1 /*
   2  * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "ci/ciUtilities.hpp"
  30 #include "code/compiledIC.hpp"
  31 #include "compiler/compileTask.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/barrierSet.hpp"
  35 #include "gc/shared/barrierSetAssembler.hpp"
  36 #include "gc/shared/cardTableBarrierSet.hpp"
  37 #include "gc/shared/cardTable.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/tlab_globals.hpp"
  40 #include "interpreter/bytecodeHistogram.hpp"
  41 #include "interpreter/interpreter.hpp"
  42 #include "interpreter/interpreterRuntime.hpp"
  43 #include "jvm.h"
  44 #include "memory/resourceArea.hpp"
  45 #include "memory/universe.hpp"
  46 #include "nativeInst_aarch64.hpp"
  47 #include "oops/accessDecorators.hpp"
  48 #include "oops/compressedKlass.inline.hpp"
  49 #include "oops/compressedOops.inline.hpp"
  50 #include "oops/klass.inline.hpp"
  51 #include "runtime/continuation.hpp"
  52 #include "runtime/icache.hpp"
  53 #include "runtime/interfaceSupport.inline.hpp"
  54 #include "runtime/javaThread.hpp"
  55 #include "runtime/jniHandles.inline.hpp"
  56 #include "runtime/sharedRuntime.hpp"
  57 #include "runtime/stubRoutines.hpp"
  58 #include "utilities/globalDefinitions.hpp"
  59 #include "utilities/powerOfTwo.hpp"
  60 #ifdef COMPILER1
  61 #include "c1/c1_LIRAssembler.hpp"
  62 #endif
  63 #ifdef COMPILER2
  64 #include "oops/oop.hpp"
  65 #include "opto/compile.hpp"
  66 #include "opto/node.hpp"
  67 #include "opto/output.hpp"
  68 #endif
  69 
  70 #include <sys/types.h>
  71 
  72 #ifdef PRODUCT
  73 #define BLOCK_COMMENT(str) /* nothing */
  74 #else
  75 #define BLOCK_COMMENT(str) block_comment(str)
  76 #endif
  77 #define STOP(str) stop(str);
  78 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  79 
  80 #ifdef ASSERT
  81 extern "C" void disnm(intptr_t p);
  82 #endif
  83 // Target-dependent relocation processing
  84 //
  85 // Instruction sequences whose target may need to be retrieved or
  86 // patched are distinguished by their leading instruction, sorting
  87 // them into three main instruction groups and related subgroups.
  88 //
  89 // 1) Branch, Exception and System (insn count = 1)
  90 //    1a) Unconditional branch (immediate):
  91 //      b/bl imm19
  92 //    1b) Compare & branch (immediate):
  93 //      cbz/cbnz Rt imm19
  94 //    1c) Test & branch (immediate):
  95 //      tbz/tbnz Rt imm14
  96 //    1d) Conditional branch (immediate):
  97 //      b.cond imm19
  98 //
  99 // 2) Loads and Stores (insn count = 1)
 100 //    2a) Load register literal:
 101 //      ldr Rt imm19
 102 //
 103 // 3) Data Processing Immediate (insn count = 2 or 3)
 104 //    3a) PC-rel. addressing
 105 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 106 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 107 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 108 //      adr/adrp Rx imm21
 109 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 110 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 111 //      The latter form can only happen when the target is an
 112 //      ExternalAddress, and (by definition) ExternalAddresses don't
 113 //      move. Because of that property, there is never any need to
 114 //      patch the last of the three instructions. However,
 115 //      MacroAssembler::target_addr_for_insn takes all three
 116 //      instructions into account and returns the correct address.
 117 //    3b) Move wide (immediate)
 118 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 119 //
 120 // A switch on a subset of the instruction's bits provides an
 121 // efficient dispatch to these subcases.
 122 //
 123 // insn[28:26] -> main group ('x' == don't care)
 124 //   00x -> UNALLOCATED
 125 //   100 -> Data Processing Immediate
 126 //   101 -> Branch, Exception and System
 127 //   x1x -> Loads and Stores
 128 //
 129 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 130 // n.b. in some cases extra bits need to be checked to verify the
 131 // instruction is as expected
 132 //
 133 // 1) ... xx101x Branch, Exception and System
 134 //   1a)  00___x Unconditional branch (immediate)
 135 //   1b)  01___0 Compare & branch (immediate)
 136 //   1c)  01___1 Test & branch (immediate)
 137 //   1d)  10___0 Conditional branch (immediate)
 138 //        other  Should not happen
 139 //
 140 // 2) ... xxx1x0 Loads and Stores
 141 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 142 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 143 //                strictly should be 64 bit non-FP/SIMD i.e.
 144 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 145 //
 146 // 3) ... xx100x Data Processing Immediate
 147 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 148 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 149 //                 strictly should be 64 bit movz #imm16<<0
 150 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 151 //
 152 
 153 static uint32_t insn_at(address insn_addr, int n) {
 154   return ((uint32_t*)insn_addr)[n];
 155 }
 156 
 157 template<typename T>
 158 class RelocActions : public AllStatic {
 159 
 160 public:
 161 
 162   static int ALWAYSINLINE run(address insn_addr, address &target) {
 163     int instructions = 1;
 164     uint32_t insn = insn_at(insn_addr, 0);
 165 
 166     uint32_t dispatch = Instruction_aarch64::extract(insn, 30, 25);
 167     switch(dispatch) {
 168       case 0b001010:
 169       case 0b001011: {
 170         instructions = T::unconditionalBranch(insn_addr, target);
 171         break;
 172       }
 173       case 0b101010:   // Conditional branch (immediate)
 174       case 0b011010: { // Compare & branch (immediate)
 175         instructions = T::conditionalBranch(insn_addr, target);
 176         break;
 177       }
 178       case 0b011011: {
 179         instructions = T::testAndBranch(insn_addr, target);
 180         break;
 181       }
 182       case 0b001100:
 183       case 0b001110:
 184       case 0b011100:
 185       case 0b011110:
 186       case 0b101100:
 187       case 0b101110:
 188       case 0b111100:
 189       case 0b111110: {
 190         // load/store
 191         if ((Instruction_aarch64::extract(insn, 29, 24) & 0b111011) == 0b011000) {
 192           // Load register (literal)
 193           instructions = T::loadStore(insn_addr, target);
 194           break;
 195         } else {
 196           // nothing to do
 197           assert(target == nullptr, "did not expect to relocate target for polling page load");
 198         }
 199         break;
 200       }
 201       case 0b001000:
 202       case 0b011000:
 203       case 0b101000:
 204       case 0b111000: {
 205         // adr/adrp
 206         assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 207         int shift = Instruction_aarch64::extract(insn, 31, 31);
 208         if (shift) {
 209           uint32_t insn2 = insn_at(insn_addr, 1);
 210           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 211               Instruction_aarch64::extract(insn, 4, 0) ==
 212               Instruction_aarch64::extract(insn2, 9, 5)) {
 213             instructions = T::adrp(insn_addr, target, T::adrpMem);
 214           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 215                      Instruction_aarch64::extract(insn, 4, 0) ==
 216                      Instruction_aarch64::extract(insn2, 4, 0)) {
 217             instructions = T::adrp(insn_addr, target, T::adrpAdd);
 218           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 219                      Instruction_aarch64::extract(insn, 4, 0) ==
 220                      Instruction_aarch64::extract(insn2, 4, 0)) {
 221             instructions = T::adrp(insn_addr, target, T::adrpMovk);
 222           } else {
 223             ShouldNotReachHere();
 224           }
 225         } else {
 226           instructions = T::adr(insn_addr, target);
 227         }
 228         break;
 229       }
 230       case 0b001001:
 231       case 0b011001:
 232       case 0b101001:
 233       case 0b111001: {
 234         instructions = T::immediate(insn_addr, target);
 235         break;
 236       }
 237       default: {
 238         ShouldNotReachHere();
 239       }
 240     }
 241 
 242     T::verify(insn_addr, target);
 243     return instructions * NativeInstruction::instruction_size;
 244   }
 245 };
 246 
 247 class Patcher : public AllStatic {
 248 public:
 249   static int unconditionalBranch(address insn_addr, address &target) {
 250     intptr_t offset = (target - insn_addr) >> 2;
 251     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 252     return 1;
 253   }
 254   static int conditionalBranch(address insn_addr, address &target) {
 255     intptr_t offset = (target - insn_addr) >> 2;
 256     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 257     return 1;
 258   }
 259   static int testAndBranch(address insn_addr, address &target) {
 260     intptr_t offset = (target - insn_addr) >> 2;
 261     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 262     return 1;
 263   }
 264   static int loadStore(address insn_addr, address &target) {
 265     intptr_t offset = (target - insn_addr) >> 2;
 266     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 267     return 1;
 268   }
 269   static int adr(address insn_addr, address &target) {
 270 #ifdef ASSERT
 271     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 272 #endif
 273     // PC-rel. addressing
 274     ptrdiff_t offset = target - insn_addr;
 275     int offset_lo = offset & 3;
 276     offset >>= 2;
 277     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 278     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 279     return 1;
 280   }
 281   template<typename U>
 282   static int adrp(address insn_addr, address &target, U inner) {
 283     int instructions = 1;
 284 #ifdef ASSERT
 285     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 286 #endif
 287     ptrdiff_t offset = target - insn_addr;
 288     instructions = 2;
 289     precond(inner != nullptr);
 290     // Give the inner reloc a chance to modify the target.
 291     address adjusted_target = target;
 292     instructions = inner(insn_addr, adjusted_target);
 293     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 294     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 295     offset = adr_page - pc_page;
 296     int offset_lo = offset & 3;
 297     offset >>= 2;
 298     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 299     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 300     return instructions;
 301   }
 302   static int adrpMem(address insn_addr, address &target) {
 303     uintptr_t dest = (uintptr_t)target;
 304     int offset_lo = dest & 0xfff;
 305     uint32_t insn2 = insn_at(insn_addr, 1);
 306     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 307     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 308     guarantee(((dest >> size) << size) == dest, "misaligned target");
 309     return 2;
 310   }
 311   static int adrpAdd(address insn_addr, address &target) {
 312     uintptr_t dest = (uintptr_t)target;
 313     int offset_lo = dest & 0xfff;
 314     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 315     return 2;
 316   }
 317   static int adrpMovk(address insn_addr, address &target) {
 318     uintptr_t dest = uintptr_t(target);
 319     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 320     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 321     target = address(dest);
 322     return 2;
 323   }
 324   static int immediate(address insn_addr, address &target) {
 325     // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 326     // We encode narrow ones by setting the upper 16 bits in the first
 327     // instruction.
 328     if (Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010101) {
 329       assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 330       narrowKlass nk = CompressedKlassPointers::encode((Klass*)target);
 331       Instruction_aarch64::patch(insn_addr, 20, 5, nk >> 16);
 332       Instruction_aarch64::patch(insn_addr+4, 20, 5, nk & 0xffff);
 333       return 2;
 334     }
 335     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010100, "must be");
 336     uint64_t dest = (uint64_t)target;
 337     // Move wide constant
 338     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 339     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 340     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 341     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 342     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 343     return 3;
 344   }
 345   static void verify(address insn_addr, address &target) {
 346 #ifdef ASSERT
 347     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 348     if (!(address_is == target)) {
 349       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 350       disnm((intptr_t)insn_addr);
 351       assert(address_is == target, "should be");
 352     }
 353 #endif
 354   }
 355 };
 356 
 357 // If insn1 and insn2 use the same register to form an address, either
 358 // by an offsetted LDR or a simple ADD, return the offset. If the
 359 // second instruction is an LDR, the offset may be scaled.
 360 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 361   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 362       Instruction_aarch64::extract(insn1, 4, 0) ==
 363       Instruction_aarch64::extract(insn2, 9, 5)) {
 364     // Load/store register (unsigned immediate)
 365     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 366     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 367     byte_offset <<= size;
 368     return true;
 369   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 370              Instruction_aarch64::extract(insn1, 4, 0) ==
 371              Instruction_aarch64::extract(insn2, 4, 0)) {
 372     // add (immediate)
 373     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 374     return true;
 375   }
 376   return false;
 377 }
 378 
 379 class AArch64Decoder : public AllStatic {
 380 public:
 381 
 382   static int loadStore(address insn_addr, address &target) {
 383     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 384     target = insn_addr + (offset << 2);
 385     return 1;
 386   }
 387   static int unconditionalBranch(address insn_addr, address &target) {
 388     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 25, 0);
 389     target = insn_addr + (offset << 2);
 390     return 1;
 391   }
 392   static int conditionalBranch(address insn_addr, address &target) {
 393     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 394     target = address(((uint64_t)insn_addr + (offset << 2)));
 395     return 1;
 396   }
 397   static int testAndBranch(address insn_addr, address &target) {
 398     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 18, 5);
 399     target = address(((uint64_t)insn_addr + (offset << 2)));
 400     return 1;
 401   }
 402   static int adr(address insn_addr, address &target) {
 403     // PC-rel. addressing
 404     uint32_t insn = insn_at(insn_addr, 0);
 405     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 406     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 407     target = address((uint64_t)insn_addr + offset);
 408     return 1;
 409   }
 410   template<typename U>
 411   static int adrp(address insn_addr, address &target, U inner) {
 412     uint32_t insn = insn_at(insn_addr, 0);
 413     assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 414     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 415     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 416     int shift = 12;
 417     offset <<= shift;
 418     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 419     target_page &= ((uint64_t)-1) << shift;
 420     uint32_t insn2 = insn_at(insn_addr, 1);
 421     target = address(target_page);
 422     precond(inner != nullptr);
 423     inner(insn_addr, target);
 424     return 2;
 425   }
 426   static int adrpMem(address insn_addr, address &target) {
 427     uint32_t insn2 = insn_at(insn_addr, 1);
 428     // Load/store register (unsigned immediate)
 429     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 430     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 431     byte_offset <<= size;
 432     target += byte_offset;
 433     return 2;
 434   }
 435   static int adrpAdd(address insn_addr, address &target) {
 436     uint32_t insn2 = insn_at(insn_addr, 1);
 437     // add (immediate)
 438     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 439     target += byte_offset;
 440     return 2;
 441   }
 442   static int adrpMovk(address insn_addr, address &target) {
 443     uint32_t insn2 = insn_at(insn_addr, 1);
 444     uint64_t dest = uint64_t(target);
 445     dest = (dest & 0xffff0000ffffffff) |
 446       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 447     target = address(dest);
 448 
 449     // We know the destination 4k page. Maybe we have a third
 450     // instruction.
 451     uint32_t insn = insn_at(insn_addr, 0);
 452     uint32_t insn3 = insn_at(insn_addr, 2);
 453     ptrdiff_t byte_offset;
 454     if (offset_for(insn, insn3, byte_offset)) {
 455       target += byte_offset;
 456       return 3;
 457     } else {
 458       return 2;
 459     }
 460   }
 461   static int immediate(address insn_addr, address &target) {
 462     uint32_t *insns = (uint32_t *)insn_addr;
 463     // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 464     // We encode narrow ones by setting the upper 16 bits in the first
 465     // instruction.
 466     if (Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010101) {
 467       assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 468       narrowKlass nk = (narrowKlass)((uint32_t(Instruction_aarch64::extract(insns[0], 20, 5)) << 16)
 469                                    +  uint32_t(Instruction_aarch64::extract(insns[1], 20, 5)));
 470       target = (address)CompressedKlassPointers::decode(nk);
 471       return 2;
 472     }
 473     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 474     // Move wide constant: movz, movk, movk.  See movptr().
 475     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 476     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 477     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 478                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 479                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 480     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 481     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 482     return 3;
 483   }
 484   static void verify(address insn_addr, address &target) {
 485   }
 486 };
 487 
 488 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 489   address target;
 490   RelocActions<AArch64Decoder>::run(insn_addr, target);
 491   return target;
 492 }
 493 
 494 // Patch any kind of instruction; there may be several instructions.
 495 // Return the total length (in bytes) of the instructions.
 496 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 497   return RelocActions<Patcher>::run(insn_addr, target);
 498 }
 499 
 500 int MacroAssembler::patch_oop(address insn_addr, address o) {
 501   int instructions;
 502   unsigned insn = *(unsigned*)insn_addr;
 503   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 504 
 505   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 506   // narrow OOPs by setting the upper 16 bits in the first
 507   // instruction.
 508   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 509     // Move narrow OOP
 510     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 511     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 512     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 513     instructions = 2;
 514   } else {
 515     // Move wide OOP
 516     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 517     uintptr_t dest = (uintptr_t)o;
 518     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 519     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 520     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 521     instructions = 3;
 522   }
 523   return instructions * NativeInstruction::instruction_size;
 524 }
 525 
 526 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 527   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 528   // We encode narrow ones by setting the upper 16 bits in the first
 529   // instruction.
 530   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 531   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 532          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 533 
 534   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 535   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 536   return 2 * NativeInstruction::instruction_size;
 537 }
 538 
 539 address MacroAssembler::target_addr_for_insn_or_null(address insn_addr) {
 540   if (NativeInstruction::is_ldrw_to_zr(insn_addr)) {
 541     return nullptr;
 542   }
 543   return MacroAssembler::target_addr_for_insn(insn_addr);
 544 }
 545 
 546 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 547   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 548   if (at_return) {
 549     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 550     // we may safely use the sp instead to perform the stack watermark check.
 551     cmp(in_nmethod ? sp : rfp, tmp);
 552     br(Assembler::HI, slow_path);
 553   } else {
 554     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 555   }
 556 }
 557 
 558 void MacroAssembler::rt_call(address dest, Register tmp) {
 559   CodeBlob *cb = CodeCache::find_blob(dest);
 560   if (cb) {
 561     far_call(RuntimeAddress(dest));
 562   } else {
 563     lea(tmp, RuntimeAddress(dest));
 564     blr(tmp);
 565   }
 566 }
 567 
 568 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 569   if (!Continuations::enabled()) return;
 570   Label done;
 571   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 572   cmp(sp, rscratch1);
 573   br(Assembler::LS, done);
 574   mov(rscratch1, sp); // we can't use sp as the source in str
 575   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 576   bind(done);
 577 }
 578 
 579 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 580   if (!Continuations::enabled()) return;
 581   Label done;
 582   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 583   cmp(sp, rscratch1);
 584   br(Assembler::LO, done);
 585   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 586   bind(done);
 587 }
 588 
 589 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 590   // we must set sp to zero to clear frame
 591   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 592 
 593   // must clear fp, so that compiled frames are not confused; it is
 594   // possible that we need it only for debugging
 595   if (clear_fp) {
 596     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 597   }
 598 
 599   // Always clear the pc because it could have been set by make_walkable()
 600   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 601 }
 602 
 603 // Calls to C land
 604 //
 605 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 606 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 607 // has to be reset to 0. This is required to allow proper stack traversal.
 608 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 609                                          Register last_java_fp,
 610                                          Register last_java_pc,
 611                                          Register scratch) {
 612 
 613   if (last_java_pc->is_valid()) {
 614       str(last_java_pc, Address(rthread,
 615                                 JavaThread::frame_anchor_offset()
 616                                 + JavaFrameAnchor::last_Java_pc_offset()));
 617     }
 618 
 619   // determine last_java_sp register
 620   if (last_java_sp == sp) {
 621     mov(scratch, sp);
 622     last_java_sp = scratch;
 623   } else if (!last_java_sp->is_valid()) {
 624     last_java_sp = esp;
 625   }
 626 
 627   // last_java_fp is optional
 628   if (last_java_fp->is_valid()) {
 629     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 630   }
 631 
 632   // We must set sp last.
 633   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 634 }
 635 
 636 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 637                                          Register last_java_fp,
 638                                          address  last_java_pc,
 639                                          Register scratch) {
 640   assert(last_java_pc != nullptr, "must provide a valid PC");
 641 
 642   adr(scratch, last_java_pc);
 643   str(scratch, Address(rthread,
 644                        JavaThread::frame_anchor_offset()
 645                        + JavaFrameAnchor::last_Java_pc_offset()));
 646 
 647   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 648 }
 649 
 650 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 651                                          Register last_java_fp,
 652                                          Label &L,
 653                                          Register scratch) {
 654   if (L.is_bound()) {
 655     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 656   } else {
 657     InstructionMark im(this);
 658     L.add_patch_at(code(), locator());
 659     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 660   }
 661 }
 662 
 663 static inline bool target_needs_far_branch(address addr) {
 664   if (AOTCodeCache::is_on_for_dump()) {
 665     return true;
 666   }
 667   // codecache size <= 128M
 668   if (!MacroAssembler::far_branches()) {
 669     return false;
 670   }
 671   // codecache size > 240M
 672   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 673     return true;
 674   }
 675   // codecache size: 128M..240M
 676   return !CodeCache::is_non_nmethod(addr);
 677 }
 678 
 679 void MacroAssembler::far_call(Address entry, Register tmp) {
 680   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 681   assert(CodeCache::find_blob(entry.target()) != nullptr,
 682          "destination of far call not found in code cache");
 683   assert(entry.rspec().type() == relocInfo::external_word_type
 684          || entry.rspec().type() == relocInfo::runtime_call_type
 685          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 686   if (target_needs_far_branch(entry.target())) {
 687     uint64_t offset;
 688     // We can use ADRP here because we know that the total size of
 689     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 690     adrp(tmp, entry, offset);
 691     add(tmp, tmp, offset);
 692     blr(tmp);
 693   } else {
 694     bl(entry);
 695   }
 696 }
 697 
 698 int MacroAssembler::far_jump(Address entry, Register tmp) {
 699   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 700   assert(CodeCache::find_blob(entry.target()) != nullptr,
 701          "destination of far call not found in code cache");
 702   assert(entry.rspec().type() == relocInfo::external_word_type
 703          || entry.rspec().type() == relocInfo::runtime_call_type
 704          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 705   address start = pc();
 706   if (target_needs_far_branch(entry.target())) {
 707     uint64_t offset;
 708     // We can use ADRP here because we know that the total size of
 709     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 710     adrp(tmp, entry, offset);
 711     add(tmp, tmp, offset);
 712     br(tmp);
 713   } else {
 714     b(entry);
 715   }
 716   return pc() - start;
 717 }
 718 
 719 void MacroAssembler::reserved_stack_check() {
 720     // testing if reserved zone needs to be enabled
 721     Label no_reserved_zone_enabling;
 722 
 723     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 724     cmp(sp, rscratch1);
 725     br(Assembler::LO, no_reserved_zone_enabling);
 726 
 727     enter();   // LR and FP are live.
 728     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 729     mov(c_rarg0, rthread);
 730     blr(rscratch1);
 731     leave();
 732 
 733     // We have already removed our own frame.
 734     // throw_delayed_StackOverflowError will think that it's been
 735     // called by our caller.
 736     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 737     br(rscratch1);
 738     should_not_reach_here();
 739 
 740     bind(no_reserved_zone_enabling);
 741 }
 742 
 743 static void pass_arg0(MacroAssembler* masm, Register arg) {
 744   if (c_rarg0 != arg ) {
 745     masm->mov(c_rarg0, arg);
 746   }
 747 }
 748 
 749 static void pass_arg1(MacroAssembler* masm, Register arg) {
 750   if (c_rarg1 != arg ) {
 751     masm->mov(c_rarg1, arg);
 752   }
 753 }
 754 
 755 static void pass_arg2(MacroAssembler* masm, Register arg) {
 756   if (c_rarg2 != arg ) {
 757     masm->mov(c_rarg2, arg);
 758   }
 759 }
 760 
 761 static void pass_arg3(MacroAssembler* masm, Register arg) {
 762   if (c_rarg3 != arg ) {
 763     masm->mov(c_rarg3, arg);
 764   }
 765 }
 766 
 767 void MacroAssembler::call_VM_base(Register oop_result,
 768                                   Register java_thread,
 769                                   Register last_java_sp,
 770                                   Label*   return_pc,
 771                                   address  entry_point,
 772                                   int      number_of_arguments,
 773                                   bool     check_exceptions) {
 774    // determine java_thread register
 775   if (!java_thread->is_valid()) {
 776     java_thread = rthread;
 777   }
 778 
 779   // determine last_java_sp register
 780   if (!last_java_sp->is_valid()) {
 781     last_java_sp = esp;
 782   }
 783 
 784   // debugging support
 785   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 786   assert(java_thread == rthread, "unexpected register");
 787 #ifdef ASSERT
 788   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 789   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 790 #endif // ASSERT
 791 
 792   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 793   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 794 
 795   // push java thread (becomes first argument of C function)
 796 
 797   mov(c_rarg0, java_thread);
 798 
 799   // set last Java frame before call
 800   assert(last_java_sp != rfp, "can't use rfp");
 801 
 802   Label l;
 803   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 804 
 805   // do the call, remove parameters
 806   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 807 
 808   // lr could be poisoned with PAC signature during throw_pending_exception
 809   // if it was tail-call optimized by compiler, since lr is not callee-saved
 810   // reload it with proper value
 811   adr(lr, l);
 812 
 813   // reset last Java frame
 814   // Only interpreter should have to clear fp
 815   reset_last_Java_frame(true);
 816 
 817    // C++ interp handles this in the interpreter
 818   check_and_handle_popframe(java_thread);
 819   check_and_handle_earlyret(java_thread);
 820 
 821   if (check_exceptions) {
 822     // check for pending exceptions (java_thread is set upon return)
 823     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 824     Label ok;
 825     cbz(rscratch1, ok);
 826     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 827     br(rscratch1);
 828     bind(ok);
 829   }
 830 
 831   // get oop result if there is one and reset the value in the thread
 832   if (oop_result->is_valid()) {
 833     get_vm_result_oop(oop_result, java_thread);
 834   }
 835 }
 836 
 837 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 838   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 839 }
 840 
 841 // Check the entry target is always reachable from any branch.
 842 static bool is_always_within_branch_range(Address entry) {
 843   if (AOTCodeCache::is_on_for_dump()) {
 844     return false;
 845   }
 846   const address target = entry.target();
 847 
 848   if (!CodeCache::contains(target)) {
 849     // We always use trampolines for callees outside CodeCache.
 850     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 851     return false;
 852   }
 853 
 854   if (!MacroAssembler::far_branches()) {
 855     return true;
 856   }
 857 
 858   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 859     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 860     // Non-compiled methods stay forever in CodeCache.
 861     // We check whether the longest possible branch is within the branch range.
 862     assert(CodeCache::find_blob(target) != nullptr &&
 863           !CodeCache::find_blob(target)->is_nmethod(),
 864           "runtime call of compiled method");
 865     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 866     const address left_longest_branch_start = CodeCache::low_bound();
 867     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 868                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 869     return is_reachable;
 870   }
 871 
 872   return false;
 873 }
 874 
 875 // Maybe emit a call via a trampoline. If the code cache is small
 876 // trampolines won't be emitted.
 877 address MacroAssembler::trampoline_call(Address entry) {
 878   assert(entry.rspec().type() == relocInfo::runtime_call_type
 879          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 880          || entry.rspec().type() == relocInfo::static_call_type
 881          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 882 
 883   address target = entry.target();
 884 
 885   if (!is_always_within_branch_range(entry)) {
 886     if (!in_scratch_emit_size()) {
 887       // We don't want to emit a trampoline if C2 is generating dummy
 888       // code during its branch shortening phase.
 889       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 890         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 891         code()->share_trampoline_for(entry.target(), offset());
 892       } else {
 893         address stub = emit_trampoline_stub(offset(), target);
 894         if (stub == nullptr) {
 895           postcond(pc() == badAddress);
 896           return nullptr; // CodeCache is full
 897         }
 898       }
 899     }
 900     target = pc();
 901   }
 902 
 903   address call_pc = pc();
 904   relocate(entry.rspec());
 905   bl(target);
 906 
 907   postcond(pc() != badAddress);
 908   return call_pc;
 909 }
 910 
 911 // Emit a trampoline stub for a call to a target which is too far away.
 912 //
 913 // code sequences:
 914 //
 915 // call-site:
 916 //   branch-and-link to <destination> or <trampoline stub>
 917 //
 918 // Related trampoline stub for this call site in the stub section:
 919 //   load the call target from the constant pool
 920 //   branch (LR still points to the call site above)
 921 
 922 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 923                                              address dest) {
 924   // Max stub size: alignment nop, TrampolineStub.
 925   address stub = start_a_stub(max_trampoline_stub_size());
 926   if (stub == nullptr) {
 927     return nullptr;  // CodeBuffer::expand failed
 928   }
 929 
 930   // Create a trampoline stub relocation which relates this trampoline stub
 931   // with the call instruction at insts_call_instruction_offset in the
 932   // instructions code-section.
 933   align(wordSize);
 934   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 935                                             + insts_call_instruction_offset));
 936   const int stub_start_offset = offset();
 937 
 938   // Now, create the trampoline stub's code:
 939   // - load the call
 940   // - call
 941   Label target;
 942   ldr(rscratch1, target);
 943   br(rscratch1);
 944   bind(target);
 945   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 946          "should be");
 947   emit_int64((int64_t)dest);
 948 
 949   const address stub_start_addr = addr_at(stub_start_offset);
 950 
 951   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 952 
 953   end_a_stub();
 954   return stub_start_addr;
 955 }
 956 
 957 int MacroAssembler::max_trampoline_stub_size() {
 958   // Max stub size: alignment nop, TrampolineStub.
 959   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 960 }
 961 
 962 void MacroAssembler::emit_static_call_stub() {
 963   // CompiledDirectCall::set_to_interpreted knows the
 964   // exact layout of this stub.
 965 
 966   isb();
 967   mov_metadata(rmethod, nullptr);
 968 
 969   // Jump to the entry point of the c2i stub.
 970   if (codestub_branch_needs_far_jump()) {
 971     movptr(rscratch1, 0);
 972     br(rscratch1);
 973   } else {
 974     b(pc());
 975   }
 976 }
 977 
 978 int MacroAssembler::max_static_call_stub_size() {
 979   // isb; movk; movz; movz; movk; movz; movz; br
 980   return 8 * NativeInstruction::instruction_size;
 981 }
 982 
 983 void MacroAssembler::c2bool(Register x) {
 984   // implements x == 0 ? 0 : 1
 985   // note: must only look at least-significant byte of x
 986   //       since C-style booleans are stored in one byte
 987   //       only! (was bug)
 988   tst(x, 0xff);
 989   cset(x, Assembler::NE);
 990 }
 991 
 992 address MacroAssembler::ic_call(address entry, jint method_index) {
 993   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 994   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 995   return trampoline_call(Address(entry, rh));
 996 }
 997 
 998 int MacroAssembler::ic_check_size() {
 999   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
1000   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
1001     return NativeInstruction::instruction_size * (7 + extra_instructions);
1002   } else {
1003     return NativeInstruction::instruction_size * (5 + extra_instructions);
1004   }
1005 }
1006 
1007 int MacroAssembler::ic_check(int end_alignment) {
1008   Register receiver = j_rarg0;
1009   Register data = rscratch2;
1010   Register tmp1 = rscratch1;
1011   Register tmp2 = r10;
1012 
1013   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
1014   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
1015   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1016   // before the inline cache check here, and not after
1017   align(end_alignment, offset() + ic_check_size());
1018 
1019   int uep_offset = offset();
1020 
1021   if (UseCompactObjectHeaders) {
1022     load_narrow_klass_compact(tmp1, receiver);
1023     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1024     cmpw(tmp1, tmp2);
1025   } else if (UseCompressedClassPointers) {
1026     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1027     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1028     cmpw(tmp1, tmp2);
1029   } else {
1030     ldr(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1031     ldr(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1032     cmp(tmp1, tmp2);
1033   }
1034 
1035   Label dont;
1036   br(Assembler::EQ, dont);
1037   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1038   bind(dont);
1039   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1040 
1041   return uep_offset;
1042 }
1043 
1044 // Implementation of call_VM versions
1045 
1046 void MacroAssembler::call_VM(Register oop_result,
1047                              address entry_point,
1048                              bool check_exceptions) {
1049   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1050 }
1051 
1052 void MacroAssembler::call_VM(Register oop_result,
1053                              address entry_point,
1054                              Register arg_1,
1055                              bool check_exceptions) {
1056   pass_arg1(this, arg_1);
1057   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1058 }
1059 
1060 void MacroAssembler::call_VM(Register oop_result,
1061                              address entry_point,
1062                              Register arg_1,
1063                              Register arg_2,
1064                              bool check_exceptions) {
1065   assert_different_registers(arg_1, c_rarg2);
1066   pass_arg2(this, arg_2);
1067   pass_arg1(this, arg_1);
1068   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1069 }
1070 
1071 void MacroAssembler::call_VM(Register oop_result,
1072                              address entry_point,
1073                              Register arg_1,
1074                              Register arg_2,
1075                              Register arg_3,
1076                              bool check_exceptions) {
1077   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1078   assert_different_registers(arg_2, c_rarg3);
1079   pass_arg3(this, arg_3);
1080 
1081   pass_arg2(this, arg_2);
1082 
1083   pass_arg1(this, arg_1);
1084   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1085 }
1086 
1087 void MacroAssembler::call_VM(Register oop_result,
1088                              Register last_java_sp,
1089                              address entry_point,
1090                              int number_of_arguments,
1091                              bool check_exceptions) {
1092   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1093 }
1094 
1095 void MacroAssembler::call_VM(Register oop_result,
1096                              Register last_java_sp,
1097                              address entry_point,
1098                              Register arg_1,
1099                              bool check_exceptions) {
1100   pass_arg1(this, arg_1);
1101   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1102 }
1103 
1104 void MacroAssembler::call_VM(Register oop_result,
1105                              Register last_java_sp,
1106                              address entry_point,
1107                              Register arg_1,
1108                              Register arg_2,
1109                              bool check_exceptions) {
1110 
1111   assert_different_registers(arg_1, c_rarg2);
1112   pass_arg2(this, arg_2);
1113   pass_arg1(this, arg_1);
1114   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1115 }
1116 
1117 void MacroAssembler::call_VM(Register oop_result,
1118                              Register last_java_sp,
1119                              address entry_point,
1120                              Register arg_1,
1121                              Register arg_2,
1122                              Register arg_3,
1123                              bool check_exceptions) {
1124   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1125   assert_different_registers(arg_2, c_rarg3);
1126   pass_arg3(this, arg_3);
1127   pass_arg2(this, arg_2);
1128   pass_arg1(this, arg_1);
1129   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1130 }
1131 
1132 
1133 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1134   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1135   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1136   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1137 }
1138 
1139 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1140   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1141   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1142 }
1143 
1144 void MacroAssembler::align(int modulus) {
1145   align(modulus, offset());
1146 }
1147 
1148 // Ensure that the code at target bytes offset from the current offset() is aligned
1149 // according to modulus.
1150 void MacroAssembler::align(int modulus, int target) {
1151   int delta = target - offset();
1152   while ((offset() + delta) % modulus != 0) nop();
1153 }
1154 
1155 void MacroAssembler::post_call_nop() {
1156   if (!Continuations::enabled()) {
1157     return;
1158   }
1159   InstructionMark im(this);
1160   relocate(post_call_nop_Relocation::spec());
1161   InlineSkippedInstructionsCounter skipCounter(this);
1162   nop();
1163   movk(zr, 0);
1164   movk(zr, 0);
1165 }
1166 
1167 // these are no-ops overridden by InterpreterMacroAssembler
1168 
1169 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1170 
1171 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1172 
1173 // Look up the method for a megamorphic invokeinterface call.
1174 // The target method is determined by <intf_klass, itable_index>.
1175 // The receiver klass is in recv_klass.
1176 // On success, the result will be in method_result, and execution falls through.
1177 // On failure, execution transfers to the given label.
1178 void MacroAssembler::lookup_interface_method(Register recv_klass,
1179                                              Register intf_klass,
1180                                              RegisterOrConstant itable_index,
1181                                              Register method_result,
1182                                              Register scan_temp,
1183                                              Label& L_no_such_interface,
1184                          bool return_method) {
1185   assert_different_registers(recv_klass, intf_klass, scan_temp);
1186   assert_different_registers(method_result, intf_klass, scan_temp);
1187   assert(recv_klass != method_result || !return_method,
1188      "recv_klass can be destroyed when method isn't needed");
1189   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1190          "caller must use same register for non-constant itable index as for method");
1191 
1192   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1193   int vtable_base = in_bytes(Klass::vtable_start_offset());
1194   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1195   int scan_step   = itableOffsetEntry::size() * wordSize;
1196   int vte_size    = vtableEntry::size_in_bytes();
1197   assert(vte_size == wordSize, "else adjust times_vte_scale");
1198 
1199   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1200 
1201   // Could store the aligned, prescaled offset in the klass.
1202   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1203   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1204   add(scan_temp, scan_temp, vtable_base);
1205 
1206   if (return_method) {
1207     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1208     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1209     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1210     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1211     if (itentry_off)
1212       add(recv_klass, recv_klass, itentry_off);
1213   }
1214 
1215   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1216   //   if (scan->interface() == intf) {
1217   //     result = (klass + scan->offset() + itable_index);
1218   //   }
1219   // }
1220   Label search, found_method;
1221 
1222   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1223   cmp(intf_klass, method_result);
1224   br(Assembler::EQ, found_method);
1225   bind(search);
1226   // Check that the previous entry is non-null.  A null entry means that
1227   // the receiver class doesn't implement the interface, and wasn't the
1228   // same as when the caller was compiled.
1229   cbz(method_result, L_no_such_interface);
1230   if (itableOffsetEntry::interface_offset() != 0) {
1231     add(scan_temp, scan_temp, scan_step);
1232     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1233   } else {
1234     ldr(method_result, Address(pre(scan_temp, scan_step)));
1235   }
1236   cmp(intf_klass, method_result);
1237   br(Assembler::NE, search);
1238 
1239   bind(found_method);
1240 
1241   // Got a hit.
1242   if (return_method) {
1243     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1244     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1245   }
1246 }
1247 
1248 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1249 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1250 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1251 // The target method is determined by <holder_klass, itable_index>.
1252 // The receiver klass is in recv_klass.
1253 // On success, the result will be in method_result, and execution falls through.
1254 // On failure, execution transfers to the given label.
1255 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1256                                                   Register holder_klass,
1257                                                   Register resolved_klass,
1258                                                   Register method_result,
1259                                                   Register temp_itbl_klass,
1260                                                   Register scan_temp,
1261                                                   int itable_index,
1262                                                   Label& L_no_such_interface) {
1263   // 'method_result' is only used as output register at the very end of this method.
1264   // Until then we can reuse it as 'holder_offset'.
1265   Register holder_offset = method_result;
1266   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1267 
1268   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1269   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1270   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1271   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1272 
1273   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1274 
1275   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1276   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1277   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1278   // temp_itbl_klass = itable[0]._interface;
1279   int vtblEntrySize = vtableEntry::size_in_bytes();
1280   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1281   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1282   mov(holder_offset, zr);
1283   // scan_temp = &(itable[0]._interface)
1284   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1285 
1286   // Initial checks:
1287   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1288   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1289   //   - if (itable[0] == 0), no such interface
1290   cmp(resolved_klass, holder_klass);
1291   br(Assembler::NE, L_loop_search_resolved_entry);
1292   cmp(holder_klass, temp_itbl_klass);
1293   br(Assembler::EQ, L_holder_found);
1294   cbz(temp_itbl_klass, L_no_such_interface);
1295 
1296   // Loop: Look for holder_klass record in itable
1297   //   do {
1298   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1299   //     if (temp_itbl_klass == holder_klass) {
1300   //       goto L_holder_found; // Found!
1301   //     }
1302   //   } while (temp_itbl_klass != 0);
1303   //   goto L_no_such_interface // Not found.
1304   Label L_search_holder;
1305   bind(L_search_holder);
1306     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1307     cmp(holder_klass, temp_itbl_klass);
1308     br(Assembler::EQ, L_holder_found);
1309     cbnz(temp_itbl_klass, L_search_holder);
1310 
1311   b(L_no_such_interface);
1312 
1313   // Loop: Look for resolved_class record in itable
1314   //   while (true) {
1315   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1316   //     if (temp_itbl_klass == 0) {
1317   //       goto L_no_such_interface;
1318   //     }
1319   //     if (temp_itbl_klass == resolved_klass) {
1320   //        goto L_resolved_found;  // Found!
1321   //     }
1322   //     if (temp_itbl_klass == holder_klass) {
1323   //        holder_offset = scan_temp;
1324   //     }
1325   //   }
1326   //
1327   Label L_loop_search_resolved;
1328   bind(L_loop_search_resolved);
1329     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1330   bind(L_loop_search_resolved_entry);
1331     cbz(temp_itbl_klass, L_no_such_interface);
1332     cmp(resolved_klass, temp_itbl_klass);
1333     br(Assembler::EQ, L_resolved_found);
1334     cmp(holder_klass, temp_itbl_klass);
1335     br(Assembler::NE, L_loop_search_resolved);
1336     mov(holder_offset, scan_temp);
1337     b(L_loop_search_resolved);
1338 
1339   // See if we already have a holder klass. If not, go and scan for it.
1340   bind(L_resolved_found);
1341   cbz(holder_offset, L_search_holder);
1342   mov(scan_temp, holder_offset);
1343 
1344   // Finally, scan_temp contains holder_klass vtable offset
1345   bind(L_holder_found);
1346   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1347   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1348     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1349   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1350 }
1351 
1352 // virtual method calling
1353 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1354                                            RegisterOrConstant vtable_index,
1355                                            Register method_result) {
1356   assert(vtableEntry::size() * wordSize == 8,
1357          "adjust the scaling in the code below");
1358   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1359 
1360   if (vtable_index.is_register()) {
1361     lea(method_result, Address(recv_klass,
1362                                vtable_index.as_register(),
1363                                Address::lsl(LogBytesPerWord)));
1364     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1365   } else {
1366     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1367     ldr(method_result,
1368         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1369   }
1370 }
1371 
1372 void MacroAssembler::check_klass_subtype(Register sub_klass,
1373                            Register super_klass,
1374                            Register temp_reg,
1375                            Label& L_success) {
1376   Label L_failure;
1377   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1378   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1379   bind(L_failure);
1380 }
1381 
1382 
1383 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1384                                                    Register super_klass,
1385                                                    Register temp_reg,
1386                                                    Label* L_success,
1387                                                    Label* L_failure,
1388                                                    Label* L_slow_path,
1389                                                    Register super_check_offset) {
1390   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1391   bool must_load_sco = ! super_check_offset->is_valid();
1392   if (must_load_sco) {
1393     assert(temp_reg != noreg, "supply either a temp or a register offset");
1394   }
1395 
1396   Label L_fallthrough;
1397   int label_nulls = 0;
1398   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1399   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1400   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1401   assert(label_nulls <= 1, "at most one null in the batch");
1402 
1403   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1404   Address super_check_offset_addr(super_klass, sco_offset);
1405 
1406   // Hacked jmp, which may only be used just before L_fallthrough.
1407 #define final_jmp(label)                                                \
1408   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1409   else                            b(label)                /*omit semi*/
1410 
1411   // If the pointers are equal, we are done (e.g., String[] elements).
1412   // This self-check enables sharing of secondary supertype arrays among
1413   // non-primary types such as array-of-interface.  Otherwise, each such
1414   // type would need its own customized SSA.
1415   // We move this check to the front of the fast path because many
1416   // type checks are in fact trivially successful in this manner,
1417   // so we get a nicely predicted branch right at the start of the check.
1418   cmp(sub_klass, super_klass);
1419   br(Assembler::EQ, *L_success);
1420 
1421   // Check the supertype display:
1422   if (must_load_sco) {
1423     ldrw(temp_reg, super_check_offset_addr);
1424     super_check_offset = temp_reg;
1425   }
1426 
1427   Address super_check_addr(sub_klass, super_check_offset);
1428   ldr(rscratch1, super_check_addr);
1429   cmp(super_klass, rscratch1); // load displayed supertype
1430   br(Assembler::EQ, *L_success);
1431 
1432   // This check has worked decisively for primary supers.
1433   // Secondary supers are sought in the super_cache ('super_cache_addr').
1434   // (Secondary supers are interfaces and very deeply nested subtypes.)
1435   // This works in the same check above because of a tricky aliasing
1436   // between the super_cache and the primary super display elements.
1437   // (The 'super_check_addr' can address either, as the case requires.)
1438   // Note that the cache is updated below if it does not help us find
1439   // what we need immediately.
1440   // So if it was a primary super, we can just fail immediately.
1441   // Otherwise, it's the slow path for us (no success at this point).
1442 
1443   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1444   if (L_failure == &L_fallthrough) {
1445     cbz(rscratch1, *L_slow_path);
1446   } else {
1447     cbnz(rscratch1, *L_failure);
1448     final_jmp(*L_slow_path);
1449   }
1450 
1451   bind(L_fallthrough);
1452 
1453 #undef final_jmp
1454 }
1455 
1456 // These two are taken from x86, but they look generally useful
1457 
1458 // scans count pointer sized words at [addr] for occurrence of value,
1459 // generic
1460 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1461                                 Register scratch) {
1462   Label Lloop, Lexit;
1463   cbz(count, Lexit);
1464   bind(Lloop);
1465   ldr(scratch, post(addr, wordSize));
1466   cmp(value, scratch);
1467   br(EQ, Lexit);
1468   sub(count, count, 1);
1469   cbnz(count, Lloop);
1470   bind(Lexit);
1471 }
1472 
1473 // scans count 4 byte words at [addr] for occurrence of value,
1474 // generic
1475 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1476                                 Register scratch) {
1477   Label Lloop, Lexit;
1478   cbz(count, Lexit);
1479   bind(Lloop);
1480   ldrw(scratch, post(addr, wordSize));
1481   cmpw(value, scratch);
1482   br(EQ, Lexit);
1483   sub(count, count, 1);
1484   cbnz(count, Lloop);
1485   bind(Lexit);
1486 }
1487 
1488 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1489                                                           Register super_klass,
1490                                                           Register temp_reg,
1491                                                           Register temp2_reg,
1492                                                           Label* L_success,
1493                                                           Label* L_failure,
1494                                                           bool set_cond_codes) {
1495   // NB! Callers may assume that, when temp2_reg is a valid register,
1496   // this code sets it to a nonzero value.
1497 
1498   assert_different_registers(sub_klass, super_klass, temp_reg);
1499   if (temp2_reg != noreg)
1500     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1501 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1502 
1503   Label L_fallthrough;
1504   int label_nulls = 0;
1505   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1506   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1507   assert(label_nulls <= 1, "at most one null in the batch");
1508 
1509   // a couple of useful fields in sub_klass:
1510   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1511   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1512   Address secondary_supers_addr(sub_klass, ss_offset);
1513   Address super_cache_addr(     sub_klass, sc_offset);
1514 
1515   BLOCK_COMMENT("check_klass_subtype_slow_path");
1516 
1517   // Do a linear scan of the secondary super-klass chain.
1518   // This code is rarely used, so simplicity is a virtue here.
1519   // The repne_scan instruction uses fixed registers, which we must spill.
1520   // Don't worry too much about pre-existing connections with the input regs.
1521 
1522   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1523   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1524 
1525   RegSet pushed_registers;
1526   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1527   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1528 
1529   if (super_klass != r0) {
1530     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1531   }
1532 
1533   push(pushed_registers, sp);
1534 
1535   // Get super_klass value into r0 (even if it was in r5 or r2).
1536   if (super_klass != r0) {
1537     mov(r0, super_klass);
1538   }
1539 
1540 #ifndef PRODUCT
1541   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1542 #endif //PRODUCT
1543 
1544   // We will consult the secondary-super array.
1545   ldr(r5, secondary_supers_addr);
1546   // Load the array length.
1547   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1548   // Skip to start of data.
1549   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1550 
1551   cmp(sp, zr); // Clear Z flag; SP is never zero
1552   // Scan R2 words at [R5] for an occurrence of R0.
1553   // Set NZ/Z based on last compare.
1554   repne_scan(r5, r0, r2, rscratch1);
1555 
1556   // Unspill the temp. registers:
1557   pop(pushed_registers, sp);
1558 
1559   br(Assembler::NE, *L_failure);
1560 
1561   // Success.  Cache the super we found and proceed in triumph.
1562 
1563   if (UseSecondarySupersCache) {
1564     str(super_klass, super_cache_addr);
1565   }
1566 
1567   if (L_success != &L_fallthrough) {
1568     b(*L_success);
1569   }
1570 
1571 #undef IS_A_TEMP
1572 
1573   bind(L_fallthrough);
1574 }
1575 
1576 // If Register r is invalid, remove a new register from
1577 // available_regs, and add new register to regs_to_push.
1578 Register MacroAssembler::allocate_if_noreg(Register r,
1579                                   RegSetIterator<Register> &available_regs,
1580                                   RegSet &regs_to_push) {
1581   if (!r->is_valid()) {
1582     r = *available_regs++;
1583     regs_to_push += r;
1584   }
1585   return r;
1586 }
1587 
1588 // check_klass_subtype_slow_path_table() looks for super_klass in the
1589 // hash table belonging to super_klass, branching to L_success or
1590 // L_failure as appropriate. This is essentially a shim which
1591 // allocates registers as necessary then calls
1592 // lookup_secondary_supers_table() to do the work. Any of the temp
1593 // regs may be noreg, in which case this logic will chooses some
1594 // registers push and pop them from the stack.
1595 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1596                                                          Register super_klass,
1597                                                          Register temp_reg,
1598                                                          Register temp2_reg,
1599                                                          Register temp3_reg,
1600                                                          Register result_reg,
1601                                                          FloatRegister vtemp,
1602                                                          Label* L_success,
1603                                                          Label* L_failure,
1604                                                          bool set_cond_codes) {
1605   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1606 
1607   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1608 
1609   Label L_fallthrough;
1610   int label_nulls = 0;
1611   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1612   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1613   assert(label_nulls <= 1, "at most one null in the batch");
1614 
1615   BLOCK_COMMENT("check_klass_subtype_slow_path");
1616 
1617   RegSetIterator<Register> available_regs
1618     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1619 
1620   RegSet pushed_regs;
1621 
1622   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1623   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1624   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1625   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1626 
1627   push(pushed_regs, sp);
1628 
1629   lookup_secondary_supers_table_var(sub_klass,
1630                                     super_klass,
1631                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1632                                     nullptr);
1633   cmp(result_reg, zr);
1634 
1635   // Unspill the temp. registers:
1636   pop(pushed_regs, sp);
1637 
1638   // NB! Callers may assume that, when set_cond_codes is true, this
1639   // code sets temp2_reg to a nonzero value.
1640   if (set_cond_codes) {
1641     mov(temp2_reg, 1);
1642   }
1643 
1644   br(Assembler::NE, *L_failure);
1645 
1646   if (L_success != &L_fallthrough) {
1647     b(*L_success);
1648   }
1649 
1650   bind(L_fallthrough);
1651 }
1652 
1653 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1654                                                    Register super_klass,
1655                                                    Register temp_reg,
1656                                                    Register temp2_reg,
1657                                                    Label* L_success,
1658                                                    Label* L_failure,
1659                                                    bool set_cond_codes) {
1660   if (UseSecondarySupersTable) {
1661     check_klass_subtype_slow_path_table
1662       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1663        /*vtemp*/fnoreg,
1664        L_success, L_failure, set_cond_codes);
1665   } else {
1666     check_klass_subtype_slow_path_linear
1667       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1668   }
1669 }
1670 
1671 
1672 // Ensure that the inline code and the stub are using the same registers.
1673 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1674 do {                                                               \
1675   assert(r_super_klass  == r0                                   && \
1676          r_array_base   == r1                                   && \
1677          r_array_length == r2                                   && \
1678          (r_array_index == r3        || r_array_index == noreg) && \
1679          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1680          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1681          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1682 } while(0)
1683 
1684 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1685                                                          Register r_super_klass,
1686                                                          Register temp1,
1687                                                          Register temp2,
1688                                                          Register temp3,
1689                                                          FloatRegister vtemp,
1690                                                          Register result,
1691                                                          u1 super_klass_slot,
1692                                                          bool stub_is_near) {
1693   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1694 
1695   Label L_fallthrough;
1696 
1697   BLOCK_COMMENT("lookup_secondary_supers_table {");
1698 
1699   const Register
1700     r_array_base   = temp1, // r1
1701     r_array_length = temp2, // r2
1702     r_array_index  = temp3, // r3
1703     r_bitmap       = rscratch2;
1704 
1705   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1706 
1707   u1 bit = super_klass_slot;
1708 
1709   // Make sure that result is nonzero if the TBZ below misses.
1710   mov(result, 1);
1711 
1712   // We're going to need the bitmap in a vector reg and in a core reg,
1713   // so load both now.
1714   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1715   if (bit != 0) {
1716     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1717   }
1718   // First check the bitmap to see if super_klass might be present. If
1719   // the bit is zero, we are certain that super_klass is not one of
1720   // the secondary supers.
1721   tbz(r_bitmap, bit, L_fallthrough);
1722 
1723   // Get the first array index that can contain super_klass into r_array_index.
1724   if (bit != 0) {
1725     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1726     cnt(vtemp, T8B, vtemp);
1727     addv(vtemp, T8B, vtemp);
1728     fmovd(r_array_index, vtemp);
1729   } else {
1730     mov(r_array_index, (u1)1);
1731   }
1732   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1733 
1734   // We will consult the secondary-super array.
1735   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1736 
1737   // The value i in r_array_index is >= 1, so even though r_array_base
1738   // points to the length, we don't need to adjust it to point to the
1739   // data.
1740   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1741   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1742 
1743   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1744   eor(result, result, r_super_klass);
1745   cbz(result, L_fallthrough); // Found a match
1746 
1747   // Is there another entry to check? Consult the bitmap.
1748   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1749 
1750   // Linear probe.
1751   if (bit != 0) {
1752     ror(r_bitmap, r_bitmap, bit);
1753   }
1754 
1755   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1756   // The next slot to be inspected, by the stub we're about to call,
1757   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1758   // have been checked.
1759   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1760   if (stub_is_near) {
1761     bl(stub);
1762   } else {
1763     address call = trampoline_call(stub);
1764     if (call == nullptr) {
1765       return false; // trampoline allocation failed
1766     }
1767   }
1768 
1769   BLOCK_COMMENT("} lookup_secondary_supers_table");
1770 
1771   bind(L_fallthrough);
1772 
1773   if (VerifySecondarySupers) {
1774     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1775                                   temp1, temp2, result);      // r1, r2, r5
1776   }
1777   return true;
1778 }
1779 
1780 // At runtime, return 0 in result if r_super_klass is a superclass of
1781 // r_sub_klass, otherwise return nonzero. Use this version of
1782 // lookup_secondary_supers_table() if you don't know ahead of time
1783 // which superclass will be searched for. Used by interpreter and
1784 // runtime stubs. It is larger and has somewhat greater latency than
1785 // the version above, which takes a constant super_klass_slot.
1786 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1787                                                        Register r_super_klass,
1788                                                        Register temp1,
1789                                                        Register temp2,
1790                                                        Register temp3,
1791                                                        FloatRegister vtemp,
1792                                                        Register result,
1793                                                        Label *L_success) {
1794   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1795 
1796   Label L_fallthrough;
1797 
1798   BLOCK_COMMENT("lookup_secondary_supers_table {");
1799 
1800   const Register
1801     r_array_index = temp3,
1802     slot          = rscratch1,
1803     r_bitmap      = rscratch2;
1804 
1805   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1806 
1807   // Make sure that result is nonzero if the test below misses.
1808   mov(result, 1);
1809 
1810   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1811 
1812   // First check the bitmap to see if super_klass might be present. If
1813   // the bit is zero, we are certain that super_klass is not one of
1814   // the secondary supers.
1815 
1816   // This next instruction is equivalent to:
1817   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1818   // sub(temp2, tmp_reg, slot);
1819   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1820   lslv(temp2, r_bitmap, temp2);
1821   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1822 
1823   bool must_save_v0 = (vtemp == fnoreg);
1824   if (must_save_v0) {
1825     // temp1 and result are free, so use them to preserve vtemp
1826     vtemp = v0;
1827     mov(temp1,  vtemp, D, 0);
1828     mov(result, vtemp, D, 1);
1829   }
1830 
1831   // Get the first array index that can contain super_klass into r_array_index.
1832   mov(vtemp, D, 0, temp2);
1833   cnt(vtemp, T8B, vtemp);
1834   addv(vtemp, T8B, vtemp);
1835   mov(r_array_index, vtemp, D, 0);
1836 
1837   if (must_save_v0) {
1838     mov(vtemp, D, 0, temp1 );
1839     mov(vtemp, D, 1, result);
1840   }
1841 
1842   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1843 
1844   const Register
1845     r_array_base   = temp1,
1846     r_array_length = temp2;
1847 
1848   // The value i in r_array_index is >= 1, so even though r_array_base
1849   // points to the length, we don't need to adjust it to point to the
1850   // data.
1851   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1852   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1853 
1854   // We will consult the secondary-super array.
1855   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1856 
1857   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1858   eor(result, result, r_super_klass);
1859   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1860 
1861   // Is there another entry to check? Consult the bitmap.
1862   rorv(r_bitmap, r_bitmap, slot);
1863   // rol(r_bitmap, r_bitmap, 1);
1864   tbz(r_bitmap, 1, L_fallthrough);
1865 
1866   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1867   // The next slot to be inspected, by the logic we're about to call,
1868   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1869   // have been checked.
1870   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1871                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1872 
1873   BLOCK_COMMENT("} lookup_secondary_supers_table");
1874 
1875   bind(L_fallthrough);
1876 
1877   if (VerifySecondarySupers) {
1878     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1879                                   temp1, temp2, result);      // r1, r2, r5
1880   }
1881 
1882   if (L_success) {
1883     cbz(result, *L_success);
1884   }
1885 }
1886 
1887 // Called by code generated by check_klass_subtype_slow_path
1888 // above. This is called when there is a collision in the hashed
1889 // lookup in the secondary supers array.
1890 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1891                                                              Register r_array_base,
1892                                                              Register r_array_index,
1893                                                              Register r_bitmap,
1894                                                              Register temp1,
1895                                                              Register result,
1896                                                              bool is_stub) {
1897   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1898 
1899   const Register
1900     r_array_length = temp1,
1901     r_sub_klass    = noreg; // unused
1902 
1903   if (is_stub) {
1904     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1905   }
1906 
1907   Label L_fallthrough, L_huge;
1908 
1909   // Load the array length.
1910   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1911   // And adjust the array base to point to the data.
1912   // NB! Effectively increments current slot index by 1.
1913   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1914   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1915 
1916   // The bitmap is full to bursting.
1917   // Implicit invariant: BITMAP_FULL implies (length > 0)
1918   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1919   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1920   br(GT, L_huge);
1921 
1922   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1923   // current slot (at secondary_supers[r_array_index]) has not yet
1924   // been inspected, and r_array_index may be out of bounds if we
1925   // wrapped around the end of the array.
1926 
1927   { // This is conventional linear probing, but instead of terminating
1928     // when a null entry is found in the table, we maintain a bitmap
1929     // in which a 0 indicates missing entries.
1930     // As long as the bitmap is not completely full,
1931     // array_length == popcount(bitmap). The array_length check above
1932     // guarantees there are 0s in the bitmap, so the loop eventually
1933     // terminates.
1934     Label L_loop;
1935     bind(L_loop);
1936 
1937     // Check for wraparound.
1938     cmp(r_array_index, r_array_length);
1939     csel(r_array_index, zr, r_array_index, GE);
1940 
1941     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1942     eor(result, rscratch1, r_super_klass);
1943     cbz(result, L_fallthrough);
1944 
1945     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1946 
1947     ror(r_bitmap, r_bitmap, 1);
1948     add(r_array_index, r_array_index, 1);
1949     b(L_loop);
1950   }
1951 
1952   { // Degenerate case: more than 64 secondary supers.
1953     // FIXME: We could do something smarter here, maybe a vectorized
1954     // comparison or a binary search, but is that worth any added
1955     // complexity?
1956     bind(L_huge);
1957     cmp(sp, zr); // Clear Z flag; SP is never zero
1958     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1959     cset(result, NE); // result == 0 iff we got a match.
1960   }
1961 
1962   bind(L_fallthrough);
1963 }
1964 
1965 // Make sure that the hashed lookup and a linear scan agree.
1966 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1967                                                    Register r_super_klass,
1968                                                    Register temp1,
1969                                                    Register temp2,
1970                                                    Register result) {
1971   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1972 
1973   const Register
1974     r_array_base   = temp1,
1975     r_array_length = temp2,
1976     r_array_index  = noreg, // unused
1977     r_bitmap       = noreg; // unused
1978 
1979   BLOCK_COMMENT("verify_secondary_supers_table {");
1980 
1981   // We will consult the secondary-super array.
1982   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1983 
1984   // Load the array length.
1985   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1986   // And adjust the array base to point to the data.
1987   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1988 
1989   cmp(sp, zr); // Clear Z flag; SP is never zero
1990   // Scan R2 words at [R5] for an occurrence of R0.
1991   // Set NZ/Z based on last compare.
1992   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1993   // rscratch1 == 0 iff we got a match.
1994   cset(rscratch1, NE);
1995 
1996   Label passed;
1997   cmp(result, zr);
1998   cset(result, NE); // normalize result to 0/1 for comparison
1999 
2000   cmp(rscratch1, result);
2001   br(EQ, passed);
2002   {
2003     mov(r0, r_super_klass);         // r0 <- r0
2004     mov(r1, r_sub_klass);           // r1 <- r4
2005     mov(r2, /*expected*/rscratch1); // r2 <- r8
2006     mov(r3, result);                // r3 <- r5
2007     mov(r4, (address)("mismatch")); // r4 <- const
2008     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
2009     should_not_reach_here();
2010   }
2011   bind(passed);
2012 
2013   BLOCK_COMMENT("} verify_secondary_supers_table");
2014 }
2015 
2016 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
2017   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
2018   assert_different_registers(klass, rthread, scratch);
2019 
2020   Label L_fallthrough, L_tmp;
2021   if (L_fast_path == nullptr) {
2022     L_fast_path = &L_fallthrough;
2023   } else if (L_slow_path == nullptr) {
2024     L_slow_path = &L_fallthrough;
2025   }
2026   // Fast path check: class is fully initialized
2027   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2028   ldarb(scratch, scratch);
2029   cmp(scratch, InstanceKlass::fully_initialized);
2030   br(Assembler::EQ, *L_fast_path);
2031 
2032   // Fast path check: current thread is initializer thread
2033   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2034   cmp(rthread, scratch);
2035 
2036   if (L_slow_path == &L_fallthrough) {
2037     br(Assembler::EQ, *L_fast_path);
2038     bind(*L_slow_path);
2039   } else if (L_fast_path == &L_fallthrough) {
2040     br(Assembler::NE, *L_slow_path);
2041     bind(*L_fast_path);
2042   } else {
2043     Unimplemented();
2044   }
2045 }
2046 
2047 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2048   if (!VerifyOops) return;
2049 
2050   // Pass register number to verify_oop_subroutine
2051   const char* b = nullptr;
2052   {
2053     ResourceMark rm;
2054     stringStream ss;
2055     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2056     b = code_string(ss.as_string());
2057   }
2058   BLOCK_COMMENT("verify_oop {");
2059 
2060   strip_return_address(); // This might happen within a stack frame.
2061   protect_return_address();
2062   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2063   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2064 
2065   mov(r0, reg);
2066   movptr(rscratch1, (uintptr_t)(address)b);
2067 
2068   // call indirectly to solve generation ordering problem
2069   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2070   ldr(rscratch2, Address(rscratch2));
2071   blr(rscratch2);
2072 
2073   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2074   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2075   authenticate_return_address();
2076 
2077   BLOCK_COMMENT("} verify_oop");
2078 }
2079 
2080 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2081   if (!VerifyOops) return;
2082 
2083   const char* b = nullptr;
2084   {
2085     ResourceMark rm;
2086     stringStream ss;
2087     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2088     b = code_string(ss.as_string());
2089   }
2090   BLOCK_COMMENT("verify_oop_addr {");
2091 
2092   strip_return_address(); // This might happen within a stack frame.
2093   protect_return_address();
2094   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2095   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2096 
2097   // addr may contain sp so we will have to adjust it based on the
2098   // pushes that we just did.
2099   if (addr.uses(sp)) {
2100     lea(r0, addr);
2101     ldr(r0, Address(r0, 4 * wordSize));
2102   } else {
2103     ldr(r0, addr);
2104   }
2105   movptr(rscratch1, (uintptr_t)(address)b);
2106 
2107   // call indirectly to solve generation ordering problem
2108   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2109   ldr(rscratch2, Address(rscratch2));
2110   blr(rscratch2);
2111 
2112   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2113   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2114   authenticate_return_address();
2115 
2116   BLOCK_COMMENT("} verify_oop_addr");
2117 }
2118 
2119 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2120                                          int extra_slot_offset) {
2121   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2122   int stackElementSize = Interpreter::stackElementSize;
2123   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2124 #ifdef ASSERT
2125   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2126   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2127 #endif
2128   if (arg_slot.is_constant()) {
2129     return Address(esp, arg_slot.as_constant() * stackElementSize
2130                    + offset);
2131   } else {
2132     add(rscratch1, esp, arg_slot.as_register(),
2133         ext::uxtx, exact_log2(stackElementSize));
2134     return Address(rscratch1, offset);
2135   }
2136 }
2137 
2138 void MacroAssembler::call_VM_leaf_base(address entry_point,
2139                                        int number_of_arguments,
2140                                        Label *retaddr) {
2141   Label E, L;
2142 
2143   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2144 
2145   mov(rscratch1, RuntimeAddress(entry_point));
2146   blr(rscratch1);
2147   if (retaddr)
2148     bind(*retaddr);
2149 
2150   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2151 }
2152 
2153 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2154   call_VM_leaf_base(entry_point, number_of_arguments);
2155 }
2156 
2157 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2158   pass_arg0(this, arg_0);
2159   call_VM_leaf_base(entry_point, 1);
2160 }
2161 
2162 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2163   assert_different_registers(arg_1, c_rarg0);
2164   pass_arg0(this, arg_0);
2165   pass_arg1(this, arg_1);
2166   call_VM_leaf_base(entry_point, 2);
2167 }
2168 
2169 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2170                                   Register arg_1, Register arg_2) {
2171   assert_different_registers(arg_1, c_rarg0);
2172   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2173   pass_arg0(this, arg_0);
2174   pass_arg1(this, arg_1);
2175   pass_arg2(this, arg_2);
2176   call_VM_leaf_base(entry_point, 3);
2177 }
2178 
2179 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2180   pass_arg0(this, arg_0);
2181   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2182 }
2183 
2184 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2185 
2186   assert_different_registers(arg_0, c_rarg1);
2187   pass_arg1(this, arg_1);
2188   pass_arg0(this, arg_0);
2189   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2190 }
2191 
2192 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2193   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2194   assert_different_registers(arg_1, c_rarg2);
2195   pass_arg2(this, arg_2);
2196   pass_arg1(this, arg_1);
2197   pass_arg0(this, arg_0);
2198   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2199 }
2200 
2201 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2202   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2203   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2204   assert_different_registers(arg_2, c_rarg3);
2205   pass_arg3(this, arg_3);
2206   pass_arg2(this, arg_2);
2207   pass_arg1(this, arg_1);
2208   pass_arg0(this, arg_0);
2209   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2210 }
2211 
2212 void MacroAssembler::null_check(Register reg, int offset) {
2213   if (needs_explicit_null_check(offset)) {
2214     // provoke OS null exception if reg is null by
2215     // accessing M[reg] w/o changing any registers
2216     // NOTE: this is plenty to provoke a segv
2217     ldr(zr, Address(reg));
2218   } else {
2219     // nothing to do, (later) access of M[reg + offset]
2220     // will provoke OS null exception if reg is null
2221   }
2222 }
2223 
2224 // MacroAssembler protected routines needed to implement
2225 // public methods
2226 
2227 void MacroAssembler::mov(Register r, Address dest) {
2228   code_section()->relocate(pc(), dest.rspec());
2229   uint64_t imm64 = (uint64_t)dest.target();
2230   movptr(r, imm64);
2231 }
2232 
2233 // Move a constant pointer into r.  In AArch64 mode the virtual
2234 // address space is 48 bits in size, so we only need three
2235 // instructions to create a patchable instruction sequence that can
2236 // reach anywhere.
2237 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2238 #ifndef PRODUCT
2239   {
2240     char buffer[64];
2241     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2242     block_comment(buffer);
2243   }
2244 #endif
2245   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2246   movz(r, imm64 & 0xffff);
2247   imm64 >>= 16;
2248   movk(r, imm64 & 0xffff, 16);
2249   imm64 >>= 16;
2250   movk(r, imm64 & 0xffff, 32);
2251 }
2252 
2253 // Macro to mov replicated immediate to vector register.
2254 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2255 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2256 // Vd will get the following values for different arrangements in T
2257 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2258 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2259 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2260 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2261 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2262 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2263 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2264 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2265 // Clobbers rscratch1
2266 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2267   assert(T != T1Q, "unsupported");
2268   if (T == T1D || T == T2D) {
2269     int imm = operand_valid_for_movi_immediate(imm64, T);
2270     if (-1 != imm) {
2271       movi(Vd, T, imm);
2272     } else {
2273       mov(rscratch1, imm64);
2274       dup(Vd, T, rscratch1);
2275     }
2276     return;
2277   }
2278 
2279 #ifdef ASSERT
2280   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2281   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2282   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2283 #endif
2284   int shift = operand_valid_for_movi_immediate(imm64, T);
2285   uint32_t imm32 = imm64 & 0xffffffffULL;
2286   if (shift >= 0) {
2287     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2288   } else {
2289     movw(rscratch1, imm32);
2290     dup(Vd, T, rscratch1);
2291   }
2292 }
2293 
2294 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2295 {
2296 #ifndef PRODUCT
2297   {
2298     char buffer[64];
2299     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2300     block_comment(buffer);
2301   }
2302 #endif
2303   if (operand_valid_for_logical_immediate(false, imm64)) {
2304     orr(dst, zr, imm64);
2305   } else {
2306     // we can use a combination of MOVZ or MOVN with
2307     // MOVK to build up the constant
2308     uint64_t imm_h[4];
2309     int zero_count = 0;
2310     int neg_count = 0;
2311     int i;
2312     for (i = 0; i < 4; i++) {
2313       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2314       if (imm_h[i] == 0) {
2315         zero_count++;
2316       } else if (imm_h[i] == 0xffffL) {
2317         neg_count++;
2318       }
2319     }
2320     if (zero_count == 4) {
2321       // one MOVZ will do
2322       movz(dst, 0);
2323     } else if (neg_count == 4) {
2324       // one MOVN will do
2325       movn(dst, 0);
2326     } else if (zero_count == 3) {
2327       for (i = 0; i < 4; i++) {
2328         if (imm_h[i] != 0L) {
2329           movz(dst, (uint32_t)imm_h[i], (i << 4));
2330           break;
2331         }
2332       }
2333     } else if (neg_count == 3) {
2334       // one MOVN will do
2335       for (int i = 0; i < 4; i++) {
2336         if (imm_h[i] != 0xffffL) {
2337           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2338           break;
2339         }
2340       }
2341     } else if (zero_count == 2) {
2342       // one MOVZ and one MOVK will do
2343       for (i = 0; i < 3; i++) {
2344         if (imm_h[i] != 0L) {
2345           movz(dst, (uint32_t)imm_h[i], (i << 4));
2346           i++;
2347           break;
2348         }
2349       }
2350       for (;i < 4; i++) {
2351         if (imm_h[i] != 0L) {
2352           movk(dst, (uint32_t)imm_h[i], (i << 4));
2353         }
2354       }
2355     } else if (neg_count == 2) {
2356       // one MOVN and one MOVK will do
2357       for (i = 0; i < 4; i++) {
2358         if (imm_h[i] != 0xffffL) {
2359           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2360           i++;
2361           break;
2362         }
2363       }
2364       for (;i < 4; i++) {
2365         if (imm_h[i] != 0xffffL) {
2366           movk(dst, (uint32_t)imm_h[i], (i << 4));
2367         }
2368       }
2369     } else if (zero_count == 1) {
2370       // one MOVZ and two MOVKs will do
2371       for (i = 0; i < 4; i++) {
2372         if (imm_h[i] != 0L) {
2373           movz(dst, (uint32_t)imm_h[i], (i << 4));
2374           i++;
2375           break;
2376         }
2377       }
2378       for (;i < 4; i++) {
2379         if (imm_h[i] != 0x0L) {
2380           movk(dst, (uint32_t)imm_h[i], (i << 4));
2381         }
2382       }
2383     } else if (neg_count == 1) {
2384       // one MOVN and two MOVKs will do
2385       for (i = 0; i < 4; i++) {
2386         if (imm_h[i] != 0xffffL) {
2387           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2388           i++;
2389           break;
2390         }
2391       }
2392       for (;i < 4; i++) {
2393         if (imm_h[i] != 0xffffL) {
2394           movk(dst, (uint32_t)imm_h[i], (i << 4));
2395         }
2396       }
2397     } else {
2398       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2399       movz(dst, (uint32_t)imm_h[0], 0);
2400       for (i = 1; i < 4; i++) {
2401         movk(dst, (uint32_t)imm_h[i], (i << 4));
2402       }
2403     }
2404   }
2405 }
2406 
2407 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2408 {
2409 #ifndef PRODUCT
2410     {
2411       char buffer[64];
2412       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2413       block_comment(buffer);
2414     }
2415 #endif
2416   if (operand_valid_for_logical_immediate(true, imm32)) {
2417     orrw(dst, zr, imm32);
2418   } else {
2419     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2420     // constant
2421     uint32_t imm_h[2];
2422     imm_h[0] = imm32 & 0xffff;
2423     imm_h[1] = ((imm32 >> 16) & 0xffff);
2424     if (imm_h[0] == 0) {
2425       movzw(dst, imm_h[1], 16);
2426     } else if (imm_h[0] == 0xffff) {
2427       movnw(dst, imm_h[1] ^ 0xffff, 16);
2428     } else if (imm_h[1] == 0) {
2429       movzw(dst, imm_h[0], 0);
2430     } else if (imm_h[1] == 0xffff) {
2431       movnw(dst, imm_h[0] ^ 0xffff, 0);
2432     } else {
2433       // use a MOVZ and MOVK (makes it easier to debug)
2434       movzw(dst, imm_h[0], 0);
2435       movkw(dst, imm_h[1], 16);
2436     }
2437   }
2438 }
2439 
2440 // Form an address from base + offset in Rd.  Rd may or may
2441 // not actually be used: you must use the Address that is returned.
2442 // It is up to you to ensure that the shift provided matches the size
2443 // of your data.
2444 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2445   if (Address::offset_ok_for_immed(byte_offset, shift))
2446     // It fits; no need for any heroics
2447     return Address(base, byte_offset);
2448 
2449   // Don't do anything clever with negative or misaligned offsets
2450   unsigned mask = (1 << shift) - 1;
2451   if (byte_offset < 0 || byte_offset & mask) {
2452     mov(Rd, byte_offset);
2453     add(Rd, base, Rd);
2454     return Address(Rd);
2455   }
2456 
2457   // See if we can do this with two 12-bit offsets
2458   {
2459     uint64_t word_offset = byte_offset >> shift;
2460     uint64_t masked_offset = word_offset & 0xfff000;
2461     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2462         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2463       add(Rd, base, masked_offset << shift);
2464       word_offset -= masked_offset;
2465       return Address(Rd, word_offset << shift);
2466     }
2467   }
2468 
2469   // Do it the hard way
2470   mov(Rd, byte_offset);
2471   add(Rd, base, Rd);
2472   return Address(Rd);
2473 }
2474 
2475 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2476                                     bool want_remainder, Register scratch)
2477 {
2478   // Full implementation of Java idiv and irem.  The function
2479   // returns the (pc) offset of the div instruction - may be needed
2480   // for implicit exceptions.
2481   //
2482   // constraint : ra/rb =/= scratch
2483   //         normal case
2484   //
2485   // input : ra: dividend
2486   //         rb: divisor
2487   //
2488   // result: either
2489   //         quotient  (= ra idiv rb)
2490   //         remainder (= ra irem rb)
2491 
2492   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2493 
2494   int idivl_offset = offset();
2495   if (! want_remainder) {
2496     sdivw(result, ra, rb);
2497   } else {
2498     sdivw(scratch, ra, rb);
2499     Assembler::msubw(result, scratch, rb, ra);
2500   }
2501 
2502   return idivl_offset;
2503 }
2504 
2505 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2506                                     bool want_remainder, Register scratch)
2507 {
2508   // Full implementation of Java ldiv and lrem.  The function
2509   // returns the (pc) offset of the div instruction - may be needed
2510   // for implicit exceptions.
2511   //
2512   // constraint : ra/rb =/= scratch
2513   //         normal case
2514   //
2515   // input : ra: dividend
2516   //         rb: divisor
2517   //
2518   // result: either
2519   //         quotient  (= ra idiv rb)
2520   //         remainder (= ra irem rb)
2521 
2522   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2523 
2524   int idivq_offset = offset();
2525   if (! want_remainder) {
2526     sdiv(result, ra, rb);
2527   } else {
2528     sdiv(scratch, ra, rb);
2529     Assembler::msub(result, scratch, rb, ra);
2530   }
2531 
2532   return idivq_offset;
2533 }
2534 
2535 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2536   address prev = pc() - NativeMembar::instruction_size;
2537   address last = code()->last_insn();
2538   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2539     NativeMembar *bar = NativeMembar_at(prev);
2540     if (AlwaysMergeDMB) {
2541       bar->set_kind(bar->get_kind() | order_constraint);
2542       BLOCK_COMMENT("merged membar(always)");
2543       return;
2544     }
2545     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2546     // doing so would introduce a StoreLoad which the caller did not
2547     // intend
2548     if (bar->get_kind() == order_constraint
2549         || bar->get_kind() == AnyAny
2550         || order_constraint == AnyAny) {
2551       // We are merging two memory barrier instructions.  On AArch64 we
2552       // can do this simply by ORing them together.
2553       bar->set_kind(bar->get_kind() | order_constraint);
2554       BLOCK_COMMENT("merged membar");
2555       return;
2556     } else {
2557       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2558       // We need check the last 2 instructions
2559       address prev2 = prev - NativeMembar::instruction_size;
2560       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2561         NativeMembar *bar2 = NativeMembar_at(prev2);
2562         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2563         BLOCK_COMMENT("merged membar(elided)");
2564         return;
2565       }
2566     }
2567   }
2568   code()->set_last_insn(pc());
2569   dmb(Assembler::barrier(order_constraint));
2570 }
2571 
2572 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2573   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2574     merge_ldst(rt, adr, size_in_bytes, is_store);
2575     code()->clear_last_insn();
2576     return true;
2577   } else {
2578     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2579     const uint64_t mask = size_in_bytes - 1;
2580     if (adr.getMode() == Address::base_plus_offset &&
2581         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2582       code()->set_last_insn(pc());
2583     }
2584     return false;
2585   }
2586 }
2587 
2588 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2589   // We always try to merge two adjacent loads into one ldp.
2590   if (!try_merge_ldst(Rx, adr, 8, false)) {
2591     Assembler::ldr(Rx, adr);
2592   }
2593 }
2594 
2595 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2596   // We always try to merge two adjacent loads into one ldp.
2597   if (!try_merge_ldst(Rw, adr, 4, false)) {
2598     Assembler::ldrw(Rw, adr);
2599   }
2600 }
2601 
2602 void MacroAssembler::str(Register Rx, const Address &adr) {
2603   // We always try to merge two adjacent stores into one stp.
2604   if (!try_merge_ldst(Rx, adr, 8, true)) {
2605     Assembler::str(Rx, adr);
2606   }
2607 }
2608 
2609 void MacroAssembler::strw(Register Rw, const Address &adr) {
2610   // We always try to merge two adjacent stores into one stp.
2611   if (!try_merge_ldst(Rw, adr, 4, true)) {
2612     Assembler::strw(Rw, adr);
2613   }
2614 }
2615 
2616 // MacroAssembler routines found actually to be needed
2617 
2618 void MacroAssembler::push(Register src)
2619 {
2620   str(src, Address(pre(esp, -1 * wordSize)));
2621 }
2622 
2623 void MacroAssembler::pop(Register dst)
2624 {
2625   ldr(dst, Address(post(esp, 1 * wordSize)));
2626 }
2627 
2628 // Note: load_unsigned_short used to be called load_unsigned_word.
2629 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2630   int off = offset();
2631   ldrh(dst, src);
2632   return off;
2633 }
2634 
2635 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2636   int off = offset();
2637   ldrb(dst, src);
2638   return off;
2639 }
2640 
2641 int MacroAssembler::load_signed_short(Register dst, Address src) {
2642   int off = offset();
2643   ldrsh(dst, src);
2644   return off;
2645 }
2646 
2647 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2648   int off = offset();
2649   ldrsb(dst, src);
2650   return off;
2651 }
2652 
2653 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2654   int off = offset();
2655   ldrshw(dst, src);
2656   return off;
2657 }
2658 
2659 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2660   int off = offset();
2661   ldrsbw(dst, src);
2662   return off;
2663 }
2664 
2665 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2666   switch (size_in_bytes) {
2667   case  8:  ldr(dst, src); break;
2668   case  4:  ldrw(dst, src); break;
2669   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2670   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2671   default:  ShouldNotReachHere();
2672   }
2673 }
2674 
2675 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2676   switch (size_in_bytes) {
2677   case  8:  str(src, dst); break;
2678   case  4:  strw(src, dst); break;
2679   case  2:  strh(src, dst); break;
2680   case  1:  strb(src, dst); break;
2681   default:  ShouldNotReachHere();
2682   }
2683 }
2684 
2685 void MacroAssembler::decrementw(Register reg, int value)
2686 {
2687   if (value < 0)  { incrementw(reg, -value);      return; }
2688   if (value == 0) {                               return; }
2689   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2690   /* else */ {
2691     guarantee(reg != rscratch2, "invalid dst for register decrement");
2692     movw(rscratch2, (unsigned)value);
2693     subw(reg, reg, rscratch2);
2694   }
2695 }
2696 
2697 void MacroAssembler::decrement(Register reg, int value)
2698 {
2699   if (value < 0)  { increment(reg, -value);      return; }
2700   if (value == 0) {                              return; }
2701   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2702   /* else */ {
2703     assert(reg != rscratch2, "invalid dst for register decrement");
2704     mov(rscratch2, (uint64_t)value);
2705     sub(reg, reg, rscratch2);
2706   }
2707 }
2708 
2709 void MacroAssembler::decrementw(Address dst, int value)
2710 {
2711   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2712   if (dst.getMode() == Address::literal) {
2713     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2714     lea(rscratch2, dst);
2715     dst = Address(rscratch2);
2716   }
2717   ldrw(rscratch1, dst);
2718   decrementw(rscratch1, value);
2719   strw(rscratch1, dst);
2720 }
2721 
2722 void MacroAssembler::decrement(Address dst, int value)
2723 {
2724   assert(!dst.uses(rscratch1), "invalid address for decrement");
2725   if (dst.getMode() == Address::literal) {
2726     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2727     lea(rscratch2, dst);
2728     dst = Address(rscratch2);
2729   }
2730   ldr(rscratch1, dst);
2731   decrement(rscratch1, value);
2732   str(rscratch1, dst);
2733 }
2734 
2735 void MacroAssembler::incrementw(Register reg, int value)
2736 {
2737   if (value < 0)  { decrementw(reg, -value);      return; }
2738   if (value == 0) {                               return; }
2739   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2740   /* else */ {
2741     assert(reg != rscratch2, "invalid dst for register increment");
2742     movw(rscratch2, (unsigned)value);
2743     addw(reg, reg, rscratch2);
2744   }
2745 }
2746 
2747 void MacroAssembler::increment(Register reg, int value)
2748 {
2749   if (value < 0)  { decrement(reg, -value);      return; }
2750   if (value == 0) {                              return; }
2751   if (value < (1 << 12)) { add(reg, reg, value); return; }
2752   /* else */ {
2753     assert(reg != rscratch2, "invalid dst for register increment");
2754     movw(rscratch2, (unsigned)value);
2755     add(reg, reg, rscratch2);
2756   }
2757 }
2758 
2759 void MacroAssembler::incrementw(Address dst, int value)
2760 {
2761   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2762   if (dst.getMode() == Address::literal) {
2763     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2764     lea(rscratch2, dst);
2765     dst = Address(rscratch2);
2766   }
2767   ldrw(rscratch1, dst);
2768   incrementw(rscratch1, value);
2769   strw(rscratch1, dst);
2770 }
2771 
2772 void MacroAssembler::increment(Address dst, int value)
2773 {
2774   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2775   if (dst.getMode() == Address::literal) {
2776     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2777     lea(rscratch2, dst);
2778     dst = Address(rscratch2);
2779   }
2780   ldr(rscratch1, dst);
2781   increment(rscratch1, value);
2782   str(rscratch1, dst);
2783 }
2784 
2785 // Push lots of registers in the bit set supplied.  Don't push sp.
2786 // Return the number of words pushed
2787 int MacroAssembler::push(unsigned int bitset, Register stack) {
2788   int words_pushed = 0;
2789 
2790   // Scan bitset to accumulate register pairs
2791   unsigned char regs[32];
2792   int count = 0;
2793   for (int reg = 0; reg <= 30; reg++) {
2794     if (1 & bitset)
2795       regs[count++] = reg;
2796     bitset >>= 1;
2797   }
2798   regs[count++] = zr->raw_encoding();
2799   count &= ~1;  // Only push an even number of regs
2800 
2801   if (count) {
2802     stp(as_Register(regs[0]), as_Register(regs[1]),
2803        Address(pre(stack, -count * wordSize)));
2804     words_pushed += 2;
2805   }
2806   for (int i = 2; i < count; i += 2) {
2807     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2808        Address(stack, i * wordSize));
2809     words_pushed += 2;
2810   }
2811 
2812   assert(words_pushed == count, "oops, pushed != count");
2813 
2814   return count;
2815 }
2816 
2817 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2818   int words_pushed = 0;
2819 
2820   // Scan bitset to accumulate register pairs
2821   unsigned char regs[32];
2822   int count = 0;
2823   for (int reg = 0; reg <= 30; reg++) {
2824     if (1 & bitset)
2825       regs[count++] = reg;
2826     bitset >>= 1;
2827   }
2828   regs[count++] = zr->raw_encoding();
2829   count &= ~1;
2830 
2831   for (int i = 2; i < count; i += 2) {
2832     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2833        Address(stack, i * wordSize));
2834     words_pushed += 2;
2835   }
2836   if (count) {
2837     ldp(as_Register(regs[0]), as_Register(regs[1]),
2838        Address(post(stack, count * wordSize)));
2839     words_pushed += 2;
2840   }
2841 
2842   assert(words_pushed == count, "oops, pushed != count");
2843 
2844   return count;
2845 }
2846 
2847 // Push lots of registers in the bit set supplied.  Don't push sp.
2848 // Return the number of dwords pushed
2849 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2850   int words_pushed = 0;
2851   bool use_sve = false;
2852   int sve_vector_size_in_bytes = 0;
2853 
2854 #ifdef COMPILER2
2855   use_sve = Matcher::supports_scalable_vector();
2856   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2857 #endif
2858 
2859   // Scan bitset to accumulate register pairs
2860   unsigned char regs[32];
2861   int count = 0;
2862   for (int reg = 0; reg <= 31; reg++) {
2863     if (1 & bitset)
2864       regs[count++] = reg;
2865     bitset >>= 1;
2866   }
2867 
2868   if (count == 0) {
2869     return 0;
2870   }
2871 
2872   if (mode == PushPopFull) {
2873     if (use_sve && sve_vector_size_in_bytes > 16) {
2874       mode = PushPopSVE;
2875     } else {
2876       mode = PushPopNeon;
2877     }
2878   }
2879 
2880 #ifndef PRODUCT
2881   {
2882     char buffer[48];
2883     if (mode == PushPopSVE) {
2884       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
2885     } else if (mode == PushPopNeon) {
2886       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
2887     } else {
2888       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
2889     }
2890     block_comment(buffer);
2891   }
2892 #endif
2893 
2894   if (mode == PushPopSVE) {
2895     sub(stack, stack, sve_vector_size_in_bytes * count);
2896     for (int i = 0; i < count; i++) {
2897       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
2898     }
2899     return count * sve_vector_size_in_bytes / 8;
2900   }
2901 
2902   if (mode == PushPopNeon) {
2903     if (count == 1) {
2904       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
2905       return 2;
2906     }
2907 
2908     bool odd = (count & 1) == 1;
2909     int push_slots = count + (odd ? 1 : 0);
2910 
2911     // Always pushing full 128 bit registers.
2912     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
2913     words_pushed += 2;
2914 
2915     for (int i = 2; i + 1 < count; i += 2) {
2916       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
2917       words_pushed += 2;
2918     }
2919 
2920     if (odd) {
2921       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
2922       words_pushed++;
2923     }
2924 
2925     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
2926     return count * 2;
2927   }
2928 
2929   if (mode == PushPopFp) {
2930     bool odd = (count & 1) == 1;
2931     int push_slots = count + (odd ? 1 : 0);
2932 
2933     if (count == 1) {
2934       // Stack pointer must be 16 bytes aligned
2935       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
2936       return 1;
2937     }
2938 
2939     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
2940     words_pushed += 2;
2941 
2942     for (int i = 2; i + 1 < count; i += 2) {
2943       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
2944       words_pushed += 2;
2945     }
2946 
2947     if (odd) {
2948       // Stack pointer must be 16 bytes aligned
2949       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
2950       words_pushed++;
2951     }
2952 
2953     assert(words_pushed == count, "oops, pushed != count");
2954 
2955     return count;
2956   }
2957 
2958   return 0;
2959 }
2960 
2961 // Return the number of dwords popped
2962 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2963   int words_pushed = 0;
2964   bool use_sve = false;
2965   int sve_vector_size_in_bytes = 0;
2966 
2967 #ifdef COMPILER2
2968   use_sve = Matcher::supports_scalable_vector();
2969   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2970 #endif
2971   // Scan bitset to accumulate register pairs
2972   unsigned char regs[32];
2973   int count = 0;
2974   for (int reg = 0; reg <= 31; reg++) {
2975     if (1 & bitset)
2976       regs[count++] = reg;
2977     bitset >>= 1;
2978   }
2979 
2980   if (count == 0) {
2981     return 0;
2982   }
2983 
2984   if (mode == PushPopFull) {
2985     if (use_sve && sve_vector_size_in_bytes > 16) {
2986       mode = PushPopSVE;
2987     } else {
2988       mode = PushPopNeon;
2989     }
2990   }
2991 
2992 #ifndef PRODUCT
2993   {
2994     char buffer[48];
2995     if (mode == PushPopSVE) {
2996       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
2997     } else if (mode == PushPopNeon) {
2998       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
2999     } else {
3000       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3001     }
3002     block_comment(buffer);
3003   }
3004 #endif
3005 
3006   if (mode == PushPopSVE) {
3007     for (int i = count - 1; i >= 0; i--) {
3008       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3009     }
3010     add(stack, stack, sve_vector_size_in_bytes * count);
3011     return count * sve_vector_size_in_bytes / 8;
3012   }
3013 
3014   if (mode == PushPopNeon) {
3015     if (count == 1) {
3016       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3017       return 2;
3018     }
3019 
3020     bool odd = (count & 1) == 1;
3021     int push_slots = count + (odd ? 1 : 0);
3022 
3023     if (odd) {
3024       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3025       words_pushed++;
3026     }
3027 
3028     for (int i = 2; i + 1 < count; i += 2) {
3029       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3030       words_pushed += 2;
3031     }
3032 
3033     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3034     words_pushed += 2;
3035 
3036     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3037 
3038     return count * 2;
3039   }
3040 
3041   if (mode == PushPopFp) {
3042     bool odd = (count & 1) == 1;
3043     int push_slots = count + (odd ? 1 : 0);
3044 
3045     if (count == 1) {
3046       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3047       return 1;
3048     }
3049 
3050     if (odd) {
3051       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3052       words_pushed++;
3053     }
3054 
3055     for (int i = 2; i + 1 < count; i += 2) {
3056       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3057       words_pushed += 2;
3058     }
3059 
3060     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3061     words_pushed += 2;
3062 
3063     assert(words_pushed == count, "oops, pushed != count");
3064 
3065     return count;
3066   }
3067 
3068   return 0;
3069 }
3070 
3071 // Return the number of dwords pushed
3072 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3073   bool use_sve = false;
3074   int sve_predicate_size_in_slots = 0;
3075 
3076 #ifdef COMPILER2
3077   use_sve = Matcher::supports_scalable_vector();
3078   if (use_sve) {
3079     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3080   }
3081 #endif
3082 
3083   if (!use_sve) {
3084     return 0;
3085   }
3086 
3087   unsigned char regs[PRegister::number_of_registers];
3088   int count = 0;
3089   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3090     if (1 & bitset)
3091       regs[count++] = reg;
3092     bitset >>= 1;
3093   }
3094 
3095   if (count == 0) {
3096     return 0;
3097   }
3098 
3099   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3100                                   VMRegImpl::stack_slot_size * count, 16);
3101   sub(stack, stack, total_push_bytes);
3102   for (int i = 0; i < count; i++) {
3103     sve_str(as_PRegister(regs[i]), Address(stack, i));
3104   }
3105   return total_push_bytes / 8;
3106 }
3107 
3108 // Return the number of dwords popped
3109 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3110   bool use_sve = false;
3111   int sve_predicate_size_in_slots = 0;
3112 
3113 #ifdef COMPILER2
3114   use_sve = Matcher::supports_scalable_vector();
3115   if (use_sve) {
3116     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3117   }
3118 #endif
3119 
3120   if (!use_sve) {
3121     return 0;
3122   }
3123 
3124   unsigned char regs[PRegister::number_of_registers];
3125   int count = 0;
3126   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3127     if (1 & bitset)
3128       regs[count++] = reg;
3129     bitset >>= 1;
3130   }
3131 
3132   if (count == 0) {
3133     return 0;
3134   }
3135 
3136   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3137                                  VMRegImpl::stack_slot_size * count, 16);
3138   for (int i = count - 1; i >= 0; i--) {
3139     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3140   }
3141   add(stack, stack, total_pop_bytes);
3142   return total_pop_bytes / 8;
3143 }
3144 
3145 #ifdef ASSERT
3146 void MacroAssembler::verify_heapbase(const char* msg) {
3147 #if 0
3148   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3149   assert (Universe::heap() != nullptr, "java heap should be initialized");
3150   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3151     // rheapbase is allocated as general register
3152     return;
3153   }
3154   if (CheckCompressedOops) {
3155     Label ok;
3156     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3157     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3158     br(Assembler::EQ, ok);
3159     stop(msg);
3160     bind(ok);
3161     pop(1 << rscratch1->encoding(), sp);
3162   }
3163 #endif
3164 }
3165 #endif
3166 
3167 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3168   assert_different_registers(value, tmp1, tmp2);
3169   Label done, tagged, weak_tagged;
3170 
3171   cbz(value, done);           // Use null as-is.
3172   tst(value, JNIHandles::tag_mask); // Test for tag.
3173   br(Assembler::NE, tagged);
3174 
3175   // Resolve local handle
3176   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3177   verify_oop(value);
3178   b(done);
3179 
3180   bind(tagged);
3181   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3182   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3183 
3184   // Resolve global handle
3185   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3186   verify_oop(value);
3187   b(done);
3188 
3189   bind(weak_tagged);
3190   // Resolve jweak.
3191   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3192                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3193   verify_oop(value);
3194 
3195   bind(done);
3196 }
3197 
3198 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3199   assert_different_registers(value, tmp1, tmp2);
3200   Label done;
3201 
3202   cbz(value, done);           // Use null as-is.
3203 
3204 #ifdef ASSERT
3205   {
3206     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3207     Label valid_global_tag;
3208     tbnz(value, 1, valid_global_tag); // Test for global tag
3209     stop("non global jobject using resolve_global_jobject");
3210     bind(valid_global_tag);
3211   }
3212 #endif
3213 
3214   // Resolve global handle
3215   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3216   verify_oop(value);
3217 
3218   bind(done);
3219 }
3220 
3221 void MacroAssembler::stop(const char* msg) {
3222   // Skip AOT caching C strings in scratch buffer.
3223   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3224   BLOCK_COMMENT(str);
3225   // load msg into r0 so we can access it from the signal handler
3226   // ExternalAddress enables saving and restoring via the code cache
3227   lea(c_rarg0, ExternalAddress((address) str));
3228   dcps1(0xdeae);
3229 }
3230 
3231 void MacroAssembler::unimplemented(const char* what) {
3232   const char* buf = nullptr;
3233   {
3234     ResourceMark rm;
3235     stringStream ss;
3236     ss.print("unimplemented: %s", what);
3237     buf = code_string(ss.as_string());
3238   }
3239   stop(buf);
3240 }
3241 
3242 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3243 #ifdef ASSERT
3244   Label OK;
3245   br(cc, OK);
3246   stop(msg);
3247   bind(OK);
3248 #endif
3249 }
3250 
3251 // If a constant does not fit in an immediate field, generate some
3252 // number of MOV instructions and then perform the operation.
3253 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3254                                            add_sub_imm_insn insn1,
3255                                            add_sub_reg_insn insn2,
3256                                            bool is32) {
3257   assert(Rd != zr, "Rd = zr and not setting flags?");
3258   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3259   if (fits) {
3260     (this->*insn1)(Rd, Rn, imm);
3261   } else {
3262     if (g_uabs(imm) < (1 << 24)) {
3263        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3264        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3265     } else {
3266        assert_different_registers(Rd, Rn);
3267        mov(Rd, imm);
3268        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3269     }
3270   }
3271 }
3272 
3273 // Separate vsn which sets the flags. Optimisations are more restricted
3274 // because we must set the flags correctly.
3275 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3276                                              add_sub_imm_insn insn1,
3277                                              add_sub_reg_insn insn2,
3278                                              bool is32) {
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     assert_different_registers(Rd, Rn);
3284     assert(Rd != zr, "overflow in immediate operand");
3285     mov(Rd, imm);
3286     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3287   }
3288 }
3289 
3290 
3291 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3292   if (increment.is_register()) {
3293     add(Rd, Rn, increment.as_register());
3294   } else {
3295     add(Rd, Rn, increment.as_constant());
3296   }
3297 }
3298 
3299 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3300   if (increment.is_register()) {
3301     addw(Rd, Rn, increment.as_register());
3302   } else {
3303     addw(Rd, Rn, increment.as_constant());
3304   }
3305 }
3306 
3307 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3308   if (decrement.is_register()) {
3309     sub(Rd, Rn, decrement.as_register());
3310   } else {
3311     sub(Rd, Rn, decrement.as_constant());
3312   }
3313 }
3314 
3315 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3316   if (decrement.is_register()) {
3317     subw(Rd, Rn, decrement.as_register());
3318   } else {
3319     subw(Rd, Rn, decrement.as_constant());
3320   }
3321 }
3322 
3323 void MacroAssembler::reinit_heapbase()
3324 {
3325   if (UseCompressedOops) {
3326     if (Universe::is_fully_initialized() && !AOTCodeCache::is_on_for_dump()) {
3327       mov(rheapbase, CompressedOops::base());
3328     } else {
3329       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3330       ldr(rheapbase, Address(rheapbase));
3331     }
3332   }
3333 }
3334 
3335 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3336 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3337 // Pass a register for the result, otherwise pass noreg.
3338 
3339 // Clobbers rscratch1
3340 void MacroAssembler::cmpxchg(Register addr, Register expected,
3341                              Register new_val,
3342                              enum operand_size size,
3343                              bool acquire, bool release,
3344                              bool weak,
3345                              Register result) {
3346   if (result == noreg)  result = rscratch1;
3347   BLOCK_COMMENT("cmpxchg {");
3348   if (UseLSE) {
3349     mov(result, expected);
3350     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3351     compare_eq(result, expected, size);
3352 #ifdef ASSERT
3353     // Poison rscratch1 which is written on !UseLSE branch
3354     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3355 #endif
3356   } else {
3357     Label retry_load, done;
3358     prfm(Address(addr), PSTL1STRM);
3359     bind(retry_load);
3360     load_exclusive(result, addr, size, acquire);
3361     compare_eq(result, expected, size);
3362     br(Assembler::NE, done);
3363     store_exclusive(rscratch1, new_val, addr, size, release);
3364     if (weak) {
3365       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3366     } else {
3367       cbnzw(rscratch1, retry_load);
3368     }
3369     bind(done);
3370   }
3371   BLOCK_COMMENT("} cmpxchg");
3372 }
3373 
3374 // A generic comparison. Only compares for equality, clobbers rscratch1.
3375 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3376   if (size == xword) {
3377     cmp(rm, rn);
3378   } else if (size == word) {
3379     cmpw(rm, rn);
3380   } else if (size == halfword) {
3381     eorw(rscratch1, rm, rn);
3382     ands(zr, rscratch1, 0xffff);
3383   } else if (size == byte) {
3384     eorw(rscratch1, rm, rn);
3385     ands(zr, rscratch1, 0xff);
3386   } else {
3387     ShouldNotReachHere();
3388   }
3389 }
3390 
3391 
3392 static bool different(Register a, RegisterOrConstant b, Register c) {
3393   if (b.is_constant())
3394     return a != c;
3395   else
3396     return a != b.as_register() && a != c && b.as_register() != c;
3397 }
3398 
3399 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3400 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3401   if (UseLSE) {                                                         \
3402     prev = prev->is_valid() ? prev : zr;                                \
3403     if (incr.is_register()) {                                           \
3404       AOP(sz, incr.as_register(), prev, addr);                          \
3405     } else {                                                            \
3406       mov(rscratch2, incr.as_constant());                               \
3407       AOP(sz, rscratch2, prev, addr);                                   \
3408     }                                                                   \
3409     return;                                                             \
3410   }                                                                     \
3411   Register result = rscratch2;                                          \
3412   if (prev->is_valid())                                                 \
3413     result = different(prev, incr, addr) ? prev : rscratch2;            \
3414                                                                         \
3415   Label retry_load;                                                     \
3416   prfm(Address(addr), PSTL1STRM);                                       \
3417   bind(retry_load);                                                     \
3418   LDXR(result, addr);                                                   \
3419   OP(rscratch1, result, incr);                                          \
3420   STXR(rscratch2, rscratch1, addr);                                     \
3421   cbnzw(rscratch2, retry_load);                                         \
3422   if (prev->is_valid() && prev != result) {                             \
3423     IOP(prev, rscratch1, incr);                                         \
3424   }                                                                     \
3425 }
3426 
3427 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3428 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3429 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3430 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3431 
3432 #undef ATOMIC_OP
3433 
3434 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3435 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3436   if (UseLSE) {                                                         \
3437     prev = prev->is_valid() ? prev : zr;                                \
3438     AOP(sz, newv, prev, addr);                                          \
3439     return;                                                             \
3440   }                                                                     \
3441   Register result = rscratch2;                                          \
3442   if (prev->is_valid())                                                 \
3443     result = different(prev, newv, addr) ? prev : rscratch2;            \
3444                                                                         \
3445   Label retry_load;                                                     \
3446   prfm(Address(addr), PSTL1STRM);                                       \
3447   bind(retry_load);                                                     \
3448   LDXR(result, addr);                                                   \
3449   STXR(rscratch1, newv, addr);                                          \
3450   cbnzw(rscratch1, retry_load);                                         \
3451   if (prev->is_valid() && prev != result)                               \
3452     mov(prev, result);                                                  \
3453 }
3454 
3455 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3456 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3457 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3458 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3459 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3460 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3461 
3462 #undef ATOMIC_XCHG
3463 
3464 #ifndef PRODUCT
3465 extern "C" void findpc(intptr_t x);
3466 #endif
3467 
3468 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3469 {
3470   // In order to get locks to work, we need to fake a in_VM state
3471   if (ShowMessageBoxOnError ) {
3472     JavaThread* thread = JavaThread::current();
3473     JavaThreadState saved_state = thread->thread_state();
3474     thread->set_thread_state(_thread_in_vm);
3475 #ifndef PRODUCT
3476     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3477       ttyLocker ttyl;
3478       BytecodeCounter::print();
3479     }
3480 #endif
3481     if (os::message_box(msg, "Execution stopped, print registers?")) {
3482       ttyLocker ttyl;
3483       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3484 #ifndef PRODUCT
3485       tty->cr();
3486       findpc(pc);
3487       tty->cr();
3488 #endif
3489       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3490       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3491       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3492       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3493       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3494       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3495       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3496       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3497       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3498       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3499       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3500       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3501       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3502       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3503       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3504       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3505       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3506       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3507       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3508       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3509       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3510       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3511       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3512       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3513       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3514       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3515       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3516       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3517       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3518       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3519       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3520       BREAKPOINT;
3521     }
3522   }
3523   fatal("DEBUG MESSAGE: %s", msg);
3524 }
3525 
3526 RegSet MacroAssembler::call_clobbered_gp_registers() {
3527   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3528 #ifndef R18_RESERVED
3529   regs += r18_tls;
3530 #endif
3531   return regs;
3532 }
3533 
3534 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3535   int step = 4 * wordSize;
3536   push(call_clobbered_gp_registers() - exclude, sp);
3537   sub(sp, sp, step);
3538   mov(rscratch1, -step);
3539   // Push v0-v7, v16-v31.
3540   for (int i = 31; i>= 4; i -= 4) {
3541     if (i <= v7->encoding() || i >= v16->encoding())
3542       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3543           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3544   }
3545   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3546       as_FloatRegister(3), T1D, Address(sp));
3547 }
3548 
3549 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3550   for (int i = 0; i < 32; i += 4) {
3551     if (i <= v7->encoding() || i >= v16->encoding())
3552       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3553           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3554   }
3555 
3556   reinitialize_ptrue();
3557 
3558   pop(call_clobbered_gp_registers() - exclude, sp);
3559 }
3560 
3561 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3562                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3563   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3564   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3565     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3566     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3567       sve_str(as_FloatRegister(i), Address(sp, i));
3568     }
3569   } else {
3570     int step = (save_vectors ? 8 : 4) * wordSize;
3571     mov(rscratch1, -step);
3572     sub(sp, sp, step);
3573     for (int i = 28; i >= 4; i -= 4) {
3574       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3575           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3576     }
3577     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3578   }
3579   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3580     sub(sp, sp, total_predicate_in_bytes);
3581     for (int i = 0; i < PRegister::number_of_registers; i++) {
3582       sve_str(as_PRegister(i), Address(sp, i));
3583     }
3584   }
3585 }
3586 
3587 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3588                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3589   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3590     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3591       sve_ldr(as_PRegister(i), Address(sp, i));
3592     }
3593     add(sp, sp, total_predicate_in_bytes);
3594   }
3595   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3596     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3597       sve_ldr(as_FloatRegister(i), Address(sp, i));
3598     }
3599     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3600   } else {
3601     int step = (restore_vectors ? 8 : 4) * wordSize;
3602     for (int i = 0; i <= 28; i += 4)
3603       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3604           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3605   }
3606 
3607   // We may use predicate registers and rely on ptrue with SVE,
3608   // regardless of wide vector (> 8 bytes) used or not.
3609   if (use_sve) {
3610     reinitialize_ptrue();
3611   }
3612 
3613   // integer registers except lr & sp
3614   pop(RegSet::range(r0, r17), sp);
3615 #ifdef R18_RESERVED
3616   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3617   pop(RegSet::range(r20, r29), sp);
3618 #else
3619   pop(RegSet::range(r18_tls, r29), sp);
3620 #endif
3621 }
3622 
3623 /**
3624  * Helpers for multiply_to_len().
3625  */
3626 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3627                                      Register src1, Register src2) {
3628   adds(dest_lo, dest_lo, src1);
3629   adc(dest_hi, dest_hi, zr);
3630   adds(dest_lo, dest_lo, src2);
3631   adc(final_dest_hi, dest_hi, zr);
3632 }
3633 
3634 // Generate an address from (r + r1 extend offset).  "size" is the
3635 // size of the operand.  The result may be in rscratch2.
3636 Address MacroAssembler::offsetted_address(Register r, Register r1,
3637                                           Address::extend ext, int offset, int size) {
3638   if (offset || (ext.shift() % size != 0)) {
3639     lea(rscratch2, Address(r, r1, ext));
3640     return Address(rscratch2, offset);
3641   } else {
3642     return Address(r, r1, ext);
3643   }
3644 }
3645 
3646 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3647 {
3648   assert(offset >= 0, "spill to negative address?");
3649   // Offset reachable ?
3650   //   Not aligned - 9 bits signed offset
3651   //   Aligned - 12 bits unsigned offset shifted
3652   Register base = sp;
3653   if ((offset & (size-1)) && offset >= (1<<8)) {
3654     add(tmp, base, offset & ((1<<12)-1));
3655     base = tmp;
3656     offset &= -1u<<12;
3657   }
3658 
3659   if (offset >= (1<<12) * size) {
3660     add(tmp, base, offset & (((1<<12)-1)<<12));
3661     base = tmp;
3662     offset &= ~(((1<<12)-1)<<12);
3663   }
3664 
3665   return Address(base, offset);
3666 }
3667 
3668 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3669   assert(offset >= 0, "spill to negative address?");
3670 
3671   Register base = sp;
3672 
3673   // An immediate offset in the range 0 to 255 which is multiplied
3674   // by the current vector or predicate register size in bytes.
3675   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3676     return Address(base, offset / sve_reg_size_in_bytes);
3677   }
3678 
3679   add(tmp, base, offset);
3680   return Address(tmp);
3681 }
3682 
3683 // Checks whether offset is aligned.
3684 // Returns true if it is, else false.
3685 bool MacroAssembler::merge_alignment_check(Register base,
3686                                            size_t size,
3687                                            int64_t cur_offset,
3688                                            int64_t prev_offset) const {
3689   if (AvoidUnalignedAccesses) {
3690     if (base == sp) {
3691       // Checks whether low offset if aligned to pair of registers.
3692       int64_t pair_mask = size * 2 - 1;
3693       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3694       return (offset & pair_mask) == 0;
3695     } else { // If base is not sp, we can't guarantee the access is aligned.
3696       return false;
3697     }
3698   } else {
3699     int64_t mask = size - 1;
3700     // Load/store pair instruction only supports element size aligned offset.
3701     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3702   }
3703 }
3704 
3705 // Checks whether current and previous loads/stores can be merged.
3706 // Returns true if it can be merged, else false.
3707 bool MacroAssembler::ldst_can_merge(Register rt,
3708                                     const Address &adr,
3709                                     size_t cur_size_in_bytes,
3710                                     bool is_store) const {
3711   address prev = pc() - NativeInstruction::instruction_size;
3712   address last = code()->last_insn();
3713 
3714   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3715     return false;
3716   }
3717 
3718   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3719     return false;
3720   }
3721 
3722   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3723   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3724 
3725   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3726   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3727 
3728   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3729     return false;
3730   }
3731 
3732   int64_t max_offset = 63 * prev_size_in_bytes;
3733   int64_t min_offset = -64 * prev_size_in_bytes;
3734 
3735   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3736 
3737   // Only same base can be merged.
3738   if (adr.base() != prev_ldst->base()) {
3739     return false;
3740   }
3741 
3742   int64_t cur_offset = adr.offset();
3743   int64_t prev_offset = prev_ldst->offset();
3744   size_t diff = abs(cur_offset - prev_offset);
3745   if (diff != prev_size_in_bytes) {
3746     return false;
3747   }
3748 
3749   // Following cases can not be merged:
3750   // ldr x2, [x2, #8]
3751   // ldr x3, [x2, #16]
3752   // or:
3753   // ldr x2, [x3, #8]
3754   // ldr x2, [x3, #16]
3755   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3756   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3757     return false;
3758   }
3759 
3760   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3761   // Offset range must be in ldp/stp instruction's range.
3762   if (low_offset > max_offset || low_offset < min_offset) {
3763     return false;
3764   }
3765 
3766   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3767     return true;
3768   }
3769 
3770   return false;
3771 }
3772 
3773 // Merge current load/store with previous load/store into ldp/stp.
3774 void MacroAssembler::merge_ldst(Register rt,
3775                                 const Address &adr,
3776                                 size_t cur_size_in_bytes,
3777                                 bool is_store) {
3778 
3779   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3780 
3781   Register rt_low, rt_high;
3782   address prev = pc() - NativeInstruction::instruction_size;
3783   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3784 
3785   int64_t offset;
3786 
3787   if (adr.offset() < prev_ldst->offset()) {
3788     offset = adr.offset();
3789     rt_low = rt;
3790     rt_high = prev_ldst->target();
3791   } else {
3792     offset = prev_ldst->offset();
3793     rt_low = prev_ldst->target();
3794     rt_high = rt;
3795   }
3796 
3797   Address adr_p = Address(prev_ldst->base(), offset);
3798   // Overwrite previous generated binary.
3799   code_section()->set_end(prev);
3800 
3801   const size_t sz = prev_ldst->size_in_bytes();
3802   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3803   if (!is_store) {
3804     BLOCK_COMMENT("merged ldr pair");
3805     if (sz == 8) {
3806       ldp(rt_low, rt_high, adr_p);
3807     } else {
3808       ldpw(rt_low, rt_high, adr_p);
3809     }
3810   } else {
3811     BLOCK_COMMENT("merged str pair");
3812     if (sz == 8) {
3813       stp(rt_low, rt_high, adr_p);
3814     } else {
3815       stpw(rt_low, rt_high, adr_p);
3816     }
3817   }
3818 }
3819 
3820 /**
3821  * Multiply 64 bit by 64 bit first loop.
3822  */
3823 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3824                                            Register y, Register y_idx, Register z,
3825                                            Register carry, Register product,
3826                                            Register idx, Register kdx) {
3827   //
3828   //  jlong carry, x[], y[], z[];
3829   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3830   //    huge_128 product = y[idx] * x[xstart] + carry;
3831   //    z[kdx] = (jlong)product;
3832   //    carry  = (jlong)(product >>> 64);
3833   //  }
3834   //  z[xstart] = carry;
3835   //
3836 
3837   Label L_first_loop, L_first_loop_exit;
3838   Label L_one_x, L_one_y, L_multiply;
3839 
3840   subsw(xstart, xstart, 1);
3841   br(Assembler::MI, L_one_x);
3842 
3843   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3844   ldr(x_xstart, Address(rscratch1));
3845   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3846 
3847   bind(L_first_loop);
3848   subsw(idx, idx, 1);
3849   br(Assembler::MI, L_first_loop_exit);
3850   subsw(idx, idx, 1);
3851   br(Assembler::MI, L_one_y);
3852   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3853   ldr(y_idx, Address(rscratch1));
3854   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
3855   bind(L_multiply);
3856 
3857   // AArch64 has a multiply-accumulate instruction that we can't use
3858   // here because it has no way to process carries, so we have to use
3859   // separate add and adc instructions.  Bah.
3860   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
3861   mul(product, x_xstart, y_idx);
3862   adds(product, product, carry);
3863   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
3864 
3865   subw(kdx, kdx, 2);
3866   ror(product, product, 32); // back to big-endian
3867   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
3868 
3869   b(L_first_loop);
3870 
3871   bind(L_one_y);
3872   ldrw(y_idx, Address(y,  0));
3873   b(L_multiply);
3874 
3875   bind(L_one_x);
3876   ldrw(x_xstart, Address(x,  0));
3877   b(L_first_loop);
3878 
3879   bind(L_first_loop_exit);
3880 }
3881 
3882 /**
3883  * Multiply 128 bit by 128. Unrolled inner loop.
3884  *
3885  */
3886 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
3887                                              Register carry, Register carry2,
3888                                              Register idx, Register jdx,
3889                                              Register yz_idx1, Register yz_idx2,
3890                                              Register tmp, Register tmp3, Register tmp4,
3891                                              Register tmp6, Register product_hi) {
3892 
3893   //   jlong carry, x[], y[], z[];
3894   //   int kdx = ystart+1;
3895   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
3896   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
3897   //     jlong carry2  = (jlong)(tmp3 >>> 64);
3898   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
3899   //     carry  = (jlong)(tmp4 >>> 64);
3900   //     z[kdx+idx+1] = (jlong)tmp3;
3901   //     z[kdx+idx] = (jlong)tmp4;
3902   //   }
3903   //   idx += 2;
3904   //   if (idx > 0) {
3905   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
3906   //     z[kdx+idx] = (jlong)yz_idx1;
3907   //     carry  = (jlong)(yz_idx1 >>> 64);
3908   //   }
3909   //
3910 
3911   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
3912 
3913   lsrw(jdx, idx, 2);
3914 
3915   bind(L_third_loop);
3916 
3917   subsw(jdx, jdx, 1);
3918   br(Assembler::MI, L_third_loop_exit);
3919   subw(idx, idx, 4);
3920 
3921   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3922 
3923   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
3924 
3925   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
3926 
3927   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
3928   ror(yz_idx2, yz_idx2, 32);
3929 
3930   ldp(rscratch2, rscratch1, Address(tmp6, 0));
3931 
3932   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
3933   umulh(tmp4, product_hi, yz_idx1);
3934 
3935   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
3936   ror(rscratch2, rscratch2, 32);
3937 
3938   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
3939   umulh(carry2, product_hi, yz_idx2);
3940 
3941   // propagate sum of both multiplications into carry:tmp4:tmp3
3942   adds(tmp3, tmp3, carry);
3943   adc(tmp4, tmp4, zr);
3944   adds(tmp3, tmp3, rscratch1);
3945   adcs(tmp4, tmp4, tmp);
3946   adc(carry, carry2, zr);
3947   adds(tmp4, tmp4, rscratch2);
3948   adc(carry, carry, zr);
3949 
3950   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
3951   ror(tmp4, tmp4, 32);
3952   stp(tmp4, tmp3, Address(tmp6, 0));
3953 
3954   b(L_third_loop);
3955   bind (L_third_loop_exit);
3956 
3957   andw (idx, idx, 0x3);
3958   cbz(idx, L_post_third_loop_done);
3959 
3960   Label L_check_1;
3961   subsw(idx, idx, 2);
3962   br(Assembler::MI, L_check_1);
3963 
3964   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3965   ldr(yz_idx1, Address(rscratch1, 0));
3966   ror(yz_idx1, yz_idx1, 32);
3967   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
3968   umulh(tmp4, product_hi, yz_idx1);
3969   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
3970   ldr(yz_idx2, Address(rscratch1, 0));
3971   ror(yz_idx2, yz_idx2, 32);
3972 
3973   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
3974 
3975   ror(tmp3, tmp3, 32);
3976   str(tmp3, Address(rscratch1, 0));
3977 
3978   bind (L_check_1);
3979 
3980   andw (idx, idx, 0x1);
3981   subsw(idx, idx, 1);
3982   br(Assembler::MI, L_post_third_loop_done);
3983   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3984   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
3985   umulh(carry2, tmp4, product_hi);
3986   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
3987 
3988   add2_with_carry(carry2, tmp3, tmp4, carry);
3989 
3990   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
3991   extr(carry, carry2, tmp3, 32);
3992 
3993   bind(L_post_third_loop_done);
3994 }
3995 
3996 /**
3997  * Code for BigInteger::multiplyToLen() intrinsic.
3998  *
3999  * r0: x
4000  * r1: xlen
4001  * r2: y
4002  * r3: ylen
4003  * r4:  z
4004  * r5: tmp0
4005  * r10: tmp1
4006  * r11: tmp2
4007  * r12: tmp3
4008  * r13: tmp4
4009  * r14: tmp5
4010  * r15: tmp6
4011  * r16: tmp7
4012  *
4013  */
4014 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4015                                      Register z, Register tmp0,
4016                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4017                                      Register tmp5, Register tmp6, Register product_hi) {
4018 
4019   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4020 
4021   const Register idx = tmp1;
4022   const Register kdx = tmp2;
4023   const Register xstart = tmp3;
4024 
4025   const Register y_idx = tmp4;
4026   const Register carry = tmp5;
4027   const Register product  = xlen;
4028   const Register x_xstart = tmp0;
4029 
4030   // First Loop.
4031   //
4032   //  final static long LONG_MASK = 0xffffffffL;
4033   //  int xstart = xlen - 1;
4034   //  int ystart = ylen - 1;
4035   //  long carry = 0;
4036   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4037   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4038   //    z[kdx] = (int)product;
4039   //    carry = product >>> 32;
4040   //  }
4041   //  z[xstart] = (int)carry;
4042   //
4043 
4044   movw(idx, ylen);       // idx = ylen;
4045   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4046   mov(carry, zr);        // carry = 0;
4047 
4048   Label L_done;
4049 
4050   movw(xstart, xlen);
4051   subsw(xstart, xstart, 1);
4052   br(Assembler::MI, L_done);
4053 
4054   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4055 
4056   Label L_second_loop;
4057   cbzw(kdx, L_second_loop);
4058 
4059   Label L_carry;
4060   subw(kdx, kdx, 1);
4061   cbzw(kdx, L_carry);
4062 
4063   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4064   lsr(carry, carry, 32);
4065   subw(kdx, kdx, 1);
4066 
4067   bind(L_carry);
4068   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4069 
4070   // Second and third (nested) loops.
4071   //
4072   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4073   //   carry = 0;
4074   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4075   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4076   //                    (z[k] & LONG_MASK) + carry;
4077   //     z[k] = (int)product;
4078   //     carry = product >>> 32;
4079   //   }
4080   //   z[i] = (int)carry;
4081   // }
4082   //
4083   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4084 
4085   const Register jdx = tmp1;
4086 
4087   bind(L_second_loop);
4088   mov(carry, zr);                // carry = 0;
4089   movw(jdx, ylen);               // j = ystart+1
4090 
4091   subsw(xstart, xstart, 1);      // i = xstart-1;
4092   br(Assembler::MI, L_done);
4093 
4094   str(z, Address(pre(sp, -4 * wordSize)));
4095 
4096   Label L_last_x;
4097   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4098   subsw(xstart, xstart, 1);       // i = xstart-1;
4099   br(Assembler::MI, L_last_x);
4100 
4101   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4102   ldr(product_hi, Address(rscratch1));
4103   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4104 
4105   Label L_third_loop_prologue;
4106   bind(L_third_loop_prologue);
4107 
4108   str(ylen, Address(sp, wordSize));
4109   stp(x, xstart, Address(sp, 2 * wordSize));
4110   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4111                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4112   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4113   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4114 
4115   addw(tmp3, xlen, 1);
4116   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4117   subsw(tmp3, tmp3, 1);
4118   br(Assembler::MI, L_done);
4119 
4120   lsr(carry, carry, 32);
4121   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4122   b(L_second_loop);
4123 
4124   // Next infrequent code is moved outside loops.
4125   bind(L_last_x);
4126   ldrw(product_hi, Address(x,  0));
4127   b(L_third_loop_prologue);
4128 
4129   bind(L_done);
4130 }
4131 
4132 // Code for BigInteger::mulAdd intrinsic
4133 // out     = r0
4134 // in      = r1
4135 // offset  = r2  (already out.length-offset)
4136 // len     = r3
4137 // k       = r4
4138 //
4139 // pseudo code from java implementation:
4140 // carry = 0;
4141 // offset = out.length-offset - 1;
4142 // for (int j=len-1; j >= 0; j--) {
4143 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4144 //     out[offset--] = (int)product;
4145 //     carry = product >>> 32;
4146 // }
4147 // return (int)carry;
4148 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4149       Register len, Register k) {
4150     Label LOOP, END;
4151     // pre-loop
4152     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4153     csel(out, zr, out, Assembler::EQ);
4154     br(Assembler::EQ, END);
4155     add(in, in, len, LSL, 2); // in[j+1] address
4156     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4157     mov(out, zr); // used to keep carry now
4158     BIND(LOOP);
4159     ldrw(rscratch1, Address(pre(in, -4)));
4160     madd(rscratch1, rscratch1, k, out);
4161     ldrw(rscratch2, Address(pre(offset, -4)));
4162     add(rscratch1, rscratch1, rscratch2);
4163     strw(rscratch1, Address(offset));
4164     lsr(out, rscratch1, 32);
4165     subs(len, len, 1);
4166     br(Assembler::NE, LOOP);
4167     BIND(END);
4168 }
4169 
4170 /**
4171  * Emits code to update CRC-32 with a byte value according to constants in table
4172  *
4173  * @param [in,out]crc   Register containing the crc.
4174  * @param [in]val       Register containing the byte to fold into the CRC.
4175  * @param [in]table     Register containing the table of crc constants.
4176  *
4177  * uint32_t crc;
4178  * val = crc_table[(val ^ crc) & 0xFF];
4179  * crc = val ^ (crc >> 8);
4180  *
4181  */
4182 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4183   eor(val, val, crc);
4184   andr(val, val, 0xff);
4185   ldrw(val, Address(table, val, Address::lsl(2)));
4186   eor(crc, val, crc, Assembler::LSR, 8);
4187 }
4188 
4189 /**
4190  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4191  *
4192  * @param [in,out]crc   Register containing the crc.
4193  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4194  * @param [in]table0    Register containing table 0 of crc constants.
4195  * @param [in]table1    Register containing table 1 of crc constants.
4196  * @param [in]table2    Register containing table 2 of crc constants.
4197  * @param [in]table3    Register containing table 3 of crc constants.
4198  *
4199  * uint32_t crc;
4200  *   v = crc ^ v
4201  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4202  *
4203  */
4204 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4205         Register table0, Register table1, Register table2, Register table3,
4206         bool upper) {
4207   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4208   uxtb(tmp, v);
4209   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4210   ubfx(tmp, v, 8, 8);
4211   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4212   eor(crc, crc, tmp);
4213   ubfx(tmp, v, 16, 8);
4214   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4215   eor(crc, crc, tmp);
4216   ubfx(tmp, v, 24, 8);
4217   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4218   eor(crc, crc, tmp);
4219 }
4220 
4221 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4222         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4223     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4224     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4225 
4226     subs(tmp0, len, 384);
4227     mvnw(crc, crc);
4228     br(Assembler::GE, CRC_by128_pre);
4229   BIND(CRC_less128);
4230     subs(len, len, 32);
4231     br(Assembler::GE, CRC_by32_loop);
4232   BIND(CRC_less32);
4233     adds(len, len, 32 - 4);
4234     br(Assembler::GE, CRC_by4_loop);
4235     adds(len, len, 4);
4236     br(Assembler::GT, CRC_by1_loop);
4237     b(L_exit);
4238 
4239   BIND(CRC_by32_loop);
4240     ldp(tmp0, tmp1, Address(buf));
4241     crc32x(crc, crc, tmp0);
4242     ldp(tmp2, tmp3, Address(buf, 16));
4243     crc32x(crc, crc, tmp1);
4244     add(buf, buf, 32);
4245     crc32x(crc, crc, tmp2);
4246     subs(len, len, 32);
4247     crc32x(crc, crc, tmp3);
4248     br(Assembler::GE, CRC_by32_loop);
4249     cmn(len, (u1)32);
4250     br(Assembler::NE, CRC_less32);
4251     b(L_exit);
4252 
4253   BIND(CRC_by4_loop);
4254     ldrw(tmp0, Address(post(buf, 4)));
4255     subs(len, len, 4);
4256     crc32w(crc, crc, tmp0);
4257     br(Assembler::GE, CRC_by4_loop);
4258     adds(len, len, 4);
4259     br(Assembler::LE, L_exit);
4260   BIND(CRC_by1_loop);
4261     ldrb(tmp0, Address(post(buf, 1)));
4262     subs(len, len, 1);
4263     crc32b(crc, crc, tmp0);
4264     br(Assembler::GT, CRC_by1_loop);
4265     b(L_exit);
4266 
4267   BIND(CRC_by128_pre);
4268     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4269       4*256*sizeof(juint) + 8*sizeof(juint));
4270     mov(crc, 0);
4271     crc32x(crc, crc, tmp0);
4272     crc32x(crc, crc, tmp1);
4273 
4274     cbnz(len, CRC_less128);
4275 
4276   BIND(L_exit);
4277     mvnw(crc, crc);
4278 }
4279 
4280 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4281         Register len, Register tmp0, Register tmp1, Register tmp2,
4282         Register tmp3) {
4283     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4284     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4285 
4286     mvnw(crc, crc);
4287 
4288     subs(len, len, 128);
4289     br(Assembler::GE, CRC_by64_pre);
4290   BIND(CRC_less64);
4291     adds(len, len, 128-32);
4292     br(Assembler::GE, CRC_by32_loop);
4293   BIND(CRC_less32);
4294     adds(len, len, 32-4);
4295     br(Assembler::GE, CRC_by4_loop);
4296     adds(len, len, 4);
4297     br(Assembler::GT, CRC_by1_loop);
4298     b(L_exit);
4299 
4300   BIND(CRC_by32_loop);
4301     ldp(tmp0, tmp1, Address(post(buf, 16)));
4302     subs(len, len, 32);
4303     crc32x(crc, crc, tmp0);
4304     ldr(tmp2, Address(post(buf, 8)));
4305     crc32x(crc, crc, tmp1);
4306     ldr(tmp3, Address(post(buf, 8)));
4307     crc32x(crc, crc, tmp2);
4308     crc32x(crc, crc, tmp3);
4309     br(Assembler::GE, CRC_by32_loop);
4310     cmn(len, (u1)32);
4311     br(Assembler::NE, CRC_less32);
4312     b(L_exit);
4313 
4314   BIND(CRC_by4_loop);
4315     ldrw(tmp0, Address(post(buf, 4)));
4316     subs(len, len, 4);
4317     crc32w(crc, crc, tmp0);
4318     br(Assembler::GE, CRC_by4_loop);
4319     adds(len, len, 4);
4320     br(Assembler::LE, L_exit);
4321   BIND(CRC_by1_loop);
4322     ldrb(tmp0, Address(post(buf, 1)));
4323     subs(len, len, 1);
4324     crc32b(crc, crc, tmp0);
4325     br(Assembler::GT, CRC_by1_loop);
4326     b(L_exit);
4327 
4328   BIND(CRC_by64_pre);
4329     sub(buf, buf, 8);
4330     ldp(tmp0, tmp1, Address(buf, 8));
4331     crc32x(crc, crc, tmp0);
4332     ldr(tmp2, Address(buf, 24));
4333     crc32x(crc, crc, tmp1);
4334     ldr(tmp3, Address(buf, 32));
4335     crc32x(crc, crc, tmp2);
4336     ldr(tmp0, Address(buf, 40));
4337     crc32x(crc, crc, tmp3);
4338     ldr(tmp1, Address(buf, 48));
4339     crc32x(crc, crc, tmp0);
4340     ldr(tmp2, Address(buf, 56));
4341     crc32x(crc, crc, tmp1);
4342     ldr(tmp3, Address(pre(buf, 64)));
4343 
4344     b(CRC_by64_loop);
4345 
4346     align(CodeEntryAlignment);
4347   BIND(CRC_by64_loop);
4348     subs(len, len, 64);
4349     crc32x(crc, crc, tmp2);
4350     ldr(tmp0, Address(buf, 8));
4351     crc32x(crc, crc, tmp3);
4352     ldr(tmp1, Address(buf, 16));
4353     crc32x(crc, crc, tmp0);
4354     ldr(tmp2, Address(buf, 24));
4355     crc32x(crc, crc, tmp1);
4356     ldr(tmp3, Address(buf, 32));
4357     crc32x(crc, crc, tmp2);
4358     ldr(tmp0, Address(buf, 40));
4359     crc32x(crc, crc, tmp3);
4360     ldr(tmp1, Address(buf, 48));
4361     crc32x(crc, crc, tmp0);
4362     ldr(tmp2, Address(buf, 56));
4363     crc32x(crc, crc, tmp1);
4364     ldr(tmp3, Address(pre(buf, 64)));
4365     br(Assembler::GE, CRC_by64_loop);
4366 
4367     // post-loop
4368     crc32x(crc, crc, tmp2);
4369     crc32x(crc, crc, tmp3);
4370 
4371     sub(len, len, 64);
4372     add(buf, buf, 8);
4373     cmn(len, (u1)128);
4374     br(Assembler::NE, CRC_less64);
4375   BIND(L_exit);
4376     mvnw(crc, crc);
4377 }
4378 
4379 /**
4380  * @param crc   register containing existing CRC (32-bit)
4381  * @param buf   register pointing to input byte buffer (byte*)
4382  * @param len   register containing number of bytes
4383  * @param table register that will contain address of CRC table
4384  * @param tmp   scratch register
4385  */
4386 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4387         Register table0, Register table1, Register table2, Register table3,
4388         Register tmp, Register tmp2, Register tmp3) {
4389   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4390 
4391   if (UseCryptoPmullForCRC32) {
4392       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4393       return;
4394   }
4395 
4396   if (UseCRC32) {
4397       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4398       return;
4399   }
4400 
4401     mvnw(crc, crc);
4402 
4403     {
4404       uint64_t offset;
4405       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4406       add(table0, table0, offset);
4407     }
4408     add(table1, table0, 1*256*sizeof(juint));
4409     add(table2, table0, 2*256*sizeof(juint));
4410     add(table3, table0, 3*256*sizeof(juint));
4411 
4412     { // Neon code start
4413       cmp(len, (u1)64);
4414       br(Assembler::LT, L_by16);
4415       eor(v16, T16B, v16, v16);
4416 
4417     Label L_fold;
4418 
4419       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4420 
4421       ld1(v0, v1, T2D, post(buf, 32));
4422       ld1r(v4, T2D, post(tmp, 8));
4423       ld1r(v5, T2D, post(tmp, 8));
4424       ld1r(v6, T2D, post(tmp, 8));
4425       ld1r(v7, T2D, post(tmp, 8));
4426       mov(v16, S, 0, crc);
4427 
4428       eor(v0, T16B, v0, v16);
4429       sub(len, len, 64);
4430 
4431     BIND(L_fold);
4432       pmull(v22, T8H, v0, v5, T8B);
4433       pmull(v20, T8H, v0, v7, T8B);
4434       pmull(v23, T8H, v0, v4, T8B);
4435       pmull(v21, T8H, v0, v6, T8B);
4436 
4437       pmull2(v18, T8H, v0, v5, T16B);
4438       pmull2(v16, T8H, v0, v7, T16B);
4439       pmull2(v19, T8H, v0, v4, T16B);
4440       pmull2(v17, T8H, v0, v6, T16B);
4441 
4442       uzp1(v24, T8H, v20, v22);
4443       uzp2(v25, T8H, v20, v22);
4444       eor(v20, T16B, v24, v25);
4445 
4446       uzp1(v26, T8H, v16, v18);
4447       uzp2(v27, T8H, v16, v18);
4448       eor(v16, T16B, v26, v27);
4449 
4450       ushll2(v22, T4S, v20, T8H, 8);
4451       ushll(v20, T4S, v20, T4H, 8);
4452 
4453       ushll2(v18, T4S, v16, T8H, 8);
4454       ushll(v16, T4S, v16, T4H, 8);
4455 
4456       eor(v22, T16B, v23, v22);
4457       eor(v18, T16B, v19, v18);
4458       eor(v20, T16B, v21, v20);
4459       eor(v16, T16B, v17, v16);
4460 
4461       uzp1(v17, T2D, v16, v20);
4462       uzp2(v21, T2D, v16, v20);
4463       eor(v17, T16B, v17, v21);
4464 
4465       ushll2(v20, T2D, v17, T4S, 16);
4466       ushll(v16, T2D, v17, T2S, 16);
4467 
4468       eor(v20, T16B, v20, v22);
4469       eor(v16, T16B, v16, v18);
4470 
4471       uzp1(v17, T2D, v20, v16);
4472       uzp2(v21, T2D, v20, v16);
4473       eor(v28, T16B, v17, v21);
4474 
4475       pmull(v22, T8H, v1, v5, T8B);
4476       pmull(v20, T8H, v1, v7, T8B);
4477       pmull(v23, T8H, v1, v4, T8B);
4478       pmull(v21, T8H, v1, v6, T8B);
4479 
4480       pmull2(v18, T8H, v1, v5, T16B);
4481       pmull2(v16, T8H, v1, v7, T16B);
4482       pmull2(v19, T8H, v1, v4, T16B);
4483       pmull2(v17, T8H, v1, v6, T16B);
4484 
4485       ld1(v0, v1, T2D, post(buf, 32));
4486 
4487       uzp1(v24, T8H, v20, v22);
4488       uzp2(v25, T8H, v20, v22);
4489       eor(v20, T16B, v24, v25);
4490 
4491       uzp1(v26, T8H, v16, v18);
4492       uzp2(v27, T8H, v16, v18);
4493       eor(v16, T16B, v26, v27);
4494 
4495       ushll2(v22, T4S, v20, T8H, 8);
4496       ushll(v20, T4S, v20, T4H, 8);
4497 
4498       ushll2(v18, T4S, v16, T8H, 8);
4499       ushll(v16, T4S, v16, T4H, 8);
4500 
4501       eor(v22, T16B, v23, v22);
4502       eor(v18, T16B, v19, v18);
4503       eor(v20, T16B, v21, v20);
4504       eor(v16, T16B, v17, v16);
4505 
4506       uzp1(v17, T2D, v16, v20);
4507       uzp2(v21, T2D, v16, v20);
4508       eor(v16, T16B, v17, v21);
4509 
4510       ushll2(v20, T2D, v16, T4S, 16);
4511       ushll(v16, T2D, v16, T2S, 16);
4512 
4513       eor(v20, T16B, v22, v20);
4514       eor(v16, T16B, v16, v18);
4515 
4516       uzp1(v17, T2D, v20, v16);
4517       uzp2(v21, T2D, v20, v16);
4518       eor(v20, T16B, v17, v21);
4519 
4520       shl(v16, T2D, v28, 1);
4521       shl(v17, T2D, v20, 1);
4522 
4523       eor(v0, T16B, v0, v16);
4524       eor(v1, T16B, v1, v17);
4525 
4526       subs(len, len, 32);
4527       br(Assembler::GE, L_fold);
4528 
4529       mov(crc, 0);
4530       mov(tmp, v0, D, 0);
4531       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4532       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4533       mov(tmp, v0, D, 1);
4534       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4535       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4536       mov(tmp, v1, D, 0);
4537       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4538       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4539       mov(tmp, v1, D, 1);
4540       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4541       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4542 
4543       add(len, len, 32);
4544     } // Neon code end
4545 
4546   BIND(L_by16);
4547     subs(len, len, 16);
4548     br(Assembler::GE, L_by16_loop);
4549     adds(len, len, 16-4);
4550     br(Assembler::GE, L_by4_loop);
4551     adds(len, len, 4);
4552     br(Assembler::GT, L_by1_loop);
4553     b(L_exit);
4554 
4555   BIND(L_by4_loop);
4556     ldrw(tmp, Address(post(buf, 4)));
4557     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4558     subs(len, len, 4);
4559     br(Assembler::GE, L_by4_loop);
4560     adds(len, len, 4);
4561     br(Assembler::LE, L_exit);
4562   BIND(L_by1_loop);
4563     subs(len, len, 1);
4564     ldrb(tmp, Address(post(buf, 1)));
4565     update_byte_crc32(crc, tmp, table0);
4566     br(Assembler::GT, L_by1_loop);
4567     b(L_exit);
4568 
4569     align(CodeEntryAlignment);
4570   BIND(L_by16_loop);
4571     subs(len, len, 16);
4572     ldp(tmp, tmp3, Address(post(buf, 16)));
4573     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4574     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4575     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4576     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4577     br(Assembler::GE, L_by16_loop);
4578     adds(len, len, 16-4);
4579     br(Assembler::GE, L_by4_loop);
4580     adds(len, len, 4);
4581     br(Assembler::GT, L_by1_loop);
4582   BIND(L_exit);
4583     mvnw(crc, crc);
4584 }
4585 
4586 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4587         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4588     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4589     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4590 
4591     subs(tmp0, len, 384);
4592     br(Assembler::GE, CRC_by128_pre);
4593   BIND(CRC_less128);
4594     subs(len, len, 32);
4595     br(Assembler::GE, CRC_by32_loop);
4596   BIND(CRC_less32);
4597     adds(len, len, 32 - 4);
4598     br(Assembler::GE, CRC_by4_loop);
4599     adds(len, len, 4);
4600     br(Assembler::GT, CRC_by1_loop);
4601     b(L_exit);
4602 
4603   BIND(CRC_by32_loop);
4604     ldp(tmp0, tmp1, Address(buf));
4605     crc32cx(crc, crc, tmp0);
4606     ldr(tmp2, Address(buf, 16));
4607     crc32cx(crc, crc, tmp1);
4608     ldr(tmp3, Address(buf, 24));
4609     crc32cx(crc, crc, tmp2);
4610     add(buf, buf, 32);
4611     subs(len, len, 32);
4612     crc32cx(crc, crc, tmp3);
4613     br(Assembler::GE, CRC_by32_loop);
4614     cmn(len, (u1)32);
4615     br(Assembler::NE, CRC_less32);
4616     b(L_exit);
4617 
4618   BIND(CRC_by4_loop);
4619     ldrw(tmp0, Address(post(buf, 4)));
4620     subs(len, len, 4);
4621     crc32cw(crc, crc, tmp0);
4622     br(Assembler::GE, CRC_by4_loop);
4623     adds(len, len, 4);
4624     br(Assembler::LE, L_exit);
4625   BIND(CRC_by1_loop);
4626     ldrb(tmp0, Address(post(buf, 1)));
4627     subs(len, len, 1);
4628     crc32cb(crc, crc, tmp0);
4629     br(Assembler::GT, CRC_by1_loop);
4630     b(L_exit);
4631 
4632   BIND(CRC_by128_pre);
4633     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4634       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4635     mov(crc, 0);
4636     crc32cx(crc, crc, tmp0);
4637     crc32cx(crc, crc, tmp1);
4638 
4639     cbnz(len, CRC_less128);
4640 
4641   BIND(L_exit);
4642 }
4643 
4644 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4645         Register len, Register tmp0, Register tmp1, Register tmp2,
4646         Register tmp3) {
4647     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4648     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4649 
4650     subs(len, len, 128);
4651     br(Assembler::GE, CRC_by64_pre);
4652   BIND(CRC_less64);
4653     adds(len, len, 128-32);
4654     br(Assembler::GE, CRC_by32_loop);
4655   BIND(CRC_less32);
4656     adds(len, len, 32-4);
4657     br(Assembler::GE, CRC_by4_loop);
4658     adds(len, len, 4);
4659     br(Assembler::GT, CRC_by1_loop);
4660     b(L_exit);
4661 
4662   BIND(CRC_by32_loop);
4663     ldp(tmp0, tmp1, Address(post(buf, 16)));
4664     subs(len, len, 32);
4665     crc32cx(crc, crc, tmp0);
4666     ldr(tmp2, Address(post(buf, 8)));
4667     crc32cx(crc, crc, tmp1);
4668     ldr(tmp3, Address(post(buf, 8)));
4669     crc32cx(crc, crc, tmp2);
4670     crc32cx(crc, crc, tmp3);
4671     br(Assembler::GE, CRC_by32_loop);
4672     cmn(len, (u1)32);
4673     br(Assembler::NE, CRC_less32);
4674     b(L_exit);
4675 
4676   BIND(CRC_by4_loop);
4677     ldrw(tmp0, Address(post(buf, 4)));
4678     subs(len, len, 4);
4679     crc32cw(crc, crc, tmp0);
4680     br(Assembler::GE, CRC_by4_loop);
4681     adds(len, len, 4);
4682     br(Assembler::LE, L_exit);
4683   BIND(CRC_by1_loop);
4684     ldrb(tmp0, Address(post(buf, 1)));
4685     subs(len, len, 1);
4686     crc32cb(crc, crc, tmp0);
4687     br(Assembler::GT, CRC_by1_loop);
4688     b(L_exit);
4689 
4690   BIND(CRC_by64_pre);
4691     sub(buf, buf, 8);
4692     ldp(tmp0, tmp1, Address(buf, 8));
4693     crc32cx(crc, crc, tmp0);
4694     ldr(tmp2, Address(buf, 24));
4695     crc32cx(crc, crc, tmp1);
4696     ldr(tmp3, Address(buf, 32));
4697     crc32cx(crc, crc, tmp2);
4698     ldr(tmp0, Address(buf, 40));
4699     crc32cx(crc, crc, tmp3);
4700     ldr(tmp1, Address(buf, 48));
4701     crc32cx(crc, crc, tmp0);
4702     ldr(tmp2, Address(buf, 56));
4703     crc32cx(crc, crc, tmp1);
4704     ldr(tmp3, Address(pre(buf, 64)));
4705 
4706     b(CRC_by64_loop);
4707 
4708     align(CodeEntryAlignment);
4709   BIND(CRC_by64_loop);
4710     subs(len, len, 64);
4711     crc32cx(crc, crc, tmp2);
4712     ldr(tmp0, Address(buf, 8));
4713     crc32cx(crc, crc, tmp3);
4714     ldr(tmp1, Address(buf, 16));
4715     crc32cx(crc, crc, tmp0);
4716     ldr(tmp2, Address(buf, 24));
4717     crc32cx(crc, crc, tmp1);
4718     ldr(tmp3, Address(buf, 32));
4719     crc32cx(crc, crc, tmp2);
4720     ldr(tmp0, Address(buf, 40));
4721     crc32cx(crc, crc, tmp3);
4722     ldr(tmp1, Address(buf, 48));
4723     crc32cx(crc, crc, tmp0);
4724     ldr(tmp2, Address(buf, 56));
4725     crc32cx(crc, crc, tmp1);
4726     ldr(tmp3, Address(pre(buf, 64)));
4727     br(Assembler::GE, CRC_by64_loop);
4728 
4729     // post-loop
4730     crc32cx(crc, crc, tmp2);
4731     crc32cx(crc, crc, tmp3);
4732 
4733     sub(len, len, 64);
4734     add(buf, buf, 8);
4735     cmn(len, (u1)128);
4736     br(Assembler::NE, CRC_less64);
4737   BIND(L_exit);
4738 }
4739 
4740 /**
4741  * @param crc   register containing existing CRC (32-bit)
4742  * @param buf   register pointing to input byte buffer (byte*)
4743  * @param len   register containing number of bytes
4744  * @param table register that will contain address of CRC table
4745  * @param tmp   scratch register
4746  */
4747 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4748         Register table0, Register table1, Register table2, Register table3,
4749         Register tmp, Register tmp2, Register tmp3) {
4750   if (UseCryptoPmullForCRC32) {
4751     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4752   } else {
4753     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4754   }
4755 }
4756 
4757 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4758         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4759     Label CRC_by128_loop;
4760     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4761 
4762     sub(len, len, 256);
4763     Register table = tmp0;
4764     {
4765       uint64_t offset;
4766       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4767       add(table, table, offset);
4768     }
4769     add(table, table, table_offset);
4770 
4771     // Registers v0..v7 are used as data registers.
4772     // Registers v16..v31 are used as tmp registers.
4773     sub(buf, buf, 0x10);
4774     ldrq(v0, Address(buf, 0x10));
4775     ldrq(v1, Address(buf, 0x20));
4776     ldrq(v2, Address(buf, 0x30));
4777     ldrq(v3, Address(buf, 0x40));
4778     ldrq(v4, Address(buf, 0x50));
4779     ldrq(v5, Address(buf, 0x60));
4780     ldrq(v6, Address(buf, 0x70));
4781     ldrq(v7, Address(pre(buf, 0x80)));
4782 
4783     movi(v31, T4S, 0);
4784     mov(v31, S, 0, crc);
4785     eor(v0, T16B, v0, v31);
4786 
4787     // Register v16 contains constants from the crc table.
4788     ldrq(v16, Address(table));
4789     b(CRC_by128_loop);
4790 
4791     align(OptoLoopAlignment);
4792   BIND(CRC_by128_loop);
4793     pmull (v17,  T1Q, v0, v16, T1D);
4794     pmull2(v18, T1Q, v0, v16, T2D);
4795     ldrq(v0, Address(buf, 0x10));
4796     eor3(v0, T16B, v17,  v18, v0);
4797 
4798     pmull (v19, T1Q, v1, v16, T1D);
4799     pmull2(v20, T1Q, v1, v16, T2D);
4800     ldrq(v1, Address(buf, 0x20));
4801     eor3(v1, T16B, v19, v20, v1);
4802 
4803     pmull (v21, T1Q, v2, v16, T1D);
4804     pmull2(v22, T1Q, v2, v16, T2D);
4805     ldrq(v2, Address(buf, 0x30));
4806     eor3(v2, T16B, v21, v22, v2);
4807 
4808     pmull (v23, T1Q, v3, v16, T1D);
4809     pmull2(v24, T1Q, v3, v16, T2D);
4810     ldrq(v3, Address(buf, 0x40));
4811     eor3(v3, T16B, v23, v24, v3);
4812 
4813     pmull (v25, T1Q, v4, v16, T1D);
4814     pmull2(v26, T1Q, v4, v16, T2D);
4815     ldrq(v4, Address(buf, 0x50));
4816     eor3(v4, T16B, v25, v26, v4);
4817 
4818     pmull (v27, T1Q, v5, v16, T1D);
4819     pmull2(v28, T1Q, v5, v16, T2D);
4820     ldrq(v5, Address(buf, 0x60));
4821     eor3(v5, T16B, v27, v28, v5);
4822 
4823     pmull (v29, T1Q, v6, v16, T1D);
4824     pmull2(v30, T1Q, v6, v16, T2D);
4825     ldrq(v6, Address(buf, 0x70));
4826     eor3(v6, T16B, v29, v30, v6);
4827 
4828     // Reuse registers v23, v24.
4829     // Using them won't block the first instruction of the next iteration.
4830     pmull (v23, T1Q, v7, v16, T1D);
4831     pmull2(v24, T1Q, v7, v16, T2D);
4832     ldrq(v7, Address(pre(buf, 0x80)));
4833     eor3(v7, T16B, v23, v24, v7);
4834 
4835     subs(len, len, 0x80);
4836     br(Assembler::GE, CRC_by128_loop);
4837 
4838     // fold into 512 bits
4839     // Use v31 for constants because v16 can be still in use.
4840     ldrq(v31, Address(table, 0x10));
4841 
4842     pmull (v17,  T1Q, v0, v31, T1D);
4843     pmull2(v18, T1Q, v0, v31, T2D);
4844     eor3(v0, T16B, v17, v18, v4);
4845 
4846     pmull (v19, T1Q, v1, v31, T1D);
4847     pmull2(v20, T1Q, v1, v31, T2D);
4848     eor3(v1, T16B, v19, v20, v5);
4849 
4850     pmull (v21, T1Q, v2, v31, T1D);
4851     pmull2(v22, T1Q, v2, v31, T2D);
4852     eor3(v2, T16B, v21, v22, v6);
4853 
4854     pmull (v23, T1Q, v3, v31, T1D);
4855     pmull2(v24, T1Q, v3, v31, T2D);
4856     eor3(v3, T16B, v23, v24, v7);
4857 
4858     // fold into 128 bits
4859     // Use v17 for constants because v31 can be still in use.
4860     ldrq(v17, Address(table, 0x20));
4861     pmull (v25, T1Q, v0, v17, T1D);
4862     pmull2(v26, T1Q, v0, v17, T2D);
4863     eor3(v3, T16B, v3, v25, v26);
4864 
4865     // Use v18 for constants because v17 can be still in use.
4866     ldrq(v18, Address(table, 0x30));
4867     pmull (v27, T1Q, v1, v18, T1D);
4868     pmull2(v28, T1Q, v1, v18, T2D);
4869     eor3(v3, T16B, v3, v27, v28);
4870 
4871     // Use v19 for constants because v18 can be still in use.
4872     ldrq(v19, Address(table, 0x40));
4873     pmull (v29, T1Q, v2, v19, T1D);
4874     pmull2(v30, T1Q, v2, v19, T2D);
4875     eor3(v0, T16B, v3, v29, v30);
4876 
4877     add(len, len, 0x80);
4878     add(buf, buf, 0x10);
4879 
4880     mov(tmp0, v0, D, 0);
4881     mov(tmp1, v0, D, 1);
4882 }
4883 
4884 void MacroAssembler::addptr(const Address &dst, int32_t src) {
4885   Address adr;
4886   switch(dst.getMode()) {
4887   case Address::base_plus_offset:
4888     // This is the expected mode, although we allow all the other
4889     // forms below.
4890     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
4891     break;
4892   default:
4893     lea(rscratch2, dst);
4894     adr = Address(rscratch2);
4895     break;
4896   }
4897   ldr(rscratch1, adr);
4898   add(rscratch1, rscratch1, src);
4899   str(rscratch1, adr);
4900 }
4901 
4902 void MacroAssembler::cmpptr(Register src1, Address src2) {
4903   uint64_t offset;
4904   adrp(rscratch1, src2, offset);
4905   ldr(rscratch1, Address(rscratch1, offset));
4906   cmp(src1, rscratch1);
4907 }
4908 
4909 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
4910   cmp(obj1, obj2);
4911 }
4912 
4913 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
4914   load_method_holder(rresult, rmethod);
4915   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
4916 }
4917 
4918 void MacroAssembler::load_method_holder(Register holder, Register method) {
4919   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
4920   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
4921   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
4922 }
4923 
4924 // Loads the obj's Klass* into dst.
4925 // Preserves all registers (incl src, rscratch1 and rscratch2).
4926 // Input:
4927 // src - the oop we want to load the klass from.
4928 // dst - output narrow klass.
4929 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
4930   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
4931   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
4932   lsr(dst, dst, markWord::klass_shift);
4933 }
4934 
4935 void MacroAssembler::load_klass(Register dst, Register src) {
4936   if (UseCompactObjectHeaders) {
4937     load_narrow_klass_compact(dst, src);
4938     decode_klass_not_null(dst);
4939   } else if (UseCompressedClassPointers) {
4940     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
4941     decode_klass_not_null(dst);
4942   } else {
4943     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
4944   }
4945 }
4946 
4947 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
4948   if (RestoreMXCSROnJNICalls) {
4949     Label OK;
4950     get_fpcr(tmp1);
4951     mov(tmp2, tmp1);
4952     // Set FPCR to the state we need. We do want Round to Nearest. We
4953     // don't want non-IEEE rounding modes or floating-point traps.
4954     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
4955     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
4956     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
4957     eor(tmp2, tmp1, tmp2);
4958     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
4959     set_fpcr(tmp1);
4960     bind(OK);
4961   }
4962 }
4963 
4964 // ((OopHandle)result).resolve();
4965 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
4966   // OopHandle::resolve is an indirection.
4967   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
4968 }
4969 
4970 // ((WeakHandle)result).resolve();
4971 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
4972   assert_different_registers(result, tmp1, tmp2);
4973   Label resolved;
4974 
4975   // A null weak handle resolves to null.
4976   cbz(result, resolved);
4977 
4978   // Only 64 bit platforms support GCs that require a tmp register
4979   // WeakHandle::resolve is an indirection like jweak.
4980   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
4981                  result, Address(result), tmp1, tmp2);
4982   bind(resolved);
4983 }
4984 
4985 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
4986   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
4987   ldr(dst, Address(rmethod, Method::const_offset()));
4988   ldr(dst, Address(dst, ConstMethod::constants_offset()));
4989   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
4990   ldr(dst, Address(dst, mirror_offset));
4991   resolve_oop_handle(dst, tmp1, tmp2);
4992 }
4993 
4994 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
4995   assert_different_registers(obj, klass, tmp);
4996   if (UseCompressedClassPointers) {
4997     if (UseCompactObjectHeaders) {
4998       load_narrow_klass_compact(tmp, obj);
4999     } else {
5000       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5001     }
5002     if (CompressedKlassPointers::base() == nullptr) {
5003       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5004       return;
5005     } else if (!AOTCodeCache::is_on_for_dump() &&
5006                ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5007                && CompressedKlassPointers::shift() == 0) {
5008       // Only the bottom 32 bits matter
5009       cmpw(klass, tmp);
5010       return;
5011     }
5012     decode_klass_not_null(tmp);
5013   } else {
5014     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5015   }
5016   cmp(klass, tmp);
5017 }
5018 
5019 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5020   if (UseCompactObjectHeaders) {
5021     load_narrow_klass_compact(tmp1, obj1);
5022     load_narrow_klass_compact(tmp2,  obj2);
5023     cmpw(tmp1, tmp2);
5024   } else if (UseCompressedClassPointers) {
5025     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5026     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5027     cmpw(tmp1, tmp2);
5028   } else {
5029     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5030     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5031     cmp(tmp1, tmp2);
5032   }
5033 }
5034 
5035 void MacroAssembler::store_klass(Register dst, Register src) {
5036   // FIXME: Should this be a store release?  concurrent gcs assumes
5037   // klass length is valid if klass field is not null.
5038   assert(!UseCompactObjectHeaders, "not with compact headers");
5039   if (UseCompressedClassPointers) {
5040     encode_klass_not_null(src);
5041     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5042   } else {
5043     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5044   }
5045 }
5046 
5047 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5048   assert(!UseCompactObjectHeaders, "not with compact headers");
5049   if (UseCompressedClassPointers) {
5050     // Store to klass gap in destination
5051     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5052   }
5053 }
5054 
5055 // Algorithm must match CompressedOops::encode.
5056 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5057 #ifdef ASSERT
5058   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5059 #endif
5060   verify_oop_msg(s, "broken oop in encode_heap_oop");
5061   if (CompressedOops::base() == nullptr) {
5062     if (CompressedOops::shift() != 0) {
5063       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5064       lsr(d, s, LogMinObjAlignmentInBytes);
5065     } else {
5066       mov(d, s);
5067     }
5068   } else {
5069     subs(d, s, rheapbase);
5070     csel(d, d, zr, Assembler::HS);
5071     lsr(d, d, LogMinObjAlignmentInBytes);
5072 
5073     /*  Old algorithm: is this any worse?
5074     Label nonnull;
5075     cbnz(r, nonnull);
5076     sub(r, r, rheapbase);
5077     bind(nonnull);
5078     lsr(r, r, LogMinObjAlignmentInBytes);
5079     */
5080   }
5081 }
5082 
5083 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5084 #ifdef ASSERT
5085   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5086   if (CheckCompressedOops) {
5087     Label ok;
5088     cbnz(r, ok);
5089     stop("null oop passed to encode_heap_oop_not_null");
5090     bind(ok);
5091   }
5092 #endif
5093   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5094   if (CompressedOops::base() != nullptr) {
5095     sub(r, r, rheapbase);
5096   }
5097   if (CompressedOops::shift() != 0) {
5098     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5099     lsr(r, r, LogMinObjAlignmentInBytes);
5100   }
5101 }
5102 
5103 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5104 #ifdef ASSERT
5105   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5106   if (CheckCompressedOops) {
5107     Label ok;
5108     cbnz(src, ok);
5109     stop("null oop passed to encode_heap_oop_not_null2");
5110     bind(ok);
5111   }
5112 #endif
5113   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5114 
5115   Register data = src;
5116   if (CompressedOops::base() != nullptr) {
5117     sub(dst, src, rheapbase);
5118     data = dst;
5119   }
5120   if (CompressedOops::shift() != 0) {
5121     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5122     lsr(dst, data, LogMinObjAlignmentInBytes);
5123     data = dst;
5124   }
5125   if (data == src)
5126     mov(dst, src);
5127 }
5128 
5129 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5130 #ifdef ASSERT
5131   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5132 #endif
5133   if (CompressedOops::base() == nullptr) {
5134     if (CompressedOops::shift() != 0) {
5135       lsl(d, s, CompressedOops::shift());
5136     } else if (d != s) {
5137       mov(d, s);
5138     }
5139   } else {
5140     Label done;
5141     if (d != s)
5142       mov(d, s);
5143     cbz(s, done);
5144     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5145     bind(done);
5146   }
5147   verify_oop_msg(d, "broken oop in decode_heap_oop");
5148 }
5149 
5150 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5151   assert (UseCompressedOops, "should only be used for compressed headers");
5152   assert (Universe::heap() != nullptr, "java heap should be initialized");
5153   // Cannot assert, unverified entry point counts instructions (see .ad file)
5154   // vtableStubs also counts instructions in pd_code_size_limit.
5155   // Also do not verify_oop as this is called by verify_oop.
5156   if (CompressedOops::shift() != 0) {
5157     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5158     if (CompressedOops::base() != nullptr) {
5159       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5160     } else {
5161       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5162     }
5163   } else {
5164     assert (CompressedOops::base() == nullptr, "sanity");
5165   }
5166 }
5167 
5168 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5169   assert (UseCompressedOops, "should only be used for compressed headers");
5170   assert (Universe::heap() != nullptr, "java heap should be initialized");
5171   // Cannot assert, unverified entry point counts instructions (see .ad file)
5172   // vtableStubs also counts instructions in pd_code_size_limit.
5173   // Also do not verify_oop as this is called by verify_oop.
5174   if (CompressedOops::shift() != 0) {
5175     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5176     if (CompressedOops::base() != nullptr) {
5177       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5178     } else {
5179       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5180     }
5181   } else {
5182     assert (CompressedOops::base() == nullptr, "sanity");
5183     if (dst != src) {
5184       mov(dst, src);
5185     }
5186   }
5187 }
5188 
5189 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5190 
5191 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5192   assert(Metaspace::initialized(), "metaspace not initialized yet");
5193   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5194   return _klass_decode_mode;
5195 }
5196 
5197 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5198   assert(UseCompressedClassPointers, "not using compressed class pointers");
5199 
5200   // KlassDecodeMode shouldn't be set already.
5201   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5202 
5203   if (base == nullptr) {
5204     return KlassDecodeZero;
5205   }
5206 
5207   if (operand_valid_for_logical_immediate(
5208         /*is32*/false, (uint64_t)base)) {
5209     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5210     if (((uint64_t)base & range_mask) == 0) {
5211       return KlassDecodeXor;
5212     }
5213   }
5214 
5215   const uint64_t shifted_base =
5216     (uint64_t)base >> shift;
5217   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5218     return KlassDecodeMovk;
5219   }
5220 
5221   // No valid encoding.
5222   return KlassDecodeNone;
5223 }
5224 
5225 // Check if one of the above decoding modes will work for given base, shift and range.
5226 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5227   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5228 }
5229 
5230 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5231   _klass_decode_mode = klass_decode_mode(base, shift, range);
5232   return _klass_decode_mode != KlassDecodeNone;
5233 }
5234 
5235 static Register pick_different_tmp(Register dst, Register src) {
5236   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5237   return *tmps.begin();
5238 }
5239 
5240 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5241   // we have to load the klass base from the AOT constants area but
5242   // not the shift because it is not allowed to change
5243   int shift = CompressedKlassPointers::shift();
5244   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5245   if (dst != src) {
5246     // we can load the base into dst, subtract it formthe src and shift down
5247     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5248     ldr(dst, dst);
5249     sub(dst, src, dst);
5250     lsr(dst, dst, shift);
5251   } else {
5252     // we need an extra register in order to load the coop base
5253     Register tmp = pick_different_tmp(dst, src);
5254     RegSet regs = RegSet::of(tmp);
5255     push(regs, sp);
5256     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5257     ldr(tmp, tmp);
5258     sub(dst, src, tmp);
5259     lsr(dst, dst, shift);
5260     pop(regs, sp);
5261   }
5262 }
5263 
5264 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5265   if (CompressedKlassPointers::base() != nullptr && AOTCodeCache::is_on_for_dump()) {
5266     encode_klass_not_null_for_aot(dst, src);
5267     return;
5268   }
5269 
5270   switch (klass_decode_mode()) {
5271   case KlassDecodeZero:
5272     if (CompressedKlassPointers::shift() != 0) {
5273       lsr(dst, src, CompressedKlassPointers::shift());
5274     } else {
5275       if (dst != src) mov(dst, src);
5276     }
5277     break;
5278 
5279   case KlassDecodeXor:
5280     if (CompressedKlassPointers::shift() != 0) {
5281       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5282       lsr(dst, dst, CompressedKlassPointers::shift());
5283     } else {
5284       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5285     }
5286     break;
5287 
5288   case KlassDecodeMovk:
5289     if (CompressedKlassPointers::shift() != 0) {
5290       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5291     } else {
5292       movw(dst, src);
5293     }
5294     break;
5295 
5296   case KlassDecodeNone:
5297     ShouldNotReachHere();
5298     break;
5299   }
5300 }
5301 
5302 void MacroAssembler::encode_klass_not_null(Register r) {
5303   encode_klass_not_null(r, r);
5304 }
5305 
5306 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5307   // we have to load the klass base from the AOT constants area but
5308   // not the shift because it is not allowed to change
5309   int shift = CompressedKlassPointers::shift();
5310   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5311   if (dst != src) {
5312     // we can load the base into dst then add the offset with a suitable shift
5313     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5314     ldr(dst, dst);
5315     add(dst, dst, src, LSL,  shift);
5316   } else {
5317     // we need an extra register in order to load the coop base
5318     Register tmp = pick_different_tmp(dst, src);
5319     RegSet regs = RegSet::of(tmp);
5320     push(regs, sp);
5321     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5322     ldr(tmp, tmp);
5323     add(dst, tmp,  src, LSL,  shift);
5324     pop(regs, sp);
5325   }
5326 }
5327 
5328 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5329   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5330 
5331   if (CompressedKlassPointers::base() != nullptr && AOTCodeCache::is_on_for_dump()) {
5332     decode_klass_not_null_for_aot(dst, src);
5333     return;
5334   }
5335 
5336   switch (klass_decode_mode()) {
5337   case KlassDecodeZero:
5338     if (CompressedKlassPointers::shift() != 0) {
5339       lsl(dst, src, CompressedKlassPointers::shift());
5340     } else {
5341       if (dst != src) mov(dst, src);
5342     }
5343     break;
5344 
5345   case KlassDecodeXor:
5346     if (CompressedKlassPointers::shift() != 0) {
5347       lsl(dst, src, CompressedKlassPointers::shift());
5348       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5349     } else {
5350       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5351     }
5352     break;
5353 
5354   case KlassDecodeMovk: {
5355     const uint64_t shifted_base =
5356       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5357 
5358     if (dst != src) movw(dst, src);
5359     movk(dst, shifted_base >> 32, 32);
5360 
5361     if (CompressedKlassPointers::shift() != 0) {
5362       lsl(dst, dst, CompressedKlassPointers::shift());
5363     }
5364 
5365     break;
5366   }
5367 
5368   case KlassDecodeNone:
5369     ShouldNotReachHere();
5370     break;
5371   }
5372 }
5373 
5374 void  MacroAssembler::decode_klass_not_null(Register r) {
5375   decode_klass_not_null(r, r);
5376 }
5377 
5378 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5379 #ifdef ASSERT
5380   {
5381     ThreadInVMfromUnknown tiv;
5382     assert (UseCompressedOops, "should only be used for compressed oops");
5383     assert (Universe::heap() != nullptr, "java heap should be initialized");
5384     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5385     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5386   }
5387 #endif
5388   int oop_index = oop_recorder()->find_index(obj);
5389   InstructionMark im(this);
5390   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5391   code_section()->relocate(inst_mark(), rspec);
5392   movz(dst, 0xDEAD, 16);
5393   movk(dst, 0xBEEF);
5394 }
5395 
5396 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5397   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5398   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5399   int index = oop_recorder()->find_index(k);
5400   assert(! Universe::heap()->is_in(k), "should not be an oop");
5401 
5402   InstructionMark im(this);
5403   RelocationHolder rspec = metadata_Relocation::spec(index);
5404   code_section()->relocate(inst_mark(), rspec);
5405   narrowKlass nk = CompressedKlassPointers::encode(k);
5406   movz(dst, (nk >> 16), 16);
5407   movk(dst, nk & 0xffff);
5408 }
5409 
5410 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5411                                     Register dst, Address src,
5412                                     Register tmp1, Register tmp2) {
5413   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5414   decorators = AccessInternal::decorator_fixup(decorators, type);
5415   bool as_raw = (decorators & AS_RAW) != 0;
5416   if (as_raw) {
5417     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5418   } else {
5419     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5420   }
5421 }
5422 
5423 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5424                                      Address dst, Register val,
5425                                      Register tmp1, Register tmp2, Register tmp3) {
5426   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5427   decorators = AccessInternal::decorator_fixup(decorators, type);
5428   bool as_raw = (decorators & AS_RAW) != 0;
5429   if (as_raw) {
5430     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5431   } else {
5432     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5433   }
5434 }
5435 
5436 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5437                                    Register tmp2, DecoratorSet decorators) {
5438   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5439 }
5440 
5441 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5442                                             Register tmp2, DecoratorSet decorators) {
5443   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5444 }
5445 
5446 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5447                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5448   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5449 }
5450 
5451 // Used for storing nulls.
5452 void MacroAssembler::store_heap_oop_null(Address dst) {
5453   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5454 }
5455 
5456 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5457   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5458   int index = oop_recorder()->allocate_metadata_index(obj);
5459   RelocationHolder rspec = metadata_Relocation::spec(index);
5460   return Address((address)obj, rspec);
5461 }
5462 
5463 // Move an oop into a register.
5464 void MacroAssembler::movoop(Register dst, jobject obj) {
5465   int oop_index;
5466   if (obj == nullptr) {
5467     oop_index = oop_recorder()->allocate_oop_index(obj);
5468   } else {
5469 #ifdef ASSERT
5470     {
5471       ThreadInVMfromUnknown tiv;
5472       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5473     }
5474 #endif
5475     oop_index = oop_recorder()->find_index(obj);
5476   }
5477   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5478 
5479   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5480     mov(dst, Address((address)obj, rspec));
5481   } else {
5482     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5483     ldr(dst, Address(dummy, rspec));
5484   }
5485 }
5486 
5487 // Move a metadata address into a register.
5488 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5489   int oop_index;
5490   if (obj == nullptr) {
5491     oop_index = oop_recorder()->allocate_metadata_index(obj);
5492   } else {
5493     oop_index = oop_recorder()->find_index(obj);
5494   }
5495   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5496   mov(dst, Address((address)obj, rspec));
5497 }
5498 
5499 Address MacroAssembler::constant_oop_address(jobject obj) {
5500 #ifdef ASSERT
5501   {
5502     ThreadInVMfromUnknown tiv;
5503     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5504     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5505   }
5506 #endif
5507   int oop_index = oop_recorder()->find_index(obj);
5508   return Address((address)obj, oop_Relocation::spec(oop_index));
5509 }
5510 
5511 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5512 void MacroAssembler::tlab_allocate(Register obj,
5513                                    Register var_size_in_bytes,
5514                                    int con_size_in_bytes,
5515                                    Register t1,
5516                                    Register t2,
5517                                    Label& slow_case) {
5518   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5519   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5520 }
5521 
5522 void MacroAssembler::verify_tlab() {
5523 #ifdef ASSERT
5524   if (UseTLAB && VerifyOops) {
5525     Label next, ok;
5526 
5527     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5528 
5529     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5530     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5531     cmp(rscratch2, rscratch1);
5532     br(Assembler::HS, next);
5533     STOP("assert(top >= start)");
5534     should_not_reach_here();
5535 
5536     bind(next);
5537     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5538     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5539     cmp(rscratch2, rscratch1);
5540     br(Assembler::HS, ok);
5541     STOP("assert(top <= end)");
5542     should_not_reach_here();
5543 
5544     bind(ok);
5545     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5546   }
5547 #endif
5548 }
5549 
5550 // Writes to stack successive pages until offset reached to check for
5551 // stack overflow + shadow pages.  This clobbers tmp.
5552 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5553   assert_different_registers(tmp, size, rscratch1);
5554   mov(tmp, sp);
5555   // Bang stack for total size given plus shadow page size.
5556   // Bang one page at a time because large size can bang beyond yellow and
5557   // red zones.
5558   Label loop;
5559   mov(rscratch1, (int)os::vm_page_size());
5560   bind(loop);
5561   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5562   subsw(size, size, rscratch1);
5563   str(size, Address(tmp));
5564   br(Assembler::GT, loop);
5565 
5566   // Bang down shadow pages too.
5567   // At this point, (tmp-0) is the last address touched, so don't
5568   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5569   // was post-decremented.)  Skip this address by starting at i=1, and
5570   // touch a few more pages below.  N.B.  It is important to touch all
5571   // the way down to and including i=StackShadowPages.
5572   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5573     // this could be any sized move but this is can be a debugging crumb
5574     // so the bigger the better.
5575     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5576     str(size, Address(tmp));
5577   }
5578 }
5579 
5580 // Move the address of the polling page into dest.
5581 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5582   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5583 }
5584 
5585 // Read the polling page.  The address of the polling page must
5586 // already be in r.
5587 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5588   address mark;
5589   {
5590     InstructionMark im(this);
5591     code_section()->relocate(inst_mark(), rtype);
5592     ldrw(zr, Address(r, 0));
5593     mark = inst_mark();
5594   }
5595   verify_cross_modify_fence_not_required();
5596   return mark;
5597 }
5598 
5599 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5600   relocInfo::relocType rtype = dest.rspec().reloc()->type();
5601   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5602   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5603   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5604   int64_t offset_low = dest_page - low_page;
5605   int64_t offset_high = dest_page - high_page;
5606 
5607   assert(is_valid_AArch64_address(dest.target()), "bad address");
5608   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5609 
5610   InstructionMark im(this);
5611   code_section()->relocate(inst_mark(), dest.rspec());
5612   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5613   // the code cache so that if it is relocated we know it will still reach
5614   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5615     _adrp(reg1, dest.target());
5616   } else {
5617     uint64_t target = (uint64_t)dest.target();
5618     uint64_t adrp_target
5619       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5620 
5621     _adrp(reg1, (address)adrp_target);
5622     movk(reg1, target >> 32, 32);
5623   }
5624   byte_offset = (uint64_t)dest.target() & 0xfff;
5625 }
5626 
5627 void MacroAssembler::load_byte_map_base(Register reg) {
5628 #if INCLUDE_CDS
5629   if (AOTCodeCache::is_on_for_dump()) {
5630     address byte_map_base_adr = AOTRuntimeConstants::card_table_address();
5631     lea(reg, ExternalAddress(byte_map_base_adr));
5632     ldr(reg, Address(reg));
5633     return;
5634   }
5635 #endif
5636   CardTable::CardValue* byte_map_base =
5637     ((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base();
5638 
5639   // Strictly speaking the byte_map_base isn't an address at all, and it might
5640   // even be negative. It is thus materialised as a constant.
5641   mov(reg, (uint64_t)byte_map_base);
5642 }
5643 
5644 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5645 #if INCLUDE_CDS
5646   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5647   if (AOTCodeCache::is_on_for_dump()) {
5648     // all aotrc field addresses should be registered in the AOTCodeCache address table
5649     lea(reg, ExternalAddress(a));
5650   } else {
5651     mov(reg, (uint64_t)a);
5652   }
5653 #else
5654   ShouldNotReachHere();
5655 #endif
5656 }
5657 
5658 void MacroAssembler::build_frame(int framesize) {
5659   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5660   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5661   protect_return_address();
5662   if (framesize < ((1 << 9) + 2 * wordSize)) {
5663     sub(sp, sp, framesize);
5664     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5665     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5666   } else {
5667     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5668     if (PreserveFramePointer) mov(rfp, sp);
5669     if (framesize < ((1 << 12) + 2 * wordSize))
5670       sub(sp, sp, framesize - 2 * wordSize);
5671     else {
5672       mov(rscratch1, framesize - 2 * wordSize);
5673       sub(sp, sp, rscratch1);
5674     }
5675   }
5676   verify_cross_modify_fence_not_required();
5677 }
5678 
5679 void MacroAssembler::remove_frame(int framesize) {
5680   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5681   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5682   if (framesize < ((1 << 9) + 2 * wordSize)) {
5683     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5684     add(sp, sp, framesize);
5685   } else {
5686     if (framesize < ((1 << 12) + 2 * wordSize))
5687       add(sp, sp, framesize - 2 * wordSize);
5688     else {
5689       mov(rscratch1, framesize - 2 * wordSize);
5690       add(sp, sp, rscratch1);
5691     }
5692     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5693   }
5694   authenticate_return_address();
5695 }
5696 
5697 
5698 // This method counts leading positive bytes (highest bit not set) in provided byte array
5699 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5700     // Simple and most common case of aligned small array which is not at the
5701     // end of memory page is placed here. All other cases are in stub.
5702     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5703     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5704     assert_different_registers(ary1, len, result);
5705 
5706     mov(result, len);
5707     cmpw(len, 0);
5708     br(LE, DONE);
5709     cmpw(len, 4 * wordSize);
5710     br(GE, STUB_LONG); // size > 32 then go to stub
5711 
5712     int shift = 64 - exact_log2(os::vm_page_size());
5713     lsl(rscratch1, ary1, shift);
5714     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5715     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5716     br(CS, STUB); // at the end of page then go to stub
5717     subs(len, len, wordSize);
5718     br(LT, END);
5719 
5720   BIND(LOOP);
5721     ldr(rscratch1, Address(post(ary1, wordSize)));
5722     tst(rscratch1, UPPER_BIT_MASK);
5723     br(NE, SET_RESULT);
5724     subs(len, len, wordSize);
5725     br(GE, LOOP);
5726     cmpw(len, -wordSize);
5727     br(EQ, DONE);
5728 
5729   BIND(END);
5730     ldr(rscratch1, Address(ary1));
5731     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5732     lslv(rscratch1, rscratch1, rscratch2);
5733     tst(rscratch1, UPPER_BIT_MASK);
5734     br(NE, SET_RESULT);
5735     b(DONE);
5736 
5737   BIND(STUB);
5738     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5739     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5740     address tpc1 = trampoline_call(count_pos);
5741     if (tpc1 == nullptr) {
5742       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5743       postcond(pc() == badAddress);
5744       return nullptr;
5745     }
5746     b(DONE);
5747 
5748   BIND(STUB_LONG);
5749     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5750     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5751     address tpc2 = trampoline_call(count_pos_long);
5752     if (tpc2 == nullptr) {
5753       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5754       postcond(pc() == badAddress);
5755       return nullptr;
5756     }
5757     b(DONE);
5758 
5759   BIND(SET_RESULT);
5760 
5761     add(len, len, wordSize);
5762     sub(result, result, len);
5763 
5764   BIND(DONE);
5765   postcond(pc() != badAddress);
5766   return pc();
5767 }
5768 
5769 // Clobbers: rscratch1, rscratch2, rflags
5770 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5771 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5772                                       Register tmp4, Register tmp5, Register result,
5773                                       Register cnt1, int elem_size) {
5774   Label DONE, SAME;
5775   Register tmp1 = rscratch1;
5776   Register tmp2 = rscratch2;
5777   int elem_per_word = wordSize/elem_size;
5778   int log_elem_size = exact_log2(elem_size);
5779   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5780   int length_offset = arrayOopDesc::length_offset_in_bytes();
5781   int base_offset
5782     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5783   // When the length offset is not aligned to 8 bytes,
5784   // then we align it down. This is valid because the new
5785   // offset will always be the klass which is the same
5786   // for type arrays.
5787   int start_offset = align_down(length_offset, BytesPerWord);
5788   int extra_length = base_offset - start_offset;
5789   assert(start_offset == length_offset || start_offset == klass_offset,
5790          "start offset must be 8-byte-aligned or be the klass offset");
5791   assert(base_offset != start_offset, "must include the length field");
5792   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5793   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5794 
5795   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5796   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5797 
5798 #ifndef PRODUCT
5799   {
5800     const char kind = (elem_size == 2) ? 'U' : 'L';
5801     char comment[64];
5802     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5803     BLOCK_COMMENT(comment);
5804   }
5805 #endif
5806 
5807   // if (a1 == a2)
5808   //     return true;
5809   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5810   br(EQ, SAME);
5811 
5812   if (UseSimpleArrayEquals) {
5813     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5814     // if (a1 == nullptr || a2 == nullptr)
5815     //     return false;
5816     // a1 & a2 == 0 means (some-pointer is null) or
5817     // (very-rare-or-even-probably-impossible-pointer-values)
5818     // so, we can save one branch in most cases
5819     tst(a1, a2);
5820     mov(result, false);
5821     br(EQ, A_MIGHT_BE_NULL);
5822     // if (a1.length != a2.length)
5823     //      return false;
5824     bind(A_IS_NOT_NULL);
5825     ldrw(cnt1, Address(a1, length_offset));
5826     // Increase loop counter by diff between base- and actual start-offset.
5827     addw(cnt1, cnt1, extra_length);
5828     lea(a1, Address(a1, start_offset));
5829     lea(a2, Address(a2, start_offset));
5830     // Check for short strings, i.e. smaller than wordSize.
5831     subs(cnt1, cnt1, elem_per_word);
5832     br(Assembler::LT, SHORT);
5833     // Main 8 byte comparison loop.
5834     bind(NEXT_WORD); {
5835       ldr(tmp1, Address(post(a1, wordSize)));
5836       ldr(tmp2, Address(post(a2, wordSize)));
5837       subs(cnt1, cnt1, elem_per_word);
5838       eor(tmp5, tmp1, tmp2);
5839       cbnz(tmp5, DONE);
5840     } br(GT, NEXT_WORD);
5841     // Last longword.  In the case where length == 4 we compare the
5842     // same longword twice, but that's still faster than another
5843     // conditional branch.
5844     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5845     // length == 4.
5846     if (log_elem_size > 0)
5847       lsl(cnt1, cnt1, log_elem_size);
5848     ldr(tmp3, Address(a1, cnt1));
5849     ldr(tmp4, Address(a2, cnt1));
5850     eor(tmp5, tmp3, tmp4);
5851     cbnz(tmp5, DONE);
5852     b(SAME);
5853     bind(A_MIGHT_BE_NULL);
5854     // in case both a1 and a2 are not-null, proceed with loads
5855     cbz(a1, DONE);
5856     cbz(a2, DONE);
5857     b(A_IS_NOT_NULL);
5858     bind(SHORT);
5859 
5860     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
5861     {
5862       ldrw(tmp1, Address(post(a1, 4)));
5863       ldrw(tmp2, Address(post(a2, 4)));
5864       eorw(tmp5, tmp1, tmp2);
5865       cbnzw(tmp5, DONE);
5866     }
5867     bind(TAIL03);
5868     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
5869     {
5870       ldrh(tmp3, Address(post(a1, 2)));
5871       ldrh(tmp4, Address(post(a2, 2)));
5872       eorw(tmp5, tmp3, tmp4);
5873       cbnzw(tmp5, DONE);
5874     }
5875     bind(TAIL01);
5876     if (elem_size == 1) { // Only needed when comparing byte arrays.
5877       tbz(cnt1, 0, SAME); // 0-1 bytes left.
5878       {
5879         ldrb(tmp1, a1);
5880         ldrb(tmp2, a2);
5881         eorw(tmp5, tmp1, tmp2);
5882         cbnzw(tmp5, DONE);
5883       }
5884     }
5885   } else {
5886     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
5887         CSET_EQ, LAST_CHECK;
5888     mov(result, false);
5889     cbz(a1, DONE);
5890     ldrw(cnt1, Address(a1, length_offset));
5891     cbz(a2, DONE);
5892     // Increase loop counter by diff between base- and actual start-offset.
5893     addw(cnt1, cnt1, extra_length);
5894 
5895     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
5896     // faster to perform another branch before comparing a1 and a2
5897     cmp(cnt1, (u1)elem_per_word);
5898     br(LE, SHORT); // short or same
5899     ldr(tmp3, Address(pre(a1, start_offset)));
5900     subs(zr, cnt1, stubBytesThreshold);
5901     br(GE, STUB);
5902     ldr(tmp4, Address(pre(a2, start_offset)));
5903     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
5904 
5905     // Main 16 byte comparison loop with 2 exits
5906     bind(NEXT_DWORD); {
5907       ldr(tmp1, Address(pre(a1, wordSize)));
5908       ldr(tmp2, Address(pre(a2, wordSize)));
5909       subs(cnt1, cnt1, 2 * elem_per_word);
5910       br(LE, TAIL);
5911       eor(tmp4, tmp3, tmp4);
5912       cbnz(tmp4, DONE);
5913       ldr(tmp3, Address(pre(a1, wordSize)));
5914       ldr(tmp4, Address(pre(a2, wordSize)));
5915       cmp(cnt1, (u1)elem_per_word);
5916       br(LE, TAIL2);
5917       cmp(tmp1, tmp2);
5918     } br(EQ, NEXT_DWORD);
5919     b(DONE);
5920 
5921     bind(TAIL);
5922     eor(tmp4, tmp3, tmp4);
5923     eor(tmp2, tmp1, tmp2);
5924     lslv(tmp2, tmp2, tmp5);
5925     orr(tmp5, tmp4, tmp2);
5926     cmp(tmp5, zr);
5927     b(CSET_EQ);
5928 
5929     bind(TAIL2);
5930     eor(tmp2, tmp1, tmp2);
5931     cbnz(tmp2, DONE);
5932     b(LAST_CHECK);
5933 
5934     bind(STUB);
5935     ldr(tmp4, Address(pre(a2, start_offset)));
5936     if (elem_size == 2) { // convert to byte counter
5937       lsl(cnt1, cnt1, 1);
5938     }
5939     eor(tmp5, tmp3, tmp4);
5940     cbnz(tmp5, DONE);
5941     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
5942     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
5943     address tpc = trampoline_call(stub);
5944     if (tpc == nullptr) {
5945       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
5946       postcond(pc() == badAddress);
5947       return nullptr;
5948     }
5949     b(DONE);
5950 
5951     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
5952     // so, if a2 == null => return false(0), else return true, so we can return a2
5953     mov(result, a2);
5954     b(DONE);
5955     bind(SHORT);
5956     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
5957     ldr(tmp3, Address(a1, start_offset));
5958     ldr(tmp4, Address(a2, start_offset));
5959     bind(LAST_CHECK);
5960     eor(tmp4, tmp3, tmp4);
5961     lslv(tmp5, tmp4, tmp5);
5962     cmp(tmp5, zr);
5963     bind(CSET_EQ);
5964     cset(result, EQ);
5965     b(DONE);
5966   }
5967 
5968   bind(SAME);
5969   mov(result, true);
5970   // That's it.
5971   bind(DONE);
5972 
5973   BLOCK_COMMENT("} array_equals");
5974   postcond(pc() != badAddress);
5975   return pc();
5976 }
5977 
5978 // Compare Strings
5979 
5980 // For Strings we're passed the address of the first characters in a1
5981 // and a2 and the length in cnt1.
5982 // There are two implementations.  For arrays >= 8 bytes, all
5983 // comparisons (including the final one, which may overlap) are
5984 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
5985 // halfword, then a short, and then a byte.
5986 
5987 void MacroAssembler::string_equals(Register a1, Register a2,
5988                                    Register result, Register cnt1)
5989 {
5990   Label SAME, DONE, SHORT, NEXT_WORD;
5991   Register tmp1 = rscratch1;
5992   Register tmp2 = rscratch2;
5993   Register cnt2 = tmp2;  // cnt2 only used in array length compare
5994 
5995   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5996 
5997 #ifndef PRODUCT
5998   {
5999     char comment[64];
6000     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6001     BLOCK_COMMENT(comment);
6002   }
6003 #endif
6004 
6005   mov(result, false);
6006 
6007   // Check for short strings, i.e. smaller than wordSize.
6008   subs(cnt1, cnt1, wordSize);
6009   br(Assembler::LT, SHORT);
6010   // Main 8 byte comparison loop.
6011   bind(NEXT_WORD); {
6012     ldr(tmp1, Address(post(a1, wordSize)));
6013     ldr(tmp2, Address(post(a2, wordSize)));
6014     subs(cnt1, cnt1, wordSize);
6015     eor(tmp1, tmp1, tmp2);
6016     cbnz(tmp1, DONE);
6017   } br(GT, NEXT_WORD);
6018   // Last longword.  In the case where length == 4 we compare the
6019   // same longword twice, but that's still faster than another
6020   // conditional branch.
6021   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6022   // length == 4.
6023   ldr(tmp1, Address(a1, cnt1));
6024   ldr(tmp2, Address(a2, cnt1));
6025   eor(tmp2, tmp1, tmp2);
6026   cbnz(tmp2, DONE);
6027   b(SAME);
6028 
6029   bind(SHORT);
6030   Label TAIL03, TAIL01;
6031 
6032   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6033   {
6034     ldrw(tmp1, Address(post(a1, 4)));
6035     ldrw(tmp2, Address(post(a2, 4)));
6036     eorw(tmp1, tmp1, tmp2);
6037     cbnzw(tmp1, DONE);
6038   }
6039   bind(TAIL03);
6040   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6041   {
6042     ldrh(tmp1, Address(post(a1, 2)));
6043     ldrh(tmp2, Address(post(a2, 2)));
6044     eorw(tmp1, tmp1, tmp2);
6045     cbnzw(tmp1, DONE);
6046   }
6047   bind(TAIL01);
6048   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6049     {
6050     ldrb(tmp1, a1);
6051     ldrb(tmp2, a2);
6052     eorw(tmp1, tmp1, tmp2);
6053     cbnzw(tmp1, DONE);
6054   }
6055   // Arrays are equal.
6056   bind(SAME);
6057   mov(result, true);
6058 
6059   // That's it.
6060   bind(DONE);
6061   BLOCK_COMMENT("} string_equals");
6062 }
6063 
6064 
6065 // The size of the blocks erased by the zero_blocks stub.  We must
6066 // handle anything smaller than this ourselves in zero_words().
6067 const int MacroAssembler::zero_words_block_size = 8;
6068 
6069 // zero_words() is used by C2 ClearArray patterns and by
6070 // C1_MacroAssembler.  It is as small as possible, handling small word
6071 // counts locally and delegating anything larger to the zero_blocks
6072 // stub.  It is expanded many times in compiled code, so it is
6073 // important to keep it short.
6074 
6075 // ptr:   Address of a buffer to be zeroed.
6076 // cnt:   Count in HeapWords.
6077 //
6078 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6079 address MacroAssembler::zero_words(Register ptr, Register cnt)
6080 {
6081   assert(is_power_of_2(zero_words_block_size), "adjust this");
6082 
6083   BLOCK_COMMENT("zero_words {");
6084   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6085   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6086   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6087 
6088   subs(rscratch1, cnt, zero_words_block_size);
6089   Label around;
6090   br(LO, around);
6091   {
6092     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6093     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6094     // Make sure this is a C2 compilation. C1 allocates space only for
6095     // trampoline stubs generated by Call LIR ops, and in any case it
6096     // makes sense for a C1 compilation task to proceed as quickly as
6097     // possible.
6098     CompileTask* task;
6099     if (StubRoutines::aarch64::complete()
6100         && Thread::current()->is_Compiler_thread()
6101         && (task = ciEnv::current()->task())
6102         && is_c2_compile(task->comp_level())) {
6103       address tpc = trampoline_call(zero_blocks);
6104       if (tpc == nullptr) {
6105         DEBUG_ONLY(reset_labels(around));
6106         return nullptr;
6107       }
6108     } else {
6109       far_call(zero_blocks);
6110     }
6111   }
6112   bind(around);
6113 
6114   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6115   // for us.
6116   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6117     Label l;
6118     tbz(cnt, exact_log2(i), l);
6119     for (int j = 0; j < i; j += 2) {
6120       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6121     }
6122     bind(l);
6123   }
6124   {
6125     Label l;
6126     tbz(cnt, 0, l);
6127     str(zr, Address(ptr));
6128     bind(l);
6129   }
6130 
6131   BLOCK_COMMENT("} zero_words");
6132   return pc();
6133 }
6134 
6135 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6136 // cnt:          Immediate count in HeapWords.
6137 //
6138 // r10, r11, rscratch1, and rscratch2 are clobbered.
6139 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6140 {
6141   assert(wordSize <= BlockZeroingLowLimit,
6142             "increase BlockZeroingLowLimit");
6143   address result = nullptr;
6144   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6145 #ifndef PRODUCT
6146     {
6147       char buf[64];
6148       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6149       BLOCK_COMMENT(buf);
6150     }
6151 #endif
6152     if (cnt >= 16) {
6153       uint64_t loops = cnt/16;
6154       if (loops > 1) {
6155         mov(rscratch2, loops - 1);
6156       }
6157       {
6158         Label loop;
6159         bind(loop);
6160         for (int i = 0; i < 16; i += 2) {
6161           stp(zr, zr, Address(base, i * BytesPerWord));
6162         }
6163         add(base, base, 16 * BytesPerWord);
6164         if (loops > 1) {
6165           subs(rscratch2, rscratch2, 1);
6166           br(GE, loop);
6167         }
6168       }
6169     }
6170     cnt %= 16;
6171     int i = cnt & 1;  // store any odd word to start
6172     if (i) str(zr, Address(base));
6173     for (; i < (int)cnt; i += 2) {
6174       stp(zr, zr, Address(base, i * wordSize));
6175     }
6176     BLOCK_COMMENT("} zero_words");
6177     result = pc();
6178   } else {
6179     mov(r10, base); mov(r11, cnt);
6180     result = zero_words(r10, r11);
6181   }
6182   return result;
6183 }
6184 
6185 // Zero blocks of memory by using DC ZVA.
6186 //
6187 // Aligns the base address first sufficiently for DC ZVA, then uses
6188 // DC ZVA repeatedly for every full block.  cnt is the size to be
6189 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6190 // in cnt.
6191 //
6192 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6193 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6194 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6195   Register tmp = rscratch1;
6196   Register tmp2 = rscratch2;
6197   int zva_length = VM_Version::zva_length();
6198   Label initial_table_end, loop_zva;
6199   Label fini;
6200 
6201   // Base must be 16 byte aligned. If not just return and let caller handle it
6202   tst(base, 0x0f);
6203   br(Assembler::NE, fini);
6204   // Align base with ZVA length.
6205   neg(tmp, base);
6206   andr(tmp, tmp, zva_length - 1);
6207 
6208   // tmp: the number of bytes to be filled to align the base with ZVA length.
6209   add(base, base, tmp);
6210   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6211   adr(tmp2, initial_table_end);
6212   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6213   br(tmp2);
6214 
6215   for (int i = -zva_length + 16; i < 0; i += 16)
6216     stp(zr, zr, Address(base, i));
6217   bind(initial_table_end);
6218 
6219   sub(cnt, cnt, zva_length >> 3);
6220   bind(loop_zva);
6221   dc(Assembler::ZVA, base);
6222   subs(cnt, cnt, zva_length >> 3);
6223   add(base, base, zva_length);
6224   br(Assembler::GE, loop_zva);
6225   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6226   bind(fini);
6227 }
6228 
6229 // base:   Address of a buffer to be filled, 8 bytes aligned.
6230 // cnt:    Count in 8-byte unit.
6231 // value:  Value to be filled with.
6232 // base will point to the end of the buffer after filling.
6233 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6234 {
6235 //  Algorithm:
6236 //
6237 //    if (cnt == 0) {
6238 //      return;
6239 //    }
6240 //    if ((p & 8) != 0) {
6241 //      *p++ = v;
6242 //    }
6243 //
6244 //    scratch1 = cnt & 14;
6245 //    cnt -= scratch1;
6246 //    p += scratch1;
6247 //    switch (scratch1 / 2) {
6248 //      do {
6249 //        cnt -= 16;
6250 //          p[-16] = v;
6251 //          p[-15] = v;
6252 //        case 7:
6253 //          p[-14] = v;
6254 //          p[-13] = v;
6255 //        case 6:
6256 //          p[-12] = v;
6257 //          p[-11] = v;
6258 //          // ...
6259 //        case 1:
6260 //          p[-2] = v;
6261 //          p[-1] = v;
6262 //        case 0:
6263 //          p += 16;
6264 //      } while (cnt);
6265 //    }
6266 //    if ((cnt & 1) == 1) {
6267 //      *p++ = v;
6268 //    }
6269 
6270   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6271 
6272   Label fini, skip, entry, loop;
6273   const int unroll = 8; // Number of stp instructions we'll unroll
6274 
6275   cbz(cnt, fini);
6276   tbz(base, 3, skip);
6277   str(value, Address(post(base, 8)));
6278   sub(cnt, cnt, 1);
6279   bind(skip);
6280 
6281   andr(rscratch1, cnt, (unroll-1) * 2);
6282   sub(cnt, cnt, rscratch1);
6283   add(base, base, rscratch1, Assembler::LSL, 3);
6284   adr(rscratch2, entry);
6285   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6286   br(rscratch2);
6287 
6288   bind(loop);
6289   add(base, base, unroll * 16);
6290   for (int i = -unroll; i < 0; i++)
6291     stp(value, value, Address(base, i * 16));
6292   bind(entry);
6293   subs(cnt, cnt, unroll * 2);
6294   br(Assembler::GE, loop);
6295 
6296   tbz(cnt, 0, fini);
6297   str(value, Address(post(base, 8)));
6298   bind(fini);
6299 }
6300 
6301 // Intrinsic for
6302 //
6303 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6304 //   Encodes char[] to byte[] in ISO-8859-1
6305 //
6306 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6307 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6308 //
6309 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6310 //   Encodes char[] to byte[] in ASCII
6311 //
6312 // This version always returns the number of characters copied, and does not
6313 // clobber the 'len' register. A successful copy will complete with the post-
6314 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6315 // post-condition: 0 <= 'res' < 'len'.
6316 //
6317 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6318 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6319 //       beyond the acceptable, even though the footprint would be smaller.
6320 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6321 //       avoid additional bloat.
6322 //
6323 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6324 void MacroAssembler::encode_iso_array(Register src, Register dst,
6325                                       Register len, Register res, bool ascii,
6326                                       FloatRegister vtmp0, FloatRegister vtmp1,
6327                                       FloatRegister vtmp2, FloatRegister vtmp3,
6328                                       FloatRegister vtmp4, FloatRegister vtmp5)
6329 {
6330   Register cnt = res;
6331   Register max = rscratch1;
6332   Register chk = rscratch2;
6333 
6334   prfm(Address(src), PLDL1STRM);
6335   movw(cnt, len);
6336 
6337 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6338 
6339   Label LOOP_32, DONE_32, FAIL_32;
6340 
6341   BIND(LOOP_32);
6342   {
6343     cmpw(cnt, 32);
6344     br(LT, DONE_32);
6345     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6346     // Extract lower bytes.
6347     FloatRegister vlo0 = vtmp4;
6348     FloatRegister vlo1 = vtmp5;
6349     uzp1(vlo0, T16B, vtmp0, vtmp1);
6350     uzp1(vlo1, T16B, vtmp2, vtmp3);
6351     // Merge bits...
6352     orr(vtmp0, T16B, vtmp0, vtmp1);
6353     orr(vtmp2, T16B, vtmp2, vtmp3);
6354     // Extract merged upper bytes.
6355     FloatRegister vhix = vtmp0;
6356     uzp2(vhix, T16B, vtmp0, vtmp2);
6357     // ISO-check on hi-parts (all zero).
6358     //                          ASCII-check on lo-parts (no sign).
6359     FloatRegister vlox = vtmp1; // Merge lower bytes.
6360                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6361     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6362     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6363     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6364                                 ASCII(orr(chk, chk, max));
6365     cbnz(chk, FAIL_32);
6366     subw(cnt, cnt, 32);
6367     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6368     b(LOOP_32);
6369   }
6370   BIND(FAIL_32);
6371   sub(src, src, 64);
6372   BIND(DONE_32);
6373 
6374   Label LOOP_8, SKIP_8;
6375 
6376   BIND(LOOP_8);
6377   {
6378     cmpw(cnt, 8);
6379     br(LT, SKIP_8);
6380     FloatRegister vhi = vtmp0;
6381     FloatRegister vlo = vtmp1;
6382     ld1(vtmp3, T8H, src);
6383     uzp1(vlo, T16B, vtmp3, vtmp3);
6384     uzp2(vhi, T16B, vtmp3, vtmp3);
6385     // ISO-check on hi-parts (all zero).
6386     //                          ASCII-check on lo-parts (no sign).
6387                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6388     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6389                                 ASCII(umov(max, vtmp2, B, 0));
6390                                 ASCII(orr(chk, chk, max));
6391     cbnz(chk, SKIP_8);
6392 
6393     strd(vlo, Address(post(dst, 8)));
6394     subw(cnt, cnt, 8);
6395     add(src, src, 16);
6396     b(LOOP_8);
6397   }
6398   BIND(SKIP_8);
6399 
6400 #undef ASCII
6401 
6402   Label LOOP, DONE;
6403 
6404   cbz(cnt, DONE);
6405   BIND(LOOP);
6406   {
6407     Register chr = rscratch1;
6408     ldrh(chr, Address(post(src, 2)));
6409     tst(chr, ascii ? 0xff80 : 0xff00);
6410     br(NE, DONE);
6411     strb(chr, Address(post(dst, 1)));
6412     subs(cnt, cnt, 1);
6413     br(GT, LOOP);
6414   }
6415   BIND(DONE);
6416   // Return index where we stopped.
6417   subw(res, len, cnt);
6418 }
6419 
6420 // Inflate byte[] array to char[].
6421 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6422 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6423                                            FloatRegister vtmp1, FloatRegister vtmp2,
6424                                            FloatRegister vtmp3, Register tmp4) {
6425   Label big, done, after_init, to_stub;
6426 
6427   assert_different_registers(src, dst, len, tmp4, rscratch1);
6428 
6429   fmovd(vtmp1, 0.0);
6430   lsrw(tmp4, len, 3);
6431   bind(after_init);
6432   cbnzw(tmp4, big);
6433   // Short string: less than 8 bytes.
6434   {
6435     Label loop, tiny;
6436 
6437     cmpw(len, 4);
6438     br(LT, tiny);
6439     // Use SIMD to do 4 bytes.
6440     ldrs(vtmp2, post(src, 4));
6441     zip1(vtmp3, T8B, vtmp2, vtmp1);
6442     subw(len, len, 4);
6443     strd(vtmp3, post(dst, 8));
6444 
6445     cbzw(len, done);
6446 
6447     // Do the remaining bytes by steam.
6448     bind(loop);
6449     ldrb(tmp4, post(src, 1));
6450     strh(tmp4, post(dst, 2));
6451     subw(len, len, 1);
6452 
6453     bind(tiny);
6454     cbnz(len, loop);
6455 
6456     b(done);
6457   }
6458 
6459   if (SoftwarePrefetchHintDistance >= 0) {
6460     bind(to_stub);
6461       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6462       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6463       address tpc = trampoline_call(stub);
6464       if (tpc == nullptr) {
6465         DEBUG_ONLY(reset_labels(big, done));
6466         postcond(pc() == badAddress);
6467         return nullptr;
6468       }
6469       b(after_init);
6470   }
6471 
6472   // Unpack the bytes 8 at a time.
6473   bind(big);
6474   {
6475     Label loop, around, loop_last, loop_start;
6476 
6477     if (SoftwarePrefetchHintDistance >= 0) {
6478       const int large_loop_threshold = (64 + 16)/8;
6479       ldrd(vtmp2, post(src, 8));
6480       andw(len, len, 7);
6481       cmp(tmp4, (u1)large_loop_threshold);
6482       br(GE, to_stub);
6483       b(loop_start);
6484 
6485       bind(loop);
6486       ldrd(vtmp2, post(src, 8));
6487       bind(loop_start);
6488       subs(tmp4, tmp4, 1);
6489       br(EQ, loop_last);
6490       zip1(vtmp2, T16B, vtmp2, vtmp1);
6491       ldrd(vtmp3, post(src, 8));
6492       st1(vtmp2, T8H, post(dst, 16));
6493       subs(tmp4, tmp4, 1);
6494       zip1(vtmp3, T16B, vtmp3, vtmp1);
6495       st1(vtmp3, T8H, post(dst, 16));
6496       br(NE, loop);
6497       b(around);
6498       bind(loop_last);
6499       zip1(vtmp2, T16B, vtmp2, vtmp1);
6500       st1(vtmp2, T8H, post(dst, 16));
6501       bind(around);
6502       cbz(len, done);
6503     } else {
6504       andw(len, len, 7);
6505       bind(loop);
6506       ldrd(vtmp2, post(src, 8));
6507       sub(tmp4, tmp4, 1);
6508       zip1(vtmp3, T16B, vtmp2, vtmp1);
6509       st1(vtmp3, T8H, post(dst, 16));
6510       cbnz(tmp4, loop);
6511     }
6512   }
6513 
6514   // Do the tail of up to 8 bytes.
6515   add(src, src, len);
6516   ldrd(vtmp3, Address(src, -8));
6517   add(dst, dst, len, ext::uxtw, 1);
6518   zip1(vtmp3, T16B, vtmp3, vtmp1);
6519   strq(vtmp3, Address(dst, -16));
6520 
6521   bind(done);
6522   postcond(pc() != badAddress);
6523   return pc();
6524 }
6525 
6526 // Compress char[] array to byte[].
6527 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6528 // Return the array length if every element in array can be encoded,
6529 // otherwise, the index of first non-latin1 (> 0xff) character.
6530 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6531                                          Register res,
6532                                          FloatRegister tmp0, FloatRegister tmp1,
6533                                          FloatRegister tmp2, FloatRegister tmp3,
6534                                          FloatRegister tmp4, FloatRegister tmp5) {
6535   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6536 }
6537 
6538 // java.math.round(double a)
6539 // Returns the closest long to the argument, with ties rounding to
6540 // positive infinity.  This requires some fiddling for corner
6541 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6542 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6543                                        FloatRegister ftmp) {
6544   Label DONE;
6545   BLOCK_COMMENT("java_round_double: { ");
6546   fmovd(rscratch1, src);
6547   // Use RoundToNearestTiesAway unless src small and -ve.
6548   fcvtasd(dst, src);
6549   // Test if src >= 0 || abs(src) >= 0x1.0p52
6550   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6551   mov(rscratch2, julong_cast(0x1.0p52));
6552   cmp(rscratch1, rscratch2);
6553   br(HS, DONE); {
6554     // src < 0 && abs(src) < 0x1.0p52
6555     // src may have a fractional part, so add 0.5
6556     fmovd(ftmp, 0.5);
6557     faddd(ftmp, src, ftmp);
6558     // Convert double to jlong, use RoundTowardsNegative
6559     fcvtmsd(dst, ftmp);
6560   }
6561   bind(DONE);
6562   BLOCK_COMMENT("} java_round_double");
6563 }
6564 
6565 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6566                                       FloatRegister ftmp) {
6567   Label DONE;
6568   BLOCK_COMMENT("java_round_float: { ");
6569   fmovs(rscratch1, src);
6570   // Use RoundToNearestTiesAway unless src small and -ve.
6571   fcvtassw(dst, src);
6572   // Test if src >= 0 || abs(src) >= 0x1.0p23
6573   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6574   mov(rscratch2, jint_cast(0x1.0p23f));
6575   cmp(rscratch1, rscratch2);
6576   br(HS, DONE); {
6577     // src < 0 && |src| < 0x1.0p23
6578     // src may have a fractional part, so add 0.5
6579     fmovs(ftmp, 0.5f);
6580     fadds(ftmp, src, ftmp);
6581     // Convert float to jint, use RoundTowardsNegative
6582     fcvtmssw(dst, ftmp);
6583   }
6584   bind(DONE);
6585   BLOCK_COMMENT("} java_round_float");
6586 }
6587 
6588 // get_thread() can be called anywhere inside generated code so we
6589 // need to save whatever non-callee save context might get clobbered
6590 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6591 // the call setup code.
6592 //
6593 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6594 // On other systems, the helper is a usual C function.
6595 //
6596 void MacroAssembler::get_thread(Register dst) {
6597   RegSet saved_regs =
6598     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6599     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6600 
6601   protect_return_address();
6602   push(saved_regs, sp);
6603 
6604   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6605   blr(lr);
6606   if (dst != c_rarg0) {
6607     mov(dst, c_rarg0);
6608   }
6609 
6610   pop(saved_regs, sp);
6611   authenticate_return_address();
6612 }
6613 
6614 void MacroAssembler::cache_wb(Address line) {
6615   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6616   assert(line.index() == noreg, "index should be noreg");
6617   assert(line.offset() == 0, "offset should be 0");
6618   // would like to assert this
6619   // assert(line._ext.shift == 0, "shift should be zero");
6620   if (VM_Version::supports_dcpop()) {
6621     // writeback using clear virtual address to point of persistence
6622     dc(Assembler::CVAP, line.base());
6623   } else {
6624     // no need to generate anything as Unsafe.writebackMemory should
6625     // never invoke this stub
6626   }
6627 }
6628 
6629 void MacroAssembler::cache_wbsync(bool is_pre) {
6630   // we only need a barrier post sync
6631   if (!is_pre) {
6632     membar(Assembler::AnyAny);
6633   }
6634 }
6635 
6636 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6637   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6638     return;
6639   }
6640   // Make sure that native code does not change SVE vector length.
6641   Label verify_ok;
6642   movw(tmp, zr);
6643   sve_inc(tmp, B);
6644   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6645   br(EQ, verify_ok);
6646   stop("Error: SVE vector length has changed since jvm startup");
6647   bind(verify_ok);
6648 }
6649 
6650 void MacroAssembler::verify_ptrue() {
6651   Label verify_ok;
6652   if (!UseSVE) {
6653     return;
6654   }
6655   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6656   sve_dec(rscratch1, B);
6657   cbz(rscratch1, verify_ok);
6658   stop("Error: the preserved predicate register (p7) elements are not all true");
6659   bind(verify_ok);
6660 }
6661 
6662 void MacroAssembler::safepoint_isb() {
6663   isb();
6664 #ifndef PRODUCT
6665   if (VerifyCrossModifyFence) {
6666     // Clear the thread state.
6667     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6668   }
6669 #endif
6670 }
6671 
6672 #ifndef PRODUCT
6673 void MacroAssembler::verify_cross_modify_fence_not_required() {
6674   if (VerifyCrossModifyFence) {
6675     // Check if thread needs a cross modify fence.
6676     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6677     Label fence_not_required;
6678     cbz(rscratch1, fence_not_required);
6679     // If it does then fail.
6680     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6681     mov(c_rarg0, rthread);
6682     blr(rscratch1);
6683     bind(fence_not_required);
6684   }
6685 }
6686 #endif
6687 
6688 void MacroAssembler::spin_wait() {
6689   block_comment("spin_wait {");
6690   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6691     switch (VM_Version::spin_wait_desc().inst()) {
6692       case SpinWait::NOP:
6693         nop();
6694         break;
6695       case SpinWait::ISB:
6696         isb();
6697         break;
6698       case SpinWait::YIELD:
6699         yield();
6700         break;
6701       case SpinWait::SB:
6702         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6703         sb();
6704         break;
6705       default:
6706         ShouldNotReachHere();
6707     }
6708   }
6709   block_comment("}");
6710 }
6711 
6712 // Stack frame creation/removal
6713 
6714 void MacroAssembler::enter(bool strip_ret_addr) {
6715   if (strip_ret_addr) {
6716     // Addresses can only be signed once. If there are multiple nested frames being created
6717     // in the same function, then the return address needs stripping first.
6718     strip_return_address();
6719   }
6720   protect_return_address();
6721   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6722   mov(rfp, sp);
6723 }
6724 
6725 void MacroAssembler::leave() {
6726   mov(sp, rfp);
6727   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6728   authenticate_return_address();
6729 }
6730 
6731 // ROP Protection
6732 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6733 // destroying stack frames or whenever directly loading/storing the LR to memory.
6734 // If ROP protection is not set then these functions are no-ops.
6735 // For more details on PAC see pauth_aarch64.hpp.
6736 
6737 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6738 // Uses value zero as the modifier.
6739 //
6740 void MacroAssembler::protect_return_address() {
6741   if (VM_Version::use_rop_protection()) {
6742     check_return_address();
6743     paciaz();
6744   }
6745 }
6746 
6747 // Sign the return value in the given register. Use before updating the LR in the existing stack
6748 // frame for the current function.
6749 // Uses value zero as the modifier.
6750 //
6751 void MacroAssembler::protect_return_address(Register return_reg) {
6752   if (VM_Version::use_rop_protection()) {
6753     check_return_address(return_reg);
6754     paciza(return_reg);
6755   }
6756 }
6757 
6758 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6759 // Uses value zero as the modifier.
6760 //
6761 void MacroAssembler::authenticate_return_address() {
6762   if (VM_Version::use_rop_protection()) {
6763     autiaz();
6764     check_return_address();
6765   }
6766 }
6767 
6768 // Authenticate the return value in the given register. Use before updating the LR in the existing
6769 // stack frame for the current function.
6770 // Uses value zero as the modifier.
6771 //
6772 void MacroAssembler::authenticate_return_address(Register return_reg) {
6773   if (VM_Version::use_rop_protection()) {
6774     autiza(return_reg);
6775     check_return_address(return_reg);
6776   }
6777 }
6778 
6779 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6780 // there is no guaranteed way of authenticating the LR.
6781 //
6782 void MacroAssembler::strip_return_address() {
6783   if (VM_Version::use_rop_protection()) {
6784     xpaclri();
6785   }
6786 }
6787 
6788 #ifndef PRODUCT
6789 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6790 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6791 // it is difficult to debug back to the callee function.
6792 // This function simply loads from the address in the given register.
6793 // Use directly after authentication to catch authentication failures.
6794 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6795 //
6796 void MacroAssembler::check_return_address(Register return_reg) {
6797   if (VM_Version::use_rop_protection()) {
6798     ldr(zr, Address(return_reg));
6799   }
6800 }
6801 #endif
6802 
6803 // The java_calling_convention describes stack locations as ideal slots on
6804 // a frame with no abi restrictions. Since we must observe abi restrictions
6805 // (like the placement of the register window) the slots must be biased by
6806 // the following value.
6807 static int reg2offset_in(VMReg r) {
6808   // Account for saved rfp and lr
6809   // This should really be in_preserve_stack_slots
6810   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6811 }
6812 
6813 static int reg2offset_out(VMReg r) {
6814   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6815 }
6816 
6817 // On 64bit we will store integer like items to the stack as
6818 // 64bits items (AArch64 ABI) even though java would only store
6819 // 32bits for a parameter. On 32bit it will simply be 32bits
6820 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6821 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6822   if (src.first()->is_stack()) {
6823     if (dst.first()->is_stack()) {
6824       // stack to stack
6825       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6826       str(tmp, Address(sp, reg2offset_out(dst.first())));
6827     } else {
6828       // stack to reg
6829       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6830     }
6831   } else if (dst.first()->is_stack()) {
6832     // reg to stack
6833     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6834   } else {
6835     if (dst.first() != src.first()) {
6836       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6837     }
6838   }
6839 }
6840 
6841 // An oop arg. Must pass a handle not the oop itself
6842 void MacroAssembler::object_move(
6843                         OopMap* map,
6844                         int oop_handle_offset,
6845                         int framesize_in_slots,
6846                         VMRegPair src,
6847                         VMRegPair dst,
6848                         bool is_receiver,
6849                         int* receiver_offset) {
6850 
6851   // must pass a handle. First figure out the location we use as a handle
6852 
6853   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
6854 
6855   // See if oop is null if it is we need no handle
6856 
6857   if (src.first()->is_stack()) {
6858 
6859     // Oop is already on the stack as an argument
6860     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
6861     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
6862     if (is_receiver) {
6863       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
6864     }
6865 
6866     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
6867     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
6868     // conditionally move a null
6869     cmp(rscratch1, zr);
6870     csel(rHandle, zr, rHandle, Assembler::EQ);
6871   } else {
6872 
6873     // Oop is in an a register we must store it to the space we reserve
6874     // on the stack for oop_handles and pass a handle if oop is non-null
6875 
6876     const Register rOop = src.first()->as_Register();
6877     int oop_slot;
6878     if (rOop == j_rarg0)
6879       oop_slot = 0;
6880     else if (rOop == j_rarg1)
6881       oop_slot = 1;
6882     else if (rOop == j_rarg2)
6883       oop_slot = 2;
6884     else if (rOop == j_rarg3)
6885       oop_slot = 3;
6886     else if (rOop == j_rarg4)
6887       oop_slot = 4;
6888     else if (rOop == j_rarg5)
6889       oop_slot = 5;
6890     else if (rOop == j_rarg6)
6891       oop_slot = 6;
6892     else {
6893       assert(rOop == j_rarg7, "wrong register");
6894       oop_slot = 7;
6895     }
6896 
6897     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
6898     int offset = oop_slot*VMRegImpl::stack_slot_size;
6899 
6900     map->set_oop(VMRegImpl::stack2reg(oop_slot));
6901     // Store oop in handle area, may be null
6902     str(rOop, Address(sp, offset));
6903     if (is_receiver) {
6904       *receiver_offset = offset;
6905     }
6906 
6907     cmp(rOop, zr);
6908     lea(rHandle, Address(sp, offset));
6909     // conditionally move a null
6910     csel(rHandle, zr, rHandle, Assembler::EQ);
6911   }
6912 
6913   // If arg is on the stack then place it otherwise it is already in correct reg.
6914   if (dst.first()->is_stack()) {
6915     str(rHandle, Address(sp, reg2offset_out(dst.first())));
6916   }
6917 }
6918 
6919 // A float arg may have to do float reg int reg conversion
6920 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
6921  if (src.first()->is_stack()) {
6922     if (dst.first()->is_stack()) {
6923       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
6924       strw(tmp, Address(sp, reg2offset_out(dst.first())));
6925     } else {
6926       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
6927     }
6928   } else if (src.first() != dst.first()) {
6929     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
6930       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
6931     else
6932       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
6933   }
6934 }
6935 
6936 // A long move
6937 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
6938   if (src.first()->is_stack()) {
6939     if (dst.first()->is_stack()) {
6940       // stack to stack
6941       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6942       str(tmp, Address(sp, reg2offset_out(dst.first())));
6943     } else {
6944       // stack to reg
6945       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6946     }
6947   } else if (dst.first()->is_stack()) {
6948     // reg to stack
6949     // Do we really have to sign extend???
6950     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
6951     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6952   } else {
6953     if (dst.first() != src.first()) {
6954       mov(dst.first()->as_Register(), src.first()->as_Register());
6955     }
6956   }
6957 }
6958 
6959 
6960 // A double move
6961 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
6962  if (src.first()->is_stack()) {
6963     if (dst.first()->is_stack()) {
6964       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6965       str(tmp, Address(sp, reg2offset_out(dst.first())));
6966     } else {
6967       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
6968     }
6969   } else if (src.first() != dst.first()) {
6970     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
6971       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
6972     else
6973       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
6974   }
6975 }
6976 
6977 // Implements fast-locking.
6978 //
6979 //  - obj: the object to be locked
6980 //  - t1, t2, t3: temporary registers, will be destroyed
6981 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
6982 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
6983   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
6984 
6985   Label push;
6986   const Register top = t1;
6987   const Register mark = t2;
6988   const Register t = t3;
6989 
6990   // Preload the markWord. It is important that this is the first
6991   // instruction emitted as it is part of C1's null check semantics.
6992   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
6993 
6994   if (UseObjectMonitorTable) {
6995     // Clear cache in case fast locking succeeds or we need to take the slow-path.
6996     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
6997   }
6998 
6999   if (DiagnoseSyncOnValueBasedClasses != 0) {
7000     load_klass(t1, obj);
7001     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7002     tst(t1, KlassFlags::_misc_is_value_based_class);
7003     br(Assembler::NE, slow);
7004   }
7005 
7006   // Check if the lock-stack is full.
7007   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7008   cmpw(top, (unsigned)LockStack::end_offset());
7009   br(Assembler::GE, slow);
7010 
7011   // Check for recursion.
7012   subw(t, top, oopSize);
7013   ldr(t, Address(rthread, t));
7014   cmp(obj, t);
7015   br(Assembler::EQ, push);
7016 
7017   // Check header for monitor (0b10).
7018   tst(mark, markWord::monitor_value);
7019   br(Assembler::NE, slow);
7020 
7021   // Try to lock. Transition lock bits 0b01 => 0b00
7022   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7023   orr(mark, mark, markWord::unlocked_value);
7024   eor(t, mark, markWord::unlocked_value);
7025   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7026           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7027   br(Assembler::NE, slow);
7028 
7029   bind(push);
7030   // After successful lock, push object on lock-stack.
7031   str(obj, Address(rthread, top));
7032   addw(top, top, oopSize);
7033   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7034 }
7035 
7036 // Implements fast-unlocking.
7037 //
7038 // - obj: the object to be unlocked
7039 // - t1, t2, t3: temporary registers
7040 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7041 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7042   // cmpxchg clobbers rscratch1.
7043   assert_different_registers(obj, t1, t2, t3, rscratch1);
7044 
7045 #ifdef ASSERT
7046   {
7047     // Check for lock-stack underflow.
7048     Label stack_ok;
7049     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7050     cmpw(t1, (unsigned)LockStack::start_offset());
7051     br(Assembler::GE, stack_ok);
7052     STOP("Lock-stack underflow");
7053     bind(stack_ok);
7054   }
7055 #endif
7056 
7057   Label unlocked, push_and_slow;
7058   const Register top = t1;
7059   const Register mark = t2;
7060   const Register t = t3;
7061 
7062   // Check if obj is top of lock-stack.
7063   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7064   subw(top, top, oopSize);
7065   ldr(t, Address(rthread, top));
7066   cmp(obj, t);
7067   br(Assembler::NE, slow);
7068 
7069   // Pop lock-stack.
7070   DEBUG_ONLY(str(zr, Address(rthread, top));)
7071   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7072 
7073   // Check if recursive.
7074   subw(t, top, oopSize);
7075   ldr(t, Address(rthread, t));
7076   cmp(obj, t);
7077   br(Assembler::EQ, unlocked);
7078 
7079   // Not recursive. Check header for monitor (0b10).
7080   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7081   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7082 
7083 #ifdef ASSERT
7084   // Check header not unlocked (0b01).
7085   Label not_unlocked;
7086   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7087   stop("fast_unlock already unlocked");
7088   bind(not_unlocked);
7089 #endif
7090 
7091   // Try to unlock. Transition lock bits 0b00 => 0b01
7092   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7093   orr(t, mark, markWord::unlocked_value);
7094   cmpxchg(obj, mark, t, Assembler::xword,
7095           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7096   br(Assembler::EQ, unlocked);
7097 
7098   bind(push_and_slow);
7099   // Restore lock-stack and handle the unlock in runtime.
7100   DEBUG_ONLY(str(obj, Address(rthread, top));)
7101   addw(top, top, oopSize);
7102   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7103   b(slow);
7104 
7105   bind(unlocked);
7106 }