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 "code/compiledIC.hpp"
  30 #include "compiler/compileTask.hpp"
  31 #include "compiler/disassembler.hpp"
  32 #include "compiler/oopMap.hpp"
  33 #include "gc/shared/barrierSet.hpp"
  34 #include "gc/shared/barrierSetAssembler.hpp"
  35 #include "gc/shared/cardTableBarrierSet.hpp"
  36 #include "gc/shared/cardTable.hpp"
  37 #include "gc/shared/collectedHeap.hpp"
  38 #include "gc/shared/tlab_globals.hpp"
  39 #include "interpreter/bytecodeHistogram.hpp"
  40 #include "interpreter/interpreter.hpp"
  41 #include "interpreter/interpreterRuntime.hpp"
  42 #include "jvm.h"
  43 #include "memory/resourceArea.hpp"
  44 #include "memory/universe.hpp"
  45 #include "nativeInst_aarch64.hpp"
  46 #include "oops/accessDecorators.hpp"
  47 #include "oops/compressedKlass.inline.hpp"
  48 #include "oops/compressedOops.inline.hpp"
  49 #include "oops/klass.inline.hpp"
  50 #include "runtime/continuation.hpp"
  51 #include "runtime/icache.hpp"
  52 #include "runtime/interfaceSupport.inline.hpp"
  53 #include "runtime/javaThread.hpp"
  54 #include "runtime/jniHandles.inline.hpp"
  55 #include "runtime/sharedRuntime.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "utilities/globalDefinitions.hpp"
  58 #include "utilities/powerOfTwo.hpp"
  59 #ifdef COMPILER1
  60 #include "c1/c1_LIRAssembler.hpp"
  61 #endif
  62 #ifdef COMPILER2
  63 #include "oops/oop.hpp"
  64 #include "opto/compile.hpp"
  65 #include "opto/node.hpp"
  66 #include "opto/output.hpp"
  67 #endif
  68 
  69 #include <sys/types.h>
  70 
  71 #ifdef PRODUCT
  72 #define BLOCK_COMMENT(str) /* nothing */
  73 #else
  74 #define BLOCK_COMMENT(str) block_comment(str)
  75 #endif
  76 #define STOP(str) stop(str);
  77 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  78 
  79 #ifdef ASSERT
  80 extern "C" void disnm(intptr_t p);
  81 #endif
  82 // Target-dependent relocation processing
  83 //
  84 // Instruction sequences whose target may need to be retrieved or
  85 // patched are distinguished by their leading instruction, sorting
  86 // them into three main instruction groups and related subgroups.
  87 //
  88 // 1) Branch, Exception and System (insn count = 1)
  89 //    1a) Unconditional branch (immediate):
  90 //      b/bl imm19
  91 //    1b) Compare & branch (immediate):
  92 //      cbz/cbnz Rt imm19
  93 //    1c) Test & branch (immediate):
  94 //      tbz/tbnz Rt imm14
  95 //    1d) Conditional branch (immediate):
  96 //      b.cond imm19
  97 //
  98 // 2) Loads and Stores (insn count = 1)
  99 //    2a) Load register literal:
 100 //      ldr Rt imm19
 101 //
 102 // 3) Data Processing Immediate (insn count = 2 or 3)
 103 //    3a) PC-rel. addressing
 104 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 105 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 106 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 107 //      adr/adrp Rx imm21
 108 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 109 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 110 //      The latter form can only happen when the target is an
 111 //      ExternalAddress, and (by definition) ExternalAddresses don't
 112 //      move. Because of that property, there is never any need to
 113 //      patch the last of the three instructions. However,
 114 //      MacroAssembler::target_addr_for_insn takes all three
 115 //      instructions into account and returns the correct address.
 116 //    3b) Move wide (immediate)
 117 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 118 //
 119 // A switch on a subset of the instruction's bits provides an
 120 // efficient dispatch to these subcases.
 121 //
 122 // insn[28:26] -> main group ('x' == don't care)
 123 //   00x -> UNALLOCATED
 124 //   100 -> Data Processing Immediate
 125 //   101 -> Branch, Exception and System
 126 //   x1x -> Loads and Stores
 127 //
 128 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 129 // n.b. in some cases extra bits need to be checked to verify the
 130 // instruction is as expected
 131 //
 132 // 1) ... xx101x Branch, Exception and System
 133 //   1a)  00___x Unconditional branch (immediate)
 134 //   1b)  01___0 Compare & branch (immediate)
 135 //   1c)  01___1 Test & branch (immediate)
 136 //   1d)  10___0 Conditional branch (immediate)
 137 //        other  Should not happen
 138 //
 139 // 2) ... xxx1x0 Loads and Stores
 140 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 141 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 142 //                strictly should be 64 bit non-FP/SIMD i.e.
 143 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 144 //
 145 // 3) ... xx100x Data Processing Immediate
 146 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 147 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 148 //                 strictly should be 64 bit movz #imm16<<0
 149 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 150 //
 151 class RelocActions {
 152 protected:
 153   typedef int (*reloc_insn)(address insn_addr, address &target);
 154 
 155   virtual reloc_insn adrpMem() = 0;
 156   virtual reloc_insn adrpAdd() = 0;
 157   virtual reloc_insn adrpMovk() = 0;
 158 
 159   const address _insn_addr;
 160   const uint32_t _insn;
 161 
 162   static uint32_t insn_at(address insn_addr, int n) {
 163     return ((uint32_t*)insn_addr)[n];
 164   }
 165   uint32_t insn_at(int n) const {
 166     return insn_at(_insn_addr, n);
 167   }
 168 
 169 public:
 170 
 171   RelocActions(address insn_addr) : _insn_addr(insn_addr), _insn(insn_at(insn_addr, 0)) {}
 172   RelocActions(address insn_addr, uint32_t insn)
 173     :  _insn_addr(insn_addr), _insn(insn) {}
 174 
 175   virtual int unconditionalBranch(address insn_addr, address &target) = 0;
 176   virtual int conditionalBranch(address insn_addr, address &target) = 0;
 177   virtual int testAndBranch(address insn_addr, address &target) = 0;
 178   virtual int loadStore(address insn_addr, address &target) = 0;
 179   virtual int adr(address insn_addr, address &target) = 0;
 180   virtual int adrp(address insn_addr, address &target, reloc_insn inner) = 0;
 181   virtual int immediate(address insn_addr, address &target) = 0;
 182   virtual void verify(address insn_addr, address &target) = 0;
 183 
 184   int ALWAYSINLINE run(address insn_addr, address &target) {
 185     int instructions = 1;
 186 
 187     uint32_t dispatch = Instruction_aarch64::extract(_insn, 30, 25);
 188     switch(dispatch) {
 189       case 0b001010:
 190       case 0b001011: {
 191         instructions = unconditionalBranch(insn_addr, target);
 192         break;
 193       }
 194       case 0b101010:   // Conditional branch (immediate)
 195       case 0b011010: { // Compare & branch (immediate)
 196         instructions = conditionalBranch(insn_addr, target);
 197           break;
 198       }
 199       case 0b011011: {
 200         instructions = testAndBranch(insn_addr, target);
 201         break;
 202       }
 203       case 0b001100:
 204       case 0b001110:
 205       case 0b011100:
 206       case 0b011110:
 207       case 0b101100:
 208       case 0b101110:
 209       case 0b111100:
 210       case 0b111110: {
 211         // load/store
 212         if ((Instruction_aarch64::extract(_insn, 29, 24) & 0b111011) == 0b011000) {
 213           // Load register (literal)
 214           instructions = loadStore(insn_addr, target);
 215           break;
 216         } else {
 217           // nothing to do
 218           assert(target == nullptr, "did not expect to relocate target for polling page load");
 219         }
 220         break;
 221       }
 222       case 0b001000:
 223       case 0b011000:
 224       case 0b101000:
 225       case 0b111000: {
 226         // adr/adrp
 227         assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 228         int shift = Instruction_aarch64::extract(_insn, 31, 31);
 229         if (shift) {
 230           uint32_t insn2 = insn_at(1);
 231           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 232               Instruction_aarch64::extract(_insn, 4, 0) ==
 233               Instruction_aarch64::extract(insn2, 9, 5)) {
 234             instructions = adrp(insn_addr, target, adrpMem());
 235           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 236                      Instruction_aarch64::extract(_insn, 4, 0) ==
 237                      Instruction_aarch64::extract(insn2, 4, 0)) {
 238             instructions = adrp(insn_addr, target, adrpAdd());
 239           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 240                      Instruction_aarch64::extract(_insn, 4, 0) ==
 241                      Instruction_aarch64::extract(insn2, 4, 0)) {
 242             instructions = adrp(insn_addr, target, adrpMovk());
 243           } else {
 244             ShouldNotReachHere();
 245           }
 246         } else {
 247           instructions = adr(insn_addr, target);
 248         }
 249         break;
 250       }
 251       case 0b001001:
 252       case 0b011001:
 253       case 0b101001:
 254       case 0b111001: {
 255         instructions = immediate(insn_addr, target);
 256         break;
 257       }
 258       default: {
 259         ShouldNotReachHere();
 260       }
 261     }
 262 
 263     verify(insn_addr, target);
 264     return instructions * NativeInstruction::instruction_size;
 265   }
 266 };
 267 
 268 class Patcher : public RelocActions {
 269   virtual reloc_insn adrpMem() { return &Patcher::adrpMem_impl; }
 270   virtual reloc_insn adrpAdd() { return &Patcher::adrpAdd_impl; }
 271   virtual reloc_insn adrpMovk() { return &Patcher::adrpMovk_impl; }
 272 
 273 public:
 274   Patcher(address insn_addr) : RelocActions(insn_addr) {}
 275 
 276   virtual int unconditionalBranch(address insn_addr, address &target) {
 277     intptr_t offset = (target - insn_addr) >> 2;
 278     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 279     return 1;
 280   }
 281   virtual int conditionalBranch(address insn_addr, address &target) {
 282     intptr_t offset = (target - insn_addr) >> 2;
 283     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 284     return 1;
 285   }
 286   virtual int testAndBranch(address insn_addr, address &target) {
 287     intptr_t offset = (target - insn_addr) >> 2;
 288     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 289     return 1;
 290   }
 291   virtual int loadStore(address insn_addr, address &target) {
 292     intptr_t offset = (target - insn_addr) >> 2;
 293     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 294     return 1;
 295   }
 296   virtual int adr(address insn_addr, address &target) {
 297 #ifdef ASSERT
 298     assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 299 #endif
 300     // PC-rel. addressing
 301     ptrdiff_t offset = target - insn_addr;
 302     int offset_lo = offset & 3;
 303     offset >>= 2;
 304     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 305     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 306     return 1;
 307   }
 308   virtual int adrp(address insn_addr, address &target, reloc_insn inner) {
 309     int instructions = 1;
 310 #ifdef ASSERT
 311     assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 312 #endif
 313     ptrdiff_t offset = target - insn_addr;
 314     instructions = 2;
 315     precond(inner != nullptr);
 316     // Give the inner reloc a chance to modify the target.
 317     address adjusted_target = target;
 318     instructions = (*inner)(insn_addr, adjusted_target);
 319     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 320     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 321     offset = adr_page - pc_page;
 322     int offset_lo = offset & 3;
 323     offset >>= 2;
 324     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 325     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 326     return instructions;
 327   }
 328   static int adrpMem_impl(address insn_addr, address &target) {
 329     uintptr_t dest = (uintptr_t)target;
 330     int offset_lo = dest & 0xfff;
 331     uint32_t insn2 = insn_at(insn_addr, 1);
 332     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 333     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 334     guarantee(((dest >> size) << size) == dest, "misaligned target");
 335     return 2;
 336   }
 337   static int adrpAdd_impl(address insn_addr, address &target) {
 338     uintptr_t dest = (uintptr_t)target;
 339     int offset_lo = dest & 0xfff;
 340     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 341     return 2;
 342   }
 343   static int adrpMovk_impl(address insn_addr, address &target) {
 344     uintptr_t dest = uintptr_t(target);
 345     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 346     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 347     target = address(dest);
 348     return 2;
 349   }
 350   virtual int immediate(address insn_addr, address &target) {
 351     assert(Instruction_aarch64::extract(_insn, 31, 21) == 0b11010010100, "must be");
 352     uint64_t dest = (uint64_t)target;
 353     // Move wide constant
 354     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 355     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 356     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 357     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 358     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 359     return 3;
 360   }
 361   virtual void verify(address insn_addr, address &target) {
 362 #ifdef ASSERT
 363     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 364     if (!(address_is == target)) {
 365       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 366       disnm((intptr_t)insn_addr);
 367       assert(address_is == target, "should be");
 368     }
 369 #endif
 370   }
 371 };
 372 
 373 // If insn1 and insn2 use the same register to form an address, either
 374 // by an offsetted LDR or a simple ADD, return the offset. If the
 375 // second instruction is an LDR, the offset may be scaled.
 376 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 377   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 378       Instruction_aarch64::extract(insn1, 4, 0) ==
 379       Instruction_aarch64::extract(insn2, 9, 5)) {
 380     // Load/store register (unsigned immediate)
 381     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 382     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 383     byte_offset <<= size;
 384     return true;
 385   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 386              Instruction_aarch64::extract(insn1, 4, 0) ==
 387              Instruction_aarch64::extract(insn2, 4, 0)) {
 388     // add (immediate)
 389     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 390     return true;
 391   }
 392   return false;
 393 }
 394 
 395 class AArch64Decoder : public RelocActions {
 396   virtual reloc_insn adrpMem() { return &AArch64Decoder::adrpMem_impl; }
 397   virtual reloc_insn adrpAdd() { return &AArch64Decoder::adrpAdd_impl; }
 398   virtual reloc_insn adrpMovk() { return &AArch64Decoder::adrpMovk_impl; }
 399 
 400 public:
 401   AArch64Decoder(address insn_addr, uint32_t insn) : RelocActions(insn_addr, insn) {}
 402 
 403   virtual int loadStore(address insn_addr, address &target) {
 404     intptr_t offset = Instruction_aarch64::sextract(_insn, 23, 5);
 405     target = insn_addr + (offset << 2);
 406     return 1;
 407   }
 408   virtual int unconditionalBranch(address insn_addr, address &target) {
 409     intptr_t offset = Instruction_aarch64::sextract(_insn, 25, 0);
 410     target = insn_addr + (offset << 2);
 411     return 1;
 412   }
 413   virtual int conditionalBranch(address insn_addr, address &target) {
 414     intptr_t offset = Instruction_aarch64::sextract(_insn, 23, 5);
 415     target = address(((uint64_t)insn_addr + (offset << 2)));
 416     return 1;
 417   }
 418   virtual int testAndBranch(address insn_addr, address &target) {
 419     intptr_t offset = Instruction_aarch64::sextract(_insn, 18, 5);
 420     target = address(((uint64_t)insn_addr + (offset << 2)));
 421     return 1;
 422   }
 423   virtual int adr(address insn_addr, address &target) {
 424     // PC-rel. addressing
 425     intptr_t offset = Instruction_aarch64::extract(_insn, 30, 29);
 426     offset |= Instruction_aarch64::sextract(_insn, 23, 5) << 2;
 427     target = address((uint64_t)insn_addr + offset);
 428     return 1;
 429   }
 430   virtual int adrp(address insn_addr, address &target, reloc_insn inner) {
 431     assert(Instruction_aarch64::extract(_insn, 28, 24) == 0b10000, "must be");
 432     intptr_t offset = Instruction_aarch64::extract(_insn, 30, 29);
 433     offset |= Instruction_aarch64::sextract(_insn, 23, 5) << 2;
 434     int shift = 12;
 435     offset <<= shift;
 436     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 437     target_page &= ((uint64_t)-1) << shift;
 438     uint32_t insn2 = insn_at(1);
 439     target = address(target_page);
 440     precond(inner != nullptr);
 441     (*inner)(insn_addr, target);
 442     return 2;
 443   }
 444   static int adrpMem_impl(address insn_addr, address &target) {
 445     uint32_t insn2 = insn_at(insn_addr, 1);
 446     // Load/store register (unsigned immediate)
 447     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 448     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 449     byte_offset <<= size;
 450     target += byte_offset;
 451     return 2;
 452   }
 453   static int adrpAdd_impl(address insn_addr, address &target) {
 454     uint32_t insn2 = insn_at(insn_addr, 1);
 455     // add (immediate)
 456     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 457     target += byte_offset;
 458     return 2;
 459   }
 460   static int adrpMovk_impl(address insn_addr, address &target) {
 461     uint32_t insn2 = insn_at(insn_addr, 1);
 462     uint64_t dest = uint64_t(target);
 463     dest = (dest & 0xffff0000ffffffff) |
 464       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 465     target = address(dest);
 466 
 467     // We know the destination 4k page. Maybe we have a third
 468     // instruction.
 469     uint32_t insn = insn_at(insn_addr, 0);
 470     uint32_t insn3 = insn_at(insn_addr, 2);
 471     ptrdiff_t byte_offset;
 472     if (offset_for(insn, insn3, byte_offset)) {
 473       target += byte_offset;
 474       return 3;
 475     } else {
 476       return 2;
 477     }
 478   }
 479   virtual int immediate(address insn_addr, address &target) {
 480     uint32_t *insns = (uint32_t *)insn_addr;
 481     assert(Instruction_aarch64::extract(_insn, 31, 21) == 0b11010010100, "must be");
 482     // Move wide constant: movz, movk, movk.  See movptr().
 483     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 484     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 485     target = address(uint64_t(Instruction_aarch64::extract(_insn, 20, 5))
 486                  + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 487                  + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 488     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 489     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 490     return 3;
 491   }
 492   virtual void verify(address insn_addr, address &target) {
 493   }
 494 };
 495 
 496 address MacroAssembler::target_addr_for_insn(address insn_addr, uint32_t insn) {
 497   AArch64Decoder decoder(insn_addr, insn);
 498   address target;
 499   decoder.run(insn_addr, target);
 500   return target;
 501 }
 502 
 503 // Patch any kind of instruction; there may be several instructions.
 504 // Return the total length (in bytes) of the instructions.
 505 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 506   Patcher patcher(insn_addr);
 507   return patcher.run(insn_addr, target);
 508 }
 509 
 510 int MacroAssembler::patch_oop(address insn_addr, address o) {
 511   int instructions;
 512   unsigned insn = *(unsigned*)insn_addr;
 513   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 514 
 515   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 516   // narrow OOPs by setting the upper 16 bits in the first
 517   // instruction.
 518   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 519     // Move narrow OOP
 520     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 521     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 522     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 523     instructions = 2;
 524   } else {
 525     // Move wide OOP
 526     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 527     uintptr_t dest = (uintptr_t)o;
 528     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 529     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 530     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 531     instructions = 3;
 532   }
 533   return instructions * NativeInstruction::instruction_size;
 534 }
 535 
 536 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 537   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 538   // We encode narrow ones by setting the upper 16 bits in the first
 539   // instruction.
 540   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 541   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 542          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 543 
 544   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 545   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 546   return 2 * NativeInstruction::instruction_size;
 547 }
 548 
 549 address MacroAssembler::target_addr_for_insn_or_null(address insn_addr, unsigned insn) {
 550   if (NativeInstruction::is_ldrw_to_zr(address(&insn))) {
 551     return nullptr;
 552   }
 553   return MacroAssembler::target_addr_for_insn(insn_addr, insn);
 554 }
 555 
 556 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 557   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 558   if (at_return) {
 559     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 560     // we may safely use the sp instead to perform the stack watermark check.
 561     cmp(in_nmethod ? sp : rfp, tmp);
 562     br(Assembler::HI, slow_path);
 563   } else {
 564     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 565   }
 566 }
 567 
 568 void MacroAssembler::rt_call(address dest, Register tmp) {
 569   CodeBlob *cb = CodeCache::find_blob(dest);
 570   if (cb) {
 571     far_call(RuntimeAddress(dest));
 572   } else {
 573     lea(tmp, RuntimeAddress(dest));
 574     blr(tmp);
 575   }
 576 }
 577 
 578 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 579   if (!Continuations::enabled()) return;
 580   Label done;
 581   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 582   cmp(sp, rscratch1);
 583   br(Assembler::LS, done);
 584   mov(rscratch1, sp); // we can't use sp as the source in str
 585   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 586   bind(done);
 587 }
 588 
 589 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 590   if (!Continuations::enabled()) return;
 591   Label done;
 592   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 593   cmp(sp, rscratch1);
 594   br(Assembler::LO, done);
 595   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 596   bind(done);
 597 }
 598 
 599 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 600   // we must set sp to zero to clear frame
 601   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 602 
 603   // must clear fp, so that compiled frames are not confused; it is
 604   // possible that we need it only for debugging
 605   if (clear_fp) {
 606     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 607   }
 608 
 609   // Always clear the pc because it could have been set by make_walkable()
 610   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 611 }
 612 
 613 // Calls to C land
 614 //
 615 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 616 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 617 // has to be reset to 0. This is required to allow proper stack traversal.
 618 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 619                                          Register last_java_fp,
 620                                          Register last_java_pc,
 621                                          Register scratch) {
 622 
 623   if (last_java_pc->is_valid()) {
 624       str(last_java_pc, Address(rthread,
 625                                 JavaThread::frame_anchor_offset()
 626                                 + JavaFrameAnchor::last_Java_pc_offset()));
 627     }
 628 
 629   // determine last_java_sp register
 630   if (last_java_sp == sp) {
 631     mov(scratch, sp);
 632     last_java_sp = scratch;
 633   } else if (!last_java_sp->is_valid()) {
 634     last_java_sp = esp;
 635   }
 636 
 637   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 638 
 639   // last_java_fp is optional
 640   if (last_java_fp->is_valid()) {
 641     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 642   }
 643 }
 644 
 645 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 646                                          Register last_java_fp,
 647                                          address  last_java_pc,
 648                                          Register scratch) {
 649   assert(last_java_pc != nullptr, "must provide a valid PC");
 650 
 651   adr(scratch, last_java_pc);
 652   str(scratch, Address(rthread,
 653                        JavaThread::frame_anchor_offset()
 654                        + JavaFrameAnchor::last_Java_pc_offset()));
 655 
 656   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 657 }
 658 
 659 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 660                                          Register last_java_fp,
 661                                          Label &L,
 662                                          Register scratch) {
 663   if (L.is_bound()) {
 664     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 665   } else {
 666     InstructionMark im(this);
 667     L.add_patch_at(code(), locator());
 668     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 669   }
 670 }
 671 
 672 static inline bool target_needs_far_branch(address addr) {
 673   if (AOTCodeCache::is_on_for_dump()) {
 674     return true;
 675   }
 676   // codecache size <= 128M
 677   if (!MacroAssembler::far_branches()) {
 678     return false;
 679   }
 680   // codecache size > 240M
 681   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 682     return true;
 683   }
 684   // codecache size: 128M..240M
 685   return !CodeCache::is_non_nmethod(addr);
 686 }
 687 
 688 void MacroAssembler::far_call(Address entry, Register tmp) {
 689   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 690   assert(CodeCache::find_blob(entry.target()) != nullptr,
 691          "destination of far call not found in code cache");
 692   assert(entry.rspec().type() == relocInfo::external_word_type
 693          || entry.rspec().type() == relocInfo::runtime_call_type
 694          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 695   if (target_needs_far_branch(entry.target())) {
 696     uint64_t offset;
 697     // We can use ADRP here because we know that the total size of
 698     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 699     adrp(tmp, entry, offset);
 700     add(tmp, tmp, offset);
 701     blr(tmp);
 702   } else {
 703     bl(entry);
 704   }
 705 }
 706 
 707 int MacroAssembler::far_jump(Address entry, Register tmp) {
 708   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 709   assert(CodeCache::find_blob(entry.target()) != nullptr,
 710          "destination of far call not found in code cache");
 711   assert(entry.rspec().type() == relocInfo::external_word_type
 712          || entry.rspec().type() == relocInfo::runtime_call_type
 713          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 714   address start = pc();
 715   if (target_needs_far_branch(entry.target())) {
 716     uint64_t offset;
 717     // We can use ADRP here because we know that the total size of
 718     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 719     adrp(tmp, entry, offset);
 720     add(tmp, tmp, offset);
 721     br(tmp);
 722   } else {
 723     b(entry);
 724   }
 725   return pc() - start;
 726 }
 727 
 728 void MacroAssembler::reserved_stack_check() {
 729     // testing if reserved zone needs to be enabled
 730     Label no_reserved_zone_enabling;
 731 
 732     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 733     cmp(sp, rscratch1);
 734     br(Assembler::LO, no_reserved_zone_enabling);
 735 
 736     enter();   // LR and FP are live.
 737     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 738     mov(c_rarg0, rthread);
 739     blr(rscratch1);
 740     leave();
 741 
 742     // We have already removed our own frame.
 743     // throw_delayed_StackOverflowError will think that it's been
 744     // called by our caller.
 745     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 746     br(rscratch1);
 747     should_not_reach_here();
 748 
 749     bind(no_reserved_zone_enabling);
 750 }
 751 
 752 static void pass_arg0(MacroAssembler* masm, Register arg) {
 753   if (c_rarg0 != arg ) {
 754     masm->mov(c_rarg0, arg);
 755   }
 756 }
 757 
 758 static void pass_arg1(MacroAssembler* masm, Register arg) {
 759   if (c_rarg1 != arg ) {
 760     masm->mov(c_rarg1, arg);
 761   }
 762 }
 763 
 764 static void pass_arg2(MacroAssembler* masm, Register arg) {
 765   if (c_rarg2 != arg ) {
 766     masm->mov(c_rarg2, arg);
 767   }
 768 }
 769 
 770 static void pass_arg3(MacroAssembler* masm, Register arg) {
 771   if (c_rarg3 != arg ) {
 772     masm->mov(c_rarg3, arg);
 773   }
 774 }
 775 
 776 static bool is_preemptable(address entry_point) {
 777   return entry_point == CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter);
 778 }
 779 
 780 void MacroAssembler::call_VM_base(Register oop_result,
 781                                   Register java_thread,
 782                                   Register last_java_sp,
 783                                   address  entry_point,
 784                                   int      number_of_arguments,
 785                                   bool     check_exceptions) {
 786    // determine java_thread register
 787   if (!java_thread->is_valid()) {
 788     java_thread = rthread;
 789   }
 790 
 791   // determine last_java_sp register
 792   if (!last_java_sp->is_valid()) {
 793     last_java_sp = esp;
 794   }
 795 
 796   // debugging support
 797   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 798   assert(java_thread == rthread, "unexpected register");
 799 #ifdef ASSERT
 800   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 801   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 802 #endif // ASSERT
 803 
 804   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 805   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 806 
 807   // push java thread (becomes first argument of C function)
 808 
 809   mov(c_rarg0, java_thread);
 810 
 811   // set last Java frame before call
 812   assert(last_java_sp != rfp, "can't use rfp");
 813 
 814   Label l;
 815   if (is_preemptable(entry_point)) {
 816     // skip setting last_pc since we already set it to desired value.
 817     set_last_Java_frame(last_java_sp, rfp, noreg, rscratch1);
 818   } else {
 819     set_last_Java_frame(last_java_sp, rfp, l, rscratch1);
 820   }
 821 
 822   // do the call, remove parameters
 823   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 824 
 825   // lr could be poisoned with PAC signature during throw_pending_exception
 826   // if it was tail-call optimized by compiler, since lr is not callee-saved
 827   // reload it with proper value
 828   adr(lr, l);
 829 
 830   // reset last Java frame
 831   // Only interpreter should have to clear fp
 832   reset_last_Java_frame(true);
 833 
 834    // C++ interp handles this in the interpreter
 835   check_and_handle_popframe(java_thread);
 836   check_and_handle_earlyret(java_thread);
 837 
 838   if (check_exceptions) {
 839     // check for pending exceptions (java_thread is set upon return)
 840     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 841     Label ok;
 842     cbz(rscratch1, ok);
 843     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 844     br(rscratch1);
 845     bind(ok);
 846   }
 847 
 848   // get oop result if there is one and reset the value in the thread
 849   if (oop_result->is_valid()) {
 850     get_vm_result_oop(oop_result, java_thread);
 851   }
 852 }
 853 
 854 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 855   call_VM_base(oop_result, noreg, noreg, entry_point, number_of_arguments, check_exceptions);
 856 }
 857 
 858 // Check the entry target is always reachable from any branch.
 859 static bool is_always_within_branch_range(Address entry) {
 860   if (AOTCodeCache::is_on_for_dump()) {
 861     return false;
 862   }
 863   const address target = entry.target();
 864 
 865   if (!CodeCache::contains(target)) {
 866     // We always use trampolines for callees outside CodeCache.
 867     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 868     return false;
 869   }
 870 
 871   if (!MacroAssembler::far_branches()) {
 872     return true;
 873   }
 874 
 875   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 876     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 877     // Non-compiled methods stay forever in CodeCache.
 878     // We check whether the longest possible branch is within the branch range.
 879     assert(CodeCache::find_blob(target) != nullptr &&
 880           !CodeCache::find_blob(target)->is_nmethod(),
 881           "runtime call of compiled method");
 882     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 883     const address left_longest_branch_start = CodeCache::low_bound();
 884     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 885                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 886     return is_reachable;
 887   }
 888 
 889   return false;
 890 }
 891 
 892 // Maybe emit a call via a trampoline. If the code cache is small
 893 // trampolines won't be emitted.
 894 address MacroAssembler::trampoline_call(Address entry) {
 895   assert(entry.rspec().type() == relocInfo::runtime_call_type
 896          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 897          || entry.rspec().type() == relocInfo::static_call_type
 898          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 899 
 900   address target = entry.target();
 901 
 902   if (!is_always_within_branch_range(entry)) {
 903     if (!in_scratch_emit_size()) {
 904       // We don't want to emit a trampoline if C2 is generating dummy
 905       // code during its branch shortening phase.
 906       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 907         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 908         code()->share_trampoline_for(entry.target(), offset());
 909       } else {
 910         address stub = emit_trampoline_stub(offset(), target);
 911         if (stub == nullptr) {
 912           postcond(pc() == badAddress);
 913           return nullptr; // CodeCache is full
 914         }
 915       }
 916     }
 917     target = pc();
 918   }
 919 
 920   address call_pc = pc();
 921   relocate(entry.rspec());
 922   bl(target);
 923 
 924   postcond(pc() != badAddress);
 925   return call_pc;
 926 }
 927 
 928 // Emit a trampoline stub for a call to a target which is too far away.
 929 //
 930 // code sequences:
 931 //
 932 // call-site:
 933 //   branch-and-link to <destination> or <trampoline stub>
 934 //
 935 // Related trampoline stub for this call site in the stub section:
 936 //   load the call target from the constant pool
 937 //   branch (LR still points to the call site above)
 938 
 939 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 940                                              address dest) {
 941   // Max stub size: alignment nop, TrampolineStub.
 942   address stub = start_a_stub(max_trampoline_stub_size());
 943   if (stub == nullptr) {
 944     return nullptr;  // CodeBuffer::expand failed
 945   }
 946 
 947   // Create a trampoline stub relocation which relates this trampoline stub
 948   // with the call instruction at insts_call_instruction_offset in the
 949   // instructions code-section.
 950   align(wordSize);
 951   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 952                                             + insts_call_instruction_offset));
 953   const int stub_start_offset = offset();
 954 
 955   // Now, create the trampoline stub's code:
 956   // - load the call
 957   // - call
 958   Label target;
 959   ldr(rscratch1, target);
 960   br(rscratch1);
 961   bind(target);
 962   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 963          "should be");
 964   emit_int64((int64_t)dest);
 965 
 966   const address stub_start_addr = addr_at(stub_start_offset);
 967 
 968   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 969 
 970   end_a_stub();
 971   return stub_start_addr;
 972 }
 973 
 974 int MacroAssembler::max_trampoline_stub_size() {
 975   // Max stub size: alignment nop, TrampolineStub.
 976   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 977 }
 978 
 979 void MacroAssembler::emit_static_call_stub() {
 980   // CompiledDirectCall::set_to_interpreted knows the
 981   // exact layout of this stub.
 982 
 983   isb();
 984   mov_metadata(rmethod, nullptr);
 985 
 986   // Jump to the entry point of the c2i stub.
 987   if (codestub_branch_needs_far_jump()) {
 988     movptr(rscratch1, 0);
 989     br(rscratch1);
 990   } else {
 991     b(pc());
 992   }
 993 }
 994 
 995 int MacroAssembler::static_call_stub_size() {
 996   if (!codestub_branch_needs_far_jump()) {
 997     // isb; movk; movz; movz; b
 998     return 5 * NativeInstruction::instruction_size;
 999   }
1000   // isb; movk; movz; movz; movk; movz; movz; br
1001   return 8 * NativeInstruction::instruction_size;
1002 }
1003 
1004 void MacroAssembler::c2bool(Register x) {
1005   // implements x == 0 ? 0 : 1
1006   // note: must only look at least-significant byte of x
1007   //       since C-style booleans are stored in one byte
1008   //       only! (was bug)
1009   tst(x, 0xff);
1010   cset(x, Assembler::NE);
1011 }
1012 
1013 address MacroAssembler::ic_call(address entry, jint method_index) {
1014   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
1015   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
1016   return trampoline_call(Address(entry, rh));
1017 }
1018 
1019 int MacroAssembler::ic_check_size() {
1020   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
1021   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
1022     return NativeInstruction::instruction_size * (7 + extra_instructions);
1023   } else {
1024     return NativeInstruction::instruction_size * (5 + extra_instructions);
1025   }
1026 }
1027 
1028 int MacroAssembler::ic_check(int end_alignment) {
1029   Register receiver = j_rarg0;
1030   Register data = rscratch2;
1031   Register tmp1 = rscratch1;
1032   Register tmp2 = r10;
1033 
1034   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
1035   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
1036   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1037   // before the inline cache check here, and not after
1038   align(end_alignment, offset() + ic_check_size());
1039 
1040   int uep_offset = offset();
1041 
1042   if (UseCompactObjectHeaders) {
1043     load_narrow_klass_compact(tmp1, receiver);
1044     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1045     cmpw(tmp1, tmp2);
1046   } else if (UseCompressedClassPointers) {
1047     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1048     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1049     cmpw(tmp1, tmp2);
1050   } else {
1051     ldr(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1052     ldr(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1053     cmp(tmp1, tmp2);
1054   }
1055 
1056   Label dont;
1057   br(Assembler::EQ, dont);
1058   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1059   bind(dont);
1060   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1061 
1062   return uep_offset;
1063 }
1064 
1065 // Implementation of call_VM versions
1066 
1067 void MacroAssembler::call_VM(Register oop_result,
1068                              address entry_point,
1069                              bool check_exceptions) {
1070   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1071 }
1072 
1073 void MacroAssembler::call_VM(Register oop_result,
1074                              address entry_point,
1075                              Register arg_1,
1076                              bool check_exceptions) {
1077   pass_arg1(this, arg_1);
1078   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1079 }
1080 
1081 void MacroAssembler::call_VM(Register oop_result,
1082                              address entry_point,
1083                              Register arg_1,
1084                              Register arg_2,
1085                              bool check_exceptions) {
1086   assert_different_registers(arg_1, c_rarg2);
1087   pass_arg2(this, arg_2);
1088   pass_arg1(this, arg_1);
1089   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1090 }
1091 
1092 void MacroAssembler::call_VM(Register oop_result,
1093                              address entry_point,
1094                              Register arg_1,
1095                              Register arg_2,
1096                              Register arg_3,
1097                              bool check_exceptions) {
1098   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1099   assert_different_registers(arg_2, c_rarg3);
1100   pass_arg3(this, arg_3);
1101 
1102   pass_arg2(this, arg_2);
1103 
1104   pass_arg1(this, arg_1);
1105   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1106 }
1107 
1108 void MacroAssembler::call_VM(Register oop_result,
1109                              Register last_java_sp,
1110                              address entry_point,
1111                              int number_of_arguments,
1112                              bool check_exceptions) {
1113   call_VM_base(oop_result, rthread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1114 }
1115 
1116 void MacroAssembler::call_VM(Register oop_result,
1117                              Register last_java_sp,
1118                              address entry_point,
1119                              Register arg_1,
1120                              bool check_exceptions) {
1121   pass_arg1(this, arg_1);
1122   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1123 }
1124 
1125 void MacroAssembler::call_VM(Register oop_result,
1126                              Register last_java_sp,
1127                              address entry_point,
1128                              Register arg_1,
1129                              Register arg_2,
1130                              bool check_exceptions) {
1131 
1132   assert_different_registers(arg_1, c_rarg2);
1133   pass_arg2(this, arg_2);
1134   pass_arg1(this, arg_1);
1135   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1136 }
1137 
1138 void MacroAssembler::call_VM(Register oop_result,
1139                              Register last_java_sp,
1140                              address entry_point,
1141                              Register arg_1,
1142                              Register arg_2,
1143                              Register arg_3,
1144                              bool check_exceptions) {
1145   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1146   assert_different_registers(arg_2, c_rarg3);
1147   pass_arg3(this, arg_3);
1148   pass_arg2(this, arg_2);
1149   pass_arg1(this, arg_1);
1150   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1151 }
1152 
1153 
1154 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1155   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1156   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1157   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1158 }
1159 
1160 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1161   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1162   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1163 }
1164 
1165 void MacroAssembler::align(int modulus) {
1166   align(modulus, offset());
1167 }
1168 
1169 // Ensure that the code at target bytes offset from the current offset() is aligned
1170 // according to modulus.
1171 void MacroAssembler::align(int modulus, int target) {
1172   int delta = target - offset();
1173   while ((offset() + delta) % modulus != 0) nop();
1174 }
1175 
1176 void MacroAssembler::post_call_nop() {
1177   if (!Continuations::enabled()) {
1178     return;
1179   }
1180   InstructionMark im(this);
1181   relocate(post_call_nop_Relocation::spec());
1182   InlineSkippedInstructionsCounter skipCounter(this);
1183   nop();
1184   movk(zr, 0);
1185   movk(zr, 0);
1186 }
1187 
1188 // these are no-ops overridden by InterpreterMacroAssembler
1189 
1190 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1191 
1192 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1193 
1194 // Look up the method for a megamorphic invokeinterface call.
1195 // The target method is determined by <intf_klass, itable_index>.
1196 // The receiver klass is in recv_klass.
1197 // On success, the result will be in method_result, and execution falls through.
1198 // On failure, execution transfers to the given label.
1199 void MacroAssembler::lookup_interface_method(Register recv_klass,
1200                                              Register intf_klass,
1201                                              RegisterOrConstant itable_index,
1202                                              Register method_result,
1203                                              Register scan_temp,
1204                                              Label& L_no_such_interface,
1205                          bool return_method) {
1206   assert_different_registers(recv_klass, intf_klass, scan_temp);
1207   assert_different_registers(method_result, intf_klass, scan_temp);
1208   assert(recv_klass != method_result || !return_method,
1209      "recv_klass can be destroyed when method isn't needed");
1210   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1211          "caller must use same register for non-constant itable index as for method");
1212 
1213   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1214   int vtable_base = in_bytes(Klass::vtable_start_offset());
1215   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1216   int scan_step   = itableOffsetEntry::size() * wordSize;
1217   int vte_size    = vtableEntry::size_in_bytes();
1218   assert(vte_size == wordSize, "else adjust times_vte_scale");
1219 
1220   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1221 
1222   // Could store the aligned, prescaled offset in the klass.
1223   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1224   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1225   add(scan_temp, scan_temp, vtable_base);
1226 
1227   if (return_method) {
1228     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1229     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1230     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1231     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1232     if (itentry_off)
1233       add(recv_klass, recv_klass, itentry_off);
1234   }
1235 
1236   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1237   //   if (scan->interface() == intf) {
1238   //     result = (klass + scan->offset() + itable_index);
1239   //   }
1240   // }
1241   Label search, found_method;
1242 
1243   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1244   cmp(intf_klass, method_result);
1245   br(Assembler::EQ, found_method);
1246   bind(search);
1247   // Check that the previous entry is non-null.  A null entry means that
1248   // the receiver class doesn't implement the interface, and wasn't the
1249   // same as when the caller was compiled.
1250   cbz(method_result, L_no_such_interface);
1251   if (itableOffsetEntry::interface_offset() != 0) {
1252     add(scan_temp, scan_temp, scan_step);
1253     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1254   } else {
1255     ldr(method_result, Address(pre(scan_temp, scan_step)));
1256   }
1257   cmp(intf_klass, method_result);
1258   br(Assembler::NE, search);
1259 
1260   bind(found_method);
1261 
1262   // Got a hit.
1263   if (return_method) {
1264     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1265     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1266   }
1267 }
1268 
1269 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1270 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1271 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1272 // The target method is determined by <holder_klass, itable_index>.
1273 // The receiver klass is in recv_klass.
1274 // On success, the result will be in method_result, and execution falls through.
1275 // On failure, execution transfers to the given label.
1276 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1277                                                   Register holder_klass,
1278                                                   Register resolved_klass,
1279                                                   Register method_result,
1280                                                   Register temp_itbl_klass,
1281                                                   Register scan_temp,
1282                                                   int itable_index,
1283                                                   Label& L_no_such_interface) {
1284   // 'method_result' is only used as output register at the very end of this method.
1285   // Until then we can reuse it as 'holder_offset'.
1286   Register holder_offset = method_result;
1287   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1288 
1289   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1290   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1291   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1292   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1293 
1294   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1295 
1296   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1297   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1298   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1299   // temp_itbl_klass = itable[0]._interface;
1300   int vtblEntrySize = vtableEntry::size_in_bytes();
1301   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1302   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1303   mov(holder_offset, zr);
1304   // scan_temp = &(itable[0]._interface)
1305   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1306 
1307   // Initial checks:
1308   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1309   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1310   //   - if (itable[0] == 0), no such interface
1311   cmp(resolved_klass, holder_klass);
1312   br(Assembler::NE, L_loop_search_resolved_entry);
1313   cmp(holder_klass, temp_itbl_klass);
1314   br(Assembler::EQ, L_holder_found);
1315   cbz(temp_itbl_klass, L_no_such_interface);
1316 
1317   // Loop: Look for holder_klass record in itable
1318   //   do {
1319   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1320   //     if (temp_itbl_klass == holder_klass) {
1321   //       goto L_holder_found; // Found!
1322   //     }
1323   //   } while (temp_itbl_klass != 0);
1324   //   goto L_no_such_interface // Not found.
1325   Label L_search_holder;
1326   bind(L_search_holder);
1327     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1328     cmp(holder_klass, temp_itbl_klass);
1329     br(Assembler::EQ, L_holder_found);
1330     cbnz(temp_itbl_klass, L_search_holder);
1331 
1332   b(L_no_such_interface);
1333 
1334   // Loop: Look for resolved_class record in itable
1335   //   while (true) {
1336   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1337   //     if (temp_itbl_klass == 0) {
1338   //       goto L_no_such_interface;
1339   //     }
1340   //     if (temp_itbl_klass == resolved_klass) {
1341   //        goto L_resolved_found;  // Found!
1342   //     }
1343   //     if (temp_itbl_klass == holder_klass) {
1344   //        holder_offset = scan_temp;
1345   //     }
1346   //   }
1347   //
1348   Label L_loop_search_resolved;
1349   bind(L_loop_search_resolved);
1350     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1351   bind(L_loop_search_resolved_entry);
1352     cbz(temp_itbl_klass, L_no_such_interface);
1353     cmp(resolved_klass, temp_itbl_klass);
1354     br(Assembler::EQ, L_resolved_found);
1355     cmp(holder_klass, temp_itbl_klass);
1356     br(Assembler::NE, L_loop_search_resolved);
1357     mov(holder_offset, scan_temp);
1358     b(L_loop_search_resolved);
1359 
1360   // See if we already have a holder klass. If not, go and scan for it.
1361   bind(L_resolved_found);
1362   cbz(holder_offset, L_search_holder);
1363   mov(scan_temp, holder_offset);
1364 
1365   // Finally, scan_temp contains holder_klass vtable offset
1366   bind(L_holder_found);
1367   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1368   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1369     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1370   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1371 }
1372 
1373 // virtual method calling
1374 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1375                                            RegisterOrConstant vtable_index,
1376                                            Register method_result) {
1377   assert(vtableEntry::size() * wordSize == 8,
1378          "adjust the scaling in the code below");
1379   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1380 
1381   if (vtable_index.is_register()) {
1382     lea(method_result, Address(recv_klass,
1383                                vtable_index.as_register(),
1384                                Address::lsl(LogBytesPerWord)));
1385     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1386   } else {
1387     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1388     ldr(method_result,
1389         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1390   }
1391 }
1392 
1393 void MacroAssembler::check_klass_subtype(Register sub_klass,
1394                            Register super_klass,
1395                            Register temp_reg,
1396                            Label& L_success) {
1397   Label L_failure;
1398   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1399   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1400   bind(L_failure);
1401 }
1402 
1403 
1404 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1405                                                    Register super_klass,
1406                                                    Register temp_reg,
1407                                                    Label* L_success,
1408                                                    Label* L_failure,
1409                                                    Label* L_slow_path,
1410                                                    Register super_check_offset) {
1411   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1412   bool must_load_sco = ! super_check_offset->is_valid();
1413   if (must_load_sco) {
1414     assert(temp_reg != noreg, "supply either a temp or a register offset");
1415   }
1416 
1417   Label L_fallthrough;
1418   int label_nulls = 0;
1419   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1420   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1421   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1422   assert(label_nulls <= 1, "at most one null in the batch");
1423 
1424   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1425   Address super_check_offset_addr(super_klass, sco_offset);
1426 
1427   // Hacked jmp, which may only be used just before L_fallthrough.
1428 #define final_jmp(label)                                                \
1429   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1430   else                            b(label)                /*omit semi*/
1431 
1432   // If the pointers are equal, we are done (e.g., String[] elements).
1433   // This self-check enables sharing of secondary supertype arrays among
1434   // non-primary types such as array-of-interface.  Otherwise, each such
1435   // type would need its own customized SSA.
1436   // We move this check to the front of the fast path because many
1437   // type checks are in fact trivially successful in this manner,
1438   // so we get a nicely predicted branch right at the start of the check.
1439   cmp(sub_klass, super_klass);
1440   br(Assembler::EQ, *L_success);
1441 
1442   // Check the supertype display:
1443   if (must_load_sco) {
1444     ldrw(temp_reg, super_check_offset_addr);
1445     super_check_offset = temp_reg;
1446   }
1447 
1448   Address super_check_addr(sub_klass, super_check_offset);
1449   ldr(rscratch1, super_check_addr);
1450   cmp(super_klass, rscratch1); // load displayed supertype
1451   br(Assembler::EQ, *L_success);
1452 
1453   // This check has worked decisively for primary supers.
1454   // Secondary supers are sought in the super_cache ('super_cache_addr').
1455   // (Secondary supers are interfaces and very deeply nested subtypes.)
1456   // This works in the same check above because of a tricky aliasing
1457   // between the super_cache and the primary super display elements.
1458   // (The 'super_check_addr' can address either, as the case requires.)
1459   // Note that the cache is updated below if it does not help us find
1460   // what we need immediately.
1461   // So if it was a primary super, we can just fail immediately.
1462   // Otherwise, it's the slow path for us (no success at this point).
1463 
1464   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1465   if (L_failure == &L_fallthrough) {
1466     cbz(rscratch1, *L_slow_path);
1467   } else {
1468     cbnz(rscratch1, *L_failure);
1469     final_jmp(*L_slow_path);
1470   }
1471 
1472   bind(L_fallthrough);
1473 
1474 #undef final_jmp
1475 }
1476 
1477 // These two are taken from x86, but they look generally useful
1478 
1479 // scans count pointer sized words at [addr] for occurrence of value,
1480 // generic
1481 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1482                                 Register scratch) {
1483   Label Lloop, Lexit;
1484   cbz(count, Lexit);
1485   bind(Lloop);
1486   ldr(scratch, post(addr, wordSize));
1487   cmp(value, scratch);
1488   br(EQ, Lexit);
1489   sub(count, count, 1);
1490   cbnz(count, Lloop);
1491   bind(Lexit);
1492 }
1493 
1494 // scans count 4 byte words at [addr] for occurrence of value,
1495 // generic
1496 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1497                                 Register scratch) {
1498   Label Lloop, Lexit;
1499   cbz(count, Lexit);
1500   bind(Lloop);
1501   ldrw(scratch, post(addr, wordSize));
1502   cmpw(value, scratch);
1503   br(EQ, Lexit);
1504   sub(count, count, 1);
1505   cbnz(count, Lloop);
1506   bind(Lexit);
1507 }
1508 
1509 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1510                                                           Register super_klass,
1511                                                           Register temp_reg,
1512                                                           Register temp2_reg,
1513                                                           Label* L_success,
1514                                                           Label* L_failure,
1515                                                           bool set_cond_codes) {
1516   // NB! Callers may assume that, when temp2_reg is a valid register,
1517   // this code sets it to a nonzero value.
1518 
1519   assert_different_registers(sub_klass, super_klass, temp_reg);
1520   if (temp2_reg != noreg)
1521     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1522 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1523 
1524   Label L_fallthrough;
1525   int label_nulls = 0;
1526   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1527   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1528   assert(label_nulls <= 1, "at most one null in the batch");
1529 
1530   // a couple of useful fields in sub_klass:
1531   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1532   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1533   Address secondary_supers_addr(sub_klass, ss_offset);
1534   Address super_cache_addr(     sub_klass, sc_offset);
1535 
1536   BLOCK_COMMENT("check_klass_subtype_slow_path");
1537 
1538   // Do a linear scan of the secondary super-klass chain.
1539   // This code is rarely used, so simplicity is a virtue here.
1540   // The repne_scan instruction uses fixed registers, which we must spill.
1541   // Don't worry too much about pre-existing connections with the input regs.
1542 
1543   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1544   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1545 
1546   RegSet pushed_registers;
1547   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1548   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1549 
1550   if (super_klass != r0) {
1551     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1552   }
1553 
1554   push(pushed_registers, sp);
1555 
1556   // Get super_klass value into r0 (even if it was in r5 or r2).
1557   if (super_klass != r0) {
1558     mov(r0, super_klass);
1559   }
1560 
1561 #ifndef PRODUCT
1562   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1563 #endif //PRODUCT
1564 
1565   // We will consult the secondary-super array.
1566   ldr(r5, secondary_supers_addr);
1567   // Load the array length.
1568   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1569   // Skip to start of data.
1570   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1571 
1572   cmp(sp, zr); // Clear Z flag; SP is never zero
1573   // Scan R2 words at [R5] for an occurrence of R0.
1574   // Set NZ/Z based on last compare.
1575   repne_scan(r5, r0, r2, rscratch1);
1576 
1577   // Unspill the temp. registers:
1578   pop(pushed_registers, sp);
1579 
1580   br(Assembler::NE, *L_failure);
1581 
1582   // Success.  Cache the super we found and proceed in triumph.
1583 
1584   if (UseSecondarySupersCache) {
1585     str(super_klass, super_cache_addr);
1586   }
1587 
1588   if (L_success != &L_fallthrough) {
1589     b(*L_success);
1590   }
1591 
1592 #undef IS_A_TEMP
1593 
1594   bind(L_fallthrough);
1595 }
1596 
1597 // If Register r is invalid, remove a new register from
1598 // available_regs, and add new register to regs_to_push.
1599 Register MacroAssembler::allocate_if_noreg(Register r,
1600                                   RegSetIterator<Register> &available_regs,
1601                                   RegSet &regs_to_push) {
1602   if (!r->is_valid()) {
1603     r = *available_regs++;
1604     regs_to_push += r;
1605   }
1606   return r;
1607 }
1608 
1609 // check_klass_subtype_slow_path_table() looks for super_klass in the
1610 // hash table belonging to super_klass, branching to L_success or
1611 // L_failure as appropriate. This is essentially a shim which
1612 // allocates registers as necessary then calls
1613 // lookup_secondary_supers_table() to do the work. Any of the temp
1614 // regs may be noreg, in which case this logic will chooses some
1615 // registers push and pop them from the stack.
1616 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1617                                                          Register super_klass,
1618                                                          Register temp_reg,
1619                                                          Register temp2_reg,
1620                                                          Register temp3_reg,
1621                                                          Register result_reg,
1622                                                          FloatRegister vtemp,
1623                                                          Label* L_success,
1624                                                          Label* L_failure,
1625                                                          bool set_cond_codes) {
1626   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1627 
1628   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1629 
1630   Label L_fallthrough;
1631   int label_nulls = 0;
1632   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1633   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1634   assert(label_nulls <= 1, "at most one null in the batch");
1635 
1636   BLOCK_COMMENT("check_klass_subtype_slow_path");
1637 
1638   RegSetIterator<Register> available_regs
1639     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1640 
1641   RegSet pushed_regs;
1642 
1643   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1644   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1645   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1646   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1647 
1648   push(pushed_regs, sp);
1649 
1650   lookup_secondary_supers_table_var(sub_klass,
1651                                     super_klass,
1652                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1653                                     nullptr);
1654   cmp(result_reg, zr);
1655 
1656   // Unspill the temp. registers:
1657   pop(pushed_regs, sp);
1658 
1659   // NB! Callers may assume that, when set_cond_codes is true, this
1660   // code sets temp2_reg to a nonzero value.
1661   if (set_cond_codes) {
1662     mov(temp2_reg, 1);
1663   }
1664 
1665   br(Assembler::NE, *L_failure);
1666 
1667   if (L_success != &L_fallthrough) {
1668     b(*L_success);
1669   }
1670 
1671   bind(L_fallthrough);
1672 }
1673 
1674 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1675                                                    Register super_klass,
1676                                                    Register temp_reg,
1677                                                    Register temp2_reg,
1678                                                    Label* L_success,
1679                                                    Label* L_failure,
1680                                                    bool set_cond_codes) {
1681   if (UseSecondarySupersTable) {
1682     check_klass_subtype_slow_path_table
1683       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1684        /*vtemp*/fnoreg,
1685        L_success, L_failure, set_cond_codes);
1686   } else {
1687     check_klass_subtype_slow_path_linear
1688       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1689   }
1690 }
1691 
1692 
1693 // Ensure that the inline code and the stub are using the same registers.
1694 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1695 do {                                                               \
1696   assert(r_super_klass  == r0                                   && \
1697          r_array_base   == r1                                   && \
1698          r_array_length == r2                                   && \
1699          (r_array_index == r3        || r_array_index == noreg) && \
1700          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1701          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1702          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1703 } while(0)
1704 
1705 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1706                                                          Register r_super_klass,
1707                                                          Register temp1,
1708                                                          Register temp2,
1709                                                          Register temp3,
1710                                                          FloatRegister vtemp,
1711                                                          Register result,
1712                                                          u1 super_klass_slot,
1713                                                          bool stub_is_near) {
1714   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1715 
1716   Label L_fallthrough;
1717 
1718   BLOCK_COMMENT("lookup_secondary_supers_table {");
1719 
1720   const Register
1721     r_array_base   = temp1, // r1
1722     r_array_length = temp2, // r2
1723     r_array_index  = temp3, // r3
1724     r_bitmap       = rscratch2;
1725 
1726   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1727 
1728   u1 bit = super_klass_slot;
1729 
1730   // Make sure that result is nonzero if the TBZ below misses.
1731   mov(result, 1);
1732 
1733   // We're going to need the bitmap in a vector reg and in a core reg,
1734   // so load both now.
1735   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1736   if (bit != 0) {
1737     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1738   }
1739   // First check the bitmap to see if super_klass might be present. If
1740   // the bit is zero, we are certain that super_klass is not one of
1741   // the secondary supers.
1742   tbz(r_bitmap, bit, L_fallthrough);
1743 
1744   // Get the first array index that can contain super_klass into r_array_index.
1745   if (bit != 0) {
1746     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1747     cnt(vtemp, T8B, vtemp);
1748     addv(vtemp, T8B, vtemp);
1749     fmovd(r_array_index, vtemp);
1750   } else {
1751     mov(r_array_index, (u1)1);
1752   }
1753   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1754 
1755   // We will consult the secondary-super array.
1756   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1757 
1758   // The value i in r_array_index is >= 1, so even though r_array_base
1759   // points to the length, we don't need to adjust it to point to the
1760   // data.
1761   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1762   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1763 
1764   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1765   eor(result, result, r_super_klass);
1766   cbz(result, L_fallthrough); // Found a match
1767 
1768   // Is there another entry to check? Consult the bitmap.
1769   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1770 
1771   // Linear probe.
1772   if (bit != 0) {
1773     ror(r_bitmap, r_bitmap, bit);
1774   }
1775 
1776   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1777   // The next slot to be inspected, by the stub we're about to call,
1778   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1779   // have been checked.
1780   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1781   if (stub_is_near) {
1782     bl(stub);
1783   } else {
1784     address call = trampoline_call(stub);
1785     if (call == nullptr) {
1786       return false; // trampoline allocation failed
1787     }
1788   }
1789 
1790   BLOCK_COMMENT("} lookup_secondary_supers_table");
1791 
1792   bind(L_fallthrough);
1793 
1794   if (VerifySecondarySupers) {
1795     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1796                                   temp1, temp2, result);      // r1, r2, r5
1797   }
1798   return true;
1799 }
1800 
1801 // At runtime, return 0 in result if r_super_klass is a superclass of
1802 // r_sub_klass, otherwise return nonzero. Use this version of
1803 // lookup_secondary_supers_table() if you don't know ahead of time
1804 // which superclass will be searched for. Used by interpreter and
1805 // runtime stubs. It is larger and has somewhat greater latency than
1806 // the version above, which takes a constant super_klass_slot.
1807 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1808                                                        Register r_super_klass,
1809                                                        Register temp1,
1810                                                        Register temp2,
1811                                                        Register temp3,
1812                                                        FloatRegister vtemp,
1813                                                        Register result,
1814                                                        Label *L_success) {
1815   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1816 
1817   Label L_fallthrough;
1818 
1819   BLOCK_COMMENT("lookup_secondary_supers_table {");
1820 
1821   const Register
1822     r_array_index = temp3,
1823     slot          = rscratch1,
1824     r_bitmap      = rscratch2;
1825 
1826   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1827 
1828   // Make sure that result is nonzero if the test below misses.
1829   mov(result, 1);
1830 
1831   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1832 
1833   // First check the bitmap to see if super_klass might be present. If
1834   // the bit is zero, we are certain that super_klass is not one of
1835   // the secondary supers.
1836 
1837   // This next instruction is equivalent to:
1838   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1839   // sub(temp2, tmp_reg, slot);
1840   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1841   lslv(temp2, r_bitmap, temp2);
1842   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1843 
1844   bool must_save_v0 = (vtemp == fnoreg);
1845   if (must_save_v0) {
1846     // temp1 and result are free, so use them to preserve vtemp
1847     vtemp = v0;
1848     mov(temp1,  vtemp, D, 0);
1849     mov(result, vtemp, D, 1);
1850   }
1851 
1852   // Get the first array index that can contain super_klass into r_array_index.
1853   mov(vtemp, D, 0, temp2);
1854   cnt(vtemp, T8B, vtemp);
1855   addv(vtemp, T8B, vtemp);
1856   mov(r_array_index, vtemp, D, 0);
1857 
1858   if (must_save_v0) {
1859     mov(vtemp, D, 0, temp1 );
1860     mov(vtemp, D, 1, result);
1861   }
1862 
1863   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1864 
1865   const Register
1866     r_array_base   = temp1,
1867     r_array_length = temp2;
1868 
1869   // The value i in r_array_index is >= 1, so even though r_array_base
1870   // points to the length, we don't need to adjust it to point to the
1871   // data.
1872   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1873   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1874 
1875   // We will consult the secondary-super array.
1876   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1877 
1878   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1879   eor(result, result, r_super_klass);
1880   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1881 
1882   // Is there another entry to check? Consult the bitmap.
1883   rorv(r_bitmap, r_bitmap, slot);
1884   // rol(r_bitmap, r_bitmap, 1);
1885   tbz(r_bitmap, 1, L_fallthrough);
1886 
1887   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1888   // The next slot to be inspected, by the logic we're about to call,
1889   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1890   // have been checked.
1891   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1892                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1893 
1894   BLOCK_COMMENT("} lookup_secondary_supers_table");
1895 
1896   bind(L_fallthrough);
1897 
1898   if (VerifySecondarySupers) {
1899     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1900                                   temp1, temp2, result);      // r1, r2, r5
1901   }
1902 
1903   if (L_success) {
1904     cbz(result, *L_success);
1905   }
1906 }
1907 
1908 // Called by code generated by check_klass_subtype_slow_path
1909 // above. This is called when there is a collision in the hashed
1910 // lookup in the secondary supers array.
1911 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1912                                                              Register r_array_base,
1913                                                              Register r_array_index,
1914                                                              Register r_bitmap,
1915                                                              Register temp1,
1916                                                              Register result,
1917                                                              bool is_stub) {
1918   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1919 
1920   const Register
1921     r_array_length = temp1,
1922     r_sub_klass    = noreg; // unused
1923 
1924   if (is_stub) {
1925     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1926   }
1927 
1928   Label L_fallthrough, L_huge;
1929 
1930   // Load the array length.
1931   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1932   // And adjust the array base to point to the data.
1933   // NB! Effectively increments current slot index by 1.
1934   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1935   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1936 
1937   // The bitmap is full to bursting.
1938   // Implicit invariant: BITMAP_FULL implies (length > 0)
1939   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1940   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1941   br(GT, L_huge);
1942 
1943   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1944   // current slot (at secondary_supers[r_array_index]) has not yet
1945   // been inspected, and r_array_index may be out of bounds if we
1946   // wrapped around the end of the array.
1947 
1948   { // This is conventional linear probing, but instead of terminating
1949     // when a null entry is found in the table, we maintain a bitmap
1950     // in which a 0 indicates missing entries.
1951     // As long as the bitmap is not completely full,
1952     // array_length == popcount(bitmap). The array_length check above
1953     // guarantees there are 0s in the bitmap, so the loop eventually
1954     // terminates.
1955     Label L_loop;
1956     bind(L_loop);
1957 
1958     // Check for wraparound.
1959     cmp(r_array_index, r_array_length);
1960     csel(r_array_index, zr, r_array_index, GE);
1961 
1962     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1963     eor(result, rscratch1, r_super_klass);
1964     cbz(result, L_fallthrough);
1965 
1966     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1967 
1968     ror(r_bitmap, r_bitmap, 1);
1969     add(r_array_index, r_array_index, 1);
1970     b(L_loop);
1971   }
1972 
1973   { // Degenerate case: more than 64 secondary supers.
1974     // FIXME: We could do something smarter here, maybe a vectorized
1975     // comparison or a binary search, but is that worth any added
1976     // complexity?
1977     bind(L_huge);
1978     cmp(sp, zr); // Clear Z flag; SP is never zero
1979     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1980     cset(result, NE); // result == 0 iff we got a match.
1981   }
1982 
1983   bind(L_fallthrough);
1984 }
1985 
1986 // Make sure that the hashed lookup and a linear scan agree.
1987 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1988                                                    Register r_super_klass,
1989                                                    Register temp1,
1990                                                    Register temp2,
1991                                                    Register result) {
1992   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1993 
1994   const Register
1995     r_array_base   = temp1,
1996     r_array_length = temp2,
1997     r_array_index  = noreg, // unused
1998     r_bitmap       = noreg; // unused
1999 
2000   BLOCK_COMMENT("verify_secondary_supers_table {");
2001 
2002   // We will consult the secondary-super array.
2003   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
2004 
2005   // Load the array length.
2006   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
2007   // And adjust the array base to point to the data.
2008   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
2009 
2010   cmp(sp, zr); // Clear Z flag; SP is never zero
2011   // Scan R2 words at [R5] for an occurrence of R0.
2012   // Set NZ/Z based on last compare.
2013   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
2014   // rscratch1 == 0 iff we got a match.
2015   cset(rscratch1, NE);
2016 
2017   Label passed;
2018   cmp(result, zr);
2019   cset(result, NE); // normalize result to 0/1 for comparison
2020 
2021   cmp(rscratch1, result);
2022   br(EQ, passed);
2023   {
2024     mov(r0, r_super_klass);         // r0 <- r0
2025     mov(r1, r_sub_klass);           // r1 <- r4
2026     mov(r2, /*expected*/rscratch1); // r2 <- r8
2027     mov(r3, result);                // r3 <- r5
2028     mov(r4, (address)("mismatch")); // r4 <- const
2029     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
2030     should_not_reach_here();
2031   }
2032   bind(passed);
2033 
2034   BLOCK_COMMENT("} verify_secondary_supers_table");
2035 }
2036 
2037 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
2038   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
2039   assert_different_registers(klass, rthread, scratch);
2040 
2041   Label L_fallthrough, L_tmp;
2042   if (L_fast_path == nullptr) {
2043     L_fast_path = &L_fallthrough;
2044   } else if (L_slow_path == nullptr) {
2045     L_slow_path = &L_fallthrough;
2046   }
2047   // Fast path check: class is fully initialized
2048   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2049   ldarb(scratch, scratch);
2050   cmp(scratch, InstanceKlass::fully_initialized);
2051   br(Assembler::EQ, *L_fast_path);
2052 
2053   // Fast path check: current thread is initializer thread
2054   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2055   cmp(rthread, scratch);
2056 
2057   if (L_slow_path == &L_fallthrough) {
2058     br(Assembler::EQ, *L_fast_path);
2059     bind(*L_slow_path);
2060   } else if (L_fast_path == &L_fallthrough) {
2061     br(Assembler::NE, *L_slow_path);
2062     bind(*L_fast_path);
2063   } else {
2064     Unimplemented();
2065   }
2066 }
2067 
2068 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2069   if (!VerifyOops) return;
2070 
2071   // Pass register number to verify_oop_subroutine
2072   const char* b = nullptr;
2073   {
2074     ResourceMark rm;
2075     stringStream ss;
2076     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2077     b = code_string(ss.as_string());
2078   }
2079   BLOCK_COMMENT("verify_oop {");
2080 
2081   strip_return_address(); // This might happen within a stack frame.
2082   protect_return_address();
2083   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2084   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2085 
2086   mov(r0, reg);
2087   movptr(rscratch1, (uintptr_t)(address)b);
2088 
2089   // call indirectly to solve generation ordering problem
2090   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2091   ldr(rscratch2, Address(rscratch2));
2092   blr(rscratch2);
2093 
2094   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2095   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2096   authenticate_return_address();
2097 
2098   BLOCK_COMMENT("} verify_oop");
2099 }
2100 
2101 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2102   if (!VerifyOops) return;
2103 
2104   const char* b = nullptr;
2105   {
2106     ResourceMark rm;
2107     stringStream ss;
2108     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2109     b = code_string(ss.as_string());
2110   }
2111   BLOCK_COMMENT("verify_oop_addr {");
2112 
2113   strip_return_address(); // This might happen within a stack frame.
2114   protect_return_address();
2115   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2116   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2117 
2118   // addr may contain sp so we will have to adjust it based on the
2119   // pushes that we just did.
2120   if (addr.uses(sp)) {
2121     lea(r0, addr);
2122     ldr(r0, Address(r0, 4 * wordSize));
2123   } else {
2124     ldr(r0, addr);
2125   }
2126   movptr(rscratch1, (uintptr_t)(address)b);
2127 
2128   // call indirectly to solve generation ordering problem
2129   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2130   ldr(rscratch2, Address(rscratch2));
2131   blr(rscratch2);
2132 
2133   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2134   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2135   authenticate_return_address();
2136 
2137   BLOCK_COMMENT("} verify_oop_addr");
2138 }
2139 
2140 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2141                                          int extra_slot_offset) {
2142   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2143   int stackElementSize = Interpreter::stackElementSize;
2144   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2145 #ifdef ASSERT
2146   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2147   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2148 #endif
2149   if (arg_slot.is_constant()) {
2150     return Address(esp, arg_slot.as_constant() * stackElementSize
2151                    + offset);
2152   } else {
2153     add(rscratch1, esp, arg_slot.as_register(),
2154         ext::uxtx, exact_log2(stackElementSize));
2155     return Address(rscratch1, offset);
2156   }
2157 }
2158 
2159 void MacroAssembler::call_VM_leaf_base(address entry_point,
2160                                        int number_of_arguments,
2161                                        Label *retaddr) {
2162   Label E, L;
2163 
2164   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2165 
2166   mov(rscratch1, RuntimeAddress(entry_point));
2167   blr(rscratch1);
2168   if (retaddr)
2169     bind(*retaddr);
2170 
2171   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2172 }
2173 
2174 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2175   call_VM_leaf_base(entry_point, number_of_arguments);
2176 }
2177 
2178 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2179   pass_arg0(this, arg_0);
2180   call_VM_leaf_base(entry_point, 1);
2181 }
2182 
2183 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2184   assert_different_registers(arg_1, c_rarg0);
2185   pass_arg0(this, arg_0);
2186   pass_arg1(this, arg_1);
2187   call_VM_leaf_base(entry_point, 2);
2188 }
2189 
2190 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2191                                   Register arg_1, Register arg_2) {
2192   assert_different_registers(arg_1, c_rarg0);
2193   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2194   pass_arg0(this, arg_0);
2195   pass_arg1(this, arg_1);
2196   pass_arg2(this, arg_2);
2197   call_VM_leaf_base(entry_point, 3);
2198 }
2199 
2200 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2201   pass_arg0(this, arg_0);
2202   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2203 }
2204 
2205 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2206 
2207   assert_different_registers(arg_0, c_rarg1);
2208   pass_arg1(this, arg_1);
2209   pass_arg0(this, arg_0);
2210   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2211 }
2212 
2213 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2214   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2215   assert_different_registers(arg_1, c_rarg2);
2216   pass_arg2(this, arg_2);
2217   pass_arg1(this, arg_1);
2218   pass_arg0(this, arg_0);
2219   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2220 }
2221 
2222 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2223   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2224   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2225   assert_different_registers(arg_2, c_rarg3);
2226   pass_arg3(this, arg_3);
2227   pass_arg2(this, arg_2);
2228   pass_arg1(this, arg_1);
2229   pass_arg0(this, arg_0);
2230   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2231 }
2232 
2233 void MacroAssembler::null_check(Register reg, int offset) {
2234   if (needs_explicit_null_check(offset)) {
2235     // provoke OS null exception if reg is null by
2236     // accessing M[reg] w/o changing any registers
2237     // NOTE: this is plenty to provoke a segv
2238     ldr(zr, Address(reg));
2239   } else {
2240     // nothing to do, (later) access of M[reg + offset]
2241     // will provoke OS null exception if reg is null
2242   }
2243 }
2244 
2245 // MacroAssembler protected routines needed to implement
2246 // public methods
2247 
2248 void MacroAssembler::mov(Register r, Address dest) {
2249   code_section()->relocate(pc(), dest.rspec());
2250   uint64_t imm64 = (uint64_t)dest.target();
2251   movptr(r, imm64);
2252 }
2253 
2254 // Move a constant pointer into r.  In AArch64 mode the virtual
2255 // address space is 48 bits in size, so we only need three
2256 // instructions to create a patchable instruction sequence that can
2257 // reach anywhere.
2258 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2259 #ifndef PRODUCT
2260   {
2261     char buffer[64];
2262     snprintf(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2263     block_comment(buffer);
2264   }
2265 #endif
2266   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2267   movz(r, imm64 & 0xffff);
2268   imm64 >>= 16;
2269   movk(r, imm64 & 0xffff, 16);
2270   imm64 >>= 16;
2271   movk(r, imm64 & 0xffff, 32);
2272 }
2273 
2274 // Macro to mov replicated immediate to vector register.
2275 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2276 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2277 // Vd will get the following values for different arrangements in T
2278 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2279 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2280 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2281 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2282 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2283 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2284 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2285 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2286 // Clobbers rscratch1
2287 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2288   assert(T != T1Q, "unsupported");
2289   if (T == T1D || T == T2D) {
2290     int imm = operand_valid_for_movi_immediate(imm64, T);
2291     if (-1 != imm) {
2292       movi(Vd, T, imm);
2293     } else {
2294       mov(rscratch1, imm64);
2295       dup(Vd, T, rscratch1);
2296     }
2297     return;
2298   }
2299 
2300 #ifdef ASSERT
2301   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2302   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2303   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2304 #endif
2305   int shift = operand_valid_for_movi_immediate(imm64, T);
2306   uint32_t imm32 = imm64 & 0xffffffffULL;
2307   if (shift >= 0) {
2308     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2309   } else {
2310     movw(rscratch1, imm32);
2311     dup(Vd, T, rscratch1);
2312   }
2313 }
2314 
2315 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2316 {
2317 #ifndef PRODUCT
2318   {
2319     char buffer[64];
2320     snprintf(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2321     block_comment(buffer);
2322   }
2323 #endif
2324   if (operand_valid_for_logical_immediate(false, imm64)) {
2325     orr(dst, zr, imm64);
2326   } else {
2327     // we can use a combination of MOVZ or MOVN with
2328     // MOVK to build up the constant
2329     uint64_t imm_h[4];
2330     int zero_count = 0;
2331     int neg_count = 0;
2332     int i;
2333     for (i = 0; i < 4; i++) {
2334       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2335       if (imm_h[i] == 0) {
2336         zero_count++;
2337       } else if (imm_h[i] == 0xffffL) {
2338         neg_count++;
2339       }
2340     }
2341     if (zero_count == 4) {
2342       // one MOVZ will do
2343       movz(dst, 0);
2344     } else if (neg_count == 4) {
2345       // one MOVN will do
2346       movn(dst, 0);
2347     } else if (zero_count == 3) {
2348       for (i = 0; i < 4; i++) {
2349         if (imm_h[i] != 0L) {
2350           movz(dst, (uint32_t)imm_h[i], (i << 4));
2351           break;
2352         }
2353       }
2354     } else if (neg_count == 3) {
2355       // one MOVN will do
2356       for (int i = 0; i < 4; i++) {
2357         if (imm_h[i] != 0xffffL) {
2358           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2359           break;
2360         }
2361       }
2362     } else if (zero_count == 2) {
2363       // one MOVZ and one MOVK will do
2364       for (i = 0; i < 3; i++) {
2365         if (imm_h[i] != 0L) {
2366           movz(dst, (uint32_t)imm_h[i], (i << 4));
2367           i++;
2368           break;
2369         }
2370       }
2371       for (;i < 4; i++) {
2372         if (imm_h[i] != 0L) {
2373           movk(dst, (uint32_t)imm_h[i], (i << 4));
2374         }
2375       }
2376     } else if (neg_count == 2) {
2377       // one MOVN and one MOVK will do
2378       for (i = 0; i < 4; i++) {
2379         if (imm_h[i] != 0xffffL) {
2380           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2381           i++;
2382           break;
2383         }
2384       }
2385       for (;i < 4; i++) {
2386         if (imm_h[i] != 0xffffL) {
2387           movk(dst, (uint32_t)imm_h[i], (i << 4));
2388         }
2389       }
2390     } else if (zero_count == 1) {
2391       // one MOVZ and two MOVKs will do
2392       for (i = 0; i < 4; i++) {
2393         if (imm_h[i] != 0L) {
2394           movz(dst, (uint32_t)imm_h[i], (i << 4));
2395           i++;
2396           break;
2397         }
2398       }
2399       for (;i < 4; i++) {
2400         if (imm_h[i] != 0x0L) {
2401           movk(dst, (uint32_t)imm_h[i], (i << 4));
2402         }
2403       }
2404     } else if (neg_count == 1) {
2405       // one MOVN and two MOVKs will do
2406       for (i = 0; i < 4; i++) {
2407         if (imm_h[i] != 0xffffL) {
2408           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2409           i++;
2410           break;
2411         }
2412       }
2413       for (;i < 4; i++) {
2414         if (imm_h[i] != 0xffffL) {
2415           movk(dst, (uint32_t)imm_h[i], (i << 4));
2416         }
2417       }
2418     } else {
2419       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2420       movz(dst, (uint32_t)imm_h[0], 0);
2421       for (i = 1; i < 4; i++) {
2422         movk(dst, (uint32_t)imm_h[i], (i << 4));
2423       }
2424     }
2425   }
2426 }
2427 
2428 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2429 {
2430 #ifndef PRODUCT
2431     {
2432       char buffer[64];
2433       snprintf(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2434       block_comment(buffer);
2435     }
2436 #endif
2437   if (operand_valid_for_logical_immediate(true, imm32)) {
2438     orrw(dst, zr, imm32);
2439   } else {
2440     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2441     // constant
2442     uint32_t imm_h[2];
2443     imm_h[0] = imm32 & 0xffff;
2444     imm_h[1] = ((imm32 >> 16) & 0xffff);
2445     if (imm_h[0] == 0) {
2446       movzw(dst, imm_h[1], 16);
2447     } else if (imm_h[0] == 0xffff) {
2448       movnw(dst, imm_h[1] ^ 0xffff, 16);
2449     } else if (imm_h[1] == 0) {
2450       movzw(dst, imm_h[0], 0);
2451     } else if (imm_h[1] == 0xffff) {
2452       movnw(dst, imm_h[0] ^ 0xffff, 0);
2453     } else {
2454       // use a MOVZ and MOVK (makes it easier to debug)
2455       movzw(dst, imm_h[0], 0);
2456       movkw(dst, imm_h[1], 16);
2457     }
2458   }
2459 }
2460 
2461 // Form an address from base + offset in Rd.  Rd may or may
2462 // not actually be used: you must use the Address that is returned.
2463 // It is up to you to ensure that the shift provided matches the size
2464 // of your data.
2465 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2466   if (Address::offset_ok_for_immed(byte_offset, shift))
2467     // It fits; no need for any heroics
2468     return Address(base, byte_offset);
2469 
2470   // Don't do anything clever with negative or misaligned offsets
2471   unsigned mask = (1 << shift) - 1;
2472   if (byte_offset < 0 || byte_offset & mask) {
2473     mov(Rd, byte_offset);
2474     add(Rd, base, Rd);
2475     return Address(Rd);
2476   }
2477 
2478   // See if we can do this with two 12-bit offsets
2479   {
2480     uint64_t word_offset = byte_offset >> shift;
2481     uint64_t masked_offset = word_offset & 0xfff000;
2482     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2483         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2484       add(Rd, base, masked_offset << shift);
2485       word_offset -= masked_offset;
2486       return Address(Rd, word_offset << shift);
2487     }
2488   }
2489 
2490   // Do it the hard way
2491   mov(Rd, byte_offset);
2492   add(Rd, base, Rd);
2493   return Address(Rd);
2494 }
2495 
2496 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2497                                     bool want_remainder, Register scratch)
2498 {
2499   // Full implementation of Java idiv and irem.  The function
2500   // returns the (pc) offset of the div instruction - may be needed
2501   // for implicit exceptions.
2502   //
2503   // constraint : ra/rb =/= scratch
2504   //         normal case
2505   //
2506   // input : ra: dividend
2507   //         rb: divisor
2508   //
2509   // result: either
2510   //         quotient  (= ra idiv rb)
2511   //         remainder (= ra irem rb)
2512 
2513   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2514 
2515   int idivl_offset = offset();
2516   if (! want_remainder) {
2517     sdivw(result, ra, rb);
2518   } else {
2519     sdivw(scratch, ra, rb);
2520     Assembler::msubw(result, scratch, rb, ra);
2521   }
2522 
2523   return idivl_offset;
2524 }
2525 
2526 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2527                                     bool want_remainder, Register scratch)
2528 {
2529   // Full implementation of Java ldiv and lrem.  The function
2530   // returns the (pc) offset of the div instruction - may be needed
2531   // for implicit exceptions.
2532   //
2533   // constraint : ra/rb =/= scratch
2534   //         normal case
2535   //
2536   // input : ra: dividend
2537   //         rb: divisor
2538   //
2539   // result: either
2540   //         quotient  (= ra idiv rb)
2541   //         remainder (= ra irem rb)
2542 
2543   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2544 
2545   int idivq_offset = offset();
2546   if (! want_remainder) {
2547     sdiv(result, ra, rb);
2548   } else {
2549     sdiv(scratch, ra, rb);
2550     Assembler::msub(result, scratch, rb, ra);
2551   }
2552 
2553   return idivq_offset;
2554 }
2555 
2556 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2557   address prev = pc() - NativeMembar::instruction_size;
2558   address last = code()->last_insn();
2559   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2560     NativeMembar *bar = NativeMembar_at(prev);
2561     if (AlwaysMergeDMB) {
2562       bar->set_kind(bar->get_kind() | order_constraint);
2563       BLOCK_COMMENT("merged membar(always)");
2564       return;
2565     }
2566     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2567     // doing so would introduce a StoreLoad which the caller did not
2568     // intend
2569     if (bar->get_kind() == order_constraint
2570         || bar->get_kind() == AnyAny
2571         || order_constraint == AnyAny) {
2572       // We are merging two memory barrier instructions.  On AArch64 we
2573       // can do this simply by ORing them together.
2574       bar->set_kind(bar->get_kind() | order_constraint);
2575       BLOCK_COMMENT("merged membar");
2576       return;
2577     } else {
2578       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2579       // We need check the last 2 instructions
2580       address prev2 = prev - NativeMembar::instruction_size;
2581       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2582         NativeMembar *bar2 = NativeMembar_at(prev2);
2583         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2584         BLOCK_COMMENT("merged membar(elided)");
2585         return;
2586       }
2587     }
2588   }
2589   code()->set_last_insn(pc());
2590   dmb(Assembler::barrier(order_constraint));
2591 }
2592 
2593 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2594   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2595     merge_ldst(rt, adr, size_in_bytes, is_store);
2596     code()->clear_last_insn();
2597     return true;
2598   } else {
2599     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2600     const uint64_t mask = size_in_bytes - 1;
2601     if (adr.getMode() == Address::base_plus_offset &&
2602         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2603       code()->set_last_insn(pc());
2604     }
2605     return false;
2606   }
2607 }
2608 
2609 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2610   // We always try to merge two adjacent loads into one ldp.
2611   if (!try_merge_ldst(Rx, adr, 8, false)) {
2612     Assembler::ldr(Rx, adr);
2613   }
2614 }
2615 
2616 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2617   // We always try to merge two adjacent loads into one ldp.
2618   if (!try_merge_ldst(Rw, adr, 4, false)) {
2619     Assembler::ldrw(Rw, adr);
2620   }
2621 }
2622 
2623 void MacroAssembler::str(Register Rx, const Address &adr) {
2624   // We always try to merge two adjacent stores into one stp.
2625   if (!try_merge_ldst(Rx, adr, 8, true)) {
2626     Assembler::str(Rx, adr);
2627   }
2628 }
2629 
2630 void MacroAssembler::strw(Register Rw, const Address &adr) {
2631   // We always try to merge two adjacent stores into one stp.
2632   if (!try_merge_ldst(Rw, adr, 4, true)) {
2633     Assembler::strw(Rw, adr);
2634   }
2635 }
2636 
2637 // MacroAssembler routines found actually to be needed
2638 
2639 void MacroAssembler::push(Register src)
2640 {
2641   str(src, Address(pre(esp, -1 * wordSize)));
2642 }
2643 
2644 void MacroAssembler::pop(Register dst)
2645 {
2646   ldr(dst, Address(post(esp, 1 * wordSize)));
2647 }
2648 
2649 // Note: load_unsigned_short used to be called load_unsigned_word.
2650 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2651   int off = offset();
2652   ldrh(dst, src);
2653   return off;
2654 }
2655 
2656 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2657   int off = offset();
2658   ldrb(dst, src);
2659   return off;
2660 }
2661 
2662 int MacroAssembler::load_signed_short(Register dst, Address src) {
2663   int off = offset();
2664   ldrsh(dst, src);
2665   return off;
2666 }
2667 
2668 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2669   int off = offset();
2670   ldrsb(dst, src);
2671   return off;
2672 }
2673 
2674 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2675   int off = offset();
2676   ldrshw(dst, src);
2677   return off;
2678 }
2679 
2680 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2681   int off = offset();
2682   ldrsbw(dst, src);
2683   return off;
2684 }
2685 
2686 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2687   switch (size_in_bytes) {
2688   case  8:  ldr(dst, src); break;
2689   case  4:  ldrw(dst, src); break;
2690   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2691   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2692   default:  ShouldNotReachHere();
2693   }
2694 }
2695 
2696 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2697   switch (size_in_bytes) {
2698   case  8:  str(src, dst); break;
2699   case  4:  strw(src, dst); break;
2700   case  2:  strh(src, dst); break;
2701   case  1:  strb(src, dst); break;
2702   default:  ShouldNotReachHere();
2703   }
2704 }
2705 
2706 void MacroAssembler::decrementw(Register reg, int value)
2707 {
2708   if (value < 0)  { incrementw(reg, -value);      return; }
2709   if (value == 0) {                               return; }
2710   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2711   /* else */ {
2712     guarantee(reg != rscratch2, "invalid dst for register decrement");
2713     movw(rscratch2, (unsigned)value);
2714     subw(reg, reg, rscratch2);
2715   }
2716 }
2717 
2718 void MacroAssembler::decrement(Register reg, int value)
2719 {
2720   if (value < 0)  { increment(reg, -value);      return; }
2721   if (value == 0) {                              return; }
2722   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2723   /* else */ {
2724     assert(reg != rscratch2, "invalid dst for register decrement");
2725     mov(rscratch2, (uint64_t)value);
2726     sub(reg, reg, rscratch2);
2727   }
2728 }
2729 
2730 void MacroAssembler::decrementw(Address dst, int value)
2731 {
2732   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2733   if (dst.getMode() == Address::literal) {
2734     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2735     lea(rscratch2, dst);
2736     dst = Address(rscratch2);
2737   }
2738   ldrw(rscratch1, dst);
2739   decrementw(rscratch1, value);
2740   strw(rscratch1, dst);
2741 }
2742 
2743 void MacroAssembler::decrement(Address dst, int value)
2744 {
2745   assert(!dst.uses(rscratch1), "invalid address for decrement");
2746   if (dst.getMode() == Address::literal) {
2747     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2748     lea(rscratch2, dst);
2749     dst = Address(rscratch2);
2750   }
2751   ldr(rscratch1, dst);
2752   decrement(rscratch1, value);
2753   str(rscratch1, dst);
2754 }
2755 
2756 void MacroAssembler::incrementw(Register reg, int value)
2757 {
2758   if (value < 0)  { decrementw(reg, -value);      return; }
2759   if (value == 0) {                               return; }
2760   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2761   /* else */ {
2762     assert(reg != rscratch2, "invalid dst for register increment");
2763     movw(rscratch2, (unsigned)value);
2764     addw(reg, reg, rscratch2);
2765   }
2766 }
2767 
2768 void MacroAssembler::increment(Register reg, int value)
2769 {
2770   if (value < 0)  { decrement(reg, -value);      return; }
2771   if (value == 0) {                              return; }
2772   if (value < (1 << 12)) { add(reg, reg, value); return; }
2773   /* else */ {
2774     assert(reg != rscratch2, "invalid dst for register increment");
2775     movw(rscratch2, (unsigned)value);
2776     add(reg, reg, rscratch2);
2777   }
2778 }
2779 
2780 void MacroAssembler::incrementw(Address dst, int value)
2781 {
2782   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2783   if (dst.getMode() == Address::literal) {
2784     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2785     lea(rscratch2, dst);
2786     dst = Address(rscratch2);
2787   }
2788   ldrw(rscratch1, dst);
2789   incrementw(rscratch1, value);
2790   strw(rscratch1, dst);
2791 }
2792 
2793 void MacroAssembler::increment(Address dst, int value)
2794 {
2795   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2796   if (dst.getMode() == Address::literal) {
2797     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2798     lea(rscratch2, dst);
2799     dst = Address(rscratch2);
2800   }
2801   ldr(rscratch1, dst);
2802   increment(rscratch1, value);
2803   str(rscratch1, dst);
2804 }
2805 
2806 // Push lots of registers in the bit set supplied.  Don't push sp.
2807 // Return the number of words pushed
2808 int MacroAssembler::push(unsigned int bitset, Register stack) {
2809   int words_pushed = 0;
2810 
2811   // Scan bitset to accumulate register pairs
2812   unsigned char regs[32];
2813   int count = 0;
2814   for (int reg = 0; reg <= 30; reg++) {
2815     if (1 & bitset)
2816       regs[count++] = reg;
2817     bitset >>= 1;
2818   }
2819   regs[count++] = zr->raw_encoding();
2820   count &= ~1;  // Only push an even number of regs
2821 
2822   if (count) {
2823     stp(as_Register(regs[0]), as_Register(regs[1]),
2824        Address(pre(stack, -count * wordSize)));
2825     words_pushed += 2;
2826   }
2827   for (int i = 2; i < count; i += 2) {
2828     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2829        Address(stack, i * wordSize));
2830     words_pushed += 2;
2831   }
2832 
2833   assert(words_pushed == count, "oops, pushed != count");
2834 
2835   return count;
2836 }
2837 
2838 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2839   int words_pushed = 0;
2840 
2841   // Scan bitset to accumulate register pairs
2842   unsigned char regs[32];
2843   int count = 0;
2844   for (int reg = 0; reg <= 30; reg++) {
2845     if (1 & bitset)
2846       regs[count++] = reg;
2847     bitset >>= 1;
2848   }
2849   regs[count++] = zr->raw_encoding();
2850   count &= ~1;
2851 
2852   for (int i = 2; i < count; i += 2) {
2853     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2854        Address(stack, i * wordSize));
2855     words_pushed += 2;
2856   }
2857   if (count) {
2858     ldp(as_Register(regs[0]), as_Register(regs[1]),
2859        Address(post(stack, count * wordSize)));
2860     words_pushed += 2;
2861   }
2862 
2863   assert(words_pushed == count, "oops, pushed != count");
2864 
2865   return count;
2866 }
2867 
2868 // Push lots of registers in the bit set supplied.  Don't push sp.
2869 // Return the number of dwords pushed
2870 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2871   int words_pushed = 0;
2872   bool use_sve = false;
2873   int sve_vector_size_in_bytes = 0;
2874 
2875 #ifdef COMPILER2
2876   use_sve = Matcher::supports_scalable_vector();
2877   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2878 #endif
2879 
2880   // Scan bitset to accumulate register pairs
2881   unsigned char regs[32];
2882   int count = 0;
2883   for (int reg = 0; reg <= 31; reg++) {
2884     if (1 & bitset)
2885       regs[count++] = reg;
2886     bitset >>= 1;
2887   }
2888 
2889   if (count == 0) {
2890     return 0;
2891   }
2892 
2893   if (mode == PushPopFull) {
2894     if (use_sve && sve_vector_size_in_bytes > 16) {
2895       mode = PushPopSVE;
2896     } else {
2897       mode = PushPopNeon;
2898     }
2899   }
2900 
2901 #ifndef PRODUCT
2902   {
2903     char buffer[48];
2904     if (mode == PushPopSVE) {
2905       snprintf(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
2906     } else if (mode == PushPopNeon) {
2907       snprintf(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
2908     } else {
2909       snprintf(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
2910     }
2911     block_comment(buffer);
2912   }
2913 #endif
2914 
2915   if (mode == PushPopSVE) {
2916     sub(stack, stack, sve_vector_size_in_bytes * count);
2917     for (int i = 0; i < count; i++) {
2918       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
2919     }
2920     return count * sve_vector_size_in_bytes / 8;
2921   }
2922 
2923   if (mode == PushPopNeon) {
2924     if (count == 1) {
2925       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
2926       return 2;
2927     }
2928 
2929     bool odd = (count & 1) == 1;
2930     int push_slots = count + (odd ? 1 : 0);
2931 
2932     // Always pushing full 128 bit registers.
2933     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
2934     words_pushed += 2;
2935 
2936     for (int i = 2; i + 1 < count; i += 2) {
2937       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
2938       words_pushed += 2;
2939     }
2940 
2941     if (odd) {
2942       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
2943       words_pushed++;
2944     }
2945 
2946     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
2947     return count * 2;
2948   }
2949 
2950   if (mode == PushPopFp) {
2951     bool odd = (count & 1) == 1;
2952     int push_slots = count + (odd ? 1 : 0);
2953 
2954     if (count == 1) {
2955       // Stack pointer must be 16 bytes aligned
2956       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
2957       return 1;
2958     }
2959 
2960     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
2961     words_pushed += 2;
2962 
2963     for (int i = 2; i + 1 < count; i += 2) {
2964       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
2965       words_pushed += 2;
2966     }
2967 
2968     if (odd) {
2969       // Stack pointer must be 16 bytes aligned
2970       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
2971       words_pushed++;
2972     }
2973 
2974     assert(words_pushed == count, "oops, pushed != count");
2975 
2976     return count;
2977   }
2978 
2979   return 0;
2980 }
2981 
2982 // Return the number of dwords popped
2983 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2984   int words_pushed = 0;
2985   bool use_sve = false;
2986   int sve_vector_size_in_bytes = 0;
2987 
2988 #ifdef COMPILER2
2989   use_sve = Matcher::supports_scalable_vector();
2990   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2991 #endif
2992   // Scan bitset to accumulate register pairs
2993   unsigned char regs[32];
2994   int count = 0;
2995   for (int reg = 0; reg <= 31; reg++) {
2996     if (1 & bitset)
2997       regs[count++] = reg;
2998     bitset >>= 1;
2999   }
3000 
3001   if (count == 0) {
3002     return 0;
3003   }
3004 
3005   if (mode == PushPopFull) {
3006     if (use_sve && sve_vector_size_in_bytes > 16) {
3007       mode = PushPopSVE;
3008     } else {
3009       mode = PushPopNeon;
3010     }
3011   }
3012 
3013 #ifndef PRODUCT
3014   {
3015     char buffer[48];
3016     if (mode == PushPopSVE) {
3017       snprintf(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3018     } else if (mode == PushPopNeon) {
3019       snprintf(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3020     } else {
3021       snprintf(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3022     }
3023     block_comment(buffer);
3024   }
3025 #endif
3026 
3027   if (mode == PushPopSVE) {
3028     for (int i = count - 1; i >= 0; i--) {
3029       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3030     }
3031     add(stack, stack, sve_vector_size_in_bytes * count);
3032     return count * sve_vector_size_in_bytes / 8;
3033   }
3034 
3035   if (mode == PushPopNeon) {
3036     if (count == 1) {
3037       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3038       return 2;
3039     }
3040 
3041     bool odd = (count & 1) == 1;
3042     int push_slots = count + (odd ? 1 : 0);
3043 
3044     if (odd) {
3045       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3046       words_pushed++;
3047     }
3048 
3049     for (int i = 2; i + 1 < count; i += 2) {
3050       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3051       words_pushed += 2;
3052     }
3053 
3054     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3055     words_pushed += 2;
3056 
3057     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3058 
3059     return count * 2;
3060   }
3061 
3062   if (mode == PushPopFp) {
3063     bool odd = (count & 1) == 1;
3064     int push_slots = count + (odd ? 1 : 0);
3065 
3066     if (count == 1) {
3067       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3068       return 1;
3069     }
3070 
3071     if (odd) {
3072       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3073       words_pushed++;
3074     }
3075 
3076     for (int i = 2; i + 1 < count; i += 2) {
3077       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3078       words_pushed += 2;
3079     }
3080 
3081     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3082     words_pushed += 2;
3083 
3084     assert(words_pushed == count, "oops, pushed != count");
3085 
3086     return count;
3087   }
3088 
3089   return 0;
3090 }
3091 
3092 // Return the number of dwords pushed
3093 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3094   bool use_sve = false;
3095   int sve_predicate_size_in_slots = 0;
3096 
3097 #ifdef COMPILER2
3098   use_sve = Matcher::supports_scalable_vector();
3099   if (use_sve) {
3100     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3101   }
3102 #endif
3103 
3104   if (!use_sve) {
3105     return 0;
3106   }
3107 
3108   unsigned char regs[PRegister::number_of_registers];
3109   int count = 0;
3110   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3111     if (1 & bitset)
3112       regs[count++] = reg;
3113     bitset >>= 1;
3114   }
3115 
3116   if (count == 0) {
3117     return 0;
3118   }
3119 
3120   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3121                                   VMRegImpl::stack_slot_size * count, 16);
3122   sub(stack, stack, total_push_bytes);
3123   for (int i = 0; i < count; i++) {
3124     sve_str(as_PRegister(regs[i]), Address(stack, i));
3125   }
3126   return total_push_bytes / 8;
3127 }
3128 
3129 // Return the number of dwords popped
3130 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3131   bool use_sve = false;
3132   int sve_predicate_size_in_slots = 0;
3133 
3134 #ifdef COMPILER2
3135   use_sve = Matcher::supports_scalable_vector();
3136   if (use_sve) {
3137     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3138   }
3139 #endif
3140 
3141   if (!use_sve) {
3142     return 0;
3143   }
3144 
3145   unsigned char regs[PRegister::number_of_registers];
3146   int count = 0;
3147   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3148     if (1 & bitset)
3149       regs[count++] = reg;
3150     bitset >>= 1;
3151   }
3152 
3153   if (count == 0) {
3154     return 0;
3155   }
3156 
3157   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3158                                  VMRegImpl::stack_slot_size * count, 16);
3159   for (int i = count - 1; i >= 0; i--) {
3160     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3161   }
3162   add(stack, stack, total_pop_bytes);
3163   return total_pop_bytes / 8;
3164 }
3165 
3166 #ifdef ASSERT
3167 void MacroAssembler::verify_heapbase(const char* msg) {
3168 #if 0
3169   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3170   assert (Universe::heap() != nullptr, "java heap should be initialized");
3171   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3172     // rheapbase is allocated as general register
3173     return;
3174   }
3175   if (CheckCompressedOops) {
3176     Label ok;
3177     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3178     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3179     br(Assembler::EQ, ok);
3180     stop(msg);
3181     bind(ok);
3182     pop(1 << rscratch1->encoding(), sp);
3183   }
3184 #endif
3185 }
3186 #endif
3187 
3188 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3189   assert_different_registers(value, tmp1, tmp2);
3190   Label done, tagged, weak_tagged;
3191 
3192   cbz(value, done);           // Use null as-is.
3193   tst(value, JNIHandles::tag_mask); // Test for tag.
3194   br(Assembler::NE, tagged);
3195 
3196   // Resolve local handle
3197   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3198   verify_oop(value);
3199   b(done);
3200 
3201   bind(tagged);
3202   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3203   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3204 
3205   // Resolve global handle
3206   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3207   verify_oop(value);
3208   b(done);
3209 
3210   bind(weak_tagged);
3211   // Resolve jweak.
3212   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3213                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3214   verify_oop(value);
3215 
3216   bind(done);
3217 }
3218 
3219 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3220   assert_different_registers(value, tmp1, tmp2);
3221   Label done;
3222 
3223   cbz(value, done);           // Use null as-is.
3224 
3225 #ifdef ASSERT
3226   {
3227     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3228     Label valid_global_tag;
3229     tbnz(value, 1, valid_global_tag); // Test for global tag
3230     stop("non global jobject using resolve_global_jobject");
3231     bind(valid_global_tag);
3232   }
3233 #endif
3234 
3235   // Resolve global handle
3236   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3237   verify_oop(value);
3238 
3239   bind(done);
3240 }
3241 
3242 void MacroAssembler::stop(const char* msg) {
3243   // Skip AOT caching C strings in scratch buffer.
3244   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3245   BLOCK_COMMENT(str);
3246   // load msg into r0 so we can access it from the signal handler
3247   // ExternalAddress enables saving and restoring via the code cache
3248   lea(c_rarg0, ExternalAddress((address) str));
3249   dcps1(0xdeae);
3250 }
3251 
3252 void MacroAssembler::unimplemented(const char* what) {
3253   const char* buf = nullptr;
3254   {
3255     ResourceMark rm;
3256     stringStream ss;
3257     ss.print("unimplemented: %s", what);
3258     buf = code_string(ss.as_string());
3259   }
3260   stop(buf);
3261 }
3262 
3263 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3264 #ifdef ASSERT
3265   Label OK;
3266   br(cc, OK);
3267   stop(msg);
3268   bind(OK);
3269 #endif
3270 }
3271 
3272 // If a constant does not fit in an immediate field, generate some
3273 // number of MOV instructions and then perform the operation.
3274 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3275                                            add_sub_imm_insn insn1,
3276                                            add_sub_reg_insn insn2,
3277                                            bool is32) {
3278   assert(Rd != zr, "Rd = zr and not setting flags?");
3279   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3280   if (fits) {
3281     (this->*insn1)(Rd, Rn, imm);
3282   } else {
3283     if (g_uabs(imm) < (1 << 24)) {
3284        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3285        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3286     } else {
3287        assert_different_registers(Rd, Rn);
3288        mov(Rd, imm);
3289        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3290     }
3291   }
3292 }
3293 
3294 // Separate vsn which sets the flags. Optimisations are more restricted
3295 // because we must set the flags correctly.
3296 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3297                                              add_sub_imm_insn insn1,
3298                                              add_sub_reg_insn insn2,
3299                                              bool is32) {
3300   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3301   if (fits) {
3302     (this->*insn1)(Rd, Rn, imm);
3303   } else {
3304     assert_different_registers(Rd, Rn);
3305     assert(Rd != zr, "overflow in immediate operand");
3306     mov(Rd, imm);
3307     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3308   }
3309 }
3310 
3311 
3312 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3313   if (increment.is_register()) {
3314     add(Rd, Rn, increment.as_register());
3315   } else {
3316     add(Rd, Rn, increment.as_constant());
3317   }
3318 }
3319 
3320 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3321   if (increment.is_register()) {
3322     addw(Rd, Rn, increment.as_register());
3323   } else {
3324     addw(Rd, Rn, increment.as_constant());
3325   }
3326 }
3327 
3328 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3329   if (decrement.is_register()) {
3330     sub(Rd, Rn, decrement.as_register());
3331   } else {
3332     sub(Rd, Rn, decrement.as_constant());
3333   }
3334 }
3335 
3336 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3337   if (decrement.is_register()) {
3338     subw(Rd, Rn, decrement.as_register());
3339   } else {
3340     subw(Rd, Rn, decrement.as_constant());
3341   }
3342 }
3343 
3344 void MacroAssembler::reinit_heapbase()
3345 {
3346   if (UseCompressedOops) {
3347     if (Universe::is_fully_initialized()) {
3348       mov(rheapbase, CompressedOops::base());
3349     } else {
3350       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3351       ldr(rheapbase, Address(rheapbase));
3352     }
3353   }
3354 }
3355 
3356 // this simulates the behaviour of the x86 cmpxchg instruction using a
3357 // load linked/store conditional pair. we use the acquire/release
3358 // versions of these instructions so that we flush pending writes as
3359 // per Java semantics.
3360 
3361 // n.b the x86 version assumes the old value to be compared against is
3362 // in rax and updates rax with the value located in memory if the
3363 // cmpxchg fails. we supply a register for the old value explicitly
3364 
3365 // the aarch64 load linked/store conditional instructions do not
3366 // accept an offset. so, unlike x86, we must provide a plain register
3367 // to identify the memory word to be compared/exchanged rather than a
3368 // register+offset Address.
3369 
3370 void MacroAssembler::cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
3371                                 Label &succeed, Label *fail) {
3372   // oldv holds comparison value
3373   // newv holds value to write in exchange
3374   // addr identifies memory word to compare against/update
3375   if (UseLSE) {
3376     mov(tmp, oldv);
3377     casal(Assembler::xword, oldv, newv, addr);
3378     cmp(tmp, oldv);
3379     br(Assembler::EQ, succeed);
3380     membar(AnyAny);
3381   } else {
3382     Label retry_load, nope;
3383     prfm(Address(addr), PSTL1STRM);
3384     bind(retry_load);
3385     // flush and load exclusive from the memory location
3386     // and fail if it is not what we expect
3387     ldaxr(tmp, addr);
3388     cmp(tmp, oldv);
3389     br(Assembler::NE, nope);
3390     // if we store+flush with no intervening write tmp will be zero
3391     stlxr(tmp, newv, addr);
3392     cbzw(tmp, succeed);
3393     // retry so we only ever return after a load fails to compare
3394     // ensures we don't return a stale value after a failed write.
3395     b(retry_load);
3396     // if the memory word differs we return it in oldv and signal a fail
3397     bind(nope);
3398     membar(AnyAny);
3399     mov(oldv, tmp);
3400   }
3401   if (fail)
3402     b(*fail);
3403 }
3404 
3405 void MacroAssembler::cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp,
3406                                         Label &succeed, Label *fail) {
3407   assert(oopDesc::mark_offset_in_bytes() == 0, "assumption");
3408   cmpxchgptr(oldv, newv, obj, tmp, succeed, fail);
3409 }
3410 
3411 void MacroAssembler::cmpxchgw(Register oldv, Register newv, Register addr, Register tmp,
3412                                 Label &succeed, Label *fail) {
3413   // oldv holds comparison value
3414   // newv holds value to write in exchange
3415   // addr identifies memory word to compare against/update
3416   // tmp returns 0/1 for success/failure
3417   if (UseLSE) {
3418     mov(tmp, oldv);
3419     casal(Assembler::word, oldv, newv, addr);
3420     cmp(tmp, oldv);
3421     br(Assembler::EQ, succeed);
3422     membar(AnyAny);
3423   } else {
3424     Label retry_load, nope;
3425     prfm(Address(addr), PSTL1STRM);
3426     bind(retry_load);
3427     // flush and load exclusive from the memory location
3428     // and fail if it is not what we expect
3429     ldaxrw(tmp, addr);
3430     cmp(tmp, oldv);
3431     br(Assembler::NE, nope);
3432     // if we store+flush with no intervening write tmp will be zero
3433     stlxrw(tmp, newv, addr);
3434     cbzw(tmp, succeed);
3435     // retry so we only ever return after a load fails to compare
3436     // ensures we don't return a stale value after a failed write.
3437     b(retry_load);
3438     // if the memory word differs we return it in oldv and signal a fail
3439     bind(nope);
3440     membar(AnyAny);
3441     mov(oldv, tmp);
3442   }
3443   if (fail)
3444     b(*fail);
3445 }
3446 
3447 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3448 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3449 // Pass a register for the result, otherwise pass noreg.
3450 
3451 // Clobbers rscratch1
3452 void MacroAssembler::cmpxchg(Register addr, Register expected,
3453                              Register new_val,
3454                              enum operand_size size,
3455                              bool acquire, bool release,
3456                              bool weak,
3457                              Register result) {
3458   if (result == noreg)  result = rscratch1;
3459   BLOCK_COMMENT("cmpxchg {");
3460   if (UseLSE) {
3461     mov(result, expected);
3462     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3463     compare_eq(result, expected, size);
3464 #ifdef ASSERT
3465     // Poison rscratch1 which is written on !UseLSE branch
3466     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3467 #endif
3468   } else {
3469     Label retry_load, done;
3470     prfm(Address(addr), PSTL1STRM);
3471     bind(retry_load);
3472     load_exclusive(result, addr, size, acquire);
3473     compare_eq(result, expected, size);
3474     br(Assembler::NE, done);
3475     store_exclusive(rscratch1, new_val, addr, size, release);
3476     if (weak) {
3477       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3478     } else {
3479       cbnzw(rscratch1, retry_load);
3480     }
3481     bind(done);
3482   }
3483   BLOCK_COMMENT("} cmpxchg");
3484 }
3485 
3486 // A generic comparison. Only compares for equality, clobbers rscratch1.
3487 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3488   if (size == xword) {
3489     cmp(rm, rn);
3490   } else if (size == word) {
3491     cmpw(rm, rn);
3492   } else if (size == halfword) {
3493     eorw(rscratch1, rm, rn);
3494     ands(zr, rscratch1, 0xffff);
3495   } else if (size == byte) {
3496     eorw(rscratch1, rm, rn);
3497     ands(zr, rscratch1, 0xff);
3498   } else {
3499     ShouldNotReachHere();
3500   }
3501 }
3502 
3503 
3504 static bool different(Register a, RegisterOrConstant b, Register c) {
3505   if (b.is_constant())
3506     return a != c;
3507   else
3508     return a != b.as_register() && a != c && b.as_register() != c;
3509 }
3510 
3511 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3512 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3513   if (UseLSE) {                                                         \
3514     prev = prev->is_valid() ? prev : zr;                                \
3515     if (incr.is_register()) {                                           \
3516       AOP(sz, incr.as_register(), prev, addr);                          \
3517     } else {                                                            \
3518       mov(rscratch2, incr.as_constant());                               \
3519       AOP(sz, rscratch2, prev, addr);                                   \
3520     }                                                                   \
3521     return;                                                             \
3522   }                                                                     \
3523   Register result = rscratch2;                                          \
3524   if (prev->is_valid())                                                 \
3525     result = different(prev, incr, addr) ? prev : rscratch2;            \
3526                                                                         \
3527   Label retry_load;                                                     \
3528   prfm(Address(addr), PSTL1STRM);                                       \
3529   bind(retry_load);                                                     \
3530   LDXR(result, addr);                                                   \
3531   OP(rscratch1, result, incr);                                          \
3532   STXR(rscratch2, rscratch1, addr);                                     \
3533   cbnzw(rscratch2, retry_load);                                         \
3534   if (prev->is_valid() && prev != result) {                             \
3535     IOP(prev, rscratch1, incr);                                         \
3536   }                                                                     \
3537 }
3538 
3539 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3540 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3541 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3542 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3543 
3544 #undef ATOMIC_OP
3545 
3546 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3547 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3548   if (UseLSE) {                                                         \
3549     prev = prev->is_valid() ? prev : zr;                                \
3550     AOP(sz, newv, prev, addr);                                          \
3551     return;                                                             \
3552   }                                                                     \
3553   Register result = rscratch2;                                          \
3554   if (prev->is_valid())                                                 \
3555     result = different(prev, newv, addr) ? prev : rscratch2;            \
3556                                                                         \
3557   Label retry_load;                                                     \
3558   prfm(Address(addr), PSTL1STRM);                                       \
3559   bind(retry_load);                                                     \
3560   LDXR(result, addr);                                                   \
3561   STXR(rscratch1, newv, addr);                                          \
3562   cbnzw(rscratch1, retry_load);                                         \
3563   if (prev->is_valid() && prev != result)                               \
3564     mov(prev, result);                                                  \
3565 }
3566 
3567 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3568 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3569 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3570 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3571 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3572 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3573 
3574 #undef ATOMIC_XCHG
3575 
3576 #ifndef PRODUCT
3577 extern "C" void findpc(intptr_t x);
3578 #endif
3579 
3580 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3581 {
3582   // In order to get locks to work, we need to fake a in_VM state
3583   if (ShowMessageBoxOnError ) {
3584     JavaThread* thread = JavaThread::current();
3585     JavaThreadState saved_state = thread->thread_state();
3586     thread->set_thread_state(_thread_in_vm);
3587 #ifndef PRODUCT
3588     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3589       ttyLocker ttyl;
3590       BytecodeCounter::print();
3591     }
3592 #endif
3593     if (os::message_box(msg, "Execution stopped, print registers?")) {
3594       ttyLocker ttyl;
3595       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3596 #ifndef PRODUCT
3597       tty->cr();
3598       findpc(pc);
3599       tty->cr();
3600 #endif
3601       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3602       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3603       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3604       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3605       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3606       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3607       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3608       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3609       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3610       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3611       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3612       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3613       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3614       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3615       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3616       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3617       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3618       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3619       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3620       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3621       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3622       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3623       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3624       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3625       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3626       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3627       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3628       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3629       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3630       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3631       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3632       BREAKPOINT;
3633     }
3634   }
3635   fatal("DEBUG MESSAGE: %s", msg);
3636 }
3637 
3638 RegSet MacroAssembler::call_clobbered_gp_registers() {
3639   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3640 #ifndef R18_RESERVED
3641   regs += r18_tls;
3642 #endif
3643   return regs;
3644 }
3645 
3646 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3647   int step = 4 * wordSize;
3648   push(call_clobbered_gp_registers() - exclude, sp);
3649   sub(sp, sp, step);
3650   mov(rscratch1, -step);
3651   // Push v0-v7, v16-v31.
3652   for (int i = 31; i>= 4; i -= 4) {
3653     if (i <= v7->encoding() || i >= v16->encoding())
3654       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3655           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3656   }
3657   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3658       as_FloatRegister(3), T1D, Address(sp));
3659 }
3660 
3661 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3662   for (int i = 0; i < 32; i += 4) {
3663     if (i <= v7->encoding() || i >= v16->encoding())
3664       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3665           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3666   }
3667 
3668   reinitialize_ptrue();
3669 
3670   pop(call_clobbered_gp_registers() - exclude, sp);
3671 }
3672 
3673 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3674                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3675   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3676   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3677     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3678     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3679       sve_str(as_FloatRegister(i), Address(sp, i));
3680     }
3681   } else {
3682     int step = (save_vectors ? 8 : 4) * wordSize;
3683     mov(rscratch1, -step);
3684     sub(sp, sp, step);
3685     for (int i = 28; i >= 4; i -= 4) {
3686       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3687           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3688     }
3689     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3690   }
3691   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3692     sub(sp, sp, total_predicate_in_bytes);
3693     for (int i = 0; i < PRegister::number_of_registers; i++) {
3694       sve_str(as_PRegister(i), Address(sp, i));
3695     }
3696   }
3697 }
3698 
3699 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3700                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3701   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3702     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3703       sve_ldr(as_PRegister(i), Address(sp, i));
3704     }
3705     add(sp, sp, total_predicate_in_bytes);
3706   }
3707   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3708     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3709       sve_ldr(as_FloatRegister(i), Address(sp, i));
3710     }
3711     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3712   } else {
3713     int step = (restore_vectors ? 8 : 4) * wordSize;
3714     for (int i = 0; i <= 28; i += 4)
3715       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3716           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3717   }
3718 
3719   // We may use predicate registers and rely on ptrue with SVE,
3720   // regardless of wide vector (> 8 bytes) used or not.
3721   if (use_sve) {
3722     reinitialize_ptrue();
3723   }
3724 
3725   // integer registers except lr & sp
3726   pop(RegSet::range(r0, r17), sp);
3727 #ifdef R18_RESERVED
3728   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3729   pop(RegSet::range(r20, r29), sp);
3730 #else
3731   pop(RegSet::range(r18_tls, r29), sp);
3732 #endif
3733 }
3734 
3735 /**
3736  * Helpers for multiply_to_len().
3737  */
3738 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3739                                      Register src1, Register src2) {
3740   adds(dest_lo, dest_lo, src1);
3741   adc(dest_hi, dest_hi, zr);
3742   adds(dest_lo, dest_lo, src2);
3743   adc(final_dest_hi, dest_hi, zr);
3744 }
3745 
3746 // Generate an address from (r + r1 extend offset).  "size" is the
3747 // size of the operand.  The result may be in rscratch2.
3748 Address MacroAssembler::offsetted_address(Register r, Register r1,
3749                                           Address::extend ext, int offset, int size) {
3750   if (offset || (ext.shift() % size != 0)) {
3751     lea(rscratch2, Address(r, r1, ext));
3752     return Address(rscratch2, offset);
3753   } else {
3754     return Address(r, r1, ext);
3755   }
3756 }
3757 
3758 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3759 {
3760   assert(offset >= 0, "spill to negative address?");
3761   // Offset reachable ?
3762   //   Not aligned - 9 bits signed offset
3763   //   Aligned - 12 bits unsigned offset shifted
3764   Register base = sp;
3765   if ((offset & (size-1)) && offset >= (1<<8)) {
3766     add(tmp, base, offset & ((1<<12)-1));
3767     base = tmp;
3768     offset &= -1u<<12;
3769   }
3770 
3771   if (offset >= (1<<12) * size) {
3772     add(tmp, base, offset & (((1<<12)-1)<<12));
3773     base = tmp;
3774     offset &= ~(((1<<12)-1)<<12);
3775   }
3776 
3777   return Address(base, offset);
3778 }
3779 
3780 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3781   assert(offset >= 0, "spill to negative address?");
3782 
3783   Register base = sp;
3784 
3785   // An immediate offset in the range 0 to 255 which is multiplied
3786   // by the current vector or predicate register size in bytes.
3787   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3788     return Address(base, offset / sve_reg_size_in_bytes);
3789   }
3790 
3791   add(tmp, base, offset);
3792   return Address(tmp);
3793 }
3794 
3795 // Checks whether offset is aligned.
3796 // Returns true if it is, else false.
3797 bool MacroAssembler::merge_alignment_check(Register base,
3798                                            size_t size,
3799                                            int64_t cur_offset,
3800                                            int64_t prev_offset) const {
3801   if (AvoidUnalignedAccesses) {
3802     if (base == sp) {
3803       // Checks whether low offset if aligned to pair of registers.
3804       int64_t pair_mask = size * 2 - 1;
3805       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3806       return (offset & pair_mask) == 0;
3807     } else { // If base is not sp, we can't guarantee the access is aligned.
3808       return false;
3809     }
3810   } else {
3811     int64_t mask = size - 1;
3812     // Load/store pair instruction only supports element size aligned offset.
3813     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3814   }
3815 }
3816 
3817 // Checks whether current and previous loads/stores can be merged.
3818 // Returns true if it can be merged, else false.
3819 bool MacroAssembler::ldst_can_merge(Register rt,
3820                                     const Address &adr,
3821                                     size_t cur_size_in_bytes,
3822                                     bool is_store) const {
3823   address prev = pc() - NativeInstruction::instruction_size;
3824   address last = code()->last_insn();
3825 
3826   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3827     return false;
3828   }
3829 
3830   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3831     return false;
3832   }
3833 
3834   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3835   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3836 
3837   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3838   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3839 
3840   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3841     return false;
3842   }
3843 
3844   int64_t max_offset = 63 * prev_size_in_bytes;
3845   int64_t min_offset = -64 * prev_size_in_bytes;
3846 
3847   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3848 
3849   // Only same base can be merged.
3850   if (adr.base() != prev_ldst->base()) {
3851     return false;
3852   }
3853 
3854   int64_t cur_offset = adr.offset();
3855   int64_t prev_offset = prev_ldst->offset();
3856   size_t diff = abs(cur_offset - prev_offset);
3857   if (diff != prev_size_in_bytes) {
3858     return false;
3859   }
3860 
3861   // Following cases can not be merged:
3862   // ldr x2, [x2, #8]
3863   // ldr x3, [x2, #16]
3864   // or:
3865   // ldr x2, [x3, #8]
3866   // ldr x2, [x3, #16]
3867   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3868   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3869     return false;
3870   }
3871 
3872   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3873   // Offset range must be in ldp/stp instruction's range.
3874   if (low_offset > max_offset || low_offset < min_offset) {
3875     return false;
3876   }
3877 
3878   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3879     return true;
3880   }
3881 
3882   return false;
3883 }
3884 
3885 // Merge current load/store with previous load/store into ldp/stp.
3886 void MacroAssembler::merge_ldst(Register rt,
3887                                 const Address &adr,
3888                                 size_t cur_size_in_bytes,
3889                                 bool is_store) {
3890 
3891   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3892 
3893   Register rt_low, rt_high;
3894   address prev = pc() - NativeInstruction::instruction_size;
3895   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3896 
3897   int64_t offset;
3898 
3899   if (adr.offset() < prev_ldst->offset()) {
3900     offset = adr.offset();
3901     rt_low = rt;
3902     rt_high = prev_ldst->target();
3903   } else {
3904     offset = prev_ldst->offset();
3905     rt_low = prev_ldst->target();
3906     rt_high = rt;
3907   }
3908 
3909   Address adr_p = Address(prev_ldst->base(), offset);
3910   // Overwrite previous generated binary.
3911   code_section()->set_end(prev);
3912 
3913   const size_t sz = prev_ldst->size_in_bytes();
3914   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3915   if (!is_store) {
3916     BLOCK_COMMENT("merged ldr pair");
3917     if (sz == 8) {
3918       ldp(rt_low, rt_high, adr_p);
3919     } else {
3920       ldpw(rt_low, rt_high, adr_p);
3921     }
3922   } else {
3923     BLOCK_COMMENT("merged str pair");
3924     if (sz == 8) {
3925       stp(rt_low, rt_high, adr_p);
3926     } else {
3927       stpw(rt_low, rt_high, adr_p);
3928     }
3929   }
3930 }
3931 
3932 /**
3933  * Multiply 64 bit by 64 bit first loop.
3934  */
3935 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3936                                            Register y, Register y_idx, Register z,
3937                                            Register carry, Register product,
3938                                            Register idx, Register kdx) {
3939   //
3940   //  jlong carry, x[], y[], z[];
3941   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3942   //    huge_128 product = y[idx] * x[xstart] + carry;
3943   //    z[kdx] = (jlong)product;
3944   //    carry  = (jlong)(product >>> 64);
3945   //  }
3946   //  z[xstart] = carry;
3947   //
3948 
3949   Label L_first_loop, L_first_loop_exit;
3950   Label L_one_x, L_one_y, L_multiply;
3951 
3952   subsw(xstart, xstart, 1);
3953   br(Assembler::MI, L_one_x);
3954 
3955   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3956   ldr(x_xstart, Address(rscratch1));
3957   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3958 
3959   bind(L_first_loop);
3960   subsw(idx, idx, 1);
3961   br(Assembler::MI, L_first_loop_exit);
3962   subsw(idx, idx, 1);
3963   br(Assembler::MI, L_one_y);
3964   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3965   ldr(y_idx, Address(rscratch1));
3966   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
3967   bind(L_multiply);
3968 
3969   // AArch64 has a multiply-accumulate instruction that we can't use
3970   // here because it has no way to process carries, so we have to use
3971   // separate add and adc instructions.  Bah.
3972   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
3973   mul(product, x_xstart, y_idx);
3974   adds(product, product, carry);
3975   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
3976 
3977   subw(kdx, kdx, 2);
3978   ror(product, product, 32); // back to big-endian
3979   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
3980 
3981   b(L_first_loop);
3982 
3983   bind(L_one_y);
3984   ldrw(y_idx, Address(y,  0));
3985   b(L_multiply);
3986 
3987   bind(L_one_x);
3988   ldrw(x_xstart, Address(x,  0));
3989   b(L_first_loop);
3990 
3991   bind(L_first_loop_exit);
3992 }
3993 
3994 /**
3995  * Multiply 128 bit by 128. Unrolled inner loop.
3996  *
3997  */
3998 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
3999                                              Register carry, Register carry2,
4000                                              Register idx, Register jdx,
4001                                              Register yz_idx1, Register yz_idx2,
4002                                              Register tmp, Register tmp3, Register tmp4,
4003                                              Register tmp6, Register product_hi) {
4004 
4005   //   jlong carry, x[], y[], z[];
4006   //   int kdx = ystart+1;
4007   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4008   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4009   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4010   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4011   //     carry  = (jlong)(tmp4 >>> 64);
4012   //     z[kdx+idx+1] = (jlong)tmp3;
4013   //     z[kdx+idx] = (jlong)tmp4;
4014   //   }
4015   //   idx += 2;
4016   //   if (idx > 0) {
4017   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4018   //     z[kdx+idx] = (jlong)yz_idx1;
4019   //     carry  = (jlong)(yz_idx1 >>> 64);
4020   //   }
4021   //
4022 
4023   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4024 
4025   lsrw(jdx, idx, 2);
4026 
4027   bind(L_third_loop);
4028 
4029   subsw(jdx, jdx, 1);
4030   br(Assembler::MI, L_third_loop_exit);
4031   subw(idx, idx, 4);
4032 
4033   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4034 
4035   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4036 
4037   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4038 
4039   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4040   ror(yz_idx2, yz_idx2, 32);
4041 
4042   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4043 
4044   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4045   umulh(tmp4, product_hi, yz_idx1);
4046 
4047   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4048   ror(rscratch2, rscratch2, 32);
4049 
4050   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4051   umulh(carry2, product_hi, yz_idx2);
4052 
4053   // propagate sum of both multiplications into carry:tmp4:tmp3
4054   adds(tmp3, tmp3, carry);
4055   adc(tmp4, tmp4, zr);
4056   adds(tmp3, tmp3, rscratch1);
4057   adcs(tmp4, tmp4, tmp);
4058   adc(carry, carry2, zr);
4059   adds(tmp4, tmp4, rscratch2);
4060   adc(carry, carry, zr);
4061 
4062   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4063   ror(tmp4, tmp4, 32);
4064   stp(tmp4, tmp3, Address(tmp6, 0));
4065 
4066   b(L_third_loop);
4067   bind (L_third_loop_exit);
4068 
4069   andw (idx, idx, 0x3);
4070   cbz(idx, L_post_third_loop_done);
4071 
4072   Label L_check_1;
4073   subsw(idx, idx, 2);
4074   br(Assembler::MI, L_check_1);
4075 
4076   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4077   ldr(yz_idx1, Address(rscratch1, 0));
4078   ror(yz_idx1, yz_idx1, 32);
4079   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4080   umulh(tmp4, product_hi, yz_idx1);
4081   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4082   ldr(yz_idx2, Address(rscratch1, 0));
4083   ror(yz_idx2, yz_idx2, 32);
4084 
4085   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4086 
4087   ror(tmp3, tmp3, 32);
4088   str(tmp3, Address(rscratch1, 0));
4089 
4090   bind (L_check_1);
4091 
4092   andw (idx, idx, 0x1);
4093   subsw(idx, idx, 1);
4094   br(Assembler::MI, L_post_third_loop_done);
4095   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4096   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4097   umulh(carry2, tmp4, product_hi);
4098   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4099 
4100   add2_with_carry(carry2, tmp3, tmp4, carry);
4101 
4102   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4103   extr(carry, carry2, tmp3, 32);
4104 
4105   bind(L_post_third_loop_done);
4106 }
4107 
4108 /**
4109  * Code for BigInteger::multiplyToLen() intrinsic.
4110  *
4111  * r0: x
4112  * r1: xlen
4113  * r2: y
4114  * r3: ylen
4115  * r4:  z
4116  * r5: tmp0
4117  * r10: tmp1
4118  * r11: tmp2
4119  * r12: tmp3
4120  * r13: tmp4
4121  * r14: tmp5
4122  * r15: tmp6
4123  * r16: tmp7
4124  *
4125  */
4126 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4127                                      Register z, Register tmp0,
4128                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4129                                      Register tmp5, Register tmp6, Register product_hi) {
4130 
4131   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4132 
4133   const Register idx = tmp1;
4134   const Register kdx = tmp2;
4135   const Register xstart = tmp3;
4136 
4137   const Register y_idx = tmp4;
4138   const Register carry = tmp5;
4139   const Register product  = xlen;
4140   const Register x_xstart = tmp0;
4141 
4142   // First Loop.
4143   //
4144   //  final static long LONG_MASK = 0xffffffffL;
4145   //  int xstart = xlen - 1;
4146   //  int ystart = ylen - 1;
4147   //  long carry = 0;
4148   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4149   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4150   //    z[kdx] = (int)product;
4151   //    carry = product >>> 32;
4152   //  }
4153   //  z[xstart] = (int)carry;
4154   //
4155 
4156   movw(idx, ylen);       // idx = ylen;
4157   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4158   mov(carry, zr);        // carry = 0;
4159 
4160   Label L_done;
4161 
4162   movw(xstart, xlen);
4163   subsw(xstart, xstart, 1);
4164   br(Assembler::MI, L_done);
4165 
4166   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4167 
4168   Label L_second_loop;
4169   cbzw(kdx, L_second_loop);
4170 
4171   Label L_carry;
4172   subw(kdx, kdx, 1);
4173   cbzw(kdx, L_carry);
4174 
4175   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4176   lsr(carry, carry, 32);
4177   subw(kdx, kdx, 1);
4178 
4179   bind(L_carry);
4180   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4181 
4182   // Second and third (nested) loops.
4183   //
4184   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4185   //   carry = 0;
4186   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4187   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4188   //                    (z[k] & LONG_MASK) + carry;
4189   //     z[k] = (int)product;
4190   //     carry = product >>> 32;
4191   //   }
4192   //   z[i] = (int)carry;
4193   // }
4194   //
4195   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4196 
4197   const Register jdx = tmp1;
4198 
4199   bind(L_second_loop);
4200   mov(carry, zr);                // carry = 0;
4201   movw(jdx, ylen);               // j = ystart+1
4202 
4203   subsw(xstart, xstart, 1);      // i = xstart-1;
4204   br(Assembler::MI, L_done);
4205 
4206   str(z, Address(pre(sp, -4 * wordSize)));
4207 
4208   Label L_last_x;
4209   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4210   subsw(xstart, xstart, 1);       // i = xstart-1;
4211   br(Assembler::MI, L_last_x);
4212 
4213   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4214   ldr(product_hi, Address(rscratch1));
4215   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4216 
4217   Label L_third_loop_prologue;
4218   bind(L_third_loop_prologue);
4219 
4220   str(ylen, Address(sp, wordSize));
4221   stp(x, xstart, Address(sp, 2 * wordSize));
4222   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4223                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4224   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4225   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4226 
4227   addw(tmp3, xlen, 1);
4228   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4229   subsw(tmp3, tmp3, 1);
4230   br(Assembler::MI, L_done);
4231 
4232   lsr(carry, carry, 32);
4233   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4234   b(L_second_loop);
4235 
4236   // Next infrequent code is moved outside loops.
4237   bind(L_last_x);
4238   ldrw(product_hi, Address(x,  0));
4239   b(L_third_loop_prologue);
4240 
4241   bind(L_done);
4242 }
4243 
4244 // Code for BigInteger::mulAdd intrinsic
4245 // out     = r0
4246 // in      = r1
4247 // offset  = r2  (already out.length-offset)
4248 // len     = r3
4249 // k       = r4
4250 //
4251 // pseudo code from java implementation:
4252 // carry = 0;
4253 // offset = out.length-offset - 1;
4254 // for (int j=len-1; j >= 0; j--) {
4255 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4256 //     out[offset--] = (int)product;
4257 //     carry = product >>> 32;
4258 // }
4259 // return (int)carry;
4260 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4261       Register len, Register k) {
4262     Label LOOP, END;
4263     // pre-loop
4264     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4265     csel(out, zr, out, Assembler::EQ);
4266     br(Assembler::EQ, END);
4267     add(in, in, len, LSL, 2); // in[j+1] address
4268     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4269     mov(out, zr); // used to keep carry now
4270     BIND(LOOP);
4271     ldrw(rscratch1, Address(pre(in, -4)));
4272     madd(rscratch1, rscratch1, k, out);
4273     ldrw(rscratch2, Address(pre(offset, -4)));
4274     add(rscratch1, rscratch1, rscratch2);
4275     strw(rscratch1, Address(offset));
4276     lsr(out, rscratch1, 32);
4277     subs(len, len, 1);
4278     br(Assembler::NE, LOOP);
4279     BIND(END);
4280 }
4281 
4282 /**
4283  * Emits code to update CRC-32 with a byte value according to constants in table
4284  *
4285  * @param [in,out]crc   Register containing the crc.
4286  * @param [in]val       Register containing the byte to fold into the CRC.
4287  * @param [in]table     Register containing the table of crc constants.
4288  *
4289  * uint32_t crc;
4290  * val = crc_table[(val ^ crc) & 0xFF];
4291  * crc = val ^ (crc >> 8);
4292  *
4293  */
4294 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4295   eor(val, val, crc);
4296   andr(val, val, 0xff);
4297   ldrw(val, Address(table, val, Address::lsl(2)));
4298   eor(crc, val, crc, Assembler::LSR, 8);
4299 }
4300 
4301 /**
4302  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4303  *
4304  * @param [in,out]crc   Register containing the crc.
4305  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4306  * @param [in]table0    Register containing table 0 of crc constants.
4307  * @param [in]table1    Register containing table 1 of crc constants.
4308  * @param [in]table2    Register containing table 2 of crc constants.
4309  * @param [in]table3    Register containing table 3 of crc constants.
4310  *
4311  * uint32_t crc;
4312  *   v = crc ^ v
4313  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4314  *
4315  */
4316 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4317         Register table0, Register table1, Register table2, Register table3,
4318         bool upper) {
4319   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4320   uxtb(tmp, v);
4321   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4322   ubfx(tmp, v, 8, 8);
4323   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4324   eor(crc, crc, tmp);
4325   ubfx(tmp, v, 16, 8);
4326   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4327   eor(crc, crc, tmp);
4328   ubfx(tmp, v, 24, 8);
4329   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4330   eor(crc, crc, tmp);
4331 }
4332 
4333 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4334         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4335     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4336     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4337 
4338     subs(tmp0, len, 384);
4339     mvnw(crc, crc);
4340     br(Assembler::GE, CRC_by128_pre);
4341   BIND(CRC_less128);
4342     subs(len, len, 32);
4343     br(Assembler::GE, CRC_by32_loop);
4344   BIND(CRC_less32);
4345     adds(len, len, 32 - 4);
4346     br(Assembler::GE, CRC_by4_loop);
4347     adds(len, len, 4);
4348     br(Assembler::GT, CRC_by1_loop);
4349     b(L_exit);
4350 
4351   BIND(CRC_by32_loop);
4352     ldp(tmp0, tmp1, Address(buf));
4353     crc32x(crc, crc, tmp0);
4354     ldp(tmp2, tmp3, Address(buf, 16));
4355     crc32x(crc, crc, tmp1);
4356     add(buf, buf, 32);
4357     crc32x(crc, crc, tmp2);
4358     subs(len, len, 32);
4359     crc32x(crc, crc, tmp3);
4360     br(Assembler::GE, CRC_by32_loop);
4361     cmn(len, (u1)32);
4362     br(Assembler::NE, CRC_less32);
4363     b(L_exit);
4364 
4365   BIND(CRC_by4_loop);
4366     ldrw(tmp0, Address(post(buf, 4)));
4367     subs(len, len, 4);
4368     crc32w(crc, crc, tmp0);
4369     br(Assembler::GE, CRC_by4_loop);
4370     adds(len, len, 4);
4371     br(Assembler::LE, L_exit);
4372   BIND(CRC_by1_loop);
4373     ldrb(tmp0, Address(post(buf, 1)));
4374     subs(len, len, 1);
4375     crc32b(crc, crc, tmp0);
4376     br(Assembler::GT, CRC_by1_loop);
4377     b(L_exit);
4378 
4379   BIND(CRC_by128_pre);
4380     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4381       4*256*sizeof(juint) + 8*sizeof(juint));
4382     mov(crc, 0);
4383     crc32x(crc, crc, tmp0);
4384     crc32x(crc, crc, tmp1);
4385 
4386     cbnz(len, CRC_less128);
4387 
4388   BIND(L_exit);
4389     mvnw(crc, crc);
4390 }
4391 
4392 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4393         Register len, Register tmp0, Register tmp1, Register tmp2,
4394         Register tmp3) {
4395     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4396     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4397 
4398     mvnw(crc, crc);
4399 
4400     subs(len, len, 128);
4401     br(Assembler::GE, CRC_by64_pre);
4402   BIND(CRC_less64);
4403     adds(len, len, 128-32);
4404     br(Assembler::GE, CRC_by32_loop);
4405   BIND(CRC_less32);
4406     adds(len, len, 32-4);
4407     br(Assembler::GE, CRC_by4_loop);
4408     adds(len, len, 4);
4409     br(Assembler::GT, CRC_by1_loop);
4410     b(L_exit);
4411 
4412   BIND(CRC_by32_loop);
4413     ldp(tmp0, tmp1, Address(post(buf, 16)));
4414     subs(len, len, 32);
4415     crc32x(crc, crc, tmp0);
4416     ldr(tmp2, Address(post(buf, 8)));
4417     crc32x(crc, crc, tmp1);
4418     ldr(tmp3, Address(post(buf, 8)));
4419     crc32x(crc, crc, tmp2);
4420     crc32x(crc, crc, tmp3);
4421     br(Assembler::GE, CRC_by32_loop);
4422     cmn(len, (u1)32);
4423     br(Assembler::NE, CRC_less32);
4424     b(L_exit);
4425 
4426   BIND(CRC_by4_loop);
4427     ldrw(tmp0, Address(post(buf, 4)));
4428     subs(len, len, 4);
4429     crc32w(crc, crc, tmp0);
4430     br(Assembler::GE, CRC_by4_loop);
4431     adds(len, len, 4);
4432     br(Assembler::LE, L_exit);
4433   BIND(CRC_by1_loop);
4434     ldrb(tmp0, Address(post(buf, 1)));
4435     subs(len, len, 1);
4436     crc32b(crc, crc, tmp0);
4437     br(Assembler::GT, CRC_by1_loop);
4438     b(L_exit);
4439 
4440   BIND(CRC_by64_pre);
4441     sub(buf, buf, 8);
4442     ldp(tmp0, tmp1, Address(buf, 8));
4443     crc32x(crc, crc, tmp0);
4444     ldr(tmp2, Address(buf, 24));
4445     crc32x(crc, crc, tmp1);
4446     ldr(tmp3, Address(buf, 32));
4447     crc32x(crc, crc, tmp2);
4448     ldr(tmp0, Address(buf, 40));
4449     crc32x(crc, crc, tmp3);
4450     ldr(tmp1, Address(buf, 48));
4451     crc32x(crc, crc, tmp0);
4452     ldr(tmp2, Address(buf, 56));
4453     crc32x(crc, crc, tmp1);
4454     ldr(tmp3, Address(pre(buf, 64)));
4455 
4456     b(CRC_by64_loop);
4457 
4458     align(CodeEntryAlignment);
4459   BIND(CRC_by64_loop);
4460     subs(len, len, 64);
4461     crc32x(crc, crc, tmp2);
4462     ldr(tmp0, Address(buf, 8));
4463     crc32x(crc, crc, tmp3);
4464     ldr(tmp1, Address(buf, 16));
4465     crc32x(crc, crc, tmp0);
4466     ldr(tmp2, Address(buf, 24));
4467     crc32x(crc, crc, tmp1);
4468     ldr(tmp3, Address(buf, 32));
4469     crc32x(crc, crc, tmp2);
4470     ldr(tmp0, Address(buf, 40));
4471     crc32x(crc, crc, tmp3);
4472     ldr(tmp1, Address(buf, 48));
4473     crc32x(crc, crc, tmp0);
4474     ldr(tmp2, Address(buf, 56));
4475     crc32x(crc, crc, tmp1);
4476     ldr(tmp3, Address(pre(buf, 64)));
4477     br(Assembler::GE, CRC_by64_loop);
4478 
4479     // post-loop
4480     crc32x(crc, crc, tmp2);
4481     crc32x(crc, crc, tmp3);
4482 
4483     sub(len, len, 64);
4484     add(buf, buf, 8);
4485     cmn(len, (u1)128);
4486     br(Assembler::NE, CRC_less64);
4487   BIND(L_exit);
4488     mvnw(crc, crc);
4489 }
4490 
4491 /**
4492  * @param crc   register containing existing CRC (32-bit)
4493  * @param buf   register pointing to input byte buffer (byte*)
4494  * @param len   register containing number of bytes
4495  * @param table register that will contain address of CRC table
4496  * @param tmp   scratch register
4497  */
4498 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4499         Register table0, Register table1, Register table2, Register table3,
4500         Register tmp, Register tmp2, Register tmp3) {
4501   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4502 
4503   if (UseCryptoPmullForCRC32) {
4504       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4505       return;
4506   }
4507 
4508   if (UseCRC32) {
4509       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4510       return;
4511   }
4512 
4513     mvnw(crc, crc);
4514 
4515     {
4516       uint64_t offset;
4517       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4518       add(table0, table0, offset);
4519     }
4520     add(table1, table0, 1*256*sizeof(juint));
4521     add(table2, table0, 2*256*sizeof(juint));
4522     add(table3, table0, 3*256*sizeof(juint));
4523 
4524     { // Neon code start
4525       cmp(len, (u1)64);
4526       br(Assembler::LT, L_by16);
4527       eor(v16, T16B, v16, v16);
4528 
4529     Label L_fold;
4530 
4531       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4532 
4533       ld1(v0, v1, T2D, post(buf, 32));
4534       ld1r(v4, T2D, post(tmp, 8));
4535       ld1r(v5, T2D, post(tmp, 8));
4536       ld1r(v6, T2D, post(tmp, 8));
4537       ld1r(v7, T2D, post(tmp, 8));
4538       mov(v16, S, 0, crc);
4539 
4540       eor(v0, T16B, v0, v16);
4541       sub(len, len, 64);
4542 
4543     BIND(L_fold);
4544       pmull(v22, T8H, v0, v5, T8B);
4545       pmull(v20, T8H, v0, v7, T8B);
4546       pmull(v23, T8H, v0, v4, T8B);
4547       pmull(v21, T8H, v0, v6, T8B);
4548 
4549       pmull2(v18, T8H, v0, v5, T16B);
4550       pmull2(v16, T8H, v0, v7, T16B);
4551       pmull2(v19, T8H, v0, v4, T16B);
4552       pmull2(v17, T8H, v0, v6, T16B);
4553 
4554       uzp1(v24, T8H, v20, v22);
4555       uzp2(v25, T8H, v20, v22);
4556       eor(v20, T16B, v24, v25);
4557 
4558       uzp1(v26, T8H, v16, v18);
4559       uzp2(v27, T8H, v16, v18);
4560       eor(v16, T16B, v26, v27);
4561 
4562       ushll2(v22, T4S, v20, T8H, 8);
4563       ushll(v20, T4S, v20, T4H, 8);
4564 
4565       ushll2(v18, T4S, v16, T8H, 8);
4566       ushll(v16, T4S, v16, T4H, 8);
4567 
4568       eor(v22, T16B, v23, v22);
4569       eor(v18, T16B, v19, v18);
4570       eor(v20, T16B, v21, v20);
4571       eor(v16, T16B, v17, v16);
4572 
4573       uzp1(v17, T2D, v16, v20);
4574       uzp2(v21, T2D, v16, v20);
4575       eor(v17, T16B, v17, v21);
4576 
4577       ushll2(v20, T2D, v17, T4S, 16);
4578       ushll(v16, T2D, v17, T2S, 16);
4579 
4580       eor(v20, T16B, v20, v22);
4581       eor(v16, T16B, v16, v18);
4582 
4583       uzp1(v17, T2D, v20, v16);
4584       uzp2(v21, T2D, v20, v16);
4585       eor(v28, T16B, v17, v21);
4586 
4587       pmull(v22, T8H, v1, v5, T8B);
4588       pmull(v20, T8H, v1, v7, T8B);
4589       pmull(v23, T8H, v1, v4, T8B);
4590       pmull(v21, T8H, v1, v6, T8B);
4591 
4592       pmull2(v18, T8H, v1, v5, T16B);
4593       pmull2(v16, T8H, v1, v7, T16B);
4594       pmull2(v19, T8H, v1, v4, T16B);
4595       pmull2(v17, T8H, v1, v6, T16B);
4596 
4597       ld1(v0, v1, T2D, post(buf, 32));
4598 
4599       uzp1(v24, T8H, v20, v22);
4600       uzp2(v25, T8H, v20, v22);
4601       eor(v20, T16B, v24, v25);
4602 
4603       uzp1(v26, T8H, v16, v18);
4604       uzp2(v27, T8H, v16, v18);
4605       eor(v16, T16B, v26, v27);
4606 
4607       ushll2(v22, T4S, v20, T8H, 8);
4608       ushll(v20, T4S, v20, T4H, 8);
4609 
4610       ushll2(v18, T4S, v16, T8H, 8);
4611       ushll(v16, T4S, v16, T4H, 8);
4612 
4613       eor(v22, T16B, v23, v22);
4614       eor(v18, T16B, v19, v18);
4615       eor(v20, T16B, v21, v20);
4616       eor(v16, T16B, v17, v16);
4617 
4618       uzp1(v17, T2D, v16, v20);
4619       uzp2(v21, T2D, v16, v20);
4620       eor(v16, T16B, v17, v21);
4621 
4622       ushll2(v20, T2D, v16, T4S, 16);
4623       ushll(v16, T2D, v16, T2S, 16);
4624 
4625       eor(v20, T16B, v22, v20);
4626       eor(v16, T16B, v16, v18);
4627 
4628       uzp1(v17, T2D, v20, v16);
4629       uzp2(v21, T2D, v20, v16);
4630       eor(v20, T16B, v17, v21);
4631 
4632       shl(v16, T2D, v28, 1);
4633       shl(v17, T2D, v20, 1);
4634 
4635       eor(v0, T16B, v0, v16);
4636       eor(v1, T16B, v1, v17);
4637 
4638       subs(len, len, 32);
4639       br(Assembler::GE, L_fold);
4640 
4641       mov(crc, 0);
4642       mov(tmp, v0, D, 0);
4643       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4644       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4645       mov(tmp, v0, D, 1);
4646       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4647       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4648       mov(tmp, v1, D, 0);
4649       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4650       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4651       mov(tmp, v1, D, 1);
4652       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4653       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4654 
4655       add(len, len, 32);
4656     } // Neon code end
4657 
4658   BIND(L_by16);
4659     subs(len, len, 16);
4660     br(Assembler::GE, L_by16_loop);
4661     adds(len, len, 16-4);
4662     br(Assembler::GE, L_by4_loop);
4663     adds(len, len, 4);
4664     br(Assembler::GT, L_by1_loop);
4665     b(L_exit);
4666 
4667   BIND(L_by4_loop);
4668     ldrw(tmp, Address(post(buf, 4)));
4669     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4670     subs(len, len, 4);
4671     br(Assembler::GE, L_by4_loop);
4672     adds(len, len, 4);
4673     br(Assembler::LE, L_exit);
4674   BIND(L_by1_loop);
4675     subs(len, len, 1);
4676     ldrb(tmp, Address(post(buf, 1)));
4677     update_byte_crc32(crc, tmp, table0);
4678     br(Assembler::GT, L_by1_loop);
4679     b(L_exit);
4680 
4681     align(CodeEntryAlignment);
4682   BIND(L_by16_loop);
4683     subs(len, len, 16);
4684     ldp(tmp, tmp3, Address(post(buf, 16)));
4685     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4686     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4687     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4688     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4689     br(Assembler::GE, L_by16_loop);
4690     adds(len, len, 16-4);
4691     br(Assembler::GE, L_by4_loop);
4692     adds(len, len, 4);
4693     br(Assembler::GT, L_by1_loop);
4694   BIND(L_exit);
4695     mvnw(crc, crc);
4696 }
4697 
4698 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4699         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4700     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4701     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4702 
4703     subs(tmp0, len, 384);
4704     br(Assembler::GE, CRC_by128_pre);
4705   BIND(CRC_less128);
4706     subs(len, len, 32);
4707     br(Assembler::GE, CRC_by32_loop);
4708   BIND(CRC_less32);
4709     adds(len, len, 32 - 4);
4710     br(Assembler::GE, CRC_by4_loop);
4711     adds(len, len, 4);
4712     br(Assembler::GT, CRC_by1_loop);
4713     b(L_exit);
4714 
4715   BIND(CRC_by32_loop);
4716     ldp(tmp0, tmp1, Address(buf));
4717     crc32cx(crc, crc, tmp0);
4718     ldr(tmp2, Address(buf, 16));
4719     crc32cx(crc, crc, tmp1);
4720     ldr(tmp3, Address(buf, 24));
4721     crc32cx(crc, crc, tmp2);
4722     add(buf, buf, 32);
4723     subs(len, len, 32);
4724     crc32cx(crc, crc, tmp3);
4725     br(Assembler::GE, CRC_by32_loop);
4726     cmn(len, (u1)32);
4727     br(Assembler::NE, CRC_less32);
4728     b(L_exit);
4729 
4730   BIND(CRC_by4_loop);
4731     ldrw(tmp0, Address(post(buf, 4)));
4732     subs(len, len, 4);
4733     crc32cw(crc, crc, tmp0);
4734     br(Assembler::GE, CRC_by4_loop);
4735     adds(len, len, 4);
4736     br(Assembler::LE, L_exit);
4737   BIND(CRC_by1_loop);
4738     ldrb(tmp0, Address(post(buf, 1)));
4739     subs(len, len, 1);
4740     crc32cb(crc, crc, tmp0);
4741     br(Assembler::GT, CRC_by1_loop);
4742     b(L_exit);
4743 
4744   BIND(CRC_by128_pre);
4745     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4746       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4747     mov(crc, 0);
4748     crc32cx(crc, crc, tmp0);
4749     crc32cx(crc, crc, tmp1);
4750 
4751     cbnz(len, CRC_less128);
4752 
4753   BIND(L_exit);
4754 }
4755 
4756 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4757         Register len, Register tmp0, Register tmp1, Register tmp2,
4758         Register tmp3) {
4759     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4760     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4761 
4762     subs(len, len, 128);
4763     br(Assembler::GE, CRC_by64_pre);
4764   BIND(CRC_less64);
4765     adds(len, len, 128-32);
4766     br(Assembler::GE, CRC_by32_loop);
4767   BIND(CRC_less32);
4768     adds(len, len, 32-4);
4769     br(Assembler::GE, CRC_by4_loop);
4770     adds(len, len, 4);
4771     br(Assembler::GT, CRC_by1_loop);
4772     b(L_exit);
4773 
4774   BIND(CRC_by32_loop);
4775     ldp(tmp0, tmp1, Address(post(buf, 16)));
4776     subs(len, len, 32);
4777     crc32cx(crc, crc, tmp0);
4778     ldr(tmp2, Address(post(buf, 8)));
4779     crc32cx(crc, crc, tmp1);
4780     ldr(tmp3, Address(post(buf, 8)));
4781     crc32cx(crc, crc, tmp2);
4782     crc32cx(crc, crc, tmp3);
4783     br(Assembler::GE, CRC_by32_loop);
4784     cmn(len, (u1)32);
4785     br(Assembler::NE, CRC_less32);
4786     b(L_exit);
4787 
4788   BIND(CRC_by4_loop);
4789     ldrw(tmp0, Address(post(buf, 4)));
4790     subs(len, len, 4);
4791     crc32cw(crc, crc, tmp0);
4792     br(Assembler::GE, CRC_by4_loop);
4793     adds(len, len, 4);
4794     br(Assembler::LE, L_exit);
4795   BIND(CRC_by1_loop);
4796     ldrb(tmp0, Address(post(buf, 1)));
4797     subs(len, len, 1);
4798     crc32cb(crc, crc, tmp0);
4799     br(Assembler::GT, CRC_by1_loop);
4800     b(L_exit);
4801 
4802   BIND(CRC_by64_pre);
4803     sub(buf, buf, 8);
4804     ldp(tmp0, tmp1, Address(buf, 8));
4805     crc32cx(crc, crc, tmp0);
4806     ldr(tmp2, Address(buf, 24));
4807     crc32cx(crc, crc, tmp1);
4808     ldr(tmp3, Address(buf, 32));
4809     crc32cx(crc, crc, tmp2);
4810     ldr(tmp0, Address(buf, 40));
4811     crc32cx(crc, crc, tmp3);
4812     ldr(tmp1, Address(buf, 48));
4813     crc32cx(crc, crc, tmp0);
4814     ldr(tmp2, Address(buf, 56));
4815     crc32cx(crc, crc, tmp1);
4816     ldr(tmp3, Address(pre(buf, 64)));
4817 
4818     b(CRC_by64_loop);
4819 
4820     align(CodeEntryAlignment);
4821   BIND(CRC_by64_loop);
4822     subs(len, len, 64);
4823     crc32cx(crc, crc, tmp2);
4824     ldr(tmp0, Address(buf, 8));
4825     crc32cx(crc, crc, tmp3);
4826     ldr(tmp1, Address(buf, 16));
4827     crc32cx(crc, crc, tmp0);
4828     ldr(tmp2, Address(buf, 24));
4829     crc32cx(crc, crc, tmp1);
4830     ldr(tmp3, Address(buf, 32));
4831     crc32cx(crc, crc, tmp2);
4832     ldr(tmp0, Address(buf, 40));
4833     crc32cx(crc, crc, tmp3);
4834     ldr(tmp1, Address(buf, 48));
4835     crc32cx(crc, crc, tmp0);
4836     ldr(tmp2, Address(buf, 56));
4837     crc32cx(crc, crc, tmp1);
4838     ldr(tmp3, Address(pre(buf, 64)));
4839     br(Assembler::GE, CRC_by64_loop);
4840 
4841     // post-loop
4842     crc32cx(crc, crc, tmp2);
4843     crc32cx(crc, crc, tmp3);
4844 
4845     sub(len, len, 64);
4846     add(buf, buf, 8);
4847     cmn(len, (u1)128);
4848     br(Assembler::NE, CRC_less64);
4849   BIND(L_exit);
4850 }
4851 
4852 /**
4853  * @param crc   register containing existing CRC (32-bit)
4854  * @param buf   register pointing to input byte buffer (byte*)
4855  * @param len   register containing number of bytes
4856  * @param table register that will contain address of CRC table
4857  * @param tmp   scratch register
4858  */
4859 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4860         Register table0, Register table1, Register table2, Register table3,
4861         Register tmp, Register tmp2, Register tmp3) {
4862   if (UseCryptoPmullForCRC32) {
4863     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4864   } else {
4865     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4866   }
4867 }
4868 
4869 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4870         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4871     Label CRC_by128_loop;
4872     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4873 
4874     sub(len, len, 256);
4875     Register table = tmp0;
4876     {
4877       uint64_t offset;
4878       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4879       add(table, table, offset);
4880     }
4881     add(table, table, table_offset);
4882 
4883     // Registers v0..v7 are used as data registers.
4884     // Registers v16..v31 are used as tmp registers.
4885     sub(buf, buf, 0x10);
4886     ldrq(v0, Address(buf, 0x10));
4887     ldrq(v1, Address(buf, 0x20));
4888     ldrq(v2, Address(buf, 0x30));
4889     ldrq(v3, Address(buf, 0x40));
4890     ldrq(v4, Address(buf, 0x50));
4891     ldrq(v5, Address(buf, 0x60));
4892     ldrq(v6, Address(buf, 0x70));
4893     ldrq(v7, Address(pre(buf, 0x80)));
4894 
4895     movi(v31, T4S, 0);
4896     mov(v31, S, 0, crc);
4897     eor(v0, T16B, v0, v31);
4898 
4899     // Register v16 contains constants from the crc table.
4900     ldrq(v16, Address(table));
4901     b(CRC_by128_loop);
4902 
4903     align(OptoLoopAlignment);
4904   BIND(CRC_by128_loop);
4905     pmull (v17,  T1Q, v0, v16, T1D);
4906     pmull2(v18, T1Q, v0, v16, T2D);
4907     ldrq(v0, Address(buf, 0x10));
4908     eor3(v0, T16B, v17,  v18, v0);
4909 
4910     pmull (v19, T1Q, v1, v16, T1D);
4911     pmull2(v20, T1Q, v1, v16, T2D);
4912     ldrq(v1, Address(buf, 0x20));
4913     eor3(v1, T16B, v19, v20, v1);
4914 
4915     pmull (v21, T1Q, v2, v16, T1D);
4916     pmull2(v22, T1Q, v2, v16, T2D);
4917     ldrq(v2, Address(buf, 0x30));
4918     eor3(v2, T16B, v21, v22, v2);
4919 
4920     pmull (v23, T1Q, v3, v16, T1D);
4921     pmull2(v24, T1Q, v3, v16, T2D);
4922     ldrq(v3, Address(buf, 0x40));
4923     eor3(v3, T16B, v23, v24, v3);
4924 
4925     pmull (v25, T1Q, v4, v16, T1D);
4926     pmull2(v26, T1Q, v4, v16, T2D);
4927     ldrq(v4, Address(buf, 0x50));
4928     eor3(v4, T16B, v25, v26, v4);
4929 
4930     pmull (v27, T1Q, v5, v16, T1D);
4931     pmull2(v28, T1Q, v5, v16, T2D);
4932     ldrq(v5, Address(buf, 0x60));
4933     eor3(v5, T16B, v27, v28, v5);
4934 
4935     pmull (v29, T1Q, v6, v16, T1D);
4936     pmull2(v30, T1Q, v6, v16, T2D);
4937     ldrq(v6, Address(buf, 0x70));
4938     eor3(v6, T16B, v29, v30, v6);
4939 
4940     // Reuse registers v23, v24.
4941     // Using them won't block the first instruction of the next iteration.
4942     pmull (v23, T1Q, v7, v16, T1D);
4943     pmull2(v24, T1Q, v7, v16, T2D);
4944     ldrq(v7, Address(pre(buf, 0x80)));
4945     eor3(v7, T16B, v23, v24, v7);
4946 
4947     subs(len, len, 0x80);
4948     br(Assembler::GE, CRC_by128_loop);
4949 
4950     // fold into 512 bits
4951     // Use v31 for constants because v16 can be still in use.
4952     ldrq(v31, Address(table, 0x10));
4953 
4954     pmull (v17,  T1Q, v0, v31, T1D);
4955     pmull2(v18, T1Q, v0, v31, T2D);
4956     eor3(v0, T16B, v17, v18, v4);
4957 
4958     pmull (v19, T1Q, v1, v31, T1D);
4959     pmull2(v20, T1Q, v1, v31, T2D);
4960     eor3(v1, T16B, v19, v20, v5);
4961 
4962     pmull (v21, T1Q, v2, v31, T1D);
4963     pmull2(v22, T1Q, v2, v31, T2D);
4964     eor3(v2, T16B, v21, v22, v6);
4965 
4966     pmull (v23, T1Q, v3, v31, T1D);
4967     pmull2(v24, T1Q, v3, v31, T2D);
4968     eor3(v3, T16B, v23, v24, v7);
4969 
4970     // fold into 128 bits
4971     // Use v17 for constants because v31 can be still in use.
4972     ldrq(v17, Address(table, 0x20));
4973     pmull (v25, T1Q, v0, v17, T1D);
4974     pmull2(v26, T1Q, v0, v17, T2D);
4975     eor3(v3, T16B, v3, v25, v26);
4976 
4977     // Use v18 for constants because v17 can be still in use.
4978     ldrq(v18, Address(table, 0x30));
4979     pmull (v27, T1Q, v1, v18, T1D);
4980     pmull2(v28, T1Q, v1, v18, T2D);
4981     eor3(v3, T16B, v3, v27, v28);
4982 
4983     // Use v19 for constants because v18 can be still in use.
4984     ldrq(v19, Address(table, 0x40));
4985     pmull (v29, T1Q, v2, v19, T1D);
4986     pmull2(v30, T1Q, v2, v19, T2D);
4987     eor3(v0, T16B, v3, v29, v30);
4988 
4989     add(len, len, 0x80);
4990     add(buf, buf, 0x10);
4991 
4992     mov(tmp0, v0, D, 0);
4993     mov(tmp1, v0, D, 1);
4994 }
4995 
4996 void MacroAssembler::addptr(const Address &dst, int32_t src) {
4997   Address adr;
4998   switch(dst.getMode()) {
4999   case Address::base_plus_offset:
5000     // This is the expected mode, although we allow all the other
5001     // forms below.
5002     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5003     break;
5004   default:
5005     lea(rscratch2, dst);
5006     adr = Address(rscratch2);
5007     break;
5008   }
5009   ldr(rscratch1, adr);
5010   add(rscratch1, rscratch1, src);
5011   str(rscratch1, adr);
5012 }
5013 
5014 void MacroAssembler::cmpptr(Register src1, Address src2) {
5015   uint64_t offset;
5016   adrp(rscratch1, src2, offset);
5017   ldr(rscratch1, Address(rscratch1, offset));
5018   cmp(src1, rscratch1);
5019 }
5020 
5021 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5022   cmp(obj1, obj2);
5023 }
5024 
5025 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5026   load_method_holder(rresult, rmethod);
5027   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5028 }
5029 
5030 void MacroAssembler::load_method_holder(Register holder, Register method) {
5031   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5032   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5033   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5034 }
5035 
5036 // Loads the obj's Klass* into dst.
5037 // Preserves all registers (incl src, rscratch1 and rscratch2).
5038 // Input:
5039 // src - the oop we want to load the klass from.
5040 // dst - output narrow klass.
5041 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5042   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5043   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5044   lsr(dst, dst, markWord::klass_shift);
5045 }
5046 
5047 void MacroAssembler::load_klass(Register dst, Register src) {
5048   if (UseCompactObjectHeaders) {
5049     load_narrow_klass_compact(dst, src);
5050     decode_klass_not_null(dst);
5051   } else if (UseCompressedClassPointers) {
5052     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5053     decode_klass_not_null(dst);
5054   } else {
5055     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5056   }
5057 }
5058 
5059 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5060   if (RestoreMXCSROnJNICalls) {
5061     Label OK;
5062     get_fpcr(tmp1);
5063     mov(tmp2, tmp1);
5064     // Set FPCR to the state we need. We do want Round to Nearest. We
5065     // don't want non-IEEE rounding modes or floating-point traps.
5066     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5067     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5068     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5069     eor(tmp2, tmp1, tmp2);
5070     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5071     set_fpcr(tmp1);
5072     bind(OK);
5073   }
5074 }
5075 
5076 // ((OopHandle)result).resolve();
5077 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5078   // OopHandle::resolve is an indirection.
5079   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5080 }
5081 
5082 // ((WeakHandle)result).resolve();
5083 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5084   assert_different_registers(result, tmp1, tmp2);
5085   Label resolved;
5086 
5087   // A null weak handle resolves to null.
5088   cbz(result, resolved);
5089 
5090   // Only 64 bit platforms support GCs that require a tmp register
5091   // WeakHandle::resolve is an indirection like jweak.
5092   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5093                  result, Address(result), tmp1, tmp2);
5094   bind(resolved);
5095 }
5096 
5097 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5098   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5099   ldr(dst, Address(rmethod, Method::const_offset()));
5100   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5101   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5102   ldr(dst, Address(dst, mirror_offset));
5103   resolve_oop_handle(dst, tmp1, tmp2);
5104 }
5105 
5106 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5107   assert_different_registers(obj, klass, tmp);
5108   if (UseCompressedClassPointers) {
5109     if (UseCompactObjectHeaders) {
5110       load_narrow_klass_compact(tmp, obj);
5111     } else {
5112       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5113     }
5114     if (CompressedKlassPointers::base() == nullptr) {
5115       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5116       return;
5117     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5118                && CompressedKlassPointers::shift() == 0) {
5119       // Only the bottom 32 bits matter
5120       cmpw(klass, tmp);
5121       return;
5122     }
5123     decode_klass_not_null(tmp);
5124   } else {
5125     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5126   }
5127   cmp(klass, tmp);
5128 }
5129 
5130 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5131   if (UseCompactObjectHeaders) {
5132     load_narrow_klass_compact(tmp1, obj1);
5133     load_narrow_klass_compact(tmp2,  obj2);
5134     cmpw(tmp1, tmp2);
5135   } else if (UseCompressedClassPointers) {
5136     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5137     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5138     cmpw(tmp1, tmp2);
5139   } else {
5140     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5141     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5142     cmp(tmp1, tmp2);
5143   }
5144 }
5145 
5146 void MacroAssembler::store_klass(Register dst, Register src) {
5147   // FIXME: Should this be a store release?  concurrent gcs assumes
5148   // klass length is valid if klass field is not null.
5149   assert(!UseCompactObjectHeaders, "not with compact headers");
5150   if (UseCompressedClassPointers) {
5151     encode_klass_not_null(src);
5152     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5153   } else {
5154     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5155   }
5156 }
5157 
5158 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5159   assert(!UseCompactObjectHeaders, "not with compact headers");
5160   if (UseCompressedClassPointers) {
5161     // Store to klass gap in destination
5162     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5163   }
5164 }
5165 
5166 // Algorithm must match CompressedOops::encode.
5167 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5168 #ifdef ASSERT
5169   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5170 #endif
5171   verify_oop_msg(s, "broken oop in encode_heap_oop");
5172   if (CompressedOops::base() == nullptr) {
5173     if (CompressedOops::shift() != 0) {
5174       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5175       lsr(d, s, LogMinObjAlignmentInBytes);
5176     } else {
5177       mov(d, s);
5178     }
5179   } else {
5180     subs(d, s, rheapbase);
5181     csel(d, d, zr, Assembler::HS);
5182     lsr(d, d, LogMinObjAlignmentInBytes);
5183 
5184     /*  Old algorithm: is this any worse?
5185     Label nonnull;
5186     cbnz(r, nonnull);
5187     sub(r, r, rheapbase);
5188     bind(nonnull);
5189     lsr(r, r, LogMinObjAlignmentInBytes);
5190     */
5191   }
5192 }
5193 
5194 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5195 #ifdef ASSERT
5196   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5197   if (CheckCompressedOops) {
5198     Label ok;
5199     cbnz(r, ok);
5200     stop("null oop passed to encode_heap_oop_not_null");
5201     bind(ok);
5202   }
5203 #endif
5204   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5205   if (CompressedOops::base() != nullptr) {
5206     sub(r, r, rheapbase);
5207   }
5208   if (CompressedOops::shift() != 0) {
5209     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5210     lsr(r, r, LogMinObjAlignmentInBytes);
5211   }
5212 }
5213 
5214 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5215 #ifdef ASSERT
5216   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5217   if (CheckCompressedOops) {
5218     Label ok;
5219     cbnz(src, ok);
5220     stop("null oop passed to encode_heap_oop_not_null2");
5221     bind(ok);
5222   }
5223 #endif
5224   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5225 
5226   Register data = src;
5227   if (CompressedOops::base() != nullptr) {
5228     sub(dst, src, rheapbase);
5229     data = dst;
5230   }
5231   if (CompressedOops::shift() != 0) {
5232     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5233     lsr(dst, data, LogMinObjAlignmentInBytes);
5234     data = dst;
5235   }
5236   if (data == src)
5237     mov(dst, src);
5238 }
5239 
5240 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5241 #ifdef ASSERT
5242   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5243 #endif
5244   if (CompressedOops::base() == nullptr) {
5245     if (CompressedOops::shift() != 0) {
5246       lsl(d, s, CompressedOops::shift());
5247     } else if (d != s) {
5248       mov(d, s);
5249     }
5250   } else {
5251     Label done;
5252     if (d != s)
5253       mov(d, s);
5254     cbz(s, done);
5255     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5256     bind(done);
5257   }
5258   verify_oop_msg(d, "broken oop in decode_heap_oop");
5259 }
5260 
5261 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5262   assert (UseCompressedOops, "should only be used for compressed headers");
5263   assert (Universe::heap() != nullptr, "java heap should be initialized");
5264   // Cannot assert, unverified entry point counts instructions (see .ad file)
5265   // vtableStubs also counts instructions in pd_code_size_limit.
5266   // Also do not verify_oop as this is called by verify_oop.
5267   if (CompressedOops::shift() != 0) {
5268     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5269     if (CompressedOops::base() != nullptr) {
5270       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5271     } else {
5272       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5273     }
5274   } else {
5275     assert (CompressedOops::base() == nullptr, "sanity");
5276   }
5277 }
5278 
5279 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5280   assert (UseCompressedOops, "should only be used for compressed headers");
5281   assert (Universe::heap() != nullptr, "java heap should be initialized");
5282   // Cannot assert, unverified entry point counts instructions (see .ad file)
5283   // vtableStubs also counts instructions in pd_code_size_limit.
5284   // Also do not verify_oop as this is called by verify_oop.
5285   if (CompressedOops::shift() != 0) {
5286     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5287     if (CompressedOops::base() != nullptr) {
5288       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5289     } else {
5290       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5291     }
5292   } else {
5293     assert (CompressedOops::base() == nullptr, "sanity");
5294     if (dst != src) {
5295       mov(dst, src);
5296     }
5297   }
5298 }
5299 
5300 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5301 
5302 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5303   assert(Metaspace::initialized(), "metaspace not initialized yet");
5304   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5305   return _klass_decode_mode;
5306 }
5307 
5308 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5309   assert(UseCompressedClassPointers, "not using compressed class pointers");
5310 
5311   // KlassDecodeMode shouldn't be set already.
5312   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5313 
5314   if (base == nullptr) {
5315     return KlassDecodeZero;
5316   }
5317 
5318   if (operand_valid_for_logical_immediate(
5319         /*is32*/false, (uint64_t)base)) {
5320     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5321     if (((uint64_t)base & range_mask) == 0) {
5322       return KlassDecodeXor;
5323     }
5324   }
5325 
5326   const uint64_t shifted_base =
5327     (uint64_t)base >> shift;
5328   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5329     return KlassDecodeMovk;
5330   }
5331 
5332   // No valid encoding.
5333   return KlassDecodeNone;
5334 }
5335 
5336 // Check if one of the above decoding modes will work for given base, shift and range.
5337 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5338   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5339 }
5340 
5341 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5342   _klass_decode_mode = klass_decode_mode(base, shift, range);
5343   return _klass_decode_mode != KlassDecodeNone;
5344 }
5345 
5346 static Register pick_different_tmp(Register dst, Register src) {
5347   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5348   return *tmps.begin();
5349 }
5350 
5351 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5352   // we have to load the klass base from the AOT constants area but
5353   // not the shift because it is not allowed to change
5354   int shift = CompressedKlassPointers::shift();
5355   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5356   if (dst != src) {
5357     // we can load the base into dst, subtract it formthe src and shift down
5358     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5359     ldr(dst, dst);
5360     sub(dst, src, dst);
5361     lsr(dst, dst, shift);
5362   } else {
5363     // we need an extra register in order to load the coop base
5364     Register tmp = pick_different_tmp(dst, src);
5365     RegSet regs = RegSet::of(tmp);
5366     push(regs, sp);
5367     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5368     ldr(tmp, tmp);
5369     sub(dst, src, tmp);
5370     lsr(dst, dst, shift);
5371     pop(regs, sp);
5372   }
5373 }
5374 
5375 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5376   if (AOTCodeCache::is_on_for_dump()) {
5377     encode_klass_not_null_for_aot(dst, src);
5378     return;
5379   }
5380 
5381   switch (klass_decode_mode()) {
5382   case KlassDecodeZero:
5383     if (CompressedKlassPointers::shift() != 0) {
5384       lsr(dst, src, CompressedKlassPointers::shift());
5385     } else {
5386       if (dst != src) mov(dst, src);
5387     }
5388     break;
5389 
5390   case KlassDecodeXor:
5391     if (CompressedKlassPointers::shift() != 0) {
5392       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5393       lsr(dst, dst, CompressedKlassPointers::shift());
5394     } else {
5395       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5396     }
5397     break;
5398 
5399   case KlassDecodeMovk:
5400     if (CompressedKlassPointers::shift() != 0) {
5401       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5402     } else {
5403       movw(dst, src);
5404     }
5405     break;
5406 
5407   case KlassDecodeNone:
5408     ShouldNotReachHere();
5409     break;
5410   }
5411 }
5412 
5413 void MacroAssembler::encode_klass_not_null(Register r) {
5414   encode_klass_not_null(r, r);
5415 }
5416 
5417 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5418   // we have to load the klass base from the AOT constants area but
5419   // not the shift because it is not allowed to change
5420   int shift = CompressedKlassPointers::shift();
5421   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5422   if (dst != src) {
5423     // we can load the base into dst then add the offset with a suitable shift
5424     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5425     ldr(dst, dst);
5426     add(dst, dst, src, LSL,  shift);
5427   } else {
5428     // we need an extra register in order to load the coop base
5429     Register tmp = pick_different_tmp(dst, src);
5430     RegSet regs = RegSet::of(tmp);
5431     push(regs, sp);
5432     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5433     ldr(tmp, tmp);
5434     add(dst, tmp,  src, LSL,  shift);
5435     pop(regs, sp);
5436   }
5437 }
5438 
5439 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5440   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5441 
5442   if (AOTCodeCache::is_on_for_dump()) {
5443     decode_klass_not_null_for_aot(dst, src);
5444     return;
5445   }
5446 
5447   switch (klass_decode_mode()) {
5448   case KlassDecodeZero:
5449     if (CompressedKlassPointers::shift() != 0) {
5450       lsl(dst, src, CompressedKlassPointers::shift());
5451     } else {
5452       if (dst != src) mov(dst, src);
5453     }
5454     break;
5455 
5456   case KlassDecodeXor:
5457     if (CompressedKlassPointers::shift() != 0) {
5458       lsl(dst, src, CompressedKlassPointers::shift());
5459       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5460     } else {
5461       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5462     }
5463     break;
5464 
5465   case KlassDecodeMovk: {
5466     const uint64_t shifted_base =
5467       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5468 
5469     if (dst != src) movw(dst, src);
5470     movk(dst, shifted_base >> 32, 32);
5471 
5472     if (CompressedKlassPointers::shift() != 0) {
5473       lsl(dst, dst, CompressedKlassPointers::shift());
5474     }
5475 
5476     break;
5477   }
5478 
5479   case KlassDecodeNone:
5480     ShouldNotReachHere();
5481     break;
5482   }
5483 }
5484 
5485 void  MacroAssembler::decode_klass_not_null(Register r) {
5486   decode_klass_not_null(r, r);
5487 }
5488 
5489 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5490 #ifdef ASSERT
5491   {
5492     ThreadInVMfromUnknown tiv;
5493     assert (UseCompressedOops, "should only be used for compressed oops");
5494     assert (Universe::heap() != nullptr, "java heap should be initialized");
5495     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5496     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5497   }
5498 #endif
5499   int oop_index = oop_recorder()->find_index(obj);
5500   InstructionMark im(this);
5501   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5502   code_section()->relocate(inst_mark(), rspec);
5503   movz(dst, 0xDEAD, 16);
5504   movk(dst, 0xBEEF);
5505 }
5506 
5507 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5508   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5509   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5510   int index = oop_recorder()->find_index(k);
5511   assert(! Universe::heap()->is_in(k), "should not be an oop");
5512 
5513   InstructionMark im(this);
5514   RelocationHolder rspec = metadata_Relocation::spec(index);
5515   code_section()->relocate(inst_mark(), rspec);
5516   narrowKlass nk = CompressedKlassPointers::encode(k);
5517   movz(dst, (nk >> 16), 16);
5518   movk(dst, nk & 0xffff);
5519 }
5520 
5521 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5522                                     Register dst, Address src,
5523                                     Register tmp1, Register tmp2) {
5524   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5525   decorators = AccessInternal::decorator_fixup(decorators, type);
5526   bool as_raw = (decorators & AS_RAW) != 0;
5527   if (as_raw) {
5528     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5529   } else {
5530     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5531   }
5532 }
5533 
5534 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5535                                      Address dst, Register val,
5536                                      Register tmp1, Register tmp2, Register tmp3) {
5537   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5538   decorators = AccessInternal::decorator_fixup(decorators, type);
5539   bool as_raw = (decorators & AS_RAW) != 0;
5540   if (as_raw) {
5541     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5542   } else {
5543     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5544   }
5545 }
5546 
5547 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5548                                    Register tmp2, DecoratorSet decorators) {
5549   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5550 }
5551 
5552 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5553                                             Register tmp2, DecoratorSet decorators) {
5554   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5555 }
5556 
5557 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5558                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5559   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5560 }
5561 
5562 // Used for storing nulls.
5563 void MacroAssembler::store_heap_oop_null(Address dst) {
5564   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5565 }
5566 
5567 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5568   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5569   int index = oop_recorder()->allocate_metadata_index(obj);
5570   RelocationHolder rspec = metadata_Relocation::spec(index);
5571   return Address((address)obj, rspec);
5572 }
5573 
5574 // Move an oop into a register.
5575 void MacroAssembler::movoop(Register dst, jobject obj) {
5576   int oop_index;
5577   if (obj == nullptr) {
5578     oop_index = oop_recorder()->allocate_oop_index(obj);
5579   } else {
5580 #ifdef ASSERT
5581     {
5582       ThreadInVMfromUnknown tiv;
5583       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5584     }
5585 #endif
5586     oop_index = oop_recorder()->find_index(obj);
5587   }
5588   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5589 
5590   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5591     mov(dst, Address((address)obj, rspec));
5592   } else {
5593     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5594     ldr(dst, Address(dummy, rspec));
5595   }
5596 }
5597 
5598 // Move a metadata address into a register.
5599 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5600   int oop_index;
5601   if (obj == nullptr) {
5602     oop_index = oop_recorder()->allocate_metadata_index(obj);
5603   } else {
5604     oop_index = oop_recorder()->find_index(obj);
5605   }
5606   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5607   mov(dst, Address((address)obj, rspec));
5608 }
5609 
5610 Address MacroAssembler::constant_oop_address(jobject obj) {
5611 #ifdef ASSERT
5612   {
5613     ThreadInVMfromUnknown tiv;
5614     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5615     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5616   }
5617 #endif
5618   int oop_index = oop_recorder()->find_index(obj);
5619   return Address((address)obj, oop_Relocation::spec(oop_index));
5620 }
5621 
5622 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5623 void MacroAssembler::tlab_allocate(Register obj,
5624                                    Register var_size_in_bytes,
5625                                    int con_size_in_bytes,
5626                                    Register t1,
5627                                    Register t2,
5628                                    Label& slow_case) {
5629   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5630   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5631 }
5632 
5633 void MacroAssembler::inc_held_monitor_count(Register tmp) {
5634   Address dst(rthread, JavaThread::held_monitor_count_offset());
5635 #ifdef ASSERT
5636   ldr(tmp, dst);
5637   increment(tmp);
5638   str(tmp, dst);
5639   Label ok;
5640   tbz(tmp, 63, ok);
5641   STOP("assert(held monitor count underflow)");
5642   should_not_reach_here();
5643   bind(ok);
5644 #else
5645   increment(dst);
5646 #endif
5647 }
5648 
5649 void MacroAssembler::dec_held_monitor_count(Register tmp) {
5650   Address dst(rthread, JavaThread::held_monitor_count_offset());
5651 #ifdef ASSERT
5652   ldr(tmp, dst);
5653   decrement(tmp);
5654   str(tmp, dst);
5655   Label ok;
5656   tbz(tmp, 63, ok);
5657   STOP("assert(held monitor count underflow)");
5658   should_not_reach_here();
5659   bind(ok);
5660 #else
5661   decrement(dst);
5662 #endif
5663 }
5664 
5665 void MacroAssembler::verify_tlab() {
5666 #ifdef ASSERT
5667   if (UseTLAB && VerifyOops) {
5668     Label next, ok;
5669 
5670     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5671 
5672     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5673     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5674     cmp(rscratch2, rscratch1);
5675     br(Assembler::HS, next);
5676     STOP("assert(top >= start)");
5677     should_not_reach_here();
5678 
5679     bind(next);
5680     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5681     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5682     cmp(rscratch2, rscratch1);
5683     br(Assembler::HS, ok);
5684     STOP("assert(top <= end)");
5685     should_not_reach_here();
5686 
5687     bind(ok);
5688     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5689   }
5690 #endif
5691 }
5692 
5693 // Writes to stack successive pages until offset reached to check for
5694 // stack overflow + shadow pages.  This clobbers tmp.
5695 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5696   assert_different_registers(tmp, size, rscratch1);
5697   mov(tmp, sp);
5698   // Bang stack for total size given plus shadow page size.
5699   // Bang one page at a time because large size can bang beyond yellow and
5700   // red zones.
5701   Label loop;
5702   mov(rscratch1, (int)os::vm_page_size());
5703   bind(loop);
5704   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5705   subsw(size, size, rscratch1);
5706   str(size, Address(tmp));
5707   br(Assembler::GT, loop);
5708 
5709   // Bang down shadow pages too.
5710   // At this point, (tmp-0) is the last address touched, so don't
5711   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5712   // was post-decremented.)  Skip this address by starting at i=1, and
5713   // touch a few more pages below.  N.B.  It is important to touch all
5714   // the way down to and including i=StackShadowPages.
5715   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5716     // this could be any sized move but this is can be a debugging crumb
5717     // so the bigger the better.
5718     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5719     str(size, Address(tmp));
5720   }
5721 }
5722 
5723 // Move the address of the polling page into dest.
5724 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5725   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5726 }
5727 
5728 // Read the polling page.  The address of the polling page must
5729 // already be in r.
5730 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5731   address mark;
5732   {
5733     InstructionMark im(this);
5734     code_section()->relocate(inst_mark(), rtype);
5735     ldrw(zr, Address(r, 0));
5736     mark = inst_mark();
5737   }
5738   verify_cross_modify_fence_not_required();
5739   return mark;
5740 }
5741 
5742 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5743   relocInfo::relocType rtype = dest.rspec().reloc()->type();
5744   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5745   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5746   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5747   int64_t offset_low = dest_page - low_page;
5748   int64_t offset_high = dest_page - high_page;
5749 
5750   assert(is_valid_AArch64_address(dest.target()), "bad address");
5751   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5752 
5753   InstructionMark im(this);
5754   code_section()->relocate(inst_mark(), dest.rspec());
5755   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5756   // the code cache so that if it is relocated we know it will still reach
5757   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5758     _adrp(reg1, dest.target());
5759   } else {
5760     uint64_t target = (uint64_t)dest.target();
5761     uint64_t adrp_target
5762       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5763 
5764     _adrp(reg1, (address)adrp_target);
5765     movk(reg1, target >> 32, 32);
5766   }
5767   byte_offset = (uint64_t)dest.target() & 0xfff;
5768 }
5769 
5770 void MacroAssembler::load_byte_map_base(Register reg) {
5771   CardTable::CardValue* byte_map_base =
5772     ((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base();
5773 
5774   // Strictly speaking the byte_map_base isn't an address at all, and it might
5775   // even be negative. It is thus materialised as a constant.
5776   mov(reg, (uint64_t)byte_map_base);
5777 }
5778 
5779 void MacroAssembler::build_frame(int framesize) {
5780   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5781   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5782   protect_return_address();
5783   if (framesize < ((1 << 9) + 2 * wordSize)) {
5784     sub(sp, sp, framesize);
5785     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5786     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5787   } else {
5788     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5789     if (PreserveFramePointer) mov(rfp, sp);
5790     if (framesize < ((1 << 12) + 2 * wordSize))
5791       sub(sp, sp, framesize - 2 * wordSize);
5792     else {
5793       mov(rscratch1, framesize - 2 * wordSize);
5794       sub(sp, sp, rscratch1);
5795     }
5796   }
5797   verify_cross_modify_fence_not_required();
5798 }
5799 
5800 void MacroAssembler::remove_frame(int framesize) {
5801   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5802   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5803   if (framesize < ((1 << 9) + 2 * wordSize)) {
5804     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5805     add(sp, sp, framesize);
5806   } else {
5807     if (framesize < ((1 << 12) + 2 * wordSize))
5808       add(sp, sp, framesize - 2 * wordSize);
5809     else {
5810       mov(rscratch1, framesize - 2 * wordSize);
5811       add(sp, sp, rscratch1);
5812     }
5813     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5814   }
5815   authenticate_return_address();
5816 }
5817 
5818 
5819 // This method counts leading positive bytes (highest bit not set) in provided byte array
5820 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5821     // Simple and most common case of aligned small array which is not at the
5822     // end of memory page is placed here. All other cases are in stub.
5823     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5824     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5825     assert_different_registers(ary1, len, result);
5826 
5827     mov(result, len);
5828     cmpw(len, 0);
5829     br(LE, DONE);
5830     cmpw(len, 4 * wordSize);
5831     br(GE, STUB_LONG); // size > 32 then go to stub
5832 
5833     int shift = 64 - exact_log2(os::vm_page_size());
5834     lsl(rscratch1, ary1, shift);
5835     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5836     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5837     br(CS, STUB); // at the end of page then go to stub
5838     subs(len, len, wordSize);
5839     br(LT, END);
5840 
5841   BIND(LOOP);
5842     ldr(rscratch1, Address(post(ary1, wordSize)));
5843     tst(rscratch1, UPPER_BIT_MASK);
5844     br(NE, SET_RESULT);
5845     subs(len, len, wordSize);
5846     br(GE, LOOP);
5847     cmpw(len, -wordSize);
5848     br(EQ, DONE);
5849 
5850   BIND(END);
5851     ldr(rscratch1, Address(ary1));
5852     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5853     lslv(rscratch1, rscratch1, rscratch2);
5854     tst(rscratch1, UPPER_BIT_MASK);
5855     br(NE, SET_RESULT);
5856     b(DONE);
5857 
5858   BIND(STUB);
5859     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5860     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5861     address tpc1 = trampoline_call(count_pos);
5862     if (tpc1 == nullptr) {
5863       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5864       postcond(pc() == badAddress);
5865       return nullptr;
5866     }
5867     b(DONE);
5868 
5869   BIND(STUB_LONG);
5870     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5871     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5872     address tpc2 = trampoline_call(count_pos_long);
5873     if (tpc2 == nullptr) {
5874       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5875       postcond(pc() == badAddress);
5876       return nullptr;
5877     }
5878     b(DONE);
5879 
5880   BIND(SET_RESULT);
5881 
5882     add(len, len, wordSize);
5883     sub(result, result, len);
5884 
5885   BIND(DONE);
5886   postcond(pc() != badAddress);
5887   return pc();
5888 }
5889 
5890 // Clobbers: rscratch1, rscratch2, rflags
5891 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5892 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5893                                       Register tmp4, Register tmp5, Register result,
5894                                       Register cnt1, int elem_size) {
5895   Label DONE, SAME;
5896   Register tmp1 = rscratch1;
5897   Register tmp2 = rscratch2;
5898   int elem_per_word = wordSize/elem_size;
5899   int log_elem_size = exact_log2(elem_size);
5900   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5901   int length_offset = arrayOopDesc::length_offset_in_bytes();
5902   int base_offset
5903     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5904   // When the length offset is not aligned to 8 bytes,
5905   // then we align it down. This is valid because the new
5906   // offset will always be the klass which is the same
5907   // for type arrays.
5908   int start_offset = align_down(length_offset, BytesPerWord);
5909   int extra_length = base_offset - start_offset;
5910   assert(start_offset == length_offset || start_offset == klass_offset,
5911          "start offset must be 8-byte-aligned or be the klass offset");
5912   assert(base_offset != start_offset, "must include the length field");
5913   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5914   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5915 
5916   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5917   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5918 
5919 #ifndef PRODUCT
5920   {
5921     const char kind = (elem_size == 2) ? 'U' : 'L';
5922     char comment[64];
5923     snprintf(comment, sizeof comment, "array_equals%c{", kind);
5924     BLOCK_COMMENT(comment);
5925   }
5926 #endif
5927 
5928   // if (a1 == a2)
5929   //     return true;
5930   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5931   br(EQ, SAME);
5932 
5933   if (UseSimpleArrayEquals) {
5934     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5935     // if (a1 == nullptr || a2 == nullptr)
5936     //     return false;
5937     // a1 & a2 == 0 means (some-pointer is null) or
5938     // (very-rare-or-even-probably-impossible-pointer-values)
5939     // so, we can save one branch in most cases
5940     tst(a1, a2);
5941     mov(result, false);
5942     br(EQ, A_MIGHT_BE_NULL);
5943     // if (a1.length != a2.length)
5944     //      return false;
5945     bind(A_IS_NOT_NULL);
5946     ldrw(cnt1, Address(a1, length_offset));
5947     // Increase loop counter by diff between base- and actual start-offset.
5948     addw(cnt1, cnt1, extra_length);
5949     lea(a1, Address(a1, start_offset));
5950     lea(a2, Address(a2, start_offset));
5951     // Check for short strings, i.e. smaller than wordSize.
5952     subs(cnt1, cnt1, elem_per_word);
5953     br(Assembler::LT, SHORT);
5954     // Main 8 byte comparison loop.
5955     bind(NEXT_WORD); {
5956       ldr(tmp1, Address(post(a1, wordSize)));
5957       ldr(tmp2, Address(post(a2, wordSize)));
5958       subs(cnt1, cnt1, elem_per_word);
5959       eor(tmp5, tmp1, tmp2);
5960       cbnz(tmp5, DONE);
5961     } br(GT, NEXT_WORD);
5962     // Last longword.  In the case where length == 4 we compare the
5963     // same longword twice, but that's still faster than another
5964     // conditional branch.
5965     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5966     // length == 4.
5967     if (log_elem_size > 0)
5968       lsl(cnt1, cnt1, log_elem_size);
5969     ldr(tmp3, Address(a1, cnt1));
5970     ldr(tmp4, Address(a2, cnt1));
5971     eor(tmp5, tmp3, tmp4);
5972     cbnz(tmp5, DONE);
5973     b(SAME);
5974     bind(A_MIGHT_BE_NULL);
5975     // in case both a1 and a2 are not-null, proceed with loads
5976     cbz(a1, DONE);
5977     cbz(a2, DONE);
5978     b(A_IS_NOT_NULL);
5979     bind(SHORT);
5980 
5981     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
5982     {
5983       ldrw(tmp1, Address(post(a1, 4)));
5984       ldrw(tmp2, Address(post(a2, 4)));
5985       eorw(tmp5, tmp1, tmp2);
5986       cbnzw(tmp5, DONE);
5987     }
5988     bind(TAIL03);
5989     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
5990     {
5991       ldrh(tmp3, Address(post(a1, 2)));
5992       ldrh(tmp4, Address(post(a2, 2)));
5993       eorw(tmp5, tmp3, tmp4);
5994       cbnzw(tmp5, DONE);
5995     }
5996     bind(TAIL01);
5997     if (elem_size == 1) { // Only needed when comparing byte arrays.
5998       tbz(cnt1, 0, SAME); // 0-1 bytes left.
5999       {
6000         ldrb(tmp1, a1);
6001         ldrb(tmp2, a2);
6002         eorw(tmp5, tmp1, tmp2);
6003         cbnzw(tmp5, DONE);
6004       }
6005     }
6006   } else {
6007     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6008         CSET_EQ, LAST_CHECK;
6009     mov(result, false);
6010     cbz(a1, DONE);
6011     ldrw(cnt1, Address(a1, length_offset));
6012     cbz(a2, DONE);
6013     // Increase loop counter by diff between base- and actual start-offset.
6014     addw(cnt1, cnt1, extra_length);
6015 
6016     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6017     // faster to perform another branch before comparing a1 and a2
6018     cmp(cnt1, (u1)elem_per_word);
6019     br(LE, SHORT); // short or same
6020     ldr(tmp3, Address(pre(a1, start_offset)));
6021     subs(zr, cnt1, stubBytesThreshold);
6022     br(GE, STUB);
6023     ldr(tmp4, Address(pre(a2, start_offset)));
6024     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6025 
6026     // Main 16 byte comparison loop with 2 exits
6027     bind(NEXT_DWORD); {
6028       ldr(tmp1, Address(pre(a1, wordSize)));
6029       ldr(tmp2, Address(pre(a2, wordSize)));
6030       subs(cnt1, cnt1, 2 * elem_per_word);
6031       br(LE, TAIL);
6032       eor(tmp4, tmp3, tmp4);
6033       cbnz(tmp4, DONE);
6034       ldr(tmp3, Address(pre(a1, wordSize)));
6035       ldr(tmp4, Address(pre(a2, wordSize)));
6036       cmp(cnt1, (u1)elem_per_word);
6037       br(LE, TAIL2);
6038       cmp(tmp1, tmp2);
6039     } br(EQ, NEXT_DWORD);
6040     b(DONE);
6041 
6042     bind(TAIL);
6043     eor(tmp4, tmp3, tmp4);
6044     eor(tmp2, tmp1, tmp2);
6045     lslv(tmp2, tmp2, tmp5);
6046     orr(tmp5, tmp4, tmp2);
6047     cmp(tmp5, zr);
6048     b(CSET_EQ);
6049 
6050     bind(TAIL2);
6051     eor(tmp2, tmp1, tmp2);
6052     cbnz(tmp2, DONE);
6053     b(LAST_CHECK);
6054 
6055     bind(STUB);
6056     ldr(tmp4, Address(pre(a2, start_offset)));
6057     if (elem_size == 2) { // convert to byte counter
6058       lsl(cnt1, cnt1, 1);
6059     }
6060     eor(tmp5, tmp3, tmp4);
6061     cbnz(tmp5, DONE);
6062     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6063     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6064     address tpc = trampoline_call(stub);
6065     if (tpc == nullptr) {
6066       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6067       postcond(pc() == badAddress);
6068       return nullptr;
6069     }
6070     b(DONE);
6071 
6072     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6073     // so, if a2 == null => return false(0), else return true, so we can return a2
6074     mov(result, a2);
6075     b(DONE);
6076     bind(SHORT);
6077     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6078     ldr(tmp3, Address(a1, start_offset));
6079     ldr(tmp4, Address(a2, start_offset));
6080     bind(LAST_CHECK);
6081     eor(tmp4, tmp3, tmp4);
6082     lslv(tmp5, tmp4, tmp5);
6083     cmp(tmp5, zr);
6084     bind(CSET_EQ);
6085     cset(result, EQ);
6086     b(DONE);
6087   }
6088 
6089   bind(SAME);
6090   mov(result, true);
6091   // That's it.
6092   bind(DONE);
6093 
6094   BLOCK_COMMENT("} array_equals");
6095   postcond(pc() != badAddress);
6096   return pc();
6097 }
6098 
6099 // Compare Strings
6100 
6101 // For Strings we're passed the address of the first characters in a1
6102 // and a2 and the length in cnt1.
6103 // There are two implementations.  For arrays >= 8 bytes, all
6104 // comparisons (including the final one, which may overlap) are
6105 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6106 // halfword, then a short, and then a byte.
6107 
6108 void MacroAssembler::string_equals(Register a1, Register a2,
6109                                    Register result, Register cnt1)
6110 {
6111   Label SAME, DONE, SHORT, NEXT_WORD;
6112   Register tmp1 = rscratch1;
6113   Register tmp2 = rscratch2;
6114   Register cnt2 = tmp2;  // cnt2 only used in array length compare
6115 
6116   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6117 
6118 #ifndef PRODUCT
6119   {
6120     char comment[64];
6121     snprintf(comment, sizeof comment, "{string_equalsL");
6122     BLOCK_COMMENT(comment);
6123   }
6124 #endif
6125 
6126   mov(result, false);
6127 
6128   // Check for short strings, i.e. smaller than wordSize.
6129   subs(cnt1, cnt1, wordSize);
6130   br(Assembler::LT, SHORT);
6131   // Main 8 byte comparison loop.
6132   bind(NEXT_WORD); {
6133     ldr(tmp1, Address(post(a1, wordSize)));
6134     ldr(tmp2, Address(post(a2, wordSize)));
6135     subs(cnt1, cnt1, wordSize);
6136     eor(tmp1, tmp1, tmp2);
6137     cbnz(tmp1, DONE);
6138   } br(GT, NEXT_WORD);
6139   // Last longword.  In the case where length == 4 we compare the
6140   // same longword twice, but that's still faster than another
6141   // conditional branch.
6142   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6143   // length == 4.
6144   ldr(tmp1, Address(a1, cnt1));
6145   ldr(tmp2, Address(a2, cnt1));
6146   eor(tmp2, tmp1, tmp2);
6147   cbnz(tmp2, DONE);
6148   b(SAME);
6149 
6150   bind(SHORT);
6151   Label TAIL03, TAIL01;
6152 
6153   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6154   {
6155     ldrw(tmp1, Address(post(a1, 4)));
6156     ldrw(tmp2, Address(post(a2, 4)));
6157     eorw(tmp1, tmp1, tmp2);
6158     cbnzw(tmp1, DONE);
6159   }
6160   bind(TAIL03);
6161   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6162   {
6163     ldrh(tmp1, Address(post(a1, 2)));
6164     ldrh(tmp2, Address(post(a2, 2)));
6165     eorw(tmp1, tmp1, tmp2);
6166     cbnzw(tmp1, DONE);
6167   }
6168   bind(TAIL01);
6169   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6170     {
6171     ldrb(tmp1, a1);
6172     ldrb(tmp2, a2);
6173     eorw(tmp1, tmp1, tmp2);
6174     cbnzw(tmp1, DONE);
6175   }
6176   // Arrays are equal.
6177   bind(SAME);
6178   mov(result, true);
6179 
6180   // That's it.
6181   bind(DONE);
6182   BLOCK_COMMENT("} string_equals");
6183 }
6184 
6185 
6186 // The size of the blocks erased by the zero_blocks stub.  We must
6187 // handle anything smaller than this ourselves in zero_words().
6188 const int MacroAssembler::zero_words_block_size = 8;
6189 
6190 // zero_words() is used by C2 ClearArray patterns and by
6191 // C1_MacroAssembler.  It is as small as possible, handling small word
6192 // counts locally and delegating anything larger to the zero_blocks
6193 // stub.  It is expanded many times in compiled code, so it is
6194 // important to keep it short.
6195 
6196 // ptr:   Address of a buffer to be zeroed.
6197 // cnt:   Count in HeapWords.
6198 //
6199 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6200 address MacroAssembler::zero_words(Register ptr, Register cnt)
6201 {
6202   assert(is_power_of_2(zero_words_block_size), "adjust this");
6203 
6204   BLOCK_COMMENT("zero_words {");
6205   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6206   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6207   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6208 
6209   subs(rscratch1, cnt, zero_words_block_size);
6210   Label around;
6211   br(LO, around);
6212   {
6213     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6214     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6215     // Make sure this is a C2 compilation. C1 allocates space only for
6216     // trampoline stubs generated by Call LIR ops, and in any case it
6217     // makes sense for a C1 compilation task to proceed as quickly as
6218     // possible.
6219     CompileTask* task;
6220     if (StubRoutines::aarch64::complete()
6221         && Thread::current()->is_Compiler_thread()
6222         && (task = ciEnv::current()->task())
6223         && is_c2_compile(task->comp_level())) {
6224       address tpc = trampoline_call(zero_blocks);
6225       if (tpc == nullptr) {
6226         DEBUG_ONLY(reset_labels(around));
6227         return nullptr;
6228       }
6229     } else {
6230       far_call(zero_blocks);
6231     }
6232   }
6233   bind(around);
6234 
6235   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6236   // for us.
6237   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6238     Label l;
6239     tbz(cnt, exact_log2(i), l);
6240     for (int j = 0; j < i; j += 2) {
6241       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6242     }
6243     bind(l);
6244   }
6245   {
6246     Label l;
6247     tbz(cnt, 0, l);
6248     str(zr, Address(ptr));
6249     bind(l);
6250   }
6251 
6252   BLOCK_COMMENT("} zero_words");
6253   return pc();
6254 }
6255 
6256 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6257 // cnt:          Immediate count in HeapWords.
6258 //
6259 // r10, r11, rscratch1, and rscratch2 are clobbered.
6260 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6261 {
6262   assert(wordSize <= BlockZeroingLowLimit,
6263             "increase BlockZeroingLowLimit");
6264   address result = nullptr;
6265   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6266 #ifndef PRODUCT
6267     {
6268       char buf[64];
6269       snprintf(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6270       BLOCK_COMMENT(buf);
6271     }
6272 #endif
6273     if (cnt >= 16) {
6274       uint64_t loops = cnt/16;
6275       if (loops > 1) {
6276         mov(rscratch2, loops - 1);
6277       }
6278       {
6279         Label loop;
6280         bind(loop);
6281         for (int i = 0; i < 16; i += 2) {
6282           stp(zr, zr, Address(base, i * BytesPerWord));
6283         }
6284         add(base, base, 16 * BytesPerWord);
6285         if (loops > 1) {
6286           subs(rscratch2, rscratch2, 1);
6287           br(GE, loop);
6288         }
6289       }
6290     }
6291     cnt %= 16;
6292     int i = cnt & 1;  // store any odd word to start
6293     if (i) str(zr, Address(base));
6294     for (; i < (int)cnt; i += 2) {
6295       stp(zr, zr, Address(base, i * wordSize));
6296     }
6297     BLOCK_COMMENT("} zero_words");
6298     result = pc();
6299   } else {
6300     mov(r10, base); mov(r11, cnt);
6301     result = zero_words(r10, r11);
6302   }
6303   return result;
6304 }
6305 
6306 // Zero blocks of memory by using DC ZVA.
6307 //
6308 // Aligns the base address first sufficiently for DC ZVA, then uses
6309 // DC ZVA repeatedly for every full block.  cnt is the size to be
6310 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6311 // in cnt.
6312 //
6313 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6314 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6315 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6316   Register tmp = rscratch1;
6317   Register tmp2 = rscratch2;
6318   int zva_length = VM_Version::zva_length();
6319   Label initial_table_end, loop_zva;
6320   Label fini;
6321 
6322   // Base must be 16 byte aligned. If not just return and let caller handle it
6323   tst(base, 0x0f);
6324   br(Assembler::NE, fini);
6325   // Align base with ZVA length.
6326   neg(tmp, base);
6327   andr(tmp, tmp, zva_length - 1);
6328 
6329   // tmp: the number of bytes to be filled to align the base with ZVA length.
6330   add(base, base, tmp);
6331   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6332   adr(tmp2, initial_table_end);
6333   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6334   br(tmp2);
6335 
6336   for (int i = -zva_length + 16; i < 0; i += 16)
6337     stp(zr, zr, Address(base, i));
6338   bind(initial_table_end);
6339 
6340   sub(cnt, cnt, zva_length >> 3);
6341   bind(loop_zva);
6342   dc(Assembler::ZVA, base);
6343   subs(cnt, cnt, zva_length >> 3);
6344   add(base, base, zva_length);
6345   br(Assembler::GE, loop_zva);
6346   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6347   bind(fini);
6348 }
6349 
6350 // base:   Address of a buffer to be filled, 8 bytes aligned.
6351 // cnt:    Count in 8-byte unit.
6352 // value:  Value to be filled with.
6353 // base will point to the end of the buffer after filling.
6354 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6355 {
6356 //  Algorithm:
6357 //
6358 //    if (cnt == 0) {
6359 //      return;
6360 //    }
6361 //    if ((p & 8) != 0) {
6362 //      *p++ = v;
6363 //    }
6364 //
6365 //    scratch1 = cnt & 14;
6366 //    cnt -= scratch1;
6367 //    p += scratch1;
6368 //    switch (scratch1 / 2) {
6369 //      do {
6370 //        cnt -= 16;
6371 //          p[-16] = v;
6372 //          p[-15] = v;
6373 //        case 7:
6374 //          p[-14] = v;
6375 //          p[-13] = v;
6376 //        case 6:
6377 //          p[-12] = v;
6378 //          p[-11] = v;
6379 //          // ...
6380 //        case 1:
6381 //          p[-2] = v;
6382 //          p[-1] = v;
6383 //        case 0:
6384 //          p += 16;
6385 //      } while (cnt);
6386 //    }
6387 //    if ((cnt & 1) == 1) {
6388 //      *p++ = v;
6389 //    }
6390 
6391   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6392 
6393   Label fini, skip, entry, loop;
6394   const int unroll = 8; // Number of stp instructions we'll unroll
6395 
6396   cbz(cnt, fini);
6397   tbz(base, 3, skip);
6398   str(value, Address(post(base, 8)));
6399   sub(cnt, cnt, 1);
6400   bind(skip);
6401 
6402   andr(rscratch1, cnt, (unroll-1) * 2);
6403   sub(cnt, cnt, rscratch1);
6404   add(base, base, rscratch1, Assembler::LSL, 3);
6405   adr(rscratch2, entry);
6406   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6407   br(rscratch2);
6408 
6409   bind(loop);
6410   add(base, base, unroll * 16);
6411   for (int i = -unroll; i < 0; i++)
6412     stp(value, value, Address(base, i * 16));
6413   bind(entry);
6414   subs(cnt, cnt, unroll * 2);
6415   br(Assembler::GE, loop);
6416 
6417   tbz(cnt, 0, fini);
6418   str(value, Address(post(base, 8)));
6419   bind(fini);
6420 }
6421 
6422 // Intrinsic for
6423 //
6424 // - sun/nio/cs/ISO_8859_1$Encoder.implEncodeISOArray
6425 //     return the number of characters copied.
6426 // - java/lang/StringUTF16.compress
6427 //     return index of non-latin1 character if copy fails, otherwise 'len'.
6428 //
6429 // This version always returns the number of characters copied, and does not
6430 // clobber the 'len' register. A successful copy will complete with the post-
6431 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6432 // post-condition: 0 <= 'res' < 'len'.
6433 //
6434 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6435 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6436 //       beyond the acceptable, even though the footprint would be smaller.
6437 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6438 //       avoid additional bloat.
6439 //
6440 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6441 void MacroAssembler::encode_iso_array(Register src, Register dst,
6442                                       Register len, Register res, bool ascii,
6443                                       FloatRegister vtmp0, FloatRegister vtmp1,
6444                                       FloatRegister vtmp2, FloatRegister vtmp3,
6445                                       FloatRegister vtmp4, FloatRegister vtmp5)
6446 {
6447   Register cnt = res;
6448   Register max = rscratch1;
6449   Register chk = rscratch2;
6450 
6451   prfm(Address(src), PLDL1STRM);
6452   movw(cnt, len);
6453 
6454 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6455 
6456   Label LOOP_32, DONE_32, FAIL_32;
6457 
6458   BIND(LOOP_32);
6459   {
6460     cmpw(cnt, 32);
6461     br(LT, DONE_32);
6462     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6463     // Extract lower bytes.
6464     FloatRegister vlo0 = vtmp4;
6465     FloatRegister vlo1 = vtmp5;
6466     uzp1(vlo0, T16B, vtmp0, vtmp1);
6467     uzp1(vlo1, T16B, vtmp2, vtmp3);
6468     // Merge bits...
6469     orr(vtmp0, T16B, vtmp0, vtmp1);
6470     orr(vtmp2, T16B, vtmp2, vtmp3);
6471     // Extract merged upper bytes.
6472     FloatRegister vhix = vtmp0;
6473     uzp2(vhix, T16B, vtmp0, vtmp2);
6474     // ISO-check on hi-parts (all zero).
6475     //                          ASCII-check on lo-parts (no sign).
6476     FloatRegister vlox = vtmp1; // Merge lower bytes.
6477                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6478     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6479     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6480     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6481                                 ASCII(orr(chk, chk, max));
6482     cbnz(chk, FAIL_32);
6483     subw(cnt, cnt, 32);
6484     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6485     b(LOOP_32);
6486   }
6487   BIND(FAIL_32);
6488   sub(src, src, 64);
6489   BIND(DONE_32);
6490 
6491   Label LOOP_8, SKIP_8;
6492 
6493   BIND(LOOP_8);
6494   {
6495     cmpw(cnt, 8);
6496     br(LT, SKIP_8);
6497     FloatRegister vhi = vtmp0;
6498     FloatRegister vlo = vtmp1;
6499     ld1(vtmp3, T8H, src);
6500     uzp1(vlo, T16B, vtmp3, vtmp3);
6501     uzp2(vhi, T16B, vtmp3, vtmp3);
6502     // ISO-check on hi-parts (all zero).
6503     //                          ASCII-check on lo-parts (no sign).
6504                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6505     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6506                                 ASCII(umov(max, vtmp2, B, 0));
6507                                 ASCII(orr(chk, chk, max));
6508     cbnz(chk, SKIP_8);
6509 
6510     strd(vlo, Address(post(dst, 8)));
6511     subw(cnt, cnt, 8);
6512     add(src, src, 16);
6513     b(LOOP_8);
6514   }
6515   BIND(SKIP_8);
6516 
6517 #undef ASCII
6518 
6519   Label LOOP, DONE;
6520 
6521   cbz(cnt, DONE);
6522   BIND(LOOP);
6523   {
6524     Register chr = rscratch1;
6525     ldrh(chr, Address(post(src, 2)));
6526     tst(chr, ascii ? 0xff80 : 0xff00);
6527     br(NE, DONE);
6528     strb(chr, Address(post(dst, 1)));
6529     subs(cnt, cnt, 1);
6530     br(GT, LOOP);
6531   }
6532   BIND(DONE);
6533   // Return index where we stopped.
6534   subw(res, len, cnt);
6535 }
6536 
6537 // Inflate byte[] array to char[].
6538 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6539 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6540                                            FloatRegister vtmp1, FloatRegister vtmp2,
6541                                            FloatRegister vtmp3, Register tmp4) {
6542   Label big, done, after_init, to_stub;
6543 
6544   assert_different_registers(src, dst, len, tmp4, rscratch1);
6545 
6546   fmovd(vtmp1, 0.0);
6547   lsrw(tmp4, len, 3);
6548   bind(after_init);
6549   cbnzw(tmp4, big);
6550   // Short string: less than 8 bytes.
6551   {
6552     Label loop, tiny;
6553 
6554     cmpw(len, 4);
6555     br(LT, tiny);
6556     // Use SIMD to do 4 bytes.
6557     ldrs(vtmp2, post(src, 4));
6558     zip1(vtmp3, T8B, vtmp2, vtmp1);
6559     subw(len, len, 4);
6560     strd(vtmp3, post(dst, 8));
6561 
6562     cbzw(len, done);
6563 
6564     // Do the remaining bytes by steam.
6565     bind(loop);
6566     ldrb(tmp4, post(src, 1));
6567     strh(tmp4, post(dst, 2));
6568     subw(len, len, 1);
6569 
6570     bind(tiny);
6571     cbnz(len, loop);
6572 
6573     b(done);
6574   }
6575 
6576   if (SoftwarePrefetchHintDistance >= 0) {
6577     bind(to_stub);
6578       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6579       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6580       address tpc = trampoline_call(stub);
6581       if (tpc == nullptr) {
6582         DEBUG_ONLY(reset_labels(big, done));
6583         postcond(pc() == badAddress);
6584         return nullptr;
6585       }
6586       b(after_init);
6587   }
6588 
6589   // Unpack the bytes 8 at a time.
6590   bind(big);
6591   {
6592     Label loop, around, loop_last, loop_start;
6593 
6594     if (SoftwarePrefetchHintDistance >= 0) {
6595       const int large_loop_threshold = (64 + 16)/8;
6596       ldrd(vtmp2, post(src, 8));
6597       andw(len, len, 7);
6598       cmp(tmp4, (u1)large_loop_threshold);
6599       br(GE, to_stub);
6600       b(loop_start);
6601 
6602       bind(loop);
6603       ldrd(vtmp2, post(src, 8));
6604       bind(loop_start);
6605       subs(tmp4, tmp4, 1);
6606       br(EQ, loop_last);
6607       zip1(vtmp2, T16B, vtmp2, vtmp1);
6608       ldrd(vtmp3, post(src, 8));
6609       st1(vtmp2, T8H, post(dst, 16));
6610       subs(tmp4, tmp4, 1);
6611       zip1(vtmp3, T16B, vtmp3, vtmp1);
6612       st1(vtmp3, T8H, post(dst, 16));
6613       br(NE, loop);
6614       b(around);
6615       bind(loop_last);
6616       zip1(vtmp2, T16B, vtmp2, vtmp1);
6617       st1(vtmp2, T8H, post(dst, 16));
6618       bind(around);
6619       cbz(len, done);
6620     } else {
6621       andw(len, len, 7);
6622       bind(loop);
6623       ldrd(vtmp2, post(src, 8));
6624       sub(tmp4, tmp4, 1);
6625       zip1(vtmp3, T16B, vtmp2, vtmp1);
6626       st1(vtmp3, T8H, post(dst, 16));
6627       cbnz(tmp4, loop);
6628     }
6629   }
6630 
6631   // Do the tail of up to 8 bytes.
6632   add(src, src, len);
6633   ldrd(vtmp3, Address(src, -8));
6634   add(dst, dst, len, ext::uxtw, 1);
6635   zip1(vtmp3, T16B, vtmp3, vtmp1);
6636   strq(vtmp3, Address(dst, -16));
6637 
6638   bind(done);
6639   postcond(pc() != badAddress);
6640   return pc();
6641 }
6642 
6643 // Compress char[] array to byte[].
6644 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6645 // Return the array length if every element in array can be encoded,
6646 // otherwise, the index of first non-latin1 (> 0xff) character.
6647 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6648                                          Register res,
6649                                          FloatRegister tmp0, FloatRegister tmp1,
6650                                          FloatRegister tmp2, FloatRegister tmp3,
6651                                          FloatRegister tmp4, FloatRegister tmp5) {
6652   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6653 }
6654 
6655 // java.math.round(double a)
6656 // Returns the closest long to the argument, with ties rounding to
6657 // positive infinity.  This requires some fiddling for corner
6658 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6659 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6660                                        FloatRegister ftmp) {
6661   Label DONE;
6662   BLOCK_COMMENT("java_round_double: { ");
6663   fmovd(rscratch1, src);
6664   // Use RoundToNearestTiesAway unless src small and -ve.
6665   fcvtasd(dst, src);
6666   // Test if src >= 0 || abs(src) >= 0x1.0p52
6667   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6668   mov(rscratch2, julong_cast(0x1.0p52));
6669   cmp(rscratch1, rscratch2);
6670   br(HS, DONE); {
6671     // src < 0 && abs(src) < 0x1.0p52
6672     // src may have a fractional part, so add 0.5
6673     fmovd(ftmp, 0.5);
6674     faddd(ftmp, src, ftmp);
6675     // Convert double to jlong, use RoundTowardsNegative
6676     fcvtmsd(dst, ftmp);
6677   }
6678   bind(DONE);
6679   BLOCK_COMMENT("} java_round_double");
6680 }
6681 
6682 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6683                                       FloatRegister ftmp) {
6684   Label DONE;
6685   BLOCK_COMMENT("java_round_float: { ");
6686   fmovs(rscratch1, src);
6687   // Use RoundToNearestTiesAway unless src small and -ve.
6688   fcvtassw(dst, src);
6689   // Test if src >= 0 || abs(src) >= 0x1.0p23
6690   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6691   mov(rscratch2, jint_cast(0x1.0p23f));
6692   cmp(rscratch1, rscratch2);
6693   br(HS, DONE); {
6694     // src < 0 && |src| < 0x1.0p23
6695     // src may have a fractional part, so add 0.5
6696     fmovs(ftmp, 0.5f);
6697     fadds(ftmp, src, ftmp);
6698     // Convert float to jint, use RoundTowardsNegative
6699     fcvtmssw(dst, ftmp);
6700   }
6701   bind(DONE);
6702   BLOCK_COMMENT("} java_round_float");
6703 }
6704 
6705 // get_thread() can be called anywhere inside generated code so we
6706 // need to save whatever non-callee save context might get clobbered
6707 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6708 // the call setup code.
6709 //
6710 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6711 // On other systems, the helper is a usual C function.
6712 //
6713 void MacroAssembler::get_thread(Register dst) {
6714   RegSet saved_regs =
6715     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6716     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6717 
6718   protect_return_address();
6719   push(saved_regs, sp);
6720 
6721   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6722   blr(lr);
6723   if (dst != c_rarg0) {
6724     mov(dst, c_rarg0);
6725   }
6726 
6727   pop(saved_regs, sp);
6728   authenticate_return_address();
6729 }
6730 
6731 void MacroAssembler::cache_wb(Address line) {
6732   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6733   assert(line.index() == noreg, "index should be noreg");
6734   assert(line.offset() == 0, "offset should be 0");
6735   // would like to assert this
6736   // assert(line._ext.shift == 0, "shift should be zero");
6737   if (VM_Version::supports_dcpop()) {
6738     // writeback using clear virtual address to point of persistence
6739     dc(Assembler::CVAP, line.base());
6740   } else {
6741     // no need to generate anything as Unsafe.writebackMemory should
6742     // never invoke this stub
6743   }
6744 }
6745 
6746 void MacroAssembler::cache_wbsync(bool is_pre) {
6747   // we only need a barrier post sync
6748   if (!is_pre) {
6749     membar(Assembler::AnyAny);
6750   }
6751 }
6752 
6753 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6754   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6755     return;
6756   }
6757   // Make sure that native code does not change SVE vector length.
6758   Label verify_ok;
6759   movw(tmp, zr);
6760   sve_inc(tmp, B);
6761   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6762   br(EQ, verify_ok);
6763   stop("Error: SVE vector length has changed since jvm startup");
6764   bind(verify_ok);
6765 }
6766 
6767 void MacroAssembler::verify_ptrue() {
6768   Label verify_ok;
6769   if (!UseSVE) {
6770     return;
6771   }
6772   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6773   sve_dec(rscratch1, B);
6774   cbz(rscratch1, verify_ok);
6775   stop("Error: the preserved predicate register (p7) elements are not all true");
6776   bind(verify_ok);
6777 }
6778 
6779 void MacroAssembler::safepoint_isb() {
6780   isb();
6781 #ifndef PRODUCT
6782   if (VerifyCrossModifyFence) {
6783     // Clear the thread state.
6784     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6785   }
6786 #endif
6787 }
6788 
6789 #ifndef PRODUCT
6790 void MacroAssembler::verify_cross_modify_fence_not_required() {
6791   if (VerifyCrossModifyFence) {
6792     // Check if thread needs a cross modify fence.
6793     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6794     Label fence_not_required;
6795     cbz(rscratch1, fence_not_required);
6796     // If it does then fail.
6797     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6798     mov(c_rarg0, rthread);
6799     blr(rscratch1);
6800     bind(fence_not_required);
6801   }
6802 }
6803 #endif
6804 
6805 void MacroAssembler::spin_wait() {
6806   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6807     switch (VM_Version::spin_wait_desc().inst()) {
6808       case SpinWait::NOP:
6809         nop();
6810         break;
6811       case SpinWait::ISB:
6812         isb();
6813         break;
6814       case SpinWait::YIELD:
6815         yield();
6816         break;
6817       case SpinWait::SB:
6818         sb();
6819         break;
6820       default:
6821         ShouldNotReachHere();
6822     }
6823   }
6824 }
6825 
6826 // Stack frame creation/removal
6827 
6828 void MacroAssembler::enter(bool strip_ret_addr) {
6829   if (strip_ret_addr) {
6830     // Addresses can only be signed once. If there are multiple nested frames being created
6831     // in the same function, then the return address needs stripping first.
6832     strip_return_address();
6833   }
6834   protect_return_address();
6835   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6836   mov(rfp, sp);
6837 }
6838 
6839 void MacroAssembler::leave() {
6840   mov(sp, rfp);
6841   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6842   authenticate_return_address();
6843 }
6844 
6845 // ROP Protection
6846 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6847 // destroying stack frames or whenever directly loading/storing the LR to memory.
6848 // If ROP protection is not set then these functions are no-ops.
6849 // For more details on PAC see pauth_aarch64.hpp.
6850 
6851 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6852 // Uses value zero as the modifier.
6853 //
6854 void MacroAssembler::protect_return_address() {
6855   if (VM_Version::use_rop_protection()) {
6856     check_return_address();
6857     paciaz();
6858   }
6859 }
6860 
6861 // Sign the return value in the given register. Use before updating the LR in the existing stack
6862 // frame for the current function.
6863 // Uses value zero as the modifier.
6864 //
6865 void MacroAssembler::protect_return_address(Register return_reg) {
6866   if (VM_Version::use_rop_protection()) {
6867     check_return_address(return_reg);
6868     paciza(return_reg);
6869   }
6870 }
6871 
6872 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6873 // Uses value zero as the modifier.
6874 //
6875 void MacroAssembler::authenticate_return_address() {
6876   if (VM_Version::use_rop_protection()) {
6877     autiaz();
6878     check_return_address();
6879   }
6880 }
6881 
6882 // Authenticate the return value in the given register. Use before updating the LR in the existing
6883 // stack frame for the current function.
6884 // Uses value zero as the modifier.
6885 //
6886 void MacroAssembler::authenticate_return_address(Register return_reg) {
6887   if (VM_Version::use_rop_protection()) {
6888     autiza(return_reg);
6889     check_return_address(return_reg);
6890   }
6891 }
6892 
6893 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6894 // there is no guaranteed way of authenticating the LR.
6895 //
6896 void MacroAssembler::strip_return_address() {
6897   if (VM_Version::use_rop_protection()) {
6898     xpaclri();
6899   }
6900 }
6901 
6902 #ifndef PRODUCT
6903 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6904 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6905 // it is difficult to debug back to the callee function.
6906 // This function simply loads from the address in the given register.
6907 // Use directly after authentication to catch authentication failures.
6908 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6909 //
6910 void MacroAssembler::check_return_address(Register return_reg) {
6911   if (VM_Version::use_rop_protection()) {
6912     ldr(zr, Address(return_reg));
6913   }
6914 }
6915 #endif
6916 
6917 // The java_calling_convention describes stack locations as ideal slots on
6918 // a frame with no abi restrictions. Since we must observe abi restrictions
6919 // (like the placement of the register window) the slots must be biased by
6920 // the following value.
6921 static int reg2offset_in(VMReg r) {
6922   // Account for saved rfp and lr
6923   // This should really be in_preserve_stack_slots
6924   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6925 }
6926 
6927 static int reg2offset_out(VMReg r) {
6928   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6929 }
6930 
6931 // On 64bit we will store integer like items to the stack as
6932 // 64bits items (AArch64 ABI) even though java would only store
6933 // 32bits for a parameter. On 32bit it will simply be 32bits
6934 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6935 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6936   if (src.first()->is_stack()) {
6937     if (dst.first()->is_stack()) {
6938       // stack to stack
6939       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6940       str(tmp, Address(sp, reg2offset_out(dst.first())));
6941     } else {
6942       // stack to reg
6943       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6944     }
6945   } else if (dst.first()->is_stack()) {
6946     // reg to stack
6947     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6948   } else {
6949     if (dst.first() != src.first()) {
6950       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6951     }
6952   }
6953 }
6954 
6955 // An oop arg. Must pass a handle not the oop itself
6956 void MacroAssembler::object_move(
6957                         OopMap* map,
6958                         int oop_handle_offset,
6959                         int framesize_in_slots,
6960                         VMRegPair src,
6961                         VMRegPair dst,
6962                         bool is_receiver,
6963                         int* receiver_offset) {
6964 
6965   // must pass a handle. First figure out the location we use as a handle
6966 
6967   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
6968 
6969   // See if oop is null if it is we need no handle
6970 
6971   if (src.first()->is_stack()) {
6972 
6973     // Oop is already on the stack as an argument
6974     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
6975     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
6976     if (is_receiver) {
6977       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
6978     }
6979 
6980     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
6981     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
6982     // conditionally move a null
6983     cmp(rscratch1, zr);
6984     csel(rHandle, zr, rHandle, Assembler::EQ);
6985   } else {
6986 
6987     // Oop is in an a register we must store it to the space we reserve
6988     // on the stack for oop_handles and pass a handle if oop is non-null
6989 
6990     const Register rOop = src.first()->as_Register();
6991     int oop_slot;
6992     if (rOop == j_rarg0)
6993       oop_slot = 0;
6994     else if (rOop == j_rarg1)
6995       oop_slot = 1;
6996     else if (rOop == j_rarg2)
6997       oop_slot = 2;
6998     else if (rOop == j_rarg3)
6999       oop_slot = 3;
7000     else if (rOop == j_rarg4)
7001       oop_slot = 4;
7002     else if (rOop == j_rarg5)
7003       oop_slot = 5;
7004     else if (rOop == j_rarg6)
7005       oop_slot = 6;
7006     else {
7007       assert(rOop == j_rarg7, "wrong register");
7008       oop_slot = 7;
7009     }
7010 
7011     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7012     int offset = oop_slot*VMRegImpl::stack_slot_size;
7013 
7014     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7015     // Store oop in handle area, may be null
7016     str(rOop, Address(sp, offset));
7017     if (is_receiver) {
7018       *receiver_offset = offset;
7019     }
7020 
7021     cmp(rOop, zr);
7022     lea(rHandle, Address(sp, offset));
7023     // conditionally move a null
7024     csel(rHandle, zr, rHandle, Assembler::EQ);
7025   }
7026 
7027   // If arg is on the stack then place it otherwise it is already in correct reg.
7028   if (dst.first()->is_stack()) {
7029     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7030   }
7031 }
7032 
7033 // A float arg may have to do float reg int reg conversion
7034 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7035  if (src.first()->is_stack()) {
7036     if (dst.first()->is_stack()) {
7037       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7038       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7039     } else {
7040       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7041     }
7042   } else if (src.first() != dst.first()) {
7043     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7044       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7045     else
7046       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7047   }
7048 }
7049 
7050 // A long move
7051 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7052   if (src.first()->is_stack()) {
7053     if (dst.first()->is_stack()) {
7054       // stack to stack
7055       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7056       str(tmp, Address(sp, reg2offset_out(dst.first())));
7057     } else {
7058       // stack to reg
7059       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7060     }
7061   } else if (dst.first()->is_stack()) {
7062     // reg to stack
7063     // Do we really have to sign extend???
7064     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7065     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7066   } else {
7067     if (dst.first() != src.first()) {
7068       mov(dst.first()->as_Register(), src.first()->as_Register());
7069     }
7070   }
7071 }
7072 
7073 
7074 // A double move
7075 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7076  if (src.first()->is_stack()) {
7077     if (dst.first()->is_stack()) {
7078       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7079       str(tmp, Address(sp, reg2offset_out(dst.first())));
7080     } else {
7081       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7082     }
7083   } else if (src.first() != dst.first()) {
7084     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7085       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7086     else
7087       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7088   }
7089 }
7090 
7091 // Implements lightweight-locking.
7092 //
7093 //  - obj: the object to be locked
7094 //  - t1, t2, t3: temporary registers, will be destroyed
7095 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7096 void MacroAssembler::lightweight_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7097   assert(LockingMode == LM_LIGHTWEIGHT, "only used with new lightweight locking");
7098   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7099 
7100   Label push;
7101   const Register top = t1;
7102   const Register mark = t2;
7103   const Register t = t3;
7104 
7105   // Preload the markWord. It is important that this is the first
7106   // instruction emitted as it is part of C1's null check semantics.
7107   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7108 
7109   if (UseObjectMonitorTable) {
7110     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7111     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7112   }
7113 
7114   if (DiagnoseSyncOnValueBasedClasses != 0) {
7115     load_klass(t1, obj);
7116     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7117     tst(t1, KlassFlags::_misc_is_value_based_class);
7118     br(Assembler::NE, slow);
7119   }
7120 
7121   // Check if the lock-stack is full.
7122   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7123   cmpw(top, (unsigned)LockStack::end_offset());
7124   br(Assembler::GE, slow);
7125 
7126   // Check for recursion.
7127   subw(t, top, oopSize);
7128   ldr(t, Address(rthread, t));
7129   cmp(obj, t);
7130   br(Assembler::EQ, push);
7131 
7132   // Check header for monitor (0b10).
7133   tst(mark, markWord::monitor_value);
7134   br(Assembler::NE, slow);
7135 
7136   // Try to lock. Transition lock bits 0b01 => 0b00
7137   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7138   orr(mark, mark, markWord::unlocked_value);
7139   eor(t, mark, markWord::unlocked_value);
7140   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7141           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7142   br(Assembler::NE, slow);
7143 
7144   bind(push);
7145   // After successful lock, push object on lock-stack.
7146   str(obj, Address(rthread, top));
7147   addw(top, top, oopSize);
7148   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7149 }
7150 
7151 // Implements lightweight-unlocking.
7152 //
7153 // - obj: the object to be unlocked
7154 // - t1, t2, t3: temporary registers
7155 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7156 void MacroAssembler::lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7157   assert(LockingMode == LM_LIGHTWEIGHT, "only used with new lightweight locking");
7158   // cmpxchg clobbers rscratch1.
7159   assert_different_registers(obj, t1, t2, t3, rscratch1);
7160 
7161 #ifdef ASSERT
7162   {
7163     // Check for lock-stack underflow.
7164     Label stack_ok;
7165     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7166     cmpw(t1, (unsigned)LockStack::start_offset());
7167     br(Assembler::GE, stack_ok);
7168     STOP("Lock-stack underflow");
7169     bind(stack_ok);
7170   }
7171 #endif
7172 
7173   Label unlocked, push_and_slow;
7174   const Register top = t1;
7175   const Register mark = t2;
7176   const Register t = t3;
7177 
7178   // Check if obj is top of lock-stack.
7179   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7180   subw(top, top, oopSize);
7181   ldr(t, Address(rthread, top));
7182   cmp(obj, t);
7183   br(Assembler::NE, slow);
7184 
7185   // Pop lock-stack.
7186   DEBUG_ONLY(str(zr, Address(rthread, top));)
7187   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7188 
7189   // Check if recursive.
7190   subw(t, top, oopSize);
7191   ldr(t, Address(rthread, t));
7192   cmp(obj, t);
7193   br(Assembler::EQ, unlocked);
7194 
7195   // Not recursive. Check header for monitor (0b10).
7196   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7197   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7198 
7199 #ifdef ASSERT
7200   // Check header not unlocked (0b01).
7201   Label not_unlocked;
7202   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7203   stop("lightweight_unlock already unlocked");
7204   bind(not_unlocked);
7205 #endif
7206 
7207   // Try to unlock. Transition lock bits 0b00 => 0b01
7208   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7209   orr(t, mark, markWord::unlocked_value);
7210   cmpxchg(obj, mark, t, Assembler::xword,
7211           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7212   br(Assembler::EQ, unlocked);
7213 
7214   bind(push_and_slow);
7215   // Restore lock-stack and handle the unlock in runtime.
7216   DEBUG_ONLY(str(obj, Address(rthread, top));)
7217   addw(top, top, oopSize);
7218   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7219   b(slow);
7220 
7221   bind(unlocked);
7222 }