1 /*
   2  * Copyright (c) 1997, 2023, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2021, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include <sys/types.h>
  27 
  28 #include "precompiled.hpp"
  29 #include "asm/assembler.hpp"
  30 #include "asm/assembler.inline.hpp"
  31 #include "ci/ciEnv.hpp"
  32 #include "compiler/compileTask.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "compiler/oopMap.hpp"
  35 #include "gc/shared/barrierSet.hpp"
  36 #include "gc/shared/barrierSetAssembler.hpp"
  37 #include "gc/shared/cardTableBarrierSet.hpp"
  38 #include "gc/shared/cardTable.hpp"
  39 #include "gc/shared/collectedHeap.hpp"
  40 #include "gc/shared/tlab_globals.hpp"
  41 #include "interpreter/bytecodeHistogram.hpp"
  42 #include "interpreter/interpreter.hpp"
  43 #include "jvm.h"
  44 #include "memory/resourceArea.hpp"
  45 #include "memory/universe.hpp"
  46 #include "nativeInst_aarch64.hpp"
  47 #include "oops/accessDecorators.hpp"
  48 #include "oops/compressedOops.inline.hpp"
  49 #include "oops/klass.inline.hpp"
  50 #include "runtime/continuation.hpp"
  51 #include "runtime/icache.hpp"
  52 #include "runtime/interfaceSupport.inline.hpp"
  53 #include "runtime/javaThread.hpp"
  54 #include "runtime/jniHandles.inline.hpp"
  55 #include "runtime/sharedRuntime.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "utilities/powerOfTwo.hpp"
  58 #ifdef COMPILER1
  59 #include "c1/c1_LIRAssembler.hpp"
  60 #endif
  61 #ifdef COMPILER2
  62 #include "oops/oop.hpp"
  63 #include "opto/compile.hpp"
  64 #include "opto/node.hpp"
  65 #include "opto/output.hpp"
  66 #endif
  67 
  68 #ifdef PRODUCT
  69 #define BLOCK_COMMENT(str) /* nothing */
  70 #else
  71 #define BLOCK_COMMENT(str) block_comment(str)
  72 #endif
  73 #define STOP(str) stop(str);
  74 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  75 
  76 #ifdef ASSERT
  77 extern "C" void disnm(intptr_t p);
  78 #endif
  79 // Target-dependent relocation processing
  80 //
  81 // Instruction sequences whose target may need to be retrieved or
  82 // patched are distinguished by their leading instruction, sorting
  83 // them into three main instruction groups and related subgroups.
  84 //
  85 // 1) Branch, Exception and System (insn count = 1)
  86 //    1a) Unconditional branch (immediate):
  87 //      b/bl imm19
  88 //    1b) Compare & branch (immediate):
  89 //      cbz/cbnz Rt imm19
  90 //    1c) Test & branch (immediate):
  91 //      tbz/tbnz Rt imm14
  92 //    1d) Conditional branch (immediate):
  93 //      b.cond imm19
  94 //
  95 // 2) Loads and Stores (insn count = 1)
  96 //    2a) Load register literal:
  97 //      ldr Rt imm19
  98 //
  99 // 3) Data Processing Immediate (insn count = 2 or 3)
 100 //    3a) PC-rel. addressing
 101 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 102 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 103 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 104 //      adr/adrp Rx imm21
 105 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 106 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 107 //      The latter form can only happen when the target is an
 108 //      ExternalAddress, and (by definition) ExternalAddresses don't
 109 //      move. Because of that property, there is never any need to
 110 //      patch the last of the three instructions. However,
 111 //      MacroAssembler::target_addr_for_insn takes all three
 112 //      instructions into account and returns the correct address.
 113 //    3b) Move wide (immediate)
 114 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 115 //
 116 // A switch on a subset of the instruction's bits provides an
 117 // efficient dispatch to these subcases.
 118 //
 119 // insn[28:26] -> main group ('x' == don't care)
 120 //   00x -> UNALLOCATED
 121 //   100 -> Data Processing Immediate
 122 //   101 -> Branch, Exception and System
 123 //   x1x -> Loads and Stores
 124 //
 125 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 126 // n.b. in some cases extra bits need to be checked to verify the
 127 // instruction is as expected
 128 //
 129 // 1) ... xx101x Branch, Exception and System
 130 //   1a)  00___x Unconditional branch (immediate)
 131 //   1b)  01___0 Compare & branch (immediate)
 132 //   1c)  01___1 Test & branch (immediate)
 133 //   1d)  10___0 Conditional branch (immediate)
 134 //        other  Should not happen
 135 //
 136 // 2) ... xxx1x0 Loads and Stores
 137 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 138 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 139 //                strictly should be 64 bit non-FP/SIMD i.e.
 140 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 141 //
 142 // 3) ... xx100x Data Processing Immediate
 143 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 144 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 145 //                 strictly should be 64 bit movz #imm16<<0
 146 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 147 //
 148 class RelocActions {
 149 protected:
 150   typedef int (*reloc_insn)(address insn_addr, address &target);
 151 
 152   virtual reloc_insn adrpMem() = 0;
 153   virtual reloc_insn adrpAdd() = 0;
 154   virtual reloc_insn adrpMovk() = 0;
 155 
 156   const address _insn_addr;
 157   const uint32_t _insn;
 158 
 159   static uint32_t insn_at(address insn_addr, int n) {
 160     return ((uint32_t*)insn_addr)[n];
 161   }
 162   uint32_t insn_at(int n) const {
 163     return insn_at(_insn_addr, n);
 164   }
 165 
 166 public:
 167 
 168   RelocActions(address insn_addr) : _insn_addr(insn_addr), _insn(insn_at(insn_addr, 0)) {}
 169   RelocActions(address insn_addr, uint32_t insn)
 170     :  _insn_addr(insn_addr), _insn(insn) {}
 171 
 172   virtual int unconditionalBranch(address insn_addr, address &target) = 0;
 173   virtual int conditionalBranch(address insn_addr, address &target) = 0;
 174   virtual int testAndBranch(address insn_addr, address &target) = 0;
 175   virtual int loadStore(address insn_addr, address &target) = 0;
 176   virtual int adr(address insn_addr, address &target) = 0;
 177   virtual int adrp(address insn_addr, address &target, reloc_insn inner) = 0;
 178   virtual int immediate(address insn_addr, address &target) = 0;
 179   virtual void verify(address insn_addr, address &target) = 0;
 180 
 181   int ALWAYSINLINE run(address insn_addr, address &target) {
 182     int instructions = 1;
 183 
 184     uint32_t dispatch = Instruction_aarch64::extract(_insn, 30, 25);
 185     switch(dispatch) {
 186       case 0b001010:
 187       case 0b001011: {
 188         instructions = unconditionalBranch(insn_addr, target);
 189         break;
 190       }
 191       case 0b101010:   // Conditional branch (immediate)
 192       case 0b011010: { // Compare & branch (immediate)
 193         instructions = conditionalBranch(insn_addr, target);
 194           break;
 195       }
 196       case 0b011011: {
 197         instructions = testAndBranch(insn_addr, target);
 198         break;
 199       }
 200       case 0b001100:
 201       case 0b001110:
 202       case 0b011100:
 203       case 0b011110:
 204       case 0b101100:
 205       case 0b101110:
 206       case 0b111100:
 207       case 0b111110: {
 208         // load/store
 209         if ((Instruction_aarch64::extract(_insn, 29, 24) & 0b111011) == 0b011000) {
 210           // Load register (literal)
 211           instructions = loadStore(insn_addr, target);
 212           break;
 213         } else {
 214           // nothing to do
 215           assert(target == 0, "did not expect to relocate target for polling page load");
 216         }
 217         break;
 218       }
 219       case 0b001000:
 220       case 0b011000:
 221       case 0b101000:
 222       case 0b111000: {
 223         // adr/adrp
 224         assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 225         int shift = Instruction_aarch64::extract(_insn, 31, 31);
 226         if (shift) {
 227           uint32_t insn2 = insn_at(1);
 228           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 229               Instruction_aarch64::extract(_insn, 4, 0) ==
 230               Instruction_aarch64::extract(insn2, 9, 5)) {
 231             instructions = adrp(insn_addr, target, adrpMem());
 232           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 233                      Instruction_aarch64::extract(_insn, 4, 0) ==
 234                      Instruction_aarch64::extract(insn2, 4, 0)) {
 235             instructions = adrp(insn_addr, target, adrpAdd());
 236           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 237                      Instruction_aarch64::extract(_insn, 4, 0) ==
 238                      Instruction_aarch64::extract(insn2, 4, 0)) {
 239             instructions = adrp(insn_addr, target, adrpMovk());
 240           } else {
 241             ShouldNotReachHere();
 242           }
 243         } else {
 244           instructions = adr(insn_addr, target);
 245         }
 246         break;
 247       }
 248       case 0b001001:
 249       case 0b011001:
 250       case 0b101001:
 251       case 0b111001: {
 252         instructions = immediate(insn_addr, target);
 253         break;
 254       }
 255       default: {
 256         ShouldNotReachHere();
 257       }
 258     }
 259 
 260     verify(insn_addr, target);
 261     return instructions * NativeInstruction::instruction_size;
 262   }
 263 };
 264 
 265 class Patcher : public RelocActions {
 266   virtual reloc_insn adrpMem() { return &Patcher::adrpMem_impl; }
 267   virtual reloc_insn adrpAdd() { return &Patcher::adrpAdd_impl; }
 268   virtual reloc_insn adrpMovk() { return &Patcher::adrpMovk_impl; }
 269 
 270 public:
 271   Patcher(address insn_addr) : RelocActions(insn_addr) {}
 272 
 273   virtual int unconditionalBranch(address insn_addr, address &target) {
 274     intptr_t offset = (target - insn_addr) >> 2;
 275     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 276     return 1;
 277   }
 278   virtual int conditionalBranch(address insn_addr, address &target) {
 279     intptr_t offset = (target - insn_addr) >> 2;
 280     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 281     return 1;
 282   }
 283   virtual int testAndBranch(address insn_addr, address &target) {
 284     intptr_t offset = (target - insn_addr) >> 2;
 285     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 286     return 1;
 287   }
 288   virtual int loadStore(address insn_addr, address &target) {
 289     intptr_t offset = (target - insn_addr) >> 2;
 290     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 291     return 1;
 292   }
 293   virtual int adr(address insn_addr, address &target) {
 294 #ifdef ASSERT
 295     assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 296 #endif
 297     // PC-rel. addressing
 298     ptrdiff_t offset = target - insn_addr;
 299     int offset_lo = offset & 3;
 300     offset >>= 2;
 301     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 302     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 303     return 1;
 304   }
 305   virtual int adrp(address insn_addr, address &target, reloc_insn inner) {
 306     int instructions = 1;
 307 #ifdef ASSERT
 308     assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 309 #endif
 310     ptrdiff_t offset = target - insn_addr;
 311     instructions = 2;
 312     precond(inner != nullptr);
 313     // Give the inner reloc a chance to modify the target.
 314     address adjusted_target = target;
 315     instructions = (*inner)(insn_addr, adjusted_target);
 316     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 317     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 318     offset = adr_page - pc_page;
 319     int offset_lo = offset & 3;
 320     offset >>= 2;
 321     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 322     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 323     return instructions;
 324   }
 325   static int adrpMem_impl(address insn_addr, address &target) {
 326     uintptr_t dest = (uintptr_t)target;
 327     int offset_lo = dest & 0xfff;
 328     uint32_t insn2 = insn_at(insn_addr, 1);
 329     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 330     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 331     guarantee(((dest >> size) << size) == dest, "misaligned target");
 332     return 2;
 333   }
 334   static int adrpAdd_impl(address insn_addr, address &target) {
 335     uintptr_t dest = (uintptr_t)target;
 336     int offset_lo = dest & 0xfff;
 337     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 338     return 2;
 339   }
 340   static int adrpMovk_impl(address insn_addr, address &target) {
 341     uintptr_t dest = uintptr_t(target);
 342     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 343     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 344     target = address(dest);
 345     return 2;
 346   }
 347   virtual int immediate(address insn_addr, address &target) {
 348     assert(Instruction_aarch64::extract(_insn, 31, 21) == 0b11010010100, "must be");
 349     uint64_t dest = (uint64_t)target;
 350     // Move wide constant
 351     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 352     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 353     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 354     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 355     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 356     return 3;
 357   }
 358   virtual void verify(address insn_addr, address &target) {
 359 #ifdef ASSERT
 360     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 361     if (!(address_is == target)) {
 362       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 363       disnm((intptr_t)insn_addr);
 364       assert(address_is == target, "should be");
 365     }
 366 #endif
 367   }
 368 };
 369 
 370 // If insn1 and insn2 use the same register to form an address, either
 371 // by an offsetted LDR or a simple ADD, return the offset. If the
 372 // second instruction is an LDR, the offset may be scaled.
 373 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 374   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 375       Instruction_aarch64::extract(insn1, 4, 0) ==
 376       Instruction_aarch64::extract(insn2, 9, 5)) {
 377     // Load/store register (unsigned immediate)
 378     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 379     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 380     byte_offset <<= size;
 381     return true;
 382   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 383              Instruction_aarch64::extract(insn1, 4, 0) ==
 384              Instruction_aarch64::extract(insn2, 4, 0)) {
 385     // add (immediate)
 386     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 387     return true;
 388   }
 389   return false;
 390 }
 391 
 392 class Decoder : public RelocActions {
 393   virtual reloc_insn adrpMem() { return &Decoder::adrpMem_impl; }
 394   virtual reloc_insn adrpAdd() { return &Decoder::adrpAdd_impl; }
 395   virtual reloc_insn adrpMovk() { return &Decoder::adrpMovk_impl; }
 396 
 397 public:
 398   Decoder(address insn_addr, uint32_t insn) : RelocActions(insn_addr, insn) {}
 399 
 400   virtual int loadStore(address insn_addr, address &target) {
 401     intptr_t offset = Instruction_aarch64::sextract(_insn, 23, 5);
 402     target = insn_addr + (offset << 2);
 403     return 1;
 404   }
 405   virtual int unconditionalBranch(address insn_addr, address &target) {
 406     intptr_t offset = Instruction_aarch64::sextract(_insn, 25, 0);
 407     target = insn_addr + (offset << 2);
 408     return 1;
 409   }
 410   virtual int conditionalBranch(address insn_addr, address &target) {
 411     intptr_t offset = Instruction_aarch64::sextract(_insn, 23, 5);
 412     target = address(((uint64_t)insn_addr + (offset << 2)));
 413     return 1;
 414   }
 415   virtual int testAndBranch(address insn_addr, address &target) {
 416     intptr_t offset = Instruction_aarch64::sextract(_insn, 18, 5);
 417     target = address(((uint64_t)insn_addr + (offset << 2)));
 418     return 1;
 419   }
 420   virtual int adr(address insn_addr, address &target) {
 421     // PC-rel. addressing
 422     intptr_t offset = Instruction_aarch64::extract(_insn, 30, 29);
 423     offset |= Instruction_aarch64::sextract(_insn, 23, 5) << 2;
 424     target = address((uint64_t)insn_addr + offset);
 425     return 1;
 426   }
 427   virtual int adrp(address insn_addr, address &target, reloc_insn inner) {
 428     assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 429     intptr_t offset = Instruction_aarch64::extract(_insn, 30, 29);
 430     offset |= Instruction_aarch64::sextract(_insn, 23, 5) << 2;
 431     int shift = 12;
 432     offset <<= shift;
 433     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 434     target_page &= ((uint64_t)-1) << shift;
 435     uint32_t insn2 = insn_at(1);
 436     target = address(target_page);
 437     precond(inner != nullptr);
 438     (*inner)(insn_addr, target);
 439     return 2;
 440   }
 441   static int adrpMem_impl(address insn_addr, address &target) {
 442     uint32_t insn2 = insn_at(insn_addr, 1);
 443     // Load/store register (unsigned immediate)
 444     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 445     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 446     byte_offset <<= size;
 447     target += byte_offset;
 448     return 2;
 449   }
 450   static int adrpAdd_impl(address insn_addr, address &target) {
 451     uint32_t insn2 = insn_at(insn_addr, 1);
 452     // add (immediate)
 453     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 454     target += byte_offset;
 455     return 2;
 456   }
 457   static int adrpMovk_impl(address insn_addr, address &target) {
 458     uint32_t insn2 = insn_at(insn_addr, 1);
 459     uint64_t dest = uint64_t(target);
 460     dest = (dest & 0xffff0000ffffffff) |
 461       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 462     target = address(dest);
 463 
 464     // We know the destination 4k page. Maybe we have a third
 465     // instruction.
 466     uint32_t insn = insn_at(insn_addr, 0);
 467     uint32_t insn3 = insn_at(insn_addr, 2);
 468     ptrdiff_t byte_offset;
 469     if (offset_for(insn, insn3, byte_offset)) {
 470       target += byte_offset;
 471       return 3;
 472     } else {
 473       return 2;
 474     }
 475   }
 476   virtual int immediate(address insn_addr, address &target) {
 477     uint32_t *insns = (uint32_t *)insn_addr;
 478     assert(Instruction_aarch64::extract(_insn, 31, 21) == 0b11010010100, "must be");
 479     // Move wide constant: movz, movk, movk.  See movptr().
 480     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 481     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 482     target = address(uint64_t(Instruction_aarch64::extract(_insn, 20, 5))
 483                  + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 484                  + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 485     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 486     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 487     return 3;
 488   }
 489   virtual void verify(address insn_addr, address &target) {
 490   }
 491 };
 492 
 493 address MacroAssembler::target_addr_for_insn(address insn_addr, uint32_t insn) {
 494   Decoder decoder(insn_addr, insn);
 495   address target;
 496   decoder.run(insn_addr, target);
 497   return target;
 498 }
 499 
 500 // Patch any kind of instruction; there may be several instructions.
 501 // Return the total length (in bytes) of the instructions.
 502 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 503   Patcher patcher(insn_addr);
 504   return patcher.run(insn_addr, target);
 505 }
 506 
 507 int MacroAssembler::patch_oop(address insn_addr, address o) {
 508   int instructions;
 509   unsigned insn = *(unsigned*)insn_addr;
 510   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 511 
 512   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 513   // narrow OOPs by setting the upper 16 bits in the first
 514   // instruction.
 515   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 516     // Move narrow OOP
 517     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 518     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 519     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 520     instructions = 2;
 521   } else {
 522     // Move wide OOP
 523     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 524     uintptr_t dest = (uintptr_t)o;
 525     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 526     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 527     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 528     instructions = 3;
 529   }
 530   return instructions * NativeInstruction::instruction_size;
 531 }
 532 
 533 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 534   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 535   // We encode narrow ones by setting the upper 16 bits in the first
 536   // instruction.
 537   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 538   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 539          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 540 
 541   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 542   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 543   return 2 * NativeInstruction::instruction_size;
 544 }
 545 
 546 address MacroAssembler::target_addr_for_insn_or_null(address insn_addr, unsigned insn) {
 547   if (NativeInstruction::is_ldrw_to_zr(address(&insn))) {
 548     return nullptr;
 549   }
 550   return MacroAssembler::target_addr_for_insn(insn_addr, insn);
 551 }
 552 
 553 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool acquire, bool in_nmethod, Register tmp) {
 554   if (acquire) {
 555     lea(tmp, Address(rthread, JavaThread::polling_word_offset()));
 556     ldar(tmp, tmp);
 557   } else {
 558     ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 559   }
 560   if (at_return) {
 561     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 562     // we may safely use the sp instead to perform the stack watermark check.
 563     cmp(in_nmethod ? sp : rfp, tmp);
 564     br(Assembler::HI, slow_path);
 565   } else {
 566     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 567   }
 568 }
 569 
 570 void MacroAssembler::rt_call(address dest, Register tmp) {
 571   CodeBlob *cb = CodeCache::find_blob(dest);
 572   if (cb) {
 573     far_call(RuntimeAddress(dest));
 574   } else {
 575     lea(tmp, RuntimeAddress(dest));
 576     blr(tmp);
 577   }
 578 }
 579 
 580 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 581   if (!Continuations::enabled()) return;
 582   Label done;
 583   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 584   cmp(sp, rscratch1);
 585   br(Assembler::LS, done);
 586   mov(rscratch1, sp); // we can't use sp as the source in str
 587   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 588   bind(done);
 589 }
 590 
 591 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 592   if (!Continuations::enabled()) return;
 593   Label done;
 594   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 595   cmp(sp, rscratch1);
 596   br(Assembler::LO, done);
 597   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 598   bind(done);
 599 }
 600 
 601 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 602   // we must set sp to zero to clear frame
 603   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 604 
 605   // must clear fp, so that compiled frames are not confused; it is
 606   // possible that we need it only for debugging
 607   if (clear_fp) {
 608     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 609   }
 610 
 611   // Always clear the pc because it could have been set by make_walkable()
 612   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 613 }
 614 
 615 // Calls to C land
 616 //
 617 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 618 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 619 // has to be reset to 0. This is required to allow proper stack traversal.
 620 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 621                                          Register last_java_fp,
 622                                          Register last_java_pc,
 623                                          Register scratch) {
 624 
 625   if (last_java_pc->is_valid()) {
 626       str(last_java_pc, Address(rthread,
 627                                 JavaThread::frame_anchor_offset()
 628                                 + JavaFrameAnchor::last_Java_pc_offset()));
 629     }
 630 
 631   // determine last_java_sp register
 632   if (last_java_sp == sp) {
 633     mov(scratch, sp);
 634     last_java_sp = scratch;
 635   } else if (!last_java_sp->is_valid()) {
 636     last_java_sp = esp;
 637   }
 638 
 639   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 640 
 641   // last_java_fp is optional
 642   if (last_java_fp->is_valid()) {
 643     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 644   }
 645 }
 646 
 647 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 648                                          Register last_java_fp,
 649                                          address  last_java_pc,
 650                                          Register scratch) {
 651   assert(last_java_pc != nullptr, "must provide a valid PC");
 652 
 653   adr(scratch, last_java_pc);
 654   str(scratch, Address(rthread,
 655                        JavaThread::frame_anchor_offset()
 656                        + JavaFrameAnchor::last_Java_pc_offset()));
 657 
 658   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 659 }
 660 
 661 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 662                                          Register last_java_fp,
 663                                          Label &L,
 664                                          Register scratch) {
 665   if (L.is_bound()) {
 666     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 667   } else {
 668     InstructionMark im(this);
 669     L.add_patch_at(code(), locator());
 670     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 671   }
 672 }
 673 
 674 static inline bool target_needs_far_branch(address addr) {
 675   // codecache size <= 128M
 676   if (!MacroAssembler::far_branches()) {
 677     return false;
 678   }
 679   // codecache size > 240M
 680   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 681     return true;
 682   }
 683   // codecache size: 128M..240M
 684   return !CodeCache::is_non_nmethod(addr);
 685 }
 686 
 687 void MacroAssembler::far_call(Address entry, Register tmp) {
 688   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 689   assert(CodeCache::find_blob(entry.target()) != nullptr,
 690          "destination of far call not found in code cache");
 691   assert(entry.rspec().type() == relocInfo::external_word_type
 692          || entry.rspec().type() == relocInfo::runtime_call_type
 693          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 694   if (target_needs_far_branch(entry.target())) {
 695     uint64_t offset;
 696     // We can use ADRP here because we know that the total size of
 697     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 698     adrp(tmp, entry, offset);
 699     add(tmp, tmp, offset);
 700     blr(tmp);
 701   } else {
 702     bl(entry);
 703   }
 704 }
 705 
 706 int MacroAssembler::far_jump(Address entry, Register tmp) {
 707   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 708   assert(CodeCache::find_blob(entry.target()) != nullptr,
 709          "destination of far call not found in code cache");
 710   assert(entry.rspec().type() == relocInfo::external_word_type
 711          || entry.rspec().type() == relocInfo::runtime_call_type
 712          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 713   address start = pc();
 714   if (target_needs_far_branch(entry.target())) {
 715     uint64_t offset;
 716     // We can use ADRP here because we know that the total size of
 717     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 718     adrp(tmp, entry, offset);
 719     add(tmp, tmp, offset);
 720     br(tmp);
 721   } else {
 722     b(entry);
 723   }
 724   return pc() - start;
 725 }
 726 
 727 void MacroAssembler::reserved_stack_check() {
 728     // testing if reserved zone needs to be enabled
 729     Label no_reserved_zone_enabling;
 730 
 731     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 732     cmp(sp, rscratch1);
 733     br(Assembler::LO, no_reserved_zone_enabling);
 734 
 735     enter();   // LR and FP are live.
 736     lea(rscratch1, CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone));
 737     mov(c_rarg0, rthread);
 738     blr(rscratch1);
 739     leave();
 740 
 741     // We have already removed our own frame.
 742     // throw_delayed_StackOverflowError will think that it's been
 743     // called by our caller.
 744     lea(rscratch1, RuntimeAddress(StubRoutines::throw_delayed_StackOverflowError_entry()));
 745     br(rscratch1);
 746     should_not_reach_here();
 747 
 748     bind(no_reserved_zone_enabling);
 749 }
 750 
 751 static void pass_arg0(MacroAssembler* masm, Register arg) {
 752   if (c_rarg0 != arg ) {
 753     masm->mov(c_rarg0, arg);
 754   }
 755 }
 756 
 757 static void pass_arg1(MacroAssembler* masm, Register arg) {
 758   if (c_rarg1 != arg ) {
 759     masm->mov(c_rarg1, arg);
 760   }
 761 }
 762 
 763 static void pass_arg2(MacroAssembler* masm, Register arg) {
 764   if (c_rarg2 != arg ) {
 765     masm->mov(c_rarg2, arg);
 766   }
 767 }
 768 
 769 static void pass_arg3(MacroAssembler* masm, Register arg) {
 770   if (c_rarg3 != arg ) {
 771     masm->mov(c_rarg3, arg);
 772   }
 773 }
 774 
 775 void MacroAssembler::call_VM_base(Register oop_result,
 776                                   Register java_thread,
 777                                   Register last_java_sp,
 778                                   address  entry_point,
 779                                   int      number_of_arguments,
 780                                   bool     check_exceptions) {
 781    // determine java_thread register
 782   if (!java_thread->is_valid()) {
 783     java_thread = rthread;
 784   }
 785 
 786   // determine last_java_sp register
 787   if (!last_java_sp->is_valid()) {
 788     last_java_sp = esp;
 789   }
 790 
 791   // debugging support
 792   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 793   assert(java_thread == rthread, "unexpected register");
 794 #ifdef ASSERT
 795   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 796   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 797 #endif // ASSERT
 798 
 799   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 800   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 801 
 802   // push java thread (becomes first argument of C function)
 803 
 804   mov(c_rarg0, java_thread);
 805 
 806   // set last Java frame before call
 807   assert(last_java_sp != rfp, "can't use rfp");
 808 
 809   Label l;
 810   set_last_Java_frame(last_java_sp, rfp, l, rscratch1);
 811 
 812   // do the call, remove parameters
 813   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 814 
 815   // lr could be poisoned with PAC signature during throw_pending_exception
 816   // if it was tail-call optimized by compiler, since lr is not callee-saved
 817   // reload it with proper value
 818   adr(lr, l);
 819 
 820   // reset last Java frame
 821   // Only interpreter should have to clear fp
 822   reset_last_Java_frame(true);
 823 
 824    // C++ interp handles this in the interpreter
 825   check_and_handle_popframe(java_thread);
 826   check_and_handle_earlyret(java_thread);
 827 
 828   if (check_exceptions) {
 829     // check for pending exceptions (java_thread is set upon return)
 830     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 831     Label ok;
 832     cbz(rscratch1, ok);
 833     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 834     br(rscratch1);
 835     bind(ok);
 836   }
 837 
 838   // get oop result if there is one and reset the value in the thread
 839   if (oop_result->is_valid()) {
 840     get_vm_result(oop_result, java_thread);
 841   }
 842 }
 843 
 844 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 845   call_VM_base(oop_result, noreg, noreg, entry_point, number_of_arguments, check_exceptions);
 846 }
 847 
 848 // Check the entry target is always reachable from any branch.
 849 static bool is_always_within_branch_range(Address entry) {
 850   const address target = entry.target();
 851 
 852   if (!CodeCache::contains(target)) {
 853     // We always use trampolines for callees outside CodeCache.
 854     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 855     return false;
 856   }
 857 
 858   if (!MacroAssembler::far_branches()) {
 859     return true;
 860   }
 861 
 862   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 863     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 864     // Non-compiled methods stay forever in CodeCache.
 865     // We check whether the longest possible branch is within the branch range.
 866     assert(CodeCache::find_blob(target) != nullptr &&
 867           !CodeCache::find_blob(target)->is_compiled(),
 868           "runtime call of compiled method");
 869     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 870     const address left_longest_branch_start = CodeCache::low_bound();
 871     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 872                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 873     return is_reachable;
 874   }
 875 
 876   return false;
 877 }
 878 
 879 // Maybe emit a call via a trampoline. If the code cache is small
 880 // trampolines won't be emitted.
 881 address MacroAssembler::trampoline_call(Address entry) {
 882   assert(entry.rspec().type() == relocInfo::runtime_call_type
 883          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 884          || entry.rspec().type() == relocInfo::static_call_type
 885          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 886 
 887   address target = entry.target();
 888 
 889   if (!is_always_within_branch_range(entry)) {
 890     if (!in_scratch_emit_size()) {
 891       // We don't want to emit a trampoline if C2 is generating dummy
 892       // code during its branch shortening phase.
 893       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 894         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 895         code()->share_trampoline_for(entry.target(), offset());
 896       } else {
 897         address stub = emit_trampoline_stub(offset(), target);
 898         if (stub == nullptr) {
 899           postcond(pc() == badAddress);
 900           return nullptr; // CodeCache is full
 901         }
 902       }
 903     }
 904     target = pc();
 905   }
 906 
 907   address call_pc = pc();
 908   relocate(entry.rspec());
 909   bl(target);
 910 
 911   postcond(pc() != badAddress);
 912   return call_pc;
 913 }
 914 
 915 // Emit a trampoline stub for a call to a target which is too far away.
 916 //
 917 // code sequences:
 918 //
 919 // call-site:
 920 //   branch-and-link to <destination> or <trampoline stub>
 921 //
 922 // Related trampoline stub for this call site in the stub section:
 923 //   load the call target from the constant pool
 924 //   branch (LR still points to the call site above)
 925 
 926 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 927                                              address dest) {
 928   // Max stub size: alignment nop, TrampolineStub.
 929   address stub = start_a_stub(max_trampoline_stub_size());
 930   if (stub == nullptr) {
 931     return nullptr;  // CodeBuffer::expand failed
 932   }
 933 
 934   // Create a trampoline stub relocation which relates this trampoline stub
 935   // with the call instruction at insts_call_instruction_offset in the
 936   // instructions code-section.
 937   align(wordSize);
 938   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 939                                             + insts_call_instruction_offset));
 940   const int stub_start_offset = offset();
 941 
 942   // Now, create the trampoline stub's code:
 943   // - load the call
 944   // - call
 945   Label target;
 946   ldr(rscratch1, target);
 947   br(rscratch1);
 948   bind(target);
 949   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 950          "should be");
 951   emit_int64((int64_t)dest);
 952 
 953   const address stub_start_addr = addr_at(stub_start_offset);
 954 
 955   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 956 
 957   end_a_stub();
 958   return stub_start_addr;
 959 }
 960 
 961 int MacroAssembler::max_trampoline_stub_size() {
 962   // Max stub size: alignment nop, TrampolineStub.
 963   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 964 }
 965 
 966 void MacroAssembler::emit_static_call_stub() {
 967   // CompiledDirectStaticCall::set_to_interpreted knows the
 968   // exact layout of this stub.
 969 
 970   isb();
 971   mov_metadata(rmethod, nullptr);
 972 
 973   // Jump to the entry point of the c2i stub.
 974   movptr(rscratch1, 0);
 975   br(rscratch1);
 976 }
 977 
 978 int MacroAssembler::static_call_stub_size() {
 979   // isb; movk; movz; movz; movk; movz; movz; br
 980   return 8 * NativeInstruction::instruction_size;
 981 }
 982 
 983 void MacroAssembler::c2bool(Register x) {
 984   // implements x == 0 ? 0 : 1
 985   // note: must only look at least-significant byte of x
 986   //       since C-style booleans are stored in one byte
 987   //       only! (was bug)
 988   tst(x, 0xff);
 989   cset(x, Assembler::NE);
 990 }
 991 
 992 address MacroAssembler::ic_call(address entry, jint method_index) {
 993   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 994   // address const_ptr = long_constant((jlong)Universe::non_oop_word());
 995   // uintptr_t offset;
 996   // ldr_constant(rscratch2, const_ptr);
 997   movptr(rscratch2, (uintptr_t)Universe::non_oop_word());
 998   return trampoline_call(Address(entry, rh));
 999 }
1000 
1001 // Implementation of call_VM versions
1002 
1003 void MacroAssembler::call_VM(Register oop_result,
1004                              address entry_point,
1005                              bool check_exceptions) {
1006   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1007 }
1008 
1009 void MacroAssembler::call_VM(Register oop_result,
1010                              address entry_point,
1011                              Register arg_1,
1012                              bool check_exceptions) {
1013   pass_arg1(this, arg_1);
1014   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1015 }
1016 
1017 void MacroAssembler::call_VM(Register oop_result,
1018                              address entry_point,
1019                              Register arg_1,
1020                              Register arg_2,
1021                              bool check_exceptions) {
1022   assert(arg_1 != c_rarg2, "smashed arg");
1023   pass_arg2(this, arg_2);
1024   pass_arg1(this, arg_1);
1025   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1026 }
1027 
1028 void MacroAssembler::call_VM(Register oop_result,
1029                              address entry_point,
1030                              Register arg_1,
1031                              Register arg_2,
1032                              Register arg_3,
1033                              bool check_exceptions) {
1034   assert(arg_1 != c_rarg3, "smashed arg");
1035   assert(arg_2 != c_rarg3, "smashed arg");
1036   pass_arg3(this, arg_3);
1037 
1038   assert(arg_1 != c_rarg2, "smashed arg");
1039   pass_arg2(this, arg_2);
1040 
1041   pass_arg1(this, arg_1);
1042   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1043 }
1044 
1045 void MacroAssembler::call_VM(Register oop_result,
1046                              Register last_java_sp,
1047                              address entry_point,
1048                              int number_of_arguments,
1049                              bool check_exceptions) {
1050   call_VM_base(oop_result, rthread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1051 }
1052 
1053 void MacroAssembler::call_VM(Register oop_result,
1054                              Register last_java_sp,
1055                              address entry_point,
1056                              Register arg_1,
1057                              bool check_exceptions) {
1058   pass_arg1(this, arg_1);
1059   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1060 }
1061 
1062 void MacroAssembler::call_VM(Register oop_result,
1063                              Register last_java_sp,
1064                              address entry_point,
1065                              Register arg_1,
1066                              Register arg_2,
1067                              bool check_exceptions) {
1068 
1069   assert(arg_1 != c_rarg2, "smashed arg");
1070   pass_arg2(this, arg_2);
1071   pass_arg1(this, arg_1);
1072   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1073 }
1074 
1075 void MacroAssembler::call_VM(Register oop_result,
1076                              Register last_java_sp,
1077                              address entry_point,
1078                              Register arg_1,
1079                              Register arg_2,
1080                              Register arg_3,
1081                              bool check_exceptions) {
1082   assert(arg_1 != c_rarg3, "smashed arg");
1083   assert(arg_2 != c_rarg3, "smashed arg");
1084   pass_arg3(this, arg_3);
1085   assert(arg_1 != c_rarg2, "smashed arg");
1086   pass_arg2(this, arg_2);
1087   pass_arg1(this, arg_1);
1088   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1089 }
1090 
1091 
1092 void MacroAssembler::get_vm_result(Register oop_result, Register java_thread) {
1093   ldr(oop_result, Address(java_thread, JavaThread::vm_result_offset()));
1094   str(zr, Address(java_thread, JavaThread::vm_result_offset()));
1095   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1096 }
1097 
1098 void MacroAssembler::get_vm_result_2(Register metadata_result, Register java_thread) {
1099   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_2_offset()));
1100   str(zr, Address(java_thread, JavaThread::vm_result_2_offset()));
1101 }
1102 
1103 void MacroAssembler::align(int modulus) {
1104   while (offset() % modulus != 0) nop();
1105 }
1106 
1107 void MacroAssembler::post_call_nop() {
1108   if (!Continuations::enabled()) {
1109     return;
1110   }
1111   InstructionMark im(this);
1112   relocate(post_call_nop_Relocation::spec());
1113   InlineSkippedInstructionsCounter skipCounter(this);
1114   nop();
1115   movk(zr, 0);
1116   movk(zr, 0);
1117 }
1118 
1119 // these are no-ops overridden by InterpreterMacroAssembler
1120 
1121 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1122 
1123 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1124 
1125 // Look up the method for a megamorphic invokeinterface call.
1126 // The target method is determined by <intf_klass, itable_index>.
1127 // The receiver klass is in recv_klass.
1128 // On success, the result will be in method_result, and execution falls through.
1129 // On failure, execution transfers to the given label.
1130 void MacroAssembler::lookup_interface_method(Register recv_klass,
1131                                              Register intf_klass,
1132                                              RegisterOrConstant itable_index,
1133                                              Register method_result,
1134                                              Register scan_temp,
1135                                              Label& L_no_such_interface,
1136                          bool return_method) {
1137   assert_different_registers(recv_klass, intf_klass, scan_temp);
1138   assert_different_registers(method_result, intf_klass, scan_temp);
1139   assert(recv_klass != method_result || !return_method,
1140      "recv_klass can be destroyed when method isn't needed");
1141   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1142          "caller must use same register for non-constant itable index as for method");
1143 
1144   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1145   int vtable_base = in_bytes(Klass::vtable_start_offset());
1146   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1147   int scan_step   = itableOffsetEntry::size() * wordSize;
1148   int vte_size    = vtableEntry::size_in_bytes();
1149   assert(vte_size == wordSize, "else adjust times_vte_scale");
1150 
1151   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1152 
1153   // %%% Could store the aligned, prescaled offset in the klassoop.
1154   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1155   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1156   add(scan_temp, scan_temp, vtable_base);
1157 
1158   if (return_method) {
1159     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1160     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1161     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1162     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1163     if (itentry_off)
1164       add(recv_klass, recv_klass, itentry_off);
1165   }
1166 
1167   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1168   //   if (scan->interface() == intf) {
1169   //     result = (klass + scan->offset() + itable_index);
1170   //   }
1171   // }
1172   Label search, found_method;
1173 
1174   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1175   cmp(intf_klass, method_result);
1176   br(Assembler::EQ, found_method);
1177   bind(search);
1178   // Check that the previous entry is non-null.  A null entry means that
1179   // the receiver class doesn't implement the interface, and wasn't the
1180   // same as when the caller was compiled.
1181   cbz(method_result, L_no_such_interface);
1182   if (itableOffsetEntry::interface_offset() != 0) {
1183     add(scan_temp, scan_temp, scan_step);
1184     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1185   } else {
1186     ldr(method_result, Address(pre(scan_temp, scan_step)));
1187   }
1188   cmp(intf_klass, method_result);
1189   br(Assembler::NE, search);
1190 
1191   bind(found_method);
1192 
1193   // Got a hit.
1194   if (return_method) {
1195     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1196     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1197   }
1198 }
1199 
1200 // virtual method calling
1201 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1202                                            RegisterOrConstant vtable_index,
1203                                            Register method_result) {
1204   assert(vtableEntry::size() * wordSize == 8,
1205          "adjust the scaling in the code below");
1206   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1207 
1208   if (vtable_index.is_register()) {
1209     lea(method_result, Address(recv_klass,
1210                                vtable_index.as_register(),
1211                                Address::lsl(LogBytesPerWord)));
1212     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1213   } else {
1214     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1215     ldr(method_result,
1216         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1217   }
1218 }
1219 
1220 void MacroAssembler::check_klass_subtype(Register sub_klass,
1221                            Register super_klass,
1222                            Register temp_reg,
1223                            Label& L_success) {
1224   Label L_failure;
1225   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1226   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1227   bind(L_failure);
1228 }
1229 
1230 
1231 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1232                                                    Register super_klass,
1233                                                    Register temp_reg,
1234                                                    Label* L_success,
1235                                                    Label* L_failure,
1236                                                    Label* L_slow_path,
1237                                         RegisterOrConstant super_check_offset) {
1238   assert_different_registers(sub_klass, super_klass, temp_reg);
1239   bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
1240   if (super_check_offset.is_register()) {
1241     assert_different_registers(sub_klass, super_klass,
1242                                super_check_offset.as_register());
1243   } else if (must_load_sco) {
1244     assert(temp_reg != noreg, "supply either a temp or a register offset");
1245   }
1246 
1247   Label L_fallthrough;
1248   int label_nulls = 0;
1249   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1250   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1251   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1252   assert(label_nulls <= 1, "at most one null in the batch");
1253 
1254   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1255   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1256   Address super_check_offset_addr(super_klass, sco_offset);
1257 
1258   // Hacked jmp, which may only be used just before L_fallthrough.
1259 #define final_jmp(label)                                                \
1260   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1261   else                            b(label)                /*omit semi*/
1262 
1263   // If the pointers are equal, we are done (e.g., String[] elements).
1264   // This self-check enables sharing of secondary supertype arrays among
1265   // non-primary types such as array-of-interface.  Otherwise, each such
1266   // type would need its own customized SSA.
1267   // We move this check to the front of the fast path because many
1268   // type checks are in fact trivially successful in this manner,
1269   // so we get a nicely predicted branch right at the start of the check.
1270   cmp(sub_klass, super_klass);
1271   br(Assembler::EQ, *L_success);
1272 
1273   // Check the supertype display:
1274   if (must_load_sco) {
1275     ldrw(temp_reg, super_check_offset_addr);
1276     super_check_offset = RegisterOrConstant(temp_reg);
1277   }
1278   Address super_check_addr(sub_klass, super_check_offset);
1279   ldr(rscratch1, super_check_addr);
1280   cmp(super_klass, rscratch1); // load displayed supertype
1281 
1282   // This check has worked decisively for primary supers.
1283   // Secondary supers are sought in the super_cache ('super_cache_addr').
1284   // (Secondary supers are interfaces and very deeply nested subtypes.)
1285   // This works in the same check above because of a tricky aliasing
1286   // between the super_cache and the primary super display elements.
1287   // (The 'super_check_addr' can address either, as the case requires.)
1288   // Note that the cache is updated below if it does not help us find
1289   // what we need immediately.
1290   // So if it was a primary super, we can just fail immediately.
1291   // Otherwise, it's the slow path for us (no success at this point).
1292 
1293   if (super_check_offset.is_register()) {
1294     br(Assembler::EQ, *L_success);
1295     subs(zr, super_check_offset.as_register(), sc_offset);
1296     if (L_failure == &L_fallthrough) {
1297       br(Assembler::EQ, *L_slow_path);
1298     } else {
1299       br(Assembler::NE, *L_failure);
1300       final_jmp(*L_slow_path);
1301     }
1302   } else if (super_check_offset.as_constant() == sc_offset) {
1303     // Need a slow path; fast failure is impossible.
1304     if (L_slow_path == &L_fallthrough) {
1305       br(Assembler::EQ, *L_success);
1306     } else {
1307       br(Assembler::NE, *L_slow_path);
1308       final_jmp(*L_success);
1309     }
1310   } else {
1311     // No slow path; it's a fast decision.
1312     if (L_failure == &L_fallthrough) {
1313       br(Assembler::EQ, *L_success);
1314     } else {
1315       br(Assembler::NE, *L_failure);
1316       final_jmp(*L_success);
1317     }
1318   }
1319 
1320   bind(L_fallthrough);
1321 
1322 #undef final_jmp
1323 }
1324 
1325 // These two are taken from x86, but they look generally useful
1326 
1327 // scans count pointer sized words at [addr] for occurrence of value,
1328 // generic
1329 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1330                                 Register scratch) {
1331   Label Lloop, Lexit;
1332   cbz(count, Lexit);
1333   bind(Lloop);
1334   ldr(scratch, post(addr, wordSize));
1335   cmp(value, scratch);
1336   br(EQ, Lexit);
1337   sub(count, count, 1);
1338   cbnz(count, Lloop);
1339   bind(Lexit);
1340 }
1341 
1342 // scans count 4 byte words at [addr] for occurrence of value,
1343 // generic
1344 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1345                                 Register scratch) {
1346   Label Lloop, Lexit;
1347   cbz(count, Lexit);
1348   bind(Lloop);
1349   ldrw(scratch, post(addr, wordSize));
1350   cmpw(value, scratch);
1351   br(EQ, Lexit);
1352   sub(count, count, 1);
1353   cbnz(count, Lloop);
1354   bind(Lexit);
1355 }
1356 
1357 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1358                                                    Register super_klass,
1359                                                    Register temp_reg,
1360                                                    Register temp2_reg,
1361                                                    Label* L_success,
1362                                                    Label* L_failure,
1363                                                    bool set_cond_codes) {
1364   assert_different_registers(sub_klass, super_klass, temp_reg);
1365   if (temp2_reg != noreg)
1366     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1367 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1368 
1369   Label L_fallthrough;
1370   int label_nulls = 0;
1371   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1372   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1373   assert(label_nulls <= 1, "at most one null in the batch");
1374 
1375   // a couple of useful fields in sub_klass:
1376   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1377   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1378   Address secondary_supers_addr(sub_klass, ss_offset);
1379   Address super_cache_addr(     sub_klass, sc_offset);
1380 
1381   BLOCK_COMMENT("check_klass_subtype_slow_path");
1382 
1383   // Do a linear scan of the secondary super-klass chain.
1384   // This code is rarely used, so simplicity is a virtue here.
1385   // The repne_scan instruction uses fixed registers, which we must spill.
1386   // Don't worry too much about pre-existing connections with the input regs.
1387 
1388   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1389   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1390 
1391   RegSet pushed_registers;
1392   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1393   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1394 
1395   if (super_klass != r0) {
1396     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1397   }
1398 
1399   push(pushed_registers, sp);
1400 
1401   // Get super_klass value into r0 (even if it was in r5 or r2).
1402   if (super_klass != r0) {
1403     mov(r0, super_klass);
1404   }
1405 
1406 #ifndef PRODUCT
1407   mov(rscratch2, (address)&SharedRuntime::_partial_subtype_ctr);
1408   Address pst_counter_addr(rscratch2);
1409   ldr(rscratch1, pst_counter_addr);
1410   add(rscratch1, rscratch1, 1);
1411   str(rscratch1, pst_counter_addr);
1412 #endif //PRODUCT
1413 
1414   // We will consult the secondary-super array.
1415   ldr(r5, secondary_supers_addr);
1416   // Load the array length.
1417   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1418   // Skip to start of data.
1419   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1420 
1421   cmp(sp, zr); // Clear Z flag; SP is never zero
1422   // Scan R2 words at [R5] for an occurrence of R0.
1423   // Set NZ/Z based on last compare.
1424   repne_scan(r5, r0, r2, rscratch1);
1425 
1426   // Unspill the temp. registers:
1427   pop(pushed_registers, sp);
1428 
1429   br(Assembler::NE, *L_failure);
1430 
1431   // Success.  Cache the super we found and proceed in triumph.
1432   str(super_klass, super_cache_addr);
1433 
1434   if (L_success != &L_fallthrough) {
1435     b(*L_success);
1436   }
1437 
1438 #undef IS_A_TEMP
1439 
1440   bind(L_fallthrough);
1441 }
1442 
1443 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
1444   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
1445   assert_different_registers(klass, rthread, scratch);
1446 
1447   Label L_fallthrough, L_tmp;
1448   if (L_fast_path == nullptr) {
1449     L_fast_path = &L_fallthrough;
1450   } else if (L_slow_path == nullptr) {
1451     L_slow_path = &L_fallthrough;
1452   }
1453   // Fast path check: class is fully initialized
1454   ldrb(scratch, Address(klass, InstanceKlass::init_state_offset()));
1455   subs(zr, scratch, InstanceKlass::fully_initialized);
1456   br(Assembler::EQ, *L_fast_path);
1457 
1458   // Fast path check: current thread is initializer thread
1459   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
1460   cmp(rthread, scratch);
1461 
1462   if (L_slow_path == &L_fallthrough) {
1463     br(Assembler::EQ, *L_fast_path);
1464     bind(*L_slow_path);
1465   } else if (L_fast_path == &L_fallthrough) {
1466     br(Assembler::NE, *L_slow_path);
1467     bind(*L_fast_path);
1468   } else {
1469     Unimplemented();
1470   }
1471 }
1472 
1473 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
1474   if (!VerifyOops) return;
1475 
1476   // Pass register number to verify_oop_subroutine
1477   const char* b = nullptr;
1478   {
1479     ResourceMark rm;
1480     stringStream ss;
1481     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
1482     b = code_string(ss.as_string());
1483   }
1484   BLOCK_COMMENT("verify_oop {");
1485 
1486   strip_return_address(); // This might happen within a stack frame.
1487   protect_return_address();
1488   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
1489   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
1490 
1491   mov(r0, reg);
1492   movptr(rscratch1, (uintptr_t)(address)b);
1493 
1494   // call indirectly to solve generation ordering problem
1495   lea(rscratch2, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
1496   ldr(rscratch2, Address(rscratch2));
1497   blr(rscratch2);
1498 
1499   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
1500   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
1501   authenticate_return_address();
1502 
1503   BLOCK_COMMENT("} verify_oop");
1504 }
1505 
1506 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
1507   if (!VerifyOops) return;
1508 
1509   const char* b = nullptr;
1510   {
1511     ResourceMark rm;
1512     stringStream ss;
1513     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
1514     b = code_string(ss.as_string());
1515   }
1516   BLOCK_COMMENT("verify_oop_addr {");
1517 
1518   strip_return_address(); // This might happen within a stack frame.
1519   protect_return_address();
1520   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
1521   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
1522 
1523   // addr may contain sp so we will have to adjust it based on the
1524   // pushes that we just did.
1525   if (addr.uses(sp)) {
1526     lea(r0, addr);
1527     ldr(r0, Address(r0, 4 * wordSize));
1528   } else {
1529     ldr(r0, addr);
1530   }
1531   movptr(rscratch1, (uintptr_t)(address)b);
1532 
1533   // call indirectly to solve generation ordering problem
1534   lea(rscratch2, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
1535   ldr(rscratch2, Address(rscratch2));
1536   blr(rscratch2);
1537 
1538   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
1539   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
1540   authenticate_return_address();
1541 
1542   BLOCK_COMMENT("} verify_oop_addr");
1543 }
1544 
1545 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
1546                                          int extra_slot_offset) {
1547   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
1548   int stackElementSize = Interpreter::stackElementSize;
1549   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
1550 #ifdef ASSERT
1551   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
1552   assert(offset1 - offset == stackElementSize, "correct arithmetic");
1553 #endif
1554   if (arg_slot.is_constant()) {
1555     return Address(esp, arg_slot.as_constant() * stackElementSize
1556                    + offset);
1557   } else {
1558     add(rscratch1, esp, arg_slot.as_register(),
1559         ext::uxtx, exact_log2(stackElementSize));
1560     return Address(rscratch1, offset);
1561   }
1562 }
1563 
1564 void MacroAssembler::call_VM_leaf_base(address entry_point,
1565                                        int number_of_arguments,
1566                                        Label *retaddr) {
1567   Label E, L;
1568 
1569   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
1570 
1571   mov(rscratch1, entry_point);
1572   blr(rscratch1);
1573   if (retaddr)
1574     bind(*retaddr);
1575 
1576   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
1577 }
1578 
1579 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
1580   call_VM_leaf_base(entry_point, number_of_arguments);
1581 }
1582 
1583 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
1584   pass_arg0(this, arg_0);
1585   call_VM_leaf_base(entry_point, 1);
1586 }
1587 
1588 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1589   pass_arg0(this, arg_0);
1590   pass_arg1(this, arg_1);
1591   call_VM_leaf_base(entry_point, 2);
1592 }
1593 
1594 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
1595                                   Register arg_1, Register arg_2) {
1596   pass_arg0(this, arg_0);
1597   pass_arg1(this, arg_1);
1598   pass_arg2(this, arg_2);
1599   call_VM_leaf_base(entry_point, 3);
1600 }
1601 
1602 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
1603   pass_arg0(this, arg_0);
1604   MacroAssembler::call_VM_leaf_base(entry_point, 1);
1605 }
1606 
1607 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1608 
1609   assert(arg_0 != c_rarg1, "smashed arg");
1610   pass_arg1(this, arg_1);
1611   pass_arg0(this, arg_0);
1612   MacroAssembler::call_VM_leaf_base(entry_point, 2);
1613 }
1614 
1615 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
1616   assert(arg_0 != c_rarg2, "smashed arg");
1617   assert(arg_1 != c_rarg2, "smashed arg");
1618   pass_arg2(this, arg_2);
1619   assert(arg_0 != c_rarg1, "smashed arg");
1620   pass_arg1(this, arg_1);
1621   pass_arg0(this, arg_0);
1622   MacroAssembler::call_VM_leaf_base(entry_point, 3);
1623 }
1624 
1625 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
1626   assert(arg_0 != c_rarg3, "smashed arg");
1627   assert(arg_1 != c_rarg3, "smashed arg");
1628   assert(arg_2 != c_rarg3, "smashed arg");
1629   pass_arg3(this, arg_3);
1630   assert(arg_0 != c_rarg2, "smashed arg");
1631   assert(arg_1 != c_rarg2, "smashed arg");
1632   pass_arg2(this, arg_2);
1633   assert(arg_0 != c_rarg1, "smashed arg");
1634   pass_arg1(this, arg_1);
1635   pass_arg0(this, arg_0);
1636   MacroAssembler::call_VM_leaf_base(entry_point, 4);
1637 }
1638 
1639 void MacroAssembler::null_check(Register reg, int offset) {
1640   if (needs_explicit_null_check(offset)) {
1641     // provoke OS null exception if reg is null by
1642     // accessing M[reg] w/o changing any registers
1643     // NOTE: this is plenty to provoke a segv
1644     ldr(zr, Address(reg));
1645   } else {
1646     // nothing to do, (later) access of M[reg + offset]
1647     // will provoke OS null exception if reg is null
1648   }
1649 }
1650 
1651 // MacroAssembler protected routines needed to implement
1652 // public methods
1653 
1654 void MacroAssembler::mov(Register r, Address dest) {
1655   code_section()->relocate(pc(), dest.rspec());
1656   uint64_t imm64 = (uint64_t)dest.target();
1657   movptr(r, imm64);
1658 }
1659 
1660 // Move a constant pointer into r.  In AArch64 mode the virtual
1661 // address space is 48 bits in size, so we only need three
1662 // instructions to create a patchable instruction sequence that can
1663 // reach anywhere.
1664 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
1665 #ifndef PRODUCT
1666   {
1667     char buffer[64];
1668     snprintf(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
1669     block_comment(buffer);
1670   }
1671 #endif
1672   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
1673   movz(r, imm64 & 0xffff);
1674   imm64 >>= 16;
1675   movk(r, imm64 & 0xffff, 16);
1676   imm64 >>= 16;
1677   movk(r, imm64 & 0xffff, 32);
1678 }
1679 
1680 // Macro to mov replicated immediate to vector register.
1681 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
1682 //        the upper 56/48/32 bits must be zeros for B/H/S type.
1683 // Vd will get the following values for different arrangements in T
1684 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
1685 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
1686 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
1687 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
1688 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
1689 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
1690 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
1691 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
1692 // Clobbers rscratch1
1693 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
1694   assert(T != T1Q, "unsupported");
1695   if (T == T1D || T == T2D) {
1696     int imm = operand_valid_for_movi_immediate(imm64, T);
1697     if (-1 != imm) {
1698       movi(Vd, T, imm);
1699     } else {
1700       mov(rscratch1, imm64);
1701       dup(Vd, T, rscratch1);
1702     }
1703     return;
1704   }
1705 
1706 #ifdef ASSERT
1707   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
1708   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
1709   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
1710 #endif
1711   int shift = operand_valid_for_movi_immediate(imm64, T);
1712   uint32_t imm32 = imm64 & 0xffffffffULL;
1713   if (shift >= 0) {
1714     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
1715   } else {
1716     movw(rscratch1, imm32);
1717     dup(Vd, T, rscratch1);
1718   }
1719 }
1720 
1721 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
1722 {
1723 #ifndef PRODUCT
1724   {
1725     char buffer[64];
1726     snprintf(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
1727     block_comment(buffer);
1728   }
1729 #endif
1730   if (operand_valid_for_logical_immediate(false, imm64)) {
1731     orr(dst, zr, imm64);
1732   } else {
1733     // we can use a combination of MOVZ or MOVN with
1734     // MOVK to build up the constant
1735     uint64_t imm_h[4];
1736     int zero_count = 0;
1737     int neg_count = 0;
1738     int i;
1739     for (i = 0; i < 4; i++) {
1740       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
1741       if (imm_h[i] == 0) {
1742         zero_count++;
1743       } else if (imm_h[i] == 0xffffL) {
1744         neg_count++;
1745       }
1746     }
1747     if (zero_count == 4) {
1748       // one MOVZ will do
1749       movz(dst, 0);
1750     } else if (neg_count == 4) {
1751       // one MOVN will do
1752       movn(dst, 0);
1753     } else if (zero_count == 3) {
1754       for (i = 0; i < 4; i++) {
1755         if (imm_h[i] != 0L) {
1756           movz(dst, (uint32_t)imm_h[i], (i << 4));
1757           break;
1758         }
1759       }
1760     } else if (neg_count == 3) {
1761       // one MOVN will do
1762       for (int i = 0; i < 4; i++) {
1763         if (imm_h[i] != 0xffffL) {
1764           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
1765           break;
1766         }
1767       }
1768     } else if (zero_count == 2) {
1769       // one MOVZ and one MOVK will do
1770       for (i = 0; i < 3; i++) {
1771         if (imm_h[i] != 0L) {
1772           movz(dst, (uint32_t)imm_h[i], (i << 4));
1773           i++;
1774           break;
1775         }
1776       }
1777       for (;i < 4; i++) {
1778         if (imm_h[i] != 0L) {
1779           movk(dst, (uint32_t)imm_h[i], (i << 4));
1780         }
1781       }
1782     } else if (neg_count == 2) {
1783       // one MOVN and one MOVK will do
1784       for (i = 0; i < 4; i++) {
1785         if (imm_h[i] != 0xffffL) {
1786           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
1787           i++;
1788           break;
1789         }
1790       }
1791       for (;i < 4; i++) {
1792         if (imm_h[i] != 0xffffL) {
1793           movk(dst, (uint32_t)imm_h[i], (i << 4));
1794         }
1795       }
1796     } else if (zero_count == 1) {
1797       // one MOVZ and two MOVKs will do
1798       for (i = 0; i < 4; i++) {
1799         if (imm_h[i] != 0L) {
1800           movz(dst, (uint32_t)imm_h[i], (i << 4));
1801           i++;
1802           break;
1803         }
1804       }
1805       for (;i < 4; i++) {
1806         if (imm_h[i] != 0x0L) {
1807           movk(dst, (uint32_t)imm_h[i], (i << 4));
1808         }
1809       }
1810     } else if (neg_count == 1) {
1811       // one MOVN and two MOVKs will do
1812       for (i = 0; i < 4; i++) {
1813         if (imm_h[i] != 0xffffL) {
1814           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
1815           i++;
1816           break;
1817         }
1818       }
1819       for (;i < 4; i++) {
1820         if (imm_h[i] != 0xffffL) {
1821           movk(dst, (uint32_t)imm_h[i], (i << 4));
1822         }
1823       }
1824     } else {
1825       // use a MOVZ and 3 MOVKs (makes it easier to debug)
1826       movz(dst, (uint32_t)imm_h[0], 0);
1827       for (i = 1; i < 4; i++) {
1828         movk(dst, (uint32_t)imm_h[i], (i << 4));
1829       }
1830     }
1831   }
1832 }
1833 
1834 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
1835 {
1836 #ifndef PRODUCT
1837     {
1838       char buffer[64];
1839       snprintf(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
1840       block_comment(buffer);
1841     }
1842 #endif
1843   if (operand_valid_for_logical_immediate(true, imm32)) {
1844     orrw(dst, zr, imm32);
1845   } else {
1846     // we can use MOVZ, MOVN or two calls to MOVK to build up the
1847     // constant
1848     uint32_t imm_h[2];
1849     imm_h[0] = imm32 & 0xffff;
1850     imm_h[1] = ((imm32 >> 16) & 0xffff);
1851     if (imm_h[0] == 0) {
1852       movzw(dst, imm_h[1], 16);
1853     } else if (imm_h[0] == 0xffff) {
1854       movnw(dst, imm_h[1] ^ 0xffff, 16);
1855     } else if (imm_h[1] == 0) {
1856       movzw(dst, imm_h[0], 0);
1857     } else if (imm_h[1] == 0xffff) {
1858       movnw(dst, imm_h[0] ^ 0xffff, 0);
1859     } else {
1860       // use a MOVZ and MOVK (makes it easier to debug)
1861       movzw(dst, imm_h[0], 0);
1862       movkw(dst, imm_h[1], 16);
1863     }
1864   }
1865 }
1866 
1867 // Form an address from base + offset in Rd.  Rd may or may
1868 // not actually be used: you must use the Address that is returned.
1869 // It is up to you to ensure that the shift provided matches the size
1870 // of your data.
1871 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
1872   if (Address::offset_ok_for_immed(byte_offset, shift))
1873     // It fits; no need for any heroics
1874     return Address(base, byte_offset);
1875 
1876   // Don't do anything clever with negative or misaligned offsets
1877   unsigned mask = (1 << shift) - 1;
1878   if (byte_offset < 0 || byte_offset & mask) {
1879     mov(Rd, byte_offset);
1880     add(Rd, base, Rd);
1881     return Address(Rd);
1882   }
1883 
1884   // See if we can do this with two 12-bit offsets
1885   {
1886     uint64_t word_offset = byte_offset >> shift;
1887     uint64_t masked_offset = word_offset & 0xfff000;
1888     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
1889         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
1890       add(Rd, base, masked_offset << shift);
1891       word_offset -= masked_offset;
1892       return Address(Rd, word_offset << shift);
1893     }
1894   }
1895 
1896   // Do it the hard way
1897   mov(Rd, byte_offset);
1898   add(Rd, base, Rd);
1899   return Address(Rd);
1900 }
1901 
1902 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
1903                                     bool want_remainder, Register scratch)
1904 {
1905   // Full implementation of Java idiv and irem.  The function
1906   // returns the (pc) offset of the div instruction - may be needed
1907   // for implicit exceptions.
1908   //
1909   // constraint : ra/rb =/= scratch
1910   //         normal case
1911   //
1912   // input : ra: dividend
1913   //         rb: divisor
1914   //
1915   // result: either
1916   //         quotient  (= ra idiv rb)
1917   //         remainder (= ra irem rb)
1918 
1919   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
1920 
1921   int idivl_offset = offset();
1922   if (! want_remainder) {
1923     sdivw(result, ra, rb);
1924   } else {
1925     sdivw(scratch, ra, rb);
1926     Assembler::msubw(result, scratch, rb, ra);
1927   }
1928 
1929   return idivl_offset;
1930 }
1931 
1932 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
1933                                     bool want_remainder, Register scratch)
1934 {
1935   // Full implementation of Java ldiv and lrem.  The function
1936   // returns the (pc) offset of the div instruction - may be needed
1937   // for implicit exceptions.
1938   //
1939   // constraint : ra/rb =/= scratch
1940   //         normal case
1941   //
1942   // input : ra: dividend
1943   //         rb: divisor
1944   //
1945   // result: either
1946   //         quotient  (= ra idiv rb)
1947   //         remainder (= ra irem rb)
1948 
1949   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
1950 
1951   int idivq_offset = offset();
1952   if (! want_remainder) {
1953     sdiv(result, ra, rb);
1954   } else {
1955     sdiv(scratch, ra, rb);
1956     Assembler::msub(result, scratch, rb, ra);
1957   }
1958 
1959   return idivq_offset;
1960 }
1961 
1962 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
1963   address prev = pc() - NativeMembar::instruction_size;
1964   address last = code()->last_insn();
1965   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
1966     NativeMembar *bar = NativeMembar_at(prev);
1967     // We are merging two memory barrier instructions.  On AArch64 we
1968     // can do this simply by ORing them together.
1969     bar->set_kind(bar->get_kind() | order_constraint);
1970     BLOCK_COMMENT("merged membar");
1971   } else {
1972     code()->set_last_insn(pc());
1973     dmb(Assembler::barrier(order_constraint));
1974   }
1975 }
1976 
1977 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
1978   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
1979     merge_ldst(rt, adr, size_in_bytes, is_store);
1980     code()->clear_last_insn();
1981     return true;
1982   } else {
1983     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
1984     const uint64_t mask = size_in_bytes - 1;
1985     if (adr.getMode() == Address::base_plus_offset &&
1986         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
1987       code()->set_last_insn(pc());
1988     }
1989     return false;
1990   }
1991 }
1992 
1993 void MacroAssembler::ldr(Register Rx, const Address &adr) {
1994   // We always try to merge two adjacent loads into one ldp.
1995   if (!try_merge_ldst(Rx, adr, 8, false)) {
1996     Assembler::ldr(Rx, adr);
1997   }
1998 }
1999 
2000 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2001   // We always try to merge two adjacent loads into one ldp.
2002   if (!try_merge_ldst(Rw, adr, 4, false)) {
2003     Assembler::ldrw(Rw, adr);
2004   }
2005 }
2006 
2007 void MacroAssembler::str(Register Rx, const Address &adr) {
2008   // We always try to merge two adjacent stores into one stp.
2009   if (!try_merge_ldst(Rx, adr, 8, true)) {
2010     Assembler::str(Rx, adr);
2011   }
2012 }
2013 
2014 void MacroAssembler::strw(Register Rw, const Address &adr) {
2015   // We always try to merge two adjacent stores into one stp.
2016   if (!try_merge_ldst(Rw, adr, 4, true)) {
2017     Assembler::strw(Rw, adr);
2018   }
2019 }
2020 
2021 // MacroAssembler routines found actually to be needed
2022 
2023 void MacroAssembler::push(Register src)
2024 {
2025   str(src, Address(pre(esp, -1 * wordSize)));
2026 }
2027 
2028 void MacroAssembler::pop(Register dst)
2029 {
2030   ldr(dst, Address(post(esp, 1 * wordSize)));
2031 }
2032 
2033 // Note: load_unsigned_short used to be called load_unsigned_word.
2034 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2035   int off = offset();
2036   ldrh(dst, src);
2037   return off;
2038 }
2039 
2040 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2041   int off = offset();
2042   ldrb(dst, src);
2043   return off;
2044 }
2045 
2046 int MacroAssembler::load_signed_short(Register dst, Address src) {
2047   int off = offset();
2048   ldrsh(dst, src);
2049   return off;
2050 }
2051 
2052 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2053   int off = offset();
2054   ldrsb(dst, src);
2055   return off;
2056 }
2057 
2058 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2059   int off = offset();
2060   ldrshw(dst, src);
2061   return off;
2062 }
2063 
2064 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2065   int off = offset();
2066   ldrsbw(dst, src);
2067   return off;
2068 }
2069 
2070 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2071   switch (size_in_bytes) {
2072   case  8:  ldr(dst, src); break;
2073   case  4:  ldrw(dst, src); break;
2074   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2075   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2076   default:  ShouldNotReachHere();
2077   }
2078 }
2079 
2080 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2081   switch (size_in_bytes) {
2082   case  8:  str(src, dst); break;
2083   case  4:  strw(src, dst); break;
2084   case  2:  strh(src, dst); break;
2085   case  1:  strb(src, dst); break;
2086   default:  ShouldNotReachHere();
2087   }
2088 }
2089 
2090 void MacroAssembler::decrementw(Register reg, int value)
2091 {
2092   if (value < 0)  { incrementw(reg, -value);      return; }
2093   if (value == 0) {                               return; }
2094   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2095   /* else */ {
2096     guarantee(reg != rscratch2, "invalid dst for register decrement");
2097     movw(rscratch2, (unsigned)value);
2098     subw(reg, reg, rscratch2);
2099   }
2100 }
2101 
2102 void MacroAssembler::decrement(Register reg, int value)
2103 {
2104   if (value < 0)  { increment(reg, -value);      return; }
2105   if (value == 0) {                              return; }
2106   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2107   /* else */ {
2108     assert(reg != rscratch2, "invalid dst for register decrement");
2109     mov(rscratch2, (uint64_t)value);
2110     sub(reg, reg, rscratch2);
2111   }
2112 }
2113 
2114 void MacroAssembler::decrementw(Address dst, int value)
2115 {
2116   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2117   if (dst.getMode() == Address::literal) {
2118     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2119     lea(rscratch2, dst);
2120     dst = Address(rscratch2);
2121   }
2122   ldrw(rscratch1, dst);
2123   decrementw(rscratch1, value);
2124   strw(rscratch1, dst);
2125 }
2126 
2127 void MacroAssembler::decrement(Address dst, int value)
2128 {
2129   assert(!dst.uses(rscratch1), "invalid address for decrement");
2130   if (dst.getMode() == Address::literal) {
2131     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2132     lea(rscratch2, dst);
2133     dst = Address(rscratch2);
2134   }
2135   ldr(rscratch1, dst);
2136   decrement(rscratch1, value);
2137   str(rscratch1, dst);
2138 }
2139 
2140 void MacroAssembler::incrementw(Register reg, int value)
2141 {
2142   if (value < 0)  { decrementw(reg, -value);      return; }
2143   if (value == 0) {                               return; }
2144   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2145   /* else */ {
2146     assert(reg != rscratch2, "invalid dst for register increment");
2147     movw(rscratch2, (unsigned)value);
2148     addw(reg, reg, rscratch2);
2149   }
2150 }
2151 
2152 void MacroAssembler::increment(Register reg, int value)
2153 {
2154   if (value < 0)  { decrement(reg, -value);      return; }
2155   if (value == 0) {                              return; }
2156   if (value < (1 << 12)) { add(reg, reg, value); return; }
2157   /* else */ {
2158     assert(reg != rscratch2, "invalid dst for register increment");
2159     movw(rscratch2, (unsigned)value);
2160     add(reg, reg, rscratch2);
2161   }
2162 }
2163 
2164 void MacroAssembler::incrementw(Address dst, int value)
2165 {
2166   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2167   if (dst.getMode() == Address::literal) {
2168     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2169     lea(rscratch2, dst);
2170     dst = Address(rscratch2);
2171   }
2172   ldrw(rscratch1, dst);
2173   incrementw(rscratch1, value);
2174   strw(rscratch1, dst);
2175 }
2176 
2177 void MacroAssembler::increment(Address dst, int value)
2178 {
2179   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2180   if (dst.getMode() == Address::literal) {
2181     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2182     lea(rscratch2, dst);
2183     dst = Address(rscratch2);
2184   }
2185   ldr(rscratch1, dst);
2186   increment(rscratch1, value);
2187   str(rscratch1, dst);
2188 }
2189 
2190 // Push lots of registers in the bit set supplied.  Don't push sp.
2191 // Return the number of words pushed
2192 int MacroAssembler::push(unsigned int bitset, Register stack) {
2193   int words_pushed = 0;
2194 
2195   // Scan bitset to accumulate register pairs
2196   unsigned char regs[32];
2197   int count = 0;
2198   for (int reg = 0; reg <= 30; reg++) {
2199     if (1 & bitset)
2200       regs[count++] = reg;
2201     bitset >>= 1;
2202   }
2203   regs[count++] = zr->raw_encoding();
2204   count &= ~1;  // Only push an even number of regs
2205 
2206   if (count) {
2207     stp(as_Register(regs[0]), as_Register(regs[1]),
2208        Address(pre(stack, -count * wordSize)));
2209     words_pushed += 2;
2210   }
2211   for (int i = 2; i < count; i += 2) {
2212     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2213        Address(stack, i * wordSize));
2214     words_pushed += 2;
2215   }
2216 
2217   assert(words_pushed == count, "oops, pushed != count");
2218 
2219   return count;
2220 }
2221 
2222 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2223   int words_pushed = 0;
2224 
2225   // Scan bitset to accumulate register pairs
2226   unsigned char regs[32];
2227   int count = 0;
2228   for (int reg = 0; reg <= 30; reg++) {
2229     if (1 & bitset)
2230       regs[count++] = reg;
2231     bitset >>= 1;
2232   }
2233   regs[count++] = zr->raw_encoding();
2234   count &= ~1;
2235 
2236   for (int i = 2; i < count; i += 2) {
2237     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2238        Address(stack, i * wordSize));
2239     words_pushed += 2;
2240   }
2241   if (count) {
2242     ldp(as_Register(regs[0]), as_Register(regs[1]),
2243        Address(post(stack, count * wordSize)));
2244     words_pushed += 2;
2245   }
2246 
2247   assert(words_pushed == count, "oops, pushed != count");
2248 
2249   return count;
2250 }
2251 
2252 // Push lots of registers in the bit set supplied.  Don't push sp.
2253 // Return the number of dwords pushed
2254 int MacroAssembler::push_fp(unsigned int bitset, Register stack) {
2255   int words_pushed = 0;
2256   bool use_sve = false;
2257   int sve_vector_size_in_bytes = 0;
2258 
2259 #ifdef COMPILER2
2260   use_sve = Matcher::supports_scalable_vector();
2261   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2262 #endif
2263 
2264   // Scan bitset to accumulate register pairs
2265   unsigned char regs[32];
2266   int count = 0;
2267   for (int reg = 0; reg <= 31; reg++) {
2268     if (1 & bitset)
2269       regs[count++] = reg;
2270     bitset >>= 1;
2271   }
2272 
2273   if (count == 0) {
2274     return 0;
2275   }
2276 
2277   // SVE
2278   if (use_sve && sve_vector_size_in_bytes > 16) {
2279     sub(stack, stack, sve_vector_size_in_bytes * count);
2280     for (int i = 0; i < count; i++) {
2281       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
2282     }
2283     return count * sve_vector_size_in_bytes / 8;
2284   }
2285 
2286   // NEON
2287   if (count == 1) {
2288     strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
2289     return 2;
2290   }
2291 
2292   bool odd = (count & 1) == 1;
2293   int push_slots = count + (odd ? 1 : 0);
2294 
2295   // Always pushing full 128 bit registers.
2296   stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
2297   words_pushed += 2;
2298 
2299   for (int i = 2; i + 1 < count; i += 2) {
2300     stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
2301     words_pushed += 2;
2302   }
2303 
2304   if (odd) {
2305     strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
2306     words_pushed++;
2307   }
2308 
2309   assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
2310   return count * 2;
2311 }
2312 
2313 // Return the number of dwords popped
2314 int MacroAssembler::pop_fp(unsigned int bitset, Register stack) {
2315   int words_pushed = 0;
2316   bool use_sve = false;
2317   int sve_vector_size_in_bytes = 0;
2318 
2319 #ifdef COMPILER2
2320   use_sve = Matcher::supports_scalable_vector();
2321   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2322 #endif
2323   // Scan bitset to accumulate register pairs
2324   unsigned char regs[32];
2325   int count = 0;
2326   for (int reg = 0; reg <= 31; reg++) {
2327     if (1 & bitset)
2328       regs[count++] = reg;
2329     bitset >>= 1;
2330   }
2331 
2332   if (count == 0) {
2333     return 0;
2334   }
2335 
2336   // SVE
2337   if (use_sve && sve_vector_size_in_bytes > 16) {
2338     for (int i = count - 1; i >= 0; i--) {
2339       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
2340     }
2341     add(stack, stack, sve_vector_size_in_bytes * count);
2342     return count * sve_vector_size_in_bytes / 8;
2343   }
2344 
2345   // NEON
2346   if (count == 1) {
2347     ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
2348     return 2;
2349   }
2350 
2351   bool odd = (count & 1) == 1;
2352   int push_slots = count + (odd ? 1 : 0);
2353 
2354   if (odd) {
2355     ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
2356     words_pushed++;
2357   }
2358 
2359   for (int i = 2; i + 1 < count; i += 2) {
2360     ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
2361     words_pushed += 2;
2362   }
2363 
2364   ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
2365   words_pushed += 2;
2366 
2367   assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
2368 
2369   return count * 2;
2370 }
2371 
2372 // Return the number of dwords pushed
2373 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
2374   bool use_sve = false;
2375   int sve_predicate_size_in_slots = 0;
2376 
2377 #ifdef COMPILER2
2378   use_sve = Matcher::supports_scalable_vector();
2379   if (use_sve) {
2380     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
2381   }
2382 #endif
2383 
2384   if (!use_sve) {
2385     return 0;
2386   }
2387 
2388   unsigned char regs[PRegister::number_of_registers];
2389   int count = 0;
2390   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
2391     if (1 & bitset)
2392       regs[count++] = reg;
2393     bitset >>= 1;
2394   }
2395 
2396   if (count == 0) {
2397     return 0;
2398   }
2399 
2400   int total_push_bytes = align_up(sve_predicate_size_in_slots *
2401                                   VMRegImpl::stack_slot_size * count, 16);
2402   sub(stack, stack, total_push_bytes);
2403   for (int i = 0; i < count; i++) {
2404     sve_str(as_PRegister(regs[i]), Address(stack, i));
2405   }
2406   return total_push_bytes / 8;
2407 }
2408 
2409 // Return the number of dwords popped
2410 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
2411   bool use_sve = false;
2412   int sve_predicate_size_in_slots = 0;
2413 
2414 #ifdef COMPILER2
2415   use_sve = Matcher::supports_scalable_vector();
2416   if (use_sve) {
2417     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
2418   }
2419 #endif
2420 
2421   if (!use_sve) {
2422     return 0;
2423   }
2424 
2425   unsigned char regs[PRegister::number_of_registers];
2426   int count = 0;
2427   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
2428     if (1 & bitset)
2429       regs[count++] = reg;
2430     bitset >>= 1;
2431   }
2432 
2433   if (count == 0) {
2434     return 0;
2435   }
2436 
2437   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
2438                                  VMRegImpl::stack_slot_size * count, 16);
2439   for (int i = count - 1; i >= 0; i--) {
2440     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
2441   }
2442   add(stack, stack, total_pop_bytes);
2443   return total_pop_bytes / 8;
2444 }
2445 
2446 #ifdef ASSERT
2447 void MacroAssembler::verify_heapbase(const char* msg) {
2448 #if 0
2449   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
2450   assert (Universe::heap() != nullptr, "java heap should be initialized");
2451   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
2452     // rheapbase is allocated as general register
2453     return;
2454   }
2455   if (CheckCompressedOops) {
2456     Label ok;
2457     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
2458     cmpptr(rheapbase, ExternalAddress(CompressedOops::ptrs_base_addr()));
2459     br(Assembler::EQ, ok);
2460     stop(msg);
2461     bind(ok);
2462     pop(1 << rscratch1->encoding(), sp);
2463   }
2464 #endif
2465 }
2466 #endif
2467 
2468 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
2469   assert_different_registers(value, tmp1, tmp2);
2470   Label done, tagged, weak_tagged;
2471 
2472   cbz(value, done);           // Use null as-is.
2473   tst(value, JNIHandles::tag_mask); // Test for tag.
2474   br(Assembler::NE, tagged);
2475 
2476   // Resolve local handle
2477   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
2478   verify_oop(value);
2479   b(done);
2480 
2481   bind(tagged);
2482   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
2483   tbnz(value, 0, weak_tagged);    // Test for weak tag.
2484 
2485   // Resolve global handle
2486   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
2487   verify_oop(value);
2488   b(done);
2489 
2490   bind(weak_tagged);
2491   // Resolve jweak.
2492   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
2493                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
2494   verify_oop(value);
2495 
2496   bind(done);
2497 }
2498 
2499 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
2500   assert_different_registers(value, tmp1, tmp2);
2501   Label done;
2502 
2503   cbz(value, done);           // Use null as-is.
2504 
2505 #ifdef ASSERT
2506   {
2507     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
2508     Label valid_global_tag;
2509     tbnz(value, 1, valid_global_tag); // Test for global tag
2510     stop("non global jobject using resolve_global_jobject");
2511     bind(valid_global_tag);
2512   }
2513 #endif
2514 
2515   // Resolve global handle
2516   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
2517   verify_oop(value);
2518 
2519   bind(done);
2520 }
2521 
2522 void MacroAssembler::stop(const char* msg) {
2523   BLOCK_COMMENT(msg);
2524   dcps1(0xdeae);
2525   emit_int64((uintptr_t)msg);
2526 }
2527 
2528 void MacroAssembler::unimplemented(const char* what) {
2529   const char* buf = nullptr;
2530   {
2531     ResourceMark rm;
2532     stringStream ss;
2533     ss.print("unimplemented: %s", what);
2534     buf = code_string(ss.as_string());
2535   }
2536   stop(buf);
2537 }
2538 
2539 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
2540 #ifdef ASSERT
2541   Label OK;
2542   br(cc, OK);
2543   stop(msg);
2544   bind(OK);
2545 #endif
2546 }
2547 
2548 // If a constant does not fit in an immediate field, generate some
2549 // number of MOV instructions and then perform the operation.
2550 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
2551                                            add_sub_imm_insn insn1,
2552                                            add_sub_reg_insn insn2,
2553                                            bool is32) {
2554   assert(Rd != zr, "Rd = zr and not setting flags?");
2555   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
2556   if (fits) {
2557     (this->*insn1)(Rd, Rn, imm);
2558   } else {
2559     if (uabs(imm) < (1 << 24)) {
2560        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
2561        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
2562     } else {
2563        assert_different_registers(Rd, Rn);
2564        mov(Rd, imm);
2565        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
2566     }
2567   }
2568 }
2569 
2570 // Separate vsn which sets the flags. Optimisations are more restricted
2571 // because we must set the flags correctly.
2572 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
2573                                              add_sub_imm_insn insn1,
2574                                              add_sub_reg_insn insn2,
2575                                              bool is32) {
2576   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
2577   if (fits) {
2578     (this->*insn1)(Rd, Rn, imm);
2579   } else {
2580     assert_different_registers(Rd, Rn);
2581     assert(Rd != zr, "overflow in immediate operand");
2582     mov(Rd, imm);
2583     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
2584   }
2585 }
2586 
2587 
2588 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
2589   if (increment.is_register()) {
2590     add(Rd, Rn, increment.as_register());
2591   } else {
2592     add(Rd, Rn, increment.as_constant());
2593   }
2594 }
2595 
2596 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
2597   if (increment.is_register()) {
2598     addw(Rd, Rn, increment.as_register());
2599   } else {
2600     addw(Rd, Rn, increment.as_constant());
2601   }
2602 }
2603 
2604 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
2605   if (decrement.is_register()) {
2606     sub(Rd, Rn, decrement.as_register());
2607   } else {
2608     sub(Rd, Rn, decrement.as_constant());
2609   }
2610 }
2611 
2612 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
2613   if (decrement.is_register()) {
2614     subw(Rd, Rn, decrement.as_register());
2615   } else {
2616     subw(Rd, Rn, decrement.as_constant());
2617   }
2618 }
2619 
2620 void MacroAssembler::reinit_heapbase()
2621 {
2622   if (UseCompressedOops) {
2623     if (Universe::is_fully_initialized()) {
2624       mov(rheapbase, CompressedOops::ptrs_base());
2625     } else {
2626       lea(rheapbase, ExternalAddress(CompressedOops::ptrs_base_addr()));
2627       ldr(rheapbase, Address(rheapbase));
2628     }
2629   }
2630 }
2631 
2632 // this simulates the behaviour of the x86 cmpxchg instruction using a
2633 // load linked/store conditional pair. we use the acquire/release
2634 // versions of these instructions so that we flush pending writes as
2635 // per Java semantics.
2636 
2637 // n.b the x86 version assumes the old value to be compared against is
2638 // in rax and updates rax with the value located in memory if the
2639 // cmpxchg fails. we supply a register for the old value explicitly
2640 
2641 // the aarch64 load linked/store conditional instructions do not
2642 // accept an offset. so, unlike x86, we must provide a plain register
2643 // to identify the memory word to be compared/exchanged rather than a
2644 // register+offset Address.
2645 
2646 void MacroAssembler::cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
2647                                 Label &succeed, Label *fail) {
2648   // oldv holds comparison value
2649   // newv holds value to write in exchange
2650   // addr identifies memory word to compare against/update
2651   if (UseLSE) {
2652     mov(tmp, oldv);
2653     casal(Assembler::xword, oldv, newv, addr);
2654     cmp(tmp, oldv);
2655     br(Assembler::EQ, succeed);
2656     membar(AnyAny);
2657   } else {
2658     Label retry_load, nope;
2659     prfm(Address(addr), PSTL1STRM);
2660     bind(retry_load);
2661     // flush and load exclusive from the memory location
2662     // and fail if it is not what we expect
2663     ldaxr(tmp, addr);
2664     cmp(tmp, oldv);
2665     br(Assembler::NE, nope);
2666     // if we store+flush with no intervening write tmp will be zero
2667     stlxr(tmp, newv, addr);
2668     cbzw(tmp, succeed);
2669     // retry so we only ever return after a load fails to compare
2670     // ensures we don't return a stale value after a failed write.
2671     b(retry_load);
2672     // if the memory word differs we return it in oldv and signal a fail
2673     bind(nope);
2674     membar(AnyAny);
2675     mov(oldv, tmp);
2676   }
2677   if (fail)
2678     b(*fail);
2679 }
2680 
2681 void MacroAssembler::cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp,
2682                                         Label &succeed, Label *fail) {
2683   assert(oopDesc::mark_offset_in_bytes() == 0, "assumption");
2684   cmpxchgptr(oldv, newv, obj, tmp, succeed, fail);
2685 }
2686 
2687 void MacroAssembler::cmpxchgw(Register oldv, Register newv, Register addr, Register tmp,
2688                                 Label &succeed, Label *fail) {
2689   // oldv holds comparison value
2690   // newv holds value to write in exchange
2691   // addr identifies memory word to compare against/update
2692   // tmp returns 0/1 for success/failure
2693   if (UseLSE) {
2694     mov(tmp, oldv);
2695     casal(Assembler::word, oldv, newv, addr);
2696     cmp(tmp, oldv);
2697     br(Assembler::EQ, succeed);
2698     membar(AnyAny);
2699   } else {
2700     Label retry_load, nope;
2701     prfm(Address(addr), PSTL1STRM);
2702     bind(retry_load);
2703     // flush and load exclusive from the memory location
2704     // and fail if it is not what we expect
2705     ldaxrw(tmp, addr);
2706     cmp(tmp, oldv);
2707     br(Assembler::NE, nope);
2708     // if we store+flush with no intervening write tmp will be zero
2709     stlxrw(tmp, newv, addr);
2710     cbzw(tmp, succeed);
2711     // retry so we only ever return after a load fails to compare
2712     // ensures we don't return a stale value after a failed write.
2713     b(retry_load);
2714     // if the memory word differs we return it in oldv and signal a fail
2715     bind(nope);
2716     membar(AnyAny);
2717     mov(oldv, tmp);
2718   }
2719   if (fail)
2720     b(*fail);
2721 }
2722 
2723 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
2724 // doesn't retry and may fail spuriously.  If the oldval is wanted,
2725 // Pass a register for the result, otherwise pass noreg.
2726 
2727 // Clobbers rscratch1
2728 void MacroAssembler::cmpxchg(Register addr, Register expected,
2729                              Register new_val,
2730                              enum operand_size size,
2731                              bool acquire, bool release,
2732                              bool weak,
2733                              Register result) {
2734   if (result == noreg)  result = rscratch1;
2735   BLOCK_COMMENT("cmpxchg {");
2736   if (UseLSE) {
2737     mov(result, expected);
2738     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
2739     compare_eq(result, expected, size);
2740   } else {
2741     Label retry_load, done;
2742     prfm(Address(addr), PSTL1STRM);
2743     bind(retry_load);
2744     load_exclusive(result, addr, size, acquire);
2745     compare_eq(result, expected, size);
2746     br(Assembler::NE, done);
2747     store_exclusive(rscratch1, new_val, addr, size, release);
2748     if (weak) {
2749       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
2750     } else {
2751       cbnzw(rscratch1, retry_load);
2752     }
2753     bind(done);
2754   }
2755   BLOCK_COMMENT("} cmpxchg");
2756 }
2757 
2758 // A generic comparison. Only compares for equality, clobbers rscratch1.
2759 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
2760   if (size == xword) {
2761     cmp(rm, rn);
2762   } else if (size == word) {
2763     cmpw(rm, rn);
2764   } else if (size == halfword) {
2765     eorw(rscratch1, rm, rn);
2766     ands(zr, rscratch1, 0xffff);
2767   } else if (size == byte) {
2768     eorw(rscratch1, rm, rn);
2769     ands(zr, rscratch1, 0xff);
2770   } else {
2771     ShouldNotReachHere();
2772   }
2773 }
2774 
2775 
2776 static bool different(Register a, RegisterOrConstant b, Register c) {
2777   if (b.is_constant())
2778     return a != c;
2779   else
2780     return a != b.as_register() && a != c && b.as_register() != c;
2781 }
2782 
2783 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
2784 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
2785   if (UseLSE) {                                                         \
2786     prev = prev->is_valid() ? prev : zr;                                \
2787     if (incr.is_register()) {                                           \
2788       AOP(sz, incr.as_register(), prev, addr);                          \
2789     } else {                                                            \
2790       mov(rscratch2, incr.as_constant());                               \
2791       AOP(sz, rscratch2, prev, addr);                                   \
2792     }                                                                   \
2793     return;                                                             \
2794   }                                                                     \
2795   Register result = rscratch2;                                          \
2796   if (prev->is_valid())                                                 \
2797     result = different(prev, incr, addr) ? prev : rscratch2;            \
2798                                                                         \
2799   Label retry_load;                                                     \
2800   prfm(Address(addr), PSTL1STRM);                                       \
2801   bind(retry_load);                                                     \
2802   LDXR(result, addr);                                                   \
2803   OP(rscratch1, result, incr);                                          \
2804   STXR(rscratch2, rscratch1, addr);                                     \
2805   cbnzw(rscratch2, retry_load);                                         \
2806   if (prev->is_valid() && prev != result) {                             \
2807     IOP(prev, rscratch1, incr);                                         \
2808   }                                                                     \
2809 }
2810 
2811 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
2812 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
2813 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
2814 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
2815 
2816 #undef ATOMIC_OP
2817 
2818 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
2819 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
2820   if (UseLSE) {                                                         \
2821     prev = prev->is_valid() ? prev : zr;                                \
2822     AOP(sz, newv, prev, addr);                                          \
2823     return;                                                             \
2824   }                                                                     \
2825   Register result = rscratch2;                                          \
2826   if (prev->is_valid())                                                 \
2827     result = different(prev, newv, addr) ? prev : rscratch2;            \
2828                                                                         \
2829   Label retry_load;                                                     \
2830   prfm(Address(addr), PSTL1STRM);                                       \
2831   bind(retry_load);                                                     \
2832   LDXR(result, addr);                                                   \
2833   STXR(rscratch1, newv, addr);                                          \
2834   cbnzw(rscratch1, retry_load);                                         \
2835   if (prev->is_valid() && prev != result)                               \
2836     mov(prev, result);                                                  \
2837 }
2838 
2839 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
2840 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
2841 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
2842 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
2843 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
2844 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
2845 
2846 #undef ATOMIC_XCHG
2847 
2848 #ifndef PRODUCT
2849 extern "C" void findpc(intptr_t x);
2850 #endif
2851 
2852 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
2853 {
2854   // In order to get locks to work, we need to fake a in_VM state
2855   if (ShowMessageBoxOnError ) {
2856     JavaThread* thread = JavaThread::current();
2857     JavaThreadState saved_state = thread->thread_state();
2858     thread->set_thread_state(_thread_in_vm);
2859 #ifndef PRODUCT
2860     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
2861       ttyLocker ttyl;
2862       BytecodeCounter::print();
2863     }
2864 #endif
2865     if (os::message_box(msg, "Execution stopped, print registers?")) {
2866       ttyLocker ttyl;
2867       tty->print_cr(" pc = 0x%016" PRIx64, pc);
2868 #ifndef PRODUCT
2869       tty->cr();
2870       findpc(pc);
2871       tty->cr();
2872 #endif
2873       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
2874       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
2875       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
2876       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
2877       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
2878       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
2879       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
2880       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
2881       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
2882       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
2883       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
2884       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
2885       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
2886       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
2887       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
2888       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
2889       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
2890       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
2891       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
2892       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
2893       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
2894       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
2895       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
2896       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
2897       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
2898       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
2899       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
2900       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
2901       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
2902       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
2903       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
2904       BREAKPOINT;
2905     }
2906   }
2907   fatal("DEBUG MESSAGE: %s", msg);
2908 }
2909 
2910 RegSet MacroAssembler::call_clobbered_gp_registers() {
2911   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
2912 #ifndef R18_RESERVED
2913   regs += r18_tls;
2914 #endif
2915   return regs;
2916 }
2917 
2918 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
2919   int step = 4 * wordSize;
2920   push(call_clobbered_gp_registers() - exclude, sp);
2921   sub(sp, sp, step);
2922   mov(rscratch1, -step);
2923   // Push v0-v7, v16-v31.
2924   for (int i = 31; i>= 4; i -= 4) {
2925     if (i <= v7->encoding() || i >= v16->encoding())
2926       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
2927           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
2928   }
2929   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
2930       as_FloatRegister(3), T1D, Address(sp));
2931 }
2932 
2933 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
2934   for (int i = 0; i < 32; i += 4) {
2935     if (i <= v7->encoding() || i >= v16->encoding())
2936       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
2937           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
2938   }
2939 
2940   reinitialize_ptrue();
2941 
2942   pop(call_clobbered_gp_registers() - exclude, sp);
2943 }
2944 
2945 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
2946                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
2947   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
2948   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
2949     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
2950     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
2951       sve_str(as_FloatRegister(i), Address(sp, i));
2952     }
2953   } else {
2954     int step = (save_vectors ? 8 : 4) * wordSize;
2955     mov(rscratch1, -step);
2956     sub(sp, sp, step);
2957     for (int i = 28; i >= 4; i -= 4) {
2958       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
2959           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
2960     }
2961     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
2962   }
2963   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
2964     sub(sp, sp, total_predicate_in_bytes);
2965     for (int i = 0; i < PRegister::number_of_registers; i++) {
2966       sve_str(as_PRegister(i), Address(sp, i));
2967     }
2968   }
2969 }
2970 
2971 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
2972                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
2973   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
2974     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
2975       sve_ldr(as_PRegister(i), Address(sp, i));
2976     }
2977     add(sp, sp, total_predicate_in_bytes);
2978   }
2979   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
2980     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
2981       sve_ldr(as_FloatRegister(i), Address(sp, i));
2982     }
2983     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
2984   } else {
2985     int step = (restore_vectors ? 8 : 4) * wordSize;
2986     for (int i = 0; i <= 28; i += 4)
2987       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
2988           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
2989   }
2990 
2991   // We may use predicate registers and rely on ptrue with SVE,
2992   // regardless of wide vector (> 8 bytes) used or not.
2993   if (use_sve) {
2994     reinitialize_ptrue();
2995   }
2996 
2997   // integer registers except lr & sp
2998   pop(RegSet::range(r0, r17), sp);
2999 #ifdef R18_RESERVED
3000   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3001   pop(RegSet::range(r20, r29), sp);
3002 #else
3003   pop(RegSet::range(r18_tls, r29), sp);
3004 #endif
3005 }
3006 
3007 /**
3008  * Helpers for multiply_to_len().
3009  */
3010 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3011                                      Register src1, Register src2) {
3012   adds(dest_lo, dest_lo, src1);
3013   adc(dest_hi, dest_hi, zr);
3014   adds(dest_lo, dest_lo, src2);
3015   adc(final_dest_hi, dest_hi, zr);
3016 }
3017 
3018 // Generate an address from (r + r1 extend offset).  "size" is the
3019 // size of the operand.  The result may be in rscratch2.
3020 Address MacroAssembler::offsetted_address(Register r, Register r1,
3021                                           Address::extend ext, int offset, int size) {
3022   if (offset || (ext.shift() % size != 0)) {
3023     lea(rscratch2, Address(r, r1, ext));
3024     return Address(rscratch2, offset);
3025   } else {
3026     return Address(r, r1, ext);
3027   }
3028 }
3029 
3030 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3031 {
3032   assert(offset >= 0, "spill to negative address?");
3033   // Offset reachable ?
3034   //   Not aligned - 9 bits signed offset
3035   //   Aligned - 12 bits unsigned offset shifted
3036   Register base = sp;
3037   if ((offset & (size-1)) && offset >= (1<<8)) {
3038     add(tmp, base, offset & ((1<<12)-1));
3039     base = tmp;
3040     offset &= -1u<<12;
3041   }
3042 
3043   if (offset >= (1<<12) * size) {
3044     add(tmp, base, offset & (((1<<12)-1)<<12));
3045     base = tmp;
3046     offset &= ~(((1<<12)-1)<<12);
3047   }
3048 
3049   return Address(base, offset);
3050 }
3051 
3052 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3053   assert(offset >= 0, "spill to negative address?");
3054 
3055   Register base = sp;
3056 
3057   // An immediate offset in the range 0 to 255 which is multiplied
3058   // by the current vector or predicate register size in bytes.
3059   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3060     return Address(base, offset / sve_reg_size_in_bytes);
3061   }
3062 
3063   add(tmp, base, offset);
3064   return Address(tmp);
3065 }
3066 
3067 // Checks whether offset is aligned.
3068 // Returns true if it is, else false.
3069 bool MacroAssembler::merge_alignment_check(Register base,
3070                                            size_t size,
3071                                            int64_t cur_offset,
3072                                            int64_t prev_offset) const {
3073   if (AvoidUnalignedAccesses) {
3074     if (base == sp) {
3075       // Checks whether low offset if aligned to pair of registers.
3076       int64_t pair_mask = size * 2 - 1;
3077       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3078       return (offset & pair_mask) == 0;
3079     } else { // If base is not sp, we can't guarantee the access is aligned.
3080       return false;
3081     }
3082   } else {
3083     int64_t mask = size - 1;
3084     // Load/store pair instruction only supports element size aligned offset.
3085     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3086   }
3087 }
3088 
3089 // Checks whether current and previous loads/stores can be merged.
3090 // Returns true if it can be merged, else false.
3091 bool MacroAssembler::ldst_can_merge(Register rt,
3092                                     const Address &adr,
3093                                     size_t cur_size_in_bytes,
3094                                     bool is_store) const {
3095   address prev = pc() - NativeInstruction::instruction_size;
3096   address last = code()->last_insn();
3097 
3098   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3099     return false;
3100   }
3101 
3102   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3103     return false;
3104   }
3105 
3106   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3107   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3108 
3109   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3110   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3111 
3112   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3113     return false;
3114   }
3115 
3116   int64_t max_offset = 63 * prev_size_in_bytes;
3117   int64_t min_offset = -64 * prev_size_in_bytes;
3118 
3119   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3120 
3121   // Only same base can be merged.
3122   if (adr.base() != prev_ldst->base()) {
3123     return false;
3124   }
3125 
3126   int64_t cur_offset = adr.offset();
3127   int64_t prev_offset = prev_ldst->offset();
3128   size_t diff = abs(cur_offset - prev_offset);
3129   if (diff != prev_size_in_bytes) {
3130     return false;
3131   }
3132 
3133   // Following cases can not be merged:
3134   // ldr x2, [x2, #8]
3135   // ldr x3, [x2, #16]
3136   // or:
3137   // ldr x2, [x3, #8]
3138   // ldr x2, [x3, #16]
3139   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3140   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3141     return false;
3142   }
3143 
3144   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3145   // Offset range must be in ldp/stp instruction's range.
3146   if (low_offset > max_offset || low_offset < min_offset) {
3147     return false;
3148   }
3149 
3150   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3151     return true;
3152   }
3153 
3154   return false;
3155 }
3156 
3157 // Merge current load/store with previous load/store into ldp/stp.
3158 void MacroAssembler::merge_ldst(Register rt,
3159                                 const Address &adr,
3160                                 size_t cur_size_in_bytes,
3161                                 bool is_store) {
3162 
3163   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3164 
3165   Register rt_low, rt_high;
3166   address prev = pc() - NativeInstruction::instruction_size;
3167   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3168 
3169   int64_t offset;
3170 
3171   if (adr.offset() < prev_ldst->offset()) {
3172     offset = adr.offset();
3173     rt_low = rt;
3174     rt_high = prev_ldst->target();
3175   } else {
3176     offset = prev_ldst->offset();
3177     rt_low = prev_ldst->target();
3178     rt_high = rt;
3179   }
3180 
3181   Address adr_p = Address(prev_ldst->base(), offset);
3182   // Overwrite previous generated binary.
3183   code_section()->set_end(prev);
3184 
3185   const size_t sz = prev_ldst->size_in_bytes();
3186   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3187   if (!is_store) {
3188     BLOCK_COMMENT("merged ldr pair");
3189     if (sz == 8) {
3190       ldp(rt_low, rt_high, adr_p);
3191     } else {
3192       ldpw(rt_low, rt_high, adr_p);
3193     }
3194   } else {
3195     BLOCK_COMMENT("merged str pair");
3196     if (sz == 8) {
3197       stp(rt_low, rt_high, adr_p);
3198     } else {
3199       stpw(rt_low, rt_high, adr_p);
3200     }
3201   }
3202 }
3203 
3204 /**
3205  * Multiply 64 bit by 64 bit first loop.
3206  */
3207 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3208                                            Register y, Register y_idx, Register z,
3209                                            Register carry, Register product,
3210                                            Register idx, Register kdx) {
3211   //
3212   //  jlong carry, x[], y[], z[];
3213   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3214   //    huge_128 product = y[idx] * x[xstart] + carry;
3215   //    z[kdx] = (jlong)product;
3216   //    carry  = (jlong)(product >>> 64);
3217   //  }
3218   //  z[xstart] = carry;
3219   //
3220 
3221   Label L_first_loop, L_first_loop_exit;
3222   Label L_one_x, L_one_y, L_multiply;
3223 
3224   subsw(xstart, xstart, 1);
3225   br(Assembler::MI, L_one_x);
3226 
3227   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3228   ldr(x_xstart, Address(rscratch1));
3229   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3230 
3231   bind(L_first_loop);
3232   subsw(idx, idx, 1);
3233   br(Assembler::MI, L_first_loop_exit);
3234   subsw(idx, idx, 1);
3235   br(Assembler::MI, L_one_y);
3236   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3237   ldr(y_idx, Address(rscratch1));
3238   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
3239   bind(L_multiply);
3240 
3241   // AArch64 has a multiply-accumulate instruction that we can't use
3242   // here because it has no way to process carries, so we have to use
3243   // separate add and adc instructions.  Bah.
3244   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
3245   mul(product, x_xstart, y_idx);
3246   adds(product, product, carry);
3247   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
3248 
3249   subw(kdx, kdx, 2);
3250   ror(product, product, 32); // back to big-endian
3251   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
3252 
3253   b(L_first_loop);
3254 
3255   bind(L_one_y);
3256   ldrw(y_idx, Address(y,  0));
3257   b(L_multiply);
3258 
3259   bind(L_one_x);
3260   ldrw(x_xstart, Address(x,  0));
3261   b(L_first_loop);
3262 
3263   bind(L_first_loop_exit);
3264 }
3265 
3266 /**
3267  * Multiply 128 bit by 128. Unrolled inner loop.
3268  *
3269  */
3270 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
3271                                              Register carry, Register carry2,
3272                                              Register idx, Register jdx,
3273                                              Register yz_idx1, Register yz_idx2,
3274                                              Register tmp, Register tmp3, Register tmp4,
3275                                              Register tmp6, Register product_hi) {
3276 
3277   //   jlong carry, x[], y[], z[];
3278   //   int kdx = ystart+1;
3279   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
3280   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
3281   //     jlong carry2  = (jlong)(tmp3 >>> 64);
3282   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
3283   //     carry  = (jlong)(tmp4 >>> 64);
3284   //     z[kdx+idx+1] = (jlong)tmp3;
3285   //     z[kdx+idx] = (jlong)tmp4;
3286   //   }
3287   //   idx += 2;
3288   //   if (idx > 0) {
3289   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
3290   //     z[kdx+idx] = (jlong)yz_idx1;
3291   //     carry  = (jlong)(yz_idx1 >>> 64);
3292   //   }
3293   //
3294 
3295   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
3296 
3297   lsrw(jdx, idx, 2);
3298 
3299   bind(L_third_loop);
3300 
3301   subsw(jdx, jdx, 1);
3302   br(Assembler::MI, L_third_loop_exit);
3303   subw(idx, idx, 4);
3304 
3305   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3306 
3307   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
3308 
3309   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
3310 
3311   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
3312   ror(yz_idx2, yz_idx2, 32);
3313 
3314   ldp(rscratch2, rscratch1, Address(tmp6, 0));
3315 
3316   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
3317   umulh(tmp4, product_hi, yz_idx1);
3318 
3319   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
3320   ror(rscratch2, rscratch2, 32);
3321 
3322   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
3323   umulh(carry2, product_hi, yz_idx2);
3324 
3325   // propagate sum of both multiplications into carry:tmp4:tmp3
3326   adds(tmp3, tmp3, carry);
3327   adc(tmp4, tmp4, zr);
3328   adds(tmp3, tmp3, rscratch1);
3329   adcs(tmp4, tmp4, tmp);
3330   adc(carry, carry2, zr);
3331   adds(tmp4, tmp4, rscratch2);
3332   adc(carry, carry, zr);
3333 
3334   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
3335   ror(tmp4, tmp4, 32);
3336   stp(tmp4, tmp3, Address(tmp6, 0));
3337 
3338   b(L_third_loop);
3339   bind (L_third_loop_exit);
3340 
3341   andw (idx, idx, 0x3);
3342   cbz(idx, L_post_third_loop_done);
3343 
3344   Label L_check_1;
3345   subsw(idx, idx, 2);
3346   br(Assembler::MI, L_check_1);
3347 
3348   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3349   ldr(yz_idx1, Address(rscratch1, 0));
3350   ror(yz_idx1, yz_idx1, 32);
3351   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
3352   umulh(tmp4, product_hi, yz_idx1);
3353   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
3354   ldr(yz_idx2, Address(rscratch1, 0));
3355   ror(yz_idx2, yz_idx2, 32);
3356 
3357   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
3358 
3359   ror(tmp3, tmp3, 32);
3360   str(tmp3, Address(rscratch1, 0));
3361 
3362   bind (L_check_1);
3363 
3364   andw (idx, idx, 0x1);
3365   subsw(idx, idx, 1);
3366   br(Assembler::MI, L_post_third_loop_done);
3367   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3368   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
3369   umulh(carry2, tmp4, product_hi);
3370   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
3371 
3372   add2_with_carry(carry2, tmp3, tmp4, carry);
3373 
3374   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
3375   extr(carry, carry2, tmp3, 32);
3376 
3377   bind(L_post_third_loop_done);
3378 }
3379 
3380 /**
3381  * Code for BigInteger::multiplyToLen() intrinsic.
3382  *
3383  * r0: x
3384  * r1: xlen
3385  * r2: y
3386  * r3: ylen
3387  * r4:  z
3388  * r5: zlen
3389  * r10: tmp1
3390  * r11: tmp2
3391  * r12: tmp3
3392  * r13: tmp4
3393  * r14: tmp5
3394  * r15: tmp6
3395  * r16: tmp7
3396  *
3397  */
3398 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
3399                                      Register z, Register zlen,
3400                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
3401                                      Register tmp5, Register tmp6, Register product_hi) {
3402 
3403   assert_different_registers(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6);
3404 
3405   const Register idx = tmp1;
3406   const Register kdx = tmp2;
3407   const Register xstart = tmp3;
3408 
3409   const Register y_idx = tmp4;
3410   const Register carry = tmp5;
3411   const Register product  = xlen;
3412   const Register x_xstart = zlen;  // reuse register
3413 
3414   // First Loop.
3415   //
3416   //  final static long LONG_MASK = 0xffffffffL;
3417   //  int xstart = xlen - 1;
3418   //  int ystart = ylen - 1;
3419   //  long carry = 0;
3420   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3421   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
3422   //    z[kdx] = (int)product;
3423   //    carry = product >>> 32;
3424   //  }
3425   //  z[xstart] = (int)carry;
3426   //
3427 
3428   movw(idx, ylen);      // idx = ylen;
3429   movw(kdx, zlen);      // kdx = xlen+ylen;
3430   mov(carry, zr);       // carry = 0;
3431 
3432   Label L_done;
3433 
3434   movw(xstart, xlen);
3435   subsw(xstart, xstart, 1);
3436   br(Assembler::MI, L_done);
3437 
3438   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
3439 
3440   Label L_second_loop;
3441   cbzw(kdx, L_second_loop);
3442 
3443   Label L_carry;
3444   subw(kdx, kdx, 1);
3445   cbzw(kdx, L_carry);
3446 
3447   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
3448   lsr(carry, carry, 32);
3449   subw(kdx, kdx, 1);
3450 
3451   bind(L_carry);
3452   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
3453 
3454   // Second and third (nested) loops.
3455   //
3456   // for (int i = xstart-1; i >= 0; i--) { // Second loop
3457   //   carry = 0;
3458   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
3459   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
3460   //                    (z[k] & LONG_MASK) + carry;
3461   //     z[k] = (int)product;
3462   //     carry = product >>> 32;
3463   //   }
3464   //   z[i] = (int)carry;
3465   // }
3466   //
3467   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
3468 
3469   const Register jdx = tmp1;
3470 
3471   bind(L_second_loop);
3472   mov(carry, zr);                // carry = 0;
3473   movw(jdx, ylen);               // j = ystart+1
3474 
3475   subsw(xstart, xstart, 1);      // i = xstart-1;
3476   br(Assembler::MI, L_done);
3477 
3478   str(z, Address(pre(sp, -4 * wordSize)));
3479 
3480   Label L_last_x;
3481   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
3482   subsw(xstart, xstart, 1);       // i = xstart-1;
3483   br(Assembler::MI, L_last_x);
3484 
3485   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
3486   ldr(product_hi, Address(rscratch1));
3487   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
3488 
3489   Label L_third_loop_prologue;
3490   bind(L_third_loop_prologue);
3491 
3492   str(ylen, Address(sp, wordSize));
3493   stp(x, xstart, Address(sp, 2 * wordSize));
3494   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
3495                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
3496   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
3497   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
3498 
3499   addw(tmp3, xlen, 1);
3500   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
3501   subsw(tmp3, tmp3, 1);
3502   br(Assembler::MI, L_done);
3503 
3504   lsr(carry, carry, 32);
3505   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
3506   b(L_second_loop);
3507 
3508   // Next infrequent code is moved outside loops.
3509   bind(L_last_x);
3510   ldrw(product_hi, Address(x,  0));
3511   b(L_third_loop_prologue);
3512 
3513   bind(L_done);
3514 }
3515 
3516 // Code for BigInteger::mulAdd intrinsic
3517 // out     = r0
3518 // in      = r1
3519 // offset  = r2  (already out.length-offset)
3520 // len     = r3
3521 // k       = r4
3522 //
3523 // pseudo code from java implementation:
3524 // carry = 0;
3525 // offset = out.length-offset - 1;
3526 // for (int j=len-1; j >= 0; j--) {
3527 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
3528 //     out[offset--] = (int)product;
3529 //     carry = product >>> 32;
3530 // }
3531 // return (int)carry;
3532 void MacroAssembler::mul_add(Register out, Register in, Register offset,
3533       Register len, Register k) {
3534     Label LOOP, END;
3535     // pre-loop
3536     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
3537     csel(out, zr, out, Assembler::EQ);
3538     br(Assembler::EQ, END);
3539     add(in, in, len, LSL, 2); // in[j+1] address
3540     add(offset, out, offset, LSL, 2); // out[offset + 1] address
3541     mov(out, zr); // used to keep carry now
3542     BIND(LOOP);
3543     ldrw(rscratch1, Address(pre(in, -4)));
3544     madd(rscratch1, rscratch1, k, out);
3545     ldrw(rscratch2, Address(pre(offset, -4)));
3546     add(rscratch1, rscratch1, rscratch2);
3547     strw(rscratch1, Address(offset));
3548     lsr(out, rscratch1, 32);
3549     subs(len, len, 1);
3550     br(Assembler::NE, LOOP);
3551     BIND(END);
3552 }
3553 
3554 /**
3555  * Emits code to update CRC-32 with a byte value according to constants in table
3556  *
3557  * @param [in,out]crc   Register containing the crc.
3558  * @param [in]val       Register containing the byte to fold into the CRC.
3559  * @param [in]table     Register containing the table of crc constants.
3560  *
3561  * uint32_t crc;
3562  * val = crc_table[(val ^ crc) & 0xFF];
3563  * crc = val ^ (crc >> 8);
3564  *
3565  */
3566 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
3567   eor(val, val, crc);
3568   andr(val, val, 0xff);
3569   ldrw(val, Address(table, val, Address::lsl(2)));
3570   eor(crc, val, crc, Assembler::LSR, 8);
3571 }
3572 
3573 /**
3574  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
3575  *
3576  * @param [in,out]crc   Register containing the crc.
3577  * @param [in]v         Register containing the 32-bit to fold into the CRC.
3578  * @param [in]table0    Register containing table 0 of crc constants.
3579  * @param [in]table1    Register containing table 1 of crc constants.
3580  * @param [in]table2    Register containing table 2 of crc constants.
3581  * @param [in]table3    Register containing table 3 of crc constants.
3582  *
3583  * uint32_t crc;
3584  *   v = crc ^ v
3585  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
3586  *
3587  */
3588 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
3589         Register table0, Register table1, Register table2, Register table3,
3590         bool upper) {
3591   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
3592   uxtb(tmp, v);
3593   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
3594   ubfx(tmp, v, 8, 8);
3595   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
3596   eor(crc, crc, tmp);
3597   ubfx(tmp, v, 16, 8);
3598   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
3599   eor(crc, crc, tmp);
3600   ubfx(tmp, v, 24, 8);
3601   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
3602   eor(crc, crc, tmp);
3603 }
3604 
3605 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
3606         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
3607     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
3608     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
3609 
3610     subs(tmp0, len, 384);
3611     mvnw(crc, crc);
3612     br(Assembler::GE, CRC_by128_pre);
3613   BIND(CRC_less128);
3614     subs(len, len, 32);
3615     br(Assembler::GE, CRC_by32_loop);
3616   BIND(CRC_less32);
3617     adds(len, len, 32 - 4);
3618     br(Assembler::GE, CRC_by4_loop);
3619     adds(len, len, 4);
3620     br(Assembler::GT, CRC_by1_loop);
3621     b(L_exit);
3622 
3623   BIND(CRC_by32_loop);
3624     ldp(tmp0, tmp1, Address(buf));
3625     crc32x(crc, crc, tmp0);
3626     ldp(tmp2, tmp3, Address(buf, 16));
3627     crc32x(crc, crc, tmp1);
3628     add(buf, buf, 32);
3629     crc32x(crc, crc, tmp2);
3630     subs(len, len, 32);
3631     crc32x(crc, crc, tmp3);
3632     br(Assembler::GE, CRC_by32_loop);
3633     cmn(len, (u1)32);
3634     br(Assembler::NE, CRC_less32);
3635     b(L_exit);
3636 
3637   BIND(CRC_by4_loop);
3638     ldrw(tmp0, Address(post(buf, 4)));
3639     subs(len, len, 4);
3640     crc32w(crc, crc, tmp0);
3641     br(Assembler::GE, CRC_by4_loop);
3642     adds(len, len, 4);
3643     br(Assembler::LE, L_exit);
3644   BIND(CRC_by1_loop);
3645     ldrb(tmp0, Address(post(buf, 1)));
3646     subs(len, len, 1);
3647     crc32b(crc, crc, tmp0);
3648     br(Assembler::GT, CRC_by1_loop);
3649     b(L_exit);
3650 
3651   BIND(CRC_by128_pre);
3652     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
3653       4*256*sizeof(juint) + 8*sizeof(juint));
3654     mov(crc, 0);
3655     crc32x(crc, crc, tmp0);
3656     crc32x(crc, crc, tmp1);
3657 
3658     cbnz(len, CRC_less128);
3659 
3660   BIND(L_exit);
3661     mvnw(crc, crc);
3662 }
3663 
3664 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
3665         Register len, Register tmp0, Register tmp1, Register tmp2,
3666         Register tmp3) {
3667     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
3668     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
3669 
3670     mvnw(crc, crc);
3671 
3672     subs(len, len, 128);
3673     br(Assembler::GE, CRC_by64_pre);
3674   BIND(CRC_less64);
3675     adds(len, len, 128-32);
3676     br(Assembler::GE, CRC_by32_loop);
3677   BIND(CRC_less32);
3678     adds(len, len, 32-4);
3679     br(Assembler::GE, CRC_by4_loop);
3680     adds(len, len, 4);
3681     br(Assembler::GT, CRC_by1_loop);
3682     b(L_exit);
3683 
3684   BIND(CRC_by32_loop);
3685     ldp(tmp0, tmp1, Address(post(buf, 16)));
3686     subs(len, len, 32);
3687     crc32x(crc, crc, tmp0);
3688     ldr(tmp2, Address(post(buf, 8)));
3689     crc32x(crc, crc, tmp1);
3690     ldr(tmp3, Address(post(buf, 8)));
3691     crc32x(crc, crc, tmp2);
3692     crc32x(crc, crc, tmp3);
3693     br(Assembler::GE, CRC_by32_loop);
3694     cmn(len, (u1)32);
3695     br(Assembler::NE, CRC_less32);
3696     b(L_exit);
3697 
3698   BIND(CRC_by4_loop);
3699     ldrw(tmp0, Address(post(buf, 4)));
3700     subs(len, len, 4);
3701     crc32w(crc, crc, tmp0);
3702     br(Assembler::GE, CRC_by4_loop);
3703     adds(len, len, 4);
3704     br(Assembler::LE, L_exit);
3705   BIND(CRC_by1_loop);
3706     ldrb(tmp0, Address(post(buf, 1)));
3707     subs(len, len, 1);
3708     crc32b(crc, crc, tmp0);
3709     br(Assembler::GT, CRC_by1_loop);
3710     b(L_exit);
3711 
3712   BIND(CRC_by64_pre);
3713     sub(buf, buf, 8);
3714     ldp(tmp0, tmp1, Address(buf, 8));
3715     crc32x(crc, crc, tmp0);
3716     ldr(tmp2, Address(buf, 24));
3717     crc32x(crc, crc, tmp1);
3718     ldr(tmp3, Address(buf, 32));
3719     crc32x(crc, crc, tmp2);
3720     ldr(tmp0, Address(buf, 40));
3721     crc32x(crc, crc, tmp3);
3722     ldr(tmp1, Address(buf, 48));
3723     crc32x(crc, crc, tmp0);
3724     ldr(tmp2, Address(buf, 56));
3725     crc32x(crc, crc, tmp1);
3726     ldr(tmp3, Address(pre(buf, 64)));
3727 
3728     b(CRC_by64_loop);
3729 
3730     align(CodeEntryAlignment);
3731   BIND(CRC_by64_loop);
3732     subs(len, len, 64);
3733     crc32x(crc, crc, tmp2);
3734     ldr(tmp0, Address(buf, 8));
3735     crc32x(crc, crc, tmp3);
3736     ldr(tmp1, Address(buf, 16));
3737     crc32x(crc, crc, tmp0);
3738     ldr(tmp2, Address(buf, 24));
3739     crc32x(crc, crc, tmp1);
3740     ldr(tmp3, Address(buf, 32));
3741     crc32x(crc, crc, tmp2);
3742     ldr(tmp0, Address(buf, 40));
3743     crc32x(crc, crc, tmp3);
3744     ldr(tmp1, Address(buf, 48));
3745     crc32x(crc, crc, tmp0);
3746     ldr(tmp2, Address(buf, 56));
3747     crc32x(crc, crc, tmp1);
3748     ldr(tmp3, Address(pre(buf, 64)));
3749     br(Assembler::GE, CRC_by64_loop);
3750 
3751     // post-loop
3752     crc32x(crc, crc, tmp2);
3753     crc32x(crc, crc, tmp3);
3754 
3755     sub(len, len, 64);
3756     add(buf, buf, 8);
3757     cmn(len, (u1)128);
3758     br(Assembler::NE, CRC_less64);
3759   BIND(L_exit);
3760     mvnw(crc, crc);
3761 }
3762 
3763 /**
3764  * @param crc   register containing existing CRC (32-bit)
3765  * @param buf   register pointing to input byte buffer (byte*)
3766  * @param len   register containing number of bytes
3767  * @param table register that will contain address of CRC table
3768  * @param tmp   scratch register
3769  */
3770 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
3771         Register table0, Register table1, Register table2, Register table3,
3772         Register tmp, Register tmp2, Register tmp3) {
3773   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
3774 
3775   if (UseCryptoPmullForCRC32) {
3776       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
3777       return;
3778   }
3779 
3780   if (UseCRC32) {
3781       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
3782       return;
3783   }
3784 
3785     mvnw(crc, crc);
3786 
3787     {
3788       uint64_t offset;
3789       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
3790       add(table0, table0, offset);
3791     }
3792     add(table1, table0, 1*256*sizeof(juint));
3793     add(table2, table0, 2*256*sizeof(juint));
3794     add(table3, table0, 3*256*sizeof(juint));
3795 
3796   if (UseNeon) {
3797       cmp(len, (u1)64);
3798       br(Assembler::LT, L_by16);
3799       eor(v16, T16B, v16, v16);
3800 
3801     Label L_fold;
3802 
3803       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
3804 
3805       ld1(v0, v1, T2D, post(buf, 32));
3806       ld1r(v4, T2D, post(tmp, 8));
3807       ld1r(v5, T2D, post(tmp, 8));
3808       ld1r(v6, T2D, post(tmp, 8));
3809       ld1r(v7, T2D, post(tmp, 8));
3810       mov(v16, S, 0, crc);
3811 
3812       eor(v0, T16B, v0, v16);
3813       sub(len, len, 64);
3814 
3815     BIND(L_fold);
3816       pmull(v22, T8H, v0, v5, T8B);
3817       pmull(v20, T8H, v0, v7, T8B);
3818       pmull(v23, T8H, v0, v4, T8B);
3819       pmull(v21, T8H, v0, v6, T8B);
3820 
3821       pmull2(v18, T8H, v0, v5, T16B);
3822       pmull2(v16, T8H, v0, v7, T16B);
3823       pmull2(v19, T8H, v0, v4, T16B);
3824       pmull2(v17, T8H, v0, v6, T16B);
3825 
3826       uzp1(v24, T8H, v20, v22);
3827       uzp2(v25, T8H, v20, v22);
3828       eor(v20, T16B, v24, v25);
3829 
3830       uzp1(v26, T8H, v16, v18);
3831       uzp2(v27, T8H, v16, v18);
3832       eor(v16, T16B, v26, v27);
3833 
3834       ushll2(v22, T4S, v20, T8H, 8);
3835       ushll(v20, T4S, v20, T4H, 8);
3836 
3837       ushll2(v18, T4S, v16, T8H, 8);
3838       ushll(v16, T4S, v16, T4H, 8);
3839 
3840       eor(v22, T16B, v23, v22);
3841       eor(v18, T16B, v19, v18);
3842       eor(v20, T16B, v21, v20);
3843       eor(v16, T16B, v17, v16);
3844 
3845       uzp1(v17, T2D, v16, v20);
3846       uzp2(v21, T2D, v16, v20);
3847       eor(v17, T16B, v17, v21);
3848 
3849       ushll2(v20, T2D, v17, T4S, 16);
3850       ushll(v16, T2D, v17, T2S, 16);
3851 
3852       eor(v20, T16B, v20, v22);
3853       eor(v16, T16B, v16, v18);
3854 
3855       uzp1(v17, T2D, v20, v16);
3856       uzp2(v21, T2D, v20, v16);
3857       eor(v28, T16B, v17, v21);
3858 
3859       pmull(v22, T8H, v1, v5, T8B);
3860       pmull(v20, T8H, v1, v7, T8B);
3861       pmull(v23, T8H, v1, v4, T8B);
3862       pmull(v21, T8H, v1, v6, T8B);
3863 
3864       pmull2(v18, T8H, v1, v5, T16B);
3865       pmull2(v16, T8H, v1, v7, T16B);
3866       pmull2(v19, T8H, v1, v4, T16B);
3867       pmull2(v17, T8H, v1, v6, T16B);
3868 
3869       ld1(v0, v1, T2D, post(buf, 32));
3870 
3871       uzp1(v24, T8H, v20, v22);
3872       uzp2(v25, T8H, v20, v22);
3873       eor(v20, T16B, v24, v25);
3874 
3875       uzp1(v26, T8H, v16, v18);
3876       uzp2(v27, T8H, v16, v18);
3877       eor(v16, T16B, v26, v27);
3878 
3879       ushll2(v22, T4S, v20, T8H, 8);
3880       ushll(v20, T4S, v20, T4H, 8);
3881 
3882       ushll2(v18, T4S, v16, T8H, 8);
3883       ushll(v16, T4S, v16, T4H, 8);
3884 
3885       eor(v22, T16B, v23, v22);
3886       eor(v18, T16B, v19, v18);
3887       eor(v20, T16B, v21, v20);
3888       eor(v16, T16B, v17, v16);
3889 
3890       uzp1(v17, T2D, v16, v20);
3891       uzp2(v21, T2D, v16, v20);
3892       eor(v16, T16B, v17, v21);
3893 
3894       ushll2(v20, T2D, v16, T4S, 16);
3895       ushll(v16, T2D, v16, T2S, 16);
3896 
3897       eor(v20, T16B, v22, v20);
3898       eor(v16, T16B, v16, v18);
3899 
3900       uzp1(v17, T2D, v20, v16);
3901       uzp2(v21, T2D, v20, v16);
3902       eor(v20, T16B, v17, v21);
3903 
3904       shl(v16, T2D, v28, 1);
3905       shl(v17, T2D, v20, 1);
3906 
3907       eor(v0, T16B, v0, v16);
3908       eor(v1, T16B, v1, v17);
3909 
3910       subs(len, len, 32);
3911       br(Assembler::GE, L_fold);
3912 
3913       mov(crc, 0);
3914       mov(tmp, v0, D, 0);
3915       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
3916       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
3917       mov(tmp, v0, D, 1);
3918       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
3919       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
3920       mov(tmp, v1, D, 0);
3921       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
3922       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
3923       mov(tmp, v1, D, 1);
3924       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
3925       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
3926 
3927       add(len, len, 32);
3928   }
3929 
3930   BIND(L_by16);
3931     subs(len, len, 16);
3932     br(Assembler::GE, L_by16_loop);
3933     adds(len, len, 16-4);
3934     br(Assembler::GE, L_by4_loop);
3935     adds(len, len, 4);
3936     br(Assembler::GT, L_by1_loop);
3937     b(L_exit);
3938 
3939   BIND(L_by4_loop);
3940     ldrw(tmp, Address(post(buf, 4)));
3941     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
3942     subs(len, len, 4);
3943     br(Assembler::GE, L_by4_loop);
3944     adds(len, len, 4);
3945     br(Assembler::LE, L_exit);
3946   BIND(L_by1_loop);
3947     subs(len, len, 1);
3948     ldrb(tmp, Address(post(buf, 1)));
3949     update_byte_crc32(crc, tmp, table0);
3950     br(Assembler::GT, L_by1_loop);
3951     b(L_exit);
3952 
3953     align(CodeEntryAlignment);
3954   BIND(L_by16_loop);
3955     subs(len, len, 16);
3956     ldp(tmp, tmp3, Address(post(buf, 16)));
3957     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
3958     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
3959     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
3960     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
3961     br(Assembler::GE, L_by16_loop);
3962     adds(len, len, 16-4);
3963     br(Assembler::GE, L_by4_loop);
3964     adds(len, len, 4);
3965     br(Assembler::GT, L_by1_loop);
3966   BIND(L_exit);
3967     mvnw(crc, crc);
3968 }
3969 
3970 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
3971         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
3972     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
3973     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
3974 
3975     subs(tmp0, len, 384);
3976     br(Assembler::GE, CRC_by128_pre);
3977   BIND(CRC_less128);
3978     subs(len, len, 32);
3979     br(Assembler::GE, CRC_by32_loop);
3980   BIND(CRC_less32);
3981     adds(len, len, 32 - 4);
3982     br(Assembler::GE, CRC_by4_loop);
3983     adds(len, len, 4);
3984     br(Assembler::GT, CRC_by1_loop);
3985     b(L_exit);
3986 
3987   BIND(CRC_by32_loop);
3988     ldp(tmp0, tmp1, Address(buf));
3989     crc32cx(crc, crc, tmp0);
3990     ldr(tmp2, Address(buf, 16));
3991     crc32cx(crc, crc, tmp1);
3992     ldr(tmp3, Address(buf, 24));
3993     crc32cx(crc, crc, tmp2);
3994     add(buf, buf, 32);
3995     subs(len, len, 32);
3996     crc32cx(crc, crc, tmp3);
3997     br(Assembler::GE, CRC_by32_loop);
3998     cmn(len, (u1)32);
3999     br(Assembler::NE, CRC_less32);
4000     b(L_exit);
4001 
4002   BIND(CRC_by4_loop);
4003     ldrw(tmp0, Address(post(buf, 4)));
4004     subs(len, len, 4);
4005     crc32cw(crc, crc, tmp0);
4006     br(Assembler::GE, CRC_by4_loop);
4007     adds(len, len, 4);
4008     br(Assembler::LE, L_exit);
4009   BIND(CRC_by1_loop);
4010     ldrb(tmp0, Address(post(buf, 1)));
4011     subs(len, len, 1);
4012     crc32cb(crc, crc, tmp0);
4013     br(Assembler::GT, CRC_by1_loop);
4014     b(L_exit);
4015 
4016   BIND(CRC_by128_pre);
4017     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4018       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4019     mov(crc, 0);
4020     crc32cx(crc, crc, tmp0);
4021     crc32cx(crc, crc, tmp1);
4022 
4023     cbnz(len, CRC_less128);
4024 
4025   BIND(L_exit);
4026 }
4027 
4028 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4029         Register len, Register tmp0, Register tmp1, Register tmp2,
4030         Register tmp3) {
4031     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4032     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4033 
4034     subs(len, len, 128);
4035     br(Assembler::GE, CRC_by64_pre);
4036   BIND(CRC_less64);
4037     adds(len, len, 128-32);
4038     br(Assembler::GE, CRC_by32_loop);
4039   BIND(CRC_less32);
4040     adds(len, len, 32-4);
4041     br(Assembler::GE, CRC_by4_loop);
4042     adds(len, len, 4);
4043     br(Assembler::GT, CRC_by1_loop);
4044     b(L_exit);
4045 
4046   BIND(CRC_by32_loop);
4047     ldp(tmp0, tmp1, Address(post(buf, 16)));
4048     subs(len, len, 32);
4049     crc32cx(crc, crc, tmp0);
4050     ldr(tmp2, Address(post(buf, 8)));
4051     crc32cx(crc, crc, tmp1);
4052     ldr(tmp3, Address(post(buf, 8)));
4053     crc32cx(crc, crc, tmp2);
4054     crc32cx(crc, crc, tmp3);
4055     br(Assembler::GE, CRC_by32_loop);
4056     cmn(len, (u1)32);
4057     br(Assembler::NE, CRC_less32);
4058     b(L_exit);
4059 
4060   BIND(CRC_by4_loop);
4061     ldrw(tmp0, Address(post(buf, 4)));
4062     subs(len, len, 4);
4063     crc32cw(crc, crc, tmp0);
4064     br(Assembler::GE, CRC_by4_loop);
4065     adds(len, len, 4);
4066     br(Assembler::LE, L_exit);
4067   BIND(CRC_by1_loop);
4068     ldrb(tmp0, Address(post(buf, 1)));
4069     subs(len, len, 1);
4070     crc32cb(crc, crc, tmp0);
4071     br(Assembler::GT, CRC_by1_loop);
4072     b(L_exit);
4073 
4074   BIND(CRC_by64_pre);
4075     sub(buf, buf, 8);
4076     ldp(tmp0, tmp1, Address(buf, 8));
4077     crc32cx(crc, crc, tmp0);
4078     ldr(tmp2, Address(buf, 24));
4079     crc32cx(crc, crc, tmp1);
4080     ldr(tmp3, Address(buf, 32));
4081     crc32cx(crc, crc, tmp2);
4082     ldr(tmp0, Address(buf, 40));
4083     crc32cx(crc, crc, tmp3);
4084     ldr(tmp1, Address(buf, 48));
4085     crc32cx(crc, crc, tmp0);
4086     ldr(tmp2, Address(buf, 56));
4087     crc32cx(crc, crc, tmp1);
4088     ldr(tmp3, Address(pre(buf, 64)));
4089 
4090     b(CRC_by64_loop);
4091 
4092     align(CodeEntryAlignment);
4093   BIND(CRC_by64_loop);
4094     subs(len, len, 64);
4095     crc32cx(crc, crc, tmp2);
4096     ldr(tmp0, Address(buf, 8));
4097     crc32cx(crc, crc, tmp3);
4098     ldr(tmp1, Address(buf, 16));
4099     crc32cx(crc, crc, tmp0);
4100     ldr(tmp2, Address(buf, 24));
4101     crc32cx(crc, crc, tmp1);
4102     ldr(tmp3, Address(buf, 32));
4103     crc32cx(crc, crc, tmp2);
4104     ldr(tmp0, Address(buf, 40));
4105     crc32cx(crc, crc, tmp3);
4106     ldr(tmp1, Address(buf, 48));
4107     crc32cx(crc, crc, tmp0);
4108     ldr(tmp2, Address(buf, 56));
4109     crc32cx(crc, crc, tmp1);
4110     ldr(tmp3, Address(pre(buf, 64)));
4111     br(Assembler::GE, CRC_by64_loop);
4112 
4113     // post-loop
4114     crc32cx(crc, crc, tmp2);
4115     crc32cx(crc, crc, tmp3);
4116 
4117     sub(len, len, 64);
4118     add(buf, buf, 8);
4119     cmn(len, (u1)128);
4120     br(Assembler::NE, CRC_less64);
4121   BIND(L_exit);
4122 }
4123 
4124 /**
4125  * @param crc   register containing existing CRC (32-bit)
4126  * @param buf   register pointing to input byte buffer (byte*)
4127  * @param len   register containing number of bytes
4128  * @param table register that will contain address of CRC table
4129  * @param tmp   scratch register
4130  */
4131 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4132         Register table0, Register table1, Register table2, Register table3,
4133         Register tmp, Register tmp2, Register tmp3) {
4134   if (UseCryptoPmullForCRC32) {
4135     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4136   } else {
4137     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4138   }
4139 }
4140 
4141 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4142         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4143     Label CRC_by128_loop;
4144     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4145 
4146     sub(len, len, 256);
4147     Register table = tmp0;
4148     {
4149       uint64_t offset;
4150       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4151       add(table, table, offset);
4152     }
4153     add(table, table, table_offset);
4154 
4155     sub(buf, buf, 0x10);
4156     ldrq(v1, Address(buf, 0x10));
4157     ldrq(v2, Address(buf, 0x20));
4158     ldrq(v3, Address(buf, 0x30));
4159     ldrq(v4, Address(buf, 0x40));
4160     ldrq(v5, Address(buf, 0x50));
4161     ldrq(v6, Address(buf, 0x60));
4162     ldrq(v7, Address(buf, 0x70));
4163     ldrq(v8, Address(pre(buf, 0x80)));
4164 
4165     movi(v25, T4S, 0);
4166     mov(v25, S, 0, crc);
4167     eor(v1, T16B, v1, v25);
4168 
4169     ldrq(v0, Address(table));
4170     b(CRC_by128_loop);
4171 
4172     align(OptoLoopAlignment);
4173   BIND(CRC_by128_loop);
4174     pmull (v9,  T1Q, v1, v0, T1D);
4175     pmull2(v10, T1Q, v1, v0, T2D);
4176     ldrq(v1, Address(buf, 0x10));
4177     eor3(v1, T16B, v9,  v10, v1);
4178 
4179     pmull (v11, T1Q, v2, v0, T1D);
4180     pmull2(v12, T1Q, v2, v0, T2D);
4181     ldrq(v2, Address(buf, 0x20));
4182     eor3(v2, T16B, v11, v12, v2);
4183 
4184     pmull (v13, T1Q, v3, v0, T1D);
4185     pmull2(v14, T1Q, v3, v0, T2D);
4186     ldrq(v3, Address(buf, 0x30));
4187     eor3(v3, T16B, v13, v14, v3);
4188 
4189     pmull (v15, T1Q, v4, v0, T1D);
4190     pmull2(v16, T1Q, v4, v0, T2D);
4191     ldrq(v4, Address(buf, 0x40));
4192     eor3(v4, T16B, v15, v16, v4);
4193 
4194     pmull (v17, T1Q, v5, v0, T1D);
4195     pmull2(v18, T1Q, v5, v0, T2D);
4196     ldrq(v5, Address(buf, 0x50));
4197     eor3(v5, T16B, v17, v18, v5);
4198 
4199     pmull (v19, T1Q, v6, v0, T1D);
4200     pmull2(v20, T1Q, v6, v0, T2D);
4201     ldrq(v6, Address(buf, 0x60));
4202     eor3(v6, T16B, v19, v20, v6);
4203 
4204     pmull (v21, T1Q, v7, v0, T1D);
4205     pmull2(v22, T1Q, v7, v0, T2D);
4206     ldrq(v7, Address(buf, 0x70));
4207     eor3(v7, T16B, v21, v22, v7);
4208 
4209     pmull (v23, T1Q, v8, v0, T1D);
4210     pmull2(v24, T1Q, v8, v0, T2D);
4211     ldrq(v8, Address(pre(buf, 0x80)));
4212     eor3(v8, T16B, v23, v24, v8);
4213 
4214     subs(len, len, 0x80);
4215     br(Assembler::GE, CRC_by128_loop);
4216 
4217     // fold into 512 bits
4218     ldrq(v0, Address(table, 0x10));
4219 
4220     pmull (v10,  T1Q, v1, v0, T1D);
4221     pmull2(v11, T1Q, v1, v0, T2D);
4222     eor3(v1, T16B, v10, v11, v5);
4223 
4224     pmull (v12, T1Q, v2, v0, T1D);
4225     pmull2(v13, T1Q, v2, v0, T2D);
4226     eor3(v2, T16B, v12, v13, v6);
4227 
4228     pmull (v14, T1Q, v3, v0, T1D);
4229     pmull2(v15, T1Q, v3, v0, T2D);
4230     eor3(v3, T16B, v14, v15, v7);
4231 
4232     pmull (v16, T1Q, v4, v0, T1D);
4233     pmull2(v17, T1Q, v4, v0, T2D);
4234     eor3(v4, T16B, v16, v17, v8);
4235 
4236     // fold into 128 bits
4237     ldrq(v5, Address(table, 0x20));
4238     pmull (v10, T1Q, v1, v5, T1D);
4239     pmull2(v11, T1Q, v1, v5, T2D);
4240     eor3(v4, T16B, v4, v10, v11);
4241 
4242     ldrq(v6, Address(table, 0x30));
4243     pmull (v12, T1Q, v2, v6, T1D);
4244     pmull2(v13, T1Q, v2, v6, T2D);
4245     eor3(v4, T16B, v4, v12, v13);
4246 
4247     ldrq(v7, Address(table, 0x40));
4248     pmull (v14, T1Q, v3, v7, T1D);
4249     pmull2(v15, T1Q, v3, v7, T2D);
4250     eor3(v1, T16B, v4, v14, v15);
4251 
4252     add(len, len, 0x80);
4253     add(buf, buf, 0x10);
4254 
4255     mov(tmp0, v1, D, 0);
4256     mov(tmp1, v1, D, 1);
4257 }
4258 
4259 SkipIfEqual::SkipIfEqual(
4260     MacroAssembler* masm, const bool* flag_addr, bool value) {
4261   _masm = masm;
4262   uint64_t offset;
4263   _masm->adrp(rscratch1, ExternalAddress((address)flag_addr), offset);
4264   _masm->ldrb(rscratch1, Address(rscratch1, offset));
4265   if (value) {
4266     _masm->cbnzw(rscratch1, _label);
4267   } else {
4268     _masm->cbzw(rscratch1, _label);
4269   }
4270 }
4271 
4272 SkipIfEqual::~SkipIfEqual() {
4273   _masm->bind(_label);
4274 }
4275 
4276 void MacroAssembler::addptr(const Address &dst, int32_t src) {
4277   Address adr;
4278   switch(dst.getMode()) {
4279   case Address::base_plus_offset:
4280     // This is the expected mode, although we allow all the other
4281     // forms below.
4282     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
4283     break;
4284   default:
4285     lea(rscratch2, dst);
4286     adr = Address(rscratch2);
4287     break;
4288   }
4289   ldr(rscratch1, adr);
4290   add(rscratch1, rscratch1, src);
4291   str(rscratch1, adr);
4292 }
4293 
4294 void MacroAssembler::cmpptr(Register src1, Address src2) {
4295   uint64_t offset;
4296   adrp(rscratch1, src2, offset);
4297   ldr(rscratch1, Address(rscratch1, offset));
4298   cmp(src1, rscratch1);
4299 }
4300 
4301 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
4302   cmp(obj1, obj2);
4303 }
4304 
4305 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
4306   load_method_holder(rresult, rmethod);
4307   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
4308 }
4309 
4310 void MacroAssembler::load_method_holder(Register holder, Register method) {
4311   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
4312   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
4313   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
4314 }
4315 
4316 void MacroAssembler::load_klass(Register dst, Register src) {
4317   if (UseCompressedClassPointers) {
4318     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
4319     decode_klass_not_null(dst);
4320   } else {
4321     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
4322   }
4323 }
4324 
4325 // ((OopHandle)result).resolve();
4326 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
4327   // OopHandle::resolve is an indirection.
4328   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
4329 }
4330 
4331 // ((WeakHandle)result).resolve();
4332 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
4333   assert_different_registers(result, tmp1, tmp2);
4334   Label resolved;
4335 
4336   // A null weak handle resolves to null.
4337   cbz(result, resolved);
4338 
4339   // Only 64 bit platforms support GCs that require a tmp register
4340   // WeakHandle::resolve is an indirection like jweak.
4341   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
4342                  result, Address(result), tmp1, tmp2);
4343   bind(resolved);
4344 }
4345 
4346 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
4347   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
4348   ldr(dst, Address(rmethod, Method::const_offset()));
4349   ldr(dst, Address(dst, ConstMethod::constants_offset()));
4350   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
4351   ldr(dst, Address(dst, mirror_offset));
4352   resolve_oop_handle(dst, tmp1, tmp2);
4353 }
4354 
4355 void MacroAssembler::cmp_klass(Register oop, Register trial_klass, Register tmp) {
4356   if (UseCompressedClassPointers) {
4357     ldrw(tmp, Address(oop, oopDesc::klass_offset_in_bytes()));
4358     if (CompressedKlassPointers::base() == nullptr) {
4359       cmp(trial_klass, tmp, LSL, CompressedKlassPointers::shift());
4360       return;
4361     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
4362                && CompressedKlassPointers::shift() == 0) {
4363       // Only the bottom 32 bits matter
4364       cmpw(trial_klass, tmp);
4365       return;
4366     }
4367     decode_klass_not_null(tmp);
4368   } else {
4369     ldr(tmp, Address(oop, oopDesc::klass_offset_in_bytes()));
4370   }
4371   cmp(trial_klass, tmp);
4372 }
4373 
4374 void MacroAssembler::store_klass(Register dst, Register src) {
4375   // FIXME: Should this be a store release?  concurrent gcs assumes
4376   // klass length is valid if klass field is not null.
4377   if (UseCompressedClassPointers) {
4378     encode_klass_not_null(src);
4379     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
4380   } else {
4381     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
4382   }
4383 }
4384 
4385 void MacroAssembler::store_klass_gap(Register dst, Register src) {
4386   if (UseCompressedClassPointers) {
4387     // Store to klass gap in destination
4388     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
4389   }
4390 }
4391 
4392 // Algorithm must match CompressedOops::encode.
4393 void MacroAssembler::encode_heap_oop(Register d, Register s) {
4394 #ifdef ASSERT
4395   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
4396 #endif
4397   verify_oop_msg(s, "broken oop in encode_heap_oop");
4398   if (CompressedOops::base() == nullptr) {
4399     if (CompressedOops::shift() != 0) {
4400       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
4401       lsr(d, s, LogMinObjAlignmentInBytes);
4402     } else {
4403       mov(d, s);
4404     }
4405   } else {
4406     subs(d, s, rheapbase);
4407     csel(d, d, zr, Assembler::HS);
4408     lsr(d, d, LogMinObjAlignmentInBytes);
4409 
4410     /*  Old algorithm: is this any worse?
4411     Label nonnull;
4412     cbnz(r, nonnull);
4413     sub(r, r, rheapbase);
4414     bind(nonnull);
4415     lsr(r, r, LogMinObjAlignmentInBytes);
4416     */
4417   }
4418 }
4419 
4420 void MacroAssembler::encode_heap_oop_not_null(Register r) {
4421 #ifdef ASSERT
4422   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
4423   if (CheckCompressedOops) {
4424     Label ok;
4425     cbnz(r, ok);
4426     stop("null oop passed to encode_heap_oop_not_null");
4427     bind(ok);
4428   }
4429 #endif
4430   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
4431   if (CompressedOops::base() != nullptr) {
4432     sub(r, r, rheapbase);
4433   }
4434   if (CompressedOops::shift() != 0) {
4435     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
4436     lsr(r, r, LogMinObjAlignmentInBytes);
4437   }
4438 }
4439 
4440 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
4441 #ifdef ASSERT
4442   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
4443   if (CheckCompressedOops) {
4444     Label ok;
4445     cbnz(src, ok);
4446     stop("null oop passed to encode_heap_oop_not_null2");
4447     bind(ok);
4448   }
4449 #endif
4450   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
4451 
4452   Register data = src;
4453   if (CompressedOops::base() != nullptr) {
4454     sub(dst, src, rheapbase);
4455     data = dst;
4456   }
4457   if (CompressedOops::shift() != 0) {
4458     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
4459     lsr(dst, data, LogMinObjAlignmentInBytes);
4460     data = dst;
4461   }
4462   if (data == src)
4463     mov(dst, src);
4464 }
4465 
4466 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
4467 #ifdef ASSERT
4468   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
4469 #endif
4470   if (CompressedOops::base() == nullptr) {
4471     if (CompressedOops::shift() != 0 || d != s) {
4472       lsl(d, s, CompressedOops::shift());
4473     }
4474   } else {
4475     Label done;
4476     if (d != s)
4477       mov(d, s);
4478     cbz(s, done);
4479     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
4480     bind(done);
4481   }
4482   verify_oop_msg(d, "broken oop in decode_heap_oop");
4483 }
4484 
4485 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
4486   assert (UseCompressedOops, "should only be used for compressed headers");
4487   assert (Universe::heap() != nullptr, "java heap should be initialized");
4488   // Cannot assert, unverified entry point counts instructions (see .ad file)
4489   // vtableStubs also counts instructions in pd_code_size_limit.
4490   // Also do not verify_oop as this is called by verify_oop.
4491   if (CompressedOops::shift() != 0) {
4492     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
4493     if (CompressedOops::base() != nullptr) {
4494       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
4495     } else {
4496       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
4497     }
4498   } else {
4499     assert (CompressedOops::base() == nullptr, "sanity");
4500   }
4501 }
4502 
4503 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
4504   assert (UseCompressedOops, "should only be used for compressed headers");
4505   assert (Universe::heap() != nullptr, "java heap should be initialized");
4506   // Cannot assert, unverified entry point counts instructions (see .ad file)
4507   // vtableStubs also counts instructions in pd_code_size_limit.
4508   // Also do not verify_oop as this is called by verify_oop.
4509   if (CompressedOops::shift() != 0) {
4510     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
4511     if (CompressedOops::base() != nullptr) {
4512       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
4513     } else {
4514       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
4515     }
4516   } else {
4517     assert (CompressedOops::base() == nullptr, "sanity");
4518     if (dst != src) {
4519       mov(dst, src);
4520     }
4521   }
4522 }
4523 
4524 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
4525 
4526 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
4527   assert(UseCompressedClassPointers, "not using compressed class pointers");
4528   assert(Metaspace::initialized(), "metaspace not initialized yet");
4529 
4530   if (_klass_decode_mode != KlassDecodeNone) {
4531     return _klass_decode_mode;
4532   }
4533 
4534   assert(LogKlassAlignmentInBytes == CompressedKlassPointers::shift()
4535          || 0 == CompressedKlassPointers::shift(), "decode alg wrong");
4536 
4537   if (CompressedKlassPointers::base() == nullptr) {
4538     return (_klass_decode_mode = KlassDecodeZero);
4539   }
4540 
4541   if (operand_valid_for_logical_immediate(
4542         /*is32*/false, (uint64_t)CompressedKlassPointers::base())) {
4543     const uint64_t range_mask =
4544       (1ULL << log2i(CompressedKlassPointers::range())) - 1;
4545     if (((uint64_t)CompressedKlassPointers::base() & range_mask) == 0) {
4546       return (_klass_decode_mode = KlassDecodeXor);
4547     }
4548   }
4549 
4550   const uint64_t shifted_base =
4551     (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
4552   guarantee((shifted_base & 0xffff0000ffffffff) == 0,
4553             "compressed class base bad alignment");
4554 
4555   return (_klass_decode_mode = KlassDecodeMovk);
4556 }
4557 
4558 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
4559   switch (klass_decode_mode()) {
4560   case KlassDecodeZero:
4561     if (CompressedKlassPointers::shift() != 0) {
4562       lsr(dst, src, LogKlassAlignmentInBytes);
4563     } else {
4564       if (dst != src) mov(dst, src);
4565     }
4566     break;
4567 
4568   case KlassDecodeXor:
4569     if (CompressedKlassPointers::shift() != 0) {
4570       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
4571       lsr(dst, dst, LogKlassAlignmentInBytes);
4572     } else {
4573       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
4574     }
4575     break;
4576 
4577   case KlassDecodeMovk:
4578     if (CompressedKlassPointers::shift() != 0) {
4579       ubfx(dst, src, LogKlassAlignmentInBytes, 32);
4580     } else {
4581       movw(dst, src);
4582     }
4583     break;
4584 
4585   case KlassDecodeNone:
4586     ShouldNotReachHere();
4587     break;
4588   }
4589 }
4590 
4591 void MacroAssembler::encode_klass_not_null(Register r) {
4592   encode_klass_not_null(r, r);
4593 }
4594 
4595 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
4596   assert (UseCompressedClassPointers, "should only be used for compressed headers");
4597 
4598   switch (klass_decode_mode()) {
4599   case KlassDecodeZero:
4600     if (CompressedKlassPointers::shift() != 0) {
4601       lsl(dst, src, LogKlassAlignmentInBytes);
4602     } else {
4603       if (dst != src) mov(dst, src);
4604     }
4605     break;
4606 
4607   case KlassDecodeXor:
4608     if (CompressedKlassPointers::shift() != 0) {
4609       lsl(dst, src, LogKlassAlignmentInBytes);
4610       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
4611     } else {
4612       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
4613     }
4614     break;
4615 
4616   case KlassDecodeMovk: {
4617     const uint64_t shifted_base =
4618       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
4619 
4620     if (dst != src) movw(dst, src);
4621     movk(dst, shifted_base >> 32, 32);
4622 
4623     if (CompressedKlassPointers::shift() != 0) {
4624       lsl(dst, dst, LogKlassAlignmentInBytes);
4625     }
4626 
4627     break;
4628   }
4629 
4630   case KlassDecodeNone:
4631     ShouldNotReachHere();
4632     break;
4633   }
4634 }
4635 
4636 void  MacroAssembler::decode_klass_not_null(Register r) {
4637   decode_klass_not_null(r, r);
4638 }
4639 
4640 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
4641 #ifdef ASSERT
4642   {
4643     ThreadInVMfromUnknown tiv;
4644     assert (UseCompressedOops, "should only be used for compressed oops");
4645     assert (Universe::heap() != nullptr, "java heap should be initialized");
4646     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
4647     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
4648   }
4649 #endif
4650   int oop_index = oop_recorder()->find_index(obj);
4651   InstructionMark im(this);
4652   RelocationHolder rspec = oop_Relocation::spec(oop_index);
4653   code_section()->relocate(inst_mark(), rspec);
4654   movz(dst, 0xDEAD, 16);
4655   movk(dst, 0xBEEF);
4656 }
4657 
4658 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
4659   assert (UseCompressedClassPointers, "should only be used for compressed headers");
4660   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
4661   int index = oop_recorder()->find_index(k);
4662   assert(! Universe::heap()->is_in(k), "should not be an oop");
4663 
4664   InstructionMark im(this);
4665   RelocationHolder rspec = metadata_Relocation::spec(index);
4666   code_section()->relocate(inst_mark(), rspec);
4667   narrowKlass nk = CompressedKlassPointers::encode(k);
4668   movz(dst, (nk >> 16), 16);
4669   movk(dst, nk & 0xffff);
4670 }
4671 
4672 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
4673                                     Register dst, Address src,
4674                                     Register tmp1, Register tmp2) {
4675   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
4676   decorators = AccessInternal::decorator_fixup(decorators, type);
4677   bool as_raw = (decorators & AS_RAW) != 0;
4678   if (as_raw) {
4679     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
4680   } else {
4681     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
4682   }
4683 }
4684 
4685 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
4686                                      Address dst, Register val,
4687                                      Register tmp1, Register tmp2, Register tmp3) {
4688   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
4689   decorators = AccessInternal::decorator_fixup(decorators, type);
4690   bool as_raw = (decorators & AS_RAW) != 0;
4691   if (as_raw) {
4692     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
4693   } else {
4694     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
4695   }
4696 }
4697 
4698 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
4699                                    Register tmp2, DecoratorSet decorators) {
4700   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
4701 }
4702 
4703 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
4704                                             Register tmp2, DecoratorSet decorators) {
4705   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
4706 }
4707 
4708 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
4709                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
4710   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
4711 }
4712 
4713 // Used for storing nulls.
4714 void MacroAssembler::store_heap_oop_null(Address dst) {
4715   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
4716 }
4717 
4718 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
4719   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
4720   int index = oop_recorder()->allocate_metadata_index(obj);
4721   RelocationHolder rspec = metadata_Relocation::spec(index);
4722   return Address((address)obj, rspec);
4723 }
4724 
4725 // Move an oop into a register.
4726 void MacroAssembler::movoop(Register dst, jobject obj) {
4727   int oop_index;
4728   if (obj == nullptr) {
4729     oop_index = oop_recorder()->allocate_oop_index(obj);
4730   } else {
4731 #ifdef ASSERT
4732     {
4733       ThreadInVMfromUnknown tiv;
4734       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
4735     }
4736 #endif
4737     oop_index = oop_recorder()->find_index(obj);
4738   }
4739   RelocationHolder rspec = oop_Relocation::spec(oop_index);
4740 
4741   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
4742     mov(dst, Address((address)obj, rspec));
4743   } else {
4744     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
4745     ldr_constant(dst, Address(dummy, rspec));
4746   }
4747 
4748 }
4749 
4750 // Move a metadata address into a register.
4751 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
4752   int oop_index;
4753   if (obj == nullptr) {
4754     oop_index = oop_recorder()->allocate_metadata_index(obj);
4755   } else {
4756     oop_index = oop_recorder()->find_index(obj);
4757   }
4758   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
4759   mov(dst, Address((address)obj, rspec));
4760 }
4761 
4762 Address MacroAssembler::constant_oop_address(jobject obj) {
4763 #ifdef ASSERT
4764   {
4765     ThreadInVMfromUnknown tiv;
4766     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
4767     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
4768   }
4769 #endif
4770   int oop_index = oop_recorder()->find_index(obj);
4771   return Address((address)obj, oop_Relocation::spec(oop_index));
4772 }
4773 
4774 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
4775 void MacroAssembler::tlab_allocate(Register obj,
4776                                    Register var_size_in_bytes,
4777                                    int con_size_in_bytes,
4778                                    Register t1,
4779                                    Register t2,
4780                                    Label& slow_case) {
4781   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
4782   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
4783 }
4784 
4785 void MacroAssembler::verify_tlab() {
4786 #ifdef ASSERT
4787   if (UseTLAB && VerifyOops) {
4788     Label next, ok;
4789 
4790     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
4791 
4792     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
4793     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
4794     cmp(rscratch2, rscratch1);
4795     br(Assembler::HS, next);
4796     STOP("assert(top >= start)");
4797     should_not_reach_here();
4798 
4799     bind(next);
4800     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
4801     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
4802     cmp(rscratch2, rscratch1);
4803     br(Assembler::HS, ok);
4804     STOP("assert(top <= end)");
4805     should_not_reach_here();
4806 
4807     bind(ok);
4808     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
4809   }
4810 #endif
4811 }
4812 
4813 // Writes to stack successive pages until offset reached to check for
4814 // stack overflow + shadow pages.  This clobbers tmp.
4815 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
4816   assert_different_registers(tmp, size, rscratch1);
4817   mov(tmp, sp);
4818   // Bang stack for total size given plus shadow page size.
4819   // Bang one page at a time because large size can bang beyond yellow and
4820   // red zones.
4821   Label loop;
4822   mov(rscratch1, (int)os::vm_page_size());
4823   bind(loop);
4824   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
4825   subsw(size, size, rscratch1);
4826   str(size, Address(tmp));
4827   br(Assembler::GT, loop);
4828 
4829   // Bang down shadow pages too.
4830   // At this point, (tmp-0) is the last address touched, so don't
4831   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
4832   // was post-decremented.)  Skip this address by starting at i=1, and
4833   // touch a few more pages below.  N.B.  It is important to touch all
4834   // the way down to and including i=StackShadowPages.
4835   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
4836     // this could be any sized move but this is can be a debugging crumb
4837     // so the bigger the better.
4838     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
4839     str(size, Address(tmp));
4840   }
4841 }
4842 
4843 // Move the address of the polling page into dest.
4844 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
4845   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
4846 }
4847 
4848 // Read the polling page.  The address of the polling page must
4849 // already be in r.
4850 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
4851   address mark;
4852   {
4853     InstructionMark im(this);
4854     code_section()->relocate(inst_mark(), rtype);
4855     ldrw(zr, Address(r, 0));
4856     mark = inst_mark();
4857   }
4858   verify_cross_modify_fence_not_required();
4859   return mark;
4860 }
4861 
4862 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
4863   relocInfo::relocType rtype = dest.rspec().reloc()->type();
4864   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
4865   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
4866   uint64_t dest_page = (uint64_t)dest.target() >> 12;
4867   int64_t offset_low = dest_page - low_page;
4868   int64_t offset_high = dest_page - high_page;
4869 
4870   assert(is_valid_AArch64_address(dest.target()), "bad address");
4871   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
4872 
4873   InstructionMark im(this);
4874   code_section()->relocate(inst_mark(), dest.rspec());
4875   // 8143067: Ensure that the adrp can reach the dest from anywhere within
4876   // the code cache so that if it is relocated we know it will still reach
4877   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
4878     _adrp(reg1, dest.target());
4879   } else {
4880     uint64_t target = (uint64_t)dest.target();
4881     uint64_t adrp_target
4882       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
4883 
4884     _adrp(reg1, (address)adrp_target);
4885     movk(reg1, target >> 32, 32);
4886   }
4887   byte_offset = (uint64_t)dest.target() & 0xfff;
4888 }
4889 
4890 void MacroAssembler::load_byte_map_base(Register reg) {
4891   CardTable::CardValue* byte_map_base =
4892     ((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base();
4893 
4894   // Strictly speaking the byte_map_base isn't an address at all, and it might
4895   // even be negative. It is thus materialised as a constant.
4896   mov(reg, (uint64_t)byte_map_base);
4897 }
4898 
4899 void MacroAssembler::build_frame(int framesize) {
4900   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
4901   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
4902   protect_return_address();
4903   if (framesize < ((1 << 9) + 2 * wordSize)) {
4904     sub(sp, sp, framesize);
4905     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
4906     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
4907   } else {
4908     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
4909     if (PreserveFramePointer) mov(rfp, sp);
4910     if (framesize < ((1 << 12) + 2 * wordSize))
4911       sub(sp, sp, framesize - 2 * wordSize);
4912     else {
4913       mov(rscratch1, framesize - 2 * wordSize);
4914       sub(sp, sp, rscratch1);
4915     }
4916   }
4917   verify_cross_modify_fence_not_required();
4918 }
4919 
4920 void MacroAssembler::remove_frame(int framesize) {
4921   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
4922   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
4923   if (framesize < ((1 << 9) + 2 * wordSize)) {
4924     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
4925     add(sp, sp, framesize);
4926   } else {
4927     if (framesize < ((1 << 12) + 2 * wordSize))
4928       add(sp, sp, framesize - 2 * wordSize);
4929     else {
4930       mov(rscratch1, framesize - 2 * wordSize);
4931       add(sp, sp, rscratch1);
4932     }
4933     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
4934   }
4935   authenticate_return_address();
4936 }
4937 
4938 
4939 // This method counts leading positive bytes (highest bit not set) in provided byte array
4940 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
4941     // Simple and most common case of aligned small array which is not at the
4942     // end of memory page is placed here. All other cases are in stub.
4943     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
4944     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
4945     assert_different_registers(ary1, len, result);
4946 
4947     mov(result, len);
4948     cmpw(len, 0);
4949     br(LE, DONE);
4950     cmpw(len, 4 * wordSize);
4951     br(GE, STUB_LONG); // size > 32 then go to stub
4952 
4953     int shift = 64 - exact_log2(os::vm_page_size());
4954     lsl(rscratch1, ary1, shift);
4955     mov(rscratch2, (size_t)(4 * wordSize) << shift);
4956     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
4957     br(CS, STUB); // at the end of page then go to stub
4958     subs(len, len, wordSize);
4959     br(LT, END);
4960 
4961   BIND(LOOP);
4962     ldr(rscratch1, Address(post(ary1, wordSize)));
4963     tst(rscratch1, UPPER_BIT_MASK);
4964     br(NE, SET_RESULT);
4965     subs(len, len, wordSize);
4966     br(GE, LOOP);
4967     cmpw(len, -wordSize);
4968     br(EQ, DONE);
4969 
4970   BIND(END);
4971     ldr(rscratch1, Address(ary1));
4972     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
4973     lslv(rscratch1, rscratch1, rscratch2);
4974     tst(rscratch1, UPPER_BIT_MASK);
4975     br(NE, SET_RESULT);
4976     b(DONE);
4977 
4978   BIND(STUB);
4979     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
4980     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
4981     address tpc1 = trampoline_call(count_pos);
4982     if (tpc1 == nullptr) {
4983       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
4984       postcond(pc() == badAddress);
4985       return nullptr;
4986     }
4987     b(DONE);
4988 
4989   BIND(STUB_LONG);
4990     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
4991     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
4992     address tpc2 = trampoline_call(count_pos_long);
4993     if (tpc2 == nullptr) {
4994       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
4995       postcond(pc() == badAddress);
4996       return nullptr;
4997     }
4998     b(DONE);
4999 
5000   BIND(SET_RESULT);
5001 
5002     add(len, len, wordSize);
5003     sub(result, result, len);
5004 
5005   BIND(DONE);
5006   postcond(pc() != badAddress);
5007   return pc();
5008 }
5009 
5010 // Clobbers: rscratch1, rscratch2, rflags
5011 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5012 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5013                                       Register tmp4, Register tmp5, Register result,
5014                                       Register cnt1, int elem_size) {
5015   Label DONE, SAME;
5016   Register tmp1 = rscratch1;
5017   Register tmp2 = rscratch2;
5018   Register cnt2 = tmp2;  // cnt2 only used in array length compare
5019   int elem_per_word = wordSize/elem_size;
5020   int log_elem_size = exact_log2(elem_size);
5021   int length_offset = arrayOopDesc::length_offset_in_bytes();
5022   int base_offset
5023     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5024   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5025 
5026   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5027   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5028 
5029 #ifndef PRODUCT
5030   {
5031     const char kind = (elem_size == 2) ? 'U' : 'L';
5032     char comment[64];
5033     snprintf(comment, sizeof comment, "array_equals%c{", kind);
5034     BLOCK_COMMENT(comment);
5035   }
5036 #endif
5037 
5038   // if (a1 == a2)
5039   //     return true;
5040   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5041   br(EQ, SAME);
5042 
5043   if (UseSimpleArrayEquals) {
5044     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5045     // if (a1 == nullptr || a2 == nullptr)
5046     //     return false;
5047     // a1 & a2 == 0 means (some-pointer is null) or
5048     // (very-rare-or-even-probably-impossible-pointer-values)
5049     // so, we can save one branch in most cases
5050     tst(a1, a2);
5051     mov(result, false);
5052     br(EQ, A_MIGHT_BE_NULL);
5053     // if (a1.length != a2.length)
5054     //      return false;
5055     bind(A_IS_NOT_NULL);
5056     ldrw(cnt1, Address(a1, length_offset));
5057     ldrw(cnt2, Address(a2, length_offset));
5058     eorw(tmp5, cnt1, cnt2);
5059     cbnzw(tmp5, DONE);
5060     lea(a1, Address(a1, base_offset));
5061     lea(a2, Address(a2, base_offset));
5062     // Check for short strings, i.e. smaller than wordSize.
5063     subs(cnt1, cnt1, elem_per_word);
5064     br(Assembler::LT, SHORT);
5065     // Main 8 byte comparison loop.
5066     bind(NEXT_WORD); {
5067       ldr(tmp1, Address(post(a1, wordSize)));
5068       ldr(tmp2, Address(post(a2, wordSize)));
5069       subs(cnt1, cnt1, elem_per_word);
5070       eor(tmp5, tmp1, tmp2);
5071       cbnz(tmp5, DONE);
5072     } br(GT, NEXT_WORD);
5073     // Last longword.  In the case where length == 4 we compare the
5074     // same longword twice, but that's still faster than another
5075     // conditional branch.
5076     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5077     // length == 4.
5078     if (log_elem_size > 0)
5079       lsl(cnt1, cnt1, log_elem_size);
5080     ldr(tmp3, Address(a1, cnt1));
5081     ldr(tmp4, Address(a2, cnt1));
5082     eor(tmp5, tmp3, tmp4);
5083     cbnz(tmp5, DONE);
5084     b(SAME);
5085     bind(A_MIGHT_BE_NULL);
5086     // in case both a1 and a2 are not-null, proceed with loads
5087     cbz(a1, DONE);
5088     cbz(a2, DONE);
5089     b(A_IS_NOT_NULL);
5090     bind(SHORT);
5091 
5092     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
5093     {
5094       ldrw(tmp1, Address(post(a1, 4)));
5095       ldrw(tmp2, Address(post(a2, 4)));
5096       eorw(tmp5, tmp1, tmp2);
5097       cbnzw(tmp5, DONE);
5098     }
5099     bind(TAIL03);
5100     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
5101     {
5102       ldrh(tmp3, Address(post(a1, 2)));
5103       ldrh(tmp4, Address(post(a2, 2)));
5104       eorw(tmp5, tmp3, tmp4);
5105       cbnzw(tmp5, DONE);
5106     }
5107     bind(TAIL01);
5108     if (elem_size == 1) { // Only needed when comparing byte arrays.
5109       tbz(cnt1, 0, SAME); // 0-1 bytes left.
5110       {
5111         ldrb(tmp1, a1);
5112         ldrb(tmp2, a2);
5113         eorw(tmp5, tmp1, tmp2);
5114         cbnzw(tmp5, DONE);
5115       }
5116     }
5117   } else {
5118     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
5119         CSET_EQ, LAST_CHECK;
5120     mov(result, false);
5121     cbz(a1, DONE);
5122     ldrw(cnt1, Address(a1, length_offset));
5123     cbz(a2, DONE);
5124     ldrw(cnt2, Address(a2, length_offset));
5125     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
5126     // faster to perform another branch before comparing a1 and a2
5127     cmp(cnt1, (u1)elem_per_word);
5128     br(LE, SHORT); // short or same
5129     ldr(tmp3, Address(pre(a1, base_offset)));
5130     subs(zr, cnt1, stubBytesThreshold);
5131     br(GE, STUB);
5132     ldr(tmp4, Address(pre(a2, base_offset)));
5133     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
5134     cmp(cnt2, cnt1);
5135     br(NE, DONE);
5136 
5137     // Main 16 byte comparison loop with 2 exits
5138     bind(NEXT_DWORD); {
5139       ldr(tmp1, Address(pre(a1, wordSize)));
5140       ldr(tmp2, Address(pre(a2, wordSize)));
5141       subs(cnt1, cnt1, 2 * elem_per_word);
5142       br(LE, TAIL);
5143       eor(tmp4, tmp3, tmp4);
5144       cbnz(tmp4, DONE);
5145       ldr(tmp3, Address(pre(a1, wordSize)));
5146       ldr(tmp4, Address(pre(a2, wordSize)));
5147       cmp(cnt1, (u1)elem_per_word);
5148       br(LE, TAIL2);
5149       cmp(tmp1, tmp2);
5150     } br(EQ, NEXT_DWORD);
5151     b(DONE);
5152 
5153     bind(TAIL);
5154     eor(tmp4, tmp3, tmp4);
5155     eor(tmp2, tmp1, tmp2);
5156     lslv(tmp2, tmp2, tmp5);
5157     orr(tmp5, tmp4, tmp2);
5158     cmp(tmp5, zr);
5159     b(CSET_EQ);
5160 
5161     bind(TAIL2);
5162     eor(tmp2, tmp1, tmp2);
5163     cbnz(tmp2, DONE);
5164     b(LAST_CHECK);
5165 
5166     bind(STUB);
5167     ldr(tmp4, Address(pre(a2, base_offset)));
5168     cmp(cnt2, cnt1);
5169     br(NE, DONE);
5170     if (elem_size == 2) { // convert to byte counter
5171       lsl(cnt1, cnt1, 1);
5172     }
5173     eor(tmp5, tmp3, tmp4);
5174     cbnz(tmp5, DONE);
5175     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
5176     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
5177     address tpc = trampoline_call(stub);
5178     if (tpc == nullptr) {
5179       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
5180       postcond(pc() == badAddress);
5181       return nullptr;
5182     }
5183     b(DONE);
5184 
5185     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
5186     // so, if a2 == null => return false(0), else return true, so we can return a2
5187     mov(result, a2);
5188     b(DONE);
5189     bind(SHORT);
5190     cmp(cnt2, cnt1);
5191     br(NE, DONE);
5192     cbz(cnt1, SAME);
5193     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
5194     ldr(tmp3, Address(a1, base_offset));
5195     ldr(tmp4, Address(a2, base_offset));
5196     bind(LAST_CHECK);
5197     eor(tmp4, tmp3, tmp4);
5198     lslv(tmp5, tmp4, tmp5);
5199     cmp(tmp5, zr);
5200     bind(CSET_EQ);
5201     cset(result, EQ);
5202     b(DONE);
5203   }
5204 
5205   bind(SAME);
5206   mov(result, true);
5207   // That's it.
5208   bind(DONE);
5209 
5210   BLOCK_COMMENT("} array_equals");
5211   postcond(pc() != badAddress);
5212   return pc();
5213 }
5214 
5215 // Compare Strings
5216 
5217 // For Strings we're passed the address of the first characters in a1
5218 // and a2 and the length in cnt1.
5219 // elem_size is the element size in bytes: either 1 or 2.
5220 // There are two implementations.  For arrays >= 8 bytes, all
5221 // comparisons (including the final one, which may overlap) are
5222 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
5223 // halfword, then a short, and then a byte.
5224 
5225 void MacroAssembler::string_equals(Register a1, Register a2,
5226                                    Register result, Register cnt1, int elem_size)
5227 {
5228   Label SAME, DONE, SHORT, NEXT_WORD;
5229   Register tmp1 = rscratch1;
5230   Register tmp2 = rscratch2;
5231   Register cnt2 = tmp2;  // cnt2 only used in array length compare
5232 
5233   assert(elem_size == 1 || elem_size == 2, "must be 2 or 1 byte");
5234   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5235 
5236 #ifndef PRODUCT
5237   {
5238     const char kind = (elem_size == 2) ? 'U' : 'L';
5239     char comment[64];
5240     snprintf(comment, sizeof comment, "{string_equals%c", kind);
5241     BLOCK_COMMENT(comment);
5242   }
5243 #endif
5244 
5245   mov(result, false);
5246 
5247   // Check for short strings, i.e. smaller than wordSize.
5248   subs(cnt1, cnt1, wordSize);
5249   br(Assembler::LT, SHORT);
5250   // Main 8 byte comparison loop.
5251   bind(NEXT_WORD); {
5252     ldr(tmp1, Address(post(a1, wordSize)));
5253     ldr(tmp2, Address(post(a2, wordSize)));
5254     subs(cnt1, cnt1, wordSize);
5255     eor(tmp1, tmp1, tmp2);
5256     cbnz(tmp1, DONE);
5257   } br(GT, NEXT_WORD);
5258   // Last longword.  In the case where length == 4 we compare the
5259   // same longword twice, but that's still faster than another
5260   // conditional branch.
5261   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5262   // length == 4.
5263   ldr(tmp1, Address(a1, cnt1));
5264   ldr(tmp2, Address(a2, cnt1));
5265   eor(tmp2, tmp1, tmp2);
5266   cbnz(tmp2, DONE);
5267   b(SAME);
5268 
5269   bind(SHORT);
5270   Label TAIL03, TAIL01;
5271 
5272   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
5273   {
5274     ldrw(tmp1, Address(post(a1, 4)));
5275     ldrw(tmp2, Address(post(a2, 4)));
5276     eorw(tmp1, tmp1, tmp2);
5277     cbnzw(tmp1, DONE);
5278   }
5279   bind(TAIL03);
5280   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
5281   {
5282     ldrh(tmp1, Address(post(a1, 2)));
5283     ldrh(tmp2, Address(post(a2, 2)));
5284     eorw(tmp1, tmp1, tmp2);
5285     cbnzw(tmp1, DONE);
5286   }
5287   bind(TAIL01);
5288   if (elem_size == 1) { // Only needed when comparing 1-byte elements
5289     tbz(cnt1, 0, SAME); // 0-1 bytes left.
5290     {
5291       ldrb(tmp1, a1);
5292       ldrb(tmp2, a2);
5293       eorw(tmp1, tmp1, tmp2);
5294       cbnzw(tmp1, DONE);
5295     }
5296   }
5297   // Arrays are equal.
5298   bind(SAME);
5299   mov(result, true);
5300 
5301   // That's it.
5302   bind(DONE);
5303   BLOCK_COMMENT("} string_equals");
5304 }
5305 
5306 
5307 // The size of the blocks erased by the zero_blocks stub.  We must
5308 // handle anything smaller than this ourselves in zero_words().
5309 const int MacroAssembler::zero_words_block_size = 8;
5310 
5311 // zero_words() is used by C2 ClearArray patterns and by
5312 // C1_MacroAssembler.  It is as small as possible, handling small word
5313 // counts locally and delegating anything larger to the zero_blocks
5314 // stub.  It is expanded many times in compiled code, so it is
5315 // important to keep it short.
5316 
5317 // ptr:   Address of a buffer to be zeroed.
5318 // cnt:   Count in HeapWords.
5319 //
5320 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
5321 address MacroAssembler::zero_words(Register ptr, Register cnt)
5322 {
5323   assert(is_power_of_2(zero_words_block_size), "adjust this");
5324 
5325   BLOCK_COMMENT("zero_words {");
5326   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
5327   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
5328   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
5329 
5330   subs(rscratch1, cnt, zero_words_block_size);
5331   Label around;
5332   br(LO, around);
5333   {
5334     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
5335     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
5336     // Make sure this is a C2 compilation. C1 allocates space only for
5337     // trampoline stubs generated by Call LIR ops, and in any case it
5338     // makes sense for a C1 compilation task to proceed as quickly as
5339     // possible.
5340     CompileTask* task;
5341     if (StubRoutines::aarch64::complete()
5342         && Thread::current()->is_Compiler_thread()
5343         && (task = ciEnv::current()->task())
5344         && is_c2_compile(task->comp_level())) {
5345       address tpc = trampoline_call(zero_blocks);
5346       if (tpc == nullptr) {
5347         DEBUG_ONLY(reset_labels(around));
5348         return nullptr;
5349       }
5350     } else {
5351       far_call(zero_blocks);
5352     }
5353   }
5354   bind(around);
5355 
5356   // We have a few words left to do. zero_blocks has adjusted r10 and r11
5357   // for us.
5358   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
5359     Label l;
5360     tbz(cnt, exact_log2(i), l);
5361     for (int j = 0; j < i; j += 2) {
5362       stp(zr, zr, post(ptr, 2 * BytesPerWord));
5363     }
5364     bind(l);
5365   }
5366   {
5367     Label l;
5368     tbz(cnt, 0, l);
5369     str(zr, Address(ptr));
5370     bind(l);
5371   }
5372 
5373   BLOCK_COMMENT("} zero_words");
5374   return pc();
5375 }
5376 
5377 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
5378 // cnt:          Immediate count in HeapWords.
5379 //
5380 // r10, r11, rscratch1, and rscratch2 are clobbered.
5381 address MacroAssembler::zero_words(Register base, uint64_t cnt)
5382 {
5383   assert(wordSize <= BlockZeroingLowLimit,
5384             "increase BlockZeroingLowLimit");
5385   address result = nullptr;
5386   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
5387 #ifndef PRODUCT
5388     {
5389       char buf[64];
5390       snprintf(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
5391       BLOCK_COMMENT(buf);
5392     }
5393 #endif
5394     if (cnt >= 16) {
5395       uint64_t loops = cnt/16;
5396       if (loops > 1) {
5397         mov(rscratch2, loops - 1);
5398       }
5399       {
5400         Label loop;
5401         bind(loop);
5402         for (int i = 0; i < 16; i += 2) {
5403           stp(zr, zr, Address(base, i * BytesPerWord));
5404         }
5405         add(base, base, 16 * BytesPerWord);
5406         if (loops > 1) {
5407           subs(rscratch2, rscratch2, 1);
5408           br(GE, loop);
5409         }
5410       }
5411     }
5412     cnt %= 16;
5413     int i = cnt & 1;  // store any odd word to start
5414     if (i) str(zr, Address(base));
5415     for (; i < (int)cnt; i += 2) {
5416       stp(zr, zr, Address(base, i * wordSize));
5417     }
5418     BLOCK_COMMENT("} zero_words");
5419     result = pc();
5420   } else {
5421     mov(r10, base); mov(r11, cnt);
5422     result = zero_words(r10, r11);
5423   }
5424   return result;
5425 }
5426 
5427 // Zero blocks of memory by using DC ZVA.
5428 //
5429 // Aligns the base address first sufficiently for DC ZVA, then uses
5430 // DC ZVA repeatedly for every full block.  cnt is the size to be
5431 // zeroed in HeapWords.  Returns the count of words left to be zeroed
5432 // in cnt.
5433 //
5434 // NOTE: This is intended to be used in the zero_blocks() stub.  If
5435 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
5436 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
5437   Register tmp = rscratch1;
5438   Register tmp2 = rscratch2;
5439   int zva_length = VM_Version::zva_length();
5440   Label initial_table_end, loop_zva;
5441   Label fini;
5442 
5443   // Base must be 16 byte aligned. If not just return and let caller handle it
5444   tst(base, 0x0f);
5445   br(Assembler::NE, fini);
5446   // Align base with ZVA length.
5447   neg(tmp, base);
5448   andr(tmp, tmp, zva_length - 1);
5449 
5450   // tmp: the number of bytes to be filled to align the base with ZVA length.
5451   add(base, base, tmp);
5452   sub(cnt, cnt, tmp, Assembler::ASR, 3);
5453   adr(tmp2, initial_table_end);
5454   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
5455   br(tmp2);
5456 
5457   for (int i = -zva_length + 16; i < 0; i += 16)
5458     stp(zr, zr, Address(base, i));
5459   bind(initial_table_end);
5460 
5461   sub(cnt, cnt, zva_length >> 3);
5462   bind(loop_zva);
5463   dc(Assembler::ZVA, base);
5464   subs(cnt, cnt, zva_length >> 3);
5465   add(base, base, zva_length);
5466   br(Assembler::GE, loop_zva);
5467   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
5468   bind(fini);
5469 }
5470 
5471 // base:   Address of a buffer to be filled, 8 bytes aligned.
5472 // cnt:    Count in 8-byte unit.
5473 // value:  Value to be filled with.
5474 // base will point to the end of the buffer after filling.
5475 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
5476 {
5477 //  Algorithm:
5478 //
5479 //    if (cnt == 0) {
5480 //      return;
5481 //    }
5482 //    if ((p & 8) != 0) {
5483 //      *p++ = v;
5484 //    }
5485 //
5486 //    scratch1 = cnt & 14;
5487 //    cnt -= scratch1;
5488 //    p += scratch1;
5489 //    switch (scratch1 / 2) {
5490 //      do {
5491 //        cnt -= 16;
5492 //          p[-16] = v;
5493 //          p[-15] = v;
5494 //        case 7:
5495 //          p[-14] = v;
5496 //          p[-13] = v;
5497 //        case 6:
5498 //          p[-12] = v;
5499 //          p[-11] = v;
5500 //          // ...
5501 //        case 1:
5502 //          p[-2] = v;
5503 //          p[-1] = v;
5504 //        case 0:
5505 //          p += 16;
5506 //      } while (cnt);
5507 //    }
5508 //    if ((cnt & 1) == 1) {
5509 //      *p++ = v;
5510 //    }
5511 
5512   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
5513 
5514   Label fini, skip, entry, loop;
5515   const int unroll = 8; // Number of stp instructions we'll unroll
5516 
5517   cbz(cnt, fini);
5518   tbz(base, 3, skip);
5519   str(value, Address(post(base, 8)));
5520   sub(cnt, cnt, 1);
5521   bind(skip);
5522 
5523   andr(rscratch1, cnt, (unroll-1) * 2);
5524   sub(cnt, cnt, rscratch1);
5525   add(base, base, rscratch1, Assembler::LSL, 3);
5526   adr(rscratch2, entry);
5527   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
5528   br(rscratch2);
5529 
5530   bind(loop);
5531   add(base, base, unroll * 16);
5532   for (int i = -unroll; i < 0; i++)
5533     stp(value, value, Address(base, i * 16));
5534   bind(entry);
5535   subs(cnt, cnt, unroll * 2);
5536   br(Assembler::GE, loop);
5537 
5538   tbz(cnt, 0, fini);
5539   str(value, Address(post(base, 8)));
5540   bind(fini);
5541 }
5542 
5543 // Intrinsic for
5544 //
5545 // - sun/nio/cs/ISO_8859_1$Encoder.implEncodeISOArray
5546 //     return the number of characters copied.
5547 // - java/lang/StringUTF16.compress
5548 //     return zero (0) if copy fails, otherwise 'len'.
5549 //
5550 // This version always returns the number of characters copied, and does not
5551 // clobber the 'len' register. A successful copy will complete with the post-
5552 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
5553 // post-condition: 0 <= 'res' < 'len'.
5554 //
5555 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
5556 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
5557 //       beyond the acceptable, even though the footprint would be smaller.
5558 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
5559 //       avoid additional bloat.
5560 //
5561 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
5562 void MacroAssembler::encode_iso_array(Register src, Register dst,
5563                                       Register len, Register res, bool ascii,
5564                                       FloatRegister vtmp0, FloatRegister vtmp1,
5565                                       FloatRegister vtmp2, FloatRegister vtmp3,
5566                                       FloatRegister vtmp4, FloatRegister vtmp5)
5567 {
5568   Register cnt = res;
5569   Register max = rscratch1;
5570   Register chk = rscratch2;
5571 
5572   prfm(Address(src), PLDL1STRM);
5573   movw(cnt, len);
5574 
5575 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
5576 
5577   Label LOOP_32, DONE_32, FAIL_32;
5578 
5579   BIND(LOOP_32);
5580   {
5581     cmpw(cnt, 32);
5582     br(LT, DONE_32);
5583     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
5584     // Extract lower bytes.
5585     FloatRegister vlo0 = vtmp4;
5586     FloatRegister vlo1 = vtmp5;
5587     uzp1(vlo0, T16B, vtmp0, vtmp1);
5588     uzp1(vlo1, T16B, vtmp2, vtmp3);
5589     // Merge bits...
5590     orr(vtmp0, T16B, vtmp0, vtmp1);
5591     orr(vtmp2, T16B, vtmp2, vtmp3);
5592     // Extract merged upper bytes.
5593     FloatRegister vhix = vtmp0;
5594     uzp2(vhix, T16B, vtmp0, vtmp2);
5595     // ISO-check on hi-parts (all zero).
5596     //                          ASCII-check on lo-parts (no sign).
5597     FloatRegister vlox = vtmp1; // Merge lower bytes.
5598                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
5599     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
5600     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
5601     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
5602                                 ASCII(orr(chk, chk, max));
5603     cbnz(chk, FAIL_32);
5604     subw(cnt, cnt, 32);
5605     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
5606     b(LOOP_32);
5607   }
5608   BIND(FAIL_32);
5609   sub(src, src, 64);
5610   BIND(DONE_32);
5611 
5612   Label LOOP_8, SKIP_8;
5613 
5614   BIND(LOOP_8);
5615   {
5616     cmpw(cnt, 8);
5617     br(LT, SKIP_8);
5618     FloatRegister vhi = vtmp0;
5619     FloatRegister vlo = vtmp1;
5620     ld1(vtmp3, T8H, src);
5621     uzp1(vlo, T16B, vtmp3, vtmp3);
5622     uzp2(vhi, T16B, vtmp3, vtmp3);
5623     // ISO-check on hi-parts (all zero).
5624     //                          ASCII-check on lo-parts (no sign).
5625                                 ASCII(cm(LT, vtmp2, T16B, vlo));
5626     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
5627                                 ASCII(umov(max, vtmp2, B, 0));
5628                                 ASCII(orr(chk, chk, max));
5629     cbnz(chk, SKIP_8);
5630 
5631     strd(vlo, Address(post(dst, 8)));
5632     subw(cnt, cnt, 8);
5633     add(src, src, 16);
5634     b(LOOP_8);
5635   }
5636   BIND(SKIP_8);
5637 
5638 #undef ASCII
5639 
5640   Label LOOP, DONE;
5641 
5642   cbz(cnt, DONE);
5643   BIND(LOOP);
5644   {
5645     Register chr = rscratch1;
5646     ldrh(chr, Address(post(src, 2)));
5647     tst(chr, ascii ? 0xff80 : 0xff00);
5648     br(NE, DONE);
5649     strb(chr, Address(post(dst, 1)));
5650     subs(cnt, cnt, 1);
5651     br(GT, LOOP);
5652   }
5653   BIND(DONE);
5654   // Return index where we stopped.
5655   subw(res, len, cnt);
5656 }
5657 
5658 // Inflate byte[] array to char[].
5659 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
5660 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
5661                                            FloatRegister vtmp1, FloatRegister vtmp2,
5662                                            FloatRegister vtmp3, Register tmp4) {
5663   Label big, done, after_init, to_stub;
5664 
5665   assert_different_registers(src, dst, len, tmp4, rscratch1);
5666 
5667   fmovd(vtmp1, 0.0);
5668   lsrw(tmp4, len, 3);
5669   bind(after_init);
5670   cbnzw(tmp4, big);
5671   // Short string: less than 8 bytes.
5672   {
5673     Label loop, tiny;
5674 
5675     cmpw(len, 4);
5676     br(LT, tiny);
5677     // Use SIMD to do 4 bytes.
5678     ldrs(vtmp2, post(src, 4));
5679     zip1(vtmp3, T8B, vtmp2, vtmp1);
5680     subw(len, len, 4);
5681     strd(vtmp3, post(dst, 8));
5682 
5683     cbzw(len, done);
5684 
5685     // Do the remaining bytes by steam.
5686     bind(loop);
5687     ldrb(tmp4, post(src, 1));
5688     strh(tmp4, post(dst, 2));
5689     subw(len, len, 1);
5690 
5691     bind(tiny);
5692     cbnz(len, loop);
5693 
5694     b(done);
5695   }
5696 
5697   if (SoftwarePrefetchHintDistance >= 0) {
5698     bind(to_stub);
5699       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
5700       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
5701       address tpc = trampoline_call(stub);
5702       if (tpc == nullptr) {
5703         DEBUG_ONLY(reset_labels(big, done));
5704         postcond(pc() == badAddress);
5705         return nullptr;
5706       }
5707       b(after_init);
5708   }
5709 
5710   // Unpack the bytes 8 at a time.
5711   bind(big);
5712   {
5713     Label loop, around, loop_last, loop_start;
5714 
5715     if (SoftwarePrefetchHintDistance >= 0) {
5716       const int large_loop_threshold = (64 + 16)/8;
5717       ldrd(vtmp2, post(src, 8));
5718       andw(len, len, 7);
5719       cmp(tmp4, (u1)large_loop_threshold);
5720       br(GE, to_stub);
5721       b(loop_start);
5722 
5723       bind(loop);
5724       ldrd(vtmp2, post(src, 8));
5725       bind(loop_start);
5726       subs(tmp4, tmp4, 1);
5727       br(EQ, loop_last);
5728       zip1(vtmp2, T16B, vtmp2, vtmp1);
5729       ldrd(vtmp3, post(src, 8));
5730       st1(vtmp2, T8H, post(dst, 16));
5731       subs(tmp4, tmp4, 1);
5732       zip1(vtmp3, T16B, vtmp3, vtmp1);
5733       st1(vtmp3, T8H, post(dst, 16));
5734       br(NE, loop);
5735       b(around);
5736       bind(loop_last);
5737       zip1(vtmp2, T16B, vtmp2, vtmp1);
5738       st1(vtmp2, T8H, post(dst, 16));
5739       bind(around);
5740       cbz(len, done);
5741     } else {
5742       andw(len, len, 7);
5743       bind(loop);
5744       ldrd(vtmp2, post(src, 8));
5745       sub(tmp4, tmp4, 1);
5746       zip1(vtmp3, T16B, vtmp2, vtmp1);
5747       st1(vtmp3, T8H, post(dst, 16));
5748       cbnz(tmp4, loop);
5749     }
5750   }
5751 
5752   // Do the tail of up to 8 bytes.
5753   add(src, src, len);
5754   ldrd(vtmp3, Address(src, -8));
5755   add(dst, dst, len, ext::uxtw, 1);
5756   zip1(vtmp3, T16B, vtmp3, vtmp1);
5757   strq(vtmp3, Address(dst, -16));
5758 
5759   bind(done);
5760   postcond(pc() != badAddress);
5761   return pc();
5762 }
5763 
5764 // Compress char[] array to byte[].
5765 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
5766                                          Register res,
5767                                          FloatRegister tmp0, FloatRegister tmp1,
5768                                          FloatRegister tmp2, FloatRegister tmp3,
5769                                          FloatRegister tmp4, FloatRegister tmp5) {
5770   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
5771   // Adjust result: res == len ? len : 0
5772   cmp(len, res);
5773   csel(res, res, zr, EQ);
5774 }
5775 
5776 // java.math.round(double a)
5777 // Returns the closest long to the argument, with ties rounding to
5778 // positive infinity.  This requires some fiddling for corner
5779 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
5780 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
5781                                        FloatRegister ftmp) {
5782   Label DONE;
5783   BLOCK_COMMENT("java_round_double: { ");
5784   fmovd(rscratch1, src);
5785   // Use RoundToNearestTiesAway unless src small and -ve.
5786   fcvtasd(dst, src);
5787   // Test if src >= 0 || abs(src) >= 0x1.0p52
5788   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
5789   mov(rscratch2, julong_cast(0x1.0p52));
5790   cmp(rscratch1, rscratch2);
5791   br(HS, DONE); {
5792     // src < 0 && abs(src) < 0x1.0p52
5793     // src may have a fractional part, so add 0.5
5794     fmovd(ftmp, 0.5);
5795     faddd(ftmp, src, ftmp);
5796     // Convert double to jlong, use RoundTowardsNegative
5797     fcvtmsd(dst, ftmp);
5798   }
5799   bind(DONE);
5800   BLOCK_COMMENT("} java_round_double");
5801 }
5802 
5803 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
5804                                       FloatRegister ftmp) {
5805   Label DONE;
5806   BLOCK_COMMENT("java_round_float: { ");
5807   fmovs(rscratch1, src);
5808   // Use RoundToNearestTiesAway unless src small and -ve.
5809   fcvtassw(dst, src);
5810   // Test if src >= 0 || abs(src) >= 0x1.0p23
5811   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
5812   mov(rscratch2, jint_cast(0x1.0p23f));
5813   cmp(rscratch1, rscratch2);
5814   br(HS, DONE); {
5815     // src < 0 && |src| < 0x1.0p23
5816     // src may have a fractional part, so add 0.5
5817     fmovs(ftmp, 0.5f);
5818     fadds(ftmp, src, ftmp);
5819     // Convert float to jint, use RoundTowardsNegative
5820     fcvtmssw(dst, ftmp);
5821   }
5822   bind(DONE);
5823   BLOCK_COMMENT("} java_round_float");
5824 }
5825 
5826 // get_thread() can be called anywhere inside generated code so we
5827 // need to save whatever non-callee save context might get clobbered
5828 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
5829 // the call setup code.
5830 //
5831 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
5832 // On other systems, the helper is a usual C function.
5833 //
5834 void MacroAssembler::get_thread(Register dst) {
5835   RegSet saved_regs =
5836     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
5837     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
5838 
5839   protect_return_address();
5840   push(saved_regs, sp);
5841 
5842   mov(lr, CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper));
5843   blr(lr);
5844   if (dst != c_rarg0) {
5845     mov(dst, c_rarg0);
5846   }
5847 
5848   pop(saved_regs, sp);
5849   authenticate_return_address();
5850 }
5851 
5852 void MacroAssembler::cache_wb(Address line) {
5853   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
5854   assert(line.index() == noreg, "index should be noreg");
5855   assert(line.offset() == 0, "offset should be 0");
5856   // would like to assert this
5857   // assert(line._ext.shift == 0, "shift should be zero");
5858   if (VM_Version::supports_dcpop()) {
5859     // writeback using clear virtual address to point of persistence
5860     dc(Assembler::CVAP, line.base());
5861   } else {
5862     // no need to generate anything as Unsafe.writebackMemory should
5863     // never invoke this stub
5864   }
5865 }
5866 
5867 void MacroAssembler::cache_wbsync(bool is_pre) {
5868   // we only need a barrier post sync
5869   if (!is_pre) {
5870     membar(Assembler::AnyAny);
5871   }
5872 }
5873 
5874 void MacroAssembler::verify_sve_vector_length(Register tmp) {
5875   // Make sure that native code does not change SVE vector length.
5876   if (!UseSVE) return;
5877   Label verify_ok;
5878   movw(tmp, zr);
5879   sve_inc(tmp, B);
5880   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
5881   br(EQ, verify_ok);
5882   stop("Error: SVE vector length has changed since jvm startup");
5883   bind(verify_ok);
5884 }
5885 
5886 void MacroAssembler::verify_ptrue() {
5887   Label verify_ok;
5888   if (!UseSVE) {
5889     return;
5890   }
5891   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
5892   sve_dec(rscratch1, B);
5893   cbz(rscratch1, verify_ok);
5894   stop("Error: the preserved predicate register (p7) elements are not all true");
5895   bind(verify_ok);
5896 }
5897 
5898 void MacroAssembler::safepoint_isb() {
5899   isb();
5900 #ifndef PRODUCT
5901   if (VerifyCrossModifyFence) {
5902     // Clear the thread state.
5903     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
5904   }
5905 #endif
5906 }
5907 
5908 #ifndef PRODUCT
5909 void MacroAssembler::verify_cross_modify_fence_not_required() {
5910   if (VerifyCrossModifyFence) {
5911     // Check if thread needs a cross modify fence.
5912     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
5913     Label fence_not_required;
5914     cbz(rscratch1, fence_not_required);
5915     // If it does then fail.
5916     lea(rscratch1, CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure));
5917     mov(c_rarg0, rthread);
5918     blr(rscratch1);
5919     bind(fence_not_required);
5920   }
5921 }
5922 #endif
5923 
5924 void MacroAssembler::spin_wait() {
5925   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
5926     switch (VM_Version::spin_wait_desc().inst()) {
5927       case SpinWait::NOP:
5928         nop();
5929         break;
5930       case SpinWait::ISB:
5931         isb();
5932         break;
5933       case SpinWait::YIELD:
5934         yield();
5935         break;
5936       default:
5937         ShouldNotReachHere();
5938     }
5939   }
5940 }
5941 
5942 // Stack frame creation/removal
5943 
5944 void MacroAssembler::enter(bool strip_ret_addr) {
5945   if (strip_ret_addr) {
5946     // Addresses can only be signed once. If there are multiple nested frames being created
5947     // in the same function, then the return address needs stripping first.
5948     strip_return_address();
5949   }
5950   protect_return_address();
5951   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5952   mov(rfp, sp);
5953 }
5954 
5955 void MacroAssembler::leave() {
5956   mov(sp, rfp);
5957   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5958   authenticate_return_address();
5959 }
5960 
5961 // ROP Protection
5962 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
5963 // destroying stack frames or whenever directly loading/storing the LR to memory.
5964 // If ROP protection is not set then these functions are no-ops.
5965 // For more details on PAC see pauth_aarch64.hpp.
5966 
5967 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
5968 // Uses the FP as the modifier.
5969 //
5970 void MacroAssembler::protect_return_address() {
5971   if (VM_Version::use_rop_protection()) {
5972     check_return_address();
5973     // The standard convention for C code is to use paciasp, which uses SP as the modifier. This
5974     // works because in C code, FP and SP match on function entry. In the JDK, SP and FP may not
5975     // match, so instead explicitly use the FP.
5976     pacia(lr, rfp);
5977   }
5978 }
5979 
5980 // Sign the return value in the given register. Use before updating the LR in the existing stack
5981 // frame for the current function.
5982 // Uses the FP from the start of the function as the modifier - which is stored at the address of
5983 // the current FP.
5984 //
5985 void MacroAssembler::protect_return_address(Register return_reg, Register temp_reg) {
5986   if (VM_Version::use_rop_protection()) {
5987     assert(PreserveFramePointer, "PreserveFramePointer must be set for ROP protection");
5988     check_return_address(return_reg);
5989     ldr(temp_reg, Address(rfp));
5990     pacia(return_reg, temp_reg);
5991   }
5992 }
5993 
5994 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
5995 //
5996 void MacroAssembler::authenticate_return_address(Register return_reg) {
5997   if (VM_Version::use_rop_protection()) {
5998     autia(return_reg, rfp);
5999     check_return_address(return_reg);
6000   }
6001 }
6002 
6003 // Authenticate the return value in the given register. Use before updating the LR in the existing
6004 // stack frame for the current function.
6005 // Uses the FP from the start of the function as the modifier - which is stored at the address of
6006 // the current FP.
6007 //
6008 void MacroAssembler::authenticate_return_address(Register return_reg, Register temp_reg) {
6009   if (VM_Version::use_rop_protection()) {
6010     assert(PreserveFramePointer, "PreserveFramePointer must be set for ROP protection");
6011     ldr(temp_reg, Address(rfp));
6012     autia(return_reg, temp_reg);
6013     check_return_address(return_reg);
6014   }
6015 }
6016 
6017 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6018 // there is no guaranteed way of authenticating the LR.
6019 //
6020 void MacroAssembler::strip_return_address() {
6021   if (VM_Version::use_rop_protection()) {
6022     xpaclri();
6023   }
6024 }
6025 
6026 #ifndef PRODUCT
6027 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6028 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6029 // it is difficult to debug back to the callee function.
6030 // This function simply loads from the address in the given register.
6031 // Use directly after authentication to catch authentication failures.
6032 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6033 //
6034 void MacroAssembler::check_return_address(Register return_reg) {
6035   if (VM_Version::use_rop_protection()) {
6036     ldr(zr, Address(return_reg));
6037   }
6038 }
6039 #endif
6040 
6041 // The java_calling_convention describes stack locations as ideal slots on
6042 // a frame with no abi restrictions. Since we must observe abi restrictions
6043 // (like the placement of the register window) the slots must be biased by
6044 // the following value.
6045 static int reg2offset_in(VMReg r) {
6046   // Account for saved rfp and lr
6047   // This should really be in_preserve_stack_slots
6048   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6049 }
6050 
6051 static int reg2offset_out(VMReg r) {
6052   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6053 }
6054 
6055 // On 64bit we will store integer like items to the stack as
6056 // 64bits items (AArch64 ABI) even though java would only store
6057 // 32bits for a parameter. On 32bit it will simply be 32bits
6058 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6059 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6060   if (src.first()->is_stack()) {
6061     if (dst.first()->is_stack()) {
6062       // stack to stack
6063       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6064       str(tmp, Address(sp, reg2offset_out(dst.first())));
6065     } else {
6066       // stack to reg
6067       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6068     }
6069   } else if (dst.first()->is_stack()) {
6070     // reg to stack
6071     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6072   } else {
6073     if (dst.first() != src.first()) {
6074       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6075     }
6076   }
6077 }
6078 
6079 // An oop arg. Must pass a handle not the oop itself
6080 void MacroAssembler::object_move(
6081                         OopMap* map,
6082                         int oop_handle_offset,
6083                         int framesize_in_slots,
6084                         VMRegPair src,
6085                         VMRegPair dst,
6086                         bool is_receiver,
6087                         int* receiver_offset) {
6088 
6089   // must pass a handle. First figure out the location we use as a handle
6090 
6091   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
6092 
6093   // See if oop is null if it is we need no handle
6094 
6095   if (src.first()->is_stack()) {
6096 
6097     // Oop is already on the stack as an argument
6098     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
6099     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
6100     if (is_receiver) {
6101       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
6102     }
6103 
6104     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
6105     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
6106     // conditionally move a null
6107     cmp(rscratch1, zr);
6108     csel(rHandle, zr, rHandle, Assembler::EQ);
6109   } else {
6110 
6111     // Oop is in an a register we must store it to the space we reserve
6112     // on the stack for oop_handles and pass a handle if oop is non-null
6113 
6114     const Register rOop = src.first()->as_Register();
6115     int oop_slot;
6116     if (rOop == j_rarg0)
6117       oop_slot = 0;
6118     else if (rOop == j_rarg1)
6119       oop_slot = 1;
6120     else if (rOop == j_rarg2)
6121       oop_slot = 2;
6122     else if (rOop == j_rarg3)
6123       oop_slot = 3;
6124     else if (rOop == j_rarg4)
6125       oop_slot = 4;
6126     else if (rOop == j_rarg5)
6127       oop_slot = 5;
6128     else if (rOop == j_rarg6)
6129       oop_slot = 6;
6130     else {
6131       assert(rOop == j_rarg7, "wrong register");
6132       oop_slot = 7;
6133     }
6134 
6135     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
6136     int offset = oop_slot*VMRegImpl::stack_slot_size;
6137 
6138     map->set_oop(VMRegImpl::stack2reg(oop_slot));
6139     // Store oop in handle area, may be null
6140     str(rOop, Address(sp, offset));
6141     if (is_receiver) {
6142       *receiver_offset = offset;
6143     }
6144 
6145     cmp(rOop, zr);
6146     lea(rHandle, Address(sp, offset));
6147     // conditionally move a null
6148     csel(rHandle, zr, rHandle, Assembler::EQ);
6149   }
6150 
6151   // If arg is on the stack then place it otherwise it is already in correct reg.
6152   if (dst.first()->is_stack()) {
6153     str(rHandle, Address(sp, reg2offset_out(dst.first())));
6154   }
6155 }
6156 
6157 // A float arg may have to do float reg int reg conversion
6158 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
6159  if (src.first()->is_stack()) {
6160     if (dst.first()->is_stack()) {
6161       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
6162       strw(tmp, Address(sp, reg2offset_out(dst.first())));
6163     } else {
6164       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
6165     }
6166   } else if (src.first() != dst.first()) {
6167     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
6168       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
6169     else
6170       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
6171   }
6172 }
6173 
6174 // A long move
6175 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
6176   if (src.first()->is_stack()) {
6177     if (dst.first()->is_stack()) {
6178       // stack to stack
6179       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6180       str(tmp, Address(sp, reg2offset_out(dst.first())));
6181     } else {
6182       // stack to reg
6183       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6184     }
6185   } else if (dst.first()->is_stack()) {
6186     // reg to stack
6187     // Do we really have to sign extend???
6188     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
6189     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6190   } else {
6191     if (dst.first() != src.first()) {
6192       mov(dst.first()->as_Register(), src.first()->as_Register());
6193     }
6194   }
6195 }
6196 
6197 
6198 // A double move
6199 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
6200  if (src.first()->is_stack()) {
6201     if (dst.first()->is_stack()) {
6202       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6203       str(tmp, Address(sp, reg2offset_out(dst.first())));
6204     } else {
6205       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
6206     }
6207   } else if (src.first() != dst.first()) {
6208     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
6209       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
6210     else
6211       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
6212   }
6213 }
6214 
6215 // Implements fast-locking.
6216 // Branches to slow upon failure to lock the object, with ZF cleared.
6217 // Falls through upon success with ZF set.
6218 //
6219 //  - obj: the object to be locked
6220 //  - hdr: the header, already loaded from obj, will be destroyed
6221 //  - t1, t2: temporary registers, will be destroyed
6222 void MacroAssembler::fast_lock(Register obj, Register hdr, Register t1, Register t2, Label& slow) {
6223   assert(LockingMode == LM_LIGHTWEIGHT, "only used with new lightweight locking");
6224   assert_different_registers(obj, hdr, t1, t2);
6225 
6226   // Check if we would have space on lock-stack for the object.
6227   ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
6228   cmpw(t1, (unsigned)LockStack::end_offset() - 1);
6229   br(Assembler::GT, slow);
6230 
6231   // Load (object->mark() | 1) into hdr
6232   orr(hdr, hdr, markWord::unlocked_value);
6233   // Clear lock-bits, into t2
6234   eor(t2, hdr, markWord::unlocked_value);
6235   // Try to swing header from unlocked to locked
6236   cmpxchg(/*addr*/ obj, /*expected*/ hdr, /*new*/ t2, Assembler::xword,
6237           /*acquire*/ true, /*release*/ true, /*weak*/ false, t1);
6238   br(Assembler::NE, slow);
6239 
6240   // After successful lock, push object on lock-stack
6241   ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
6242   str(obj, Address(rthread, t1));
6243   addw(t1, t1, oopSize);
6244   strw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
6245 }
6246 
6247 // Implements fast-unlocking.
6248 // Branches to slow upon failure, with ZF cleared.
6249 // Falls through upon success, with ZF set.
6250 //
6251 // - obj: the object to be unlocked
6252 // - hdr: the (pre-loaded) header of the object
6253 // - t1, t2: temporary registers
6254 void MacroAssembler::fast_unlock(Register obj, Register hdr, Register t1, Register t2, Label& slow) {
6255   assert(LockingMode == LM_LIGHTWEIGHT, "only used with new lightweight locking");
6256   assert_different_registers(obj, hdr, t1, t2);
6257 
6258 #ifdef ASSERT
6259   {
6260     // The following checks rely on the fact that LockStack is only ever modified by
6261     // its owning thread, even if the lock got inflated concurrently; removal of LockStack
6262     // entries after inflation will happen delayed in that case.
6263 
6264     // Check for lock-stack underflow.
6265     Label stack_ok;
6266     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
6267     cmpw(t1, (unsigned)LockStack::start_offset());
6268     br(Assembler::GT, stack_ok);
6269     STOP("Lock-stack underflow");
6270     bind(stack_ok);
6271   }
6272   {
6273     // Check if the top of the lock-stack matches the unlocked object.
6274     Label tos_ok;
6275     subw(t1, t1, oopSize);
6276     ldr(t1, Address(rthread, t1));
6277     cmpoop(t1, obj);
6278     br(Assembler::EQ, tos_ok);
6279     STOP("Top of lock-stack does not match the unlocked object");
6280     bind(tos_ok);
6281   }
6282   {
6283     // Check that hdr is fast-locked.
6284     Label hdr_ok;
6285     tst(hdr, markWord::lock_mask_in_place);
6286     br(Assembler::EQ, hdr_ok);
6287     STOP("Header is not fast-locked");
6288     bind(hdr_ok);
6289   }
6290 #endif
6291 
6292   // Load the new header (unlocked) into t1
6293   orr(t1, hdr, markWord::unlocked_value);
6294 
6295   // Try to swing header from locked to unlocked
6296   cmpxchg(obj, hdr, t1, Assembler::xword,
6297           /*acquire*/ true, /*release*/ true, /*weak*/ false, t2);
6298   br(Assembler::NE, slow);
6299 
6300   // After successful unlock, pop object from lock-stack
6301   ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
6302   subw(t1, t1, oopSize);
6303 #ifdef ASSERT
6304   str(zr, Address(rthread, t1));
6305 #endif
6306   strw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
6307 }