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/macros.hpp"
   63 #include "vmreg_x86.inline.hpp"
   64 #ifdef COMPILER2
   65 #include "opto/output.hpp"
   66 #endif
   67 
   68 #ifdef PRODUCT
   69 #define BLOCK_COMMENT(str) /* nothing */
   70 #define STOP(error) stop(error)
   71 #else
   72 #define BLOCK_COMMENT(str) block_comment(str)
   73 #define STOP(error) block_comment(error); stop(error)
   74 #endif
   75 
   76 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   77 
   78 #ifdef ASSERT
   79 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   80 #endif
   81 
   82 static const Assembler::Condition reverse[] = {
   83     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   84     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   85     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   86     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,
   87     Assembler::notZero        /* zero          = 0x4, equal         = 0x4 */ ,
   88     Assembler::zero           /* notZero       = 0x5, notEqual      = 0x5 */ ,
   89     Assembler::above          /* belowEqual    = 0x6 */ ,
   90     Assembler::belowEqual     /* above         = 0x7 */ ,
   91     Assembler::positive       /* negative      = 0x8 */ ,
   92     Assembler::negative       /* positive      = 0x9 */ ,
   93     Assembler::noParity       /* parity        = 0xa */ ,
   94     Assembler::parity         /* noParity      = 0xb */ ,
   95     Assembler::greaterEqual   /* less          = 0xc */ ,
   96     Assembler::less           /* greaterEqual  = 0xd */ ,
   97     Assembler::greater        /* lessEqual     = 0xe */ ,
   98     Assembler::lessEqual      /* greater       = 0xf, */
   99 
  100 };
  101 
  102 
  103 // Implementation of MacroAssembler
  104 
  105 Address MacroAssembler::as_Address(AddressLiteral adr) {
  106   // amd64 always does this as a pc-rel
  107   // we can be absolute or disp based on the instruction type
  108   // jmp/call are displacements others are absolute
  109   assert(!adr.is_lval(), "must be rval");
  110   assert(reachable(adr), "must be");
  111   return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
  112 
  113 }
  114 
  115 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
  116   AddressLiteral base = adr.base();
  117   lea(rscratch, base);
  118   Address index = adr.index();
  119   assert(index._disp == 0, "must not have disp"); // maybe it can?
  120   Address array(rscratch, index._index, index._scale, index._disp);
  121   return array;
  122 }
  123 
  124 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
  125   Label L, E;
  126 
  127 #ifdef _WIN64
  128   // Windows always allocates space for it's register args
  129   assert(num_args <= 4, "only register arguments supported");
  130   subq(rsp,  frame::arg_reg_save_area_bytes);
  131 #endif
  132 
  133   // Align stack if necessary
  134   testl(rsp, 15);
  135   jcc(Assembler::zero, L);
  136 
  137   subq(rsp, 8);
  138   call(RuntimeAddress(entry_point));
  139   addq(rsp, 8);
  140   jmp(E);
  141 
  142   bind(L);
  143   call(RuntimeAddress(entry_point));
  144 
  145   bind(E);
  146 
  147 #ifdef _WIN64
  148   // restore stack pointer
  149   addq(rsp, frame::arg_reg_save_area_bytes);
  150 #endif
  151 }
  152 
  153 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
  154   assert(!src2.is_lval(), "should use cmpptr");
  155   assert(rscratch != noreg || always_reachable(src2), "missing");
  156 
  157   if (reachable(src2)) {
  158     cmpq(src1, as_Address(src2));
  159   } else {
  160     lea(rscratch, src2);
  161     Assembler::cmpq(src1, Address(rscratch, 0));
  162   }
  163 }
  164 
  165 int MacroAssembler::corrected_idivq(Register reg) {
  166   // Full implementation of Java ldiv and lrem; checks for special
  167   // case as described in JVM spec., p.243 & p.271.  The function
  168   // returns the (pc) offset of the idivl instruction - may be needed
  169   // for implicit exceptions.
  170   //
  171   //         normal case                           special case
  172   //
  173   // input : rax: dividend                         min_long
  174   //         reg: divisor   (may not be eax/edx)   -1
  175   //
  176   // output: rax: quotient  (= rax idiv reg)       min_long
  177   //         rdx: remainder (= rax irem reg)       0
  178   assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
  179   static const int64_t min_long = 0x8000000000000000;
  180   Label normal_case, special_case;
  181 
  182   // check for special case
  183   cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
  184   jcc(Assembler::notEqual, normal_case);
  185   xorl(rdx, rdx); // prepare rdx for possible special case (where
  186                   // remainder = 0)
  187   cmpq(reg, -1);
  188   jcc(Assembler::equal, special_case);
  189 
  190   // handle normal case
  191   bind(normal_case);
  192   cdqq();
  193   int idivq_offset = offset();
  194   idivq(reg);
  195 
  196   // normal and special case exit
  197   bind(special_case);
  198 
  199   return idivq_offset;
  200 }
  201 
  202 void MacroAssembler::decrementq(Register reg, int value) {
  203   if (value == min_jint) { subq(reg, value); return; }
  204   if (value <  0) { incrementq(reg, -value); return; }
  205   if (value == 0) {                        ; return; }
  206   if (value == 1 && UseIncDec) { decq(reg) ; return; }
  207   /* else */      { subq(reg, value)       ; return; }
  208 }
  209 
  210 void MacroAssembler::decrementq(Address dst, int value) {
  211   if (value == min_jint) { subq(dst, value); return; }
  212   if (value <  0) { incrementq(dst, -value); return; }
  213   if (value == 0) {                        ; return; }
  214   if (value == 1 && UseIncDec) { decq(dst) ; return; }
  215   /* else */      { subq(dst, value)       ; return; }
  216 }
  217 
  218 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
  219   assert(rscratch != noreg || always_reachable(dst), "missing");
  220 
  221   if (reachable(dst)) {
  222     incrementq(as_Address(dst));
  223   } else {
  224     lea(rscratch, dst);
  225     incrementq(Address(rscratch, 0));
  226   }
  227 }
  228 
  229 void MacroAssembler::incrementq(Register reg, int value) {
  230   if (value == min_jint) { addq(reg, value); return; }
  231   if (value <  0) { decrementq(reg, -value); return; }
  232   if (value == 0) {                        ; return; }
  233   if (value == 1 && UseIncDec) { incq(reg) ; return; }
  234   /* else */      { addq(reg, value)       ; return; }
  235 }
  236 
  237 void MacroAssembler::incrementq(Address dst, int value) {
  238   if (value == min_jint) { addq(dst, value); return; }
  239   if (value <  0) { decrementq(dst, -value); return; }
  240   if (value == 0) {                        ; return; }
  241   if (value == 1 && UseIncDec) { incq(dst) ; return; }
  242   /* else */      { addq(dst, value)       ; return; }
  243 }
  244 
  245 // 32bit can do a case table jump in one instruction but we no longer allow the base
  246 // to be installed in the Address class
  247 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
  248   lea(rscratch, entry.base());
  249   Address dispatch = entry.index();
  250   assert(dispatch._base == noreg, "must be");
  251   dispatch._base = rscratch;
  252   jmp(dispatch);
  253 }
  254 
  255 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
  256   ShouldNotReachHere(); // 64bit doesn't use two regs
  257   cmpq(x_lo, y_lo);
  258 }
  259 
  260 void MacroAssembler::lea(Register dst, AddressLiteral src) {
  261   mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  262 }
  263 
  264 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
  265   lea(rscratch, adr);
  266   movptr(dst, rscratch);
  267 }
  268 
  269 void MacroAssembler::leave() {
  270   // %%% is this really better? Why not on 32bit too?
  271   emit_int8((unsigned char)0xC9); // LEAVE
  272 }
  273 
  274 void MacroAssembler::lneg(Register hi, Register lo) {
  275   ShouldNotReachHere(); // 64bit doesn't use two regs
  276   negq(lo);
  277 }
  278 
  279 void MacroAssembler::movoop(Register dst, jobject obj) {
  280   mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  281 }
  282 
  283 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
  284   mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
  285   movq(dst, rscratch);
  286 }
  287 
  288 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
  289   mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  290 }
  291 
  292 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
  293   mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
  294   movq(dst, rscratch);
  295 }
  296 
  297 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
  298   if (src.is_lval()) {
  299     mov_literal64(dst, (intptr_t)src.target(), src.rspec());
  300   } else {
  301     if (reachable(src)) {
  302       movq(dst, as_Address(src));
  303     } else {
  304       lea(dst, src);
  305       movq(dst, Address(dst, 0));
  306     }
  307   }
  308 }
  309 
  310 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
  311   movq(as_Address(dst, rscratch), src);
  312 }
  313 
  314 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
  315   movq(dst, as_Address(src, dst /*rscratch*/));
  316 }
  317 
  318 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
  319 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
  320   if (is_simm32(src)) {
  321     movptr(dst, checked_cast<int32_t>(src));
  322   } else {
  323     mov64(rscratch, src);
  324     movq(dst, rscratch);
  325   }
  326 }
  327 
  328 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
  329   movoop(rscratch, obj);
  330   push(rscratch);
  331 }
  332 
  333 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
  334   mov_metadata(rscratch, obj);
  335   push(rscratch);
  336 }
  337 
  338 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
  339   lea(rscratch, src);
  340   if (src.is_lval()) {
  341     push(rscratch);
  342   } else {
  343     pushq(Address(rscratch, 0));
  344   }
  345 }
  346 
  347 static void pass_arg0(MacroAssembler* masm, Register arg) {
  348   if (c_rarg0 != arg ) {
  349     masm->mov(c_rarg0, arg);
  350   }
  351 }
  352 
  353 static void pass_arg1(MacroAssembler* masm, Register arg) {
  354   if (c_rarg1 != arg ) {
  355     masm->mov(c_rarg1, arg);
  356   }
  357 }
  358 
  359 static void pass_arg2(MacroAssembler* masm, Register arg) {
  360   if (c_rarg2 != arg ) {
  361     masm->mov(c_rarg2, arg);
  362   }
  363 }
  364 
  365 static void pass_arg3(MacroAssembler* masm, Register arg) {
  366   if (c_rarg3 != arg ) {
  367     masm->mov(c_rarg3, arg);
  368   }
  369 }
  370 
  371 void MacroAssembler::stop(const char* msg) {
  372   if (ShowMessageBoxOnError) {
  373     address rip = pc();
  374     pusha(); // get regs on stack
  375     lea(c_rarg1, InternalAddress(rip));
  376     movq(c_rarg2, rsp); // pass pointer to regs array
  377   }
  378   // Skip AOT caching C strings in scratch buffer.
  379   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
  380   lea(c_rarg0, ExternalAddress((address) str));
  381   andq(rsp, -16); // align stack as required by ABI
  382   call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
  383   hlt();
  384 }
  385 
  386 void MacroAssembler::warn(const char* msg) {
  387   push(rbp);
  388   movq(rbp, rsp);
  389   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  390   push_CPU_state();   // keeps alignment at 16 bytes
  391 
  392 #ifdef _WIN64
  393   // Windows always allocates space for its register args
  394   subq(rsp,  frame::arg_reg_save_area_bytes);
  395 #endif
  396   lea(c_rarg0, ExternalAddress((address) msg));
  397   call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
  398 
  399 #ifdef _WIN64
  400   // restore stack pointer
  401   addq(rsp, frame::arg_reg_save_area_bytes);
  402 #endif
  403   pop_CPU_state();
  404   mov(rsp, rbp);
  405   pop(rbp);
  406 }
  407 
  408 void MacroAssembler::print_state() {
  409   address rip = pc();
  410   pusha();            // get regs on stack
  411   push(rbp);
  412   movq(rbp, rsp);
  413   andq(rsp, -16);     // align stack as required by push_CPU_state and call
  414   push_CPU_state();   // keeps alignment at 16 bytes
  415 
  416   lea(c_rarg0, InternalAddress(rip));
  417   lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
  418   call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
  419 
  420   pop_CPU_state();
  421   mov(rsp, rbp);
  422   pop(rbp);
  423   popa();
  424 }
  425 
  426 #ifndef PRODUCT
  427 extern "C" void findpc(intptr_t x);
  428 #endif
  429 
  430 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
  431   // In order to get locks to work, we need to fake a in_VM state
  432   if (ShowMessageBoxOnError) {
  433     JavaThread* thread = JavaThread::current();
  434     JavaThreadState saved_state = thread->thread_state();
  435     thread->set_thread_state(_thread_in_vm);
  436 #ifndef PRODUCT
  437     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
  438       ttyLocker ttyl;
  439       BytecodeCounter::print();
  440     }
  441 #endif
  442     // To see where a verify_oop failed, get $ebx+40/X for this frame.
  443     // XXX correct this offset for amd64
  444     // This is the value of eip which points to where verify_oop will return.
  445     if (os::message_box(msg, "Execution stopped, print registers?")) {
  446       print_state64(pc, regs);
  447       BREAKPOINT;
  448     }
  449   }
  450   fatal("DEBUG MESSAGE: %s", msg);
  451 }
  452 
  453 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
  454   ttyLocker ttyl;
  455   DebuggingContext debugging{};
  456   tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
  457 #ifndef PRODUCT
  458   tty->cr();
  459   findpc(pc);
  460   tty->cr();
  461 #endif
  462 #define PRINT_REG(rax, value) \
  463   { tty->print("%s = ", #rax); os::print_location(tty, value); }
  464   PRINT_REG(rax, regs[15]);
  465   PRINT_REG(rbx, regs[12]);
  466   PRINT_REG(rcx, regs[14]);
  467   PRINT_REG(rdx, regs[13]);
  468   PRINT_REG(rdi, regs[8]);
  469   PRINT_REG(rsi, regs[9]);
  470   PRINT_REG(rbp, regs[10]);
  471   // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
  472   PRINT_REG(rsp, (intptr_t)(&regs[16]));
  473   PRINT_REG(r8 , regs[7]);
  474   PRINT_REG(r9 , regs[6]);
  475   PRINT_REG(r10, regs[5]);
  476   PRINT_REG(r11, regs[4]);
  477   PRINT_REG(r12, regs[3]);
  478   PRINT_REG(r13, regs[2]);
  479   PRINT_REG(r14, regs[1]);
  480   PRINT_REG(r15, regs[0]);
  481 #undef PRINT_REG
  482   // Print some words near the top of the stack.
  483   int64_t* rsp = &regs[16];
  484   int64_t* dump_sp = rsp;
  485   for (int col1 = 0; col1 < 8; col1++) {
  486     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  487     os::print_location(tty, *dump_sp++);
  488   }
  489   for (int row = 0; row < 25; row++) {
  490     tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
  491     for (int col = 0; col < 4; col++) {
  492       tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
  493     }
  494     tty->cr();
  495   }
  496   // Print some instructions around pc:
  497   Disassembler::decode((address)pc-64, (address)pc);
  498   tty->print_cr("--------");
  499   Disassembler::decode((address)pc, (address)pc+32);
  500 }
  501 
  502 // The java_calling_convention describes stack locations as ideal slots on
  503 // a frame with no abi restrictions. Since we must observe abi restrictions
  504 // (like the placement of the register window) the slots must be biased by
  505 // the following value.
  506 static int reg2offset_in(VMReg r) {
  507   // Account for saved rbp and return address
  508   // This should really be in_preserve_stack_slots
  509   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
  510 }
  511 
  512 static int reg2offset_out(VMReg r) {
  513   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
  514 }
  515 
  516 // A long move
  517 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  518 
  519   // The calling conventions assures us that each VMregpair is either
  520   // all really one physical register or adjacent stack slots.
  521 
  522   if (src.is_single_phys_reg() ) {
  523     if (dst.is_single_phys_reg()) {
  524       if (dst.first() != src.first()) {
  525         mov(dst.first()->as_Register(), src.first()->as_Register());
  526       }
  527     } else {
  528       assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
  529              src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
  530       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  531     }
  532   } else if (dst.is_single_phys_reg()) {
  533     assert(src.is_single_reg(),  "not a stack pair");
  534     movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  535   } else {
  536     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  537     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  538     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  539   }
  540 }
  541 
  542 // A double move
  543 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  544 
  545   // The calling conventions assures us that each VMregpair is either
  546   // all really one physical register or adjacent stack slots.
  547 
  548   if (src.is_single_phys_reg() ) {
  549     if (dst.is_single_phys_reg()) {
  550       // In theory these overlap but the ordering is such that this is likely a nop
  551       if ( src.first() != dst.first()) {
  552         movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
  553       }
  554     } else {
  555       assert(dst.is_single_reg(), "not a stack pair");
  556       movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  557     }
  558   } else if (dst.is_single_phys_reg()) {
  559     assert(src.is_single_reg(),  "not a stack pair");
  560     movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  561   } else {
  562     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
  563     movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  564     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  565   }
  566 }
  567 
  568 
  569 // A float arg may have to do float reg int reg conversion
  570 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  571   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
  572 
  573   // The calling conventions assures us that each VMregpair is either
  574   // all really one physical register or adjacent stack slots.
  575 
  576   if (src.first()->is_stack()) {
  577     if (dst.first()->is_stack()) {
  578       movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  579       movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  580     } else {
  581       // stack to reg
  582       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  583       movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  584     }
  585   } else if (dst.first()->is_stack()) {
  586     // reg to stack
  587     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
  588     movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
  589   } else {
  590     // reg to reg
  591     // In theory these overlap but the ordering is such that this is likely a nop
  592     if ( src.first() != dst.first()) {
  593       movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
  594     }
  595   }
  596 }
  597 
  598 // On 64 bit we will store integer like items to the stack as
  599 // 64 bits items (x86_32/64 abi) even though java would only store
  600 // 32bits for a parameter. On 32bit it will simply be 32 bits
  601 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
  602 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
  603   if (src.first()->is_stack()) {
  604     if (dst.first()->is_stack()) {
  605       // stack to stack
  606       movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  607       movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
  608     } else {
  609       // stack to reg
  610       movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
  611     }
  612   } else if (dst.first()->is_stack()) {
  613     // reg to stack
  614     // Do we really have to sign extend???
  615     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
  616     movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
  617   } else {
  618     // Do we really have to sign extend???
  619     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
  620     if (dst.first() != src.first()) {
  621       movq(dst.first()->as_Register(), src.first()->as_Register());
  622     }
  623   }
  624 }
  625 
  626 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
  627   if (src.first()->is_stack()) {
  628     if (dst.first()->is_stack()) {
  629       // stack to stack
  630       movq(rax, Address(rbp, reg2offset_in(src.first())));
  631       movq(Address(rsp, reg2offset_out(dst.first())), rax);
  632     } else {
  633       // stack to reg
  634       movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
  635     }
  636   } else if (dst.first()->is_stack()) {
  637     // reg to stack
  638     movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
  639   } else {
  640     if (dst.first() != src.first()) {
  641       movq(dst.first()->as_Register(), src.first()->as_Register());
  642     }
  643   }
  644 }
  645 
  646 // An oop arg. Must pass a handle not the oop itself
  647 void MacroAssembler::object_move(OopMap* map,
  648                         int oop_handle_offset,
  649                         int framesize_in_slots,
  650                         VMRegPair src,
  651                         VMRegPair dst,
  652                         bool is_receiver,
  653                         int* receiver_offset) {
  654 
  655   // must pass a handle. First figure out the location we use as a handle
  656 
  657   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
  658 
  659   // See if oop is null if it is we need no handle
  660 
  661   if (src.first()->is_stack()) {
  662 
  663     // Oop is already on the stack as an argument
  664     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
  665     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
  666     if (is_receiver) {
  667       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
  668     }
  669 
  670     cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
  671     lea(rHandle, Address(rbp, reg2offset_in(src.first())));
  672     // conditionally move a null
  673     cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
  674   } else {
  675 
  676     // Oop is in a register we must store it to the space we reserve
  677     // on the stack for oop_handles and pass a handle if oop is non-null
  678 
  679     const Register rOop = src.first()->as_Register();
  680     int oop_slot;
  681     if (rOop == j_rarg0)
  682       oop_slot = 0;
  683     else if (rOop == j_rarg1)
  684       oop_slot = 1;
  685     else if (rOop == j_rarg2)
  686       oop_slot = 2;
  687     else if (rOop == j_rarg3)
  688       oop_slot = 3;
  689     else if (rOop == j_rarg4)
  690       oop_slot = 4;
  691     else {
  692       assert(rOop == j_rarg5, "wrong register");
  693       oop_slot = 5;
  694     }
  695 
  696     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
  697     int offset = oop_slot*VMRegImpl::stack_slot_size;
  698 
  699     map->set_oop(VMRegImpl::stack2reg(oop_slot));
  700     // Store oop in handle area, may be null
  701     movptr(Address(rsp, offset), rOop);
  702     if (is_receiver) {
  703       *receiver_offset = offset;
  704     }
  705 
  706     cmpptr(rOop, NULL_WORD);
  707     lea(rHandle, Address(rsp, offset));
  708     // conditionally move a null from the handle area where it was just stored
  709     cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
  710   }
  711 
  712   // If arg is on the stack then place it otherwise it is already in correct reg.
  713   if (dst.first()->is_stack()) {
  714     movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
  715   }
  716 }
  717 
  718 void MacroAssembler::addptr(Register dst, int32_t imm32) {
  719   addq(dst, imm32);
  720 }
  721 
  722 void MacroAssembler::addptr(Register dst, Register src) {
  723   addq(dst, src);
  724 }
  725 
  726 void MacroAssembler::addptr(Address dst, Register src) {
  727   addq(dst, src);
  728 }
  729 
  730 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  731   assert(rscratch != noreg || always_reachable(src), "missing");
  732 
  733   if (reachable(src)) {
  734     Assembler::addsd(dst, as_Address(src));
  735   } else {
  736     lea(rscratch, src);
  737     Assembler::addsd(dst, Address(rscratch, 0));
  738   }
  739 }
  740 
  741 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
  742   assert(rscratch != noreg || always_reachable(src), "missing");
  743 
  744   if (reachable(src)) {
  745     addss(dst, as_Address(src));
  746   } else {
  747     lea(rscratch, src);
  748     addss(dst, Address(rscratch, 0));
  749   }
  750 }
  751 
  752 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  753   assert(rscratch != noreg || always_reachable(src), "missing");
  754 
  755   if (reachable(src)) {
  756     Assembler::addpd(dst, as_Address(src));
  757   } else {
  758     lea(rscratch, src);
  759     Assembler::addpd(dst, Address(rscratch, 0));
  760   }
  761 }
  762 
  763 // See 8273459.  Function for ensuring 64-byte alignment, intended for stubs only.
  764 // Stub code is generated once and never copied.
  765 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
  766 void MacroAssembler::align64() {
  767   align(64, (uint)(uintptr_t)pc());
  768 }
  769 
  770 void MacroAssembler::align32() {
  771   align(32, (uint)(uintptr_t)pc());
  772 }
  773 
  774 void MacroAssembler::align(uint modulus) {
  775   // 8273459: Ensure alignment is possible with current segment alignment
  776   assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
  777   align(modulus, offset());
  778 }
  779 
  780 void MacroAssembler::align(uint modulus, uint target) {
  781   if (target % modulus != 0) {
  782     nop(modulus - (target % modulus));
  783   }
  784 }
  785 
  786 void MacroAssembler::push_f(XMMRegister r) {
  787   subptr(rsp, wordSize);
  788   movflt(Address(rsp, 0), r);
  789 }
  790 
  791 void MacroAssembler::pop_f(XMMRegister r) {
  792   movflt(r, Address(rsp, 0));
  793   addptr(rsp, wordSize);
  794 }
  795 
  796 void MacroAssembler::push_d(XMMRegister r) {
  797   subptr(rsp, 2 * wordSize);
  798   movdbl(Address(rsp, 0), r);
  799 }
  800 
  801 void MacroAssembler::pop_d(XMMRegister r) {
  802   movdbl(r, Address(rsp, 0));
  803   addptr(rsp, 2 * Interpreter::stackElementSize);
  804 }
  805 
  806 void MacroAssembler::push_ppx(Register src) {
  807   if (VM_Version::supports_apx_f()) {
  808     pushp(src);
  809   } else {
  810     Assembler::push(src);
  811   }
  812 }
  813 
  814 void MacroAssembler::pop_ppx(Register dst) {
  815   if (VM_Version::supports_apx_f()) {
  816     popp(dst);
  817   } else {
  818     Assembler::pop(dst);
  819   }
  820 }
  821 
  822 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
  823   // Used in sign-masking with aligned address.
  824   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  825   assert(rscratch != noreg || always_reachable(src), "missing");
  826 
  827   if (UseAVX > 2 &&
  828       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
  829       (dst->encoding() >= 16)) {
  830     vpand(dst, dst, src, AVX_512bit, rscratch);
  831   } else if (reachable(src)) {
  832     Assembler::andpd(dst, as_Address(src));
  833   } else {
  834     lea(rscratch, src);
  835     Assembler::andpd(dst, Address(rscratch, 0));
  836   }
  837 }
  838 
  839 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
  840   // Used in sign-masking with aligned address.
  841   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
  842   assert(rscratch != noreg || always_reachable(src), "missing");
  843 
  844   if (reachable(src)) {
  845     Assembler::andps(dst, as_Address(src));
  846   } else {
  847     lea(rscratch, src);
  848     Assembler::andps(dst, Address(rscratch, 0));
  849   }
  850 }
  851 
  852 void MacroAssembler::andptr(Register dst, int32_t imm32) {
  853   andq(dst, imm32);
  854 }
  855 
  856 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
  857   assert(rscratch != noreg || always_reachable(src), "missing");
  858 
  859   if (reachable(src)) {
  860     andq(dst, as_Address(src));
  861   } else {
  862     lea(rscratch, src);
  863     andq(dst, Address(rscratch, 0));
  864   }
  865 }
  866 
  867 void MacroAssembler::atomic_incl(Address counter_addr) {
  868   lock();
  869   incrementl(counter_addr);
  870 }
  871 
  872 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
  873   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  874 
  875   if (reachable(counter_addr)) {
  876     atomic_incl(as_Address(counter_addr));
  877   } else {
  878     lea(rscratch, counter_addr);
  879     atomic_incl(Address(rscratch, 0));
  880   }
  881 }
  882 
  883 void MacroAssembler::atomic_incq(Address counter_addr) {
  884   lock();
  885   incrementq(counter_addr);
  886 }
  887 
  888 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
  889   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
  890 
  891   if (reachable(counter_addr)) {
  892     atomic_incq(as_Address(counter_addr));
  893   } else {
  894     lea(rscratch, counter_addr);
  895     atomic_incq(Address(rscratch, 0));
  896   }
  897 }
  898 
  899 // Writes to stack successive pages until offset reached to check for
  900 // stack overflow + shadow pages.  This clobbers tmp.
  901 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
  902   movptr(tmp, rsp);
  903   // Bang stack for total size given plus shadow page size.
  904   // Bang one page at a time because large size can bang beyond yellow and
  905   // red zones.
  906   Label loop;
  907   bind(loop);
  908   movl(Address(tmp, (-(int)os::vm_page_size())), size );
  909   subptr(tmp, (int)os::vm_page_size());
  910   subl(size, (int)os::vm_page_size());
  911   jcc(Assembler::greater, loop);
  912 
  913   // Bang down shadow pages too.
  914   // At this point, (tmp-0) is the last address touched, so don't
  915   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
  916   // was post-decremented.)  Skip this address by starting at i=1, and
  917   // touch a few more pages below.  N.B.  It is important to touch all
  918   // the way down including all pages in the shadow zone.
  919   for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
  920     // this could be any sized move but this is can be a debugging crumb
  921     // so the bigger the better.
  922     movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
  923   }
  924 }
  925 
  926 void MacroAssembler::reserved_stack_check() {
  927   // testing if reserved zone needs to be enabled
  928   Label no_reserved_zone_enabling;
  929 
  930   cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
  931   jcc(Assembler::below, no_reserved_zone_enabling);
  932 
  933   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
  934   jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
  935   should_not_reach_here();
  936 
  937   bind(no_reserved_zone_enabling);
  938 }
  939 
  940 void MacroAssembler::c2bool(Register x) {
  941   // implements x == 0 ? 0 : 1
  942   // note: must only look at least-significant byte of x
  943   //       since C-style booleans are stored in one byte
  944   //       only! (was bug)
  945   andl(x, 0xFF);
  946   setb(Assembler::notZero, x);
  947 }
  948 
  949 // Wouldn't need if AddressLiteral version had new name
  950 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
  951   Assembler::call(L, rtype);
  952 }
  953 
  954 void MacroAssembler::call(Register entry) {
  955   Assembler::call(entry);
  956 }
  957 
  958 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
  959   assert(rscratch != noreg || always_reachable(entry), "missing");
  960 
  961   if (reachable(entry)) {
  962     Assembler::call_literal(entry.target(), entry.rspec());
  963   } else {
  964     lea(rscratch, entry);
  965     Assembler::call(rscratch);
  966   }
  967 }
  968 
  969 void MacroAssembler::ic_call(address entry, jint method_index) {
  970   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
  971   // Needs full 64-bit immediate for later patching.
  972   mov64(rax, (int64_t)Universe::non_oop_word());
  973   call(AddressLiteral(entry, rh));
  974 }
  975 
  976 int MacroAssembler::ic_check_size() {
  977   return UseCompactObjectHeaders ? 17 : 14;
  978 }
  979 
  980 int MacroAssembler::ic_check(int end_alignment) {
  981   Register receiver = j_rarg0;
  982   Register data = rax;
  983   Register temp = rscratch1;
  984 
  985   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
  986   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
  987   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
  988   // before the inline cache check here, and not after
  989   align(end_alignment, offset() + ic_check_size());
  990 
  991   int uep_offset = offset();
  992 
  993   if (UseCompactObjectHeaders) {
  994     load_narrow_klass_compact(temp, receiver);
  995     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  996   } else if (UseCompressedClassPointers) {
  997     movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
  998     cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
  999   } else {
 1000     movptr(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
 1001     cmpptr(temp, Address(data, CompiledICData::speculated_klass_offset()));
 1002   }
 1003 
 1004   // if inline cache check fails, then jump to runtime routine
 1005   jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 1006   assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
 1007 
 1008   return uep_offset;
 1009 }
 1010 
 1011 void MacroAssembler::emit_static_call_stub() {
 1012   // Static stub relocation also tags the Method* in the code-stream.
 1013   mov_metadata(rbx, (Metadata*) nullptr);  // Method is zapped till fixup time.
 1014   // This is recognized as unresolved by relocs/nativeinst/ic code.
 1015   jump(RuntimeAddress(pc()));
 1016 }
 1017 
 1018 // Implementation of call_VM versions
 1019 
 1020 void MacroAssembler::call_VM(Register oop_result,
 1021                              address entry_point,
 1022                              bool check_exceptions) {
 1023   Label C, E;
 1024   call(C, relocInfo::none);
 1025   jmp(E);
 1026 
 1027   bind(C);
 1028   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
 1029   ret(0);
 1030 
 1031   bind(E);
 1032 }
 1033 
 1034 void MacroAssembler::call_VM(Register oop_result,
 1035                              address entry_point,
 1036                              Register arg_1,
 1037                              bool check_exceptions) {
 1038   Label C, E;
 1039   call(C, relocInfo::none);
 1040   jmp(E);
 1041 
 1042   bind(C);
 1043   pass_arg1(this, arg_1);
 1044   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
 1045   ret(0);
 1046 
 1047   bind(E);
 1048 }
 1049 
 1050 void MacroAssembler::call_VM(Register oop_result,
 1051                              address entry_point,
 1052                              Register arg_1,
 1053                              Register arg_2,
 1054                              bool check_exceptions) {
 1055   Label C, E;
 1056   call(C, relocInfo::none);
 1057   jmp(E);
 1058 
 1059   bind(C);
 1060 
 1061   assert_different_registers(arg_1, c_rarg2);
 1062 
 1063   pass_arg2(this, arg_2);
 1064   pass_arg1(this, arg_1);
 1065   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
 1066   ret(0);
 1067 
 1068   bind(E);
 1069 }
 1070 
 1071 void MacroAssembler::call_VM(Register oop_result,
 1072                              address entry_point,
 1073                              Register arg_1,
 1074                              Register arg_2,
 1075                              Register arg_3,
 1076                              bool check_exceptions) {
 1077   Label C, E;
 1078   call(C, relocInfo::none);
 1079   jmp(E);
 1080 
 1081   bind(C);
 1082 
 1083   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1084   assert_different_registers(arg_2, c_rarg3);
 1085   pass_arg3(this, arg_3);
 1086   pass_arg2(this, arg_2);
 1087   pass_arg1(this, arg_1);
 1088   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
 1089   ret(0);
 1090 
 1091   bind(E);
 1092 }
 1093 
 1094 void MacroAssembler::call_VM(Register oop_result,
 1095                              Register last_java_sp,
 1096                              address entry_point,
 1097                              int number_of_arguments,
 1098                              bool check_exceptions) {
 1099   call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1100 }
 1101 
 1102 void MacroAssembler::call_VM(Register oop_result,
 1103                              Register last_java_sp,
 1104                              address entry_point,
 1105                              Register arg_1,
 1106                              bool check_exceptions) {
 1107   pass_arg1(this, arg_1);
 1108   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1109 }
 1110 
 1111 void MacroAssembler::call_VM(Register oop_result,
 1112                              Register last_java_sp,
 1113                              address entry_point,
 1114                              Register arg_1,
 1115                              Register arg_2,
 1116                              bool check_exceptions) {
 1117 
 1118   assert_different_registers(arg_1, c_rarg2);
 1119   pass_arg2(this, arg_2);
 1120   pass_arg1(this, arg_1);
 1121   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1122 }
 1123 
 1124 void MacroAssembler::call_VM(Register oop_result,
 1125                              Register last_java_sp,
 1126                              address entry_point,
 1127                              Register arg_1,
 1128                              Register arg_2,
 1129                              Register arg_3,
 1130                              bool check_exceptions) {
 1131   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1132   assert_different_registers(arg_2, c_rarg3);
 1133   pass_arg3(this, arg_3);
 1134   pass_arg2(this, arg_2);
 1135   pass_arg1(this, arg_1);
 1136   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1137 }
 1138 
 1139 void MacroAssembler::super_call_VM(Register oop_result,
 1140                                    Register last_java_sp,
 1141                                    address entry_point,
 1142                                    int number_of_arguments,
 1143                                    bool check_exceptions) {
 1144   MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
 1145 }
 1146 
 1147 void MacroAssembler::super_call_VM(Register oop_result,
 1148                                    Register last_java_sp,
 1149                                    address entry_point,
 1150                                    Register arg_1,
 1151                                    bool check_exceptions) {
 1152   pass_arg1(this, arg_1);
 1153   super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
 1154 }
 1155 
 1156 void MacroAssembler::super_call_VM(Register oop_result,
 1157                                    Register last_java_sp,
 1158                                    address entry_point,
 1159                                    Register arg_1,
 1160                                    Register arg_2,
 1161                                    bool check_exceptions) {
 1162 
 1163   assert_different_registers(arg_1, c_rarg2);
 1164   pass_arg2(this, arg_2);
 1165   pass_arg1(this, arg_1);
 1166   super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
 1167 }
 1168 
 1169 void MacroAssembler::super_call_VM(Register oop_result,
 1170                                    Register last_java_sp,
 1171                                    address entry_point,
 1172                                    Register arg_1,
 1173                                    Register arg_2,
 1174                                    Register arg_3,
 1175                                    bool check_exceptions) {
 1176   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1177   assert_different_registers(arg_2, c_rarg3);
 1178   pass_arg3(this, arg_3);
 1179   pass_arg2(this, arg_2);
 1180   pass_arg1(this, arg_1);
 1181   super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
 1182 }
 1183 
 1184 void MacroAssembler::call_VM_base(Register oop_result,
 1185                                   Register last_java_sp,
 1186                                   address  entry_point,
 1187                                   int      number_of_arguments,
 1188                                   bool     check_exceptions) {
 1189   Register java_thread = r15_thread;
 1190 
 1191   // determine last_java_sp register
 1192   if (!last_java_sp->is_valid()) {
 1193     last_java_sp = rsp;
 1194   }
 1195   // debugging support
 1196   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 1197 #ifdef ASSERT
 1198   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 1199   // r12 is the heapbase.
 1200   if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 1201 #endif // ASSERT
 1202 
 1203   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 1204   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 1205 
 1206   // push java thread (becomes first argument of C function)
 1207 
 1208   mov(c_rarg0, r15_thread);
 1209 
 1210   // set last Java frame before call
 1211   assert(last_java_sp != rbp, "can't use ebp/rbp");
 1212 
 1213   // Only interpreter should have to set fp
 1214   set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
 1215 
 1216   // do the call, remove parameters
 1217   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 1218 
 1219 #ifdef ASSERT
 1220   // Check that thread register is not clobbered.
 1221   guarantee(java_thread != rax, "change this code");
 1222   push(rax);
 1223   { Label L;
 1224     get_thread_slow(rax);
 1225     cmpptr(java_thread, rax);
 1226     jcc(Assembler::equal, L);
 1227     STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
 1228     bind(L);
 1229   }
 1230   pop(rax);
 1231 #endif
 1232 
 1233   // reset last Java frame
 1234   // Only interpreter should have to clear fp
 1235   reset_last_Java_frame(true);
 1236 
 1237    // C++ interp handles this in the interpreter
 1238   check_and_handle_popframe();
 1239   check_and_handle_earlyret();
 1240 
 1241   if (check_exceptions) {
 1242     // check for pending exceptions (java_thread is set upon return)
 1243     cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
 1244     // This used to conditionally jump to forward_exception however it is
 1245     // possible if we relocate that the branch will not reach. So we must jump
 1246     // around so we can always reach
 1247 
 1248     Label ok;
 1249     jcc(Assembler::equal, ok);
 1250     jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 1251     bind(ok);
 1252   }
 1253 
 1254   // get oop result if there is one and reset the value in the thread
 1255   if (oop_result->is_valid()) {
 1256     get_vm_result_oop(oop_result);
 1257   }
 1258 }
 1259 
 1260 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 1261   // Calculate the value for last_Java_sp somewhat subtle.
 1262   // call_VM does an intermediate call which places a return address on
 1263   // the stack just under the stack pointer as the user finished with it.
 1264   // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
 1265 
 1266   // We've pushed one address, correct last_Java_sp
 1267   lea(rax, Address(rsp, wordSize));
 1268 
 1269   call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
 1270 }
 1271 
 1272 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
 1273 void MacroAssembler::call_VM_leaf0(address entry_point) {
 1274   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 1275 }
 1276 
 1277 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
 1278   call_VM_leaf_base(entry_point, number_of_arguments);
 1279 }
 1280 
 1281 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
 1282   pass_arg0(this, arg_0);
 1283   call_VM_leaf(entry_point, 1);
 1284 }
 1285 
 1286 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1287 
 1288   assert_different_registers(arg_0, c_rarg1);
 1289   pass_arg1(this, arg_1);
 1290   pass_arg0(this, arg_0);
 1291   call_VM_leaf(entry_point, 2);
 1292 }
 1293 
 1294 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1295   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1296   assert_different_registers(arg_1, c_rarg2);
 1297   pass_arg2(this, arg_2);
 1298   pass_arg1(this, arg_1);
 1299   pass_arg0(this, arg_0);
 1300   call_VM_leaf(entry_point, 3);
 1301 }
 1302 
 1303 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1304   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1305   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1306   assert_different_registers(arg_2, c_rarg3);
 1307   pass_arg3(this, arg_3);
 1308   pass_arg2(this, arg_2);
 1309   pass_arg1(this, arg_1);
 1310   pass_arg0(this, arg_0);
 1311   call_VM_leaf(entry_point, 3);
 1312 }
 1313 
 1314 void MacroAssembler::super_call_VM_leaf(address entry_point) {
 1315   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1316 }
 1317 
 1318 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1319   pass_arg0(this, arg_0);
 1320   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1321 }
 1322 
 1323 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1324   assert_different_registers(arg_0, c_rarg1);
 1325   pass_arg1(this, arg_1);
 1326   pass_arg0(this, arg_0);
 1327   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1328 }
 1329 
 1330 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1331   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1332   assert_different_registers(arg_1, c_rarg2);
 1333   pass_arg2(this, arg_2);
 1334   pass_arg1(this, arg_1);
 1335   pass_arg0(this, arg_0);
 1336   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1337 }
 1338 
 1339 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1340   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1341   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1342   assert_different_registers(arg_2, c_rarg3);
 1343   pass_arg3(this, arg_3);
 1344   pass_arg2(this, arg_2);
 1345   pass_arg1(this, arg_1);
 1346   pass_arg0(this, arg_0);
 1347   MacroAssembler::call_VM_leaf_base(entry_point, 4);
 1348 }
 1349 
 1350 void MacroAssembler::get_vm_result_oop(Register oop_result) {
 1351   movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
 1352   movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
 1353   verify_oop_msg(oop_result, "broken oop in call_VM_base");
 1354 }
 1355 
 1356 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
 1357   movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
 1358   movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
 1359 }
 1360 
 1361 void MacroAssembler::check_and_handle_earlyret() {
 1362 }
 1363 
 1364 void MacroAssembler::check_and_handle_popframe() {
 1365 }
 1366 
 1367 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
 1368   assert(rscratch != noreg || always_reachable(src1), "missing");
 1369 
 1370   if (reachable(src1)) {
 1371     cmpl(as_Address(src1), imm);
 1372   } else {
 1373     lea(rscratch, src1);
 1374     cmpl(Address(rscratch, 0), imm);
 1375   }
 1376 }
 1377 
 1378 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
 1379   assert(!src2.is_lval(), "use cmpptr");
 1380   assert(rscratch != noreg || always_reachable(src2), "missing");
 1381 
 1382   if (reachable(src2)) {
 1383     cmpl(src1, as_Address(src2));
 1384   } else {
 1385     lea(rscratch, src2);
 1386     cmpl(src1, Address(rscratch, 0));
 1387   }
 1388 }
 1389 
 1390 void MacroAssembler::cmp32(Register src1, int32_t imm) {
 1391   Assembler::cmpl(src1, imm);
 1392 }
 1393 
 1394 void MacroAssembler::cmp32(Register src1, Address src2) {
 1395   Assembler::cmpl(src1, src2);
 1396 }
 1397 
 1398 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1399   ucomisd(opr1, opr2);
 1400 
 1401   Label L;
 1402   if (unordered_is_less) {
 1403     movl(dst, -1);
 1404     jcc(Assembler::parity, L);
 1405     jcc(Assembler::below , L);
 1406     movl(dst, 0);
 1407     jcc(Assembler::equal , L);
 1408     increment(dst);
 1409   } else { // unordered is greater
 1410     movl(dst, 1);
 1411     jcc(Assembler::parity, L);
 1412     jcc(Assembler::above , L);
 1413     movl(dst, 0);
 1414     jcc(Assembler::equal , L);
 1415     decrementl(dst);
 1416   }
 1417   bind(L);
 1418 }
 1419 
 1420 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
 1421   ucomiss(opr1, opr2);
 1422 
 1423   Label L;
 1424   if (unordered_is_less) {
 1425     movl(dst, -1);
 1426     jcc(Assembler::parity, L);
 1427     jcc(Assembler::below , L);
 1428     movl(dst, 0);
 1429     jcc(Assembler::equal , L);
 1430     increment(dst);
 1431   } else { // unordered is greater
 1432     movl(dst, 1);
 1433     jcc(Assembler::parity, L);
 1434     jcc(Assembler::above , L);
 1435     movl(dst, 0);
 1436     jcc(Assembler::equal , L);
 1437     decrementl(dst);
 1438   }
 1439   bind(L);
 1440 }
 1441 
 1442 
 1443 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
 1444   assert(rscratch != noreg || always_reachable(src1), "missing");
 1445 
 1446   if (reachable(src1)) {
 1447     cmpb(as_Address(src1), imm);
 1448   } else {
 1449     lea(rscratch, src1);
 1450     cmpb(Address(rscratch, 0), imm);
 1451   }
 1452 }
 1453 
 1454 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
 1455   assert(rscratch != noreg || always_reachable(src2), "missing");
 1456 
 1457   if (src2.is_lval()) {
 1458     movptr(rscratch, src2);
 1459     Assembler::cmpq(src1, rscratch);
 1460   } else if (reachable(src2)) {
 1461     cmpq(src1, as_Address(src2));
 1462   } else {
 1463     lea(rscratch, src2);
 1464     Assembler::cmpq(src1, Address(rscratch, 0));
 1465   }
 1466 }
 1467 
 1468 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
 1469   assert(src2.is_lval(), "not a mem-mem compare");
 1470   // moves src2's literal address
 1471   movptr(rscratch, src2);
 1472   Assembler::cmpq(src1, rscratch);
 1473 }
 1474 
 1475 void MacroAssembler::cmpoop(Register src1, Register src2) {
 1476   cmpptr(src1, src2);
 1477 }
 1478 
 1479 void MacroAssembler::cmpoop(Register src1, Address src2) {
 1480   cmpptr(src1, src2);
 1481 }
 1482 
 1483 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
 1484   movoop(rscratch, src2);
 1485   cmpptr(src1, rscratch);
 1486 }
 1487 
 1488 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
 1489   assert(rscratch != noreg || always_reachable(adr), "missing");
 1490 
 1491   if (reachable(adr)) {
 1492     lock();
 1493     cmpxchgptr(reg, as_Address(adr));
 1494   } else {
 1495     lea(rscratch, adr);
 1496     lock();
 1497     cmpxchgptr(reg, Address(rscratch, 0));
 1498   }
 1499 }
 1500 
 1501 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
 1502   cmpxchgq(reg, adr);
 1503 }
 1504 
 1505 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1506   assert(rscratch != noreg || always_reachable(src), "missing");
 1507 
 1508   if (reachable(src)) {
 1509     Assembler::comisd(dst, as_Address(src));
 1510   } else {
 1511     lea(rscratch, src);
 1512     Assembler::comisd(dst, Address(rscratch, 0));
 1513   }
 1514 }
 1515 
 1516 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1517   assert(rscratch != noreg || always_reachable(src), "missing");
 1518 
 1519   if (reachable(src)) {
 1520     Assembler::comiss(dst, as_Address(src));
 1521   } else {
 1522     lea(rscratch, src);
 1523     Assembler::comiss(dst, Address(rscratch, 0));
 1524   }
 1525 }
 1526 
 1527 
 1528 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
 1529   assert(rscratch != noreg || always_reachable(counter_addr), "missing");
 1530 
 1531   Condition negated_cond = negate_condition(cond);
 1532   Label L;
 1533   jcc(negated_cond, L);
 1534   pushf(); // Preserve flags
 1535   atomic_incl(counter_addr, rscratch);
 1536   popf();
 1537   bind(L);
 1538 }
 1539 
 1540 int MacroAssembler::corrected_idivl(Register reg) {
 1541   // Full implementation of Java idiv and irem; checks for
 1542   // special case as described in JVM spec., p.243 & p.271.
 1543   // The function returns the (pc) offset of the idivl
 1544   // instruction - may be needed for implicit exceptions.
 1545   //
 1546   //         normal case                           special case
 1547   //
 1548   // input : rax,: dividend                         min_int
 1549   //         reg: divisor   (may not be rax,/rdx)   -1
 1550   //
 1551   // output: rax,: quotient  (= rax, idiv reg)       min_int
 1552   //         rdx: remainder (= rax, irem reg)       0
 1553   assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
 1554   const int min_int = 0x80000000;
 1555   Label normal_case, special_case;
 1556 
 1557   // check for special case
 1558   cmpl(rax, min_int);
 1559   jcc(Assembler::notEqual, normal_case);
 1560   xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
 1561   cmpl(reg, -1);
 1562   jcc(Assembler::equal, special_case);
 1563 
 1564   // handle normal case
 1565   bind(normal_case);
 1566   cdql();
 1567   int idivl_offset = offset();
 1568   idivl(reg);
 1569 
 1570   // normal and special case exit
 1571   bind(special_case);
 1572 
 1573   return idivl_offset;
 1574 }
 1575 
 1576 
 1577 
 1578 void MacroAssembler::decrementl(Register reg, int value) {
 1579   if (value == min_jint) {subl(reg, value) ; return; }
 1580   if (value <  0) { incrementl(reg, -value); return; }
 1581   if (value == 0) {                        ; return; }
 1582   if (value == 1 && UseIncDec) { decl(reg) ; return; }
 1583   /* else */      { subl(reg, value)       ; return; }
 1584 }
 1585 
 1586 void MacroAssembler::decrementl(Address dst, int value) {
 1587   if (value == min_jint) {subl(dst, value) ; return; }
 1588   if (value <  0) { incrementl(dst, -value); return; }
 1589   if (value == 0) {                        ; return; }
 1590   if (value == 1 && UseIncDec) { decl(dst) ; return; }
 1591   /* else */      { subl(dst, value)       ; return; }
 1592 }
 1593 
 1594 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
 1595   assert(shift_value > 0, "illegal shift value");
 1596   Label _is_positive;
 1597   testl (reg, reg);
 1598   jcc (Assembler::positive, _is_positive);
 1599   int offset = (1 << shift_value) - 1 ;
 1600 
 1601   if (offset == 1) {
 1602     incrementl(reg);
 1603   } else {
 1604     addl(reg, offset);
 1605   }
 1606 
 1607   bind (_is_positive);
 1608   sarl(reg, shift_value);
 1609 }
 1610 
 1611 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1612   assert(rscratch != noreg || always_reachable(src), "missing");
 1613 
 1614   if (reachable(src)) {
 1615     Assembler::divsd(dst, as_Address(src));
 1616   } else {
 1617     lea(rscratch, src);
 1618     Assembler::divsd(dst, Address(rscratch, 0));
 1619   }
 1620 }
 1621 
 1622 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1623   assert(rscratch != noreg || always_reachable(src), "missing");
 1624 
 1625   if (reachable(src)) {
 1626     Assembler::divss(dst, as_Address(src));
 1627   } else {
 1628     lea(rscratch, src);
 1629     Assembler::divss(dst, Address(rscratch, 0));
 1630   }
 1631 }
 1632 
 1633 void MacroAssembler::enter() {
 1634   push(rbp);
 1635   mov(rbp, rsp);
 1636 }
 1637 
 1638 void MacroAssembler::post_call_nop() {
 1639   if (!Continuations::enabled()) {
 1640     return;
 1641   }
 1642   InstructionMark im(this);
 1643   relocate(post_call_nop_Relocation::spec());
 1644   InlineSkippedInstructionsCounter skipCounter(this);
 1645   emit_int8((uint8_t)0x0f);
 1646   emit_int8((uint8_t)0x1f);
 1647   emit_int8((uint8_t)0x84);
 1648   emit_int8((uint8_t)0x00);
 1649   emit_int32(0x00);
 1650 }
 1651 
 1652 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1653   assert(rscratch != noreg || always_reachable(src), "missing");
 1654   if (reachable(src)) {
 1655     Assembler::mulpd(dst, as_Address(src));
 1656   } else {
 1657     lea(rscratch, src);
 1658     Assembler::mulpd(dst, Address(rscratch, 0));
 1659   }
 1660 }
 1661 
 1662 // dst = c = a * b + c
 1663 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1664   Assembler::vfmadd231sd(c, a, b);
 1665   if (dst != c) {
 1666     movdbl(dst, c);
 1667   }
 1668 }
 1669 
 1670 // dst = c = a * b + c
 1671 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
 1672   Assembler::vfmadd231ss(c, a, b);
 1673   if (dst != c) {
 1674     movflt(dst, c);
 1675   }
 1676 }
 1677 
 1678 // dst = c = a * b + c
 1679 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1680   Assembler::vfmadd231pd(c, a, b, vector_len);
 1681   if (dst != c) {
 1682     vmovdqu(dst, c);
 1683   }
 1684 }
 1685 
 1686 // dst = c = a * b + c
 1687 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
 1688   Assembler::vfmadd231ps(c, a, b, vector_len);
 1689   if (dst != c) {
 1690     vmovdqu(dst, c);
 1691   }
 1692 }
 1693 
 1694 // dst = c = a * b + c
 1695 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1696   Assembler::vfmadd231pd(c, a, b, vector_len);
 1697   if (dst != c) {
 1698     vmovdqu(dst, c);
 1699   }
 1700 }
 1701 
 1702 // dst = c = a * b + c
 1703 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
 1704   Assembler::vfmadd231ps(c, a, b, vector_len);
 1705   if (dst != c) {
 1706     vmovdqu(dst, c);
 1707   }
 1708 }
 1709 
 1710 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
 1711   assert(rscratch != noreg || always_reachable(dst), "missing");
 1712 
 1713   if (reachable(dst)) {
 1714     incrementl(as_Address(dst));
 1715   } else {
 1716     lea(rscratch, dst);
 1717     incrementl(Address(rscratch, 0));
 1718   }
 1719 }
 1720 
 1721 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
 1722   incrementl(as_Address(dst, rscratch));
 1723 }
 1724 
 1725 void MacroAssembler::incrementl(Register reg, int value) {
 1726   if (value == min_jint) {addl(reg, value) ; return; }
 1727   if (value <  0) { decrementl(reg, -value); return; }
 1728   if (value == 0) {                        ; return; }
 1729   if (value == 1 && UseIncDec) { incl(reg) ; return; }
 1730   /* else */      { addl(reg, value)       ; return; }
 1731 }
 1732 
 1733 void MacroAssembler::incrementl(Address dst, int value) {
 1734   if (value == min_jint) {addl(dst, value) ; return; }
 1735   if (value <  0) { decrementl(dst, -value); return; }
 1736   if (value == 0) {                        ; return; }
 1737   if (value == 1 && UseIncDec) { incl(dst) ; return; }
 1738   /* else */      { addl(dst, value)       ; return; }
 1739 }
 1740 
 1741 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
 1742   assert(rscratch != noreg || always_reachable(dst), "missing");
 1743   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
 1744   if (reachable(dst)) {
 1745     jmp_literal(dst.target(), dst.rspec());
 1746   } else {
 1747     lea(rscratch, dst);
 1748     jmp(rscratch);
 1749   }
 1750 }
 1751 
 1752 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
 1753   assert(rscratch != noreg || always_reachable(dst), "missing");
 1754   assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump_cc");
 1755   if (reachable(dst)) {
 1756     InstructionMark im(this);
 1757     relocate(dst.reloc());
 1758     const int short_size = 2;
 1759     const int long_size = 6;
 1760     int offs = (intptr_t)dst.target() - ((intptr_t)pc());
 1761     if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
 1762       // 0111 tttn #8-bit disp
 1763       emit_int8(0x70 | cc);
 1764       emit_int8((offs - short_size) & 0xFF);
 1765     } else {
 1766       // 0000 1111 1000 tttn #32-bit disp
 1767       emit_int8(0x0F);
 1768       emit_int8((unsigned char)(0x80 | cc));
 1769       emit_int32(offs - long_size);
 1770     }
 1771   } else {
 1772 #ifdef ASSERT
 1773     warning("reversing conditional branch");
 1774 #endif /* ASSERT */
 1775     Label skip;
 1776     jccb(reverse[cc], skip);
 1777     lea(rscratch, dst);
 1778     Assembler::jmp(rscratch);
 1779     bind(skip);
 1780   }
 1781 }
 1782 
 1783 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
 1784   ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
 1785   assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
 1786 
 1787   stmxcsr(mxcsr_save);
 1788   movl(tmp, mxcsr_save);
 1789   if (EnableX86ECoreOpts) {
 1790     // The mxcsr_std has status bits set for performance on ECore
 1791     orl(tmp, 0x003f);
 1792   } else {
 1793     // Mask out status bits (only check control and mask bits)
 1794     andl(tmp, 0xFFC0);
 1795   }
 1796   cmp32(tmp, mxcsr_std, rscratch);
 1797 }
 1798 
 1799 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
 1800   assert(rscratch != noreg || always_reachable(src), "missing");
 1801 
 1802   if (reachable(src)) {
 1803     Assembler::ldmxcsr(as_Address(src));
 1804   } else {
 1805     lea(rscratch, src);
 1806     Assembler::ldmxcsr(Address(rscratch, 0));
 1807   }
 1808 }
 1809 
 1810 int MacroAssembler::load_signed_byte(Register dst, Address src) {
 1811   int off = offset();
 1812   movsbl(dst, src); // movsxb
 1813   return off;
 1814 }
 1815 
 1816 // Note: load_signed_short used to be called load_signed_word.
 1817 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
 1818 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
 1819 // The term "word" in HotSpot means a 32- or 64-bit machine word.
 1820 int MacroAssembler::load_signed_short(Register dst, Address src) {
 1821   // This is dubious to me since it seems safe to do a signed 16 => 64 bit
 1822   // version but this is what 64bit has always done. This seems to imply
 1823   // that users are only using 32bits worth.
 1824   int off = offset();
 1825   movswl(dst, src); // movsxw
 1826   return off;
 1827 }
 1828 
 1829 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
 1830   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1831   // and "3.9 Partial Register Penalties", p. 22).
 1832   int off = offset();
 1833   movzbl(dst, src); // movzxb
 1834   return off;
 1835 }
 1836 
 1837 // Note: load_unsigned_short used to be called load_unsigned_word.
 1838 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
 1839   // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
 1840   // and "3.9 Partial Register Penalties", p. 22).
 1841   int off = offset();
 1842   movzwl(dst, src); // movzxw
 1843   return off;
 1844 }
 1845 
 1846 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
 1847   switch (size_in_bytes) {
 1848   case  8:  movq(dst, src); break;
 1849   case  4:  movl(dst, src); break;
 1850   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
 1851   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
 1852   default:  ShouldNotReachHere();
 1853   }
 1854 }
 1855 
 1856 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
 1857   switch (size_in_bytes) {
 1858   case  8:  movq(dst, src); break;
 1859   case  4:  movl(dst, src); break;
 1860   case  2:  movw(dst, src); break;
 1861   case  1:  movb(dst, src); break;
 1862   default:  ShouldNotReachHere();
 1863   }
 1864 }
 1865 
 1866 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
 1867   assert(rscratch != noreg || always_reachable(dst), "missing");
 1868 
 1869   if (reachable(dst)) {
 1870     movl(as_Address(dst), src);
 1871   } else {
 1872     lea(rscratch, dst);
 1873     movl(Address(rscratch, 0), src);
 1874   }
 1875 }
 1876 
 1877 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
 1878   if (reachable(src)) {
 1879     movl(dst, as_Address(src));
 1880   } else {
 1881     lea(dst, src);
 1882     movl(dst, Address(dst, 0));
 1883   }
 1884 }
 1885 
 1886 // C++ bool manipulation
 1887 
 1888 void MacroAssembler::movbool(Register dst, Address src) {
 1889   if(sizeof(bool) == 1)
 1890     movb(dst, src);
 1891   else if(sizeof(bool) == 2)
 1892     movw(dst, src);
 1893   else if(sizeof(bool) == 4)
 1894     movl(dst, src);
 1895   else
 1896     // unsupported
 1897     ShouldNotReachHere();
 1898 }
 1899 
 1900 void MacroAssembler::movbool(Address dst, bool boolconst) {
 1901   if(sizeof(bool) == 1)
 1902     movb(dst, (int) boolconst);
 1903   else if(sizeof(bool) == 2)
 1904     movw(dst, (int) boolconst);
 1905   else if(sizeof(bool) == 4)
 1906     movl(dst, (int) boolconst);
 1907   else
 1908     // unsupported
 1909     ShouldNotReachHere();
 1910 }
 1911 
 1912 void MacroAssembler::movbool(Address dst, Register src) {
 1913   if(sizeof(bool) == 1)
 1914     movb(dst, src);
 1915   else if(sizeof(bool) == 2)
 1916     movw(dst, src);
 1917   else if(sizeof(bool) == 4)
 1918     movl(dst, src);
 1919   else
 1920     // unsupported
 1921     ShouldNotReachHere();
 1922 }
 1923 
 1924 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1925   assert(rscratch != noreg || always_reachable(src), "missing");
 1926 
 1927   if (reachable(src)) {
 1928     movdl(dst, as_Address(src));
 1929   } else {
 1930     lea(rscratch, src);
 1931     movdl(dst, Address(rscratch, 0));
 1932   }
 1933 }
 1934 
 1935 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1936   assert(rscratch != noreg || always_reachable(src), "missing");
 1937 
 1938   if (reachable(src)) {
 1939     movq(dst, as_Address(src));
 1940   } else {
 1941     lea(rscratch, src);
 1942     movq(dst, Address(rscratch, 0));
 1943   }
 1944 }
 1945 
 1946 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1947   assert(rscratch != noreg || always_reachable(src), "missing");
 1948 
 1949   if (reachable(src)) {
 1950     if (UseXmmLoadAndClearUpper) {
 1951       movsd (dst, as_Address(src));
 1952     } else {
 1953       movlpd(dst, as_Address(src));
 1954     }
 1955   } else {
 1956     lea(rscratch, src);
 1957     if (UseXmmLoadAndClearUpper) {
 1958       movsd (dst, Address(rscratch, 0));
 1959     } else {
 1960       movlpd(dst, Address(rscratch, 0));
 1961     }
 1962   }
 1963 }
 1964 
 1965 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
 1966   assert(rscratch != noreg || always_reachable(src), "missing");
 1967 
 1968   if (reachable(src)) {
 1969     movss(dst, as_Address(src));
 1970   } else {
 1971     lea(rscratch, src);
 1972     movss(dst, Address(rscratch, 0));
 1973   }
 1974 }
 1975 
 1976 void MacroAssembler::movptr(Register dst, Register src) {
 1977   movq(dst, src);
 1978 }
 1979 
 1980 void MacroAssembler::movptr(Register dst, Address src) {
 1981   movq(dst, src);
 1982 }
 1983 
 1984 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
 1985 void MacroAssembler::movptr(Register dst, intptr_t src) {
 1986   if (is_uimm32(src)) {
 1987     movl(dst, checked_cast<uint32_t>(src));
 1988   } else if (is_simm32(src)) {
 1989     movq(dst, checked_cast<int32_t>(src));
 1990   } else {
 1991     mov64(dst, src);
 1992   }
 1993 }
 1994 
 1995 void MacroAssembler::movptr(Address dst, Register src) {
 1996   movq(dst, src);
 1997 }
 1998 
 1999 void MacroAssembler::movptr(Address dst, int32_t src) {
 2000   movslq(dst, src);
 2001 }
 2002 
 2003 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
 2004   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2005   Assembler::movdqu(dst, src);
 2006 }
 2007 
 2008 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
 2009   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2010   Assembler::movdqu(dst, src);
 2011 }
 2012 
 2013 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
 2014   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2015   Assembler::movdqu(dst, src);
 2016 }
 2017 
 2018 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2019   assert(rscratch != noreg || always_reachable(src), "missing");
 2020 
 2021   if (reachable(src)) {
 2022     movdqu(dst, as_Address(src));
 2023   } else {
 2024     lea(rscratch, src);
 2025     movdqu(dst, Address(rscratch, 0));
 2026   }
 2027 }
 2028 
 2029 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
 2030   assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2031   Assembler::vmovdqu(dst, src);
 2032 }
 2033 
 2034 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
 2035   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2036   Assembler::vmovdqu(dst, src);
 2037 }
 2038 
 2039 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
 2040   assert(((dst->encoding() < 16  && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 2041   Assembler::vmovdqu(dst, src);
 2042 }
 2043 
 2044 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2045   assert(rscratch != noreg || always_reachable(src), "missing");
 2046 
 2047   if (reachable(src)) {
 2048     vmovdqu(dst, as_Address(src));
 2049   }
 2050   else {
 2051     lea(rscratch, src);
 2052     vmovdqu(dst, Address(rscratch, 0));
 2053   }
 2054 }
 2055 
 2056 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2057   assert(rscratch != noreg || always_reachable(src), "missing");
 2058 
 2059   if (vector_len == AVX_512bit) {
 2060     evmovdquq(dst, src, AVX_512bit, rscratch);
 2061   } else if (vector_len == AVX_256bit) {
 2062     vmovdqu(dst, src, rscratch);
 2063   } else {
 2064     movdqu(dst, src, rscratch);
 2065   }
 2066 }
 2067 
 2068 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
 2069   if (vector_len == AVX_512bit) {
 2070     evmovdquq(dst, src, AVX_512bit);
 2071   } else if (vector_len == AVX_256bit) {
 2072     vmovdqu(dst, src);
 2073   } else {
 2074     movdqu(dst, src);
 2075   }
 2076 }
 2077 
 2078 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
 2079   if (vector_len == AVX_512bit) {
 2080     evmovdquq(dst, src, AVX_512bit);
 2081   } else if (vector_len == AVX_256bit) {
 2082     vmovdqu(dst, src);
 2083   } else {
 2084     movdqu(dst, src);
 2085   }
 2086 }
 2087 
 2088 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
 2089   if (vector_len == AVX_512bit) {
 2090     evmovdquq(dst, src, AVX_512bit);
 2091   } else if (vector_len == AVX_256bit) {
 2092     vmovdqu(dst, src);
 2093   } else {
 2094     movdqu(dst, src);
 2095   }
 2096 }
 2097 
 2098 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2099   assert(rscratch != noreg || always_reachable(src), "missing");
 2100 
 2101   if (reachable(src)) {
 2102     vmovdqa(dst, as_Address(src));
 2103   }
 2104   else {
 2105     lea(rscratch, src);
 2106     vmovdqa(dst, Address(rscratch, 0));
 2107   }
 2108 }
 2109 
 2110 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2111   assert(rscratch != noreg || always_reachable(src), "missing");
 2112 
 2113   if (vector_len == AVX_512bit) {
 2114     evmovdqaq(dst, src, AVX_512bit, rscratch);
 2115   } else if (vector_len == AVX_256bit) {
 2116     vmovdqa(dst, src, rscratch);
 2117   } else {
 2118     movdqa(dst, src, rscratch);
 2119   }
 2120 }
 2121 
 2122 void MacroAssembler::kmov(KRegister dst, Address src) {
 2123   if (VM_Version::supports_avx512bw()) {
 2124     kmovql(dst, src);
 2125   } else {
 2126     assert(VM_Version::supports_evex(), "");
 2127     kmovwl(dst, src);
 2128   }
 2129 }
 2130 
 2131 void MacroAssembler::kmov(Address dst, KRegister src) {
 2132   if (VM_Version::supports_avx512bw()) {
 2133     kmovql(dst, src);
 2134   } else {
 2135     assert(VM_Version::supports_evex(), "");
 2136     kmovwl(dst, src);
 2137   }
 2138 }
 2139 
 2140 void MacroAssembler::kmov(KRegister dst, KRegister src) {
 2141   if (VM_Version::supports_avx512bw()) {
 2142     kmovql(dst, src);
 2143   } else {
 2144     assert(VM_Version::supports_evex(), "");
 2145     kmovwl(dst, src);
 2146   }
 2147 }
 2148 
 2149 void MacroAssembler::kmov(Register dst, KRegister src) {
 2150   if (VM_Version::supports_avx512bw()) {
 2151     kmovql(dst, src);
 2152   } else {
 2153     assert(VM_Version::supports_evex(), "");
 2154     kmovwl(dst, src);
 2155   }
 2156 }
 2157 
 2158 void MacroAssembler::kmov(KRegister dst, Register src) {
 2159   if (VM_Version::supports_avx512bw()) {
 2160     kmovql(dst, src);
 2161   } else {
 2162     assert(VM_Version::supports_evex(), "");
 2163     kmovwl(dst, src);
 2164   }
 2165 }
 2166 
 2167 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
 2168   assert(rscratch != noreg || always_reachable(src), "missing");
 2169 
 2170   if (reachable(src)) {
 2171     kmovql(dst, as_Address(src));
 2172   } else {
 2173     lea(rscratch, src);
 2174     kmovql(dst, Address(rscratch, 0));
 2175   }
 2176 }
 2177 
 2178 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
 2179   assert(rscratch != noreg || always_reachable(src), "missing");
 2180 
 2181   if (reachable(src)) {
 2182     kmovwl(dst, as_Address(src));
 2183   } else {
 2184     lea(rscratch, src);
 2185     kmovwl(dst, Address(rscratch, 0));
 2186   }
 2187 }
 2188 
 2189 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2190                                int vector_len, Register rscratch) {
 2191   assert(rscratch != noreg || always_reachable(src), "missing");
 2192 
 2193   if (reachable(src)) {
 2194     Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
 2195   } else {
 2196     lea(rscratch, src);
 2197     Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
 2198   }
 2199 }
 2200 
 2201 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
 2202                                int vector_len, Register rscratch) {
 2203   assert(rscratch != noreg || always_reachable(src), "missing");
 2204 
 2205   if (reachable(src)) {
 2206     Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
 2207   } else {
 2208     lea(rscratch, src);
 2209     Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
 2210   }
 2211 }
 2212 
 2213 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2214   assert(rscratch != noreg || always_reachable(src), "missing");
 2215 
 2216   if (reachable(src)) {
 2217     Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
 2218   } else {
 2219     lea(rscratch, src);
 2220     Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
 2221   }
 2222 }
 2223 
 2224 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2225   assert(rscratch != noreg || always_reachable(src), "missing");
 2226 
 2227   if (reachable(src)) {
 2228     Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
 2229   } else {
 2230     lea(rscratch, src);
 2231     Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2232   }
 2233 }
 2234 
 2235 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2236   assert(rscratch != noreg || always_reachable(src), "missing");
 2237 
 2238   if (reachable(src)) {
 2239     Assembler::evmovdquq(dst, as_Address(src), vector_len);
 2240   } else {
 2241     lea(rscratch, src);
 2242     Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
 2243   }
 2244 }
 2245 
 2246 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
 2247   assert(rscratch != noreg || always_reachable(src), "missing");
 2248 
 2249   if (reachable(src)) {
 2250     Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
 2251   } else {
 2252     lea(rscratch, src);
 2253     Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
 2254   }
 2255 }
 2256 
 2257 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2258   assert(rscratch != noreg || always_reachable(src), "missing");
 2259 
 2260   if (reachable(src)) {
 2261     Assembler::evmovdqaq(dst, as_Address(src), vector_len);
 2262   } else {
 2263     lea(rscratch, src);
 2264     Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
 2265   }
 2266 }
 2267 
 2268 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2269   assert(rscratch != noreg || always_reachable(src), "missing");
 2270 
 2271   if (reachable(src)) {
 2272     Assembler::movapd(dst, as_Address(src));
 2273   } else {
 2274     lea(rscratch, src);
 2275     Assembler::movapd(dst, Address(rscratch, 0));
 2276   }
 2277 }
 2278 
 2279 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2280   assert(rscratch != noreg || always_reachable(src), "missing");
 2281 
 2282   if (reachable(src)) {
 2283     Assembler::movdqa(dst, as_Address(src));
 2284   } else {
 2285     lea(rscratch, src);
 2286     Assembler::movdqa(dst, Address(rscratch, 0));
 2287   }
 2288 }
 2289 
 2290 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2291   assert(rscratch != noreg || always_reachable(src), "missing");
 2292 
 2293   if (reachable(src)) {
 2294     Assembler::movsd(dst, as_Address(src));
 2295   } else {
 2296     lea(rscratch, src);
 2297     Assembler::movsd(dst, Address(rscratch, 0));
 2298   }
 2299 }
 2300 
 2301 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2302   assert(rscratch != noreg || always_reachable(src), "missing");
 2303 
 2304   if (reachable(src)) {
 2305     Assembler::movss(dst, as_Address(src));
 2306   } else {
 2307     lea(rscratch, src);
 2308     Assembler::movss(dst, Address(rscratch, 0));
 2309   }
 2310 }
 2311 
 2312 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2313   assert(rscratch != noreg || always_reachable(src), "missing");
 2314 
 2315   if (reachable(src)) {
 2316     Assembler::movddup(dst, as_Address(src));
 2317   } else {
 2318     lea(rscratch, src);
 2319     Assembler::movddup(dst, Address(rscratch, 0));
 2320   }
 2321 }
 2322 
 2323 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2324   assert(rscratch != noreg || always_reachable(src), "missing");
 2325 
 2326   if (reachable(src)) {
 2327     Assembler::vmovddup(dst, as_Address(src), vector_len);
 2328   } else {
 2329     lea(rscratch, src);
 2330     Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
 2331   }
 2332 }
 2333 
 2334 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2335   assert(rscratch != noreg || always_reachable(src), "missing");
 2336 
 2337   if (reachable(src)) {
 2338     Assembler::mulsd(dst, as_Address(src));
 2339   } else {
 2340     lea(rscratch, src);
 2341     Assembler::mulsd(dst, Address(rscratch, 0));
 2342   }
 2343 }
 2344 
 2345 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2346   assert(rscratch != noreg || always_reachable(src), "missing");
 2347 
 2348   if (reachable(src)) {
 2349     Assembler::mulss(dst, as_Address(src));
 2350   } else {
 2351     lea(rscratch, src);
 2352     Assembler::mulss(dst, Address(rscratch, 0));
 2353   }
 2354 }
 2355 
 2356 void MacroAssembler::null_check(Register reg, int offset) {
 2357   if (needs_explicit_null_check(offset)) {
 2358     // provoke OS null exception if reg is null by
 2359     // accessing M[reg] w/o changing any (non-CC) registers
 2360     // NOTE: cmpl is plenty here to provoke a segv
 2361     cmpptr(rax, Address(reg, 0));
 2362     // Note: should probably use testl(rax, Address(reg, 0));
 2363     //       may be shorter code (however, this version of
 2364     //       testl needs to be implemented first)
 2365   } else {
 2366     // nothing to do, (later) access of M[reg + offset]
 2367     // will provoke OS null exception if reg is null
 2368   }
 2369 }
 2370 
 2371 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
 2372   andptr(markword, markWord::inline_type_mask_in_place);
 2373   cmpptr(markword, markWord::inline_type_pattern);
 2374   jcc(Assembler::equal, is_inline_type);
 2375 }
 2376 
 2377 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
 2378   if (can_be_null) {
 2379     testptr(object, object);
 2380     jcc(Assembler::zero, not_inline_type);
 2381   }
 2382   const int is_inline_type_mask = markWord::inline_type_pattern;
 2383   movptr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
 2384   andptr(tmp, is_inline_type_mask);
 2385   cmpptr(tmp, is_inline_type_mask);
 2386   jcc(Assembler::notEqual, not_inline_type);
 2387 }
 2388 
 2389 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
 2390   movl(temp_reg, flags);
 2391   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2392   jcc(Assembler::notEqual, is_null_free_inline_type);
 2393 }
 2394 
 2395 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
 2396   movl(temp_reg, flags);
 2397   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2398   jcc(Assembler::equal, not_null_free_inline_type);
 2399 }
 2400 
 2401 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
 2402   movl(temp_reg, flags);
 2403   testl(temp_reg, 1 << ResolvedFieldEntry::is_flat_shift);
 2404   jcc(Assembler::notEqual, is_flat);
 2405 }
 2406 
 2407 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
 2408   movl(temp_reg, flags);
 2409   testl(temp_reg, 1 << ResolvedFieldEntry::has_null_marker_shift);
 2410   jcc(Assembler::notEqual, has_null_marker);
 2411 }
 2412 
 2413 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
 2414   Label test_mark_word;
 2415   // load mark word
 2416   movptr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
 2417   // check displaced
 2418   testl(temp_reg, markWord::unlocked_value);
 2419   jccb(Assembler::notZero, test_mark_word);
 2420   // slow path use klass prototype
 2421   push(rscratch1);
 2422   load_prototype_header(temp_reg, oop, rscratch1);
 2423   pop(rscratch1);
 2424 
 2425   bind(test_mark_word);
 2426   testl(temp_reg, test_bit);
 2427   jcc((jmp_set) ? Assembler::notZero : Assembler::zero, jmp_label);
 2428 }
 2429 
 2430 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg,
 2431                                          Label& is_flat_array) {
 2432 #ifdef _LP64
 2433   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
 2434 #else
 2435   load_klass(temp_reg, oop, noreg);
 2436   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 2437   test_flat_array_layout(temp_reg, is_flat_array);
 2438 #endif
 2439 }
 2440 
 2441 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
 2442                                              Label& is_non_flat_array) {
 2443 #ifdef _LP64
 2444   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
 2445 #else
 2446   load_klass(temp_reg, oop, noreg);
 2447   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 2448   test_non_flat_array_layout(temp_reg, is_non_flat_array);
 2449 #endif
 2450 }
 2451 
 2452 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label&is_null_free_array) {
 2453 #ifdef _LP64
 2454   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
 2455 #else
 2456   Unimplemented();
 2457 #endif
 2458 }
 2459 
 2460 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
 2461 #ifdef _LP64
 2462   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
 2463 #else
 2464   Unimplemented();
 2465 #endif
 2466 }
 2467 
 2468 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
 2469   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2470   jcc(Assembler::notZero, is_flat_array);
 2471 }
 2472 
 2473 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
 2474   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2475   jcc(Assembler::zero, is_non_flat_array);
 2476 }
 2477 
 2478 void MacroAssembler::os_breakpoint() {
 2479   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2480   // (e.g., MSVC can't call ps() otherwise)
 2481   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2482 }
 2483 
 2484 void MacroAssembler::unimplemented(const char* what) {
 2485   const char* buf = nullptr;
 2486   {
 2487     ResourceMark rm;
 2488     stringStream ss;
 2489     ss.print("unimplemented: %s", what);
 2490     buf = code_string(ss.as_string());
 2491   }
 2492   stop(buf);
 2493 }
 2494 
 2495 #define XSTATE_BV 0x200
 2496 
 2497 void MacroAssembler::pop_CPU_state() {
 2498   pop_FPU_state();
 2499   pop_IU_state();
 2500 }
 2501 
 2502 void MacroAssembler::pop_FPU_state() {
 2503   fxrstor(Address(rsp, 0));
 2504   addptr(rsp, FPUStateSizeInWords * wordSize);
 2505 }
 2506 
 2507 void MacroAssembler::pop_IU_state() {
 2508   popa();
 2509   addq(rsp, 8);
 2510   popf();
 2511 }
 2512 
 2513 // Save Integer and Float state
 2514 // Warning: Stack must be 16 byte aligned (64bit)
 2515 void MacroAssembler::push_CPU_state() {
 2516   push_IU_state();
 2517   push_FPU_state();
 2518 }
 2519 
 2520 void MacroAssembler::push_FPU_state() {
 2521   subptr(rsp, FPUStateSizeInWords * wordSize);
 2522   fxsave(Address(rsp, 0));
 2523 }
 2524 
 2525 void MacroAssembler::push_IU_state() {
 2526   // Push flags first because pusha kills them
 2527   pushf();
 2528   // Make sure rsp stays 16-byte aligned
 2529   subq(rsp, 8);
 2530   pusha();
 2531 }
 2532 
 2533 void MacroAssembler::push_cont_fastpath() {
 2534   if (!Continuations::enabled()) return;
 2535 
 2536   Label L_done;
 2537   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2538   jccb(Assembler::belowEqual, L_done);
 2539   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
 2540   bind(L_done);
 2541 }
 2542 
 2543 void MacroAssembler::pop_cont_fastpath() {
 2544   if (!Continuations::enabled()) return;
 2545 
 2546   Label L_done;
 2547   cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
 2548   jccb(Assembler::below, L_done);
 2549   movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
 2550   bind(L_done);
 2551 }
 2552 
 2553 #ifdef ASSERT
 2554 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
 2555   Label no_cont;
 2556   movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
 2557   testl(cont, cont);
 2558   jcc(Assembler::zero, no_cont);
 2559   stop(name);
 2560   bind(no_cont);
 2561 }
 2562 #endif
 2563 
 2564 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
 2565   // we must set sp to zero to clear frame
 2566   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
 2567   // must clear fp, so that compiled frames are not confused; it is
 2568   // possible that we need it only for debugging
 2569   if (clear_fp) {
 2570     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
 2571   }
 2572   // Always clear the pc because it could have been set by make_walkable()
 2573   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
 2574   vzeroupper();
 2575 }
 2576 
 2577 void MacroAssembler::round_to(Register reg, int modulus) {
 2578   addptr(reg, modulus - 1);
 2579   andptr(reg, -modulus);
 2580 }
 2581 
 2582 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
 2583   if (at_return) {
 2584     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 2585     // we may safely use rsp instead to perform the stack watermark check.
 2586     cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
 2587     jcc(Assembler::above, slow_path);
 2588     return;
 2589   }
 2590   testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 2591   jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
 2592 }
 2593 
 2594 // Calls to C land
 2595 //
 2596 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
 2597 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 2598 // has to be reset to 0. This is required to allow proper stack traversal.
 2599 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2600                                          Register last_java_fp,
 2601                                          address  last_java_pc,
 2602                                          Register rscratch) {
 2603   vzeroupper();
 2604   // determine last_java_sp register
 2605   if (!last_java_sp->is_valid()) {
 2606     last_java_sp = rsp;
 2607   }
 2608   // last_java_fp is optional
 2609   if (last_java_fp->is_valid()) {
 2610     movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
 2611   }
 2612   // last_java_pc is optional
 2613   if (last_java_pc != nullptr) {
 2614     Address java_pc(r15_thread,
 2615                     JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
 2616     lea(java_pc, InternalAddress(last_java_pc), rscratch);
 2617   }
 2618   movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
 2619 }
 2620 
 2621 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 2622                                          Register last_java_fp,
 2623                                          Label &L,
 2624                                          Register scratch) {
 2625   lea(scratch, L);
 2626   movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
 2627   set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
 2628 }
 2629 
 2630 void MacroAssembler::shlptr(Register dst, int imm8) {
 2631   shlq(dst, imm8);
 2632 }
 2633 
 2634 void MacroAssembler::shrptr(Register dst, int imm8) {
 2635   shrq(dst, imm8);
 2636 }
 2637 
 2638 void MacroAssembler::sign_extend_byte(Register reg) {
 2639   movsbl(reg, reg); // movsxb
 2640 }
 2641 
 2642 void MacroAssembler::sign_extend_short(Register reg) {
 2643   movswl(reg, reg); // movsxw
 2644 }
 2645 
 2646 void MacroAssembler::testl(Address dst, int32_t imm32) {
 2647   if (imm32 >= 0 && is8bit(imm32)) {
 2648     testb(dst, imm32);
 2649   } else {
 2650     Assembler::testl(dst, imm32);
 2651   }
 2652 }
 2653 
 2654 void MacroAssembler::testl(Register dst, int32_t imm32) {
 2655   if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
 2656     testb(dst, imm32);
 2657   } else {
 2658     Assembler::testl(dst, imm32);
 2659   }
 2660 }
 2661 
 2662 void MacroAssembler::testl(Register dst, AddressLiteral src) {
 2663   assert(always_reachable(src), "Address should be reachable");
 2664   testl(dst, as_Address(src));
 2665 }
 2666 
 2667 void MacroAssembler::testq(Address dst, int32_t imm32) {
 2668   if (imm32 >= 0) {
 2669     testl(dst, imm32);
 2670   } else {
 2671     Assembler::testq(dst, imm32);
 2672   }
 2673 }
 2674 
 2675 void MacroAssembler::testq(Register dst, int32_t imm32) {
 2676   if (imm32 >= 0) {
 2677     testl(dst, imm32);
 2678   } else {
 2679     Assembler::testq(dst, imm32);
 2680   }
 2681 }
 2682 
 2683 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
 2684   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2685   Assembler::pcmpeqb(dst, src);
 2686 }
 2687 
 2688 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
 2689   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2690   Assembler::pcmpeqw(dst, src);
 2691 }
 2692 
 2693 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
 2694   assert((dst->encoding() < 16),"XMM register should be 0-15");
 2695   Assembler::pcmpestri(dst, src, imm8);
 2696 }
 2697 
 2698 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
 2699   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2700   Assembler::pcmpestri(dst, src, imm8);
 2701 }
 2702 
 2703 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
 2704   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2705   Assembler::pmovzxbw(dst, src);
 2706 }
 2707 
 2708 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
 2709   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2710   Assembler::pmovzxbw(dst, src);
 2711 }
 2712 
 2713 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
 2714   assert((src->encoding() < 16),"XMM register should be 0-15");
 2715   Assembler::pmovmskb(dst, src);
 2716 }
 2717 
 2718 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
 2719   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 2720   Assembler::ptest(dst, src);
 2721 }
 2722 
 2723 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2724   assert(rscratch != noreg || always_reachable(src), "missing");
 2725 
 2726   if (reachable(src)) {
 2727     Assembler::sqrtss(dst, as_Address(src));
 2728   } else {
 2729     lea(rscratch, src);
 2730     Assembler::sqrtss(dst, Address(rscratch, 0));
 2731   }
 2732 }
 2733 
 2734 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2735   assert(rscratch != noreg || always_reachable(src), "missing");
 2736 
 2737   if (reachable(src)) {
 2738     Assembler::subsd(dst, as_Address(src));
 2739   } else {
 2740     lea(rscratch, src);
 2741     Assembler::subsd(dst, Address(rscratch, 0));
 2742   }
 2743 }
 2744 
 2745 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
 2746   assert(rscratch != noreg || always_reachable(src), "missing");
 2747 
 2748   if (reachable(src)) {
 2749     Assembler::roundsd(dst, as_Address(src), rmode);
 2750   } else {
 2751     lea(rscratch, src);
 2752     Assembler::roundsd(dst, Address(rscratch, 0), rmode);
 2753   }
 2754 }
 2755 
 2756 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2757   assert(rscratch != noreg || always_reachable(src), "missing");
 2758 
 2759   if (reachable(src)) {
 2760     Assembler::subss(dst, as_Address(src));
 2761   } else {
 2762     lea(rscratch, src);
 2763     Assembler::subss(dst, Address(rscratch, 0));
 2764   }
 2765 }
 2766 
 2767 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2768   assert(rscratch != noreg || always_reachable(src), "missing");
 2769 
 2770   if (reachable(src)) {
 2771     Assembler::ucomisd(dst, as_Address(src));
 2772   } else {
 2773     lea(rscratch, src);
 2774     Assembler::ucomisd(dst, Address(rscratch, 0));
 2775   }
 2776 }
 2777 
 2778 void MacroAssembler::vucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2779   assert(rscratch != noreg || always_reachable(src), "missing");
 2780 
 2781   if (reachable(src)) {
 2782     Assembler::vucomxsd(dst, as_Address(src));
 2783   } else {
 2784     lea(rscratch, src);
 2785     Assembler::vucomxsd(dst, Address(rscratch, 0));
 2786   }
 2787 }
 2788 
 2789 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2790   assert(rscratch != noreg || always_reachable(src), "missing");
 2791 
 2792   if (reachable(src)) {
 2793     Assembler::ucomiss(dst, as_Address(src));
 2794   } else {
 2795     lea(rscratch, src);
 2796     Assembler::ucomiss(dst, Address(rscratch, 0));
 2797   }
 2798 }
 2799 
 2800 void MacroAssembler::vucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2801   assert(rscratch != noreg || always_reachable(src), "missing");
 2802 
 2803   if (reachable(src)) {
 2804     Assembler::vucomxss(dst, as_Address(src));
 2805   } else {
 2806     lea(rscratch, src);
 2807     Assembler::vucomxss(dst, Address(rscratch, 0));
 2808   }
 2809 }
 2810 
 2811 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2812   assert(rscratch != noreg || always_reachable(src), "missing");
 2813 
 2814   // Used in sign-bit flipping with aligned address.
 2815   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2816 
 2817   if (UseAVX > 2 &&
 2818       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2819       (dst->encoding() >= 16)) {
 2820     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2821   } else if (reachable(src)) {
 2822     Assembler::xorpd(dst, as_Address(src));
 2823   } else {
 2824     lea(rscratch, src);
 2825     Assembler::xorpd(dst, Address(rscratch, 0));
 2826   }
 2827 }
 2828 
 2829 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
 2830   if (UseAVX > 2 &&
 2831       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2832       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2833     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2834   } else {
 2835     Assembler::xorpd(dst, src);
 2836   }
 2837 }
 2838 
 2839 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
 2840   if (UseAVX > 2 &&
 2841       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2842       ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
 2843     Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
 2844   } else {
 2845     Assembler::xorps(dst, src);
 2846   }
 2847 }
 2848 
 2849 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2850   assert(rscratch != noreg || always_reachable(src), "missing");
 2851 
 2852   // Used in sign-bit flipping with aligned address.
 2853   assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
 2854 
 2855   if (UseAVX > 2 &&
 2856       (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
 2857       (dst->encoding() >= 16)) {
 2858     vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
 2859   } else if (reachable(src)) {
 2860     Assembler::xorps(dst, as_Address(src));
 2861   } else {
 2862     lea(rscratch, src);
 2863     Assembler::xorps(dst, Address(rscratch, 0));
 2864   }
 2865 }
 2866 
 2867 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
 2868   assert(rscratch != noreg || always_reachable(src), "missing");
 2869 
 2870   // Used in sign-bit flipping with aligned address.
 2871   bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
 2872   assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
 2873   if (reachable(src)) {
 2874     Assembler::pshufb(dst, as_Address(src));
 2875   } else {
 2876     lea(rscratch, src);
 2877     Assembler::pshufb(dst, Address(rscratch, 0));
 2878   }
 2879 }
 2880 
 2881 // AVX 3-operands instructions
 2882 
 2883 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2884   assert(rscratch != noreg || always_reachable(src), "missing");
 2885 
 2886   if (reachable(src)) {
 2887     vaddsd(dst, nds, as_Address(src));
 2888   } else {
 2889     lea(rscratch, src);
 2890     vaddsd(dst, nds, Address(rscratch, 0));
 2891   }
 2892 }
 2893 
 2894 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 2895   assert(rscratch != noreg || always_reachable(src), "missing");
 2896 
 2897   if (reachable(src)) {
 2898     vaddss(dst, nds, as_Address(src));
 2899   } else {
 2900     lea(rscratch, src);
 2901     vaddss(dst, nds, Address(rscratch, 0));
 2902   }
 2903 }
 2904 
 2905 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2906   assert(UseAVX > 0, "requires some form of AVX");
 2907   assert(rscratch != noreg || always_reachable(src), "missing");
 2908 
 2909   if (reachable(src)) {
 2910     Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
 2911   } else {
 2912     lea(rscratch, src);
 2913     Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
 2914   }
 2915 }
 2916 
 2917 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2918   assert(UseAVX > 0, "requires some form of AVX");
 2919   assert(rscratch != noreg || always_reachable(src), "missing");
 2920 
 2921   if (reachable(src)) {
 2922     Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
 2923   } else {
 2924     lea(rscratch, src);
 2925     Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
 2926   }
 2927 }
 2928 
 2929 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2930   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2931   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2932 
 2933   vandps(dst, nds, negate_field, vector_len, rscratch);
 2934 }
 2935 
 2936 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
 2937   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 2938   assert(rscratch != noreg || always_reachable(negate_field), "missing");
 2939 
 2940   vandpd(dst, nds, negate_field, vector_len, rscratch);
 2941 }
 2942 
 2943 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2944   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2945   Assembler::vpaddb(dst, nds, src, vector_len);
 2946 }
 2947 
 2948 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2949   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2950   Assembler::vpaddb(dst, nds, src, vector_len);
 2951 }
 2952 
 2953 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 2954   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2955   Assembler::vpaddw(dst, nds, src, vector_len);
 2956 }
 2957 
 2958 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 2959   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 2960   Assembler::vpaddw(dst, nds, src, vector_len);
 2961 }
 2962 
 2963 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 2964   assert(rscratch != noreg || always_reachable(src), "missing");
 2965 
 2966   if (reachable(src)) {
 2967     Assembler::vpand(dst, nds, as_Address(src), vector_len);
 2968   } else {
 2969     lea(rscratch, src);
 2970     Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
 2971   }
 2972 }
 2973 
 2974 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2975   assert(rscratch != noreg || always_reachable(src), "missing");
 2976 
 2977   if (reachable(src)) {
 2978     Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
 2979   } else {
 2980     lea(rscratch, src);
 2981     Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
 2982   }
 2983 }
 2984 
 2985 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2986   assert(rscratch != noreg || always_reachable(src), "missing");
 2987 
 2988   if (reachable(src)) {
 2989     Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
 2990   } else {
 2991     lea(rscratch, src);
 2992     Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
 2993   }
 2994 }
 2995 
 2996 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 2997   assert(rscratch != noreg || always_reachable(src), "missing");
 2998 
 2999   if (reachable(src)) {
 3000     Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
 3001   } else {
 3002     lea(rscratch, src);
 3003     Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
 3004   }
 3005 }
 3006 
 3007 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3008   assert(rscratch != noreg || always_reachable(src), "missing");
 3009 
 3010   if (reachable(src)) {
 3011     Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
 3012   } else {
 3013     lea(rscratch, src);
 3014     Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
 3015   }
 3016 }
 3017 
 3018 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
 3019   assert(rscratch != noreg || always_reachable(src), "missing");
 3020 
 3021   if (reachable(src)) {
 3022     Assembler::vbroadcastss(dst, as_Address(src), vector_len);
 3023   } else {
 3024     lea(rscratch, src);
 3025     Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
 3026   }
 3027 }
 3028 
 3029 // Vector float blend
 3030 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3031 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3032   // WARN: Allow dst == (src1|src2), mask == scratch
 3033   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3034                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3035   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
 3036   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3037   if (blend_emulation && scratch_available && dst_available) {
 3038     if (compute_mask) {
 3039       vpsrad(scratch, mask, 32, vector_len);
 3040       mask = scratch;
 3041     }
 3042     if (dst == src1) {
 3043       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src1
 3044       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3045     } else {
 3046       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3047       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
 3048     }
 3049     vpor(dst, dst, scratch, vector_len);
 3050   } else {
 3051     Assembler::vblendvps(dst, src1, src2, mask, vector_len);
 3052   }
 3053 }
 3054 
 3055 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
 3056 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
 3057   // WARN: Allow dst == (src1|src2), mask == scratch
 3058   bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1 &&
 3059                          !(VM_Version::is_intel_darkmont() && (dst == src1)); // partially fixed on Darkmont
 3060   bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
 3061   bool dst_available = dst != mask && (dst != src1 || dst != src2);
 3062   if (blend_emulation && scratch_available && dst_available) {
 3063     if (compute_mask) {
 3064       vpxor(scratch, scratch, scratch, vector_len);
 3065       vpcmpgtq(scratch, scratch, mask, vector_len);
 3066       mask = scratch;
 3067     }
 3068     if (dst == src1) {
 3069       vpandn(dst,     mask, src1, vector_len); // if mask == 0, src
 3070       vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
 3071     } else {
 3072       vpand (dst,     mask, src2, vector_len); // if mask == 1, src2
 3073       vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
 3074     }
 3075     vpor(dst, dst, scratch, vector_len);
 3076   } else {
 3077     Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
 3078   }
 3079 }
 3080 
 3081 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3082   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3083   Assembler::vpcmpeqb(dst, nds, src, vector_len);
 3084 }
 3085 
 3086 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
 3087   assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3088   Assembler::vpcmpeqb(dst, src1, src2, vector_len);
 3089 }
 3090 
 3091 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3092   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3093   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3094 }
 3095 
 3096 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3097   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3098   Assembler::vpcmpeqw(dst, nds, src, vector_len);
 3099 }
 3100 
 3101 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3102   assert(rscratch != noreg || always_reachable(src), "missing");
 3103 
 3104   if (reachable(src)) {
 3105     Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
 3106   } else {
 3107     lea(rscratch, src);
 3108     Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
 3109   }
 3110 }
 3111 
 3112 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3113                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3114   assert(rscratch != noreg || always_reachable(src), "missing");
 3115 
 3116   if (reachable(src)) {
 3117     Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3118   } else {
 3119     lea(rscratch, src);
 3120     Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3121   }
 3122 }
 3123 
 3124 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3125                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3126   assert(rscratch != noreg || always_reachable(src), "missing");
 3127 
 3128   if (reachable(src)) {
 3129     Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3130   } else {
 3131     lea(rscratch, src);
 3132     Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3133   }
 3134 }
 3135 
 3136 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3137                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3138   assert(rscratch != noreg || always_reachable(src), "missing");
 3139 
 3140   if (reachable(src)) {
 3141     Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3142   } else {
 3143     lea(rscratch, src);
 3144     Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3145   }
 3146 }
 3147 
 3148 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3149                              int comparison, bool is_signed, int vector_len, Register rscratch) {
 3150   assert(rscratch != noreg || always_reachable(src), "missing");
 3151 
 3152   if (reachable(src)) {
 3153     Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
 3154   } else {
 3155     lea(rscratch, src);
 3156     Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
 3157   }
 3158 }
 3159 
 3160 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
 3161   if (width == Assembler::Q) {
 3162     Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
 3163   } else {
 3164     Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
 3165   }
 3166 }
 3167 
 3168 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
 3169   int eq_cond_enc = 0x29;
 3170   int gt_cond_enc = 0x37;
 3171   if (width != Assembler::Q) {
 3172     eq_cond_enc = 0x74 + width;
 3173     gt_cond_enc = 0x64 + width;
 3174   }
 3175   switch (cond) {
 3176   case eq:
 3177     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3178     break;
 3179   case neq:
 3180     vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
 3181     vallones(xtmp, vector_len);
 3182     vpxor(dst, xtmp, dst, vector_len);
 3183     break;
 3184   case le:
 3185     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3186     vallones(xtmp, vector_len);
 3187     vpxor(dst, xtmp, dst, vector_len);
 3188     break;
 3189   case nlt:
 3190     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3191     vallones(xtmp, vector_len);
 3192     vpxor(dst, xtmp, dst, vector_len);
 3193     break;
 3194   case lt:
 3195     vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
 3196     break;
 3197   case nle:
 3198     vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
 3199     break;
 3200   default:
 3201     assert(false, "Should not reach here");
 3202   }
 3203 }
 3204 
 3205 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
 3206   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3207   Assembler::vpmovzxbw(dst, src, vector_len);
 3208 }
 3209 
 3210 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
 3211   assert((src->encoding() < 16),"XMM register should be 0-15");
 3212   Assembler::vpmovmskb(dst, src, vector_len);
 3213 }
 3214 
 3215 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3216   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3217   Assembler::vpmullw(dst, nds, src, vector_len);
 3218 }
 3219 
 3220 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3221   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3222   Assembler::vpmullw(dst, nds, src, vector_len);
 3223 }
 3224 
 3225 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3226   assert((UseAVX > 0), "AVX support is needed");
 3227   assert(rscratch != noreg || always_reachable(src), "missing");
 3228 
 3229   if (reachable(src)) {
 3230     Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
 3231   } else {
 3232     lea(rscratch, src);
 3233     Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
 3234   }
 3235 }
 3236 
 3237 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3238   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3239   Assembler::vpsubb(dst, nds, src, vector_len);
 3240 }
 3241 
 3242 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3243   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3244   Assembler::vpsubb(dst, nds, src, vector_len);
 3245 }
 3246 
 3247 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
 3248   assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3249   Assembler::vpsubw(dst, nds, src, vector_len);
 3250 }
 3251 
 3252 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
 3253   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3254   Assembler::vpsubw(dst, nds, src, vector_len);
 3255 }
 3256 
 3257 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3258   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3259   Assembler::vpsraw(dst, nds, shift, vector_len);
 3260 }
 3261 
 3262 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3263   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3264   Assembler::vpsraw(dst, nds, shift, vector_len);
 3265 }
 3266 
 3267 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3268   assert(UseAVX > 2,"");
 3269   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3270      vector_len = 2;
 3271   }
 3272   Assembler::evpsraq(dst, nds, shift, vector_len);
 3273 }
 3274 
 3275 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3276   assert(UseAVX > 2,"");
 3277   if (!VM_Version::supports_avx512vl() && vector_len < 2) {
 3278      vector_len = 2;
 3279   }
 3280   Assembler::evpsraq(dst, nds, shift, vector_len);
 3281 }
 3282 
 3283 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3284   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3285   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3286 }
 3287 
 3288 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3289   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3290   Assembler::vpsrlw(dst, nds, shift, vector_len);
 3291 }
 3292 
 3293 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
 3294   assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3295   Assembler::vpsllw(dst, nds, shift, vector_len);
 3296 }
 3297 
 3298 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
 3299   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3300   Assembler::vpsllw(dst, nds, shift, vector_len);
 3301 }
 3302 
 3303 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
 3304   assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
 3305   Assembler::vptest(dst, src);
 3306 }
 3307 
 3308 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
 3309   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3310   Assembler::punpcklbw(dst, src);
 3311 }
 3312 
 3313 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
 3314   assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
 3315   Assembler::pshufd(dst, src, mode);
 3316 }
 3317 
 3318 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
 3319   assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
 3320   Assembler::pshuflw(dst, src, mode);
 3321 }
 3322 
 3323 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3324   assert(rscratch != noreg || always_reachable(src), "missing");
 3325 
 3326   if (reachable(src)) {
 3327     vandpd(dst, nds, as_Address(src), vector_len);
 3328   } else {
 3329     lea(rscratch, src);
 3330     vandpd(dst, nds, Address(rscratch, 0), vector_len);
 3331   }
 3332 }
 3333 
 3334 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3335   assert(rscratch != noreg || always_reachable(src), "missing");
 3336 
 3337   if (reachable(src)) {
 3338     vandps(dst, nds, as_Address(src), vector_len);
 3339   } else {
 3340     lea(rscratch, src);
 3341     vandps(dst, nds, Address(rscratch, 0), vector_len);
 3342   }
 3343 }
 3344 
 3345 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
 3346                             bool merge, int vector_len, Register rscratch) {
 3347   assert(rscratch != noreg || always_reachable(src), "missing");
 3348 
 3349   if (reachable(src)) {
 3350     Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
 3351   } else {
 3352     lea(rscratch, src);
 3353     Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
 3354   }
 3355 }
 3356 
 3357 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3358   assert(rscratch != noreg || always_reachable(src), "missing");
 3359 
 3360   if (reachable(src)) {
 3361     vdivsd(dst, nds, as_Address(src));
 3362   } else {
 3363     lea(rscratch, src);
 3364     vdivsd(dst, nds, Address(rscratch, 0));
 3365   }
 3366 }
 3367 
 3368 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3369   assert(rscratch != noreg || always_reachable(src), "missing");
 3370 
 3371   if (reachable(src)) {
 3372     vdivss(dst, nds, as_Address(src));
 3373   } else {
 3374     lea(rscratch, src);
 3375     vdivss(dst, nds, Address(rscratch, 0));
 3376   }
 3377 }
 3378 
 3379 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3380   assert(rscratch != noreg || always_reachable(src), "missing");
 3381 
 3382   if (reachable(src)) {
 3383     vmulsd(dst, nds, as_Address(src));
 3384   } else {
 3385     lea(rscratch, src);
 3386     vmulsd(dst, nds, Address(rscratch, 0));
 3387   }
 3388 }
 3389 
 3390 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3391   assert(rscratch != noreg || always_reachable(src), "missing");
 3392 
 3393   if (reachable(src)) {
 3394     vmulss(dst, nds, as_Address(src));
 3395   } else {
 3396     lea(rscratch, src);
 3397     vmulss(dst, nds, Address(rscratch, 0));
 3398   }
 3399 }
 3400 
 3401 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3402   assert(rscratch != noreg || always_reachable(src), "missing");
 3403 
 3404   if (reachable(src)) {
 3405     vsubsd(dst, nds, as_Address(src));
 3406   } else {
 3407     lea(rscratch, src);
 3408     vsubsd(dst, nds, Address(rscratch, 0));
 3409   }
 3410 }
 3411 
 3412 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3413   assert(rscratch != noreg || always_reachable(src), "missing");
 3414 
 3415   if (reachable(src)) {
 3416     vsubss(dst, nds, as_Address(src));
 3417   } else {
 3418     lea(rscratch, src);
 3419     vsubss(dst, nds, Address(rscratch, 0));
 3420   }
 3421 }
 3422 
 3423 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3424   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3425   assert(rscratch != noreg || always_reachable(src), "missing");
 3426 
 3427   vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3428 }
 3429 
 3430 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
 3431   assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
 3432   assert(rscratch != noreg || always_reachable(src), "missing");
 3433 
 3434   vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
 3435 }
 3436 
 3437 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3438   assert(rscratch != noreg || always_reachable(src), "missing");
 3439 
 3440   if (reachable(src)) {
 3441     vxorpd(dst, nds, as_Address(src), vector_len);
 3442   } else {
 3443     lea(rscratch, src);
 3444     vxorpd(dst, nds, Address(rscratch, 0), vector_len);
 3445   }
 3446 }
 3447 
 3448 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3449   assert(rscratch != noreg || always_reachable(src), "missing");
 3450 
 3451   if (reachable(src)) {
 3452     vxorps(dst, nds, as_Address(src), vector_len);
 3453   } else {
 3454     lea(rscratch, src);
 3455     vxorps(dst, nds, Address(rscratch, 0), vector_len);
 3456   }
 3457 }
 3458 
 3459 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3460   assert(rscratch != noreg || always_reachable(src), "missing");
 3461 
 3462   if (UseAVX > 1 || (vector_len < 1)) {
 3463     if (reachable(src)) {
 3464       Assembler::vpxor(dst, nds, as_Address(src), vector_len);
 3465     } else {
 3466       lea(rscratch, src);
 3467       Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
 3468     }
 3469   } else {
 3470     MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
 3471   }
 3472 }
 3473 
 3474 void MacroAssembler::vpermd(XMMRegister dst,  XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
 3475   assert(rscratch != noreg || always_reachable(src), "missing");
 3476 
 3477   if (reachable(src)) {
 3478     Assembler::vpermd(dst, nds, as_Address(src), vector_len);
 3479   } else {
 3480     lea(rscratch, src);
 3481     Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
 3482   }
 3483 }
 3484 
 3485 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
 3486   const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
 3487   STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
 3488   // The inverted mask is sign-extended
 3489   andptr(possibly_non_local, inverted_mask);
 3490 }
 3491 
 3492 void MacroAssembler::resolve_jobject(Register value,
 3493                                      Register tmp) {
 3494   Register thread = r15_thread;
 3495   assert_different_registers(value, thread, tmp);
 3496   Label done, tagged, weak_tagged;
 3497   testptr(value, value);
 3498   jcc(Assembler::zero, done);           // Use null as-is.
 3499   testptr(value, JNIHandles::tag_mask); // Test for tag.
 3500   jcc(Assembler::notZero, tagged);
 3501 
 3502   // Resolve local handle
 3503   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
 3504   verify_oop(value);
 3505   jmp(done);
 3506 
 3507   bind(tagged);
 3508   testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
 3509   jcc(Assembler::notZero, weak_tagged);
 3510 
 3511   // Resolve global handle
 3512   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3513   verify_oop(value);
 3514   jmp(done);
 3515 
 3516   bind(weak_tagged);
 3517   // Resolve jweak.
 3518   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 3519                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
 3520   verify_oop(value);
 3521 
 3522   bind(done);
 3523 }
 3524 
 3525 void MacroAssembler::resolve_global_jobject(Register value,
 3526                                             Register tmp) {
 3527   Register thread = r15_thread;
 3528   assert_different_registers(value, thread, tmp);
 3529   Label done;
 3530 
 3531   testptr(value, value);
 3532   jcc(Assembler::zero, done);           // Use null as-is.
 3533 
 3534 #ifdef ASSERT
 3535   {
 3536     Label valid_global_tag;
 3537     testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
 3538     jcc(Assembler::notZero, valid_global_tag);
 3539     stop("non global jobject using resolve_global_jobject");
 3540     bind(valid_global_tag);
 3541   }
 3542 #endif
 3543 
 3544   // Resolve global handle
 3545   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
 3546   verify_oop(value);
 3547 
 3548   bind(done);
 3549 }
 3550 
 3551 void MacroAssembler::subptr(Register dst, int32_t imm32) {
 3552   subq(dst, imm32);
 3553 }
 3554 
 3555 // Force generation of a 4 byte immediate value even if it fits into 8bit
 3556 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
 3557   subq_imm32(dst, imm32);
 3558 }
 3559 
 3560 void MacroAssembler::subptr(Register dst, Register src) {
 3561   subq(dst, src);
 3562 }
 3563 
 3564 // C++ bool manipulation
 3565 void MacroAssembler::testbool(Register dst) {
 3566   if(sizeof(bool) == 1)
 3567     testb(dst, 0xff);
 3568   else if(sizeof(bool) == 2) {
 3569     // testw implementation needed for two byte bools
 3570     ShouldNotReachHere();
 3571   } else if(sizeof(bool) == 4)
 3572     testl(dst, dst);
 3573   else
 3574     // unsupported
 3575     ShouldNotReachHere();
 3576 }
 3577 
 3578 void MacroAssembler::testptr(Register dst, Register src) {
 3579   testq(dst, src);
 3580 }
 3581 
 3582 // Object / value buffer allocation...
 3583 //
 3584 // Kills klass and rsi on LP64
 3585 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
 3586                                        Register t1, Register t2,
 3587                                        bool clear_fields, Label& alloc_failed)
 3588 {
 3589   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
 3590   Register layout_size = t1;
 3591   assert(new_obj == rax, "needs to be rax");
 3592   assert_different_registers(klass, new_obj, t1, t2);
 3593 
 3594   // get instance_size in InstanceKlass (scaled to a count of bytes)
 3595   movl(layout_size, Address(klass, Klass::layout_helper_offset()));
 3596   // test to see if it is malformed in some way
 3597   testl(layout_size, Klass::_lh_instance_slow_path_bit);
 3598   jcc(Assembler::notZero, slow_case_no_pop);
 3599 
 3600   // Allocate the instance:
 3601   //  If TLAB is enabled:
 3602   //    Try to allocate in the TLAB.
 3603   //    If fails, go to the slow path.
 3604   //  Else If inline contiguous allocations are enabled:
 3605   //    Try to allocate in eden.
 3606   //    If fails due to heap end, go to slow path.
 3607   //
 3608   //  If TLAB is enabled OR inline contiguous is enabled:
 3609   //    Initialize the allocation.
 3610   //    Exit.
 3611   //
 3612   //  Go to slow path.
 3613 
 3614   push(klass);
 3615   if (UseTLAB) {
 3616     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
 3617     if (ZeroTLAB || (!clear_fields)) {
 3618       // the fields have been already cleared
 3619       jmp(initialize_header);
 3620     } else {
 3621       // initialize both the header and fields
 3622       jmp(initialize_object);
 3623     }
 3624   } else {
 3625     jmp(slow_case);
 3626   }
 3627 
 3628   // If UseTLAB is true, the object is created above and there is an initialize need.
 3629   // Otherwise, skip and go to the slow path.
 3630   if (UseTLAB) {
 3631     if (clear_fields) {
 3632       // The object is initialized before the header.  If the object size is
 3633       // zero, go directly to the header initialization.
 3634       bind(initialize_object);
 3635       if (UseCompactObjectHeaders) {
 3636         assert(is_aligned(oopDesc::base_offset_in_bytes(), BytesPerLong), "oop base offset must be 8-byte-aligned");
 3637         decrement(layout_size, oopDesc::base_offset_in_bytes());
 3638       } else {
 3639         decrement(layout_size, sizeof(oopDesc));
 3640       }
 3641       jcc(Assembler::zero, initialize_header);
 3642 
 3643       // Initialize topmost object field, divide size by 8, check if odd and
 3644       // test if zero.
 3645       Register zero = klass;
 3646       xorl(zero, zero);    // use zero reg to clear memory (shorter code)
 3647       shrl(layout_size, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
 3648 
 3649   #ifdef ASSERT
 3650       // make sure instance_size was multiple of 8
 3651       Label L;
 3652       // Ignore partial flag stall after shrl() since it is debug VM
 3653       jcc(Assembler::carryClear, L);
 3654       stop("object size is not multiple of 2 - adjust this code");
 3655       bind(L);
 3656       // must be > 0, no extra check needed here
 3657   #endif
 3658 
 3659       // initialize remaining object fields: instance_size was a multiple of 8
 3660       {
 3661         Label loop;
 3662         bind(loop);
 3663         int header_size_bytes = oopDesc::header_size() * HeapWordSize;
 3664         assert(is_aligned(header_size_bytes, BytesPerLong), "oop header size must be 8-byte-aligned");
 3665         movptr(Address(new_obj, layout_size, Address::times_8, header_size_bytes - 1*oopSize), zero);
 3666         decrement(layout_size);
 3667         jcc(Assembler::notZero, loop);
 3668       }
 3669     } // clear_fields
 3670 
 3671     // initialize object header only.
 3672     bind(initialize_header);
 3673     if (UseCompactObjectHeaders || Arguments::is_valhalla_enabled()) {
 3674       pop(klass);
 3675       Register mark_word = t2;
 3676       movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 3677       movptr(Address(new_obj, oopDesc::mark_offset_in_bytes ()), mark_word);
 3678     } else {
 3679      movptr(Address(new_obj, oopDesc::mark_offset_in_bytes()),
 3680             (intptr_t)markWord::prototype().value()); // header
 3681      pop(klass);   // get saved klass back in the register.
 3682     }
 3683     if (!UseCompactObjectHeaders) {
 3684       xorl(rsi, rsi);                 // use zero reg to clear memory (shorter code)
 3685       store_klass_gap(new_obj, rsi);  // zero klass gap for compressed oops
 3686       movptr(t2, klass);         // preserve klass
 3687       store_klass(new_obj, t2, rscratch1);  // src klass reg is potentially compressed
 3688     }
 3689     jmp(done);
 3690   }
 3691 
 3692   bind(slow_case);
 3693   pop(klass);
 3694   bind(slow_case_no_pop);
 3695   jmp(alloc_failed);
 3696 
 3697   bind(done);
 3698 }
 3699 
 3700 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3701 void MacroAssembler::tlab_allocate(Register obj,
 3702                                    Register var_size_in_bytes,
 3703                                    int con_size_in_bytes,
 3704                                    Register t1,
 3705                                    Register t2,
 3706                                    Label& slow_case) {
 3707   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3708   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3709 }
 3710 
 3711 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3712   RegSet regs;
 3713   regs += RegSet::of(rax, rcx, rdx);
 3714 #ifndef _WINDOWS
 3715   regs += RegSet::of(rsi, rdi);
 3716 #endif
 3717   regs += RegSet::range(r8, r11);
 3718   if (UseAPX) {
 3719     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
 3720   }
 3721   return regs;
 3722 }
 3723 
 3724 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
 3725   int num_xmm_registers = XMMRegister::available_xmm_registers();
 3726 #if defined(_WINDOWS)
 3727   XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
 3728   if (num_xmm_registers > 16) {
 3729      result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
 3730   }
 3731   return result;
 3732 #else
 3733   return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
 3734 #endif
 3735 }
 3736 
 3737 // C1 only ever uses the first double/float of the XMM register.
 3738 static int xmm_save_size() { return sizeof(double); }
 3739 
 3740 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3741   masm->movdbl(Address(rsp, offset), reg);
 3742 }
 3743 
 3744 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
 3745   masm->movdbl(reg, Address(rsp, offset));
 3746 }
 3747 
 3748 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
 3749                                   bool save_fpu, int& gp_area_size, int& xmm_area_size) {
 3750 
 3751   gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
 3752                          StackAlignmentInBytes);
 3753   xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
 3754 
 3755   return gp_area_size + xmm_area_size;
 3756 }
 3757 
 3758 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
 3759   block_comment("push_call_clobbered_registers start");
 3760   // Regular registers
 3761   RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
 3762 
 3763   int gp_area_size;
 3764   int xmm_area_size;
 3765   int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
 3766                                                gp_area_size, xmm_area_size);
 3767   subptr(rsp, total_save_size);
 3768 
 3769   push_set(gp_registers_to_push, 0);
 3770 
 3771   if (save_fpu) {
 3772     push_set(call_clobbered_xmm_registers(), gp_area_size);
 3773   }
 3774 
 3775   block_comment("push_call_clobbered_registers end");
 3776 }
 3777 
 3778 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
 3779   block_comment("pop_call_clobbered_registers start");
 3780 
 3781   RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
 3782 
 3783   int gp_area_size;
 3784   int xmm_area_size;
 3785   int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
 3786                                                gp_area_size, xmm_area_size);
 3787 
 3788   if (restore_fpu) {
 3789     pop_set(call_clobbered_xmm_registers(), gp_area_size);
 3790   }
 3791 
 3792   pop_set(gp_registers_to_pop, 0);
 3793 
 3794   addptr(rsp, total_save_size);
 3795 
 3796   vzeroupper();
 3797 
 3798   block_comment("pop_call_clobbered_registers end");
 3799 }
 3800 
 3801 void MacroAssembler::push_set(XMMRegSet set, int offset) {
 3802   assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
 3803   int spill_offset = offset;
 3804 
 3805   for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
 3806     save_xmm_register(this, spill_offset, *it);
 3807     spill_offset += xmm_save_size();
 3808   }
 3809 }
 3810 
 3811 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
 3812   int restore_size = set.size() * xmm_save_size();
 3813   assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
 3814 
 3815   int restore_offset = offset + restore_size - xmm_save_size();
 3816 
 3817   for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
 3818     restore_xmm_register(this, restore_offset, *it);
 3819     restore_offset -= xmm_save_size();
 3820   }
 3821 }
 3822 
 3823 void MacroAssembler::push_set(RegSet set, int offset) {
 3824   int spill_offset;
 3825   if (offset == -1) {
 3826     int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3827     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 3828     subptr(rsp, aligned_size);
 3829     spill_offset = 0;
 3830   } else {
 3831     spill_offset = offset;
 3832   }
 3833 
 3834   for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
 3835     movptr(Address(rsp, spill_offset), *it);
 3836     spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3837   }
 3838 }
 3839 
 3840 void MacroAssembler::pop_set(RegSet set, int offset) {
 3841 
 3842   int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 3843   int restore_size = set.size() * gp_reg_size;
 3844   int aligned_size = align_up(restore_size, StackAlignmentInBytes);
 3845 
 3846   int restore_offset;
 3847   if (offset == -1) {
 3848     restore_offset = restore_size - gp_reg_size;
 3849   } else {
 3850     restore_offset = offset + restore_size - gp_reg_size;
 3851   }
 3852   for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
 3853     movptr(*it, Address(rsp, restore_offset));
 3854     restore_offset -= gp_reg_size;
 3855   }
 3856 
 3857   if (offset == -1) {
 3858     addptr(rsp, aligned_size);
 3859   }
 3860 }
 3861 
 3862 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
 3863 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
 3864   assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
 3865   assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
 3866   Label done;
 3867 
 3868   testptr(length_in_bytes, length_in_bytes);
 3869   jcc(Assembler::zero, done);
 3870 
 3871   // initialize topmost word, divide index by 2, check if odd and test if zero
 3872   // note: for the remaining code to work, index must be a multiple of BytesPerWord
 3873 #ifdef ASSERT
 3874   {
 3875     Label L;
 3876     testptr(length_in_bytes, BytesPerWord - 1);
 3877     jcc(Assembler::zero, L);
 3878     stop("length must be a multiple of BytesPerWord");
 3879     bind(L);
 3880   }
 3881 #endif
 3882   Register index = length_in_bytes;
 3883   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3884   if (UseIncDec) {
 3885     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3886   } else {
 3887     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3888     shrptr(index, 1);
 3889   }
 3890 
 3891   // initialize remaining object fields: index is a multiple of 2 now
 3892   {
 3893     Label loop;
 3894     bind(loop);
 3895     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3896     decrement(index);
 3897     jcc(Assembler::notZero, loop);
 3898   }
 3899 
 3900   bind(done);
 3901 }
 3902 
 3903 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
 3904   movptr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
 3905 #ifdef ASSERT
 3906   {
 3907     Label done;
 3908     cmpptr(layout_info, 0);
 3909     jcc(Assembler::notEqual, done);
 3910     stop("inline_layout_info_array is null");
 3911     bind(done);
 3912   }
 3913 #endif
 3914 
 3915   InlineLayoutInfo array[2];
 3916   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
 3917   if (is_power_of_2(size)) {
 3918     shll(index, log2i_exact(size)); // Scale index by power of 2
 3919   } else {
 3920     imull(index, index, size); // Scale the index to be the entry index * array_element_size
 3921   }
 3922   lea(layout_info, Address(layout_info, index, Address::times_1, Array<InlineLayoutInfo>::base_offset_in_bytes()));
 3923 }
 3924 
 3925 // Look up the method for a megamorphic invokeinterface call.
 3926 // The target method is determined by <intf_klass, itable_index>.
 3927 // The receiver klass is in recv_klass.
 3928 // On success, the result will be in method_result, and execution falls through.
 3929 // On failure, execution transfers to the given label.
 3930 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3931                                              Register intf_klass,
 3932                                              RegisterOrConstant itable_index,
 3933                                              Register method_result,
 3934                                              Register scan_temp,
 3935                                              Label& L_no_such_interface,
 3936                                              bool return_method) {
 3937   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3938   assert_different_registers(method_result, intf_klass, scan_temp);
 3939   assert(recv_klass != method_result || !return_method,
 3940          "recv_klass can be destroyed when method isn't needed");
 3941 
 3942   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3943          "caller must use same register for non-constant itable index as for method");
 3944 
 3945   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
 3946   int vtable_base = in_bytes(Klass::vtable_start_offset());
 3947   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 3948   int scan_step   = itableOffsetEntry::size() * wordSize;
 3949   int vte_size    = vtableEntry::size_in_bytes();
 3950   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 3951   assert(vte_size == wordSize, "else adjust times_vte_scale");
 3952 
 3953   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 3954 
 3955   // Could store the aligned, prescaled offset in the klass.
 3956   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
 3957 
 3958   if (return_method) {
 3959     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
 3960     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 3961     lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
 3962   }
 3963 
 3964   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
 3965   //   if (scan->interface() == intf) {
 3966   //     result = (klass + scan->offset() + itable_index);
 3967   //   }
 3968   // }
 3969   Label search, found_method;
 3970 
 3971   for (int peel = 1; peel >= 0; peel--) {
 3972     movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
 3973     cmpptr(intf_klass, method_result);
 3974 
 3975     if (peel) {
 3976       jccb(Assembler::equal, found_method);
 3977     } else {
 3978       jccb(Assembler::notEqual, search);
 3979       // (invert the test to fall through to found_method...)
 3980     }
 3981 
 3982     if (!peel)  break;
 3983 
 3984     bind(search);
 3985 
 3986     // Check that the previous entry is non-null.  A null entry means that
 3987     // the receiver class doesn't implement the interface, and wasn't the
 3988     // same as when the caller was compiled.
 3989     testptr(method_result, method_result);
 3990     jcc(Assembler::zero, L_no_such_interface);
 3991     addptr(scan_temp, scan_step);
 3992   }
 3993 
 3994   bind(found_method);
 3995 
 3996   if (return_method) {
 3997     // Got a hit.
 3998     movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
 3999     movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
 4000   }
 4001 }
 4002 
 4003 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
 4004 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
 4005 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
 4006 // The target method is determined by <holder_klass, itable_index>.
 4007 // The receiver klass is in recv_klass.
 4008 // On success, the result will be in method_result, and execution falls through.
 4009 // On failure, execution transfers to the given label.
 4010 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
 4011                                                   Register holder_klass,
 4012                                                   Register resolved_klass,
 4013                                                   Register method_result,
 4014                                                   Register scan_temp,
 4015                                                   Register temp_reg2,
 4016                                                   Register receiver,
 4017                                                   int itable_index,
 4018                                                   Label& L_no_such_interface) {
 4019   assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
 4020   Register temp_itbl_klass = method_result;
 4021   Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
 4022 
 4023   int vtable_base = in_bytes(Klass::vtable_start_offset());
 4024   int itentry_off = in_bytes(itableMethodEntry::method_offset());
 4025   int scan_step = itableOffsetEntry::size() * wordSize;
 4026   int vte_size = vtableEntry::size_in_bytes();
 4027   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
 4028   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
 4029   Address::ScaleFactor times_vte_scale = Address::times_ptr;
 4030   assert(vte_size == wordSize, "adjust times_vte_scale");
 4031 
 4032   Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
 4033 
 4034   // temp_itbl_klass = recv_klass.itable[0]
 4035   // scan_temp = &recv_klass.itable[0] + step
 4036   movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
 4037   movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
 4038   lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
 4039   xorptr(temp_reg, temp_reg);
 4040 
 4041   // Initial checks:
 4042   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
 4043   //   - if (itable[0] == 0), no such interface
 4044   //   - if (itable[0] == holder_klass), shortcut to "holder found"
 4045   cmpptr(holder_klass, resolved_klass);
 4046   jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
 4047   testptr(temp_itbl_klass, temp_itbl_klass);
 4048   jccb(Assembler::zero, L_no_such_interface);
 4049   cmpptr(holder_klass, temp_itbl_klass);
 4050   jccb(Assembler::equal, L_holder_found);
 4051 
 4052   // Loop: Look for holder_klass record in itable
 4053   //   do {
 4054   //     tmp = itable[index];
 4055   //     index += step;
 4056   //     if (tmp == holder_klass) {
 4057   //       goto L_holder_found; // Found!
 4058   //     }
 4059   //   } while (tmp != 0);
 4060   //   goto L_no_such_interface // Not found.
 4061   Label L_scan_holder;
 4062   bind(L_scan_holder);
 4063     movptr(temp_itbl_klass, Address(scan_temp, 0));
 4064     addptr(scan_temp, scan_step);
 4065     cmpptr(holder_klass, temp_itbl_klass);
 4066     jccb(Assembler::equal, L_holder_found);
 4067     testptr(temp_itbl_klass, temp_itbl_klass);
 4068     jccb(Assembler::notZero, L_scan_holder);
 4069 
 4070   jmpb(L_no_such_interface);
 4071 
 4072   // Loop: Look for resolved_class record in itable
 4073   //   do {
 4074   //     tmp = itable[index];
 4075   //     index += step;
 4076   //     if (tmp == holder_klass) {
 4077   //        // Also check if we have met a holder klass
 4078   //        holder_tmp = itable[index-step-ioffset];
 4079   //     }
 4080   //     if (tmp == resolved_klass) {
 4081   //        goto L_resolved_found;  // Found!
 4082   //     }
 4083   //   } while (tmp != 0);
 4084   //   goto L_no_such_interface // Not found.
 4085   //
 4086   Label L_loop_scan_resolved;
 4087   bind(L_loop_scan_resolved);
 4088     movptr(temp_itbl_klass, Address(scan_temp, 0));
 4089     addptr(scan_temp, scan_step);
 4090     bind(L_loop_scan_resolved_entry);
 4091     cmpptr(holder_klass, temp_itbl_klass);
 4092     cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4093     cmpptr(resolved_klass, temp_itbl_klass);
 4094     jccb(Assembler::equal, L_resolved_found);
 4095     testptr(temp_itbl_klass, temp_itbl_klass);
 4096     jccb(Assembler::notZero, L_loop_scan_resolved);
 4097 
 4098   jmpb(L_no_such_interface);
 4099 
 4100   Label L_ready;
 4101 
 4102   // See if we already have a holder klass. If not, go and scan for it.
 4103   bind(L_resolved_found);
 4104   testptr(temp_reg, temp_reg);
 4105   jccb(Assembler::zero, L_scan_holder);
 4106   jmpb(L_ready);
 4107 
 4108   bind(L_holder_found);
 4109   movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
 4110 
 4111   // Finally, temp_reg contains holder_klass vtable offset
 4112   bind(L_ready);
 4113   assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 4114   if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
 4115     load_klass(scan_temp, receiver, noreg);
 4116     movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4117   } else {
 4118     movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
 4119   }
 4120 }
 4121 
 4122 
 4123 // virtual method calling
 4124 void MacroAssembler::lookup_virtual_method(Register recv_klass,
 4125                                            RegisterOrConstant vtable_index,
 4126                                            Register method_result) {
 4127   const ByteSize base = Klass::vtable_start_offset();
 4128   assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
 4129   Address vtable_entry_addr(recv_klass,
 4130                             vtable_index, Address::times_ptr,
 4131                             base + vtableEntry::method_offset());
 4132   movptr(method_result, vtable_entry_addr);
 4133 }
 4134 
 4135 
 4136 void MacroAssembler::check_klass_subtype(Register sub_klass,
 4137                            Register super_klass,
 4138                            Register temp_reg,
 4139                            Label& L_success) {
 4140   Label L_failure;
 4141   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
 4142   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
 4143   bind(L_failure);
 4144 }
 4145 
 4146 
 4147 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
 4148                                                    Register super_klass,
 4149                                                    Register temp_reg,
 4150                                                    Label* L_success,
 4151                                                    Label* L_failure,
 4152                                                    Label* L_slow_path,
 4153                                         RegisterOrConstant super_check_offset) {
 4154   assert_different_registers(sub_klass, super_klass, temp_reg);
 4155   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
 4156   if (super_check_offset.is_register()) {
 4157     assert_different_registers(sub_klass, super_klass,
 4158                                super_check_offset.as_register());
 4159   } else if (must_load_sco) {
 4160     assert(temp_reg != noreg, "supply either a temp or a register offset");
 4161   }
 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   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
 4168   assert(label_nulls <= 1, "at most one null in the batch");
 4169 
 4170   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4171   int sco_offset = in_bytes(Klass::super_check_offset_offset());
 4172   Address super_check_offset_addr(super_klass, sco_offset);
 4173 
 4174   // Hacked jcc, which "knows" that L_fallthrough, at least, is in
 4175   // range of a jccb.  If this routine grows larger, reconsider at
 4176   // least some of these.
 4177 #define local_jcc(assembler_cond, label)                                \
 4178   if (&(label) == &L_fallthrough)  jccb(assembler_cond, label);         \
 4179   else                             jcc( assembler_cond, label) /*omit semi*/
 4180 
 4181   // Hacked jmp, which may only be used just before L_fallthrough.
 4182 #define final_jmp(label)                                                \
 4183   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
 4184   else                            jmp(label)                /*omit semi*/
 4185 
 4186   // If the pointers are equal, we are done (e.g., String[] elements).
 4187   // This self-check enables sharing of secondary supertype arrays among
 4188   // non-primary types such as array-of-interface.  Otherwise, each such
 4189   // type would need its own customized SSA.
 4190   // We move this check to the front of the fast path because many
 4191   // type checks are in fact trivially successful in this manner,
 4192   // so we get a nicely predicted branch right at the start of the check.
 4193   cmpptr(sub_klass, super_klass);
 4194   local_jcc(Assembler::equal, *L_success);
 4195 
 4196   // Check the supertype display:
 4197   if (must_load_sco) {
 4198     // Positive movl does right thing on LP64.
 4199     movl(temp_reg, super_check_offset_addr);
 4200     super_check_offset = RegisterOrConstant(temp_reg);
 4201   }
 4202   Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
 4203   cmpptr(super_klass, super_check_addr); // load displayed supertype
 4204 
 4205   // This check has worked decisively for primary supers.
 4206   // Secondary supers are sought in the super_cache ('super_cache_addr').
 4207   // (Secondary supers are interfaces and very deeply nested subtypes.)
 4208   // This works in the same check above because of a tricky aliasing
 4209   // between the super_cache and the primary super display elements.
 4210   // (The 'super_check_addr' can address either, as the case requires.)
 4211   // Note that the cache is updated below if it does not help us find
 4212   // what we need immediately.
 4213   // So if it was a primary super, we can just fail immediately.
 4214   // Otherwise, it's the slow path for us (no success at this point).
 4215 
 4216   if (super_check_offset.is_register()) {
 4217     local_jcc(Assembler::equal, *L_success);
 4218     cmpl(super_check_offset.as_register(), sc_offset);
 4219     if (L_failure == &L_fallthrough) {
 4220       local_jcc(Assembler::equal, *L_slow_path);
 4221     } else {
 4222       local_jcc(Assembler::notEqual, *L_failure);
 4223       final_jmp(*L_slow_path);
 4224     }
 4225   } else if (super_check_offset.as_constant() == sc_offset) {
 4226     // Need a slow path; fast failure is impossible.
 4227     if (L_slow_path == &L_fallthrough) {
 4228       local_jcc(Assembler::equal, *L_success);
 4229     } else {
 4230       local_jcc(Assembler::notEqual, *L_slow_path);
 4231       final_jmp(*L_success);
 4232     }
 4233   } else {
 4234     // No slow path; it's a fast decision.
 4235     if (L_failure == &L_fallthrough) {
 4236       local_jcc(Assembler::equal, *L_success);
 4237     } else {
 4238       local_jcc(Assembler::notEqual, *L_failure);
 4239       final_jmp(*L_success);
 4240     }
 4241   }
 4242 
 4243   bind(L_fallthrough);
 4244 
 4245 #undef local_jcc
 4246 #undef final_jmp
 4247 }
 4248 
 4249 
 4250 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
 4251                                                           Register super_klass,
 4252                                                           Register temp_reg,
 4253                                                           Register temp2_reg,
 4254                                                           Label* L_success,
 4255                                                           Label* L_failure,
 4256                                                           bool set_cond_codes) {
 4257   assert_different_registers(sub_klass, super_klass, temp_reg);
 4258   if (temp2_reg != noreg)
 4259     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
 4260 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
 4261 
 4262   Label L_fallthrough;
 4263   int label_nulls = 0;
 4264   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4265   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4266   assert(label_nulls <= 1, "at most one null in the batch");
 4267 
 4268   // a couple of useful fields in sub_klass:
 4269   int ss_offset = in_bytes(Klass::secondary_supers_offset());
 4270   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
 4271   Address secondary_supers_addr(sub_klass, ss_offset);
 4272   Address super_cache_addr(     sub_klass, sc_offset);
 4273 
 4274   // Do a linear scan of the secondary super-klass chain.
 4275   // This code is rarely used, so simplicity is a virtue here.
 4276   // The repne_scan instruction uses fixed registers, which we must spill.
 4277   // Don't worry too much about pre-existing connections with the input regs.
 4278 
 4279   assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
 4280   assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
 4281 
 4282   // Get super_klass value into rax (even if it was in rdi or rcx).
 4283   bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
 4284   if (super_klass != rax) {
 4285     if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
 4286     mov(rax, super_klass);
 4287   }
 4288   if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
 4289   if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
 4290 
 4291 #ifndef PRODUCT
 4292   uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
 4293   ExternalAddress pst_counter_addr((address) pst_counter);
 4294   lea(rcx, pst_counter_addr);
 4295   incrementl(Address(rcx, 0));
 4296 #endif //PRODUCT
 4297 
 4298   // We will consult the secondary-super array.
 4299   movptr(rdi, secondary_supers_addr);
 4300   // Load the array length.  (Positive movl does right thing on LP64.)
 4301   movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
 4302   // Skip to start of data.
 4303   addptr(rdi, Array<Klass*>::base_offset_in_bytes());
 4304 
 4305   // Scan RCX words at [RDI] for an occurrence of RAX.
 4306   // Set NZ/Z based on last compare.
 4307   // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
 4308   // not change flags (only scas instruction which is repeated sets flags).
 4309   // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
 4310 
 4311     testptr(rax,rax); // Set Z = 0
 4312     repne_scan();
 4313 
 4314   // Unspill the temp. registers:
 4315   if (pushed_rdi)  pop(rdi);
 4316   if (pushed_rcx)  pop(rcx);
 4317   if (pushed_rax)  pop(rax);
 4318 
 4319   if (set_cond_codes) {
 4320     // Special hack for the AD files:  rdi is guaranteed non-zero.
 4321     assert(!pushed_rdi, "rdi must be left non-null");
 4322     // Also, the condition codes are properly set Z/NZ on succeed/failure.
 4323   }
 4324 
 4325   if (L_failure == &L_fallthrough)
 4326         jccb(Assembler::notEqual, *L_failure);
 4327   else  jcc(Assembler::notEqual, *L_failure);
 4328 
 4329   // Success.  Cache the super we found and proceed in triumph.
 4330   movptr(super_cache_addr, super_klass);
 4331 
 4332   if (L_success != &L_fallthrough) {
 4333     jmp(*L_success);
 4334   }
 4335 
 4336 #undef IS_A_TEMP
 4337 
 4338   bind(L_fallthrough);
 4339 }
 4340 
 4341 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4342                                                    Register super_klass,
 4343                                                    Register temp_reg,
 4344                                                    Register temp2_reg,
 4345                                                    Label* L_success,
 4346                                                    Label* L_failure,
 4347                                                    bool set_cond_codes) {
 4348   assert(set_cond_codes == false, "must be false on 64-bit x86");
 4349   check_klass_subtype_slow_path
 4350     (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
 4351      L_success, L_failure);
 4352 }
 4353 
 4354 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 4355                                                    Register super_klass,
 4356                                                    Register temp_reg,
 4357                                                    Register temp2_reg,
 4358                                                    Register temp3_reg,
 4359                                                    Register temp4_reg,
 4360                                                    Label* L_success,
 4361                                                    Label* L_failure) {
 4362   if (UseSecondarySupersTable) {
 4363     check_klass_subtype_slow_path_table
 4364       (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
 4365        L_success, L_failure);
 4366   } else {
 4367     check_klass_subtype_slow_path_linear
 4368       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
 4369   }
 4370 }
 4371 
 4372 Register MacroAssembler::allocate_if_noreg(Register r,
 4373                                   RegSetIterator<Register> &available_regs,
 4374                                   RegSet &regs_to_push) {
 4375   if (!r->is_valid()) {
 4376     r = *available_regs++;
 4377     regs_to_push += r;
 4378   }
 4379   return r;
 4380 }
 4381 
 4382 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
 4383                                                          Register super_klass,
 4384                                                          Register temp_reg,
 4385                                                          Register temp2_reg,
 4386                                                          Register temp3_reg,
 4387                                                          Register result_reg,
 4388                                                          Label* L_success,
 4389                                                          Label* L_failure) {
 4390   // NB! Callers may assume that, when temp2_reg is a valid register,
 4391   // this code sets it to a nonzero value.
 4392   bool temp2_reg_was_valid = temp2_reg->is_valid();
 4393 
 4394   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
 4395 
 4396   Label L_fallthrough;
 4397   int label_nulls = 0;
 4398   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4399   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4400   assert(label_nulls <= 1, "at most one null in the batch");
 4401 
 4402   BLOCK_COMMENT("check_klass_subtype_slow_path_table");
 4403 
 4404   RegSetIterator<Register> available_regs
 4405     = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
 4406 
 4407   RegSet pushed_regs;
 4408 
 4409   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
 4410   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
 4411   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
 4412   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
 4413   Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
 4414 
 4415   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
 4416 
 4417   {
 4418 
 4419     int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
 4420     int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
 4421     subptr(rsp, aligned_size);
 4422     push_set(pushed_regs, 0);
 4423 
 4424     lookup_secondary_supers_table_var(sub_klass,
 4425                                       super_klass,
 4426                                       temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
 4427     cmpq(result_reg, 0);
 4428 
 4429     // Unspill the temp. registers:
 4430     pop_set(pushed_regs, 0);
 4431     // Increment SP but do not clobber flags.
 4432     lea(rsp, Address(rsp, aligned_size));
 4433   }
 4434 
 4435   if (temp2_reg_was_valid) {
 4436     movq(temp2_reg, 1);
 4437   }
 4438 
 4439   jcc(Assembler::notEqual, *L_failure);
 4440 
 4441   if (L_success != &L_fallthrough) {
 4442     jmp(*L_success);
 4443   }
 4444 
 4445   bind(L_fallthrough);
 4446 }
 4447 
 4448 // population_count variant for running without the POPCNT
 4449 // instruction, which was introduced with SSE4.2 in 2008.
 4450 void MacroAssembler::population_count(Register dst, Register src,
 4451                                       Register scratch1, Register scratch2) {
 4452   assert_different_registers(src, scratch1, scratch2);
 4453   if (UsePopCountInstruction) {
 4454     Assembler::popcntq(dst, src);
 4455   } else {
 4456     assert_different_registers(src, scratch1, scratch2);
 4457     assert_different_registers(dst, scratch1, scratch2);
 4458     Label loop, done;
 4459 
 4460     mov(scratch1, src);
 4461     // dst = 0;
 4462     // while(scratch1 != 0) {
 4463     //   dst++;
 4464     //   scratch1 &= (scratch1 - 1);
 4465     // }
 4466     xorl(dst, dst);
 4467     testq(scratch1, scratch1);
 4468     jccb(Assembler::equal, done);
 4469     {
 4470       bind(loop);
 4471       incq(dst);
 4472       movq(scratch2, scratch1);
 4473       decq(scratch2);
 4474       andq(scratch1, scratch2);
 4475       jccb(Assembler::notEqual, loop);
 4476     }
 4477     bind(done);
 4478   }
 4479 #ifdef ASSERT
 4480   mov64(scratch1, 0xCafeBabeDeadBeef);
 4481   movq(scratch2, scratch1);
 4482 #endif
 4483 }
 4484 
 4485 // Ensure that the inline code and the stub are using the same registers.
 4486 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                      \
 4487 do {                                                                 \
 4488   assert(r_super_klass  == rax, "mismatch");                         \
 4489   assert(r_array_base   == rbx, "mismatch");                         \
 4490   assert(r_array_length == rcx, "mismatch");                         \
 4491   assert(r_array_index  == rdx, "mismatch");                         \
 4492   assert(r_sub_klass    == rsi || r_sub_klass == noreg, "mismatch"); \
 4493   assert(r_bitmap       == r11 || r_bitmap    == noreg, "mismatch"); \
 4494   assert(result         == rdi || result      == noreg, "mismatch"); \
 4495 } while(0)
 4496 
 4497 // Versions of salq and rorq that don't need count to be in rcx
 4498 
 4499 void MacroAssembler::salq(Register dest, Register count) {
 4500   if (count == rcx) {
 4501     Assembler::salq(dest);
 4502   } else {
 4503     assert_different_registers(rcx, dest);
 4504     xchgq(rcx, count);
 4505     Assembler::salq(dest);
 4506     xchgq(rcx, count);
 4507   }
 4508 }
 4509 
 4510 void MacroAssembler::rorq(Register dest, Register count) {
 4511   if (count == rcx) {
 4512     Assembler::rorq(dest);
 4513   } else {
 4514     assert_different_registers(rcx, dest);
 4515     xchgq(rcx, count);
 4516     Assembler::rorq(dest);
 4517     xchgq(rcx, count);
 4518   }
 4519 }
 4520 
 4521 // Return true: we succeeded in generating this code
 4522 //
 4523 // At runtime, return 0 in result if r_super_klass is a superclass of
 4524 // r_sub_klass, otherwise return nonzero. Use this if you know the
 4525 // super_klass_slot of the class you're looking for. This is always
 4526 // the case for instanceof and checkcast.
 4527 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
 4528                                                          Register r_super_klass,
 4529                                                          Register temp1,
 4530                                                          Register temp2,
 4531                                                          Register temp3,
 4532                                                          Register temp4,
 4533                                                          Register result,
 4534                                                          u1 super_klass_slot) {
 4535   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4536 
 4537   Label L_fallthrough, L_success, L_failure;
 4538 
 4539   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4540 
 4541   const Register
 4542     r_array_index  = temp1,
 4543     r_array_length = temp2,
 4544     r_array_base   = temp3,
 4545     r_bitmap       = temp4;
 4546 
 4547   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
 4548 
 4549   xorq(result, result); // = 0
 4550 
 4551   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4552   movq(r_array_index, r_bitmap);
 4553 
 4554   // First check the bitmap to see if super_klass might be present. If
 4555   // the bit is zero, we are certain that super_klass is not one of
 4556   // the secondary supers.
 4557   u1 bit = super_klass_slot;
 4558   {
 4559     // NB: If the count in a x86 shift instruction is 0, the flags are
 4560     // not affected, so we do a testq instead.
 4561     int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
 4562     if (shift_count != 0) {
 4563       salq(r_array_index, shift_count);
 4564     } else {
 4565       testq(r_array_index, r_array_index);
 4566     }
 4567   }
 4568   // We test the MSB of r_array_index, i.e. its sign bit
 4569   jcc(Assembler::positive, L_failure);
 4570 
 4571   // Get the first array index that can contain super_klass into r_array_index.
 4572   if (bit != 0) {
 4573     population_count(r_array_index, r_array_index, temp2, temp3);
 4574   } else {
 4575     movl(r_array_index, 1);
 4576   }
 4577   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4578 
 4579   // We will consult the secondary-super array.
 4580   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4581 
 4582   // We're asserting that the first word in an Array<Klass*> is the
 4583   // length, and the second word is the first word of the data. If
 4584   // that ever changes, r_array_base will have to be adjusted here.
 4585   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4586   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4587 
 4588   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4589   jccb(Assembler::equal, L_success);
 4590 
 4591   // Is there another entry to check? Consult the bitmap.
 4592   btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
 4593   jccb(Assembler::carryClear, L_failure);
 4594 
 4595   // Linear probe. Rotate the bitmap so that the next bit to test is
 4596   // in Bit 1.
 4597   if (bit != 0) {
 4598     rorq(r_bitmap, bit);
 4599   }
 4600 
 4601   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4602   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4603   // Kills: r_array_length.
 4604   // Returns: result.
 4605   call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
 4606   // Result (0/1) is in rdi
 4607   jmpb(L_fallthrough);
 4608 
 4609   bind(L_failure);
 4610   incq(result); // 0 => 1
 4611 
 4612   bind(L_success);
 4613   // result = 0;
 4614 
 4615   bind(L_fallthrough);
 4616   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4617 
 4618   if (VerifySecondarySupers) {
 4619     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4620                                   temp1, temp2, temp3);
 4621   }
 4622 }
 4623 
 4624 // At runtime, return 0 in result if r_super_klass is a superclass of
 4625 // r_sub_klass, otherwise return nonzero. Use this version of
 4626 // lookup_secondary_supers_table() if you don't know ahead of time
 4627 // which superclass will be searched for. Used by interpreter and
 4628 // runtime stubs. It is larger and has somewhat greater latency than
 4629 // the version above, which takes a constant super_klass_slot.
 4630 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
 4631                                                        Register r_super_klass,
 4632                                                        Register temp1,
 4633                                                        Register temp2,
 4634                                                        Register temp3,
 4635                                                        Register temp4,
 4636                                                        Register result) {
 4637   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
 4638   assert_different_registers(r_sub_klass, r_super_klass, rcx);
 4639   RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
 4640 
 4641   Label L_fallthrough, L_success, L_failure;
 4642 
 4643   BLOCK_COMMENT("lookup_secondary_supers_table {");
 4644 
 4645   RegSetIterator<Register> available_regs = (temps - rcx).begin();
 4646 
 4647   // FIXME. Once we are sure that all paths reaching this point really
 4648   // do pass rcx as one of our temps we can get rid of the following
 4649   // workaround.
 4650   assert(temps.contains(rcx), "fix this code");
 4651 
 4652   // We prefer to have our shift count in rcx. If rcx is one of our
 4653   // temps, use it for slot. If not, pick any of our temps.
 4654   Register slot;
 4655   if (!temps.contains(rcx)) {
 4656     slot = *available_regs++;
 4657   } else {
 4658     slot = rcx;
 4659   }
 4660 
 4661   const Register r_array_index = *available_regs++;
 4662   const Register r_bitmap      = *available_regs++;
 4663 
 4664   // The logic above guarantees this property, but we state it here.
 4665   assert_different_registers(r_array_index, r_bitmap, rcx);
 4666 
 4667   movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
 4668   movq(r_array_index, r_bitmap);
 4669 
 4670   // First check the bitmap to see if super_klass might be present. If
 4671   // the bit is zero, we are certain that super_klass is not one of
 4672   // the secondary supers.
 4673   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4674   xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
 4675   salq(r_array_index, slot);
 4676 
 4677   testq(r_array_index, r_array_index);
 4678   // We test the MSB of r_array_index, i.e. its sign bit
 4679   jcc(Assembler::positive, L_failure);
 4680 
 4681   const Register r_array_base = *available_regs++;
 4682 
 4683   // Get the first array index that can contain super_klass into r_array_index.
 4684   // Note: Clobbers r_array_base and slot.
 4685   population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
 4686 
 4687   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
 4688 
 4689   // We will consult the secondary-super array.
 4690   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4691 
 4692   // We're asserting that the first word in an Array<Klass*> is the
 4693   // length, and the second word is the first word of the data. If
 4694   // that ever changes, r_array_base will have to be adjusted here.
 4695   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
 4696   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
 4697 
 4698   cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4699   jccb(Assembler::equal, L_success);
 4700 
 4701   // Restore slot to its true value
 4702   movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
 4703 
 4704   // Linear probe. Rotate the bitmap so that the next bit to test is
 4705   // in Bit 1.
 4706   rorq(r_bitmap, slot);
 4707 
 4708   // Is there another entry to check? Consult the bitmap.
 4709   btq(r_bitmap, 1);
 4710   jccb(Assembler::carryClear, L_failure);
 4711 
 4712   // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
 4713   // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
 4714   // Kills: r_array_length.
 4715   // Returns: result.
 4716   lookup_secondary_supers_table_slow_path(r_super_klass,
 4717                                           r_array_base,
 4718                                           r_array_index,
 4719                                           r_bitmap,
 4720                                           /*temp1*/result,
 4721                                           /*temp2*/slot,
 4722                                           &L_success,
 4723                                           nullptr);
 4724 
 4725   bind(L_failure);
 4726   movq(result, 1);
 4727   jmpb(L_fallthrough);
 4728 
 4729   bind(L_success);
 4730   xorq(result, result); // = 0
 4731 
 4732   bind(L_fallthrough);
 4733   BLOCK_COMMENT("} lookup_secondary_supers_table");
 4734 
 4735   if (VerifySecondarySupers) {
 4736     verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
 4737                                   temp1, temp2, temp3);
 4738   }
 4739 }
 4740 
 4741 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
 4742                                  Label* L_success, Label* L_failure) {
 4743   Label L_loop, L_fallthrough;
 4744   {
 4745     int label_nulls = 0;
 4746     if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
 4747     if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
 4748     assert(label_nulls <= 1, "at most one null in the batch");
 4749   }
 4750   bind(L_loop);
 4751   cmpq(value, Address(addr, count, Address::times_8));
 4752   jcc(Assembler::equal, *L_success);
 4753   addl(count, 1);
 4754   cmpl(count, limit);
 4755   jcc(Assembler::less, L_loop);
 4756 
 4757   if (&L_fallthrough != L_failure) {
 4758     jmp(*L_failure);
 4759   }
 4760   bind(L_fallthrough);
 4761 }
 4762 
 4763 // Called by code generated by check_klass_subtype_slow_path
 4764 // above. This is called when there is a collision in the hashed
 4765 // lookup in the secondary supers array.
 4766 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
 4767                                                              Register r_array_base,
 4768                                                              Register r_array_index,
 4769                                                              Register r_bitmap,
 4770                                                              Register temp1,
 4771                                                              Register temp2,
 4772                                                              Label* L_success,
 4773                                                              Label* L_failure) {
 4774   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
 4775 
 4776   const Register
 4777     r_array_length = temp1,
 4778     r_sub_klass    = noreg,
 4779     result         = noreg;
 4780 
 4781   Label L_fallthrough;
 4782   int label_nulls = 0;
 4783   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
 4784   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
 4785   assert(label_nulls <= 1, "at most one null in the batch");
 4786 
 4787   // Load the array length.
 4788   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4789   // And adjust the array base to point to the data.
 4790   // NB! Effectively increments current slot index by 1.
 4791   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
 4792   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4793 
 4794   // Linear probe
 4795   Label L_huge;
 4796 
 4797   // The bitmap is full to bursting.
 4798   // Implicit invariant: BITMAP_FULL implies (length > 0)
 4799   cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
 4800   jcc(Assembler::greater, L_huge);
 4801 
 4802   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
 4803   // current slot (at secondary_supers[r_array_index]) has not yet
 4804   // been inspected, and r_array_index may be out of bounds if we
 4805   // wrapped around the end of the array.
 4806 
 4807   { // This is conventional linear probing, but instead of terminating
 4808     // when a null entry is found in the table, we maintain a bitmap
 4809     // in which a 0 indicates missing entries.
 4810     // The check above guarantees there are 0s in the bitmap, so the loop
 4811     // eventually terminates.
 4812 
 4813     xorl(temp2, temp2); // = 0;
 4814 
 4815     Label L_again;
 4816     bind(L_again);
 4817 
 4818     // Check for array wraparound.
 4819     cmpl(r_array_index, r_array_length);
 4820     cmovl(Assembler::greaterEqual, r_array_index, temp2);
 4821 
 4822     cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
 4823     jcc(Assembler::equal, *L_success);
 4824 
 4825     // If the next bit in bitmap is zero, we're done.
 4826     btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
 4827     jcc(Assembler::carryClear, *L_failure);
 4828 
 4829     rorq(r_bitmap, 1); // Bits 1/2 => 0/1
 4830     addl(r_array_index, 1);
 4831 
 4832     jmp(L_again);
 4833   }
 4834 
 4835   { // Degenerate case: more than 64 secondary supers.
 4836     // FIXME: We could do something smarter here, maybe a vectorized
 4837     // comparison or a binary search, but is that worth any added
 4838     // complexity?
 4839     bind(L_huge);
 4840     xorl(r_array_index, r_array_index); // = 0
 4841     repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
 4842                 L_success,
 4843                 (&L_fallthrough != L_failure ? L_failure : nullptr));
 4844 
 4845     bind(L_fallthrough);
 4846   }
 4847 }
 4848 
 4849 struct VerifyHelperArguments {
 4850   Klass* _super;
 4851   Klass* _sub;
 4852   intptr_t _linear_result;
 4853   intptr_t _table_result;
 4854 };
 4855 
 4856 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
 4857   Klass::on_secondary_supers_verification_failure(args->_super,
 4858                                                   args->_sub,
 4859                                                   args->_linear_result,
 4860                                                   args->_table_result,
 4861                                                   msg);
 4862 }
 4863 
 4864 // Make sure that the hashed lookup and a linear scan agree.
 4865 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
 4866                                                    Register r_super_klass,
 4867                                                    Register result,
 4868                                                    Register temp1,
 4869                                                    Register temp2,
 4870                                                    Register temp3) {
 4871   const Register
 4872       r_array_index  = temp1,
 4873       r_array_length = temp2,
 4874       r_array_base   = temp3,
 4875       r_bitmap       = noreg;
 4876 
 4877   BLOCK_COMMENT("verify_secondary_supers_table {");
 4878 
 4879   Label L_success, L_failure, L_check, L_done;
 4880 
 4881   movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
 4882   movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
 4883   // And adjust the array base to point to the data.
 4884   addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
 4885 
 4886   testl(r_array_length, r_array_length); // array_length == 0?
 4887   jcc(Assembler::zero, L_failure);
 4888 
 4889   movl(r_array_index, 0);
 4890   repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
 4891   // fall through to L_failure
 4892 
 4893   const Register linear_result = r_array_index; // reuse temp1
 4894 
 4895   bind(L_failure); // not present
 4896   movl(linear_result, 1);
 4897   jmp(L_check);
 4898 
 4899   bind(L_success); // present
 4900   movl(linear_result, 0);
 4901 
 4902   bind(L_check);
 4903   cmpl(linear_result, result);
 4904   jcc(Assembler::equal, L_done);
 4905 
 4906   { // To avoid calling convention issues, build a record on the stack
 4907     // and pass the pointer to that instead.
 4908     push(result);
 4909     push(linear_result);
 4910     push(r_sub_klass);
 4911     push(r_super_klass);
 4912     movptr(c_rarg1, rsp);
 4913     movptr(c_rarg0, (uintptr_t) "mismatch");
 4914     call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
 4915     should_not_reach_here();
 4916   }
 4917   bind(L_done);
 4918 
 4919   BLOCK_COMMENT("} verify_secondary_supers_table");
 4920 }
 4921 
 4922 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
 4923 
 4924 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
 4925   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
 4926 
 4927   Label L_fallthrough;
 4928   if (L_fast_path == nullptr) {
 4929     L_fast_path = &L_fallthrough;
 4930   } else if (L_slow_path == nullptr) {
 4931     L_slow_path = &L_fallthrough;
 4932   }
 4933 
 4934   // Fast path check: class is fully initialized.
 4935   // init_state needs acquire, but x86 is TSO, and so we are already good.
 4936   cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
 4937   jcc(Assembler::equal, *L_fast_path);
 4938 
 4939   // Fast path check: current thread is initializer thread
 4940   cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
 4941   if (L_slow_path == &L_fallthrough) {
 4942     jcc(Assembler::equal, *L_fast_path);
 4943     bind(*L_slow_path);
 4944   } else if (L_fast_path == &L_fallthrough) {
 4945     jcc(Assembler::notEqual, *L_slow_path);
 4946     bind(*L_fast_path);
 4947   } else {
 4948     Unimplemented();
 4949   }
 4950 }
 4951 
 4952 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
 4953   if (VM_Version::supports_cmov()) {
 4954     cmovl(cc, dst, src);
 4955   } else {
 4956     Label L;
 4957     jccb(negate_condition(cc), L);
 4958     movl(dst, src);
 4959     bind(L);
 4960   }
 4961 }
 4962 
 4963 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4964   if (VM_Version::supports_cmov()) {
 4965     cmovl(cc, dst, src);
 4966   } else {
 4967     Label L;
 4968     jccb(negate_condition(cc), L);
 4969     movl(dst, src);
 4970     bind(L);
 4971   }
 4972 }
 4973 
 4974 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4975   if (!VerifyOops || VerifyAdapterSharing) {
 4976     // Below address of the code string confuses VerifyAdapterSharing
 4977     // because it may differ between otherwise equivalent adapters.
 4978     return;
 4979   }
 4980 
 4981   BLOCK_COMMENT("verify_oop {");
 4982   push(rscratch1);
 4983   push(rax);                          // save rax
 4984   push(reg);                          // pass register argument
 4985 
 4986   // Pass register number to verify_oop_subroutine
 4987   const char* b = nullptr;
 4988   {
 4989     ResourceMark rm;
 4990     stringStream ss;
 4991     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4992     b = code_string(ss.as_string());
 4993   }
 4994   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4995   pushptr(buffer.addr(), rscratch1);
 4996 
 4997   // call indirectly to solve generation ordering problem
 4998   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4999   call(rax);
 5000   // Caller pops the arguments (oop, message) and restores rax, r10
 5001   BLOCK_COMMENT("} verify_oop");
 5002 }
 5003 
 5004 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
 5005   if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
 5006     // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
 5007     // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
 5008     vpternlogd(dst, 0xFF, dst, dst, vector_len);
 5009   } else if (VM_Version::supports_avx()) {
 5010     vpcmpeqd(dst, dst, dst, vector_len);
 5011   } else {
 5012     pcmpeqd(dst, dst);
 5013   }
 5014 }
 5015 
 5016 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
 5017                                          int extra_slot_offset) {
 5018   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
 5019   int stackElementSize = Interpreter::stackElementSize;
 5020   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
 5021 #ifdef ASSERT
 5022   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
 5023   assert(offset1 - offset == stackElementSize, "correct arithmetic");
 5024 #endif
 5025   Register             scale_reg    = noreg;
 5026   Address::ScaleFactor scale_factor = Address::no_scale;
 5027   if (arg_slot.is_constant()) {
 5028     offset += arg_slot.as_constant() * stackElementSize;
 5029   } else {
 5030     scale_reg    = arg_slot.as_register();
 5031     scale_factor = Address::times(stackElementSize);
 5032   }
 5033   offset += wordSize;           // return PC is on stack
 5034   return Address(rsp, scale_reg, scale_factor, offset);
 5035 }
 5036 
 5037 // Handle the receiver type profile update given the "recv" klass.
 5038 //
 5039 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
 5040 // If there are no matching or claimable receiver entries in RD, updates
 5041 // the polymorphic counter.
 5042 //
 5043 // This code expected to run by either the interpreter or JIT-ed code, without
 5044 // extra synchronization. For safety, receiver cells are claimed atomically, which
 5045 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
 5046 // counter updates are not atomic.
 5047 //
 5048 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
 5049   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
 5050   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
 5051   int poly_count_offset      = in_bytes(CounterData::count_offset());
 5052   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
 5053   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
 5054 
 5055   // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
 5056   assert(is_aligned(mdp_offset, BytesPerWord), "sanity");
 5057   base_receiver_offset += mdp_offset;
 5058   end_receiver_offset  += mdp_offset;
 5059   poly_count_offset    += mdp_offset;
 5060 
 5061   // Scale down to optimize encoding. Slots are pointer-sized.
 5062   assert(is_aligned(base_receiver_offset,   BytesPerWord), "sanity");
 5063   assert(is_aligned(end_receiver_offset,    BytesPerWord), "sanity");
 5064   assert(is_aligned(poly_count_offset,      BytesPerWord), "sanity");
 5065   assert(is_aligned(receiver_step,          BytesPerWord), "sanity");
 5066   assert(is_aligned(receiver_to_count_step, BytesPerWord), "sanity");
 5067   base_receiver_offset   >>= LogBytesPerWord;
 5068   end_receiver_offset    >>= LogBytesPerWord;
 5069   poly_count_offset      >>= LogBytesPerWord;
 5070   receiver_step          >>= LogBytesPerWord;
 5071   receiver_to_count_step >>= LogBytesPerWord;
 5072 
 5073 #ifdef ASSERT
 5074   // We are about to walk the MDO slots without asking for offsets.
 5075   // Check that our math hits all the right spots.
 5076   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
 5077     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
 5078     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
 5079     int offset = base_receiver_offset + receiver_step*c;
 5080     int count_offset = offset + receiver_to_count_step;
 5081     assert((offset << LogBytesPerWord) == real_recv_offset, "receiver slot math");
 5082     assert((count_offset << LogBytesPerWord) == real_count_offset, "receiver count math");
 5083   }
 5084   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
 5085   assert(poly_count_offset << LogBytesPerWord == real_poly_count_offset, "poly counter math");
 5086 #endif
 5087 
 5088   // Corner case: no profile table. Increment poly counter and exit.
 5089   if (ReceiverTypeData::row_limit() == 0) {
 5090     addptr(Address(mdp, poly_count_offset, Address::times_ptr), DataLayout::counter_increment);
 5091     return;
 5092   }
 5093 
 5094   Register offset = rscratch1;
 5095 
 5096   Label L_loop_search_receiver, L_loop_search_empty;
 5097   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
 5098 
 5099   // The code here recognizes three major cases:
 5100   //   A. Fastest: receiver found in the table
 5101   //   B. Fast: no receiver in the table, and the table is full
 5102   //   C. Slow: no receiver in the table, free slots in the table
 5103   //
 5104   // The case A performance is most important, as perfectly-behaved code would end up
 5105   // there, especially with larger TypeProfileWidth. The case B performance is
 5106   // important as well, this is where bulk of code would land for normally megamorphic
 5107   // cases. The case C performance is not essential, its job is to deal with installation
 5108   // races, we optimize for code density instead. Case C needs to make sure that receiver
 5109   // rows are only claimed once. This makes sure we never overwrite a row for another
 5110   // receiver and never duplicate the receivers in the list, making profile type-accurate.
 5111   //
 5112   // It is very tempting to handle these cases in a single loop, and claim the first slot
 5113   // without checking the rest of the table. But, profiling code should tolerate free slots
 5114   // in the table, as class unloading can clear them. After such cleanup, the receiver
 5115   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
 5116   // to complete, before trying to install new slots. Splitting the code in several tight
 5117   // loops also helpfully optimizes for cases A and B.
 5118   //
 5119   // This code is effectively:
 5120   //
 5121   // restart:
 5122   //   // Fastest: receiver is already installed
 5123   //   for (i = 0; i < receiver_count(); i++) {
 5124   //     if (receiver(i) == recv) goto found_recv(i);
 5125   //   }
 5126   //
 5127   //   // Fast: no receiver, but profile is full
 5128   //   for (i = 0; i < receiver_count(); i++) {
 5129   //     if (receiver(i) == null) goto found_null(i);
 5130   //   }
 5131   //   goto polymorphic
 5132   //
 5133   //   // Slow: try to install receiver
 5134   // found_null(i):
 5135   //   CAS(&receiver(i), null, recv);
 5136   //   goto restart
 5137   //
 5138   // polymorphic:
 5139   //   count++;
 5140   //   return
 5141   //
 5142   // found_recv(i):
 5143   //   *receiver_count(i)++
 5144   //
 5145 
 5146   bind(L_restart);
 5147 
 5148   // Fastest: receiver is already installed
 5149   movptr(offset, base_receiver_offset);
 5150   bind(L_loop_search_receiver);
 5151     cmpptr(recv, Address(mdp, offset, Address::times_ptr));
 5152     jccb(Assembler::equal, L_found_recv);
 5153   addptr(offset, receiver_step);
 5154   cmpptr(offset, end_receiver_offset);
 5155   jccb(Assembler::notEqual, L_loop_search_receiver);
 5156 
 5157   // Fast: no receiver, but profile is full
 5158   movptr(offset, base_receiver_offset);
 5159   bind(L_loop_search_empty);
 5160     cmpptr(Address(mdp, offset, Address::times_ptr), NULL_WORD);
 5161     jccb(Assembler::equal, L_found_empty);
 5162   addptr(offset, receiver_step);
 5163   cmpptr(offset, end_receiver_offset);
 5164   jccb(Assembler::notEqual, L_loop_search_empty);
 5165   jmpb(L_polymorphic);
 5166 
 5167   // Slow: try to install receiver
 5168   bind(L_found_empty);
 5169 
 5170   // Atomically swing receiver slot: null -> recv.
 5171   //
 5172   // The update code uses CAS, which wants RAX register specifically, *and* it needs
 5173   // other important registers untouched, as they form the address. Therefore, we need
 5174   // to shift any important registers from RAX into some other spare register. If we
 5175   // have a spare register, we are forced to save it on stack here.
 5176 
 5177   Register spare_reg = noreg;
 5178   Register shifted_mdp = mdp;
 5179   Register shifted_recv = recv;
 5180   if (recv == rax || mdp == rax) {
 5181     spare_reg = (recv != rbx && mdp != rbx) ? rbx :
 5182                 (recv != rcx && mdp != rcx) ? rcx :
 5183                 rdx;
 5184     assert_different_registers(mdp, recv, offset, spare_reg);
 5185 
 5186     push(spare_reg);
 5187     if (recv == rax) {
 5188       movptr(spare_reg, recv);
 5189       shifted_recv = spare_reg;
 5190     } else {
 5191       assert(mdp == rax, "Remaining case");
 5192       movptr(spare_reg, mdp);
 5193       shifted_mdp = spare_reg;
 5194     }
 5195   } else {
 5196     push(rax);
 5197   }
 5198 
 5199   // None of the important registers are in RAX after this shuffle.
 5200   assert_different_registers(rax, shifted_mdp, shifted_recv, offset);
 5201 
 5202   xorptr(rax, rax);
 5203   cmpxchgptr(shifted_recv, Address(shifted_mdp, offset, Address::times_ptr));
 5204 
 5205   // Unshift registers.
 5206   if (recv == rax || mdp == rax) {
 5207     movptr(rax, spare_reg);
 5208     pop(spare_reg);
 5209   } else {
 5210     pop(rax);
 5211   }
 5212 
 5213   // CAS success means the slot now has the receiver we want. CAS failure means
 5214   // something had claimed the slot concurrently: it can be the same receiver we want,
 5215   // or something else. Since this is a slow path, we can optimize for code density,
 5216   // and just restart the search from the beginning.
 5217   jmpb(L_restart);
 5218 
 5219   // Counter updates:
 5220 
 5221   // Increment polymorphic counter instead of receiver slot.
 5222   bind(L_polymorphic);
 5223   movptr(offset, poly_count_offset);
 5224   jmpb(L_count_update);
 5225 
 5226   // Found a receiver, convert its slot offset to corresponding count offset.
 5227   bind(L_found_recv);
 5228   addptr(offset, receiver_to_count_step);
 5229 
 5230   bind(L_count_update);
 5231   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 5232 }
 5233 
 5234 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5235   if (!VerifyOops || VerifyAdapterSharing) {
 5236     // Below address of the code string confuses VerifyAdapterSharing
 5237     // because it may differ between otherwise equivalent adapters.
 5238     return;
 5239   }
 5240 
 5241   push(rscratch1);
 5242   push(rax); // save rax,
 5243   // addr may contain rsp so we will have to adjust it based on the push
 5244   // we just did (and on 64 bit we do two pushes)
 5245   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5246   // stores rax into addr which is backwards of what was intended.
 5247   if (addr.uses(rsp)) {
 5248     lea(rax, addr);
 5249     pushptr(Address(rax, 2 * BytesPerWord));
 5250   } else {
 5251     pushptr(addr);
 5252   }
 5253 
 5254   // Pass register number to verify_oop_subroutine
 5255   const char* b = nullptr;
 5256   {
 5257     ResourceMark rm;
 5258     stringStream ss;
 5259     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
 5260     b = code_string(ss.as_string());
 5261   }
 5262   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 5263   pushptr(buffer.addr(), rscratch1);
 5264 
 5265   // call indirectly to solve generation ordering problem
 5266   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 5267   call(rax);
 5268   // Caller pops the arguments (addr, message) and restores rax, r10.
 5269 }
 5270 
 5271 void MacroAssembler::verify_tlab() {
 5272 #ifdef ASSERT
 5273   if (UseTLAB && VerifyOops) {
 5274     Label next, ok;
 5275     Register t1 = rsi;
 5276 
 5277     push(t1);
 5278 
 5279     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5280     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
 5281     jcc(Assembler::aboveEqual, next);
 5282     STOP("assert(top >= start)");
 5283     should_not_reach_here();
 5284 
 5285     bind(next);
 5286     movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
 5287     cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
 5288     jcc(Assembler::aboveEqual, ok);
 5289     STOP("assert(top <= end)");
 5290     should_not_reach_here();
 5291 
 5292     bind(ok);
 5293     pop(t1);
 5294   }
 5295 #endif
 5296 }
 5297 
 5298 class ControlWord {
 5299  public:
 5300   int32_t _value;
 5301 
 5302   int  rounding_control() const        { return  (_value >> 10) & 3      ; }
 5303   int  precision_control() const       { return  (_value >>  8) & 3      ; }
 5304   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5305   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5306   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5307   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5308   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5309   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5310 
 5311   void print() const {
 5312     // rounding control
 5313     const char* rc;
 5314     switch (rounding_control()) {
 5315       case 0: rc = "round near"; break;
 5316       case 1: rc = "round down"; break;
 5317       case 2: rc = "round up  "; break;
 5318       case 3: rc = "chop      "; break;
 5319       default:
 5320         rc = nullptr; // silence compiler warnings
 5321         fatal("Unknown rounding control: %d", rounding_control());
 5322     };
 5323     // precision control
 5324     const char* pc;
 5325     switch (precision_control()) {
 5326       case 0: pc = "24 bits "; break;
 5327       case 1: pc = "reserved"; break;
 5328       case 2: pc = "53 bits "; break;
 5329       case 3: pc = "64 bits "; break;
 5330       default:
 5331         pc = nullptr; // silence compiler warnings
 5332         fatal("Unknown precision control: %d", precision_control());
 5333     };
 5334     // flags
 5335     char f[9];
 5336     f[0] = ' ';
 5337     f[1] = ' ';
 5338     f[2] = (precision   ()) ? 'P' : 'p';
 5339     f[3] = (underflow   ()) ? 'U' : 'u';
 5340     f[4] = (overflow    ()) ? 'O' : 'o';
 5341     f[5] = (zero_divide ()) ? 'Z' : 'z';
 5342     f[6] = (denormalized()) ? 'D' : 'd';
 5343     f[7] = (invalid     ()) ? 'I' : 'i';
 5344     f[8] = '\x0';
 5345     // output
 5346     printf("%04x  masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
 5347   }
 5348 
 5349 };
 5350 
 5351 class StatusWord {
 5352  public:
 5353   int32_t _value;
 5354 
 5355   bool busy() const                    { return ((_value >> 15) & 1) != 0; }
 5356   bool C3() const                      { return ((_value >> 14) & 1) != 0; }
 5357   bool C2() const                      { return ((_value >> 10) & 1) != 0; }
 5358   bool C1() const                      { return ((_value >>  9) & 1) != 0; }
 5359   bool C0() const                      { return ((_value >>  8) & 1) != 0; }
 5360   int  top() const                     { return  (_value >> 11) & 7      ; }
 5361   bool error_status() const            { return ((_value >>  7) & 1) != 0; }
 5362   bool stack_fault() const             { return ((_value >>  6) & 1) != 0; }
 5363   bool precision() const               { return ((_value >>  5) & 1) != 0; }
 5364   bool underflow() const               { return ((_value >>  4) & 1) != 0; }
 5365   bool overflow() const                { return ((_value >>  3) & 1) != 0; }
 5366   bool zero_divide() const             { return ((_value >>  2) & 1) != 0; }
 5367   bool denormalized() const            { return ((_value >>  1) & 1) != 0; }
 5368   bool invalid() const                 { return ((_value >>  0) & 1) != 0; }
 5369 
 5370   void print() const {
 5371     // condition codes
 5372     char c[5];
 5373     c[0] = (C3()) ? '3' : '-';
 5374     c[1] = (C2()) ? '2' : '-';
 5375     c[2] = (C1()) ? '1' : '-';
 5376     c[3] = (C0()) ? '0' : '-';
 5377     c[4] = '\x0';
 5378     // flags
 5379     char f[9];
 5380     f[0] = (error_status()) ? 'E' : '-';
 5381     f[1] = (stack_fault ()) ? 'S' : '-';
 5382     f[2] = (precision   ()) ? 'P' : '-';
 5383     f[3] = (underflow   ()) ? 'U' : '-';
 5384     f[4] = (overflow    ()) ? 'O' : '-';
 5385     f[5] = (zero_divide ()) ? 'Z' : '-';
 5386     f[6] = (denormalized()) ? 'D' : '-';
 5387     f[7] = (invalid     ()) ? 'I' : '-';
 5388     f[8] = '\x0';
 5389     // output
 5390     printf("%04x  flags = %s, cc =  %s, top = %d", _value & 0xFFFF, f, c, top());
 5391   }
 5392 
 5393 };
 5394 
 5395 class TagWord {
 5396  public:
 5397   int32_t _value;
 5398 
 5399   int tag_at(int i) const              { return (_value >> (i*2)) & 3; }
 5400 
 5401   void print() const {
 5402     printf("%04x", _value & 0xFFFF);
 5403   }
 5404 
 5405 };
 5406 
 5407 class FPU_Register {
 5408  public:
 5409   int32_t _m0;
 5410   int32_t _m1;
 5411   int16_t _ex;
 5412 
 5413   bool is_indefinite() const           {
 5414     return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
 5415   }
 5416 
 5417   void print() const {
 5418     char  sign = (_ex < 0) ? '-' : '+';
 5419     const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : "   ";
 5420     printf("%c%04hx.%08x%08x  %s", sign, _ex, _m1, _m0, kind);
 5421   };
 5422 
 5423 };
 5424 
 5425 class FPU_State {
 5426  public:
 5427   enum {
 5428     register_size       = 10,
 5429     number_of_registers =  8,
 5430     register_mask       =  7
 5431   };
 5432 
 5433   ControlWord  _control_word;
 5434   StatusWord   _status_word;
 5435   TagWord      _tag_word;
 5436   int32_t      _error_offset;
 5437   int32_t      _error_selector;
 5438   int32_t      _data_offset;
 5439   int32_t      _data_selector;
 5440   int8_t       _register[register_size * number_of_registers];
 5441 
 5442   int tag_for_st(int i) const          { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
 5443   FPU_Register* st(int i) const        { return (FPU_Register*)&_register[register_size * i]; }
 5444 
 5445   const char* tag_as_string(int tag) const {
 5446     switch (tag) {
 5447       case 0: return "valid";
 5448       case 1: return "zero";
 5449       case 2: return "special";
 5450       case 3: return "empty";
 5451     }
 5452     ShouldNotReachHere();
 5453     return nullptr;
 5454   }
 5455 
 5456   void print() const {
 5457     // print computation registers
 5458     { int t = _status_word.top();
 5459       for (int i = 0; i < number_of_registers; i++) {
 5460         int j = (i - t) & register_mask;
 5461         printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
 5462         st(j)->print();
 5463         printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
 5464       }
 5465     }
 5466     printf("\n");
 5467     // print control registers
 5468     printf("ctrl = "); _control_word.print(); printf("\n");
 5469     printf("stat = "); _status_word .print(); printf("\n");
 5470     printf("tags = "); _tag_word    .print(); printf("\n");
 5471   }
 5472 
 5473 };
 5474 
 5475 class Flag_Register {
 5476  public:
 5477   int32_t _value;
 5478 
 5479   bool overflow() const                { return ((_value >> 11) & 1) != 0; }
 5480   bool direction() const               { return ((_value >> 10) & 1) != 0; }
 5481   bool sign() const                    { return ((_value >>  7) & 1) != 0; }
 5482   bool zero() const                    { return ((_value >>  6) & 1) != 0; }
 5483   bool auxiliary_carry() const         { return ((_value >>  4) & 1) != 0; }
 5484   bool parity() const                  { return ((_value >>  2) & 1) != 0; }
 5485   bool carry() const                   { return ((_value >>  0) & 1) != 0; }
 5486 
 5487   void print() const {
 5488     // flags
 5489     char f[8];
 5490     f[0] = (overflow       ()) ? 'O' : '-';
 5491     f[1] = (direction      ()) ? 'D' : '-';
 5492     f[2] = (sign           ()) ? 'S' : '-';
 5493     f[3] = (zero           ()) ? 'Z' : '-';
 5494     f[4] = (auxiliary_carry()) ? 'A' : '-';
 5495     f[5] = (parity         ()) ? 'P' : '-';
 5496     f[6] = (carry          ()) ? 'C' : '-';
 5497     f[7] = '\x0';
 5498     // output
 5499     printf("%08x  flags = %s", _value, f);
 5500   }
 5501 
 5502 };
 5503 
 5504 class IU_Register {
 5505  public:
 5506   int32_t _value;
 5507 
 5508   void print() const {
 5509     printf("%08x  %11d", _value, _value);
 5510   }
 5511 
 5512 };
 5513 
 5514 class IU_State {
 5515  public:
 5516   Flag_Register _eflags;
 5517   IU_Register   _rdi;
 5518   IU_Register   _rsi;
 5519   IU_Register   _rbp;
 5520   IU_Register   _rsp;
 5521   IU_Register   _rbx;
 5522   IU_Register   _rdx;
 5523   IU_Register   _rcx;
 5524   IU_Register   _rax;
 5525 
 5526   void print() const {
 5527     // computation registers
 5528     printf("rax,  = "); _rax.print(); printf("\n");
 5529     printf("rbx,  = "); _rbx.print(); printf("\n");
 5530     printf("rcx  = "); _rcx.print(); printf("\n");
 5531     printf("rdx  = "); _rdx.print(); printf("\n");
 5532     printf("rdi  = "); _rdi.print(); printf("\n");
 5533     printf("rsi  = "); _rsi.print(); printf("\n");
 5534     printf("rbp,  = "); _rbp.print(); printf("\n");
 5535     printf("rsp  = "); _rsp.print(); printf("\n");
 5536     printf("\n");
 5537     // control registers
 5538     printf("flgs = "); _eflags.print(); printf("\n");
 5539   }
 5540 };
 5541 
 5542 
 5543 class CPU_State {
 5544  public:
 5545   FPU_State _fpu_state;
 5546   IU_State  _iu_state;
 5547 
 5548   void print() const {
 5549     printf("--------------------------------------------------\n");
 5550     _iu_state .print();
 5551     printf("\n");
 5552     _fpu_state.print();
 5553     printf("--------------------------------------------------\n");
 5554   }
 5555 
 5556 };
 5557 
 5558 
 5559 static void _print_CPU_state(CPU_State* state) {
 5560   state->print();
 5561 };
 5562 
 5563 
 5564 void MacroAssembler::print_CPU_state() {
 5565   push_CPU_state();
 5566   push(rsp);                // pass CPU state
 5567   call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
 5568   addptr(rsp, wordSize);       // discard argument
 5569   pop_CPU_state();
 5570 }
 5571 
 5572 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
 5573   // Either restore the MXCSR register after returning from the JNI Call
 5574   // or verify that it wasn't changed (with -Xcheck:jni flag).
 5575   if (VM_Version::supports_sse()) {
 5576     if (RestoreMXCSROnJNICalls) {
 5577       ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
 5578     } else if (CheckJNICalls) {
 5579       call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 5580     }
 5581   }
 5582   // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
 5583   vzeroupper();
 5584 }
 5585 
 5586 // ((OopHandle)result).resolve();
 5587 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
 5588   assert_different_registers(result, tmp);
 5589 
 5590   // Only 64 bit platforms support GCs that require a tmp register
 5591   // Only IN_HEAP loads require a thread_tmp register
 5592   // OopHandle::resolve is an indirection like jobject.
 5593   access_load_at(T_OBJECT, IN_NATIVE,
 5594                  result, Address(result, 0), tmp);
 5595 }
 5596 
 5597 // ((WeakHandle)result).resolve();
 5598 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
 5599   assert_different_registers(rresult, rtmp);
 5600   Label resolved;
 5601 
 5602   // A null weak handle resolves to null.
 5603   cmpptr(rresult, 0);
 5604   jcc(Assembler::equal, resolved);
 5605 
 5606   // Only 64 bit platforms support GCs that require a tmp register
 5607   // Only IN_HEAP loads require a thread_tmp register
 5608   // WeakHandle::resolve is an indirection like jweak.
 5609   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
 5610                  rresult, Address(rresult, 0), rtmp);
 5611   bind(resolved);
 5612 }
 5613 
 5614 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5615   // get mirror
 5616   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5617   load_method_holder(mirror, method);
 5618   movptr(mirror, Address(mirror, mirror_offset));
 5619   resolve_oop_handle(mirror, tmp);
 5620 }
 5621 
 5622 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5623   load_method_holder(rresult, rmethod);
 5624   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5625 }
 5626 
 5627 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5628   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5629   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5630   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5631 }
 5632 
 5633 void MacroAssembler::load_metadata(Register dst, Register src) {
 5634   if (UseCompactObjectHeaders) {
 5635     load_narrow_klass_compact(dst, src);
 5636   } else if (UseCompressedClassPointers) {
 5637     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5638   } else {
 5639     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5640   }
 5641 }
 5642 
 5643 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5644   assert(UseCompactObjectHeaders, "expect compact object headers");
 5645   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5646   shrq(dst, markWord::klass_shift);
 5647 }
 5648 
 5649 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5650   assert_different_registers(src, tmp);
 5651   assert_different_registers(dst, tmp);
 5652 
 5653   if (UseCompactObjectHeaders) {
 5654     load_narrow_klass_compact(dst, src);
 5655     decode_klass_not_null(dst, tmp);
 5656   } else if (UseCompressedClassPointers) {
 5657     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5658     decode_klass_not_null(dst, tmp);
 5659   } else {
 5660     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5661   }
 5662 }
 5663 
 5664 void MacroAssembler::load_prototype_header(Register dst, Register src, Register tmp) {
 5665   load_klass(dst, src, tmp);
 5666   movptr(dst, Address(dst, Klass::prototype_header_offset()));
 5667 }
 5668 
 5669 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5670   assert(!UseCompactObjectHeaders, "not with compact headers");
 5671   assert_different_registers(src, tmp);
 5672   assert_different_registers(dst, tmp);
 5673   if (UseCompressedClassPointers) {
 5674     encode_klass_not_null(src, tmp);
 5675     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5676   } else {
 5677     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5678   }
 5679 }
 5680 
 5681 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5682   if (UseCompactObjectHeaders) {
 5683     assert(tmp != noreg, "need tmp");
 5684     assert_different_registers(klass, obj, tmp);
 5685     load_narrow_klass_compact(tmp, obj);
 5686     cmpl(klass, tmp);
 5687   } else if (UseCompressedClassPointers) {
 5688     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5689   } else {
 5690     cmpptr(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
 5691   }
 5692 }
 5693 
 5694 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
 5695   if (UseCompactObjectHeaders) {
 5696     assert(tmp2 != noreg, "need tmp2");
 5697     assert_different_registers(obj1, obj2, tmp1, tmp2);
 5698     load_narrow_klass_compact(tmp1, obj1);
 5699     load_narrow_klass_compact(tmp2, obj2);
 5700     cmpl(tmp1, tmp2);
 5701   } else if (UseCompressedClassPointers) {
 5702     movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5703     cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5704   } else {
 5705     movptr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
 5706     cmpptr(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
 5707   }
 5708 }
 5709 
 5710 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 5711                                     Register tmp1) {
 5712   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5713   decorators = AccessInternal::decorator_fixup(decorators, type);
 5714   bool as_raw = (decorators & AS_RAW) != 0;
 5715   if (as_raw) {
 5716     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5717   } else {
 5718     bs->load_at(this, decorators, type, dst, src, tmp1);
 5719   }
 5720 }
 5721 
 5722 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5723                                      Register tmp1, Register tmp2, Register tmp3) {
 5724   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5725   decorators = AccessInternal::decorator_fixup(decorators, type);
 5726   bool as_raw = (decorators & AS_RAW) != 0;
 5727   if (as_raw) {
 5728     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5729   } else {
 5730     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5731   }
 5732 }
 5733 
 5734 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
 5735                                      Register inline_layout_info) {
 5736   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5737   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
 5738 }
 5739 
 5740 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
 5741   movptr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
 5742   movl(offset, Address(offset, InlineKlass::payload_offset_offset()));
 5743 }
 5744 
 5745 void MacroAssembler::payload_addr(Register oop, Register data, Register inline_klass) {
 5746   // ((address) (void*) o) + vk->payload_offset();
 5747   Register offset = (data == oop) ? rscratch1 : data;
 5748   payload_offset(inline_klass, offset);
 5749   if (data == oop) {
 5750     addptr(data, offset);
 5751   } else {
 5752     lea(data, Address(oop, offset));
 5753   }
 5754 }
 5755 
 5756 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
 5757                                                 Register index, Register data) {
 5758   assert(index != rcx, "index needs to shift by rcx");
 5759   assert_different_registers(array, array_klass, index);
 5760   assert_different_registers(rcx, array, index);
 5761 
 5762   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
 5763   movl(rcx, Address(array_klass, Klass::layout_helper_offset()));
 5764 
 5765   // Klass::layout_helper_log2_element_size(lh)
 5766   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
 5767   shrl(rcx, Klass::_lh_log2_element_size_shift);
 5768   andl(rcx, Klass::_lh_log2_element_size_mask);
 5769   shlptr(index); // index << rcx
 5770 
 5771   lea(data, Address(array, index, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT)));
 5772 }
 5773 
 5774 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5775   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5776 }
 5777 
 5778 // Doesn't do verification, generates fixed size code
 5779 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5780   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5781 }
 5782 
 5783 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5784                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5785   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5786 }
 5787 
 5788 // Used for storing nulls.
 5789 void MacroAssembler::store_heap_oop_null(Address dst) {
 5790   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5791 }
 5792 
 5793 void MacroAssembler::store_klass_gap(Register dst, Register src) {
 5794   assert(!UseCompactObjectHeaders, "Don't use with compact headers");
 5795   if (UseCompressedClassPointers) {
 5796     // Store to klass gap in destination
 5797     movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
 5798   }
 5799 }
 5800 
 5801 #ifdef ASSERT
 5802 void MacroAssembler::verify_heapbase(const char* msg) {
 5803   assert (UseCompressedOops, "should be compressed");
 5804   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5805   if (CheckCompressedOops) {
 5806     Label ok;
 5807     ExternalAddress src2(CompressedOops::base_addr());
 5808     const bool is_src2_reachable = reachable(src2);
 5809     if (!is_src2_reachable) {
 5810       push(rscratch1);  // cmpptr trashes rscratch1
 5811     }
 5812     cmpptr(r12_heapbase, src2, rscratch1);
 5813     jcc(Assembler::equal, ok);
 5814     STOP(msg);
 5815     bind(ok);
 5816     if (!is_src2_reachable) {
 5817       pop(rscratch1);
 5818     }
 5819   }
 5820 }
 5821 #endif
 5822 
 5823 // Algorithm must match oop.inline.hpp encode_heap_oop.
 5824 void MacroAssembler::encode_heap_oop(Register r) {
 5825 #ifdef ASSERT
 5826   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
 5827 #endif
 5828   verify_oop_msg(r, "broken oop in encode_heap_oop");
 5829   if (CompressedOops::base() == nullptr) {
 5830     if (CompressedOops::shift() != 0) {
 5831       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5832       shrq(r, LogMinObjAlignmentInBytes);
 5833     }
 5834     return;
 5835   }
 5836   testq(r, r);
 5837   cmovq(Assembler::equal, r, r12_heapbase);
 5838   subq(r, r12_heapbase);
 5839   shrq(r, LogMinObjAlignmentInBytes);
 5840 }
 5841 
 5842 void MacroAssembler::encode_heap_oop_not_null(Register r) {
 5843 #ifdef ASSERT
 5844   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
 5845   if (CheckCompressedOops) {
 5846     Label ok;
 5847     testq(r, r);
 5848     jcc(Assembler::notEqual, ok);
 5849     STOP("null oop passed to encode_heap_oop_not_null");
 5850     bind(ok);
 5851   }
 5852 #endif
 5853   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
 5854   if (CompressedOops::base() != nullptr) {
 5855     subq(r, r12_heapbase);
 5856   }
 5857   if (CompressedOops::shift() != 0) {
 5858     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5859     shrq(r, LogMinObjAlignmentInBytes);
 5860   }
 5861 }
 5862 
 5863 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
 5864 #ifdef ASSERT
 5865   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
 5866   if (CheckCompressedOops) {
 5867     Label ok;
 5868     testq(src, src);
 5869     jcc(Assembler::notEqual, ok);
 5870     STOP("null oop passed to encode_heap_oop_not_null2");
 5871     bind(ok);
 5872   }
 5873 #endif
 5874   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
 5875   if (dst != src) {
 5876     movq(dst, src);
 5877   }
 5878   if (CompressedOops::base() != nullptr) {
 5879     subq(dst, r12_heapbase);
 5880   }
 5881   if (CompressedOops::shift() != 0) {
 5882     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5883     shrq(dst, LogMinObjAlignmentInBytes);
 5884   }
 5885 }
 5886 
 5887 void  MacroAssembler::decode_heap_oop(Register r) {
 5888 #ifdef ASSERT
 5889   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
 5890 #endif
 5891   if (CompressedOops::base() == nullptr) {
 5892     if (CompressedOops::shift() != 0) {
 5893       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5894       shlq(r, LogMinObjAlignmentInBytes);
 5895     }
 5896   } else {
 5897     Label done;
 5898     shlq(r, LogMinObjAlignmentInBytes);
 5899     jccb(Assembler::equal, done);
 5900     addq(r, r12_heapbase);
 5901     bind(done);
 5902   }
 5903   verify_oop_msg(r, "broken oop in decode_heap_oop");
 5904 }
 5905 
 5906 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
 5907   // Note: it will change flags
 5908   assert (UseCompressedOops, "should only be used for compressed headers");
 5909   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5910   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5911   // vtableStubs also counts instructions in pd_code_size_limit.
 5912   // Also do not verify_oop as this is called by verify_oop.
 5913   if (CompressedOops::shift() != 0) {
 5914     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5915     shlq(r, LogMinObjAlignmentInBytes);
 5916     if (CompressedOops::base() != nullptr) {
 5917       addq(r, r12_heapbase);
 5918     }
 5919   } else {
 5920     assert (CompressedOops::base() == nullptr, "sanity");
 5921   }
 5922 }
 5923 
 5924 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
 5925   // Note: it will change flags
 5926   assert (UseCompressedOops, "should only be used for compressed headers");
 5927   assert (Universe::heap() != nullptr, "java heap should be initialized");
 5928   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5929   // vtableStubs also counts instructions in pd_code_size_limit.
 5930   // Also do not verify_oop as this is called by verify_oop.
 5931   if (CompressedOops::shift() != 0) {
 5932     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
 5933     if (LogMinObjAlignmentInBytes == Address::times_8) {
 5934       leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
 5935     } else {
 5936       if (dst != src) {
 5937         movq(dst, src);
 5938       }
 5939       shlq(dst, LogMinObjAlignmentInBytes);
 5940       if (CompressedOops::base() != nullptr) {
 5941         addq(dst, r12_heapbase);
 5942       }
 5943     }
 5944   } else {
 5945     assert (CompressedOops::base() == nullptr, "sanity");
 5946     if (dst != src) {
 5947       movq(dst, src);
 5948     }
 5949   }
 5950 }
 5951 
 5952 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
 5953   BLOCK_COMMENT("encode_klass_not_null {");
 5954   assert_different_registers(r, tmp);
 5955   if (CompressedKlassPointers::base() != nullptr) {
 5956     if (AOTCodeCache::is_on_for_dump()) {
 5957       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5958     } else {
 5959       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 5960     }
 5961     subq(r, tmp);
 5962   }
 5963   if (CompressedKlassPointers::shift() != 0) {
 5964     shrq(r, CompressedKlassPointers::shift());
 5965   }
 5966   BLOCK_COMMENT("} encode_klass_not_null");
 5967 }
 5968 
 5969 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
 5970   BLOCK_COMMENT("encode_and_move_klass_not_null {");
 5971   assert_different_registers(src, dst);
 5972   if (CompressedKlassPointers::base() != nullptr) {
 5973     movptr(dst, -(intptr_t)CompressedKlassPointers::base());
 5974     addq(dst, src);
 5975   } else {
 5976     movptr(dst, src);
 5977   }
 5978   if (CompressedKlassPointers::shift() != 0) {
 5979     shrq(dst, CompressedKlassPointers::shift());
 5980   }
 5981   BLOCK_COMMENT("} encode_and_move_klass_not_null");
 5982 }
 5983 
 5984 void  MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
 5985   BLOCK_COMMENT("decode_klass_not_null {");
 5986   assert_different_registers(r, tmp);
 5987   // Note: it will change flags
 5988   assert(UseCompressedClassPointers, "should only be used for compressed headers");
 5989   // Cannot assert, unverified entry point counts instructions (see .ad file)
 5990   // vtableStubs also counts instructions in pd_code_size_limit.
 5991   // Also do not verify_oop as this is called by verify_oop.
 5992   if (CompressedKlassPointers::shift() != 0) {
 5993     shlq(r, CompressedKlassPointers::shift());
 5994   }
 5995   if (CompressedKlassPointers::base() != nullptr) {
 5996     if (AOTCodeCache::is_on_for_dump()) {
 5997       movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
 5998     } else {
 5999       movptr(tmp, (intptr_t)CompressedKlassPointers::base());
 6000     }
 6001     addq(r, tmp);
 6002   }
 6003   BLOCK_COMMENT("} decode_klass_not_null");
 6004 }
 6005 
 6006 void  MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
 6007   BLOCK_COMMENT("decode_and_move_klass_not_null {");
 6008   assert_different_registers(src, dst);
 6009   // Note: it will change flags
 6010   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 6011   // Cannot assert, unverified entry point counts instructions (see .ad file)
 6012   // vtableStubs also counts instructions in pd_code_size_limit.
 6013   // Also do not verify_oop as this is called by verify_oop.
 6014 
 6015   if (CompressedKlassPointers::base() == nullptr &&
 6016       CompressedKlassPointers::shift() == 0) {
 6017     // The best case scenario is that there is no base or shift. Then it is already
 6018     // a pointer that needs nothing but a register rename.
 6019     movl(dst, src);
 6020   } else {
 6021     if (CompressedKlassPointers::shift() <= Address::times_8) {
 6022       if (CompressedKlassPointers::base() != nullptr) {
 6023         movptr(dst, (intptr_t)CompressedKlassPointers::base());
 6024       } else {
 6025         xorq(dst, dst);
 6026       }
 6027       if (CompressedKlassPointers::shift() != 0) {
 6028         assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
 6029         leaq(dst, Address(dst, src, Address::times_8, 0));
 6030       } else {
 6031         addq(dst, src);
 6032       }
 6033     } else {
 6034       if (CompressedKlassPointers::base() != nullptr) {
 6035         const intptr_t base_right_shifted =
 6036             (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
 6037         movptr(dst, base_right_shifted);
 6038       } else {
 6039         xorq(dst, dst);
 6040       }
 6041       addq(dst, src);
 6042       shlq(dst, CompressedKlassPointers::shift());
 6043     }
 6044   }
 6045   BLOCK_COMMENT("} decode_and_move_klass_not_null");
 6046 }
 6047 
 6048 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
 6049   assert (UseCompressedOops, "should only be used for compressed headers");
 6050   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6051   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6052   int oop_index = oop_recorder()->find_index(obj);
 6053   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6054   mov_narrow_oop(dst, oop_index, rspec);
 6055 }
 6056 
 6057 void  MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
 6058   assert (UseCompressedOops, "should only be used for compressed headers");
 6059   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6060   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6061   int oop_index = oop_recorder()->find_index(obj);
 6062   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6063   mov_narrow_oop(dst, oop_index, rspec);
 6064 }
 6065 
 6066 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
 6067   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 6068   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6069   int klass_index = oop_recorder()->find_index(k);
 6070   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6071   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6072 }
 6073 
 6074 void  MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
 6075   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 6076   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6077   int klass_index = oop_recorder()->find_index(k);
 6078   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6079   mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6080 }
 6081 
 6082 void  MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
 6083   assert (UseCompressedOops, "should only be used for compressed headers");
 6084   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6085   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6086   int oop_index = oop_recorder()->find_index(obj);
 6087   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6088   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 6089 }
 6090 
 6091 void  MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
 6092   assert (UseCompressedOops, "should only be used for compressed headers");
 6093   assert (Universe::heap() != nullptr, "java heap should be initialized");
 6094   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6095   int oop_index = oop_recorder()->find_index(obj);
 6096   RelocationHolder rspec = oop_Relocation::spec(oop_index);
 6097   Assembler::cmp_narrow_oop(dst, oop_index, rspec);
 6098 }
 6099 
 6100 void  MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
 6101   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 6102   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6103   int klass_index = oop_recorder()->find_index(k);
 6104   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6105   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6106 }
 6107 
 6108 void  MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
 6109   assert (UseCompressedClassPointers, "should only be used for compressed headers");
 6110   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6111   int klass_index = oop_recorder()->find_index(k);
 6112   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6113   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6114 }
 6115 
 6116 void MacroAssembler::reinit_heapbase() {
 6117   if (UseCompressedOops) {
 6118     if (Universe::heap() != nullptr) {
 6119       if (CompressedOops::base() == nullptr) {
 6120         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 6121       } else {
 6122         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 6123       }
 6124     } else {
 6125       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 6126     }
 6127   }
 6128 }
 6129 
 6130 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
 6131   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
 6132   // An inline type might be returned. If fields are in registers we
 6133   // need to allocate an inline type instance and initialize it with
 6134   // the value of the fields.
 6135   Label skip;
 6136   // We only need a new buffered inline type if a new one is not returned
 6137   testptr(rax, 1);
 6138   jcc(Assembler::zero, skip);
 6139   int call_offset = -1;
 6140 
 6141 #ifdef _LP64
 6142   // The following code is similar to allocate_instance but has some slight differences,
 6143   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
 6144   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
 6145   Label slow_case;
 6146   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
 6147   mov(rscratch1, rax); // save rax for slow_case since *_allocate may corrupt it when allocation failed
 6148   if (vk != nullptr) {
 6149     // Called from C1, where the return type is statically known.
 6150     movptr(rbx, (intptr_t)vk->get_InlineKlass());
 6151     jint lh = vk->layout_helper();
 6152     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
 6153     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
 6154       tlab_allocate(rax, noreg, lh, r13, r14, slow_case);
 6155     } else {
 6156       jmp(slow_case);
 6157     }
 6158   } else {
 6159     // Call from interpreter. RAX contains ((the InlineKlass* of the return type) | 0x01)
 6160     mov(rbx, rax);
 6161     andptr(rbx, -2);
 6162     if (UseTLAB) {
 6163       movl(r14, Address(rbx, Klass::layout_helper_offset()));
 6164       testl(r14, Klass::_lh_instance_slow_path_bit);
 6165       jcc(Assembler::notZero, slow_case);
 6166       tlab_allocate(rax, r14, 0, r13, r14, slow_case);
 6167     } else {
 6168       jmp(slow_case);
 6169     }
 6170   }
 6171   if (UseTLAB) {
 6172     // 2. Initialize buffered inline instance header
 6173     Register buffer_obj = rax;
 6174     Register klass = rbx;
 6175     if (UseCompactObjectHeaders) {
 6176       Register mark_word = r13;
 6177       movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 6178       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), mark_word);
 6179     } else {
 6180       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), (intptr_t)markWord::inline_type_prototype().value());
 6181       xorl(r13, r13);
 6182       store_klass_gap(buffer_obj, r13);
 6183       if (vk == nullptr) {
 6184         // store_klass corrupts rbx(klass), so save it in r13 for later use (interpreter case only).
 6185         mov(r13, klass);
 6186       }
 6187       store_klass(buffer_obj, klass, rscratch1);
 6188       klass = r13;
 6189     }
 6190     // 3. Initialize its fields with an inline class specific handler
 6191     if (vk != nullptr) {
 6192       call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
 6193     } else {
 6194       movptr(rbx, Address(klass, InlineKlass::adr_members_offset()));
 6195       movptr(rbx, Address(rbx, InlineKlass::pack_handler_offset()));
 6196       call(rbx);
 6197     }
 6198     jmp(skip);
 6199   }
 6200   bind(slow_case);
 6201   // We failed to allocate a new inline type, fall back to a runtime
 6202   // call. Some oop field may be live in some registers but we can't
 6203   // tell. That runtime call will take care of preserving them
 6204   // across a GC if there's one.
 6205   mov(rax, rscratch1);
 6206 #endif
 6207 
 6208   if (from_interpreter) {
 6209     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
 6210   } else {
 6211     call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
 6212     call_offset = offset();
 6213   }
 6214 
 6215   bind(skip);
 6216   return call_offset;
 6217 }
 6218 
 6219 // Move a value between registers/stack slots and update the reg_state
 6220 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
 6221   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
 6222   if (reg_state[to->value()] == reg_written) {
 6223     return true; // Already written
 6224   }
 6225   if (from != to && bt != T_VOID) {
 6226     if (reg_state[to->value()] == reg_readonly) {
 6227       return false; // Not yet writable
 6228     }
 6229     if (from->is_reg()) {
 6230       if (to->is_reg()) {
 6231         if (from->is_XMMRegister()) {
 6232           if (bt == T_DOUBLE) {
 6233             movdbl(to->as_XMMRegister(), from->as_XMMRegister());
 6234           } else {
 6235             assert(bt == T_FLOAT, "must be float");
 6236             movflt(to->as_XMMRegister(), from->as_XMMRegister());
 6237           }
 6238         } else {
 6239           movq(to->as_Register(), from->as_Register());
 6240         }
 6241       } else {
 6242         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6243         Address to_addr = Address(rsp, st_off);
 6244         if (from->is_XMMRegister()) {
 6245           if (bt == T_DOUBLE) {
 6246             movdbl(to_addr, from->as_XMMRegister());
 6247           } else {
 6248             assert(bt == T_FLOAT, "must be float");
 6249             movflt(to_addr, from->as_XMMRegister());
 6250           }
 6251         } else {
 6252           movq(to_addr, from->as_Register());
 6253         }
 6254       }
 6255     } else {
 6256       Address from_addr = Address(rsp, from->reg2stack() * VMRegImpl::stack_slot_size + wordSize);
 6257       if (to->is_reg()) {
 6258         if (to->is_XMMRegister()) {
 6259           if (bt == T_DOUBLE) {
 6260             movdbl(to->as_XMMRegister(), from_addr);
 6261           } else {
 6262             assert(bt == T_FLOAT, "must be float");
 6263             movflt(to->as_XMMRegister(), from_addr);
 6264           }
 6265         } else {
 6266           movq(to->as_Register(), from_addr);
 6267         }
 6268       } else {
 6269         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6270         movq(r13, from_addr);
 6271         movq(Address(rsp, st_off), r13);
 6272       }
 6273     }
 6274   }
 6275   // Update register states
 6276   reg_state[from->value()] = reg_writable;
 6277   reg_state[to->value()] = reg_written;
 6278   return true;
 6279 }
 6280 
 6281 // Calculate the extra stack space required for packing or unpacking inline
 6282 // args and adjust the stack pointer (see MacroAssembler::remove_frame).
 6283 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
 6284   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
 6285   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
 6286   assert(sp_inc > 0, "sanity");
 6287   // Two additional slots to account for return address
 6288   sp_inc +=  2 * VMRegImpl::stack_slot_size;
 6289 
 6290   push(rbp);
 6291   subptr(rsp, sp_inc);
 6292 #ifdef ASSERT
 6293   movl(Address(rsp, 0), badRegWordVal);
 6294   movl(Address(rsp, VMRegImpl::stack_slot_size), badRegWordVal);
 6295 #endif
 6296   return sp_inc + wordSize; // account for rbp space
 6297 }
 6298 
 6299 // Read all fields from an inline type buffer and store the field values in registers/stack slots.
 6300 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
 6301                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
 6302                                           RegState reg_state[]) {
 6303   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
 6304   assert(from->is_valid(), "source must be valid");
 6305   bool progress = false;
 6306 #ifdef ASSERT
 6307   const int start_offset = offset();
 6308 #endif
 6309 
 6310   Label L_null, L_notNull;
 6311   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
 6312   Register tmp1 = r10;
 6313   Register tmp2 = r13;
 6314   Register fromReg = noreg;
 6315   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
 6316   bool done = true;
 6317   bool mark_done = true;
 6318   VMReg toReg;
 6319   BasicType bt;
 6320   // Check if argument requires a null check
 6321   bool null_check = false;
 6322   VMReg nullCheckReg;
 6323   while (stream.next(nullCheckReg, bt)) {
 6324     if (sig->at(stream.sig_index())._offset == -1) {
 6325       null_check = true;
 6326       break;
 6327     }
 6328   }
 6329   stream.reset(sig_index, to_index);
 6330   while (stream.next(toReg, bt)) {
 6331     assert(toReg->is_valid(), "destination must be valid");
 6332     int idx = (int)toReg->value();
 6333     if (reg_state[idx] == reg_readonly) {
 6334       if (idx != from->value()) {
 6335         mark_done = false;
 6336       }
 6337       done = false;
 6338       continue;
 6339     } else if (reg_state[idx] == reg_written) {
 6340       continue;
 6341     }
 6342     assert(reg_state[idx] == reg_writable, "must be writable");
 6343     reg_state[idx] = reg_written;
 6344     progress = true;
 6345 
 6346     if (fromReg == noreg) {
 6347       if (from->is_reg()) {
 6348         fromReg = from->as_Register();
 6349       } else {
 6350         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6351         movq(tmp1, Address(rsp, st_off));
 6352         fromReg = tmp1;
 6353       }
 6354       if (null_check) {
 6355         // Nullable inline type argument, emit null check
 6356         testptr(fromReg, fromReg);
 6357         jcc(Assembler::zero, L_null);
 6358       }
 6359     }
 6360     int off = sig->at(stream.sig_index())._offset;
 6361     if (off == -1) {
 6362       assert(null_check, "Missing null check at");
 6363       if (toReg->is_stack()) {
 6364         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6365         movq(Address(rsp, st_off), 1);
 6366       } else {
 6367         movq(toReg->as_Register(), 1);
 6368       }
 6369       continue;
 6370     }
 6371     assert(off > 0, "offset in object should be positive");
 6372     Address fromAddr = Address(fromReg, off);
 6373     if (!toReg->is_XMMRegister()) {
 6374       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
 6375       if (is_reference_type(bt)) {
 6376         load_heap_oop(dst, fromAddr);
 6377       } else {
 6378         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
 6379         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
 6380       }
 6381       if (toReg->is_stack()) {
 6382         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6383         movq(Address(rsp, st_off), dst);
 6384       }
 6385     } else if (bt == T_DOUBLE) {
 6386       movdbl(toReg->as_XMMRegister(), fromAddr);
 6387     } else {
 6388       assert(bt == T_FLOAT, "must be float");
 6389       movflt(toReg->as_XMMRegister(), fromAddr);
 6390     }
 6391   }
 6392   if (progress && null_check) {
 6393     if (done) {
 6394       jmp(L_notNull);
 6395       bind(L_null);
 6396       // Set null marker to zero to signal that the argument is null.
 6397       // Also set all fields to zero since the runtime requires a canonical
 6398       // representation of a flat null.
 6399       stream.reset(sig_index, to_index);
 6400       while (stream.next(toReg, bt)) {
 6401         if (toReg->is_stack()) {
 6402           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6403           movq(Address(rsp, st_off), 0);
 6404         } else if (toReg->is_XMMRegister()) {
 6405           xorps(toReg->as_XMMRegister(), toReg->as_XMMRegister());
 6406         } else {
 6407           xorl(toReg->as_Register(), toReg->as_Register());
 6408         }
 6409       }
 6410       bind(L_notNull);
 6411     } else {
 6412       bind(L_null);
 6413     }
 6414   }
 6415 
 6416   sig_index = stream.sig_index();
 6417   to_index = stream.regs_index();
 6418 
 6419   if (mark_done && reg_state[from->value()] != reg_written) {
 6420     // This is okay because no one else will write to that slot
 6421     reg_state[from->value()] = reg_writable;
 6422   }
 6423   from_index--;
 6424   assert(progress || (start_offset == offset()), "should not emit code");
 6425   return done;
 6426 }
 6427 
 6428 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
 6429                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
 6430                                         RegState reg_state[], Register val_array) {
 6431   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
 6432   assert(to->is_valid(), "destination must be valid");
 6433 
 6434   if (reg_state[to->value()] == reg_written) {
 6435     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6436     return true; // Already written
 6437   }
 6438 
 6439   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
 6440   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
 6441   Register val_obj_tmp = r11;
 6442   Register from_reg_tmp = r14;
 6443   Register tmp1 = r10;
 6444   Register tmp2 = r13;
 6445   Register tmp3 = rbx;
 6446   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
 6447 
 6448   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
 6449 
 6450   if (reg_state[to->value()] == reg_readonly) {
 6451     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
 6452       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6453       return false; // Not yet writable
 6454     }
 6455     val_obj = val_obj_tmp;
 6456   }
 6457 
 6458   int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
 6459   load_heap_oop(val_obj, Address(val_array, index));
 6460 
 6461   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
 6462   VMReg fromReg;
 6463   BasicType bt;
 6464   Label L_null;
 6465   while (stream.next(fromReg, bt)) {
 6466     assert(fromReg->is_valid(), "source must be valid");
 6467     reg_state[fromReg->value()] = reg_writable;
 6468 
 6469     int off = sig->at(stream.sig_index())._offset;
 6470     if (off == -1) {
 6471       // Nullable inline type argument, emit null check
 6472       Label L_notNull;
 6473       if (fromReg->is_stack()) {
 6474         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6475         testb(Address(rsp, ld_off), 1);
 6476       } else {
 6477         testb(fromReg->as_Register(), 1);
 6478       }
 6479       jcc(Assembler::notZero, L_notNull);
 6480       movptr(val_obj, 0);
 6481       jmp(L_null);
 6482       bind(L_notNull);
 6483       continue;
 6484     }
 6485 
 6486     assert(off > 0, "offset in object should be positive");
 6487     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
 6488 
 6489     // Pack the scalarized field into the value object.
 6490     Address dst(val_obj, off);
 6491     if (!fromReg->is_XMMRegister()) {
 6492       Register src;
 6493       if (fromReg->is_stack()) {
 6494         src = from_reg_tmp;
 6495         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6496         load_sized_value(src, Address(rsp, ld_off), size_in_bytes, /* is_signed */ false);
 6497       } else {
 6498         src = fromReg->as_Register();
 6499       }
 6500       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
 6501       if (is_reference_type(bt)) {
 6502         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
 6503         mov(tmp3, val_obj);
 6504         Address dst_with_tmp3(tmp3, off);
 6505         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
 6506       } else {
 6507         store_sized_value(dst, src, size_in_bytes);
 6508       }
 6509     } else if (bt == T_DOUBLE) {
 6510       movdbl(dst, fromReg->as_XMMRegister());
 6511     } else {
 6512       assert(bt == T_FLOAT, "must be float");
 6513       movflt(dst, fromReg->as_XMMRegister());
 6514     }
 6515   }
 6516   bind(L_null);
 6517   sig_index = stream.sig_index();
 6518   from_index = stream.regs_index();
 6519 
 6520   assert(reg_state[to->value()] == reg_writable, "must have already been read");
 6521   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
 6522   assert(success, "to register must be writeable");
 6523   return true;
 6524 }
 6525 
 6526 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
 6527   return reg->is_XMMRegister() ? xmm8->as_VMReg() : r14->as_VMReg();
 6528 }
 6529 
 6530 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
 6531   assert((initial_framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 6532   if (needs_stack_repair) {
 6533     // The method has a scalarized entry point (where fields of value object arguments
 6534     // are passed through registers and stack), and a non-scalarized entry point (where
 6535     // value object arguments are given as oops). The non-scalarized entry point will
 6536     // first load each field of value object arguments and store them in registers and on
 6537     // the stack in a way compatible with the scalarized entry point. To do so, some extra
 6538     // stack space might be reserved (if argument registers are not enough). On leaving the
 6539     // method, this space must be freed.
 6540     //
 6541     // In case we used the non-scalarized entry point the stack looks like this:
 6542     //
 6543     // | Arguments from caller     |
 6544     // |---------------------------|  <-- caller's SP
 6545     // | Return address #1         |
 6546     // | Saved RBP #1              |
 6547     // |---------------------------|
 6548     // | Extension space for       |
 6549     // |   inline arg (un)packing  |
 6550     // |---------------------------|  <-- start of this method's frame
 6551     // | Return address #2         |
 6552     // | Saved RBP #2              |
 6553     // |---------------------------|  <-- RBP (with -XX:+PreserveFramePointer)
 6554     // | sp_inc                    |
 6555     // | method locals             |
 6556     // |---------------------------|  <-- SP
 6557     //
 6558     // Space for the return pc and saved rbp is reserved twice. But only the #1 copies
 6559     // contain the real values of return pc and saved rbp. The #2 copies are not reliable
 6560     // and should not be used. They are mostly needed to add space between the  extension
 6561     // space and the locals, as there would be between the real arguments and the locals
 6562     // if we don't need to do unpacking (from the scalarized entry point).
 6563     //
 6564     // When leaving, one must load RBP #1 into RBP, and use the copy #1 of the return address,
 6565     // while keeping in mind that from the scalarized entry point, there will be only one
 6566     // copy. Indeed, in the case we used the scalarized calling convention, the stack looks like this:
 6567     //
 6568     // | Arguments from caller     |
 6569     // |---------------------------|  <-- caller's SP
 6570     // | Return address            |
 6571     // | Saved RBP                 |
 6572     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
 6573     // | sp_inc                    |
 6574     // | method locals             |
 6575     // |---------------------------|  <-- SP
 6576     //
 6577     // The sp_inc stack slot holds the total size of the frame, including the extension
 6578     // space and copies #2 of the return address and the saved RBP (but never the copies
 6579     // #1 of the return address and saved RBP). That is how to find the copies #1 of the
 6580     // return address and saved rbp. This size is expressed in bytes. Be careful when using
 6581     // it from C++ in pointer arithmetic you might need to divide it by wordSize.
 6582 
 6583     // The stack increment resides just below the saved rbp
 6584     addq(rsp, Address(rsp, initial_framesize - wordSize));
 6585     pop(rbp);
 6586   } else {
 6587     if (initial_framesize > 0) {
 6588       addq(rsp, initial_framesize);
 6589     }
 6590     pop(rbp);
 6591   }
 6592 }
 6593 
 6594 #if COMPILER2_OR_JVMCI
 6595 
 6596 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 6597 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp, KRegister mask) {
 6598   // cnt - number of qwords (8-byte words).
 6599   // base - start address, qword aligned.
 6600   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 6601   bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
 6602   if (use64byteVector) {
 6603     evpbroadcastq(xtmp, val, AVX_512bit);
 6604   } else if (MaxVectorSize >= 32) {
 6605     movdq(xtmp, val);
 6606     punpcklqdq(xtmp, xtmp);
 6607     vinserti128_high(xtmp, xtmp);
 6608   } else {
 6609     movdq(xtmp, val);
 6610     punpcklqdq(xtmp, xtmp);
 6611   }
 6612   jmp(L_zero_64_bytes);
 6613 
 6614   BIND(L_loop);
 6615   if (MaxVectorSize >= 32) {
 6616     fill64(base, 0, xtmp, use64byteVector);
 6617   } else {
 6618     movdqu(Address(base,  0), xtmp);
 6619     movdqu(Address(base, 16), xtmp);
 6620     movdqu(Address(base, 32), xtmp);
 6621     movdqu(Address(base, 48), xtmp);
 6622   }
 6623   addptr(base, 64);
 6624 
 6625   BIND(L_zero_64_bytes);
 6626   subptr(cnt, 8);
 6627   jccb(Assembler::greaterEqual, L_loop);
 6628 
 6629   // Copy trailing 64 bytes
 6630   if (use64byteVector) {
 6631     addptr(cnt, 8);
 6632     jccb(Assembler::equal, L_end);
 6633     fill64_masked(3, base, 0, xtmp, mask, cnt, val, true);
 6634     jmp(L_end);
 6635   } else {
 6636     addptr(cnt, 4);
 6637     jccb(Assembler::less, L_tail);
 6638     if (MaxVectorSize >= 32) {
 6639       vmovdqu(Address(base, 0), xtmp);
 6640     } else {
 6641       movdqu(Address(base,  0), xtmp);
 6642       movdqu(Address(base, 16), xtmp);
 6643     }
 6644   }
 6645   addptr(base, 32);
 6646   subptr(cnt, 4);
 6647 
 6648   BIND(L_tail);
 6649   addptr(cnt, 4);
 6650   jccb(Assembler::lessEqual, L_end);
 6651   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 6652     fill32_masked(3, base, 0, xtmp, mask, cnt, val);
 6653   } else {
 6654     decrement(cnt);
 6655 
 6656     BIND(L_sloop);
 6657     movq(Address(base, 0), xtmp);
 6658     addptr(base, 8);
 6659     decrement(cnt);
 6660     jccb(Assembler::greaterEqual, L_sloop);
 6661   }
 6662   BIND(L_end);
 6663 }
 6664 
 6665 // Clearing constant sized memory using YMM/ZMM registers.
 6666 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 6667   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 6668   bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
 6669 
 6670   int vector64_count = (cnt & (~0x7)) >> 3;
 6671   cnt = cnt & 0x7;
 6672   const int fill64_per_loop = 4;
 6673   const int max_unrolled_fill64 = 8;
 6674 
 6675   // 64 byte initialization loop.
 6676   vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
 6677   int start64 = 0;
 6678   if (vector64_count > max_unrolled_fill64) {
 6679     Label LOOP;
 6680     Register index = rtmp;
 6681 
 6682     start64 = vector64_count - (vector64_count % fill64_per_loop);
 6683 
 6684     movl(index, 0);
 6685     BIND(LOOP);
 6686     for (int i = 0; i < fill64_per_loop; i++) {
 6687       fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
 6688     }
 6689     addl(index, fill64_per_loop * 64);
 6690     cmpl(index, start64 * 64);
 6691     jccb(Assembler::less, LOOP);
 6692   }
 6693   for (int i = start64; i < vector64_count; i++) {
 6694     fill64(base, i * 64, xtmp, use64byteVector);
 6695   }
 6696 
 6697   // Clear remaining 64 byte tail.
 6698   int disp = vector64_count * 64;
 6699   if (cnt) {
 6700     switch (cnt) {
 6701       case 1:
 6702         movq(Address(base, disp), xtmp);
 6703         break;
 6704       case 2:
 6705         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
 6706         break;
 6707       case 3:
 6708         movl(rtmp, 0x7);
 6709         kmovwl(mask, rtmp);
 6710         evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
 6711         break;
 6712       case 4:
 6713         evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6714         break;
 6715       case 5:
 6716         if (use64byteVector) {
 6717           movl(rtmp, 0x1F);
 6718           kmovwl(mask, rtmp);
 6719           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6720         } else {
 6721           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6722           movq(Address(base, disp + 32), xtmp);
 6723         }
 6724         break;
 6725       case 6:
 6726         if (use64byteVector) {
 6727           movl(rtmp, 0x3F);
 6728           kmovwl(mask, rtmp);
 6729           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6730         } else {
 6731           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6732           evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
 6733         }
 6734         break;
 6735       case 7:
 6736         if (use64byteVector) {
 6737           movl(rtmp, 0x7F);
 6738           kmovwl(mask, rtmp);
 6739           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6740         } else {
 6741           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6742           movl(rtmp, 0x7);
 6743           kmovwl(mask, rtmp);
 6744           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 6745         }
 6746         break;
 6747       default:
 6748         fatal("Unexpected length : %d\n",cnt);
 6749         break;
 6750     }
 6751   }
 6752 }
 6753 
 6754 void MacroAssembler::clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp,
 6755                                bool is_large, bool word_copy_only, KRegister mask) {
 6756   // cnt      - number of qwords (8-byte words).
 6757   // base     - start address, qword aligned.
 6758   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6759   assert(base==rdi, "base register must be edi for rep stos");
 6760   assert(val==rax,   "val register must be eax for rep stos");
 6761   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6762   assert(InitArrayShortSize % BytesPerLong == 0,
 6763     "InitArrayShortSize should be the multiple of BytesPerLong");
 6764 
 6765   Label DONE;
 6766 
 6767   if (!is_large) {
 6768     Label LOOP, LONG;
 6769     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6770     jccb(Assembler::greater, LONG);
 6771 
 6772     decrement(cnt);
 6773     jccb(Assembler::negative, DONE); // Zero length
 6774 
 6775     // Use individual pointer-sized stores for small counts:
 6776     BIND(LOOP);
 6777     movptr(Address(base, cnt, Address::times_ptr), val);
 6778     decrement(cnt);
 6779     jccb(Assembler::greaterEqual, LOOP);
 6780     jmpb(DONE);
 6781 
 6782     BIND(LONG);
 6783   }
 6784 
 6785   // Use longer rep-prefixed ops for non-small counts:
 6786   if (UseFastStosb && !word_copy_only) {
 6787     shlptr(cnt, 3); // convert to number of bytes
 6788     rep_stosb();
 6789   } else if (UseXMMForObjInit) {
 6790     xmm_clear_mem(base, cnt, val, xtmp, mask);
 6791   } else {
 6792     rep_stos();
 6793   }
 6794 
 6795   BIND(DONE);
 6796 }
 6797 
 6798 #endif //COMPILER2_OR_JVMCI
 6799 
 6800 
 6801 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6802                                    Register to, Register value, Register count,
 6803                                    Register rtmp, XMMRegister xtmp) {
 6804   ShortBranchVerifier sbv(this);
 6805   assert_different_registers(to, value, count, rtmp);
 6806   Label L_exit;
 6807   Label L_fill_2_bytes, L_fill_4_bytes;
 6808 
 6809 #if defined(COMPILER2)
 6810   if(MaxVectorSize >=32 &&
 6811      VM_Version::supports_avx512vlbw() &&
 6812      VM_Version::supports_bmi2()) {
 6813     generate_fill_avx3(t, to, value, count, rtmp, xtmp);
 6814     return;
 6815   }
 6816 #endif
 6817 
 6818   int shift = -1;
 6819   switch (t) {
 6820     case T_BYTE:
 6821       shift = 2;
 6822       break;
 6823     case T_SHORT:
 6824       shift = 1;
 6825       break;
 6826     case T_INT:
 6827       shift = 0;
 6828       break;
 6829     default: ShouldNotReachHere();
 6830   }
 6831 
 6832   if (t == T_BYTE) {
 6833     andl(value, 0xff);
 6834     movl(rtmp, value);
 6835     shll(rtmp, 8);
 6836     orl(value, rtmp);
 6837   }
 6838   if (t == T_SHORT) {
 6839     andl(value, 0xffff);
 6840   }
 6841   if (t == T_BYTE || t == T_SHORT) {
 6842     movl(rtmp, value);
 6843     shll(rtmp, 16);
 6844     orl(value, rtmp);
 6845   }
 6846 
 6847   cmpptr(count, 8 << shift); // Short arrays (< 32 bytes) fill by element
 6848   jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
 6849   if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
 6850     Label L_skip_align2;
 6851     // align source address at 4 bytes address boundary
 6852     if (t == T_BYTE) {
 6853       Label L_skip_align1;
 6854       // One byte misalignment happens only for byte arrays
 6855       testptr(to, 1);
 6856       jccb(Assembler::zero, L_skip_align1);
 6857       movb(Address(to, 0), value);
 6858       increment(to);
 6859       decrement(count);
 6860       BIND(L_skip_align1);
 6861     }
 6862     // Two bytes misalignment happens only for byte and short (char) arrays
 6863     testptr(to, 2);
 6864     jccb(Assembler::zero, L_skip_align2);
 6865     movw(Address(to, 0), value);
 6866     addptr(to, 2);
 6867     subptr(count, 1<<(shift-1));
 6868     BIND(L_skip_align2);
 6869   }
 6870   {
 6871     Label L_fill_32_bytes;
 6872     if (!UseUnalignedLoadStores) {
 6873       // align to 8 bytes, we know we are 4 byte aligned to start
 6874       testptr(to, 4);
 6875       jccb(Assembler::zero, L_fill_32_bytes);
 6876       movl(Address(to, 0), value);
 6877       addptr(to, 4);
 6878       subptr(count, 1<<shift);
 6879     }
 6880     BIND(L_fill_32_bytes);
 6881     {
 6882       Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
 6883       movdl(xtmp, value);
 6884       if (UseAVX >= 2 && UseUnalignedLoadStores) {
 6885         Label L_check_fill_32_bytes;
 6886         if (UseAVX > 2) {
 6887           // Fill 64-byte chunks
 6888           Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
 6889 
 6890           // If number of bytes to fill < VM_Version::avx3_threshold(), perform fill using AVX2
 6891           cmpptr(count, VM_Version::avx3_threshold());
 6892           jccb(Assembler::below, L_check_fill_64_bytes_avx2);
 6893 
 6894           vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
 6895 
 6896           subptr(count, 16 << shift);
 6897           jcc(Assembler::less, L_check_fill_32_bytes);
 6898           align(16);
 6899 
 6900           BIND(L_fill_64_bytes_loop_avx3);
 6901           evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
 6902           addptr(to, 64);
 6903           subptr(count, 16 << shift);
 6904           jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
 6905           jmpb(L_check_fill_32_bytes);
 6906 
 6907           BIND(L_check_fill_64_bytes_avx2);
 6908         }
 6909         // Fill 64-byte chunks
 6910         vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
 6911 
 6912         subptr(count, 16 << shift);
 6913         jcc(Assembler::less, L_check_fill_32_bytes);
 6914 
 6915         // align data for 64-byte chunks
 6916         Label L_fill_64_bytes_loop, L_align_64_bytes_loop;
 6917         if (EnableX86ECoreOpts) {
 6918             // align 'big' arrays to cache lines to minimize split_stores
 6919             cmpptr(count, 96 << shift);
 6920             jcc(Assembler::below, L_fill_64_bytes_loop);
 6921 
 6922             // Find the bytes needed for alignment
 6923             movptr(rtmp, to);
 6924             andptr(rtmp, 0x1c);
 6925             jcc(Assembler::zero, L_fill_64_bytes_loop);
 6926             negptr(rtmp);           // number of bytes to fill 32-rtmp. it filled by 2 mov by 32
 6927             addptr(rtmp, 32);
 6928             shrptr(rtmp, 2 - shift);// get number of elements from bytes
 6929             subptr(count, rtmp);    // adjust count by number of elements
 6930 
 6931             align(16);
 6932             BIND(L_align_64_bytes_loop);
 6933             movdl(Address(to, 0), xtmp);
 6934             addptr(to, 4);
 6935             subptr(rtmp, 1 << shift);
 6936             jcc(Assembler::greater, L_align_64_bytes_loop);
 6937         }
 6938 
 6939         align(16);
 6940         BIND(L_fill_64_bytes_loop);
 6941         vmovdqu(Address(to, 0), xtmp);
 6942         vmovdqu(Address(to, 32), xtmp);
 6943         addptr(to, 64);
 6944         subptr(count, 16 << shift);
 6945         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
 6946 
 6947         align(16);
 6948         BIND(L_check_fill_32_bytes);
 6949         addptr(count, 8 << shift);
 6950         jccb(Assembler::less, L_check_fill_8_bytes);
 6951         vmovdqu(Address(to, 0), xtmp);
 6952         addptr(to, 32);
 6953         subptr(count, 8 << shift);
 6954 
 6955         BIND(L_check_fill_8_bytes);
 6956         // clean upper bits of YMM registers
 6957         movdl(xtmp, value);
 6958         pshufd(xtmp, xtmp, 0);
 6959       } else {
 6960         // Fill 32-byte chunks
 6961         pshufd(xtmp, xtmp, 0);
 6962 
 6963         subptr(count, 8 << shift);
 6964         jcc(Assembler::less, L_check_fill_8_bytes);
 6965         align(16);
 6966 
 6967         BIND(L_fill_32_bytes_loop);
 6968 
 6969         if (UseUnalignedLoadStores) {
 6970           movdqu(Address(to, 0), xtmp);
 6971           movdqu(Address(to, 16), xtmp);
 6972         } else {
 6973           movq(Address(to, 0), xtmp);
 6974           movq(Address(to, 8), xtmp);
 6975           movq(Address(to, 16), xtmp);
 6976           movq(Address(to, 24), xtmp);
 6977         }
 6978 
 6979         addptr(to, 32);
 6980         subptr(count, 8 << shift);
 6981         jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
 6982 
 6983         BIND(L_check_fill_8_bytes);
 6984       }
 6985       addptr(count, 8 << shift);
 6986       jccb(Assembler::zero, L_exit);
 6987       jmpb(L_fill_8_bytes);
 6988 
 6989       //
 6990       // length is too short, just fill qwords
 6991       //
 6992       align(16);
 6993       BIND(L_fill_8_bytes_loop);
 6994       movq(Address(to, 0), xtmp);
 6995       addptr(to, 8);
 6996       BIND(L_fill_8_bytes);
 6997       subptr(count, 1 << (shift + 1));
 6998       jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
 6999     }
 7000   }
 7001 
 7002   Label L_fill_4_bytes_loop;
 7003   testl(count, 1 << shift);
 7004   jccb(Assembler::zero, L_fill_2_bytes);
 7005 
 7006   align(16);
 7007   BIND(L_fill_4_bytes_loop);
 7008   movl(Address(to, 0), value);
 7009   addptr(to, 4);
 7010 
 7011   BIND(L_fill_4_bytes);
 7012   subptr(count, 1 << shift);
 7013   jccb(Assembler::greaterEqual, L_fill_4_bytes_loop);
 7014 
 7015   if (t == T_BYTE || t == T_SHORT) {
 7016     Label L_fill_byte;
 7017     BIND(L_fill_2_bytes);
 7018     // fill trailing 2 bytes
 7019     testl(count, 1<<(shift-1));
 7020     jccb(Assembler::zero, L_fill_byte);
 7021     movw(Address(to, 0), value);
 7022     if (t == T_BYTE) {
 7023       addptr(to, 2);
 7024       BIND(L_fill_byte);
 7025       // fill trailing byte
 7026       testl(count, 1);
 7027       jccb(Assembler::zero, L_exit);
 7028       movb(Address(to, 0), value);
 7029     } else {
 7030       BIND(L_fill_byte);
 7031     }
 7032   } else {
 7033     BIND(L_fill_2_bytes);
 7034   }
 7035   BIND(L_exit);
 7036 }
 7037 
 7038 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
 7039   switch(type) {
 7040     case T_BYTE:
 7041     case T_BOOLEAN:
 7042       evpbroadcastb(dst, src, vector_len);
 7043       break;
 7044     case T_SHORT:
 7045     case T_CHAR:
 7046       evpbroadcastw(dst, src, vector_len);
 7047       break;
 7048     case T_INT:
 7049     case T_FLOAT:
 7050       evpbroadcastd(dst, src, vector_len);
 7051       break;
 7052     case T_LONG:
 7053     case T_DOUBLE:
 7054       evpbroadcastq(dst, src, vector_len);
 7055       break;
 7056     default:
 7057       fatal("Unhandled type : %s", type2name(type));
 7058       break;
 7059   }
 7060 }
 7061 
 7062 // Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
 7063 //
 7064 // @IntrinsicCandidate
 7065 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
 7066 //         char[] sa, int sp, byte[] da, int dp, int len) {
 7067 //     int i = 0;
 7068 //     for (; i < len; i++) {
 7069 //         char c = sa[sp++];
 7070 //         if (c > '\u00FF')
 7071 //             break;
 7072 //         da[dp++] = (byte) c;
 7073 //     }
 7074 //     return i;
 7075 // }
 7076 //
 7077 // @IntrinsicCandidate
 7078 // int java.lang.StringCoding.encodeISOArray0(
 7079 //         byte[] sa, int sp, byte[] da, int dp, int len) {
 7080 //   int i = 0;
 7081 //   for (; i < len; i++) {
 7082 //     char c = StringUTF16.getChar(sa, sp++);
 7083 //     if (c > '\u00FF')
 7084 //       break;
 7085 //     da[dp++] = (byte) c;
 7086 //   }
 7087 //   return i;
 7088 // }
 7089 //
 7090 // @IntrinsicCandidate
 7091 // int java.lang.StringCoding.encodeAsciiArray0(
 7092 //         char[] sa, int sp, byte[] da, int dp, int len) {
 7093 //   int i = 0;
 7094 //   for (; i < len; i++) {
 7095 //     char c = sa[sp++];
 7096 //     if (c >= '\u0080')
 7097 //       break;
 7098 //     da[dp++] = (byte) c;
 7099 //   }
 7100 //   return i;
 7101 // }
 7102 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
 7103   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 7104   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 7105   Register tmp5, Register result, bool ascii) {
 7106 
 7107   // rsi: src
 7108   // rdi: dst
 7109   // rdx: len
 7110   // rcx: tmp5
 7111   // rax: result
 7112   ShortBranchVerifier sbv(this);
 7113   assert_different_registers(src, dst, len, tmp5, result);
 7114   Label L_done, L_copy_1_char, L_copy_1_char_exit;
 7115 
 7116   int mask = ascii ? 0xff80ff80 : 0xff00ff00;
 7117   int short_mask = ascii ? 0xff80 : 0xff00;
 7118 
 7119   // set result
 7120   xorl(result, result);
 7121   // check for zero length
 7122   testl(len, len);
 7123   jcc(Assembler::zero, L_done);
 7124 
 7125   movl(result, len);
 7126 
 7127   // Setup pointers
 7128   lea(src, Address(src, len, Address::times_2)); // char[]
 7129   lea(dst, Address(dst, len, Address::times_1)); // byte[]
 7130   negptr(len);
 7131 
 7132   if (UseSSE42Intrinsics || UseAVX >= 2) {
 7133     Label L_copy_8_chars, L_copy_8_chars_exit;
 7134     Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
 7135 
 7136     if (UseAVX >= 2) {
 7137       Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
 7138       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7139       movdl(tmp1Reg, tmp5);
 7140       vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
 7141       jmp(L_chars_32_check);
 7142 
 7143       bind(L_copy_32_chars);
 7144       vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
 7145       vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
 7146       vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7147       vptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7148       jccb(Assembler::notZero, L_copy_32_chars_exit);
 7149       vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
 7150       vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
 7151       vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
 7152 
 7153       bind(L_chars_32_check);
 7154       addptr(len, 32);
 7155       jcc(Assembler::lessEqual, L_copy_32_chars);
 7156 
 7157       bind(L_copy_32_chars_exit);
 7158       subptr(len, 16);
 7159       jccb(Assembler::greater, L_copy_16_chars_exit);
 7160 
 7161     } else if (UseSSE42Intrinsics) {
 7162       movl(tmp5, mask);   // create mask to test for Unicode or non-ASCII chars in vector
 7163       movdl(tmp1Reg, tmp5);
 7164       pshufd(tmp1Reg, tmp1Reg, 0);
 7165       jmpb(L_chars_16_check);
 7166     }
 7167 
 7168     bind(L_copy_16_chars);
 7169     if (UseAVX >= 2) {
 7170       vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
 7171       vptest(tmp2Reg, tmp1Reg);
 7172       jcc(Assembler::notZero, L_copy_16_chars_exit);
 7173       vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
 7174       vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
 7175     } else {
 7176       if (UseAVX > 0) {
 7177         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7178         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7179         vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
 7180       } else {
 7181         movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
 7182         por(tmp2Reg, tmp3Reg);
 7183         movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
 7184         por(tmp2Reg, tmp4Reg);
 7185       }
 7186       ptest(tmp2Reg, tmp1Reg);       // check for Unicode or non-ASCII chars in vector
 7187       jccb(Assembler::notZero, L_copy_16_chars_exit);
 7188       packuswb(tmp3Reg, tmp4Reg);
 7189     }
 7190     movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
 7191 
 7192     bind(L_chars_16_check);
 7193     addptr(len, 16);
 7194     jcc(Assembler::lessEqual, L_copy_16_chars);
 7195 
 7196     bind(L_copy_16_chars_exit);
 7197     if (UseAVX >= 2) {
 7198       // clean upper bits of YMM registers
 7199       vpxor(tmp2Reg, tmp2Reg);
 7200       vpxor(tmp3Reg, tmp3Reg);
 7201       vpxor(tmp4Reg, tmp4Reg);
 7202       movdl(tmp1Reg, tmp5);
 7203       pshufd(tmp1Reg, tmp1Reg, 0);
 7204     }
 7205     subptr(len, 8);
 7206     jccb(Assembler::greater, L_copy_8_chars_exit);
 7207 
 7208     bind(L_copy_8_chars);
 7209     movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
 7210     ptest(tmp3Reg, tmp1Reg);
 7211     jccb(Assembler::notZero, L_copy_8_chars_exit);
 7212     packuswb(tmp3Reg, tmp1Reg);
 7213     movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
 7214     addptr(len, 8);
 7215     jccb(Assembler::lessEqual, L_copy_8_chars);
 7216 
 7217     bind(L_copy_8_chars_exit);
 7218     subptr(len, 8);
 7219     jccb(Assembler::zero, L_done);
 7220   }
 7221 
 7222   bind(L_copy_1_char);
 7223   load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
 7224   testl(tmp5, short_mask);      // check if Unicode or non-ASCII char
 7225   jccb(Assembler::notZero, L_copy_1_char_exit);
 7226   movb(Address(dst, len, Address::times_1, 0), tmp5);
 7227   addptr(len, 1);
 7228   jccb(Assembler::less, L_copy_1_char);
 7229 
 7230   bind(L_copy_1_char_exit);
 7231   addptr(result, len); // len is negative count of not processed elements
 7232 
 7233   bind(L_done);
 7234 }
 7235 
 7236 /**
 7237  * Helper for multiply_to_len().
 7238  */
 7239 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
 7240   addq(dest_lo, src1);
 7241   adcq(dest_hi, 0);
 7242   addq(dest_lo, src2);
 7243   adcq(dest_hi, 0);
 7244 }
 7245 
 7246 /**
 7247  * Multiply 64 bit by 64 bit first loop.
 7248  */
 7249 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 7250                                            Register y, Register y_idx, Register z,
 7251                                            Register carry, Register product,
 7252                                            Register idx, Register kdx) {
 7253   //
 7254   //  jlong carry, x[], y[], z[];
 7255   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7256   //    huge_128 product = y[idx] * x[xstart] + carry;
 7257   //    z[kdx] = (jlong)product;
 7258   //    carry  = (jlong)(product >>> 64);
 7259   //  }
 7260   //  z[xstart] = carry;
 7261   //
 7262 
 7263   Label L_first_loop, L_first_loop_exit;
 7264   Label L_one_x, L_one_y, L_multiply;
 7265 
 7266   decrementl(xstart);
 7267   jcc(Assembler::negative, L_one_x);
 7268 
 7269   movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7270   rorq(x_xstart, 32); // convert big-endian to little-endian
 7271 
 7272   bind(L_first_loop);
 7273   decrementl(idx);
 7274   jcc(Assembler::negative, L_first_loop_exit);
 7275   decrementl(idx);
 7276   jcc(Assembler::negative, L_one_y);
 7277   movq(y_idx, Address(y, idx, Address::times_4,  0));
 7278   rorq(y_idx, 32); // convert big-endian to little-endian
 7279   bind(L_multiply);
 7280   movq(product, x_xstart);
 7281   mulq(y_idx); // product(rax) * y_idx -> rdx:rax
 7282   addq(product, carry);
 7283   adcq(rdx, 0);
 7284   subl(kdx, 2);
 7285   movl(Address(z, kdx, Address::times_4,  4), product);
 7286   shrq(product, 32);
 7287   movl(Address(z, kdx, Address::times_4,  0), product);
 7288   movq(carry, rdx);
 7289   jmp(L_first_loop);
 7290 
 7291   bind(L_one_y);
 7292   movl(y_idx, Address(y,  0));
 7293   jmp(L_multiply);
 7294 
 7295   bind(L_one_x);
 7296   movl(x_xstart, Address(x,  0));
 7297   jmp(L_first_loop);
 7298 
 7299   bind(L_first_loop_exit);
 7300 }
 7301 
 7302 /**
 7303  * Multiply 64 bit by 64 bit and add 128 bit.
 7304  */
 7305 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 7306                                             Register yz_idx, Register idx,
 7307                                             Register carry, Register product, int offset) {
 7308   //     huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
 7309   //     z[kdx] = (jlong)product;
 7310 
 7311   movq(yz_idx, Address(y, idx, Address::times_4,  offset));
 7312   rorq(yz_idx, 32); // convert big-endian to little-endian
 7313   movq(product, x_xstart);
 7314   mulq(yz_idx);     // product(rax) * yz_idx -> rdx:product(rax)
 7315   movq(yz_idx, Address(z, idx, Address::times_4,  offset));
 7316   rorq(yz_idx, 32); // convert big-endian to little-endian
 7317 
 7318   add2_with_carry(rdx, product, carry, yz_idx);
 7319 
 7320   movl(Address(z, idx, Address::times_4,  offset+4), product);
 7321   shrq(product, 32);
 7322   movl(Address(z, idx, Address::times_4,  offset), product);
 7323 
 7324 }
 7325 
 7326 /**
 7327  * Multiply 128 bit by 128 bit. Unrolled inner loop.
 7328  */
 7329 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
 7330                                              Register yz_idx, Register idx, Register jdx,
 7331                                              Register carry, Register product,
 7332                                              Register carry2) {
 7333   //   jlong carry, x[], y[], z[];
 7334   //   int kdx = ystart+1;
 7335   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7336   //     huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
 7337   //     z[kdx+idx+1] = (jlong)product;
 7338   //     jlong carry2  = (jlong)(product >>> 64);
 7339   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
 7340   //     z[kdx+idx] = (jlong)product;
 7341   //     carry  = (jlong)(product >>> 64);
 7342   //   }
 7343   //   idx += 2;
 7344   //   if (idx > 0) {
 7345   //     product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
 7346   //     z[kdx+idx] = (jlong)product;
 7347   //     carry  = (jlong)(product >>> 64);
 7348   //   }
 7349   //
 7350 
 7351   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7352 
 7353   movl(jdx, idx);
 7354   andl(jdx, 0xFFFFFFFC);
 7355   shrl(jdx, 2);
 7356 
 7357   bind(L_third_loop);
 7358   subl(jdx, 1);
 7359   jcc(Assembler::negative, L_third_loop_exit);
 7360   subl(idx, 4);
 7361 
 7362   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
 7363   movq(carry2, rdx);
 7364 
 7365   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
 7366   movq(carry, rdx);
 7367   jmp(L_third_loop);
 7368 
 7369   bind (L_third_loop_exit);
 7370 
 7371   andl (idx, 0x3);
 7372   jcc(Assembler::zero, L_post_third_loop_done);
 7373 
 7374   Label L_check_1;
 7375   subl(idx, 2);
 7376   jcc(Assembler::negative, L_check_1);
 7377 
 7378   multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
 7379   movq(carry, rdx);
 7380 
 7381   bind (L_check_1);
 7382   addl (idx, 0x2);
 7383   andl (idx, 0x1);
 7384   subl(idx, 1);
 7385   jcc(Assembler::negative, L_post_third_loop_done);
 7386 
 7387   movl(yz_idx, Address(y, idx, Address::times_4,  0));
 7388   movq(product, x_xstart);
 7389   mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
 7390   movl(yz_idx, Address(z, idx, Address::times_4,  0));
 7391 
 7392   add2_with_carry(rdx, product, yz_idx, carry);
 7393 
 7394   movl(Address(z, idx, Address::times_4,  0), product);
 7395   shrq(product, 32);
 7396 
 7397   shlq(rdx, 32);
 7398   orq(product, rdx);
 7399   movq(carry, product);
 7400 
 7401   bind(L_post_third_loop_done);
 7402 }
 7403 
 7404 /**
 7405  * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
 7406  *
 7407  */
 7408 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
 7409                                                   Register carry, Register carry2,
 7410                                                   Register idx, Register jdx,
 7411                                                   Register yz_idx1, Register yz_idx2,
 7412                                                   Register tmp, Register tmp3, Register tmp4) {
 7413   assert(UseBMI2Instructions, "should be used only when BMI2 is available");
 7414 
 7415   //   jlong carry, x[], y[], z[];
 7416   //   int kdx = ystart+1;
 7417   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
 7418   //     huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
 7419   //     jlong carry2  = (jlong)(tmp3 >>> 64);
 7420   //     huge_128 tmp4 = (y[idx]   * rdx) + z[kdx+idx] + carry2;
 7421   //     carry  = (jlong)(tmp4 >>> 64);
 7422   //     z[kdx+idx+1] = (jlong)tmp3;
 7423   //     z[kdx+idx] = (jlong)tmp4;
 7424   //   }
 7425   //   idx += 2;
 7426   //   if (idx > 0) {
 7427   //     yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
 7428   //     z[kdx+idx] = (jlong)yz_idx1;
 7429   //     carry  = (jlong)(yz_idx1 >>> 64);
 7430   //   }
 7431   //
 7432 
 7433   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
 7434 
 7435   movl(jdx, idx);
 7436   andl(jdx, 0xFFFFFFFC);
 7437   shrl(jdx, 2);
 7438 
 7439   bind(L_third_loop);
 7440   subl(jdx, 1);
 7441   jcc(Assembler::negative, L_third_loop_exit);
 7442   subl(idx, 4);
 7443 
 7444   movq(yz_idx1,  Address(y, idx, Address::times_4,  8));
 7445   rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
 7446   movq(yz_idx2, Address(y, idx, Address::times_4,  0));
 7447   rorxq(yz_idx2, yz_idx2, 32);
 7448 
 7449   mulxq(tmp4, tmp3, yz_idx1);  //  yz_idx1 * rdx -> tmp4:tmp3
 7450   mulxq(carry2, tmp, yz_idx2); //  yz_idx2 * rdx -> carry2:tmp
 7451 
 7452   movq(yz_idx1,  Address(z, idx, Address::times_4,  8));
 7453   rorxq(yz_idx1, yz_idx1, 32);
 7454   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7455   rorxq(yz_idx2, yz_idx2, 32);
 7456 
 7457   if (VM_Version::supports_adx()) {
 7458     adcxq(tmp3, carry);
 7459     adoxq(tmp3, yz_idx1);
 7460 
 7461     adcxq(tmp4, tmp);
 7462     adoxq(tmp4, yz_idx2);
 7463 
 7464     movl(carry, 0); // does not affect flags
 7465     adcxq(carry2, carry);
 7466     adoxq(carry2, carry);
 7467   } else {
 7468     add2_with_carry(tmp4, tmp3, carry, yz_idx1);
 7469     add2_with_carry(carry2, tmp4, tmp, yz_idx2);
 7470   }
 7471   movq(carry, carry2);
 7472 
 7473   movl(Address(z, idx, Address::times_4, 12), tmp3);
 7474   shrq(tmp3, 32);
 7475   movl(Address(z, idx, Address::times_4,  8), tmp3);
 7476 
 7477   movl(Address(z, idx, Address::times_4,  4), tmp4);
 7478   shrq(tmp4, 32);
 7479   movl(Address(z, idx, Address::times_4,  0), tmp4);
 7480 
 7481   jmp(L_third_loop);
 7482 
 7483   bind (L_third_loop_exit);
 7484 
 7485   andl (idx, 0x3);
 7486   jcc(Assembler::zero, L_post_third_loop_done);
 7487 
 7488   Label L_check_1;
 7489   subl(idx, 2);
 7490   jcc(Assembler::negative, L_check_1);
 7491 
 7492   movq(yz_idx1, Address(y, idx, Address::times_4,  0));
 7493   rorxq(yz_idx1, yz_idx1, 32);
 7494   mulxq(tmp4, tmp3, yz_idx1); //  yz_idx1 * rdx -> tmp4:tmp3
 7495   movq(yz_idx2, Address(z, idx, Address::times_4,  0));
 7496   rorxq(yz_idx2, yz_idx2, 32);
 7497 
 7498   add2_with_carry(tmp4, tmp3, carry, yz_idx2);
 7499 
 7500   movl(Address(z, idx, Address::times_4,  4), tmp3);
 7501   shrq(tmp3, 32);
 7502   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7503   movq(carry, tmp4);
 7504 
 7505   bind (L_check_1);
 7506   addl (idx, 0x2);
 7507   andl (idx, 0x1);
 7508   subl(idx, 1);
 7509   jcc(Assembler::negative, L_post_third_loop_done);
 7510   movl(tmp4, Address(y, idx, Address::times_4,  0));
 7511   mulxq(carry2, tmp3, tmp4);  //  tmp4 * rdx -> carry2:tmp3
 7512   movl(tmp4, Address(z, idx, Address::times_4,  0));
 7513 
 7514   add2_with_carry(carry2, tmp3, tmp4, carry);
 7515 
 7516   movl(Address(z, idx, Address::times_4,  0), tmp3);
 7517   shrq(tmp3, 32);
 7518 
 7519   shlq(carry2, 32);
 7520   orq(tmp3, carry2);
 7521   movq(carry, tmp3);
 7522 
 7523   bind(L_post_third_loop_done);
 7524 }
 7525 
 7526 /**
 7527  * Code for BigInteger::multiplyToLen() intrinsic.
 7528  *
 7529  * rdi: x
 7530  * rax: xlen
 7531  * rsi: y
 7532  * rcx: ylen
 7533  * r8:  z
 7534  * r11: tmp0
 7535  * r12: tmp1
 7536  * r13: tmp2
 7537  * r14: tmp3
 7538  * r15: tmp4
 7539  * rbx: tmp5
 7540  *
 7541  */
 7542 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
 7543                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
 7544   ShortBranchVerifier sbv(this);
 7545   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
 7546 
 7547   push(tmp0);
 7548   push(tmp1);
 7549   push(tmp2);
 7550   push(tmp3);
 7551   push(tmp4);
 7552   push(tmp5);
 7553 
 7554   push(xlen);
 7555 
 7556   const Register idx = tmp1;
 7557   const Register kdx = tmp2;
 7558   const Register xstart = tmp3;
 7559 
 7560   const Register y_idx = tmp4;
 7561   const Register carry = tmp5;
 7562   const Register product  = xlen;
 7563   const Register x_xstart = tmp0;
 7564 
 7565   // First Loop.
 7566   //
 7567   //  final static long LONG_MASK = 0xffffffffL;
 7568   //  int xstart = xlen - 1;
 7569   //  int ystart = ylen - 1;
 7570   //  long carry = 0;
 7571   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
 7572   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
 7573   //    z[kdx] = (int)product;
 7574   //    carry = product >>> 32;
 7575   //  }
 7576   //  z[xstart] = (int)carry;
 7577   //
 7578 
 7579   movl(idx, ylen);               // idx = ylen;
 7580   lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
 7581   xorq(carry, carry);            // carry = 0;
 7582 
 7583   Label L_done;
 7584 
 7585   movl(xstart, xlen);
 7586   decrementl(xstart);
 7587   jcc(Assembler::negative, L_done);
 7588 
 7589   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
 7590 
 7591   Label L_second_loop;
 7592   testl(kdx, kdx);
 7593   jcc(Assembler::zero, L_second_loop);
 7594 
 7595   Label L_carry;
 7596   subl(kdx, 1);
 7597   jcc(Assembler::zero, L_carry);
 7598 
 7599   movl(Address(z, kdx, Address::times_4,  0), carry);
 7600   shrq(carry, 32);
 7601   subl(kdx, 1);
 7602 
 7603   bind(L_carry);
 7604   movl(Address(z, kdx, Address::times_4,  0), carry);
 7605 
 7606   // Second and third (nested) loops.
 7607   //
 7608   // for (int i = xstart-1; i >= 0; i--) { // Second loop
 7609   //   carry = 0;
 7610   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
 7611   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
 7612   //                    (z[k] & LONG_MASK) + carry;
 7613   //     z[k] = (int)product;
 7614   //     carry = product >>> 32;
 7615   //   }
 7616   //   z[i] = (int)carry;
 7617   // }
 7618   //
 7619   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
 7620 
 7621   const Register jdx = tmp1;
 7622 
 7623   bind(L_second_loop);
 7624   xorl(carry, carry);    // carry = 0;
 7625   movl(jdx, ylen);       // j = ystart+1
 7626 
 7627   subl(xstart, 1);       // i = xstart-1;
 7628   jcc(Assembler::negative, L_done);
 7629 
 7630   push (z);
 7631 
 7632   Label L_last_x;
 7633   lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
 7634   subl(xstart, 1);       // i = xstart-1;
 7635   jcc(Assembler::negative, L_last_x);
 7636 
 7637   if (UseBMI2Instructions) {
 7638     movq(rdx,  Address(x, xstart, Address::times_4,  0));
 7639     rorxq(rdx, rdx, 32); // convert big-endian to little-endian
 7640   } else {
 7641     movq(x_xstart, Address(x, xstart, Address::times_4,  0));
 7642     rorq(x_xstart, 32);  // convert big-endian to little-endian
 7643   }
 7644 
 7645   Label L_third_loop_prologue;
 7646   bind(L_third_loop_prologue);
 7647 
 7648   push (x);
 7649   push (xstart);
 7650   push (ylen);
 7651 
 7652 
 7653   if (UseBMI2Instructions) {
 7654     multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
 7655   } else { // !UseBMI2Instructions
 7656     multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
 7657   }
 7658 
 7659   pop(ylen);
 7660   pop(xlen);
 7661   pop(x);
 7662   pop(z);
 7663 
 7664   movl(tmp3, xlen);
 7665   addl(tmp3, 1);
 7666   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7667   subl(tmp3, 1);
 7668   jccb(Assembler::negative, L_done);
 7669 
 7670   shrq(carry, 32);
 7671   movl(Address(z, tmp3, Address::times_4,  0), carry);
 7672   jmp(L_second_loop);
 7673 
 7674   // Next infrequent code is moved outside loops.
 7675   bind(L_last_x);
 7676   if (UseBMI2Instructions) {
 7677     movl(rdx, Address(x,  0));
 7678   } else {
 7679     movl(x_xstart, Address(x,  0));
 7680   }
 7681   jmp(L_third_loop_prologue);
 7682 
 7683   bind(L_done);
 7684 
 7685   pop(xlen);
 7686 
 7687   pop(tmp5);
 7688   pop(tmp4);
 7689   pop(tmp3);
 7690   pop(tmp2);
 7691   pop(tmp1);
 7692   pop(tmp0);
 7693 }
 7694 
 7695 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
 7696   Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
 7697   assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
 7698   Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
 7699   Label VECTOR8_TAIL, VECTOR4_TAIL;
 7700   Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
 7701   Label SAME_TILL_END, DONE;
 7702   Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
 7703 
 7704   //scale is in rcx in both Win64 and Unix
 7705   ShortBranchVerifier sbv(this);
 7706 
 7707   shlq(length);
 7708   xorq(result, result);
 7709 
 7710   if ((AVX3Threshold == 0) && (UseAVX > 2) &&
 7711       VM_Version::supports_avx512vlbw()) {
 7712     Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
 7713 
 7714     cmpq(length, 64);
 7715     jcc(Assembler::less, VECTOR32_TAIL);
 7716 
 7717     movq(tmp1, length);
 7718     andq(tmp1, 0x3F);      // tail count
 7719     andq(length, ~(0x3F)); //vector count
 7720 
 7721     bind(VECTOR64_LOOP);
 7722     // AVX512 code to compare 64 byte vectors.
 7723     evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
 7724     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7725     kortestql(k7, k7);
 7726     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
 7727     addq(result, 64);
 7728     subq(length, 64);
 7729     jccb(Assembler::notZero, VECTOR64_LOOP);
 7730 
 7731     //bind(VECTOR64_TAIL);
 7732     testq(tmp1, tmp1);
 7733     jcc(Assembler::zero, SAME_TILL_END);
 7734 
 7735     //bind(VECTOR64_TAIL);
 7736     // AVX512 code to compare up to 63 byte vectors.
 7737     mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
 7738     shlxq(tmp2, tmp2, tmp1);
 7739     notq(tmp2);
 7740     kmovql(k3, tmp2);
 7741 
 7742     evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
 7743     evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
 7744 
 7745     ktestql(k7, k3);
 7746     jcc(Assembler::below, SAME_TILL_END);     // not mismatch
 7747 
 7748     bind(VECTOR64_NOT_EQUAL);
 7749     kmovql(tmp1, k7);
 7750     notq(tmp1);
 7751     tzcntq(tmp1, tmp1);
 7752     addq(result, tmp1);
 7753     shrq(result);
 7754     jmp(DONE);
 7755     bind(VECTOR32_TAIL);
 7756   }
 7757 
 7758   cmpq(length, 8);
 7759   jcc(Assembler::equal, VECTOR8_LOOP);
 7760   jcc(Assembler::less, VECTOR4_TAIL);
 7761 
 7762   if (UseAVX >= 2) {
 7763     Label VECTOR16_TAIL, VECTOR32_LOOP;
 7764 
 7765     cmpq(length, 16);
 7766     jcc(Assembler::equal, VECTOR16_LOOP);
 7767     jcc(Assembler::less, VECTOR8_LOOP);
 7768 
 7769     cmpq(length, 32);
 7770     jccb(Assembler::less, VECTOR16_TAIL);
 7771 
 7772     subq(length, 32);
 7773     bind(VECTOR32_LOOP);
 7774     vmovdqu(rymm0, Address(obja, result));
 7775     vmovdqu(rymm1, Address(objb, result));
 7776     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
 7777     vptest(rymm2, rymm2);
 7778     jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
 7779     addq(result, 32);
 7780     subq(length, 32);
 7781     jcc(Assembler::greaterEqual, VECTOR32_LOOP);
 7782     addq(length, 32);
 7783     jcc(Assembler::equal, SAME_TILL_END);
 7784     //falling through if less than 32 bytes left //close the branch here.
 7785 
 7786     bind(VECTOR16_TAIL);
 7787     cmpq(length, 16);
 7788     jccb(Assembler::less, VECTOR8_TAIL);
 7789     bind(VECTOR16_LOOP);
 7790     movdqu(rymm0, Address(obja, result));
 7791     movdqu(rymm1, Address(objb, result));
 7792     vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
 7793     ptest(rymm2, rymm2);
 7794     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7795     addq(result, 16);
 7796     subq(length, 16);
 7797     jcc(Assembler::equal, SAME_TILL_END);
 7798     //falling through if less than 16 bytes left
 7799   } else {//regular intrinsics
 7800 
 7801     cmpq(length, 16);
 7802     jccb(Assembler::less, VECTOR8_TAIL);
 7803 
 7804     subq(length, 16);
 7805     bind(VECTOR16_LOOP);
 7806     movdqu(rymm0, Address(obja, result));
 7807     movdqu(rymm1, Address(objb, result));
 7808     pxor(rymm0, rymm1);
 7809     ptest(rymm0, rymm0);
 7810     jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
 7811     addq(result, 16);
 7812     subq(length, 16);
 7813     jccb(Assembler::greaterEqual, VECTOR16_LOOP);
 7814     addq(length, 16);
 7815     jcc(Assembler::equal, SAME_TILL_END);
 7816     //falling through if less than 16 bytes left
 7817   }
 7818 
 7819   bind(VECTOR8_TAIL);
 7820   cmpq(length, 8);
 7821   jccb(Assembler::less, VECTOR4_TAIL);
 7822   bind(VECTOR8_LOOP);
 7823   movq(tmp1, Address(obja, result));
 7824   movq(tmp2, Address(objb, result));
 7825   xorq(tmp1, tmp2);
 7826   testq(tmp1, tmp1);
 7827   jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
 7828   addq(result, 8);
 7829   subq(length, 8);
 7830   jcc(Assembler::equal, SAME_TILL_END);
 7831   //falling through if less than 8 bytes left
 7832 
 7833   bind(VECTOR4_TAIL);
 7834   cmpq(length, 4);
 7835   jccb(Assembler::less, BYTES_TAIL);
 7836   bind(VECTOR4_LOOP);
 7837   movl(tmp1, Address(obja, result));
 7838   xorl(tmp1, Address(objb, result));
 7839   testl(tmp1, tmp1);
 7840   jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
 7841   addq(result, 4);
 7842   subq(length, 4);
 7843   jcc(Assembler::equal, SAME_TILL_END);
 7844   //falling through if less than 4 bytes left
 7845 
 7846   bind(BYTES_TAIL);
 7847   bind(BYTES_LOOP);
 7848   load_unsigned_byte(tmp1, Address(obja, result));
 7849   load_unsigned_byte(tmp2, Address(objb, result));
 7850   xorl(tmp1, tmp2);
 7851   testl(tmp1, tmp1);
 7852   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7853   decq(length);
 7854   jcc(Assembler::zero, SAME_TILL_END);
 7855   incq(result);
 7856   load_unsigned_byte(tmp1, Address(obja, result));
 7857   load_unsigned_byte(tmp2, Address(objb, result));
 7858   xorl(tmp1, tmp2);
 7859   testl(tmp1, tmp1);
 7860   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7861   decq(length);
 7862   jcc(Assembler::zero, SAME_TILL_END);
 7863   incq(result);
 7864   load_unsigned_byte(tmp1, Address(obja, result));
 7865   load_unsigned_byte(tmp2, Address(objb, result));
 7866   xorl(tmp1, tmp2);
 7867   testl(tmp1, tmp1);
 7868   jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
 7869   jmp(SAME_TILL_END);
 7870 
 7871   if (UseAVX >= 2) {
 7872     bind(VECTOR32_NOT_EQUAL);
 7873     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
 7874     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
 7875     vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
 7876     vpmovmskb(tmp1, rymm0);
 7877     bsfq(tmp1, tmp1);
 7878     addq(result, tmp1);
 7879     shrq(result);
 7880     jmp(DONE);
 7881   }
 7882 
 7883   bind(VECTOR16_NOT_EQUAL);
 7884   if (UseAVX >= 2) {
 7885     vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
 7886     vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
 7887     pxor(rymm0, rymm2);
 7888   } else {
 7889     pcmpeqb(rymm2, rymm2);
 7890     pxor(rymm0, rymm1);
 7891     pcmpeqb(rymm0, rymm1);
 7892     pxor(rymm0, rymm2);
 7893   }
 7894   pmovmskb(tmp1, rymm0);
 7895   bsfq(tmp1, tmp1);
 7896   addq(result, tmp1);
 7897   shrq(result);
 7898   jmpb(DONE);
 7899 
 7900   bind(VECTOR8_NOT_EQUAL);
 7901   bind(VECTOR4_NOT_EQUAL);
 7902   bsfq(tmp1, tmp1);
 7903   shrq(tmp1, 3);
 7904   addq(result, tmp1);
 7905   bind(BYTES_NOT_EQUAL);
 7906   shrq(result);
 7907   jmpb(DONE);
 7908 
 7909   bind(SAME_TILL_END);
 7910   mov64(result, -1);
 7911 
 7912   bind(DONE);
 7913 }
 7914 
 7915 //Helper functions for square_to_len()
 7916 
 7917 /**
 7918  * Store the squares of x[], right shifted one bit (divided by 2) into z[]
 7919  * Preserves x and z and modifies rest of the registers.
 7920  */
 7921 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 7922   // Perform square and right shift by 1
 7923   // Handle odd xlen case first, then for even xlen do the following
 7924   // jlong carry = 0;
 7925   // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
 7926   //     huge_128 product = x[j:j+1] * x[j:j+1];
 7927   //     z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
 7928   //     z[i+2:i+3] = (jlong)(product >>> 1);
 7929   //     carry = (jlong)product;
 7930   // }
 7931 
 7932   xorq(tmp5, tmp5);     // carry
 7933   xorq(rdxReg, rdxReg);
 7934   xorl(tmp1, tmp1);     // index for x
 7935   xorl(tmp4, tmp4);     // index for z
 7936 
 7937   Label L_first_loop, L_first_loop_exit;
 7938 
 7939   testl(xlen, 1);
 7940   jccb(Assembler::zero, L_first_loop); //jump if xlen is even
 7941 
 7942   // Square and right shift by 1 the odd element using 32 bit multiply
 7943   movl(raxReg, Address(x, tmp1, Address::times_4, 0));
 7944   imulq(raxReg, raxReg);
 7945   shrq(raxReg, 1);
 7946   adcq(tmp5, 0);
 7947   movq(Address(z, tmp4, Address::times_4, 0), raxReg);
 7948   incrementl(tmp1);
 7949   addl(tmp4, 2);
 7950 
 7951   // Square and  right shift by 1 the rest using 64 bit multiply
 7952   bind(L_first_loop);
 7953   cmpptr(tmp1, xlen);
 7954   jccb(Assembler::equal, L_first_loop_exit);
 7955 
 7956   // Square
 7957   movq(raxReg, Address(x, tmp1, Address::times_4,  0));
 7958   rorq(raxReg, 32);    // convert big-endian to little-endian
 7959   mulq(raxReg);        // 64-bit multiply rax * rax -> rdx:rax
 7960 
 7961   // Right shift by 1 and save carry
 7962   shrq(tmp5, 1);       // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
 7963   rcrq(rdxReg, 1);
 7964   rcrq(raxReg, 1);
 7965   adcq(tmp5, 0);
 7966 
 7967   // Store result in z
 7968   movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
 7969   movq(Address(z, tmp4, Address::times_4, 8), raxReg);
 7970 
 7971   // Update indices for x and z
 7972   addl(tmp1, 2);
 7973   addl(tmp4, 4);
 7974   jmp(L_first_loop);
 7975 
 7976   bind(L_first_loop_exit);
 7977 }
 7978 
 7979 
 7980 /**
 7981  * Perform the following multiply add operation using BMI2 instructions
 7982  * carry:sum = sum + op1*op2 + carry
 7983  * op2 should be in rdx
 7984  * op2 is preserved, all other registers are modified
 7985  */
 7986 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
 7987   // assert op2 is rdx
 7988   mulxq(tmp2, op1, op1);  //  op1 * op2 -> tmp2:op1
 7989   addq(sum, carry);
 7990   adcq(tmp2, 0);
 7991   addq(sum, op1);
 7992   adcq(tmp2, 0);
 7993   movq(carry, tmp2);
 7994 }
 7995 
 7996 /**
 7997  * Perform the following multiply add operation:
 7998  * carry:sum = sum + op1*op2 + carry
 7999  * Preserves op1, op2 and modifies rest of registers
 8000  */
 8001 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
 8002   // rdx:rax = op1 * op2
 8003   movq(raxReg, op2);
 8004   mulq(op1);
 8005 
 8006   //  rdx:rax = sum + carry + rdx:rax
 8007   addq(sum, carry);
 8008   adcq(rdxReg, 0);
 8009   addq(sum, raxReg);
 8010   adcq(rdxReg, 0);
 8011 
 8012   // carry:sum = rdx:sum
 8013   movq(carry, rdxReg);
 8014 }
 8015 
 8016 /**
 8017  * Add 64 bit long carry into z[] with carry propagation.
 8018  * Preserves z and carry register values and modifies rest of registers.
 8019  *
 8020  */
 8021 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
 8022   Label L_fourth_loop, L_fourth_loop_exit;
 8023 
 8024   movl(tmp1, 1);
 8025   subl(zlen, 2);
 8026   addq(Address(z, zlen, Address::times_4, 0), carry);
 8027 
 8028   bind(L_fourth_loop);
 8029   jccb(Assembler::carryClear, L_fourth_loop_exit);
 8030   subl(zlen, 2);
 8031   jccb(Assembler::negative, L_fourth_loop_exit);
 8032   addq(Address(z, zlen, Address::times_4, 0), tmp1);
 8033   jmp(L_fourth_loop);
 8034   bind(L_fourth_loop_exit);
 8035 }
 8036 
 8037 /**
 8038  * Shift z[] left by 1 bit.
 8039  * Preserves x, len, z and zlen registers and modifies rest of the registers.
 8040  *
 8041  */
 8042 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
 8043 
 8044   Label L_fifth_loop, L_fifth_loop_exit;
 8045 
 8046   // Fifth loop
 8047   // Perform primitiveLeftShift(z, zlen, 1)
 8048 
 8049   const Register prev_carry = tmp1;
 8050   const Register new_carry = tmp4;
 8051   const Register value = tmp2;
 8052   const Register zidx = tmp3;
 8053 
 8054   // int zidx, carry;
 8055   // long value;
 8056   // carry = 0;
 8057   // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
 8058   //    (carry:value)  = (z[i] << 1) | carry ;
 8059   //    z[i] = value;
 8060   // }
 8061 
 8062   movl(zidx, zlen);
 8063   xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
 8064 
 8065   bind(L_fifth_loop);
 8066   decl(zidx);  // Use decl to preserve carry flag
 8067   decl(zidx);
 8068   jccb(Assembler::negative, L_fifth_loop_exit);
 8069 
 8070   if (UseBMI2Instructions) {
 8071      movq(value, Address(z, zidx, Address::times_4, 0));
 8072      rclq(value, 1);
 8073      rorxq(value, value, 32);
 8074      movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 8075   }
 8076   else {
 8077     // clear new_carry
 8078     xorl(new_carry, new_carry);
 8079 
 8080     // Shift z[i] by 1, or in previous carry and save new carry
 8081     movq(value, Address(z, zidx, Address::times_4, 0));
 8082     shlq(value, 1);
 8083     adcl(new_carry, 0);
 8084 
 8085     orq(value, prev_carry);
 8086     rorq(value, 0x20);
 8087     movq(Address(z, zidx, Address::times_4,  0), value);  // Store back in big endian form
 8088 
 8089     // Set previous carry = new carry
 8090     movl(prev_carry, new_carry);
 8091   }
 8092   jmp(L_fifth_loop);
 8093 
 8094   bind(L_fifth_loop_exit);
 8095 }
 8096 
 8097 
 8098 /**
 8099  * Code for BigInteger::squareToLen() intrinsic
 8100  *
 8101  * rdi: x
 8102  * rsi: len
 8103  * r8:  z
 8104  * rcx: zlen
 8105  * r12: tmp1
 8106  * r13: tmp2
 8107  * r14: tmp3
 8108  * r15: tmp4
 8109  * rbx: tmp5
 8110  *
 8111  */
 8112 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) {
 8113 
 8114   Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
 8115   push(tmp1);
 8116   push(tmp2);
 8117   push(tmp3);
 8118   push(tmp4);
 8119   push(tmp5);
 8120 
 8121   // First loop
 8122   // Store the squares, right shifted one bit (i.e., divided by 2).
 8123   square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8124 
 8125   // Add in off-diagonal sums.
 8126   //
 8127   // Second, third (nested) and fourth loops.
 8128   // zlen +=2;
 8129   // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
 8130   //    carry = 0;
 8131   //    long op2 = x[xidx:xidx+1];
 8132   //    for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
 8133   //       k -= 2;
 8134   //       long op1 = x[j:j+1];
 8135   //       long sum = z[k:k+1];
 8136   //       carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
 8137   //       z[k:k+1] = sum;
 8138   //    }
 8139   //    add_one_64(z, k, carry, tmp_regs);
 8140   // }
 8141 
 8142   const Register carry = tmp5;
 8143   const Register sum = tmp3;
 8144   const Register op1 = tmp4;
 8145   Register op2 = tmp2;
 8146 
 8147   push(zlen);
 8148   push(len);
 8149   addl(zlen,2);
 8150   bind(L_second_loop);
 8151   xorq(carry, carry);
 8152   subl(zlen, 4);
 8153   subl(len, 2);
 8154   push(zlen);
 8155   push(len);
 8156   cmpl(len, 0);
 8157   jccb(Assembler::lessEqual, L_second_loop_exit);
 8158 
 8159   // Multiply an array by one 64 bit long.
 8160   if (UseBMI2Instructions) {
 8161     op2 = rdxReg;
 8162     movq(op2, Address(x, len, Address::times_4,  0));
 8163     rorxq(op2, op2, 32);
 8164   }
 8165   else {
 8166     movq(op2, Address(x, len, Address::times_4,  0));
 8167     rorq(op2, 32);
 8168   }
 8169 
 8170   bind(L_third_loop);
 8171   decrementl(len);
 8172   jccb(Assembler::negative, L_third_loop_exit);
 8173   decrementl(len);
 8174   jccb(Assembler::negative, L_last_x);
 8175 
 8176   movq(op1, Address(x, len, Address::times_4,  0));
 8177   rorq(op1, 32);
 8178 
 8179   bind(L_multiply);
 8180   subl(zlen, 2);
 8181   movq(sum, Address(z, zlen, Address::times_4,  0));
 8182 
 8183   // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
 8184   if (UseBMI2Instructions) {
 8185     multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
 8186   }
 8187   else {
 8188     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8189   }
 8190 
 8191   movq(Address(z, zlen, Address::times_4, 0), sum);
 8192 
 8193   jmp(L_third_loop);
 8194   bind(L_third_loop_exit);
 8195 
 8196   // Fourth loop
 8197   // Add 64 bit long carry into z with carry propagation.
 8198   // Uses offsetted zlen.
 8199   add_one_64(z, zlen, carry, tmp1);
 8200 
 8201   pop(len);
 8202   pop(zlen);
 8203   jmp(L_second_loop);
 8204 
 8205   // Next infrequent code is moved outside loops.
 8206   bind(L_last_x);
 8207   movl(op1, Address(x, 0));
 8208   jmp(L_multiply);
 8209 
 8210   bind(L_second_loop_exit);
 8211   pop(len);
 8212   pop(zlen);
 8213   pop(len);
 8214   pop(zlen);
 8215 
 8216   // Fifth loop
 8217   // Shift z left 1 bit.
 8218   lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
 8219 
 8220   // z[zlen-1] |= x[len-1] & 1;
 8221   movl(tmp3, Address(x, len, Address::times_4, -4));
 8222   andl(tmp3, 1);
 8223   orl(Address(z, zlen, Address::times_4,  -4), tmp3);
 8224 
 8225   pop(tmp5);
 8226   pop(tmp4);
 8227   pop(tmp3);
 8228   pop(tmp2);
 8229   pop(tmp1);
 8230 }
 8231 
 8232 /**
 8233  * Helper function for mul_add()
 8234  * Multiply the in[] by int k and add to out[] starting at offset offs using
 8235  * 128 bit by 32 bit multiply and return the carry in tmp5.
 8236  * Only quad int aligned length of in[] is operated on in this function.
 8237  * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
 8238  * This function preserves out, in and k registers.
 8239  * len and offset point to the appropriate index in "in" & "out" correspondingly
 8240  * tmp5 has the carry.
 8241  * other registers are temporary and are modified.
 8242  *
 8243  */
 8244 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
 8245   Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
 8246   Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8247 
 8248   Label L_first_loop, L_first_loop_exit;
 8249 
 8250   movl(tmp1, len);
 8251   shrl(tmp1, 2);
 8252 
 8253   bind(L_first_loop);
 8254   subl(tmp1, 1);
 8255   jccb(Assembler::negative, L_first_loop_exit);
 8256 
 8257   subl(len, 4);
 8258   subl(offset, 4);
 8259 
 8260   Register op2 = tmp2;
 8261   const Register sum = tmp3;
 8262   const Register op1 = tmp4;
 8263   const Register carry = tmp5;
 8264 
 8265   if (UseBMI2Instructions) {
 8266     op2 = rdxReg;
 8267   }
 8268 
 8269   movq(op1, Address(in, len, Address::times_4,  8));
 8270   rorq(op1, 32);
 8271   movq(sum, Address(out, offset, Address::times_4,  8));
 8272   rorq(sum, 32);
 8273   if (UseBMI2Instructions) {
 8274     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8275   }
 8276   else {
 8277     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8278   }
 8279   // Store back in big endian from little endian
 8280   rorq(sum, 0x20);
 8281   movq(Address(out, offset, Address::times_4,  8), sum);
 8282 
 8283   movq(op1, Address(in, len, Address::times_4,  0));
 8284   rorq(op1, 32);
 8285   movq(sum, Address(out, offset, Address::times_4,  0));
 8286   rorq(sum, 32);
 8287   if (UseBMI2Instructions) {
 8288     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8289   }
 8290   else {
 8291     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8292   }
 8293   // Store back in big endian from little endian
 8294   rorq(sum, 0x20);
 8295   movq(Address(out, offset, Address::times_4,  0), sum);
 8296 
 8297   jmp(L_first_loop);
 8298   bind(L_first_loop_exit);
 8299 }
 8300 
 8301 /**
 8302  * Code for BigInteger::mulAdd() intrinsic
 8303  *
 8304  * rdi: out
 8305  * rsi: in
 8306  * r11: offs (out.length - offset)
 8307  * rcx: len
 8308  * r8:  k
 8309  * r12: tmp1
 8310  * r13: tmp2
 8311  * r14: tmp3
 8312  * r15: tmp4
 8313  * rbx: tmp5
 8314  * Multiply the in[] by word k and add to out[], return the carry in rax
 8315  */
 8316 void MacroAssembler::mul_add(Register out, Register in, Register offs,
 8317    Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
 8318    Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
 8319 
 8320   Label L_carry, L_last_in, L_done;
 8321 
 8322 // carry = 0;
 8323 // for (int j=len-1; j >= 0; j--) {
 8324 //    long product = (in[j] & LONG_MASK) * kLong +
 8325 //                   (out[offs] & LONG_MASK) + carry;
 8326 //    out[offs--] = (int)product;
 8327 //    carry = product >>> 32;
 8328 // }
 8329 //
 8330   push(tmp1);
 8331   push(tmp2);
 8332   push(tmp3);
 8333   push(tmp4);
 8334   push(tmp5);
 8335 
 8336   Register op2 = tmp2;
 8337   const Register sum = tmp3;
 8338   const Register op1 = tmp4;
 8339   const Register carry =  tmp5;
 8340 
 8341   if (UseBMI2Instructions) {
 8342     op2 = rdxReg;
 8343     movl(op2, k);
 8344   }
 8345   else {
 8346     movl(op2, k);
 8347   }
 8348 
 8349   xorq(carry, carry);
 8350 
 8351   //First loop
 8352 
 8353   //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
 8354   //The carry is in tmp5
 8355   mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
 8356 
 8357   //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
 8358   decrementl(len);
 8359   jccb(Assembler::negative, L_carry);
 8360   decrementl(len);
 8361   jccb(Assembler::negative, L_last_in);
 8362 
 8363   movq(op1, Address(in, len, Address::times_4,  0));
 8364   rorq(op1, 32);
 8365 
 8366   subl(offs, 2);
 8367   movq(sum, Address(out, offs, Address::times_4,  0));
 8368   rorq(sum, 32);
 8369 
 8370   if (UseBMI2Instructions) {
 8371     multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
 8372   }
 8373   else {
 8374     multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
 8375   }
 8376 
 8377   // Store back in big endian from little endian
 8378   rorq(sum, 0x20);
 8379   movq(Address(out, offs, Address::times_4,  0), sum);
 8380 
 8381   testl(len, len);
 8382   jccb(Assembler::zero, L_carry);
 8383 
 8384   //Multiply the last in[] entry, if any
 8385   bind(L_last_in);
 8386   movl(op1, Address(in, 0));
 8387   movl(sum, Address(out, offs, Address::times_4,  -4));
 8388 
 8389   movl(raxReg, k);
 8390   mull(op1); //tmp4 * eax -> edx:eax
 8391   addl(sum, carry);
 8392   adcl(rdxReg, 0);
 8393   addl(sum, raxReg);
 8394   adcl(rdxReg, 0);
 8395   movl(carry, rdxReg);
 8396 
 8397   movl(Address(out, offs, Address::times_4,  -4), sum);
 8398 
 8399   bind(L_carry);
 8400   //return tmp5/carry as carry in rax
 8401   movl(rax, carry);
 8402 
 8403   bind(L_done);
 8404   pop(tmp5);
 8405   pop(tmp4);
 8406   pop(tmp3);
 8407   pop(tmp2);
 8408   pop(tmp1);
 8409 }
 8410 
 8411 /**
 8412  * Emits code to update CRC-32 with a byte value according to constants in table
 8413  *
 8414  * @param [in,out]crc   Register containing the crc.
 8415  * @param [in]val       Register containing the byte to fold into the CRC.
 8416  * @param [in]table     Register containing the table of crc constants.
 8417  *
 8418  * uint32_t crc;
 8419  * val = crc_table[(val ^ crc) & 0xFF];
 8420  * crc = val ^ (crc >> 8);
 8421  *
 8422  */
 8423 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
 8424   xorl(val, crc);
 8425   andl(val, 0xFF);
 8426   shrl(crc, 8); // unsigned shift
 8427   xorl(crc, Address(table, val, Address::times_4, 0));
 8428 }
 8429 
 8430 /**
 8431  * Fold 128-bit data chunk
 8432  */
 8433 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
 8434   if (UseAVX > 0) {
 8435     vpclmulhdq(xtmp, xK, xcrc); // [123:64]
 8436     vpclmulldq(xcrc, xK, xcrc); // [63:0]
 8437     vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
 8438     pxor(xcrc, xtmp);
 8439   } else {
 8440     movdqa(xtmp, xcrc);
 8441     pclmulhdq(xtmp, xK);   // [123:64]
 8442     pclmulldq(xcrc, xK);   // [63:0]
 8443     pxor(xcrc, xtmp);
 8444     movdqu(xtmp, Address(buf, offset));
 8445     pxor(xcrc, xtmp);
 8446   }
 8447 }
 8448 
 8449 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
 8450   if (UseAVX > 0) {
 8451     vpclmulhdq(xtmp, xK, xcrc);
 8452     vpclmulldq(xcrc, xK, xcrc);
 8453     pxor(xcrc, xbuf);
 8454     pxor(xcrc, xtmp);
 8455   } else {
 8456     movdqa(xtmp, xcrc);
 8457     pclmulhdq(xtmp, xK);
 8458     pclmulldq(xcrc, xK);
 8459     pxor(xcrc, xbuf);
 8460     pxor(xcrc, xtmp);
 8461   }
 8462 }
 8463 
 8464 /**
 8465  * 8-bit folds to compute 32-bit CRC
 8466  *
 8467  * uint64_t xcrc;
 8468  * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
 8469  */
 8470 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
 8471   movdl(tmp, xcrc);
 8472   andl(tmp, 0xFF);
 8473   movdl(xtmp, Address(table, tmp, Address::times_4, 0));
 8474   psrldq(xcrc, 1); // unsigned shift one byte
 8475   pxor(xcrc, xtmp);
 8476 }
 8477 
 8478 /**
 8479  * uint32_t crc;
 8480  * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
 8481  */
 8482 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
 8483   movl(tmp, crc);
 8484   andl(tmp, 0xFF);
 8485   shrl(crc, 8);
 8486   xorl(crc, Address(table, tmp, Address::times_4, 0));
 8487 }
 8488 
 8489 /**
 8490  * @param crc   register containing existing CRC (32-bit)
 8491  * @param buf   register pointing to input byte buffer (byte*)
 8492  * @param len   register containing number of bytes
 8493  * @param table register that will contain address of CRC table
 8494  * @param tmp   scratch register
 8495  */
 8496 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
 8497   assert_different_registers(crc, buf, len, table, tmp, rax);
 8498 
 8499   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8500   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8501 
 8502   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8503   // context for the registers used, where all instructions below are using 128-bit mode
 8504   // On EVEX without VL and BW, these instructions will all be AVX.
 8505   lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
 8506   notl(crc); // ~crc
 8507   cmpl(len, 16);
 8508   jcc(Assembler::less, L_tail);
 8509 
 8510   // Align buffer to 16 bytes
 8511   movl(tmp, buf);
 8512   andl(tmp, 0xF);
 8513   jccb(Assembler::zero, L_aligned);
 8514   subl(tmp,  16);
 8515   addl(len, tmp);
 8516 
 8517   align(4);
 8518   BIND(L_align_loop);
 8519   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8520   update_byte_crc32(crc, rax, table);
 8521   increment(buf);
 8522   incrementl(tmp);
 8523   jccb(Assembler::less, L_align_loop);
 8524 
 8525   BIND(L_aligned);
 8526   movl(tmp, len); // save
 8527   shrl(len, 4);
 8528   jcc(Assembler::zero, L_tail_restore);
 8529 
 8530   // Fold crc into first bytes of vector
 8531   movdqa(xmm1, Address(buf, 0));
 8532   movdl(rax, xmm1);
 8533   xorl(crc, rax);
 8534   if (VM_Version::supports_sse4_1()) {
 8535     pinsrd(xmm1, crc, 0);
 8536   } else {
 8537     pinsrw(xmm1, crc, 0);
 8538     shrl(crc, 16);
 8539     pinsrw(xmm1, crc, 1);
 8540   }
 8541   addptr(buf, 16);
 8542   subl(len, 4); // len > 0
 8543   jcc(Assembler::less, L_fold_tail);
 8544 
 8545   movdqa(xmm2, Address(buf,  0));
 8546   movdqa(xmm3, Address(buf, 16));
 8547   movdqa(xmm4, Address(buf, 32));
 8548   addptr(buf, 48);
 8549   subl(len, 3);
 8550   jcc(Assembler::lessEqual, L_fold_512b);
 8551 
 8552   // Fold total 512 bits of polynomial on each iteration,
 8553   // 128 bits per each of 4 parallel streams.
 8554   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
 8555 
 8556   align32();
 8557   BIND(L_fold_512b_loop);
 8558   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8559   fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
 8560   fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
 8561   fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
 8562   addptr(buf, 64);
 8563   subl(len, 4);
 8564   jcc(Assembler::greater, L_fold_512b_loop);
 8565 
 8566   // Fold 512 bits to 128 bits.
 8567   BIND(L_fold_512b);
 8568   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8569   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
 8570   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
 8571   fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
 8572 
 8573   // Fold the rest of 128 bits data chunks
 8574   BIND(L_fold_tail);
 8575   addl(len, 3);
 8576   jccb(Assembler::lessEqual, L_fold_128b);
 8577   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
 8578 
 8579   BIND(L_fold_tail_loop);
 8580   fold_128bit_crc32(xmm1, xmm0, xmm5, buf,  0);
 8581   addptr(buf, 16);
 8582   decrementl(len);
 8583   jccb(Assembler::greater, L_fold_tail_loop);
 8584 
 8585   // Fold 128 bits in xmm1 down into 32 bits in crc register.
 8586   BIND(L_fold_128b);
 8587   movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
 8588   if (UseAVX > 0) {
 8589     vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
 8590     vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
 8591     vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
 8592   } else {
 8593     movdqa(xmm2, xmm0);
 8594     pclmulqdq(xmm2, xmm1, 0x1);
 8595     movdqa(xmm3, xmm0);
 8596     pand(xmm3, xmm2);
 8597     pclmulqdq(xmm0, xmm3, 0x1);
 8598   }
 8599   psrldq(xmm1, 8);
 8600   psrldq(xmm2, 4);
 8601   pxor(xmm0, xmm1);
 8602   pxor(xmm0, xmm2);
 8603 
 8604   // 8 8-bit folds to compute 32-bit CRC.
 8605   for (int j = 0; j < 4; j++) {
 8606     fold_8bit_crc32(xmm0, table, xmm1, rax);
 8607   }
 8608   movdl(crc, xmm0); // mov 32 bits to general register
 8609   for (int j = 0; j < 4; j++) {
 8610     fold_8bit_crc32(crc, table, rax);
 8611   }
 8612 
 8613   BIND(L_tail_restore);
 8614   movl(len, tmp); // restore
 8615   BIND(L_tail);
 8616   andl(len, 0xf);
 8617   jccb(Assembler::zero, L_exit);
 8618 
 8619   // Fold the rest of bytes
 8620   align(4);
 8621   BIND(L_tail_loop);
 8622   movsbl(rax, Address(buf, 0)); // load byte with sign extension
 8623   update_byte_crc32(crc, rax, table);
 8624   increment(buf);
 8625   decrementl(len);
 8626   jccb(Assembler::greater, L_tail_loop);
 8627 
 8628   BIND(L_exit);
 8629   notl(crc); // ~c
 8630 }
 8631 
 8632 // Helper function for AVX 512 CRC32
 8633 // Fold 512-bit data chunks
 8634 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
 8635                                              Register pos, int offset) {
 8636   evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
 8637   evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
 8638   evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
 8639   evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
 8640   evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
 8641 }
 8642 
 8643 // Helper function for AVX 512 CRC32
 8644 // Compute CRC32 for < 256B buffers
 8645 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
 8646                                               Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
 8647                                               Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
 8648 
 8649   Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
 8650   Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
 8651   Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
 8652 
 8653   // check if there is enough buffer to be able to fold 16B at a time
 8654   cmpl(len, 32);
 8655   jcc(Assembler::less, L_less_than_32);
 8656 
 8657   // if there is, load the constants
 8658   movdqu(xmm10, Address(table, 1 * 16));    //rk1 and rk2 in xmm10
 8659   movdl(xmm0, crc);                        // get the initial crc value
 8660   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8661   pxor(xmm7, xmm0);
 8662 
 8663   // update the buffer pointer
 8664   addl(pos, 16);
 8665   //update the counter.subtract 32 instead of 16 to save one instruction from the loop
 8666   subl(len, 32);
 8667   jmp(L_16B_reduction_loop);
 8668 
 8669   bind(L_less_than_32);
 8670   //mov initial crc to the return value. this is necessary for zero - length buffers.
 8671   movl(rax, crc);
 8672   testl(len, len);
 8673   jcc(Assembler::equal, L_cleanup);
 8674 
 8675   movdl(xmm0, crc);                        //get the initial crc value
 8676 
 8677   cmpl(len, 16);
 8678   jcc(Assembler::equal, L_exact_16_left);
 8679   jcc(Assembler::less, L_less_than_16_left);
 8680 
 8681   movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
 8682   pxor(xmm7, xmm0);                       //xor the initial crc value
 8683   addl(pos, 16);
 8684   subl(len, 16);
 8685   movdqu(xmm10, Address(table, 1 * 16));    // rk1 and rk2 in xmm10
 8686   jmp(L_get_last_two_xmms);
 8687 
 8688   bind(L_less_than_16_left);
 8689   //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
 8690   pxor(xmm1, xmm1);
 8691   movptr(tmp1, rsp);
 8692   movdqu(Address(tmp1, 0 * 16), xmm1);
 8693 
 8694   cmpl(len, 4);
 8695   jcc(Assembler::less, L_only_less_than_4);
 8696 
 8697   //backup the counter value
 8698   movl(tmp2, len);
 8699   cmpl(len, 8);
 8700   jcc(Assembler::less, L_less_than_8_left);
 8701 
 8702   //load 8 Bytes
 8703   movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
 8704   movq(Address(tmp1, 0 * 16), rax);
 8705   addptr(tmp1, 8);
 8706   subl(len, 8);
 8707   addl(pos, 8);
 8708 
 8709   bind(L_less_than_8_left);
 8710   cmpl(len, 4);
 8711   jcc(Assembler::less, L_less_than_4_left);
 8712 
 8713   //load 4 Bytes
 8714   movl(rax, Address(buf, pos, Address::times_1, 0));
 8715   movl(Address(tmp1, 0 * 16), rax);
 8716   addptr(tmp1, 4);
 8717   subl(len, 4);
 8718   addl(pos, 4);
 8719 
 8720   bind(L_less_than_4_left);
 8721   cmpl(len, 2);
 8722   jcc(Assembler::less, L_less_than_2_left);
 8723 
 8724   // load 2 Bytes
 8725   movw(rax, Address(buf, pos, Address::times_1, 0));
 8726   movl(Address(tmp1, 0 * 16), rax);
 8727   addptr(tmp1, 2);
 8728   subl(len, 2);
 8729   addl(pos, 2);
 8730 
 8731   bind(L_less_than_2_left);
 8732   cmpl(len, 1);
 8733   jcc(Assembler::less, L_zero_left);
 8734 
 8735   // load 1 Byte
 8736   movb(rax, Address(buf, pos, Address::times_1, 0));
 8737   movb(Address(tmp1, 0 * 16), rax);
 8738 
 8739   bind(L_zero_left);
 8740   movdqu(xmm7, Address(rsp, 0));
 8741   pxor(xmm7, xmm0);                       //xor the initial crc value
 8742 
 8743   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8744   movdqu(xmm0, Address(rax, tmp2));
 8745   pshufb(xmm7, xmm0);
 8746   jmp(L_128_done);
 8747 
 8748   bind(L_exact_16_left);
 8749   movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
 8750   pxor(xmm7, xmm0);                       //xor the initial crc value
 8751   jmp(L_128_done);
 8752 
 8753   bind(L_only_less_than_4);
 8754   cmpl(len, 3);
 8755   jcc(Assembler::less, L_only_less_than_3);
 8756 
 8757   // load 3 Bytes
 8758   movb(rax, Address(buf, pos, Address::times_1, 0));
 8759   movb(Address(tmp1, 0), rax);
 8760 
 8761   movb(rax, Address(buf, pos, Address::times_1, 1));
 8762   movb(Address(tmp1, 1), rax);
 8763 
 8764   movb(rax, Address(buf, pos, Address::times_1, 2));
 8765   movb(Address(tmp1, 2), rax);
 8766 
 8767   movdqu(xmm7, Address(rsp, 0));
 8768   pxor(xmm7, xmm0);                     //xor the initial crc value
 8769 
 8770   pslldq(xmm7, 0x5);
 8771   jmp(L_barrett);
 8772   bind(L_only_less_than_3);
 8773   cmpl(len, 2);
 8774   jcc(Assembler::less, L_only_less_than_2);
 8775 
 8776   // load 2 Bytes
 8777   movb(rax, Address(buf, pos, Address::times_1, 0));
 8778   movb(Address(tmp1, 0), rax);
 8779 
 8780   movb(rax, Address(buf, pos, Address::times_1, 1));
 8781   movb(Address(tmp1, 1), rax);
 8782 
 8783   movdqu(xmm7, Address(rsp, 0));
 8784   pxor(xmm7, xmm0);                     //xor the initial crc value
 8785 
 8786   pslldq(xmm7, 0x6);
 8787   jmp(L_barrett);
 8788 
 8789   bind(L_only_less_than_2);
 8790   //load 1 Byte
 8791   movb(rax, Address(buf, pos, Address::times_1, 0));
 8792   movb(Address(tmp1, 0), rax);
 8793 
 8794   movdqu(xmm7, Address(rsp, 0));
 8795   pxor(xmm7, xmm0);                     //xor the initial crc value
 8796 
 8797   pslldq(xmm7, 0x7);
 8798 }
 8799 
 8800 /**
 8801 * Compute CRC32 using AVX512 instructions
 8802 * param crc   register containing existing CRC (32-bit)
 8803 * param buf   register pointing to input byte buffer (byte*)
 8804 * param len   register containing number of bytes
 8805 * param table address of crc or crc32c table
 8806 * param tmp1  scratch register
 8807 * param tmp2  scratch register
 8808 * return rax  result register
 8809 *
 8810 * This routine is identical for crc32c with the exception of the precomputed constant
 8811 * table which will be passed as the table argument.  The calculation steps are
 8812 * the same for both variants.
 8813 */
 8814 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
 8815   assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
 8816 
 8817   Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
 8818   Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
 8819   Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
 8820   Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
 8821   Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
 8822 
 8823   const Register pos = r12;
 8824   push(r12);
 8825   subptr(rsp, 16 * 2 + 8);
 8826 
 8827   // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
 8828   // context for the registers used, where all instructions below are using 128-bit mode
 8829   // On EVEX without VL and BW, these instructions will all be AVX.
 8830   movl(pos, 0);
 8831 
 8832   // check if smaller than 256B
 8833   cmpl(len, 256);
 8834   jcc(Assembler::less, L_less_than_256);
 8835 
 8836   // load the initial crc value
 8837   movdl(xmm10, crc);
 8838 
 8839   // receive the initial 64B data, xor the initial crc value
 8840   evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
 8841   evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
 8842   evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
 8843   evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
 8844 
 8845   subl(len, 256);
 8846   cmpl(len, 256);
 8847   jcc(Assembler::less, L_fold_128_B_loop);
 8848 
 8849   evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
 8850   evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
 8851   evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
 8852   subl(len, 256);
 8853 
 8854   bind(L_fold_256_B_loop);
 8855   addl(pos, 256);
 8856   fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
 8857   fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
 8858   fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
 8859   fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
 8860 
 8861   subl(len, 256);
 8862   jcc(Assembler::greaterEqual, L_fold_256_B_loop);
 8863 
 8864   // Fold 256 into 128
 8865   addl(pos, 256);
 8866   evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
 8867   evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
 8868   vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
 8869 
 8870   evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
 8871   evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
 8872   vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
 8873 
 8874   evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
 8875   evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
 8876 
 8877   addl(len, 128);
 8878   jmp(L_fold_128_B_register);
 8879 
 8880   // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
 8881   // loop will fold 128B at a time until we have 128 + y Bytes of buffer
 8882 
 8883   // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
 8884   bind(L_fold_128_B_loop);
 8885   addl(pos, 128);
 8886   fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
 8887   fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
 8888 
 8889   subl(len, 128);
 8890   jcc(Assembler::greaterEqual, L_fold_128_B_loop);
 8891 
 8892   addl(pos, 128);
 8893 
 8894   // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
 8895   // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
 8896   bind(L_fold_128_B_register);
 8897   evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
 8898   evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
 8899   evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
 8900   evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
 8901   // save last that has no multiplicand
 8902   vextracti64x2(xmm7, xmm4, 3);
 8903 
 8904   evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
 8905   evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
 8906   // Needed later in reduction loop
 8907   movdqu(xmm10, Address(table, 1 * 16));
 8908   vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
 8909   vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
 8910 
 8911   // Swap 1,0,3,2 - 01 00 11 10
 8912   evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
 8913   evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
 8914   vextracti128(xmm5, xmm8, 1);
 8915   evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
 8916 
 8917   // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
 8918   // instead of a cmp instruction, we use the negative flag with the jl instruction
 8919   addl(len, 128 - 16);
 8920   jcc(Assembler::less, L_final_reduction_for_128);
 8921 
 8922   bind(L_16B_reduction_loop);
 8923   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8924   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8925   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8926   movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
 8927   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8928   addl(pos, 16);
 8929   subl(len, 16);
 8930   jcc(Assembler::greaterEqual, L_16B_reduction_loop);
 8931 
 8932   bind(L_final_reduction_for_128);
 8933   addl(len, 16);
 8934   jcc(Assembler::equal, L_128_done);
 8935 
 8936   bind(L_get_last_two_xmms);
 8937   movdqu(xmm2, xmm7);
 8938   addl(pos, len);
 8939   movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
 8940   subl(pos, len);
 8941 
 8942   // get rid of the extra data that was loaded before
 8943   // load the shift constant
 8944   lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
 8945   movdqu(xmm0, Address(rax, len));
 8946   addl(rax, len);
 8947 
 8948   vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8949   //Change mask to 512
 8950   vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
 8951   vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
 8952 
 8953   blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
 8954   vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
 8955   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8956   vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
 8957   vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
 8958 
 8959   bind(L_128_done);
 8960   // compute crc of a 128-bit value
 8961   movdqu(xmm10, Address(table, 3 * 16));
 8962   movdqu(xmm0, xmm7);
 8963 
 8964   // 64b fold
 8965   vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
 8966   vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
 8967   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8968 
 8969   // 32b fold
 8970   movdqu(xmm0, xmm7);
 8971   vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
 8972   vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
 8973   vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
 8974   jmp(L_barrett);
 8975 
 8976   bind(L_less_than_256);
 8977   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);
 8978 
 8979   //barrett reduction
 8980   bind(L_barrett);
 8981   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
 8982   movdqu(xmm1, xmm7);
 8983   movdqu(xmm2, xmm7);
 8984   movdqu(xmm10, Address(table, 4 * 16));
 8985 
 8986   pclmulqdq(xmm7, xmm10, 0x0);
 8987   pxor(xmm7, xmm2);
 8988   vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
 8989   movdqu(xmm2, xmm7);
 8990   pclmulqdq(xmm7, xmm10, 0x10);
 8991   pxor(xmm7, xmm2);
 8992   pxor(xmm7, xmm1);
 8993   pextrd(crc, xmm7, 2);
 8994 
 8995   bind(L_cleanup);
 8996   addptr(rsp, 16 * 2 + 8);
 8997   pop(r12);
 8998 }
 8999 
 9000 // S. Gueron / Information Processing Letters 112 (2012) 184
 9001 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
 9002 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
 9003 // Output: the 64-bit carry-less product of B * CONST
 9004 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
 9005                                      Register tmp1, Register tmp2, Register tmp3) {
 9006   lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
 9007   if (n > 0) {
 9008     addq(tmp3, n * 256 * 8);
 9009   }
 9010   //    Q1 = TABLEExt[n][B & 0xFF];
 9011   movl(tmp1, in);
 9012   andl(tmp1, 0x000000FF);
 9013   shll(tmp1, 3);
 9014   addq(tmp1, tmp3);
 9015   movq(tmp1, Address(tmp1, 0));
 9016 
 9017   //    Q2 = TABLEExt[n][B >> 8 & 0xFF];
 9018   movl(tmp2, in);
 9019   shrl(tmp2, 8);
 9020   andl(tmp2, 0x000000FF);
 9021   shll(tmp2, 3);
 9022   addq(tmp2, tmp3);
 9023   movq(tmp2, Address(tmp2, 0));
 9024 
 9025   shlq(tmp2, 8);
 9026   xorq(tmp1, tmp2);
 9027 
 9028   //    Q3 = TABLEExt[n][B >> 16 & 0xFF];
 9029   movl(tmp2, in);
 9030   shrl(tmp2, 16);
 9031   andl(tmp2, 0x000000FF);
 9032   shll(tmp2, 3);
 9033   addq(tmp2, tmp3);
 9034   movq(tmp2, Address(tmp2, 0));
 9035 
 9036   shlq(tmp2, 16);
 9037   xorq(tmp1, tmp2);
 9038 
 9039   //    Q4 = TABLEExt[n][B >> 24 & 0xFF];
 9040   shrl(in, 24);
 9041   andl(in, 0x000000FF);
 9042   shll(in, 3);
 9043   addq(in, tmp3);
 9044   movq(in, Address(in, 0));
 9045 
 9046   shlq(in, 24);
 9047   xorq(in, tmp1);
 9048   //    return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
 9049 }
 9050 
 9051 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
 9052                                       Register in_out,
 9053                                       uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
 9054                                       XMMRegister w_xtmp2,
 9055                                       Register tmp1,
 9056                                       Register n_tmp2, Register n_tmp3) {
 9057   if (is_pclmulqdq_supported) {
 9058     movdl(w_xtmp1, in_out); // modified blindly
 9059 
 9060     movl(tmp1, const_or_pre_comp_const_index);
 9061     movdl(w_xtmp2, tmp1);
 9062     pclmulqdq(w_xtmp1, w_xtmp2, 0);
 9063 
 9064     movdq(in_out, w_xtmp1);
 9065   } else {
 9066     crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
 9067   }
 9068 }
 9069 
 9070 // Recombination Alternative 2: No bit-reflections
 9071 // T1 = (CRC_A * U1) << 1
 9072 // T2 = (CRC_B * U2) << 1
 9073 // C1 = T1 >> 32
 9074 // C2 = T2 >> 32
 9075 // T1 = T1 & 0xFFFFFFFF
 9076 // T2 = T2 & 0xFFFFFFFF
 9077 // T1 = CRC32(0, T1)
 9078 // T2 = CRC32(0, T2)
 9079 // C1 = C1 ^ T1
 9080 // C2 = C2 ^ T2
 9081 // CRC = C1 ^ C2 ^ CRC_C
 9082 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,
 9083                                      XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9084                                      Register tmp1, Register tmp2,
 9085                                      Register n_tmp3) {
 9086   crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 9087   crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
 9088   shlq(in_out, 1);
 9089   movl(tmp1, in_out);
 9090   shrq(in_out, 32);
 9091   xorl(tmp2, tmp2);
 9092   crc32(tmp2, tmp1, 4);
 9093   xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
 9094   shlq(in1, 1);
 9095   movl(tmp1, in1);
 9096   shrq(in1, 32);
 9097   xorl(tmp2, tmp2);
 9098   crc32(tmp2, tmp1, 4);
 9099   xorl(in1, tmp2);
 9100   xorl(in_out, in1);
 9101   xorl(in_out, in2);
 9102 }
 9103 
 9104 // Set N to predefined value
 9105 // Subtract from a length of a buffer
 9106 // execute in a loop:
 9107 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
 9108 // for i = 1 to N do
 9109 //  CRC_A = CRC32(CRC_A, A[i])
 9110 //  CRC_B = CRC32(CRC_B, B[i])
 9111 //  CRC_C = CRC32(CRC_C, C[i])
 9112 // end for
 9113 // Recombine
 9114 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,
 9115                                        Register in_out1, Register in_out2, Register in_out3,
 9116                                        Register tmp1, Register tmp2, Register tmp3,
 9117                                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9118                                        Register tmp4, Register tmp5,
 9119                                        Register n_tmp6) {
 9120   Label L_processPartitions;
 9121   Label L_processPartition;
 9122   Label L_exit;
 9123 
 9124   bind(L_processPartitions);
 9125   cmpl(in_out1, 3 * size);
 9126   jcc(Assembler::less, L_exit);
 9127     xorl(tmp1, tmp1);
 9128     xorl(tmp2, tmp2);
 9129     movq(tmp3, in_out2);
 9130     addq(tmp3, size);
 9131 
 9132     bind(L_processPartition);
 9133       crc32(in_out3, Address(in_out2, 0), 8);
 9134       crc32(tmp1, Address(in_out2, size), 8);
 9135       crc32(tmp2, Address(in_out2, size * 2), 8);
 9136       addq(in_out2, 8);
 9137       cmpq(in_out2, tmp3);
 9138       jcc(Assembler::less, L_processPartition);
 9139     crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
 9140             w_xtmp1, w_xtmp2, w_xtmp3,
 9141             tmp4, tmp5,
 9142             n_tmp6);
 9143     addq(in_out2, 2 * size);
 9144     subl(in_out1, 3 * size);
 9145     jmp(L_processPartitions);
 9146 
 9147   bind(L_exit);
 9148 }
 9149 
 9150 // Algorithm 2: Pipelined usage of the CRC32 instruction.
 9151 // Input: A buffer I of L bytes.
 9152 // Output: the CRC32C value of the buffer.
 9153 // Notations:
 9154 // Write L = 24N + r, with N = floor (L/24).
 9155 // r = L mod 24 (0 <= r < 24).
 9156 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
 9157 // N quadwords, and R consists of r bytes.
 9158 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
 9159 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
 9160 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
 9161 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
 9162 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
 9163                                           Register tmp1, Register tmp2, Register tmp3,
 9164                                           Register tmp4, Register tmp5, Register tmp6,
 9165                                           XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
 9166                                           bool is_pclmulqdq_supported) {
 9167   uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
 9168   Label L_wordByWord;
 9169   Label L_byteByByteProlog;
 9170   Label L_byteByByte;
 9171   Label L_exit;
 9172 
 9173   if (is_pclmulqdq_supported ) {
 9174     const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
 9175     const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
 9176 
 9177     const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
 9178     const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
 9179 
 9180     const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
 9181     const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
 9182     assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
 9183   } else {
 9184     const_or_pre_comp_const_index[0] = 1;
 9185     const_or_pre_comp_const_index[1] = 0;
 9186 
 9187     const_or_pre_comp_const_index[2] = 3;
 9188     const_or_pre_comp_const_index[3] = 2;
 9189 
 9190     const_or_pre_comp_const_index[4] = 5;
 9191     const_or_pre_comp_const_index[5] = 4;
 9192    }
 9193   crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
 9194                     in2, in1, in_out,
 9195                     tmp1, tmp2, tmp3,
 9196                     w_xtmp1, w_xtmp2, w_xtmp3,
 9197                     tmp4, tmp5,
 9198                     tmp6);
 9199   crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
 9200                     in2, in1, in_out,
 9201                     tmp1, tmp2, tmp3,
 9202                     w_xtmp1, w_xtmp2, w_xtmp3,
 9203                     tmp4, tmp5,
 9204                     tmp6);
 9205   crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
 9206                     in2, in1, in_out,
 9207                     tmp1, tmp2, tmp3,
 9208                     w_xtmp1, w_xtmp2, w_xtmp3,
 9209                     tmp4, tmp5,
 9210                     tmp6);
 9211   movl(tmp1, in2);
 9212   andl(tmp1, 0x00000007);
 9213   negl(tmp1);
 9214   addl(tmp1, in2);
 9215   addq(tmp1, in1);
 9216 
 9217   cmpq(in1, tmp1);
 9218   jccb(Assembler::greaterEqual, L_byteByByteProlog);
 9219   align(16);
 9220   BIND(L_wordByWord);
 9221     crc32(in_out, Address(in1, 0), 8);
 9222     addq(in1, 8);
 9223     cmpq(in1, tmp1);
 9224     jcc(Assembler::less, L_wordByWord);
 9225 
 9226   BIND(L_byteByByteProlog);
 9227   andl(in2, 0x00000007);
 9228   movl(tmp2, 1);
 9229 
 9230   cmpl(tmp2, in2);
 9231   jccb(Assembler::greater, L_exit);
 9232   BIND(L_byteByByte);
 9233     crc32(in_out, Address(in1, 0), 1);
 9234     incq(in1);
 9235     incl(tmp2);
 9236     cmpl(tmp2, in2);
 9237     jcc(Assembler::lessEqual, L_byteByByte);
 9238 
 9239   BIND(L_exit);
 9240 }
 9241 #undef BIND
 9242 #undef BLOCK_COMMENT
 9243 
 9244 // Compress char[] array to byte[].
 9245 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
 9246 // Return the array length if every element in array can be encoded,
 9247 // otherwise, the index of first non-latin1 (> 0xff) character.
 9248 //   @IntrinsicCandidate
 9249 //   public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
 9250 //     for (int i = 0; i < len; i++) {
 9251 //       char c = src[srcOff];
 9252 //       if (c > 0xff) {
 9253 //           return i;  // return index of non-latin1 char
 9254 //       }
 9255 //       dst[dstOff] = (byte)c;
 9256 //       srcOff++;
 9257 //       dstOff++;
 9258 //     }
 9259 //     return len;
 9260 //   }
 9261 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
 9262   XMMRegister tmp1Reg, XMMRegister tmp2Reg,
 9263   XMMRegister tmp3Reg, XMMRegister tmp4Reg,
 9264   Register tmp5, Register result, KRegister mask1, KRegister mask2) {
 9265   Label copy_chars_loop, done, reset_sp, copy_tail;
 9266 
 9267   // rsi: src
 9268   // rdi: dst
 9269   // rdx: len
 9270   // rcx: tmp5
 9271   // rax: result
 9272 
 9273   // rsi holds start addr of source char[] to be compressed
 9274   // rdi holds start addr of destination byte[]
 9275   // rdx holds length
 9276 
 9277   assert(len != result, "");
 9278 
 9279   // save length for return
 9280   movl(result, len);
 9281 
 9282   if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
 9283     VM_Version::supports_avx512vlbw() &&
 9284     VM_Version::supports_bmi2()) {
 9285 
 9286     Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
 9287 
 9288     // alignment
 9289     Label post_alignment;
 9290 
 9291     // if length of the string is less than 32, handle it the old fashioned way
 9292     testl(len, -32);
 9293     jcc(Assembler::zero, below_threshold);
 9294 
 9295     // First check whether a character is compressible ( <= 0xFF).
 9296     // Create mask to test for Unicode chars inside zmm vector
 9297     movl(tmp5, 0x00FF);
 9298     evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
 9299 
 9300     testl(len, -64);
 9301     jccb(Assembler::zero, post_alignment);
 9302 
 9303     movl(tmp5, dst);
 9304     andl(tmp5, (32 - 1));
 9305     negl(tmp5);
 9306     andl(tmp5, (32 - 1));
 9307 
 9308     // bail out when there is nothing to be done
 9309     testl(tmp5, 0xFFFFFFFF);
 9310     jccb(Assembler::zero, post_alignment);
 9311 
 9312     // ~(~0 << len), where len is the # of remaining elements to process
 9313     movl(len, 0xFFFFFFFF);
 9314     shlxl(len, len, tmp5);
 9315     notl(len);
 9316     kmovdl(mask2, len);
 9317     movl(len, result);
 9318 
 9319     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9320     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9321     ktestd(mask1, mask2);
 9322     jcc(Assembler::carryClear, copy_tail);
 9323 
 9324     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9325 
 9326     addptr(src, tmp5);
 9327     addptr(src, tmp5);
 9328     addptr(dst, tmp5);
 9329     subl(len, tmp5);
 9330 
 9331     bind(post_alignment);
 9332     // end of alignment
 9333 
 9334     movl(tmp5, len);
 9335     andl(tmp5, (32 - 1));    // tail count (in chars)
 9336     andl(len, ~(32 - 1));    // vector count (in chars)
 9337     jccb(Assembler::zero, copy_loop_tail);
 9338 
 9339     lea(src, Address(src, len, Address::times_2));
 9340     lea(dst, Address(dst, len, Address::times_1));
 9341     negptr(len);
 9342 
 9343     bind(copy_32_loop);
 9344     evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
 9345     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
 9346     kortestdl(mask1, mask1);
 9347     jccb(Assembler::carryClear, reset_for_copy_tail);
 9348 
 9349     // All elements in current processed chunk are valid candidates for
 9350     // compression. Write a truncated byte elements to the memory.
 9351     evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
 9352     addptr(len, 32);
 9353     jccb(Assembler::notZero, copy_32_loop);
 9354 
 9355     bind(copy_loop_tail);
 9356     // bail out when there is nothing to be done
 9357     testl(tmp5, 0xFFFFFFFF);
 9358     jcc(Assembler::zero, done);
 9359 
 9360     movl(len, tmp5);
 9361 
 9362     // ~(~0 << len), where len is the # of remaining elements to process
 9363     movl(tmp5, 0xFFFFFFFF);
 9364     shlxl(tmp5, tmp5, len);
 9365     notl(tmp5);
 9366 
 9367     kmovdl(mask2, tmp5);
 9368 
 9369     evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
 9370     evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
 9371     ktestd(mask1, mask2);
 9372     jcc(Assembler::carryClear, copy_tail);
 9373 
 9374     evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
 9375     jmp(done);
 9376 
 9377     bind(reset_for_copy_tail);
 9378     lea(src, Address(src, tmp5, Address::times_2));
 9379     lea(dst, Address(dst, tmp5, Address::times_1));
 9380     subptr(len, tmp5);
 9381     jmp(copy_chars_loop);
 9382 
 9383     bind(below_threshold);
 9384   }
 9385 
 9386   if (UseSSE42Intrinsics) {
 9387     Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
 9388 
 9389     // vectored compression
 9390     testl(len, 0xfffffff8);
 9391     jcc(Assembler::zero, copy_tail);
 9392 
 9393     movl(tmp5, 0xff00ff00);   // create mask to test for Unicode chars in vectors
 9394     movdl(tmp1Reg, tmp5);
 9395     pshufd(tmp1Reg, tmp1Reg, 0);   // store Unicode mask in tmp1Reg
 9396 
 9397     andl(len, 0xfffffff0);
 9398     jccb(Assembler::zero, copy_16);
 9399 
 9400     // compress 16 chars per iter
 9401     pxor(tmp4Reg, tmp4Reg);
 9402 
 9403     lea(src, Address(src, len, Address::times_2));
 9404     lea(dst, Address(dst, len, Address::times_1));
 9405     negptr(len);
 9406 
 9407     bind(copy_32_loop);
 9408     movdqu(tmp2Reg, Address(src, len, Address::times_2));     // load 1st 8 characters
 9409     por(tmp4Reg, tmp2Reg);
 9410     movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
 9411     por(tmp4Reg, tmp3Reg);
 9412     ptest(tmp4Reg, tmp1Reg);       // check for Unicode chars in next vector
 9413     jccb(Assembler::notZero, reset_for_copy_tail);
 9414     packuswb(tmp2Reg, tmp3Reg);    // only ASCII chars; compress each to 1 byte
 9415     movdqu(Address(dst, len, Address::times_1), tmp2Reg);
 9416     addptr(len, 16);
 9417     jccb(Assembler::notZero, copy_32_loop);
 9418 
 9419     // compress next vector of 8 chars (if any)
 9420     bind(copy_16);
 9421     // len = 0
 9422     testl(result, 0x00000008);     // check if there's a block of 8 chars to compress
 9423     jccb(Assembler::zero, copy_tail_sse);
 9424 
 9425     pxor(tmp3Reg, tmp3Reg);
 9426 
 9427     movdqu(tmp2Reg, Address(src, 0));
 9428     ptest(tmp2Reg, tmp1Reg);       // check for Unicode chars in vector
 9429     jccb(Assembler::notZero, reset_for_copy_tail);
 9430     packuswb(tmp2Reg, tmp3Reg);    // only LATIN1 chars; compress each to 1 byte
 9431     movq(Address(dst, 0), tmp2Reg);
 9432     addptr(src, 16);
 9433     addptr(dst, 8);
 9434     jmpb(copy_tail_sse);
 9435 
 9436     bind(reset_for_copy_tail);
 9437     movl(tmp5, result);
 9438     andl(tmp5, 0x0000000f);
 9439     lea(src, Address(src, tmp5, Address::times_2));
 9440     lea(dst, Address(dst, tmp5, Address::times_1));
 9441     subptr(len, tmp5);
 9442     jmpb(copy_chars_loop);
 9443 
 9444     bind(copy_tail_sse);
 9445     movl(len, result);
 9446     andl(len, 0x00000007);    // tail count (in chars)
 9447   }
 9448   // compress 1 char per iter
 9449   bind(copy_tail);
 9450   testl(len, len);
 9451   jccb(Assembler::zero, done);
 9452   lea(src, Address(src, len, Address::times_2));
 9453   lea(dst, Address(dst, len, Address::times_1));
 9454   negptr(len);
 9455 
 9456   bind(copy_chars_loop);
 9457   load_unsigned_short(tmp5, Address(src, len, Address::times_2));
 9458   testl(tmp5, 0xff00);      // check if Unicode char
 9459   jccb(Assembler::notZero, reset_sp);
 9460   movb(Address(dst, len, Address::times_1), tmp5);  // ASCII char; compress to 1 byte
 9461   increment(len);
 9462   jccb(Assembler::notZero, copy_chars_loop);
 9463 
 9464   // add len then return (len will be zero if compress succeeded, otherwise negative)
 9465   bind(reset_sp);
 9466   addl(result, len);
 9467 
 9468   bind(done);
 9469 }
 9470 
 9471 // Inflate byte[] array to char[].
 9472 //   ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
 9473 //   @IntrinsicCandidate
 9474 //   private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
 9475 //     for (int i = 0; i < len; i++) {
 9476 //       dst[dstOff++] = (char)(src[srcOff++] & 0xff);
 9477 //     }
 9478 //   }
 9479 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
 9480   XMMRegister tmp1, Register tmp2, KRegister mask) {
 9481   Label copy_chars_loop, done, below_threshold, avx3_threshold;
 9482   // rsi: src
 9483   // rdi: dst
 9484   // rdx: len
 9485   // rcx: tmp2
 9486 
 9487   // rsi holds start addr of source byte[] to be inflated
 9488   // rdi holds start addr of destination char[]
 9489   // rdx holds length
 9490   assert_different_registers(src, dst, len, tmp2);
 9491   movl(tmp2, len);
 9492   if ((UseAVX > 2) && // AVX512
 9493     VM_Version::supports_avx512vlbw() &&
 9494     VM_Version::supports_bmi2()) {
 9495 
 9496     Label copy_32_loop, copy_tail;
 9497     Register tmp3_aliased = len;
 9498 
 9499     // if length of the string is less than 16, handle it in an old fashioned way
 9500     testl(len, -16);
 9501     jcc(Assembler::zero, below_threshold);
 9502 
 9503     testl(len, -1 * AVX3Threshold);
 9504     jcc(Assembler::zero, avx3_threshold);
 9505 
 9506     // In order to use only one arithmetic operation for the main loop we use
 9507     // this pre-calculation
 9508     andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
 9509     andl(len, -32);     // vector count
 9510     jccb(Assembler::zero, copy_tail);
 9511 
 9512     lea(src, Address(src, len, Address::times_1));
 9513     lea(dst, Address(dst, len, Address::times_2));
 9514     negptr(len);
 9515 
 9516 
 9517     // inflate 32 chars per iter
 9518     bind(copy_32_loop);
 9519     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
 9520     evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
 9521     addptr(len, 32);
 9522     jcc(Assembler::notZero, copy_32_loop);
 9523 
 9524     bind(copy_tail);
 9525     // bail out when there is nothing to be done
 9526     testl(tmp2, -1); // we don't destroy the contents of tmp2 here
 9527     jcc(Assembler::zero, done);
 9528 
 9529     // ~(~0 << length), where length is the # of remaining elements to process
 9530     movl(tmp3_aliased, -1);
 9531     shlxl(tmp3_aliased, tmp3_aliased, tmp2);
 9532     notl(tmp3_aliased);
 9533     kmovdl(mask, tmp3_aliased);
 9534     evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
 9535     evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
 9536 
 9537     jmp(done);
 9538     bind(avx3_threshold);
 9539   }
 9540   if (UseSSE42Intrinsics) {
 9541     Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
 9542 
 9543     if (UseAVX > 1) {
 9544       andl(tmp2, (16 - 1));
 9545       andl(len, -16);
 9546       jccb(Assembler::zero, copy_new_tail);
 9547     } else {
 9548       andl(tmp2, 0x00000007);   // tail count (in chars)
 9549       andl(len, 0xfffffff8);    // vector count (in chars)
 9550       jccb(Assembler::zero, copy_tail);
 9551     }
 9552 
 9553     // vectored inflation
 9554     lea(src, Address(src, len, Address::times_1));
 9555     lea(dst, Address(dst, len, Address::times_2));
 9556     negptr(len);
 9557 
 9558     if (UseAVX > 1) {
 9559       bind(copy_16_loop);
 9560       vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
 9561       vmovdqu(Address(dst, len, Address::times_2), tmp1);
 9562       addptr(len, 16);
 9563       jcc(Assembler::notZero, copy_16_loop);
 9564 
 9565       bind(below_threshold);
 9566       bind(copy_new_tail);
 9567       movl(len, tmp2);
 9568       andl(tmp2, 0x00000007);
 9569       andl(len, 0xFFFFFFF8);
 9570       jccb(Assembler::zero, copy_tail);
 9571 
 9572       pmovzxbw(tmp1, Address(src, 0));
 9573       movdqu(Address(dst, 0), tmp1);
 9574       addptr(src, 8);
 9575       addptr(dst, 2 * 8);
 9576 
 9577       jmp(copy_tail, true);
 9578     }
 9579 
 9580     // inflate 8 chars per iter
 9581     bind(copy_8_loop);
 9582     pmovzxbw(tmp1, Address(src, len, Address::times_1));  // unpack to 8 words
 9583     movdqu(Address(dst, len, Address::times_2), tmp1);
 9584     addptr(len, 8);
 9585     jcc(Assembler::notZero, copy_8_loop);
 9586 
 9587     bind(copy_tail);
 9588     movl(len, tmp2);
 9589 
 9590     cmpl(len, 4);
 9591     jccb(Assembler::less, copy_bytes);
 9592 
 9593     movdl(tmp1, Address(src, 0));  // load 4 byte chars
 9594     pmovzxbw(tmp1, tmp1);
 9595     movq(Address(dst, 0), tmp1);
 9596     subptr(len, 4);
 9597     addptr(src, 4);
 9598     addptr(dst, 8);
 9599 
 9600     bind(copy_bytes);
 9601   } else {
 9602     bind(below_threshold);
 9603   }
 9604 
 9605   testl(len, len);
 9606   jccb(Assembler::zero, done);
 9607   lea(src, Address(src, len, Address::times_1));
 9608   lea(dst, Address(dst, len, Address::times_2));
 9609   negptr(len);
 9610 
 9611   // inflate 1 char per iter
 9612   bind(copy_chars_loop);
 9613   load_unsigned_byte(tmp2, Address(src, len, Address::times_1));  // load byte char
 9614   movw(Address(dst, len, Address::times_2), tmp2);  // inflate byte char to word
 9615   increment(len);
 9616   jcc(Assembler::notZero, copy_chars_loop);
 9617 
 9618   bind(done);
 9619 }
 9620 
 9621 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
 9622   switch(type) {
 9623     case T_BYTE:
 9624     case T_BOOLEAN:
 9625       evmovdqub(dst, kmask, src, merge, vector_len);
 9626       break;
 9627     case T_CHAR:
 9628     case T_SHORT:
 9629       evmovdquw(dst, kmask, src, merge, vector_len);
 9630       break;
 9631     case T_INT:
 9632     case T_FLOAT:
 9633       evmovdqul(dst, kmask, src, merge, vector_len);
 9634       break;
 9635     case T_LONG:
 9636     case T_DOUBLE:
 9637       evmovdquq(dst, kmask, src, merge, vector_len);
 9638       break;
 9639     default:
 9640       fatal("Unexpected type argument %s", type2name(type));
 9641       break;
 9642   }
 9643 }
 9644 
 9645 
 9646 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
 9647   switch(type) {
 9648     case T_BYTE:
 9649     case T_BOOLEAN:
 9650       evmovdqub(dst, kmask, src, merge, vector_len);
 9651       break;
 9652     case T_CHAR:
 9653     case T_SHORT:
 9654       evmovdquw(dst, kmask, src, merge, vector_len);
 9655       break;
 9656     case T_INT:
 9657     case T_FLOAT:
 9658       evmovdqul(dst, kmask, src, merge, vector_len);
 9659       break;
 9660     case T_LONG:
 9661     case T_DOUBLE:
 9662       evmovdquq(dst, kmask, src, merge, vector_len);
 9663       break;
 9664     default:
 9665       fatal("Unexpected type argument %s", type2name(type));
 9666       break;
 9667   }
 9668 }
 9669 
 9670 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
 9671   switch(type) {
 9672     case T_BYTE:
 9673     case T_BOOLEAN:
 9674       evmovdqub(dst, kmask, src, merge, vector_len);
 9675       break;
 9676     case T_CHAR:
 9677     case T_SHORT:
 9678       evmovdquw(dst, kmask, src, merge, vector_len);
 9679       break;
 9680     case T_INT:
 9681     case T_FLOAT:
 9682       evmovdqul(dst, kmask, src, merge, vector_len);
 9683       break;
 9684     case T_LONG:
 9685     case T_DOUBLE:
 9686       evmovdquq(dst, kmask, src, merge, vector_len);
 9687       break;
 9688     default:
 9689       fatal("Unexpected type argument %s", type2name(type));
 9690       break;
 9691   }
 9692 }
 9693 
 9694 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
 9695   switch(masklen) {
 9696     case 2:
 9697        knotbl(dst, src);
 9698        movl(rtmp, 3);
 9699        kmovbl(ktmp, rtmp);
 9700        kandbl(dst, ktmp, dst);
 9701        break;
 9702     case 4:
 9703        knotbl(dst, src);
 9704        movl(rtmp, 15);
 9705        kmovbl(ktmp, rtmp);
 9706        kandbl(dst, ktmp, dst);
 9707        break;
 9708     case 8:
 9709        knotbl(dst, src);
 9710        break;
 9711     case 16:
 9712        knotwl(dst, src);
 9713        break;
 9714     case 32:
 9715        knotdl(dst, src);
 9716        break;
 9717     case 64:
 9718        knotql(dst, src);
 9719        break;
 9720     default:
 9721       fatal("Unexpected vector length %d", masklen);
 9722       break;
 9723   }
 9724 }
 9725 
 9726 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9727   switch(type) {
 9728     case T_BOOLEAN:
 9729     case T_BYTE:
 9730        kandbl(dst, src1, src2);
 9731        break;
 9732     case T_CHAR:
 9733     case T_SHORT:
 9734        kandwl(dst, src1, src2);
 9735        break;
 9736     case T_INT:
 9737     case T_FLOAT:
 9738        kanddl(dst, src1, src2);
 9739        break;
 9740     case T_LONG:
 9741     case T_DOUBLE:
 9742        kandql(dst, src1, src2);
 9743        break;
 9744     default:
 9745       fatal("Unexpected type argument %s", type2name(type));
 9746       break;
 9747   }
 9748 }
 9749 
 9750 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9751   switch(type) {
 9752     case T_BOOLEAN:
 9753     case T_BYTE:
 9754        korbl(dst, src1, src2);
 9755        break;
 9756     case T_CHAR:
 9757     case T_SHORT:
 9758        korwl(dst, src1, src2);
 9759        break;
 9760     case T_INT:
 9761     case T_FLOAT:
 9762        kordl(dst, src1, src2);
 9763        break;
 9764     case T_LONG:
 9765     case T_DOUBLE:
 9766        korql(dst, src1, src2);
 9767        break;
 9768     default:
 9769       fatal("Unexpected type argument %s", type2name(type));
 9770       break;
 9771   }
 9772 }
 9773 
 9774 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
 9775   switch(type) {
 9776     case T_BOOLEAN:
 9777     case T_BYTE:
 9778        kxorbl(dst, src1, src2);
 9779        break;
 9780     case T_CHAR:
 9781     case T_SHORT:
 9782        kxorwl(dst, src1, src2);
 9783        break;
 9784     case T_INT:
 9785     case T_FLOAT:
 9786        kxordl(dst, src1, src2);
 9787        break;
 9788     case T_LONG:
 9789     case T_DOUBLE:
 9790        kxorql(dst, src1, src2);
 9791        break;
 9792     default:
 9793       fatal("Unexpected type argument %s", type2name(type));
 9794       break;
 9795   }
 9796 }
 9797 
 9798 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9799   switch(type) {
 9800     case T_BOOLEAN:
 9801     case T_BYTE:
 9802       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9803     case T_CHAR:
 9804     case T_SHORT:
 9805       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9806     case T_INT:
 9807     case T_FLOAT:
 9808       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9809     case T_LONG:
 9810     case T_DOUBLE:
 9811       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9812     default:
 9813       fatal("Unexpected type argument %s", type2name(type)); break;
 9814   }
 9815 }
 9816 
 9817 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9818   switch(type) {
 9819     case T_BOOLEAN:
 9820     case T_BYTE:
 9821       evpermb(dst, mask, nds, src, merge, vector_len); break;
 9822     case T_CHAR:
 9823     case T_SHORT:
 9824       evpermw(dst, mask, nds, src, merge, vector_len); break;
 9825     case T_INT:
 9826     case T_FLOAT:
 9827       evpermd(dst, mask, nds, src, merge, vector_len); break;
 9828     case T_LONG:
 9829     case T_DOUBLE:
 9830       evpermq(dst, mask, nds, src, merge, vector_len); break;
 9831     default:
 9832       fatal("Unexpected type argument %s", type2name(type)); break;
 9833   }
 9834 }
 9835 
 9836 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9837   switch(type) {
 9838     case T_BYTE:
 9839       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9840     case T_SHORT:
 9841       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9842     case T_INT:
 9843       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9844     case T_LONG:
 9845       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9846     default:
 9847       fatal("Unexpected type argument %s", type2name(type)); break;
 9848   }
 9849 }
 9850 
 9851 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9852   switch(type) {
 9853     case T_BYTE:
 9854       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9855     case T_SHORT:
 9856       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9857     case T_INT:
 9858       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9859     case T_LONG:
 9860       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9861     default:
 9862       fatal("Unexpected type argument %s", type2name(type)); break;
 9863   }
 9864 }
 9865 
 9866 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9867   switch(type) {
 9868     case T_BYTE:
 9869       evpminub(dst, mask, nds, src, merge, vector_len); break;
 9870     case T_SHORT:
 9871       evpminuw(dst, mask, nds, src, merge, vector_len); break;
 9872     case T_INT:
 9873       evpminud(dst, mask, nds, src, merge, vector_len); break;
 9874     case T_LONG:
 9875       evpminuq(dst, mask, nds, src, merge, vector_len); break;
 9876     default:
 9877       fatal("Unexpected type argument %s", type2name(type)); break;
 9878   }
 9879 }
 9880 
 9881 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9882   switch(type) {
 9883     case T_BYTE:
 9884       evpmaxub(dst, mask, nds, src, merge, vector_len); break;
 9885     case T_SHORT:
 9886       evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
 9887     case T_INT:
 9888       evpmaxud(dst, mask, nds, src, merge, vector_len); break;
 9889     case T_LONG:
 9890       evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
 9891     default:
 9892       fatal("Unexpected type argument %s", type2name(type)); break;
 9893   }
 9894 }
 9895 
 9896 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9897   switch(type) {
 9898     case T_BYTE:
 9899       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9900     case T_SHORT:
 9901       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9902     case T_INT:
 9903       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9904     case T_LONG:
 9905       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9906     case T_FLOAT:
 9907       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9908     case T_DOUBLE:
 9909       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9910     default:
 9911       fatal("Unexpected type argument %s", type2name(type)); break;
 9912   }
 9913 }
 9914 
 9915 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9916   switch(type) {
 9917     case T_BYTE:
 9918       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9919     case T_SHORT:
 9920       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9921     case T_INT:
 9922       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9923     case T_LONG:
 9924       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9925     case T_FLOAT:
 9926       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9927     case T_DOUBLE:
 9928       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9929     default:
 9930       fatal("Unexpected type argument %s", type2name(type)); break;
 9931   }
 9932 }
 9933 
 9934 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9935   switch(type) {
 9936     case T_BYTE:
 9937       evpminsb(dst, mask, nds, src, merge, vector_len); break;
 9938     case T_SHORT:
 9939       evpminsw(dst, mask, nds, src, merge, vector_len); break;
 9940     case T_INT:
 9941       evpminsd(dst, mask, nds, src, merge, vector_len); break;
 9942     case T_LONG:
 9943       evpminsq(dst, mask, nds, src, merge, vector_len); break;
 9944     case T_FLOAT:
 9945       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9946     case T_DOUBLE:
 9947       evminmaxpd(dst, mask, nds, src, merge, AVX10_2_MINMAX_MIN_COMPARE_SIGN, vector_len); break;
 9948     default:
 9949       fatal("Unexpected type argument %s", type2name(type)); break;
 9950   }
 9951 }
 9952 
 9953 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9954   switch(type) {
 9955     case T_BYTE:
 9956       evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
 9957     case T_SHORT:
 9958       evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
 9959     case T_INT:
 9960       evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
 9961     case T_LONG:
 9962       evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
 9963     case T_FLOAT:
 9964       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9965     case T_DOUBLE:
 9966       evminmaxps(dst, mask, nds, src, merge, AVX10_2_MINMAX_MAX_COMPARE_SIGN, vector_len); break;
 9967     default:
 9968       fatal("Unexpected type argument %s", type2name(type)); break;
 9969   }
 9970 }
 9971 
 9972 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9973   switch(type) {
 9974     case T_INT:
 9975       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9976     case T_LONG:
 9977       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9978     default:
 9979       fatal("Unexpected type argument %s", type2name(type)); break;
 9980   }
 9981 }
 9982 
 9983 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
 9984   switch(type) {
 9985     case T_INT:
 9986       evpxord(dst, mask, nds, src, merge, vector_len); break;
 9987     case T_LONG:
 9988       evpxorq(dst, mask, nds, src, merge, vector_len); break;
 9989     default:
 9990       fatal("Unexpected type argument %s", type2name(type)); break;
 9991   }
 9992 }
 9993 
 9994 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
 9995   switch(type) {
 9996     case T_INT:
 9997       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
 9998     case T_LONG:
 9999       evporq(dst, mask, nds, src, merge, vector_len); break;
10000     default:
10001       fatal("Unexpected type argument %s", type2name(type)); break;
10002   }
10003 }
10004 
10005 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
10006   switch(type) {
10007     case T_INT:
10008       Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
10009     case T_LONG:
10010       evporq(dst, mask, nds, src, merge, vector_len); break;
10011     default:
10012       fatal("Unexpected type argument %s", type2name(type)); break;
10013   }
10014 }
10015 
10016 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
10017   switch(type) {
10018     case T_INT:
10019       evpandd(dst, mask, nds, src, merge, vector_len); break;
10020     case T_LONG:
10021       evpandq(dst, mask, nds, src, merge, vector_len); break;
10022     default:
10023       fatal("Unexpected type argument %s", type2name(type)); break;
10024   }
10025 }
10026 
10027 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
10028   switch(type) {
10029     case T_INT:
10030       evpandd(dst, mask, nds, src, merge, vector_len); break;
10031     case T_LONG:
10032       evpandq(dst, mask, nds, src, merge, vector_len); break;
10033     default:
10034       fatal("Unexpected type argument %s", type2name(type)); break;
10035   }
10036 }
10037 
10038 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
10039   switch(masklen) {
10040     case 8:
10041        kortestbl(src1, src2);
10042        break;
10043     case 16:
10044        kortestwl(src1, src2);
10045        break;
10046     case 32:
10047        kortestdl(src1, src2);
10048        break;
10049     case 64:
10050        kortestql(src1, src2);
10051        break;
10052     default:
10053       fatal("Unexpected mask length %d", masklen);
10054       break;
10055   }
10056 }
10057 
10058 
10059 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
10060   switch(masklen)  {
10061     case 8:
10062        ktestbl(src1, src2);
10063        break;
10064     case 16:
10065        ktestwl(src1, src2);
10066        break;
10067     case 32:
10068        ktestdl(src1, src2);
10069        break;
10070     case 64:
10071        ktestql(src1, src2);
10072        break;
10073     default:
10074       fatal("Unexpected mask length %d", masklen);
10075       break;
10076   }
10077 }
10078 
10079 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
10080   switch(type) {
10081     case T_INT:
10082       evprold(dst, mask, src, shift, merge, vlen_enc); break;
10083     case T_LONG:
10084       evprolq(dst, mask, src, shift, merge, vlen_enc); break;
10085     default:
10086       fatal("Unexpected type argument %s", type2name(type)); break;
10087       break;
10088   }
10089 }
10090 
10091 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
10092   switch(type) {
10093     case T_INT:
10094       evprord(dst, mask, src, shift, merge, vlen_enc); break;
10095     case T_LONG:
10096       evprorq(dst, mask, src, shift, merge, vlen_enc); break;
10097     default:
10098       fatal("Unexpected type argument %s", type2name(type)); break;
10099   }
10100 }
10101 
10102 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
10103   switch(type) {
10104     case T_INT:
10105       evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
10106     case T_LONG:
10107       evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
10108     default:
10109       fatal("Unexpected type argument %s", type2name(type)); break;
10110   }
10111 }
10112 
10113 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
10114   switch(type) {
10115     case T_INT:
10116       evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
10117     case T_LONG:
10118       evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
10119     default:
10120       fatal("Unexpected type argument %s", type2name(type)); break;
10121   }
10122 }
10123 
10124 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10125   assert(rscratch != noreg || always_reachable(src), "missing");
10126 
10127   if (reachable(src)) {
10128     evpandq(dst, nds, as_Address(src), vector_len);
10129   } else {
10130     lea(rscratch, src);
10131     evpandq(dst, nds, Address(rscratch, 0), vector_len);
10132   }
10133 }
10134 
10135 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
10136   assert(rscratch != noreg || always_reachable(src), "missing");
10137 
10138   if (reachable(src)) {
10139     Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
10140   } else {
10141     lea(rscratch, src);
10142     Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
10143   }
10144 }
10145 
10146 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10147   assert(rscratch != noreg || always_reachable(src), "missing");
10148 
10149   if (reachable(src)) {
10150     evporq(dst, nds, as_Address(src), vector_len);
10151   } else {
10152     lea(rscratch, src);
10153     evporq(dst, nds, Address(rscratch, 0), vector_len);
10154   }
10155 }
10156 
10157 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10158   assert(rscratch != noreg || always_reachable(src), "missing");
10159 
10160   if (reachable(src)) {
10161     vpshufb(dst, nds, as_Address(src), vector_len);
10162   } else {
10163     lea(rscratch, src);
10164     vpshufb(dst, nds, Address(rscratch, 0), vector_len);
10165   }
10166 }
10167 
10168 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
10169   assert(rscratch != noreg || always_reachable(src), "missing");
10170 
10171   if (reachable(src)) {
10172     Assembler::vpor(dst, nds, as_Address(src), vector_len);
10173   } else {
10174     lea(rscratch, src);
10175     Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
10176   }
10177 }
10178 
10179 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
10180   assert(rscratch != noreg || always_reachable(src3), "missing");
10181 
10182   if (reachable(src3)) {
10183     vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
10184   } else {
10185     lea(rscratch, src3);
10186     vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
10187   }
10188 }
10189 
10190 #if COMPILER2_OR_JVMCI
10191 
10192 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
10193                                  Register length, Register temp, int vec_enc) {
10194   // Computing mask for predicated vector store.
10195   movptr(temp, -1);
10196   bzhiq(temp, temp, length);
10197   kmov(mask, temp);
10198   evmovdqu(bt, mask, dst, xmm, true, vec_enc);
10199 }
10200 
10201 // Set memory operation for length "less than" 64 bytes.
10202 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
10203                                        XMMRegister xmm, KRegister mask, Register length,
10204                                        Register temp, bool use64byteVector) {
10205   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10206   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10207   if (!use64byteVector) {
10208     fill32(dst, disp, xmm);
10209     subptr(length, 32 >> shift);
10210     fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
10211   } else {
10212     assert(MaxVectorSize == 64, "vector length != 64");
10213     fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
10214   }
10215 }
10216 
10217 
10218 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
10219                                        XMMRegister xmm, KRegister mask, Register length,
10220                                        Register temp) {
10221   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10222   const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
10223   fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
10224 }
10225 
10226 
10227 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
10228   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10229   vmovdqu(dst, xmm);
10230 }
10231 
10232 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
10233   fill32(Address(dst, disp), xmm);
10234 }
10235 
10236 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
10237   assert(MaxVectorSize >= 32, "vector length should be >= 32");
10238   if (!use64byteVector) {
10239     fill32(dst, xmm);
10240     fill32(dst.plus_disp(32), xmm);
10241   } else {
10242     evmovdquq(dst, xmm, Assembler::AVX_512bit);
10243   }
10244 }
10245 
10246 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
10247   fill64(Address(dst, disp), xmm, use64byteVector);
10248 }
10249 
10250 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
10251                                         Register count, Register rtmp, XMMRegister xtmp) {
10252   Label L_exit;
10253   Label L_fill_start;
10254   Label L_fill_64_bytes;
10255   Label L_fill_96_bytes;
10256   Label L_fill_128_bytes;
10257   Label L_fill_128_bytes_loop;
10258   Label L_fill_128_loop_header;
10259   Label L_fill_128_bytes_loop_header;
10260   Label L_fill_128_bytes_loop_pre_header;
10261   Label L_fill_zmm_sequence;
10262 
10263   int shift = -1;
10264   int avx3threshold = VM_Version::avx3_threshold();
10265   switch(type) {
10266     case T_BYTE:  shift = 0;
10267       break;
10268     case T_SHORT: shift = 1;
10269       break;
10270     case T_INT:   shift = 2;
10271       break;
10272     /* Uncomment when LONG fill stubs are supported.
10273     case T_LONG:  shift = 3;
10274       break;
10275     */
10276     default:
10277       fatal("Unhandled type: %s\n", type2name(type));
10278   }
10279 
10280   if ((avx3threshold != 0)  || (MaxVectorSize == 32)) {
10281 
10282     if (MaxVectorSize == 64) {
10283       cmpq(count, avx3threshold >> shift);
10284       jcc(Assembler::greater, L_fill_zmm_sequence);
10285     }
10286 
10287     evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
10288 
10289     bind(L_fill_start);
10290 
10291     cmpq(count, 32 >> shift);
10292     jccb(Assembler::greater, L_fill_64_bytes);
10293     fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
10294     jmp(L_exit);
10295 
10296     bind(L_fill_64_bytes);
10297     cmpq(count, 64 >> shift);
10298     jccb(Assembler::greater, L_fill_96_bytes);
10299     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
10300     jmp(L_exit);
10301 
10302     bind(L_fill_96_bytes);
10303     cmpq(count, 96 >> shift);
10304     jccb(Assembler::greater, L_fill_128_bytes);
10305     fill64(to, 0, xtmp);
10306     subq(count, 64 >> shift);
10307     fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
10308     jmp(L_exit);
10309 
10310     bind(L_fill_128_bytes);
10311     cmpq(count, 128 >> shift);
10312     jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
10313     fill64(to, 0, xtmp);
10314     fill32(to, 64, xtmp);
10315     subq(count, 96 >> shift);
10316     fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
10317     jmp(L_exit);
10318 
10319     bind(L_fill_128_bytes_loop_pre_header);
10320     {
10321       mov(rtmp, to);
10322       andq(rtmp, 31);
10323       jccb(Assembler::zero, L_fill_128_bytes_loop_header);
10324       negq(rtmp);
10325       addq(rtmp, 32);
10326       mov64(r8, -1L);
10327       bzhiq(r8, r8, rtmp);
10328       kmovql(k2, r8);
10329       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
10330       addq(to, rtmp);
10331       shrq(rtmp, shift);
10332       subq(count, rtmp);
10333     }
10334 
10335     cmpq(count, 128 >> shift);
10336     jcc(Assembler::less, L_fill_start);
10337 
10338     bind(L_fill_128_bytes_loop_header);
10339     subq(count, 128 >> shift);
10340 
10341     align32();
10342     bind(L_fill_128_bytes_loop);
10343       fill64(to, 0, xtmp);
10344       fill64(to, 64, xtmp);
10345       addq(to, 128);
10346       subq(count, 128 >> shift);
10347       jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
10348 
10349     addq(count, 128 >> shift);
10350     jcc(Assembler::zero, L_exit);
10351     jmp(L_fill_start);
10352   }
10353 
10354   if (MaxVectorSize == 64) {
10355     // Sequence using 64 byte ZMM register.
10356     Label L_fill_128_bytes_zmm;
10357     Label L_fill_192_bytes_zmm;
10358     Label L_fill_192_bytes_loop_zmm;
10359     Label L_fill_192_bytes_loop_header_zmm;
10360     Label L_fill_192_bytes_loop_pre_header_zmm;
10361     Label L_fill_start_zmm_sequence;
10362 
10363     bind(L_fill_zmm_sequence);
10364     evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
10365 
10366     bind(L_fill_start_zmm_sequence);
10367     cmpq(count, 64 >> shift);
10368     jccb(Assembler::greater, L_fill_128_bytes_zmm);
10369     fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
10370     jmp(L_exit);
10371 
10372     bind(L_fill_128_bytes_zmm);
10373     cmpq(count, 128 >> shift);
10374     jccb(Assembler::greater, L_fill_192_bytes_zmm);
10375     fill64(to, 0, xtmp, true);
10376     subq(count, 64 >> shift);
10377     fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
10378     jmp(L_exit);
10379 
10380     bind(L_fill_192_bytes_zmm);
10381     cmpq(count, 192 >> shift);
10382     jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
10383     fill64(to, 0, xtmp, true);
10384     fill64(to, 64, xtmp, true);
10385     subq(count, 128 >> shift);
10386     fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
10387     jmp(L_exit);
10388 
10389     bind(L_fill_192_bytes_loop_pre_header_zmm);
10390     {
10391       movq(rtmp, to);
10392       andq(rtmp, 63);
10393       jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
10394       negq(rtmp);
10395       addq(rtmp, 64);
10396       mov64(r8, -1L);
10397       bzhiq(r8, r8, rtmp);
10398       kmovql(k2, r8);
10399       evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
10400       addq(to, rtmp);
10401       shrq(rtmp, shift);
10402       subq(count, rtmp);
10403     }
10404 
10405     cmpq(count, 192 >> shift);
10406     jcc(Assembler::less, L_fill_start_zmm_sequence);
10407 
10408     bind(L_fill_192_bytes_loop_header_zmm);
10409     subq(count, 192 >> shift);
10410 
10411     align32();
10412     bind(L_fill_192_bytes_loop_zmm);
10413       fill64(to, 0, xtmp, true);
10414       fill64(to, 64, xtmp, true);
10415       fill64(to, 128, xtmp, true);
10416       addq(to, 192);
10417       subq(count, 192 >> shift);
10418       jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
10419 
10420     addq(count, 192 >> shift);
10421     jcc(Assembler::zero, L_exit);
10422     jmp(L_fill_start_zmm_sequence);
10423   }
10424   bind(L_exit);
10425 }
10426 #endif //COMPILER2_OR_JVMCI
10427 
10428 
10429 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
10430   Label done;
10431   cvttss2sil(dst, src);
10432   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10433   cmpl(dst, 0x80000000); // float_sign_flip
10434   jccb(Assembler::notEqual, done);
10435   subptr(rsp, 8);
10436   movflt(Address(rsp, 0), src);
10437   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
10438   pop(dst);
10439   bind(done);
10440 }
10441 
10442 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
10443   Label done;
10444   cvttsd2sil(dst, src);
10445   // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
10446   cmpl(dst, 0x80000000); // float_sign_flip
10447   jccb(Assembler::notEqual, done);
10448   subptr(rsp, 8);
10449   movdbl(Address(rsp, 0), src);
10450   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
10451   pop(dst);
10452   bind(done);
10453 }
10454 
10455 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
10456   Label done;
10457   cvttss2siq(dst, src);
10458   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10459   jccb(Assembler::notEqual, done);
10460   subptr(rsp, 8);
10461   movflt(Address(rsp, 0), src);
10462   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
10463   pop(dst);
10464   bind(done);
10465 }
10466 
10467 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10468   // Following code is line by line assembly translation rounding algorithm.
10469   // Please refer to java.lang.Math.round(float) algorithm for details.
10470   const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
10471   const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
10472   const int32_t FloatConsts_EXP_BIAS = 127;
10473   const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
10474   const int32_t MINUS_32 = 0xFFFFFFE0;
10475   Label L_special_case, L_block1, L_exit;
10476   movl(rtmp, FloatConsts_EXP_BIT_MASK);
10477   movdl(dst, src);
10478   andl(dst, rtmp);
10479   sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
10480   movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
10481   subl(rtmp, dst);
10482   movl(rcx, rtmp);
10483   movl(dst, MINUS_32);
10484   testl(rtmp, dst);
10485   jccb(Assembler::notEqual, L_special_case);
10486   movdl(dst, src);
10487   andl(dst, FloatConsts_SIGNIF_BIT_MASK);
10488   orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
10489   movdl(rtmp, src);
10490   testl(rtmp, rtmp);
10491   jccb(Assembler::greaterEqual, L_block1);
10492   negl(dst);
10493   bind(L_block1);
10494   sarl(dst);
10495   addl(dst, 0x1);
10496   sarl(dst, 0x1);
10497   jmp(L_exit);
10498   bind(L_special_case);
10499   convert_f2i(dst, src);
10500   bind(L_exit);
10501 }
10502 
10503 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
10504   // Following code is line by line assembly translation rounding algorithm.
10505   // Please refer to java.lang.Math.round(double) algorithm for details.
10506   const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
10507   const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
10508   const int64_t DoubleConsts_EXP_BIAS = 1023;
10509   const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
10510   const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
10511   Label L_special_case, L_block1, L_exit;
10512   mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
10513   movq(dst, src);
10514   andq(dst, rtmp);
10515   sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
10516   mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
10517   subq(rtmp, dst);
10518   movq(rcx, rtmp);
10519   mov64(dst, MINUS_64);
10520   testq(rtmp, dst);
10521   jccb(Assembler::notEqual, L_special_case);
10522   movq(dst, src);
10523   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
10524   andq(dst, rtmp);
10525   mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
10526   orq(dst, rtmp);
10527   movq(rtmp, src);
10528   testq(rtmp, rtmp);
10529   jccb(Assembler::greaterEqual, L_block1);
10530   negq(dst);
10531   bind(L_block1);
10532   sarq(dst);
10533   addq(dst, 0x1);
10534   sarq(dst, 0x1);
10535   jmp(L_exit);
10536   bind(L_special_case);
10537   convert_d2l(dst, src);
10538   bind(L_exit);
10539 }
10540 
10541 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
10542   Label done;
10543   cvttsd2siq(dst, src);
10544   cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
10545   jccb(Assembler::notEqual, done);
10546   subptr(rsp, 8);
10547   movdbl(Address(rsp, 0), src);
10548   call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
10549   pop(dst);
10550   bind(done);
10551 }
10552 
10553 void MacroAssembler::cache_wb(Address line)
10554 {
10555   // 64 bit cpus always support clflush
10556   assert(VM_Version::supports_clflush(), "clflush should be available");
10557   bool optimized = VM_Version::supports_clflushopt();
10558   bool no_evict = VM_Version::supports_clwb();
10559 
10560   // prefer clwb (writeback without evict) otherwise
10561   // prefer clflushopt (potentially parallel writeback with evict)
10562   // otherwise fallback on clflush (serial writeback with evict)
10563 
10564   if (optimized) {
10565     if (no_evict) {
10566       clwb(line);
10567     } else {
10568       clflushopt(line);
10569     }
10570   } else {
10571     // no need for fence when using CLFLUSH
10572     clflush(line);
10573   }
10574 }
10575 
10576 void MacroAssembler::cache_wbsync(bool is_pre)
10577 {
10578   assert(VM_Version::supports_clflush(), "clflush should be available");
10579   bool optimized = VM_Version::supports_clflushopt();
10580   bool no_evict = VM_Version::supports_clwb();
10581 
10582   // pick the correct implementation
10583 
10584   if (!is_pre && (optimized || no_evict)) {
10585     // need an sfence for post flush when using clflushopt or clwb
10586     // otherwise no no need for any synchroniaztion
10587 
10588     sfence();
10589   }
10590 }
10591 
10592 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
10593   switch (cond) {
10594     // Note some conditions are synonyms for others
10595     case Assembler::zero:         return Assembler::notZero;
10596     case Assembler::notZero:      return Assembler::zero;
10597     case Assembler::less:         return Assembler::greaterEqual;
10598     case Assembler::lessEqual:    return Assembler::greater;
10599     case Assembler::greater:      return Assembler::lessEqual;
10600     case Assembler::greaterEqual: return Assembler::less;
10601     case Assembler::below:        return Assembler::aboveEqual;
10602     case Assembler::belowEqual:   return Assembler::above;
10603     case Assembler::above:        return Assembler::belowEqual;
10604     case Assembler::aboveEqual:   return Assembler::below;
10605     case Assembler::overflow:     return Assembler::noOverflow;
10606     case Assembler::noOverflow:   return Assembler::overflow;
10607     case Assembler::negative:     return Assembler::positive;
10608     case Assembler::positive:     return Assembler::negative;
10609     case Assembler::parity:       return Assembler::noParity;
10610     case Assembler::noParity:     return Assembler::parity;
10611   }
10612   ShouldNotReachHere(); return Assembler::overflow;
10613 }
10614 
10615 // This is simply a call to Thread::current()
10616 void MacroAssembler::get_thread_slow(Register thread) {
10617   if (thread != rax) {
10618     push(rax);
10619   }
10620   push(rdi);
10621   push(rsi);
10622   push(rdx);
10623   push(rcx);
10624   push(r8);
10625   push(r9);
10626   push(r10);
10627   push(r11);
10628 
10629   MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
10630 
10631   pop(r11);
10632   pop(r10);
10633   pop(r9);
10634   pop(r8);
10635   pop(rcx);
10636   pop(rdx);
10637   pop(rsi);
10638   pop(rdi);
10639   if (thread != rax) {
10640     mov(thread, rax);
10641     pop(rax);
10642   }
10643 }
10644 
10645 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
10646   Label L_stack_ok;
10647   if (bias == 0) {
10648     testptr(sp, 2 * wordSize - 1);
10649   } else {
10650     // lea(tmp, Address(rsp, bias);
10651     mov(tmp, sp);
10652     addptr(tmp, bias);
10653     testptr(tmp, 2 * wordSize - 1);
10654   }
10655   jcc(Assembler::equal, L_stack_ok);
10656   block_comment(msg);
10657   stop(msg);
10658   bind(L_stack_ok);
10659 }
10660 
10661 // Implements fast-locking.
10662 //
10663 // obj: the object to be locked
10664 // reg_rax: rax
10665 // thread: the thread which attempts to lock obj
10666 // tmp: a temporary register
10667 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
10668   Register thread = r15_thread;
10669 
10670   assert(reg_rax == rax, "");
10671   assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
10672 
10673   Label push;
10674   const Register top = tmp;
10675 
10676   // Preload the markWord. It is important that this is the first
10677   // instruction emitted as it is part of C1's null check semantics.
10678   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10679 
10680   if (UseObjectMonitorTable) {
10681     // Clear cache in case fast locking succeeds or we need to take the slow-path.
10682     movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
10683   }
10684 
10685   if (DiagnoseSyncOnValueBasedClasses != 0) {
10686     load_klass(tmp, obj, rscratch1);
10687     testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
10688     jcc(Assembler::notZero, slow);
10689   }
10690 
10691   // Load top.
10692   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10693 
10694   // Check if the lock-stack is full.
10695   cmpl(top, LockStack::end_offset());
10696   jcc(Assembler::greaterEqual, slow);
10697 
10698   // Check for recursion.
10699   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10700   jcc(Assembler::equal, push);
10701 
10702   // Check header for monitor (0b10).
10703   testptr(reg_rax, markWord::monitor_value);
10704   jcc(Assembler::notZero, slow);
10705 
10706   // Try to lock. Transition lock bits 0b01 => 0b00
10707   movptr(tmp, reg_rax);
10708   andptr(tmp, ~(int32_t)markWord::unlocked_value);
10709   orptr(reg_rax, markWord::unlocked_value);
10710   // Mask inline_type bit such that we go to the slow path if object is an inline type
10711   andptr(reg_rax, ~((int) markWord::inline_type_bit_in_place));
10712 
10713   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10714   jcc(Assembler::notEqual, slow);
10715 
10716   // Restore top, CAS clobbers register.
10717   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10718 
10719   bind(push);
10720   // After successful lock, push object on lock-stack.
10721   movptr(Address(thread, top), obj);
10722   incrementl(top, oopSize);
10723   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
10724 }
10725 
10726 // Implements fast-unlocking.
10727 //
10728 // obj: the object to be unlocked
10729 // reg_rax: rax
10730 // thread: the thread
10731 // tmp: a temporary register
10732 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
10733   Register thread = r15_thread;
10734 
10735   assert(reg_rax == rax, "");
10736   assert_different_registers(obj, reg_rax, thread, tmp);
10737 
10738   Label unlocked, push_and_slow;
10739   const Register top = tmp;
10740 
10741   // Check if obj is top of lock-stack.
10742   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10743   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10744   jcc(Assembler::notEqual, slow);
10745 
10746   // Pop lock-stack.
10747   DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
10748   subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10749 
10750   // Check if recursive.
10751   cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
10752   jcc(Assembler::equal, unlocked);
10753 
10754   // Not recursive. Check header for monitor (0b10).
10755   movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
10756   testptr(reg_rax, markWord::monitor_value);
10757   jcc(Assembler::notZero, push_and_slow);
10758 
10759 #ifdef ASSERT
10760   // Check header not unlocked (0b01).
10761   Label not_unlocked;
10762   testptr(reg_rax, markWord::unlocked_value);
10763   jcc(Assembler::zero, not_unlocked);
10764   stop("fast_unlock already unlocked");
10765   bind(not_unlocked);
10766 #endif
10767 
10768   // Try to unlock. Transition lock bits 0b00 => 0b01
10769   movptr(tmp, reg_rax);
10770   orptr(tmp, markWord::unlocked_value);
10771   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10772   jcc(Assembler::equal, unlocked);
10773 
10774   bind(push_and_slow);
10775   // Restore lock-stack and handle the unlock in runtime.
10776 #ifdef ASSERT
10777   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10778   movptr(Address(thread, top), obj);
10779 #endif
10780   addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
10781   jmp(slow);
10782 
10783   bind(unlocked);
10784 }
10785 
10786 // Saves legacy GPRs state on stack.
10787 void MacroAssembler::save_legacy_gprs() {
10788   subq(rsp, 16 * wordSize);
10789   movq(Address(rsp, 15 * wordSize), rax);
10790   movq(Address(rsp, 14 * wordSize), rcx);
10791   movq(Address(rsp, 13 * wordSize), rdx);
10792   movq(Address(rsp, 12 * wordSize), rbx);
10793   movq(Address(rsp, 10 * wordSize), rbp);
10794   movq(Address(rsp, 9 * wordSize), rsi);
10795   movq(Address(rsp, 8 * wordSize), rdi);
10796   movq(Address(rsp, 7 * wordSize), r8);
10797   movq(Address(rsp, 6 * wordSize), r9);
10798   movq(Address(rsp, 5 * wordSize), r10);
10799   movq(Address(rsp, 4 * wordSize), r11);
10800   movq(Address(rsp, 3 * wordSize), r12);
10801   movq(Address(rsp, 2 * wordSize), r13);
10802   movq(Address(rsp, wordSize), r14);
10803   movq(Address(rsp, 0), r15);
10804 }
10805 
10806 // Resotres back legacy GPRs state from stack.
10807 void MacroAssembler::restore_legacy_gprs() {
10808   movq(r15, Address(rsp, 0));
10809   movq(r14, Address(rsp, wordSize));
10810   movq(r13, Address(rsp, 2 * wordSize));
10811   movq(r12, Address(rsp, 3 * wordSize));
10812   movq(r11, Address(rsp, 4 * wordSize));
10813   movq(r10, Address(rsp, 5 * wordSize));
10814   movq(r9,  Address(rsp, 6 * wordSize));
10815   movq(r8,  Address(rsp, 7 * wordSize));
10816   movq(rdi, Address(rsp, 8 * wordSize));
10817   movq(rsi, Address(rsp, 9 * wordSize));
10818   movq(rbp, Address(rsp, 10 * wordSize));
10819   movq(rbx, Address(rsp, 12 * wordSize));
10820   movq(rdx, Address(rsp, 13 * wordSize));
10821   movq(rcx, Address(rsp, 14 * wordSize));
10822   movq(rax, Address(rsp, 15 * wordSize));
10823   addq(rsp, 16 * wordSize);
10824 }
10825 
10826 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
10827   if (VM_Version::supports_apx_f()) {
10828     esetzucc(comparison, dst);
10829   } else {
10830     setb(comparison, dst);
10831     movzbl(dst, dst);
10832   }
10833 }