1 /*
   2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "code/compiledIC.hpp"
  30 #include "compiler/compileTask.hpp"
  31 #include "compiler/disassembler.hpp"
  32 #include "compiler/oopMap.hpp"
  33 #include "gc/shared/barrierSet.hpp"
  34 #include "gc/shared/barrierSetAssembler.hpp"
  35 #include "gc/shared/cardTableBarrierSet.hpp"
  36 #include "gc/shared/cardTable.hpp"
  37 #include "gc/shared/collectedHeap.hpp"
  38 #include "gc/shared/tlab_globals.hpp"
  39 #include "interpreter/bytecodeHistogram.hpp"
  40 #include "interpreter/interpreter.hpp"
  41 #include "interpreter/interpreterRuntime.hpp"
  42 #include "jvm.h"
  43 #include "memory/resourceArea.hpp"
  44 #include "memory/universe.hpp"
  45 #include "nativeInst_aarch64.hpp"
  46 #include "oops/accessDecorators.hpp"
  47 #include "oops/compressedKlass.inline.hpp"
  48 #include "oops/compressedOops.inline.hpp"
  49 #include "oops/klass.inline.hpp"
  50 #include "runtime/continuation.hpp"
  51 #include "runtime/icache.hpp"
  52 #include "runtime/interfaceSupport.inline.hpp"
  53 #include "runtime/javaThread.hpp"
  54 #include "runtime/jniHandles.inline.hpp"
  55 #include "runtime/sharedRuntime.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "utilities/globalDefinitions.hpp"
  58 #include "utilities/integerCast.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     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010100, "must be");
 326     uint64_t dest = (uint64_t)target;
 327     // Move wide constant
 328     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 329     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 330     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 331     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 332     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 333     return 3;
 334   }
 335   static void verify(address insn_addr, address &target) {
 336 #ifdef ASSERT
 337     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 338     if (!(address_is == target)) {
 339       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 340       disnm((intptr_t)insn_addr);
 341       assert(address_is == target, "should be");
 342     }
 343 #endif
 344   }
 345 };
 346 
 347 // If insn1 and insn2 use the same register to form an address, either
 348 // by an offsetted LDR or a simple ADD, return the offset. If the
 349 // second instruction is an LDR, the offset may be scaled.
 350 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 351   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 352       Instruction_aarch64::extract(insn1, 4, 0) ==
 353       Instruction_aarch64::extract(insn2, 9, 5)) {
 354     // Load/store register (unsigned immediate)
 355     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 356     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 357     byte_offset <<= size;
 358     return true;
 359   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 360              Instruction_aarch64::extract(insn1, 4, 0) ==
 361              Instruction_aarch64::extract(insn2, 4, 0)) {
 362     // add (immediate)
 363     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 364     return true;
 365   }
 366   return false;
 367 }
 368 
 369 class AArch64Decoder : public AllStatic {
 370 public:
 371 
 372   static int loadStore(address insn_addr, address &target) {
 373     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 374     target = insn_addr + (offset << 2);
 375     return 1;
 376   }
 377   static int unconditionalBranch(address insn_addr, address &target) {
 378     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 25, 0);
 379     target = insn_addr + (offset << 2);
 380     return 1;
 381   }
 382   static int conditionalBranch(address insn_addr, address &target) {
 383     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 384     target = address(((uint64_t)insn_addr + (offset << 2)));
 385     return 1;
 386   }
 387   static int testAndBranch(address insn_addr, address &target) {
 388     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 18, 5);
 389     target = address(((uint64_t)insn_addr + (offset << 2)));
 390     return 1;
 391   }
 392   static int adr(address insn_addr, address &target) {
 393     // PC-rel. addressing
 394     uint32_t insn = insn_at(insn_addr, 0);
 395     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 396     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 397     target = address((uint64_t)insn_addr + offset);
 398     return 1;
 399   }
 400   template<typename U>
 401   static int adrp(address insn_addr, address &target, U inner) {
 402     uint32_t insn = insn_at(insn_addr, 0);
 403     assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 404     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 405     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 406     int shift = 12;
 407     offset <<= shift;
 408     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 409     target_page &= ((uint64_t)-1) << shift;
 410     target = address(target_page);
 411     precond(inner != nullptr);
 412     inner(insn_addr, target);
 413     return 2;
 414   }
 415   static int adrpMem(address insn_addr, address &target) {
 416     uint32_t insn2 = insn_at(insn_addr, 1);
 417     // Load/store register (unsigned immediate)
 418     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 419     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 420     byte_offset <<= size;
 421     target += byte_offset;
 422     return 2;
 423   }
 424   static int adrpAdd(address insn_addr, address &target) {
 425     uint32_t insn2 = insn_at(insn_addr, 1);
 426     // add (immediate)
 427     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 428     target += byte_offset;
 429     return 2;
 430   }
 431   static int adrpMovk(address insn_addr, address &target) {
 432     uint32_t insn2 = insn_at(insn_addr, 1);
 433     uint64_t dest = uint64_t(target);
 434     dest = (dest & 0xffff0000ffffffff) |
 435       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 436     target = address(dest);
 437 
 438     // We know the destination 4k page. Maybe we have a third
 439     // instruction.
 440     uint32_t insn = insn_at(insn_addr, 0);
 441     uint32_t insn3 = insn_at(insn_addr, 2);
 442     ptrdiff_t byte_offset;
 443     if (offset_for(insn, insn3, byte_offset)) {
 444       target += byte_offset;
 445       return 3;
 446     } else {
 447       return 2;
 448     }
 449   }
 450   static int immediate(address insn_addr, address &target) {
 451     uint32_t *insns = (uint32_t *)insn_addr;
 452     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 453     // Move wide constant: movz, movk, movk.  See movptr().
 454     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 455     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 456     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 457                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 458                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 459     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 460     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 461     return 3;
 462   }
 463   static void verify(address insn_addr, address &target) {
 464   }
 465 };
 466 
 467 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 468   address target;
 469   RelocActions<AArch64Decoder>::run(insn_addr, target);
 470   return target;
 471 }
 472 
 473 // Patch any kind of instruction; there may be several instructions.
 474 // Return the total length (in bytes) of the instructions.
 475 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 476   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 477   return RelocActions<Patcher>::run(insn_addr, target);
 478 }
 479 
 480 int MacroAssembler::patch_oop(address insn_addr, address o) {
 481   int instructions;
 482   unsigned insn = *(unsigned*)insn_addr;
 483   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 484 
 485   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 486 
 487   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 488   // narrow OOPs by setting the upper 16 bits in the first
 489   // instruction.
 490   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 491     // Move narrow OOP
 492     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 493     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 494     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 495     instructions = 2;
 496   } else {
 497     // Move wide OOP
 498     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 499     uintptr_t dest = (uintptr_t)o;
 500     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 501     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 502     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 503     instructions = 3;
 504   }
 505   return instructions * NativeInstruction::instruction_size;
 506 }
 507 
 508 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 509   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 510   // We encode narrow ones by setting the upper 16 bits in the first
 511   // instruction.
 512   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 513   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 514          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 515 
 516   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 517 
 518   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 519   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 520   return 2 * NativeInstruction::instruction_size;
 521 }
 522 
 523 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 524   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 525   if (at_return) {
 526     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 527     // we may safely use the sp instead to perform the stack watermark check.
 528     cmp(in_nmethod ? sp : rfp, tmp);
 529     br(Assembler::HI, slow_path);
 530   } else {
 531     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 532   }
 533 }
 534 
 535 void MacroAssembler::rt_call(address dest, Register tmp) {
 536   CodeBlob *cb = CodeCache::find_blob(dest);
 537   if (cb) {
 538     far_call(RuntimeAddress(dest));
 539   } else {
 540     lea(tmp, RuntimeAddress(dest));
 541     blr(tmp);
 542   }
 543 }
 544 
 545 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 546   if (!Continuations::enabled()) return;
 547   Label done;
 548   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 549   cmp(sp, rscratch1);
 550   br(Assembler::LS, done);
 551   mov(rscratch1, sp); // we can't use sp as the source in str
 552   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 553   bind(done);
 554 }
 555 
 556 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 557   if (!Continuations::enabled()) return;
 558   Label done;
 559   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 560   cmp(sp, rscratch1);
 561   br(Assembler::LO, done);
 562   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 563   bind(done);
 564 }
 565 
 566 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 567   // we must set sp to zero to clear frame
 568   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 569 
 570   // must clear fp, so that compiled frames are not confused; it is
 571   // possible that we need it only for debugging
 572   if (clear_fp) {
 573     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 574   }
 575 
 576   // Always clear the pc because it could have been set by make_walkable()
 577   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 578 }
 579 
 580 // Calls to C land
 581 //
 582 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 583 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 584 // has to be reset to 0. This is required to allow proper stack traversal.
 585 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 586                                          Register last_java_fp,
 587                                          Register last_java_pc,
 588                                          Register scratch) {
 589 
 590   if (last_java_pc->is_valid()) {
 591       str(last_java_pc, Address(rthread,
 592                                 JavaThread::frame_anchor_offset()
 593                                 + JavaFrameAnchor::last_Java_pc_offset()));
 594     }
 595 
 596   // determine last_java_sp register
 597   if (last_java_sp == sp) {
 598     mov(scratch, sp);
 599     last_java_sp = scratch;
 600   } else if (!last_java_sp->is_valid()) {
 601     last_java_sp = esp;
 602   }
 603 
 604   // last_java_fp is optional
 605   if (last_java_fp->is_valid()) {
 606     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 607   }
 608 
 609   // We must set sp last.
 610   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 611 }
 612 
 613 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 614                                          Register last_java_fp,
 615                                          address  last_java_pc,
 616                                          Register scratch) {
 617   assert(last_java_pc != nullptr, "must provide a valid PC");
 618 
 619   adr(scratch, last_java_pc);
 620   str(scratch, Address(rthread,
 621                        JavaThread::frame_anchor_offset()
 622                        + JavaFrameAnchor::last_Java_pc_offset()));
 623 
 624   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 625 }
 626 
 627 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 628                                          Register last_java_fp,
 629                                          Label &L,
 630                                          Register scratch) {
 631   if (L.is_bound()) {
 632     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 633   } else {
 634     InstructionMark im(this);
 635     L.add_patch_at(code(), locator());
 636     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 637   }
 638 }
 639 
 640 static inline bool target_needs_far_branch(address addr) {
 641   if (AOTCodeCache::is_on_for_dump()) {
 642     return true;
 643   }
 644   // codecache size <= 128M
 645   if (!MacroAssembler::far_branches()) {
 646     return false;
 647   }
 648   // codecache size > 240M
 649   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 650     return true;
 651   }
 652   // codecache size: 128M..240M
 653   return !CodeCache::is_non_nmethod(addr);
 654 }
 655 
 656 void MacroAssembler::far_call(Address entry, Register tmp) {
 657   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 658   assert(CodeCache::find_blob(entry.target()) != nullptr,
 659          "destination of far call not found in code cache");
 660   assert(entry.rspec().type() == relocInfo::external_word_type
 661          || entry.rspec().type() == relocInfo::runtime_call_type
 662          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 663   if (target_needs_far_branch(entry.target())) {
 664     uint64_t offset;
 665     // We can use ADRP here because we know that the total size of
 666     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 667     adrp(tmp, entry, offset);
 668     add(tmp, tmp, offset);
 669     blr(tmp);
 670   } else {
 671     bl(entry);
 672   }
 673 }
 674 
 675 int MacroAssembler::far_jump(Address entry, Register tmp) {
 676   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 677   assert(CodeCache::find_blob(entry.target()) != nullptr,
 678          "destination of far call not found in code cache");
 679   assert(entry.rspec().type() == relocInfo::external_word_type
 680          || entry.rspec().type() == relocInfo::runtime_call_type
 681          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 682   address start = pc();
 683   if (target_needs_far_branch(entry.target())) {
 684     uint64_t offset;
 685     // We can use ADRP here because we know that the total size of
 686     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 687     adrp(tmp, entry, offset);
 688     add(tmp, tmp, offset);
 689     br(tmp);
 690   } else {
 691     b(entry);
 692   }
 693   return pc() - start;
 694 }
 695 
 696 void MacroAssembler::reserved_stack_check() {
 697     // testing if reserved zone needs to be enabled
 698     Label no_reserved_zone_enabling;
 699 
 700     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 701     cmp(sp, rscratch1);
 702     br(Assembler::LO, no_reserved_zone_enabling);
 703 
 704     enter();   // LR and FP are live.
 705     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 706     mov(c_rarg0, rthread);
 707     blr(rscratch1);
 708     leave();
 709 
 710     // We have already removed our own frame.
 711     // throw_delayed_StackOverflowError will think that it's been
 712     // called by our caller.
 713     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 714     br(rscratch1);
 715     should_not_reach_here();
 716 
 717     bind(no_reserved_zone_enabling);
 718 }
 719 
 720 static void pass_arg0(MacroAssembler* masm, Register arg) {
 721   if (c_rarg0 != arg ) {
 722     masm->mov(c_rarg0, arg);
 723   }
 724 }
 725 
 726 static void pass_arg1(MacroAssembler* masm, Register arg) {
 727   if (c_rarg1 != arg ) {
 728     masm->mov(c_rarg1, arg);
 729   }
 730 }
 731 
 732 static void pass_arg2(MacroAssembler* masm, Register arg) {
 733   if (c_rarg2 != arg ) {
 734     masm->mov(c_rarg2, arg);
 735   }
 736 }
 737 
 738 static void pass_arg3(MacroAssembler* masm, Register arg) {
 739   if (c_rarg3 != arg ) {
 740     masm->mov(c_rarg3, arg);
 741   }
 742 }
 743 
 744 void MacroAssembler::call_VM_base(Register oop_result,
 745                                   Register java_thread,
 746                                   Register last_java_sp,
 747                                   Label*   return_pc,
 748                                   address  entry_point,
 749                                   int      number_of_arguments,
 750                                   bool     check_exceptions) {
 751    // determine java_thread register
 752   if (!java_thread->is_valid()) {
 753     java_thread = rthread;
 754   }
 755 
 756   // determine last_java_sp register
 757   if (!last_java_sp->is_valid()) {
 758     last_java_sp = esp;
 759   }
 760 
 761   // debugging support
 762   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 763   assert(java_thread == rthread, "unexpected register");
 764 #ifdef ASSERT
 765   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 766   // if (!TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 767 #endif // ASSERT
 768 
 769   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 770   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 771 
 772   // push java thread (becomes first argument of C function)
 773 
 774   mov(c_rarg0, java_thread);
 775 
 776   // set last Java frame before call
 777   assert(last_java_sp != rfp, "can't use rfp");
 778 
 779   Label l;
 780   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 781 
 782   // do the call, remove parameters
 783   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 784 
 785   // lr could be poisoned with PAC signature during throw_pending_exception
 786   // if it was tail-call optimized by compiler, since lr is not callee-saved
 787   // reload it with proper value
 788   adr(lr, l);
 789 
 790   // reset last Java frame
 791   // Only interpreter should have to clear fp
 792   reset_last_Java_frame(true);
 793 
 794    // C++ interp handles this in the interpreter
 795   check_and_handle_popframe(java_thread);
 796   check_and_handle_earlyret(java_thread);
 797 
 798   if (check_exceptions) {
 799     // check for pending exceptions (java_thread is set upon return)
 800     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 801     Label ok;
 802     cbz(rscratch1, ok);
 803     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 804     br(rscratch1);
 805     bind(ok);
 806   }
 807 
 808   // get oop result if there is one and reset the value in the thread
 809   if (oop_result->is_valid()) {
 810     get_vm_result_oop(oop_result, java_thread);
 811   }
 812 }
 813 
 814 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 815   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 816 }
 817 
 818 // Check the entry target is always reachable from any branch.
 819 static bool is_always_within_branch_range(Address entry) {
 820   if (AOTCodeCache::is_on_for_dump()) {
 821     return false;
 822   }
 823   const address target = entry.target();
 824 
 825   if (!CodeCache::contains(target)) {
 826     // We always use trampolines for callees outside CodeCache.
 827     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 828     return false;
 829   }
 830 
 831   if (!MacroAssembler::far_branches()) {
 832     return true;
 833   }
 834 
 835   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 836     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 837     // Non-compiled methods stay forever in CodeCache.
 838     // We check whether the longest possible branch is within the branch range.
 839     assert(CodeCache::find_blob(target) != nullptr &&
 840           !CodeCache::find_blob(target)->is_nmethod(),
 841           "runtime call of compiled method");
 842     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 843     const address left_longest_branch_start = CodeCache::low_bound();
 844     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 845                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 846     return is_reachable;
 847   }
 848 
 849   return false;
 850 }
 851 
 852 // Maybe emit a call via a trampoline. If the code cache is small
 853 // trampolines won't be emitted.
 854 address MacroAssembler::trampoline_call(Address entry) {
 855   assert(entry.rspec().type() == relocInfo::runtime_call_type
 856          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 857          || entry.rspec().type() == relocInfo::static_call_type
 858          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 859 
 860   address target = entry.target();
 861 
 862   if (!is_always_within_branch_range(entry)) {
 863     if (!in_scratch_emit_size()) {
 864       // We don't want to emit a trampoline if C2 is generating dummy
 865       // code during its branch shortening phase.
 866       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 867         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 868         code()->share_trampoline_for(entry.target(), offset());
 869       } else {
 870         address stub = emit_trampoline_stub(offset(), target);
 871         if (stub == nullptr) {
 872           postcond(pc() == badAddress);
 873           return nullptr; // CodeCache is full
 874         }
 875       }
 876     }
 877     target = pc();
 878   }
 879 
 880   address call_pc = pc();
 881   relocate(entry.rspec());
 882   bl(target);
 883 
 884   postcond(pc() != badAddress);
 885   return call_pc;
 886 }
 887 
 888 // Emit a trampoline stub for a call to a target which is too far away.
 889 //
 890 // code sequences:
 891 //
 892 // call-site:
 893 //   branch-and-link to <destination> or <trampoline stub>
 894 //
 895 // Related trampoline stub for this call site in the stub section:
 896 //   load the call target from the constant pool
 897 //   branch (LR still points to the call site above)
 898 
 899 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 900                                              address dest) {
 901   // Max stub size: alignment nop, TrampolineStub.
 902   address stub = start_a_stub(max_trampoline_stub_size());
 903   if (stub == nullptr) {
 904     return nullptr;  // CodeBuffer::expand failed
 905   }
 906 
 907   // Create a trampoline stub relocation which relates this trampoline stub
 908   // with the call instruction at insts_call_instruction_offset in the
 909   // instructions code-section.
 910   align(wordSize);
 911   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 912                                             + insts_call_instruction_offset));
 913   const int stub_start_offset = offset();
 914 
 915   // Now, create the trampoline stub's code:
 916   // - load the call
 917   // - call
 918   Label target;
 919   ldr(rscratch1, target);
 920   br(rscratch1);
 921   bind(target);
 922   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 923          "should be");
 924   emit_int64((int64_t)dest);
 925 
 926   const address stub_start_addr = addr_at(stub_start_offset);
 927 
 928   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 929 
 930   end_a_stub();
 931   return stub_start_addr;
 932 }
 933 
 934 int MacroAssembler::max_trampoline_stub_size() {
 935   // Max stub size: alignment nop, TrampolineStub.
 936   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 937 }
 938 
 939 void MacroAssembler::emit_static_call_stub() {
 940   // CompiledDirectCall::set_to_interpreted knows the
 941   // exact layout of this stub.
 942 
 943   isb();
 944   mov_metadata(rmethod, nullptr);
 945 
 946   // Jump to the entry point of the c2i stub.
 947   if (codestub_branch_needs_far_jump()) {
 948     movptr(rscratch1, 0);
 949     br(rscratch1);
 950   } else {
 951     b(pc());
 952   }
 953 }
 954 
 955 int MacroAssembler::static_call_stub_size() {
 956   // During AOT production run AOT and JIT compiled code
 957   // are used at the same time. We need this size
 958   // to be the same for both types of code.
 959   if (!codestub_branch_needs_far_jump() && !AOTCodeCache::is_on_for_use()) {
 960     // isb; movk; movz; movz; b
 961     return 5 * NativeInstruction::instruction_size;
 962   }
 963   // isb; movk; movz; movz; movk; movz; movz; br
 964   return 8 * NativeInstruction::instruction_size;
 965 }
 966 
 967 void MacroAssembler::c2bool(Register x) {
 968   // implements x == 0 ? 0 : 1
 969   // note: must only look at least-significant byte of x
 970   //       since C-style booleans are stored in one byte
 971   //       only! (was bug)
 972   tst(x, 0xff);
 973   cset(x, Assembler::NE);
 974 }
 975 
 976 address MacroAssembler::ic_call(address entry, jint method_index) {
 977   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 978   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 979   return trampoline_call(Address(entry, rh));
 980 }
 981 
 982 int MacroAssembler::ic_check_size() {
 983   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 984   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 985     return NativeInstruction::instruction_size * (7 + extra_instructions);
 986   } else {
 987     return NativeInstruction::instruction_size * (5 + extra_instructions);
 988   }
 989 }
 990 
 991 int MacroAssembler::ic_check(int end_alignment) {
 992   Register receiver = j_rarg0;
 993   Register data = rscratch2;
 994   Register tmp1 = rscratch1;
 995   Register tmp2 = r10;
 996 
 997   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
 998   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
 999   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1000   // before the inline cache check here, and not after
1001   align(end_alignment, offset() + ic_check_size());
1002 
1003   int uep_offset = offset();
1004 
1005   if (UseCompactObjectHeaders) {
1006     load_narrow_klass_compact(tmp1, receiver);
1007     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1008     cmpw(tmp1, tmp2);
1009   } else {
1010     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1011     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1012     cmpw(tmp1, tmp2);
1013   }
1014 
1015   Label dont;
1016   br(Assembler::EQ, dont);
1017   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1018   bind(dont);
1019   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1020 
1021   return uep_offset;
1022 }
1023 
1024 // Implementation of call_VM versions
1025 
1026 void MacroAssembler::call_VM(Register oop_result,
1027                              address entry_point,
1028                              bool check_exceptions) {
1029   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1030 }
1031 
1032 void MacroAssembler::call_VM(Register oop_result,
1033                              address entry_point,
1034                              Register arg_1,
1035                              bool check_exceptions) {
1036   pass_arg1(this, arg_1);
1037   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1038 }
1039 
1040 void MacroAssembler::call_VM(Register oop_result,
1041                              address entry_point,
1042                              Register arg_1,
1043                              Register arg_2,
1044                              bool check_exceptions) {
1045   assert_different_registers(arg_1, c_rarg2);
1046   pass_arg2(this, arg_2);
1047   pass_arg1(this, arg_1);
1048   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1049 }
1050 
1051 void MacroAssembler::call_VM(Register oop_result,
1052                              address entry_point,
1053                              Register arg_1,
1054                              Register arg_2,
1055                              Register arg_3,
1056                              bool check_exceptions) {
1057   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1058   assert_different_registers(arg_2, c_rarg3);
1059   pass_arg3(this, arg_3);
1060 
1061   pass_arg2(this, arg_2);
1062 
1063   pass_arg1(this, arg_1);
1064   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1065 }
1066 
1067 void MacroAssembler::call_VM(Register oop_result,
1068                              Register last_java_sp,
1069                              address entry_point,
1070                              int number_of_arguments,
1071                              bool check_exceptions) {
1072   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1073 }
1074 
1075 void MacroAssembler::call_VM(Register oop_result,
1076                              Register last_java_sp,
1077                              address entry_point,
1078                              Register arg_1,
1079                              bool check_exceptions) {
1080   pass_arg1(this, arg_1);
1081   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1082 }
1083 
1084 void MacroAssembler::call_VM(Register oop_result,
1085                              Register last_java_sp,
1086                              address entry_point,
1087                              Register arg_1,
1088                              Register arg_2,
1089                              bool check_exceptions) {
1090 
1091   assert_different_registers(arg_1, c_rarg2);
1092   pass_arg2(this, arg_2);
1093   pass_arg1(this, arg_1);
1094   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1095 }
1096 
1097 void MacroAssembler::call_VM(Register oop_result,
1098                              Register last_java_sp,
1099                              address entry_point,
1100                              Register arg_1,
1101                              Register arg_2,
1102                              Register arg_3,
1103                              bool check_exceptions) {
1104   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1105   assert_different_registers(arg_2, c_rarg3);
1106   pass_arg3(this, arg_3);
1107   pass_arg2(this, arg_2);
1108   pass_arg1(this, arg_1);
1109   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1110 }
1111 
1112 
1113 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1114   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1115   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1116   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1117 }
1118 
1119 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1120   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1121   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1122 }
1123 
1124 void MacroAssembler::align(int modulus) {
1125   align(modulus, offset());
1126 }
1127 
1128 // Ensure that the code at target bytes offset from the current offset() is aligned
1129 // according to modulus.
1130 void MacroAssembler::align(int modulus, int target) {
1131   int delta = target - offset();
1132   while ((offset() + delta) % modulus != 0) nop();
1133 }
1134 
1135 void MacroAssembler::post_call_nop() {
1136   if (!Continuations::enabled()) {
1137     return;
1138   }
1139   InstructionMark im(this);
1140   relocate(post_call_nop_Relocation::spec());
1141   InlineSkippedInstructionsCounter skipCounter(this);
1142   nop();
1143   movk(zr, 0);
1144   movk(zr, 0);
1145 }
1146 
1147 // these are no-ops overridden by InterpreterMacroAssembler
1148 
1149 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1150 
1151 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1152 
1153 // Look up the method for a megamorphic invokeinterface call.
1154 // The target method is determined by <intf_klass, itable_index>.
1155 // The receiver klass is in recv_klass.
1156 // On success, the result will be in method_result, and execution falls through.
1157 // On failure, execution transfers to the given label.
1158 void MacroAssembler::lookup_interface_method(Register recv_klass,
1159                                              Register intf_klass,
1160                                              RegisterOrConstant itable_index,
1161                                              Register method_result,
1162                                              Register scan_temp,
1163                                              Label& L_no_such_interface,
1164                          bool return_method) {
1165   assert_different_registers(recv_klass, intf_klass, scan_temp);
1166   assert_different_registers(method_result, intf_klass, scan_temp);
1167   assert(recv_klass != method_result || !return_method,
1168      "recv_klass can be destroyed when method isn't needed");
1169   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1170          "caller must use same register for non-constant itable index as for method");
1171 
1172   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1173   int vtable_base = in_bytes(Klass::vtable_start_offset());
1174   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1175   int scan_step   = itableOffsetEntry::size() * wordSize;
1176   int vte_size    = vtableEntry::size_in_bytes();
1177   assert(vte_size == wordSize, "else adjust times_vte_scale");
1178 
1179   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1180 
1181   // Could store the aligned, prescaled offset in the klass.
1182   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1183   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1184   add(scan_temp, scan_temp, vtable_base);
1185 
1186   if (return_method) {
1187     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1188     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1189     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1190     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1191     if (itentry_off)
1192       add(recv_klass, recv_klass, itentry_off);
1193   }
1194 
1195   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1196   //   if (scan->interface() == intf) {
1197   //     result = (klass + scan->offset() + itable_index);
1198   //   }
1199   // }
1200   Label search, found_method;
1201 
1202   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1203   cmp(intf_klass, method_result);
1204   br(Assembler::EQ, found_method);
1205   bind(search);
1206   // Check that the previous entry is non-null.  A null entry means that
1207   // the receiver class doesn't implement the interface, and wasn't the
1208   // same as when the caller was compiled.
1209   cbz(method_result, L_no_such_interface);
1210   if (itableOffsetEntry::interface_offset() != 0) {
1211     add(scan_temp, scan_temp, scan_step);
1212     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1213   } else {
1214     ldr(method_result, Address(pre(scan_temp, scan_step)));
1215   }
1216   cmp(intf_klass, method_result);
1217   br(Assembler::NE, search);
1218 
1219   bind(found_method);
1220 
1221   // Got a hit.
1222   if (return_method) {
1223     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1224     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1225   }
1226 }
1227 
1228 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1229 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1230 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1231 // The target method is determined by <holder_klass, itable_index>.
1232 // The receiver klass is in recv_klass.
1233 // On success, the result will be in method_result, and execution falls through.
1234 // On failure, execution transfers to the given label.
1235 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1236                                                   Register holder_klass,
1237                                                   Register resolved_klass,
1238                                                   Register method_result,
1239                                                   Register temp_itbl_klass,
1240                                                   Register scan_temp,
1241                                                   int itable_index,
1242                                                   Label& L_no_such_interface) {
1243   // 'method_result' is only used as output register at the very end of this method.
1244   // Until then we can reuse it as 'holder_offset'.
1245   Register holder_offset = method_result;
1246   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1247 
1248   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1249   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1250   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1251   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1252 
1253   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1254 
1255   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1256   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1257   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1258   // temp_itbl_klass = itable[0]._interface;
1259   int vtblEntrySize = vtableEntry::size_in_bytes();
1260   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1261   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1262   mov(holder_offset, zr);
1263   // scan_temp = &(itable[0]._interface)
1264   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1265 
1266   // Initial checks:
1267   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1268   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1269   //   - if (itable[0] == 0), no such interface
1270   cmp(resolved_klass, holder_klass);
1271   br(Assembler::NE, L_loop_search_resolved_entry);
1272   cmp(holder_klass, temp_itbl_klass);
1273   br(Assembler::EQ, L_holder_found);
1274   cbz(temp_itbl_klass, L_no_such_interface);
1275 
1276   // Loop: Look for holder_klass record in itable
1277   //   do {
1278   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1279   //     if (temp_itbl_klass == holder_klass) {
1280   //       goto L_holder_found; // Found!
1281   //     }
1282   //   } while (temp_itbl_klass != 0);
1283   //   goto L_no_such_interface // Not found.
1284   Label L_search_holder;
1285   bind(L_search_holder);
1286     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1287     cmp(holder_klass, temp_itbl_klass);
1288     br(Assembler::EQ, L_holder_found);
1289     cbnz(temp_itbl_klass, L_search_holder);
1290 
1291   b(L_no_such_interface);
1292 
1293   // Loop: Look for resolved_class record in itable
1294   //   while (true) {
1295   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1296   //     if (temp_itbl_klass == 0) {
1297   //       goto L_no_such_interface;
1298   //     }
1299   //     if (temp_itbl_klass == resolved_klass) {
1300   //        goto L_resolved_found;  // Found!
1301   //     }
1302   //     if (temp_itbl_klass == holder_klass) {
1303   //        holder_offset = scan_temp;
1304   //     }
1305   //   }
1306   //
1307   Label L_loop_search_resolved;
1308   bind(L_loop_search_resolved);
1309     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1310   bind(L_loop_search_resolved_entry);
1311     cbz(temp_itbl_klass, L_no_such_interface);
1312     cmp(resolved_klass, temp_itbl_klass);
1313     br(Assembler::EQ, L_resolved_found);
1314     cmp(holder_klass, temp_itbl_klass);
1315     br(Assembler::NE, L_loop_search_resolved);
1316     mov(holder_offset, scan_temp);
1317     b(L_loop_search_resolved);
1318 
1319   // See if we already have a holder klass. If not, go and scan for it.
1320   bind(L_resolved_found);
1321   cbz(holder_offset, L_search_holder);
1322   mov(scan_temp, holder_offset);
1323 
1324   // Finally, scan_temp contains holder_klass vtable offset
1325   bind(L_holder_found);
1326   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1327   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1328     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1329   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1330 }
1331 
1332 // virtual method calling
1333 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1334                                            RegisterOrConstant vtable_index,
1335                                            Register method_result) {
1336   assert(vtableEntry::size() * wordSize == 8,
1337          "adjust the scaling in the code below");
1338   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1339 
1340   if (vtable_index.is_register()) {
1341     lea(method_result, Address(recv_klass,
1342                                vtable_index.as_register(),
1343                                Address::lsl(LogBytesPerWord)));
1344     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1345   } else {
1346     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1347     ldr(method_result,
1348         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1349   }
1350 }
1351 
1352 void MacroAssembler::check_klass_subtype(Register sub_klass,
1353                            Register super_klass,
1354                            Register temp_reg,
1355                            Label& L_success) {
1356   Label L_failure;
1357   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1358   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1359   bind(L_failure);
1360 }
1361 
1362 
1363 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1364                                                    Register super_klass,
1365                                                    Register temp_reg,
1366                                                    Label* L_success,
1367                                                    Label* L_failure,
1368                                                    Label* L_slow_path,
1369                                                    Register super_check_offset) {
1370   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1371   bool must_load_sco = ! super_check_offset->is_valid();
1372   if (must_load_sco) {
1373     assert(temp_reg != noreg, "supply either a temp or a register offset");
1374   }
1375 
1376   Label L_fallthrough;
1377   int label_nulls = 0;
1378   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1379   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1380   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1381   assert(label_nulls <= 1, "at most one null in the batch");
1382 
1383   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1384   Address super_check_offset_addr(super_klass, sco_offset);
1385 
1386   // Hacked jmp, which may only be used just before L_fallthrough.
1387 #define final_jmp(label)                                                \
1388   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1389   else                            b(label)                /*omit semi*/
1390 
1391   // If the pointers are equal, we are done (e.g., String[] elements).
1392   // This self-check enables sharing of secondary supertype arrays among
1393   // non-primary types such as array-of-interface.  Otherwise, each such
1394   // type would need its own customized SSA.
1395   // We move this check to the front of the fast path because many
1396   // type checks are in fact trivially successful in this manner,
1397   // so we get a nicely predicted branch right at the start of the check.
1398   cmp(sub_klass, super_klass);
1399   br(Assembler::EQ, *L_success);
1400 
1401   // Check the supertype display:
1402   if (must_load_sco) {
1403     ldrw(temp_reg, super_check_offset_addr);
1404     super_check_offset = temp_reg;
1405   }
1406 
1407   Address super_check_addr(sub_klass, super_check_offset);
1408   ldr(rscratch1, super_check_addr);
1409   cmp(super_klass, rscratch1); // load displayed supertype
1410   br(Assembler::EQ, *L_success);
1411 
1412   // This check has worked decisively for primary supers.
1413   // Secondary supers are sought in the super_cache ('super_cache_addr').
1414   // (Secondary supers are interfaces and very deeply nested subtypes.)
1415   // This works in the same check above because of a tricky aliasing
1416   // between the super_cache and the primary super display elements.
1417   // (The 'super_check_addr' can address either, as the case requires.)
1418   // Note that the cache is updated below if it does not help us find
1419   // what we need immediately.
1420   // So if it was a primary super, we can just fail immediately.
1421   // Otherwise, it's the slow path for us (no success at this point).
1422 
1423   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1424   if (L_failure == &L_fallthrough) {
1425     cbz(rscratch1, *L_slow_path);
1426   } else {
1427     cbnz(rscratch1, *L_failure);
1428     final_jmp(*L_slow_path);
1429   }
1430 
1431   bind(L_fallthrough);
1432 
1433 #undef final_jmp
1434 }
1435 
1436 // These two are taken from x86, but they look generally useful
1437 
1438 // scans count pointer sized words at [addr] for occurrence of value,
1439 // generic
1440 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1441                                 Register scratch) {
1442   Label Lloop, Lexit;
1443   cbz(count, Lexit);
1444   bind(Lloop);
1445   ldr(scratch, post(addr, wordSize));
1446   cmp(value, scratch);
1447   br(EQ, Lexit);
1448   sub(count, count, 1);
1449   cbnz(count, Lloop);
1450   bind(Lexit);
1451 }
1452 
1453 // scans count 4 byte words at [addr] for occurrence of value,
1454 // generic
1455 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1456                                 Register scratch) {
1457   Label Lloop, Lexit;
1458   cbz(count, Lexit);
1459   bind(Lloop);
1460   ldrw(scratch, post(addr, wordSize));
1461   cmpw(value, scratch);
1462   br(EQ, Lexit);
1463   sub(count, count, 1);
1464   cbnz(count, Lloop);
1465   bind(Lexit);
1466 }
1467 
1468 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1469                                                           Register super_klass,
1470                                                           Register temp_reg,
1471                                                           Register temp2_reg,
1472                                                           Label* L_success,
1473                                                           Label* L_failure,
1474                                                           bool set_cond_codes) {
1475   // NB! Callers may assume that, when temp2_reg is a valid register,
1476   // this code sets it to a nonzero value.
1477 
1478   assert_different_registers(sub_klass, super_klass, temp_reg);
1479   if (temp2_reg != noreg)
1480     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1481 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1482 
1483   Label L_fallthrough;
1484   int label_nulls = 0;
1485   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1486   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1487   assert(label_nulls <= 1, "at most one null in the batch");
1488 
1489   // a couple of useful fields in sub_klass:
1490   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1491   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1492   Address secondary_supers_addr(sub_klass, ss_offset);
1493   Address super_cache_addr(     sub_klass, sc_offset);
1494 
1495   BLOCK_COMMENT("check_klass_subtype_slow_path");
1496 
1497   // Do a linear scan of the secondary super-klass chain.
1498   // This code is rarely used, so simplicity is a virtue here.
1499   // The repne_scan instruction uses fixed registers, which we must spill.
1500   // Don't worry too much about pre-existing connections with the input regs.
1501 
1502   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1503   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1504 
1505   RegSet pushed_registers;
1506   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1507   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1508 
1509   if (super_klass != r0) {
1510     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1511   }
1512 
1513   push(pushed_registers, sp);
1514 
1515   // Get super_klass value into r0 (even if it was in r5 or r2).
1516   if (super_klass != r0) {
1517     mov(r0, super_klass);
1518   }
1519 
1520 #ifndef PRODUCT
1521   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1522 #endif //PRODUCT
1523 
1524   // We will consult the secondary-super array.
1525   ldr(r5, secondary_supers_addr);
1526   // Load the array length.
1527   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1528   // Skip to start of data.
1529   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1530 
1531   cmp(sp, zr); // Clear Z flag; SP is never zero
1532   // Scan R2 words at [R5] for an occurrence of R0.
1533   // Set NZ/Z based on last compare.
1534   repne_scan(r5, r0, r2, rscratch1);
1535 
1536   // Unspill the temp. registers:
1537   pop(pushed_registers, sp);
1538 
1539   br(Assembler::NE, *L_failure);
1540 
1541   // Success.  Cache the super we found and proceed in triumph.
1542 
1543   if (UseSecondarySupersCache) {
1544     str(super_klass, super_cache_addr);
1545   }
1546 
1547   if (L_success != &L_fallthrough) {
1548     b(*L_success);
1549   }
1550 
1551 #undef IS_A_TEMP
1552 
1553   bind(L_fallthrough);
1554 }
1555 
1556 // If Register r is invalid, remove a new register from
1557 // available_regs, and add new register to regs_to_push.
1558 Register MacroAssembler::allocate_if_noreg(Register r,
1559                                   RegSetIterator<Register> &available_regs,
1560                                   RegSet &regs_to_push) {
1561   if (!r->is_valid()) {
1562     r = *available_regs++;
1563     regs_to_push += r;
1564   }
1565   return r;
1566 }
1567 
1568 // check_klass_subtype_slow_path_table() looks for super_klass in the
1569 // hash table belonging to super_klass, branching to L_success or
1570 // L_failure as appropriate. This is essentially a shim which
1571 // allocates registers as necessary then calls
1572 // lookup_secondary_supers_table() to do the work. Any of the temp
1573 // regs may be noreg, in which case this logic will chooses some
1574 // registers push and pop them from the stack.
1575 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1576                                                          Register super_klass,
1577                                                          Register temp_reg,
1578                                                          Register temp2_reg,
1579                                                          Register temp3_reg,
1580                                                          Register result_reg,
1581                                                          FloatRegister vtemp,
1582                                                          Label* L_success,
1583                                                          Label* L_failure,
1584                                                          bool set_cond_codes) {
1585   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1586 
1587   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1588 
1589   Label L_fallthrough;
1590   int label_nulls = 0;
1591   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1592   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1593   assert(label_nulls <= 1, "at most one null in the batch");
1594 
1595   BLOCK_COMMENT("check_klass_subtype_slow_path");
1596 
1597   RegSetIterator<Register> available_regs
1598     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1599 
1600   RegSet pushed_regs;
1601 
1602   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1603   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1604   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1605   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1606 
1607   push(pushed_regs, sp);
1608 
1609   lookup_secondary_supers_table_var(sub_klass,
1610                                     super_klass,
1611                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1612                                     nullptr);
1613   cmp(result_reg, zr);
1614 
1615   // Unspill the temp. registers:
1616   pop(pushed_regs, sp);
1617 
1618   // NB! Callers may assume that, when set_cond_codes is true, this
1619   // code sets temp2_reg to a nonzero value.
1620   if (set_cond_codes) {
1621     mov(temp2_reg, 1);
1622   }
1623 
1624   br(Assembler::NE, *L_failure);
1625 
1626   if (L_success != &L_fallthrough) {
1627     b(*L_success);
1628   }
1629 
1630   bind(L_fallthrough);
1631 }
1632 
1633 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1634                                                    Register super_klass,
1635                                                    Register temp_reg,
1636                                                    Register temp2_reg,
1637                                                    Label* L_success,
1638                                                    Label* L_failure,
1639                                                    bool set_cond_codes) {
1640   if (UseSecondarySupersTable) {
1641     check_klass_subtype_slow_path_table
1642       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1643        /*vtemp*/fnoreg,
1644        L_success, L_failure, set_cond_codes);
1645   } else {
1646     check_klass_subtype_slow_path_linear
1647       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1648   }
1649 }
1650 
1651 
1652 // Ensure that the inline code and the stub are using the same registers.
1653 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1654 do {                                                               \
1655   assert(r_super_klass  == r0                                   && \
1656          r_array_base   == r1                                   && \
1657          r_array_length == r2                                   && \
1658          (r_array_index == r3        || r_array_index == noreg) && \
1659          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1660          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1661          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1662 } while(0)
1663 
1664 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1665                                                          Register r_super_klass,
1666                                                          Register temp1,
1667                                                          Register temp2,
1668                                                          Register temp3,
1669                                                          FloatRegister vtemp,
1670                                                          Register result,
1671                                                          u1 super_klass_slot,
1672                                                          bool stub_is_near) {
1673   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1674 
1675   Label L_fallthrough;
1676 
1677   BLOCK_COMMENT("lookup_secondary_supers_table {");
1678 
1679   const Register
1680     r_array_base   = temp1, // r1
1681     r_array_length = temp2, // r2
1682     r_array_index  = temp3, // r3
1683     r_bitmap       = rscratch2;
1684 
1685   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1686 
1687   u1 bit = super_klass_slot;
1688 
1689   // Make sure that result is nonzero if the TBZ below misses.
1690   mov(result, 1);
1691 
1692   // We're going to need the bitmap in a vector reg and in a core reg,
1693   // so load both now.
1694   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1695   if (bit != 0) {
1696     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1697   }
1698   // First check the bitmap to see if super_klass might be present. If
1699   // the bit is zero, we are certain that super_klass is not one of
1700   // the secondary supers.
1701   tbz(r_bitmap, bit, L_fallthrough);
1702 
1703   // Get the first array index that can contain super_klass into r_array_index.
1704   if (bit != 0) {
1705     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1706     cnt(vtemp, T8B, vtemp);
1707     addv(vtemp, T8B, vtemp);
1708     fmovd(r_array_index, vtemp);
1709   } else {
1710     mov(r_array_index, (u1)1);
1711   }
1712   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1713 
1714   // We will consult the secondary-super array.
1715   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1716 
1717   // The value i in r_array_index is >= 1, so even though r_array_base
1718   // points to the length, we don't need to adjust it to point to the
1719   // data.
1720   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1721   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1722 
1723   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1724   eor(result, result, r_super_klass);
1725   cbz(result, L_fallthrough); // Found a match
1726 
1727   // Is there another entry to check? Consult the bitmap.
1728   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1729 
1730   // Linear probe.
1731   if (bit != 0) {
1732     ror(r_bitmap, r_bitmap, bit);
1733   }
1734 
1735   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1736   // The next slot to be inspected, by the stub we're about to call,
1737   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1738   // have been checked.
1739   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1740   if (stub_is_near) {
1741     bl(stub);
1742   } else {
1743     address call = trampoline_call(stub);
1744     if (call == nullptr) {
1745       return false; // trampoline allocation failed
1746     }
1747   }
1748 
1749   BLOCK_COMMENT("} lookup_secondary_supers_table");
1750 
1751   bind(L_fallthrough);
1752 
1753   if (VerifySecondarySupers) {
1754     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1755                                   temp1, temp2, result);      // r1, r2, r5
1756   }
1757   return true;
1758 }
1759 
1760 // At runtime, return 0 in result if r_super_klass is a superclass of
1761 // r_sub_klass, otherwise return nonzero. Use this version of
1762 // lookup_secondary_supers_table() if you don't know ahead of time
1763 // which superclass will be searched for. Used by interpreter and
1764 // runtime stubs. It is larger and has somewhat greater latency than
1765 // the version above, which takes a constant super_klass_slot.
1766 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1767                                                        Register r_super_klass,
1768                                                        Register temp1,
1769                                                        Register temp2,
1770                                                        Register temp3,
1771                                                        FloatRegister vtemp,
1772                                                        Register result,
1773                                                        Label *L_success) {
1774   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1775 
1776   Label L_fallthrough;
1777 
1778   BLOCK_COMMENT("lookup_secondary_supers_table {");
1779 
1780   const Register
1781     r_array_index = temp3,
1782     slot          = rscratch1,
1783     r_bitmap      = rscratch2;
1784 
1785   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1786 
1787   // Make sure that result is nonzero if the test below misses.
1788   mov(result, 1);
1789 
1790   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1791 
1792   // First check the bitmap to see if super_klass might be present. If
1793   // the bit is zero, we are certain that super_klass is not one of
1794   // the secondary supers.
1795 
1796   // This next instruction is equivalent to:
1797   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1798   // sub(temp2, tmp_reg, slot);
1799   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1800   lslv(temp2, r_bitmap, temp2);
1801   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1802 
1803   bool must_save_v0 = (vtemp == fnoreg);
1804   if (must_save_v0) {
1805     // temp1 and result are free, so use them to preserve vtemp
1806     vtemp = v0;
1807     mov(temp1,  vtemp, D, 0);
1808     mov(result, vtemp, D, 1);
1809   }
1810 
1811   // Get the first array index that can contain super_klass into r_array_index.
1812   mov(vtemp, D, 0, temp2);
1813   cnt(vtemp, T8B, vtemp);
1814   addv(vtemp, T8B, vtemp);
1815   mov(r_array_index, vtemp, D, 0);
1816 
1817   if (must_save_v0) {
1818     mov(vtemp, D, 0, temp1 );
1819     mov(vtemp, D, 1, result);
1820   }
1821 
1822   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1823 
1824   const Register
1825     r_array_base   = temp1,
1826     r_array_length = temp2;
1827 
1828   // The value i in r_array_index is >= 1, so even though r_array_base
1829   // points to the length, we don't need to adjust it to point to the
1830   // data.
1831   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1832   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1833 
1834   // We will consult the secondary-super array.
1835   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1836 
1837   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1838   eor(result, result, r_super_klass);
1839   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1840 
1841   // Is there another entry to check? Consult the bitmap.
1842   rorv(r_bitmap, r_bitmap, slot);
1843   // rol(r_bitmap, r_bitmap, 1);
1844   tbz(r_bitmap, 1, L_fallthrough);
1845 
1846   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1847   // The next slot to be inspected, by the logic we're about to call,
1848   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1849   // have been checked.
1850   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1851                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1852 
1853   BLOCK_COMMENT("} lookup_secondary_supers_table");
1854 
1855   bind(L_fallthrough);
1856 
1857   if (VerifySecondarySupers) {
1858     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1859                                   temp1, temp2, result);      // r1, r2, r5
1860   }
1861 
1862   if (L_success) {
1863     cbz(result, *L_success);
1864   }
1865 }
1866 
1867 // Called by code generated by check_klass_subtype_slow_path
1868 // above. This is called when there is a collision in the hashed
1869 // lookup in the secondary supers array.
1870 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1871                                                              Register r_array_base,
1872                                                              Register r_array_index,
1873                                                              Register r_bitmap,
1874                                                              Register temp1,
1875                                                              Register result,
1876                                                              bool is_stub) {
1877   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1878 
1879   const Register
1880     r_array_length = temp1,
1881     r_sub_klass    = noreg; // unused
1882 
1883   if (is_stub) {
1884     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1885   }
1886 
1887   Label L_fallthrough, L_huge;
1888 
1889   // Load the array length.
1890   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1891   // And adjust the array base to point to the data.
1892   // NB! Effectively increments current slot index by 1.
1893   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1894   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1895 
1896   // The bitmap is full to bursting.
1897   // Implicit invariant: BITMAP_FULL implies (length > 0)
1898   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1899   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1900   br(GT, L_huge);
1901 
1902   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1903   // current slot (at secondary_supers[r_array_index]) has not yet
1904   // been inspected, and r_array_index may be out of bounds if we
1905   // wrapped around the end of the array.
1906 
1907   { // This is conventional linear probing, but instead of terminating
1908     // when a null entry is found in the table, we maintain a bitmap
1909     // in which a 0 indicates missing entries.
1910     // As long as the bitmap is not completely full,
1911     // array_length == popcount(bitmap). The array_length check above
1912     // guarantees there are 0s in the bitmap, so the loop eventually
1913     // terminates.
1914     Label L_loop;
1915     bind(L_loop);
1916 
1917     // Check for wraparound.
1918     cmp(r_array_index, r_array_length);
1919     csel(r_array_index, zr, r_array_index, GE);
1920 
1921     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1922     eor(result, rscratch1, r_super_klass);
1923     cbz(result, L_fallthrough);
1924 
1925     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1926 
1927     ror(r_bitmap, r_bitmap, 1);
1928     add(r_array_index, r_array_index, 1);
1929     b(L_loop);
1930   }
1931 
1932   { // Degenerate case: more than 64 secondary supers.
1933     // FIXME: We could do something smarter here, maybe a vectorized
1934     // comparison or a binary search, but is that worth any added
1935     // complexity?
1936     bind(L_huge);
1937     cmp(sp, zr); // Clear Z flag; SP is never zero
1938     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1939     cset(result, NE); // result == 0 iff we got a match.
1940   }
1941 
1942   bind(L_fallthrough);
1943 }
1944 
1945 // Make sure that the hashed lookup and a linear scan agree.
1946 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1947                                                    Register r_super_klass,
1948                                                    Register temp1,
1949                                                    Register temp2,
1950                                                    Register result) {
1951   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1952 
1953   const Register
1954     r_array_base   = temp1,
1955     r_array_length = temp2;
1956 
1957   BLOCK_COMMENT("verify_secondary_supers_table {");
1958 
1959   // We will consult the secondary-super array.
1960   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1961 
1962   // Load the array length.
1963   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1964   // And adjust the array base to point to the data.
1965   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1966 
1967   cmp(sp, zr); // Clear Z flag; SP is never zero
1968   // Scan R2 words at [R5] for an occurrence of R0.
1969   // Set NZ/Z based on last compare.
1970   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1971   // rscratch1 == 0 iff we got a match.
1972   cset(rscratch1, NE);
1973 
1974   Label passed;
1975   cmp(result, zr);
1976   cset(result, NE); // normalize result to 0/1 for comparison
1977 
1978   cmp(rscratch1, result);
1979   br(EQ, passed);
1980   {
1981     mov(r0, r_super_klass);         // r0 <- r0
1982     mov(r1, r_sub_klass);           // r1 <- r4
1983     mov(r2, /*expected*/rscratch1); // r2 <- r8
1984     mov(r3, result);                // r3 <- r5
1985     mov(r4, (address)("mismatch")); // r4 <- const
1986     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
1987     should_not_reach_here();
1988   }
1989   bind(passed);
1990 
1991   BLOCK_COMMENT("} verify_secondary_supers_table");
1992 }
1993 
1994 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
1995   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
1996   assert_different_registers(klass, rthread, scratch);
1997 
1998   Label L_fallthrough, L_tmp;
1999   if (L_fast_path == nullptr) {
2000     L_fast_path = &L_fallthrough;
2001   } else if (L_slow_path == nullptr) {
2002     L_slow_path = &L_fallthrough;
2003   }
2004   // Fast path check: class is fully initialized
2005   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2006   ldarb(scratch, scratch);
2007   cmp(scratch, InstanceKlass::fully_initialized);
2008   br(Assembler::EQ, *L_fast_path);
2009 
2010   // Fast path check: current thread is initializer thread
2011   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2012   cmp(rthread, scratch);
2013 
2014   if (L_slow_path == &L_fallthrough) {
2015     br(Assembler::EQ, *L_fast_path);
2016     bind(*L_slow_path);
2017   } else if (L_fast_path == &L_fallthrough) {
2018     br(Assembler::NE, *L_slow_path);
2019     bind(*L_fast_path);
2020   } else {
2021     Unimplemented();
2022   }
2023 }
2024 
2025 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2026   if (!VerifyOops) return;
2027 
2028   // Pass register number to verify_oop_subroutine
2029   const char* b = nullptr;
2030   {
2031     ResourceMark rm;
2032     stringStream ss;
2033     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2034     b = code_string(ss.as_string());
2035   }
2036   BLOCK_COMMENT("verify_oop {");
2037 
2038   strip_return_address(); // This might happen within a stack frame.
2039   protect_return_address();
2040   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2041   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2042 
2043   mov(r0, reg);
2044   movptr(rscratch1, (uintptr_t)(address)b);
2045 
2046   // call indirectly to solve generation ordering problem
2047   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2048   ldr(rscratch2, Address(rscratch2));
2049   blr(rscratch2);
2050 
2051   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2052   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2053   authenticate_return_address();
2054 
2055   BLOCK_COMMENT("} verify_oop");
2056 }
2057 
2058 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2059   if (!VerifyOops) return;
2060 
2061   const char* b = nullptr;
2062   {
2063     ResourceMark rm;
2064     stringStream ss;
2065     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2066     b = code_string(ss.as_string());
2067   }
2068   BLOCK_COMMENT("verify_oop_addr {");
2069 
2070   strip_return_address(); // This might happen within a stack frame.
2071   protect_return_address();
2072   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2073   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2074 
2075   // addr may contain sp so we will have to adjust it based on the
2076   // pushes that we just did.
2077   if (addr.uses(sp)) {
2078     lea(r0, addr);
2079     ldr(r0, Address(r0, 4 * wordSize));
2080   } else {
2081     ldr(r0, addr);
2082   }
2083   movptr(rscratch1, (uintptr_t)(address)b);
2084 
2085   // call indirectly to solve generation ordering problem
2086   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2087   ldr(rscratch2, Address(rscratch2));
2088   blr(rscratch2);
2089 
2090   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2091   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2092   authenticate_return_address();
2093 
2094   BLOCK_COMMENT("} verify_oop_addr");
2095 }
2096 
2097 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2098                                          int extra_slot_offset) {
2099   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2100   int stackElementSize = Interpreter::stackElementSize;
2101   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2102 #ifdef ASSERT
2103   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2104   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2105 #endif
2106   if (arg_slot.is_constant()) {
2107     return Address(esp, arg_slot.as_constant() * stackElementSize
2108                    + offset);
2109   } else {
2110     add(rscratch1, esp, arg_slot.as_register(),
2111         ext::uxtx, exact_log2(stackElementSize));
2112     return Address(rscratch1, offset);
2113   }
2114 }
2115 
2116 // Handle the receiver type profile update given the "recv" klass.
2117 //
2118 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2119 // If there are no matching or claimable receiver entries in RD, updates
2120 // the polymorphic counter.
2121 //
2122 // This code expected to run by either the interpreter or JIT-ed code, without
2123 // extra synchronization. For safety, receiver cells are claimed atomically, which
2124 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2125 // counter updates are not atomic.
2126 //
2127 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2128   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2129 
2130   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2131   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2132   int poly_count_offset      = in_bytes(CounterData::count_offset());
2133   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2134   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2135 
2136   // Adjust for MDP offsets.
2137   base_receiver_offset += mdp_offset;
2138   end_receiver_offset  += mdp_offset;
2139   poly_count_offset    += mdp_offset;
2140 
2141 #ifdef ASSERT
2142   // We are about to walk the MDO slots without asking for offsets.
2143   // Check that our math hits all the right spots.
2144   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2145     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2146     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2147     int offset = base_receiver_offset + receiver_step*c;
2148     int count_offset = offset + receiver_to_count_step;
2149     assert(offset == real_recv_offset, "receiver slot math");
2150     assert(count_offset == real_count_offset, "receiver count math");
2151   }
2152   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2153   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2154 #endif
2155 
2156   // Corner case: no profile table. Increment poly counter and exit.
2157   if (ReceiverTypeData::row_limit() == 0) {
2158     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2159     return;
2160   }
2161 
2162   Register offset = rscratch2;
2163 
2164   Label L_loop_search_receiver, L_loop_search_empty;
2165   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
2166 
2167   // The code here recognizes three major cases:
2168   //   A. Fastest: receiver found in the table
2169   //   B. Fast: no receiver in the table, and the table is full
2170   //   C. Slow: no receiver in the table, free slots in the table
2171   //
2172   // The case A performance is most important, as perfectly-behaved code would end up
2173   // there, especially with larger TypeProfileWidth. The case B performance is
2174   // important as well, this is where bulk of code would land for normally megamorphic
2175   // cases. The case C performance is not essential, its job is to deal with installation
2176   // races, we optimize for code density instead. Case C needs to make sure that receiver
2177   // rows are only claimed once. This makes sure we never overwrite a row for another
2178   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2179   //
2180   // It is very tempting to handle these cases in a single loop, and claim the first slot
2181   // without checking the rest of the table. But, profiling code should tolerate free slots
2182   // in the table, as class unloading can clear them. After such cleanup, the receiver
2183   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2184   // to complete, before trying to install new slots. Splitting the code in several tight
2185   // loops also helpfully optimizes for cases A and B.
2186   //
2187   // This code is effectively:
2188   //
2189   // restart:
2190   //   // Fastest: receiver is already installed
2191   //   for (i = 0; i < receiver_count(); i++) {
2192   //     if (receiver(i) == recv) goto found_recv(i);
2193   //   }
2194   //
2195   //   // Fast: no receiver, but profile is full
2196   //   for (i = 0; i < receiver_count(); i++) {
2197   //     if (receiver(i) == null) goto found_null(i);
2198   //   }
2199   //   goto polymorphic
2200   //
2201   //   // Slow: try to install receiver
2202   // found_null(i):
2203   //   CAS(&receiver(i), null, recv);
2204   //   goto restart
2205   //
2206   // polymorphic:
2207   //   count++;
2208   //   return
2209   //
2210   // found_recv(i):
2211   //   *receiver_count(i)++
2212   //
2213 
2214   bind(L_restart);
2215 
2216   // Fastest: receiver is already installed
2217   mov(offset, base_receiver_offset);
2218   bind(L_loop_search_receiver);
2219     ldr(rscratch1, Address(mdp, offset));
2220     cmp(rscratch1, recv);
2221     br(Assembler::EQ, L_found_recv);
2222   add(offset, offset, receiver_step);
2223   sub(rscratch1, offset, end_receiver_offset);
2224   cbnz(rscratch1, L_loop_search_receiver);
2225 
2226   // Fast: no receiver, but profile is full
2227   mov(offset, base_receiver_offset);
2228   bind(L_loop_search_empty);
2229     ldr(rscratch1, Address(mdp, offset));
2230     cbz(rscratch1, L_found_empty);
2231   add(offset, offset, receiver_step);
2232   sub(rscratch1, offset, end_receiver_offset);
2233   cbnz(rscratch1, L_loop_search_empty);
2234   b(L_polymorphic);
2235 
2236   // Slow: try to install receiver
2237   bind(L_found_empty);
2238 
2239   // Atomically swing receiver slot: null -> recv.
2240   //
2241   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2242   // is used to hold the destination address. This is safe because the
2243   // offset is no longer needed after the address is computed.
2244 
2245   lea(rscratch2, Address(mdp, offset));
2246   cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
2247           /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
2248 
2249   // CAS success means the slot now has the receiver we want. CAS failure means
2250   // something had claimed the slot concurrently: it can be the same receiver we want,
2251   // or something else. Since this is a slow path, we can optimize for code density,
2252   // and just restart the search from the beginning.
2253   b(L_restart);
2254 
2255   // Counter updates:
2256 
2257   // Increment polymorphic counter instead of receiver slot.
2258   bind(L_polymorphic);
2259   mov(offset, poly_count_offset);
2260   b(L_count_update);
2261 
2262   // Found a receiver, convert its slot offset to corresponding count offset.
2263   bind(L_found_recv);
2264   add(offset, offset, receiver_to_count_step);
2265 
2266   bind(L_count_update);
2267   increment(Address(mdp, offset), DataLayout::counter_increment);
2268 }
2269 
2270 
2271 void MacroAssembler::call_VM_leaf_base(address entry_point,
2272                                        int number_of_arguments,
2273                                        Label *retaddr) {
2274   Label E, L;
2275 
2276   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2277 
2278   mov(rscratch1, RuntimeAddress(entry_point));
2279   blr(rscratch1);
2280   if (retaddr)
2281     bind(*retaddr);
2282 
2283   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2284 }
2285 
2286 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2287   call_VM_leaf_base(entry_point, number_of_arguments);
2288 }
2289 
2290 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2291   pass_arg0(this, arg_0);
2292   call_VM_leaf_base(entry_point, 1);
2293 }
2294 
2295 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2296   assert_different_registers(arg_1, c_rarg0);
2297   pass_arg0(this, arg_0);
2298   pass_arg1(this, arg_1);
2299   call_VM_leaf_base(entry_point, 2);
2300 }
2301 
2302 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2303                                   Register arg_1, Register arg_2) {
2304   assert_different_registers(arg_1, c_rarg0);
2305   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2306   pass_arg0(this, arg_0);
2307   pass_arg1(this, arg_1);
2308   pass_arg2(this, arg_2);
2309   call_VM_leaf_base(entry_point, 3);
2310 }
2311 
2312 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2313   pass_arg0(this, arg_0);
2314   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2315 }
2316 
2317 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2318 
2319   assert_different_registers(arg_0, c_rarg1);
2320   pass_arg1(this, arg_1);
2321   pass_arg0(this, arg_0);
2322   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2323 }
2324 
2325 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2326   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2327   assert_different_registers(arg_1, c_rarg2);
2328   pass_arg2(this, arg_2);
2329   pass_arg1(this, arg_1);
2330   pass_arg0(this, arg_0);
2331   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2332 }
2333 
2334 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2335   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2336   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2337   assert_different_registers(arg_2, c_rarg3);
2338   pass_arg3(this, arg_3);
2339   pass_arg2(this, arg_2);
2340   pass_arg1(this, arg_1);
2341   pass_arg0(this, arg_0);
2342   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2343 }
2344 
2345 void MacroAssembler::null_check(Register reg, int offset) {
2346   if (needs_explicit_null_check(offset)) {
2347     // provoke OS null exception if reg is null by
2348     // accessing M[reg] w/o changing any registers
2349     // NOTE: this is plenty to provoke a segv
2350     ldr(zr, Address(reg));
2351   } else {
2352     // nothing to do, (later) access of M[reg + offset]
2353     // will provoke OS null exception if reg is null
2354   }
2355 }
2356 
2357 // MacroAssembler protected routines needed to implement
2358 // public methods
2359 
2360 void MacroAssembler::mov(Register r, Address dest) {
2361   code_section()->relocate(pc(), dest.rspec());
2362   uint64_t imm64 = (uint64_t)dest.target();
2363   movptr(r, imm64);
2364 }
2365 
2366 // Move a constant pointer into r.  In AArch64 mode the virtual
2367 // address space is 48 bits in size, so we only need three
2368 // instructions to create a patchable instruction sequence that can
2369 // reach anywhere.
2370 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2371 #ifndef PRODUCT
2372   {
2373     char buffer[64];
2374     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2375     block_comment(buffer);
2376   }
2377 #endif
2378   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2379   movz(r, imm64 & 0xffff);
2380   imm64 >>= 16;
2381   movk(r, imm64 & 0xffff, 16);
2382   imm64 >>= 16;
2383   movk(r, imm64 & 0xffff, 32);
2384 }
2385 
2386 // Macro to mov replicated immediate to vector register.
2387 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2388 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2389 // Vd will get the following values for different arrangements in T
2390 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2391 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2392 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2393 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2394 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2395 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2396 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2397 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2398 // Clobbers rscratch1
2399 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2400   assert(T != T1Q, "unsupported");
2401   if (T == T1D || T == T2D) {
2402     int imm = operand_valid_for_movi_immediate(imm64, T);
2403     if (-1 != imm) {
2404       movi(Vd, T, imm);
2405     } else {
2406       mov(rscratch1, imm64);
2407       dup(Vd, T, rscratch1);
2408     }
2409     return;
2410   }
2411 
2412 #ifdef ASSERT
2413   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2414   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2415   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2416 #endif
2417   int shift = operand_valid_for_movi_immediate(imm64, T);
2418   uint32_t imm32 = imm64 & 0xffffffffULL;
2419   if (shift >= 0) {
2420     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2421   } else {
2422     movw(rscratch1, imm32);
2423     dup(Vd, T, rscratch1);
2424   }
2425 }
2426 
2427 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2428 {
2429 #ifndef PRODUCT
2430   {
2431     char buffer[64];
2432     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2433     block_comment(buffer);
2434   }
2435 #endif
2436   if (operand_valid_for_logical_immediate(false, imm64)) {
2437     orr(dst, zr, imm64);
2438   } else {
2439     // we can use a combination of MOVZ or MOVN with
2440     // MOVK to build up the constant
2441     uint64_t imm_h[4];
2442     int zero_count = 0;
2443     int neg_count = 0;
2444     int i;
2445     for (i = 0; i < 4; i++) {
2446       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2447       if (imm_h[i] == 0) {
2448         zero_count++;
2449       } else if (imm_h[i] == 0xffffL) {
2450         neg_count++;
2451       }
2452     }
2453     if (zero_count == 4) {
2454       // one MOVZ will do
2455       movz(dst, 0);
2456     } else if (neg_count == 4) {
2457       // one MOVN will do
2458       movn(dst, 0);
2459     } else if (zero_count == 3) {
2460       for (i = 0; i < 4; i++) {
2461         if (imm_h[i] != 0L) {
2462           movz(dst, (uint32_t)imm_h[i], (i << 4));
2463           break;
2464         }
2465       }
2466     } else if (neg_count == 3) {
2467       // one MOVN will do
2468       for (int i = 0; i < 4; i++) {
2469         if (imm_h[i] != 0xffffL) {
2470           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2471           break;
2472         }
2473       }
2474     } else if (zero_count == 2) {
2475       // one MOVZ and one MOVK will do
2476       for (i = 0; i < 3; i++) {
2477         if (imm_h[i] != 0L) {
2478           movz(dst, (uint32_t)imm_h[i], (i << 4));
2479           i++;
2480           break;
2481         }
2482       }
2483       for (;i < 4; i++) {
2484         if (imm_h[i] != 0L) {
2485           movk(dst, (uint32_t)imm_h[i], (i << 4));
2486         }
2487       }
2488     } else if (neg_count == 2) {
2489       // one MOVN and one MOVK will do
2490       for (i = 0; i < 4; i++) {
2491         if (imm_h[i] != 0xffffL) {
2492           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2493           i++;
2494           break;
2495         }
2496       }
2497       for (;i < 4; i++) {
2498         if (imm_h[i] != 0xffffL) {
2499           movk(dst, (uint32_t)imm_h[i], (i << 4));
2500         }
2501       }
2502     } else if (zero_count == 1) {
2503       // one MOVZ and two MOVKs will do
2504       for (i = 0; i < 4; i++) {
2505         if (imm_h[i] != 0L) {
2506           movz(dst, (uint32_t)imm_h[i], (i << 4));
2507           i++;
2508           break;
2509         }
2510       }
2511       for (;i < 4; i++) {
2512         if (imm_h[i] != 0x0L) {
2513           movk(dst, (uint32_t)imm_h[i], (i << 4));
2514         }
2515       }
2516     } else if (neg_count == 1) {
2517       // one MOVN and two MOVKs will do
2518       for (i = 0; i < 4; i++) {
2519         if (imm_h[i] != 0xffffL) {
2520           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2521           i++;
2522           break;
2523         }
2524       }
2525       for (;i < 4; i++) {
2526         if (imm_h[i] != 0xffffL) {
2527           movk(dst, (uint32_t)imm_h[i], (i << 4));
2528         }
2529       }
2530     } else {
2531       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2532       movz(dst, (uint32_t)imm_h[0], 0);
2533       for (i = 1; i < 4; i++) {
2534         movk(dst, (uint32_t)imm_h[i], (i << 4));
2535       }
2536     }
2537   }
2538 }
2539 
2540 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2541 {
2542 #ifndef PRODUCT
2543     {
2544       char buffer[64];
2545       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2546       block_comment(buffer);
2547     }
2548 #endif
2549   if (operand_valid_for_logical_immediate(true, imm32)) {
2550     orrw(dst, zr, imm32);
2551   } else {
2552     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2553     // constant
2554     uint32_t imm_h[2];
2555     imm_h[0] = imm32 & 0xffff;
2556     imm_h[1] = ((imm32 >> 16) & 0xffff);
2557     if (imm_h[0] == 0) {
2558       movzw(dst, imm_h[1], 16);
2559     } else if (imm_h[0] == 0xffff) {
2560       movnw(dst, imm_h[1] ^ 0xffff, 16);
2561     } else if (imm_h[1] == 0) {
2562       movzw(dst, imm_h[0], 0);
2563     } else if (imm_h[1] == 0xffff) {
2564       movnw(dst, imm_h[0] ^ 0xffff, 0);
2565     } else {
2566       // use a MOVZ and MOVK (makes it easier to debug)
2567       movzw(dst, imm_h[0], 0);
2568       movkw(dst, imm_h[1], 16);
2569     }
2570   }
2571 }
2572 
2573 // Form an address from base + offset in Rd.  Rd may or may
2574 // not actually be used: you must use the Address that is returned.
2575 // It is up to you to ensure that the shift provided matches the size
2576 // of your data.
2577 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2578   if (Address::offset_ok_for_immed(byte_offset, shift))
2579     // It fits; no need for any heroics
2580     return Address(base, byte_offset);
2581 
2582   // Don't do anything clever with negative or misaligned offsets
2583   unsigned mask = (1 << shift) - 1;
2584   if (byte_offset < 0 || byte_offset & mask) {
2585     mov(Rd, byte_offset);
2586     add(Rd, base, Rd);
2587     return Address(Rd);
2588   }
2589 
2590   // See if we can do this with two 12-bit offsets
2591   {
2592     uint64_t word_offset = byte_offset >> shift;
2593     uint64_t masked_offset = word_offset & 0xfff000;
2594     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2595         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2596       add(Rd, base, masked_offset << shift);
2597       word_offset -= masked_offset;
2598       return Address(Rd, word_offset << shift);
2599     }
2600   }
2601 
2602   // Do it the hard way
2603   mov(Rd, byte_offset);
2604   add(Rd, base, Rd);
2605   return Address(Rd);
2606 }
2607 
2608 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2609                                     bool want_remainder, Register scratch)
2610 {
2611   // Full implementation of Java idiv and irem.  The function
2612   // returns the (pc) offset of the div instruction - may be needed
2613   // for implicit exceptions.
2614   //
2615   // constraint : ra/rb =/= scratch
2616   //         normal case
2617   //
2618   // input : ra: dividend
2619   //         rb: divisor
2620   //
2621   // result: either
2622   //         quotient  (= ra idiv rb)
2623   //         remainder (= ra irem rb)
2624 
2625   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2626 
2627   int idivl_offset = offset();
2628   if (! want_remainder) {
2629     sdivw(result, ra, rb);
2630   } else {
2631     sdivw(scratch, ra, rb);
2632     Assembler::msubw(result, scratch, rb, ra);
2633   }
2634 
2635   return idivl_offset;
2636 }
2637 
2638 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2639                                     bool want_remainder, Register scratch)
2640 {
2641   // Full implementation of Java ldiv and lrem.  The function
2642   // returns the (pc) offset of the div instruction - may be needed
2643   // for implicit exceptions.
2644   //
2645   // constraint : ra/rb =/= scratch
2646   //         normal case
2647   //
2648   // input : ra: dividend
2649   //         rb: divisor
2650   //
2651   // result: either
2652   //         quotient  (= ra idiv rb)
2653   //         remainder (= ra irem rb)
2654 
2655   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2656 
2657   int idivq_offset = offset();
2658   if (! want_remainder) {
2659     sdiv(result, ra, rb);
2660   } else {
2661     sdiv(scratch, ra, rb);
2662     Assembler::msub(result, scratch, rb, ra);
2663   }
2664 
2665   return idivq_offset;
2666 }
2667 
2668 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2669   address prev = pc() - NativeMembar::instruction_size;
2670   address last = code()->last_insn();
2671   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2672     NativeMembar *bar = NativeMembar_at(prev);
2673     if (AlwaysMergeDMB) {
2674       bar->set_kind(bar->get_kind() | order_constraint);
2675       BLOCK_COMMENT("merged membar(always)");
2676       return;
2677     }
2678     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2679     // doing so would introduce a StoreLoad which the caller did not
2680     // intend
2681     if (bar->get_kind() == order_constraint
2682         || bar->get_kind() == AnyAny
2683         || order_constraint == AnyAny) {
2684       // We are merging two memory barrier instructions.  On AArch64 we
2685       // can do this simply by ORing them together.
2686       bar->set_kind(bar->get_kind() | order_constraint);
2687       BLOCK_COMMENT("merged membar");
2688       return;
2689     } else {
2690       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped.
2691       // We need to check the second-to-last instruction, only if it is inside
2692       // the current code section.
2693       address prev2 = prev - NativeMembar::instruction_size;
2694       if (prev2 >= begin() && last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2695         NativeMembar *bar2 = NativeMembar_at(prev2);
2696         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2697         BLOCK_COMMENT("merged membar(elided)");
2698         return;
2699       }
2700     }
2701   }
2702   code()->set_last_insn(pc());
2703   dmb(Assembler::barrier(order_constraint));
2704 }
2705 
2706 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2707   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2708     merge_ldst(rt, adr, size_in_bytes, is_store);
2709     code()->clear_last_insn();
2710     return true;
2711   } else {
2712     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2713     const uint64_t mask = size_in_bytes - 1;
2714     if (adr.getMode() == Address::base_plus_offset &&
2715         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2716       code()->set_last_insn(pc());
2717     }
2718     return false;
2719   }
2720 }
2721 
2722 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2723   // We always try to merge two adjacent loads into one ldp.
2724   if (!try_merge_ldst(Rx, adr, 8, false)) {
2725     Assembler::ldr(Rx, adr);
2726   }
2727 }
2728 
2729 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2730   // We always try to merge two adjacent loads into one ldp.
2731   if (!try_merge_ldst(Rw, adr, 4, false)) {
2732     Assembler::ldrw(Rw, adr);
2733   }
2734 }
2735 
2736 void MacroAssembler::str(Register Rx, const Address &adr) {
2737   // We always try to merge two adjacent stores into one stp.
2738   if (!try_merge_ldst(Rx, adr, 8, true)) {
2739     Assembler::str(Rx, adr);
2740   }
2741 }
2742 
2743 void MacroAssembler::strw(Register Rw, const Address &adr) {
2744   // We always try to merge two adjacent stores into one stp.
2745   if (!try_merge_ldst(Rw, adr, 4, true)) {
2746     Assembler::strw(Rw, adr);
2747   }
2748 }
2749 
2750 // MacroAssembler routines found actually to be needed
2751 
2752 void MacroAssembler::push(Register src)
2753 {
2754   str(src, Address(pre(esp, -1 * wordSize)));
2755 }
2756 
2757 void MacroAssembler::pop(Register dst)
2758 {
2759   ldr(dst, Address(post(esp, 1 * wordSize)));
2760 }
2761 
2762 // Note: load_unsigned_short used to be called load_unsigned_word.
2763 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2764   int off = offset();
2765   ldrh(dst, src);
2766   return off;
2767 }
2768 
2769 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2770   int off = offset();
2771   ldrb(dst, src);
2772   return off;
2773 }
2774 
2775 int MacroAssembler::load_signed_short(Register dst, Address src) {
2776   int off = offset();
2777   ldrsh(dst, src);
2778   return off;
2779 }
2780 
2781 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2782   int off = offset();
2783   ldrsb(dst, src);
2784   return off;
2785 }
2786 
2787 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2788   int off = offset();
2789   ldrshw(dst, src);
2790   return off;
2791 }
2792 
2793 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2794   int off = offset();
2795   ldrsbw(dst, src);
2796   return off;
2797 }
2798 
2799 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2800   switch (size_in_bytes) {
2801   case  8:  ldr(dst, src); break;
2802   case  4:  ldrw(dst, src); break;
2803   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2804   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2805   default:  ShouldNotReachHere();
2806   }
2807 }
2808 
2809 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2810   switch (size_in_bytes) {
2811   case  8:  str(src, dst); break;
2812   case  4:  strw(src, dst); break;
2813   case  2:  strh(src, dst); break;
2814   case  1:  strb(src, dst); break;
2815   default:  ShouldNotReachHere();
2816   }
2817 }
2818 
2819 void MacroAssembler::narrow_subword_type(Register reg, BasicType bt) {
2820   assert(is_subword_type(bt), "required");
2821   switch (bt) {
2822   case T_BOOLEAN: andw(reg, reg, 1); break;
2823   case T_BYTE:    sxtbw(reg, reg); break;
2824   case T_CHAR:    uxthw(reg, reg); break;
2825   case T_SHORT:   sxthw(reg, reg); break;
2826   default:        ShouldNotReachHere();
2827   }
2828 }
2829 
2830 void MacroAssembler::decrementw(Register reg, int value)
2831 {
2832   if (value < 0)  { incrementw(reg, -value);      return; }
2833   if (value == 0) {                               return; }
2834   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2835   /* else */ {
2836     guarantee(reg != rscratch2, "invalid dst for register decrement");
2837     movw(rscratch2, (unsigned)value);
2838     subw(reg, reg, rscratch2);
2839   }
2840 }
2841 
2842 void MacroAssembler::decrement(Register reg, int value)
2843 {
2844   if (value < 0)  { increment(reg, -value);      return; }
2845   if (value == 0) {                              return; }
2846   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2847   /* else */ {
2848     assert(reg != rscratch2, "invalid dst for register decrement");
2849     mov(rscratch2, (uint64_t)value);
2850     sub(reg, reg, rscratch2);
2851   }
2852 }
2853 
2854 void MacroAssembler::decrementw(Address dst, int value)
2855 {
2856   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2857   if (dst.getMode() == Address::literal) {
2858     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2859     lea(rscratch2, dst);
2860     dst = Address(rscratch2);
2861   }
2862   ldrw(rscratch1, dst);
2863   decrementw(rscratch1, value);
2864   strw(rscratch1, dst);
2865 }
2866 
2867 void MacroAssembler::decrement(Address dst, int value)
2868 {
2869   assert(!dst.uses(rscratch1), "invalid address for decrement");
2870   if (dst.getMode() == Address::literal) {
2871     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2872     lea(rscratch2, dst);
2873     dst = Address(rscratch2);
2874   }
2875   ldr(rscratch1, dst);
2876   decrement(rscratch1, value);
2877   str(rscratch1, dst);
2878 }
2879 
2880 void MacroAssembler::incrementw(Register reg, int value)
2881 {
2882   if (value < 0)  { decrementw(reg, -value);      return; }
2883   if (value == 0) {                               return; }
2884   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2885   /* else */ {
2886     assert(reg != rscratch2, "invalid dst for register increment");
2887     movw(rscratch2, (unsigned)value);
2888     addw(reg, reg, rscratch2);
2889   }
2890 }
2891 
2892 void MacroAssembler::increment(Register reg, int value)
2893 {
2894   if (value < 0)  { decrement(reg, -value);      return; }
2895   if (value == 0) {                              return; }
2896   if (value < (1 << 12)) { add(reg, reg, value); return; }
2897   /* else */ {
2898     assert(reg != rscratch2, "invalid dst for register increment");
2899     movw(rscratch2, (unsigned)value);
2900     add(reg, reg, rscratch2);
2901   }
2902 }
2903 
2904 void MacroAssembler::incrementw(Address dst, int value)
2905 {
2906   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2907   if (dst.getMode() == Address::literal) {
2908     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2909     lea(rscratch2, dst);
2910     dst = Address(rscratch2);
2911   }
2912   ldrw(rscratch1, dst);
2913   incrementw(rscratch1, value);
2914   strw(rscratch1, dst);
2915 }
2916 
2917 void MacroAssembler::increment(Address dst, int value)
2918 {
2919   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2920   if (dst.getMode() == Address::literal) {
2921     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2922     lea(rscratch2, dst);
2923     dst = Address(rscratch2);
2924   }
2925   ldr(rscratch1, dst);
2926   increment(rscratch1, value);
2927   str(rscratch1, dst);
2928 }
2929 
2930 // Push lots of registers in the bit set supplied.  Don't push sp.
2931 // Return the number of words pushed
2932 int MacroAssembler::push(RegSet regset, Register stack) {
2933   if (regset.bits() == 0) {
2934     return 0;
2935   }
2936   auto bitset = integer_cast<unsigned int>(regset.bits());
2937   int words_pushed = 0;
2938 
2939   // Scan bitset to accumulate register pairs
2940   unsigned char regs[32];
2941   int count = 0;
2942   for (int reg = 0; reg <= 30; reg++) {
2943     if (1 & bitset)
2944       regs[count++] = reg;
2945     bitset >>= 1;
2946   }
2947   regs[count++] = zr->raw_encoding();
2948   count &= ~1;  // Only push an even number of regs
2949 
2950   if (count) {
2951     stp(as_Register(regs[0]), as_Register(regs[1]),
2952        Address(pre(stack, -count * wordSize)));
2953     words_pushed += 2;
2954   }
2955   for (int i = 2; i < count; i += 2) {
2956     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2957        Address(stack, i * wordSize));
2958     words_pushed += 2;
2959   }
2960 
2961   assert(words_pushed == count, "oops, pushed != count");
2962 
2963   return count;
2964 }
2965 
2966 int MacroAssembler::pop(RegSet regset, Register stack) {
2967   if (regset.bits() == 0) {
2968     return 0;
2969   }
2970   auto bitset = integer_cast<unsigned int>(regset.bits());
2971   int words_pushed = 0;
2972 
2973   // Scan bitset to accumulate register pairs
2974   unsigned char regs[32];
2975   int count = 0;
2976   for (int reg = 0; reg <= 30; reg++) {
2977     if (1 & bitset)
2978       regs[count++] = reg;
2979     bitset >>= 1;
2980   }
2981   regs[count++] = zr->raw_encoding();
2982   count &= ~1;
2983 
2984   for (int i = 2; i < count; i += 2) {
2985     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2986        Address(stack, i * wordSize));
2987     words_pushed += 2;
2988   }
2989   if (count) {
2990     ldp(as_Register(regs[0]), as_Register(regs[1]),
2991        Address(post(stack, count * wordSize)));
2992     words_pushed += 2;
2993   }
2994 
2995   assert(words_pushed == count, "oops, pushed != count");
2996 
2997   return count;
2998 }
2999 
3000 // Push lots of registers in the bit set supplied.  Don't push sp.
3001 // Return the number of dwords pushed
3002 int MacroAssembler::push_fp(FloatRegSet regset, Register stack, FpPushPopMode mode) {
3003   if (regset.bits() == 0) {
3004     return 0;
3005   }
3006   auto bitset = integer_cast<unsigned int>(regset.bits());
3007   int words_pushed = 0;
3008   bool use_sve = false;
3009   int sve_vector_size_in_bytes = 0;
3010 
3011 #ifdef COMPILER2
3012   use_sve = Matcher::supports_scalable_vector();
3013   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3014 #endif
3015 
3016   // Scan bitset to accumulate register pairs
3017   unsigned char regs[32];
3018   int count = 0;
3019   for (int reg = 0; reg <= 31; reg++) {
3020     if (1 & bitset)
3021       regs[count++] = reg;
3022     bitset >>= 1;
3023   }
3024 
3025   if (count == 0) {
3026     return 0;
3027   }
3028 
3029   if (mode == PushPopFull) {
3030     if (use_sve && sve_vector_size_in_bytes > 16) {
3031       mode = PushPopSVE;
3032     } else {
3033       mode = PushPopNeon;
3034     }
3035   }
3036 
3037 #ifndef PRODUCT
3038   {
3039     char buffer[48];
3040     if (mode == PushPopSVE) {
3041       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3042     } else if (mode == PushPopNeon) {
3043       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3044     } else {
3045       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3046     }
3047     block_comment(buffer);
3048   }
3049 #endif
3050 
3051   if (mode == PushPopSVE) {
3052     sub(stack, stack, sve_vector_size_in_bytes * count);
3053     for (int i = 0; i < count; i++) {
3054       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3055     }
3056     return count * sve_vector_size_in_bytes / 8;
3057   }
3058 
3059   if (mode == PushPopNeon) {
3060     if (count == 1) {
3061       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3062       return 2;
3063     }
3064 
3065     bool odd = (count & 1) == 1;
3066     int push_slots = count + (odd ? 1 : 0);
3067 
3068     // Always pushing full 128 bit registers.
3069     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3070     words_pushed += 2;
3071 
3072     for (int i = 2; i + 1 < count; i += 2) {
3073       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3074       words_pushed += 2;
3075     }
3076 
3077     if (odd) {
3078       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3079       words_pushed++;
3080     }
3081 
3082     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3083     return count * 2;
3084   }
3085 
3086   if (mode == PushPopFp) {
3087     bool odd = (count & 1) == 1;
3088     int push_slots = count + (odd ? 1 : 0);
3089 
3090     if (count == 1) {
3091       // Stack pointer must be 16 bytes aligned
3092       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3093       return 1;
3094     }
3095 
3096     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3097     words_pushed += 2;
3098 
3099     for (int i = 2; i + 1 < count; i += 2) {
3100       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3101       words_pushed += 2;
3102     }
3103 
3104     if (odd) {
3105       // Stack pointer must be 16 bytes aligned
3106       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3107       words_pushed++;
3108     }
3109 
3110     assert(words_pushed == count, "oops, pushed != count");
3111 
3112     return count;
3113   }
3114 
3115   return 0;
3116 }
3117 
3118 // Return the number of dwords popped
3119 int MacroAssembler::pop_fp(FloatRegSet regset, Register stack, FpPushPopMode mode) {
3120   if (regset.bits() == 0) {
3121     return 0;
3122   }
3123   auto bitset = integer_cast<unsigned int>(regset.bits());
3124   int words_pushed = 0;
3125   bool use_sve = false;
3126   int sve_vector_size_in_bytes = 0;
3127 
3128 #ifdef COMPILER2
3129   use_sve = Matcher::supports_scalable_vector();
3130   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3131 #endif
3132   // Scan bitset to accumulate register pairs
3133   unsigned char regs[32];
3134   int count = 0;
3135   for (int reg = 0; reg <= 31; reg++) {
3136     if (1 & bitset)
3137       regs[count++] = reg;
3138     bitset >>= 1;
3139   }
3140 
3141   if (count == 0) {
3142     return 0;
3143   }
3144 
3145   if (mode == PushPopFull) {
3146     if (use_sve && sve_vector_size_in_bytes > 16) {
3147       mode = PushPopSVE;
3148     } else {
3149       mode = PushPopNeon;
3150     }
3151   }
3152 
3153 #ifndef PRODUCT
3154   {
3155     char buffer[48];
3156     if (mode == PushPopSVE) {
3157       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3158     } else if (mode == PushPopNeon) {
3159       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3160     } else {
3161       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3162     }
3163     block_comment(buffer);
3164   }
3165 #endif
3166 
3167   if (mode == PushPopSVE) {
3168     for (int i = count - 1; i >= 0; i--) {
3169       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3170     }
3171     add(stack, stack, sve_vector_size_in_bytes * count);
3172     return count * sve_vector_size_in_bytes / 8;
3173   }
3174 
3175   if (mode == PushPopNeon) {
3176     if (count == 1) {
3177       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3178       return 2;
3179     }
3180 
3181     bool odd = (count & 1) == 1;
3182     int push_slots = count + (odd ? 1 : 0);
3183 
3184     if (odd) {
3185       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3186       words_pushed++;
3187     }
3188 
3189     for (int i = 2; i + 1 < count; i += 2) {
3190       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3191       words_pushed += 2;
3192     }
3193 
3194     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3195     words_pushed += 2;
3196 
3197     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3198 
3199     return count * 2;
3200   }
3201 
3202   if (mode == PushPopFp) {
3203     bool odd = (count & 1) == 1;
3204     int push_slots = count + (odd ? 1 : 0);
3205 
3206     if (count == 1) {
3207       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3208       return 1;
3209     }
3210 
3211     if (odd) {
3212       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3213       words_pushed++;
3214     }
3215 
3216     for (int i = 2; i + 1 < count; i += 2) {
3217       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3218       words_pushed += 2;
3219     }
3220 
3221     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3222     words_pushed += 2;
3223 
3224     assert(words_pushed == count, "oops, pushed != count");
3225 
3226     return count;
3227   }
3228 
3229   return 0;
3230 }
3231 
3232 // Return the number of dwords pushed
3233 int MacroAssembler::push_p(PRegSet regset, Register stack) {
3234   if (regset.bits() == 0) {
3235     return 0;
3236   }
3237   auto bitset = integer_cast<unsigned int>(regset.bits());
3238   bool use_sve = false;
3239   int sve_predicate_size_in_slots = 0;
3240 
3241 #ifdef COMPILER2
3242   use_sve = Matcher::supports_scalable_vector();
3243   if (use_sve) {
3244     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3245   }
3246 #endif
3247 
3248   if (!use_sve) {
3249     return 0;
3250   }
3251 
3252   unsigned char regs[PRegister::number_of_registers];
3253   int count = 0;
3254   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3255     if (1 & bitset)
3256       regs[count++] = reg;
3257     bitset >>= 1;
3258   }
3259 
3260   if (count == 0) {
3261     return 0;
3262   }
3263 
3264   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3265                                   VMRegImpl::stack_slot_size * count, 16);
3266   sub(stack, stack, total_push_bytes);
3267   for (int i = 0; i < count; i++) {
3268     sve_str(as_PRegister(regs[i]), Address(stack, i));
3269   }
3270   return total_push_bytes / 8;
3271 }
3272 
3273 // Return the number of dwords popped
3274 int MacroAssembler::pop_p(PRegSet regset, Register stack) {
3275   if (regset.bits() == 0) {
3276     return 0;
3277   }
3278   auto bitset = integer_cast<unsigned int>(regset.bits());
3279   bool use_sve = false;
3280   int sve_predicate_size_in_slots = 0;
3281 
3282 #ifdef COMPILER2
3283   use_sve = Matcher::supports_scalable_vector();
3284   if (use_sve) {
3285     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3286   }
3287 #endif
3288 
3289   if (!use_sve) {
3290     return 0;
3291   }
3292 
3293   unsigned char regs[PRegister::number_of_registers];
3294   int count = 0;
3295   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3296     if (1 & bitset)
3297       regs[count++] = reg;
3298     bitset >>= 1;
3299   }
3300 
3301   if (count == 0) {
3302     return 0;
3303   }
3304 
3305   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3306                                  VMRegImpl::stack_slot_size * count, 16);
3307   for (int i = count - 1; i >= 0; i--) {
3308     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3309   }
3310   add(stack, stack, total_pop_bytes);
3311   return total_pop_bytes / 8;
3312 }
3313 
3314 #ifdef ASSERT
3315 void MacroAssembler::verify_heapbase(const char* msg) {
3316 #if 0
3317   assert (Universe::heap() != nullptr, "java heap should be initialized");
3318   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3319     // rheapbase is allocated as general register
3320     return;
3321   }
3322   if (CheckCompressedOops) {
3323     Label ok;
3324     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3325     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3326     br(Assembler::EQ, ok);
3327     stop(msg);
3328     bind(ok);
3329     pop(1 << rscratch1->encoding(), sp);
3330   }
3331 #endif
3332 }
3333 #endif
3334 
3335 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3336   assert_different_registers(value, tmp1, tmp2);
3337   Label done, tagged, weak_tagged;
3338 
3339   cbz(value, done);           // Use null as-is.
3340   tst(value, JNIHandles::tag_mask); // Test for tag.
3341   br(Assembler::NE, tagged);
3342 
3343   // Resolve local handle
3344   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3345   verify_oop(value);
3346   b(done);
3347 
3348   bind(tagged);
3349   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3350   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3351 
3352   // Resolve global handle
3353   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3354   verify_oop(value);
3355   b(done);
3356 
3357   bind(weak_tagged);
3358   // Resolve jweak.
3359   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3360                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3361   verify_oop(value);
3362 
3363   bind(done);
3364 }
3365 
3366 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3367   assert_different_registers(value, tmp1, tmp2);
3368   Label done;
3369 
3370   cbz(value, done);           // Use null as-is.
3371 
3372 #ifdef ASSERT
3373   {
3374     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3375     Label valid_global_tag;
3376     tbnz(value, 1, valid_global_tag); // Test for global tag
3377     stop("non global jobject using resolve_global_jobject");
3378     bind(valid_global_tag);
3379   }
3380 #endif
3381 
3382   // Resolve global handle
3383   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3384   verify_oop(value);
3385 
3386   bind(done);
3387 }
3388 
3389 void MacroAssembler::stop(const char* msg) {
3390   // Skip AOT caching C strings in scratch buffer.
3391   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3392   BLOCK_COMMENT(str);
3393   // load msg into r0 so we can access it from the signal handler
3394   // ExternalAddress enables saving and restoring via the code cache
3395   lea(c_rarg0, ExternalAddress((address) str));
3396   dcps1(0xdeae);
3397 }
3398 
3399 void MacroAssembler::unimplemented(const char* what) {
3400   const char* buf = nullptr;
3401   {
3402     ResourceMark rm;
3403     stringStream ss;
3404     ss.print("unimplemented: %s", what);
3405     buf = code_string(ss.as_string());
3406   }
3407   stop(buf);
3408 }
3409 
3410 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3411 #ifdef ASSERT
3412   Label OK;
3413   br(cc, OK);
3414   stop(msg);
3415   bind(OK);
3416 #endif
3417 }
3418 
3419 // If a constant does not fit in an immediate field, generate some
3420 // number of MOV instructions and then perform the operation.
3421 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3422                                            add_sub_imm_insn insn1,
3423                                            add_sub_reg_insn insn2,
3424                                            bool is32) {
3425   assert(Rd != zr, "Rd = zr and not setting flags?");
3426   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3427   if (fits) {
3428     (this->*insn1)(Rd, Rn, imm);
3429   } else {
3430     if (g_uabs(imm) < (1 << 24)) {
3431        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3432        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3433     } else {
3434        assert_different_registers(Rd, Rn);
3435        mov(Rd, imm);
3436        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3437     }
3438   }
3439 }
3440 
3441 // Separate vsn which sets the flags. Optimisations are more restricted
3442 // because we must set the flags correctly.
3443 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3444                                              add_sub_imm_insn insn1,
3445                                              add_sub_reg_insn insn2,
3446                                              bool is32) {
3447   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3448   if (fits) {
3449     (this->*insn1)(Rd, Rn, imm);
3450   } else {
3451     assert_different_registers(Rd, Rn);
3452     assert(Rd != zr, "overflow in immediate operand");
3453     mov(Rd, imm);
3454     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3455   }
3456 }
3457 
3458 
3459 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3460   if (increment.is_register()) {
3461     add(Rd, Rn, increment.as_register());
3462   } else {
3463     add(Rd, Rn, increment.as_constant());
3464   }
3465 }
3466 
3467 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3468   if (increment.is_register()) {
3469     addw(Rd, Rn, increment.as_register());
3470   } else {
3471     addw(Rd, Rn, increment.as_constant());
3472   }
3473 }
3474 
3475 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3476   if (decrement.is_register()) {
3477     sub(Rd, Rn, decrement.as_register());
3478   } else {
3479     sub(Rd, Rn, decrement.as_constant());
3480   }
3481 }
3482 
3483 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3484   if (decrement.is_register()) {
3485     subw(Rd, Rn, decrement.as_register());
3486   } else {
3487     subw(Rd, Rn, decrement.as_constant());
3488   }
3489 }
3490 
3491 void MacroAssembler::reinit_heapbase()
3492 {
3493   if (UseCompressedOops) {
3494     if (Universe::is_fully_initialized() && !AOTCodeCache::is_on_for_dump()) {
3495       mov(rheapbase, CompressedOops::base());
3496     } else {
3497       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3498       ldr(rheapbase, Address(rheapbase));
3499     }
3500   }
3501 }
3502 
3503 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3504 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3505 // Pass a register for the result, otherwise pass noreg.
3506 
3507 // Clobbers rscratch1
3508 void MacroAssembler::cmpxchg(Register addr, Register expected,
3509                              Register new_val,
3510                              enum operand_size size,
3511                              bool acquire, bool release,
3512                              bool weak,
3513                              Register result) {
3514   if (result == noreg)  result = rscratch1;
3515   BLOCK_COMMENT("cmpxchg {");
3516   if (UseLSE) {
3517     mov(result, expected);
3518     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3519     compare_eq(result, expected, size);
3520 #ifdef ASSERT
3521     // Poison rscratch1 which is written on !UseLSE branch
3522     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3523 #endif
3524   } else {
3525     Label retry_load, done;
3526     prfm(Address(addr), PSTL1STRM);
3527     bind(retry_load);
3528     load_exclusive(result, addr, size, acquire);
3529     compare_eq(result, expected, size);
3530     br(Assembler::NE, done);
3531     store_exclusive(rscratch1, new_val, addr, size, release);
3532     if (weak) {
3533       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3534     } else {
3535       cbnzw(rscratch1, retry_load);
3536     }
3537     bind(done);
3538   }
3539   BLOCK_COMMENT("} cmpxchg");
3540 }
3541 
3542 // A generic comparison. Only compares for equality, clobbers rscratch1.
3543 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3544   if (size == xword) {
3545     cmp(rm, rn);
3546   } else if (size == word) {
3547     cmpw(rm, rn);
3548   } else if (size == halfword) {
3549     eorw(rscratch1, rm, rn);
3550     ands(zr, rscratch1, 0xffff);
3551   } else if (size == byte) {
3552     eorw(rscratch1, rm, rn);
3553     ands(zr, rscratch1, 0xff);
3554   } else {
3555     ShouldNotReachHere();
3556   }
3557 }
3558 
3559 
3560 static bool different(Register a, RegisterOrConstant b, Register c) {
3561   if (b.is_constant())
3562     return a != c;
3563   else
3564     return a != b.as_register() && a != c && b.as_register() != c;
3565 }
3566 
3567 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3568 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3569   if (UseLSE) {                                                         \
3570     prev = prev->is_valid() ? prev : zr;                                \
3571     if (incr.is_register()) {                                           \
3572       AOP(sz, incr.as_register(), prev, addr);                          \
3573     } else {                                                            \
3574       mov(rscratch2, incr.as_constant());                               \
3575       AOP(sz, rscratch2, prev, addr);                                   \
3576     }                                                                   \
3577     return;                                                             \
3578   }                                                                     \
3579   Register result = rscratch2;                                          \
3580   if (prev->is_valid())                                                 \
3581     result = different(prev, incr, addr) ? prev : rscratch2;            \
3582                                                                         \
3583   Label retry_load;                                                     \
3584   prfm(Address(addr), PSTL1STRM);                                       \
3585   bind(retry_load);                                                     \
3586   LDXR(result, addr);                                                   \
3587   OP(rscratch1, result, incr);                                          \
3588   STXR(rscratch2, rscratch1, addr);                                     \
3589   cbnzw(rscratch2, retry_load);                                         \
3590   if (prev->is_valid() && prev != result) {                             \
3591     IOP(prev, rscratch1, incr);                                         \
3592   }                                                                     \
3593 }
3594 
3595 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3596 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3597 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3598 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3599 
3600 #undef ATOMIC_OP
3601 
3602 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3603 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3604   if (UseLSE) {                                                         \
3605     prev = prev->is_valid() ? prev : zr;                                \
3606     AOP(sz, newv, prev, addr);                                          \
3607     return;                                                             \
3608   }                                                                     \
3609   Register result = rscratch2;                                          \
3610   if (prev->is_valid())                                                 \
3611     result = different(prev, newv, addr) ? prev : rscratch2;            \
3612                                                                         \
3613   Label retry_load;                                                     \
3614   prfm(Address(addr), PSTL1STRM);                                       \
3615   bind(retry_load);                                                     \
3616   LDXR(result, addr);                                                   \
3617   STXR(rscratch1, newv, addr);                                          \
3618   cbnzw(rscratch1, retry_load);                                         \
3619   if (prev->is_valid() && prev != result)                               \
3620     mov(prev, result);                                                  \
3621 }
3622 
3623 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3624 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3625 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3626 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3627 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3628 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3629 
3630 #undef ATOMIC_XCHG
3631 
3632 #ifndef PRODUCT
3633 extern "C" void findpc(intptr_t x);
3634 #endif
3635 
3636 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3637 {
3638   // In order to get locks to work, we need to fake a in_VM state
3639   if (ShowMessageBoxOnError) {
3640     JavaThread* thread = JavaThread::current();
3641     thread->set_thread_state(_thread_in_vm);
3642 #ifndef PRODUCT
3643     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3644       ttyLocker ttyl;
3645       BytecodeCounter::print();
3646     }
3647 #endif
3648     if (os::message_box(msg, "Execution stopped, print registers?")) {
3649       ttyLocker ttyl;
3650       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3651 #ifndef PRODUCT
3652       tty->cr();
3653       findpc(pc);
3654       tty->cr();
3655 #endif
3656       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3657       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3658       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3659       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3660       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3661       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3662       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3663       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3664       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3665       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3666       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3667       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3668       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3669       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3670       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3671       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3672       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3673       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3674       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3675       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3676       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3677       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3678       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3679       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3680       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3681       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3682       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3683       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3684       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3685       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3686       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3687       BREAKPOINT;
3688     }
3689   }
3690   fatal("DEBUG MESSAGE: %s", msg);
3691 }
3692 
3693 RegSet MacroAssembler::call_clobbered_gp_registers() {
3694   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3695 #ifndef R18_RESERVED
3696   regs += r18_tls;
3697 #endif
3698   return regs;
3699 }
3700 
3701 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3702   int step = 4 * wordSize;
3703   push(call_clobbered_gp_registers() - exclude, sp);
3704   sub(sp, sp, step);
3705   mov(rscratch1, -step);
3706   // Push v0-v7, v16-v31.
3707   for (int i = 31; i>= 4; i -= 4) {
3708     if (i <= v7->encoding() || i >= v16->encoding())
3709       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3710           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3711   }
3712   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3713       as_FloatRegister(3), T1D, Address(sp));
3714 }
3715 
3716 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3717   for (int i = 0; i < 32; i += 4) {
3718     if (i <= v7->encoding() || i >= v16->encoding())
3719       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3720           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3721   }
3722 
3723   reinitialize_ptrue();
3724 
3725   pop(call_clobbered_gp_registers() - exclude, sp);
3726 }
3727 
3728 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3729                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3730   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3731   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3732     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3733     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3734       sve_str(as_FloatRegister(i), Address(sp, i));
3735     }
3736   } else {
3737     int step = (save_vectors ? 8 : 4) * wordSize;
3738     mov(rscratch1, -step);
3739     sub(sp, sp, step);
3740     for (int i = 28; i >= 4; i -= 4) {
3741       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3742           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3743     }
3744     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3745   }
3746   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3747     sub(sp, sp, total_predicate_in_bytes);
3748     for (int i = 0; i < PRegister::number_of_registers; i++) {
3749       sve_str(as_PRegister(i), Address(sp, i));
3750     }
3751   }
3752 }
3753 
3754 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3755                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3756   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3757     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3758       sve_ldr(as_PRegister(i), Address(sp, i));
3759     }
3760     add(sp, sp, total_predicate_in_bytes);
3761   }
3762   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3763     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3764       sve_ldr(as_FloatRegister(i), Address(sp, i));
3765     }
3766     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3767   } else {
3768     int step = (restore_vectors ? 8 : 4) * wordSize;
3769     for (int i = 0; i <= 28; i += 4)
3770       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3771           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3772   }
3773 
3774   // We may use predicate registers and rely on ptrue with SVE,
3775   // regardless of wide vector (> 8 bytes) used or not.
3776   if (use_sve) {
3777     reinitialize_ptrue();
3778   }
3779 
3780   // integer registers except lr & sp
3781   pop(RegSet::range(r0, r17), sp);
3782 #ifdef R18_RESERVED
3783   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3784   pop(RegSet::range(r20, r29), sp);
3785 #else
3786   pop(RegSet::range(r18_tls, r29), sp);
3787 #endif
3788 }
3789 
3790 /**
3791  * Helpers for multiply_to_len().
3792  */
3793 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3794                                      Register src1, Register src2) {
3795   adds(dest_lo, dest_lo, src1);
3796   adc(dest_hi, dest_hi, zr);
3797   adds(dest_lo, dest_lo, src2);
3798   adc(final_dest_hi, dest_hi, zr);
3799 }
3800 
3801 // Generate an address from (r + r1 extend offset).  "size" is the
3802 // size of the operand.  The result may be in rscratch2.
3803 Address MacroAssembler::offsetted_address(Register r, Register r1,
3804                                           Address::extend ext, int offset, int size) {
3805   if (offset || (ext.shift() % size != 0)) {
3806     lea(rscratch2, Address(r, r1, ext));
3807     return Address(rscratch2, offset);
3808   } else {
3809     return Address(r, r1, ext);
3810   }
3811 }
3812 
3813 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3814 {
3815   assert(offset >= 0, "spill to negative address?");
3816   // Offset reachable ?
3817   //   Not aligned - 9 bits signed offset
3818   //   Aligned - 12 bits unsigned offset shifted
3819   Register base = sp;
3820   if ((offset & (size-1)) && offset >= (1<<8)) {
3821     add(tmp, base, offset & ((1<<12)-1));
3822     base = tmp;
3823     offset &= -1u<<12;
3824   }
3825 
3826   if (offset >= (1<<12) * size) {
3827     add(tmp, base, offset & (((1<<12)-1)<<12));
3828     base = tmp;
3829     offset &= ~(((1<<12)-1)<<12);
3830   }
3831 
3832   return Address(base, offset);
3833 }
3834 
3835 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3836   assert(offset >= 0, "spill to negative address?");
3837 
3838   Register base = sp;
3839 
3840   // An immediate offset in the range 0 to 255 which is multiplied
3841   // by the current vector or predicate register size in bytes.
3842   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3843     return Address(base, offset / sve_reg_size_in_bytes);
3844   }
3845 
3846   add(tmp, base, offset);
3847   return Address(tmp);
3848 }
3849 
3850 // Checks whether offset is aligned.
3851 // Returns true if it is, else false.
3852 bool MacroAssembler::merge_alignment_check(Register base,
3853                                            size_t size,
3854                                            int64_t cur_offset,
3855                                            int64_t prev_offset) const {
3856   if (AvoidUnalignedAccesses) {
3857     if (base == sp) {
3858       // Checks whether low offset if aligned to pair of registers.
3859       int64_t pair_mask = size * 2 - 1;
3860       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3861       return (offset & pair_mask) == 0;
3862     } else { // If base is not sp, we can't guarantee the access is aligned.
3863       return false;
3864     }
3865   } else {
3866     int64_t mask = size - 1;
3867     // Load/store pair instruction only supports element size aligned offset.
3868     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3869   }
3870 }
3871 
3872 // Checks whether current and previous loads/stores can be merged.
3873 // Returns true if it can be merged, else false.
3874 bool MacroAssembler::ldst_can_merge(Register rt,
3875                                     const Address &adr,
3876                                     size_t cur_size_in_bytes,
3877                                     bool is_store) const {
3878   address prev = pc() - NativeInstruction::instruction_size;
3879   address last = code()->last_insn();
3880 
3881   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3882     return false;
3883   }
3884 
3885   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3886     return false;
3887   }
3888 
3889   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3890   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3891 
3892   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3893   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3894 
3895   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3896     return false;
3897   }
3898 
3899   int64_t max_offset = 63 * prev_size_in_bytes;
3900   int64_t min_offset = -64 * prev_size_in_bytes;
3901 
3902   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3903 
3904   // Only same base can be merged.
3905   if (adr.base() != prev_ldst->base()) {
3906     return false;
3907   }
3908 
3909   int64_t cur_offset = adr.offset();
3910   int64_t prev_offset = prev_ldst->offset();
3911   size_t diff = abs(cur_offset - prev_offset);
3912   if (diff != prev_size_in_bytes) {
3913     return false;
3914   }
3915 
3916   // Following cases can not be merged:
3917   // ldr x2, [x2, #8]
3918   // ldr x3, [x2, #16]
3919   // or:
3920   // ldr x2, [x3, #8]
3921   // ldr x2, [x3, #16]
3922   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3923   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3924     return false;
3925   }
3926 
3927   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3928   // Offset range must be in ldp/stp instruction's range.
3929   if (low_offset > max_offset || low_offset < min_offset) {
3930     return false;
3931   }
3932 
3933   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3934     return true;
3935   }
3936 
3937   return false;
3938 }
3939 
3940 // Merge current load/store with previous load/store into ldp/stp.
3941 void MacroAssembler::merge_ldst(Register rt,
3942                                 const Address &adr,
3943                                 size_t cur_size_in_bytes,
3944                                 bool is_store) {
3945 
3946   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3947 
3948   Register rt_low, rt_high;
3949   address prev = pc() - NativeInstruction::instruction_size;
3950   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3951 
3952   int64_t offset;
3953 
3954   if (adr.offset() < prev_ldst->offset()) {
3955     offset = adr.offset();
3956     rt_low = rt;
3957     rt_high = prev_ldst->target();
3958   } else {
3959     offset = prev_ldst->offset();
3960     rt_low = prev_ldst->target();
3961     rt_high = rt;
3962   }
3963 
3964   Address adr_p = Address(prev_ldst->base(), offset);
3965   // Overwrite previous generated binary.
3966   code_section()->set_end(prev);
3967 
3968   const size_t sz = prev_ldst->size_in_bytes();
3969   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3970   if (!is_store) {
3971     BLOCK_COMMENT("merged ldr pair");
3972     if (sz == 8) {
3973       ldp(rt_low, rt_high, adr_p);
3974     } else {
3975       ldpw(rt_low, rt_high, adr_p);
3976     }
3977   } else {
3978     BLOCK_COMMENT("merged str pair");
3979     if (sz == 8) {
3980       stp(rt_low, rt_high, adr_p);
3981     } else {
3982       stpw(rt_low, rt_high, adr_p);
3983     }
3984   }
3985 }
3986 
3987 /**
3988  * Multiply 64 bit by 64 bit first loop.
3989  */
3990 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3991                                            Register y, Register y_idx, Register z,
3992                                            Register carry, Register product,
3993                                            Register idx, Register kdx) {
3994   //
3995   //  jlong carry, x[], y[], z[];
3996   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3997   //    huge_128 product = y[idx] * x[xstart] + carry;
3998   //    z[kdx] = (jlong)product;
3999   //    carry  = (jlong)(product >>> 64);
4000   //  }
4001   //  z[xstart] = carry;
4002   //
4003 
4004   Label L_first_loop, L_first_loop_exit;
4005   Label L_one_x, L_one_y, L_multiply;
4006 
4007   subsw(xstart, xstart, 1);
4008   br(Assembler::MI, L_one_x);
4009 
4010   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
4011   ldr(x_xstart, Address(rscratch1));
4012   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
4013 
4014   bind(L_first_loop);
4015   subsw(idx, idx, 1);
4016   br(Assembler::MI, L_first_loop_exit);
4017   subsw(idx, idx, 1);
4018   br(Assembler::MI, L_one_y);
4019   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4020   ldr(y_idx, Address(rscratch1));
4021   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
4022   bind(L_multiply);
4023 
4024   // AArch64 has a multiply-accumulate instruction that we can't use
4025   // here because it has no way to process carries, so we have to use
4026   // separate add and adc instructions.  Bah.
4027   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
4028   mul(product, x_xstart, y_idx);
4029   adds(product, product, carry);
4030   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
4031 
4032   subw(kdx, kdx, 2);
4033   ror(product, product, 32); // back to big-endian
4034   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4035 
4036   b(L_first_loop);
4037 
4038   bind(L_one_y);
4039   ldrw(y_idx, Address(y,  0));
4040   b(L_multiply);
4041 
4042   bind(L_one_x);
4043   ldrw(x_xstart, Address(x,  0));
4044   b(L_first_loop);
4045 
4046   bind(L_first_loop_exit);
4047 }
4048 
4049 /**
4050  * Multiply 128 bit by 128. Unrolled inner loop.
4051  *
4052  */
4053 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4054                                              Register carry, Register carry2,
4055                                              Register idx, Register jdx,
4056                                              Register yz_idx1, Register yz_idx2,
4057                                              Register tmp, Register tmp3, Register tmp4,
4058                                              Register tmp6, Register product_hi) {
4059 
4060   //   jlong carry, x[], y[], z[];
4061   //   int kdx = ystart+1;
4062   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4063   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4064   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4065   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4066   //     carry  = (jlong)(tmp4 >>> 64);
4067   //     z[kdx+idx+1] = (jlong)tmp3;
4068   //     z[kdx+idx] = (jlong)tmp4;
4069   //   }
4070   //   idx += 2;
4071   //   if (idx > 0) {
4072   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4073   //     z[kdx+idx] = (jlong)yz_idx1;
4074   //     carry  = (jlong)(yz_idx1 >>> 64);
4075   //   }
4076   //
4077 
4078   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4079 
4080   lsrw(jdx, idx, 2);
4081 
4082   bind(L_third_loop);
4083 
4084   subsw(jdx, jdx, 1);
4085   br(Assembler::MI, L_third_loop_exit);
4086   subw(idx, idx, 4);
4087 
4088   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4089 
4090   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4091 
4092   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4093 
4094   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4095   ror(yz_idx2, yz_idx2, 32);
4096 
4097   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4098 
4099   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4100   umulh(tmp4, product_hi, yz_idx1);
4101 
4102   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4103   ror(rscratch2, rscratch2, 32);
4104 
4105   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4106   umulh(carry2, product_hi, yz_idx2);
4107 
4108   // propagate sum of both multiplications into carry:tmp4:tmp3
4109   adds(tmp3, tmp3, carry);
4110   adc(tmp4, tmp4, zr);
4111   adds(tmp3, tmp3, rscratch1);
4112   adcs(tmp4, tmp4, tmp);
4113   adc(carry, carry2, zr);
4114   adds(tmp4, tmp4, rscratch2);
4115   adc(carry, carry, zr);
4116 
4117   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4118   ror(tmp4, tmp4, 32);
4119   stp(tmp4, tmp3, Address(tmp6, 0));
4120 
4121   b(L_third_loop);
4122   bind (L_third_loop_exit);
4123 
4124   andw (idx, idx, 0x3);
4125   cbz(idx, L_post_third_loop_done);
4126 
4127   Label L_check_1;
4128   subsw(idx, idx, 2);
4129   br(Assembler::MI, L_check_1);
4130 
4131   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4132   ldr(yz_idx1, Address(rscratch1, 0));
4133   ror(yz_idx1, yz_idx1, 32);
4134   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4135   umulh(tmp4, product_hi, yz_idx1);
4136   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4137   ldr(yz_idx2, Address(rscratch1, 0));
4138   ror(yz_idx2, yz_idx2, 32);
4139 
4140   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4141 
4142   ror(tmp3, tmp3, 32);
4143   str(tmp3, Address(rscratch1, 0));
4144 
4145   bind (L_check_1);
4146 
4147   andw (idx, idx, 0x1);
4148   subsw(idx, idx, 1);
4149   br(Assembler::MI, L_post_third_loop_done);
4150   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4151   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4152   umulh(carry2, tmp4, product_hi);
4153   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4154 
4155   add2_with_carry(carry2, tmp3, tmp4, carry);
4156 
4157   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4158   extr(carry, carry2, tmp3, 32);
4159 
4160   bind(L_post_third_loop_done);
4161 }
4162 
4163 /**
4164  * Code for BigInteger::multiplyToLen() intrinsic.
4165  *
4166  * r0: x
4167  * r1: xlen
4168  * r2: y
4169  * r3: ylen
4170  * r4:  z
4171  * r5: tmp0
4172  * r10: tmp1
4173  * r11: tmp2
4174  * r12: tmp3
4175  * r13: tmp4
4176  * r14: tmp5
4177  * r15: tmp6
4178  * r16: tmp7
4179  *
4180  */
4181 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4182                                      Register z, Register tmp0,
4183                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4184                                      Register tmp5, Register tmp6, Register product_hi) {
4185 
4186   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4187 
4188   const Register idx = tmp1;
4189   const Register kdx = tmp2;
4190   const Register xstart = tmp3;
4191 
4192   const Register y_idx = tmp4;
4193   const Register carry = tmp5;
4194   const Register product  = xlen;
4195   const Register x_xstart = tmp0;
4196 
4197   // First Loop.
4198   //
4199   //  final static long LONG_MASK = 0xffffffffL;
4200   //  int xstart = xlen - 1;
4201   //  int ystart = ylen - 1;
4202   //  long carry = 0;
4203   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4204   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4205   //    z[kdx] = (int)product;
4206   //    carry = product >>> 32;
4207   //  }
4208   //  z[xstart] = (int)carry;
4209   //
4210 
4211   movw(idx, ylen);       // idx = ylen;
4212   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4213   mov(carry, zr);        // carry = 0;
4214 
4215   Label L_done;
4216 
4217   movw(xstart, xlen);
4218   subsw(xstart, xstart, 1);
4219   br(Assembler::MI, L_done);
4220 
4221   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4222 
4223   Label L_second_loop;
4224   cbzw(kdx, L_second_loop);
4225 
4226   Label L_carry;
4227   subw(kdx, kdx, 1);
4228   cbzw(kdx, L_carry);
4229 
4230   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4231   lsr(carry, carry, 32);
4232   subw(kdx, kdx, 1);
4233 
4234   bind(L_carry);
4235   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4236 
4237   // Second and third (nested) loops.
4238   //
4239   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4240   //   carry = 0;
4241   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4242   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4243   //                    (z[k] & LONG_MASK) + carry;
4244   //     z[k] = (int)product;
4245   //     carry = product >>> 32;
4246   //   }
4247   //   z[i] = (int)carry;
4248   // }
4249   //
4250   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4251 
4252   const Register jdx = tmp1;
4253 
4254   bind(L_second_loop);
4255   mov(carry, zr);                // carry = 0;
4256   movw(jdx, ylen);               // j = ystart+1
4257 
4258   subsw(xstart, xstart, 1);      // i = xstart-1;
4259   br(Assembler::MI, L_done);
4260 
4261   str(z, Address(pre(sp, -4 * wordSize)));
4262 
4263   Label L_last_x;
4264   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4265   subsw(xstart, xstart, 1);       // i = xstart-1;
4266   br(Assembler::MI, L_last_x);
4267 
4268   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4269   ldr(product_hi, Address(rscratch1));
4270   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4271 
4272   Label L_third_loop_prologue;
4273   bind(L_third_loop_prologue);
4274 
4275   str(ylen, Address(sp, wordSize));
4276   stp(x, xstart, Address(sp, 2 * wordSize));
4277   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4278                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4279   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4280   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4281 
4282   addw(tmp3, xlen, 1);
4283   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4284   subsw(tmp3, tmp3, 1);
4285   br(Assembler::MI, L_done);
4286 
4287   lsr(carry, carry, 32);
4288   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4289   b(L_second_loop);
4290 
4291   // Next infrequent code is moved outside loops.
4292   bind(L_last_x);
4293   ldrw(product_hi, Address(x,  0));
4294   b(L_third_loop_prologue);
4295 
4296   bind(L_done);
4297 }
4298 
4299 // Code for BigInteger::mulAdd intrinsic
4300 // out     = r0
4301 // in      = r1
4302 // offset  = r2  (already out.length-offset)
4303 // len     = r3
4304 // k       = r4
4305 //
4306 // pseudo code from java implementation:
4307 // carry = 0;
4308 // offset = out.length-offset - 1;
4309 // for (int j=len-1; j >= 0; j--) {
4310 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4311 //     out[offset--] = (int)product;
4312 //     carry = product >>> 32;
4313 // }
4314 // return (int)carry;
4315 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4316       Register len, Register k) {
4317     Label LOOP, END;
4318     // pre-loop
4319     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4320     csel(out, zr, out, Assembler::EQ);
4321     br(Assembler::EQ, END);
4322     add(in, in, len, LSL, 2); // in[j+1] address
4323     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4324     mov(out, zr); // used to keep carry now
4325     BIND(LOOP);
4326     ldrw(rscratch1, Address(pre(in, -4)));
4327     madd(rscratch1, rscratch1, k, out);
4328     ldrw(rscratch2, Address(pre(offset, -4)));
4329     add(rscratch1, rscratch1, rscratch2);
4330     strw(rscratch1, Address(offset));
4331     lsr(out, rscratch1, 32);
4332     subs(len, len, 1);
4333     br(Assembler::NE, LOOP);
4334     BIND(END);
4335 }
4336 
4337 /**
4338  * Emits code to update CRC-32 with a byte value according to constants in table
4339  *
4340  * @param [in,out]crc   Register containing the crc.
4341  * @param [in]val       Register containing the byte to fold into the CRC.
4342  * @param [in]table     Register containing the table of crc constants.
4343  *
4344  * uint32_t crc;
4345  * val = crc_table[(val ^ crc) & 0xFF];
4346  * crc = val ^ (crc >> 8);
4347  *
4348  */
4349 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4350   eor(val, val, crc);
4351   andr(val, val, 0xff);
4352   ldrw(val, Address(table, val, Address::lsl(2)));
4353   eor(crc, val, crc, Assembler::LSR, 8);
4354 }
4355 
4356 /**
4357  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4358  *
4359  * @param [in,out]crc   Register containing the crc.
4360  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4361  * @param [in]table0    Register containing table 0 of crc constants.
4362  * @param [in]table1    Register containing table 1 of crc constants.
4363  * @param [in]table2    Register containing table 2 of crc constants.
4364  * @param [in]table3    Register containing table 3 of crc constants.
4365  *
4366  * uint32_t crc;
4367  *   v = crc ^ v
4368  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4369  *
4370  */
4371 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4372         Register table0, Register table1, Register table2, Register table3,
4373         bool upper) {
4374   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4375   uxtb(tmp, v);
4376   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4377   ubfx(tmp, v, 8, 8);
4378   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4379   eor(crc, crc, tmp);
4380   ubfx(tmp, v, 16, 8);
4381   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4382   eor(crc, crc, tmp);
4383   ubfx(tmp, v, 24, 8);
4384   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4385   eor(crc, crc, tmp);
4386 }
4387 
4388 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4389         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4390     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4391     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4392 
4393     subs(tmp0, len, 384);
4394     mvnw(crc, crc);
4395     br(Assembler::GE, CRC_by128_pre);
4396   BIND(CRC_less128);
4397     subs(len, len, 32);
4398     br(Assembler::GE, CRC_by32_loop);
4399   BIND(CRC_less32);
4400     adds(len, len, 32 - 4);
4401     br(Assembler::GE, CRC_by4_loop);
4402     adds(len, len, 4);
4403     br(Assembler::GT, CRC_by1_loop);
4404     b(L_exit);
4405 
4406   BIND(CRC_by32_loop);
4407     ldp(tmp0, tmp1, Address(buf));
4408     crc32x(crc, crc, tmp0);
4409     ldp(tmp2, tmp3, Address(buf, 16));
4410     crc32x(crc, crc, tmp1);
4411     add(buf, buf, 32);
4412     crc32x(crc, crc, tmp2);
4413     subs(len, len, 32);
4414     crc32x(crc, crc, tmp3);
4415     br(Assembler::GE, CRC_by32_loop);
4416     cmn(len, (u1)32);
4417     br(Assembler::NE, CRC_less32);
4418     b(L_exit);
4419 
4420   BIND(CRC_by4_loop);
4421     ldrw(tmp0, Address(post(buf, 4)));
4422     subs(len, len, 4);
4423     crc32w(crc, crc, tmp0);
4424     br(Assembler::GE, CRC_by4_loop);
4425     adds(len, len, 4);
4426     br(Assembler::LE, L_exit);
4427   BIND(CRC_by1_loop);
4428     ldrb(tmp0, Address(post(buf, 1)));
4429     subs(len, len, 1);
4430     crc32b(crc, crc, tmp0);
4431     br(Assembler::GT, CRC_by1_loop);
4432     b(L_exit);
4433 
4434   BIND(CRC_by128_pre);
4435     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4436       4*256*sizeof(juint) + 8*sizeof(juint));
4437     mov(crc, 0);
4438     crc32x(crc, crc, tmp0);
4439     crc32x(crc, crc, tmp1);
4440 
4441     cbnz(len, CRC_less128);
4442 
4443   BIND(L_exit);
4444     mvnw(crc, crc);
4445 }
4446 
4447 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4448         Register len, Register tmp0, Register tmp1, Register tmp2,
4449         Register tmp3) {
4450     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4451     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4452 
4453     mvnw(crc, crc);
4454 
4455     subs(len, len, 128);
4456     br(Assembler::GE, CRC_by64_pre);
4457   BIND(CRC_less64);
4458     adds(len, len, 128-32);
4459     br(Assembler::GE, CRC_by32_loop);
4460   BIND(CRC_less32);
4461     adds(len, len, 32-4);
4462     br(Assembler::GE, CRC_by4_loop);
4463     adds(len, len, 4);
4464     br(Assembler::GT, CRC_by1_loop);
4465     b(L_exit);
4466 
4467   BIND(CRC_by32_loop);
4468     ldp(tmp0, tmp1, Address(post(buf, 16)));
4469     subs(len, len, 32);
4470     crc32x(crc, crc, tmp0);
4471     ldr(tmp2, Address(post(buf, 8)));
4472     crc32x(crc, crc, tmp1);
4473     ldr(tmp3, Address(post(buf, 8)));
4474     crc32x(crc, crc, tmp2);
4475     crc32x(crc, crc, tmp3);
4476     br(Assembler::GE, CRC_by32_loop);
4477     cmn(len, (u1)32);
4478     br(Assembler::NE, CRC_less32);
4479     b(L_exit);
4480 
4481   BIND(CRC_by4_loop);
4482     ldrw(tmp0, Address(post(buf, 4)));
4483     subs(len, len, 4);
4484     crc32w(crc, crc, tmp0);
4485     br(Assembler::GE, CRC_by4_loop);
4486     adds(len, len, 4);
4487     br(Assembler::LE, L_exit);
4488   BIND(CRC_by1_loop);
4489     ldrb(tmp0, Address(post(buf, 1)));
4490     subs(len, len, 1);
4491     crc32b(crc, crc, tmp0);
4492     br(Assembler::GT, CRC_by1_loop);
4493     b(L_exit);
4494 
4495   BIND(CRC_by64_pre);
4496     sub(buf, buf, 8);
4497     ldp(tmp0, tmp1, Address(buf, 8));
4498     crc32x(crc, crc, tmp0);
4499     ldr(tmp2, Address(buf, 24));
4500     crc32x(crc, crc, tmp1);
4501     ldr(tmp3, Address(buf, 32));
4502     crc32x(crc, crc, tmp2);
4503     ldr(tmp0, Address(buf, 40));
4504     crc32x(crc, crc, tmp3);
4505     ldr(tmp1, Address(buf, 48));
4506     crc32x(crc, crc, tmp0);
4507     ldr(tmp2, Address(buf, 56));
4508     crc32x(crc, crc, tmp1);
4509     ldr(tmp3, Address(pre(buf, 64)));
4510 
4511     b(CRC_by64_loop);
4512 
4513     align(CodeEntryAlignment);
4514   BIND(CRC_by64_loop);
4515     subs(len, len, 64);
4516     crc32x(crc, crc, tmp2);
4517     ldr(tmp0, Address(buf, 8));
4518     crc32x(crc, crc, tmp3);
4519     ldr(tmp1, Address(buf, 16));
4520     crc32x(crc, crc, tmp0);
4521     ldr(tmp2, Address(buf, 24));
4522     crc32x(crc, crc, tmp1);
4523     ldr(tmp3, Address(buf, 32));
4524     crc32x(crc, crc, tmp2);
4525     ldr(tmp0, Address(buf, 40));
4526     crc32x(crc, crc, tmp3);
4527     ldr(tmp1, Address(buf, 48));
4528     crc32x(crc, crc, tmp0);
4529     ldr(tmp2, Address(buf, 56));
4530     crc32x(crc, crc, tmp1);
4531     ldr(tmp3, Address(pre(buf, 64)));
4532     br(Assembler::GE, CRC_by64_loop);
4533 
4534     // post-loop
4535     crc32x(crc, crc, tmp2);
4536     crc32x(crc, crc, tmp3);
4537 
4538     sub(len, len, 64);
4539     add(buf, buf, 8);
4540     cmn(len, (u1)128);
4541     br(Assembler::NE, CRC_less64);
4542   BIND(L_exit);
4543     mvnw(crc, crc);
4544 }
4545 
4546 /**
4547  * @param crc   register containing existing CRC (32-bit)
4548  * @param buf   register pointing to input byte buffer (byte*)
4549  * @param len   register containing number of bytes
4550  * @param table register that will contain address of CRC table
4551  * @param tmp   scratch register
4552  */
4553 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4554         Register table0, Register table1, Register table2, Register table3,
4555         Register tmp, Register tmp2, Register tmp3) {
4556   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4557 
4558   if (UseCryptoPmullForCRC32) {
4559       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4560       return;
4561   }
4562 
4563   if (UseCRC32) {
4564       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4565       return;
4566   }
4567 
4568     mvnw(crc, crc);
4569 
4570     {
4571       uint64_t offset;
4572       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4573       add(table0, table0, offset);
4574     }
4575     add(table1, table0, 1*256*sizeof(juint));
4576     add(table2, table0, 2*256*sizeof(juint));
4577     add(table3, table0, 3*256*sizeof(juint));
4578 
4579     { // Neon code start
4580       cmp(len, (u1)64);
4581       br(Assembler::LT, L_by16);
4582       eor(v16, T16B, v16, v16);
4583 
4584     Label L_fold;
4585 
4586       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4587 
4588       ld1(v0, v1, T2D, post(buf, 32));
4589       ld1r(v4, T2D, post(tmp, 8));
4590       ld1r(v5, T2D, post(tmp, 8));
4591       ld1r(v6, T2D, post(tmp, 8));
4592       ld1r(v7, T2D, post(tmp, 8));
4593       mov(v16, S, 0, crc);
4594 
4595       eor(v0, T16B, v0, v16);
4596       sub(len, len, 64);
4597 
4598     BIND(L_fold);
4599       pmull(v22, T8H, v0, v5, T8B);
4600       pmull(v20, T8H, v0, v7, T8B);
4601       pmull(v23, T8H, v0, v4, T8B);
4602       pmull(v21, T8H, v0, v6, T8B);
4603 
4604       pmull2(v18, T8H, v0, v5, T16B);
4605       pmull2(v16, T8H, v0, v7, T16B);
4606       pmull2(v19, T8H, v0, v4, T16B);
4607       pmull2(v17, T8H, v0, v6, T16B);
4608 
4609       uzp1(v24, T8H, v20, v22);
4610       uzp2(v25, T8H, v20, v22);
4611       eor(v20, T16B, v24, v25);
4612 
4613       uzp1(v26, T8H, v16, v18);
4614       uzp2(v27, T8H, v16, v18);
4615       eor(v16, T16B, v26, v27);
4616 
4617       ushll2(v22, T4S, v20, T8H, 8);
4618       ushll(v20, T4S, v20, T4H, 8);
4619 
4620       ushll2(v18, T4S, v16, T8H, 8);
4621       ushll(v16, T4S, v16, T4H, 8);
4622 
4623       eor(v22, T16B, v23, v22);
4624       eor(v18, T16B, v19, v18);
4625       eor(v20, T16B, v21, v20);
4626       eor(v16, T16B, v17, v16);
4627 
4628       uzp1(v17, T2D, v16, v20);
4629       uzp2(v21, T2D, v16, v20);
4630       eor(v17, T16B, v17, v21);
4631 
4632       ushll2(v20, T2D, v17, T4S, 16);
4633       ushll(v16, T2D, v17, T2S, 16);
4634 
4635       eor(v20, T16B, v20, v22);
4636       eor(v16, T16B, v16, v18);
4637 
4638       uzp1(v17, T2D, v20, v16);
4639       uzp2(v21, T2D, v20, v16);
4640       eor(v28, T16B, v17, v21);
4641 
4642       pmull(v22, T8H, v1, v5, T8B);
4643       pmull(v20, T8H, v1, v7, T8B);
4644       pmull(v23, T8H, v1, v4, T8B);
4645       pmull(v21, T8H, v1, v6, T8B);
4646 
4647       pmull2(v18, T8H, v1, v5, T16B);
4648       pmull2(v16, T8H, v1, v7, T16B);
4649       pmull2(v19, T8H, v1, v4, T16B);
4650       pmull2(v17, T8H, v1, v6, T16B);
4651 
4652       ld1(v0, v1, T2D, post(buf, 32));
4653 
4654       uzp1(v24, T8H, v20, v22);
4655       uzp2(v25, T8H, v20, v22);
4656       eor(v20, T16B, v24, v25);
4657 
4658       uzp1(v26, T8H, v16, v18);
4659       uzp2(v27, T8H, v16, v18);
4660       eor(v16, T16B, v26, v27);
4661 
4662       ushll2(v22, T4S, v20, T8H, 8);
4663       ushll(v20, T4S, v20, T4H, 8);
4664 
4665       ushll2(v18, T4S, v16, T8H, 8);
4666       ushll(v16, T4S, v16, T4H, 8);
4667 
4668       eor(v22, T16B, v23, v22);
4669       eor(v18, T16B, v19, v18);
4670       eor(v20, T16B, v21, v20);
4671       eor(v16, T16B, v17, v16);
4672 
4673       uzp1(v17, T2D, v16, v20);
4674       uzp2(v21, T2D, v16, v20);
4675       eor(v16, T16B, v17, v21);
4676 
4677       ushll2(v20, T2D, v16, T4S, 16);
4678       ushll(v16, T2D, v16, T2S, 16);
4679 
4680       eor(v20, T16B, v22, v20);
4681       eor(v16, T16B, v16, v18);
4682 
4683       uzp1(v17, T2D, v20, v16);
4684       uzp2(v21, T2D, v20, v16);
4685       eor(v20, T16B, v17, v21);
4686 
4687       shl(v16, T2D, v28, 1);
4688       shl(v17, T2D, v20, 1);
4689 
4690       eor(v0, T16B, v0, v16);
4691       eor(v1, T16B, v1, v17);
4692 
4693       subs(len, len, 32);
4694       br(Assembler::GE, L_fold);
4695 
4696       mov(crc, 0);
4697       mov(tmp, v0, D, 0);
4698       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4699       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4700       mov(tmp, v0, D, 1);
4701       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4702       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4703       mov(tmp, v1, D, 0);
4704       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4705       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4706       mov(tmp, v1, D, 1);
4707       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4708       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4709 
4710       add(len, len, 32);
4711     } // Neon code end
4712 
4713   BIND(L_by16);
4714     subs(len, len, 16);
4715     br(Assembler::GE, L_by16_loop);
4716     adds(len, len, 16-4);
4717     br(Assembler::GE, L_by4_loop);
4718     adds(len, len, 4);
4719     br(Assembler::GT, L_by1_loop);
4720     b(L_exit);
4721 
4722   BIND(L_by4_loop);
4723     ldrw(tmp, Address(post(buf, 4)));
4724     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4725     subs(len, len, 4);
4726     br(Assembler::GE, L_by4_loop);
4727     adds(len, len, 4);
4728     br(Assembler::LE, L_exit);
4729   BIND(L_by1_loop);
4730     subs(len, len, 1);
4731     ldrb(tmp, Address(post(buf, 1)));
4732     update_byte_crc32(crc, tmp, table0);
4733     br(Assembler::GT, L_by1_loop);
4734     b(L_exit);
4735 
4736     align(CodeEntryAlignment);
4737   BIND(L_by16_loop);
4738     subs(len, len, 16);
4739     ldp(tmp, tmp3, Address(post(buf, 16)));
4740     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4741     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4742     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4743     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4744     br(Assembler::GE, L_by16_loop);
4745     adds(len, len, 16-4);
4746     br(Assembler::GE, L_by4_loop);
4747     adds(len, len, 4);
4748     br(Assembler::GT, L_by1_loop);
4749   BIND(L_exit);
4750     mvnw(crc, crc);
4751 }
4752 
4753 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4754         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4755     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4756     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4757 
4758     subs(tmp0, len, 384);
4759     br(Assembler::GE, CRC_by128_pre);
4760   BIND(CRC_less128);
4761     subs(len, len, 32);
4762     br(Assembler::GE, CRC_by32_loop);
4763   BIND(CRC_less32);
4764     adds(len, len, 32 - 4);
4765     br(Assembler::GE, CRC_by4_loop);
4766     adds(len, len, 4);
4767     br(Assembler::GT, CRC_by1_loop);
4768     b(L_exit);
4769 
4770   BIND(CRC_by32_loop);
4771     ldp(tmp0, tmp1, Address(buf));
4772     crc32cx(crc, crc, tmp0);
4773     ldr(tmp2, Address(buf, 16));
4774     crc32cx(crc, crc, tmp1);
4775     ldr(tmp3, Address(buf, 24));
4776     crc32cx(crc, crc, tmp2);
4777     add(buf, buf, 32);
4778     subs(len, len, 32);
4779     crc32cx(crc, crc, tmp3);
4780     br(Assembler::GE, CRC_by32_loop);
4781     cmn(len, (u1)32);
4782     br(Assembler::NE, CRC_less32);
4783     b(L_exit);
4784 
4785   BIND(CRC_by4_loop);
4786     ldrw(tmp0, Address(post(buf, 4)));
4787     subs(len, len, 4);
4788     crc32cw(crc, crc, tmp0);
4789     br(Assembler::GE, CRC_by4_loop);
4790     adds(len, len, 4);
4791     br(Assembler::LE, L_exit);
4792   BIND(CRC_by1_loop);
4793     ldrb(tmp0, Address(post(buf, 1)));
4794     subs(len, len, 1);
4795     crc32cb(crc, crc, tmp0);
4796     br(Assembler::GT, CRC_by1_loop);
4797     b(L_exit);
4798 
4799   BIND(CRC_by128_pre);
4800     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4801       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4802     mov(crc, 0);
4803     crc32cx(crc, crc, tmp0);
4804     crc32cx(crc, crc, tmp1);
4805 
4806     cbnz(len, CRC_less128);
4807 
4808   BIND(L_exit);
4809 }
4810 
4811 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4812         Register len, Register tmp0, Register tmp1, Register tmp2,
4813         Register tmp3) {
4814     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4815     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4816 
4817     subs(len, len, 128);
4818     br(Assembler::GE, CRC_by64_pre);
4819   BIND(CRC_less64);
4820     adds(len, len, 128-32);
4821     br(Assembler::GE, CRC_by32_loop);
4822   BIND(CRC_less32);
4823     adds(len, len, 32-4);
4824     br(Assembler::GE, CRC_by4_loop);
4825     adds(len, len, 4);
4826     br(Assembler::GT, CRC_by1_loop);
4827     b(L_exit);
4828 
4829   BIND(CRC_by32_loop);
4830     ldp(tmp0, tmp1, Address(post(buf, 16)));
4831     subs(len, len, 32);
4832     crc32cx(crc, crc, tmp0);
4833     ldr(tmp2, Address(post(buf, 8)));
4834     crc32cx(crc, crc, tmp1);
4835     ldr(tmp3, Address(post(buf, 8)));
4836     crc32cx(crc, crc, tmp2);
4837     crc32cx(crc, crc, tmp3);
4838     br(Assembler::GE, CRC_by32_loop);
4839     cmn(len, (u1)32);
4840     br(Assembler::NE, CRC_less32);
4841     b(L_exit);
4842 
4843   BIND(CRC_by4_loop);
4844     ldrw(tmp0, Address(post(buf, 4)));
4845     subs(len, len, 4);
4846     crc32cw(crc, crc, tmp0);
4847     br(Assembler::GE, CRC_by4_loop);
4848     adds(len, len, 4);
4849     br(Assembler::LE, L_exit);
4850   BIND(CRC_by1_loop);
4851     ldrb(tmp0, Address(post(buf, 1)));
4852     subs(len, len, 1);
4853     crc32cb(crc, crc, tmp0);
4854     br(Assembler::GT, CRC_by1_loop);
4855     b(L_exit);
4856 
4857   BIND(CRC_by64_pre);
4858     sub(buf, buf, 8);
4859     ldp(tmp0, tmp1, Address(buf, 8));
4860     crc32cx(crc, crc, tmp0);
4861     ldr(tmp2, Address(buf, 24));
4862     crc32cx(crc, crc, tmp1);
4863     ldr(tmp3, Address(buf, 32));
4864     crc32cx(crc, crc, tmp2);
4865     ldr(tmp0, Address(buf, 40));
4866     crc32cx(crc, crc, tmp3);
4867     ldr(tmp1, Address(buf, 48));
4868     crc32cx(crc, crc, tmp0);
4869     ldr(tmp2, Address(buf, 56));
4870     crc32cx(crc, crc, tmp1);
4871     ldr(tmp3, Address(pre(buf, 64)));
4872 
4873     b(CRC_by64_loop);
4874 
4875     align(CodeEntryAlignment);
4876   BIND(CRC_by64_loop);
4877     subs(len, len, 64);
4878     crc32cx(crc, crc, tmp2);
4879     ldr(tmp0, Address(buf, 8));
4880     crc32cx(crc, crc, tmp3);
4881     ldr(tmp1, Address(buf, 16));
4882     crc32cx(crc, crc, tmp0);
4883     ldr(tmp2, Address(buf, 24));
4884     crc32cx(crc, crc, tmp1);
4885     ldr(tmp3, Address(buf, 32));
4886     crc32cx(crc, crc, tmp2);
4887     ldr(tmp0, Address(buf, 40));
4888     crc32cx(crc, crc, tmp3);
4889     ldr(tmp1, Address(buf, 48));
4890     crc32cx(crc, crc, tmp0);
4891     ldr(tmp2, Address(buf, 56));
4892     crc32cx(crc, crc, tmp1);
4893     ldr(tmp3, Address(pre(buf, 64)));
4894     br(Assembler::GE, CRC_by64_loop);
4895 
4896     // post-loop
4897     crc32cx(crc, crc, tmp2);
4898     crc32cx(crc, crc, tmp3);
4899 
4900     sub(len, len, 64);
4901     add(buf, buf, 8);
4902     cmn(len, (u1)128);
4903     br(Assembler::NE, CRC_less64);
4904   BIND(L_exit);
4905 }
4906 
4907 /**
4908  * @param crc   register containing existing CRC (32-bit)
4909  * @param buf   register pointing to input byte buffer (byte*)
4910  * @param len   register containing number of bytes
4911  * @param table register that will contain address of CRC table
4912  * @param tmp   scratch register
4913  */
4914 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4915         Register table0, Register table1, Register table2, Register table3,
4916         Register tmp, Register tmp2, Register tmp3) {
4917   if (UseCryptoPmullForCRC32) {
4918     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4919   } else {
4920     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4921   }
4922 }
4923 
4924 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4925         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4926     Label CRC_by128_loop;
4927     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4928 
4929     sub(len, len, 256);
4930     Register table = tmp0;
4931     {
4932       uint64_t offset;
4933       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4934       add(table, table, offset);
4935     }
4936     add(table, table, table_offset);
4937 
4938     // Registers v0..v7 are used as data registers.
4939     // Registers v16..v31 are used as tmp registers.
4940     sub(buf, buf, 0x10);
4941     ldrq(v0, Address(buf, 0x10));
4942     ldrq(v1, Address(buf, 0x20));
4943     ldrq(v2, Address(buf, 0x30));
4944     ldrq(v3, Address(buf, 0x40));
4945     ldrq(v4, Address(buf, 0x50));
4946     ldrq(v5, Address(buf, 0x60));
4947     ldrq(v6, Address(buf, 0x70));
4948     ldrq(v7, Address(pre(buf, 0x80)));
4949 
4950     movi(v31, T4S, 0);
4951     mov(v31, S, 0, crc);
4952     eor(v0, T16B, v0, v31);
4953 
4954     // Register v16 contains constants from the crc table.
4955     ldrq(v16, Address(table));
4956     b(CRC_by128_loop);
4957 
4958     align(OptoLoopAlignment);
4959   BIND(CRC_by128_loop);
4960     pmull (v17,  T1Q, v0, v16, T1D);
4961     pmull2(v18, T1Q, v0, v16, T2D);
4962     ldrq(v0, Address(buf, 0x10));
4963     eor3(v0, T16B, v17,  v18, v0);
4964 
4965     pmull (v19, T1Q, v1, v16, T1D);
4966     pmull2(v20, T1Q, v1, v16, T2D);
4967     ldrq(v1, Address(buf, 0x20));
4968     eor3(v1, T16B, v19, v20, v1);
4969 
4970     pmull (v21, T1Q, v2, v16, T1D);
4971     pmull2(v22, T1Q, v2, v16, T2D);
4972     ldrq(v2, Address(buf, 0x30));
4973     eor3(v2, T16B, v21, v22, v2);
4974 
4975     pmull (v23, T1Q, v3, v16, T1D);
4976     pmull2(v24, T1Q, v3, v16, T2D);
4977     ldrq(v3, Address(buf, 0x40));
4978     eor3(v3, T16B, v23, v24, v3);
4979 
4980     pmull (v25, T1Q, v4, v16, T1D);
4981     pmull2(v26, T1Q, v4, v16, T2D);
4982     ldrq(v4, Address(buf, 0x50));
4983     eor3(v4, T16B, v25, v26, v4);
4984 
4985     pmull (v27, T1Q, v5, v16, T1D);
4986     pmull2(v28, T1Q, v5, v16, T2D);
4987     ldrq(v5, Address(buf, 0x60));
4988     eor3(v5, T16B, v27, v28, v5);
4989 
4990     pmull (v29, T1Q, v6, v16, T1D);
4991     pmull2(v30, T1Q, v6, v16, T2D);
4992     ldrq(v6, Address(buf, 0x70));
4993     eor3(v6, T16B, v29, v30, v6);
4994 
4995     // Reuse registers v23, v24.
4996     // Using them won't block the first instruction of the next iteration.
4997     pmull (v23, T1Q, v7, v16, T1D);
4998     pmull2(v24, T1Q, v7, v16, T2D);
4999     ldrq(v7, Address(pre(buf, 0x80)));
5000     eor3(v7, T16B, v23, v24, v7);
5001 
5002     subs(len, len, 0x80);
5003     br(Assembler::GE, CRC_by128_loop);
5004 
5005     // fold into 512 bits
5006     // Use v31 for constants because v16 can be still in use.
5007     ldrq(v31, Address(table, 0x10));
5008 
5009     pmull (v17,  T1Q, v0, v31, T1D);
5010     pmull2(v18, T1Q, v0, v31, T2D);
5011     eor3(v0, T16B, v17, v18, v4);
5012 
5013     pmull (v19, T1Q, v1, v31, T1D);
5014     pmull2(v20, T1Q, v1, v31, T2D);
5015     eor3(v1, T16B, v19, v20, v5);
5016 
5017     pmull (v21, T1Q, v2, v31, T1D);
5018     pmull2(v22, T1Q, v2, v31, T2D);
5019     eor3(v2, T16B, v21, v22, v6);
5020 
5021     pmull (v23, T1Q, v3, v31, T1D);
5022     pmull2(v24, T1Q, v3, v31, T2D);
5023     eor3(v3, T16B, v23, v24, v7);
5024 
5025     // fold into 128 bits
5026     // Use v17 for constants because v31 can be still in use.
5027     ldrq(v17, Address(table, 0x20));
5028     pmull (v25, T1Q, v0, v17, T1D);
5029     pmull2(v26, T1Q, v0, v17, T2D);
5030     eor3(v3, T16B, v3, v25, v26);
5031 
5032     // Use v18 for constants because v17 can be still in use.
5033     ldrq(v18, Address(table, 0x30));
5034     pmull (v27, T1Q, v1, v18, T1D);
5035     pmull2(v28, T1Q, v1, v18, T2D);
5036     eor3(v3, T16B, v3, v27, v28);
5037 
5038     // Use v19 for constants because v18 can be still in use.
5039     ldrq(v19, Address(table, 0x40));
5040     pmull (v29, T1Q, v2, v19, T1D);
5041     pmull2(v30, T1Q, v2, v19, T2D);
5042     eor3(v0, T16B, v3, v29, v30);
5043 
5044     add(len, len, 0x80);
5045     add(buf, buf, 0x10);
5046 
5047     mov(tmp0, v0, D, 0);
5048     mov(tmp1, v0, D, 1);
5049 }
5050 
5051 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5052   Address adr;
5053   switch(dst.getMode()) {
5054   case Address::base_plus_offset:
5055     // This is the expected mode, although we allow all the other
5056     // forms below.
5057     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5058     break;
5059   default:
5060     lea(rscratch2, dst);
5061     adr = Address(rscratch2);
5062     break;
5063   }
5064   ldr(rscratch1, adr);
5065   add(rscratch1, rscratch1, src);
5066   str(rscratch1, adr);
5067 }
5068 
5069 void MacroAssembler::cmpptr(Register src1, Address src2) {
5070   uint64_t offset;
5071   adrp(rscratch1, src2, offset);
5072   ldr(rscratch1, Address(rscratch1, offset));
5073   cmp(src1, rscratch1);
5074 }
5075 
5076 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5077   cmp(obj1, obj2);
5078 }
5079 
5080 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5081   load_method_holder(rresult, rmethod);
5082   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5083 }
5084 
5085 void MacroAssembler::load_method_holder(Register holder, Register method) {
5086   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5087   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5088   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5089 }
5090 
5091 // Loads the obj's Klass* into dst.
5092 // Preserves all registers (incl src, rscratch1 and rscratch2).
5093 // Input:
5094 // src - the oop we want to load the klass from.
5095 // dst - output narrow klass.
5096 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5097   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5098   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5099   lsr(dst, dst, markWord::klass_shift);
5100 }
5101 
5102 void MacroAssembler::load_klass(Register dst, Register src) {
5103   if (UseCompactObjectHeaders) {
5104     load_narrow_klass_compact(dst, src);
5105   } else {
5106     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5107   }
5108   decode_klass_not_null(dst);
5109 }
5110 
5111 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5112   if (RestoreMXCSROnJNICalls) {
5113     Label OK;
5114     get_fpcr(tmp1);
5115     mov(tmp2, tmp1);
5116     // Set FPCR to the state we need. We do want Round to Nearest. We
5117     // don't want non-IEEE rounding modes or floating-point traps.
5118     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5119     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5120     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5121     eor(tmp2, tmp1, tmp2);
5122     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5123     set_fpcr(tmp1);
5124     bind(OK);
5125   }
5126 }
5127 
5128 // ((OopHandle)result).resolve();
5129 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5130   // OopHandle::resolve is an indirection.
5131   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5132 }
5133 
5134 // ((WeakHandle)result).resolve();
5135 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5136   assert_different_registers(result, tmp1, tmp2);
5137   Label resolved;
5138 
5139   // A null weak handle resolves to null.
5140   cbz(result, resolved);
5141 
5142   // Only 64 bit platforms support GCs that require a tmp register
5143   // WeakHandle::resolve is an indirection like jweak.
5144   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5145                  result, Address(result), tmp1, tmp2);
5146   bind(resolved);
5147 }
5148 
5149 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5150   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5151   ldr(dst, Address(rmethod, Method::const_offset()));
5152   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5153   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5154   ldr(dst, Address(dst, mirror_offset));
5155   resolve_oop_handle(dst, tmp1, tmp2);
5156 }
5157 
5158 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5159   assert_different_registers(obj, klass, tmp);
5160   if (UseCompactObjectHeaders) {
5161     load_narrow_klass_compact(tmp, obj);
5162   } else {
5163     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5164   }
5165   if (CompressedKlassPointers::base() == nullptr) {
5166     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5167     return;
5168   } else if (!AOTCodeCache::is_on_for_dump() &&
5169              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5170              && CompressedKlassPointers::shift() == 0) {
5171     // Only the bottom 32 bits matter
5172     cmpw(klass, tmp);
5173     return;
5174   }
5175   decode_klass_not_null(tmp);
5176   cmp(klass, tmp);
5177 }
5178 
5179 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5180   if (UseCompactObjectHeaders) {
5181     load_narrow_klass_compact(tmp1, obj1);
5182     load_narrow_klass_compact(tmp2,  obj2);
5183   } else {
5184     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5185     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5186   }
5187   cmpw(tmp1, tmp2);
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   encode_klass_not_null(src);
5195   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5196 }
5197 
5198 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5199   assert(!UseCompactObjectHeaders, "not with compact headers");
5200   // Store to klass gap in destination
5201   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5202 }
5203 
5204 // Algorithm must match CompressedOops::encode.
5205 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5206 #ifdef ASSERT
5207   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5208 #endif
5209   verify_oop_msg(s, "broken oop in encode_heap_oop");
5210   if (CompressedOops::base() == nullptr) {
5211     if (CompressedOops::shift() != 0) {
5212       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5213       lsr(d, s, LogMinObjAlignmentInBytes);
5214     } else {
5215       mov(d, s);
5216     }
5217   } else {
5218     subs(d, s, rheapbase);
5219     csel(d, d, zr, Assembler::HS);
5220     lsr(d, d, LogMinObjAlignmentInBytes);
5221 
5222     /*  Old algorithm: is this any worse?
5223     Label nonnull;
5224     cbnz(r, nonnull);
5225     sub(r, r, rheapbase);
5226     bind(nonnull);
5227     lsr(r, r, LogMinObjAlignmentInBytes);
5228     */
5229   }
5230 }
5231 
5232 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5233 #ifdef ASSERT
5234   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5235   if (CheckCompressedOops) {
5236     Label ok;
5237     cbnz(r, ok);
5238     stop("null oop passed to encode_heap_oop_not_null");
5239     bind(ok);
5240   }
5241 #endif
5242   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5243   if (CompressedOops::base() != nullptr) {
5244     sub(r, r, rheapbase);
5245   }
5246   if (CompressedOops::shift() != 0) {
5247     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5248     lsr(r, r, LogMinObjAlignmentInBytes);
5249   }
5250 }
5251 
5252 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5253 #ifdef ASSERT
5254   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5255   if (CheckCompressedOops) {
5256     Label ok;
5257     cbnz(src, ok);
5258     stop("null oop passed to encode_heap_oop_not_null2");
5259     bind(ok);
5260   }
5261 #endif
5262   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5263 
5264   Register data = src;
5265   if (CompressedOops::base() != nullptr) {
5266     sub(dst, src, rheapbase);
5267     data = dst;
5268   }
5269   if (CompressedOops::shift() != 0) {
5270     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5271     lsr(dst, data, LogMinObjAlignmentInBytes);
5272     data = dst;
5273   }
5274   if (data == src)
5275     mov(dst, src);
5276 }
5277 
5278 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5279 #ifdef ASSERT
5280   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5281 #endif
5282   if (CompressedOops::base() == nullptr) {
5283     if (CompressedOops::shift() != 0) {
5284       lsl(d, s, CompressedOops::shift());
5285     } else if (d != s) {
5286       mov(d, s);
5287     }
5288   } else {
5289     Label done;
5290     if (d != s)
5291       mov(d, s);
5292     cbz(s, done);
5293     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5294     bind(done);
5295   }
5296   verify_oop_msg(d, "broken oop in decode_heap_oop");
5297 }
5298 
5299 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5300   assert (UseCompressedOops, "should only be used for compressed headers");
5301   assert (Universe::heap() != nullptr, "java heap should be initialized");
5302   // Cannot assert, unverified entry point counts instructions (see .ad file)
5303   // vtableStubs also counts instructions in pd_code_size_limit.
5304   // Also do not verify_oop as this is called by verify_oop.
5305   if (CompressedOops::shift() != 0) {
5306     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5307     if (CompressedOops::base() != nullptr) {
5308       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5309     } else {
5310       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5311     }
5312   } else {
5313     assert (CompressedOops::base() == nullptr, "sanity");
5314   }
5315 }
5316 
5317 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5318   assert (UseCompressedOops, "should only be used for compressed headers");
5319   assert (Universe::heap() != nullptr, "java heap should be initialized");
5320   // Cannot assert, unverified entry point counts instructions (see .ad file)
5321   // vtableStubs also counts instructions in pd_code_size_limit.
5322   // Also do not verify_oop as this is called by verify_oop.
5323   if (CompressedOops::shift() != 0) {
5324     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5325     if (CompressedOops::base() != nullptr) {
5326       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5327     } else {
5328       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5329     }
5330   } else {
5331     assert (CompressedOops::base() == nullptr, "sanity");
5332     if (dst != src) {
5333       mov(dst, src);
5334     }
5335   }
5336 }
5337 
5338 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5339 
5340 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5341   assert(Metaspace::initialized(), "metaspace not initialized yet");
5342   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5343   return _klass_decode_mode;
5344 }
5345 
5346 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5347   // KlassDecodeMode shouldn't be set already.
5348   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5349 
5350   if (base == nullptr) {
5351     return KlassDecodeZero;
5352   }
5353 
5354   if (operand_valid_for_logical_immediate(
5355         /*is32*/false, (uint64_t)base)) {
5356     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5357     if (((uint64_t)base & range_mask) == 0) {
5358       return KlassDecodeXor;
5359     }
5360   }
5361 
5362   const uint64_t shifted_base =
5363     (uint64_t)base >> shift;
5364   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5365     return KlassDecodeMovk;
5366   }
5367 
5368   // No valid encoding.
5369   return KlassDecodeNone;
5370 }
5371 
5372 // Check if one of the above decoding modes will work for given base, shift and range.
5373 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5374   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5375 }
5376 
5377 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5378   _klass_decode_mode = klass_decode_mode(base, shift, range);
5379   return _klass_decode_mode != KlassDecodeNone;
5380 }
5381 
5382 static Register pick_different_tmp(Register dst, Register src) {
5383   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5384   return *tmps.begin();
5385 }
5386 
5387 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5388   // we have to load the klass base from the AOT constants area but
5389   // not the shift because it is not allowed to change
5390   int shift = CompressedKlassPointers::shift();
5391   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5392   if (dst != src) {
5393     // we can load the base into dst, subtract it formthe src and shift down
5394     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5395     ldr(dst, dst);
5396     sub(dst, src, dst);
5397     lsr(dst, dst, shift);
5398   } else {
5399     // we need an extra register in order to load the coop base
5400     Register tmp = pick_different_tmp(dst, src);
5401     RegSet regs = RegSet::of(tmp);
5402     push(regs, sp);
5403     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5404     ldr(tmp, tmp);
5405     sub(dst, src, tmp);
5406     lsr(dst, dst, shift);
5407     pop(regs, sp);
5408   }
5409 }
5410 
5411 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5412   if (CompressedKlassPointers::base() != nullptr && AOTCodeCache::is_on_for_dump()) {
5413     encode_klass_not_null_for_aot(dst, src);
5414     return;
5415   }
5416 
5417   switch (klass_decode_mode()) {
5418   case KlassDecodeZero:
5419     if (CompressedKlassPointers::shift() != 0) {
5420       lsr(dst, src, CompressedKlassPointers::shift());
5421     } else {
5422       if (dst != src) mov(dst, src);
5423     }
5424     break;
5425 
5426   case KlassDecodeXor:
5427     if (CompressedKlassPointers::shift() != 0) {
5428       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5429       lsr(dst, dst, CompressedKlassPointers::shift());
5430     } else {
5431       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5432     }
5433     break;
5434 
5435   case KlassDecodeMovk:
5436     if (CompressedKlassPointers::shift() != 0) {
5437       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5438     } else {
5439       movw(dst, src);
5440     }
5441     break;
5442 
5443   case KlassDecodeNone:
5444     ShouldNotReachHere();
5445     break;
5446   }
5447 }
5448 
5449 void MacroAssembler::encode_klass_not_null(Register r) {
5450   encode_klass_not_null(r, r);
5451 }
5452 
5453 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5454   // we have to load the klass base from the AOT constants area but
5455   // not the shift because it is not allowed to change
5456   int shift = CompressedKlassPointers::shift();
5457   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5458   if (dst != src) {
5459     // we can load the base into dst then add the offset with a suitable shift
5460     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5461     ldr(dst, dst);
5462     add(dst, dst, src, LSL,  shift);
5463   } else {
5464     // we need an extra register in order to load the coop base
5465     Register tmp = pick_different_tmp(dst, src);
5466     RegSet regs = RegSet::of(tmp);
5467     push(regs, sp);
5468     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5469     ldr(tmp, tmp);
5470     add(dst, tmp,  src, LSL,  shift);
5471     pop(regs, sp);
5472   }
5473 }
5474 
5475 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5476   if (AOTCodeCache::is_on_for_dump()) {
5477     decode_klass_not_null_for_aot(dst, src);
5478     return;
5479   }
5480 
5481   switch (klass_decode_mode()) {
5482   case KlassDecodeZero:
5483     if (CompressedKlassPointers::shift() != 0) {
5484       lsl(dst, src, CompressedKlassPointers::shift());
5485     } else {
5486       if (dst != src) mov(dst, src);
5487     }
5488     break;
5489 
5490   case KlassDecodeXor:
5491     if (CompressedKlassPointers::shift() != 0) {
5492       lsl(dst, src, CompressedKlassPointers::shift());
5493       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5494     } else {
5495       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5496     }
5497     break;
5498 
5499   case KlassDecodeMovk: {
5500     const uint64_t shifted_base =
5501       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5502 
5503     if (dst != src) movw(dst, src);
5504     movk(dst, shifted_base >> 32, 32);
5505 
5506     if (CompressedKlassPointers::shift() != 0) {
5507       lsl(dst, dst, CompressedKlassPointers::shift());
5508     }
5509 
5510     break;
5511   }
5512 
5513   case KlassDecodeNone:
5514     ShouldNotReachHere();
5515     break;
5516   }
5517 }
5518 
5519 void  MacroAssembler::decode_klass_not_null(Register r) {
5520   decode_klass_not_null(r, r);
5521 }
5522 
5523 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5524 #ifdef ASSERT
5525   {
5526     ThreadInVMfromUnknown tiv;
5527     assert (UseCompressedOops, "should only be used for compressed oops");
5528     assert (Universe::heap() != nullptr, "java heap should be initialized");
5529     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5530     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5531   }
5532 #endif
5533   int oop_index = oop_recorder()->find_index(obj);
5534   InstructionMark im(this);
5535   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5536   code_section()->relocate(inst_mark(), rspec);
5537   movz(dst, 0xDEAD, 16);
5538   movk(dst, 0xBEEF);
5539 }
5540 
5541 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5542   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5543   int index = oop_recorder()->find_index(k);
5544 
5545   InstructionMark im(this);
5546   RelocationHolder rspec = metadata_Relocation::spec(index);
5547   code_section()->relocate(inst_mark(), rspec);
5548   narrowKlass nk = CompressedKlassPointers::encode(k);
5549   movz(dst, (nk >> 16), 16);
5550   movk(dst, nk & 0xffff);
5551 }
5552 
5553 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5554                                     Register dst, Address src,
5555                                     Register tmp1, Register tmp2) {
5556   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5557   decorators = AccessInternal::decorator_fixup(decorators, type);
5558   bool as_raw = (decorators & AS_RAW) != 0;
5559   if (as_raw) {
5560     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5561   } else {
5562     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5563   }
5564 }
5565 
5566 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5567                                      Address dst, Register val,
5568                                      Register tmp1, Register tmp2, Register tmp3) {
5569   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5570   decorators = AccessInternal::decorator_fixup(decorators, type);
5571   bool as_raw = (decorators & AS_RAW) != 0;
5572   if (as_raw) {
5573     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5574   } else {
5575     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5576   }
5577 }
5578 
5579 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5580                                    Register tmp2, DecoratorSet decorators) {
5581   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5582 }
5583 
5584 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5585                                             Register tmp2, DecoratorSet decorators) {
5586   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5587 }
5588 
5589 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5590                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5591   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5592 }
5593 
5594 // Used for storing nulls.
5595 void MacroAssembler::store_heap_oop_null(Address dst) {
5596   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5597 }
5598 
5599 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5600   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5601   int index = oop_recorder()->allocate_metadata_index(obj);
5602   RelocationHolder rspec = metadata_Relocation::spec(index);
5603   return Address((address)obj, rspec);
5604 }
5605 
5606 // Move an oop into a register.
5607 void MacroAssembler::movoop(Register dst, jobject obj) {
5608   int oop_index;
5609   if (obj == nullptr) {
5610     oop_index = oop_recorder()->allocate_oop_index(obj);
5611   } else {
5612 #ifdef ASSERT
5613     {
5614       ThreadInVMfromUnknown tiv;
5615       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5616     }
5617 #endif
5618     oop_index = oop_recorder()->find_index(obj);
5619   }
5620   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5621 
5622   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5623     mov(dst, Address((address)obj, rspec));
5624   } else {
5625     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5626     ldr(dst, Address(dummy, rspec));
5627   }
5628 }
5629 
5630 // Move a metadata address into a register.
5631 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5632   int oop_index;
5633   if (obj == nullptr) {
5634     oop_index = oop_recorder()->allocate_metadata_index(obj);
5635   } else {
5636     oop_index = oop_recorder()->find_index(obj);
5637   }
5638   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5639   mov(dst, Address((address)obj, rspec));
5640 }
5641 
5642 Address MacroAssembler::constant_oop_address(jobject obj) {
5643 #ifdef ASSERT
5644   {
5645     ThreadInVMfromUnknown tiv;
5646     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5647     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5648   }
5649 #endif
5650   int oop_index = oop_recorder()->find_index(obj);
5651   return Address((address)obj, oop_Relocation::spec(oop_index));
5652 }
5653 
5654 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5655 void MacroAssembler::tlab_allocate(Register obj,
5656                                    Register var_size_in_bytes,
5657                                    int con_size_in_bytes,
5658                                    Register t1,
5659                                    Register t2,
5660                                    Label& slow_case) {
5661   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5662   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5663 }
5664 
5665 void MacroAssembler::verify_tlab() {
5666 #ifdef ASSERT
5667   if (UseTLAB && VerifyOops) {
5668     Label next, ok;
5669 
5670     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5671 
5672     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5673     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5674     cmp(rscratch2, rscratch1);
5675     br(Assembler::HS, next);
5676     STOP("assert(top >= start)");
5677     should_not_reach_here();
5678 
5679     bind(next);
5680     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5681     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5682     cmp(rscratch2, rscratch1);
5683     br(Assembler::HS, ok);
5684     STOP("assert(top <= end)");
5685     should_not_reach_here();
5686 
5687     bind(ok);
5688     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5689   }
5690 #endif
5691 }
5692 
5693 // Writes to stack successive pages until offset reached to check for
5694 // stack overflow + shadow pages.  This clobbers tmp.
5695 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5696   assert_different_registers(tmp, size, rscratch1);
5697   mov(tmp, sp);
5698   // Bang stack for total size given plus shadow page size.
5699   // Bang one page at a time because large size can bang beyond yellow and
5700   // red zones.
5701   Label loop;
5702   mov(rscratch1, (int)os::vm_page_size());
5703   bind(loop);
5704   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5705   subsw(size, size, rscratch1);
5706   str(size, Address(tmp));
5707   br(Assembler::GT, loop);
5708 
5709   // Bang down shadow pages too.
5710   // At this point, (tmp-0) is the last address touched, so don't
5711   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5712   // was post-decremented.)  Skip this address by starting at i=1, and
5713   // touch a few more pages below.  N.B.  It is important to touch all
5714   // the way down to and including i=StackShadowPages.
5715   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5716     // this could be any sized move but this is can be a debugging crumb
5717     // so the bigger the better.
5718     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5719     str(size, Address(tmp));
5720   }
5721 }
5722 
5723 // Move the address of the polling page into dest.
5724 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5725   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5726 }
5727 
5728 // Read the polling page.  The address of the polling page must
5729 // already be in r.
5730 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5731   address mark;
5732   {
5733     InstructionMark im(this);
5734     code_section()->relocate(inst_mark(), rtype);
5735     ldrw(zr, Address(r, 0));
5736     mark = inst_mark();
5737   }
5738   verify_cross_modify_fence_not_required();
5739   return mark;
5740 }
5741 
5742 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5743   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5744   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5745   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5746   int64_t offset_low = dest_page - low_page;
5747   int64_t offset_high = dest_page - high_page;
5748 
5749   assert(is_valid_AArch64_address(dest.target()), "bad address");
5750   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5751 
5752   InstructionMark im(this);
5753   code_section()->relocate(inst_mark(), dest.rspec());
5754   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5755   // the code cache so that if it is relocated we know it will still reach
5756   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5757     _adrp(reg1, dest.target());
5758   } else {
5759     uint64_t target = (uint64_t)dest.target();
5760     uint64_t adrp_target
5761       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5762 
5763     _adrp(reg1, (address)adrp_target);
5764     movk(reg1, target >> 32, 32);
5765   }
5766   byte_offset = (uint64_t)dest.target() & 0xfff;
5767 }
5768 
5769 void MacroAssembler::load_byte_map_base(Register reg) {
5770 #if INCLUDE_CDS
5771   if (AOTCodeCache::is_on_for_dump()) {
5772     address byte_map_base_adr = AOTRuntimeConstants::card_table_base_address();
5773     lea(reg, ExternalAddress(byte_map_base_adr));
5774     ldr(reg, Address(reg));
5775     return;
5776   }
5777 #endif
5778   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
5779 
5780   // Strictly speaking the card table base isn't an address at all, and it might
5781   // even be negative. It is thus materialised as a constant.
5782   mov(reg, (uint64_t)ctbs->card_table_base_const());
5783 }
5784 
5785 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5786 #if INCLUDE_CDS
5787   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5788   if (AOTCodeCache::is_on_for_dump()) {
5789     // all aotrc field addresses should be registered in the AOTCodeCache address table
5790     lea(reg, ExternalAddress(a));
5791   } else {
5792     mov(reg, (uint64_t)a);
5793   }
5794 #else
5795   ShouldNotReachHere();
5796 #endif
5797 }
5798 
5799 void MacroAssembler::build_frame(int framesize) {
5800   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5801   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5802   protect_return_address();
5803   if (framesize < ((1 << 9) + 2 * wordSize)) {
5804     sub(sp, sp, framesize);
5805     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5806     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5807   } else {
5808     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5809     if (PreserveFramePointer) mov(rfp, sp);
5810     if (framesize < ((1 << 12) + 2 * wordSize))
5811       sub(sp, sp, framesize - 2 * wordSize);
5812     else {
5813       mov(rscratch1, framesize - 2 * wordSize);
5814       sub(sp, sp, rscratch1);
5815     }
5816   }
5817   verify_cross_modify_fence_not_required();
5818 }
5819 
5820 void MacroAssembler::remove_frame(int framesize) {
5821   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5822   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5823   if (framesize < ((1 << 9) + 2 * wordSize)) {
5824     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5825     add(sp, sp, framesize);
5826   } else {
5827     if (framesize < ((1 << 12) + 2 * wordSize))
5828       add(sp, sp, framesize - 2 * wordSize);
5829     else {
5830       mov(rscratch1, framesize - 2 * wordSize);
5831       add(sp, sp, rscratch1);
5832     }
5833     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5834   }
5835   authenticate_return_address();
5836 }
5837 
5838 
5839 // This method counts leading positive bytes (highest bit not set) in provided byte array
5840 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5841     // Simple and most common case of aligned small array which is not at the
5842     // end of memory page is placed here. All other cases are in stub.
5843     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5844     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5845     assert_different_registers(ary1, len, result);
5846 
5847     mov(result, len);
5848     cmpw(len, 0);
5849     br(LE, DONE);
5850     cmpw(len, 4 * wordSize);
5851     br(GE, STUB_LONG); // size > 32 then go to stub
5852 
5853     int shift = 64 - exact_log2(os::vm_page_size());
5854     lsl(rscratch1, ary1, shift);
5855     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5856     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5857     br(CS, STUB); // at the end of page then go to stub
5858     subs(len, len, wordSize);
5859     br(LT, END);
5860 
5861   BIND(LOOP);
5862     ldr(rscratch1, Address(post(ary1, wordSize)));
5863     tst(rscratch1, UPPER_BIT_MASK);
5864     br(NE, SET_RESULT);
5865     subs(len, len, wordSize);
5866     br(GE, LOOP);
5867     cmpw(len, -wordSize);
5868     br(EQ, DONE);
5869 
5870   BIND(END);
5871     ldr(rscratch1, Address(ary1));
5872     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5873     lslv(rscratch1, rscratch1, rscratch2);
5874     tst(rscratch1, UPPER_BIT_MASK);
5875     br(NE, SET_RESULT);
5876     b(DONE);
5877 
5878   BIND(STUB);
5879     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5880     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5881     address tpc1 = trampoline_call(count_pos);
5882     if (tpc1 == nullptr) {
5883       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5884       postcond(pc() == badAddress);
5885       return nullptr;
5886     }
5887     b(DONE);
5888 
5889   BIND(STUB_LONG);
5890     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5891     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5892     address tpc2 = trampoline_call(count_pos_long);
5893     if (tpc2 == nullptr) {
5894       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5895       postcond(pc() == badAddress);
5896       return nullptr;
5897     }
5898     b(DONE);
5899 
5900   BIND(SET_RESULT);
5901 
5902     add(len, len, wordSize);
5903     sub(result, result, len);
5904 
5905   BIND(DONE);
5906   postcond(pc() != badAddress);
5907   return pc();
5908 }
5909 
5910 // Clobbers: rscratch1, rscratch2, rflags
5911 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5912 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5913                                       Register tmp4, Register tmp5, Register result,
5914                                       Register cnt1, int elem_size) {
5915   Label DONE, SAME;
5916   Register tmp1 = rscratch1;
5917   Register tmp2 = rscratch2;
5918   int elem_per_word = wordSize/elem_size;
5919   int log_elem_size = exact_log2(elem_size);
5920   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5921   int length_offset = arrayOopDesc::length_offset_in_bytes();
5922   int base_offset
5923     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5924   // When the length offset is not aligned to 8 bytes,
5925   // then we align it down. This is valid because the new
5926   // offset will always be the klass which is the same
5927   // for type arrays.
5928   int start_offset = align_down(length_offset, BytesPerWord);
5929   int extra_length = base_offset - start_offset;
5930   assert(start_offset == length_offset || start_offset == klass_offset,
5931          "start offset must be 8-byte-aligned or be the klass offset");
5932   assert(base_offset != start_offset, "must include the length field");
5933   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5934   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5935 
5936   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5937   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5938 
5939 #ifndef PRODUCT
5940   {
5941     const char kind = (elem_size == 2) ? 'U' : 'L';
5942     char comment[64];
5943     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5944     BLOCK_COMMENT(comment);
5945   }
5946 #endif
5947 
5948   // if (a1 == a2)
5949   //     return true;
5950   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5951   br(EQ, SAME);
5952 
5953   if (UseSimpleArrayEquals) {
5954     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5955     // if (a1 == nullptr || a2 == nullptr)
5956     //     return false;
5957     // a1 & a2 == 0 means (some-pointer is null) or
5958     // (very-rare-or-even-probably-impossible-pointer-values)
5959     // so, we can save one branch in most cases
5960     tst(a1, a2);
5961     mov(result, false);
5962     br(EQ, A_MIGHT_BE_NULL);
5963     // if (a1.length != a2.length)
5964     //      return false;
5965     bind(A_IS_NOT_NULL);
5966     ldrw(cnt1, Address(a1, length_offset));
5967     ldrw(tmp5, Address(a2, length_offset));
5968     cmp(cnt1, tmp5);
5969     br(NE, DONE); // If lengths differ, return false
5970     // Increase loop counter by diff between base- and actual start-offset.
5971     addw(cnt1, cnt1, extra_length);
5972     lea(a1, Address(a1, start_offset));
5973     lea(a2, Address(a2, start_offset));
5974     // Check for short strings, i.e. smaller than wordSize.
5975     subs(cnt1, cnt1, elem_per_word);
5976     br(Assembler::LT, SHORT);
5977     // Main 8 byte comparison loop.
5978     bind(NEXT_WORD); {
5979       ldr(tmp1, Address(post(a1, wordSize)));
5980       ldr(tmp2, Address(post(a2, wordSize)));
5981       subs(cnt1, cnt1, elem_per_word);
5982       eor(tmp5, tmp1, tmp2);
5983       cbnz(tmp5, DONE);
5984     } br(GT, NEXT_WORD);
5985     // Last longword.  In the case where length == 4 we compare the
5986     // same longword twice, but that's still faster than another
5987     // conditional branch.
5988     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5989     // length == 4.
5990     if (log_elem_size > 0)
5991       lsl(cnt1, cnt1, log_elem_size);
5992     ldr(tmp3, Address(a1, cnt1));
5993     ldr(tmp4, Address(a2, cnt1));
5994     eor(tmp5, tmp3, tmp4);
5995     cbnz(tmp5, DONE);
5996     b(SAME);
5997     bind(A_MIGHT_BE_NULL);
5998     // in case both a1 and a2 are not-null, proceed with loads
5999     cbz(a1, DONE);
6000     cbz(a2, DONE);
6001     b(A_IS_NOT_NULL);
6002     bind(SHORT);
6003 
6004     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
6005     {
6006       ldrw(tmp1, Address(post(a1, 4)));
6007       ldrw(tmp2, Address(post(a2, 4)));
6008       eorw(tmp5, tmp1, tmp2);
6009       cbnzw(tmp5, DONE);
6010     }
6011     bind(TAIL03);
6012     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6013     {
6014       ldrh(tmp3, Address(post(a1, 2)));
6015       ldrh(tmp4, Address(post(a2, 2)));
6016       eorw(tmp5, tmp3, tmp4);
6017       cbnzw(tmp5, DONE);
6018     }
6019     bind(TAIL01);
6020     if (elem_size == 1) { // Only needed when comparing byte arrays.
6021       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6022       {
6023         ldrb(tmp1, a1);
6024         ldrb(tmp2, a2);
6025         eorw(tmp5, tmp1, tmp2);
6026         cbnzw(tmp5, DONE);
6027       }
6028     }
6029   } else {
6030     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6031         CSET_EQ, LAST_CHECK;
6032     mov(result, false);
6033     cbz(a1, DONE);
6034     ldrw(cnt1, Address(a1, length_offset));
6035     cbz(a2, DONE);
6036     ldrw(tmp5, Address(a2, length_offset));
6037     cmp(cnt1, tmp5);
6038     br(NE, DONE); // If lengths differ, return false
6039     // Increase loop counter by diff between base- and actual start-offset.
6040     addw(cnt1, cnt1, extra_length);
6041 
6042     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6043     // faster to perform another branch before comparing a1 and a2
6044     cmp(cnt1, (u1)elem_per_word);
6045     br(LE, SHORT); // short or same
6046     ldr(tmp3, Address(pre(a1, start_offset)));
6047     subs(zr, cnt1, stubBytesThreshold);
6048     br(GE, STUB);
6049     ldr(tmp4, Address(pre(a2, start_offset)));
6050     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6051 
6052     // Main 16 byte comparison loop with 2 exits
6053     bind(NEXT_DWORD); {
6054       ldr(tmp1, Address(pre(a1, wordSize)));
6055       ldr(tmp2, Address(pre(a2, wordSize)));
6056       subs(cnt1, cnt1, 2 * elem_per_word);
6057       br(LE, TAIL);
6058       eor(tmp4, tmp3, tmp4);
6059       cbnz(tmp4, DONE);
6060       ldr(tmp3, Address(pre(a1, wordSize)));
6061       ldr(tmp4, Address(pre(a2, wordSize)));
6062       cmp(cnt1, (u1)elem_per_word);
6063       br(LE, TAIL2);
6064       cmp(tmp1, tmp2);
6065     } br(EQ, NEXT_DWORD);
6066     b(DONE);
6067 
6068     bind(TAIL);
6069     eor(tmp4, tmp3, tmp4);
6070     eor(tmp2, tmp1, tmp2);
6071     lslv(tmp2, tmp2, tmp5);
6072     orr(tmp5, tmp4, tmp2);
6073     cmp(tmp5, zr);
6074     b(CSET_EQ);
6075 
6076     bind(TAIL2);
6077     eor(tmp2, tmp1, tmp2);
6078     cbnz(tmp2, DONE);
6079     b(LAST_CHECK);
6080 
6081     bind(STUB);
6082     ldr(tmp4, Address(pre(a2, start_offset)));
6083     if (elem_size == 2) { // convert to byte counter
6084       lsl(cnt1, cnt1, 1);
6085     }
6086     eor(tmp5, tmp3, tmp4);
6087     cbnz(tmp5, DONE);
6088     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6089     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6090     address tpc = trampoline_call(stub);
6091     if (tpc == nullptr) {
6092       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6093       postcond(pc() == badAddress);
6094       return nullptr;
6095     }
6096     b(DONE);
6097 
6098     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6099     // so, if a2 == null => return false(0), else return true, so we can return a2
6100     mov(result, a2);
6101     b(DONE);
6102     bind(SHORT);
6103     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6104     ldr(tmp3, Address(a1, start_offset));
6105     ldr(tmp4, Address(a2, start_offset));
6106     bind(LAST_CHECK);
6107     eor(tmp4, tmp3, tmp4);
6108     lslv(tmp5, tmp4, tmp5);
6109     cmp(tmp5, zr);
6110     bind(CSET_EQ);
6111     cset(result, EQ);
6112     b(DONE);
6113   }
6114 
6115   bind(SAME);
6116   mov(result, true);
6117   // That's it.
6118   bind(DONE);
6119 
6120   BLOCK_COMMENT("} array_equals");
6121   postcond(pc() != badAddress);
6122   return pc();
6123 }
6124 
6125 // Compare Strings
6126 
6127 // For Strings we're passed the address of the first characters in a1
6128 // and a2 and the length in cnt1.
6129 // There are two implementations.  For arrays >= 8 bytes, all
6130 // comparisons (including the final one, which may overlap) are
6131 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6132 // halfword, then a short, and then a byte.
6133 
6134 void MacroAssembler::string_equals(Register a1, Register a2,
6135                                    Register result, Register cnt1)
6136 {
6137   Label SAME, DONE, SHORT, NEXT_WORD;
6138   Register tmp1 = rscratch1;
6139   Register tmp2 = rscratch2;
6140 
6141   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6142 
6143 #ifndef PRODUCT
6144   {
6145     char comment[64];
6146     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6147     BLOCK_COMMENT(comment);
6148   }
6149 #endif
6150 
6151   mov(result, false);
6152 
6153   // Check for short strings, i.e. smaller than wordSize.
6154   subs(cnt1, cnt1, wordSize);
6155   br(Assembler::LT, SHORT);
6156   // Main 8 byte comparison loop.
6157   bind(NEXT_WORD); {
6158     ldr(tmp1, Address(post(a1, wordSize)));
6159     ldr(tmp2, Address(post(a2, wordSize)));
6160     subs(cnt1, cnt1, wordSize);
6161     eor(tmp1, tmp1, tmp2);
6162     cbnz(tmp1, DONE);
6163   } br(GT, NEXT_WORD);
6164   // Last longword.  In the case where length == 4 we compare the
6165   // same longword twice, but that's still faster than another
6166   // conditional branch.
6167   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6168   // length == 4.
6169   ldr(tmp1, Address(a1, cnt1));
6170   ldr(tmp2, Address(a2, cnt1));
6171   eor(tmp2, tmp1, tmp2);
6172   cbnz(tmp2, DONE);
6173   b(SAME);
6174 
6175   bind(SHORT);
6176   Label TAIL03, TAIL01;
6177 
6178   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6179   {
6180     ldrw(tmp1, Address(post(a1, 4)));
6181     ldrw(tmp2, Address(post(a2, 4)));
6182     eorw(tmp1, tmp1, tmp2);
6183     cbnzw(tmp1, DONE);
6184   }
6185   bind(TAIL03);
6186   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6187   {
6188     ldrh(tmp1, Address(post(a1, 2)));
6189     ldrh(tmp2, Address(post(a2, 2)));
6190     eorw(tmp1, tmp1, tmp2);
6191     cbnzw(tmp1, DONE);
6192   }
6193   bind(TAIL01);
6194   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6195     {
6196     ldrb(tmp1, a1);
6197     ldrb(tmp2, a2);
6198     eorw(tmp1, tmp1, tmp2);
6199     cbnzw(tmp1, DONE);
6200   }
6201   // Arrays are equal.
6202   bind(SAME);
6203   mov(result, true);
6204 
6205   // That's it.
6206   bind(DONE);
6207   BLOCK_COMMENT("} string_equals");
6208 }
6209 
6210 
6211 // The size of the blocks erased by the zero_blocks stub.  We must
6212 // handle anything smaller than this ourselves in zero_words().
6213 const int MacroAssembler::zero_words_block_size = 8;
6214 
6215 // zero_words() is used by C2 ClearArray patterns and by
6216 // C1_MacroAssembler.  It is as small as possible, handling small word
6217 // counts locally and delegating anything larger to the zero_blocks
6218 // stub.  It is expanded many times in compiled code, so it is
6219 // important to keep it short.
6220 
6221 // ptr:   Address of a buffer to be zeroed.
6222 // cnt:   Count in HeapWords.
6223 //
6224 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6225 address MacroAssembler::zero_words(Register ptr, Register cnt)
6226 {
6227   assert(is_power_of_2(zero_words_block_size), "adjust this");
6228 
6229   BLOCK_COMMENT("zero_words {");
6230   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6231   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6232   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6233 
6234   subs(rscratch1, cnt, zero_words_block_size);
6235   Label around;
6236   br(LO, around);
6237   {
6238     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6239     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6240     // Make sure this is a C2 compilation. C1 allocates space only for
6241     // trampoline stubs generated by Call LIR ops, and in any case it
6242     // makes sense for a C1 compilation task to proceed as quickly as
6243     // possible.
6244     CompileTask* task;
6245     if (StubRoutines::aarch64::complete()
6246         && Thread::current()->is_Compiler_thread()
6247         && (task = ciEnv::current()->task())
6248         && is_c2_compile(task->comp_level())) {
6249       address tpc = trampoline_call(zero_blocks);
6250       if (tpc == nullptr) {
6251         DEBUG_ONLY(reset_labels(around));
6252         return nullptr;
6253       }
6254     } else {
6255       far_call(zero_blocks);
6256     }
6257   }
6258   bind(around);
6259 
6260   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6261   // for us.
6262   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6263     Label l;
6264     tbz(cnt, exact_log2(i), l);
6265     for (int j = 0; j < i; j += 2) {
6266       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6267     }
6268     bind(l);
6269   }
6270   {
6271     Label l;
6272     tbz(cnt, 0, l);
6273     str(zr, Address(ptr));
6274     bind(l);
6275   }
6276 
6277   BLOCK_COMMENT("} zero_words");
6278   return pc();
6279 }
6280 
6281 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6282 // cnt:          Immediate count in HeapWords.
6283 //
6284 // r10, r11, rscratch1, and rscratch2 are clobbered.
6285 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6286 {
6287   assert(wordSize <= BlockZeroingLowLimit,
6288             "increase BlockZeroingLowLimit");
6289   address result = nullptr;
6290   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6291 #ifndef PRODUCT
6292     {
6293       char buf[64];
6294       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6295       BLOCK_COMMENT(buf);
6296     }
6297 #endif
6298     if (cnt >= 16) {
6299       uint64_t loops = cnt/16;
6300       if (loops > 1) {
6301         mov(rscratch2, loops - 1);
6302       }
6303       {
6304         Label loop;
6305         bind(loop);
6306         for (int i = 0; i < 16; i += 2) {
6307           stp(zr, zr, Address(base, i * BytesPerWord));
6308         }
6309         add(base, base, 16 * BytesPerWord);
6310         if (loops > 1) {
6311           subs(rscratch2, rscratch2, 1);
6312           br(GE, loop);
6313         }
6314       }
6315     }
6316     cnt %= 16;
6317     int i = cnt & 1;  // store any odd word to start
6318     if (i) str(zr, Address(base));
6319     for (; i < (int)cnt; i += 2) {
6320       stp(zr, zr, Address(base, i * wordSize));
6321     }
6322     BLOCK_COMMENT("} zero_words");
6323     result = pc();
6324   } else {
6325     mov(r10, base); mov(r11, cnt);
6326     result = zero_words(r10, r11);
6327   }
6328   return result;
6329 }
6330 
6331 // Zero blocks of memory by using DC ZVA.
6332 //
6333 // Aligns the base address first sufficiently for DC ZVA, then uses
6334 // DC ZVA repeatedly for every full block.  cnt is the size to be
6335 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6336 // in cnt.
6337 //
6338 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6339 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6340 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6341   Register tmp = rscratch1;
6342   Register tmp2 = rscratch2;
6343   int zva_length = VM_Version::zva_length();
6344   Label initial_table_end, loop_zva;
6345   Label fini;
6346 
6347   // Base must be 16 byte aligned. If not just return and let caller handle it
6348   tst(base, 0x0f);
6349   br(Assembler::NE, fini);
6350   // Align base with ZVA length.
6351   neg(tmp, base);
6352   andr(tmp, tmp, zva_length - 1);
6353 
6354   // tmp: the number of bytes to be filled to align the base with ZVA length.
6355   add(base, base, tmp);
6356   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6357   adr(tmp2, initial_table_end);
6358   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6359   br(tmp2);
6360 
6361   for (int i = -zva_length + 16; i < 0; i += 16)
6362     stp(zr, zr, Address(base, i));
6363   bind(initial_table_end);
6364 
6365   sub(cnt, cnt, zva_length >> 3);
6366   bind(loop_zva);
6367   dc(Assembler::ZVA, base);
6368   subs(cnt, cnt, zva_length >> 3);
6369   add(base, base, zva_length);
6370   br(Assembler::GE, loop_zva);
6371   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6372   bind(fini);
6373 }
6374 
6375 // base:   Address of a buffer to be filled, 8 bytes aligned.
6376 // cnt:    Count in 8-byte unit.
6377 // value:  Value to be filled with.
6378 // base will point to the end of the buffer after filling.
6379 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6380 {
6381 //  Algorithm:
6382 //
6383 //    if (cnt == 0) {
6384 //      return;
6385 //    }
6386 //    if ((p & 8) != 0) {
6387 //      *p++ = v;
6388 //    }
6389 //
6390 //    scratch1 = cnt & 14;
6391 //    cnt -= scratch1;
6392 //    p += scratch1;
6393 //    switch (scratch1 / 2) {
6394 //      do {
6395 //        cnt -= 16;
6396 //          p[-16] = v;
6397 //          p[-15] = v;
6398 //        case 7:
6399 //          p[-14] = v;
6400 //          p[-13] = v;
6401 //        case 6:
6402 //          p[-12] = v;
6403 //          p[-11] = v;
6404 //          // ...
6405 //        case 1:
6406 //          p[-2] = v;
6407 //          p[-1] = v;
6408 //        case 0:
6409 //          p += 16;
6410 //      } while (cnt);
6411 //    }
6412 //    if ((cnt & 1) == 1) {
6413 //      *p++ = v;
6414 //    }
6415 
6416   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6417 
6418   Label fini, skip, entry, loop;
6419   const int unroll = 8; // Number of stp instructions we'll unroll
6420 
6421   cbz(cnt, fini);
6422   tbz(base, 3, skip);
6423   str(value, Address(post(base, 8)));
6424   sub(cnt, cnt, 1);
6425   bind(skip);
6426 
6427   andr(rscratch1, cnt, (unroll-1) * 2);
6428   sub(cnt, cnt, rscratch1);
6429   add(base, base, rscratch1, Assembler::LSL, 3);
6430   adr(rscratch2, entry);
6431   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6432   br(rscratch2);
6433 
6434   bind(loop);
6435   add(base, base, unroll * 16);
6436   for (int i = -unroll; i < 0; i++)
6437     stp(value, value, Address(base, i * 16));
6438   bind(entry);
6439   subs(cnt, cnt, unroll * 2);
6440   br(Assembler::GE, loop);
6441 
6442   tbz(cnt, 0, fini);
6443   str(value, Address(post(base, 8)));
6444   bind(fini);
6445 }
6446 
6447 // Intrinsic for
6448 //
6449 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6450 //   Encodes char[] to byte[] in ISO-8859-1
6451 //
6452 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6453 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6454 //
6455 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6456 //   Encodes char[] to byte[] in ASCII
6457 //
6458 // This version always returns the number of characters copied, and does not
6459 // clobber the 'len' register. A successful copy will complete with the post-
6460 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6461 // post-condition: 0 <= 'res' < 'len'.
6462 //
6463 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6464 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6465 //       beyond the acceptable, even though the footprint would be smaller.
6466 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6467 //       avoid additional bloat.
6468 //
6469 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6470 void MacroAssembler::encode_iso_array(Register src, Register dst,
6471                                       Register len, Register res, bool ascii,
6472                                       FloatRegister vtmp0, FloatRegister vtmp1,
6473                                       FloatRegister vtmp2, FloatRegister vtmp3,
6474                                       FloatRegister vtmp4, FloatRegister vtmp5)
6475 {
6476   Register cnt = res;
6477   Register max = rscratch1;
6478   Register chk = rscratch2;
6479 
6480   prfm(Address(src), PLDL1STRM);
6481   movw(cnt, len);
6482 
6483 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6484 
6485   Label LOOP_32, DONE_32, FAIL_32;
6486 
6487   BIND(LOOP_32);
6488   {
6489     cmpw(cnt, 32);
6490     br(LT, DONE_32);
6491     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6492     // Extract lower bytes.
6493     FloatRegister vlo0 = vtmp4;
6494     FloatRegister vlo1 = vtmp5;
6495     uzp1(vlo0, T16B, vtmp0, vtmp1);
6496     uzp1(vlo1, T16B, vtmp2, vtmp3);
6497     // Merge bits...
6498     orr(vtmp0, T16B, vtmp0, vtmp1);
6499     orr(vtmp2, T16B, vtmp2, vtmp3);
6500     // Extract merged upper bytes.
6501     FloatRegister vhix = vtmp0;
6502     uzp2(vhix, T16B, vtmp0, vtmp2);
6503     // ISO-check on hi-parts (all zero).
6504     //                          ASCII-check on lo-parts (no sign).
6505     FloatRegister vlox = vtmp1; // Merge lower bytes.
6506                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6507     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6508     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6509     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6510                                 ASCII(orr(chk, chk, max));
6511     cbnz(chk, FAIL_32);
6512     subw(cnt, cnt, 32);
6513     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6514     b(LOOP_32);
6515   }
6516   BIND(FAIL_32);
6517   sub(src, src, 64);
6518   BIND(DONE_32);
6519 
6520   Label LOOP_8, SKIP_8;
6521 
6522   BIND(LOOP_8);
6523   {
6524     cmpw(cnt, 8);
6525     br(LT, SKIP_8);
6526     FloatRegister vhi = vtmp0;
6527     FloatRegister vlo = vtmp1;
6528     ld1(vtmp3, T8H, src);
6529     uzp1(vlo, T16B, vtmp3, vtmp3);
6530     uzp2(vhi, T16B, vtmp3, vtmp3);
6531     // ISO-check on hi-parts (all zero).
6532     //                          ASCII-check on lo-parts (no sign).
6533                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6534     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6535                                 ASCII(umov(max, vtmp2, B, 0));
6536                                 ASCII(orr(chk, chk, max));
6537     cbnz(chk, SKIP_8);
6538 
6539     strd(vlo, Address(post(dst, 8)));
6540     subw(cnt, cnt, 8);
6541     add(src, src, 16);
6542     b(LOOP_8);
6543   }
6544   BIND(SKIP_8);
6545 
6546 #undef ASCII
6547 
6548   Label LOOP, DONE;
6549 
6550   cbz(cnt, DONE);
6551   BIND(LOOP);
6552   {
6553     Register chr = rscratch1;
6554     ldrh(chr, Address(post(src, 2)));
6555     tst(chr, ascii ? 0xff80 : 0xff00);
6556     br(NE, DONE);
6557     strb(chr, Address(post(dst, 1)));
6558     subs(cnt, cnt, 1);
6559     br(GT, LOOP);
6560   }
6561   BIND(DONE);
6562   // Return index where we stopped.
6563   subw(res, len, cnt);
6564 }
6565 
6566 // Inflate byte[] array to char[].
6567 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6568 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6569                                            FloatRegister vtmp1, FloatRegister vtmp2,
6570                                            FloatRegister vtmp3, Register tmp4) {
6571   Label big, done, after_init, to_stub;
6572 
6573   assert_different_registers(src, dst, len, tmp4, rscratch1);
6574 
6575   fmovd(vtmp1, 0.0);
6576   lsrw(tmp4, len, 3);
6577   bind(after_init);
6578   cbnzw(tmp4, big);
6579   // Short string: less than 8 bytes.
6580   {
6581     Label loop, tiny;
6582 
6583     cmpw(len, 4);
6584     br(LT, tiny);
6585     // Use SIMD to do 4 bytes.
6586     ldrs(vtmp2, post(src, 4));
6587     zip1(vtmp3, T8B, vtmp2, vtmp1);
6588     subw(len, len, 4);
6589     strd(vtmp3, post(dst, 8));
6590 
6591     cbzw(len, done);
6592 
6593     // Do the remaining bytes by steam.
6594     bind(loop);
6595     ldrb(tmp4, post(src, 1));
6596     strh(tmp4, post(dst, 2));
6597     subw(len, len, 1);
6598 
6599     bind(tiny);
6600     cbnz(len, loop);
6601 
6602     b(done);
6603   }
6604 
6605   if (SoftwarePrefetchHintDistance >= 0) {
6606     bind(to_stub);
6607       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6608       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6609       address tpc = trampoline_call(stub);
6610       if (tpc == nullptr) {
6611         DEBUG_ONLY(reset_labels(big, done));
6612         postcond(pc() == badAddress);
6613         return nullptr;
6614       }
6615       b(after_init);
6616   }
6617 
6618   // Unpack the bytes 8 at a time.
6619   bind(big);
6620   {
6621     Label loop, around, loop_last, loop_start;
6622 
6623     if (SoftwarePrefetchHintDistance >= 0) {
6624       const int large_loop_threshold = (64 + 16)/8;
6625       ldrd(vtmp2, post(src, 8));
6626       andw(len, len, 7);
6627       cmp(tmp4, (u1)large_loop_threshold);
6628       br(GE, to_stub);
6629       b(loop_start);
6630 
6631       bind(loop);
6632       ldrd(vtmp2, post(src, 8));
6633       bind(loop_start);
6634       subs(tmp4, tmp4, 1);
6635       br(EQ, loop_last);
6636       zip1(vtmp2, T16B, vtmp2, vtmp1);
6637       ldrd(vtmp3, post(src, 8));
6638       st1(vtmp2, T8H, post(dst, 16));
6639       subs(tmp4, tmp4, 1);
6640       zip1(vtmp3, T16B, vtmp3, vtmp1);
6641       st1(vtmp3, T8H, post(dst, 16));
6642       br(NE, loop);
6643       b(around);
6644       bind(loop_last);
6645       zip1(vtmp2, T16B, vtmp2, vtmp1);
6646       st1(vtmp2, T8H, post(dst, 16));
6647       bind(around);
6648       cbz(len, done);
6649     } else {
6650       andw(len, len, 7);
6651       bind(loop);
6652       ldrd(vtmp2, post(src, 8));
6653       sub(tmp4, tmp4, 1);
6654       zip1(vtmp3, T16B, vtmp2, vtmp1);
6655       st1(vtmp3, T8H, post(dst, 16));
6656       cbnz(tmp4, loop);
6657     }
6658   }
6659 
6660   // Do the tail of up to 8 bytes.
6661   add(src, src, len);
6662   ldrd(vtmp3, Address(src, -8));
6663   add(dst, dst, len, ext::uxtw, 1);
6664   zip1(vtmp3, T16B, vtmp3, vtmp1);
6665   strq(vtmp3, Address(dst, -16));
6666 
6667   bind(done);
6668   postcond(pc() != badAddress);
6669   return pc();
6670 }
6671 
6672 // Compress char[] array to byte[].
6673 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6674 // Return the array length if every element in array can be encoded,
6675 // otherwise, the index of first non-latin1 (> 0xff) character.
6676 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6677                                          Register res,
6678                                          FloatRegister tmp0, FloatRegister tmp1,
6679                                          FloatRegister tmp2, FloatRegister tmp3,
6680                                          FloatRegister tmp4, FloatRegister tmp5) {
6681   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6682 }
6683 
6684 // java.math.round(double a)
6685 // Returns the closest long to the argument, with ties rounding to
6686 // positive infinity.  This requires some fiddling for corner
6687 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6688 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6689                                        FloatRegister ftmp) {
6690   Label DONE;
6691   BLOCK_COMMENT("java_round_double: { ");
6692   fmovd(rscratch1, src);
6693   // Use RoundToNearestTiesAway unless src small and -ve.
6694   fcvtasd(dst, src);
6695   // Test if src >= 0 || abs(src) >= 0x1.0p52
6696   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6697   mov(rscratch2, julong_cast(0x1.0p52));
6698   cmp(rscratch1, rscratch2);
6699   br(HS, DONE); {
6700     // src < 0 && abs(src) < 0x1.0p52
6701     // src may have a fractional part, so add 0.5
6702     fmovd(ftmp, 0.5);
6703     faddd(ftmp, src, ftmp);
6704     // Convert double to jlong, use RoundTowardsNegative
6705     fcvtmsd(dst, ftmp);
6706   }
6707   bind(DONE);
6708   BLOCK_COMMENT("} java_round_double");
6709 }
6710 
6711 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6712                                       FloatRegister ftmp) {
6713   Label DONE;
6714   BLOCK_COMMENT("java_round_float: { ");
6715   fmovs(rscratch1, src);
6716   // Use RoundToNearestTiesAway unless src small and -ve.
6717   fcvtassw(dst, src);
6718   // Test if src >= 0 || abs(src) >= 0x1.0p23
6719   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6720   mov(rscratch2, jint_cast(0x1.0p23f));
6721   cmp(rscratch1, rscratch2);
6722   br(HS, DONE); {
6723     // src < 0 && |src| < 0x1.0p23
6724     // src may have a fractional part, so add 0.5
6725     fmovs(ftmp, 0.5f);
6726     fadds(ftmp, src, ftmp);
6727     // Convert float to jint, use RoundTowardsNegative
6728     fcvtmssw(dst, ftmp);
6729   }
6730   bind(DONE);
6731   BLOCK_COMMENT("} java_round_float");
6732 }
6733 
6734 // get_thread() can be called anywhere inside generated code so we
6735 // need to save whatever non-callee save context might get clobbered
6736 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6737 // the call setup code.
6738 //
6739 // On Linux and Windows, aarch64_get_thread_helper() is implemented in
6740 // assembly and clobbers only r0, r1, and flags.
6741 // On other systems, the helper is a usual C function.
6742 //
6743 void MacroAssembler::get_thread(Register dst) {
6744   RegSet saved_regs =
6745     BSD_ONLY(RegSet::range(r0, r17) + lr - dst)
6746     NOT_BSD (RegSet::range(r0, r1)  + lr - dst);
6747 
6748   protect_return_address();
6749   push(saved_regs, sp);
6750 
6751   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6752   blr(lr);
6753   if (dst != c_rarg0) {
6754     mov(dst, c_rarg0);
6755   }
6756 
6757   pop(saved_regs, sp);
6758   authenticate_return_address();
6759 }
6760 
6761 void MacroAssembler::cache_wb(Address line) {
6762   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6763   assert(line.index() == noreg, "index should be noreg");
6764   assert(line.offset() == 0, "offset should be 0");
6765   // would like to assert this
6766   // assert(line._ext.shift == 0, "shift should be zero");
6767   if (VM_Version::supports_dcpop()) {
6768     // writeback using clear virtual address to point of persistence
6769     dc(Assembler::CVAP, line.base());
6770   } else {
6771     // no need to generate anything as Unsafe.writebackMemory should
6772     // never invoke this stub
6773   }
6774 }
6775 
6776 void MacroAssembler::cache_wbsync(bool is_pre) {
6777   // we only need a barrier post sync
6778   if (!is_pre) {
6779     membar(Assembler::AnyAny);
6780   }
6781 }
6782 
6783 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6784   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6785     return;
6786   }
6787   // Make sure that native code does not change SVE vector length.
6788   Label verify_ok;
6789   movw(tmp, zr);
6790   sve_inc(tmp, B);
6791   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6792   br(EQ, verify_ok);
6793   stop("Error: SVE vector length has changed since jvm startup");
6794   bind(verify_ok);
6795 }
6796 
6797 void MacroAssembler::verify_ptrue() {
6798   Label verify_ok;
6799   if (!UseSVE) {
6800     return;
6801   }
6802   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6803   sve_dec(rscratch1, B);
6804   cbz(rscratch1, verify_ok);
6805   stop("Error: the preserved predicate register (p7) elements are not all true");
6806   bind(verify_ok);
6807 }
6808 
6809 void MacroAssembler::safepoint_isb() {
6810   isb();
6811 #ifndef PRODUCT
6812   if (VerifyCrossModifyFence) {
6813     // Clear the thread state.
6814     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6815   }
6816 #endif
6817 }
6818 
6819 #ifndef PRODUCT
6820 void MacroAssembler::verify_cross_modify_fence_not_required() {
6821   if (VerifyCrossModifyFence) {
6822     // Check if thread needs a cross modify fence.
6823     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6824     Label fence_not_required;
6825     cbz(rscratch1, fence_not_required);
6826     // If it does then fail.
6827     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6828     mov(c_rarg0, rthread);
6829     blr(rscratch1);
6830     bind(fence_not_required);
6831   }
6832 }
6833 #endif
6834 
6835 void MacroAssembler::spin_wait() {
6836   block_comment("spin_wait {");
6837   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6838     switch (VM_Version::spin_wait_desc().inst()) {
6839       case SpinWait::NOP:
6840         nop();
6841         break;
6842       case SpinWait::ISB:
6843         isb();
6844         break;
6845       case SpinWait::YIELD:
6846         yield();
6847         break;
6848       case SpinWait::SB:
6849         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6850         sb();
6851         break;
6852       case SpinWait::WFET:
6853         spin_wait_wfet(VM_Version::spin_wait_desc().delay());
6854         break;
6855       default:
6856         ShouldNotReachHere();
6857     }
6858   }
6859   block_comment("}");
6860 }
6861 
6862 void MacroAssembler::spin_wait_wfet(int delay_ns) {
6863   // The sequence assumes CNTFRQ_EL0 is fixed to 1GHz. The assumption is valid
6864   // starting from Armv8.6, according to the "D12.1.2 The system counter" of the
6865   // Arm Architecture Reference Manual for A-profile architecture version M.a.a.
6866   // This is sufficient because FEAT_WFXT is introduced from Armv8.6.
6867   Register target = rscratch1;
6868   Register current = rscratch2;
6869   get_cntvctss_el0(current);
6870   add(target, current, delay_ns);
6871 
6872   Label L_wait_loop;
6873   bind(L_wait_loop);
6874 
6875   wfet(target);
6876   get_cntvctss_el0(current);
6877 
6878   cmp(current, target);
6879   br(LT, L_wait_loop);
6880 
6881   sb();
6882 }
6883 
6884 // Stack frame creation/removal
6885 
6886 void MacroAssembler::enter(bool strip_ret_addr) {
6887   if (strip_ret_addr) {
6888     // Addresses can only be signed once. If there are multiple nested frames being created
6889     // in the same function, then the return address needs stripping first.
6890     strip_return_address();
6891   }
6892   protect_return_address();
6893   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6894   mov(rfp, sp);
6895 }
6896 
6897 void MacroAssembler::leave() {
6898   mov(sp, rfp);
6899   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6900   authenticate_return_address();
6901 }
6902 
6903 // ROP Protection
6904 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6905 // destroying stack frames or whenever directly loading/storing the LR to memory.
6906 // If ROP protection is not set then these functions are no-ops.
6907 // For more details on PAC see pauth_aarch64.hpp.
6908 
6909 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6910 // Uses value zero as the modifier.
6911 //
6912 void MacroAssembler::protect_return_address() {
6913   if (VM_Version::use_rop_protection()) {
6914     check_return_address();
6915     paciaz();
6916   }
6917 }
6918 
6919 // Sign the return value in the given register. Use before updating the LR in the existing stack
6920 // frame for the current function.
6921 // Uses value zero as the modifier.
6922 //
6923 void MacroAssembler::protect_return_address(Register return_reg) {
6924   if (VM_Version::use_rop_protection()) {
6925     check_return_address(return_reg);
6926     paciza(return_reg);
6927   }
6928 }
6929 
6930 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6931 // Uses value zero as the modifier.
6932 //
6933 void MacroAssembler::authenticate_return_address() {
6934   if (VM_Version::use_rop_protection()) {
6935     autiaz();
6936     check_return_address();
6937   }
6938 }
6939 
6940 // Authenticate the return value in the given register. Use before updating the LR in the existing
6941 // stack frame for the current function.
6942 // Uses value zero as the modifier.
6943 //
6944 void MacroAssembler::authenticate_return_address(Register return_reg) {
6945   if (VM_Version::use_rop_protection()) {
6946     autiza(return_reg);
6947     check_return_address(return_reg);
6948   }
6949 }
6950 
6951 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6952 // there is no guaranteed way of authenticating the LR.
6953 //
6954 void MacroAssembler::strip_return_address() {
6955   if (VM_Version::use_rop_protection()) {
6956     xpaclri();
6957   }
6958 }
6959 
6960 #ifndef PRODUCT
6961 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6962 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6963 // it is difficult to debug back to the callee function.
6964 // This function simply loads from the address in the given register.
6965 // Use directly after authentication to catch authentication failures.
6966 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6967 //
6968 void MacroAssembler::check_return_address(Register return_reg) {
6969   if (VM_Version::use_rop_protection()) {
6970     ldr(zr, Address(return_reg));
6971   }
6972 }
6973 #endif
6974 
6975 // The java_calling_convention describes stack locations as ideal slots on
6976 // a frame with no abi restrictions. Since we must observe abi restrictions
6977 // (like the placement of the register window) the slots must be biased by
6978 // the following value.
6979 static int reg2offset_in(VMReg r) {
6980   // Account for saved rfp and lr
6981   // This should really be in_preserve_stack_slots
6982   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6983 }
6984 
6985 static int reg2offset_out(VMReg r) {
6986   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6987 }
6988 
6989 // On 64bit we will store integer like items to the stack as
6990 // 64bits items (AArch64 ABI) even though java would only store
6991 // 32bits for a parameter. On 32bit it will simply be 32bits
6992 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6993 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6994   if (src.first()->is_stack()) {
6995     if (dst.first()->is_stack()) {
6996       // stack to stack
6997       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6998       str(tmp, Address(sp, reg2offset_out(dst.first())));
6999     } else {
7000       // stack to reg
7001       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7002     }
7003   } else if (dst.first()->is_stack()) {
7004     // reg to stack
7005     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7006   } else {
7007     if (dst.first() != src.first()) {
7008       sxtw(dst.first()->as_Register(), src.first()->as_Register());
7009     }
7010   }
7011 }
7012 
7013 // An oop arg. Must pass a handle not the oop itself
7014 void MacroAssembler::object_move(
7015                         OopMap* map,
7016                         int oop_handle_offset,
7017                         int framesize_in_slots,
7018                         VMRegPair src,
7019                         VMRegPair dst,
7020                         bool is_receiver,
7021                         int* receiver_offset) {
7022 
7023   // must pass a handle. First figure out the location we use as a handle
7024 
7025   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
7026 
7027   // See if oop is null if it is we need no handle
7028 
7029   if (src.first()->is_stack()) {
7030 
7031     // Oop is already on the stack as an argument
7032     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
7033     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
7034     if (is_receiver) {
7035       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
7036     }
7037 
7038     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7039     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7040     // conditionally move a null
7041     cmp(rscratch1, zr);
7042     csel(rHandle, zr, rHandle, Assembler::EQ);
7043   } else {
7044 
7045     // Oop is in an a register we must store it to the space we reserve
7046     // on the stack for oop_handles and pass a handle if oop is non-null
7047 
7048     const Register rOop = src.first()->as_Register();
7049     int oop_slot;
7050     if (rOop == j_rarg0)
7051       oop_slot = 0;
7052     else if (rOop == j_rarg1)
7053       oop_slot = 1;
7054     else if (rOop == j_rarg2)
7055       oop_slot = 2;
7056     else if (rOop == j_rarg3)
7057       oop_slot = 3;
7058     else if (rOop == j_rarg4)
7059       oop_slot = 4;
7060     else if (rOop == j_rarg5)
7061       oop_slot = 5;
7062     else if (rOop == j_rarg6)
7063       oop_slot = 6;
7064     else {
7065       assert(rOop == j_rarg7, "wrong register");
7066       oop_slot = 7;
7067     }
7068 
7069     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7070     int offset = oop_slot*VMRegImpl::stack_slot_size;
7071 
7072     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7073     // Store oop in handle area, may be null
7074     str(rOop, Address(sp, offset));
7075     if (is_receiver) {
7076       *receiver_offset = offset;
7077     }
7078 
7079     cmp(rOop, zr);
7080     lea(rHandle, Address(sp, offset));
7081     // conditionally move a null
7082     csel(rHandle, zr, rHandle, Assembler::EQ);
7083   }
7084 
7085   // If arg is on the stack then place it otherwise it is already in correct reg.
7086   if (dst.first()->is_stack()) {
7087     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7088   }
7089 }
7090 
7091 // A float arg may have to do float reg int reg conversion
7092 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7093  if (src.first()->is_stack()) {
7094     if (dst.first()->is_stack()) {
7095       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7096       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7097     } else {
7098       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7099     }
7100   } else if (src.first() != dst.first()) {
7101     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7102       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7103     else
7104       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7105   }
7106 }
7107 
7108 // A long move
7109 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7110   if (src.first()->is_stack()) {
7111     if (dst.first()->is_stack()) {
7112       // stack to stack
7113       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7114       str(tmp, Address(sp, reg2offset_out(dst.first())));
7115     } else {
7116       // stack to reg
7117       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7118     }
7119   } else if (dst.first()->is_stack()) {
7120     // reg to stack
7121     // Do we really have to sign extend???
7122     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7123     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7124   } else {
7125     if (dst.first() != src.first()) {
7126       mov(dst.first()->as_Register(), src.first()->as_Register());
7127     }
7128   }
7129 }
7130 
7131 
7132 // A double move
7133 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7134  if (src.first()->is_stack()) {
7135     if (dst.first()->is_stack()) {
7136       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7137       str(tmp, Address(sp, reg2offset_out(dst.first())));
7138     } else {
7139       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7140     }
7141   } else if (src.first() != dst.first()) {
7142     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7143       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7144     else
7145       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7146   }
7147 }
7148 
7149 // Implements fast-locking.
7150 //
7151 //  - obj: the object to be locked
7152 //  - t1, t2, t3: temporary registers, will be destroyed
7153 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7154 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7155   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7156 
7157   Label push;
7158   const Register top = t1;
7159   const Register mark = t2;
7160   const Register t = t3;
7161 
7162   // Preload the markWord. It is important that this is the first
7163   // instruction emitted as it is part of C1's null check semantics.
7164   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7165 
7166   if (UseObjectMonitorTable) {
7167     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7168     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7169   }
7170 
7171   if (DiagnoseSyncOnValueBasedClasses != 0) {
7172     load_klass(t1, obj);
7173     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7174     tst(t1, KlassFlags::_misc_is_value_based_class);
7175     br(Assembler::NE, slow);
7176   }
7177 
7178   // Check if the lock-stack is full.
7179   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7180   cmpw(top, (unsigned)LockStack::end_offset());
7181   br(Assembler::GE, slow);
7182 
7183   // Check for recursion.
7184   subw(t, top, oopSize);
7185   ldr(t, Address(rthread, t));
7186   cmp(obj, t);
7187   br(Assembler::EQ, push);
7188 
7189   // Check header for monitor (0b10).
7190   tst(mark, markWord::monitor_value);
7191   br(Assembler::NE, slow);
7192 
7193   // Try to lock. Transition lock bits 0b01 => 0b00
7194   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7195   orr(mark, mark, markWord::unlocked_value);
7196   eor(t, mark, markWord::unlocked_value);
7197   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7198           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7199   br(Assembler::NE, slow);
7200 
7201   bind(push);
7202   // After successful lock, push object on lock-stack.
7203   str(obj, Address(rthread, top));
7204   addw(top, top, oopSize);
7205   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7206 }
7207 
7208 // Implements fast-unlocking.
7209 //
7210 // - obj: the object to be unlocked
7211 // - t1, t2, t3: temporary registers
7212 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7213 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7214   // cmpxchg clobbers rscratch1.
7215   assert_different_registers(obj, t1, t2, t3, rscratch1);
7216 
7217 #ifdef ASSERT
7218   {
7219     // Check for lock-stack underflow.
7220     Label stack_ok;
7221     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7222     cmpw(t1, (unsigned)LockStack::start_offset());
7223     br(Assembler::GE, stack_ok);
7224     STOP("Lock-stack underflow");
7225     bind(stack_ok);
7226   }
7227 #endif
7228 
7229   Label unlocked, push_and_slow;
7230   const Register top = t1;
7231   const Register mark = t2;
7232   const Register t = t3;
7233 
7234   // Check if obj is top of lock-stack.
7235   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7236   subw(top, top, oopSize);
7237   ldr(t, Address(rthread, top));
7238   cmp(obj, t);
7239   br(Assembler::NE, slow);
7240 
7241   // Pop lock-stack.
7242   DEBUG_ONLY(str(zr, Address(rthread, top));)
7243   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7244 
7245   // Check if recursive.
7246   subw(t, top, oopSize);
7247   ldr(t, Address(rthread, t));
7248   cmp(obj, t);
7249   br(Assembler::EQ, unlocked);
7250 
7251   // Not recursive. Check header for monitor (0b10).
7252   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7253   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7254 
7255 #ifdef ASSERT
7256   // Check header not unlocked (0b01).
7257   Label not_unlocked;
7258   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7259   stop("fast_unlock already unlocked");
7260   bind(not_unlocked);
7261 #endif
7262 
7263   // Try to unlock. Transition lock bits 0b00 => 0b01
7264   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7265   orr(t, mark, markWord::unlocked_value);
7266   cmpxchg(obj, mark, t, Assembler::xword,
7267           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7268   br(Assembler::EQ, unlocked);
7269 
7270   bind(push_and_slow);
7271   // Restore lock-stack and handle the unlock in runtime.
7272   DEBUG_ONLY(str(obj, Address(rthread, top));)
7273   addw(top, top, oopSize);
7274   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7275   b(slow);
7276 
7277   bind(unlocked);
7278 }
7279 
7280 // Rotate using USHR and SLI instructions (or copy, if rotate count is zero)
7281 void MacroAssembler::neon_vector_rotate(FloatRegister dst, SIMD_Arrangement T,
7282                                         FloatRegister src, int shift_amount) {
7283   assert(src != dst, "did not expect src and dst to be the same register");
7284 
7285   int esize = BitsPerByte << (T / 2);
7286   int lshift = shift_amount & (esize - 1);
7287 
7288   if (lshift == 0) {
7289     // T & 1 == 0 => 64-bit arrangements, else 128-bit arrangements
7290     orr(dst, (T & 1) == 0 ? T8B : T16B, src, src);
7291   } else {
7292     ushr(dst, T, src, esize - lshift);
7293     sli(dst, T, src, lshift);
7294   }
7295 }