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 // Handle the receiver type profile update given the "recv" klass.
2139 //
2140 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2141 // If there are no matching or claimable receiver entries in RD, updates
2142 // the polymorphic counter.
2143 //
2144 // This code expected to run by either the interpreter or JIT-ed code, without
2145 // extra synchronization. For safety, receiver cells are claimed atomically, which
2146 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2147 // counter updates are not atomic.
2148 //
2149 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2150   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2151 
2152   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2153   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2154   int poly_count_offset      = in_bytes(CounterData::count_offset());
2155   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2156   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2157 
2158   // Adjust for MDP offsets.
2159   base_receiver_offset += mdp_offset;
2160   end_receiver_offset  += mdp_offset;
2161   poly_count_offset    += mdp_offset;
2162 
2163 #ifdef ASSERT
2164   // We are about to walk the MDO slots without asking for offsets.
2165   // Check that our math hits all the right spots.
2166   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2167     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2168     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2169     int offset = base_receiver_offset + receiver_step*c;
2170     int count_offset = offset + receiver_to_count_step;
2171     assert(offset == real_recv_offset, "receiver slot math");
2172     assert(count_offset == real_count_offset, "receiver count math");
2173   }
2174   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2175   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2176 #endif
2177 
2178   // Corner case: no profile table. Increment poly counter and exit.
2179   if (ReceiverTypeData::row_limit() == 0) {
2180     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2181     return;
2182   }
2183 
2184   Register offset = rscratch2;
2185 
2186   Label L_loop_search_receiver, L_loop_search_empty;
2187   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
2188 
2189   // The code here recognizes three major cases:
2190   //   A. Fastest: receiver found in the table
2191   //   B. Fast: no receiver in the table, and the table is full
2192   //   C. Slow: no receiver in the table, free slots in the table
2193   //
2194   // The case A performance is most important, as perfectly-behaved code would end up
2195   // there, especially with larger TypeProfileWidth. The case B performance is
2196   // important as well, this is where bulk of code would land for normally megamorphic
2197   // cases. The case C performance is not essential, its job is to deal with installation
2198   // races, we optimize for code density instead. Case C needs to make sure that receiver
2199   // rows are only claimed once. This makes sure we never overwrite a row for another
2200   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2201   //
2202   // It is very tempting to handle these cases in a single loop, and claim the first slot
2203   // without checking the rest of the table. But, profiling code should tolerate free slots
2204   // in the table, as class unloading can clear them. After such cleanup, the receiver
2205   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2206   // to complete, before trying to install new slots. Splitting the code in several tight
2207   // loops also helpfully optimizes for cases A and B.
2208   //
2209   // This code is effectively:
2210   //
2211   // restart:
2212   //   // Fastest: receiver is already installed
2213   //   for (i = 0; i < receiver_count(); i++) {
2214   //     if (receiver(i) == recv) goto found_recv(i);
2215   //   }
2216   //
2217   //   // Fast: no receiver, but profile is full
2218   //   for (i = 0; i < receiver_count(); i++) {
2219   //     if (receiver(i) == null) goto found_null(i);
2220   //   }
2221   //   goto polymorphic
2222   //
2223   //   // Slow: try to install receiver
2224   // found_null(i):
2225   //   CAS(&receiver(i), null, recv);
2226   //   goto restart
2227   //
2228   // polymorphic:
2229   //   count++;
2230   //   return
2231   //
2232   // found_recv(i):
2233   //   *receiver_count(i)++
2234   //
2235 
2236   bind(L_restart);
2237 
2238   // Fastest: receiver is already installed
2239   mov(offset, base_receiver_offset);
2240   bind(L_loop_search_receiver);
2241     ldr(rscratch1, Address(mdp, offset));
2242     cmp(rscratch1, recv);
2243     br(Assembler::EQ, L_found_recv);
2244   add(offset, offset, receiver_step);
2245   sub(rscratch1, offset, end_receiver_offset);
2246   cbnz(rscratch1, L_loop_search_receiver);
2247 
2248   // Fast: no receiver, but profile is full
2249   mov(offset, base_receiver_offset);
2250   bind(L_loop_search_empty);
2251     ldr(rscratch1, Address(mdp, offset));
2252     cbz(rscratch1, L_found_empty);
2253   add(offset, offset, receiver_step);
2254   sub(rscratch1, offset, end_receiver_offset);
2255   cbnz(rscratch1, L_loop_search_empty);
2256   b(L_polymorphic);
2257 
2258   // Slow: try to install receiver
2259   bind(L_found_empty);
2260 
2261   // Atomically swing receiver slot: null -> recv.
2262   //
2263   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2264   // is used to hold the destination address. This is safe because the
2265   // offset is no longer needed after the address is computed.
2266 
2267   lea(rscratch2, Address(mdp, offset));
2268   cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
2269           /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
2270 
2271   // CAS success means the slot now has the receiver we want. CAS failure means
2272   // something had claimed the slot concurrently: it can be the same receiver we want,
2273   // or something else. Since this is a slow path, we can optimize for code density,
2274   // and just restart the search from the beginning.
2275   b(L_restart);
2276 
2277   // Counter updates:
2278 
2279   // Increment polymorphic counter instead of receiver slot.
2280   bind(L_polymorphic);
2281   mov(offset, poly_count_offset);
2282   b(L_count_update);
2283 
2284   // Found a receiver, convert its slot offset to corresponding count offset.
2285   bind(L_found_recv);
2286   add(offset, offset, receiver_to_count_step);
2287 
2288   bind(L_count_update);
2289   increment(Address(mdp, offset), DataLayout::counter_increment);
2290 }
2291 
2292 
2293 void MacroAssembler::call_VM_leaf_base(address entry_point,
2294                                        int number_of_arguments,
2295                                        Label *retaddr) {
2296   Label E, L;
2297 
2298   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2299 
2300   mov(rscratch1, RuntimeAddress(entry_point));
2301   blr(rscratch1);
2302   if (retaddr)
2303     bind(*retaddr);
2304 
2305   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2306 }
2307 
2308 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2309   call_VM_leaf_base(entry_point, number_of_arguments);
2310 }
2311 
2312 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2313   pass_arg0(this, arg_0);
2314   call_VM_leaf_base(entry_point, 1);
2315 }
2316 
2317 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2318   assert_different_registers(arg_1, c_rarg0);
2319   pass_arg0(this, arg_0);
2320   pass_arg1(this, arg_1);
2321   call_VM_leaf_base(entry_point, 2);
2322 }
2323 
2324 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2325                                   Register arg_1, Register arg_2) {
2326   assert_different_registers(arg_1, c_rarg0);
2327   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2328   pass_arg0(this, arg_0);
2329   pass_arg1(this, arg_1);
2330   pass_arg2(this, arg_2);
2331   call_VM_leaf_base(entry_point, 3);
2332 }
2333 
2334 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2335   pass_arg0(this, arg_0);
2336   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2337 }
2338 
2339 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2340 
2341   assert_different_registers(arg_0, c_rarg1);
2342   pass_arg1(this, arg_1);
2343   pass_arg0(this, arg_0);
2344   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2345 }
2346 
2347 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2348   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2349   assert_different_registers(arg_1, c_rarg2);
2350   pass_arg2(this, arg_2);
2351   pass_arg1(this, arg_1);
2352   pass_arg0(this, arg_0);
2353   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2354 }
2355 
2356 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2357   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2358   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2359   assert_different_registers(arg_2, c_rarg3);
2360   pass_arg3(this, arg_3);
2361   pass_arg2(this, arg_2);
2362   pass_arg1(this, arg_1);
2363   pass_arg0(this, arg_0);
2364   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2365 }
2366 
2367 void MacroAssembler::null_check(Register reg, int offset) {
2368   if (needs_explicit_null_check(offset)) {
2369     // provoke OS null exception if reg is null by
2370     // accessing M[reg] w/o changing any registers
2371     // NOTE: this is plenty to provoke a segv
2372     ldr(zr, Address(reg));
2373   } else {
2374     // nothing to do, (later) access of M[reg + offset]
2375     // will provoke OS null exception if reg is null
2376   }
2377 }
2378 
2379 // MacroAssembler protected routines needed to implement
2380 // public methods
2381 
2382 void MacroAssembler::mov(Register r, Address dest) {
2383   code_section()->relocate(pc(), dest.rspec());
2384   uint64_t imm64 = (uint64_t)dest.target();
2385   movptr(r, imm64);
2386 }
2387 
2388 // Move a constant pointer into r.  In AArch64 mode the virtual
2389 // address space is 48 bits in size, so we only need three
2390 // instructions to create a patchable instruction sequence that can
2391 // reach anywhere.
2392 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2393 #ifndef PRODUCT
2394   {
2395     char buffer[64];
2396     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2397     block_comment(buffer);
2398   }
2399 #endif
2400   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2401   movz(r, imm64 & 0xffff);
2402   imm64 >>= 16;
2403   movk(r, imm64 & 0xffff, 16);
2404   imm64 >>= 16;
2405   movk(r, imm64 & 0xffff, 32);
2406 }
2407 
2408 // Macro to mov replicated immediate to vector register.
2409 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2410 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2411 // Vd will get the following values for different arrangements in T
2412 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2413 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2414 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2415 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2416 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2417 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2418 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2419 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2420 // Clobbers rscratch1
2421 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2422   assert(T != T1Q, "unsupported");
2423   if (T == T1D || T == T2D) {
2424     int imm = operand_valid_for_movi_immediate(imm64, T);
2425     if (-1 != imm) {
2426       movi(Vd, T, imm);
2427     } else {
2428       mov(rscratch1, imm64);
2429       dup(Vd, T, rscratch1);
2430     }
2431     return;
2432   }
2433 
2434 #ifdef ASSERT
2435   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2436   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2437   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2438 #endif
2439   int shift = operand_valid_for_movi_immediate(imm64, T);
2440   uint32_t imm32 = imm64 & 0xffffffffULL;
2441   if (shift >= 0) {
2442     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2443   } else {
2444     movw(rscratch1, imm32);
2445     dup(Vd, T, rscratch1);
2446   }
2447 }
2448 
2449 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2450 {
2451 #ifndef PRODUCT
2452   {
2453     char buffer[64];
2454     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2455     block_comment(buffer);
2456   }
2457 #endif
2458   if (operand_valid_for_logical_immediate(false, imm64)) {
2459     orr(dst, zr, imm64);
2460   } else {
2461     // we can use a combination of MOVZ or MOVN with
2462     // MOVK to build up the constant
2463     uint64_t imm_h[4];
2464     int zero_count = 0;
2465     int neg_count = 0;
2466     int i;
2467     for (i = 0; i < 4; i++) {
2468       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2469       if (imm_h[i] == 0) {
2470         zero_count++;
2471       } else if (imm_h[i] == 0xffffL) {
2472         neg_count++;
2473       }
2474     }
2475     if (zero_count == 4) {
2476       // one MOVZ will do
2477       movz(dst, 0);
2478     } else if (neg_count == 4) {
2479       // one MOVN will do
2480       movn(dst, 0);
2481     } else if (zero_count == 3) {
2482       for (i = 0; i < 4; i++) {
2483         if (imm_h[i] != 0L) {
2484           movz(dst, (uint32_t)imm_h[i], (i << 4));
2485           break;
2486         }
2487       }
2488     } else if (neg_count == 3) {
2489       // one MOVN will do
2490       for (int i = 0; i < 4; i++) {
2491         if (imm_h[i] != 0xffffL) {
2492           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2493           break;
2494         }
2495       }
2496     } else if (zero_count == 2) {
2497       // one MOVZ and one MOVK will do
2498       for (i = 0; i < 3; i++) {
2499         if (imm_h[i] != 0L) {
2500           movz(dst, (uint32_t)imm_h[i], (i << 4));
2501           i++;
2502           break;
2503         }
2504       }
2505       for (;i < 4; i++) {
2506         if (imm_h[i] != 0L) {
2507           movk(dst, (uint32_t)imm_h[i], (i << 4));
2508         }
2509       }
2510     } else if (neg_count == 2) {
2511       // one MOVN and one MOVK will do
2512       for (i = 0; i < 4; i++) {
2513         if (imm_h[i] != 0xffffL) {
2514           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2515           i++;
2516           break;
2517         }
2518       }
2519       for (;i < 4; i++) {
2520         if (imm_h[i] != 0xffffL) {
2521           movk(dst, (uint32_t)imm_h[i], (i << 4));
2522         }
2523       }
2524     } else if (zero_count == 1) {
2525       // one MOVZ and two MOVKs will do
2526       for (i = 0; i < 4; i++) {
2527         if (imm_h[i] != 0L) {
2528           movz(dst, (uint32_t)imm_h[i], (i << 4));
2529           i++;
2530           break;
2531         }
2532       }
2533       for (;i < 4; i++) {
2534         if (imm_h[i] != 0x0L) {
2535           movk(dst, (uint32_t)imm_h[i], (i << 4));
2536         }
2537       }
2538     } else if (neg_count == 1) {
2539       // one MOVN and two MOVKs will do
2540       for (i = 0; i < 4; i++) {
2541         if (imm_h[i] != 0xffffL) {
2542           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2543           i++;
2544           break;
2545         }
2546       }
2547       for (;i < 4; i++) {
2548         if (imm_h[i] != 0xffffL) {
2549           movk(dst, (uint32_t)imm_h[i], (i << 4));
2550         }
2551       }
2552     } else {
2553       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2554       movz(dst, (uint32_t)imm_h[0], 0);
2555       for (i = 1; i < 4; i++) {
2556         movk(dst, (uint32_t)imm_h[i], (i << 4));
2557       }
2558     }
2559   }
2560 }
2561 
2562 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2563 {
2564 #ifndef PRODUCT
2565     {
2566       char buffer[64];
2567       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2568       block_comment(buffer);
2569     }
2570 #endif
2571   if (operand_valid_for_logical_immediate(true, imm32)) {
2572     orrw(dst, zr, imm32);
2573   } else {
2574     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2575     // constant
2576     uint32_t imm_h[2];
2577     imm_h[0] = imm32 & 0xffff;
2578     imm_h[1] = ((imm32 >> 16) & 0xffff);
2579     if (imm_h[0] == 0) {
2580       movzw(dst, imm_h[1], 16);
2581     } else if (imm_h[0] == 0xffff) {
2582       movnw(dst, imm_h[1] ^ 0xffff, 16);
2583     } else if (imm_h[1] == 0) {
2584       movzw(dst, imm_h[0], 0);
2585     } else if (imm_h[1] == 0xffff) {
2586       movnw(dst, imm_h[0] ^ 0xffff, 0);
2587     } else {
2588       // use a MOVZ and MOVK (makes it easier to debug)
2589       movzw(dst, imm_h[0], 0);
2590       movkw(dst, imm_h[1], 16);
2591     }
2592   }
2593 }
2594 
2595 // Form an address from base + offset in Rd.  Rd may or may
2596 // not actually be used: you must use the Address that is returned.
2597 // It is up to you to ensure that the shift provided matches the size
2598 // of your data.
2599 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2600   if (Address::offset_ok_for_immed(byte_offset, shift))
2601     // It fits; no need for any heroics
2602     return Address(base, byte_offset);
2603 
2604   // Don't do anything clever with negative or misaligned offsets
2605   unsigned mask = (1 << shift) - 1;
2606   if (byte_offset < 0 || byte_offset & mask) {
2607     mov(Rd, byte_offset);
2608     add(Rd, base, Rd);
2609     return Address(Rd);
2610   }
2611 
2612   // See if we can do this with two 12-bit offsets
2613   {
2614     uint64_t word_offset = byte_offset >> shift;
2615     uint64_t masked_offset = word_offset & 0xfff000;
2616     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2617         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2618       add(Rd, base, masked_offset << shift);
2619       word_offset -= masked_offset;
2620       return Address(Rd, word_offset << shift);
2621     }
2622   }
2623 
2624   // Do it the hard way
2625   mov(Rd, byte_offset);
2626   add(Rd, base, Rd);
2627   return Address(Rd);
2628 }
2629 
2630 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2631                                     bool want_remainder, Register scratch)
2632 {
2633   // Full implementation of Java idiv and irem.  The function
2634   // returns the (pc) offset of the div instruction - may be needed
2635   // for implicit exceptions.
2636   //
2637   // constraint : ra/rb =/= scratch
2638   //         normal case
2639   //
2640   // input : ra: dividend
2641   //         rb: divisor
2642   //
2643   // result: either
2644   //         quotient  (= ra idiv rb)
2645   //         remainder (= ra irem rb)
2646 
2647   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2648 
2649   int idivl_offset = offset();
2650   if (! want_remainder) {
2651     sdivw(result, ra, rb);
2652   } else {
2653     sdivw(scratch, ra, rb);
2654     Assembler::msubw(result, scratch, rb, ra);
2655   }
2656 
2657   return idivl_offset;
2658 }
2659 
2660 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2661                                     bool want_remainder, Register scratch)
2662 {
2663   // Full implementation of Java ldiv and lrem.  The function
2664   // returns the (pc) offset of the div instruction - may be needed
2665   // for implicit exceptions.
2666   //
2667   // constraint : ra/rb =/= scratch
2668   //         normal case
2669   //
2670   // input : ra: dividend
2671   //         rb: divisor
2672   //
2673   // result: either
2674   //         quotient  (= ra idiv rb)
2675   //         remainder (= ra irem rb)
2676 
2677   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2678 
2679   int idivq_offset = offset();
2680   if (! want_remainder) {
2681     sdiv(result, ra, rb);
2682   } else {
2683     sdiv(scratch, ra, rb);
2684     Assembler::msub(result, scratch, rb, ra);
2685   }
2686 
2687   return idivq_offset;
2688 }
2689 
2690 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2691   address prev = pc() - NativeMembar::instruction_size;
2692   address last = code()->last_insn();
2693   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2694     NativeMembar *bar = NativeMembar_at(prev);
2695     if (AlwaysMergeDMB) {
2696       bar->set_kind(bar->get_kind() | order_constraint);
2697       BLOCK_COMMENT("merged membar(always)");
2698       return;
2699     }
2700     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2701     // doing so would introduce a StoreLoad which the caller did not
2702     // intend
2703     if (bar->get_kind() == order_constraint
2704         || bar->get_kind() == AnyAny
2705         || order_constraint == AnyAny) {
2706       // We are merging two memory barrier instructions.  On AArch64 we
2707       // can do this simply by ORing them together.
2708       bar->set_kind(bar->get_kind() | order_constraint);
2709       BLOCK_COMMENT("merged membar");
2710       return;
2711     } else {
2712       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2713       // We need check the last 2 instructions
2714       address prev2 = prev - NativeMembar::instruction_size;
2715       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2716         NativeMembar *bar2 = NativeMembar_at(prev2);
2717         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2718         BLOCK_COMMENT("merged membar(elided)");
2719         return;
2720       }
2721     }
2722   }
2723   code()->set_last_insn(pc());
2724   dmb(Assembler::barrier(order_constraint));
2725 }
2726 
2727 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2728   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2729     merge_ldst(rt, adr, size_in_bytes, is_store);
2730     code()->clear_last_insn();
2731     return true;
2732   } else {
2733     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2734     const uint64_t mask = size_in_bytes - 1;
2735     if (adr.getMode() == Address::base_plus_offset &&
2736         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2737       code()->set_last_insn(pc());
2738     }
2739     return false;
2740   }
2741 }
2742 
2743 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2744   // We always try to merge two adjacent loads into one ldp.
2745   if (!try_merge_ldst(Rx, adr, 8, false)) {
2746     Assembler::ldr(Rx, adr);
2747   }
2748 }
2749 
2750 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2751   // We always try to merge two adjacent loads into one ldp.
2752   if (!try_merge_ldst(Rw, adr, 4, false)) {
2753     Assembler::ldrw(Rw, adr);
2754   }
2755 }
2756 
2757 void MacroAssembler::str(Register Rx, const Address &adr) {
2758   // We always try to merge two adjacent stores into one stp.
2759   if (!try_merge_ldst(Rx, adr, 8, true)) {
2760     Assembler::str(Rx, adr);
2761   }
2762 }
2763 
2764 void MacroAssembler::strw(Register Rw, const Address &adr) {
2765   // We always try to merge two adjacent stores into one stp.
2766   if (!try_merge_ldst(Rw, adr, 4, true)) {
2767     Assembler::strw(Rw, adr);
2768   }
2769 }
2770 
2771 // MacroAssembler routines found actually to be needed
2772 
2773 void MacroAssembler::push(Register src)
2774 {
2775   str(src, Address(pre(esp, -1 * wordSize)));
2776 }
2777 
2778 void MacroAssembler::pop(Register dst)
2779 {
2780   ldr(dst, Address(post(esp, 1 * wordSize)));
2781 }
2782 
2783 // Note: load_unsigned_short used to be called load_unsigned_word.
2784 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2785   int off = offset();
2786   ldrh(dst, src);
2787   return off;
2788 }
2789 
2790 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2791   int off = offset();
2792   ldrb(dst, src);
2793   return off;
2794 }
2795 
2796 int MacroAssembler::load_signed_short(Register dst, Address src) {
2797   int off = offset();
2798   ldrsh(dst, src);
2799   return off;
2800 }
2801 
2802 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2803   int off = offset();
2804   ldrsb(dst, src);
2805   return off;
2806 }
2807 
2808 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2809   int off = offset();
2810   ldrshw(dst, src);
2811   return off;
2812 }
2813 
2814 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2815   int off = offset();
2816   ldrsbw(dst, src);
2817   return off;
2818 }
2819 
2820 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2821   switch (size_in_bytes) {
2822   case  8:  ldr(dst, src); break;
2823   case  4:  ldrw(dst, src); break;
2824   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2825   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2826   default:  ShouldNotReachHere();
2827   }
2828 }
2829 
2830 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2831   switch (size_in_bytes) {
2832   case  8:  str(src, dst); break;
2833   case  4:  strw(src, dst); break;
2834   case  2:  strh(src, dst); break;
2835   case  1:  strb(src, dst); break;
2836   default:  ShouldNotReachHere();
2837   }
2838 }
2839 
2840 void MacroAssembler::decrementw(Register reg, int value)
2841 {
2842   if (value < 0)  { incrementw(reg, -value);      return; }
2843   if (value == 0) {                               return; }
2844   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2845   /* else */ {
2846     guarantee(reg != rscratch2, "invalid dst for register decrement");
2847     movw(rscratch2, (unsigned)value);
2848     subw(reg, reg, rscratch2);
2849   }
2850 }
2851 
2852 void MacroAssembler::decrement(Register reg, int value)
2853 {
2854   if (value < 0)  { increment(reg, -value);      return; }
2855   if (value == 0) {                              return; }
2856   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2857   /* else */ {
2858     assert(reg != rscratch2, "invalid dst for register decrement");
2859     mov(rscratch2, (uint64_t)value);
2860     sub(reg, reg, rscratch2);
2861   }
2862 }
2863 
2864 void MacroAssembler::decrementw(Address dst, int value)
2865 {
2866   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2867   if (dst.getMode() == Address::literal) {
2868     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2869     lea(rscratch2, dst);
2870     dst = Address(rscratch2);
2871   }
2872   ldrw(rscratch1, dst);
2873   decrementw(rscratch1, value);
2874   strw(rscratch1, dst);
2875 }
2876 
2877 void MacroAssembler::decrement(Address dst, int value)
2878 {
2879   assert(!dst.uses(rscratch1), "invalid address for decrement");
2880   if (dst.getMode() == Address::literal) {
2881     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2882     lea(rscratch2, dst);
2883     dst = Address(rscratch2);
2884   }
2885   ldr(rscratch1, dst);
2886   decrement(rscratch1, value);
2887   str(rscratch1, dst);
2888 }
2889 
2890 void MacroAssembler::incrementw(Register reg, int value)
2891 {
2892   if (value < 0)  { decrementw(reg, -value);      return; }
2893   if (value == 0) {                               return; }
2894   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2895   /* else */ {
2896     assert(reg != rscratch2, "invalid dst for register increment");
2897     movw(rscratch2, (unsigned)value);
2898     addw(reg, reg, rscratch2);
2899   }
2900 }
2901 
2902 void MacroAssembler::increment(Register reg, int value)
2903 {
2904   if (value < 0)  { decrement(reg, -value);      return; }
2905   if (value == 0) {                              return; }
2906   if (value < (1 << 12)) { add(reg, reg, value); return; }
2907   /* else */ {
2908     assert(reg != rscratch2, "invalid dst for register increment");
2909     movw(rscratch2, (unsigned)value);
2910     add(reg, reg, rscratch2);
2911   }
2912 }
2913 
2914 void MacroAssembler::incrementw(Address dst, int value)
2915 {
2916   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2917   if (dst.getMode() == Address::literal) {
2918     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2919     lea(rscratch2, dst);
2920     dst = Address(rscratch2);
2921   }
2922   ldrw(rscratch1, dst);
2923   incrementw(rscratch1, value);
2924   strw(rscratch1, dst);
2925 }
2926 
2927 void MacroAssembler::increment(Address dst, int value)
2928 {
2929   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2930   if (dst.getMode() == Address::literal) {
2931     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2932     lea(rscratch2, dst);
2933     dst = Address(rscratch2);
2934   }
2935   ldr(rscratch1, dst);
2936   increment(rscratch1, value);
2937   str(rscratch1, dst);
2938 }
2939 
2940 // Push lots of registers in the bit set supplied.  Don't push sp.
2941 // Return the number of words pushed
2942 int MacroAssembler::push(unsigned int bitset, Register stack) {
2943   int words_pushed = 0;
2944 
2945   // Scan bitset to accumulate register pairs
2946   unsigned char regs[32];
2947   int count = 0;
2948   for (int reg = 0; reg <= 30; reg++) {
2949     if (1 & bitset)
2950       regs[count++] = reg;
2951     bitset >>= 1;
2952   }
2953   regs[count++] = zr->raw_encoding();
2954   count &= ~1;  // Only push an even number of regs
2955 
2956   if (count) {
2957     stp(as_Register(regs[0]), as_Register(regs[1]),
2958        Address(pre(stack, -count * wordSize)));
2959     words_pushed += 2;
2960   }
2961   for (int i = 2; i < count; i += 2) {
2962     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2963        Address(stack, i * wordSize));
2964     words_pushed += 2;
2965   }
2966 
2967   assert(words_pushed == count, "oops, pushed != count");
2968 
2969   return count;
2970 }
2971 
2972 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2973   int words_pushed = 0;
2974 
2975   // Scan bitset to accumulate register pairs
2976   unsigned char regs[32];
2977   int count = 0;
2978   for (int reg = 0; reg <= 30; reg++) {
2979     if (1 & bitset)
2980       regs[count++] = reg;
2981     bitset >>= 1;
2982   }
2983   regs[count++] = zr->raw_encoding();
2984   count &= ~1;
2985 
2986   for (int i = 2; i < count; i += 2) {
2987     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2988        Address(stack, i * wordSize));
2989     words_pushed += 2;
2990   }
2991   if (count) {
2992     ldp(as_Register(regs[0]), as_Register(regs[1]),
2993        Address(post(stack, count * wordSize)));
2994     words_pushed += 2;
2995   }
2996 
2997   assert(words_pushed == count, "oops, pushed != count");
2998 
2999   return count;
3000 }
3001 
3002 // Push lots of registers in the bit set supplied.  Don't push sp.
3003 // Return the number of dwords pushed
3004 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3005   int words_pushed = 0;
3006   bool use_sve = false;
3007   int sve_vector_size_in_bytes = 0;
3008 
3009 #ifdef COMPILER2
3010   use_sve = Matcher::supports_scalable_vector();
3011   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3012 #endif
3013 
3014   // Scan bitset to accumulate register pairs
3015   unsigned char regs[32];
3016   int count = 0;
3017   for (int reg = 0; reg <= 31; reg++) {
3018     if (1 & bitset)
3019       regs[count++] = reg;
3020     bitset >>= 1;
3021   }
3022 
3023   if (count == 0) {
3024     return 0;
3025   }
3026 
3027   if (mode == PushPopFull) {
3028     if (use_sve && sve_vector_size_in_bytes > 16) {
3029       mode = PushPopSVE;
3030     } else {
3031       mode = PushPopNeon;
3032     }
3033   }
3034 
3035 #ifndef PRODUCT
3036   {
3037     char buffer[48];
3038     if (mode == PushPopSVE) {
3039       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3040     } else if (mode == PushPopNeon) {
3041       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3042     } else {
3043       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3044     }
3045     block_comment(buffer);
3046   }
3047 #endif
3048 
3049   if (mode == PushPopSVE) {
3050     sub(stack, stack, sve_vector_size_in_bytes * count);
3051     for (int i = 0; i < count; i++) {
3052       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3053     }
3054     return count * sve_vector_size_in_bytes / 8;
3055   }
3056 
3057   if (mode == PushPopNeon) {
3058     if (count == 1) {
3059       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3060       return 2;
3061     }
3062 
3063     bool odd = (count & 1) == 1;
3064     int push_slots = count + (odd ? 1 : 0);
3065 
3066     // Always pushing full 128 bit registers.
3067     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3068     words_pushed += 2;
3069 
3070     for (int i = 2; i + 1 < count; i += 2) {
3071       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3072       words_pushed += 2;
3073     }
3074 
3075     if (odd) {
3076       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3077       words_pushed++;
3078     }
3079 
3080     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3081     return count * 2;
3082   }
3083 
3084   if (mode == PushPopFp) {
3085     bool odd = (count & 1) == 1;
3086     int push_slots = count + (odd ? 1 : 0);
3087 
3088     if (count == 1) {
3089       // Stack pointer must be 16 bytes aligned
3090       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3091       return 1;
3092     }
3093 
3094     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3095     words_pushed += 2;
3096 
3097     for (int i = 2; i + 1 < count; i += 2) {
3098       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3099       words_pushed += 2;
3100     }
3101 
3102     if (odd) {
3103       // Stack pointer must be 16 bytes aligned
3104       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3105       words_pushed++;
3106     }
3107 
3108     assert(words_pushed == count, "oops, pushed != count");
3109 
3110     return count;
3111   }
3112 
3113   return 0;
3114 }
3115 
3116 // Return the number of dwords popped
3117 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3118   int words_pushed = 0;
3119   bool use_sve = false;
3120   int sve_vector_size_in_bytes = 0;
3121 
3122 #ifdef COMPILER2
3123   use_sve = Matcher::supports_scalable_vector();
3124   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3125 #endif
3126   // Scan bitset to accumulate register pairs
3127   unsigned char regs[32];
3128   int count = 0;
3129   for (int reg = 0; reg <= 31; reg++) {
3130     if (1 & bitset)
3131       regs[count++] = reg;
3132     bitset >>= 1;
3133   }
3134 
3135   if (count == 0) {
3136     return 0;
3137   }
3138 
3139   if (mode == PushPopFull) {
3140     if (use_sve && sve_vector_size_in_bytes > 16) {
3141       mode = PushPopSVE;
3142     } else {
3143       mode = PushPopNeon;
3144     }
3145   }
3146 
3147 #ifndef PRODUCT
3148   {
3149     char buffer[48];
3150     if (mode == PushPopSVE) {
3151       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3152     } else if (mode == PushPopNeon) {
3153       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3154     } else {
3155       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3156     }
3157     block_comment(buffer);
3158   }
3159 #endif
3160 
3161   if (mode == PushPopSVE) {
3162     for (int i = count - 1; i >= 0; i--) {
3163       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3164     }
3165     add(stack, stack, sve_vector_size_in_bytes * count);
3166     return count * sve_vector_size_in_bytes / 8;
3167   }
3168 
3169   if (mode == PushPopNeon) {
3170     if (count == 1) {
3171       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3172       return 2;
3173     }
3174 
3175     bool odd = (count & 1) == 1;
3176     int push_slots = count + (odd ? 1 : 0);
3177 
3178     if (odd) {
3179       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3180       words_pushed++;
3181     }
3182 
3183     for (int i = 2; i + 1 < count; i += 2) {
3184       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3185       words_pushed += 2;
3186     }
3187 
3188     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3189     words_pushed += 2;
3190 
3191     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3192 
3193     return count * 2;
3194   }
3195 
3196   if (mode == PushPopFp) {
3197     bool odd = (count & 1) == 1;
3198     int push_slots = count + (odd ? 1 : 0);
3199 
3200     if (count == 1) {
3201       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3202       return 1;
3203     }
3204 
3205     if (odd) {
3206       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3207       words_pushed++;
3208     }
3209 
3210     for (int i = 2; i + 1 < count; i += 2) {
3211       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3212       words_pushed += 2;
3213     }
3214 
3215     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3216     words_pushed += 2;
3217 
3218     assert(words_pushed == count, "oops, pushed != count");
3219 
3220     return count;
3221   }
3222 
3223   return 0;
3224 }
3225 
3226 // Return the number of dwords pushed
3227 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3228   bool use_sve = false;
3229   int sve_predicate_size_in_slots = 0;
3230 
3231 #ifdef COMPILER2
3232   use_sve = Matcher::supports_scalable_vector();
3233   if (use_sve) {
3234     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3235   }
3236 #endif
3237 
3238   if (!use_sve) {
3239     return 0;
3240   }
3241 
3242   unsigned char regs[PRegister::number_of_registers];
3243   int count = 0;
3244   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3245     if (1 & bitset)
3246       regs[count++] = reg;
3247     bitset >>= 1;
3248   }
3249 
3250   if (count == 0) {
3251     return 0;
3252   }
3253 
3254   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3255                                   VMRegImpl::stack_slot_size * count, 16);
3256   sub(stack, stack, total_push_bytes);
3257   for (int i = 0; i < count; i++) {
3258     sve_str(as_PRegister(regs[i]), Address(stack, i));
3259   }
3260   return total_push_bytes / 8;
3261 }
3262 
3263 // Return the number of dwords popped
3264 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3265   bool use_sve = false;
3266   int sve_predicate_size_in_slots = 0;
3267 
3268 #ifdef COMPILER2
3269   use_sve = Matcher::supports_scalable_vector();
3270   if (use_sve) {
3271     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3272   }
3273 #endif
3274 
3275   if (!use_sve) {
3276     return 0;
3277   }
3278 
3279   unsigned char regs[PRegister::number_of_registers];
3280   int count = 0;
3281   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3282     if (1 & bitset)
3283       regs[count++] = reg;
3284     bitset >>= 1;
3285   }
3286 
3287   if (count == 0) {
3288     return 0;
3289   }
3290 
3291   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3292                                  VMRegImpl::stack_slot_size * count, 16);
3293   for (int i = count - 1; i >= 0; i--) {
3294     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3295   }
3296   add(stack, stack, total_pop_bytes);
3297   return total_pop_bytes / 8;
3298 }
3299 
3300 #ifdef ASSERT
3301 void MacroAssembler::verify_heapbase(const char* msg) {
3302 #if 0
3303   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3304   assert (Universe::heap() != nullptr, "java heap should be initialized");
3305   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3306     // rheapbase is allocated as general register
3307     return;
3308   }
3309   if (CheckCompressedOops) {
3310     Label ok;
3311     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3312     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3313     br(Assembler::EQ, ok);
3314     stop(msg);
3315     bind(ok);
3316     pop(1 << rscratch1->encoding(), sp);
3317   }
3318 #endif
3319 }
3320 #endif
3321 
3322 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3323   assert_different_registers(value, tmp1, tmp2);
3324   Label done, tagged, weak_tagged;
3325 
3326   cbz(value, done);           // Use null as-is.
3327   tst(value, JNIHandles::tag_mask); // Test for tag.
3328   br(Assembler::NE, tagged);
3329 
3330   // Resolve local handle
3331   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3332   verify_oop(value);
3333   b(done);
3334 
3335   bind(tagged);
3336   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3337   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3338 
3339   // Resolve global handle
3340   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3341   verify_oop(value);
3342   b(done);
3343 
3344   bind(weak_tagged);
3345   // Resolve jweak.
3346   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3347                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3348   verify_oop(value);
3349 
3350   bind(done);
3351 }
3352 
3353 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3354   assert_different_registers(value, tmp1, tmp2);
3355   Label done;
3356 
3357   cbz(value, done);           // Use null as-is.
3358 
3359 #ifdef ASSERT
3360   {
3361     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3362     Label valid_global_tag;
3363     tbnz(value, 1, valid_global_tag); // Test for global tag
3364     stop("non global jobject using resolve_global_jobject");
3365     bind(valid_global_tag);
3366   }
3367 #endif
3368 
3369   // Resolve global handle
3370   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3371   verify_oop(value);
3372 
3373   bind(done);
3374 }
3375 
3376 void MacroAssembler::stop(const char* msg) {
3377   // Skip AOT caching C strings in scratch buffer.
3378   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3379   BLOCK_COMMENT(str);
3380   // load msg into r0 so we can access it from the signal handler
3381   // ExternalAddress enables saving and restoring via the code cache
3382   lea(c_rarg0, ExternalAddress((address) str));
3383   dcps1(0xdeae);
3384 }
3385 
3386 void MacroAssembler::unimplemented(const char* what) {
3387   const char* buf = nullptr;
3388   {
3389     ResourceMark rm;
3390     stringStream ss;
3391     ss.print("unimplemented: %s", what);
3392     buf = code_string(ss.as_string());
3393   }
3394   stop(buf);
3395 }
3396 
3397 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3398 #ifdef ASSERT
3399   Label OK;
3400   br(cc, OK);
3401   stop(msg);
3402   bind(OK);
3403 #endif
3404 }
3405 
3406 // If a constant does not fit in an immediate field, generate some
3407 // number of MOV instructions and then perform the operation.
3408 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3409                                            add_sub_imm_insn insn1,
3410                                            add_sub_reg_insn insn2,
3411                                            bool is32) {
3412   assert(Rd != zr, "Rd = zr and not setting flags?");
3413   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3414   if (fits) {
3415     (this->*insn1)(Rd, Rn, imm);
3416   } else {
3417     if (g_uabs(imm) < (1 << 24)) {
3418        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3419        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3420     } else {
3421        assert_different_registers(Rd, Rn);
3422        mov(Rd, imm);
3423        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3424     }
3425   }
3426 }
3427 
3428 // Separate vsn which sets the flags. Optimisations are more restricted
3429 // because we must set the flags correctly.
3430 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3431                                              add_sub_imm_insn insn1,
3432                                              add_sub_reg_insn insn2,
3433                                              bool is32) {
3434   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3435   if (fits) {
3436     (this->*insn1)(Rd, Rn, imm);
3437   } else {
3438     assert_different_registers(Rd, Rn);
3439     assert(Rd != zr, "overflow in immediate operand");
3440     mov(Rd, imm);
3441     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3442   }
3443 }
3444 
3445 
3446 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3447   if (increment.is_register()) {
3448     add(Rd, Rn, increment.as_register());
3449   } else {
3450     add(Rd, Rn, increment.as_constant());
3451   }
3452 }
3453 
3454 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3455   if (increment.is_register()) {
3456     addw(Rd, Rn, increment.as_register());
3457   } else {
3458     addw(Rd, Rn, increment.as_constant());
3459   }
3460 }
3461 
3462 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3463   if (decrement.is_register()) {
3464     sub(Rd, Rn, decrement.as_register());
3465   } else {
3466     sub(Rd, Rn, decrement.as_constant());
3467   }
3468 }
3469 
3470 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3471   if (decrement.is_register()) {
3472     subw(Rd, Rn, decrement.as_register());
3473   } else {
3474     subw(Rd, Rn, decrement.as_constant());
3475   }
3476 }
3477 
3478 void MacroAssembler::reinit_heapbase()
3479 {
3480   if (UseCompressedOops) {
3481     if (Universe::is_fully_initialized() && !AOTCodeCache::is_on_for_dump()) {
3482       mov(rheapbase, CompressedOops::base());
3483     } else {
3484       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3485       ldr(rheapbase, Address(rheapbase));
3486     }
3487   }
3488 }
3489 
3490 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3491 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3492 // Pass a register for the result, otherwise pass noreg.
3493 
3494 // Clobbers rscratch1
3495 void MacroAssembler::cmpxchg(Register addr, Register expected,
3496                              Register new_val,
3497                              enum operand_size size,
3498                              bool acquire, bool release,
3499                              bool weak,
3500                              Register result) {
3501   if (result == noreg)  result = rscratch1;
3502   BLOCK_COMMENT("cmpxchg {");
3503   if (UseLSE) {
3504     mov(result, expected);
3505     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3506     compare_eq(result, expected, size);
3507 #ifdef ASSERT
3508     // Poison rscratch1 which is written on !UseLSE branch
3509     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3510 #endif
3511   } else {
3512     Label retry_load, done;
3513     prfm(Address(addr), PSTL1STRM);
3514     bind(retry_load);
3515     load_exclusive(result, addr, size, acquire);
3516     compare_eq(result, expected, size);
3517     br(Assembler::NE, done);
3518     store_exclusive(rscratch1, new_val, addr, size, release);
3519     if (weak) {
3520       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3521     } else {
3522       cbnzw(rscratch1, retry_load);
3523     }
3524     bind(done);
3525   }
3526   BLOCK_COMMENT("} cmpxchg");
3527 }
3528 
3529 // A generic comparison. Only compares for equality, clobbers rscratch1.
3530 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3531   if (size == xword) {
3532     cmp(rm, rn);
3533   } else if (size == word) {
3534     cmpw(rm, rn);
3535   } else if (size == halfword) {
3536     eorw(rscratch1, rm, rn);
3537     ands(zr, rscratch1, 0xffff);
3538   } else if (size == byte) {
3539     eorw(rscratch1, rm, rn);
3540     ands(zr, rscratch1, 0xff);
3541   } else {
3542     ShouldNotReachHere();
3543   }
3544 }
3545 
3546 
3547 static bool different(Register a, RegisterOrConstant b, Register c) {
3548   if (b.is_constant())
3549     return a != c;
3550   else
3551     return a != b.as_register() && a != c && b.as_register() != c;
3552 }
3553 
3554 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3555 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3556   if (UseLSE) {                                                         \
3557     prev = prev->is_valid() ? prev : zr;                                \
3558     if (incr.is_register()) {                                           \
3559       AOP(sz, incr.as_register(), prev, addr);                          \
3560     } else {                                                            \
3561       mov(rscratch2, incr.as_constant());                               \
3562       AOP(sz, rscratch2, prev, addr);                                   \
3563     }                                                                   \
3564     return;                                                             \
3565   }                                                                     \
3566   Register result = rscratch2;                                          \
3567   if (prev->is_valid())                                                 \
3568     result = different(prev, incr, addr) ? prev : rscratch2;            \
3569                                                                         \
3570   Label retry_load;                                                     \
3571   prfm(Address(addr), PSTL1STRM);                                       \
3572   bind(retry_load);                                                     \
3573   LDXR(result, addr);                                                   \
3574   OP(rscratch1, result, incr);                                          \
3575   STXR(rscratch2, rscratch1, addr);                                     \
3576   cbnzw(rscratch2, retry_load);                                         \
3577   if (prev->is_valid() && prev != result) {                             \
3578     IOP(prev, rscratch1, incr);                                         \
3579   }                                                                     \
3580 }
3581 
3582 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3583 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3584 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3585 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3586 
3587 #undef ATOMIC_OP
3588 
3589 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3590 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3591   if (UseLSE) {                                                         \
3592     prev = prev->is_valid() ? prev : zr;                                \
3593     AOP(sz, newv, prev, addr);                                          \
3594     return;                                                             \
3595   }                                                                     \
3596   Register result = rscratch2;                                          \
3597   if (prev->is_valid())                                                 \
3598     result = different(prev, newv, addr) ? prev : rscratch2;            \
3599                                                                         \
3600   Label retry_load;                                                     \
3601   prfm(Address(addr), PSTL1STRM);                                       \
3602   bind(retry_load);                                                     \
3603   LDXR(result, addr);                                                   \
3604   STXR(rscratch1, newv, addr);                                          \
3605   cbnzw(rscratch1, retry_load);                                         \
3606   if (prev->is_valid() && prev != result)                               \
3607     mov(prev, result);                                                  \
3608 }
3609 
3610 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3611 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3612 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3613 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3614 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3615 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3616 
3617 #undef ATOMIC_XCHG
3618 
3619 #ifndef PRODUCT
3620 extern "C" void findpc(intptr_t x);
3621 #endif
3622 
3623 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3624 {
3625   // In order to get locks to work, we need to fake a in_VM state
3626   if (ShowMessageBoxOnError ) {
3627     JavaThread* thread = JavaThread::current();
3628     JavaThreadState saved_state = thread->thread_state();
3629     thread->set_thread_state(_thread_in_vm);
3630 #ifndef PRODUCT
3631     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3632       ttyLocker ttyl;
3633       BytecodeCounter::print();
3634     }
3635 #endif
3636     if (os::message_box(msg, "Execution stopped, print registers?")) {
3637       ttyLocker ttyl;
3638       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3639 #ifndef PRODUCT
3640       tty->cr();
3641       findpc(pc);
3642       tty->cr();
3643 #endif
3644       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3645       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3646       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3647       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3648       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3649       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3650       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3651       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3652       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3653       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3654       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3655       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3656       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3657       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3658       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3659       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3660       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3661       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3662       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3663       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3664       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3665       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3666       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3667       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3668       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3669       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3670       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3671       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3672       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3673       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3674       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3675       BREAKPOINT;
3676     }
3677   }
3678   fatal("DEBUG MESSAGE: %s", msg);
3679 }
3680 
3681 RegSet MacroAssembler::call_clobbered_gp_registers() {
3682   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3683 #ifndef R18_RESERVED
3684   regs += r18_tls;
3685 #endif
3686   return regs;
3687 }
3688 
3689 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3690   int step = 4 * wordSize;
3691   push(call_clobbered_gp_registers() - exclude, sp);
3692   sub(sp, sp, step);
3693   mov(rscratch1, -step);
3694   // Push v0-v7, v16-v31.
3695   for (int i = 31; i>= 4; i -= 4) {
3696     if (i <= v7->encoding() || i >= v16->encoding())
3697       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3698           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3699   }
3700   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3701       as_FloatRegister(3), T1D, Address(sp));
3702 }
3703 
3704 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3705   for (int i = 0; i < 32; i += 4) {
3706     if (i <= v7->encoding() || i >= v16->encoding())
3707       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3708           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3709   }
3710 
3711   reinitialize_ptrue();
3712 
3713   pop(call_clobbered_gp_registers() - exclude, sp);
3714 }
3715 
3716 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3717                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3718   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3719   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3720     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3721     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3722       sve_str(as_FloatRegister(i), Address(sp, i));
3723     }
3724   } else {
3725     int step = (save_vectors ? 8 : 4) * wordSize;
3726     mov(rscratch1, -step);
3727     sub(sp, sp, step);
3728     for (int i = 28; i >= 4; i -= 4) {
3729       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3730           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3731     }
3732     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3733   }
3734   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3735     sub(sp, sp, total_predicate_in_bytes);
3736     for (int i = 0; i < PRegister::number_of_registers; i++) {
3737       sve_str(as_PRegister(i), Address(sp, i));
3738     }
3739   }
3740 }
3741 
3742 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3743                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3744   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3745     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3746       sve_ldr(as_PRegister(i), Address(sp, i));
3747     }
3748     add(sp, sp, total_predicate_in_bytes);
3749   }
3750   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3751     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3752       sve_ldr(as_FloatRegister(i), Address(sp, i));
3753     }
3754     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3755   } else {
3756     int step = (restore_vectors ? 8 : 4) * wordSize;
3757     for (int i = 0; i <= 28; i += 4)
3758       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3759           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3760   }
3761 
3762   // We may use predicate registers and rely on ptrue with SVE,
3763   // regardless of wide vector (> 8 bytes) used or not.
3764   if (use_sve) {
3765     reinitialize_ptrue();
3766   }
3767 
3768   // integer registers except lr & sp
3769   pop(RegSet::range(r0, r17), sp);
3770 #ifdef R18_RESERVED
3771   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3772   pop(RegSet::range(r20, r29), sp);
3773 #else
3774   pop(RegSet::range(r18_tls, r29), sp);
3775 #endif
3776 }
3777 
3778 /**
3779  * Helpers for multiply_to_len().
3780  */
3781 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3782                                      Register src1, Register src2) {
3783   adds(dest_lo, dest_lo, src1);
3784   adc(dest_hi, dest_hi, zr);
3785   adds(dest_lo, dest_lo, src2);
3786   adc(final_dest_hi, dest_hi, zr);
3787 }
3788 
3789 // Generate an address from (r + r1 extend offset).  "size" is the
3790 // size of the operand.  The result may be in rscratch2.
3791 Address MacroAssembler::offsetted_address(Register r, Register r1,
3792                                           Address::extend ext, int offset, int size) {
3793   if (offset || (ext.shift() % size != 0)) {
3794     lea(rscratch2, Address(r, r1, ext));
3795     return Address(rscratch2, offset);
3796   } else {
3797     return Address(r, r1, ext);
3798   }
3799 }
3800 
3801 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3802 {
3803   assert(offset >= 0, "spill to negative address?");
3804   // Offset reachable ?
3805   //   Not aligned - 9 bits signed offset
3806   //   Aligned - 12 bits unsigned offset shifted
3807   Register base = sp;
3808   if ((offset & (size-1)) && offset >= (1<<8)) {
3809     add(tmp, base, offset & ((1<<12)-1));
3810     base = tmp;
3811     offset &= -1u<<12;
3812   }
3813 
3814   if (offset >= (1<<12) * size) {
3815     add(tmp, base, offset & (((1<<12)-1)<<12));
3816     base = tmp;
3817     offset &= ~(((1<<12)-1)<<12);
3818   }
3819 
3820   return Address(base, offset);
3821 }
3822 
3823 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3824   assert(offset >= 0, "spill to negative address?");
3825 
3826   Register base = sp;
3827 
3828   // An immediate offset in the range 0 to 255 which is multiplied
3829   // by the current vector or predicate register size in bytes.
3830   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3831     return Address(base, offset / sve_reg_size_in_bytes);
3832   }
3833 
3834   add(tmp, base, offset);
3835   return Address(tmp);
3836 }
3837 
3838 // Checks whether offset is aligned.
3839 // Returns true if it is, else false.
3840 bool MacroAssembler::merge_alignment_check(Register base,
3841                                            size_t size,
3842                                            int64_t cur_offset,
3843                                            int64_t prev_offset) const {
3844   if (AvoidUnalignedAccesses) {
3845     if (base == sp) {
3846       // Checks whether low offset if aligned to pair of registers.
3847       int64_t pair_mask = size * 2 - 1;
3848       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3849       return (offset & pair_mask) == 0;
3850     } else { // If base is not sp, we can't guarantee the access is aligned.
3851       return false;
3852     }
3853   } else {
3854     int64_t mask = size - 1;
3855     // Load/store pair instruction only supports element size aligned offset.
3856     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3857   }
3858 }
3859 
3860 // Checks whether current and previous loads/stores can be merged.
3861 // Returns true if it can be merged, else false.
3862 bool MacroAssembler::ldst_can_merge(Register rt,
3863                                     const Address &adr,
3864                                     size_t cur_size_in_bytes,
3865                                     bool is_store) const {
3866   address prev = pc() - NativeInstruction::instruction_size;
3867   address last = code()->last_insn();
3868 
3869   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3870     return false;
3871   }
3872 
3873   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3874     return false;
3875   }
3876 
3877   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3878   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3879 
3880   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3881   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3882 
3883   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3884     return false;
3885   }
3886 
3887   int64_t max_offset = 63 * prev_size_in_bytes;
3888   int64_t min_offset = -64 * prev_size_in_bytes;
3889 
3890   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3891 
3892   // Only same base can be merged.
3893   if (adr.base() != prev_ldst->base()) {
3894     return false;
3895   }
3896 
3897   int64_t cur_offset = adr.offset();
3898   int64_t prev_offset = prev_ldst->offset();
3899   size_t diff = abs(cur_offset - prev_offset);
3900   if (diff != prev_size_in_bytes) {
3901     return false;
3902   }
3903 
3904   // Following cases can not be merged:
3905   // ldr x2, [x2, #8]
3906   // ldr x3, [x2, #16]
3907   // or:
3908   // ldr x2, [x3, #8]
3909   // ldr x2, [x3, #16]
3910   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3911   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3912     return false;
3913   }
3914 
3915   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3916   // Offset range must be in ldp/stp instruction's range.
3917   if (low_offset > max_offset || low_offset < min_offset) {
3918     return false;
3919   }
3920 
3921   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3922     return true;
3923   }
3924 
3925   return false;
3926 }
3927 
3928 // Merge current load/store with previous load/store into ldp/stp.
3929 void MacroAssembler::merge_ldst(Register rt,
3930                                 const Address &adr,
3931                                 size_t cur_size_in_bytes,
3932                                 bool is_store) {
3933 
3934   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3935 
3936   Register rt_low, rt_high;
3937   address prev = pc() - NativeInstruction::instruction_size;
3938   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3939 
3940   int64_t offset;
3941 
3942   if (adr.offset() < prev_ldst->offset()) {
3943     offset = adr.offset();
3944     rt_low = rt;
3945     rt_high = prev_ldst->target();
3946   } else {
3947     offset = prev_ldst->offset();
3948     rt_low = prev_ldst->target();
3949     rt_high = rt;
3950   }
3951 
3952   Address adr_p = Address(prev_ldst->base(), offset);
3953   // Overwrite previous generated binary.
3954   code_section()->set_end(prev);
3955 
3956   const size_t sz = prev_ldst->size_in_bytes();
3957   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3958   if (!is_store) {
3959     BLOCK_COMMENT("merged ldr pair");
3960     if (sz == 8) {
3961       ldp(rt_low, rt_high, adr_p);
3962     } else {
3963       ldpw(rt_low, rt_high, adr_p);
3964     }
3965   } else {
3966     BLOCK_COMMENT("merged str pair");
3967     if (sz == 8) {
3968       stp(rt_low, rt_high, adr_p);
3969     } else {
3970       stpw(rt_low, rt_high, adr_p);
3971     }
3972   }
3973 }
3974 
3975 /**
3976  * Multiply 64 bit by 64 bit first loop.
3977  */
3978 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3979                                            Register y, Register y_idx, Register z,
3980                                            Register carry, Register product,
3981                                            Register idx, Register kdx) {
3982   //
3983   //  jlong carry, x[], y[], z[];
3984   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3985   //    huge_128 product = y[idx] * x[xstart] + carry;
3986   //    z[kdx] = (jlong)product;
3987   //    carry  = (jlong)(product >>> 64);
3988   //  }
3989   //  z[xstart] = carry;
3990   //
3991 
3992   Label L_first_loop, L_first_loop_exit;
3993   Label L_one_x, L_one_y, L_multiply;
3994 
3995   subsw(xstart, xstart, 1);
3996   br(Assembler::MI, L_one_x);
3997 
3998   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3999   ldr(x_xstart, Address(rscratch1));
4000   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
4001 
4002   bind(L_first_loop);
4003   subsw(idx, idx, 1);
4004   br(Assembler::MI, L_first_loop_exit);
4005   subsw(idx, idx, 1);
4006   br(Assembler::MI, L_one_y);
4007   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4008   ldr(y_idx, Address(rscratch1));
4009   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
4010   bind(L_multiply);
4011 
4012   // AArch64 has a multiply-accumulate instruction that we can't use
4013   // here because it has no way to process carries, so we have to use
4014   // separate add and adc instructions.  Bah.
4015   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
4016   mul(product, x_xstart, y_idx);
4017   adds(product, product, carry);
4018   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
4019 
4020   subw(kdx, kdx, 2);
4021   ror(product, product, 32); // back to big-endian
4022   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4023 
4024   b(L_first_loop);
4025 
4026   bind(L_one_y);
4027   ldrw(y_idx, Address(y,  0));
4028   b(L_multiply);
4029 
4030   bind(L_one_x);
4031   ldrw(x_xstart, Address(x,  0));
4032   b(L_first_loop);
4033 
4034   bind(L_first_loop_exit);
4035 }
4036 
4037 /**
4038  * Multiply 128 bit by 128. Unrolled inner loop.
4039  *
4040  */
4041 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4042                                              Register carry, Register carry2,
4043                                              Register idx, Register jdx,
4044                                              Register yz_idx1, Register yz_idx2,
4045                                              Register tmp, Register tmp3, Register tmp4,
4046                                              Register tmp6, Register product_hi) {
4047 
4048   //   jlong carry, x[], y[], z[];
4049   //   int kdx = ystart+1;
4050   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4051   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4052   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4053   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4054   //     carry  = (jlong)(tmp4 >>> 64);
4055   //     z[kdx+idx+1] = (jlong)tmp3;
4056   //     z[kdx+idx] = (jlong)tmp4;
4057   //   }
4058   //   idx += 2;
4059   //   if (idx > 0) {
4060   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4061   //     z[kdx+idx] = (jlong)yz_idx1;
4062   //     carry  = (jlong)(yz_idx1 >>> 64);
4063   //   }
4064   //
4065 
4066   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4067 
4068   lsrw(jdx, idx, 2);
4069 
4070   bind(L_third_loop);
4071 
4072   subsw(jdx, jdx, 1);
4073   br(Assembler::MI, L_third_loop_exit);
4074   subw(idx, idx, 4);
4075 
4076   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4077 
4078   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4079 
4080   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4081 
4082   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4083   ror(yz_idx2, yz_idx2, 32);
4084 
4085   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4086 
4087   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4088   umulh(tmp4, product_hi, yz_idx1);
4089 
4090   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4091   ror(rscratch2, rscratch2, 32);
4092 
4093   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4094   umulh(carry2, product_hi, yz_idx2);
4095 
4096   // propagate sum of both multiplications into carry:tmp4:tmp3
4097   adds(tmp3, tmp3, carry);
4098   adc(tmp4, tmp4, zr);
4099   adds(tmp3, tmp3, rscratch1);
4100   adcs(tmp4, tmp4, tmp);
4101   adc(carry, carry2, zr);
4102   adds(tmp4, tmp4, rscratch2);
4103   adc(carry, carry, zr);
4104 
4105   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4106   ror(tmp4, tmp4, 32);
4107   stp(tmp4, tmp3, Address(tmp6, 0));
4108 
4109   b(L_third_loop);
4110   bind (L_third_loop_exit);
4111 
4112   andw (idx, idx, 0x3);
4113   cbz(idx, L_post_third_loop_done);
4114 
4115   Label L_check_1;
4116   subsw(idx, idx, 2);
4117   br(Assembler::MI, L_check_1);
4118 
4119   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4120   ldr(yz_idx1, Address(rscratch1, 0));
4121   ror(yz_idx1, yz_idx1, 32);
4122   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4123   umulh(tmp4, product_hi, yz_idx1);
4124   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4125   ldr(yz_idx2, Address(rscratch1, 0));
4126   ror(yz_idx2, yz_idx2, 32);
4127 
4128   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4129 
4130   ror(tmp3, tmp3, 32);
4131   str(tmp3, Address(rscratch1, 0));
4132 
4133   bind (L_check_1);
4134 
4135   andw (idx, idx, 0x1);
4136   subsw(idx, idx, 1);
4137   br(Assembler::MI, L_post_third_loop_done);
4138   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4139   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4140   umulh(carry2, tmp4, product_hi);
4141   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4142 
4143   add2_with_carry(carry2, tmp3, tmp4, carry);
4144 
4145   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4146   extr(carry, carry2, tmp3, 32);
4147 
4148   bind(L_post_third_loop_done);
4149 }
4150 
4151 /**
4152  * Code for BigInteger::multiplyToLen() intrinsic.
4153  *
4154  * r0: x
4155  * r1: xlen
4156  * r2: y
4157  * r3: ylen
4158  * r4:  z
4159  * r5: tmp0
4160  * r10: tmp1
4161  * r11: tmp2
4162  * r12: tmp3
4163  * r13: tmp4
4164  * r14: tmp5
4165  * r15: tmp6
4166  * r16: tmp7
4167  *
4168  */
4169 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4170                                      Register z, Register tmp0,
4171                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4172                                      Register tmp5, Register tmp6, Register product_hi) {
4173 
4174   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4175 
4176   const Register idx = tmp1;
4177   const Register kdx = tmp2;
4178   const Register xstart = tmp3;
4179 
4180   const Register y_idx = tmp4;
4181   const Register carry = tmp5;
4182   const Register product  = xlen;
4183   const Register x_xstart = tmp0;
4184 
4185   // First Loop.
4186   //
4187   //  final static long LONG_MASK = 0xffffffffL;
4188   //  int xstart = xlen - 1;
4189   //  int ystart = ylen - 1;
4190   //  long carry = 0;
4191   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4192   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4193   //    z[kdx] = (int)product;
4194   //    carry = product >>> 32;
4195   //  }
4196   //  z[xstart] = (int)carry;
4197   //
4198 
4199   movw(idx, ylen);       // idx = ylen;
4200   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4201   mov(carry, zr);        // carry = 0;
4202 
4203   Label L_done;
4204 
4205   movw(xstart, xlen);
4206   subsw(xstart, xstart, 1);
4207   br(Assembler::MI, L_done);
4208 
4209   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4210 
4211   Label L_second_loop;
4212   cbzw(kdx, L_second_loop);
4213 
4214   Label L_carry;
4215   subw(kdx, kdx, 1);
4216   cbzw(kdx, L_carry);
4217 
4218   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4219   lsr(carry, carry, 32);
4220   subw(kdx, kdx, 1);
4221 
4222   bind(L_carry);
4223   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4224 
4225   // Second and third (nested) loops.
4226   //
4227   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4228   //   carry = 0;
4229   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4230   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4231   //                    (z[k] & LONG_MASK) + carry;
4232   //     z[k] = (int)product;
4233   //     carry = product >>> 32;
4234   //   }
4235   //   z[i] = (int)carry;
4236   // }
4237   //
4238   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4239 
4240   const Register jdx = tmp1;
4241 
4242   bind(L_second_loop);
4243   mov(carry, zr);                // carry = 0;
4244   movw(jdx, ylen);               // j = ystart+1
4245 
4246   subsw(xstart, xstart, 1);      // i = xstart-1;
4247   br(Assembler::MI, L_done);
4248 
4249   str(z, Address(pre(sp, -4 * wordSize)));
4250 
4251   Label L_last_x;
4252   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4253   subsw(xstart, xstart, 1);       // i = xstart-1;
4254   br(Assembler::MI, L_last_x);
4255 
4256   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4257   ldr(product_hi, Address(rscratch1));
4258   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4259 
4260   Label L_third_loop_prologue;
4261   bind(L_third_loop_prologue);
4262 
4263   str(ylen, Address(sp, wordSize));
4264   stp(x, xstart, Address(sp, 2 * wordSize));
4265   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4266                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4267   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4268   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4269 
4270   addw(tmp3, xlen, 1);
4271   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4272   subsw(tmp3, tmp3, 1);
4273   br(Assembler::MI, L_done);
4274 
4275   lsr(carry, carry, 32);
4276   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4277   b(L_second_loop);
4278 
4279   // Next infrequent code is moved outside loops.
4280   bind(L_last_x);
4281   ldrw(product_hi, Address(x,  0));
4282   b(L_third_loop_prologue);
4283 
4284   bind(L_done);
4285 }
4286 
4287 // Code for BigInteger::mulAdd intrinsic
4288 // out     = r0
4289 // in      = r1
4290 // offset  = r2  (already out.length-offset)
4291 // len     = r3
4292 // k       = r4
4293 //
4294 // pseudo code from java implementation:
4295 // carry = 0;
4296 // offset = out.length-offset - 1;
4297 // for (int j=len-1; j >= 0; j--) {
4298 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4299 //     out[offset--] = (int)product;
4300 //     carry = product >>> 32;
4301 // }
4302 // return (int)carry;
4303 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4304       Register len, Register k) {
4305     Label LOOP, END;
4306     // pre-loop
4307     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4308     csel(out, zr, out, Assembler::EQ);
4309     br(Assembler::EQ, END);
4310     add(in, in, len, LSL, 2); // in[j+1] address
4311     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4312     mov(out, zr); // used to keep carry now
4313     BIND(LOOP);
4314     ldrw(rscratch1, Address(pre(in, -4)));
4315     madd(rscratch1, rscratch1, k, out);
4316     ldrw(rscratch2, Address(pre(offset, -4)));
4317     add(rscratch1, rscratch1, rscratch2);
4318     strw(rscratch1, Address(offset));
4319     lsr(out, rscratch1, 32);
4320     subs(len, len, 1);
4321     br(Assembler::NE, LOOP);
4322     BIND(END);
4323 }
4324 
4325 /**
4326  * Emits code to update CRC-32 with a byte value according to constants in table
4327  *
4328  * @param [in,out]crc   Register containing the crc.
4329  * @param [in]val       Register containing the byte to fold into the CRC.
4330  * @param [in]table     Register containing the table of crc constants.
4331  *
4332  * uint32_t crc;
4333  * val = crc_table[(val ^ crc) & 0xFF];
4334  * crc = val ^ (crc >> 8);
4335  *
4336  */
4337 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4338   eor(val, val, crc);
4339   andr(val, val, 0xff);
4340   ldrw(val, Address(table, val, Address::lsl(2)));
4341   eor(crc, val, crc, Assembler::LSR, 8);
4342 }
4343 
4344 /**
4345  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4346  *
4347  * @param [in,out]crc   Register containing the crc.
4348  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4349  * @param [in]table0    Register containing table 0 of crc constants.
4350  * @param [in]table1    Register containing table 1 of crc constants.
4351  * @param [in]table2    Register containing table 2 of crc constants.
4352  * @param [in]table3    Register containing table 3 of crc constants.
4353  *
4354  * uint32_t crc;
4355  *   v = crc ^ v
4356  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4357  *
4358  */
4359 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4360         Register table0, Register table1, Register table2, Register table3,
4361         bool upper) {
4362   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4363   uxtb(tmp, v);
4364   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4365   ubfx(tmp, v, 8, 8);
4366   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4367   eor(crc, crc, tmp);
4368   ubfx(tmp, v, 16, 8);
4369   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4370   eor(crc, crc, tmp);
4371   ubfx(tmp, v, 24, 8);
4372   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4373   eor(crc, crc, tmp);
4374 }
4375 
4376 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4377         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4378     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4379     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4380 
4381     subs(tmp0, len, 384);
4382     mvnw(crc, crc);
4383     br(Assembler::GE, CRC_by128_pre);
4384   BIND(CRC_less128);
4385     subs(len, len, 32);
4386     br(Assembler::GE, CRC_by32_loop);
4387   BIND(CRC_less32);
4388     adds(len, len, 32 - 4);
4389     br(Assembler::GE, CRC_by4_loop);
4390     adds(len, len, 4);
4391     br(Assembler::GT, CRC_by1_loop);
4392     b(L_exit);
4393 
4394   BIND(CRC_by32_loop);
4395     ldp(tmp0, tmp1, Address(buf));
4396     crc32x(crc, crc, tmp0);
4397     ldp(tmp2, tmp3, Address(buf, 16));
4398     crc32x(crc, crc, tmp1);
4399     add(buf, buf, 32);
4400     crc32x(crc, crc, tmp2);
4401     subs(len, len, 32);
4402     crc32x(crc, crc, tmp3);
4403     br(Assembler::GE, CRC_by32_loop);
4404     cmn(len, (u1)32);
4405     br(Assembler::NE, CRC_less32);
4406     b(L_exit);
4407 
4408   BIND(CRC_by4_loop);
4409     ldrw(tmp0, Address(post(buf, 4)));
4410     subs(len, len, 4);
4411     crc32w(crc, crc, tmp0);
4412     br(Assembler::GE, CRC_by4_loop);
4413     adds(len, len, 4);
4414     br(Assembler::LE, L_exit);
4415   BIND(CRC_by1_loop);
4416     ldrb(tmp0, Address(post(buf, 1)));
4417     subs(len, len, 1);
4418     crc32b(crc, crc, tmp0);
4419     br(Assembler::GT, CRC_by1_loop);
4420     b(L_exit);
4421 
4422   BIND(CRC_by128_pre);
4423     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4424       4*256*sizeof(juint) + 8*sizeof(juint));
4425     mov(crc, 0);
4426     crc32x(crc, crc, tmp0);
4427     crc32x(crc, crc, tmp1);
4428 
4429     cbnz(len, CRC_less128);
4430 
4431   BIND(L_exit);
4432     mvnw(crc, crc);
4433 }
4434 
4435 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4436         Register len, Register tmp0, Register tmp1, Register tmp2,
4437         Register tmp3) {
4438     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4439     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4440 
4441     mvnw(crc, crc);
4442 
4443     subs(len, len, 128);
4444     br(Assembler::GE, CRC_by64_pre);
4445   BIND(CRC_less64);
4446     adds(len, len, 128-32);
4447     br(Assembler::GE, CRC_by32_loop);
4448   BIND(CRC_less32);
4449     adds(len, len, 32-4);
4450     br(Assembler::GE, CRC_by4_loop);
4451     adds(len, len, 4);
4452     br(Assembler::GT, CRC_by1_loop);
4453     b(L_exit);
4454 
4455   BIND(CRC_by32_loop);
4456     ldp(tmp0, tmp1, Address(post(buf, 16)));
4457     subs(len, len, 32);
4458     crc32x(crc, crc, tmp0);
4459     ldr(tmp2, Address(post(buf, 8)));
4460     crc32x(crc, crc, tmp1);
4461     ldr(tmp3, Address(post(buf, 8)));
4462     crc32x(crc, crc, tmp2);
4463     crc32x(crc, crc, tmp3);
4464     br(Assembler::GE, CRC_by32_loop);
4465     cmn(len, (u1)32);
4466     br(Assembler::NE, CRC_less32);
4467     b(L_exit);
4468 
4469   BIND(CRC_by4_loop);
4470     ldrw(tmp0, Address(post(buf, 4)));
4471     subs(len, len, 4);
4472     crc32w(crc, crc, tmp0);
4473     br(Assembler::GE, CRC_by4_loop);
4474     adds(len, len, 4);
4475     br(Assembler::LE, L_exit);
4476   BIND(CRC_by1_loop);
4477     ldrb(tmp0, Address(post(buf, 1)));
4478     subs(len, len, 1);
4479     crc32b(crc, crc, tmp0);
4480     br(Assembler::GT, CRC_by1_loop);
4481     b(L_exit);
4482 
4483   BIND(CRC_by64_pre);
4484     sub(buf, buf, 8);
4485     ldp(tmp0, tmp1, Address(buf, 8));
4486     crc32x(crc, crc, tmp0);
4487     ldr(tmp2, Address(buf, 24));
4488     crc32x(crc, crc, tmp1);
4489     ldr(tmp3, Address(buf, 32));
4490     crc32x(crc, crc, tmp2);
4491     ldr(tmp0, Address(buf, 40));
4492     crc32x(crc, crc, tmp3);
4493     ldr(tmp1, Address(buf, 48));
4494     crc32x(crc, crc, tmp0);
4495     ldr(tmp2, Address(buf, 56));
4496     crc32x(crc, crc, tmp1);
4497     ldr(tmp3, Address(pre(buf, 64)));
4498 
4499     b(CRC_by64_loop);
4500 
4501     align(CodeEntryAlignment);
4502   BIND(CRC_by64_loop);
4503     subs(len, len, 64);
4504     crc32x(crc, crc, tmp2);
4505     ldr(tmp0, Address(buf, 8));
4506     crc32x(crc, crc, tmp3);
4507     ldr(tmp1, Address(buf, 16));
4508     crc32x(crc, crc, tmp0);
4509     ldr(tmp2, Address(buf, 24));
4510     crc32x(crc, crc, tmp1);
4511     ldr(tmp3, Address(buf, 32));
4512     crc32x(crc, crc, tmp2);
4513     ldr(tmp0, Address(buf, 40));
4514     crc32x(crc, crc, tmp3);
4515     ldr(tmp1, Address(buf, 48));
4516     crc32x(crc, crc, tmp0);
4517     ldr(tmp2, Address(buf, 56));
4518     crc32x(crc, crc, tmp1);
4519     ldr(tmp3, Address(pre(buf, 64)));
4520     br(Assembler::GE, CRC_by64_loop);
4521 
4522     // post-loop
4523     crc32x(crc, crc, tmp2);
4524     crc32x(crc, crc, tmp3);
4525 
4526     sub(len, len, 64);
4527     add(buf, buf, 8);
4528     cmn(len, (u1)128);
4529     br(Assembler::NE, CRC_less64);
4530   BIND(L_exit);
4531     mvnw(crc, crc);
4532 }
4533 
4534 /**
4535  * @param crc   register containing existing CRC (32-bit)
4536  * @param buf   register pointing to input byte buffer (byte*)
4537  * @param len   register containing number of bytes
4538  * @param table register that will contain address of CRC table
4539  * @param tmp   scratch register
4540  */
4541 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4542         Register table0, Register table1, Register table2, Register table3,
4543         Register tmp, Register tmp2, Register tmp3) {
4544   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4545 
4546   if (UseCryptoPmullForCRC32) {
4547       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4548       return;
4549   }
4550 
4551   if (UseCRC32) {
4552       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4553       return;
4554   }
4555 
4556     mvnw(crc, crc);
4557 
4558     {
4559       uint64_t offset;
4560       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4561       add(table0, table0, offset);
4562     }
4563     add(table1, table0, 1*256*sizeof(juint));
4564     add(table2, table0, 2*256*sizeof(juint));
4565     add(table3, table0, 3*256*sizeof(juint));
4566 
4567     { // Neon code start
4568       cmp(len, (u1)64);
4569       br(Assembler::LT, L_by16);
4570       eor(v16, T16B, v16, v16);
4571 
4572     Label L_fold;
4573 
4574       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4575 
4576       ld1(v0, v1, T2D, post(buf, 32));
4577       ld1r(v4, T2D, post(tmp, 8));
4578       ld1r(v5, T2D, post(tmp, 8));
4579       ld1r(v6, T2D, post(tmp, 8));
4580       ld1r(v7, T2D, post(tmp, 8));
4581       mov(v16, S, 0, crc);
4582 
4583       eor(v0, T16B, v0, v16);
4584       sub(len, len, 64);
4585 
4586     BIND(L_fold);
4587       pmull(v22, T8H, v0, v5, T8B);
4588       pmull(v20, T8H, v0, v7, T8B);
4589       pmull(v23, T8H, v0, v4, T8B);
4590       pmull(v21, T8H, v0, v6, T8B);
4591 
4592       pmull2(v18, T8H, v0, v5, T16B);
4593       pmull2(v16, T8H, v0, v7, T16B);
4594       pmull2(v19, T8H, v0, v4, T16B);
4595       pmull2(v17, T8H, v0, v6, T16B);
4596 
4597       uzp1(v24, T8H, v20, v22);
4598       uzp2(v25, T8H, v20, v22);
4599       eor(v20, T16B, v24, v25);
4600 
4601       uzp1(v26, T8H, v16, v18);
4602       uzp2(v27, T8H, v16, v18);
4603       eor(v16, T16B, v26, v27);
4604 
4605       ushll2(v22, T4S, v20, T8H, 8);
4606       ushll(v20, T4S, v20, T4H, 8);
4607 
4608       ushll2(v18, T4S, v16, T8H, 8);
4609       ushll(v16, T4S, v16, T4H, 8);
4610 
4611       eor(v22, T16B, v23, v22);
4612       eor(v18, T16B, v19, v18);
4613       eor(v20, T16B, v21, v20);
4614       eor(v16, T16B, v17, v16);
4615 
4616       uzp1(v17, T2D, v16, v20);
4617       uzp2(v21, T2D, v16, v20);
4618       eor(v17, T16B, v17, v21);
4619 
4620       ushll2(v20, T2D, v17, T4S, 16);
4621       ushll(v16, T2D, v17, T2S, 16);
4622 
4623       eor(v20, T16B, v20, v22);
4624       eor(v16, T16B, v16, v18);
4625 
4626       uzp1(v17, T2D, v20, v16);
4627       uzp2(v21, T2D, v20, v16);
4628       eor(v28, T16B, v17, v21);
4629 
4630       pmull(v22, T8H, v1, v5, T8B);
4631       pmull(v20, T8H, v1, v7, T8B);
4632       pmull(v23, T8H, v1, v4, T8B);
4633       pmull(v21, T8H, v1, v6, T8B);
4634 
4635       pmull2(v18, T8H, v1, v5, T16B);
4636       pmull2(v16, T8H, v1, v7, T16B);
4637       pmull2(v19, T8H, v1, v4, T16B);
4638       pmull2(v17, T8H, v1, v6, T16B);
4639 
4640       ld1(v0, v1, T2D, post(buf, 32));
4641 
4642       uzp1(v24, T8H, v20, v22);
4643       uzp2(v25, T8H, v20, v22);
4644       eor(v20, T16B, v24, v25);
4645 
4646       uzp1(v26, T8H, v16, v18);
4647       uzp2(v27, T8H, v16, v18);
4648       eor(v16, T16B, v26, v27);
4649 
4650       ushll2(v22, T4S, v20, T8H, 8);
4651       ushll(v20, T4S, v20, T4H, 8);
4652 
4653       ushll2(v18, T4S, v16, T8H, 8);
4654       ushll(v16, T4S, v16, T4H, 8);
4655 
4656       eor(v22, T16B, v23, v22);
4657       eor(v18, T16B, v19, v18);
4658       eor(v20, T16B, v21, v20);
4659       eor(v16, T16B, v17, v16);
4660 
4661       uzp1(v17, T2D, v16, v20);
4662       uzp2(v21, T2D, v16, v20);
4663       eor(v16, T16B, v17, v21);
4664 
4665       ushll2(v20, T2D, v16, T4S, 16);
4666       ushll(v16, T2D, v16, T2S, 16);
4667 
4668       eor(v20, T16B, v22, v20);
4669       eor(v16, T16B, v16, v18);
4670 
4671       uzp1(v17, T2D, v20, v16);
4672       uzp2(v21, T2D, v20, v16);
4673       eor(v20, T16B, v17, v21);
4674 
4675       shl(v16, T2D, v28, 1);
4676       shl(v17, T2D, v20, 1);
4677 
4678       eor(v0, T16B, v0, v16);
4679       eor(v1, T16B, v1, v17);
4680 
4681       subs(len, len, 32);
4682       br(Assembler::GE, L_fold);
4683 
4684       mov(crc, 0);
4685       mov(tmp, v0, D, 0);
4686       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4687       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4688       mov(tmp, v0, D, 1);
4689       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4690       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4691       mov(tmp, v1, D, 0);
4692       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4693       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4694       mov(tmp, v1, D, 1);
4695       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4696       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4697 
4698       add(len, len, 32);
4699     } // Neon code end
4700 
4701   BIND(L_by16);
4702     subs(len, len, 16);
4703     br(Assembler::GE, L_by16_loop);
4704     adds(len, len, 16-4);
4705     br(Assembler::GE, L_by4_loop);
4706     adds(len, len, 4);
4707     br(Assembler::GT, L_by1_loop);
4708     b(L_exit);
4709 
4710   BIND(L_by4_loop);
4711     ldrw(tmp, Address(post(buf, 4)));
4712     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4713     subs(len, len, 4);
4714     br(Assembler::GE, L_by4_loop);
4715     adds(len, len, 4);
4716     br(Assembler::LE, L_exit);
4717   BIND(L_by1_loop);
4718     subs(len, len, 1);
4719     ldrb(tmp, Address(post(buf, 1)));
4720     update_byte_crc32(crc, tmp, table0);
4721     br(Assembler::GT, L_by1_loop);
4722     b(L_exit);
4723 
4724     align(CodeEntryAlignment);
4725   BIND(L_by16_loop);
4726     subs(len, len, 16);
4727     ldp(tmp, tmp3, Address(post(buf, 16)));
4728     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4729     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4730     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4731     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4732     br(Assembler::GE, L_by16_loop);
4733     adds(len, len, 16-4);
4734     br(Assembler::GE, L_by4_loop);
4735     adds(len, len, 4);
4736     br(Assembler::GT, L_by1_loop);
4737   BIND(L_exit);
4738     mvnw(crc, crc);
4739 }
4740 
4741 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4742         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4743     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4744     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4745 
4746     subs(tmp0, len, 384);
4747     br(Assembler::GE, CRC_by128_pre);
4748   BIND(CRC_less128);
4749     subs(len, len, 32);
4750     br(Assembler::GE, CRC_by32_loop);
4751   BIND(CRC_less32);
4752     adds(len, len, 32 - 4);
4753     br(Assembler::GE, CRC_by4_loop);
4754     adds(len, len, 4);
4755     br(Assembler::GT, CRC_by1_loop);
4756     b(L_exit);
4757 
4758   BIND(CRC_by32_loop);
4759     ldp(tmp0, tmp1, Address(buf));
4760     crc32cx(crc, crc, tmp0);
4761     ldr(tmp2, Address(buf, 16));
4762     crc32cx(crc, crc, tmp1);
4763     ldr(tmp3, Address(buf, 24));
4764     crc32cx(crc, crc, tmp2);
4765     add(buf, buf, 32);
4766     subs(len, len, 32);
4767     crc32cx(crc, crc, tmp3);
4768     br(Assembler::GE, CRC_by32_loop);
4769     cmn(len, (u1)32);
4770     br(Assembler::NE, CRC_less32);
4771     b(L_exit);
4772 
4773   BIND(CRC_by4_loop);
4774     ldrw(tmp0, Address(post(buf, 4)));
4775     subs(len, len, 4);
4776     crc32cw(crc, crc, tmp0);
4777     br(Assembler::GE, CRC_by4_loop);
4778     adds(len, len, 4);
4779     br(Assembler::LE, L_exit);
4780   BIND(CRC_by1_loop);
4781     ldrb(tmp0, Address(post(buf, 1)));
4782     subs(len, len, 1);
4783     crc32cb(crc, crc, tmp0);
4784     br(Assembler::GT, CRC_by1_loop);
4785     b(L_exit);
4786 
4787   BIND(CRC_by128_pre);
4788     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4789       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4790     mov(crc, 0);
4791     crc32cx(crc, crc, tmp0);
4792     crc32cx(crc, crc, tmp1);
4793 
4794     cbnz(len, CRC_less128);
4795 
4796   BIND(L_exit);
4797 }
4798 
4799 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4800         Register len, Register tmp0, Register tmp1, Register tmp2,
4801         Register tmp3) {
4802     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4803     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4804 
4805     subs(len, len, 128);
4806     br(Assembler::GE, CRC_by64_pre);
4807   BIND(CRC_less64);
4808     adds(len, len, 128-32);
4809     br(Assembler::GE, CRC_by32_loop);
4810   BIND(CRC_less32);
4811     adds(len, len, 32-4);
4812     br(Assembler::GE, CRC_by4_loop);
4813     adds(len, len, 4);
4814     br(Assembler::GT, CRC_by1_loop);
4815     b(L_exit);
4816 
4817   BIND(CRC_by32_loop);
4818     ldp(tmp0, tmp1, Address(post(buf, 16)));
4819     subs(len, len, 32);
4820     crc32cx(crc, crc, tmp0);
4821     ldr(tmp2, Address(post(buf, 8)));
4822     crc32cx(crc, crc, tmp1);
4823     ldr(tmp3, Address(post(buf, 8)));
4824     crc32cx(crc, crc, tmp2);
4825     crc32cx(crc, crc, tmp3);
4826     br(Assembler::GE, CRC_by32_loop);
4827     cmn(len, (u1)32);
4828     br(Assembler::NE, CRC_less32);
4829     b(L_exit);
4830 
4831   BIND(CRC_by4_loop);
4832     ldrw(tmp0, Address(post(buf, 4)));
4833     subs(len, len, 4);
4834     crc32cw(crc, crc, tmp0);
4835     br(Assembler::GE, CRC_by4_loop);
4836     adds(len, len, 4);
4837     br(Assembler::LE, L_exit);
4838   BIND(CRC_by1_loop);
4839     ldrb(tmp0, Address(post(buf, 1)));
4840     subs(len, len, 1);
4841     crc32cb(crc, crc, tmp0);
4842     br(Assembler::GT, CRC_by1_loop);
4843     b(L_exit);
4844 
4845   BIND(CRC_by64_pre);
4846     sub(buf, buf, 8);
4847     ldp(tmp0, tmp1, Address(buf, 8));
4848     crc32cx(crc, crc, tmp0);
4849     ldr(tmp2, Address(buf, 24));
4850     crc32cx(crc, crc, tmp1);
4851     ldr(tmp3, Address(buf, 32));
4852     crc32cx(crc, crc, tmp2);
4853     ldr(tmp0, Address(buf, 40));
4854     crc32cx(crc, crc, tmp3);
4855     ldr(tmp1, Address(buf, 48));
4856     crc32cx(crc, crc, tmp0);
4857     ldr(tmp2, Address(buf, 56));
4858     crc32cx(crc, crc, tmp1);
4859     ldr(tmp3, Address(pre(buf, 64)));
4860 
4861     b(CRC_by64_loop);
4862 
4863     align(CodeEntryAlignment);
4864   BIND(CRC_by64_loop);
4865     subs(len, len, 64);
4866     crc32cx(crc, crc, tmp2);
4867     ldr(tmp0, Address(buf, 8));
4868     crc32cx(crc, crc, tmp3);
4869     ldr(tmp1, Address(buf, 16));
4870     crc32cx(crc, crc, tmp0);
4871     ldr(tmp2, Address(buf, 24));
4872     crc32cx(crc, crc, tmp1);
4873     ldr(tmp3, Address(buf, 32));
4874     crc32cx(crc, crc, tmp2);
4875     ldr(tmp0, Address(buf, 40));
4876     crc32cx(crc, crc, tmp3);
4877     ldr(tmp1, Address(buf, 48));
4878     crc32cx(crc, crc, tmp0);
4879     ldr(tmp2, Address(buf, 56));
4880     crc32cx(crc, crc, tmp1);
4881     ldr(tmp3, Address(pre(buf, 64)));
4882     br(Assembler::GE, CRC_by64_loop);
4883 
4884     // post-loop
4885     crc32cx(crc, crc, tmp2);
4886     crc32cx(crc, crc, tmp3);
4887 
4888     sub(len, len, 64);
4889     add(buf, buf, 8);
4890     cmn(len, (u1)128);
4891     br(Assembler::NE, CRC_less64);
4892   BIND(L_exit);
4893 }
4894 
4895 /**
4896  * @param crc   register containing existing CRC (32-bit)
4897  * @param buf   register pointing to input byte buffer (byte*)
4898  * @param len   register containing number of bytes
4899  * @param table register that will contain address of CRC table
4900  * @param tmp   scratch register
4901  */
4902 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4903         Register table0, Register table1, Register table2, Register table3,
4904         Register tmp, Register tmp2, Register tmp3) {
4905   if (UseCryptoPmullForCRC32) {
4906     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4907   } else {
4908     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4909   }
4910 }
4911 
4912 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4913         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4914     Label CRC_by128_loop;
4915     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4916 
4917     sub(len, len, 256);
4918     Register table = tmp0;
4919     {
4920       uint64_t offset;
4921       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4922       add(table, table, offset);
4923     }
4924     add(table, table, table_offset);
4925 
4926     // Registers v0..v7 are used as data registers.
4927     // Registers v16..v31 are used as tmp registers.
4928     sub(buf, buf, 0x10);
4929     ldrq(v0, Address(buf, 0x10));
4930     ldrq(v1, Address(buf, 0x20));
4931     ldrq(v2, Address(buf, 0x30));
4932     ldrq(v3, Address(buf, 0x40));
4933     ldrq(v4, Address(buf, 0x50));
4934     ldrq(v5, Address(buf, 0x60));
4935     ldrq(v6, Address(buf, 0x70));
4936     ldrq(v7, Address(pre(buf, 0x80)));
4937 
4938     movi(v31, T4S, 0);
4939     mov(v31, S, 0, crc);
4940     eor(v0, T16B, v0, v31);
4941 
4942     // Register v16 contains constants from the crc table.
4943     ldrq(v16, Address(table));
4944     b(CRC_by128_loop);
4945 
4946     align(OptoLoopAlignment);
4947   BIND(CRC_by128_loop);
4948     pmull (v17,  T1Q, v0, v16, T1D);
4949     pmull2(v18, T1Q, v0, v16, T2D);
4950     ldrq(v0, Address(buf, 0x10));
4951     eor3(v0, T16B, v17,  v18, v0);
4952 
4953     pmull (v19, T1Q, v1, v16, T1D);
4954     pmull2(v20, T1Q, v1, v16, T2D);
4955     ldrq(v1, Address(buf, 0x20));
4956     eor3(v1, T16B, v19, v20, v1);
4957 
4958     pmull (v21, T1Q, v2, v16, T1D);
4959     pmull2(v22, T1Q, v2, v16, T2D);
4960     ldrq(v2, Address(buf, 0x30));
4961     eor3(v2, T16B, v21, v22, v2);
4962 
4963     pmull (v23, T1Q, v3, v16, T1D);
4964     pmull2(v24, T1Q, v3, v16, T2D);
4965     ldrq(v3, Address(buf, 0x40));
4966     eor3(v3, T16B, v23, v24, v3);
4967 
4968     pmull (v25, T1Q, v4, v16, T1D);
4969     pmull2(v26, T1Q, v4, v16, T2D);
4970     ldrq(v4, Address(buf, 0x50));
4971     eor3(v4, T16B, v25, v26, v4);
4972 
4973     pmull (v27, T1Q, v5, v16, T1D);
4974     pmull2(v28, T1Q, v5, v16, T2D);
4975     ldrq(v5, Address(buf, 0x60));
4976     eor3(v5, T16B, v27, v28, v5);
4977 
4978     pmull (v29, T1Q, v6, v16, T1D);
4979     pmull2(v30, T1Q, v6, v16, T2D);
4980     ldrq(v6, Address(buf, 0x70));
4981     eor3(v6, T16B, v29, v30, v6);
4982 
4983     // Reuse registers v23, v24.
4984     // Using them won't block the first instruction of the next iteration.
4985     pmull (v23, T1Q, v7, v16, T1D);
4986     pmull2(v24, T1Q, v7, v16, T2D);
4987     ldrq(v7, Address(pre(buf, 0x80)));
4988     eor3(v7, T16B, v23, v24, v7);
4989 
4990     subs(len, len, 0x80);
4991     br(Assembler::GE, CRC_by128_loop);
4992 
4993     // fold into 512 bits
4994     // Use v31 for constants because v16 can be still in use.
4995     ldrq(v31, Address(table, 0x10));
4996 
4997     pmull (v17,  T1Q, v0, v31, T1D);
4998     pmull2(v18, T1Q, v0, v31, T2D);
4999     eor3(v0, T16B, v17, v18, v4);
5000 
5001     pmull (v19, T1Q, v1, v31, T1D);
5002     pmull2(v20, T1Q, v1, v31, T2D);
5003     eor3(v1, T16B, v19, v20, v5);
5004 
5005     pmull (v21, T1Q, v2, v31, T1D);
5006     pmull2(v22, T1Q, v2, v31, T2D);
5007     eor3(v2, T16B, v21, v22, v6);
5008 
5009     pmull (v23, T1Q, v3, v31, T1D);
5010     pmull2(v24, T1Q, v3, v31, T2D);
5011     eor3(v3, T16B, v23, v24, v7);
5012 
5013     // fold into 128 bits
5014     // Use v17 for constants because v31 can be still in use.
5015     ldrq(v17, Address(table, 0x20));
5016     pmull (v25, T1Q, v0, v17, T1D);
5017     pmull2(v26, T1Q, v0, v17, T2D);
5018     eor3(v3, T16B, v3, v25, v26);
5019 
5020     // Use v18 for constants because v17 can be still in use.
5021     ldrq(v18, Address(table, 0x30));
5022     pmull (v27, T1Q, v1, v18, T1D);
5023     pmull2(v28, T1Q, v1, v18, T2D);
5024     eor3(v3, T16B, v3, v27, v28);
5025 
5026     // Use v19 for constants because v18 can be still in use.
5027     ldrq(v19, Address(table, 0x40));
5028     pmull (v29, T1Q, v2, v19, T1D);
5029     pmull2(v30, T1Q, v2, v19, T2D);
5030     eor3(v0, T16B, v3, v29, v30);
5031 
5032     add(len, len, 0x80);
5033     add(buf, buf, 0x10);
5034 
5035     mov(tmp0, v0, D, 0);
5036     mov(tmp1, v0, D, 1);
5037 }
5038 
5039 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5040   Address adr;
5041   switch(dst.getMode()) {
5042   case Address::base_plus_offset:
5043     // This is the expected mode, although we allow all the other
5044     // forms below.
5045     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5046     break;
5047   default:
5048     lea(rscratch2, dst);
5049     adr = Address(rscratch2);
5050     break;
5051   }
5052   ldr(rscratch1, adr);
5053   add(rscratch1, rscratch1, src);
5054   str(rscratch1, adr);
5055 }
5056 
5057 void MacroAssembler::cmpptr(Register src1, Address src2) {
5058   uint64_t offset;
5059   adrp(rscratch1, src2, offset);
5060   ldr(rscratch1, Address(rscratch1, offset));
5061   cmp(src1, rscratch1);
5062 }
5063 
5064 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5065   cmp(obj1, obj2);
5066 }
5067 
5068 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5069   load_method_holder(rresult, rmethod);
5070   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5071 }
5072 
5073 void MacroAssembler::load_method_holder(Register holder, Register method) {
5074   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5075   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5076   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5077 }
5078 
5079 // Loads the obj's Klass* into dst.
5080 // Preserves all registers (incl src, rscratch1 and rscratch2).
5081 // Input:
5082 // src - the oop we want to load the klass from.
5083 // dst - output narrow klass.
5084 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5085   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5086   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5087   lsr(dst, dst, markWord::klass_shift);
5088 }
5089 
5090 void MacroAssembler::load_klass(Register dst, Register src) {
5091   if (UseCompactObjectHeaders) {
5092     load_narrow_klass_compact(dst, src);
5093     decode_klass_not_null(dst);
5094   } else if (UseCompressedClassPointers) {
5095     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5096     decode_klass_not_null(dst);
5097   } else {
5098     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5099   }
5100 }
5101 
5102 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5103   if (RestoreMXCSROnJNICalls) {
5104     Label OK;
5105     get_fpcr(tmp1);
5106     mov(tmp2, tmp1);
5107     // Set FPCR to the state we need. We do want Round to Nearest. We
5108     // don't want non-IEEE rounding modes or floating-point traps.
5109     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5110     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5111     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5112     eor(tmp2, tmp1, tmp2);
5113     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5114     set_fpcr(tmp1);
5115     bind(OK);
5116   }
5117 }
5118 
5119 // ((OopHandle)result).resolve();
5120 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5121   // OopHandle::resolve is an indirection.
5122   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5123 }
5124 
5125 // ((WeakHandle)result).resolve();
5126 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5127   assert_different_registers(result, tmp1, tmp2);
5128   Label resolved;
5129 
5130   // A null weak handle resolves to null.
5131   cbz(result, resolved);
5132 
5133   // Only 64 bit platforms support GCs that require a tmp register
5134   // WeakHandle::resolve is an indirection like jweak.
5135   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5136                  result, Address(result), tmp1, tmp2);
5137   bind(resolved);
5138 }
5139 
5140 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5141   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5142   ldr(dst, Address(rmethod, Method::const_offset()));
5143   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5144   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5145   ldr(dst, Address(dst, mirror_offset));
5146   resolve_oop_handle(dst, tmp1, tmp2);
5147 }
5148 
5149 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5150   assert_different_registers(obj, klass, tmp);
5151   if (UseCompressedClassPointers) {
5152     if (UseCompactObjectHeaders) {
5153       load_narrow_klass_compact(tmp, obj);
5154     } else {
5155       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5156     }
5157     if (CompressedKlassPointers::base() == nullptr) {
5158       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5159       return;
5160     } else if (!AOTCodeCache::is_on_for_dump() &&
5161                ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5162                && CompressedKlassPointers::shift() == 0) {
5163       // Only the bottom 32 bits matter
5164       cmpw(klass, tmp);
5165       return;
5166     }
5167     decode_klass_not_null(tmp);
5168   } else {
5169     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5170   }
5171   cmp(klass, tmp);
5172 }
5173 
5174 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5175   if (UseCompactObjectHeaders) {
5176     load_narrow_klass_compact(tmp1, obj1);
5177     load_narrow_klass_compact(tmp2,  obj2);
5178     cmpw(tmp1, tmp2);
5179   } else if (UseCompressedClassPointers) {
5180     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5181     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5182     cmpw(tmp1, tmp2);
5183   } else {
5184     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5185     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5186     cmp(tmp1, tmp2);
5187   }
5188 }
5189 
5190 void MacroAssembler::store_klass(Register dst, Register src) {
5191   // FIXME: Should this be a store release?  concurrent gcs assumes
5192   // klass length is valid if klass field is not null.
5193   assert(!UseCompactObjectHeaders, "not with compact headers");
5194   if (UseCompressedClassPointers) {
5195     encode_klass_not_null(src);
5196     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5197   } else {
5198     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5199   }
5200 }
5201 
5202 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5203   assert(!UseCompactObjectHeaders, "not with compact headers");
5204   if (UseCompressedClassPointers) {
5205     // Store to klass gap in destination
5206     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5207   }
5208 }
5209 
5210 // Algorithm must match CompressedOops::encode.
5211 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5212 #ifdef ASSERT
5213   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5214 #endif
5215   verify_oop_msg(s, "broken oop in encode_heap_oop");
5216   if (CompressedOops::base() == nullptr) {
5217     if (CompressedOops::shift() != 0) {
5218       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5219       lsr(d, s, LogMinObjAlignmentInBytes);
5220     } else {
5221       mov(d, s);
5222     }
5223   } else {
5224     subs(d, s, rheapbase);
5225     csel(d, d, zr, Assembler::HS);
5226     lsr(d, d, LogMinObjAlignmentInBytes);
5227 
5228     /*  Old algorithm: is this any worse?
5229     Label nonnull;
5230     cbnz(r, nonnull);
5231     sub(r, r, rheapbase);
5232     bind(nonnull);
5233     lsr(r, r, LogMinObjAlignmentInBytes);
5234     */
5235   }
5236 }
5237 
5238 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5239 #ifdef ASSERT
5240   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5241   if (CheckCompressedOops) {
5242     Label ok;
5243     cbnz(r, ok);
5244     stop("null oop passed to encode_heap_oop_not_null");
5245     bind(ok);
5246   }
5247 #endif
5248   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5249   if (CompressedOops::base() != nullptr) {
5250     sub(r, r, rheapbase);
5251   }
5252   if (CompressedOops::shift() != 0) {
5253     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5254     lsr(r, r, LogMinObjAlignmentInBytes);
5255   }
5256 }
5257 
5258 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5259 #ifdef ASSERT
5260   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5261   if (CheckCompressedOops) {
5262     Label ok;
5263     cbnz(src, ok);
5264     stop("null oop passed to encode_heap_oop_not_null2");
5265     bind(ok);
5266   }
5267 #endif
5268   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5269 
5270   Register data = src;
5271   if (CompressedOops::base() != nullptr) {
5272     sub(dst, src, rheapbase);
5273     data = dst;
5274   }
5275   if (CompressedOops::shift() != 0) {
5276     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5277     lsr(dst, data, LogMinObjAlignmentInBytes);
5278     data = dst;
5279   }
5280   if (data == src)
5281     mov(dst, src);
5282 }
5283 
5284 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5285 #ifdef ASSERT
5286   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5287 #endif
5288   if (CompressedOops::base() == nullptr) {
5289     if (CompressedOops::shift() != 0) {
5290       lsl(d, s, CompressedOops::shift());
5291     } else if (d != s) {
5292       mov(d, s);
5293     }
5294   } else {
5295     Label done;
5296     if (d != s)
5297       mov(d, s);
5298     cbz(s, done);
5299     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5300     bind(done);
5301   }
5302   verify_oop_msg(d, "broken oop in decode_heap_oop");
5303 }
5304 
5305 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5306   assert (UseCompressedOops, "should only be used for compressed headers");
5307   assert (Universe::heap() != nullptr, "java heap should be initialized");
5308   // Cannot assert, unverified entry point counts instructions (see .ad file)
5309   // vtableStubs also counts instructions in pd_code_size_limit.
5310   // Also do not verify_oop as this is called by verify_oop.
5311   if (CompressedOops::shift() != 0) {
5312     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5313     if (CompressedOops::base() != nullptr) {
5314       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5315     } else {
5316       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5317     }
5318   } else {
5319     assert (CompressedOops::base() == nullptr, "sanity");
5320   }
5321 }
5322 
5323 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5324   assert (UseCompressedOops, "should only be used for compressed headers");
5325   assert (Universe::heap() != nullptr, "java heap should be initialized");
5326   // Cannot assert, unverified entry point counts instructions (see .ad file)
5327   // vtableStubs also counts instructions in pd_code_size_limit.
5328   // Also do not verify_oop as this is called by verify_oop.
5329   if (CompressedOops::shift() != 0) {
5330     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5331     if (CompressedOops::base() != nullptr) {
5332       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5333     } else {
5334       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5335     }
5336   } else {
5337     assert (CompressedOops::base() == nullptr, "sanity");
5338     if (dst != src) {
5339       mov(dst, src);
5340     }
5341   }
5342 }
5343 
5344 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5345 
5346 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5347   assert(Metaspace::initialized(), "metaspace not initialized yet");
5348   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5349   return _klass_decode_mode;
5350 }
5351 
5352 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5353   assert(UseCompressedClassPointers, "not using compressed class pointers");
5354 
5355   // KlassDecodeMode shouldn't be set already.
5356   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5357 
5358   if (base == nullptr) {
5359     return KlassDecodeZero;
5360   }
5361 
5362   if (operand_valid_for_logical_immediate(
5363         /*is32*/false, (uint64_t)base)) {
5364     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5365     if (((uint64_t)base & range_mask) == 0) {
5366       return KlassDecodeXor;
5367     }
5368   }
5369 
5370   const uint64_t shifted_base =
5371     (uint64_t)base >> shift;
5372   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5373     return KlassDecodeMovk;
5374   }
5375 
5376   // No valid encoding.
5377   return KlassDecodeNone;
5378 }
5379 
5380 // Check if one of the above decoding modes will work for given base, shift and range.
5381 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5382   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5383 }
5384 
5385 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5386   _klass_decode_mode = klass_decode_mode(base, shift, range);
5387   return _klass_decode_mode != KlassDecodeNone;
5388 }
5389 
5390 static Register pick_different_tmp(Register dst, Register src) {
5391   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5392   return *tmps.begin();
5393 }
5394 
5395 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5396   // we have to load the klass base from the AOT constants area but
5397   // not the shift because it is not allowed to change
5398   int shift = CompressedKlassPointers::shift();
5399   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5400   if (dst != src) {
5401     // we can load the base into dst, subtract it formthe src and shift down
5402     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5403     ldr(dst, dst);
5404     sub(dst, src, dst);
5405     lsr(dst, dst, shift);
5406   } else {
5407     // we need an extra register in order to load the coop base
5408     Register tmp = pick_different_tmp(dst, src);
5409     RegSet regs = RegSet::of(tmp);
5410     push(regs, sp);
5411     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5412     ldr(tmp, tmp);
5413     sub(dst, src, tmp);
5414     lsr(dst, dst, shift);
5415     pop(regs, sp);
5416   }
5417 }
5418 
5419 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5420   if (CompressedKlassPointers::base() != nullptr && AOTCodeCache::is_on_for_dump()) {
5421     encode_klass_not_null_for_aot(dst, src);
5422     return;
5423   }
5424 
5425   switch (klass_decode_mode()) {
5426   case KlassDecodeZero:
5427     if (CompressedKlassPointers::shift() != 0) {
5428       lsr(dst, src, CompressedKlassPointers::shift());
5429     } else {
5430       if (dst != src) mov(dst, src);
5431     }
5432     break;
5433 
5434   case KlassDecodeXor:
5435     if (CompressedKlassPointers::shift() != 0) {
5436       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5437       lsr(dst, dst, CompressedKlassPointers::shift());
5438     } else {
5439       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5440     }
5441     break;
5442 
5443   case KlassDecodeMovk:
5444     if (CompressedKlassPointers::shift() != 0) {
5445       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5446     } else {
5447       movw(dst, src);
5448     }
5449     break;
5450 
5451   case KlassDecodeNone:
5452     ShouldNotReachHere();
5453     break;
5454   }
5455 }
5456 
5457 void MacroAssembler::encode_klass_not_null(Register r) {
5458   encode_klass_not_null(r, r);
5459 }
5460 
5461 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5462   // we have to load the klass base from the AOT constants area but
5463   // not the shift because it is not allowed to change
5464   int shift = CompressedKlassPointers::shift();
5465   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5466   if (dst != src) {
5467     // we can load the base into dst then add the offset with a suitable shift
5468     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5469     ldr(dst, dst);
5470     add(dst, dst, src, LSL,  shift);
5471   } else {
5472     // we need an extra register in order to load the coop base
5473     Register tmp = pick_different_tmp(dst, src);
5474     RegSet regs = RegSet::of(tmp);
5475     push(regs, sp);
5476     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5477     ldr(tmp, tmp);
5478     add(dst, tmp,  src, LSL,  shift);
5479     pop(regs, sp);
5480   }
5481 }
5482 
5483 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5484   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5485 
5486   if (CompressedKlassPointers::base() != nullptr && AOTCodeCache::is_on_for_dump()) {
5487     decode_klass_not_null_for_aot(dst, src);
5488     return;
5489   }
5490 
5491   switch (klass_decode_mode()) {
5492   case KlassDecodeZero:
5493     if (CompressedKlassPointers::shift() != 0) {
5494       lsl(dst, src, CompressedKlassPointers::shift());
5495     } else {
5496       if (dst != src) mov(dst, src);
5497     }
5498     break;
5499 
5500   case KlassDecodeXor:
5501     if (CompressedKlassPointers::shift() != 0) {
5502       lsl(dst, src, CompressedKlassPointers::shift());
5503       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5504     } else {
5505       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5506     }
5507     break;
5508 
5509   case KlassDecodeMovk: {
5510     const uint64_t shifted_base =
5511       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5512 
5513     if (dst != src) movw(dst, src);
5514     movk(dst, shifted_base >> 32, 32);
5515 
5516     if (CompressedKlassPointers::shift() != 0) {
5517       lsl(dst, dst, CompressedKlassPointers::shift());
5518     }
5519 
5520     break;
5521   }
5522 
5523   case KlassDecodeNone:
5524     ShouldNotReachHere();
5525     break;
5526   }
5527 }
5528 
5529 void  MacroAssembler::decode_klass_not_null(Register r) {
5530   decode_klass_not_null(r, r);
5531 }
5532 
5533 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5534 #ifdef ASSERT
5535   {
5536     ThreadInVMfromUnknown tiv;
5537     assert (UseCompressedOops, "should only be used for compressed oops");
5538     assert (Universe::heap() != nullptr, "java heap should be initialized");
5539     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5540     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5541   }
5542 #endif
5543   int oop_index = oop_recorder()->find_index(obj);
5544   InstructionMark im(this);
5545   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5546   code_section()->relocate(inst_mark(), rspec);
5547   movz(dst, 0xDEAD, 16);
5548   movk(dst, 0xBEEF);
5549 }
5550 
5551 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5552   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5553   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5554   int index = oop_recorder()->find_index(k);
5555 
5556   InstructionMark im(this);
5557   RelocationHolder rspec = metadata_Relocation::spec(index);
5558   code_section()->relocate(inst_mark(), rspec);
5559   narrowKlass nk = CompressedKlassPointers::encode(k);
5560   movz(dst, (nk >> 16), 16);
5561   movk(dst, nk & 0xffff);
5562 }
5563 
5564 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5565                                     Register dst, Address src,
5566                                     Register tmp1, Register tmp2) {
5567   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5568   decorators = AccessInternal::decorator_fixup(decorators, type);
5569   bool as_raw = (decorators & AS_RAW) != 0;
5570   if (as_raw) {
5571     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5572   } else {
5573     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5574   }
5575 }
5576 
5577 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5578                                      Address dst, Register val,
5579                                      Register tmp1, Register tmp2, Register tmp3) {
5580   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5581   decorators = AccessInternal::decorator_fixup(decorators, type);
5582   bool as_raw = (decorators & AS_RAW) != 0;
5583   if (as_raw) {
5584     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5585   } else {
5586     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5587   }
5588 }
5589 
5590 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5591                                    Register tmp2, DecoratorSet decorators) {
5592   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5593 }
5594 
5595 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5596                                             Register tmp2, DecoratorSet decorators) {
5597   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5598 }
5599 
5600 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5601                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5602   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5603 }
5604 
5605 // Used for storing nulls.
5606 void MacroAssembler::store_heap_oop_null(Address dst) {
5607   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5608 }
5609 
5610 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5611   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5612   int index = oop_recorder()->allocate_metadata_index(obj);
5613   RelocationHolder rspec = metadata_Relocation::spec(index);
5614   return Address((address)obj, rspec);
5615 }
5616 
5617 // Move an oop into a register.
5618 void MacroAssembler::movoop(Register dst, jobject obj) {
5619   int oop_index;
5620   if (obj == nullptr) {
5621     oop_index = oop_recorder()->allocate_oop_index(obj);
5622   } else {
5623 #ifdef ASSERT
5624     {
5625       ThreadInVMfromUnknown tiv;
5626       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5627     }
5628 #endif
5629     oop_index = oop_recorder()->find_index(obj);
5630   }
5631   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5632 
5633   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5634     mov(dst, Address((address)obj, rspec));
5635   } else {
5636     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5637     ldr(dst, Address(dummy, rspec));
5638   }
5639 }
5640 
5641 // Move a metadata address into a register.
5642 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5643   int oop_index;
5644   if (obj == nullptr) {
5645     oop_index = oop_recorder()->allocate_metadata_index(obj);
5646   } else {
5647     oop_index = oop_recorder()->find_index(obj);
5648   }
5649   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5650   mov(dst, Address((address)obj, rspec));
5651 }
5652 
5653 Address MacroAssembler::constant_oop_address(jobject obj) {
5654 #ifdef ASSERT
5655   {
5656     ThreadInVMfromUnknown tiv;
5657     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5658     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5659   }
5660 #endif
5661   int oop_index = oop_recorder()->find_index(obj);
5662   return Address((address)obj, oop_Relocation::spec(oop_index));
5663 }
5664 
5665 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5666 void MacroAssembler::tlab_allocate(Register obj,
5667                                    Register var_size_in_bytes,
5668                                    int con_size_in_bytes,
5669                                    Register t1,
5670                                    Register t2,
5671                                    Label& slow_case) {
5672   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5673   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5674 }
5675 
5676 void MacroAssembler::verify_tlab() {
5677 #ifdef ASSERT
5678   if (UseTLAB && VerifyOops) {
5679     Label next, ok;
5680 
5681     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5682 
5683     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5684     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5685     cmp(rscratch2, rscratch1);
5686     br(Assembler::HS, next);
5687     STOP("assert(top >= start)");
5688     should_not_reach_here();
5689 
5690     bind(next);
5691     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5692     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5693     cmp(rscratch2, rscratch1);
5694     br(Assembler::HS, ok);
5695     STOP("assert(top <= end)");
5696     should_not_reach_here();
5697 
5698     bind(ok);
5699     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5700   }
5701 #endif
5702 }
5703 
5704 // Writes to stack successive pages until offset reached to check for
5705 // stack overflow + shadow pages.  This clobbers tmp.
5706 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5707   assert_different_registers(tmp, size, rscratch1);
5708   mov(tmp, sp);
5709   // Bang stack for total size given plus shadow page size.
5710   // Bang one page at a time because large size can bang beyond yellow and
5711   // red zones.
5712   Label loop;
5713   mov(rscratch1, (int)os::vm_page_size());
5714   bind(loop);
5715   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5716   subsw(size, size, rscratch1);
5717   str(size, Address(tmp));
5718   br(Assembler::GT, loop);
5719 
5720   // Bang down shadow pages too.
5721   // At this point, (tmp-0) is the last address touched, so don't
5722   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5723   // was post-decremented.)  Skip this address by starting at i=1, and
5724   // touch a few more pages below.  N.B.  It is important to touch all
5725   // the way down to and including i=StackShadowPages.
5726   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5727     // this could be any sized move but this is can be a debugging crumb
5728     // so the bigger the better.
5729     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5730     str(size, Address(tmp));
5731   }
5732 }
5733 
5734 // Move the address of the polling page into dest.
5735 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5736   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5737 }
5738 
5739 // Read the polling page.  The address of the polling page must
5740 // already be in r.
5741 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5742   address mark;
5743   {
5744     InstructionMark im(this);
5745     code_section()->relocate(inst_mark(), rtype);
5746     ldrw(zr, Address(r, 0));
5747     mark = inst_mark();
5748   }
5749   verify_cross_modify_fence_not_required();
5750   return mark;
5751 }
5752 
5753 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5754   relocInfo::relocType rtype = dest.rspec().reloc()->type();
5755   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5756   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5757   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5758   int64_t offset_low = dest_page - low_page;
5759   int64_t offset_high = dest_page - high_page;
5760 
5761   assert(is_valid_AArch64_address(dest.target()), "bad address");
5762   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5763 
5764   InstructionMark im(this);
5765   code_section()->relocate(inst_mark(), dest.rspec());
5766   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5767   // the code cache so that if it is relocated we know it will still reach
5768   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5769     _adrp(reg1, dest.target());
5770   } else {
5771     uint64_t target = (uint64_t)dest.target();
5772     uint64_t adrp_target
5773       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5774 
5775     _adrp(reg1, (address)adrp_target);
5776     movk(reg1, target >> 32, 32);
5777   }
5778   byte_offset = (uint64_t)dest.target() & 0xfff;
5779 }
5780 
5781 void MacroAssembler::load_byte_map_base(Register reg) {
5782 #if INCLUDE_CDS
5783   if (AOTCodeCache::is_on_for_dump()) {
5784     address byte_map_base_adr = AOTRuntimeConstants::card_table_address();
5785     lea(reg, ExternalAddress(byte_map_base_adr));
5786     ldr(reg, Address(reg));
5787     return;
5788   }
5789 #endif
5790   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
5791 
5792   // Strictly speaking the card table base isn't an address at all, and it might
5793   // even be negative. It is thus materialised as a constant.
5794   mov(reg, (uint64_t)ctbs->card_table_base_const());
5795 }
5796 
5797 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5798 #if INCLUDE_CDS
5799   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5800   if (AOTCodeCache::is_on_for_dump()) {
5801     // all aotrc field addresses should be registered in the AOTCodeCache address table
5802     lea(reg, ExternalAddress(a));
5803   } else {
5804     mov(reg, (uint64_t)a);
5805   }
5806 #else
5807   ShouldNotReachHere();
5808 #endif
5809 }
5810 
5811 void MacroAssembler::build_frame(int framesize) {
5812   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5813   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5814   protect_return_address();
5815   if (framesize < ((1 << 9) + 2 * wordSize)) {
5816     sub(sp, sp, framesize);
5817     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5818     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5819   } else {
5820     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5821     if (PreserveFramePointer) mov(rfp, sp);
5822     if (framesize < ((1 << 12) + 2 * wordSize))
5823       sub(sp, sp, framesize - 2 * wordSize);
5824     else {
5825       mov(rscratch1, framesize - 2 * wordSize);
5826       sub(sp, sp, rscratch1);
5827     }
5828   }
5829   verify_cross_modify_fence_not_required();
5830 }
5831 
5832 void MacroAssembler::remove_frame(int framesize) {
5833   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5834   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5835   if (framesize < ((1 << 9) + 2 * wordSize)) {
5836     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5837     add(sp, sp, framesize);
5838   } else {
5839     if (framesize < ((1 << 12) + 2 * wordSize))
5840       add(sp, sp, framesize - 2 * wordSize);
5841     else {
5842       mov(rscratch1, framesize - 2 * wordSize);
5843       add(sp, sp, rscratch1);
5844     }
5845     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5846   }
5847   authenticate_return_address();
5848 }
5849 
5850 
5851 // This method counts leading positive bytes (highest bit not set) in provided byte array
5852 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5853     // Simple and most common case of aligned small array which is not at the
5854     // end of memory page is placed here. All other cases are in stub.
5855     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5856     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5857     assert_different_registers(ary1, len, result);
5858 
5859     mov(result, len);
5860     cmpw(len, 0);
5861     br(LE, DONE);
5862     cmpw(len, 4 * wordSize);
5863     br(GE, STUB_LONG); // size > 32 then go to stub
5864 
5865     int shift = 64 - exact_log2(os::vm_page_size());
5866     lsl(rscratch1, ary1, shift);
5867     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5868     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5869     br(CS, STUB); // at the end of page then go to stub
5870     subs(len, len, wordSize);
5871     br(LT, END);
5872 
5873   BIND(LOOP);
5874     ldr(rscratch1, Address(post(ary1, wordSize)));
5875     tst(rscratch1, UPPER_BIT_MASK);
5876     br(NE, SET_RESULT);
5877     subs(len, len, wordSize);
5878     br(GE, LOOP);
5879     cmpw(len, -wordSize);
5880     br(EQ, DONE);
5881 
5882   BIND(END);
5883     ldr(rscratch1, Address(ary1));
5884     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5885     lslv(rscratch1, rscratch1, rscratch2);
5886     tst(rscratch1, UPPER_BIT_MASK);
5887     br(NE, SET_RESULT);
5888     b(DONE);
5889 
5890   BIND(STUB);
5891     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5892     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5893     address tpc1 = trampoline_call(count_pos);
5894     if (tpc1 == nullptr) {
5895       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5896       postcond(pc() == badAddress);
5897       return nullptr;
5898     }
5899     b(DONE);
5900 
5901   BIND(STUB_LONG);
5902     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5903     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5904     address tpc2 = trampoline_call(count_pos_long);
5905     if (tpc2 == nullptr) {
5906       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5907       postcond(pc() == badAddress);
5908       return nullptr;
5909     }
5910     b(DONE);
5911 
5912   BIND(SET_RESULT);
5913 
5914     add(len, len, wordSize);
5915     sub(result, result, len);
5916 
5917   BIND(DONE);
5918   postcond(pc() != badAddress);
5919   return pc();
5920 }
5921 
5922 // Clobbers: rscratch1, rscratch2, rflags
5923 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5924 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5925                                       Register tmp4, Register tmp5, Register result,
5926                                       Register cnt1, int elem_size) {
5927   Label DONE, SAME;
5928   Register tmp1 = rscratch1;
5929   Register tmp2 = rscratch2;
5930   int elem_per_word = wordSize/elem_size;
5931   int log_elem_size = exact_log2(elem_size);
5932   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5933   int length_offset = arrayOopDesc::length_offset_in_bytes();
5934   int base_offset
5935     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5936   // When the length offset is not aligned to 8 bytes,
5937   // then we align it down. This is valid because the new
5938   // offset will always be the klass which is the same
5939   // for type arrays.
5940   int start_offset = align_down(length_offset, BytesPerWord);
5941   int extra_length = base_offset - start_offset;
5942   assert(start_offset == length_offset || start_offset == klass_offset,
5943          "start offset must be 8-byte-aligned or be the klass offset");
5944   assert(base_offset != start_offset, "must include the length field");
5945   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5946   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5947 
5948   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5949   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5950 
5951 #ifndef PRODUCT
5952   {
5953     const char kind = (elem_size == 2) ? 'U' : 'L';
5954     char comment[64];
5955     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5956     BLOCK_COMMENT(comment);
5957   }
5958 #endif
5959 
5960   // if (a1 == a2)
5961   //     return true;
5962   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5963   br(EQ, SAME);
5964 
5965   if (UseSimpleArrayEquals) {
5966     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5967     // if (a1 == nullptr || a2 == nullptr)
5968     //     return false;
5969     // a1 & a2 == 0 means (some-pointer is null) or
5970     // (very-rare-or-even-probably-impossible-pointer-values)
5971     // so, we can save one branch in most cases
5972     tst(a1, a2);
5973     mov(result, false);
5974     br(EQ, A_MIGHT_BE_NULL);
5975     // if (a1.length != a2.length)
5976     //      return false;
5977     bind(A_IS_NOT_NULL);
5978     ldrw(cnt1, Address(a1, length_offset));
5979     ldrw(tmp5, Address(a2, length_offset));
5980     cmp(cnt1, tmp5);
5981     br(NE, DONE); // If lengths differ, return false
5982     // Increase loop counter by diff between base- and actual start-offset.
5983     addw(cnt1, cnt1, extra_length);
5984     lea(a1, Address(a1, start_offset));
5985     lea(a2, Address(a2, start_offset));
5986     // Check for short strings, i.e. smaller than wordSize.
5987     subs(cnt1, cnt1, elem_per_word);
5988     br(Assembler::LT, SHORT);
5989     // Main 8 byte comparison loop.
5990     bind(NEXT_WORD); {
5991       ldr(tmp1, Address(post(a1, wordSize)));
5992       ldr(tmp2, Address(post(a2, wordSize)));
5993       subs(cnt1, cnt1, elem_per_word);
5994       eor(tmp5, tmp1, tmp2);
5995       cbnz(tmp5, DONE);
5996     } br(GT, NEXT_WORD);
5997     // Last longword.  In the case where length == 4 we compare the
5998     // same longword twice, but that's still faster than another
5999     // conditional branch.
6000     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6001     // length == 4.
6002     if (log_elem_size > 0)
6003       lsl(cnt1, cnt1, log_elem_size);
6004     ldr(tmp3, Address(a1, cnt1));
6005     ldr(tmp4, Address(a2, cnt1));
6006     eor(tmp5, tmp3, tmp4);
6007     cbnz(tmp5, DONE);
6008     b(SAME);
6009     bind(A_MIGHT_BE_NULL);
6010     // in case both a1 and a2 are not-null, proceed with loads
6011     cbz(a1, DONE);
6012     cbz(a2, DONE);
6013     b(A_IS_NOT_NULL);
6014     bind(SHORT);
6015 
6016     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
6017     {
6018       ldrw(tmp1, Address(post(a1, 4)));
6019       ldrw(tmp2, Address(post(a2, 4)));
6020       eorw(tmp5, tmp1, tmp2);
6021       cbnzw(tmp5, DONE);
6022     }
6023     bind(TAIL03);
6024     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6025     {
6026       ldrh(tmp3, Address(post(a1, 2)));
6027       ldrh(tmp4, Address(post(a2, 2)));
6028       eorw(tmp5, tmp3, tmp4);
6029       cbnzw(tmp5, DONE);
6030     }
6031     bind(TAIL01);
6032     if (elem_size == 1) { // Only needed when comparing byte arrays.
6033       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6034       {
6035         ldrb(tmp1, a1);
6036         ldrb(tmp2, a2);
6037         eorw(tmp5, tmp1, tmp2);
6038         cbnzw(tmp5, DONE);
6039       }
6040     }
6041   } else {
6042     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6043         CSET_EQ, LAST_CHECK;
6044     mov(result, false);
6045     cbz(a1, DONE);
6046     ldrw(cnt1, Address(a1, length_offset));
6047     cbz(a2, DONE);
6048     ldrw(tmp5, Address(a2, length_offset));
6049     cmp(cnt1, tmp5);
6050     br(NE, DONE); // If lengths differ, return false
6051     // Increase loop counter by diff between base- and actual start-offset.
6052     addw(cnt1, cnt1, extra_length);
6053 
6054     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6055     // faster to perform another branch before comparing a1 and a2
6056     cmp(cnt1, (u1)elem_per_word);
6057     br(LE, SHORT); // short or same
6058     ldr(tmp3, Address(pre(a1, start_offset)));
6059     subs(zr, cnt1, stubBytesThreshold);
6060     br(GE, STUB);
6061     ldr(tmp4, Address(pre(a2, start_offset)));
6062     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6063 
6064     // Main 16 byte comparison loop with 2 exits
6065     bind(NEXT_DWORD); {
6066       ldr(tmp1, Address(pre(a1, wordSize)));
6067       ldr(tmp2, Address(pre(a2, wordSize)));
6068       subs(cnt1, cnt1, 2 * elem_per_word);
6069       br(LE, TAIL);
6070       eor(tmp4, tmp3, tmp4);
6071       cbnz(tmp4, DONE);
6072       ldr(tmp3, Address(pre(a1, wordSize)));
6073       ldr(tmp4, Address(pre(a2, wordSize)));
6074       cmp(cnt1, (u1)elem_per_word);
6075       br(LE, TAIL2);
6076       cmp(tmp1, tmp2);
6077     } br(EQ, NEXT_DWORD);
6078     b(DONE);
6079 
6080     bind(TAIL);
6081     eor(tmp4, tmp3, tmp4);
6082     eor(tmp2, tmp1, tmp2);
6083     lslv(tmp2, tmp2, tmp5);
6084     orr(tmp5, tmp4, tmp2);
6085     cmp(tmp5, zr);
6086     b(CSET_EQ);
6087 
6088     bind(TAIL2);
6089     eor(tmp2, tmp1, tmp2);
6090     cbnz(tmp2, DONE);
6091     b(LAST_CHECK);
6092 
6093     bind(STUB);
6094     ldr(tmp4, Address(pre(a2, start_offset)));
6095     if (elem_size == 2) { // convert to byte counter
6096       lsl(cnt1, cnt1, 1);
6097     }
6098     eor(tmp5, tmp3, tmp4);
6099     cbnz(tmp5, DONE);
6100     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6101     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6102     address tpc = trampoline_call(stub);
6103     if (tpc == nullptr) {
6104       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6105       postcond(pc() == badAddress);
6106       return nullptr;
6107     }
6108     b(DONE);
6109 
6110     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6111     // so, if a2 == null => return false(0), else return true, so we can return a2
6112     mov(result, a2);
6113     b(DONE);
6114     bind(SHORT);
6115     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6116     ldr(tmp3, Address(a1, start_offset));
6117     ldr(tmp4, Address(a2, start_offset));
6118     bind(LAST_CHECK);
6119     eor(tmp4, tmp3, tmp4);
6120     lslv(tmp5, tmp4, tmp5);
6121     cmp(tmp5, zr);
6122     bind(CSET_EQ);
6123     cset(result, EQ);
6124     b(DONE);
6125   }
6126 
6127   bind(SAME);
6128   mov(result, true);
6129   // That's it.
6130   bind(DONE);
6131 
6132   BLOCK_COMMENT("} array_equals");
6133   postcond(pc() != badAddress);
6134   return pc();
6135 }
6136 
6137 // Compare Strings
6138 
6139 // For Strings we're passed the address of the first characters in a1
6140 // and a2 and the length in cnt1.
6141 // There are two implementations.  For arrays >= 8 bytes, all
6142 // comparisons (including the final one, which may overlap) are
6143 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6144 // halfword, then a short, and then a byte.
6145 
6146 void MacroAssembler::string_equals(Register a1, Register a2,
6147                                    Register result, Register cnt1)
6148 {
6149   Label SAME, DONE, SHORT, NEXT_WORD;
6150   Register tmp1 = rscratch1;
6151   Register tmp2 = rscratch2;
6152   Register cnt2 = tmp2;  // cnt2 only used in array length compare
6153 
6154   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6155 
6156 #ifndef PRODUCT
6157   {
6158     char comment[64];
6159     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6160     BLOCK_COMMENT(comment);
6161   }
6162 #endif
6163 
6164   mov(result, false);
6165 
6166   // Check for short strings, i.e. smaller than wordSize.
6167   subs(cnt1, cnt1, wordSize);
6168   br(Assembler::LT, SHORT);
6169   // Main 8 byte comparison loop.
6170   bind(NEXT_WORD); {
6171     ldr(tmp1, Address(post(a1, wordSize)));
6172     ldr(tmp2, Address(post(a2, wordSize)));
6173     subs(cnt1, cnt1, wordSize);
6174     eor(tmp1, tmp1, tmp2);
6175     cbnz(tmp1, DONE);
6176   } br(GT, NEXT_WORD);
6177   // Last longword.  In the case where length == 4 we compare the
6178   // same longword twice, but that's still faster than another
6179   // conditional branch.
6180   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6181   // length == 4.
6182   ldr(tmp1, Address(a1, cnt1));
6183   ldr(tmp2, Address(a2, cnt1));
6184   eor(tmp2, tmp1, tmp2);
6185   cbnz(tmp2, DONE);
6186   b(SAME);
6187 
6188   bind(SHORT);
6189   Label TAIL03, TAIL01;
6190 
6191   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6192   {
6193     ldrw(tmp1, Address(post(a1, 4)));
6194     ldrw(tmp2, Address(post(a2, 4)));
6195     eorw(tmp1, tmp1, tmp2);
6196     cbnzw(tmp1, DONE);
6197   }
6198   bind(TAIL03);
6199   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6200   {
6201     ldrh(tmp1, Address(post(a1, 2)));
6202     ldrh(tmp2, Address(post(a2, 2)));
6203     eorw(tmp1, tmp1, tmp2);
6204     cbnzw(tmp1, DONE);
6205   }
6206   bind(TAIL01);
6207   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6208     {
6209     ldrb(tmp1, a1);
6210     ldrb(tmp2, a2);
6211     eorw(tmp1, tmp1, tmp2);
6212     cbnzw(tmp1, DONE);
6213   }
6214   // Arrays are equal.
6215   bind(SAME);
6216   mov(result, true);
6217 
6218   // That's it.
6219   bind(DONE);
6220   BLOCK_COMMENT("} string_equals");
6221 }
6222 
6223 
6224 // The size of the blocks erased by the zero_blocks stub.  We must
6225 // handle anything smaller than this ourselves in zero_words().
6226 const int MacroAssembler::zero_words_block_size = 8;
6227 
6228 // zero_words() is used by C2 ClearArray patterns and by
6229 // C1_MacroAssembler.  It is as small as possible, handling small word
6230 // counts locally and delegating anything larger to the zero_blocks
6231 // stub.  It is expanded many times in compiled code, so it is
6232 // important to keep it short.
6233 
6234 // ptr:   Address of a buffer to be zeroed.
6235 // cnt:   Count in HeapWords.
6236 //
6237 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6238 address MacroAssembler::zero_words(Register ptr, Register cnt)
6239 {
6240   assert(is_power_of_2(zero_words_block_size), "adjust this");
6241 
6242   BLOCK_COMMENT("zero_words {");
6243   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6244   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6245   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6246 
6247   subs(rscratch1, cnt, zero_words_block_size);
6248   Label around;
6249   br(LO, around);
6250   {
6251     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6252     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6253     // Make sure this is a C2 compilation. C1 allocates space only for
6254     // trampoline stubs generated by Call LIR ops, and in any case it
6255     // makes sense for a C1 compilation task to proceed as quickly as
6256     // possible.
6257     CompileTask* task;
6258     if (StubRoutines::aarch64::complete()
6259         && Thread::current()->is_Compiler_thread()
6260         && (task = ciEnv::current()->task())
6261         && is_c2_compile(task->comp_level())) {
6262       address tpc = trampoline_call(zero_blocks);
6263       if (tpc == nullptr) {
6264         DEBUG_ONLY(reset_labels(around));
6265         return nullptr;
6266       }
6267     } else {
6268       far_call(zero_blocks);
6269     }
6270   }
6271   bind(around);
6272 
6273   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6274   // for us.
6275   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6276     Label l;
6277     tbz(cnt, exact_log2(i), l);
6278     for (int j = 0; j < i; j += 2) {
6279       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6280     }
6281     bind(l);
6282   }
6283   {
6284     Label l;
6285     tbz(cnt, 0, l);
6286     str(zr, Address(ptr));
6287     bind(l);
6288   }
6289 
6290   BLOCK_COMMENT("} zero_words");
6291   return pc();
6292 }
6293 
6294 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6295 // cnt:          Immediate count in HeapWords.
6296 //
6297 // r10, r11, rscratch1, and rscratch2 are clobbered.
6298 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6299 {
6300   assert(wordSize <= BlockZeroingLowLimit,
6301             "increase BlockZeroingLowLimit");
6302   address result = nullptr;
6303   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6304 #ifndef PRODUCT
6305     {
6306       char buf[64];
6307       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6308       BLOCK_COMMENT(buf);
6309     }
6310 #endif
6311     if (cnt >= 16) {
6312       uint64_t loops = cnt/16;
6313       if (loops > 1) {
6314         mov(rscratch2, loops - 1);
6315       }
6316       {
6317         Label loop;
6318         bind(loop);
6319         for (int i = 0; i < 16; i += 2) {
6320           stp(zr, zr, Address(base, i * BytesPerWord));
6321         }
6322         add(base, base, 16 * BytesPerWord);
6323         if (loops > 1) {
6324           subs(rscratch2, rscratch2, 1);
6325           br(GE, loop);
6326         }
6327       }
6328     }
6329     cnt %= 16;
6330     int i = cnt & 1;  // store any odd word to start
6331     if (i) str(zr, Address(base));
6332     for (; i < (int)cnt; i += 2) {
6333       stp(zr, zr, Address(base, i * wordSize));
6334     }
6335     BLOCK_COMMENT("} zero_words");
6336     result = pc();
6337   } else {
6338     mov(r10, base); mov(r11, cnt);
6339     result = zero_words(r10, r11);
6340   }
6341   return result;
6342 }
6343 
6344 // Zero blocks of memory by using DC ZVA.
6345 //
6346 // Aligns the base address first sufficiently for DC ZVA, then uses
6347 // DC ZVA repeatedly for every full block.  cnt is the size to be
6348 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6349 // in cnt.
6350 //
6351 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6352 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6353 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6354   Register tmp = rscratch1;
6355   Register tmp2 = rscratch2;
6356   int zva_length = VM_Version::zva_length();
6357   Label initial_table_end, loop_zva;
6358   Label fini;
6359 
6360   // Base must be 16 byte aligned. If not just return and let caller handle it
6361   tst(base, 0x0f);
6362   br(Assembler::NE, fini);
6363   // Align base with ZVA length.
6364   neg(tmp, base);
6365   andr(tmp, tmp, zva_length - 1);
6366 
6367   // tmp: the number of bytes to be filled to align the base with ZVA length.
6368   add(base, base, tmp);
6369   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6370   adr(tmp2, initial_table_end);
6371   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6372   br(tmp2);
6373 
6374   for (int i = -zva_length + 16; i < 0; i += 16)
6375     stp(zr, zr, Address(base, i));
6376   bind(initial_table_end);
6377 
6378   sub(cnt, cnt, zva_length >> 3);
6379   bind(loop_zva);
6380   dc(Assembler::ZVA, base);
6381   subs(cnt, cnt, zva_length >> 3);
6382   add(base, base, zva_length);
6383   br(Assembler::GE, loop_zva);
6384   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6385   bind(fini);
6386 }
6387 
6388 // base:   Address of a buffer to be filled, 8 bytes aligned.
6389 // cnt:    Count in 8-byte unit.
6390 // value:  Value to be filled with.
6391 // base will point to the end of the buffer after filling.
6392 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6393 {
6394 //  Algorithm:
6395 //
6396 //    if (cnt == 0) {
6397 //      return;
6398 //    }
6399 //    if ((p & 8) != 0) {
6400 //      *p++ = v;
6401 //    }
6402 //
6403 //    scratch1 = cnt & 14;
6404 //    cnt -= scratch1;
6405 //    p += scratch1;
6406 //    switch (scratch1 / 2) {
6407 //      do {
6408 //        cnt -= 16;
6409 //          p[-16] = v;
6410 //          p[-15] = v;
6411 //        case 7:
6412 //          p[-14] = v;
6413 //          p[-13] = v;
6414 //        case 6:
6415 //          p[-12] = v;
6416 //          p[-11] = v;
6417 //          // ...
6418 //        case 1:
6419 //          p[-2] = v;
6420 //          p[-1] = v;
6421 //        case 0:
6422 //          p += 16;
6423 //      } while (cnt);
6424 //    }
6425 //    if ((cnt & 1) == 1) {
6426 //      *p++ = v;
6427 //    }
6428 
6429   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6430 
6431   Label fini, skip, entry, loop;
6432   const int unroll = 8; // Number of stp instructions we'll unroll
6433 
6434   cbz(cnt, fini);
6435   tbz(base, 3, skip);
6436   str(value, Address(post(base, 8)));
6437   sub(cnt, cnt, 1);
6438   bind(skip);
6439 
6440   andr(rscratch1, cnt, (unroll-1) * 2);
6441   sub(cnt, cnt, rscratch1);
6442   add(base, base, rscratch1, Assembler::LSL, 3);
6443   adr(rscratch2, entry);
6444   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6445   br(rscratch2);
6446 
6447   bind(loop);
6448   add(base, base, unroll * 16);
6449   for (int i = -unroll; i < 0; i++)
6450     stp(value, value, Address(base, i * 16));
6451   bind(entry);
6452   subs(cnt, cnt, unroll * 2);
6453   br(Assembler::GE, loop);
6454 
6455   tbz(cnt, 0, fini);
6456   str(value, Address(post(base, 8)));
6457   bind(fini);
6458 }
6459 
6460 // Intrinsic for
6461 //
6462 // - sun/nio/cs/ISO_8859_1$Encoder.implEncodeISOArray
6463 //     return the number of characters copied.
6464 // - java/lang/StringUTF16.compress
6465 //     return index of non-latin1 character if copy fails, otherwise 'len'.
6466 //
6467 // This version always returns the number of characters copied, and does not
6468 // clobber the 'len' register. A successful copy will complete with the post-
6469 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6470 // post-condition: 0 <= 'res' < 'len'.
6471 //
6472 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6473 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6474 //       beyond the acceptable, even though the footprint would be smaller.
6475 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6476 //       avoid additional bloat.
6477 //
6478 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6479 void MacroAssembler::encode_iso_array(Register src, Register dst,
6480                                       Register len, Register res, bool ascii,
6481                                       FloatRegister vtmp0, FloatRegister vtmp1,
6482                                       FloatRegister vtmp2, FloatRegister vtmp3,
6483                                       FloatRegister vtmp4, FloatRegister vtmp5)
6484 {
6485   Register cnt = res;
6486   Register max = rscratch1;
6487   Register chk = rscratch2;
6488 
6489   prfm(Address(src), PLDL1STRM);
6490   movw(cnt, len);
6491 
6492 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6493 
6494   Label LOOP_32, DONE_32, FAIL_32;
6495 
6496   BIND(LOOP_32);
6497   {
6498     cmpw(cnt, 32);
6499     br(LT, DONE_32);
6500     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6501     // Extract lower bytes.
6502     FloatRegister vlo0 = vtmp4;
6503     FloatRegister vlo1 = vtmp5;
6504     uzp1(vlo0, T16B, vtmp0, vtmp1);
6505     uzp1(vlo1, T16B, vtmp2, vtmp3);
6506     // Merge bits...
6507     orr(vtmp0, T16B, vtmp0, vtmp1);
6508     orr(vtmp2, T16B, vtmp2, vtmp3);
6509     // Extract merged upper bytes.
6510     FloatRegister vhix = vtmp0;
6511     uzp2(vhix, T16B, vtmp0, vtmp2);
6512     // ISO-check on hi-parts (all zero).
6513     //                          ASCII-check on lo-parts (no sign).
6514     FloatRegister vlox = vtmp1; // Merge lower bytes.
6515                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6516     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6517     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6518     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6519                                 ASCII(orr(chk, chk, max));
6520     cbnz(chk, FAIL_32);
6521     subw(cnt, cnt, 32);
6522     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6523     b(LOOP_32);
6524   }
6525   BIND(FAIL_32);
6526   sub(src, src, 64);
6527   BIND(DONE_32);
6528 
6529   Label LOOP_8, SKIP_8;
6530 
6531   BIND(LOOP_8);
6532   {
6533     cmpw(cnt, 8);
6534     br(LT, SKIP_8);
6535     FloatRegister vhi = vtmp0;
6536     FloatRegister vlo = vtmp1;
6537     ld1(vtmp3, T8H, src);
6538     uzp1(vlo, T16B, vtmp3, vtmp3);
6539     uzp2(vhi, T16B, vtmp3, vtmp3);
6540     // ISO-check on hi-parts (all zero).
6541     //                          ASCII-check on lo-parts (no sign).
6542                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6543     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6544                                 ASCII(umov(max, vtmp2, B, 0));
6545                                 ASCII(orr(chk, chk, max));
6546     cbnz(chk, SKIP_8);
6547 
6548     strd(vlo, Address(post(dst, 8)));
6549     subw(cnt, cnt, 8);
6550     add(src, src, 16);
6551     b(LOOP_8);
6552   }
6553   BIND(SKIP_8);
6554 
6555 #undef ASCII
6556 
6557   Label LOOP, DONE;
6558 
6559   cbz(cnt, DONE);
6560   BIND(LOOP);
6561   {
6562     Register chr = rscratch1;
6563     ldrh(chr, Address(post(src, 2)));
6564     tst(chr, ascii ? 0xff80 : 0xff00);
6565     br(NE, DONE);
6566     strb(chr, Address(post(dst, 1)));
6567     subs(cnt, cnt, 1);
6568     br(GT, LOOP);
6569   }
6570   BIND(DONE);
6571   // Return index where we stopped.
6572   subw(res, len, cnt);
6573 }
6574 
6575 // Inflate byte[] array to char[].
6576 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6577 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6578                                            FloatRegister vtmp1, FloatRegister vtmp2,
6579                                            FloatRegister vtmp3, Register tmp4) {
6580   Label big, done, after_init, to_stub;
6581 
6582   assert_different_registers(src, dst, len, tmp4, rscratch1);
6583 
6584   fmovd(vtmp1, 0.0);
6585   lsrw(tmp4, len, 3);
6586   bind(after_init);
6587   cbnzw(tmp4, big);
6588   // Short string: less than 8 bytes.
6589   {
6590     Label loop, tiny;
6591 
6592     cmpw(len, 4);
6593     br(LT, tiny);
6594     // Use SIMD to do 4 bytes.
6595     ldrs(vtmp2, post(src, 4));
6596     zip1(vtmp3, T8B, vtmp2, vtmp1);
6597     subw(len, len, 4);
6598     strd(vtmp3, post(dst, 8));
6599 
6600     cbzw(len, done);
6601 
6602     // Do the remaining bytes by steam.
6603     bind(loop);
6604     ldrb(tmp4, post(src, 1));
6605     strh(tmp4, post(dst, 2));
6606     subw(len, len, 1);
6607 
6608     bind(tiny);
6609     cbnz(len, loop);
6610 
6611     b(done);
6612   }
6613 
6614   if (SoftwarePrefetchHintDistance >= 0) {
6615     bind(to_stub);
6616       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6617       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6618       address tpc = trampoline_call(stub);
6619       if (tpc == nullptr) {
6620         DEBUG_ONLY(reset_labels(big, done));
6621         postcond(pc() == badAddress);
6622         return nullptr;
6623       }
6624       b(after_init);
6625   }
6626 
6627   // Unpack the bytes 8 at a time.
6628   bind(big);
6629   {
6630     Label loop, around, loop_last, loop_start;
6631 
6632     if (SoftwarePrefetchHintDistance >= 0) {
6633       const int large_loop_threshold = (64 + 16)/8;
6634       ldrd(vtmp2, post(src, 8));
6635       andw(len, len, 7);
6636       cmp(tmp4, (u1)large_loop_threshold);
6637       br(GE, to_stub);
6638       b(loop_start);
6639 
6640       bind(loop);
6641       ldrd(vtmp2, post(src, 8));
6642       bind(loop_start);
6643       subs(tmp4, tmp4, 1);
6644       br(EQ, loop_last);
6645       zip1(vtmp2, T16B, vtmp2, vtmp1);
6646       ldrd(vtmp3, post(src, 8));
6647       st1(vtmp2, T8H, post(dst, 16));
6648       subs(tmp4, tmp4, 1);
6649       zip1(vtmp3, T16B, vtmp3, vtmp1);
6650       st1(vtmp3, T8H, post(dst, 16));
6651       br(NE, loop);
6652       b(around);
6653       bind(loop_last);
6654       zip1(vtmp2, T16B, vtmp2, vtmp1);
6655       st1(vtmp2, T8H, post(dst, 16));
6656       bind(around);
6657       cbz(len, done);
6658     } else {
6659       andw(len, len, 7);
6660       bind(loop);
6661       ldrd(vtmp2, post(src, 8));
6662       sub(tmp4, tmp4, 1);
6663       zip1(vtmp3, T16B, vtmp2, vtmp1);
6664       st1(vtmp3, T8H, post(dst, 16));
6665       cbnz(tmp4, loop);
6666     }
6667   }
6668 
6669   // Do the tail of up to 8 bytes.
6670   add(src, src, len);
6671   ldrd(vtmp3, Address(src, -8));
6672   add(dst, dst, len, ext::uxtw, 1);
6673   zip1(vtmp3, T16B, vtmp3, vtmp1);
6674   strq(vtmp3, Address(dst, -16));
6675 
6676   bind(done);
6677   postcond(pc() != badAddress);
6678   return pc();
6679 }
6680 
6681 // Compress char[] array to byte[].
6682 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6683 // Return the array length if every element in array can be encoded,
6684 // otherwise, the index of first non-latin1 (> 0xff) character.
6685 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6686                                          Register res,
6687                                          FloatRegister tmp0, FloatRegister tmp1,
6688                                          FloatRegister tmp2, FloatRegister tmp3,
6689                                          FloatRegister tmp4, FloatRegister tmp5) {
6690   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6691 }
6692 
6693 // java.math.round(double a)
6694 // Returns the closest long to the argument, with ties rounding to
6695 // positive infinity.  This requires some fiddling for corner
6696 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6697 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6698                                        FloatRegister ftmp) {
6699   Label DONE;
6700   BLOCK_COMMENT("java_round_double: { ");
6701   fmovd(rscratch1, src);
6702   // Use RoundToNearestTiesAway unless src small and -ve.
6703   fcvtasd(dst, src);
6704   // Test if src >= 0 || abs(src) >= 0x1.0p52
6705   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6706   mov(rscratch2, julong_cast(0x1.0p52));
6707   cmp(rscratch1, rscratch2);
6708   br(HS, DONE); {
6709     // src < 0 && abs(src) < 0x1.0p52
6710     // src may have a fractional part, so add 0.5
6711     fmovd(ftmp, 0.5);
6712     faddd(ftmp, src, ftmp);
6713     // Convert double to jlong, use RoundTowardsNegative
6714     fcvtmsd(dst, ftmp);
6715   }
6716   bind(DONE);
6717   BLOCK_COMMENT("} java_round_double");
6718 }
6719 
6720 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6721                                       FloatRegister ftmp) {
6722   Label DONE;
6723   BLOCK_COMMENT("java_round_float: { ");
6724   fmovs(rscratch1, src);
6725   // Use RoundToNearestTiesAway unless src small and -ve.
6726   fcvtassw(dst, src);
6727   // Test if src >= 0 || abs(src) >= 0x1.0p23
6728   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6729   mov(rscratch2, jint_cast(0x1.0p23f));
6730   cmp(rscratch1, rscratch2);
6731   br(HS, DONE); {
6732     // src < 0 && |src| < 0x1.0p23
6733     // src may have a fractional part, so add 0.5
6734     fmovs(ftmp, 0.5f);
6735     fadds(ftmp, src, ftmp);
6736     // Convert float to jint, use RoundTowardsNegative
6737     fcvtmssw(dst, ftmp);
6738   }
6739   bind(DONE);
6740   BLOCK_COMMENT("} java_round_float");
6741 }
6742 
6743 // get_thread() can be called anywhere inside generated code so we
6744 // need to save whatever non-callee save context might get clobbered
6745 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6746 // the call setup code.
6747 //
6748 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6749 // On other systems, the helper is a usual C function.
6750 //
6751 void MacroAssembler::get_thread(Register dst) {
6752   RegSet saved_regs =
6753     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6754     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6755 
6756   protect_return_address();
6757   push(saved_regs, sp);
6758 
6759   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6760   blr(lr);
6761   if (dst != c_rarg0) {
6762     mov(dst, c_rarg0);
6763   }
6764 
6765   pop(saved_regs, sp);
6766   authenticate_return_address();
6767 }
6768 
6769 void MacroAssembler::cache_wb(Address line) {
6770   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6771   assert(line.index() == noreg, "index should be noreg");
6772   assert(line.offset() == 0, "offset should be 0");
6773   // would like to assert this
6774   // assert(line._ext.shift == 0, "shift should be zero");
6775   if (VM_Version::supports_dcpop()) {
6776     // writeback using clear virtual address to point of persistence
6777     dc(Assembler::CVAP, line.base());
6778   } else {
6779     // no need to generate anything as Unsafe.writebackMemory should
6780     // never invoke this stub
6781   }
6782 }
6783 
6784 void MacroAssembler::cache_wbsync(bool is_pre) {
6785   // we only need a barrier post sync
6786   if (!is_pre) {
6787     membar(Assembler::AnyAny);
6788   }
6789 }
6790 
6791 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6792   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6793     return;
6794   }
6795   // Make sure that native code does not change SVE vector length.
6796   Label verify_ok;
6797   movw(tmp, zr);
6798   sve_inc(tmp, B);
6799   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6800   br(EQ, verify_ok);
6801   stop("Error: SVE vector length has changed since jvm startup");
6802   bind(verify_ok);
6803 }
6804 
6805 void MacroAssembler::verify_ptrue() {
6806   Label verify_ok;
6807   if (!UseSVE) {
6808     return;
6809   }
6810   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6811   sve_dec(rscratch1, B);
6812   cbz(rscratch1, verify_ok);
6813   stop("Error: the preserved predicate register (p7) elements are not all true");
6814   bind(verify_ok);
6815 }
6816 
6817 void MacroAssembler::safepoint_isb() {
6818   isb();
6819 #ifndef PRODUCT
6820   if (VerifyCrossModifyFence) {
6821     // Clear the thread state.
6822     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6823   }
6824 #endif
6825 }
6826 
6827 #ifndef PRODUCT
6828 void MacroAssembler::verify_cross_modify_fence_not_required() {
6829   if (VerifyCrossModifyFence) {
6830     // Check if thread needs a cross modify fence.
6831     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6832     Label fence_not_required;
6833     cbz(rscratch1, fence_not_required);
6834     // If it does then fail.
6835     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6836     mov(c_rarg0, rthread);
6837     blr(rscratch1);
6838     bind(fence_not_required);
6839   }
6840 }
6841 #endif
6842 
6843 void MacroAssembler::spin_wait() {
6844   block_comment("spin_wait {");
6845   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6846     switch (VM_Version::spin_wait_desc().inst()) {
6847       case SpinWait::NOP:
6848         nop();
6849         break;
6850       case SpinWait::ISB:
6851         isb();
6852         break;
6853       case SpinWait::YIELD:
6854         yield();
6855         break;
6856       case SpinWait::SB:
6857         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6858         sb();
6859         break;
6860       default:
6861         ShouldNotReachHere();
6862     }
6863   }
6864   block_comment("}");
6865 }
6866 
6867 // Stack frame creation/removal
6868 
6869 void MacroAssembler::enter(bool strip_ret_addr) {
6870   if (strip_ret_addr) {
6871     // Addresses can only be signed once. If there are multiple nested frames being created
6872     // in the same function, then the return address needs stripping first.
6873     strip_return_address();
6874   }
6875   protect_return_address();
6876   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6877   mov(rfp, sp);
6878 }
6879 
6880 void MacroAssembler::leave() {
6881   mov(sp, rfp);
6882   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6883   authenticate_return_address();
6884 }
6885 
6886 // ROP Protection
6887 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6888 // destroying stack frames or whenever directly loading/storing the LR to memory.
6889 // If ROP protection is not set then these functions are no-ops.
6890 // For more details on PAC see pauth_aarch64.hpp.
6891 
6892 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6893 // Uses value zero as the modifier.
6894 //
6895 void MacroAssembler::protect_return_address() {
6896   if (VM_Version::use_rop_protection()) {
6897     check_return_address();
6898     paciaz();
6899   }
6900 }
6901 
6902 // Sign the return value in the given register. Use before updating the LR in the existing stack
6903 // frame for the current function.
6904 // Uses value zero as the modifier.
6905 //
6906 void MacroAssembler::protect_return_address(Register return_reg) {
6907   if (VM_Version::use_rop_protection()) {
6908     check_return_address(return_reg);
6909     paciza(return_reg);
6910   }
6911 }
6912 
6913 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6914 // Uses value zero as the modifier.
6915 //
6916 void MacroAssembler::authenticate_return_address() {
6917   if (VM_Version::use_rop_protection()) {
6918     autiaz();
6919     check_return_address();
6920   }
6921 }
6922 
6923 // Authenticate the return value in the given register. Use before updating the LR in the existing
6924 // stack frame for the current function.
6925 // Uses value zero as the modifier.
6926 //
6927 void MacroAssembler::authenticate_return_address(Register return_reg) {
6928   if (VM_Version::use_rop_protection()) {
6929     autiza(return_reg);
6930     check_return_address(return_reg);
6931   }
6932 }
6933 
6934 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6935 // there is no guaranteed way of authenticating the LR.
6936 //
6937 void MacroAssembler::strip_return_address() {
6938   if (VM_Version::use_rop_protection()) {
6939     xpaclri();
6940   }
6941 }
6942 
6943 #ifndef PRODUCT
6944 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6945 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6946 // it is difficult to debug back to the callee function.
6947 // This function simply loads from the address in the given register.
6948 // Use directly after authentication to catch authentication failures.
6949 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6950 //
6951 void MacroAssembler::check_return_address(Register return_reg) {
6952   if (VM_Version::use_rop_protection()) {
6953     ldr(zr, Address(return_reg));
6954   }
6955 }
6956 #endif
6957 
6958 // The java_calling_convention describes stack locations as ideal slots on
6959 // a frame with no abi restrictions. Since we must observe abi restrictions
6960 // (like the placement of the register window) the slots must be biased by
6961 // the following value.
6962 static int reg2offset_in(VMReg r) {
6963   // Account for saved rfp and lr
6964   // This should really be in_preserve_stack_slots
6965   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6966 }
6967 
6968 static int reg2offset_out(VMReg r) {
6969   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6970 }
6971 
6972 // On 64bit we will store integer like items to the stack as
6973 // 64bits items (AArch64 ABI) even though java would only store
6974 // 32bits for a parameter. On 32bit it will simply be 32bits
6975 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6976 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6977   if (src.first()->is_stack()) {
6978     if (dst.first()->is_stack()) {
6979       // stack to stack
6980       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6981       str(tmp, Address(sp, reg2offset_out(dst.first())));
6982     } else {
6983       // stack to reg
6984       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6985     }
6986   } else if (dst.first()->is_stack()) {
6987     // reg to stack
6988     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6989   } else {
6990     if (dst.first() != src.first()) {
6991       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6992     }
6993   }
6994 }
6995 
6996 // An oop arg. Must pass a handle not the oop itself
6997 void MacroAssembler::object_move(
6998                         OopMap* map,
6999                         int oop_handle_offset,
7000                         int framesize_in_slots,
7001                         VMRegPair src,
7002                         VMRegPair dst,
7003                         bool is_receiver,
7004                         int* receiver_offset) {
7005 
7006   // must pass a handle. First figure out the location we use as a handle
7007 
7008   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
7009 
7010   // See if oop is null if it is we need no handle
7011 
7012   if (src.first()->is_stack()) {
7013 
7014     // Oop is already on the stack as an argument
7015     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
7016     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
7017     if (is_receiver) {
7018       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
7019     }
7020 
7021     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7022     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7023     // conditionally move a null
7024     cmp(rscratch1, zr);
7025     csel(rHandle, zr, rHandle, Assembler::EQ);
7026   } else {
7027 
7028     // Oop is in an a register we must store it to the space we reserve
7029     // on the stack for oop_handles and pass a handle if oop is non-null
7030 
7031     const Register rOop = src.first()->as_Register();
7032     int oop_slot;
7033     if (rOop == j_rarg0)
7034       oop_slot = 0;
7035     else if (rOop == j_rarg1)
7036       oop_slot = 1;
7037     else if (rOop == j_rarg2)
7038       oop_slot = 2;
7039     else if (rOop == j_rarg3)
7040       oop_slot = 3;
7041     else if (rOop == j_rarg4)
7042       oop_slot = 4;
7043     else if (rOop == j_rarg5)
7044       oop_slot = 5;
7045     else if (rOop == j_rarg6)
7046       oop_slot = 6;
7047     else {
7048       assert(rOop == j_rarg7, "wrong register");
7049       oop_slot = 7;
7050     }
7051 
7052     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7053     int offset = oop_slot*VMRegImpl::stack_slot_size;
7054 
7055     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7056     // Store oop in handle area, may be null
7057     str(rOop, Address(sp, offset));
7058     if (is_receiver) {
7059       *receiver_offset = offset;
7060     }
7061 
7062     cmp(rOop, zr);
7063     lea(rHandle, Address(sp, offset));
7064     // conditionally move a null
7065     csel(rHandle, zr, rHandle, Assembler::EQ);
7066   }
7067 
7068   // If arg is on the stack then place it otherwise it is already in correct reg.
7069   if (dst.first()->is_stack()) {
7070     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7071   }
7072 }
7073 
7074 // A float arg may have to do float reg int reg conversion
7075 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7076  if (src.first()->is_stack()) {
7077     if (dst.first()->is_stack()) {
7078       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7079       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7080     } else {
7081       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7082     }
7083   } else if (src.first() != dst.first()) {
7084     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7085       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7086     else
7087       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7088   }
7089 }
7090 
7091 // A long move
7092 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7093   if (src.first()->is_stack()) {
7094     if (dst.first()->is_stack()) {
7095       // stack to stack
7096       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7097       str(tmp, Address(sp, reg2offset_out(dst.first())));
7098     } else {
7099       // stack to reg
7100       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7101     }
7102   } else if (dst.first()->is_stack()) {
7103     // reg to stack
7104     // Do we really have to sign extend???
7105     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7106     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7107   } else {
7108     if (dst.first() != src.first()) {
7109       mov(dst.first()->as_Register(), src.first()->as_Register());
7110     }
7111   }
7112 }
7113 
7114 
7115 // A double move
7116 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7117  if (src.first()->is_stack()) {
7118     if (dst.first()->is_stack()) {
7119       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7120       str(tmp, Address(sp, reg2offset_out(dst.first())));
7121     } else {
7122       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7123     }
7124   } else if (src.first() != dst.first()) {
7125     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7126       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7127     else
7128       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7129   }
7130 }
7131 
7132 // Implements fast-locking.
7133 //
7134 //  - obj: the object to be locked
7135 //  - t1, t2, t3: temporary registers, will be destroyed
7136 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7137 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7138   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7139 
7140   Label push;
7141   const Register top = t1;
7142   const Register mark = t2;
7143   const Register t = t3;
7144 
7145   // Preload the markWord. It is important that this is the first
7146   // instruction emitted as it is part of C1's null check semantics.
7147   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7148 
7149   if (UseObjectMonitorTable) {
7150     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7151     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7152   }
7153 
7154   if (DiagnoseSyncOnValueBasedClasses != 0) {
7155     load_klass(t1, obj);
7156     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7157     tst(t1, KlassFlags::_misc_is_value_based_class);
7158     br(Assembler::NE, slow);
7159   }
7160 
7161   // Check if the lock-stack is full.
7162   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7163   cmpw(top, (unsigned)LockStack::end_offset());
7164   br(Assembler::GE, slow);
7165 
7166   // Check for recursion.
7167   subw(t, top, oopSize);
7168   ldr(t, Address(rthread, t));
7169   cmp(obj, t);
7170   br(Assembler::EQ, push);
7171 
7172   // Check header for monitor (0b10).
7173   tst(mark, markWord::monitor_value);
7174   br(Assembler::NE, slow);
7175 
7176   // Try to lock. Transition lock bits 0b01 => 0b00
7177   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7178   orr(mark, mark, markWord::unlocked_value);
7179   eor(t, mark, markWord::unlocked_value);
7180   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7181           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7182   br(Assembler::NE, slow);
7183 
7184   bind(push);
7185   // After successful lock, push object on lock-stack.
7186   str(obj, Address(rthread, top));
7187   addw(top, top, oopSize);
7188   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7189 }
7190 
7191 // Implements fast-unlocking.
7192 //
7193 // - obj: the object to be unlocked
7194 // - t1, t2, t3: temporary registers
7195 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7196 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7197   // cmpxchg clobbers rscratch1.
7198   assert_different_registers(obj, t1, t2, t3, rscratch1);
7199 
7200 #ifdef ASSERT
7201   {
7202     // Check for lock-stack underflow.
7203     Label stack_ok;
7204     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7205     cmpw(t1, (unsigned)LockStack::start_offset());
7206     br(Assembler::GE, stack_ok);
7207     STOP("Lock-stack underflow");
7208     bind(stack_ok);
7209   }
7210 #endif
7211 
7212   Label unlocked, push_and_slow;
7213   const Register top = t1;
7214   const Register mark = t2;
7215   const Register t = t3;
7216 
7217   // Check if obj is top of lock-stack.
7218   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7219   subw(top, top, oopSize);
7220   ldr(t, Address(rthread, top));
7221   cmp(obj, t);
7222   br(Assembler::NE, slow);
7223 
7224   // Pop lock-stack.
7225   DEBUG_ONLY(str(zr, Address(rthread, top));)
7226   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7227 
7228   // Check if recursive.
7229   subw(t, top, oopSize);
7230   ldr(t, Address(rthread, t));
7231   cmp(obj, t);
7232   br(Assembler::EQ, unlocked);
7233 
7234   // Not recursive. Check header for monitor (0b10).
7235   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7236   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7237 
7238 #ifdef ASSERT
7239   // Check header not unlocked (0b01).
7240   Label not_unlocked;
7241   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7242   stop("fast_unlock already unlocked");
7243   bind(not_unlocked);
7244 #endif
7245 
7246   // Try to unlock. Transition lock bits 0b00 => 0b01
7247   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7248   orr(t, mark, markWord::unlocked_value);
7249   cmpxchg(obj, mark, t, Assembler::xword,
7250           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7251   br(Assembler::EQ, unlocked);
7252 
7253   bind(push_and_slow);
7254   // Restore lock-stack and handle the unlock in runtime.
7255   DEBUG_ONLY(str(obj, Address(rthread, top));)
7256   addw(top, top, oopSize);
7257   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7258   b(slow);
7259 
7260   bind(unlocked);
7261 }