE);
 7773   }
 7774 
 7775   bind(VECTOR16_NOT_EQUAL);
 7776   if (UseAVX >= 2) {
 7777     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7778     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7779     pxor(rymm0, rymm2);
 7780   } else {
 7781     pcmpeqb(rymm2, rymm2);
 7782     pxor(rymm0, rymm1);
 7783     pcmpeqb(rymm0, rymm1);
 7784     pxor(rymm0, rymm2);
 7785   }
 7786   pmovmskb(tmp1, rymm0);
 7787   bsfq(tmp1, tmp1);
 7788   addq(result, tmp1);
 7789   shrq(result);
 7790   jmpb(DONE);
 7791 
 7792   bind(VECTOR8_NOT_EQUAL);
 7793   bind(VECTOR4_NOT_EQUAL);
 7794   bsfq(tmp1, tmp1);
 7795   shrq(tmp1, 3);
 7796   addq(result, tmp1);
 7797   bind(BYTES_NOT_EQUAL);
 7798   shrq(result);
 7799   jmpb(DONE);
 7800 
 7801   bind(SAME_TILL_END);
 7802   mov64(result, -1);
 7803 
 7804   bind(DONE);
 7805 }
 7806 
 7807 //Helper functions for square_to_len()
 7808 
 7809 /**
 7810  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7811  * Preserves x and z and modifies rest of the registers.
 7812  */
 7813 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7814   // Perform square and right shift by 1
 7815   // Handle odd xlen case first, then for even xlen do the following
 7816   // jlong carry = 0;
 7817   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7818   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7819   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7820   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7821   //     carry = (jlong)product;
 7822   // }
 7823 
 7824   xorq(tmp5, tmp5);     // carry
 7825   xorq(rdxReg, rdxReg);
 7826   xorl(tmp1, tmp1);     // index for x
 7827   xorl(tmp4, tmp4);     // index for z
 7828 
 7829   Label L_first_loop, L_first_loop_exit;
 7830 
 7831   testl(xlen, 1);
 7832   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7833 
 7834   // Square and right shift by 1 the odd element using 32 bit multiply
 7835   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7836   imulq(raxReg, raxReg);
 7837   shrq(raxReg, 1);
 7838   adcq(tmp5, 0);
 7839   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7840   incrementl(tmp1);
 7841   addl(tmp4, 2);
 7842 
 7843   // Square and  right shift by 1 the rest using 64 bit multiply
 7844   bind(L_first_loop);
 7845   cmpptr(tmp1, xlen);
 7846   jccb(Assembler::equal, L_first_loop_exit);
 7847 
 7848   // Square
 7849   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7850   rorq(raxReg, 32);    // convert big-endian to little-endian
 7851   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7852 
 7853   // Right shift by 1 and save carry
 7854   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7855   rcrq(rdxReg, 1);
 7856   rcrq(raxReg, 1);
 7857   adcq(tmp5, 0);
 7858 
 7859   // Store result in z
 7860   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7861   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7862 
 7863   // Update indices for x and z
 7864   addl(tmp1, 2);
 7865   addl(tmp4, 4);
 7866   jmp(L_first_loop);
 7867 
 7868   bind(L_first_loop_exit);
 7869 }
 7870 
 7871 
 7872 /**
 7873  * Perform the following multiply add operation using BMI2 instructions
 7874  * carry:sum = sum + op1*op2 + carry
 7875  * op2 should be in rdx
 7876  * op2 is preserved, all other registers are modified
 7877  */
 7878 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7879   // assert op2 is rdx
 7880   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7881   addq(sum, carry);
 7882   adcq(tmp2, 0);
 7883   addq(sum, op1);
 7884   adcq(tmp2, 0);
 7885   movq(carry, tmp2);
 7886 }
 7887 
 7888 /**
 7889  * Perform the following multiply add operation:
 7890  * carry:sum = sum + op1*op2 + carry
 7891  * Preserves op1, op2 and modifies rest of registers
 7892  */
 7893 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 7894   // rdx:rax = op1 * op2
 7895   movq(raxReg, op2);
 7896   mulq(op1);
 7897 
 7898   //  rdx:rax = sum + carry + rdx:rax
 7899   addq(sum, carry);
 7900   adcq(rdxReg, 0);
 7901   addq(sum, raxReg);
 7902   adcq(rdxReg, 0);
 7903 
 7904   // carry:sum = rdx:sum
 7905   movq(carry, rdxReg);
 7906 }
 7907 
 7908 /**
 7909  * Add 64 bit long carry into z[] with carry propagation.
 7910  * Preserves z and carry register values and modifies rest of registers.
 7911  *
 7912  */
 7913 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 7914   Label L_fourth_loop, L_fourth_loop_exit;
 7915 
 7916   movl(tmp1, 1);
 7917   subl(zlen, 2);
 7918   addq(Address(z, zlen, Address::times_4, 0), carry);
 7919 
 7920   bind(L_fourth_loop);
 7921   jccb(Assembler::carryClear, L_fourth_loop_exit);
 7922   subl(zlen, 2);
 7923   jccb(Assembler::negative, L_fourth_loop_exit);
 7924   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 7925   jmp(L_fourth_loop);
 7926   bind(L_fourth_loop_exit);
 7927 }
 7928 
 7929 /**
 7930  * Shift z[] left by 1 bit.
 7931  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 7932  *
 7933  */
 7934 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 7935 
 7936   Label L_fifth_loop, L_fifth_loop_exit;
 7937 
 7938   // Fifth loop
 7939   // Perform primitiveLeftShift(z, zlen, 1)
 7940 
 7941   const Register prev_carry = tmp1;
 7942   const Register new_carry = tmp4;
 7943   const Register value = tmp2;
 7944   const Register zidx = tmp3;
 7945 
 7946   // int zidx, carry;
 7947   // long value;
 7948   // carry = 0;
 7949   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 7950   //    (carry:value)  = (z[i] << 1) | carry ;
 7951   //    z[i] = value;
 7952   // }
 7953 
 7954   movl(zidx, zlen);
 7955   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 7956 
 7957   bind(L_fifth_loop);
 7958   decl(zidx);  // Use decl to preserve carry flag
 7959   decl(zidx);
 7960   jccb(Assembler::negative, L_fifth_loop_exit);
 7961 
 7962   if (UseBMI2Instructions) {
 7963      movq(value, Address(z, zidx, Address::times_4, 0));
 7964      rclq(value, 1);
 7965      rorxq(value, value, 32);
 7966      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7967   }
 7968   else {
 7969     // clear new_carry
 7970     xorl(new_carry, new_carry);
 7971 
 7972     // Shift z[i] by 1, or in previous carry and save new carry
 7973     movq(value, Address(z, zidx, Address::times_4, 0));
 7974     shlq(value, 1);
 7975     adcl(new_carry, 0);
 7976 
 7977     orq(value, prev_carry);
 7978     rorq(value, 0x20);
 7979     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 7980 
 7981     // Set previous carry = new carry
 7982     movl(prev_carry, new_carry);
 7983   }
 7984   jmp(L_fifth_loop);
 7985 
 7986   bind(L_fifth_loop_exit);
 7987 }
 7988 
 7989 
 7990 /**
 7991  * Code for BigInteger::squareToLen() intrinsic
 7992  *
 7993  * rdi: x
 7994  * rsi: len
 7995  * r8:  z
 7996  * rcx: zlen
 7997  * r12: tmp1
 7998  * r13: tmp2
 7999  * r14: tmp3
 8000  * r15: tmp4
 8001  * rbx: tmp5
 8002  *
 8003  */
 8004 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) {
 8005 
 8006   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 8007   push(tmp1);
 8008   push(tmp2);
 8009   push(tmp3);
 8010   push(tmp4);
 8011   push(tmp5);
 8012 
 8013   // First loop
 8014   // Store the squares, right shifted one bit (i.e., divided by 2).
 8015   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8016 
 8017   // Add in off-diagonal sums.
 8018   //
 8019   // Second, third (nested) and fourth loops.
 8020   // zlen +=2;
 8021   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 8022   //    carry = 0;
 8023   //    long op2 = x[xidx:xidx+1];
 8024   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 8025   //       k -= 2;
 8026   //       long op1 = x[j:j+1];
 8027   //       long sum = z[k:k+1];
 8028   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 8029   //       z[k:k+1] = sum;
 8030   //    }
 8031   //    add_one_64(z, k, carry, tmp_regs);
 8032   // }
 8033 
 8034   const Register carry = tmp5;
 8035   const Register sum = tmp3;
 8036   const Register op1 = tmp4;
 8037   Register op2 = tmp2;
 8038 
 8039   push(zlen);
 8040   push(len);
 8041   addl(zlen,2);
 8042   bind(L_second_loop);
 8043   xorq(carry, carry);
 8044   subl(zlen, 4);
 8045   subl(len, 2);
 8046   push(zlen);
 8047   push(len);
 8048   cmpl(len, 0);
 8049   jccb(Assembler::lessEqual, L_second_loop_exit);
 8050 
 8051   // Multiply an array by one 64 bit long.
 8052   if (UseBMI2Instructions) {
 8053     op2 = rdxReg;
 8054     movq(op2, Address(x, len, Address::times_4,  0));
 8055     rorxq(op2, op2, 32);
 8056   }
 8057   else {
 8058     movq(op2, Address(x, len, Address::times_4,  0));
 8059     rorq(op2, 32);
 8060   }
 8061 
 8062   bind(L_third_loop);
 8063   decrementl(len);
 8064   jccb(Assembler::negative, L_third_loop_exit);
 8065   decrementl(len);
 8066   jccb(Assembler::negative, L_last_x);
 8067 
 8068   movq(op1, Address(x, len, Address::times_4,  0));
 8069   rorq(op1, 32);
 8070 
 8071   bind(L_multiply);
 8072   subl(zlen, 2);
 8073   movq(sum, Address(z, zlen, Address::times_4,  0));
 8074 
 8075   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 8076   if (UseBMI2Instructions) {
 8077     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 8078   }
 8079   else {
 8080     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8081   }
 8082 
 8083   movq(Address(z, zlen, Address::times_4, 0), sum);
 8084 
 8085   jmp(L_third_loop);
 8086   bind(L_third_loop_exit);
 8087 
 8088   // Fourth loop
 8089   // Add 64 bit long carry into z with carry propagation.
 8090   // Uses offsetted zlen.
 8091   add_one_64(z, zlen, carry, tmp1);
 8092 
 8093   pop(len);
 8094   pop(zlen);
 8095   jmp(L_second_loop);
 8096 
 8097   // Next infrequent code is moved outside loops.
 8098   bind(L_last_x);
 8099   movl(op1, Address(x, 0));
 8100   jmp(L_multiply);
 8101 
 8102   bind(L_second_loop_exit);
 8103   pop(len);
 8104   pop(zlen);
 8105   pop(len);
 8106   pop(zlen);
 8107 
 8108   // Fifth loop
 8109   // Shift z left 1 bit.
 8110   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 8111 
 8112   // z[zlen-1] |= x[len-1] & 1;
 8113   movl(tmp3, Address(x, len, Address::times_4, -4));
 8114   andl(tmp3, 1);
 8115   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 8116 
 8117   pop(tmp5);
 8118   pop(tmp4);
 8119   pop(tmp3);
 8120   pop(tmp2);
 8121   pop(tmp1);
 8122 }
 8123 
 8124 /**
 8125  * Helper function for mul_add()
 8126  * Multiply the in[] by int k and add to out[] starting at offset offs using
 8127  * 128 bit by 32 bit multiply and return the carry in tmp5.
 8128  * Only quad int aligned length of in[] is operated on in this function.
 8129  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 8130  * This function preserves out, in and k registers.
 8131  * len and offset point to the appropriate index in "in" & "out" correspondingly
 8132  * tmp5 has the carry.
 8133  * other registers are temporary and are modified.
 8134  *
 8135  */
 8136 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 8137   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 8138   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8139 
 8140   Label L_first_loop, L_first_loop_exit;
 8141 
 8142   movl(tmp1, len);
 8143   shrl(tmp1, 2);
 8144 
 8145   bind(L_first_loop);
 8146   subl(tmp1, 1);
 8147   jccb(Assembler::negative, L_first_loop_exit);
 8148 
 8149   subl(len, 4);
 8150   subl(offset, 4);
 8151 
 8152   Register op2 = tmp2;
 8153   const Register sum = tmp3;
 8154   const Register op1 = tmp4;
 8155   const Register carry = tmp5;
 8156 
 8157   if (UseBMI2Instructions) {
 8158     op2 = rdxReg;
 8159   }
 8160 
 8161   movq(op1, Address(in, len, Address::times_4,  8));
 8162   rorq(op1, 32);
 8163   movq(sum, Address(out, offset, Address::times_4,  8));
 8164   rorq(sum, 32);
 8165   if (UseBMI2Instructions) {
 8166     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8167   }
 8168   else {
 8169     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8170   }
 8171   // Store back in big endian from little endian
 8172   rorq(sum, 0x20);
 8173   movq(Address(out, offset, Address::times_4,  8), sum);
 8174 
 8175   movq(op1, Address(in, len, Address::times_4,  0));
 8176   rorq(op1, 32);
 8177   movq(sum, Address(out, offset, Address::times_4,  0));
 8178   rorq(sum, 32);
 8179   if (UseBMI2Instructions) {
 8180     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8181   }
 8182   else {
 8183     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8184   }
 8185   // Store back in big endian from little endian
 8186   rorq(sum, 0x20);
 8187   movq(Address(out, offset, Address::times_4,  0), sum);
 8188 
 8189   jmp(L_first_loop);
 8190   bind(L_first_loop_exit);
 8191 }
 8192 
 8193 /**
 8194  * Code for BigInteger::mulAdd() intrinsic
 8195  *
 8196  * rdi: out
 8197  * rsi: in
 8198  * r11: offs (out.length - offset)
 8199  * rcx: len
 8200  * r8:  k
 8201  * r12: tmp1
 8202  * r13: tmp2
 8203  * r14: tmp3
 8204  * r15: tmp4
 8205  * rbx: tmp5
 8206  * Multiply the in[] by word k and add to out[], return the carry in rax
 8207  */
 8208 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 8209    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 8210    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8211 
 8212   Label L_carry, L_last_in, L_done;
 8213 
 8214 // carry = 0;
 8215 // for (int j=len-1; j >= 0; j--) {
 8216 //    long product = (in[j] & LONG_MASK) * kLong +
 8217 //                   (out[offs] & LONG_MASK) + carry;
 8218 //    out[offs--] = (int)product;
 8219 //    carry = product >>> 32;
 8220 // }
 8221 //
 8222   push(tmp1);
 8223   push(tmp2);
 8224   push(tmp3);
 8225   push(tmp4);
 8226   push(tmp5);
 8227 
 8228   Register op2 = tmp2;
 8229   const Register sum = tmp3;
 8230   const Register op1 = tmp4;
 8231   const Register carry =  tmp5;
 8232 
 8233   if (UseBMI2Instructions) {
 8234     op2 = rdxReg;
 8235     movl(op2, k);
 8236   }
 8237   else {
 8238     movl(op2, k);
 8239   }
 8240 
 8241   xorq(carry, carry);
 8242 
 8243   //First loop
 8244 
 8245   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 8246   //The carry is in tmp5
 8247   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8248 
 8249   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 8250   decrementl(len);
 8251   jccb(Assembler::negative, L_carry);
 8252   decrementl(len);
 8253   jccb(Assembler::negative, L_last_in);
 8254 
 8255   movq(op1, Address(in, len, Address::times_4,  0));
 8256   rorq(op1, 32);
 8257 
 8258   subl(offs, 2);
 8259   movq(sum, Address(out, offs, Address::times_4,  0));
 8260   rorq(sum, 32);
 8261 
 8262   if (UseBMI2Instructions) {
 8263     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8264   }
 8265   else {
 8266     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8267   }
 8268 
 8269   // Store back in big endian from little endian
 8270   rorq(sum, 0x20);
 8271   movq(Address(out, offs, Address::times_4,  0), sum);
 8272 
 8273   testl(len, len);
 8274   jccb(Assembler::zero, L_carry);
 8275 
 8276   //Multiply the last in[] entry, if any
 8277   bind(L_last_in);
 8278   movl(op1, Address(in, 0));
 8279   movl(sum, Address(out, offs, Address::times_4,  -4));
 8280 
 8281   movl(raxReg, k);
 8282   mull(op1); //tmp4 * eax -> edx:eax
 8283   addl(sum, carry);
 8284   adcl(rdxReg, 0);
 8285   addl(sum, raxReg);
 8286   adcl(rdxReg, 0);
 8287   movl(carry, rdxReg);
 8288 
 8289   movl(Address(out, offs, Address::times_4,  -4), sum);
 8290 
 8291   bind(L_carry);
 8292   //return tmp5/carry as carry in rax
 8293   movl(rax, carry);
 8294 
 8295   bind(L_done);
 8296   pop(tmp5);
 8297   pop(tmp4);
 8298   pop(tmp3);
 8299   pop(tmp2);
 8300   pop(tmp1);
 8301 }
 8302 
 8303 /**
 8304  * Emits code to update CRC-32 with a byte value according to constants in table
 8305  *
 8306  * @param [in,out]crc   Register containing the crc.
 8307  * @param [in]val       Register containing the byte to fold into the CRC.
 8308  * @param [in]table     Register containing the table of crc constants.
 8309  *
 8310  * uint32_t crc;
 8311  * val = crc_table[(val ^ crc) & 0xFF];
 8312  * crc = val ^ (crc >> 8);
 8313  *
 8314  */
 8315 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 8316   xorl(val, crc);
 8317   andl(val, 0xFF);
 8318   shrl(crc, 8); // unsigned shift
 8319   xorl(crc, Address(table, val, Address::times_4, 0));
 8320 }
 8321 
 8322 /**
 8323  * Fold 128-bit data chunk
 8324  */
 8325 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 8326   if (UseAVX > 0) {
 8327     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 8328     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 8329     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 8330     pxor(xcrc, xtmp);
 8331   } else {
 8332     movdqa(xtmp, xcrc);
 8333     pclmulhdq(xtmp, xK);   // [123:64]
 8334     pclmulldq(xcrc, xK);   // [63:0]
 8335     pxor(xcrc, xtmp);
 8336     movdqu(xtmp, Address(buf, offset));
 8337     pxor(xcrc, xtmp);
 8338   }
 8339 }
 8340 
 8341 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 8342   if (UseAVX > 0) {
 8343     vpclmulhdq(xtmp, xK, xcrc);
 8344     vpclmulldq(xcrc, xK, xcrc);
 8345     pxor(xcrc, xbuf);
 8346     pxor(xcrc, xtmp);
 8347   } else {
 8348     movdqa(xtmp, xcrc);
 8349     pclmulhdq(xtmp, xK);
 8350     pclmulldq(xcrc, xK);
 8351     pxor(xcrc, xbuf);
 8352     pxor(xcrc, xtmp);
 8353   }
 8354 }
 8355 
 8356 /**
 8357  * 8-bit folds to compute 32-bit CRC
 8358  *
 8359  * uint64_t xcrc;
 8360  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 8361  */
 8362 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 8363   movdl(tmp, xcrc);
 8364   andl(tmp, 0xFF);
 8365   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 8366   psrldq(xcrc, 1); // unsigned shift one byte
 8367   pxor(xcrc, xtmp);
 8368 }
 8369 
 8370 /**
 8371  * uint32_t crc;
 8372  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 8373  */
 8374 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 8375   movl(tmp, crc);
 8376   andl(tmp, 0xFF);
 8377   shrl(crc, 8);
 8378   xorl(crc, Address(table, tmp, Address::times_4, 0));
 8379 }
 8380 
 8381 /**
 8382  * @param crc   register containing existing CRC (32-bit)
 8383  * @param buf   register pointing to input byte buffer (byte*)
 8384  * @param len   register containing number of bytes
 8385  * @param table register that will contain address of CRC table
 8386  * @param tmp   scratch register
 8387  */
 8388 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 8389   assert_different_registers(crc, buf, len, table, tmp, rax);
 8390 
 8391   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8392   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8393 
 8394   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8395   // context for the registers used, where all instructions below are using 128-bit mode
 8396   // On EVEX without VL and BW, these instructions will all be AVX.
 8397   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 8398   notl(crc); // ~crc
 8399   cmpl(len, 16);
 8400   jcc(Assembler::less, L_tail);
 8401 
 8402   // Align buffer to 16 bytes
 8403   movl(tmp, buf);
 8404   andl(tmp, 0xF);
 8405   jccb(Assembler::zero, L_aligned);
 8406   subl(tmp,  16);
 8407   addl(len, tmp);
 8408 
 8409   align(4);
 8410   BIND(L_align_loop);
 8411   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8412   update_byte_crc32(crc, rax, table);
 8413   increment(buf);
 8414   incrementl(tmp);
 8415   jccb(Assembler::less, L_align_loop);
 8416 
 8417   BIND(L_aligned);
 8418   movl(tmp, len); // save
 8419   shrl(len, 4);
 8420   jcc(Assembler::zero, L_tail_restore);
 8421 
 8422   // Fold crc into first bytes of vector
 8423   movdqa(xmm1, Address(buf, 0));
 8424   movdl(rax, xmm1);
 8425   xorl(crc, rax);
 8426   if (VM_Version::supports_sse4_1()) {
 8427     pinsrd(xmm1, crc, 0);
 8428   } else {
 8429     pinsrw(xmm1, crc, 0);
 8430     shrl(crc, 16);
 8431     pinsrw(xmm1, crc, 1);
 8432   }
 8433   addptr(buf, 16);
 8434   subl(len, 4); // len > 0
 8435   jcc(Assembler::less, L_fold_tail);
 8436 
 8437   movdqa(xmm2, Address(buf,  0));
 8438   movdqa(xmm3, Address(buf, 16));
 8439   movdqa(xmm4, Address(buf, 32));
 8440   addptr(buf, 48);
 8441   subl(len, 3);
 8442   jcc(Assembler::lessEqual, L_fold_512b);
 8443 
 8444   // Fold total 512 bits of polynomial on each iteration,
 8445   // 128 bits per each of 4 parallel streams.
 8446   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 8447 
 8448   align32();
 8449   BIND(L_fold_512b_loop);
 8450   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8451   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 8452   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 8453   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 8454   addptr(buf, 64);
 8455   subl(len, 4);
 8456   jcc(Assembler::greater, L_fold_512b_loop);
 8457 
 8458   // Fold 512 bits to 128 bits.
 8459   BIND(L_fold_512b);
 8460   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8461   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 8462   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 8463   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 8464 
 8465   // Fold the rest of 128 bits data chunks
 8466   BIND(L_fold_tail);
 8467   addl(len, 3);
 8468   jccb(Assembler::lessEqual, L_fold_128b);
 8469   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8470 
 8471   BIND(L_fold_tail_loop);
 8472   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8473   addptr(buf, 16);
 8474   decrementl(len);
 8475   jccb(Assembler::greater, L_fold_tail_loop);
 8476 
 8477   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 8478   BIND(L_fold_128b);
 8479   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 8480   if (UseAVX > 0) {
 8481     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 8482     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 8483     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 8484   } else {
 8485     movdqa(xmm2, xmm0);
 8486     pclmulqdq(xmm2, xmm1, 0x1);
 8487     movdqa(xmm3, xmm0);
 8488     pand(xmm3, xmm2);
 8489     pclmulqdq(xmm0, xmm3, 0x1);
 8490   }
 8491   psrldq(xmm1, 8);
 8492   psrldq(xmm2, 4);
 8493   pxor(xmm0, xmm1);
 8494   pxor(xmm0, xmm2);
 8495 
 8496   // 8 8-bit folds to compute 32-bit CRC.
 8497   for (int j = 0; j < 4; j++) {
 8498     fold_8bit_crc32(xmm0, table, xmm1, rax);
 8499   }
 8500   movdl(crc, xmm0); // mov 32 bits to general register
 8501   for (int j = 0; j < 4; j++) {
 8502     fold_8bit_crc32(crc, table, rax);
 8503   }
 8504 
 8505   BIND(L_tail_restore);
 8506   movl(len, tmp); // restore
 8507   BIND(L_tail);
 8508   andl(len, 0xf);
 8509   jccb(Assembler::zero, L_exit);
 8510 
 8511   // Fold the rest of bytes
 8512   align(4);
 8513   BIND(L_tail_loop);
 8514   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8515   update_byte_crc32(crc, rax, table);
 8516   increment(buf);
 8517   decrementl(len);
 8518   jccb(Assembler::greater, L_tail_loop);
 8519 
 8520   BIND(L_exit);
 8521   notl(crc); // ~c
 8522 }
 8523 
 8524 // Helper function for AVX 512 CRC32
 8525 // Fold 512-bit data chunks
 8526 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 8527                                              Register pos, int offset) {
 8528   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 8529   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 8530   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 8531   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 8532   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 8533 }
 8534 
 8535 // Helper function for AVX 512 CRC32
 8536 // Compute CRC32 for < 256B buffers
 8537 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 8538                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 8539                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 8540 
 8541   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 8542   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 8543   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 8544 
 8545   // check if there is enough buffer to be able to fold 16B at a time
 8546   cmpl(len, 32);
 8547   jcc(Assembler::less, L_less_than_32);
 8548 
 8549   // if there is, load the constants
 8550   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 8551   movdl(xmm0, crc);                        // get the initial crc value
 8552   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8553   pxor(xmm7, xmm0);
 8554 
 8555   // update the buffer pointer
 8556   addl(pos, 16);
 8557   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 8558   subl(len, 32);
 8559   jmp(L_16B_reduction_loop);
 8560 
 8561   bind(L_less_than_32);
 8562   //mov initial crc to the return value. this is necessary for zero - length buffers.
 8563   movl(rax, crc);
 8564   testl(len, len);
 8565   jcc(Assembler::equal, L_cleanup);
 8566 
 8567   movdl(xmm0, crc);                        //get the initial crc value
 8568 
 8569   cmpl(len, 16);
 8570   jcc(Assembler::equal, L_exact_16_left);
 8571   jcc(Assembler::less, L_less_than_16_left);
 8572 
 8573   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8574   pxor(xmm7, xmm0);                       //xor the initial crc value
 8575   addl(pos, 16);
 8576   subl(len, 16);
 8577   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 8578   jmp(L_get_last_two_xmms);
 8579 
 8580   bind(L_less_than_16_left);
 8581   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 8582   pxor(xmm1, xmm1);
 8583   movptr(tmp1, rsp);
 8584   movdqu(Address(tmp1, 0 * 16), xmm1);
 8585 
 8586   cmpl(len, 4);
 8587   jcc(Assembler::less, L_only_less_than_4);
 8588 
 8589   //backup the counter value
 8590   movl(tmp2, len);
 8591   cmpl(len, 8);
 8592   jcc(Assembler::less, L_less_than_8_left);
 8593 
 8594   //load 8 Bytes
 8595   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 8596   movq(Address(tmp1, 0 * 16), rax);
 8597   addptr(tmp1, 8);
 8598   subl(len, 8);
 8599   addl(pos, 8);
 8600 
 8601   bind(L_less_than_8_left);
 8602   cmpl(len, 4);
 8603   jcc(Assembler::less, L_less_than_4_left);
 8604 
 8605   //load 4 Bytes
 8606   movl(rax, Address(buf, pos, Address::times_1, 0));
 8607   movl(Address(tmp1, 0 * 16), rax);
 8608   addptr(tmp1, 4);
 8609   subl(len, 4);
 8610   addl(pos, 4);
 8611 
 8612   bind(L_less_than_4_left);
 8613   cmpl(len, 2);
 8614   jcc(Assembler::less, L_less_than_2_left);
 8615 
 8616   // load 2 Bytes
 8617   movw(rax, Address(buf, pos, Address::times_1, 0));
 8618   movl(Address(tmp1, 0 * 16), rax);
 8619   addptr(tmp1, 2);
 8620   subl(len, 2);
 8621   addl(pos, 2);
 8622 
 8623   bind(L_less_than_2_left);
 8624   cmpl(len, 1);
 8625   jcc(Assembler::less, L_zero_left);
 8626 
 8627   // load 1 Byte
 8628   movb(rax, Address(buf, pos, Address::times_1, 0));
 8629   movb(Address(tmp1, 0 * 16), rax);
 8630 
 8631   bind(L_zero_left);
 8632   movdqu(xmm7, Address(rsp, 0));
 8633   pxor(xmm7, xmm0);                       //xor the initial crc value
 8634 
 8635   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8636   movdqu(xmm0, Address(rax, tmp2));
 8637   pshufb(xmm7, xmm0);
 8638   jmp(L_128_done);
 8639 
 8640   bind(L_exact_16_left);
 8641   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 8642   pxor(xmm7, xmm0);                       //xor the initial crc value
 8643   jmp(L_128_done);
 8644 
 8645   bind(L_only_less_than_4);
 8646   cmpl(len, 3);
 8647   jcc(Assembler::less, L_only_less_than_3);
 8648 
 8649   // load 3 Bytes
 8650   movb(rax, Address(buf, pos, Address::times_1, 0));
 8651   movb(Address(tmp1, 0), rax);
 8652 
 8653   movb(rax, Address(buf, pos, Address::times_1, 1));
 8654   movb(Address(tmp1, 1), rax);
 8655 
 8656   movb(rax, Address(buf, pos, Address::times_1, 2));
 8657   movb(Address(tmp1, 2), rax);
 8658 
 8659   movdqu(xmm7, Address(rsp, 0));
 8660   pxor(xmm7, xmm0);                     //xor the initial crc value
 8661 
 8662   pslldq(xmm7, 0x5);
 8663   jmp(L_barrett);
 8664   bind(L_only_less_than_3);
 8665   cmpl(len, 2);
 8666   jcc(Assembler::less, L_only_less_than_2);
 8667 
 8668   // load 2 Bytes
 8669   movb(rax, Address(buf, pos, Address::times_1, 0));
 8670   movb(Address(tmp1, 0), rax);
 8671 
 8672   movb(rax, Address(buf, pos, Address::times_1, 1));
 8673   movb(Address(tmp1, 1), rax);
 8674 
 8675   movdqu(xmm7, Address(rsp, 0));
 8676   pxor(xmm7, xmm0);                     //xor the initial crc value
 8677 
 8678   pslldq(xmm7, 0x6);
 8679   jmp(L_barrett);
 8680 
 8681   bind(L_only_less_than_2);
 8682   //load 1 Byte
 8683   movb(rax, Address(buf, pos, Address::times_1, 0));
 8684   movb(Address(tmp1, 0), rax);
 8685 
 8686   movdqu(xmm7, Address(rsp, 0));
 8687   pxor(xmm7, xmm0);                     //xor the initial crc value
 8688 
 8689   pslldq(xmm7, 0x7);
 8690 }
 8691 
 8692 /**
 8693 * Compute CRC32 using AVX512 instructions
 8694 * param crc   register containing existing CRC (32-bit)
 8695 * param buf   register pointing to input byte buffer (byte*)
 8696 * param len   register containing number of bytes
 8697 * param table address of crc or crc32c table
 8698 * param tmp1  scratch register
 8699 * param tmp2  scratch register
 8700 * return rax  result register
 8701 *
 8702 * This routine is identical for crc32c with the exception of the precomputed constant
 8703 * table which will be passed as the table argument.  The calculation steps are
 8704 * the same for both variants.
 8705 */
 8706 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8707   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8708 
 8709   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8710   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8711   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8712   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8713   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8714 
 8715   const Register pos = r12;
 8716   push(r12);
 8717   subptr(rsp, 16 * 2 + 8);
 8718 
 8719   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8720   // context for the registers used, where all instructions below are using 128-bit mode
 8721   // On EVEX without VL and BW, these instructions will all be AVX.
 8722   movl(pos, 0);
 8723 
 8724   // check if smaller than 256B
 8725   cmpl(len, 256);
 8726   jcc(Assembler::less, L_less_than_256);
 8727 
 8728   // load the initial crc value
 8729   movdl(xmm10, crc);
 8730 
 8731   // receive the initial 64B data, xor the initial crc value
 8732   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8733   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8734   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8735   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8736 
 8737   subl(len, 256);
 8738   cmpl(len, 256);
 8739   jcc(Assembler::less, L_fold_128_B_loop);
 8740 
 8741   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8742   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8743   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8744   subl(len, 256);
 8745 
 8746   bind(L_fold_256_B_loop);
 8747   addl(pos, 256);
 8748   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8749   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8750   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8751   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8752 
 8753   subl(len, 256);
 8754   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8755 
 8756   // Fold 256 into 128
 8757   addl(pos, 256);
 8758   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8759   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8760   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8761 
 8762   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8763   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8764   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8765 
 8766   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8767   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8768 
 8769   addl(len, 128);
 8770   jmp(L_fold_128_B_register);
 8771 
 8772   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8773   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8774 
 8775   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8776   bind(L_fold_128_B_loop);
 8777   addl(pos, 128);
 8778   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8779   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8780 
 8781   subl(len, 128);
 8782   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8783 
 8784   addl(pos, 128);
 8785 
 8786   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8787   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8788   bind(L_fold_128_B_register);
 8789   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8790   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8791   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8792   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8793   // save last that has no multiplicand
 8794   vextracti64x2(xmm7, xmm4, 3);
 8795 
 8796   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8797   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8798   // Needed later in reduction loop
 8799   movdqu(xmm10, Address(table, 1 * 16));
 8800   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8801   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8802 
 8803   // Swap 1,0,3,2 - 01 00 11 10
 8804   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8805   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8806   vextracti128(xmm5, xmm8, 1);
 8807   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8808 
 8809   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8810   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8811   addl(len, 128 - 16);
 8812   jcc(Assembler::less, L_final_reduction_for_128);
 8813 
 8814   bind(L_16B_reduction_loop);
 8815   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8816   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8817   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8818   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8819   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8820   addl(pos, 16);
 8821   subl(len, 16);
 8822   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8823 
 8824   bind(L_final_reduction_for_128);
 8825   addl(len, 16);
 8826   jcc(Assembler::equal, L_128_done);
 8827 
 8828   bind(L_get_last_two_xmms);
 8829   movdqu(xmm2, xmm7);
 8830   addl(pos, len);
 8831   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8832   subl(pos, len);
 8833 
 8834   // get rid of the extra data that was loaded before
 8835   // load the shift constant
 8836   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8837   movdqu(xmm0, Address(rax, len));
 8838   addl(rax, len);
 8839 
 8840   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8841   //Change mask to 512
 8842   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8843   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8844 
 8845   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8846   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8847   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8848   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8849   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8850 
 8851   bind(L_128_done);
 8852   // compute crc of a 128-bit value
 8853   movdqu(xmm10, Address(table, 3 * 16));
 8854   movdqu(xmm0, xmm7);
 8855 
 8856   // 64b fold
 8857   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8858   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8859   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8860 
 8861   // 32b fold
 8862   movdqu(xmm0, xmm7);
 8863   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8864   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8865   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8866   jmp(L_barrett);
 8867 
 8868   bind(L_less_than_256);
 8869   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);
 8870 
 8871   //barrett reduction
 8872   bind(L_barrett);
 8873   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8874   movdqu(xmm1, xmm7);
 8875   movdqu(xmm2, xmm7);
 8876   movdqu(xmm10, Address(table, 4 * 16));
 8877 
 8878   pclmulqdq(xmm7, xmm10, 0x0);
 8879   pxor(xmm7, xmm2);
 8880   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8881   movdqu(xmm2, xmm7);
 8882   pclmulqdq(xmm7, xmm10, 0x10);
 8883   pxor(xmm7, xmm2);
 8884   pxor(xmm7, xmm1);
 8885   pextrd(crc, xmm7, 2);
 8886 
 8887   bind(L_cleanup);
 8888   addptr(rsp, 16 * 2 + 8);
 8889   pop(r12);
 8890 }
 8891 
 8892 // S. Gueron / Information Processing Letters 112 (2012) 184
 8893 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 8894 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 8895 // Output: the 64-bit carry-less product of B * CONST
 8896 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 8897                                      Register tmp1, Register tmp2, Register tmp3) {
 8898   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 8899   if (n > 0) {
 8900     addq(tmp3, n * 256 * 8);
 8901   }
 8902   //    Q1 = TABLEExt[n][B & 0xFF];
 8903   movl(tmp1, in);
 8904   andl(tmp1, 0x000000FF);
 8905   shll(tmp1, 3);
 8906   addq(tmp1, tmp3);
 8907   movq(tmp1, Address(tmp1, 0));
 8908 
 8909   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 8910   movl(tmp2, in);
 8911   shrl(tmp2, 8);
 8912   andl(tmp2, 0x000000FF);
 8913   shll(tmp2, 3);
 8914   addq(tmp2, tmp3);
 8915   movq(tmp2, Address(tmp2, 0));
 8916 
 8917   shlq(tmp2, 8);
 8918   xorq(tmp1, tmp2);
 8919 
 8920   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 8921   movl(tmp2, in);
 8922   shrl(tmp2, 16);
 8923   andl(tmp2, 0x000000FF);
 8924   shll(tmp2, 3);
 8925   addq(tmp2, tmp3);
 8926   movq(tmp2, Address(tmp2, 0));
 8927 
 8928   shlq(tmp2, 16);
 8929   xorq(tmp1, tmp2);
 8930 
 8931   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 8932   shrl(in, 24);
 8933   andl(in, 0x000000FF);
 8934   shll(in, 3);
 8935   addq(in, tmp3);
 8936   movq(in, Address(in, 0));
 8937 
 8938   shlq(in, 24);
 8939   xorq(in, tmp1);
 8940   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 8941 }
 8942 
 8943 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 8944                                       Register in_out,
 8945                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 8946                                       XMMRegister w_xtmp2,
 8947                                       Register tmp1,
 8948                                       Register n_tmp2, Register n_tmp3) {
 8949   if (is_pclmulqdq_supported) {
 8950     movdl(w_xtmp1, in_out); // modified blindly
 8951 
 8952     movl(tmp1, const_or_pre_comp_const_index);
 8953     movdl(w_xtmp2, tmp1);
 8954     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 8955 
 8956     movdq(in_out, w_xtmp1);
 8957   } else {
 8958     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 8959   }
 8960 }
 8961 
 8962 // Recombination Alternative 2: No bit-reflections
 8963 // T1 = (CRC_A * U1) << 1
 8964 // T2 = (CRC_B * U2) << 1
 8965 // C1 = T1 >> 32
 8966 // C2 = T2 >> 32
 8967 // T1 = T1 & 0xFFFFFFFF
 8968 // T2 = T2 & 0xFFFFFFFF
 8969 // T1 = CRC32(0, T1)
 8970 // T2 = CRC32(0, T2)
 8971 // C1 = C1 ^ T1
 8972 // C2 = C2 ^ T2
 8973 // CRC = C1 ^ C2 ^ CRC_C
 8974 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,
 8975                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 8976                                      Register tmp1, Register tmp2,
 8977                                      Register n_tmp3) {
 8978   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8979   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 8980   shlq(in_out, 1);
 8981   movl(tmp1, in_out);
 8982   shrq(in_out, 32);
 8983   xorl(tmp2, tmp2);
 8984   crc32(tmp2, tmp1, 4);
 8985   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 8986   shlq(in1, 1);
 8987   movl(tmp1, in1);
 8988   shrq(in1, 32);
 8989   xorl(tmp2, tmp2);
 8990   crc32(tmp2, tmp1, 4);
 8991   xorl(in1, tmp2);
 8992   xorl(in_out, in1);
 8993   xorl(in_out, in2);
 8994 }
 8995 
 8996 // Set N to predefined value
 8997 // Subtract from a length of a buffer
 8998 // execute in a loop:
 8999 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 9000 // for i = 1 to N do
 9001 //  CRC_A = CRC32(CRC_A, A[i])
 9002 //  CRC_B = CRC32(CRC_B, B[i])
 9003 //  CRC_C = CRC32(CRC_C, C[i])
 9004 // end for
 9005 // Recombine
 9006 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,
 9007                                        Register in_out1, Register in_out2, Register in_out3,
 9008                                        Register tmp1, Register tmp2, Register tmp3,
 9009                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9010                                        Register tmp4, Register tmp5,
 9011                                        Register n_tmp6) {
 9012   Label L_processPartitions;
 9013   Label L_processPartition;
 9014   Label L_exit;
 9015 
 9016   bind(L_processPartitions);
 9017   cmpl(in_out1, 3 * size);
 9018   jcc(Assembler::less, L_exit);
 9019     xorl(tmp1, tmp1);
 9020     xorl(tmp2, tmp2);
 9021     movq(tmp3, in_out2);
 9022     addq(tmp3, size);
 9023 
 9024     bind(L_processPartition);
 9025       crc32(in_out3, Address(in_out2, 0), 8);
 9026       crc32(tmp1, Address(in_out2, size), 8);
 9027       crc32(tmp2, Address(in_out2, size * 2), 8);
 9028       addq(in_out2, 8);
 9029       cmpq(in_out2, tmp3);
 9030       jcc(Assembler::less, L_processPartition);
 9031     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 9032             w_xtmp1, w_xtmp2, w_xtmp3,
 9033             tmp4, tmp5,
 9034             n_tmp6);
 9035     addq(in_out2, 2 * size);
 9036     subl(in_out1, 3 * size);
 9037     jmp(L_processPartitions);
 9038 
 9039   bind(L_exit);
 9040 }
 9041 
 9042 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 9043 // Input: A buffer I of L bytes.
 9044 // Output: the CRC32C value of the buffer.
 9045 // Notations:
 9046 // Write L = 24N + r, with N = floor (L/24).
 9047 // r = L mod 24 (0 <= r < 24).
 9048 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 9049 // N quadwords, and R consists of r bytes.
 9050 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 9051 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 9052 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 9053 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 9054 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 9055                                           Register tmp1, Register tmp2, Register tmp3,
 9056                                           Register tmp4, Register tmp5, Register tmp6,
 9057                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9058                                           bool is_pclmulqdq_supported) {
 9059   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 9060   Label L_wordByWord;
 9061   Label L_byteByByteProlog;
 9062   Label L_byteByByte;
 9063   Label L_exit;
 9064 
 9065   if (is_pclmulqdq_supported ) {
 9066     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 9067     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 9068 
 9069     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 9070     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 9071 
 9072     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 9073     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 9074     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 9075   } else {
 9076     const_or_pre_comp_const_index[0] = 1;
 9077     const_or_pre_comp_const_index[1] = 0;
 9078 
 9079     const_or_pre_comp_const_index[2] = 3;
 9080     const_or_pre_comp_const_index[3] = 2;
 9081 
 9082     const_or_pre_comp_const_index[4] = 5;
 9083     const_or_pre_comp_const_index[5] = 4;
 9084    }
 9085   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 9086                     in2, in1, in_out,
 9087                     tmp1, tmp2, tmp3,
 9088                     w_xtmp1, w_xtmp2, w_xtmp3,
 9089                     tmp4, tmp5,
 9090                     tmp6);
 9091   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 9092                     in2, in1, in_out,
 9093                     tmp1, tmp2, tmp3,
 9094                     w_xtmp1, w_xtmp2, w_xtmp3,
 9095                     tmp4, tmp5,
 9096                     tmp6);
 9097   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 9098                     in2, in1, in_out,
 9099                     tmp1, tmp2, tmp3,
 9100                     w_xtmp1, w_xtmp2, w_xtmp3,
 9101                     tmp4, tmp5,
 9102                     tmp6);
 9103   movl(tmp1, in2);
 9104   andl(tmp1, 0x00000007);
 9105   negl(tmp1);
 9106   addl(tmp1, in2);
 9107   addq(tmp1, in1);
 9108 
 9109   cmpq(in1, tmp1);
 9110   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 9111   align(16);
 9112   BIND(L_wordByWord);
 9113     crc32(in_out, Address(in1, 0), 8);
 9114     addq(in1, 8);
 9115     cmpq(in1, tmp1);
 9116     jcc(Assembler::less, L_wordByWord);
 9117 
 9118   BIND(L_byteByByteProlog);
 9119   andl(in2, 0x00000007);
 9120   movl(tmp2, 1);
 9121 
 9122   cmpl(tmp2, in2);
 9123   jccb(Assembler::greater, L_exit);
 9124   BIND(L_byteByByte);
 9125     crc32(in_out, Address(in1, 0), 1);
 9126     incq(in1);
 9127     incl(tmp2);
 9128     cmpl(tmp2, in2);
 9129     jcc(Assembler::lessEqual, L_byteByByte);
 9130 
 9131   BIND(L_exit);
 9132 }
 9133 #undef BIND
 9134 #undef BLOCK_COMMENT
 9135 
 9136 // Compress char[] array to byte[].
 9137 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 9138 // Return the array length if every element in array can be encoded,
 9139 // otherwise, the index of first non-latin1 (> 0xff) character.
 9140 //   @IntrinsicCandidate
 9141 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 9142 //     for (int i = 0; i < len; i++) {
 9143 //       char c = src[srcOff];
 9144 //       if (c > 0xff) {
 9145 //           return i;  // return index of non-latin1 char
 9146 //       }
 9147 //       dst[dstOff] = (byte)c;
 9148 //       srcOff++;
 9149 //       dstOff++;
 9150 //     }
 9151 //     return len;
 9152 //   }
 9153 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 9154   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 9155   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 9156   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 9157   Label copy_chars_loop, done, reset_sp, copy_tail;
 9158 
 9159   // rsi: src
 9160   // rdi: dst
 9161   // rdx: len
 9162   // rcx: tmp5
 9163   // rax: result
 9164 
 9165   // rsi holds start addr of source char[] to be compressed
 9166   // rdi holds start addr of destination byte[]
 9167   // rdx holds length
 9168 
 9169   assert(len != result, "");
 9170 
 9171   // save length for return
 9172   movl(result, len);
 9173 
 9174   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 9175     VM_Version::supports_avx512vlbw() &&
 9176     VM_Version::supports_bmi2()) {
 9177 
 9178     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 9179 
 9180     // alignment
 9181     Label post_alignment;
 9182 
 9183     // if length of the string is less than 32, handle it the old fashioned way
 9184     testl(len, -32);
 9185     jcc(Assembler::zero, below_threshold);
 9186 
 9187     // First check whether a character is compressible ( <= 0xFF).
 9188     // Create mask to test for Unicode chars inside zmm vector
 9189     movl(tmp5, 0x00FF);
 9190     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 9191 
 9192     testl(len, -64);
 9193     jccb(Assembler::zero, post_alignment);
 9194 
 9195     movl(tmp5, dst);
 9196     andl(tmp5, (32 - 1));
 9197     negl(tmp5);
 9198     andl(tmp5, (32 - 1));
 9199 
 9200     // bail out when there is nothing to be done
 9201     testl(tmp5, 0xFFFFFFFF);
 9202     jccb(Assembler::zero, post_alignment);
 9203 
 9204     // ~(~0 << len), where len is the # of remaining elements to process
 9205     movl(len, 0xFFFFFFFF);
 9206     shlxl(len, len, tmp5);
 9207     notl(len);
 9208     kmovdl(mask2, len);
 9209     movl(len, result);
 9210 
 9211     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9212     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9213     ktestd(mask1, mask2);
 9214     jcc(Assembler::carryClear, copy_tail);
 9215 
 9216     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9217 
 9218     addptr(src, tmp5);
 9219     addptr(src, tmp5);
 9220     addptr(dst, tmp5);
 9221     subl(len, tmp5);
 9222 
 9223     bind(post_alignment);
 9224     // end of alignment
 9225 
 9226     movl(tmp5, len);
 9227     andl(tmp5, (32 - 1));    // tail count (in chars)
 9228     andl(len, ~(32 - 1));    // vector count (in chars)
 9229     jccb(Assembler::zero, copy_loop_tail);
 9230 
 9231     lea(src, Address(src, len, Address::times_2));
 9232     lea(dst, Address(dst, len, Address::times_1));
 9233     negptr(len);
 9234 
 9235     bind(copy_32_loop);
 9236     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 9237     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 9238     kortestdl(mask1, mask1);
 9239     jccb(Assembler::carryClear, reset_for_copy_tail);
 9240 
 9241     // All elements in current processed chunk are valid candidates for
 9242     // compression. Write a truncated byte elements to the memory.
 9243     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 9244     addptr(len, 32);
 9245     jccb(Assembler::notZero, copy_32_loop);
 9246 
 9247     bind(copy_loop_tail);
 9248     // bail out when there is nothing to be done
 9249     testl(tmp5, 0xFFFFFFFF);
 9250     jcc(Assembler::zero, done);
 9251 
 9252     movl(len, tmp5);
 9253 
 9254     // ~(~0 << len), where len is the # of remaining elements to process
 9255     movl(tmp5, 0xFFFFFFFF);
 9256     shlxl(tmp5, tmp5, len);
 9257     notl(tmp5);
 9258 
 9259     kmovdl(mask2, tmp5);
 9260 
 9261     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9262     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9263     ktestd(mask1, mask2);
 9264     jcc(Assembler::carryClear, copy_tail);
 9265 
 9266     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9267     jmp(done);
 9268 
 9269     bind(reset_for_copy_tail);
 9270     lea(src, Address(src, tmp5, Address::times_2));
 9271     lea(dst, Address(dst, tmp5, Address::times_1));
 9272     subptr(len, tmp5);
 9273     jmp(copy_chars_loop);
 9274 
 9275     bind(below_threshold);
 9276   }
 9277 
 9278   if (UseSSE42Intrinsics) {
 9279     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 9280 
 9281     // vectored compression
 9282     testl(len, 0xfffffff8);
 9283     jcc(Assembler::zero, copy_tail);
 9284 
 9285     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 9286     movdl(tmp1Reg, tmp5);
 9287     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 9288 
 9289     andl(len, 0xfffffff0);
 9290     jccb(Assembler::zero, copy_16);
 9291 
 9292     // compress 16 chars per iter
 9293     pxor(tmp4Reg, tmp4Reg);
 9294 
 9295     lea(src, Address(src, len, Address::times_2));
 9296     lea(dst, Address(dst, len, Address::times_1));
 9297     negptr(len);
 9298 
 9299     bind(copy_32_loop);
 9300     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 9301     por(tmp4Reg, tmp2Reg);
 9302     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 9303     por(tmp4Reg, tmp3Reg);
 9304     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 9305     jccb(Assembler::notZero, reset_for_copy_tail);
 9306     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 9307     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 9308     addptr(len, 16);
 9309     jccb(Assembler::notZero, copy_32_loop);
 9310 
 9311     // compress next vector of 8 chars (if any)
 9312     bind(copy_16);
 9313     // len = 0
 9314     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 9315     jccb(Assembler::zero, copy_tail_sse);
 9316 
 9317     pxor(tmp3Reg, tmp3Reg);
 9318 
 9319     movdqu(tmp2Reg, Address(src, 0));
 9320     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 9321     jccb(Assembler::notZero, reset_for_copy_tail);
 9322     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 9323     movq(Address(dst, 0), tmp2Reg);
 9324     addptr(src, 16);
 9325     addptr(dst, 8);
 9326     jmpb(copy_tail_sse);
 9327 
 9328     bind(reset_for_copy_tail);
 9329     movl(tmp5, result);
 9330     andl(tmp5, 0x0000000f);
 9331     lea(src, Address(src, tmp5, Address::times_2));
 9332     lea(dst, Address(dst, tmp5, Address::times_1));
 9333     subptr(len, tmp5);
 9334     jmpb(copy_chars_loop);
 9335 
 9336     bind(copy_tail_sse);
 9337     movl(len, result);
 9338     andl(len, 0x00000007);    // tail count (in chars)
 9339   }
 9340   // compress 1 char per iter
 9341   bind(copy_tail);
 9342   testl(len, len);
 9343   jccb(Assembler::zero, done);
 9344   lea(src, Address(src, len, Address::times_2));
 9345   lea(dst, Address(dst, len, Address::times_1));
 9346   negptr(len);
 9347 
 9348   bind(copy_chars_loop);
 9349   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 9350   testl(tmp5, 0xff00);      // check if Unicode char
 9351   jccb(Assembler::notZero, reset_sp);
 9352   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 9353   increment(len);
 9354   jccb(Assembler::notZero, copy_chars_loop);
 9355 
 9356   // add len then return (len will be zero if compress succeeded, otherwise negative)
 9357   bind(reset_sp);
 9358   addl(result, len);
 9359 
 9360   bind(done);
 9361 }
 9362 
 9363 // Inflate byte[] array to char[].
 9364 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 9365 //   @IntrinsicCandidate
 9366 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 9367 //     for (int i = 0; i < len; i++) {
 9368 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 9369 //     }
 9370 //   }
 9371 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 9372   XMMRegister tmp1, Register tmp2, KRegister mask) {
 9373   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 9374   // rsi: src
 9375   // rdi: dst
 9376   // rdx: len
 9377   // rcx: tmp2
 9378 
 9379   // rsi holds start addr of source byte[] to be inflated
 9380   // rdi holds start addr of destination char[]
 9381   // rdx holds length
 9382   assert_different_registers(src, dst, len, tmp2);
 9383   movl(tmp2, len);
 9384   if ((UseAVX > 2) && // AVX512
 9385     VM_Version::supports_avx512vlbw() &&
 9386     VM_Version::supports_bmi2()) {
 9387 
 9388     Label copy_32_loop, copy_tail;
 9389     Register tmp3_aliased = len;
 9390 
 9391     // if length of the string is less than 16, handle it in an old fashioned way
 9392     testl(len, -16);
 9393     jcc(Assembler::zero, below_threshold);
 9394 
 9395     testl(len, -1 * AVX3Threshold);
 9396     jcc(Assembler::zero, avx3_threshold);
 9397 
 9398     // In order to use only one arithmetic operation for the main loop we use
 9399     // this pre-calculation
 9400     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 9401     andl(len, -32);     // vector count
 9402     jccb(Assembler::zero, copy_tail);
 9403 
 9404     lea(src, Address(src, len, Address::times_1));
 9405     lea(dst, Address(dst, len, Address::times_2));
 9406     negptr(len);
 9407 
 9408 
 9409     // inflate 32 chars per iter
 9410     bind(copy_32_loop);
 9411     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 9412     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 9413     addptr(len, 32);
 9414     jcc(Assembler::notZero, copy_32_loop);
 9415 
 9416     bind(copy_tail);
 9417     // bail out when there is nothing to be done
 9418     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 9419     jcc(Assembler::zero, done);
 9420 
 9421     // ~(~0 << length), where length is the # of remaining elements to process
 9422     movl(tmp3_aliased, -1);
 9423     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 9424     notl(tmp3_aliased);
 9425     kmovdl(mask, tmp3_aliased);
 9426     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 9427     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 9428 
 9429     jmp(done);
 9430     bind(avx3_threshold);
 9431   }
 9432   if (UseSSE42Intrinsics) {
 9433     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 9434 
 9435     if (UseAVX > 1) {
 9436       andl(tmp2, (16 - 1));
 9437       andl(len, -16);
 9438       jccb(Assembler::zero, copy_new_tail);
 9439     } else {
 9440       andl(tmp2, 0x00000007);   // tail count (in chars)
 9441       andl(len, 0xfffffff8);    // vector count (in chars)
 9442       jccb(Assembler::zero, copy_tail);
 9443     }
 9444 
 9445     // vectored inflation
 9446     lea(src, Address(src, len, Address::times_1));
 9447     lea(dst, Address(dst, len, Address::times_2));
 9448     negptr(len);
 9449 
 9450     if (UseAVX > 1) {
 9451       bind(copy_16_loop);
 9452       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 9453       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 9454       addptr(len, 16);
 9455       jcc(Assembler::notZero, copy_16_loop);
 9456 
 9457       bind(below_threshold);
 9458       bind(copy_new_tail);
 9459       movl(len, tmp2);
 9460       andl(tmp2, 0x00000007);
 9461       andl(len, 0xFFFFFFF8);
 9462       jccb(Assembler::zero, copy_tail);
 9463 
 9464       pmovzxbw(tmp1, Address(src, 0));
 9465       movdqu(Address(dst, 0), tmp1);
 9466       addptr(src, 8);
 9467       addptr(dst, 2 * 8);
 9468 
 9469       jmp(copy_tail, true);
 9470     }
 9471 
 9472     // inflate 8 chars per iter
 9473     bind(copy_8_loop);
 9474     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 9475     movdqu(Address(dst, len, Address::times_2), tmp1);
 9476     addptr(len, 8);
 9477     jcc(Assembler::notZero, copy_8_loop);
 9478 
 9479     bind(copy_tail);
 9480     movl(len, tmp2);
 9481 
 9482     cmpl(len, 4);
 9483     jccb(Assembler::less, copy_bytes);
 9484 
 9485     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 9486     pmovzxbw(tmp1, tmp1);
 9487     movq(Address(dst, 0), tmp1);
 9488     subptr(len, 4);
 9489     addptr(src, 4);
 9490     addptr(dst, 8);
 9491 
 9492     bind(copy_bytes);
 9493   } else {
 9494     bind(below_threshold);
 9495   }
 9496 
 9497   testl(len, len);
 9498   jccb(Assembler::zero, done);
 9499   lea(src, Address(src, len, Address::times_1));
 9500   lea(dst, Address(dst, len, Address::times_2));
 9501   negptr(len);
 9502 
 9503   // inflate 1 char per iter
 9504   bind(copy_chars_loop);
 9505   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 9506   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 9507   increment(len);
 9508   jcc(Assembler::notZero, copy_chars_loop);
 9509 
 9510   bind(done);
 9511 }
 9512 
 9513 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 9514   switch(type) {
 9515     case T_BYTE:
 9516     case T_BOOLEAN:
 9517       evmovdqub(dst, kmask, src, merge, vector_len);
 9518       break;
 9519     case T_CHAR:
 9520     case T_SHORT:
 9521       evmovdquw(dst, kmask, src, merge, vector_len);
 9522       break;
 9523     case T_INT:
 9524     case T_FLOAT:
 9525       evmovdqul(dst, kmask, src, merge, vector_len);
 9526       break;
 9527     case T_LONG:
 9528     case T_DOUBLE:
 9529       evmovdquq(dst, kmask, src, merge, vector_len);
 9530       break;
 9531     default:
 9532       fatal("Unexpected type argument %s", type2name(type));
 9533       break;
 9534   }
 9535 }
 9536 
 9537 
 9538 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 9539   switch(type) {
 9540     case T_BYTE:
 9541     case T_BOOLEAN:
 9542       evmovdqub(dst, kmask, src, merge, vector_len);
 9543       break;
 9544     case T_CHAR:
 9545     case T_SHORT:
 9546       evmovdquw(dst, kmask, src, merge, vector_len);
 9547       break;
 9548     case T_INT:
 9549     case T_FLOAT:
 9550       evmovdqul(dst, kmask, src, merge, vector_len);
 9551       break;
 9552     case T_LONG:
 9553     case T_DOUBLE:
 9554       evmovdquq(dst, kmask, src, merge, vector_len);
 9555       break;
 9556     default:
 9557       fatal("Unexpected type argument %s", type2name(type));
 9558       break;
 9559   }
 9560 }
 9561 
 9562 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 9563   switch(type) {
 9564     case T_BYTE:
 9565     case T_BOOLEAN:
 9566       evmovdqub(dst, kmask, src, merge, vector_len);
 9567       break;
 9568     case T_CHAR:
 9569     case T_SHORT:
 9570       evmovdquw(dst, kmask, src, merge, vector_len);
 9571       break;
 9572     case T_INT:
 9573     case T_FLOAT:
 9574       evmovdqul(dst, kmask, src, merge, vector_len);
 9575       break;
 9576     case T_LONG:
 9577     case T_DOUBLE:
 9578       evmovdquq(dst, kmask, src, merge, vector_len);
 9579       break;
 9580     default:
 9581       fatal("Unexpected type argument %s", type2name(type));
 9582       break;
 9583   }
 9584 }
 9585 
 9586 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 9587   switch(masklen) {
 9588     case 2:
 9589        knotbl(dst, src);
 9590        movl(rtmp, 3);
 9591        kmovbl(ktmp, rtmp);
 9592        kandbl(dst, ktmp, dst);
 9593        break;
 9594     case 4:
 9595        knotbl(dst, src);
 9596        movl(rtmp, 15);
 9597        kmovbl(ktmp, rtmp);
 9598        kandbl(dst, ktmp, dst);
 9599        break;
 9600     case 8:
 9601        knotbl(dst, src);
 9602        break;
 9603     case 16:
 9604        knotwl(dst, src);
 9605        break;
 9606     case 32:
 9607        knotdl(dst, src);
 9608        break;
 9609     case 64:
 9610        knotql(dst, src);
 9611        break;
 9612     default:
 9613       fatal("Unexpected vector length %d", masklen);
 9614       break;
 9615   }
 9616 }
 9617 
 9618 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9619   switch(type) {
 9620     case T_BOOLEAN:
 9621     case T_BYTE:
 9622        kandbl(dst, src1, src2);
 9623        break;
 9624     case T_CHAR:
 9625     case T_SHORT:
 9626        kandwl(dst, src1, src2);
 9627        break;
 9628     case T_INT:
 9629     case T_FLOAT:
 9630        kanddl(dst, src1, src2);
 9631        break;
 9632     case T_LONG:
 9633     case T_DOUBLE:
 9634        kandql(dst, src1, src2);
 9635        break;
 9636     default:
 9637       fatal("Unexpected type argument %s", type2name(type));
 9638       break;
 9639   }
 9640 }
 9641 
 9642 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9643   switch(type) {
 9644     case T_BOOLEAN:
 9645     case T_BYTE:
 9646        korbl(dst, src1, src2);
 9647        break;
 9648     case T_CHAR:
 9649     case T_SHORT:
 9650        korwl(dst, src1, src2);
 9651        break;
 9652     case T_INT:
 9653     case T_FLOAT:
 9654        kordl(dst, src1, src2);
 9655        break;
 9656     case T_LONG:
 9657     case T_DOUBLE:
 9658        korql(dst, src1, src2);
 9659        break;
 9660     default:
 9661       fatal("Unexpected type argument %s", type2name(type));
 9662       break;
 9663   }
 9664 }
 9665 
 9666 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9667   switch(type) {
 9668     case T_BOOLEAN:
 9669     case T_BYTE:
 9670        kxorbl(dst, src1, src2);
 9671        break;
 9672     case T_CHAR:
 9673     case T_SHORT:
 9674        kxorwl(dst, src1, src2);
 9675        break;
 9676     case T_INT:
 9677     case T_FLOAT:
 9678        kxordl(dst, src1, src2);
 9679        break;
 9680     case T_LONG:
 9681     case T_DOUBLE:
 9682        kxorql(dst, src1, src2);
 9683        break;
 9684     default:
 9685       fatal("Unexpected type argument %s", type2name(type));
 9686       break;
 9687   }
 9688 }
 9689 
 9690 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9691   switch(type) {
 9692     case T_BOOLEAN:
 9693     case T_BYTE:
 9694       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9695     case T_CHAR:
 9696     case T_SHORT:
 9697       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9698     case T_INT:
 9699     case T_FLOAT:
 9700       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9701     case T_LONG:
 9702     case T_DOUBLE:
 9703       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9704     default:
 9705       fatal("Unexpected type argument %s", type2name(type)); break;
 9706   }
 9707 }
 9708 
 9709 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9710   switch(type) {
 9711     case T_BOOLEAN:
 9712     case T_BYTE:
 9713       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9714     case T_CHAR:
 9715     case T_SHORT:
 9716       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9717     case T_INT:
 9718     case T_FLOAT:
 9719       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9720     case T_LONG:
 9721     case T_DOUBLE:
 9722       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9723     default:
 9724       fatal("Unexpected type argument %s", type2name(type)); break;
 9725   }
 9726 }
 9727 
 9728 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9729   switch(type) {
 9730     case T_BYTE:
 9731       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9732     case T_SHORT:
 9733       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9734     case T_INT:
 9735       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9736     case T_LONG:
 9737       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9738     default:
 9739       fatal("Unexpected type argument %s", type2name(type)); break;
 9740   }
 9741 }
 9742 
 9743 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9744   switch(type) {
 9745     case T_BYTE:
 9746       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9747     case T_SHORT:
 9748       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9749     case T_INT:
 9750       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9751     case T_LONG:
 9752       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9753     default:
 9754       fatal("Unexpected type argument %s", type2name(type)); break;
 9755   }
 9756 }
 9757 
 9758 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9759   switch(type) {
 9760     case T_BYTE:
 9761       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9762     case T_SHORT:
 9763       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9764     case T_INT:
 9765       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9766     case T_LONG:
 9767       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9768     default:
 9769       fatal("Unexpected type argument %s", type2name(type)); break;
 9770   }
 9771 }
 9772 
 9773 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9774   switch(type) {
 9775     case T_BYTE:
 9776       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9777     case T_SHORT:
 9778       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9779     case T_INT:
 9780       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9781     case T_LONG:
 9782       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9783     default:
 9784       fatal("Unexpected type argument %s", type2name(type)); break;
 9785   }
 9786 }
 9787 
 9788 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9789   switch(type) {
 9790     case T_BYTE:
 9791       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9792     case T_SHORT:
 9793       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9794     case T_INT:
 9795       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9796     case T_LONG:
 9797       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9798     case T_FLOAT:
 9799       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9800     case T_DOUBLE:
 9801       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9802     default:
 9803       fatal("Unexpected type argument %s", type2name(type)); break;
 9804   }
 9805 }
 9806 
 9807 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9808   switch(type) {
 9809     case T_BYTE:
 9810       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9811     case T_SHORT:
 9812       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9813     case T_INT:
 9814       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9815     case T_LONG:
 9816       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9817     case T_FLOAT:
 9818       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9819     case T_DOUBLE:
 9820       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9821     default:
 9822       fatal("Unexpected type argument %s", type2name(type)); break;
 9823   }
 9824 }
 9825 
 9826 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9827   switch(type) {
 9828     case T_BYTE:
 9829       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9830     case T_SHORT:
 9831       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9832     case T_INT:
 9833       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9834     case T_LONG:
 9835       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9836     case T_FLOAT:
 9837       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9838     case T_DOUBLE:
 9839       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9840     default:
 9841       fatal("Unexpected type argument %s", type2name(type)); break;
 9842   }
 9843 }
 9844 
 9845 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9846   switch(type) {
 9847     case T_BYTE:
 9848       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9849     case T_SHORT:
 9850       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9851     case T_INT:
 9852       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9853     case T_LONG:
 9854       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9855     case T_FLOAT:
 9856       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9857     case T_DOUBLE:
 9858       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9859     default:
 9860       fatal("Unexpected type argument %s", type2name(type)); break;
 9861   }
 9862 }
 9863 
 9864 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9865   switch(type) {
 9866     case T_INT:
 9867       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9868     case T_LONG:
 9869       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9870     default:
 9871       fatal("Unexpected type argument %s", type2name(type)); break;
 9872   }
 9873 }
 9874 
 9875 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9876   switch(type) {
 9877     case T_INT:
 9878       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9879     case T_LONG:
 9880       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9881     default:
 9882       fatal("Unexpected type argument %s", type2name(type)); break;
 9883   }
 9884 }
 9885 
 9886 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9887   switch(type) {
 9888     case T_INT:
 9889       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9890     case T_LONG:
 9891       evporq(dst, mask, nds, src, merge, vector_len); break;
 9892     default:
 9893       fatal("Unexpected type argument %s", type2name(type)); break;
 9894   }
 9895 }
 9896 
 9897 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9898   switch(type) {
 9899     case T_INT:
 9900       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9901     case T_LONG:
 9902       evporq(dst, mask, nds, src, merge, vector_len); break;
 9903     default:
 9904       fatal("Unexpected type argument %s", type2name(type)); break;
 9905   }
 9906 }
 9907 
 9908 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9909   switch(type) {
 9910     case T_INT:
 9911       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9912     case T_LONG:
 9913       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9914     default:
 9915       fatal("Unexpected type argument %s", type2name(type)); break;
 9916   }
 9917 }
 9918 
 9919 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9920   switch(type) {
 9921     case T_INT:
 9922       evpandd(dst, mask, nds, src, merge, vector_len); break;
 9923     case T_LONG:
 9924       evpandq(dst, mask, nds, src, merge, vector_len); break;
 9925     default:
 9926       fatal("Unexpected type argument %s", type2name(type)); break;
 9927   }
 9928 }
 9929 
 9930 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
 9931   switch(masklen) {
 9932     case 8:
 9933        kortestbl(src1, src2);
 9934        break;
 9935     case 16:
 9936        kortestwl(src1, src2);
 9937        break;
 9938     case 32:
 9939        kortestdl(src1, src2);
 9940        break;
 9941     case 64:
 9942        kortestql(src1, src2);
 9943        break;
 9944     default:
 9945       fatal("Unexpected mask length %d", masklen);
 9946       break;
 9947   }
 9948 }
 9949 
 9950 
 9951 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
 9952   switch(masklen)  {
 9953     case 8:
 9954        ktestbl(src1, src2);
 9955        break;
 9956     case 16:
 9957        ktestwl(src1, src2);
 9958        break;
 9959     case 32:
 9960        ktestdl(src1, src2);
 9961        break;
 9962     case 64:
 9963        ktestql(src1, src2);
 9964        break;
 9965     default:
 9966       fatal("Unexpected mask length %d", masklen);
 9967       break;
 9968   }
 9969 }
 9970 
 9971 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9972   switch(type) {
 9973     case T_INT:
 9974       evprold(dst, mask, src, shift, merge, vlen_enc); break;
 9975     case T_LONG:
 9976       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
 9977     default:
 9978       fatal("Unexpected type argument %s", type2name(type)); break;
 9979       break;
 9980   }
 9981 }
 9982 
 9983 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
 9984   switch(type) {
 9985     case T_INT:
 9986       evprord(dst, mask, src, shift, merge, vlen_enc); break;
 9987     case T_LONG:
 9988       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
 9989     default:
 9990       fatal("Unexpected type argument %s", type2name(type)); break;
 9991   }
 9992 }
 9993 
 9994 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
 9995   switch(type) {
 9996     case T_INT:
 9997       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
 9998     case T_LONG:
 9999       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
10000     default:
10001       fatal("Unexpected type argument %s", type2name(type)); break;
10002   }
10003 }
10004 
10005 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
10006   switch(type) {
10007     case T_INT:
10008       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
10009     case T_LONG:
10010       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
10011     default:
10012       fatal("Unexpected type argument %s", type2name(type)); break;
10013   }
10014 }
10015 
10016 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10017   assert(rscratch != noreg || always_reachable(src), "missing");
10018 
10019   if (reachable(src)) {
10020     evpandq(dst, nds, as_Address(src), vector_len);
10021   } else {
10022     lea(rscratch, src);
10023     evpandq(dst, nds, Address(rscratch, 0), vector_len);
10024   }
10025 }
10026 
10027 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
10028   assert(rscratch != noreg || always_reachable(src), "missing");
10029 
10030   if (reachable(src)) {
10031     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
10032   } else {
10033     lea(rscratch, src);
10034     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
10035   }
10036 }
10037 
10038 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10039   assert(rscratch != noreg || always_reachable(src), "missing");
10040 
10041   if (reachable(src)) {
10042     evporq(dst, nds, as_Address(src), vector_len);
10043   } else {
10044     lea(rscratch, src);
10045     evporq(dst, nds, Address(rscratch, 0), vector_len);
10046   }
10047 }
10048 
10049 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10050   assert(rscratch != noreg || always_reachable(src), "missing");
10051 
10052   if (reachable(src)) {
10053     vpshufb(dst, nds, as_Address(src), vector_len);
10054   } else {
10055     lea(rscratch, src);
10056     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
10057   }
10058 }
10059 
10060 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10061   assert(rscratch != noreg || always_reachable(src), "missing");
10062 
10063   if (reachable(src)) {
10064     Assembler::vpor(dst, nds, as_Address(src), vector_len);
10065   } else {
10066     lea(rscratch, src);
10067     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
10068   }
10069 }
10070 
10071 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
10072   assert(rscratch != noreg || always_reachable(src3), "missing");
10073 
10074   if (reachable(src3)) {
10075     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
10076   } else {
10077     lea(rscratch, src3);
10078     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
10079   }
10080 }
10081 
10082 #if COMPILER2_OR_JVMCI
10083 
10084 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
10085                                  Register length, Register temp, int vec_enc) {
10086   // Computing mask for predicated vector store.
10087   movptr(temp, -1);
10088   bzhiq(temp, temp, length);
10089   kmov(mask, temp);
10090   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
10091 }
10092 
10093 // Set memory operation for length "less than" 64 bytes.
10094 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
10095                                        XMMRegister xmm, KRegister mask, Register length,
10096                                        Register temp, bool use64byteVector) {
10097   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10098   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10099   if (!use64byteVector) {
10100     fill32(dst, disp, xmm);
10101     subptr(length, 32 >> shift);
10102     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
10103   } else {
10104     assert(MaxVectorSize == 64, "vector length != 64");
10105     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
10106   }
10107 }
10108 
10109 
10110 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
10111                                        XMMRegister xmm, KRegister mask, Register length,
10112                                        Register temp) {
10113   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10114   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10115   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
10116 }
10117 
10118 
10119 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
10120   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10121   vmovdqu(dst, xmm);
10122 }
10123 
10124 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
10125   fill32(Address(dst, disp), xmm);
10126 }
10127 
10128 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
10129   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10130   if (!use64byteVector) {
10131     fill32(dst, xmm);
10132     fill32(dst.plus_disp(32), xmm);
10133   } else {
10134     evmovdquq(dst, xmm, Assembler::AVX_512bit);
10135   }
10136 }
10137 
10138 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
10139   fill64(Address(dst, disp), xmm, use64byteVector);
10140 }
10141 
10142 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
10143                                         Register count, Register rtmp, XMMRegister xtmp) {
10144   Label L_exit;
10145   Label L_fill_start;
10146   Label L_fill_64_bytes;
10147   Label L_fill_96_bytes;
10148   Label L_fill_128_bytes;
10149   Label L_fill_128_bytes_loop;
10150   Label L_fill_128_loop_header;
10151   Label L_fill_128_bytes_loop_header;
10152   Label L_fill_128_bytes_loop_pre_header;
10153   Label L_fill_zmm_sequence;
10154 
10155   int shift = -1;
10156   switch(type) {
10157     case T_BYTE:  shift = 0;
10158       break;
10159     case T_SHORT: shift = 1;
10160       break;
10161     case T_INT:   shift = 2;
10162       break;
10163     /* Uncomment when LONG fill stubs are supported.
10164     case T_LONG:  shift = 3;
10165       break;
10166     */
10167     default:
10168       fatal("Unhandled type: %s\n", type2name(type));
10169   }
10170 
10171   if ((CopyAVX3Threshold != 0)  || (MaxVectorSize == 32)) {
10172 
10173     if (MaxVectorSize == 64) {
10174       cmpq(count, CopyAVX3Threshold >> shift);
10175       jcc(Assembler::greater, L_fill_zmm_sequence);
10176     }
10177 
10178     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
10179 
10180     bind(L_fill_start);
10181 
10182     cmpq(count, 32 >> shift);
10183     jccb(Assembler::greater, L_fill_64_bytes);
10184     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
10185     jmp(L_exit);
10186 
10187     bind(L_fill_64_bytes);
10188     cmpq(count, 64 >> shift);
10189     jccb(Assembler::greater, L_fill_96_bytes);
10190     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
10191     jmp(L_exit);
10192 
10193     bind(L_fill_96_bytes);
10194     cmpq(count, 96 >> shift);
10195     jccb(Assembler::greater, L_fill_128_bytes);
10196     fill64(to, 0, xtmp);
10197     subq(count, 64 >> shift);
10198     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
10199     jmp(L_exit);
10200 
10201     bind(L_fill_128_bytes);
10202     cmpq(count, 128 >> shift);
10203     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
10204     fill64(to, 0, xtmp);
10205     fill32(to, 64, xtmp);
10206     subq(count, 96 >> shift);
10207     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
10208     jmp(L_exit);
10209 
10210     bind(L_fill_128_bytes_loop_pre_header);
10211     {
10212       mov(rtmp, to);
10213       andq(rtmp, 31);
10214       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
10215       negq(rtmp);
10216       addq(rtmp, 32);
10217       mov64(r8, -1L);
10218       bzhiq(r8, r8, rtmp);
10219       kmovql(k2, r8);
10220       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
10221       addq(to, rtmp);
10222       shrq(rtmp, shift);
10223       subq(count, rtmp);
10224     }
10225 
10226     cmpq(count, 128 >> shift);
10227     jcc(Assembler::less, L_fill_start);
10228 
10229     bind(L_fill_128_bytes_loop_header);
10230     subq(count, 128 >> shift);
10231 
10232     align32();
10233     bind(L_fill_128_bytes_loop);
10234       fill64(to, 0, xtmp);
10235       fill64(to, 64, xtmp);
10236       addq(to, 128);
10237       subq(count, 128 >> shift);
10238       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
10239 
10240     addq(count, 128 >> shift);
10241     jcc(Assembler::zero, L_exit);
10242     jmp(L_fill_start);
10243   }
10244 
10245   if (MaxVectorSize == 64) {
10246     // Sequence using 64 byte ZMM register.
10247     Label L_fill_128_bytes_zmm;
10248     Label L_fill_192_bytes_zmm;
10249     Label L_fill_192_bytes_loop_zmm;
10250     Label L_fill_192_bytes_loop_header_zmm;
10251     Label L_fill_192_bytes_loop_pre_header_zmm;
10252     Label L_fill_start_zmm_sequence;
10253 
10254     bind(L_fill_zmm_sequence);
10255     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
10256 
10257     bind(L_fill_start_zmm_sequence);
10258     cmpq(count, 64 >> shift);
10259     jccb(Assembler::greater, L_fill_128_bytes_zmm);
10260     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
10261     jmp(L_exit);
10262 
10263     bind(L_fill_128_bytes_zmm);
10264     cmpq(count, 128 >> shift);
10265     jccb(Assembler::greater, L_fill_192_bytes_zmm);
10266     fill64(to, 0, xtmp, true);
10267     subq(count, 64 >> shift);
10268     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
10269     jmp(L_exit);
10270 
10271     bind(L_fill_192_bytes_zmm);
10272     cmpq(count, 192 >> shift);
10273     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
10274     fill64(to, 0, xtmp, true);
10275     fill64(to, 64, xtmp, true);
10276     subq(count, 128 >> shift);
10277     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
10278     jmp(L_exit);
10279 
10280     bind(L_fill_192_bytes_loop_pre_header_zmm);
10281     {
10282       movq(rtmp, to);
10283       andq(rtmp, 63);
10284       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
10285       negq(rtmp);
10286       addq(rtmp, 64);
10287       mov64(r8, -1L);
10288       bzhiq(r8, r8, rtmp);
10289       kmovql(k2, r8);
10290       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
10291       addq(to, rtmp);
10292       shrq(rtmp, shift);
10293       subq(count, rtmp);
10294     }
10295 
10296     cmpq(count, 192 >> shift);
10297     jcc(Assembler::less, L_fill_start_zmm_sequence);
10298 
10299     bind(L_fill_192_bytes_loop_header_zmm);
10300     subq(count, 192 >> shift);
10301 
10302     align32();
10303     bind(L_fill_192_bytes_loop_zmm);
10304       fill64(to, 0, xtmp, true);
10305       fill64(to, 64, xtmp, true);
10306       fill64(to, 128, xtmp, true);
10307       addq(to, 192);
10308       subq(count, 192 >> shift);
10309       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
10310 
10311     addq(count, 192 >> shift);
10312     jcc(Assembler::zero, L_exit);
10313     jmp(L_fill_start_zmm_sequence);
10314   }
10315   bind(L_exit);
10316 }
10317 #endif //COMPILER2_OR_JVMCI
10318 
10319 
10320 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
10321   Label done;
10322   cvttss2sil(dst, src);
10323   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10324   cmpl(dst, 0x80000000); // float_sign_flip
10325   jccb(Assembler::notEqual, done);
10326   subptr(rsp, 8);
10327   movflt(Address(rsp, 0), src);
10328   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
10329   pop(dst);
10330   bind(done);
10331 }
10332 
10333 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
10334   Label done;
10335   cvttsd2sil(dst, src);
10336   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10337   cmpl(dst, 0x80000000); // float_sign_flip
10338   jccb(Assembler::notEqual, done);
10339   subptr(rsp, 8);
10340   movdbl(Address(rsp, 0), src);
10341   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
10342   pop(dst);
10343   bind(done);
10344 }
10345 
10346 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
10347   Label done;
10348   cvttss2siq(dst, src);
10349   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10350   jccb(Assembler::notEqual, done);
10351   subptr(rsp, 8);
10352   movflt(Address(rsp, 0), src);
10353   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
10354   pop(dst);
10355   bind(done);
10356 }
10357 
10358 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10359   // Following code is line by line assembly translation rounding algorithm.
10360   // Please refer to java.lang.Math.round(float) algorithm for details.
10361   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
10362   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
10363   const int32_t FloatConsts_EXP_BIAS = 127;
10364   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
10365   const int32_t MINUS_32 = 0xFFFFFFE0;
10366   Label L_special_case, L_block1, L_exit;
10367   movl(rtmp, FloatConsts_EXP_BIT_MASK);
10368   movdl(dst, src);
10369   andl(dst, rtmp);
10370   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
10371   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
10372   subl(rtmp, dst);
10373   movl(rcx, rtmp);
10374   movl(dst, MINUS_32);
10375   testl(rtmp, dst);
10376   jccb(Assembler::notEqual, L_special_case);
10377   movdl(dst, src);
10378   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
10379   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
10380   movdl(rtmp, src);
10381   testl(rtmp, rtmp);
10382   jccb(Assembler::greaterEqual, L_block1);
10383   negl(dst);
10384   bind(L_block1);
10385   sarl(dst);
10386   addl(dst, 0x1);
10387   sarl(dst, 0x1);
10388   jmp(L_exit);
10389   bind(L_special_case);
10390   convert_f2i(dst, src);
10391   bind(L_exit);
10392 }
10393 
10394 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10395   // Following code is line by line assembly translation rounding algorithm.
10396   // Please refer to java.lang.Math.round(double) algorithm for details.
10397   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
10398   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
10399   const int64_t DoubleConsts_EXP_BIAS = 1023;
10400   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
10401   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
10402   Label L_special_case, L_block1, L_exit;
10403   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
10404   movq(dst, src);
10405   andq(dst, rtmp);
10406   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
10407   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
10408   subq(rtmp, dst);
10409   movq(rcx, rtmp);
10410   mov64(dst, MINUS_64);
10411   testq(rtmp, dst);
10412   jccb(Assembler::notEqual, L_special_case);
10413   movq(dst, src);
10414   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
10415   andq(dst, rtmp);
10416   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
10417   orq(dst, rtmp);
10418   movq(rtmp, src);
10419   testq(rtmp, rtmp);
10420   jccb(Assembler::greaterEqual, L_block1);
10421   negq(dst);
10422   bind(L_block1);
10423   sarq(dst);
10424   addq(dst, 0x1);
10425   sarq(dst, 0x1);
10426   jmp(L_exit);
10427   bind(L_special_case);
10428   convert_d2l(dst, src);
10429   bind(L_exit);
10430 }
10431 
10432 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
10433   Label done;
10434   cvttsd2siq(dst, src);
10435   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10436   jccb(Assembler::notEqual, done);
10437   subptr(rsp, 8);
10438   movdbl(Address(rsp, 0), src);
10439   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
10440   pop(dst);
10441   bind(done);
10442 }
10443 
10444 void MacroAssembler::cache_wb(Address line)
10445 {
10446   // 64 bit cpus always support clflush
10447   bool optimized = VM_Version::supports_clflushopt();
10448   bool no_evict = VM_Version::supports_clwb();
10449 
10450   // prefer clwb (writeback without evict) otherwise
10451   // prefer clflushopt (potentially parallel writeback with evict)
10452   // otherwise fallback on clflush (serial writeback with evict)
10453 
10454   if (optimized) {
10455     if (no_evict) {
10456       clwb(line);
10457     } else {
10458       clflushopt(line);
10459     }
10460   } else {
10461     // no need for fence when using CLFLUSH
10462     clflush(line);
10463   }
10464 }
10465 
10466 void MacroAssembler::cache_wbsync(bool is_pre)
10467 {
10468   bool optimized = VM_Version::supports_clflushopt();
10469   bool no_evict = VM_Version::supports_clwb();
10470 
10471   // pick the correct implementation
10472 
10473   if (!is_pre && (optimized || no_evict)) {
10474     // need an sfence for post flush when using clflushopt or clwb
10475     // otherwise no no need for any synchroniaztion
10476 
10477     sfence();
10478   }
10479 }
10480 
10481 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
10482   switch (cond) {
10483     // Note some conditions are synonyms for others
10484     case Assembler::zero:         return Assembler::notZero;
10485     case Assembler::notZero:      return Assembler::zero;
10486     case Assembler::less:         return Assembler::greaterEqual;
10487     case Assembler::lessEqual:    return Assembler::greater;
10488     case Assembler::greater:      return Assembler::lessEqual;
10489     case Assembler::greaterEqual: return Assembler::less;
10490     case Assembler::below:        return Assembler::aboveEqual;
10491     case Assembler::belowEqual:   return Assembler::above;
10492     case Assembler::above:        return Assembler::belowEqual;
10493     case Assembler::aboveEqual:   return Assembler::below;
10494     case Assembler::overflow:     return Assembler::noOverflow;
10495     case Assembler::noOverflow:   return Assembler::overflow;
10496     case Assembler::negative:     return Assembler::positive;
10497     case Assembler::positive:     return Assembler::negative;
10498     case Assembler::parity:       return Assembler::noParity;
10499     case Assembler::noParity:     return Assembler::parity;
10500   }
10501   ShouldNotReachHere(); return Assembler::overflow;
10502 }
10503 
10504 // This is simply a call to Thread::current()
10505 void MacroAssembler::get_thread_slow(Register thread) {
10506   if (thread != rax) {
10507     push(rax);
10508   }
10509   push(rdi);
10510   push(rsi);
10511   push(rdx);
10512   push(rcx);
10513   push(r8);
10514   push(r9);
10515   push(r10);
10516   push(r11);
10517 
10518   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
10519 
10520   pop(r11);
10521   pop(r10);
10522   pop(r9);
10523   pop(r8);
10524   pop(rcx);
10525   pop(rdx);
10526   pop(rsi);
10527   pop(rdi);
10528   if (thread != rax) {
10529     mov(thread, rax);
10530     pop(rax);
10531   }
10532 }
10533 
10534 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
10535   Label L_stack_ok;
10536   if (bias == 0) {
10537     testptr(sp, 2 * wordSize - 1);
10538   } else {
10539     // lea(tmp, Address(rsp, bias);
10540     mov(tmp, sp);
10541     addptr(tmp, bias);
10542     testptr(tmp, 2 * wordSize - 1);
10543   }
10544   jcc(Assembler::equal, L_stack_ok);
10545   block_comment(msg);
10546   stop(msg);
10547   bind(L_stack_ok);
10548 }
10549 
10550 // Implements fast-locking.
10551 //
10552 // obj: the object to be locked
10553 // reg_rax: rax
10554 // thread: the thread which attempts to lock obj
10555 // tmp: a temporary register
10556 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
10557   Register thread = r15_thread;
10558 
10559   assert(reg_rax == rax, "");
10560   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
10561 
10562   Label push;
10563   const Register top = tmp;
10564 
10565   // Preload the markWord. It is important that this is the first
10566   // instruction emitted as it is part of C1's null check semantics.
10567   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10568 
10569   if (UseObjectMonitorTable) {
10570     // Clear cache in case fast locking succeeds or we need to take the slow-path.
10571     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
10572   }
10573 
10574   if (DiagnoseSyncOnValueBasedClasses != 0) {
10575     load_klass(tmp, obj, rscratch1);
10576     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
10577     jcc(Assembler::notZero, slow);
10578   }
10579 
10580   // Load top.
10581   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10582 
10583   // Check if the lock-stack is full.
10584   cmpl(top, LockStack::end_offset());
10585   jcc(Assembler::greaterEqual, slow);
10586 
10587   // Check for recursion.
10588   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10589   jcc(Assembler::equal, push);
10590 
10591   // Check header for monitor (0b10).
10592   testptr(reg_rax, markWord::monitor_value);
10593   jcc(Assembler::notZero, slow);
10594 
10595   // Try to lock. Transition lock bits 0b01 => 0b00
10596   movptr(tmp, reg_rax);
10597   andptr(tmp, ~(int32_t)markWord::unlocked_value);
10598   orptr(reg_rax, markWord::unlocked_value);
10599   // Mask inline_type bit such that we go to the slow path if object is an inline type
10600   andptr(reg_rax, ~((int) markWord::inline_type_bit_in_place));
10601 
10602   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10603   jcc(Assembler::notEqual, slow);
10604 
10605   // Restore top, CAS clobbers register.
10606   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10607 
10608   bind(push);
10609   // After successful lock, push object on lock-stack.
10610   movptr(Address(thread, top), obj);
10611   incrementl(top, oopSize);
10612   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
10613 }
10614 
10615 // Implements fast-unlocking.
10616 //
10617 // obj: the object to be unlocked
10618 // reg_rax: rax
10619 // thread: the thread
10620 // tmp: a temporary register
10621 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
10622   Register thread = r15_thread;
10623 
10624   assert(reg_rax == rax, "");
10625   assert_different_registers(obj, reg_rax, thread, tmp);
10626 
10627   Label unlocked, push_and_slow;
10628   const Register top = tmp;
10629 
10630   // Check if obj is top of lock-stack.
10631   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10632   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10633   jcc(Assembler::notEqual, slow);
10634 
10635   // Pop lock-stack.
10636   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
10637   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10638 
10639   // Check if recursive.
10640   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
10641   jcc(Assembler::equal, unlocked);
10642 
10643   // Not recursive. Check header for monitor (0b10).
10644   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10645   testptr(reg_rax, markWord::monitor_value);
10646   jcc(Assembler::notZero, push_and_slow);
10647 
10648 #ifdef ASSERT
10649   // Check header not unlocked (0b01).
10650   Label not_unlocked;
10651   testptr(reg_rax, markWord::unlocked_value);
10652   jcc(Assembler::zero, not_unlocked);
10653   stop("fast_unlock already unlocked");
10654   bind(not_unlocked);
10655 #endif
10656 
10657   // Try to unlock. Transition lock bits 0b00 => 0b01
10658   movptr(tmp, reg_rax);
10659   orptr(tmp, markWord::unlocked_value);
10660   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10661   jcc(Assembler::equal, unlocked);
10662 
10663   bind(push_and_slow);
10664   // Restore lock-stack and handle the unlock in runtime.
10665 #ifdef ASSERT
10666   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10667   movptr(Address(thread, top), obj);
10668 #endif
10669   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10670   jmp(slow);
10671 
10672   bind(unlocked);
10673 }
10674 
10675 // Saves legacy GPRs state on stack.
10676 void MacroAssembler::save_legacy_gprs() {
10677   subq(rsp, 16 * wordSize);
10678   movq(Address(rsp, 15 * wordSize), rax);
10679   movq(Address(rsp, 14 * wordSize), rcx);
10680   movq(Address(rsp, 13 * wordSize), rdx);
10681   movq(Address(rsp, 12 * wordSize), rbx);
10682   movq(Address(rsp, 10 * wordSize), rbp);
10683   movq(Address(rsp, 9 * wordSize), rsi);
10684   movq(Address(rsp, 8 * wordSize), rdi);
10685   movq(Address(rsp, 7 * wordSize), r8);
10686   movq(Address(rsp, 6 * wordSize), r9);
10687   movq(Address(rsp, 5 * wordSize), r10);
10688   movq(Address(rsp, 4 * wordSize), r11);
10689   movq(Address(rsp, 3 * wordSize), r12);
10690   movq(Address(rsp, 2 * wordSize), r13);
10691   movq(Address(rsp, wordSize), r14);
10692   movq(Address(rsp, 0), r15);
10693 }
10694 
10695 // Resotres back legacy GPRs state from stack.
10696 void MacroAssembler::restore_legacy_gprs() {
10697   movq(r15, Address(rsp, 0));
10698   movq(r14, Address(rsp, wordSize));
10699   movq(r13, Address(rsp, 2 * wordSize));
10700   movq(r12, Address(rsp, 3 * wordSize));
10701   movq(r11, Address(rsp, 4 * wordSize));
10702   movq(r10, Address(rsp, 5 * wordSize));
10703   movq(r9,  Address(rsp, 6 * wordSize));
10704   movq(r8,  Address(rsp, 7 * wordSize));
10705   movq(rdi, Address(rsp, 8 * wordSize));
10706   movq(rsi, Address(rsp, 9 * wordSize));
10707   movq(rbp, Address(rsp, 10 * wordSize));
10708   movq(rbx, Address(rsp, 12 * wordSize));
10709   movq(rdx, Address(rsp, 13 * wordSize));
10710   movq(rcx, Address(rsp, 14 * wordSize));
10711   movq(rax, Address(rsp, 15 * wordSize));
10712   addq(rsp, 16 * wordSize);
10713 }
10714 
10715 void MacroAssembler::load_aotrc_address(Register reg, address a) {
10716 #if INCLUDE_CDS
10717   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
10718   if (AOTCodeCache::is_on_for_dump()) {
10719     // all aotrc field addresses should be registered in the AOTCodeCache address table
10720     lea(reg, ExternalAddress(a));
10721   } else {
10722     mov64(reg, (uint64_t)a);
10723   }
10724 #else
10725   ShouldNotReachHere();
10726 #endif
10727 }
10728 
10729 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10730   if (VM_Version::supports_apx_f()) {
10731     esetzucc(comparison, dst);
10732   } else {
10733     setb(comparison, dst);
10734     movzbl(dst, dst);
10735   }
10736 }