1 /*
   2  * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, 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 "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "ci/ciInlineKlass.hpp"
  30 #include "code/compiledIC.hpp"
  31 #include "compiler/compileTask.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/barrierSet.hpp"
  35 #include "gc/shared/barrierSetAssembler.hpp"
  36 #include "gc/shared/cardTableBarrierSet.hpp"
  37 #include "gc/shared/cardTable.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/tlab_globals.hpp"
  40 #include "interpreter/bytecodeHistogram.hpp"
  41 #include "interpreter/interpreter.hpp"
  42 #include "interpreter/interpreterRuntime.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/compressedKlass.inline.hpp"
  49 #include "oops/compressedOops.inline.hpp"
  50 #include "oops/klass.inline.hpp"
  51 #include "oops/resolvedFieldEntry.hpp"
  52 #include "runtime/continuation.hpp"
  53 #include "runtime/globals.hpp"
  54 #include "runtime/icache.hpp"
  55 #include "runtime/interfaceSupport.inline.hpp"
  56 #include "runtime/javaThread.hpp"
  57 #include "runtime/jniHandles.inline.hpp"
  58 #include "runtime/sharedRuntime.hpp"
  59 #include "runtime/signature_cc.hpp"
  60 #include "runtime/stubRoutines.hpp"
  61 #include "utilities/globalDefinitions.hpp"
  62 #include "utilities/powerOfTwo.hpp"
  63 #include "vmreg_aarch64.inline.hpp"
  64 #ifdef COMPILER1
  65 #include "c1/c1_LIRAssembler.hpp"
  66 #endif
  67 #ifdef COMPILER2
  68 #include "oops/oop.hpp"
  69 #include "opto/compile.hpp"
  70 #include "opto/node.hpp"
  71 #include "opto/output.hpp"
  72 #endif
  73 
  74 #include <sys/types.h>
  75 
  76 #ifdef PRODUCT
  77 #define BLOCK_COMMENT(str) /* nothing */
  78 #else
  79 #define BLOCK_COMMENT(str) block_comment(str)
  80 #endif
  81 #define STOP(str) stop(str);
  82 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  83 
  84 #ifdef ASSERT
  85 extern "C" void disnm(intptr_t p);
  86 #endif
  87 // Target-dependent relocation processing
  88 //
  89 // Instruction sequences whose target may need to be retrieved or
  90 // patched are distinguished by their leading instruction, sorting
  91 // them into three main instruction groups and related subgroups.
  92 //
  93 // 1) Branch, Exception and System (insn count = 1)
  94 //    1a) Unconditional branch (immediate):
  95 //      b/bl imm19
  96 //    1b) Compare & branch (immediate):
  97 //      cbz/cbnz Rt imm19
  98 //    1c) Test & branch (immediate):
  99 //      tbz/tbnz Rt imm14
 100 //    1d) Conditional branch (immediate):
 101 //      b.cond imm19
 102 //
 103 // 2) Loads and Stores (insn count = 1)
 104 //    2a) Load register literal:
 105 //      ldr Rt imm19
 106 //
 107 // 3) Data Processing Immediate (insn count = 2 or 3)
 108 //    3a) PC-rel. addressing
 109 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 110 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 111 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 112 //      adr/adrp Rx imm21
 113 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 114 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 115 //      The latter form can only happen when the target is an
 116 //      ExternalAddress, and (by definition) ExternalAddresses don't
 117 //      move. Because of that property, there is never any need to
 118 //      patch the last of the three instructions. However,
 119 //      MacroAssembler::target_addr_for_insn takes all three
 120 //      instructions into account and returns the correct address.
 121 //    3b) Move wide (immediate)
 122 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 123 //
 124 // A switch on a subset of the instruction's bits provides an
 125 // efficient dispatch to these subcases.
 126 //
 127 // insn[28:26] -> main group ('x' == don't care)
 128 //   00x -> UNALLOCATED
 129 //   100 -> Data Processing Immediate
 130 //   101 -> Branch, Exception and System
 131 //   x1x -> Loads and Stores
 132 //
 133 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 134 // n.b. in some cases extra bits need to be checked to verify the
 135 // instruction is as expected
 136 //
 137 // 1) ... xx101x Branch, Exception and System
 138 //   1a)  00___x Unconditional branch (immediate)
 139 //   1b)  01___0 Compare & branch (immediate)
 140 //   1c)  01___1 Test & branch (immediate)
 141 //   1d)  10___0 Conditional branch (immediate)
 142 //        other  Should not happen
 143 //
 144 // 2) ... xxx1x0 Loads and Stores
 145 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 146 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 147 //                strictly should be 64 bit non-FP/SIMD i.e.
 148 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 149 //
 150 // 3) ... xx100x Data Processing Immediate
 151 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 152 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 153 //                 strictly should be 64 bit movz #imm16<<0
 154 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 155 //
 156 class RelocActions {
 157 protected:
 158   typedef int (*reloc_insn)(address insn_addr, address &target);
 159 
 160   virtual reloc_insn adrpMem() = 0;
 161   virtual reloc_insn adrpAdd() = 0;
 162   virtual reloc_insn adrpMovk() = 0;
 163 
 164   const address _insn_addr;
 165   const uint32_t _insn;
 166 
 167   static uint32_t insn_at(address insn_addr, int n) {
 168     return ((uint32_t*)insn_addr)[n];
 169   }
 170   uint32_t insn_at(int n) const {
 171     return insn_at(_insn_addr, n);
 172   }
 173 
 174 public:
 175 
 176   RelocActions(address insn_addr) : _insn_addr(insn_addr), _insn(insn_at(insn_addr, 0)) {}
 177   RelocActions(address insn_addr, uint32_t insn)
 178     :  _insn_addr(insn_addr), _insn(insn) {}
 179 
 180   virtual int unconditionalBranch(address insn_addr, address &target) = 0;
 181   virtual int conditionalBranch(address insn_addr, address &target) = 0;
 182   virtual int testAndBranch(address insn_addr, address &target) = 0;
 183   virtual int loadStore(address insn_addr, address &target) = 0;
 184   virtual int adr(address insn_addr, address &target) = 0;
 185   virtual int adrp(address insn_addr, address &target, reloc_insn inner) = 0;
 186   virtual int immediate(address insn_addr, address &target) = 0;
 187   virtual void verify(address insn_addr, address &target) = 0;
 188 
 189   int ALWAYSINLINE run(address insn_addr, address &target) {
 190     int instructions = 1;
 191 
 192     uint32_t dispatch = Instruction_aarch64::extract(_insn, 30, 25);
 193     switch(dispatch) {
 194       case 0b001010:
 195       case 0b001011: {
 196         instructions = unconditionalBranch(insn_addr, target);
 197         break;
 198       }
 199       case 0b101010:   // Conditional branch (immediate)
 200       case 0b011010: { // Compare & branch (immediate)
 201         instructions = conditionalBranch(insn_addr, target);
 202           break;
 203       }
 204       case 0b011011: {
 205         instructions = testAndBranch(insn_addr, target);
 206         break;
 207       }
 208       case 0b001100:
 209       case 0b001110:
 210       case 0b011100:
 211       case 0b011110:
 212       case 0b101100:
 213       case 0b101110:
 214       case 0b111100:
 215       case 0b111110: {
 216         // load/store
 217         if ((Instruction_aarch64::extract(_insn, 29, 24) & 0b111011) == 0b011000) {
 218           // Load register (literal)
 219           instructions = loadStore(insn_addr, target);
 220           break;
 221         } else {
 222           // nothing to do
 223           assert(target == nullptr, "did not expect to relocate target for polling page load");
 224         }
 225         break;
 226       }
 227       case 0b001000:
 228       case 0b011000:
 229       case 0b101000:
 230       case 0b111000: {
 231         // adr/adrp
 232         assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 233         int shift = Instruction_aarch64::extract(_insn, 31, 31);
 234         if (shift) {
 235           uint32_t insn2 = insn_at(1);
 236           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 237               Instruction_aarch64::extract(_insn, 4, 0) ==
 238               Instruction_aarch64::extract(insn2, 9, 5)) {
 239             instructions = adrp(insn_addr, target, adrpMem());
 240           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 241                      Instruction_aarch64::extract(_insn, 4, 0) ==
 242                      Instruction_aarch64::extract(insn2, 4, 0)) {
 243             instructions = adrp(insn_addr, target, adrpAdd());
 244           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 245                      Instruction_aarch64::extract(_insn, 4, 0) ==
 246                      Instruction_aarch64::extract(insn2, 4, 0)) {
 247             instructions = adrp(insn_addr, target, adrpMovk());
 248           } else {
 249             ShouldNotReachHere();
 250           }
 251         } else {
 252           instructions = adr(insn_addr, target);
 253         }
 254         break;
 255       }
 256       case 0b001001:
 257       case 0b011001:
 258       case 0b101001:
 259       case 0b111001: {
 260         instructions = immediate(insn_addr, target);
 261         break;
 262       }
 263       default: {
 264         ShouldNotReachHere();
 265       }
 266     }
 267 
 268     verify(insn_addr, target);
 269     return instructions * NativeInstruction::instruction_size;
 270   }
 271 };
 272 
 273 class Patcher : public RelocActions {
 274   virtual reloc_insn adrpMem() { return &Patcher::adrpMem_impl; }
 275   virtual reloc_insn adrpAdd() { return &Patcher::adrpAdd_impl; }
 276   virtual reloc_insn adrpMovk() { return &Patcher::adrpMovk_impl; }
 277 
 278 public:
 279   Patcher(address insn_addr) : RelocActions(insn_addr) {}
 280 
 281   virtual int unconditionalBranch(address insn_addr, address &target) {
 282     intptr_t offset = (target - insn_addr) >> 2;
 283     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 284     return 1;
 285   }
 286   virtual int conditionalBranch(address insn_addr, address &target) {
 287     intptr_t offset = (target - insn_addr) >> 2;
 288     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 289     return 1;
 290   }
 291   virtual int testAndBranch(address insn_addr, address &target) {
 292     intptr_t offset = (target - insn_addr) >> 2;
 293     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 294     return 1;
 295   }
 296   virtual int loadStore(address insn_addr, address &target) {
 297     intptr_t offset = (target - insn_addr) >> 2;
 298     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 299     return 1;
 300   }
 301   virtual int adr(address insn_addr, address &target) {
 302 #ifdef ASSERT
 303     assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 304 #endif
 305     // PC-rel. addressing
 306     ptrdiff_t offset = target - insn_addr;
 307     int offset_lo = offset & 3;
 308     offset >>= 2;
 309     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 310     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 311     return 1;
 312   }
 313   virtual int adrp(address insn_addr, address &target, reloc_insn inner) {
 314     int instructions = 1;
 315 #ifdef ASSERT
 316     assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 317 #endif
 318     ptrdiff_t offset = target - insn_addr;
 319     instructions = 2;
 320     precond(inner != nullptr);
 321     // Give the inner reloc a chance to modify the target.
 322     address adjusted_target = target;
 323     instructions = (*inner)(insn_addr, adjusted_target);
 324     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 325     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 326     offset = adr_page - pc_page;
 327     int offset_lo = offset & 3;
 328     offset >>= 2;
 329     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 330     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 331     return instructions;
 332   }
 333   static int adrpMem_impl(address insn_addr, address &target) {
 334     uintptr_t dest = (uintptr_t)target;
 335     int offset_lo = dest & 0xfff;
 336     uint32_t insn2 = insn_at(insn_addr, 1);
 337     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 338     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 339     guarantee(((dest >> size) << size) == dest, "misaligned target");
 340     return 2;
 341   }
 342   static int adrpAdd_impl(address insn_addr, address &target) {
 343     uintptr_t dest = (uintptr_t)target;
 344     int offset_lo = dest & 0xfff;
 345     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 346     return 2;
 347   }
 348   static int adrpMovk_impl(address insn_addr, address &target) {
 349     uintptr_t dest = uintptr_t(target);
 350     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 351     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 352     target = address(dest);
 353     return 2;
 354   }
 355   virtual int immediate(address insn_addr, address &target) {
 356     assert(Instruction_aarch64::extract(_insn, 31, 21) == 0b11010010100, "must be");
 357     uint64_t dest = (uint64_t)target;
 358     // Move wide constant
 359     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 360     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 361     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 362     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 363     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 364     return 3;
 365   }
 366   virtual void verify(address insn_addr, address &target) {
 367 #ifdef ASSERT
 368     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 369     if (!(address_is == target)) {
 370       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 371       disnm((intptr_t)insn_addr);
 372       assert(address_is == target, "should be");
 373     }
 374 #endif
 375   }
 376 };
 377 
 378 // If insn1 and insn2 use the same register to form an address, either
 379 // by an offsetted LDR or a simple ADD, return the offset. If the
 380 // second instruction is an LDR, the offset may be scaled.
 381 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 382   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 383       Instruction_aarch64::extract(insn1, 4, 0) ==
 384       Instruction_aarch64::extract(insn2, 9, 5)) {
 385     // Load/store register (unsigned immediate)
 386     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 387     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 388     byte_offset <<= size;
 389     return true;
 390   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 391              Instruction_aarch64::extract(insn1, 4, 0) ==
 392              Instruction_aarch64::extract(insn2, 4, 0)) {
 393     // add (immediate)
 394     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 395     return true;
 396   }
 397   return false;
 398 }
 399 
 400 class AArch64Decoder : public RelocActions {
 401   virtual reloc_insn adrpMem() { return &AArch64Decoder::adrpMem_impl; }
 402   virtual reloc_insn adrpAdd() { return &AArch64Decoder::adrpAdd_impl; }
 403   virtual reloc_insn adrpMovk() { return &AArch64Decoder::adrpMovk_impl; }
 404 
 405 public:
 406   AArch64Decoder(address insn_addr, uint32_t insn) : RelocActions(insn_addr, insn) {}
 407 
 408   virtual int loadStore(address insn_addr, address &target) {
 409     intptr_t offset = Instruction_aarch64::sextract(_insn, 23, 5);
 410     target = insn_addr + (offset << 2);
 411     return 1;
 412   }
 413   virtual int unconditionalBranch(address insn_addr, address &target) {
 414     intptr_t offset = Instruction_aarch64::sextract(_insn, 25, 0);
 415     target = insn_addr + (offset << 2);
 416     return 1;
 417   }
 418   virtual int conditionalBranch(address insn_addr, address &target) {
 419     intptr_t offset = Instruction_aarch64::sextract(_insn, 23, 5);
 420     target = address(((uint64_t)insn_addr + (offset << 2)));
 421     return 1;
 422   }
 423   virtual int testAndBranch(address insn_addr, address &target) {
 424     intptr_t offset = Instruction_aarch64::sextract(_insn, 18, 5);
 425     target = address(((uint64_t)insn_addr + (offset << 2)));
 426     return 1;
 427   }
 428   virtual int adr(address insn_addr, address &target) {
 429     // PC-rel. addressing
 430     intptr_t offset = Instruction_aarch64::extract(_insn, 30, 29);
 431     offset |= Instruction_aarch64::sextract(_insn, 23, 5) << 2;
 432     target = address((uint64_t)insn_addr + offset);
 433     return 1;
 434   }
 435   virtual int adrp(address insn_addr, address &target, reloc_insn inner) {
 436     assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 437     intptr_t offset = Instruction_aarch64::extract(_insn, 30, 29);
 438     offset |= Instruction_aarch64::sextract(_insn, 23, 5) << 2;
 439     int shift = 12;
 440     offset <<= shift;
 441     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 442     target_page &= ((uint64_t)-1) << shift;
 443     uint32_t insn2 = insn_at(1);
 444     target = address(target_page);
 445     precond(inner != nullptr);
 446     (*inner)(insn_addr, target);
 447     return 2;
 448   }
 449   static int adrpMem_impl(address insn_addr, address &target) {
 450     uint32_t insn2 = insn_at(insn_addr, 1);
 451     // Load/store register (unsigned immediate)
 452     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 453     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 454     byte_offset <<= size;
 455     target += byte_offset;
 456     return 2;
 457   }
 458   static int adrpAdd_impl(address insn_addr, address &target) {
 459     uint32_t insn2 = insn_at(insn_addr, 1);
 460     // add (immediate)
 461     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 462     target += byte_offset;
 463     return 2;
 464   }
 465   static int adrpMovk_impl(address insn_addr, address &target) {
 466     uint32_t insn2 = insn_at(insn_addr, 1);
 467     uint64_t dest = uint64_t(target);
 468     dest = (dest & 0xffff0000ffffffff) |
 469       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 470     target = address(dest);
 471 
 472     // We know the destination 4k page. Maybe we have a third
 473     // instruction.
 474     uint32_t insn = insn_at(insn_addr, 0);
 475     uint32_t insn3 = insn_at(insn_addr, 2);
 476     ptrdiff_t byte_offset;
 477     if (offset_for(insn, insn3, byte_offset)) {
 478       target += byte_offset;
 479       return 3;
 480     } else {
 481       return 2;
 482     }
 483   }
 484   virtual int immediate(address insn_addr, address &target) {
 485     uint32_t *insns = (uint32_t *)insn_addr;
 486     assert(Instruction_aarch64::extract(_insn, 31, 21) == 0b11010010100, "must be");
 487     // Move wide constant: movz, movk, movk.  See movptr().
 488     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 489     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 490     target = address(uint64_t(Instruction_aarch64::extract(_insn, 20, 5))
 491                  + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 492                  + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 493     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 494     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 495     return 3;
 496   }
 497   virtual void verify(address insn_addr, address &target) {
 498   }
 499 };
 500 
 501 address MacroAssembler::target_addr_for_insn(address insn_addr, uint32_t insn) {
 502   AArch64Decoder decoder(insn_addr, insn);
 503   address target;
 504   decoder.run(insn_addr, target);
 505   return target;
 506 }
 507 
 508 // Patch any kind of instruction; there may be several instructions.
 509 // Return the total length (in bytes) of the instructions.
 510 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 511   Patcher patcher(insn_addr);
 512   return patcher.run(insn_addr, target);
 513 }
 514 
 515 int MacroAssembler::patch_oop(address insn_addr, address o) {
 516   int instructions;
 517   unsigned insn = *(unsigned*)insn_addr;
 518   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 519 
 520   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 521   // narrow OOPs by setting the upper 16 bits in the first
 522   // instruction.
 523   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 524     // Move narrow OOP
 525     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 526     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 527     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 528     instructions = 2;
 529   } else {
 530     // Move wide OOP
 531     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 532     uintptr_t dest = (uintptr_t)o;
 533     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 534     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 535     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 536     instructions = 3;
 537   }
 538   return instructions * NativeInstruction::instruction_size;
 539 }
 540 
 541 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 542   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 543   // We encode narrow ones by setting the upper 16 bits in the first
 544   // instruction.
 545   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 546   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 547          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 548 
 549   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 550   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 551   return 2 * NativeInstruction::instruction_size;
 552 }
 553 
 554 address MacroAssembler::target_addr_for_insn_or_null(address insn_addr, unsigned insn) {
 555   if (NativeInstruction::is_ldrw_to_zr(address(&insn))) {
 556     return nullptr;
 557   }
 558   return MacroAssembler::target_addr_for_insn(insn_addr, insn);
 559 }
 560 
 561 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 562   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 563   if (at_return) {
 564     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 565     // we may safely use the sp instead to perform the stack watermark check.
 566     cmp(in_nmethod ? sp : rfp, tmp);
 567     br(Assembler::HI, slow_path);
 568   } else {
 569     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 570   }
 571 }
 572 
 573 void MacroAssembler::rt_call(address dest, Register tmp) {
 574   CodeBlob *cb = CodeCache::find_blob(dest);
 575   if (cb) {
 576     far_call(RuntimeAddress(dest));
 577   } else {
 578     lea(tmp, RuntimeAddress(dest));
 579     blr(tmp);
 580   }
 581 }
 582 
 583 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 584   if (!Continuations::enabled()) return;
 585   Label done;
 586   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 587   cmp(sp, rscratch1);
 588   br(Assembler::LS, done);
 589   mov(rscratch1, sp); // we can't use sp as the source in str
 590   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 591   bind(done);
 592 }
 593 
 594 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 595   if (!Continuations::enabled()) return;
 596   Label done;
 597   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 598   cmp(sp, rscratch1);
 599   br(Assembler::LO, done);
 600   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 601   bind(done);
 602 }
 603 
 604 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 605   // we must set sp to zero to clear frame
 606   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 607 
 608   // must clear fp, so that compiled frames are not confused; it is
 609   // possible that we need it only for debugging
 610   if (clear_fp) {
 611     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 612   }
 613 
 614   // Always clear the pc because it could have been set by make_walkable()
 615   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 616 }
 617 
 618 // Calls to C land
 619 //
 620 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 621 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 622 // has to be reset to 0. This is required to allow proper stack traversal.
 623 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 624                                          Register last_java_fp,
 625                                          Register last_java_pc,
 626                                          Register scratch) {
 627 
 628   if (last_java_pc->is_valid()) {
 629       str(last_java_pc, Address(rthread,
 630                                 JavaThread::frame_anchor_offset()
 631                                 + JavaFrameAnchor::last_Java_pc_offset()));
 632     }
 633 
 634   // determine last_java_sp register
 635   if (last_java_sp == sp) {
 636     mov(scratch, sp);
 637     last_java_sp = scratch;
 638   } else if (!last_java_sp->is_valid()) {
 639     last_java_sp = esp;
 640   }
 641 
 642   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 643 
 644   // last_java_fp is optional
 645   if (last_java_fp->is_valid()) {
 646     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 647   }
 648 }
 649 
 650 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 651                                          Register last_java_fp,
 652                                          address  last_java_pc,
 653                                          Register scratch) {
 654   assert(last_java_pc != nullptr, "must provide a valid PC");
 655 
 656   adr(scratch, last_java_pc);
 657   str(scratch, Address(rthread,
 658                        JavaThread::frame_anchor_offset()
 659                        + JavaFrameAnchor::last_Java_pc_offset()));
 660 
 661   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 662 }
 663 
 664 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 665                                          Register last_java_fp,
 666                                          Label &L,
 667                                          Register scratch) {
 668   if (L.is_bound()) {
 669     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 670   } else {
 671     InstructionMark im(this);
 672     L.add_patch_at(code(), locator());
 673     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 674   }
 675 }
 676 
 677 static inline bool target_needs_far_branch(address addr) {
 678   if (AOTCodeCache::is_on_for_dump()) {
 679     return true;
 680   }
 681   // codecache size <= 128M
 682   if (!MacroAssembler::far_branches()) {
 683     return false;
 684   }
 685   // codecache size > 240M
 686   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 687     return true;
 688   }
 689   // codecache size: 128M..240M
 690   return !CodeCache::is_non_nmethod(addr);
 691 }
 692 
 693 void MacroAssembler::far_call(Address entry, Register tmp) {
 694   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 695   assert(CodeCache::find_blob(entry.target()) != nullptr,
 696          "destination of far call not found in code cache");
 697   assert(entry.rspec().type() == relocInfo::external_word_type
 698          || entry.rspec().type() == relocInfo::runtime_call_type
 699          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 700   if (target_needs_far_branch(entry.target())) {
 701     uint64_t offset;
 702     // We can use ADRP here because we know that the total size of
 703     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 704     adrp(tmp, entry, offset);
 705     add(tmp, tmp, offset);
 706     blr(tmp);
 707   } else {
 708     bl(entry);
 709   }
 710 }
 711 
 712 int MacroAssembler::far_jump(Address entry, Register tmp) {
 713   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 714   assert(CodeCache::find_blob(entry.target()) != nullptr,
 715          "destination of far call not found in code cache");
 716   assert(entry.rspec().type() == relocInfo::external_word_type
 717          || entry.rspec().type() == relocInfo::runtime_call_type
 718          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 719   address start = pc();
 720   if (target_needs_far_branch(entry.target())) {
 721     uint64_t offset;
 722     // We can use ADRP here because we know that the total size of
 723     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 724     adrp(tmp, entry, offset);
 725     add(tmp, tmp, offset);
 726     br(tmp);
 727   } else {
 728     b(entry);
 729   }
 730   return pc() - start;
 731 }
 732 
 733 void MacroAssembler::reserved_stack_check() {
 734     // testing if reserved zone needs to be enabled
 735     Label no_reserved_zone_enabling;
 736 
 737     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 738     cmp(sp, rscratch1);
 739     br(Assembler::LO, no_reserved_zone_enabling);
 740 
 741     enter();   // LR and FP are live.
 742     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 743     mov(c_rarg0, rthread);
 744     blr(rscratch1);
 745     leave();
 746 
 747     // We have already removed our own frame.
 748     // throw_delayed_StackOverflowError will think that it's been
 749     // called by our caller.
 750     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 751     br(rscratch1);
 752     should_not_reach_here();
 753 
 754     bind(no_reserved_zone_enabling);
 755 }
 756 
 757 static void pass_arg0(MacroAssembler* masm, Register arg) {
 758   if (c_rarg0 != arg ) {
 759     masm->mov(c_rarg0, arg);
 760   }
 761 }
 762 
 763 static void pass_arg1(MacroAssembler* masm, Register arg) {
 764   if (c_rarg1 != arg ) {
 765     masm->mov(c_rarg1, arg);
 766   }
 767 }
 768 
 769 static void pass_arg2(MacroAssembler* masm, Register arg) {
 770   if (c_rarg2 != arg ) {
 771     masm->mov(c_rarg2, arg);
 772   }
 773 }
 774 
 775 static void pass_arg3(MacroAssembler* masm, Register arg) {
 776   if (c_rarg3 != arg ) {
 777     masm->mov(c_rarg3, arg);
 778   }
 779 }
 780 
 781 static bool is_preemptable(address entry_point) {
 782   return entry_point == CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter);
 783 }
 784 
 785 void MacroAssembler::call_VM_base(Register oop_result,
 786                                   Register java_thread,
 787                                   Register last_java_sp,
 788                                   address  entry_point,
 789                                   int      number_of_arguments,
 790                                   bool     check_exceptions) {
 791    // determine java_thread register
 792   if (!java_thread->is_valid()) {
 793     java_thread = rthread;
 794   }
 795 
 796   // determine last_java_sp register
 797   if (!last_java_sp->is_valid()) {
 798     last_java_sp = esp;
 799   }
 800 
 801   // debugging support
 802   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 803   assert(java_thread == rthread, "unexpected register");
 804 #ifdef ASSERT
 805   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 806   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 807 #endif // ASSERT
 808 
 809   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 810   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 811 
 812   // push java thread (becomes first argument of C function)
 813 
 814   mov(c_rarg0, java_thread);
 815 
 816   // set last Java frame before call
 817   assert(last_java_sp != rfp, "can't use rfp");
 818 
 819   Label l;
 820   if (is_preemptable(entry_point)) {
 821     // skip setting last_pc since we already set it to desired value.
 822     set_last_Java_frame(last_java_sp, rfp, noreg, rscratch1);
 823   } else {
 824     set_last_Java_frame(last_java_sp, rfp, l, rscratch1);
 825   }
 826 
 827   // do the call, remove parameters
 828   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 829 
 830   // lr could be poisoned with PAC signature during throw_pending_exception
 831   // if it was tail-call optimized by compiler, since lr is not callee-saved
 832   // reload it with proper value
 833   adr(lr, l);
 834 
 835   // reset last Java frame
 836   // Only interpreter should have to clear fp
 837   reset_last_Java_frame(true);
 838 
 839    // C++ interp handles this in the interpreter
 840   check_and_handle_popframe(java_thread);
 841   check_and_handle_earlyret(java_thread);
 842 
 843   if (check_exceptions) {
 844     // check for pending exceptions (java_thread is set upon return)
 845     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 846     Label ok;
 847     cbz(rscratch1, ok);
 848     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 849     br(rscratch1);
 850     bind(ok);
 851   }
 852 
 853   // get oop result if there is one and reset the value in the thread
 854   if (oop_result->is_valid()) {
 855     get_vm_result_oop(oop_result, java_thread);
 856   }
 857 }
 858 
 859 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 860   call_VM_base(oop_result, noreg, noreg, entry_point, number_of_arguments, check_exceptions);
 861 }
 862 
 863 // Check the entry target is always reachable from any branch.
 864 static bool is_always_within_branch_range(Address entry) {
 865   if (AOTCodeCache::is_on_for_dump()) {
 866     return false;
 867   }
 868   const address target = entry.target();
 869 
 870   if (!CodeCache::contains(target)) {
 871     // We always use trampolines for callees outside CodeCache.
 872     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 873     return false;
 874   }
 875 
 876   if (!MacroAssembler::far_branches()) {
 877     return true;
 878   }
 879 
 880   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 881     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 882     // Non-compiled methods stay forever in CodeCache.
 883     // We check whether the longest possible branch is within the branch range.
 884     assert(CodeCache::find_blob(target) != nullptr &&
 885           !CodeCache::find_blob(target)->is_nmethod(),
 886           "runtime call of compiled method");
 887     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 888     const address left_longest_branch_start = CodeCache::low_bound();
 889     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 890                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 891     return is_reachable;
 892   }
 893 
 894   return false;
 895 }
 896 
 897 // Maybe emit a call via a trampoline. If the code cache is small
 898 // trampolines won't be emitted.
 899 address MacroAssembler::trampoline_call(Address entry) {
 900   assert(entry.rspec().type() == relocInfo::runtime_call_type
 901          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 902          || entry.rspec().type() == relocInfo::static_call_type
 903          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 904 
 905   address target = entry.target();
 906 
 907   if (!is_always_within_branch_range(entry)) {
 908     if (!in_scratch_emit_size()) {
 909       // We don't want to emit a trampoline if C2 is generating dummy
 910       // code during its branch shortening phase.
 911       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 912         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 913         code()->share_trampoline_for(entry.target(), offset());
 914       } else {
 915         address stub = emit_trampoline_stub(offset(), target);
 916         if (stub == nullptr) {
 917           postcond(pc() == badAddress);
 918           return nullptr; // CodeCache is full
 919         }
 920       }
 921     }
 922     target = pc();
 923   }
 924 
 925   address call_pc = pc();
 926   relocate(entry.rspec());
 927   bl(target);
 928 
 929   postcond(pc() != badAddress);
 930   return call_pc;
 931 }
 932 
 933 // Emit a trampoline stub for a call to a target which is too far away.
 934 //
 935 // code sequences:
 936 //
 937 // call-site:
 938 //   branch-and-link to <destination> or <trampoline stub>
 939 //
 940 // Related trampoline stub for this call site in the stub section:
 941 //   load the call target from the constant pool
 942 //   branch (LR still points to the call site above)
 943 
 944 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 945                                              address dest) {
 946   // Max stub size: alignment nop, TrampolineStub.
 947   address stub = start_a_stub(max_trampoline_stub_size());
 948   if (stub == nullptr) {
 949     return nullptr;  // CodeBuffer::expand failed
 950   }
 951 
 952   // Create a trampoline stub relocation which relates this trampoline stub
 953   // with the call instruction at insts_call_instruction_offset in the
 954   // instructions code-section.
 955   align(wordSize);
 956   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 957                                             + insts_call_instruction_offset));
 958   const int stub_start_offset = offset();
 959 
 960   // Now, create the trampoline stub's code:
 961   // - load the call
 962   // - call
 963   Label target;
 964   ldr(rscratch1, target);
 965   br(rscratch1);
 966   bind(target);
 967   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 968          "should be");
 969   emit_int64((int64_t)dest);
 970 
 971   const address stub_start_addr = addr_at(stub_start_offset);
 972 
 973   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 974 
 975   end_a_stub();
 976   return stub_start_addr;
 977 }
 978 
 979 int MacroAssembler::max_trampoline_stub_size() {
 980   // Max stub size: alignment nop, TrampolineStub.
 981   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 982 }
 983 
 984 void MacroAssembler::emit_static_call_stub() {
 985   // CompiledDirectCall::set_to_interpreted knows the
 986   // exact layout of this stub.
 987 
 988   isb();
 989   mov_metadata(rmethod, nullptr);
 990 
 991   // Jump to the entry point of the c2i stub.
 992   if (codestub_branch_needs_far_jump()) {
 993     movptr(rscratch1, 0);
 994     br(rscratch1);
 995   } else {
 996     b(pc());
 997   }
 998 }
 999 
1000 int MacroAssembler::static_call_stub_size() {
1001   if (!codestub_branch_needs_far_jump()) {
1002     // isb; movk; movz; movz; b
1003     return 5 * NativeInstruction::instruction_size;
1004   }
1005   // isb; movk; movz; movz; movk; movz; movz; br
1006   return 8 * NativeInstruction::instruction_size;
1007 }
1008 
1009 void MacroAssembler::c2bool(Register x) {
1010   // implements x == 0 ? 0 : 1
1011   // note: must only look at least-significant byte of x
1012   //       since C-style booleans are stored in one byte
1013   //       only! (was bug)
1014   tst(x, 0xff);
1015   cset(x, Assembler::NE);
1016 }
1017 
1018 address MacroAssembler::ic_call(address entry, jint method_index) {
1019   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
1020   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
1021   return trampoline_call(Address(entry, rh));
1022 }
1023 
1024 int MacroAssembler::ic_check_size() {
1025   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
1026   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
1027     return NativeInstruction::instruction_size * (7 + extra_instructions);
1028   } else {
1029     return NativeInstruction::instruction_size * (5 + extra_instructions);
1030   }
1031 }
1032 
1033 int MacroAssembler::ic_check(int end_alignment) {
1034   Register receiver = j_rarg0;
1035   Register data = rscratch2;
1036   Register tmp1 = rscratch1;
1037   Register tmp2 = r10;
1038 
1039   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
1040   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
1041   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1042   // before the inline cache check here, and not after
1043   align(end_alignment, offset() + ic_check_size());
1044 
1045   int uep_offset = offset();
1046 
1047   if (UseCompactObjectHeaders) {
1048     load_narrow_klass_compact(tmp1, receiver);
1049     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1050     cmpw(tmp1, tmp2);
1051   } else if (UseCompressedClassPointers) {
1052     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1053     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1054     cmpw(tmp1, tmp2);
1055   } else {
1056     ldr(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1057     ldr(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1058     cmp(tmp1, tmp2);
1059   }
1060 
1061   Label dont;
1062   br(Assembler::EQ, dont);
1063   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1064   bind(dont);
1065   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1066 
1067   return uep_offset;
1068 }
1069 
1070 // Implementation of call_VM versions
1071 
1072 void MacroAssembler::call_VM(Register oop_result,
1073                              address entry_point,
1074                              bool check_exceptions) {
1075   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1076 }
1077 
1078 void MacroAssembler::call_VM(Register oop_result,
1079                              address entry_point,
1080                              Register arg_1,
1081                              bool check_exceptions) {
1082   pass_arg1(this, arg_1);
1083   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1084 }
1085 
1086 void MacroAssembler::call_VM(Register oop_result,
1087                              address entry_point,
1088                              Register arg_1,
1089                              Register arg_2,
1090                              bool check_exceptions) {
1091   assert_different_registers(arg_1, c_rarg2);
1092   pass_arg2(this, arg_2);
1093   pass_arg1(this, arg_1);
1094   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1095 }
1096 
1097 void MacroAssembler::call_VM(Register oop_result,
1098                              address entry_point,
1099                              Register arg_1,
1100                              Register arg_2,
1101                              Register arg_3,
1102                              bool check_exceptions) {
1103   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1104   assert_different_registers(arg_2, c_rarg3);
1105   pass_arg3(this, arg_3);
1106 
1107   pass_arg2(this, arg_2);
1108 
1109   pass_arg1(this, arg_1);
1110   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1111 }
1112 
1113 void MacroAssembler::call_VM(Register oop_result,
1114                              Register last_java_sp,
1115                              address entry_point,
1116                              int number_of_arguments,
1117                              bool check_exceptions) {
1118   call_VM_base(oop_result, rthread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1119 }
1120 
1121 void MacroAssembler::call_VM(Register oop_result,
1122                              Register last_java_sp,
1123                              address entry_point,
1124                              Register arg_1,
1125                              bool check_exceptions) {
1126   pass_arg1(this, arg_1);
1127   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1128 }
1129 
1130 void MacroAssembler::call_VM(Register oop_result,
1131                              Register last_java_sp,
1132                              address entry_point,
1133                              Register arg_1,
1134                              Register arg_2,
1135                              bool check_exceptions) {
1136 
1137   assert_different_registers(arg_1, c_rarg2);
1138   pass_arg2(this, arg_2);
1139   pass_arg1(this, arg_1);
1140   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1141 }
1142 
1143 void MacroAssembler::call_VM(Register oop_result,
1144                              Register last_java_sp,
1145                              address entry_point,
1146                              Register arg_1,
1147                              Register arg_2,
1148                              Register arg_3,
1149                              bool check_exceptions) {
1150   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1151   assert_different_registers(arg_2, c_rarg3);
1152   pass_arg3(this, arg_3);
1153   pass_arg2(this, arg_2);
1154   pass_arg1(this, arg_1);
1155   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1156 }
1157 
1158 
1159 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1160   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1161   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1162   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1163 }
1164 
1165 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1166   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1167   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1168 }
1169 
1170 void MacroAssembler::align(int modulus) {
1171   align(modulus, offset());
1172 }
1173 
1174 // Ensure that the code at target bytes offset from the current offset() is aligned
1175 // according to modulus.
1176 void MacroAssembler::align(int modulus, int target) {
1177   int delta = target - offset();
1178   while ((offset() + delta) % modulus != 0) nop();
1179 }
1180 
1181 void MacroAssembler::post_call_nop() {
1182   if (!Continuations::enabled()) {
1183     return;
1184   }
1185   InstructionMark im(this);
1186   relocate(post_call_nop_Relocation::spec());
1187   InlineSkippedInstructionsCounter skipCounter(this);
1188   nop();
1189   movk(zr, 0);
1190   movk(zr, 0);
1191 }
1192 
1193 // these are no-ops overridden by InterpreterMacroAssembler
1194 
1195 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1196 
1197 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1198 
1199 // Look up the method for a megamorphic invokeinterface call.
1200 // The target method is determined by <intf_klass, itable_index>.
1201 // The receiver klass is in recv_klass.
1202 // On success, the result will be in method_result, and execution falls through.
1203 // On failure, execution transfers to the given label.
1204 void MacroAssembler::lookup_interface_method(Register recv_klass,
1205                                              Register intf_klass,
1206                                              RegisterOrConstant itable_index,
1207                                              Register method_result,
1208                                              Register scan_temp,
1209                                              Label& L_no_such_interface,
1210                          bool return_method) {
1211   assert_different_registers(recv_klass, intf_klass, scan_temp);
1212   assert_different_registers(method_result, intf_klass, scan_temp);
1213   assert(recv_klass != method_result || !return_method,
1214      "recv_klass can be destroyed when method isn't needed");
1215   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1216          "caller must use same register for non-constant itable index as for method");
1217 
1218   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1219   int vtable_base = in_bytes(Klass::vtable_start_offset());
1220   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1221   int scan_step   = itableOffsetEntry::size() * wordSize;
1222   int vte_size    = vtableEntry::size_in_bytes();
1223   assert(vte_size == wordSize, "else adjust times_vte_scale");
1224 
1225   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1226 
1227   // Could store the aligned, prescaled offset in the klass.
1228   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1229   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1230   add(scan_temp, scan_temp, vtable_base);
1231 
1232   if (return_method) {
1233     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1234     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1235     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1236     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1237     if (itentry_off)
1238       add(recv_klass, recv_klass, itentry_off);
1239   }
1240 
1241   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1242   //   if (scan->interface() == intf) {
1243   //     result = (klass + scan->offset() + itable_index);
1244   //   }
1245   // }
1246   Label search, found_method;
1247 
1248   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1249   cmp(intf_klass, method_result);
1250   br(Assembler::EQ, found_method);
1251   bind(search);
1252   // Check that the previous entry is non-null.  A null entry means that
1253   // the receiver class doesn't implement the interface, and wasn't the
1254   // same as when the caller was compiled.
1255   cbz(method_result, L_no_such_interface);
1256   if (itableOffsetEntry::interface_offset() != 0) {
1257     add(scan_temp, scan_temp, scan_step);
1258     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1259   } else {
1260     ldr(method_result, Address(pre(scan_temp, scan_step)));
1261   }
1262   cmp(intf_klass, method_result);
1263   br(Assembler::NE, search);
1264 
1265   bind(found_method);
1266 
1267   // Got a hit.
1268   if (return_method) {
1269     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1270     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1271   }
1272 }
1273 
1274 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1275 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1276 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1277 // The target method is determined by <holder_klass, itable_index>.
1278 // The receiver klass is in recv_klass.
1279 // On success, the result will be in method_result, and execution falls through.
1280 // On failure, execution transfers to the given label.
1281 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1282                                                   Register holder_klass,
1283                                                   Register resolved_klass,
1284                                                   Register method_result,
1285                                                   Register temp_itbl_klass,
1286                                                   Register scan_temp,
1287                                                   int itable_index,
1288                                                   Label& L_no_such_interface) {
1289   // 'method_result' is only used as output register at the very end of this method.
1290   // Until then we can reuse it as 'holder_offset'.
1291   Register holder_offset = method_result;
1292   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1293 
1294   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1295   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1296   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1297   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1298 
1299   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1300 
1301   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1302   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1303   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1304   // temp_itbl_klass = itable[0]._interface;
1305   int vtblEntrySize = vtableEntry::size_in_bytes();
1306   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1307   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1308   mov(holder_offset, zr);
1309   // scan_temp = &(itable[0]._interface)
1310   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1311 
1312   // Initial checks:
1313   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1314   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1315   //   - if (itable[0] == 0), no such interface
1316   cmp(resolved_klass, holder_klass);
1317   br(Assembler::NE, L_loop_search_resolved_entry);
1318   cmp(holder_klass, temp_itbl_klass);
1319   br(Assembler::EQ, L_holder_found);
1320   cbz(temp_itbl_klass, L_no_such_interface);
1321 
1322   // Loop: Look for holder_klass record in itable
1323   //   do {
1324   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1325   //     if (temp_itbl_klass == holder_klass) {
1326   //       goto L_holder_found; // Found!
1327   //     }
1328   //   } while (temp_itbl_klass != 0);
1329   //   goto L_no_such_interface // Not found.
1330   Label L_search_holder;
1331   bind(L_search_holder);
1332     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1333     cmp(holder_klass, temp_itbl_klass);
1334     br(Assembler::EQ, L_holder_found);
1335     cbnz(temp_itbl_klass, L_search_holder);
1336 
1337   b(L_no_such_interface);
1338 
1339   // Loop: Look for resolved_class record in itable
1340   //   while (true) {
1341   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1342   //     if (temp_itbl_klass == 0) {
1343   //       goto L_no_such_interface;
1344   //     }
1345   //     if (temp_itbl_klass == resolved_klass) {
1346   //        goto L_resolved_found;  // Found!
1347   //     }
1348   //     if (temp_itbl_klass == holder_klass) {
1349   //        holder_offset = scan_temp;
1350   //     }
1351   //   }
1352   //
1353   Label L_loop_search_resolved;
1354   bind(L_loop_search_resolved);
1355     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1356   bind(L_loop_search_resolved_entry);
1357     cbz(temp_itbl_klass, L_no_such_interface);
1358     cmp(resolved_klass, temp_itbl_klass);
1359     br(Assembler::EQ, L_resolved_found);
1360     cmp(holder_klass, temp_itbl_klass);
1361     br(Assembler::NE, L_loop_search_resolved);
1362     mov(holder_offset, scan_temp);
1363     b(L_loop_search_resolved);
1364 
1365   // See if we already have a holder klass. If not, go and scan for it.
1366   bind(L_resolved_found);
1367   cbz(holder_offset, L_search_holder);
1368   mov(scan_temp, holder_offset);
1369 
1370   // Finally, scan_temp contains holder_klass vtable offset
1371   bind(L_holder_found);
1372   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1373   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1374     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1375   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1376 }
1377 
1378 // virtual method calling
1379 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1380                                            RegisterOrConstant vtable_index,
1381                                            Register method_result) {
1382   assert(vtableEntry::size() * wordSize == 8,
1383          "adjust the scaling in the code below");
1384   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1385 
1386   if (vtable_index.is_register()) {
1387     lea(method_result, Address(recv_klass,
1388                                vtable_index.as_register(),
1389                                Address::lsl(LogBytesPerWord)));
1390     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1391   } else {
1392     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1393     ldr(method_result,
1394         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1395   }
1396 }
1397 
1398 void MacroAssembler::check_klass_subtype(Register sub_klass,
1399                            Register super_klass,
1400                            Register temp_reg,
1401                            Label& L_success) {
1402   Label L_failure;
1403   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1404   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1405   bind(L_failure);
1406 }
1407 
1408 
1409 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1410                                                    Register super_klass,
1411                                                    Register temp_reg,
1412                                                    Label* L_success,
1413                                                    Label* L_failure,
1414                                                    Label* L_slow_path,
1415                                                    Register super_check_offset) {
1416   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1417   bool must_load_sco = ! super_check_offset->is_valid();
1418   if (must_load_sco) {
1419     assert(temp_reg != noreg, "supply either a temp or a register offset");
1420   }
1421 
1422   Label L_fallthrough;
1423   int label_nulls = 0;
1424   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1425   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1426   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1427   assert(label_nulls <= 1, "at most one null in the batch");
1428 
1429   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1430   Address super_check_offset_addr(super_klass, sco_offset);
1431 
1432   // Hacked jmp, which may only be used just before L_fallthrough.
1433 #define final_jmp(label)                                                \
1434   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1435   else                            b(label)                /*omit semi*/
1436 
1437   // If the pointers are equal, we are done (e.g., String[] elements).
1438   // This self-check enables sharing of secondary supertype arrays among
1439   // non-primary types such as array-of-interface.  Otherwise, each such
1440   // type would need its own customized SSA.
1441   // We move this check to the front of the fast path because many
1442   // type checks are in fact trivially successful in this manner,
1443   // so we get a nicely predicted branch right at the start of the check.
1444   // TODO 8370341 For a direct pointer comparison, we need the refined array klass pointer
1445   cmp(sub_klass, super_klass);
1446   br(Assembler::EQ, *L_success);
1447 
1448   // Check the supertype display:
1449   if (must_load_sco) {
1450     ldrw(temp_reg, super_check_offset_addr);
1451     super_check_offset = temp_reg;
1452   }
1453 
1454   Address super_check_addr(sub_klass, super_check_offset);
1455   ldr(rscratch1, super_check_addr);
1456   cmp(super_klass, rscratch1); // load displayed supertype
1457   br(Assembler::EQ, *L_success);
1458 
1459   // This check has worked decisively for primary supers.
1460   // Secondary supers are sought in the super_cache ('super_cache_addr').
1461   // (Secondary supers are interfaces and very deeply nested subtypes.)
1462   // This works in the same check above because of a tricky aliasing
1463   // between the super_cache and the primary super display elements.
1464   // (The 'super_check_addr' can address either, as the case requires.)
1465   // Note that the cache is updated below if it does not help us find
1466   // what we need immediately.
1467   // So if it was a primary super, we can just fail immediately.
1468   // Otherwise, it's the slow path for us (no success at this point).
1469 
1470   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1471   if (L_failure == &L_fallthrough) {
1472     cbz(rscratch1, *L_slow_path);
1473   } else {
1474     cbnz(rscratch1, *L_failure);
1475     final_jmp(*L_slow_path);
1476   }
1477 
1478   bind(L_fallthrough);
1479 
1480 #undef final_jmp
1481 }
1482 
1483 // These two are taken from x86, but they look generally useful
1484 
1485 // scans count pointer sized words at [addr] for occurrence of value,
1486 // generic
1487 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1488                                 Register scratch) {
1489   Label Lloop, Lexit;
1490   cbz(count, Lexit);
1491   bind(Lloop);
1492   ldr(scratch, post(addr, wordSize));
1493   cmp(value, scratch);
1494   br(EQ, Lexit);
1495   sub(count, count, 1);
1496   cbnz(count, Lloop);
1497   bind(Lexit);
1498 }
1499 
1500 // scans count 4 byte words at [addr] for occurrence of value,
1501 // generic
1502 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1503                                 Register scratch) {
1504   Label Lloop, Lexit;
1505   cbz(count, Lexit);
1506   bind(Lloop);
1507   ldrw(scratch, post(addr, wordSize));
1508   cmpw(value, scratch);
1509   br(EQ, Lexit);
1510   sub(count, count, 1);
1511   cbnz(count, Lloop);
1512   bind(Lexit);
1513 }
1514 
1515 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1516                                                           Register super_klass,
1517                                                           Register temp_reg,
1518                                                           Register temp2_reg,
1519                                                           Label* L_success,
1520                                                           Label* L_failure,
1521                                                           bool set_cond_codes) {
1522   // NB! Callers may assume that, when temp2_reg is a valid register,
1523   // this code sets it to a nonzero value.
1524 
1525   assert_different_registers(sub_klass, super_klass, temp_reg);
1526   if (temp2_reg != noreg)
1527     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1528 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1529 
1530   Label L_fallthrough;
1531   int label_nulls = 0;
1532   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1533   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1534   assert(label_nulls <= 1, "at most one null in the batch");
1535 
1536   // a couple of useful fields in sub_klass:
1537   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1538   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1539   Address secondary_supers_addr(sub_klass, ss_offset);
1540   Address super_cache_addr(     sub_klass, sc_offset);
1541 
1542   BLOCK_COMMENT("check_klass_subtype_slow_path");
1543 
1544   // Do a linear scan of the secondary super-klass chain.
1545   // This code is rarely used, so simplicity is a virtue here.
1546   // The repne_scan instruction uses fixed registers, which we must spill.
1547   // Don't worry too much about pre-existing connections with the input regs.
1548 
1549   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1550   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1551 
1552   RegSet pushed_registers;
1553   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1554   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1555 
1556   if (super_klass != r0) {
1557     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1558   }
1559 
1560   push(pushed_registers, sp);
1561 
1562   // Get super_klass value into r0 (even if it was in r5 or r2).
1563   if (super_klass != r0) {
1564     mov(r0, super_klass);
1565   }
1566 
1567 #ifndef PRODUCT
1568   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1569 #endif //PRODUCT
1570 
1571   // We will consult the secondary-super array.
1572   ldr(r5, secondary_supers_addr);
1573   // Load the array length.
1574   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1575   // Skip to start of data.
1576   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1577 
1578   cmp(sp, zr); // Clear Z flag; SP is never zero
1579   // Scan R2 words at [R5] for an occurrence of R0.
1580   // Set NZ/Z based on last compare.
1581   repne_scan(r5, r0, r2, rscratch1);
1582 
1583   // Unspill the temp. registers:
1584   pop(pushed_registers, sp);
1585 
1586   br(Assembler::NE, *L_failure);
1587 
1588   // Success.  Cache the super we found and proceed in triumph.
1589 
1590   if (UseSecondarySupersCache) {
1591     str(super_klass, super_cache_addr);
1592   }
1593 
1594   if (L_success != &L_fallthrough) {
1595     b(*L_success);
1596   }
1597 
1598 #undef IS_A_TEMP
1599 
1600   bind(L_fallthrough);
1601 }
1602 
1603 // If Register r is invalid, remove a new register from
1604 // available_regs, and add new register to regs_to_push.
1605 Register MacroAssembler::allocate_if_noreg(Register r,
1606                                   RegSetIterator<Register> &available_regs,
1607                                   RegSet &regs_to_push) {
1608   if (!r->is_valid()) {
1609     r = *available_regs++;
1610     regs_to_push += r;
1611   }
1612   return r;
1613 }
1614 
1615 // check_klass_subtype_slow_path_table() looks for super_klass in the
1616 // hash table belonging to super_klass, branching to L_success or
1617 // L_failure as appropriate. This is essentially a shim which
1618 // allocates registers as necessary then calls
1619 // lookup_secondary_supers_table() to do the work. Any of the temp
1620 // regs may be noreg, in which case this logic will chooses some
1621 // registers push and pop them from the stack.
1622 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1623                                                          Register super_klass,
1624                                                          Register temp_reg,
1625                                                          Register temp2_reg,
1626                                                          Register temp3_reg,
1627                                                          Register result_reg,
1628                                                          FloatRegister vtemp,
1629                                                          Label* L_success,
1630                                                          Label* L_failure,
1631                                                          bool set_cond_codes) {
1632   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1633 
1634   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1635 
1636   Label L_fallthrough;
1637   int label_nulls = 0;
1638   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1639   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1640   assert(label_nulls <= 1, "at most one null in the batch");
1641 
1642   BLOCK_COMMENT("check_klass_subtype_slow_path");
1643 
1644   RegSetIterator<Register> available_regs
1645     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1646 
1647   RegSet pushed_regs;
1648 
1649   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1650   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1651   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1652   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1653 
1654   push(pushed_regs, sp);
1655 
1656   lookup_secondary_supers_table_var(sub_klass,
1657                                     super_klass,
1658                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1659                                     nullptr);
1660   cmp(result_reg, zr);
1661 
1662   // Unspill the temp. registers:
1663   pop(pushed_regs, sp);
1664 
1665   // NB! Callers may assume that, when set_cond_codes is true, this
1666   // code sets temp2_reg to a nonzero value.
1667   if (set_cond_codes) {
1668     mov(temp2_reg, 1);
1669   }
1670 
1671   br(Assembler::NE, *L_failure);
1672 
1673   if (L_success != &L_fallthrough) {
1674     b(*L_success);
1675   }
1676 
1677   bind(L_fallthrough);
1678 }
1679 
1680 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1681                                                    Register super_klass,
1682                                                    Register temp_reg,
1683                                                    Register temp2_reg,
1684                                                    Label* L_success,
1685                                                    Label* L_failure,
1686                                                    bool set_cond_codes) {
1687   if (UseSecondarySupersTable) {
1688     check_klass_subtype_slow_path_table
1689       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1690        /*vtemp*/fnoreg,
1691        L_success, L_failure, set_cond_codes);
1692   } else {
1693     check_klass_subtype_slow_path_linear
1694       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1695   }
1696 }
1697 
1698 
1699 // Ensure that the inline code and the stub are using the same registers.
1700 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1701 do {                                                               \
1702   assert(r_super_klass  == r0                                   && \
1703          r_array_base   == r1                                   && \
1704          r_array_length == r2                                   && \
1705          (r_array_index == r3        || r_array_index == noreg) && \
1706          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1707          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1708          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1709 } while(0)
1710 
1711 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1712                                                          Register r_super_klass,
1713                                                          Register temp1,
1714                                                          Register temp2,
1715                                                          Register temp3,
1716                                                          FloatRegister vtemp,
1717                                                          Register result,
1718                                                          u1 super_klass_slot,
1719                                                          bool stub_is_near) {
1720   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1721 
1722   Label L_fallthrough;
1723 
1724   BLOCK_COMMENT("lookup_secondary_supers_table {");
1725 
1726   const Register
1727     r_array_base   = temp1, // r1
1728     r_array_length = temp2, // r2
1729     r_array_index  = temp3, // r3
1730     r_bitmap       = rscratch2;
1731 
1732   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1733 
1734   u1 bit = super_klass_slot;
1735 
1736   // Make sure that result is nonzero if the TBZ below misses.
1737   mov(result, 1);
1738 
1739   // We're going to need the bitmap in a vector reg and in a core reg,
1740   // so load both now.
1741   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1742   if (bit != 0) {
1743     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1744   }
1745   // First check the bitmap to see if super_klass might be present. If
1746   // the bit is zero, we are certain that super_klass is not one of
1747   // the secondary supers.
1748   tbz(r_bitmap, bit, L_fallthrough);
1749 
1750   // Get the first array index that can contain super_klass into r_array_index.
1751   if (bit != 0) {
1752     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1753     cnt(vtemp, T8B, vtemp);
1754     addv(vtemp, T8B, vtemp);
1755     fmovd(r_array_index, vtemp);
1756   } else {
1757     mov(r_array_index, (u1)1);
1758   }
1759   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1760 
1761   // We will consult the secondary-super array.
1762   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1763 
1764   // The value i in r_array_index is >= 1, so even though r_array_base
1765   // points to the length, we don't need to adjust it to point to the
1766   // data.
1767   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1768   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1769 
1770   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1771   eor(result, result, r_super_klass);
1772   cbz(result, L_fallthrough); // Found a match
1773 
1774   // Is there another entry to check? Consult the bitmap.
1775   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1776 
1777   // Linear probe.
1778   if (bit != 0) {
1779     ror(r_bitmap, r_bitmap, bit);
1780   }
1781 
1782   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1783   // The next slot to be inspected, by the stub we're about to call,
1784   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1785   // have been checked.
1786   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1787   if (stub_is_near) {
1788     bl(stub);
1789   } else {
1790     address call = trampoline_call(stub);
1791     if (call == nullptr) {
1792       return false; // trampoline allocation failed
1793     }
1794   }
1795 
1796   BLOCK_COMMENT("} lookup_secondary_supers_table");
1797 
1798   bind(L_fallthrough);
1799 
1800   if (VerifySecondarySupers) {
1801     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1802                                   temp1, temp2, result);      // r1, r2, r5
1803   }
1804   return true;
1805 }
1806 
1807 // At runtime, return 0 in result if r_super_klass is a superclass of
1808 // r_sub_klass, otherwise return nonzero. Use this version of
1809 // lookup_secondary_supers_table() if you don't know ahead of time
1810 // which superclass will be searched for. Used by interpreter and
1811 // runtime stubs. It is larger and has somewhat greater latency than
1812 // the version above, which takes a constant super_klass_slot.
1813 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1814                                                        Register r_super_klass,
1815                                                        Register temp1,
1816                                                        Register temp2,
1817                                                        Register temp3,
1818                                                        FloatRegister vtemp,
1819                                                        Register result,
1820                                                        Label *L_success) {
1821   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1822 
1823   Label L_fallthrough;
1824 
1825   BLOCK_COMMENT("lookup_secondary_supers_table {");
1826 
1827   const Register
1828     r_array_index = temp3,
1829     slot          = rscratch1,
1830     r_bitmap      = rscratch2;
1831 
1832   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1833 
1834   // Make sure that result is nonzero if the test below misses.
1835   mov(result, 1);
1836 
1837   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1838 
1839   // First check the bitmap to see if super_klass might be present. If
1840   // the bit is zero, we are certain that super_klass is not one of
1841   // the secondary supers.
1842 
1843   // This next instruction is equivalent to:
1844   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1845   // sub(temp2, tmp_reg, slot);
1846   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1847   lslv(temp2, r_bitmap, temp2);
1848   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1849 
1850   bool must_save_v0 = (vtemp == fnoreg);
1851   if (must_save_v0) {
1852     // temp1 and result are free, so use them to preserve vtemp
1853     vtemp = v0;
1854     mov(temp1,  vtemp, D, 0);
1855     mov(result, vtemp, D, 1);
1856   }
1857 
1858   // Get the first array index that can contain super_klass into r_array_index.
1859   mov(vtemp, D, 0, temp2);
1860   cnt(vtemp, T8B, vtemp);
1861   addv(vtemp, T8B, vtemp);
1862   mov(r_array_index, vtemp, D, 0);
1863 
1864   if (must_save_v0) {
1865     mov(vtemp, D, 0, temp1 );
1866     mov(vtemp, D, 1, result);
1867   }
1868 
1869   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1870 
1871   const Register
1872     r_array_base   = temp1,
1873     r_array_length = temp2;
1874 
1875   // The value i in r_array_index is >= 1, so even though r_array_base
1876   // points to the length, we don't need to adjust it to point to the
1877   // data.
1878   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1879   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1880 
1881   // We will consult the secondary-super array.
1882   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1883 
1884   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1885   eor(result, result, r_super_klass);
1886   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1887 
1888   // Is there another entry to check? Consult the bitmap.
1889   rorv(r_bitmap, r_bitmap, slot);
1890   // rol(r_bitmap, r_bitmap, 1);
1891   tbz(r_bitmap, 1, L_fallthrough);
1892 
1893   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1894   // The next slot to be inspected, by the logic we're about to call,
1895   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1896   // have been checked.
1897   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1898                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1899 
1900   BLOCK_COMMENT("} lookup_secondary_supers_table");
1901 
1902   bind(L_fallthrough);
1903 
1904   if (VerifySecondarySupers) {
1905     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1906                                   temp1, temp2, result);      // r1, r2, r5
1907   }
1908 
1909   if (L_success) {
1910     cbz(result, *L_success);
1911   }
1912 }
1913 
1914 // Called by code generated by check_klass_subtype_slow_path
1915 // above. This is called when there is a collision in the hashed
1916 // lookup in the secondary supers array.
1917 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1918                                                              Register r_array_base,
1919                                                              Register r_array_index,
1920                                                              Register r_bitmap,
1921                                                              Register temp1,
1922                                                              Register result,
1923                                                              bool is_stub) {
1924   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1925 
1926   const Register
1927     r_array_length = temp1,
1928     r_sub_klass    = noreg; // unused
1929 
1930   if (is_stub) {
1931     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1932   }
1933 
1934   Label L_fallthrough, L_huge;
1935 
1936   // Load the array length.
1937   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1938   // And adjust the array base to point to the data.
1939   // NB! Effectively increments current slot index by 1.
1940   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1941   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1942 
1943   // The bitmap is full to bursting.
1944   // Implicit invariant: BITMAP_FULL implies (length > 0)
1945   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1946   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1947   br(GT, L_huge);
1948 
1949   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1950   // current slot (at secondary_supers[r_array_index]) has not yet
1951   // been inspected, and r_array_index may be out of bounds if we
1952   // wrapped around the end of the array.
1953 
1954   { // This is conventional linear probing, but instead of terminating
1955     // when a null entry is found in the table, we maintain a bitmap
1956     // in which a 0 indicates missing entries.
1957     // As long as the bitmap is not completely full,
1958     // array_length == popcount(bitmap). The array_length check above
1959     // guarantees there are 0s in the bitmap, so the loop eventually
1960     // terminates.
1961     Label L_loop;
1962     bind(L_loop);
1963 
1964     // Check for wraparound.
1965     cmp(r_array_index, r_array_length);
1966     csel(r_array_index, zr, r_array_index, GE);
1967 
1968     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1969     eor(result, rscratch1, r_super_klass);
1970     cbz(result, L_fallthrough);
1971 
1972     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1973 
1974     ror(r_bitmap, r_bitmap, 1);
1975     add(r_array_index, r_array_index, 1);
1976     b(L_loop);
1977   }
1978 
1979   { // Degenerate case: more than 64 secondary supers.
1980     // FIXME: We could do something smarter here, maybe a vectorized
1981     // comparison or a binary search, but is that worth any added
1982     // complexity?
1983     bind(L_huge);
1984     cmp(sp, zr); // Clear Z flag; SP is never zero
1985     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1986     cset(result, NE); // result == 0 iff we got a match.
1987   }
1988 
1989   bind(L_fallthrough);
1990 }
1991 
1992 // Make sure that the hashed lookup and a linear scan agree.
1993 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1994                                                    Register r_super_klass,
1995                                                    Register temp1,
1996                                                    Register temp2,
1997                                                    Register result) {
1998   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1999 
2000   const Register
2001     r_array_base   = temp1,
2002     r_array_length = temp2,
2003     r_array_index  = noreg, // unused
2004     r_bitmap       = noreg; // unused
2005 
2006   BLOCK_COMMENT("verify_secondary_supers_table {");
2007 
2008   // We will consult the secondary-super array.
2009   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
2010 
2011   // Load the array length.
2012   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
2013   // And adjust the array base to point to the data.
2014   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
2015 
2016   cmp(sp, zr); // Clear Z flag; SP is never zero
2017   // Scan R2 words at [R5] for an occurrence of R0.
2018   // Set NZ/Z based on last compare.
2019   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
2020   // rscratch1 == 0 iff we got a match.
2021   cset(rscratch1, NE);
2022 
2023   Label passed;
2024   cmp(result, zr);
2025   cset(result, NE); // normalize result to 0/1 for comparison
2026 
2027   cmp(rscratch1, result);
2028   br(EQ, passed);
2029   {
2030     mov(r0, r_super_klass);         // r0 <- r0
2031     mov(r1, r_sub_klass);           // r1 <- r4
2032     mov(r2, /*expected*/rscratch1); // r2 <- r8
2033     mov(r3, result);                // r3 <- r5
2034     mov(r4, (address)("mismatch")); // r4 <- const
2035     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
2036     should_not_reach_here();
2037   }
2038   bind(passed);
2039 
2040   BLOCK_COMMENT("} verify_secondary_supers_table");
2041 }
2042 
2043 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
2044   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
2045   assert_different_registers(klass, rthread, scratch);
2046 
2047   Label L_fallthrough, L_tmp;
2048   if (L_fast_path == nullptr) {
2049     L_fast_path = &L_fallthrough;
2050   } else if (L_slow_path == nullptr) {
2051     L_slow_path = &L_fallthrough;
2052   }
2053   // Fast path check: class is fully initialized
2054   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2055   ldarb(scratch, scratch);
2056   cmp(scratch, InstanceKlass::fully_initialized);
2057   br(Assembler::EQ, *L_fast_path);
2058 
2059   // Fast path check: current thread is initializer thread
2060   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2061   cmp(rthread, scratch);
2062 
2063   if (L_slow_path == &L_fallthrough) {
2064     br(Assembler::EQ, *L_fast_path);
2065     bind(*L_slow_path);
2066   } else if (L_fast_path == &L_fallthrough) {
2067     br(Assembler::NE, *L_slow_path);
2068     bind(*L_fast_path);
2069   } else {
2070     Unimplemented();
2071   }
2072 }
2073 
2074 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2075   if (!VerifyOops || VerifyAdapterSharing) {
2076     // Below address of the code string confuses VerifyAdapterSharing
2077     // because it may differ between otherwise equivalent adapters.
2078     return;
2079   }
2080 
2081   // Pass register number to verify_oop_subroutine
2082   const char* b = nullptr;
2083   {
2084     ResourceMark rm;
2085     stringStream ss;
2086     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2087     b = code_string(ss.as_string());
2088   }
2089   BLOCK_COMMENT("verify_oop {");
2090 
2091   strip_return_address(); // This might happen within a stack frame.
2092   protect_return_address();
2093   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2094   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2095 
2096   mov(r0, reg);
2097   movptr(rscratch1, (uintptr_t)(address)b);
2098 
2099   // call indirectly to solve generation ordering problem
2100   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2101   ldr(rscratch2, Address(rscratch2));
2102   blr(rscratch2);
2103 
2104   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2105   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2106   authenticate_return_address();
2107 
2108   BLOCK_COMMENT("} verify_oop");
2109 }
2110 
2111 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2112   if (!VerifyOops || VerifyAdapterSharing) {
2113     // Below address of the code string confuses VerifyAdapterSharing
2114     // because it may differ between otherwise equivalent adapters.
2115     return;
2116   }
2117 
2118   const char* b = nullptr;
2119   {
2120     ResourceMark rm;
2121     stringStream ss;
2122     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2123     b = code_string(ss.as_string());
2124   }
2125   BLOCK_COMMENT("verify_oop_addr {");
2126 
2127   strip_return_address(); // This might happen within a stack frame.
2128   protect_return_address();
2129   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2130   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2131 
2132   // addr may contain sp so we will have to adjust it based on the
2133   // pushes that we just did.
2134   if (addr.uses(sp)) {
2135     lea(r0, addr);
2136     ldr(r0, Address(r0, 4 * wordSize));
2137   } else {
2138     ldr(r0, addr);
2139   }
2140   movptr(rscratch1, (uintptr_t)(address)b);
2141 
2142   // call indirectly to solve generation ordering problem
2143   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2144   ldr(rscratch2, Address(rscratch2));
2145   blr(rscratch2);
2146 
2147   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2148   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2149   authenticate_return_address();
2150 
2151   BLOCK_COMMENT("} verify_oop_addr");
2152 }
2153 
2154 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2155                                          int extra_slot_offset) {
2156   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2157   int stackElementSize = Interpreter::stackElementSize;
2158   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2159 #ifdef ASSERT
2160   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2161   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2162 #endif
2163   if (arg_slot.is_constant()) {
2164     return Address(esp, arg_slot.as_constant() * stackElementSize
2165                    + offset);
2166   } else {
2167     add(rscratch1, esp, arg_slot.as_register(),
2168         ext::uxtx, exact_log2(stackElementSize));
2169     return Address(rscratch1, offset);
2170   }
2171 }
2172 
2173 void MacroAssembler::call_VM_leaf_base(address entry_point,
2174                                        int number_of_arguments,
2175                                        Label *retaddr) {
2176   Label E, L;
2177 
2178   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2179 
2180   mov(rscratch1, RuntimeAddress(entry_point));
2181   blr(rscratch1);
2182   if (retaddr)
2183     bind(*retaddr);
2184 
2185   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2186 }
2187 
2188 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2189   call_VM_leaf_base(entry_point, number_of_arguments);
2190 }
2191 
2192 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2193   pass_arg0(this, arg_0);
2194   call_VM_leaf_base(entry_point, 1);
2195 }
2196 
2197 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2198   assert_different_registers(arg_1, c_rarg0);
2199   pass_arg0(this, arg_0);
2200   pass_arg1(this, arg_1);
2201   call_VM_leaf_base(entry_point, 2);
2202 }
2203 
2204 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2205                                   Register arg_1, Register arg_2) {
2206   assert_different_registers(arg_1, c_rarg0);
2207   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2208   pass_arg0(this, arg_0);
2209   pass_arg1(this, arg_1);
2210   pass_arg2(this, arg_2);
2211   call_VM_leaf_base(entry_point, 3);
2212 }
2213 
2214 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2215   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2216 }
2217 
2218 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2219   pass_arg0(this, arg_0);
2220   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2221 }
2222 
2223 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2224 
2225   assert_different_registers(arg_0, c_rarg1);
2226   pass_arg1(this, arg_1);
2227   pass_arg0(this, arg_0);
2228   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2229 }
2230 
2231 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2232   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2233   assert_different_registers(arg_1, c_rarg2);
2234   pass_arg2(this, arg_2);
2235   pass_arg1(this, arg_1);
2236   pass_arg0(this, arg_0);
2237   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2238 }
2239 
2240 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2241   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2242   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2243   assert_different_registers(arg_2, c_rarg3);
2244   pass_arg3(this, arg_3);
2245   pass_arg2(this, arg_2);
2246   pass_arg1(this, arg_1);
2247   pass_arg0(this, arg_0);
2248   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2249 }
2250 
2251 void MacroAssembler::null_check(Register reg, int offset) {
2252   if (needs_explicit_null_check(offset)) {
2253     // provoke OS null exception if reg is null by
2254     // accessing M[reg] w/o changing any registers
2255     // NOTE: this is plenty to provoke a segv
2256     ldr(zr, Address(reg));
2257   } else {
2258     // nothing to do, (later) access of M[reg + offset]
2259     // will provoke OS null exception if reg is null
2260   }
2261 }
2262 
2263 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2264   assert_different_registers(markword, rscratch2);
2265   mov(rscratch2, markWord::inline_type_mask_in_place);
2266   andr(markword, markword, rscratch2);
2267   mov(rscratch2, markWord::inline_type_pattern);
2268   cmp(markword, rscratch2);
2269   br(Assembler::EQ, is_inline_type);
2270 }
2271 
2272 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2273   assert_different_registers(tmp, rscratch1);
2274   if (can_be_null) {
2275     cbz(object, not_inline_type);
2276   }
2277   const int is_inline_type_mask = markWord::inline_type_pattern;
2278   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2279   mov(rscratch1, is_inline_type_mask);
2280   andr(tmp, tmp, rscratch1);
2281   cmp(tmp, rscratch1);
2282   br(Assembler::NE, not_inline_type);
2283 }
2284 
2285 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2286   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2287   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2288 }
2289 
2290 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2291   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2292   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2293 }
2294 
2295 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2296   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2297   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2298 }
2299 
2300 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
2301   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2302   tbnz(flags, ResolvedFieldEntry::has_null_marker_shift, has_null_marker);
2303 }
2304 
2305 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2306   Label test_mark_word;
2307   // load mark word
2308   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2309   // check displaced
2310   tst(temp_reg, markWord::unlocked_value);
2311   br(Assembler::NE, test_mark_word);
2312   // slow path use klass prototype
2313   load_prototype_header(temp_reg, oop);
2314 
2315   bind(test_mark_word);
2316   andr(temp_reg, temp_reg, test_bit);
2317   if (jmp_set) {
2318     cbnz(temp_reg, jmp_label);
2319   } else {
2320     cbz(temp_reg, jmp_label);
2321   }
2322 }
2323 
2324 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2325   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2326 }
2327 
2328 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2329                                                   Label&is_non_flat_array) {
2330   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2331 }
2332 
2333 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2334   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2335 }
2336 
2337 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2338   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2339 }
2340 
2341 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2342   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2343   br(Assembler::NE, is_flat_array);
2344 }
2345 
2346 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
2347   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2348   br(Assembler::EQ, is_non_flat_array);
2349 }
2350 
2351 // MacroAssembler protected routines needed to implement
2352 // public methods
2353 
2354 void MacroAssembler::mov(Register r, Address dest) {
2355   code_section()->relocate(pc(), dest.rspec());
2356   uint64_t imm64 = (uint64_t)dest.target();
2357   movptr(r, imm64);
2358 }
2359 
2360 // Move a constant pointer into r.  In AArch64 mode the virtual
2361 // address space is 48 bits in size, so we only need three
2362 // instructions to create a patchable instruction sequence that can
2363 // reach anywhere.
2364 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2365 #ifndef PRODUCT
2366   {
2367     char buffer[64];
2368     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2369     block_comment(buffer);
2370   }
2371 #endif
2372   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2373   movz(r, imm64 & 0xffff);
2374   imm64 >>= 16;
2375   movk(r, imm64 & 0xffff, 16);
2376   imm64 >>= 16;
2377   movk(r, imm64 & 0xffff, 32);
2378 }
2379 
2380 // Macro to mov replicated immediate to vector register.
2381 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2382 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2383 // Vd will get the following values for different arrangements in T
2384 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2385 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2386 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2387 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2388 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2389 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2390 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2391 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2392 // Clobbers rscratch1
2393 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2394   assert(T != T1Q, "unsupported");
2395   if (T == T1D || T == T2D) {
2396     int imm = operand_valid_for_movi_immediate(imm64, T);
2397     if (-1 != imm) {
2398       movi(Vd, T, imm);
2399     } else {
2400       mov(rscratch1, imm64);
2401       dup(Vd, T, rscratch1);
2402     }
2403     return;
2404   }
2405 
2406 #ifdef ASSERT
2407   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2408   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2409   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2410 #endif
2411   int shift = operand_valid_for_movi_immediate(imm64, T);
2412   uint32_t imm32 = imm64 & 0xffffffffULL;
2413   if (shift >= 0) {
2414     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2415   } else {
2416     movw(rscratch1, imm32);
2417     dup(Vd, T, rscratch1);
2418   }
2419 }
2420 
2421 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2422 {
2423 #ifndef PRODUCT
2424   {
2425     char buffer[64];
2426     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2427     block_comment(buffer);
2428   }
2429 #endif
2430   if (operand_valid_for_logical_immediate(false, imm64)) {
2431     orr(dst, zr, imm64);
2432   } else {
2433     // we can use a combination of MOVZ or MOVN with
2434     // MOVK to build up the constant
2435     uint64_t imm_h[4];
2436     int zero_count = 0;
2437     int neg_count = 0;
2438     int i;
2439     for (i = 0; i < 4; i++) {
2440       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2441       if (imm_h[i] == 0) {
2442         zero_count++;
2443       } else if (imm_h[i] == 0xffffL) {
2444         neg_count++;
2445       }
2446     }
2447     if (zero_count == 4) {
2448       // one MOVZ will do
2449       movz(dst, 0);
2450     } else if (neg_count == 4) {
2451       // one MOVN will do
2452       movn(dst, 0);
2453     } else if (zero_count == 3) {
2454       for (i = 0; i < 4; i++) {
2455         if (imm_h[i] != 0L) {
2456           movz(dst, (uint32_t)imm_h[i], (i << 4));
2457           break;
2458         }
2459       }
2460     } else if (neg_count == 3) {
2461       // one MOVN will do
2462       for (int i = 0; i < 4; i++) {
2463         if (imm_h[i] != 0xffffL) {
2464           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2465           break;
2466         }
2467       }
2468     } else if (zero_count == 2) {
2469       // one MOVZ and one MOVK will do
2470       for (i = 0; i < 3; i++) {
2471         if (imm_h[i] != 0L) {
2472           movz(dst, (uint32_t)imm_h[i], (i << 4));
2473           i++;
2474           break;
2475         }
2476       }
2477       for (;i < 4; i++) {
2478         if (imm_h[i] != 0L) {
2479           movk(dst, (uint32_t)imm_h[i], (i << 4));
2480         }
2481       }
2482     } else if (neg_count == 2) {
2483       // one MOVN and one MOVK will do
2484       for (i = 0; i < 4; i++) {
2485         if (imm_h[i] != 0xffffL) {
2486           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2487           i++;
2488           break;
2489         }
2490       }
2491       for (;i < 4; i++) {
2492         if (imm_h[i] != 0xffffL) {
2493           movk(dst, (uint32_t)imm_h[i], (i << 4));
2494         }
2495       }
2496     } else if (zero_count == 1) {
2497       // one MOVZ and two MOVKs will do
2498       for (i = 0; i < 4; i++) {
2499         if (imm_h[i] != 0L) {
2500           movz(dst, (uint32_t)imm_h[i], (i << 4));
2501           i++;
2502           break;
2503         }
2504       }
2505       for (;i < 4; i++) {
2506         if (imm_h[i] != 0x0L) {
2507           movk(dst, (uint32_t)imm_h[i], (i << 4));
2508         }
2509       }
2510     } else if (neg_count == 1) {
2511       // one MOVN and two MOVKs will do
2512       for (i = 0; i < 4; i++) {
2513         if (imm_h[i] != 0xffffL) {
2514           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2515           i++;
2516           break;
2517         }
2518       }
2519       for (;i < 4; i++) {
2520         if (imm_h[i] != 0xffffL) {
2521           movk(dst, (uint32_t)imm_h[i], (i << 4));
2522         }
2523       }
2524     } else {
2525       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2526       movz(dst, (uint32_t)imm_h[0], 0);
2527       for (i = 1; i < 4; i++) {
2528         movk(dst, (uint32_t)imm_h[i], (i << 4));
2529       }
2530     }
2531   }
2532 }
2533 
2534 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2535 {
2536 #ifndef PRODUCT
2537     {
2538       char buffer[64];
2539       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2540       block_comment(buffer);
2541     }
2542 #endif
2543   if (operand_valid_for_logical_immediate(true, imm32)) {
2544     orrw(dst, zr, imm32);
2545   } else {
2546     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2547     // constant
2548     uint32_t imm_h[2];
2549     imm_h[0] = imm32 & 0xffff;
2550     imm_h[1] = ((imm32 >> 16) & 0xffff);
2551     if (imm_h[0] == 0) {
2552       movzw(dst, imm_h[1], 16);
2553     } else if (imm_h[0] == 0xffff) {
2554       movnw(dst, imm_h[1] ^ 0xffff, 16);
2555     } else if (imm_h[1] == 0) {
2556       movzw(dst, imm_h[0], 0);
2557     } else if (imm_h[1] == 0xffff) {
2558       movnw(dst, imm_h[0] ^ 0xffff, 0);
2559     } else {
2560       // use a MOVZ and MOVK (makes it easier to debug)
2561       movzw(dst, imm_h[0], 0);
2562       movkw(dst, imm_h[1], 16);
2563     }
2564   }
2565 }
2566 
2567 // Form an address from base + offset in Rd.  Rd may or may
2568 // not actually be used: you must use the Address that is returned.
2569 // It is up to you to ensure that the shift provided matches the size
2570 // of your data.
2571 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2572   if (Address::offset_ok_for_immed(byte_offset, shift))
2573     // It fits; no need for any heroics
2574     return Address(base, byte_offset);
2575 
2576   // Don't do anything clever with negative or misaligned offsets
2577   unsigned mask = (1 << shift) - 1;
2578   if (byte_offset < 0 || byte_offset & mask) {
2579     mov(Rd, byte_offset);
2580     add(Rd, base, Rd);
2581     return Address(Rd);
2582   }
2583 
2584   // See if we can do this with two 12-bit offsets
2585   {
2586     uint64_t word_offset = byte_offset >> shift;
2587     uint64_t masked_offset = word_offset & 0xfff000;
2588     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2589         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2590       add(Rd, base, masked_offset << shift);
2591       word_offset -= masked_offset;
2592       return Address(Rd, word_offset << shift);
2593     }
2594   }
2595 
2596   // Do it the hard way
2597   mov(Rd, byte_offset);
2598   add(Rd, base, Rd);
2599   return Address(Rd);
2600 }
2601 
2602 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2603                                     bool want_remainder, Register scratch)
2604 {
2605   // Full implementation of Java idiv and irem.  The function
2606   // returns the (pc) offset of the div instruction - may be needed
2607   // for implicit exceptions.
2608   //
2609   // constraint : ra/rb =/= scratch
2610   //         normal case
2611   //
2612   // input : ra: dividend
2613   //         rb: divisor
2614   //
2615   // result: either
2616   //         quotient  (= ra idiv rb)
2617   //         remainder (= ra irem rb)
2618 
2619   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2620 
2621   int idivl_offset = offset();
2622   if (! want_remainder) {
2623     sdivw(result, ra, rb);
2624   } else {
2625     sdivw(scratch, ra, rb);
2626     Assembler::msubw(result, scratch, rb, ra);
2627   }
2628 
2629   return idivl_offset;
2630 }
2631 
2632 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2633                                     bool want_remainder, Register scratch)
2634 {
2635   // Full implementation of Java ldiv and lrem.  The function
2636   // returns the (pc) offset of the div instruction - may be needed
2637   // for implicit exceptions.
2638   //
2639   // constraint : ra/rb =/= scratch
2640   //         normal case
2641   //
2642   // input : ra: dividend
2643   //         rb: divisor
2644   //
2645   // result: either
2646   //         quotient  (= ra idiv rb)
2647   //         remainder (= ra irem rb)
2648 
2649   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2650 
2651   int idivq_offset = offset();
2652   if (! want_remainder) {
2653     sdiv(result, ra, rb);
2654   } else {
2655     sdiv(scratch, ra, rb);
2656     Assembler::msub(result, scratch, rb, ra);
2657   }
2658 
2659   return idivq_offset;
2660 }
2661 
2662 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2663   address prev = pc() - NativeMembar::instruction_size;
2664   address last = code()->last_insn();
2665   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2666     NativeMembar *bar = NativeMembar_at(prev);
2667     if (AlwaysMergeDMB) {
2668       bar->set_kind(bar->get_kind() | order_constraint);
2669       BLOCK_COMMENT("merged membar(always)");
2670       return;
2671     }
2672     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2673     // doing so would introduce a StoreLoad which the caller did not
2674     // intend
2675     if (bar->get_kind() == order_constraint
2676         || bar->get_kind() == AnyAny
2677         || order_constraint == AnyAny) {
2678       // We are merging two memory barrier instructions.  On AArch64 we
2679       // can do this simply by ORing them together.
2680       bar->set_kind(bar->get_kind() | order_constraint);
2681       BLOCK_COMMENT("merged membar");
2682       return;
2683     } else {
2684       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2685       // We need check the last 2 instructions
2686       address prev2 = prev - NativeMembar::instruction_size;
2687       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2688         NativeMembar *bar2 = NativeMembar_at(prev2);
2689         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2690         BLOCK_COMMENT("merged membar(elided)");
2691         return;
2692       }
2693     }
2694   }
2695   code()->set_last_insn(pc());
2696   dmb(Assembler::barrier(order_constraint));
2697 }
2698 
2699 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2700   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2701     merge_ldst(rt, adr, size_in_bytes, is_store);
2702     code()->clear_last_insn();
2703     return true;
2704   } else {
2705     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2706     const uint64_t mask = size_in_bytes - 1;
2707     if (adr.getMode() == Address::base_plus_offset &&
2708         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2709       code()->set_last_insn(pc());
2710     }
2711     return false;
2712   }
2713 }
2714 
2715 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2716   // We always try to merge two adjacent loads into one ldp.
2717   if (!try_merge_ldst(Rx, adr, 8, false)) {
2718     Assembler::ldr(Rx, adr);
2719   }
2720 }
2721 
2722 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2723   // We always try to merge two adjacent loads into one ldp.
2724   if (!try_merge_ldst(Rw, adr, 4, false)) {
2725     Assembler::ldrw(Rw, adr);
2726   }
2727 }
2728 
2729 void MacroAssembler::str(Register Rx, const Address &adr) {
2730   // We always try to merge two adjacent stores into one stp.
2731   if (!try_merge_ldst(Rx, adr, 8, true)) {
2732     Assembler::str(Rx, adr);
2733   }
2734 }
2735 
2736 void MacroAssembler::strw(Register Rw, const Address &adr) {
2737   // We always try to merge two adjacent stores into one stp.
2738   if (!try_merge_ldst(Rw, adr, 4, true)) {
2739     Assembler::strw(Rw, adr);
2740   }
2741 }
2742 
2743 // MacroAssembler routines found actually to be needed
2744 
2745 void MacroAssembler::push(Register src)
2746 {
2747   str(src, Address(pre(esp, -1 * wordSize)));
2748 }
2749 
2750 void MacroAssembler::pop(Register dst)
2751 {
2752   ldr(dst, Address(post(esp, 1 * wordSize)));
2753 }
2754 
2755 // Note: load_unsigned_short used to be called load_unsigned_word.
2756 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2757   int off = offset();
2758   ldrh(dst, src);
2759   return off;
2760 }
2761 
2762 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2763   int off = offset();
2764   ldrb(dst, src);
2765   return off;
2766 }
2767 
2768 int MacroAssembler::load_signed_short(Register dst, Address src) {
2769   int off = offset();
2770   ldrsh(dst, src);
2771   return off;
2772 }
2773 
2774 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2775   int off = offset();
2776   ldrsb(dst, src);
2777   return off;
2778 }
2779 
2780 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2781   int off = offset();
2782   ldrshw(dst, src);
2783   return off;
2784 }
2785 
2786 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2787   int off = offset();
2788   ldrsbw(dst, src);
2789   return off;
2790 }
2791 
2792 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2793   switch (size_in_bytes) {
2794   case  8:  ldr(dst, src); break;
2795   case  4:  ldrw(dst, src); break;
2796   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2797   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2798   default:  ShouldNotReachHere();
2799   }
2800 }
2801 
2802 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2803   switch (size_in_bytes) {
2804   case  8:  str(src, dst); break;
2805   case  4:  strw(src, dst); break;
2806   case  2:  strh(src, dst); break;
2807   case  1:  strb(src, dst); break;
2808   default:  ShouldNotReachHere();
2809   }
2810 }
2811 
2812 void MacroAssembler::decrementw(Register reg, int value)
2813 {
2814   if (value < 0)  { incrementw(reg, -value);      return; }
2815   if (value == 0) {                               return; }
2816   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2817   /* else */ {
2818     guarantee(reg != rscratch2, "invalid dst for register decrement");
2819     movw(rscratch2, (unsigned)value);
2820     subw(reg, reg, rscratch2);
2821   }
2822 }
2823 
2824 void MacroAssembler::decrement(Register reg, int value)
2825 {
2826   if (value < 0)  { increment(reg, -value);      return; }
2827   if (value == 0) {                              return; }
2828   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2829   /* else */ {
2830     assert(reg != rscratch2, "invalid dst for register decrement");
2831     mov(rscratch2, (uint64_t)value);
2832     sub(reg, reg, rscratch2);
2833   }
2834 }
2835 
2836 void MacroAssembler::decrementw(Address dst, int value)
2837 {
2838   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2839   if (dst.getMode() == Address::literal) {
2840     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2841     lea(rscratch2, dst);
2842     dst = Address(rscratch2);
2843   }
2844   ldrw(rscratch1, dst);
2845   decrementw(rscratch1, value);
2846   strw(rscratch1, dst);
2847 }
2848 
2849 void MacroAssembler::decrement(Address dst, int value)
2850 {
2851   assert(!dst.uses(rscratch1), "invalid address for decrement");
2852   if (dst.getMode() == Address::literal) {
2853     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2854     lea(rscratch2, dst);
2855     dst = Address(rscratch2);
2856   }
2857   ldr(rscratch1, dst);
2858   decrement(rscratch1, value);
2859   str(rscratch1, dst);
2860 }
2861 
2862 void MacroAssembler::incrementw(Register reg, int value)
2863 {
2864   if (value < 0)  { decrementw(reg, -value);      return; }
2865   if (value == 0) {                               return; }
2866   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2867   /* else */ {
2868     assert(reg != rscratch2, "invalid dst for register increment");
2869     movw(rscratch2, (unsigned)value);
2870     addw(reg, reg, rscratch2);
2871   }
2872 }
2873 
2874 void MacroAssembler::increment(Register reg, int value)
2875 {
2876   if (value < 0)  { decrement(reg, -value);      return; }
2877   if (value == 0) {                              return; }
2878   if (value < (1 << 12)) { add(reg, reg, value); return; }
2879   /* else */ {
2880     assert(reg != rscratch2, "invalid dst for register increment");
2881     movw(rscratch2, (unsigned)value);
2882     add(reg, reg, rscratch2);
2883   }
2884 }
2885 
2886 void MacroAssembler::incrementw(Address dst, int value)
2887 {
2888   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2889   if (dst.getMode() == Address::literal) {
2890     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2891     lea(rscratch2, dst);
2892     dst = Address(rscratch2);
2893   }
2894   ldrw(rscratch1, dst);
2895   incrementw(rscratch1, value);
2896   strw(rscratch1, dst);
2897 }
2898 
2899 void MacroAssembler::increment(Address dst, int value)
2900 {
2901   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2902   if (dst.getMode() == Address::literal) {
2903     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2904     lea(rscratch2, dst);
2905     dst = Address(rscratch2);
2906   }
2907   ldr(rscratch1, dst);
2908   increment(rscratch1, value);
2909   str(rscratch1, dst);
2910 }
2911 
2912 // Push lots of registers in the bit set supplied.  Don't push sp.
2913 // Return the number of words pushed
2914 int MacroAssembler::push(unsigned int bitset, Register stack) {
2915   int words_pushed = 0;
2916 
2917   // Scan bitset to accumulate register pairs
2918   unsigned char regs[32];
2919   int count = 0;
2920   for (int reg = 0; reg <= 30; reg++) {
2921     if (1 & bitset)
2922       regs[count++] = reg;
2923     bitset >>= 1;
2924   }
2925   regs[count++] = zr->raw_encoding();
2926   count &= ~1;  // Only push an even number of regs
2927 
2928   if (count) {
2929     stp(as_Register(regs[0]), as_Register(regs[1]),
2930        Address(pre(stack, -count * wordSize)));
2931     words_pushed += 2;
2932   }
2933   for (int i = 2; i < count; i += 2) {
2934     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2935        Address(stack, i * wordSize));
2936     words_pushed += 2;
2937   }
2938 
2939   assert(words_pushed == count, "oops, pushed != count");
2940 
2941   return count;
2942 }
2943 
2944 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2945   int words_pushed = 0;
2946 
2947   // Scan bitset to accumulate register pairs
2948   unsigned char regs[32];
2949   int count = 0;
2950   for (int reg = 0; reg <= 30; reg++) {
2951     if (1 & bitset)
2952       regs[count++] = reg;
2953     bitset >>= 1;
2954   }
2955   regs[count++] = zr->raw_encoding();
2956   count &= ~1;
2957 
2958   for (int i = 2; i < count; i += 2) {
2959     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2960        Address(stack, i * wordSize));
2961     words_pushed += 2;
2962   }
2963   if (count) {
2964     ldp(as_Register(regs[0]), as_Register(regs[1]),
2965        Address(post(stack, count * wordSize)));
2966     words_pushed += 2;
2967   }
2968 
2969   assert(words_pushed == count, "oops, pushed != count");
2970 
2971   return count;
2972 }
2973 
2974 // Push lots of registers in the bit set supplied.  Don't push sp.
2975 // Return the number of dwords pushed
2976 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2977   int words_pushed = 0;
2978   bool use_sve = false;
2979   int sve_vector_size_in_bytes = 0;
2980 
2981 #ifdef COMPILER2
2982   use_sve = Matcher::supports_scalable_vector();
2983   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2984 #endif
2985 
2986   // Scan bitset to accumulate register pairs
2987   unsigned char regs[32];
2988   int count = 0;
2989   for (int reg = 0; reg <= 31; reg++) {
2990     if (1 & bitset)
2991       regs[count++] = reg;
2992     bitset >>= 1;
2993   }
2994 
2995   if (count == 0) {
2996     return 0;
2997   }
2998 
2999   if (mode == PushPopFull) {
3000     if (use_sve && sve_vector_size_in_bytes > 16) {
3001       mode = PushPopSVE;
3002     } else {
3003       mode = PushPopNeon;
3004     }
3005   }
3006 
3007 #ifndef PRODUCT
3008   {
3009     char buffer[48];
3010     if (mode == PushPopSVE) {
3011       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3012     } else if (mode == PushPopNeon) {
3013       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3014     } else {
3015       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3016     }
3017     block_comment(buffer);
3018   }
3019 #endif
3020 
3021   if (mode == PushPopSVE) {
3022     sub(stack, stack, sve_vector_size_in_bytes * count);
3023     for (int i = 0; i < count; i++) {
3024       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3025     }
3026     return count * sve_vector_size_in_bytes / 8;
3027   }
3028 
3029   if (mode == PushPopNeon) {
3030     if (count == 1) {
3031       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3032       return 2;
3033     }
3034 
3035     bool odd = (count & 1) == 1;
3036     int push_slots = count + (odd ? 1 : 0);
3037 
3038     // Always pushing full 128 bit registers.
3039     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3040     words_pushed += 2;
3041 
3042     for (int i = 2; i + 1 < count; i += 2) {
3043       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3044       words_pushed += 2;
3045     }
3046 
3047     if (odd) {
3048       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3049       words_pushed++;
3050     }
3051 
3052     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3053     return count * 2;
3054   }
3055 
3056   if (mode == PushPopFp) {
3057     bool odd = (count & 1) == 1;
3058     int push_slots = count + (odd ? 1 : 0);
3059 
3060     if (count == 1) {
3061       // Stack pointer must be 16 bytes aligned
3062       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3063       return 1;
3064     }
3065 
3066     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3067     words_pushed += 2;
3068 
3069     for (int i = 2; i + 1 < count; i += 2) {
3070       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3071       words_pushed += 2;
3072     }
3073 
3074     if (odd) {
3075       // Stack pointer must be 16 bytes aligned
3076       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3077       words_pushed++;
3078     }
3079 
3080     assert(words_pushed == count, "oops, pushed != count");
3081 
3082     return count;
3083   }
3084 
3085   return 0;
3086 }
3087 
3088 // Return the number of dwords popped
3089 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3090   int words_pushed = 0;
3091   bool use_sve = false;
3092   int sve_vector_size_in_bytes = 0;
3093 
3094 #ifdef COMPILER2
3095   use_sve = Matcher::supports_scalable_vector();
3096   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3097 #endif
3098   // Scan bitset to accumulate register pairs
3099   unsigned char regs[32];
3100   int count = 0;
3101   for (int reg = 0; reg <= 31; reg++) {
3102     if (1 & bitset)
3103       regs[count++] = reg;
3104     bitset >>= 1;
3105   }
3106 
3107   if (count == 0) {
3108     return 0;
3109   }
3110 
3111   if (mode == PushPopFull) {
3112     if (use_sve && sve_vector_size_in_bytes > 16) {
3113       mode = PushPopSVE;
3114     } else {
3115       mode = PushPopNeon;
3116     }
3117   }
3118 
3119 #ifndef PRODUCT
3120   {
3121     char buffer[48];
3122     if (mode == PushPopSVE) {
3123       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3124     } else if (mode == PushPopNeon) {
3125       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3126     } else {
3127       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3128     }
3129     block_comment(buffer);
3130   }
3131 #endif
3132 
3133   if (mode == PushPopSVE) {
3134     for (int i = count - 1; i >= 0; i--) {
3135       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3136     }
3137     add(stack, stack, sve_vector_size_in_bytes * count);
3138     return count * sve_vector_size_in_bytes / 8;
3139   }
3140 
3141   if (mode == PushPopNeon) {
3142     if (count == 1) {
3143       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3144       return 2;
3145     }
3146 
3147     bool odd = (count & 1) == 1;
3148     int push_slots = count + (odd ? 1 : 0);
3149 
3150     if (odd) {
3151       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3152       words_pushed++;
3153     }
3154 
3155     for (int i = 2; i + 1 < count; i += 2) {
3156       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3157       words_pushed += 2;
3158     }
3159 
3160     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3161     words_pushed += 2;
3162 
3163     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3164 
3165     return count * 2;
3166   }
3167 
3168   if (mode == PushPopFp) {
3169     bool odd = (count & 1) == 1;
3170     int push_slots = count + (odd ? 1 : 0);
3171 
3172     if (count == 1) {
3173       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3174       return 1;
3175     }
3176 
3177     if (odd) {
3178       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3179       words_pushed++;
3180     }
3181 
3182     for (int i = 2; i + 1 < count; i += 2) {
3183       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3184       words_pushed += 2;
3185     }
3186 
3187     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3188     words_pushed += 2;
3189 
3190     assert(words_pushed == count, "oops, pushed != count");
3191 
3192     return count;
3193   }
3194 
3195   return 0;
3196 }
3197 
3198 // Return the number of dwords pushed
3199 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3200   bool use_sve = false;
3201   int sve_predicate_size_in_slots = 0;
3202 
3203 #ifdef COMPILER2
3204   use_sve = Matcher::supports_scalable_vector();
3205   if (use_sve) {
3206     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3207   }
3208 #endif
3209 
3210   if (!use_sve) {
3211     return 0;
3212   }
3213 
3214   unsigned char regs[PRegister::number_of_registers];
3215   int count = 0;
3216   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3217     if (1 & bitset)
3218       regs[count++] = reg;
3219     bitset >>= 1;
3220   }
3221 
3222   if (count == 0) {
3223     return 0;
3224   }
3225 
3226   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3227                                   VMRegImpl::stack_slot_size * count, 16);
3228   sub(stack, stack, total_push_bytes);
3229   for (int i = 0; i < count; i++) {
3230     sve_str(as_PRegister(regs[i]), Address(stack, i));
3231   }
3232   return total_push_bytes / 8;
3233 }
3234 
3235 // Return the number of dwords popped
3236 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3237   bool use_sve = false;
3238   int sve_predicate_size_in_slots = 0;
3239 
3240 #ifdef COMPILER2
3241   use_sve = Matcher::supports_scalable_vector();
3242   if (use_sve) {
3243     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3244   }
3245 #endif
3246 
3247   if (!use_sve) {
3248     return 0;
3249   }
3250 
3251   unsigned char regs[PRegister::number_of_registers];
3252   int count = 0;
3253   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3254     if (1 & bitset)
3255       regs[count++] = reg;
3256     bitset >>= 1;
3257   }
3258 
3259   if (count == 0) {
3260     return 0;
3261   }
3262 
3263   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3264                                  VMRegImpl::stack_slot_size * count, 16);
3265   for (int i = count - 1; i >= 0; i--) {
3266     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3267   }
3268   add(stack, stack, total_pop_bytes);
3269   return total_pop_bytes / 8;
3270 }
3271 
3272 #ifdef ASSERT
3273 void MacroAssembler::verify_heapbase(const char* msg) {
3274 #if 0
3275   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3276   assert (Universe::heap() != nullptr, "java heap should be initialized");
3277   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3278     // rheapbase is allocated as general register
3279     return;
3280   }
3281   if (CheckCompressedOops) {
3282     Label ok;
3283     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3284     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3285     br(Assembler::EQ, ok);
3286     stop(msg);
3287     bind(ok);
3288     pop(1 << rscratch1->encoding(), sp);
3289   }
3290 #endif
3291 }
3292 #endif
3293 
3294 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3295   assert_different_registers(value, tmp1, tmp2);
3296   Label done, tagged, weak_tagged;
3297 
3298   cbz(value, done);           // Use null as-is.
3299   tst(value, JNIHandles::tag_mask); // Test for tag.
3300   br(Assembler::NE, tagged);
3301 
3302   // Resolve local handle
3303   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3304   verify_oop(value);
3305   b(done);
3306 
3307   bind(tagged);
3308   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3309   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3310 
3311   // Resolve global handle
3312   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3313   verify_oop(value);
3314   b(done);
3315 
3316   bind(weak_tagged);
3317   // Resolve jweak.
3318   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3319                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3320   verify_oop(value);
3321 
3322   bind(done);
3323 }
3324 
3325 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3326   assert_different_registers(value, tmp1, tmp2);
3327   Label done;
3328 
3329   cbz(value, done);           // Use null as-is.
3330 
3331 #ifdef ASSERT
3332   {
3333     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3334     Label valid_global_tag;
3335     tbnz(value, 1, valid_global_tag); // Test for global tag
3336     stop("non global jobject using resolve_global_jobject");
3337     bind(valid_global_tag);
3338   }
3339 #endif
3340 
3341   // Resolve global handle
3342   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3343   verify_oop(value);
3344 
3345   bind(done);
3346 }
3347 
3348 void MacroAssembler::stop(const char* msg) {
3349   // Skip AOT caching C strings in scratch buffer.
3350   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3351   BLOCK_COMMENT(str);
3352   // load msg into r0 so we can access it from the signal handler
3353   // ExternalAddress enables saving and restoring via the code cache
3354   lea(c_rarg0, ExternalAddress((address) str));
3355   dcps1(0xdeae);
3356 }
3357 
3358 void MacroAssembler::unimplemented(const char* what) {
3359   const char* buf = nullptr;
3360   {
3361     ResourceMark rm;
3362     stringStream ss;
3363     ss.print("unimplemented: %s", what);
3364     buf = code_string(ss.as_string());
3365   }
3366   stop(buf);
3367 }
3368 
3369 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3370 #ifdef ASSERT
3371   Label OK;
3372   br(cc, OK);
3373   stop(msg);
3374   bind(OK);
3375 #endif
3376 }
3377 
3378 // If a constant does not fit in an immediate field, generate some
3379 // number of MOV instructions and then perform the operation.
3380 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3381                                            add_sub_imm_insn insn1,
3382                                            add_sub_reg_insn insn2,
3383                                            bool is32) {
3384   assert(Rd != zr, "Rd = zr and not setting flags?");
3385   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3386   if (fits) {
3387     (this->*insn1)(Rd, Rn, imm);
3388   } else {
3389     if (g_uabs(imm) < (1 << 24)) {
3390        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3391        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3392     } else {
3393        assert_different_registers(Rd, Rn);
3394        mov(Rd, imm);
3395        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3396     }
3397   }
3398 }
3399 
3400 // Separate vsn which sets the flags. Optimisations are more restricted
3401 // because we must set the flags correctly.
3402 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3403                                              add_sub_imm_insn insn1,
3404                                              add_sub_reg_insn insn2,
3405                                              bool is32) {
3406   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3407   if (fits) {
3408     (this->*insn1)(Rd, Rn, imm);
3409   } else {
3410     assert_different_registers(Rd, Rn);
3411     assert(Rd != zr, "overflow in immediate operand");
3412     mov(Rd, imm);
3413     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3414   }
3415 }
3416 
3417 
3418 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3419   if (increment.is_register()) {
3420     add(Rd, Rn, increment.as_register());
3421   } else {
3422     add(Rd, Rn, increment.as_constant());
3423   }
3424 }
3425 
3426 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3427   if (increment.is_register()) {
3428     addw(Rd, Rn, increment.as_register());
3429   } else {
3430     addw(Rd, Rn, increment.as_constant());
3431   }
3432 }
3433 
3434 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3435   if (decrement.is_register()) {
3436     sub(Rd, Rn, decrement.as_register());
3437   } else {
3438     sub(Rd, Rn, decrement.as_constant());
3439   }
3440 }
3441 
3442 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3443   if (decrement.is_register()) {
3444     subw(Rd, Rn, decrement.as_register());
3445   } else {
3446     subw(Rd, Rn, decrement.as_constant());
3447   }
3448 }
3449 
3450 void MacroAssembler::reinit_heapbase()
3451 {
3452   if (UseCompressedOops) {
3453     if (Universe::is_fully_initialized()) {
3454       mov(rheapbase, CompressedOops::base());
3455     } else {
3456       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3457       ldr(rheapbase, Address(rheapbase));
3458     }
3459   }
3460 }
3461 
3462 // this simulates the behaviour of the x86 cmpxchg instruction using a
3463 // load linked/store conditional pair. we use the acquire/release
3464 // versions of these instructions so that we flush pending writes as
3465 // per Java semantics.
3466 
3467 // n.b the x86 version assumes the old value to be compared against is
3468 // in rax and updates rax with the value located in memory if the
3469 // cmpxchg fails. we supply a register for the old value explicitly
3470 
3471 // the aarch64 load linked/store conditional instructions do not
3472 // accept an offset. so, unlike x86, we must provide a plain register
3473 // to identify the memory word to be compared/exchanged rather than a
3474 // register+offset Address.
3475 
3476 void MacroAssembler::cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
3477                                 Label &succeed, Label *fail) {
3478   // oldv holds comparison value
3479   // newv holds value to write in exchange
3480   // addr identifies memory word to compare against/update
3481   if (UseLSE) {
3482     mov(tmp, oldv);
3483     casal(Assembler::xword, oldv, newv, addr);
3484     cmp(tmp, oldv);
3485     br(Assembler::EQ, succeed);
3486     membar(AnyAny);
3487   } else {
3488     Label retry_load, nope;
3489     prfm(Address(addr), PSTL1STRM);
3490     bind(retry_load);
3491     // flush and load exclusive from the memory location
3492     // and fail if it is not what we expect
3493     ldaxr(tmp, addr);
3494     cmp(tmp, oldv);
3495     br(Assembler::NE, nope);
3496     // if we store+flush with no intervening write tmp will be zero
3497     stlxr(tmp, newv, addr);
3498     cbzw(tmp, succeed);
3499     // retry so we only ever return after a load fails to compare
3500     // ensures we don't return a stale value after a failed write.
3501     b(retry_load);
3502     // if the memory word differs we return it in oldv and signal a fail
3503     bind(nope);
3504     membar(AnyAny);
3505     mov(oldv, tmp);
3506   }
3507   if (fail)
3508     b(*fail);
3509 }
3510 
3511 void MacroAssembler::cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp,
3512                                         Label &succeed, Label *fail) {
3513   assert(oopDesc::mark_offset_in_bytes() == 0, "assumption");
3514   cmpxchgptr(oldv, newv, obj, tmp, succeed, fail);
3515 }
3516 
3517 void MacroAssembler::cmpxchgw(Register oldv, Register newv, Register addr, Register tmp,
3518                                 Label &succeed, Label *fail) {
3519   // oldv holds comparison value
3520   // newv holds value to write in exchange
3521   // addr identifies memory word to compare against/update
3522   // tmp returns 0/1 for success/failure
3523   if (UseLSE) {
3524     mov(tmp, oldv);
3525     casal(Assembler::word, oldv, newv, addr);
3526     cmp(tmp, oldv);
3527     br(Assembler::EQ, succeed);
3528     membar(AnyAny);
3529   } else {
3530     Label retry_load, nope;
3531     prfm(Address(addr), PSTL1STRM);
3532     bind(retry_load);
3533     // flush and load exclusive from the memory location
3534     // and fail if it is not what we expect
3535     ldaxrw(tmp, addr);
3536     cmp(tmp, oldv);
3537     br(Assembler::NE, nope);
3538     // if we store+flush with no intervening write tmp will be zero
3539     stlxrw(tmp, newv, addr);
3540     cbzw(tmp, succeed);
3541     // retry so we only ever return after a load fails to compare
3542     // ensures we don't return a stale value after a failed write.
3543     b(retry_load);
3544     // if the memory word differs we return it in oldv and signal a fail
3545     bind(nope);
3546     membar(AnyAny);
3547     mov(oldv, tmp);
3548   }
3549   if (fail)
3550     b(*fail);
3551 }
3552 
3553 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3554 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3555 // Pass a register for the result, otherwise pass noreg.
3556 
3557 // Clobbers rscratch1
3558 void MacroAssembler::cmpxchg(Register addr, Register expected,
3559                              Register new_val,
3560                              enum operand_size size,
3561                              bool acquire, bool release,
3562                              bool weak,
3563                              Register result) {
3564   if (result == noreg)  result = rscratch1;
3565   BLOCK_COMMENT("cmpxchg {");
3566   if (UseLSE) {
3567     mov(result, expected);
3568     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3569     compare_eq(result, expected, size);
3570 #ifdef ASSERT
3571     // Poison rscratch1 which is written on !UseLSE branch
3572     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3573 #endif
3574   } else {
3575     Label retry_load, done;
3576     prfm(Address(addr), PSTL1STRM);
3577     bind(retry_load);
3578     load_exclusive(result, addr, size, acquire);
3579     compare_eq(result, expected, size);
3580     br(Assembler::NE, done);
3581     store_exclusive(rscratch1, new_val, addr, size, release);
3582     if (weak) {
3583       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3584     } else {
3585       cbnzw(rscratch1, retry_load);
3586     }
3587     bind(done);
3588   }
3589   BLOCK_COMMENT("} cmpxchg");
3590 }
3591 
3592 // A generic comparison. Only compares for equality, clobbers rscratch1.
3593 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3594   if (size == xword) {
3595     cmp(rm, rn);
3596   } else if (size == word) {
3597     cmpw(rm, rn);
3598   } else if (size == halfword) {
3599     eorw(rscratch1, rm, rn);
3600     ands(zr, rscratch1, 0xffff);
3601   } else if (size == byte) {
3602     eorw(rscratch1, rm, rn);
3603     ands(zr, rscratch1, 0xff);
3604   } else {
3605     ShouldNotReachHere();
3606   }
3607 }
3608 
3609 
3610 static bool different(Register a, RegisterOrConstant b, Register c) {
3611   if (b.is_constant())
3612     return a != c;
3613   else
3614     return a != b.as_register() && a != c && b.as_register() != c;
3615 }
3616 
3617 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3618 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3619   if (UseLSE) {                                                         \
3620     prev = prev->is_valid() ? prev : zr;                                \
3621     if (incr.is_register()) {                                           \
3622       AOP(sz, incr.as_register(), prev, addr);                          \
3623     } else {                                                            \
3624       mov(rscratch2, incr.as_constant());                               \
3625       AOP(sz, rscratch2, prev, addr);                                   \
3626     }                                                                   \
3627     return;                                                             \
3628   }                                                                     \
3629   Register result = rscratch2;                                          \
3630   if (prev->is_valid())                                                 \
3631     result = different(prev, incr, addr) ? prev : rscratch2;            \
3632                                                                         \
3633   Label retry_load;                                                     \
3634   prfm(Address(addr), PSTL1STRM);                                       \
3635   bind(retry_load);                                                     \
3636   LDXR(result, addr);                                                   \
3637   OP(rscratch1, result, incr);                                          \
3638   STXR(rscratch2, rscratch1, addr);                                     \
3639   cbnzw(rscratch2, retry_load);                                         \
3640   if (prev->is_valid() && prev != result) {                             \
3641     IOP(prev, rscratch1, incr);                                         \
3642   }                                                                     \
3643 }
3644 
3645 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3646 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3647 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3648 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3649 
3650 #undef ATOMIC_OP
3651 
3652 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3653 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3654   if (UseLSE) {                                                         \
3655     prev = prev->is_valid() ? prev : zr;                                \
3656     AOP(sz, newv, prev, addr);                                          \
3657     return;                                                             \
3658   }                                                                     \
3659   Register result = rscratch2;                                          \
3660   if (prev->is_valid())                                                 \
3661     result = different(prev, newv, addr) ? prev : rscratch2;            \
3662                                                                         \
3663   Label retry_load;                                                     \
3664   prfm(Address(addr), PSTL1STRM);                                       \
3665   bind(retry_load);                                                     \
3666   LDXR(result, addr);                                                   \
3667   STXR(rscratch1, newv, addr);                                          \
3668   cbnzw(rscratch1, retry_load);                                         \
3669   if (prev->is_valid() && prev != result)                               \
3670     mov(prev, result);                                                  \
3671 }
3672 
3673 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3674 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3675 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3676 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3677 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3678 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3679 
3680 #undef ATOMIC_XCHG
3681 
3682 #ifndef PRODUCT
3683 extern "C" void findpc(intptr_t x);
3684 #endif
3685 
3686 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3687 {
3688   // In order to get locks to work, we need to fake a in_VM state
3689   if (ShowMessageBoxOnError ) {
3690     JavaThread* thread = JavaThread::current();
3691     JavaThreadState saved_state = thread->thread_state();
3692     thread->set_thread_state(_thread_in_vm);
3693 #ifndef PRODUCT
3694     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3695       ttyLocker ttyl;
3696       BytecodeCounter::print();
3697     }
3698 #endif
3699     if (os::message_box(msg, "Execution stopped, print registers?")) {
3700       ttyLocker ttyl;
3701       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3702 #ifndef PRODUCT
3703       tty->cr();
3704       findpc(pc);
3705       tty->cr();
3706 #endif
3707       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3708       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3709       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3710       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3711       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3712       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3713       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3714       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3715       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3716       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3717       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3718       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3719       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3720       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3721       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3722       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3723       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3724       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3725       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3726       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3727       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3728       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3729       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3730       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3731       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3732       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3733       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3734       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3735       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3736       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3737       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3738       BREAKPOINT;
3739     }
3740   }
3741   fatal("DEBUG MESSAGE: %s", msg);
3742 }
3743 
3744 RegSet MacroAssembler::call_clobbered_gp_registers() {
3745   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3746 #ifndef R18_RESERVED
3747   regs += r18_tls;
3748 #endif
3749   return regs;
3750 }
3751 
3752 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3753   int step = 4 * wordSize;
3754   push(call_clobbered_gp_registers() - exclude, sp);
3755   sub(sp, sp, step);
3756   mov(rscratch1, -step);
3757   // Push v0-v7, v16-v31.
3758   for (int i = 31; i>= 4; i -= 4) {
3759     if (i <= v7->encoding() || i >= v16->encoding())
3760       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3761           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3762   }
3763   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3764       as_FloatRegister(3), T1D, Address(sp));
3765 }
3766 
3767 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3768   for (int i = 0; i < 32; i += 4) {
3769     if (i <= v7->encoding() || i >= v16->encoding())
3770       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3771           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3772   }
3773 
3774   reinitialize_ptrue();
3775 
3776   pop(call_clobbered_gp_registers() - exclude, sp);
3777 }
3778 
3779 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3780                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3781   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3782   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3783     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3784     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3785       sve_str(as_FloatRegister(i), Address(sp, i));
3786     }
3787   } else {
3788     int step = (save_vectors ? 8 : 4) * wordSize;
3789     mov(rscratch1, -step);
3790     sub(sp, sp, step);
3791     for (int i = 28; i >= 4; i -= 4) {
3792       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3793           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3794     }
3795     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3796   }
3797   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3798     sub(sp, sp, total_predicate_in_bytes);
3799     for (int i = 0; i < PRegister::number_of_registers; i++) {
3800       sve_str(as_PRegister(i), Address(sp, i));
3801     }
3802   }
3803 }
3804 
3805 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3806                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3807   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3808     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3809       sve_ldr(as_PRegister(i), Address(sp, i));
3810     }
3811     add(sp, sp, total_predicate_in_bytes);
3812   }
3813   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3814     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3815       sve_ldr(as_FloatRegister(i), Address(sp, i));
3816     }
3817     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3818   } else {
3819     int step = (restore_vectors ? 8 : 4) * wordSize;
3820     for (int i = 0; i <= 28; i += 4)
3821       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3822           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3823   }
3824 
3825   // We may use predicate registers and rely on ptrue with SVE,
3826   // regardless of wide vector (> 8 bytes) used or not.
3827   if (use_sve) {
3828     reinitialize_ptrue();
3829   }
3830 
3831   // integer registers except lr & sp
3832   pop(RegSet::range(r0, r17), sp);
3833 #ifdef R18_RESERVED
3834   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3835   pop(RegSet::range(r20, r29), sp);
3836 #else
3837   pop(RegSet::range(r18_tls, r29), sp);
3838 #endif
3839 }
3840 
3841 /**
3842  * Helpers for multiply_to_len().
3843  */
3844 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3845                                      Register src1, Register src2) {
3846   adds(dest_lo, dest_lo, src1);
3847   adc(dest_hi, dest_hi, zr);
3848   adds(dest_lo, dest_lo, src2);
3849   adc(final_dest_hi, dest_hi, zr);
3850 }
3851 
3852 // Generate an address from (r + r1 extend offset).  "size" is the
3853 // size of the operand.  The result may be in rscratch2.
3854 Address MacroAssembler::offsetted_address(Register r, Register r1,
3855                                           Address::extend ext, int offset, int size) {
3856   if (offset || (ext.shift() % size != 0)) {
3857     lea(rscratch2, Address(r, r1, ext));
3858     return Address(rscratch2, offset);
3859   } else {
3860     return Address(r, r1, ext);
3861   }
3862 }
3863 
3864 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3865 {
3866   assert(offset >= 0, "spill to negative address?");
3867   // Offset reachable ?
3868   //   Not aligned - 9 bits signed offset
3869   //   Aligned - 12 bits unsigned offset shifted
3870   Register base = sp;
3871   if ((offset & (size-1)) && offset >= (1<<8)) {
3872     add(tmp, base, offset & ((1<<12)-1));
3873     base = tmp;
3874     offset &= -1u<<12;
3875   }
3876 
3877   if (offset >= (1<<12) * size) {
3878     add(tmp, base, offset & (((1<<12)-1)<<12));
3879     base = tmp;
3880     offset &= ~(((1<<12)-1)<<12);
3881   }
3882 
3883   return Address(base, offset);
3884 }
3885 
3886 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3887   assert(offset >= 0, "spill to negative address?");
3888 
3889   Register base = sp;
3890 
3891   // An immediate offset in the range 0 to 255 which is multiplied
3892   // by the current vector or predicate register size in bytes.
3893   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3894     return Address(base, offset / sve_reg_size_in_bytes);
3895   }
3896 
3897   add(tmp, base, offset);
3898   return Address(tmp);
3899 }
3900 
3901 // Checks whether offset is aligned.
3902 // Returns true if it is, else false.
3903 bool MacroAssembler::merge_alignment_check(Register base,
3904                                            size_t size,
3905                                            int64_t cur_offset,
3906                                            int64_t prev_offset) const {
3907   if (AvoidUnalignedAccesses) {
3908     if (base == sp) {
3909       // Checks whether low offset if aligned to pair of registers.
3910       int64_t pair_mask = size * 2 - 1;
3911       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3912       return (offset & pair_mask) == 0;
3913     } else { // If base is not sp, we can't guarantee the access is aligned.
3914       return false;
3915     }
3916   } else {
3917     int64_t mask = size - 1;
3918     // Load/store pair instruction only supports element size aligned offset.
3919     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3920   }
3921 }
3922 
3923 // Checks whether current and previous loads/stores can be merged.
3924 // Returns true if it can be merged, else false.
3925 bool MacroAssembler::ldst_can_merge(Register rt,
3926                                     const Address &adr,
3927                                     size_t cur_size_in_bytes,
3928                                     bool is_store) const {
3929   address prev = pc() - NativeInstruction::instruction_size;
3930   address last = code()->last_insn();
3931 
3932   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3933     return false;
3934   }
3935 
3936   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3937     return false;
3938   }
3939 
3940   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3941   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3942 
3943   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3944   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3945 
3946   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3947     return false;
3948   }
3949 
3950   int64_t max_offset = 63 * prev_size_in_bytes;
3951   int64_t min_offset = -64 * prev_size_in_bytes;
3952 
3953   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3954 
3955   // Only same base can be merged.
3956   if (adr.base() != prev_ldst->base()) {
3957     return false;
3958   }
3959 
3960   int64_t cur_offset = adr.offset();
3961   int64_t prev_offset = prev_ldst->offset();
3962   size_t diff = abs(cur_offset - prev_offset);
3963   if (diff != prev_size_in_bytes) {
3964     return false;
3965   }
3966 
3967   // Following cases can not be merged:
3968   // ldr x2, [x2, #8]
3969   // ldr x3, [x2, #16]
3970   // or:
3971   // ldr x2, [x3, #8]
3972   // ldr x2, [x3, #16]
3973   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3974   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3975     return false;
3976   }
3977 
3978   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3979   // Offset range must be in ldp/stp instruction's range.
3980   if (low_offset > max_offset || low_offset < min_offset) {
3981     return false;
3982   }
3983 
3984   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3985     return true;
3986   }
3987 
3988   return false;
3989 }
3990 
3991 // Merge current load/store with previous load/store into ldp/stp.
3992 void MacroAssembler::merge_ldst(Register rt,
3993                                 const Address &adr,
3994                                 size_t cur_size_in_bytes,
3995                                 bool is_store) {
3996 
3997   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3998 
3999   Register rt_low, rt_high;
4000   address prev = pc() - NativeInstruction::instruction_size;
4001   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
4002 
4003   int64_t offset;
4004 
4005   if (adr.offset() < prev_ldst->offset()) {
4006     offset = adr.offset();
4007     rt_low = rt;
4008     rt_high = prev_ldst->target();
4009   } else {
4010     offset = prev_ldst->offset();
4011     rt_low = prev_ldst->target();
4012     rt_high = rt;
4013   }
4014 
4015   Address adr_p = Address(prev_ldst->base(), offset);
4016   // Overwrite previous generated binary.
4017   code_section()->set_end(prev);
4018 
4019   const size_t sz = prev_ldst->size_in_bytes();
4020   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
4021   if (!is_store) {
4022     BLOCK_COMMENT("merged ldr pair");
4023     if (sz == 8) {
4024       ldp(rt_low, rt_high, adr_p);
4025     } else {
4026       ldpw(rt_low, rt_high, adr_p);
4027     }
4028   } else {
4029     BLOCK_COMMENT("merged str pair");
4030     if (sz == 8) {
4031       stp(rt_low, rt_high, adr_p);
4032     } else {
4033       stpw(rt_low, rt_high, adr_p);
4034     }
4035   }
4036 }
4037 
4038 /**
4039  * Multiply 64 bit by 64 bit first loop.
4040  */
4041 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
4042                                            Register y, Register y_idx, Register z,
4043                                            Register carry, Register product,
4044                                            Register idx, Register kdx) {
4045   //
4046   //  jlong carry, x[], y[], z[];
4047   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4048   //    huge_128 product = y[idx] * x[xstart] + carry;
4049   //    z[kdx] = (jlong)product;
4050   //    carry  = (jlong)(product >>> 64);
4051   //  }
4052   //  z[xstart] = carry;
4053   //
4054 
4055   Label L_first_loop, L_first_loop_exit;
4056   Label L_one_x, L_one_y, L_multiply;
4057 
4058   subsw(xstart, xstart, 1);
4059   br(Assembler::MI, L_one_x);
4060 
4061   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
4062   ldr(x_xstart, Address(rscratch1));
4063   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
4064 
4065   bind(L_first_loop);
4066   subsw(idx, idx, 1);
4067   br(Assembler::MI, L_first_loop_exit);
4068   subsw(idx, idx, 1);
4069   br(Assembler::MI, L_one_y);
4070   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4071   ldr(y_idx, Address(rscratch1));
4072   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
4073   bind(L_multiply);
4074 
4075   // AArch64 has a multiply-accumulate instruction that we can't use
4076   // here because it has no way to process carries, so we have to use
4077   // separate add and adc instructions.  Bah.
4078   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
4079   mul(product, x_xstart, y_idx);
4080   adds(product, product, carry);
4081   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
4082 
4083   subw(kdx, kdx, 2);
4084   ror(product, product, 32); // back to big-endian
4085   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4086 
4087   b(L_first_loop);
4088 
4089   bind(L_one_y);
4090   ldrw(y_idx, Address(y,  0));
4091   b(L_multiply);
4092 
4093   bind(L_one_x);
4094   ldrw(x_xstart, Address(x,  0));
4095   b(L_first_loop);
4096 
4097   bind(L_first_loop_exit);
4098 }
4099 
4100 /**
4101  * Multiply 128 bit by 128. Unrolled inner loop.
4102  *
4103  */
4104 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4105                                              Register carry, Register carry2,
4106                                              Register idx, Register jdx,
4107                                              Register yz_idx1, Register yz_idx2,
4108                                              Register tmp, Register tmp3, Register tmp4,
4109                                              Register tmp6, Register product_hi) {
4110 
4111   //   jlong carry, x[], y[], z[];
4112   //   int kdx = ystart+1;
4113   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4114   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4115   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4116   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4117   //     carry  = (jlong)(tmp4 >>> 64);
4118   //     z[kdx+idx+1] = (jlong)tmp3;
4119   //     z[kdx+idx] = (jlong)tmp4;
4120   //   }
4121   //   idx += 2;
4122   //   if (idx > 0) {
4123   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4124   //     z[kdx+idx] = (jlong)yz_idx1;
4125   //     carry  = (jlong)(yz_idx1 >>> 64);
4126   //   }
4127   //
4128 
4129   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4130 
4131   lsrw(jdx, idx, 2);
4132 
4133   bind(L_third_loop);
4134 
4135   subsw(jdx, jdx, 1);
4136   br(Assembler::MI, L_third_loop_exit);
4137   subw(idx, idx, 4);
4138 
4139   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4140 
4141   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4142 
4143   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4144 
4145   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4146   ror(yz_idx2, yz_idx2, 32);
4147 
4148   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4149 
4150   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4151   umulh(tmp4, product_hi, yz_idx1);
4152 
4153   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4154   ror(rscratch2, rscratch2, 32);
4155 
4156   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4157   umulh(carry2, product_hi, yz_idx2);
4158 
4159   // propagate sum of both multiplications into carry:tmp4:tmp3
4160   adds(tmp3, tmp3, carry);
4161   adc(tmp4, tmp4, zr);
4162   adds(tmp3, tmp3, rscratch1);
4163   adcs(tmp4, tmp4, tmp);
4164   adc(carry, carry2, zr);
4165   adds(tmp4, tmp4, rscratch2);
4166   adc(carry, carry, zr);
4167 
4168   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4169   ror(tmp4, tmp4, 32);
4170   stp(tmp4, tmp3, Address(tmp6, 0));
4171 
4172   b(L_third_loop);
4173   bind (L_third_loop_exit);
4174 
4175   andw (idx, idx, 0x3);
4176   cbz(idx, L_post_third_loop_done);
4177 
4178   Label L_check_1;
4179   subsw(idx, idx, 2);
4180   br(Assembler::MI, L_check_1);
4181 
4182   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4183   ldr(yz_idx1, Address(rscratch1, 0));
4184   ror(yz_idx1, yz_idx1, 32);
4185   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4186   umulh(tmp4, product_hi, yz_idx1);
4187   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4188   ldr(yz_idx2, Address(rscratch1, 0));
4189   ror(yz_idx2, yz_idx2, 32);
4190 
4191   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4192 
4193   ror(tmp3, tmp3, 32);
4194   str(tmp3, Address(rscratch1, 0));
4195 
4196   bind (L_check_1);
4197 
4198   andw (idx, idx, 0x1);
4199   subsw(idx, idx, 1);
4200   br(Assembler::MI, L_post_third_loop_done);
4201   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4202   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4203   umulh(carry2, tmp4, product_hi);
4204   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4205 
4206   add2_with_carry(carry2, tmp3, tmp4, carry);
4207 
4208   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4209   extr(carry, carry2, tmp3, 32);
4210 
4211   bind(L_post_third_loop_done);
4212 }
4213 
4214 /**
4215  * Code for BigInteger::multiplyToLen() intrinsic.
4216  *
4217  * r0: x
4218  * r1: xlen
4219  * r2: y
4220  * r3: ylen
4221  * r4:  z
4222  * r5: tmp0
4223  * r10: tmp1
4224  * r11: tmp2
4225  * r12: tmp3
4226  * r13: tmp4
4227  * r14: tmp5
4228  * r15: tmp6
4229  * r16: tmp7
4230  *
4231  */
4232 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4233                                      Register z, Register tmp0,
4234                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4235                                      Register tmp5, Register tmp6, Register product_hi) {
4236 
4237   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4238 
4239   const Register idx = tmp1;
4240   const Register kdx = tmp2;
4241   const Register xstart = tmp3;
4242 
4243   const Register y_idx = tmp4;
4244   const Register carry = tmp5;
4245   const Register product  = xlen;
4246   const Register x_xstart = tmp0;
4247 
4248   // First Loop.
4249   //
4250   //  final static long LONG_MASK = 0xffffffffL;
4251   //  int xstart = xlen - 1;
4252   //  int ystart = ylen - 1;
4253   //  long carry = 0;
4254   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4255   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4256   //    z[kdx] = (int)product;
4257   //    carry = product >>> 32;
4258   //  }
4259   //  z[xstart] = (int)carry;
4260   //
4261 
4262   movw(idx, ylen);       // idx = ylen;
4263   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4264   mov(carry, zr);        // carry = 0;
4265 
4266   Label L_done;
4267 
4268   movw(xstart, xlen);
4269   subsw(xstart, xstart, 1);
4270   br(Assembler::MI, L_done);
4271 
4272   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4273 
4274   Label L_second_loop;
4275   cbzw(kdx, L_second_loop);
4276 
4277   Label L_carry;
4278   subw(kdx, kdx, 1);
4279   cbzw(kdx, L_carry);
4280 
4281   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4282   lsr(carry, carry, 32);
4283   subw(kdx, kdx, 1);
4284 
4285   bind(L_carry);
4286   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4287 
4288   // Second and third (nested) loops.
4289   //
4290   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4291   //   carry = 0;
4292   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4293   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4294   //                    (z[k] & LONG_MASK) + carry;
4295   //     z[k] = (int)product;
4296   //     carry = product >>> 32;
4297   //   }
4298   //   z[i] = (int)carry;
4299   // }
4300   //
4301   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4302 
4303   const Register jdx = tmp1;
4304 
4305   bind(L_second_loop);
4306   mov(carry, zr);                // carry = 0;
4307   movw(jdx, ylen);               // j = ystart+1
4308 
4309   subsw(xstart, xstart, 1);      // i = xstart-1;
4310   br(Assembler::MI, L_done);
4311 
4312   str(z, Address(pre(sp, -4 * wordSize)));
4313 
4314   Label L_last_x;
4315   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4316   subsw(xstart, xstart, 1);       // i = xstart-1;
4317   br(Assembler::MI, L_last_x);
4318 
4319   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4320   ldr(product_hi, Address(rscratch1));
4321   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4322 
4323   Label L_third_loop_prologue;
4324   bind(L_third_loop_prologue);
4325 
4326   str(ylen, Address(sp, wordSize));
4327   stp(x, xstart, Address(sp, 2 * wordSize));
4328   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4329                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4330   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4331   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4332 
4333   addw(tmp3, xlen, 1);
4334   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4335   subsw(tmp3, tmp3, 1);
4336   br(Assembler::MI, L_done);
4337 
4338   lsr(carry, carry, 32);
4339   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4340   b(L_second_loop);
4341 
4342   // Next infrequent code is moved outside loops.
4343   bind(L_last_x);
4344   ldrw(product_hi, Address(x,  0));
4345   b(L_third_loop_prologue);
4346 
4347   bind(L_done);
4348 }
4349 
4350 // Code for BigInteger::mulAdd intrinsic
4351 // out     = r0
4352 // in      = r1
4353 // offset  = r2  (already out.length-offset)
4354 // len     = r3
4355 // k       = r4
4356 //
4357 // pseudo code from java implementation:
4358 // carry = 0;
4359 // offset = out.length-offset - 1;
4360 // for (int j=len-1; j >= 0; j--) {
4361 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4362 //     out[offset--] = (int)product;
4363 //     carry = product >>> 32;
4364 // }
4365 // return (int)carry;
4366 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4367       Register len, Register k) {
4368     Label LOOP, END;
4369     // pre-loop
4370     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4371     csel(out, zr, out, Assembler::EQ);
4372     br(Assembler::EQ, END);
4373     add(in, in, len, LSL, 2); // in[j+1] address
4374     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4375     mov(out, zr); // used to keep carry now
4376     BIND(LOOP);
4377     ldrw(rscratch1, Address(pre(in, -4)));
4378     madd(rscratch1, rscratch1, k, out);
4379     ldrw(rscratch2, Address(pre(offset, -4)));
4380     add(rscratch1, rscratch1, rscratch2);
4381     strw(rscratch1, Address(offset));
4382     lsr(out, rscratch1, 32);
4383     subs(len, len, 1);
4384     br(Assembler::NE, LOOP);
4385     BIND(END);
4386 }
4387 
4388 /**
4389  * Emits code to update CRC-32 with a byte value according to constants in table
4390  *
4391  * @param [in,out]crc   Register containing the crc.
4392  * @param [in]val       Register containing the byte to fold into the CRC.
4393  * @param [in]table     Register containing the table of crc constants.
4394  *
4395  * uint32_t crc;
4396  * val = crc_table[(val ^ crc) & 0xFF];
4397  * crc = val ^ (crc >> 8);
4398  *
4399  */
4400 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4401   eor(val, val, crc);
4402   andr(val, val, 0xff);
4403   ldrw(val, Address(table, val, Address::lsl(2)));
4404   eor(crc, val, crc, Assembler::LSR, 8);
4405 }
4406 
4407 /**
4408  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4409  *
4410  * @param [in,out]crc   Register containing the crc.
4411  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4412  * @param [in]table0    Register containing table 0 of crc constants.
4413  * @param [in]table1    Register containing table 1 of crc constants.
4414  * @param [in]table2    Register containing table 2 of crc constants.
4415  * @param [in]table3    Register containing table 3 of crc constants.
4416  *
4417  * uint32_t crc;
4418  *   v = crc ^ v
4419  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4420  *
4421  */
4422 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4423         Register table0, Register table1, Register table2, Register table3,
4424         bool upper) {
4425   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4426   uxtb(tmp, v);
4427   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4428   ubfx(tmp, v, 8, 8);
4429   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4430   eor(crc, crc, tmp);
4431   ubfx(tmp, v, 16, 8);
4432   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4433   eor(crc, crc, tmp);
4434   ubfx(tmp, v, 24, 8);
4435   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4436   eor(crc, crc, tmp);
4437 }
4438 
4439 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4440         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4441     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4442     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4443 
4444     subs(tmp0, len, 384);
4445     mvnw(crc, crc);
4446     br(Assembler::GE, CRC_by128_pre);
4447   BIND(CRC_less128);
4448     subs(len, len, 32);
4449     br(Assembler::GE, CRC_by32_loop);
4450   BIND(CRC_less32);
4451     adds(len, len, 32 - 4);
4452     br(Assembler::GE, CRC_by4_loop);
4453     adds(len, len, 4);
4454     br(Assembler::GT, CRC_by1_loop);
4455     b(L_exit);
4456 
4457   BIND(CRC_by32_loop);
4458     ldp(tmp0, tmp1, Address(buf));
4459     crc32x(crc, crc, tmp0);
4460     ldp(tmp2, tmp3, Address(buf, 16));
4461     crc32x(crc, crc, tmp1);
4462     add(buf, buf, 32);
4463     crc32x(crc, crc, tmp2);
4464     subs(len, len, 32);
4465     crc32x(crc, crc, tmp3);
4466     br(Assembler::GE, CRC_by32_loop);
4467     cmn(len, (u1)32);
4468     br(Assembler::NE, CRC_less32);
4469     b(L_exit);
4470 
4471   BIND(CRC_by4_loop);
4472     ldrw(tmp0, Address(post(buf, 4)));
4473     subs(len, len, 4);
4474     crc32w(crc, crc, tmp0);
4475     br(Assembler::GE, CRC_by4_loop);
4476     adds(len, len, 4);
4477     br(Assembler::LE, L_exit);
4478   BIND(CRC_by1_loop);
4479     ldrb(tmp0, Address(post(buf, 1)));
4480     subs(len, len, 1);
4481     crc32b(crc, crc, tmp0);
4482     br(Assembler::GT, CRC_by1_loop);
4483     b(L_exit);
4484 
4485   BIND(CRC_by128_pre);
4486     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4487       4*256*sizeof(juint) + 8*sizeof(juint));
4488     mov(crc, 0);
4489     crc32x(crc, crc, tmp0);
4490     crc32x(crc, crc, tmp1);
4491 
4492     cbnz(len, CRC_less128);
4493 
4494   BIND(L_exit);
4495     mvnw(crc, crc);
4496 }
4497 
4498 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4499         Register len, Register tmp0, Register tmp1, Register tmp2,
4500         Register tmp3) {
4501     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4502     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4503 
4504     mvnw(crc, crc);
4505 
4506     subs(len, len, 128);
4507     br(Assembler::GE, CRC_by64_pre);
4508   BIND(CRC_less64);
4509     adds(len, len, 128-32);
4510     br(Assembler::GE, CRC_by32_loop);
4511   BIND(CRC_less32);
4512     adds(len, len, 32-4);
4513     br(Assembler::GE, CRC_by4_loop);
4514     adds(len, len, 4);
4515     br(Assembler::GT, CRC_by1_loop);
4516     b(L_exit);
4517 
4518   BIND(CRC_by32_loop);
4519     ldp(tmp0, tmp1, Address(post(buf, 16)));
4520     subs(len, len, 32);
4521     crc32x(crc, crc, tmp0);
4522     ldr(tmp2, Address(post(buf, 8)));
4523     crc32x(crc, crc, tmp1);
4524     ldr(tmp3, Address(post(buf, 8)));
4525     crc32x(crc, crc, tmp2);
4526     crc32x(crc, crc, tmp3);
4527     br(Assembler::GE, CRC_by32_loop);
4528     cmn(len, (u1)32);
4529     br(Assembler::NE, CRC_less32);
4530     b(L_exit);
4531 
4532   BIND(CRC_by4_loop);
4533     ldrw(tmp0, Address(post(buf, 4)));
4534     subs(len, len, 4);
4535     crc32w(crc, crc, tmp0);
4536     br(Assembler::GE, CRC_by4_loop);
4537     adds(len, len, 4);
4538     br(Assembler::LE, L_exit);
4539   BIND(CRC_by1_loop);
4540     ldrb(tmp0, Address(post(buf, 1)));
4541     subs(len, len, 1);
4542     crc32b(crc, crc, tmp0);
4543     br(Assembler::GT, CRC_by1_loop);
4544     b(L_exit);
4545 
4546   BIND(CRC_by64_pre);
4547     sub(buf, buf, 8);
4548     ldp(tmp0, tmp1, Address(buf, 8));
4549     crc32x(crc, crc, tmp0);
4550     ldr(tmp2, Address(buf, 24));
4551     crc32x(crc, crc, tmp1);
4552     ldr(tmp3, Address(buf, 32));
4553     crc32x(crc, crc, tmp2);
4554     ldr(tmp0, Address(buf, 40));
4555     crc32x(crc, crc, tmp3);
4556     ldr(tmp1, Address(buf, 48));
4557     crc32x(crc, crc, tmp0);
4558     ldr(tmp2, Address(buf, 56));
4559     crc32x(crc, crc, tmp1);
4560     ldr(tmp3, Address(pre(buf, 64)));
4561 
4562     b(CRC_by64_loop);
4563 
4564     align(CodeEntryAlignment);
4565   BIND(CRC_by64_loop);
4566     subs(len, len, 64);
4567     crc32x(crc, crc, tmp2);
4568     ldr(tmp0, Address(buf, 8));
4569     crc32x(crc, crc, tmp3);
4570     ldr(tmp1, Address(buf, 16));
4571     crc32x(crc, crc, tmp0);
4572     ldr(tmp2, Address(buf, 24));
4573     crc32x(crc, crc, tmp1);
4574     ldr(tmp3, Address(buf, 32));
4575     crc32x(crc, crc, tmp2);
4576     ldr(tmp0, Address(buf, 40));
4577     crc32x(crc, crc, tmp3);
4578     ldr(tmp1, Address(buf, 48));
4579     crc32x(crc, crc, tmp0);
4580     ldr(tmp2, Address(buf, 56));
4581     crc32x(crc, crc, tmp1);
4582     ldr(tmp3, Address(pre(buf, 64)));
4583     br(Assembler::GE, CRC_by64_loop);
4584 
4585     // post-loop
4586     crc32x(crc, crc, tmp2);
4587     crc32x(crc, crc, tmp3);
4588 
4589     sub(len, len, 64);
4590     add(buf, buf, 8);
4591     cmn(len, (u1)128);
4592     br(Assembler::NE, CRC_less64);
4593   BIND(L_exit);
4594     mvnw(crc, crc);
4595 }
4596 
4597 /**
4598  * @param crc   register containing existing CRC (32-bit)
4599  * @param buf   register pointing to input byte buffer (byte*)
4600  * @param len   register containing number of bytes
4601  * @param table register that will contain address of CRC table
4602  * @param tmp   scratch register
4603  */
4604 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4605         Register table0, Register table1, Register table2, Register table3,
4606         Register tmp, Register tmp2, Register tmp3) {
4607   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4608 
4609   if (UseCryptoPmullForCRC32) {
4610       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4611       return;
4612   }
4613 
4614   if (UseCRC32) {
4615       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4616       return;
4617   }
4618 
4619     mvnw(crc, crc);
4620 
4621     {
4622       uint64_t offset;
4623       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4624       add(table0, table0, offset);
4625     }
4626     add(table1, table0, 1*256*sizeof(juint));
4627     add(table2, table0, 2*256*sizeof(juint));
4628     add(table3, table0, 3*256*sizeof(juint));
4629 
4630     { // Neon code start
4631       cmp(len, (u1)64);
4632       br(Assembler::LT, L_by16);
4633       eor(v16, T16B, v16, v16);
4634 
4635     Label L_fold;
4636 
4637       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4638 
4639       ld1(v0, v1, T2D, post(buf, 32));
4640       ld1r(v4, T2D, post(tmp, 8));
4641       ld1r(v5, T2D, post(tmp, 8));
4642       ld1r(v6, T2D, post(tmp, 8));
4643       ld1r(v7, T2D, post(tmp, 8));
4644       mov(v16, S, 0, crc);
4645 
4646       eor(v0, T16B, v0, v16);
4647       sub(len, len, 64);
4648 
4649     BIND(L_fold);
4650       pmull(v22, T8H, v0, v5, T8B);
4651       pmull(v20, T8H, v0, v7, T8B);
4652       pmull(v23, T8H, v0, v4, T8B);
4653       pmull(v21, T8H, v0, v6, T8B);
4654 
4655       pmull2(v18, T8H, v0, v5, T16B);
4656       pmull2(v16, T8H, v0, v7, T16B);
4657       pmull2(v19, T8H, v0, v4, T16B);
4658       pmull2(v17, T8H, v0, v6, T16B);
4659 
4660       uzp1(v24, T8H, v20, v22);
4661       uzp2(v25, T8H, v20, v22);
4662       eor(v20, T16B, v24, v25);
4663 
4664       uzp1(v26, T8H, v16, v18);
4665       uzp2(v27, T8H, v16, v18);
4666       eor(v16, T16B, v26, v27);
4667 
4668       ushll2(v22, T4S, v20, T8H, 8);
4669       ushll(v20, T4S, v20, T4H, 8);
4670 
4671       ushll2(v18, T4S, v16, T8H, 8);
4672       ushll(v16, T4S, v16, T4H, 8);
4673 
4674       eor(v22, T16B, v23, v22);
4675       eor(v18, T16B, v19, v18);
4676       eor(v20, T16B, v21, v20);
4677       eor(v16, T16B, v17, v16);
4678 
4679       uzp1(v17, T2D, v16, v20);
4680       uzp2(v21, T2D, v16, v20);
4681       eor(v17, T16B, v17, v21);
4682 
4683       ushll2(v20, T2D, v17, T4S, 16);
4684       ushll(v16, T2D, v17, T2S, 16);
4685 
4686       eor(v20, T16B, v20, v22);
4687       eor(v16, T16B, v16, v18);
4688 
4689       uzp1(v17, T2D, v20, v16);
4690       uzp2(v21, T2D, v20, v16);
4691       eor(v28, T16B, v17, v21);
4692 
4693       pmull(v22, T8H, v1, v5, T8B);
4694       pmull(v20, T8H, v1, v7, T8B);
4695       pmull(v23, T8H, v1, v4, T8B);
4696       pmull(v21, T8H, v1, v6, T8B);
4697 
4698       pmull2(v18, T8H, v1, v5, T16B);
4699       pmull2(v16, T8H, v1, v7, T16B);
4700       pmull2(v19, T8H, v1, v4, T16B);
4701       pmull2(v17, T8H, v1, v6, T16B);
4702 
4703       ld1(v0, v1, T2D, post(buf, 32));
4704 
4705       uzp1(v24, T8H, v20, v22);
4706       uzp2(v25, T8H, v20, v22);
4707       eor(v20, T16B, v24, v25);
4708 
4709       uzp1(v26, T8H, v16, v18);
4710       uzp2(v27, T8H, v16, v18);
4711       eor(v16, T16B, v26, v27);
4712 
4713       ushll2(v22, T4S, v20, T8H, 8);
4714       ushll(v20, T4S, v20, T4H, 8);
4715 
4716       ushll2(v18, T4S, v16, T8H, 8);
4717       ushll(v16, T4S, v16, T4H, 8);
4718 
4719       eor(v22, T16B, v23, v22);
4720       eor(v18, T16B, v19, v18);
4721       eor(v20, T16B, v21, v20);
4722       eor(v16, T16B, v17, v16);
4723 
4724       uzp1(v17, T2D, v16, v20);
4725       uzp2(v21, T2D, v16, v20);
4726       eor(v16, T16B, v17, v21);
4727 
4728       ushll2(v20, T2D, v16, T4S, 16);
4729       ushll(v16, T2D, v16, T2S, 16);
4730 
4731       eor(v20, T16B, v22, v20);
4732       eor(v16, T16B, v16, v18);
4733 
4734       uzp1(v17, T2D, v20, v16);
4735       uzp2(v21, T2D, v20, v16);
4736       eor(v20, T16B, v17, v21);
4737 
4738       shl(v16, T2D, v28, 1);
4739       shl(v17, T2D, v20, 1);
4740 
4741       eor(v0, T16B, v0, v16);
4742       eor(v1, T16B, v1, v17);
4743 
4744       subs(len, len, 32);
4745       br(Assembler::GE, L_fold);
4746 
4747       mov(crc, 0);
4748       mov(tmp, v0, D, 0);
4749       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4750       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4751       mov(tmp, v0, D, 1);
4752       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4753       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4754       mov(tmp, v1, D, 0);
4755       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4756       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4757       mov(tmp, v1, D, 1);
4758       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4759       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4760 
4761       add(len, len, 32);
4762     } // Neon code end
4763 
4764   BIND(L_by16);
4765     subs(len, len, 16);
4766     br(Assembler::GE, L_by16_loop);
4767     adds(len, len, 16-4);
4768     br(Assembler::GE, L_by4_loop);
4769     adds(len, len, 4);
4770     br(Assembler::GT, L_by1_loop);
4771     b(L_exit);
4772 
4773   BIND(L_by4_loop);
4774     ldrw(tmp, Address(post(buf, 4)));
4775     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4776     subs(len, len, 4);
4777     br(Assembler::GE, L_by4_loop);
4778     adds(len, len, 4);
4779     br(Assembler::LE, L_exit);
4780   BIND(L_by1_loop);
4781     subs(len, len, 1);
4782     ldrb(tmp, Address(post(buf, 1)));
4783     update_byte_crc32(crc, tmp, table0);
4784     br(Assembler::GT, L_by1_loop);
4785     b(L_exit);
4786 
4787     align(CodeEntryAlignment);
4788   BIND(L_by16_loop);
4789     subs(len, len, 16);
4790     ldp(tmp, tmp3, Address(post(buf, 16)));
4791     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4792     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4793     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4794     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4795     br(Assembler::GE, L_by16_loop);
4796     adds(len, len, 16-4);
4797     br(Assembler::GE, L_by4_loop);
4798     adds(len, len, 4);
4799     br(Assembler::GT, L_by1_loop);
4800   BIND(L_exit);
4801     mvnw(crc, crc);
4802 }
4803 
4804 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4805         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4806     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4807     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4808 
4809     subs(tmp0, len, 384);
4810     br(Assembler::GE, CRC_by128_pre);
4811   BIND(CRC_less128);
4812     subs(len, len, 32);
4813     br(Assembler::GE, CRC_by32_loop);
4814   BIND(CRC_less32);
4815     adds(len, len, 32 - 4);
4816     br(Assembler::GE, CRC_by4_loop);
4817     adds(len, len, 4);
4818     br(Assembler::GT, CRC_by1_loop);
4819     b(L_exit);
4820 
4821   BIND(CRC_by32_loop);
4822     ldp(tmp0, tmp1, Address(buf));
4823     crc32cx(crc, crc, tmp0);
4824     ldr(tmp2, Address(buf, 16));
4825     crc32cx(crc, crc, tmp1);
4826     ldr(tmp3, Address(buf, 24));
4827     crc32cx(crc, crc, tmp2);
4828     add(buf, buf, 32);
4829     subs(len, len, 32);
4830     crc32cx(crc, crc, tmp3);
4831     br(Assembler::GE, CRC_by32_loop);
4832     cmn(len, (u1)32);
4833     br(Assembler::NE, CRC_less32);
4834     b(L_exit);
4835 
4836   BIND(CRC_by4_loop);
4837     ldrw(tmp0, Address(post(buf, 4)));
4838     subs(len, len, 4);
4839     crc32cw(crc, crc, tmp0);
4840     br(Assembler::GE, CRC_by4_loop);
4841     adds(len, len, 4);
4842     br(Assembler::LE, L_exit);
4843   BIND(CRC_by1_loop);
4844     ldrb(tmp0, Address(post(buf, 1)));
4845     subs(len, len, 1);
4846     crc32cb(crc, crc, tmp0);
4847     br(Assembler::GT, CRC_by1_loop);
4848     b(L_exit);
4849 
4850   BIND(CRC_by128_pre);
4851     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4852       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4853     mov(crc, 0);
4854     crc32cx(crc, crc, tmp0);
4855     crc32cx(crc, crc, tmp1);
4856 
4857     cbnz(len, CRC_less128);
4858 
4859   BIND(L_exit);
4860 }
4861 
4862 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4863         Register len, Register tmp0, Register tmp1, Register tmp2,
4864         Register tmp3) {
4865     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4866     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4867 
4868     subs(len, len, 128);
4869     br(Assembler::GE, CRC_by64_pre);
4870   BIND(CRC_less64);
4871     adds(len, len, 128-32);
4872     br(Assembler::GE, CRC_by32_loop);
4873   BIND(CRC_less32);
4874     adds(len, len, 32-4);
4875     br(Assembler::GE, CRC_by4_loop);
4876     adds(len, len, 4);
4877     br(Assembler::GT, CRC_by1_loop);
4878     b(L_exit);
4879 
4880   BIND(CRC_by32_loop);
4881     ldp(tmp0, tmp1, Address(post(buf, 16)));
4882     subs(len, len, 32);
4883     crc32cx(crc, crc, tmp0);
4884     ldr(tmp2, Address(post(buf, 8)));
4885     crc32cx(crc, crc, tmp1);
4886     ldr(tmp3, Address(post(buf, 8)));
4887     crc32cx(crc, crc, tmp2);
4888     crc32cx(crc, crc, tmp3);
4889     br(Assembler::GE, CRC_by32_loop);
4890     cmn(len, (u1)32);
4891     br(Assembler::NE, CRC_less32);
4892     b(L_exit);
4893 
4894   BIND(CRC_by4_loop);
4895     ldrw(tmp0, Address(post(buf, 4)));
4896     subs(len, len, 4);
4897     crc32cw(crc, crc, tmp0);
4898     br(Assembler::GE, CRC_by4_loop);
4899     adds(len, len, 4);
4900     br(Assembler::LE, L_exit);
4901   BIND(CRC_by1_loop);
4902     ldrb(tmp0, Address(post(buf, 1)));
4903     subs(len, len, 1);
4904     crc32cb(crc, crc, tmp0);
4905     br(Assembler::GT, CRC_by1_loop);
4906     b(L_exit);
4907 
4908   BIND(CRC_by64_pre);
4909     sub(buf, buf, 8);
4910     ldp(tmp0, tmp1, Address(buf, 8));
4911     crc32cx(crc, crc, tmp0);
4912     ldr(tmp2, Address(buf, 24));
4913     crc32cx(crc, crc, tmp1);
4914     ldr(tmp3, Address(buf, 32));
4915     crc32cx(crc, crc, tmp2);
4916     ldr(tmp0, Address(buf, 40));
4917     crc32cx(crc, crc, tmp3);
4918     ldr(tmp1, Address(buf, 48));
4919     crc32cx(crc, crc, tmp0);
4920     ldr(tmp2, Address(buf, 56));
4921     crc32cx(crc, crc, tmp1);
4922     ldr(tmp3, Address(pre(buf, 64)));
4923 
4924     b(CRC_by64_loop);
4925 
4926     align(CodeEntryAlignment);
4927   BIND(CRC_by64_loop);
4928     subs(len, len, 64);
4929     crc32cx(crc, crc, tmp2);
4930     ldr(tmp0, Address(buf, 8));
4931     crc32cx(crc, crc, tmp3);
4932     ldr(tmp1, Address(buf, 16));
4933     crc32cx(crc, crc, tmp0);
4934     ldr(tmp2, Address(buf, 24));
4935     crc32cx(crc, crc, tmp1);
4936     ldr(tmp3, Address(buf, 32));
4937     crc32cx(crc, crc, tmp2);
4938     ldr(tmp0, Address(buf, 40));
4939     crc32cx(crc, crc, tmp3);
4940     ldr(tmp1, Address(buf, 48));
4941     crc32cx(crc, crc, tmp0);
4942     ldr(tmp2, Address(buf, 56));
4943     crc32cx(crc, crc, tmp1);
4944     ldr(tmp3, Address(pre(buf, 64)));
4945     br(Assembler::GE, CRC_by64_loop);
4946 
4947     // post-loop
4948     crc32cx(crc, crc, tmp2);
4949     crc32cx(crc, crc, tmp3);
4950 
4951     sub(len, len, 64);
4952     add(buf, buf, 8);
4953     cmn(len, (u1)128);
4954     br(Assembler::NE, CRC_less64);
4955   BIND(L_exit);
4956 }
4957 
4958 /**
4959  * @param crc   register containing existing CRC (32-bit)
4960  * @param buf   register pointing to input byte buffer (byte*)
4961  * @param len   register containing number of bytes
4962  * @param table register that will contain address of CRC table
4963  * @param tmp   scratch register
4964  */
4965 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4966         Register table0, Register table1, Register table2, Register table3,
4967         Register tmp, Register tmp2, Register tmp3) {
4968   if (UseCryptoPmullForCRC32) {
4969     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4970   } else {
4971     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4972   }
4973 }
4974 
4975 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4976         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4977     Label CRC_by128_loop;
4978     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4979 
4980     sub(len, len, 256);
4981     Register table = tmp0;
4982     {
4983       uint64_t offset;
4984       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4985       add(table, table, offset);
4986     }
4987     add(table, table, table_offset);
4988 
4989     // Registers v0..v7 are used as data registers.
4990     // Registers v16..v31 are used as tmp registers.
4991     sub(buf, buf, 0x10);
4992     ldrq(v0, Address(buf, 0x10));
4993     ldrq(v1, Address(buf, 0x20));
4994     ldrq(v2, Address(buf, 0x30));
4995     ldrq(v3, Address(buf, 0x40));
4996     ldrq(v4, Address(buf, 0x50));
4997     ldrq(v5, Address(buf, 0x60));
4998     ldrq(v6, Address(buf, 0x70));
4999     ldrq(v7, Address(pre(buf, 0x80)));
5000 
5001     movi(v31, T4S, 0);
5002     mov(v31, S, 0, crc);
5003     eor(v0, T16B, v0, v31);
5004 
5005     // Register v16 contains constants from the crc table.
5006     ldrq(v16, Address(table));
5007     b(CRC_by128_loop);
5008 
5009     align(OptoLoopAlignment);
5010   BIND(CRC_by128_loop);
5011     pmull (v17,  T1Q, v0, v16, T1D);
5012     pmull2(v18, T1Q, v0, v16, T2D);
5013     ldrq(v0, Address(buf, 0x10));
5014     eor3(v0, T16B, v17,  v18, v0);
5015 
5016     pmull (v19, T1Q, v1, v16, T1D);
5017     pmull2(v20, T1Q, v1, v16, T2D);
5018     ldrq(v1, Address(buf, 0x20));
5019     eor3(v1, T16B, v19, v20, v1);
5020 
5021     pmull (v21, T1Q, v2, v16, T1D);
5022     pmull2(v22, T1Q, v2, v16, T2D);
5023     ldrq(v2, Address(buf, 0x30));
5024     eor3(v2, T16B, v21, v22, v2);
5025 
5026     pmull (v23, T1Q, v3, v16, T1D);
5027     pmull2(v24, T1Q, v3, v16, T2D);
5028     ldrq(v3, Address(buf, 0x40));
5029     eor3(v3, T16B, v23, v24, v3);
5030 
5031     pmull (v25, T1Q, v4, v16, T1D);
5032     pmull2(v26, T1Q, v4, v16, T2D);
5033     ldrq(v4, Address(buf, 0x50));
5034     eor3(v4, T16B, v25, v26, v4);
5035 
5036     pmull (v27, T1Q, v5, v16, T1D);
5037     pmull2(v28, T1Q, v5, v16, T2D);
5038     ldrq(v5, Address(buf, 0x60));
5039     eor3(v5, T16B, v27, v28, v5);
5040 
5041     pmull (v29, T1Q, v6, v16, T1D);
5042     pmull2(v30, T1Q, v6, v16, T2D);
5043     ldrq(v6, Address(buf, 0x70));
5044     eor3(v6, T16B, v29, v30, v6);
5045 
5046     // Reuse registers v23, v24.
5047     // Using them won't block the first instruction of the next iteration.
5048     pmull (v23, T1Q, v7, v16, T1D);
5049     pmull2(v24, T1Q, v7, v16, T2D);
5050     ldrq(v7, Address(pre(buf, 0x80)));
5051     eor3(v7, T16B, v23, v24, v7);
5052 
5053     subs(len, len, 0x80);
5054     br(Assembler::GE, CRC_by128_loop);
5055 
5056     // fold into 512 bits
5057     // Use v31 for constants because v16 can be still in use.
5058     ldrq(v31, Address(table, 0x10));
5059 
5060     pmull (v17,  T1Q, v0, v31, T1D);
5061     pmull2(v18, T1Q, v0, v31, T2D);
5062     eor3(v0, T16B, v17, v18, v4);
5063 
5064     pmull (v19, T1Q, v1, v31, T1D);
5065     pmull2(v20, T1Q, v1, v31, T2D);
5066     eor3(v1, T16B, v19, v20, v5);
5067 
5068     pmull (v21, T1Q, v2, v31, T1D);
5069     pmull2(v22, T1Q, v2, v31, T2D);
5070     eor3(v2, T16B, v21, v22, v6);
5071 
5072     pmull (v23, T1Q, v3, v31, T1D);
5073     pmull2(v24, T1Q, v3, v31, T2D);
5074     eor3(v3, T16B, v23, v24, v7);
5075 
5076     // fold into 128 bits
5077     // Use v17 for constants because v31 can be still in use.
5078     ldrq(v17, Address(table, 0x20));
5079     pmull (v25, T1Q, v0, v17, T1D);
5080     pmull2(v26, T1Q, v0, v17, T2D);
5081     eor3(v3, T16B, v3, v25, v26);
5082 
5083     // Use v18 for constants because v17 can be still in use.
5084     ldrq(v18, Address(table, 0x30));
5085     pmull (v27, T1Q, v1, v18, T1D);
5086     pmull2(v28, T1Q, v1, v18, T2D);
5087     eor3(v3, T16B, v3, v27, v28);
5088 
5089     // Use v19 for constants because v18 can be still in use.
5090     ldrq(v19, Address(table, 0x40));
5091     pmull (v29, T1Q, v2, v19, T1D);
5092     pmull2(v30, T1Q, v2, v19, T2D);
5093     eor3(v0, T16B, v3, v29, v30);
5094 
5095     add(len, len, 0x80);
5096     add(buf, buf, 0x10);
5097 
5098     mov(tmp0, v0, D, 0);
5099     mov(tmp1, v0, D, 1);
5100 }
5101 
5102 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5103   Address adr;
5104   switch(dst.getMode()) {
5105   case Address::base_plus_offset:
5106     // This is the expected mode, although we allow all the other
5107     // forms below.
5108     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5109     break;
5110   default:
5111     lea(rscratch2, dst);
5112     adr = Address(rscratch2);
5113     break;
5114   }
5115   ldr(rscratch1, adr);
5116   add(rscratch1, rscratch1, src);
5117   str(rscratch1, adr);
5118 }
5119 
5120 void MacroAssembler::cmpptr(Register src1, Address src2) {
5121   uint64_t offset;
5122   adrp(rscratch1, src2, offset);
5123   ldr(rscratch1, Address(rscratch1, offset));
5124   cmp(src1, rscratch1);
5125 }
5126 
5127 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5128   cmp(obj1, obj2);
5129 }
5130 
5131 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5132   load_method_holder(rresult, rmethod);
5133   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5134 }
5135 
5136 void MacroAssembler::load_method_holder(Register holder, Register method) {
5137   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5138   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5139   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5140 }
5141 
5142 void MacroAssembler::load_metadata(Register dst, Register src) {
5143   if (UseCompactObjectHeaders) {
5144     load_narrow_klass_compact(dst, src);
5145   } else if (UseCompressedClassPointers) {
5146     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5147   } else {
5148     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5149   }
5150 }
5151 
5152 // Loads the obj's Klass* into dst.
5153 // Preserves all registers (incl src, rscratch1 and rscratch2).
5154 // Input:
5155 // src - the oop we want to load the klass from.
5156 // dst - output narrow klass.
5157 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5158   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5159   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5160   lsr(dst, dst, markWord::klass_shift);
5161 }
5162 
5163 void MacroAssembler::load_klass(Register dst, Register src) {
5164   if (UseCompactObjectHeaders) {
5165     load_narrow_klass_compact(dst, src);
5166     decode_klass_not_null(dst);
5167   } else if (UseCompressedClassPointers) {
5168     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5169     decode_klass_not_null(dst);
5170   } else {
5171     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5172   }
5173 }
5174 
5175 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5176   if (RestoreMXCSROnJNICalls) {
5177     Label OK;
5178     get_fpcr(tmp1);
5179     mov(tmp2, tmp1);
5180     // Set FPCR to the state we need. We do want Round to Nearest. We
5181     // don't want non-IEEE rounding modes or floating-point traps.
5182     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5183     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5184     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5185     eor(tmp2, tmp1, tmp2);
5186     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5187     set_fpcr(tmp1);
5188     bind(OK);
5189   }
5190 }
5191 
5192 // ((OopHandle)result).resolve();
5193 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5194   // OopHandle::resolve is an indirection.
5195   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5196 }
5197 
5198 // ((WeakHandle)result).resolve();
5199 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5200   assert_different_registers(result, tmp1, tmp2);
5201   Label resolved;
5202 
5203   // A null weak handle resolves to null.
5204   cbz(result, resolved);
5205 
5206   // Only 64 bit platforms support GCs that require a tmp register
5207   // WeakHandle::resolve is an indirection like jweak.
5208   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5209                  result, Address(result), tmp1, tmp2);
5210   bind(resolved);
5211 }
5212 
5213 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5214   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5215   ldr(dst, Address(rmethod, Method::const_offset()));
5216   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5217   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5218   ldr(dst, Address(dst, mirror_offset));
5219   resolve_oop_handle(dst, tmp1, tmp2);
5220 }
5221 
5222 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5223   assert_different_registers(obj, klass, tmp);
5224   if (UseCompressedClassPointers) {
5225     if (UseCompactObjectHeaders) {
5226       load_narrow_klass_compact(tmp, obj);
5227     } else {
5228       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5229     }
5230     if (CompressedKlassPointers::base() == nullptr) {
5231       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5232       return;
5233     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5234                && CompressedKlassPointers::shift() == 0) {
5235       // Only the bottom 32 bits matter
5236       cmpw(klass, tmp);
5237       return;
5238     }
5239     decode_klass_not_null(tmp);
5240   } else {
5241     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5242   }
5243   cmp(klass, tmp);
5244 }
5245 
5246 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5247   if (UseCompactObjectHeaders) {
5248     load_narrow_klass_compact(tmp1, obj1);
5249     load_narrow_klass_compact(tmp2,  obj2);
5250     cmpw(tmp1, tmp2);
5251   } else if (UseCompressedClassPointers) {
5252     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5253     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5254     cmpw(tmp1, tmp2);
5255   } else {
5256     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5257     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5258     cmp(tmp1, tmp2);
5259   }
5260 }
5261 
5262 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5263   load_klass(dst, src);
5264   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5265 }
5266 
5267 void MacroAssembler::store_klass(Register dst, Register src) {
5268   // FIXME: Should this be a store release?  concurrent gcs assumes
5269   // klass length is valid if klass field is not null.
5270   assert(!UseCompactObjectHeaders, "not with compact headers");
5271   if (UseCompressedClassPointers) {
5272     encode_klass_not_null(src);
5273     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5274   } else {
5275     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5276   }
5277 }
5278 
5279 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5280   assert(!UseCompactObjectHeaders, "not with compact headers");
5281   if (UseCompressedClassPointers) {
5282     // Store to klass gap in destination
5283     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5284   }
5285 }
5286 
5287 // Algorithm must match CompressedOops::encode.
5288 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5289 #ifdef ASSERT
5290   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5291 #endif
5292   verify_oop_msg(s, "broken oop in encode_heap_oop");
5293   if (CompressedOops::base() == nullptr) {
5294     if (CompressedOops::shift() != 0) {
5295       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5296       lsr(d, s, LogMinObjAlignmentInBytes);
5297     } else {
5298       mov(d, s);
5299     }
5300   } else {
5301     subs(d, s, rheapbase);
5302     csel(d, d, zr, Assembler::HS);
5303     lsr(d, d, LogMinObjAlignmentInBytes);
5304 
5305     /*  Old algorithm: is this any worse?
5306     Label nonnull;
5307     cbnz(r, nonnull);
5308     sub(r, r, rheapbase);
5309     bind(nonnull);
5310     lsr(r, r, LogMinObjAlignmentInBytes);
5311     */
5312   }
5313 }
5314 
5315 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5316 #ifdef ASSERT
5317   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5318   if (CheckCompressedOops) {
5319     Label ok;
5320     cbnz(r, ok);
5321     stop("null oop passed to encode_heap_oop_not_null");
5322     bind(ok);
5323   }
5324 #endif
5325   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5326   if (CompressedOops::base() != nullptr) {
5327     sub(r, r, rheapbase);
5328   }
5329   if (CompressedOops::shift() != 0) {
5330     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5331     lsr(r, r, LogMinObjAlignmentInBytes);
5332   }
5333 }
5334 
5335 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5336 #ifdef ASSERT
5337   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5338   if (CheckCompressedOops) {
5339     Label ok;
5340     cbnz(src, ok);
5341     stop("null oop passed to encode_heap_oop_not_null2");
5342     bind(ok);
5343   }
5344 #endif
5345   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5346 
5347   Register data = src;
5348   if (CompressedOops::base() != nullptr) {
5349     sub(dst, src, rheapbase);
5350     data = dst;
5351   }
5352   if (CompressedOops::shift() != 0) {
5353     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5354     lsr(dst, data, LogMinObjAlignmentInBytes);
5355     data = dst;
5356   }
5357   if (data == src)
5358     mov(dst, src);
5359 }
5360 
5361 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5362 #ifdef ASSERT
5363   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5364 #endif
5365   if (CompressedOops::base() == nullptr) {
5366     if (CompressedOops::shift() != 0) {
5367       lsl(d, s, CompressedOops::shift());
5368     } else if (d != s) {
5369       mov(d, s);
5370     }
5371   } else {
5372     Label done;
5373     if (d != s)
5374       mov(d, s);
5375     cbz(s, done);
5376     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5377     bind(done);
5378   }
5379   verify_oop_msg(d, "broken oop in decode_heap_oop");
5380 }
5381 
5382 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5383   assert (UseCompressedOops, "should only be used for compressed headers");
5384   assert (Universe::heap() != nullptr, "java heap should be initialized");
5385   // Cannot assert, unverified entry point counts instructions (see .ad file)
5386   // vtableStubs also counts instructions in pd_code_size_limit.
5387   // Also do not verify_oop as this is called by verify_oop.
5388   if (CompressedOops::shift() != 0) {
5389     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5390     if (CompressedOops::base() != nullptr) {
5391       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5392     } else {
5393       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5394     }
5395   } else {
5396     assert (CompressedOops::base() == nullptr, "sanity");
5397   }
5398 }
5399 
5400 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5401   assert (UseCompressedOops, "should only be used for compressed headers");
5402   assert (Universe::heap() != nullptr, "java heap should be initialized");
5403   // Cannot assert, unverified entry point counts instructions (see .ad file)
5404   // vtableStubs also counts instructions in pd_code_size_limit.
5405   // Also do not verify_oop as this is called by verify_oop.
5406   if (CompressedOops::shift() != 0) {
5407     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5408     if (CompressedOops::base() != nullptr) {
5409       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5410     } else {
5411       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5412     }
5413   } else {
5414     assert (CompressedOops::base() == nullptr, "sanity");
5415     if (dst != src) {
5416       mov(dst, src);
5417     }
5418   }
5419 }
5420 
5421 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5422 
5423 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5424   assert(Metaspace::initialized(), "metaspace not initialized yet");
5425   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5426   return _klass_decode_mode;
5427 }
5428 
5429 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5430   assert(UseCompressedClassPointers, "not using compressed class pointers");
5431 
5432   // KlassDecodeMode shouldn't be set already.
5433   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5434 
5435   if (base == nullptr) {
5436     return KlassDecodeZero;
5437   }
5438 
5439   if (operand_valid_for_logical_immediate(
5440         /*is32*/false, (uint64_t)base)) {
5441     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5442     if (((uint64_t)base & range_mask) == 0) {
5443       return KlassDecodeXor;
5444     }
5445   }
5446 
5447   const uint64_t shifted_base =
5448     (uint64_t)base >> shift;
5449   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5450     return KlassDecodeMovk;
5451   }
5452 
5453   // No valid encoding.
5454   return KlassDecodeNone;
5455 }
5456 
5457 // Check if one of the above decoding modes will work for given base, shift and range.
5458 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5459   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5460 }
5461 
5462 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5463   _klass_decode_mode = klass_decode_mode(base, shift, range);
5464   return _klass_decode_mode != KlassDecodeNone;
5465 }
5466 
5467 static Register pick_different_tmp(Register dst, Register src) {
5468   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5469   return *tmps.begin();
5470 }
5471 
5472 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5473   // we have to load the klass base from the AOT constants area but
5474   // not the shift because it is not allowed to change
5475   int shift = CompressedKlassPointers::shift();
5476   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5477   if (dst != src) {
5478     // we can load the base into dst, subtract it formthe src and shift down
5479     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5480     ldr(dst, dst);
5481     sub(dst, src, dst);
5482     lsr(dst, dst, shift);
5483   } else {
5484     // we need an extra register in order to load the coop base
5485     Register tmp = pick_different_tmp(dst, src);
5486     RegSet regs = RegSet::of(tmp);
5487     push(regs, sp);
5488     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5489     ldr(tmp, tmp);
5490     sub(dst, src, tmp);
5491     lsr(dst, dst, shift);
5492     pop(regs, sp);
5493   }
5494 }
5495 
5496 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5497   if (AOTCodeCache::is_on_for_dump()) {
5498     encode_klass_not_null_for_aot(dst, src);
5499     return;
5500   }
5501 
5502   switch (klass_decode_mode()) {
5503   case KlassDecodeZero:
5504     if (CompressedKlassPointers::shift() != 0) {
5505       lsr(dst, src, CompressedKlassPointers::shift());
5506     } else {
5507       if (dst != src) mov(dst, src);
5508     }
5509     break;
5510 
5511   case KlassDecodeXor:
5512     if (CompressedKlassPointers::shift() != 0) {
5513       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5514       lsr(dst, dst, CompressedKlassPointers::shift());
5515     } else {
5516       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5517     }
5518     break;
5519 
5520   case KlassDecodeMovk:
5521     if (CompressedKlassPointers::shift() != 0) {
5522       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5523     } else {
5524       movw(dst, src);
5525     }
5526     break;
5527 
5528   case KlassDecodeNone:
5529     ShouldNotReachHere();
5530     break;
5531   }
5532 }
5533 
5534 void MacroAssembler::encode_klass_not_null(Register r) {
5535   encode_klass_not_null(r, r);
5536 }
5537 
5538 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5539   // we have to load the klass base from the AOT constants area but
5540   // not the shift because it is not allowed to change
5541   int shift = CompressedKlassPointers::shift();
5542   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5543   if (dst != src) {
5544     // we can load the base into dst then add the offset with a suitable shift
5545     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5546     ldr(dst, dst);
5547     add(dst, dst, src, LSL,  shift);
5548   } else {
5549     // we need an extra register in order to load the coop base
5550     Register tmp = pick_different_tmp(dst, src);
5551     RegSet regs = RegSet::of(tmp);
5552     push(regs, sp);
5553     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5554     ldr(tmp, tmp);
5555     add(dst, tmp,  src, LSL,  shift);
5556     pop(regs, sp);
5557   }
5558 }
5559 
5560 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5561   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5562 
5563   if (AOTCodeCache::is_on_for_dump()) {
5564     decode_klass_not_null_for_aot(dst, src);
5565     return;
5566   }
5567 
5568   switch (klass_decode_mode()) {
5569   case KlassDecodeZero:
5570     if (CompressedKlassPointers::shift() != 0) {
5571       lsl(dst, src, CompressedKlassPointers::shift());
5572     } else {
5573       if (dst != src) mov(dst, src);
5574     }
5575     break;
5576 
5577   case KlassDecodeXor:
5578     if (CompressedKlassPointers::shift() != 0) {
5579       lsl(dst, src, CompressedKlassPointers::shift());
5580       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5581     } else {
5582       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5583     }
5584     break;
5585 
5586   case KlassDecodeMovk: {
5587     const uint64_t shifted_base =
5588       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5589 
5590     if (dst != src) movw(dst, src);
5591     movk(dst, shifted_base >> 32, 32);
5592 
5593     if (CompressedKlassPointers::shift() != 0) {
5594       lsl(dst, dst, CompressedKlassPointers::shift());
5595     }
5596 
5597     break;
5598   }
5599 
5600   case KlassDecodeNone:
5601     ShouldNotReachHere();
5602     break;
5603   }
5604 }
5605 
5606 void  MacroAssembler::decode_klass_not_null(Register r) {
5607   decode_klass_not_null(r, r);
5608 }
5609 
5610 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5611 #ifdef ASSERT
5612   {
5613     ThreadInVMfromUnknown tiv;
5614     assert (UseCompressedOops, "should only be used for compressed oops");
5615     assert (Universe::heap() != nullptr, "java heap should be initialized");
5616     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5617     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5618   }
5619 #endif
5620   int oop_index = oop_recorder()->find_index(obj);
5621   InstructionMark im(this);
5622   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5623   code_section()->relocate(inst_mark(), rspec);
5624   movz(dst, 0xDEAD, 16);
5625   movk(dst, 0xBEEF);
5626 }
5627 
5628 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5629   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5630   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5631   int index = oop_recorder()->find_index(k);
5632   assert(! Universe::heap()->is_in(k), "should not be an oop");
5633 
5634   InstructionMark im(this);
5635   RelocationHolder rspec = metadata_Relocation::spec(index);
5636   code_section()->relocate(inst_mark(), rspec);
5637   narrowKlass nk = CompressedKlassPointers::encode(k);
5638   movz(dst, (nk >> 16), 16);
5639   movk(dst, nk & 0xffff);
5640 }
5641 
5642 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5643                                     Register dst, Address src,
5644                                     Register tmp1, Register tmp2) {
5645   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5646   decorators = AccessInternal::decorator_fixup(decorators, type);
5647   bool as_raw = (decorators & AS_RAW) != 0;
5648   if (as_raw) {
5649     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5650   } else {
5651     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5652   }
5653 }
5654 
5655 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5656                                      Address dst, Register val,
5657                                      Register tmp1, Register tmp2, Register tmp3) {
5658   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5659   decorators = AccessInternal::decorator_fixup(decorators, type);
5660   bool as_raw = (decorators & AS_RAW) != 0;
5661   if (as_raw) {
5662     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5663   } else {
5664     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5665   }
5666 }
5667 
5668 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5669                                      Register inline_layout_info) {
5670   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5671   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5672 }
5673 
5674 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5675   ldr(offset, Address(inline_klass, InstanceKlass::adr_inlineklass_fixed_block_offset()));
5676   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5677 }
5678 
5679 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5680   // ((address) (void*) o) + vk->payload_offset();
5681   Register offset = (data == oop) ? rscratch1 : data;
5682   payload_offset(inline_klass, offset);
5683   if (data == oop) {
5684     add(data, data, offset);
5685   } else {
5686     lea(data, Address(oop, offset));
5687   }
5688 }
5689 
5690 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
5691                                                 Register index, Register data) {
5692   assert_different_registers(array, array_klass, index);
5693   assert_different_registers(rscratch1, array, index);
5694 
5695   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
5696   ldrw(rscratch1, Address(array_klass, Klass::layout_helper_offset()));
5697 
5698   // Klass::layout_helper_log2_element_size(lh)
5699   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
5700   lsr(rscratch1, rscratch1, Klass::_lh_log2_element_size_shift);
5701   andr(rscratch1, rscratch1, Klass::_lh_log2_element_size_mask);
5702   lslv(index, index, rscratch1);
5703 
5704   add(data, array, index);
5705   add(data, data, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT));
5706 }
5707 
5708 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5709                                    Register tmp2, DecoratorSet decorators) {
5710   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5711 }
5712 
5713 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5714                                             Register tmp2, DecoratorSet decorators) {
5715   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5716 }
5717 
5718 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5719                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5720   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5721 }
5722 
5723 // Used for storing nulls.
5724 void MacroAssembler::store_heap_oop_null(Address dst) {
5725   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5726 }
5727 
5728 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5729   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5730   int index = oop_recorder()->allocate_metadata_index(obj);
5731   RelocationHolder rspec = metadata_Relocation::spec(index);
5732   return Address((address)obj, rspec);
5733 }
5734 
5735 // Move an oop into a register.
5736 void MacroAssembler::movoop(Register dst, jobject obj) {
5737   int oop_index;
5738   if (obj == nullptr) {
5739     oop_index = oop_recorder()->allocate_oop_index(obj);
5740   } else {
5741 #ifdef ASSERT
5742     {
5743       ThreadInVMfromUnknown tiv;
5744       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5745     }
5746 #endif
5747     oop_index = oop_recorder()->find_index(obj);
5748   }
5749   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5750 
5751   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5752     mov(dst, Address((address)obj, rspec));
5753   } else {
5754     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5755     ldr(dst, Address(dummy, rspec));
5756   }
5757 }
5758 
5759 // Move a metadata address into a register.
5760 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5761   int oop_index;
5762   if (obj == nullptr) {
5763     oop_index = oop_recorder()->allocate_metadata_index(obj);
5764   } else {
5765     oop_index = oop_recorder()->find_index(obj);
5766   }
5767   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5768   mov(dst, Address((address)obj, rspec));
5769 }
5770 
5771 Address MacroAssembler::constant_oop_address(jobject obj) {
5772 #ifdef ASSERT
5773   {
5774     ThreadInVMfromUnknown tiv;
5775     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5776     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5777   }
5778 #endif
5779   int oop_index = oop_recorder()->find_index(obj);
5780   return Address((address)obj, oop_Relocation::spec(oop_index));
5781 }
5782 
5783 // Object / value buffer allocation...
5784 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
5785                                        Register t1, Register t2,
5786                                        bool clear_fields, Label& alloc_failed)
5787 {
5788   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
5789   Register layout_size = t1;
5790   assert(new_obj == r0, "needs to be r0");
5791   assert_different_registers(klass, new_obj, t1, t2);
5792 
5793   // get instance_size in InstanceKlass (scaled to a count of bytes)
5794   ldrw(layout_size, Address(klass, Klass::layout_helper_offset()));
5795   // test to see if it is malformed in some way
5796   tst(layout_size, Klass::_lh_instance_slow_path_bit);
5797   br(Assembler::NE, slow_case_no_pop);
5798 
5799   // Allocate the instance:
5800   //  If TLAB is enabled:
5801   //    Try to allocate in the TLAB.
5802   //    If fails, go to the slow path.
5803   //    Initialize the allocation.
5804   //    Exit.
5805   //
5806   //  Go to slow path.
5807 
5808   if (UseTLAB) {
5809     push(klass);
5810     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
5811     if (ZeroTLAB || (!clear_fields)) {
5812       // the fields have been already cleared
5813       b(initialize_header);
5814     } else {
5815       // initialize both the header and fields
5816       b(initialize_object);
5817     }
5818 
5819     if (clear_fields) {
5820       // The object is initialized before the header.  If the object size is
5821       // zero, go directly to the header initialization.
5822       bind(initialize_object);
5823       int header_size = oopDesc::header_size() * HeapWordSize;
5824       assert(is_aligned(header_size, BytesPerLong), "oop header size must be 8-byte-aligned");
5825       subs(layout_size, layout_size, header_size);
5826       br(Assembler::EQ, initialize_header);
5827 
5828       // Initialize topmost object field, divide size by 8, check if odd and
5829       // test if zero.
5830 
5831   #ifdef ASSERT
5832       // make sure instance_size was multiple of 8
5833       Label L;
5834       tst(layout_size, 7);
5835       br(Assembler::EQ, L);
5836       stop("object size is not multiple of 8 - adjust this code");
5837       bind(L);
5838       // must be > 0, no extra check needed here
5839   #endif
5840 
5841       lsr(layout_size, layout_size, LogBytesPerLong);
5842 
5843       // initialize remaining object fields: instance_size was a multiple of 8
5844       {
5845         Label loop;
5846         Register base = t2;
5847 
5848         bind(loop);
5849         add(rscratch1, new_obj, layout_size, Assembler::LSL, LogBytesPerLong);
5850         str(zr, Address(rscratch1, header_size - 1*oopSize));
5851         subs(layout_size, layout_size, 1);
5852         br(Assembler::NE, loop);
5853       }
5854     } // clear_fields
5855 
5856     // initialize object header only.
5857     bind(initialize_header);
5858     pop(klass);
5859     Register mark_word = t2;
5860     if (UseCompactObjectHeaders || EnableValhalla) {
5861       ldr(mark_word, Address(klass, Klass::prototype_header_offset()));
5862       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5863     } else {
5864       mov(mark_word, (intptr_t)markWord::prototype().value());
5865       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5866     }
5867     if (!UseCompactObjectHeaders) {
5868       store_klass_gap(new_obj, zr);  // zero klass gap for compressed oops
5869       mov(t2, klass);                // preserve klass
5870       store_klass(new_obj, t2);      // src klass reg is potentially compressed
5871     }
5872     b(done);
5873   }
5874 
5875   if (UseTLAB) {
5876     bind(slow_case);
5877     pop(klass);
5878   }
5879   bind(slow_case_no_pop);
5880   b(alloc_failed);
5881 
5882   bind(done);
5883 }
5884 
5885 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5886 void MacroAssembler::tlab_allocate(Register obj,
5887                                    Register var_size_in_bytes,
5888                                    int con_size_in_bytes,
5889                                    Register t1,
5890                                    Register t2,
5891                                    Label& slow_case) {
5892   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5893   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5894 }
5895 
5896 void MacroAssembler::verify_tlab() {
5897 #ifdef ASSERT
5898   if (UseTLAB && VerifyOops) {
5899     Label next, ok;
5900 
5901     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5902 
5903     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5904     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5905     cmp(rscratch2, rscratch1);
5906     br(Assembler::HS, next);
5907     STOP("assert(top >= start)");
5908     should_not_reach_here();
5909 
5910     bind(next);
5911     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5912     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5913     cmp(rscratch2, rscratch1);
5914     br(Assembler::HS, ok);
5915     STOP("assert(top <= end)");
5916     should_not_reach_here();
5917 
5918     bind(ok);
5919     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5920   }
5921 #endif
5922 }
5923 
5924 void MacroAssembler::get_inline_type_field_klass(Register holder_klass, Register index, Register inline_klass) {
5925   inline_layout_info(holder_klass, index, inline_klass);
5926   ldr(inline_klass, Address(inline_klass, InlineLayoutInfo::klass_offset()));
5927 }
5928 
5929 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5930   assert_different_registers(holder_klass, index, layout_info);
5931   InlineLayoutInfo array[2];
5932   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5933   if (is_power_of_2(size)) {
5934     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5935   } else {
5936     mov(layout_info, size);
5937     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5938   }
5939   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5940   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5941   lea(layout_info, Address(layout_info, index));
5942 }
5943 
5944 // Writes to stack successive pages until offset reached to check for
5945 // stack overflow + shadow pages.  This clobbers tmp.
5946 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5947   assert_different_registers(tmp, size, rscratch1);
5948   mov(tmp, sp);
5949   // Bang stack for total size given plus shadow page size.
5950   // Bang one page at a time because large size can bang beyond yellow and
5951   // red zones.
5952   Label loop;
5953   mov(rscratch1, (int)os::vm_page_size());
5954   bind(loop);
5955   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5956   subsw(size, size, rscratch1);
5957   str(size, Address(tmp));
5958   br(Assembler::GT, loop);
5959 
5960   // Bang down shadow pages too.
5961   // At this point, (tmp-0) is the last address touched, so don't
5962   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5963   // was post-decremented.)  Skip this address by starting at i=1, and
5964   // touch a few more pages below.  N.B.  It is important to touch all
5965   // the way down to and including i=StackShadowPages.
5966   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5967     // this could be any sized move but this is can be a debugging crumb
5968     // so the bigger the better.
5969     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5970     str(size, Address(tmp));
5971   }
5972 }
5973 
5974 // Move the address of the polling page into dest.
5975 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5976   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5977 }
5978 
5979 // Read the polling page.  The address of the polling page must
5980 // already be in r.
5981 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5982   address mark;
5983   {
5984     InstructionMark im(this);
5985     code_section()->relocate(inst_mark(), rtype);
5986     ldrw(zr, Address(r, 0));
5987     mark = inst_mark();
5988   }
5989   verify_cross_modify_fence_not_required();
5990   return mark;
5991 }
5992 
5993 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5994   relocInfo::relocType rtype = dest.rspec().reloc()->type();
5995   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5996   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5997   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5998   int64_t offset_low = dest_page - low_page;
5999   int64_t offset_high = dest_page - high_page;
6000 
6001   assert(is_valid_AArch64_address(dest.target()), "bad address");
6002   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
6003 
6004   InstructionMark im(this);
6005   code_section()->relocate(inst_mark(), dest.rspec());
6006   // 8143067: Ensure that the adrp can reach the dest from anywhere within
6007   // the code cache so that if it is relocated we know it will still reach
6008   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
6009     _adrp(reg1, dest.target());
6010   } else {
6011     uint64_t target = (uint64_t)dest.target();
6012     uint64_t adrp_target
6013       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
6014 
6015     _adrp(reg1, (address)adrp_target);
6016     movk(reg1, target >> 32, 32);
6017   }
6018   byte_offset = (uint64_t)dest.target() & 0xfff;
6019 }
6020 
6021 void MacroAssembler::load_byte_map_base(Register reg) {
6022   CardTable::CardValue* byte_map_base =
6023     ((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base();
6024 
6025   // Strictly speaking the byte_map_base isn't an address at all, and it might
6026   // even be negative. It is thus materialised as a constant.
6027   mov(reg, (uint64_t)byte_map_base);
6028 }
6029 
6030 void MacroAssembler::build_frame(int framesize) {
6031   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6032   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6033   protect_return_address();
6034   if (framesize < ((1 << 9) + 2 * wordSize)) {
6035     sub(sp, sp, framesize);
6036     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6037     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
6038   } else {
6039     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6040     if (PreserveFramePointer) mov(rfp, sp);
6041     if (framesize < ((1 << 12) + 2 * wordSize))
6042       sub(sp, sp, framesize - 2 * wordSize);
6043     else {
6044       mov(rscratch1, framesize - 2 * wordSize);
6045       sub(sp, sp, rscratch1);
6046     }
6047   }
6048   verify_cross_modify_fence_not_required();
6049 }
6050 
6051 void MacroAssembler::remove_frame(int framesize) {
6052   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6053   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6054   if (framesize < ((1 << 9) + 2 * wordSize)) {
6055     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6056     add(sp, sp, framesize);
6057   } else {
6058     if (framesize < ((1 << 12) + 2 * wordSize))
6059       add(sp, sp, framesize - 2 * wordSize);
6060     else {
6061       mov(rscratch1, framesize - 2 * wordSize);
6062       add(sp, sp, rscratch1);
6063     }
6064     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6065   }
6066   authenticate_return_address();
6067 }
6068 
6069 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
6070   if (needs_stack_repair) {
6071     // Remove the extension of the caller's frame used for inline type unpacking
6072     //
6073     // Right now the stack looks like this:
6074     //
6075     // | Arguments from caller     |
6076     // |---------------------------|  <-- caller's SP
6077     // | Saved LR #1               |
6078     // | Saved FP #1               |
6079     // |---------------------------|
6080     // | Extension space for       |
6081     // |   inline arg (un)packing  |
6082     // |---------------------------|  <-- start of this method's frame
6083     // | Saved LR #2               |
6084     // | Saved FP #2               |
6085     // |---------------------------|  <-- FP
6086     // | sp_inc                    |
6087     // | method locals             |
6088     // |---------------------------|  <-- SP
6089     //
6090     // There are two copies of FP and LR on the stack. They will be identical at
6091     // first, but that can change.
6092     // If the caller has been deoptimized, LR #1 will be patched to point at the
6093     // deopt blob, and LR #2 will still point into the old method.
6094     // If the saved FP (x29) was not used as the frame pointer, but to store an
6095     // oop, the GC will be aware only of FP #2 as the spilled location of x29 and
6096     // will fix only this one.
6097     //
6098     // When restoring, one must then load FP #2 into x29, and LR #1 into x30,
6099     // while keeping in mind that from the scalarized entry point, there will be
6100     // only one copy of each.
6101     //
6102     // The sp_inc stack slot holds the total size of the frame including the
6103     // extension space minus two words for the saved FP and LR. That is how to
6104     // find LR #1. FP #2 is always located just after sp_inc.
6105 
6106     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6107 
6108     ldr(rscratch1, Address(sp, sp_inc_offset));
6109     ldr(rfp, Address(sp, sp_inc_offset + wordSize));
6110     add(sp, sp, rscratch1);
6111     ldr(lr, Address(sp, wordSize));
6112     add(sp, sp, 2 * wordSize);
6113   } else {
6114     remove_frame(initial_framesize);
6115   }
6116 }
6117 
6118 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6119   int real_frame_size = frame_size + sp_inc;
6120   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6121   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6122   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6123 
6124   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6125 
6126   // Subtract two words for the saved FP and LR as these will be popped
6127   // separately. See remove_frame above.
6128   mov(rscratch1, real_frame_size - 2*wordSize);
6129   str(rscratch1, Address(sp, sp_inc_offset));
6130 }
6131 
6132 // This method counts leading positive bytes (highest bit not set) in provided byte array
6133 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6134     // Simple and most common case of aligned small array which is not at the
6135     // end of memory page is placed here. All other cases are in stub.
6136     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6137     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6138     assert_different_registers(ary1, len, result);
6139 
6140     mov(result, len);
6141     cmpw(len, 0);
6142     br(LE, DONE);
6143     cmpw(len, 4 * wordSize);
6144     br(GE, STUB_LONG); // size > 32 then go to stub
6145 
6146     int shift = 64 - exact_log2(os::vm_page_size());
6147     lsl(rscratch1, ary1, shift);
6148     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6149     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6150     br(CS, STUB); // at the end of page then go to stub
6151     subs(len, len, wordSize);
6152     br(LT, END);
6153 
6154   BIND(LOOP);
6155     ldr(rscratch1, Address(post(ary1, wordSize)));
6156     tst(rscratch1, UPPER_BIT_MASK);
6157     br(NE, SET_RESULT);
6158     subs(len, len, wordSize);
6159     br(GE, LOOP);
6160     cmpw(len, -wordSize);
6161     br(EQ, DONE);
6162 
6163   BIND(END);
6164     ldr(rscratch1, Address(ary1));
6165     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
6166     lslv(rscratch1, rscratch1, rscratch2);
6167     tst(rscratch1, UPPER_BIT_MASK);
6168     br(NE, SET_RESULT);
6169     b(DONE);
6170 
6171   BIND(STUB);
6172     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
6173     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
6174     address tpc1 = trampoline_call(count_pos);
6175     if (tpc1 == nullptr) {
6176       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
6177       postcond(pc() == badAddress);
6178       return nullptr;
6179     }
6180     b(DONE);
6181 
6182   BIND(STUB_LONG);
6183     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
6184     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
6185     address tpc2 = trampoline_call(count_pos_long);
6186     if (tpc2 == nullptr) {
6187       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
6188       postcond(pc() == badAddress);
6189       return nullptr;
6190     }
6191     b(DONE);
6192 
6193   BIND(SET_RESULT);
6194 
6195     add(len, len, wordSize);
6196     sub(result, result, len);
6197 
6198   BIND(DONE);
6199   postcond(pc() != badAddress);
6200   return pc();
6201 }
6202 
6203 // Clobbers: rscratch1, rscratch2, rflags
6204 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
6205 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
6206                                       Register tmp4, Register tmp5, Register result,
6207                                       Register cnt1, int elem_size) {
6208   Label DONE, SAME;
6209   Register tmp1 = rscratch1;
6210   Register tmp2 = rscratch2;
6211   int elem_per_word = wordSize/elem_size;
6212   int log_elem_size = exact_log2(elem_size);
6213   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
6214   int length_offset = arrayOopDesc::length_offset_in_bytes();
6215   int base_offset
6216     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
6217   // When the length offset is not aligned to 8 bytes,
6218   // then we align it down. This is valid because the new
6219   // offset will always be the klass which is the same
6220   // for type arrays.
6221   int start_offset = align_down(length_offset, BytesPerWord);
6222   int extra_length = base_offset - start_offset;
6223   assert(start_offset == length_offset || start_offset == klass_offset,
6224          "start offset must be 8-byte-aligned or be the klass offset");
6225   assert(base_offset != start_offset, "must include the length field");
6226   extra_length = extra_length / elem_size; // We count in elements, not bytes.
6227   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
6228 
6229   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
6230   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6231 
6232 #ifndef PRODUCT
6233   {
6234     const char kind = (elem_size == 2) ? 'U' : 'L';
6235     char comment[64];
6236     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
6237     BLOCK_COMMENT(comment);
6238   }
6239 #endif
6240 
6241   // if (a1 == a2)
6242   //     return true;
6243   cmpoop(a1, a2); // May have read barriers for a1 and a2.
6244   br(EQ, SAME);
6245 
6246   if (UseSimpleArrayEquals) {
6247     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
6248     // if (a1 == nullptr || a2 == nullptr)
6249     //     return false;
6250     // a1 & a2 == 0 means (some-pointer is null) or
6251     // (very-rare-or-even-probably-impossible-pointer-values)
6252     // so, we can save one branch in most cases
6253     tst(a1, a2);
6254     mov(result, false);
6255     br(EQ, A_MIGHT_BE_NULL);
6256     // if (a1.length != a2.length)
6257     //      return false;
6258     bind(A_IS_NOT_NULL);
6259     ldrw(cnt1, Address(a1, length_offset));
6260     // Increase loop counter by diff between base- and actual start-offset.
6261     addw(cnt1, cnt1, extra_length);
6262     lea(a1, Address(a1, start_offset));
6263     lea(a2, Address(a2, start_offset));
6264     // Check for short strings, i.e. smaller than wordSize.
6265     subs(cnt1, cnt1, elem_per_word);
6266     br(Assembler::LT, SHORT);
6267     // Main 8 byte comparison loop.
6268     bind(NEXT_WORD); {
6269       ldr(tmp1, Address(post(a1, wordSize)));
6270       ldr(tmp2, Address(post(a2, wordSize)));
6271       subs(cnt1, cnt1, elem_per_word);
6272       eor(tmp5, tmp1, tmp2);
6273       cbnz(tmp5, DONE);
6274     } br(GT, NEXT_WORD);
6275     // Last longword.  In the case where length == 4 we compare the
6276     // same longword twice, but that's still faster than another
6277     // conditional branch.
6278     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6279     // length == 4.
6280     if (log_elem_size > 0)
6281       lsl(cnt1, cnt1, log_elem_size);
6282     ldr(tmp3, Address(a1, cnt1));
6283     ldr(tmp4, Address(a2, cnt1));
6284     eor(tmp5, tmp3, tmp4);
6285     cbnz(tmp5, DONE);
6286     b(SAME);
6287     bind(A_MIGHT_BE_NULL);
6288     // in case both a1 and a2 are not-null, proceed with loads
6289     cbz(a1, DONE);
6290     cbz(a2, DONE);
6291     b(A_IS_NOT_NULL);
6292     bind(SHORT);
6293 
6294     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
6295     {
6296       ldrw(tmp1, Address(post(a1, 4)));
6297       ldrw(tmp2, Address(post(a2, 4)));
6298       eorw(tmp5, tmp1, tmp2);
6299       cbnzw(tmp5, DONE);
6300     }
6301     bind(TAIL03);
6302     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6303     {
6304       ldrh(tmp3, Address(post(a1, 2)));
6305       ldrh(tmp4, Address(post(a2, 2)));
6306       eorw(tmp5, tmp3, tmp4);
6307       cbnzw(tmp5, DONE);
6308     }
6309     bind(TAIL01);
6310     if (elem_size == 1) { // Only needed when comparing byte arrays.
6311       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6312       {
6313         ldrb(tmp1, a1);
6314         ldrb(tmp2, a2);
6315         eorw(tmp5, tmp1, tmp2);
6316         cbnzw(tmp5, DONE);
6317       }
6318     }
6319   } else {
6320     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6321         CSET_EQ, LAST_CHECK;
6322     mov(result, false);
6323     cbz(a1, DONE);
6324     ldrw(cnt1, Address(a1, length_offset));
6325     cbz(a2, DONE);
6326     // Increase loop counter by diff between base- and actual start-offset.
6327     addw(cnt1, cnt1, extra_length);
6328 
6329     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6330     // faster to perform another branch before comparing a1 and a2
6331     cmp(cnt1, (u1)elem_per_word);
6332     br(LE, SHORT); // short or same
6333     ldr(tmp3, Address(pre(a1, start_offset)));
6334     subs(zr, cnt1, stubBytesThreshold);
6335     br(GE, STUB);
6336     ldr(tmp4, Address(pre(a2, start_offset)));
6337     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6338 
6339     // Main 16 byte comparison loop with 2 exits
6340     bind(NEXT_DWORD); {
6341       ldr(tmp1, Address(pre(a1, wordSize)));
6342       ldr(tmp2, Address(pre(a2, wordSize)));
6343       subs(cnt1, cnt1, 2 * elem_per_word);
6344       br(LE, TAIL);
6345       eor(tmp4, tmp3, tmp4);
6346       cbnz(tmp4, DONE);
6347       ldr(tmp3, Address(pre(a1, wordSize)));
6348       ldr(tmp4, Address(pre(a2, wordSize)));
6349       cmp(cnt1, (u1)elem_per_word);
6350       br(LE, TAIL2);
6351       cmp(tmp1, tmp2);
6352     } br(EQ, NEXT_DWORD);
6353     b(DONE);
6354 
6355     bind(TAIL);
6356     eor(tmp4, tmp3, tmp4);
6357     eor(tmp2, tmp1, tmp2);
6358     lslv(tmp2, tmp2, tmp5);
6359     orr(tmp5, tmp4, tmp2);
6360     cmp(tmp5, zr);
6361     b(CSET_EQ);
6362 
6363     bind(TAIL2);
6364     eor(tmp2, tmp1, tmp2);
6365     cbnz(tmp2, DONE);
6366     b(LAST_CHECK);
6367 
6368     bind(STUB);
6369     ldr(tmp4, Address(pre(a2, start_offset)));
6370     if (elem_size == 2) { // convert to byte counter
6371       lsl(cnt1, cnt1, 1);
6372     }
6373     eor(tmp5, tmp3, tmp4);
6374     cbnz(tmp5, DONE);
6375     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6376     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6377     address tpc = trampoline_call(stub);
6378     if (tpc == nullptr) {
6379       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6380       postcond(pc() == badAddress);
6381       return nullptr;
6382     }
6383     b(DONE);
6384 
6385     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6386     // so, if a2 == null => return false(0), else return true, so we can return a2
6387     mov(result, a2);
6388     b(DONE);
6389     bind(SHORT);
6390     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6391     ldr(tmp3, Address(a1, start_offset));
6392     ldr(tmp4, Address(a2, start_offset));
6393     bind(LAST_CHECK);
6394     eor(tmp4, tmp3, tmp4);
6395     lslv(tmp5, tmp4, tmp5);
6396     cmp(tmp5, zr);
6397     bind(CSET_EQ);
6398     cset(result, EQ);
6399     b(DONE);
6400   }
6401 
6402   bind(SAME);
6403   mov(result, true);
6404   // That's it.
6405   bind(DONE);
6406 
6407   BLOCK_COMMENT("} array_equals");
6408   postcond(pc() != badAddress);
6409   return pc();
6410 }
6411 
6412 // Compare Strings
6413 
6414 // For Strings we're passed the address of the first characters in a1
6415 // and a2 and the length in cnt1.
6416 // There are two implementations.  For arrays >= 8 bytes, all
6417 // comparisons (including the final one, which may overlap) are
6418 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6419 // halfword, then a short, and then a byte.
6420 
6421 void MacroAssembler::string_equals(Register a1, Register a2,
6422                                    Register result, Register cnt1)
6423 {
6424   Label SAME, DONE, SHORT, NEXT_WORD;
6425   Register tmp1 = rscratch1;
6426   Register tmp2 = rscratch2;
6427   Register cnt2 = tmp2;  // cnt2 only used in array length compare
6428 
6429   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6430 
6431 #ifndef PRODUCT
6432   {
6433     char comment[64];
6434     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6435     BLOCK_COMMENT(comment);
6436   }
6437 #endif
6438 
6439   mov(result, false);
6440 
6441   // Check for short strings, i.e. smaller than wordSize.
6442   subs(cnt1, cnt1, wordSize);
6443   br(Assembler::LT, SHORT);
6444   // Main 8 byte comparison loop.
6445   bind(NEXT_WORD); {
6446     ldr(tmp1, Address(post(a1, wordSize)));
6447     ldr(tmp2, Address(post(a2, wordSize)));
6448     subs(cnt1, cnt1, wordSize);
6449     eor(tmp1, tmp1, tmp2);
6450     cbnz(tmp1, DONE);
6451   } br(GT, NEXT_WORD);
6452   // Last longword.  In the case where length == 4 we compare the
6453   // same longword twice, but that's still faster than another
6454   // conditional branch.
6455   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6456   // length == 4.
6457   ldr(tmp1, Address(a1, cnt1));
6458   ldr(tmp2, Address(a2, cnt1));
6459   eor(tmp2, tmp1, tmp2);
6460   cbnz(tmp2, DONE);
6461   b(SAME);
6462 
6463   bind(SHORT);
6464   Label TAIL03, TAIL01;
6465 
6466   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6467   {
6468     ldrw(tmp1, Address(post(a1, 4)));
6469     ldrw(tmp2, Address(post(a2, 4)));
6470     eorw(tmp1, tmp1, tmp2);
6471     cbnzw(tmp1, DONE);
6472   }
6473   bind(TAIL03);
6474   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6475   {
6476     ldrh(tmp1, Address(post(a1, 2)));
6477     ldrh(tmp2, Address(post(a2, 2)));
6478     eorw(tmp1, tmp1, tmp2);
6479     cbnzw(tmp1, DONE);
6480   }
6481   bind(TAIL01);
6482   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6483     {
6484     ldrb(tmp1, a1);
6485     ldrb(tmp2, a2);
6486     eorw(tmp1, tmp1, tmp2);
6487     cbnzw(tmp1, DONE);
6488   }
6489   // Arrays are equal.
6490   bind(SAME);
6491   mov(result, true);
6492 
6493   // That's it.
6494   bind(DONE);
6495   BLOCK_COMMENT("} string_equals");
6496 }
6497 
6498 
6499 // The size of the blocks erased by the zero_blocks stub.  We must
6500 // handle anything smaller than this ourselves in zero_words().
6501 const int MacroAssembler::zero_words_block_size = 8;
6502 
6503 // zero_words() is used by C2 ClearArray patterns and by
6504 // C1_MacroAssembler.  It is as small as possible, handling small word
6505 // counts locally and delegating anything larger to the zero_blocks
6506 // stub.  It is expanded many times in compiled code, so it is
6507 // important to keep it short.
6508 
6509 // ptr:   Address of a buffer to be zeroed.
6510 // cnt:   Count in HeapWords.
6511 //
6512 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6513 address MacroAssembler::zero_words(Register ptr, Register cnt)
6514 {
6515   assert(is_power_of_2(zero_words_block_size), "adjust this");
6516 
6517   BLOCK_COMMENT("zero_words {");
6518   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6519   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6520   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6521 
6522   subs(rscratch1, cnt, zero_words_block_size);
6523   Label around;
6524   br(LO, around);
6525   {
6526     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6527     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6528     // Make sure this is a C2 compilation. C1 allocates space only for
6529     // trampoline stubs generated by Call LIR ops, and in any case it
6530     // makes sense for a C1 compilation task to proceed as quickly as
6531     // possible.
6532     CompileTask* task;
6533     if (StubRoutines::aarch64::complete()
6534         && Thread::current()->is_Compiler_thread()
6535         && (task = ciEnv::current()->task())
6536         && is_c2_compile(task->comp_level())) {
6537       address tpc = trampoline_call(zero_blocks);
6538       if (tpc == nullptr) {
6539         DEBUG_ONLY(reset_labels(around));
6540         return nullptr;
6541       }
6542     } else {
6543       far_call(zero_blocks);
6544     }
6545   }
6546   bind(around);
6547 
6548   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6549   // for us.
6550   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6551     Label l;
6552     tbz(cnt, exact_log2(i), l);
6553     for (int j = 0; j < i; j += 2) {
6554       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6555     }
6556     bind(l);
6557   }
6558   {
6559     Label l;
6560     tbz(cnt, 0, l);
6561     str(zr, Address(ptr));
6562     bind(l);
6563   }
6564 
6565   BLOCK_COMMENT("} zero_words");
6566   return pc();
6567 }
6568 
6569 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6570 // cnt:          Immediate count in HeapWords.
6571 //
6572 // r10, r11, rscratch1, and rscratch2 are clobbered.
6573 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6574 {
6575   assert(wordSize <= BlockZeroingLowLimit,
6576             "increase BlockZeroingLowLimit");
6577   address result = nullptr;
6578   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6579 #ifndef PRODUCT
6580     {
6581       char buf[64];
6582       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6583       BLOCK_COMMENT(buf);
6584     }
6585 #endif
6586     if (cnt >= 16) {
6587       uint64_t loops = cnt/16;
6588       if (loops > 1) {
6589         mov(rscratch2, loops - 1);
6590       }
6591       {
6592         Label loop;
6593         bind(loop);
6594         for (int i = 0; i < 16; i += 2) {
6595           stp(zr, zr, Address(base, i * BytesPerWord));
6596         }
6597         add(base, base, 16 * BytesPerWord);
6598         if (loops > 1) {
6599           subs(rscratch2, rscratch2, 1);
6600           br(GE, loop);
6601         }
6602       }
6603     }
6604     cnt %= 16;
6605     int i = cnt & 1;  // store any odd word to start
6606     if (i) str(zr, Address(base));
6607     for (; i < (int)cnt; i += 2) {
6608       stp(zr, zr, Address(base, i * wordSize));
6609     }
6610     BLOCK_COMMENT("} zero_words");
6611     result = pc();
6612   } else {
6613     mov(r10, base); mov(r11, cnt);
6614     result = zero_words(r10, r11);
6615   }
6616   return result;
6617 }
6618 
6619 // Zero blocks of memory by using DC ZVA.
6620 //
6621 // Aligns the base address first sufficiently for DC ZVA, then uses
6622 // DC ZVA repeatedly for every full block.  cnt is the size to be
6623 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6624 // in cnt.
6625 //
6626 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6627 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6628 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6629   Register tmp = rscratch1;
6630   Register tmp2 = rscratch2;
6631   int zva_length = VM_Version::zva_length();
6632   Label initial_table_end, loop_zva;
6633   Label fini;
6634 
6635   // Base must be 16 byte aligned. If not just return and let caller handle it
6636   tst(base, 0x0f);
6637   br(Assembler::NE, fini);
6638   // Align base with ZVA length.
6639   neg(tmp, base);
6640   andr(tmp, tmp, zva_length - 1);
6641 
6642   // tmp: the number of bytes to be filled to align the base with ZVA length.
6643   add(base, base, tmp);
6644   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6645   adr(tmp2, initial_table_end);
6646   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6647   br(tmp2);
6648 
6649   for (int i = -zva_length + 16; i < 0; i += 16)
6650     stp(zr, zr, Address(base, i));
6651   bind(initial_table_end);
6652 
6653   sub(cnt, cnt, zva_length >> 3);
6654   bind(loop_zva);
6655   dc(Assembler::ZVA, base);
6656   subs(cnt, cnt, zva_length >> 3);
6657   add(base, base, zva_length);
6658   br(Assembler::GE, loop_zva);
6659   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6660   bind(fini);
6661 }
6662 
6663 // base:   Address of a buffer to be filled, 8 bytes aligned.
6664 // cnt:    Count in 8-byte unit.
6665 // value:  Value to be filled with.
6666 // base will point to the end of the buffer after filling.
6667 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6668 {
6669 //  Algorithm:
6670 //
6671 //    if (cnt == 0) {
6672 //      return;
6673 //    }
6674 //    if ((p & 8) != 0) {
6675 //      *p++ = v;
6676 //    }
6677 //
6678 //    scratch1 = cnt & 14;
6679 //    cnt -= scratch1;
6680 //    p += scratch1;
6681 //    switch (scratch1 / 2) {
6682 //      do {
6683 //        cnt -= 16;
6684 //          p[-16] = v;
6685 //          p[-15] = v;
6686 //        case 7:
6687 //          p[-14] = v;
6688 //          p[-13] = v;
6689 //        case 6:
6690 //          p[-12] = v;
6691 //          p[-11] = v;
6692 //          // ...
6693 //        case 1:
6694 //          p[-2] = v;
6695 //          p[-1] = v;
6696 //        case 0:
6697 //          p += 16;
6698 //      } while (cnt);
6699 //    }
6700 //    if ((cnt & 1) == 1) {
6701 //      *p++ = v;
6702 //    }
6703 
6704   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6705 
6706   Label fini, skip, entry, loop;
6707   const int unroll = 8; // Number of stp instructions we'll unroll
6708 
6709   cbz(cnt, fini);
6710   tbz(base, 3, skip);
6711   str(value, Address(post(base, 8)));
6712   sub(cnt, cnt, 1);
6713   bind(skip);
6714 
6715   andr(rscratch1, cnt, (unroll-1) * 2);
6716   sub(cnt, cnt, rscratch1);
6717   add(base, base, rscratch1, Assembler::LSL, 3);
6718   adr(rscratch2, entry);
6719   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6720   br(rscratch2);
6721 
6722   bind(loop);
6723   add(base, base, unroll * 16);
6724   for (int i = -unroll; i < 0; i++)
6725     stp(value, value, Address(base, i * 16));
6726   bind(entry);
6727   subs(cnt, cnt, unroll * 2);
6728   br(Assembler::GE, loop);
6729 
6730   tbz(cnt, 0, fini);
6731   str(value, Address(post(base, 8)));
6732   bind(fini);
6733 }
6734 
6735 // Intrinsic for
6736 //
6737 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6738 //   Encodes char[] to byte[] in ISO-8859-1
6739 //
6740 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6741 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6742 //
6743 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6744 //   Encodes char[] to byte[] in ASCII
6745 //
6746 // This version always returns the number of characters copied, and does not
6747 // clobber the 'len' register. A successful copy will complete with the post-
6748 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6749 // post-condition: 0 <= 'res' < 'len'.
6750 //
6751 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6752 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6753 //       beyond the acceptable, even though the footprint would be smaller.
6754 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6755 //       avoid additional bloat.
6756 //
6757 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6758 void MacroAssembler::encode_iso_array(Register src, Register dst,
6759                                       Register len, Register res, bool ascii,
6760                                       FloatRegister vtmp0, FloatRegister vtmp1,
6761                                       FloatRegister vtmp2, FloatRegister vtmp3,
6762                                       FloatRegister vtmp4, FloatRegister vtmp5)
6763 {
6764   Register cnt = res;
6765   Register max = rscratch1;
6766   Register chk = rscratch2;
6767 
6768   prfm(Address(src), PLDL1STRM);
6769   movw(cnt, len);
6770 
6771 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6772 
6773   Label LOOP_32, DONE_32, FAIL_32;
6774 
6775   BIND(LOOP_32);
6776   {
6777     cmpw(cnt, 32);
6778     br(LT, DONE_32);
6779     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6780     // Extract lower bytes.
6781     FloatRegister vlo0 = vtmp4;
6782     FloatRegister vlo1 = vtmp5;
6783     uzp1(vlo0, T16B, vtmp0, vtmp1);
6784     uzp1(vlo1, T16B, vtmp2, vtmp3);
6785     // Merge bits...
6786     orr(vtmp0, T16B, vtmp0, vtmp1);
6787     orr(vtmp2, T16B, vtmp2, vtmp3);
6788     // Extract merged upper bytes.
6789     FloatRegister vhix = vtmp0;
6790     uzp2(vhix, T16B, vtmp0, vtmp2);
6791     // ISO-check on hi-parts (all zero).
6792     //                          ASCII-check on lo-parts (no sign).
6793     FloatRegister vlox = vtmp1; // Merge lower bytes.
6794                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6795     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6796     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6797     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6798                                 ASCII(orr(chk, chk, max));
6799     cbnz(chk, FAIL_32);
6800     subw(cnt, cnt, 32);
6801     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6802     b(LOOP_32);
6803   }
6804   BIND(FAIL_32);
6805   sub(src, src, 64);
6806   BIND(DONE_32);
6807 
6808   Label LOOP_8, SKIP_8;
6809 
6810   BIND(LOOP_8);
6811   {
6812     cmpw(cnt, 8);
6813     br(LT, SKIP_8);
6814     FloatRegister vhi = vtmp0;
6815     FloatRegister vlo = vtmp1;
6816     ld1(vtmp3, T8H, src);
6817     uzp1(vlo, T16B, vtmp3, vtmp3);
6818     uzp2(vhi, T16B, vtmp3, vtmp3);
6819     // ISO-check on hi-parts (all zero).
6820     //                          ASCII-check on lo-parts (no sign).
6821                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6822     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6823                                 ASCII(umov(max, vtmp2, B, 0));
6824                                 ASCII(orr(chk, chk, max));
6825     cbnz(chk, SKIP_8);
6826 
6827     strd(vlo, Address(post(dst, 8)));
6828     subw(cnt, cnt, 8);
6829     add(src, src, 16);
6830     b(LOOP_8);
6831   }
6832   BIND(SKIP_8);
6833 
6834 #undef ASCII
6835 
6836   Label LOOP, DONE;
6837 
6838   cbz(cnt, DONE);
6839   BIND(LOOP);
6840   {
6841     Register chr = rscratch1;
6842     ldrh(chr, Address(post(src, 2)));
6843     tst(chr, ascii ? 0xff80 : 0xff00);
6844     br(NE, DONE);
6845     strb(chr, Address(post(dst, 1)));
6846     subs(cnt, cnt, 1);
6847     br(GT, LOOP);
6848   }
6849   BIND(DONE);
6850   // Return index where we stopped.
6851   subw(res, len, cnt);
6852 }
6853 
6854 // Inflate byte[] array to char[].
6855 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6856 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6857                                            FloatRegister vtmp1, FloatRegister vtmp2,
6858                                            FloatRegister vtmp3, Register tmp4) {
6859   Label big, done, after_init, to_stub;
6860 
6861   assert_different_registers(src, dst, len, tmp4, rscratch1);
6862 
6863   fmovd(vtmp1, 0.0);
6864   lsrw(tmp4, len, 3);
6865   bind(after_init);
6866   cbnzw(tmp4, big);
6867   // Short string: less than 8 bytes.
6868   {
6869     Label loop, tiny;
6870 
6871     cmpw(len, 4);
6872     br(LT, tiny);
6873     // Use SIMD to do 4 bytes.
6874     ldrs(vtmp2, post(src, 4));
6875     zip1(vtmp3, T8B, vtmp2, vtmp1);
6876     subw(len, len, 4);
6877     strd(vtmp3, post(dst, 8));
6878 
6879     cbzw(len, done);
6880 
6881     // Do the remaining bytes by steam.
6882     bind(loop);
6883     ldrb(tmp4, post(src, 1));
6884     strh(tmp4, post(dst, 2));
6885     subw(len, len, 1);
6886 
6887     bind(tiny);
6888     cbnz(len, loop);
6889 
6890     b(done);
6891   }
6892 
6893   if (SoftwarePrefetchHintDistance >= 0) {
6894     bind(to_stub);
6895       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6896       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6897       address tpc = trampoline_call(stub);
6898       if (tpc == nullptr) {
6899         DEBUG_ONLY(reset_labels(big, done));
6900         postcond(pc() == badAddress);
6901         return nullptr;
6902       }
6903       b(after_init);
6904   }
6905 
6906   // Unpack the bytes 8 at a time.
6907   bind(big);
6908   {
6909     Label loop, around, loop_last, loop_start;
6910 
6911     if (SoftwarePrefetchHintDistance >= 0) {
6912       const int large_loop_threshold = (64 + 16)/8;
6913       ldrd(vtmp2, post(src, 8));
6914       andw(len, len, 7);
6915       cmp(tmp4, (u1)large_loop_threshold);
6916       br(GE, to_stub);
6917       b(loop_start);
6918 
6919       bind(loop);
6920       ldrd(vtmp2, post(src, 8));
6921       bind(loop_start);
6922       subs(tmp4, tmp4, 1);
6923       br(EQ, loop_last);
6924       zip1(vtmp2, T16B, vtmp2, vtmp1);
6925       ldrd(vtmp3, post(src, 8));
6926       st1(vtmp2, T8H, post(dst, 16));
6927       subs(tmp4, tmp4, 1);
6928       zip1(vtmp3, T16B, vtmp3, vtmp1);
6929       st1(vtmp3, T8H, post(dst, 16));
6930       br(NE, loop);
6931       b(around);
6932       bind(loop_last);
6933       zip1(vtmp2, T16B, vtmp2, vtmp1);
6934       st1(vtmp2, T8H, post(dst, 16));
6935       bind(around);
6936       cbz(len, done);
6937     } else {
6938       andw(len, len, 7);
6939       bind(loop);
6940       ldrd(vtmp2, post(src, 8));
6941       sub(tmp4, tmp4, 1);
6942       zip1(vtmp3, T16B, vtmp2, vtmp1);
6943       st1(vtmp3, T8H, post(dst, 16));
6944       cbnz(tmp4, loop);
6945     }
6946   }
6947 
6948   // Do the tail of up to 8 bytes.
6949   add(src, src, len);
6950   ldrd(vtmp3, Address(src, -8));
6951   add(dst, dst, len, ext::uxtw, 1);
6952   zip1(vtmp3, T16B, vtmp3, vtmp1);
6953   strq(vtmp3, Address(dst, -16));
6954 
6955   bind(done);
6956   postcond(pc() != badAddress);
6957   return pc();
6958 }
6959 
6960 // Compress char[] array to byte[].
6961 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6962 // Return the array length if every element in array can be encoded,
6963 // otherwise, the index of first non-latin1 (> 0xff) character.
6964 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6965                                          Register res,
6966                                          FloatRegister tmp0, FloatRegister tmp1,
6967                                          FloatRegister tmp2, FloatRegister tmp3,
6968                                          FloatRegister tmp4, FloatRegister tmp5) {
6969   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6970 }
6971 
6972 // java.math.round(double a)
6973 // Returns the closest long to the argument, with ties rounding to
6974 // positive infinity.  This requires some fiddling for corner
6975 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6976 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6977                                        FloatRegister ftmp) {
6978   Label DONE;
6979   BLOCK_COMMENT("java_round_double: { ");
6980   fmovd(rscratch1, src);
6981   // Use RoundToNearestTiesAway unless src small and -ve.
6982   fcvtasd(dst, src);
6983   // Test if src >= 0 || abs(src) >= 0x1.0p52
6984   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6985   mov(rscratch2, julong_cast(0x1.0p52));
6986   cmp(rscratch1, rscratch2);
6987   br(HS, DONE); {
6988     // src < 0 && abs(src) < 0x1.0p52
6989     // src may have a fractional part, so add 0.5
6990     fmovd(ftmp, 0.5);
6991     faddd(ftmp, src, ftmp);
6992     // Convert double to jlong, use RoundTowardsNegative
6993     fcvtmsd(dst, ftmp);
6994   }
6995   bind(DONE);
6996   BLOCK_COMMENT("} java_round_double");
6997 }
6998 
6999 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
7000                                       FloatRegister ftmp) {
7001   Label DONE;
7002   BLOCK_COMMENT("java_round_float: { ");
7003   fmovs(rscratch1, src);
7004   // Use RoundToNearestTiesAway unless src small and -ve.
7005   fcvtassw(dst, src);
7006   // Test if src >= 0 || abs(src) >= 0x1.0p23
7007   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
7008   mov(rscratch2, jint_cast(0x1.0p23f));
7009   cmp(rscratch1, rscratch2);
7010   br(HS, DONE); {
7011     // src < 0 && |src| < 0x1.0p23
7012     // src may have a fractional part, so add 0.5
7013     fmovs(ftmp, 0.5f);
7014     fadds(ftmp, src, ftmp);
7015     // Convert float to jint, use RoundTowardsNegative
7016     fcvtmssw(dst, ftmp);
7017   }
7018   bind(DONE);
7019   BLOCK_COMMENT("} java_round_float");
7020 }
7021 
7022 // get_thread() can be called anywhere inside generated code so we
7023 // need to save whatever non-callee save context might get clobbered
7024 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
7025 // the call setup code.
7026 //
7027 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
7028 // On other systems, the helper is a usual C function.
7029 //
7030 void MacroAssembler::get_thread(Register dst) {
7031   RegSet saved_regs =
7032     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
7033     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
7034 
7035   protect_return_address();
7036   push(saved_regs, sp);
7037 
7038   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
7039   blr(lr);
7040   if (dst != c_rarg0) {
7041     mov(dst, c_rarg0);
7042   }
7043 
7044   pop(saved_regs, sp);
7045   authenticate_return_address();
7046 }
7047 
7048 #ifdef COMPILER2
7049 // C2 compiled method's prolog code
7050 // Moved here from aarch64.ad to support Valhalla code belows
7051 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
7052   if (C->clinit_barrier_on_entry()) {
7053     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
7054 
7055     Label L_skip_barrier;
7056 
7057     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7058     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7059     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7060     bind(L_skip_barrier);
7061   }
7062 
7063   if (C->max_vector_size() > 0) {
7064     reinitialize_ptrue();
7065   }
7066 
7067   int bangsize = C->output()->bang_size_in_bytes();
7068   if (C->output()->need_stack_bang(bangsize))
7069     generate_stack_overflow_check(bangsize);
7070 
7071   // n.b. frame size includes space for return pc and rfp
7072   const long framesize = C->output()->frame_size_in_bytes();
7073   build_frame(framesize);
7074 
7075   if (C->needs_stack_repair()) {
7076     save_stack_increment(sp_inc, framesize);
7077   }
7078 
7079   if (VerifyStackAtCalls) {
7080     Unimplemented();
7081   }
7082 }
7083 #endif // COMPILER2
7084 
7085 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7086   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7087   // An inline type might be returned. If fields are in registers we
7088   // need to allocate an inline type instance and initialize it with
7089   // the value of the fields.
7090   Label skip;
7091   // We only need a new buffered inline type if a new one is not returned
7092   tbz(r0, 0, skip);
7093   int call_offset = -1;
7094 
7095   // Be careful not to clobber r1-7 which hold returned fields
7096   // Also do not use callee-saved registers as these may be live in the interpreter
7097   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7098 
7099   // The following code is similar to allocate_instance but has some slight differences,
7100   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7101   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
7102   Label slow_case;
7103   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7104   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7105 
7106   if (vk != nullptr) {
7107     // Called from C1, where the return type is statically known.
7108     movptr(klass, (intptr_t)vk->get_InlineKlass());
7109     jint lh = vk->layout_helper();
7110     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7111     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7112       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7113     } else {
7114       b(slow_case);
7115     }
7116   } else {
7117     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7118     andr(klass, r0, -2);
7119     if (UseTLAB) {
7120       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7121       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7122       br(Assembler::NE, slow_case);
7123       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7124     } else {
7125       b(slow_case);
7126     }
7127   }
7128   if (UseTLAB) {
7129     // 2. Initialize buffered inline instance header
7130     Register buffer_obj = r0;
7131     if (UseCompactObjectHeaders) {
7132       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7133       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7134     } else {
7135       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7136       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7137       store_klass_gap(buffer_obj, zr);
7138       if (vk == nullptr) {
7139         // store_klass corrupts klass, so save it for later use (interpreter case only).
7140         mov(tmp1, klass);
7141       }
7142       store_klass(buffer_obj, klass);
7143       klass = tmp1;
7144     }
7145     // 3. Initialize its fields with an inline class specific handler
7146     if (vk != nullptr) {
7147       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7148     } else {
7149       ldr(tmp1, Address(klass, InstanceKlass::adr_inlineklass_fixed_block_offset()));
7150       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7151       blr(tmp1);
7152     }
7153 
7154     membar(Assembler::StoreStore);
7155     b(skip);
7156   } else {
7157     // Must have already branched to slow_case above.
7158     DEBUG_ONLY(should_not_reach_here());
7159   }
7160   bind(slow_case);
7161   // We failed to allocate a new inline type, fall back to a runtime
7162   // call. Some oop field may be live in some registers but we can't
7163   // tell. That runtime call will take care of preserving them
7164   // across a GC if there's one.
7165   mov(r0, r0_preserved);
7166 
7167   if (from_interpreter) {
7168     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7169   } else {
7170     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7171     call_offset = offset();
7172   }
7173   membar(Assembler::StoreStore);
7174 
7175   bind(skip);
7176   return call_offset;
7177 }
7178 
7179 // Move a value between registers/stack slots and update the reg_state
7180 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7181   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7182   if (reg_state[to->value()] == reg_written) {
7183     return true; // Already written
7184   }
7185 
7186   if (from != to && bt != T_VOID) {
7187     if (reg_state[to->value()] == reg_readonly) {
7188       return false; // Not yet writable
7189     }
7190     if (from->is_reg()) {
7191       if (to->is_reg()) {
7192         if (from->is_Register() && to->is_Register()) {
7193           mov(to->as_Register(), from->as_Register());
7194         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7195           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7196         } else {
7197           ShouldNotReachHere();
7198         }
7199       } else {
7200         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7201         Address to_addr = Address(sp, st_off);
7202         if (from->is_FloatRegister()) {
7203           if (bt == T_DOUBLE) {
7204              strd(from->as_FloatRegister(), to_addr);
7205           } else {
7206              assert(bt == T_FLOAT, "must be float");
7207              strs(from->as_FloatRegister(), to_addr);
7208           }
7209         } else {
7210           str(from->as_Register(), to_addr);
7211         }
7212       }
7213     } else {
7214       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7215       if (to->is_reg()) {
7216         if (to->is_FloatRegister()) {
7217           if (bt == T_DOUBLE) {
7218             ldrd(to->as_FloatRegister(), from_addr);
7219           } else {
7220             assert(bt == T_FLOAT, "must be float");
7221             ldrs(to->as_FloatRegister(), from_addr);
7222           }
7223         } else {
7224           ldr(to->as_Register(), from_addr);
7225         }
7226       } else {
7227         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7228         ldr(rscratch1, from_addr);
7229         str(rscratch1, Address(sp, st_off));
7230       }
7231     }
7232   }
7233 
7234   // Update register states
7235   reg_state[from->value()] = reg_writable;
7236   reg_state[to->value()] = reg_written;
7237   return true;
7238 }
7239 
7240 // Calculate the extra stack space required for packing or unpacking inline
7241 // args and adjust the stack pointer
7242 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7243   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7244   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7245   assert(sp_inc > 0, "sanity");
7246 
7247   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7248   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7249 
7250   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7251   if (sp_inc < (1 << 9)) {
7252     sub(sp, sp, sp_inc);   // Fits in an immediate
7253   } else {
7254     mov(rscratch1, sp_inc);
7255     sub(sp, sp, rscratch1);
7256   }
7257 
7258   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7259 }
7260 
7261 // Read all fields from an inline type oop and store the values in registers/stack slots
7262 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7263                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7264                                           RegState reg_state[]) {
7265   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7266   assert(from->is_valid(), "source must be valid");
7267   bool progress = false;
7268 #ifdef ASSERT
7269   const int start_offset = offset();
7270 #endif
7271 
7272   Label L_null, L_notNull;
7273   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7274   // TODO 8366717 We need to make sure that r14 (and potentially other long-life regs) are kept live in slowpath runtime calls in GC barriers
7275   Register tmp1 = r10;
7276   Register tmp2 = r11;
7277   Register fromReg = noreg;
7278   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, -1);
7279   bool done = true;
7280   bool mark_done = true;
7281   VMReg toReg;
7282   BasicType bt;
7283   // Check if argument requires a null check
7284   bool null_check = false;
7285   VMReg nullCheckReg;
7286   while (stream.next(nullCheckReg, bt)) {
7287     if (sig->at(stream.sig_index())._offset == -1) {
7288       null_check = true;
7289       break;
7290     }
7291   }
7292   stream.reset(sig_index, to_index);
7293   while (stream.next(toReg, bt)) {
7294     assert(toReg->is_valid(), "destination must be valid");
7295     int idx = (int)toReg->value();
7296     if (reg_state[idx] == reg_readonly) {
7297       if (idx != from->value()) {
7298         mark_done = false;
7299       }
7300       done = false;
7301       continue;
7302     } else if (reg_state[idx] == reg_written) {
7303       continue;
7304     }
7305     assert(reg_state[idx] == reg_writable, "must be writable");
7306     reg_state[idx] = reg_written;
7307     progress = true;
7308 
7309     if (fromReg == noreg) {
7310       if (from->is_reg()) {
7311         fromReg = from->as_Register();
7312       } else {
7313         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7314         ldr(tmp1, Address(sp, st_off));
7315         fromReg = tmp1;
7316       }
7317       if (null_check) {
7318         // Nullable inline type argument, emit null check
7319         cbz(fromReg, L_null);
7320       }
7321     }
7322     int off = sig->at(stream.sig_index())._offset;
7323     if (off == -1) {
7324       assert(null_check, "Missing null check at");
7325       if (toReg->is_stack()) {
7326         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7327         mov(tmp2, 1);
7328         str(tmp2, Address(sp, st_off));
7329       } else {
7330         mov(toReg->as_Register(), 1);
7331       }
7332       continue;
7333     }
7334     assert(off > 0, "offset in object should be positive");
7335     Address fromAddr = Address(fromReg, off);
7336     if (!toReg->is_FloatRegister()) {
7337       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7338       if (is_reference_type(bt)) {
7339         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7340       } else {
7341         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7342         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7343       }
7344       if (toReg->is_stack()) {
7345         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7346         str(dst, Address(sp, st_off));
7347       }
7348     } else if (bt == T_DOUBLE) {
7349       ldrd(toReg->as_FloatRegister(), fromAddr);
7350     } else {
7351       assert(bt == T_FLOAT, "must be float");
7352       ldrs(toReg->as_FloatRegister(), fromAddr);
7353     }
7354   }
7355   if (progress && null_check) {
7356     if (done) {
7357       b(L_notNull);
7358       bind(L_null);
7359       // Set null marker to zero to signal that the argument is null.
7360       // Also set all oop fields to zero to make the GC happy.
7361       stream.reset(sig_index, to_index);
7362       while (stream.next(toReg, bt)) {
7363         if (sig->at(stream.sig_index())._offset == -1 ||
7364             bt == T_OBJECT || bt == T_ARRAY) {
7365           if (toReg->is_stack()) {
7366             int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7367             str(zr, Address(sp, st_off));
7368           } else {
7369             mov(toReg->as_Register(), zr);
7370           }
7371         }
7372       }
7373       bind(L_notNull);
7374     } else {
7375       bind(L_null);
7376     }
7377   }
7378 
7379   // TODO 8366717 This is probably okay but looks fishy because stream is reset in the "Set null marker to zero" case just above. Same on x64.
7380   sig_index = stream.sig_index();
7381   to_index = stream.regs_index();
7382 
7383   if (mark_done && reg_state[from->value()] != reg_written) {
7384     // This is okay because no one else will write to that slot
7385     reg_state[from->value()] = reg_writable;
7386   }
7387   from_index--;
7388   assert(progress || (start_offset == offset()), "should not emit code");
7389   return done;
7390 }
7391 
7392 // Pack fields back into an inline type oop
7393 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7394                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7395                                         RegState reg_state[], Register val_array) {
7396   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7397   assert(to->is_valid(), "destination must be valid");
7398 
7399   if (reg_state[to->value()] == reg_written) {
7400     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7401     return true; // Already written
7402   }
7403 
7404   // The GC barrier expanded by store_heap_oop below may call into the
7405   // runtime so use callee-saved registers for any values that need to be
7406   // preserved. The GC barrier assembler should take care of saving the
7407   // Java argument registers.
7408   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
7409   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7410   Register val_obj_tmp = r21;
7411   Register from_reg_tmp = r22;
7412   Register tmp1 = r14;
7413   Register tmp2 = r13;
7414   Register tmp3 = r12;
7415   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7416 
7417   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7418 
7419   if (reg_state[to->value()] == reg_readonly) {
7420     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7421       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7422       return false; // Not yet writable
7423     }
7424     val_obj = val_obj_tmp;
7425   }
7426 
7427   int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7428   load_heap_oop(val_obj, Address(val_array, index), tmp1, tmp2);
7429 
7430   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7431   VMReg fromReg;
7432   BasicType bt;
7433   Label L_null;
7434   while (stream.next(fromReg, bt)) {
7435     assert(fromReg->is_valid(), "source must be valid");
7436     reg_state[fromReg->value()] = reg_writable;
7437 
7438     int off = sig->at(stream.sig_index())._offset;
7439     if (off == -1) {
7440       // Nullable inline type argument, emit null check
7441       Label L_notNull;
7442       if (fromReg->is_stack()) {
7443         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7444         ldrb(tmp2, Address(sp, ld_off));
7445         cbnz(tmp2, L_notNull);
7446       } else {
7447         cbnz(fromReg->as_Register(), L_notNull);
7448       }
7449       mov(val_obj, 0);
7450       b(L_null);
7451       bind(L_notNull);
7452       continue;
7453     }
7454 
7455     assert(off > 0, "offset in object should be positive");
7456     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7457 
7458     // Pack the scalarized field into the value object.
7459     Address dst(val_obj, off);
7460     if (!fromReg->is_FloatRegister()) {
7461       Register src;
7462       if (fromReg->is_stack()) {
7463         src = from_reg_tmp;
7464         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7465         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7466       } else {
7467         src = fromReg->as_Register();
7468       }
7469       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7470       if (is_reference_type(bt)) {
7471         store_heap_oop(dst, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7472       } else {
7473         store_sized_value(dst, src, size_in_bytes);
7474       }
7475     } else if (bt == T_DOUBLE) {
7476       strd(fromReg->as_FloatRegister(), dst);
7477     } else {
7478       assert(bt == T_FLOAT, "must be float");
7479       strs(fromReg->as_FloatRegister(), dst);
7480     }
7481   }
7482   bind(L_null);
7483   sig_index = stream.sig_index();
7484   from_index = stream.regs_index();
7485 
7486   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7487   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7488   assert(success, "to register must be writeable");
7489   return true;
7490 }
7491 
7492 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7493   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7494 }
7495 
7496 void MacroAssembler::cache_wb(Address line) {
7497   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7498   assert(line.index() == noreg, "index should be noreg");
7499   assert(line.offset() == 0, "offset should be 0");
7500   // would like to assert this
7501   // assert(line._ext.shift == 0, "shift should be zero");
7502   if (VM_Version::supports_dcpop()) {
7503     // writeback using clear virtual address to point of persistence
7504     dc(Assembler::CVAP, line.base());
7505   } else {
7506     // no need to generate anything as Unsafe.writebackMemory should
7507     // never invoke this stub
7508   }
7509 }
7510 
7511 void MacroAssembler::cache_wbsync(bool is_pre) {
7512   // we only need a barrier post sync
7513   if (!is_pre) {
7514     membar(Assembler::AnyAny);
7515   }
7516 }
7517 
7518 void MacroAssembler::verify_sve_vector_length(Register tmp) {
7519   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
7520     return;
7521   }
7522   // Make sure that native code does not change SVE vector length.
7523   Label verify_ok;
7524   movw(tmp, zr);
7525   sve_inc(tmp, B);
7526   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
7527   br(EQ, verify_ok);
7528   stop("Error: SVE vector length has changed since jvm startup");
7529   bind(verify_ok);
7530 }
7531 
7532 void MacroAssembler::verify_ptrue() {
7533   Label verify_ok;
7534   if (!UseSVE) {
7535     return;
7536   }
7537   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
7538   sve_dec(rscratch1, B);
7539   cbz(rscratch1, verify_ok);
7540   stop("Error: the preserved predicate register (p7) elements are not all true");
7541   bind(verify_ok);
7542 }
7543 
7544 void MacroAssembler::safepoint_isb() {
7545   isb();
7546 #ifndef PRODUCT
7547   if (VerifyCrossModifyFence) {
7548     // Clear the thread state.
7549     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7550   }
7551 #endif
7552 }
7553 
7554 #ifndef PRODUCT
7555 void MacroAssembler::verify_cross_modify_fence_not_required() {
7556   if (VerifyCrossModifyFence) {
7557     // Check if thread needs a cross modify fence.
7558     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7559     Label fence_not_required;
7560     cbz(rscratch1, fence_not_required);
7561     // If it does then fail.
7562     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
7563     mov(c_rarg0, rthread);
7564     blr(rscratch1);
7565     bind(fence_not_required);
7566   }
7567 }
7568 #endif
7569 
7570 void MacroAssembler::spin_wait() {
7571   block_comment("spin_wait {");
7572   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
7573     switch (VM_Version::spin_wait_desc().inst()) {
7574       case SpinWait::NOP:
7575         nop();
7576         break;
7577       case SpinWait::ISB:
7578         isb();
7579         break;
7580       case SpinWait::YIELD:
7581         yield();
7582         break;
7583       case SpinWait::SB:
7584         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
7585         sb();
7586         break;
7587       default:
7588         ShouldNotReachHere();
7589     }
7590   }
7591   block_comment("}");
7592 }
7593 
7594 // Stack frame creation/removal
7595 
7596 void MacroAssembler::enter(bool strip_ret_addr) {
7597   if (strip_ret_addr) {
7598     // Addresses can only be signed once. If there are multiple nested frames being created
7599     // in the same function, then the return address needs stripping first.
7600     strip_return_address();
7601   }
7602   protect_return_address();
7603   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7604   mov(rfp, sp);
7605 }
7606 
7607 void MacroAssembler::leave() {
7608   mov(sp, rfp);
7609   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
7610   authenticate_return_address();
7611 }
7612 
7613 // ROP Protection
7614 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
7615 // destroying stack frames or whenever directly loading/storing the LR to memory.
7616 // If ROP protection is not set then these functions are no-ops.
7617 // For more details on PAC see pauth_aarch64.hpp.
7618 
7619 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
7620 // Uses value zero as the modifier.
7621 //
7622 void MacroAssembler::protect_return_address() {
7623   if (VM_Version::use_rop_protection()) {
7624     check_return_address();
7625     paciaz();
7626   }
7627 }
7628 
7629 // Sign the return value in the given register. Use before updating the LR in the existing stack
7630 // frame for the current function.
7631 // Uses value zero as the modifier.
7632 //
7633 void MacroAssembler::protect_return_address(Register return_reg) {
7634   if (VM_Version::use_rop_protection()) {
7635     check_return_address(return_reg);
7636     paciza(return_reg);
7637   }
7638 }
7639 
7640 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
7641 // Uses value zero as the modifier.
7642 //
7643 void MacroAssembler::authenticate_return_address() {
7644   if (VM_Version::use_rop_protection()) {
7645     autiaz();
7646     check_return_address();
7647   }
7648 }
7649 
7650 // Authenticate the return value in the given register. Use before updating the LR in the existing
7651 // stack frame for the current function.
7652 // Uses value zero as the modifier.
7653 //
7654 void MacroAssembler::authenticate_return_address(Register return_reg) {
7655   if (VM_Version::use_rop_protection()) {
7656     autiza(return_reg);
7657     check_return_address(return_reg);
7658   }
7659 }
7660 
7661 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
7662 // there is no guaranteed way of authenticating the LR.
7663 //
7664 void MacroAssembler::strip_return_address() {
7665   if (VM_Version::use_rop_protection()) {
7666     xpaclri();
7667   }
7668 }
7669 
7670 #ifndef PRODUCT
7671 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
7672 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
7673 // it is difficult to debug back to the callee function.
7674 // This function simply loads from the address in the given register.
7675 // Use directly after authentication to catch authentication failures.
7676 // Also use before signing to check that the pointer is valid and hasn't already been signed.
7677 //
7678 void MacroAssembler::check_return_address(Register return_reg) {
7679   if (VM_Version::use_rop_protection()) {
7680     ldr(zr, Address(return_reg));
7681   }
7682 }
7683 #endif
7684 
7685 // The java_calling_convention describes stack locations as ideal slots on
7686 // a frame with no abi restrictions. Since we must observe abi restrictions
7687 // (like the placement of the register window) the slots must be biased by
7688 // the following value.
7689 static int reg2offset_in(VMReg r) {
7690   // Account for saved rfp and lr
7691   // This should really be in_preserve_stack_slots
7692   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
7693 }
7694 
7695 static int reg2offset_out(VMReg r) {
7696   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
7697 }
7698 
7699 // On 64bit we will store integer like items to the stack as
7700 // 64bits items (AArch64 ABI) even though java would only store
7701 // 32bits for a parameter. On 32bit it will simply be 32bits
7702 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
7703 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
7704   if (src.first()->is_stack()) {
7705     if (dst.first()->is_stack()) {
7706       // stack to stack
7707       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7708       str(tmp, Address(sp, reg2offset_out(dst.first())));
7709     } else {
7710       // stack to reg
7711       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7712     }
7713   } else if (dst.first()->is_stack()) {
7714     // reg to stack
7715     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7716   } else {
7717     if (dst.first() != src.first()) {
7718       sxtw(dst.first()->as_Register(), src.first()->as_Register());
7719     }
7720   }
7721 }
7722 
7723 // An oop arg. Must pass a handle not the oop itself
7724 void MacroAssembler::object_move(
7725                         OopMap* map,
7726                         int oop_handle_offset,
7727                         int framesize_in_slots,
7728                         VMRegPair src,
7729                         VMRegPair dst,
7730                         bool is_receiver,
7731                         int* receiver_offset) {
7732 
7733   // must pass a handle. First figure out the location we use as a handle
7734 
7735   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
7736 
7737   // See if oop is null if it is we need no handle
7738 
7739   if (src.first()->is_stack()) {
7740 
7741     // Oop is already on the stack as an argument
7742     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
7743     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
7744     if (is_receiver) {
7745       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
7746     }
7747 
7748     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7749     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7750     // conditionally move a null
7751     cmp(rscratch1, zr);
7752     csel(rHandle, zr, rHandle, Assembler::EQ);
7753   } else {
7754 
7755     // Oop is in an a register we must store it to the space we reserve
7756     // on the stack for oop_handles and pass a handle if oop is non-null
7757 
7758     const Register rOop = src.first()->as_Register();
7759     int oop_slot;
7760     if (rOop == j_rarg0)
7761       oop_slot = 0;
7762     else if (rOop == j_rarg1)
7763       oop_slot = 1;
7764     else if (rOop == j_rarg2)
7765       oop_slot = 2;
7766     else if (rOop == j_rarg3)
7767       oop_slot = 3;
7768     else if (rOop == j_rarg4)
7769       oop_slot = 4;
7770     else if (rOop == j_rarg5)
7771       oop_slot = 5;
7772     else if (rOop == j_rarg6)
7773       oop_slot = 6;
7774     else {
7775       assert(rOop == j_rarg7, "wrong register");
7776       oop_slot = 7;
7777     }
7778 
7779     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7780     int offset = oop_slot*VMRegImpl::stack_slot_size;
7781 
7782     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7783     // Store oop in handle area, may be null
7784     str(rOop, Address(sp, offset));
7785     if (is_receiver) {
7786       *receiver_offset = offset;
7787     }
7788 
7789     cmp(rOop, zr);
7790     lea(rHandle, Address(sp, offset));
7791     // conditionally move a null
7792     csel(rHandle, zr, rHandle, Assembler::EQ);
7793   }
7794 
7795   // If arg is on the stack then place it otherwise it is already in correct reg.
7796   if (dst.first()->is_stack()) {
7797     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7798   }
7799 }
7800 
7801 // A float arg may have to do float reg int reg conversion
7802 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7803  if (src.first()->is_stack()) {
7804     if (dst.first()->is_stack()) {
7805       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7806       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7807     } else {
7808       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7809     }
7810   } else if (src.first() != dst.first()) {
7811     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7812       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7813     else
7814       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7815   }
7816 }
7817 
7818 // A long move
7819 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7820   if (src.first()->is_stack()) {
7821     if (dst.first()->is_stack()) {
7822       // stack to stack
7823       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7824       str(tmp, Address(sp, reg2offset_out(dst.first())));
7825     } else {
7826       // stack to reg
7827       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7828     }
7829   } else if (dst.first()->is_stack()) {
7830     // reg to stack
7831     // Do we really have to sign extend???
7832     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7833     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7834   } else {
7835     if (dst.first() != src.first()) {
7836       mov(dst.first()->as_Register(), src.first()->as_Register());
7837     }
7838   }
7839 }
7840 
7841 
7842 // A double move
7843 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7844  if (src.first()->is_stack()) {
7845     if (dst.first()->is_stack()) {
7846       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7847       str(tmp, Address(sp, reg2offset_out(dst.first())));
7848     } else {
7849       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7850     }
7851   } else if (src.first() != dst.first()) {
7852     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7853       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7854     else
7855       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7856   }
7857 }
7858 
7859 // Implements lightweight-locking.
7860 //
7861 //  - obj: the object to be locked
7862 //  - t1, t2, t3: temporary registers, will be destroyed
7863 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7864 void MacroAssembler::lightweight_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7865   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7866 
7867   Label push;
7868   const Register top = t1;
7869   const Register mark = t2;
7870   const Register t = t3;
7871 
7872   // Preload the markWord. It is important that this is the first
7873   // instruction emitted as it is part of C1's null check semantics.
7874   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7875 
7876   if (UseObjectMonitorTable) {
7877     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7878     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7879   }
7880 
7881   if (DiagnoseSyncOnValueBasedClasses != 0) {
7882     load_klass(t1, obj);
7883     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7884     tst(t1, KlassFlags::_misc_is_value_based_class);
7885     br(Assembler::NE, slow);
7886   }
7887 
7888   // Check if the lock-stack is full.
7889   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7890   cmpw(top, (unsigned)LockStack::end_offset());
7891   br(Assembler::GE, slow);
7892 
7893   // Check for recursion.
7894   subw(t, top, oopSize);
7895   ldr(t, Address(rthread, t));
7896   cmp(obj, t);
7897   br(Assembler::EQ, push);
7898 
7899   // Check header for monitor (0b10).
7900   tst(mark, markWord::monitor_value);
7901   br(Assembler::NE, slow);
7902 
7903   // Try to lock. Transition lock bits 0b01 => 0b00
7904   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7905   orr(mark, mark, markWord::unlocked_value);
7906   if (EnableValhalla) {
7907     // Mask inline_type bit such that we go to the slow path if object is an inline type
7908     andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7909   }
7910   eor(t, mark, markWord::unlocked_value);
7911   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7912           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7913   br(Assembler::NE, slow);
7914 
7915   bind(push);
7916   // After successful lock, push object on lock-stack.
7917   str(obj, Address(rthread, top));
7918   addw(top, top, oopSize);
7919   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7920 }
7921 
7922 // Implements lightweight-unlocking.
7923 //
7924 // - obj: the object to be unlocked
7925 // - t1, t2, t3: temporary registers
7926 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7927 void MacroAssembler::lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7928   // cmpxchg clobbers rscratch1.
7929   assert_different_registers(obj, t1, t2, t3, rscratch1);
7930 
7931 #ifdef ASSERT
7932   {
7933     // Check for lock-stack underflow.
7934     Label stack_ok;
7935     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7936     cmpw(t1, (unsigned)LockStack::start_offset());
7937     br(Assembler::GE, stack_ok);
7938     STOP("Lock-stack underflow");
7939     bind(stack_ok);
7940   }
7941 #endif
7942 
7943   Label unlocked, push_and_slow;
7944   const Register top = t1;
7945   const Register mark = t2;
7946   const Register t = t3;
7947 
7948   // Check if obj is top of lock-stack.
7949   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7950   subw(top, top, oopSize);
7951   ldr(t, Address(rthread, top));
7952   cmp(obj, t);
7953   br(Assembler::NE, slow);
7954 
7955   // Pop lock-stack.
7956   DEBUG_ONLY(str(zr, Address(rthread, top));)
7957   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7958 
7959   // Check if recursive.
7960   subw(t, top, oopSize);
7961   ldr(t, Address(rthread, t));
7962   cmp(obj, t);
7963   br(Assembler::EQ, unlocked);
7964 
7965   // Not recursive. Check header for monitor (0b10).
7966   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7967   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7968 
7969 #ifdef ASSERT
7970   // Check header not unlocked (0b01).
7971   Label not_unlocked;
7972   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7973   stop("lightweight_unlock already unlocked");
7974   bind(not_unlocked);
7975 #endif
7976 
7977   // Try to unlock. Transition lock bits 0b00 => 0b01
7978   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7979   orr(t, mark, markWord::unlocked_value);
7980   cmpxchg(obj, mark, t, Assembler::xword,
7981           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7982   br(Assembler::EQ, unlocked);
7983 
7984   bind(push_and_slow);
7985   // Restore lock-stack and handle the unlock in runtime.
7986   DEBUG_ONLY(str(obj, Address(rthread, top));)
7987   addw(top, top, oopSize);
7988   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7989   b(slow);
7990 
7991   bind(unlocked);
7992 }