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/powerOfTwo.hpp"
  59 #ifdef COMPILER1
  60 #include "c1/c1_LIRAssembler.hpp"
  61 #endif
  62 #ifdef COMPILER2
  63 #include "oops/oop.hpp"
  64 #include "opto/compile.hpp"
  65 #include "opto/node.hpp"
  66 #include "opto/output.hpp"
  67 #endif
  68 
  69 #include <sys/types.h>
  70 
  71 #ifdef PRODUCT
  72 #define BLOCK_COMMENT(str) /* nothing */
  73 #else
  74 #define BLOCK_COMMENT(str) block_comment(str)
  75 #endif
  76 #define STOP(str) stop(str);
  77 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  78 
  79 #ifdef ASSERT
  80 extern "C" void disnm(intptr_t p);
  81 #endif
  82 // Target-dependent relocation processing
  83 //
  84 // Instruction sequences whose target may need to be retrieved or
  85 // patched are distinguished by their leading instruction, sorting
  86 // them into three main instruction groups and related subgroups.
  87 //
  88 // 1) Branch, Exception and System (insn count = 1)
  89 //    1a) Unconditional branch (immediate):
  90 //      b/bl imm19
  91 //    1b) Compare & branch (immediate):
  92 //      cbz/cbnz Rt imm19
  93 //    1c) Test & branch (immediate):
  94 //      tbz/tbnz Rt imm14
  95 //    1d) Conditional branch (immediate):
  96 //      b.cond imm19
  97 //
  98 // 2) Loads and Stores (insn count = 1)
  99 //    2a) Load register literal:
 100 //      ldr Rt imm19
 101 //
 102 // 3) Data Processing Immediate (insn count = 2 or 3)
 103 //    3a) PC-rel. addressing
 104 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 105 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 106 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 107 //      adr/adrp Rx imm21
 108 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 109 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 110 //      The latter form can only happen when the target is an
 111 //      ExternalAddress, and (by definition) ExternalAddresses don't
 112 //      move. Because of that property, there is never any need to
 113 //      patch the last of the three instructions. However,
 114 //      MacroAssembler::target_addr_for_insn takes all three
 115 //      instructions into account and returns the correct address.
 116 //    3b) Move wide (immediate)
 117 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 118 //
 119 // A switch on a subset of the instruction's bits provides an
 120 // efficient dispatch to these subcases.
 121 //
 122 // insn[28:26] -> main group ('x' == don't care)
 123 //   00x -> UNALLOCATED
 124 //   100 -> Data Processing Immediate
 125 //   101 -> Branch, Exception and System
 126 //   x1x -> Loads and Stores
 127 //
 128 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 129 // n.b. in some cases extra bits need to be checked to verify the
 130 // instruction is as expected
 131 //
 132 // 1) ... xx101x Branch, Exception and System
 133 //   1a)  00___x Unconditional branch (immediate)
 134 //   1b)  01___0 Compare & branch (immediate)
 135 //   1c)  01___1 Test & branch (immediate)
 136 //   1d)  10___0 Conditional branch (immediate)
 137 //        other  Should not happen
 138 //
 139 // 2) ... xxx1x0 Loads and Stores
 140 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 141 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 142 //                strictly should be 64 bit non-FP/SIMD i.e.
 143 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 144 //
 145 // 3) ... xx100x Data Processing Immediate
 146 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 147 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 148 //                 strictly should be 64 bit movz #imm16<<0
 149 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 150 //
 151 
 152 static uint32_t insn_at(address insn_addr, int n) {
 153   return ((uint32_t*)insn_addr)[n];
 154 }
 155 
 156 template<typename T>
 157 class RelocActions : public AllStatic {
 158 
 159 public:
 160 
 161   static int ALWAYSINLINE run(address insn_addr, address &target) {
 162     int instructions = 1;
 163     uint32_t insn = insn_at(insn_addr, 0);
 164 
 165     uint32_t dispatch = Instruction_aarch64::extract(insn, 30, 25);
 166     switch(dispatch) {
 167       case 0b001010:
 168       case 0b001011: {
 169         instructions = T::unconditionalBranch(insn_addr, target);
 170         break;
 171       }
 172       case 0b101010:   // Conditional branch (immediate)
 173       case 0b011010: { // Compare & branch (immediate)
 174         instructions = T::conditionalBranch(insn_addr, target);
 175         break;
 176       }
 177       case 0b011011: {
 178         instructions = T::testAndBranch(insn_addr, target);
 179         break;
 180       }
 181       case 0b001100:
 182       case 0b001110:
 183       case 0b011100:
 184       case 0b011110:
 185       case 0b101100:
 186       case 0b101110:
 187       case 0b111100:
 188       case 0b111110: {
 189         // load/store
 190         if ((Instruction_aarch64::extract(insn, 29, 24) & 0b111011) == 0b011000) {
 191           // Load register (literal)
 192           instructions = T::loadStore(insn_addr, target);
 193           break;
 194         } else {
 195           // nothing to do
 196           assert(target == nullptr, "did not expect to relocate target for polling page load");
 197         }
 198         break;
 199       }
 200       case 0b001000:
 201       case 0b011000:
 202       case 0b101000:
 203       case 0b111000: {
 204         // adr/adrp
 205         assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 206         int shift = Instruction_aarch64::extract(insn, 31, 31);
 207         if (shift) {
 208           uint32_t insn2 = insn_at(insn_addr, 1);
 209           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 210               Instruction_aarch64::extract(insn, 4, 0) ==
 211               Instruction_aarch64::extract(insn2, 9, 5)) {
 212             instructions = T::adrp(insn_addr, target, T::adrpMem);
 213           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 214                      Instruction_aarch64::extract(insn, 4, 0) ==
 215                      Instruction_aarch64::extract(insn2, 4, 0)) {
 216             instructions = T::adrp(insn_addr, target, T::adrpAdd);
 217           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 218                      Instruction_aarch64::extract(insn, 4, 0) ==
 219                      Instruction_aarch64::extract(insn2, 4, 0)) {
 220             instructions = T::adrp(insn_addr, target, T::adrpMovk);
 221           } else {
 222             ShouldNotReachHere();
 223           }
 224         } else {
 225           instructions = T::adr(insn_addr, target);
 226         }
 227         break;
 228       }
 229       case 0b001001:
 230       case 0b011001:
 231       case 0b101001:
 232       case 0b111001: {
 233         instructions = T::immediate(insn_addr, target);
 234         break;
 235       }
 236       default: {
 237         ShouldNotReachHere();
 238       }
 239     }
 240 
 241     T::verify(insn_addr, target);
 242     return instructions * NativeInstruction::instruction_size;
 243   }
 244 };
 245 
 246 class Patcher : public AllStatic {
 247 public:
 248   static int unconditionalBranch(address insn_addr, address &target) {
 249     intptr_t offset = (target - insn_addr) >> 2;
 250     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 251     return 1;
 252   }
 253   static int conditionalBranch(address insn_addr, address &target) {
 254     intptr_t offset = (target - insn_addr) >> 2;
 255     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 256     return 1;
 257   }
 258   static int testAndBranch(address insn_addr, address &target) {
 259     intptr_t offset = (target - insn_addr) >> 2;
 260     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 261     return 1;
 262   }
 263   static int loadStore(address insn_addr, address &target) {
 264     intptr_t offset = (target - insn_addr) >> 2;
 265     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 266     return 1;
 267   }
 268   static int adr(address insn_addr, address &target) {
 269 #ifdef ASSERT
 270     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 271 #endif
 272     // PC-rel. addressing
 273     ptrdiff_t offset = target - insn_addr;
 274     int offset_lo = offset & 3;
 275     offset >>= 2;
 276     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 277     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 278     return 1;
 279   }
 280   template<typename U>
 281   static int adrp(address insn_addr, address &target, U inner) {
 282     int instructions = 1;
 283 #ifdef ASSERT
 284     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 285 #endif
 286     ptrdiff_t offset = target - insn_addr;
 287     instructions = 2;
 288     precond(inner != nullptr);
 289     // Give the inner reloc a chance to modify the target.
 290     address adjusted_target = target;
 291     instructions = inner(insn_addr, adjusted_target);
 292     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 293     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 294     offset = adr_page - pc_page;
 295     int offset_lo = offset & 3;
 296     offset >>= 2;
 297     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 298     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 299     return instructions;
 300   }
 301   static int adrpMem(address insn_addr, address &target) {
 302     uintptr_t dest = (uintptr_t)target;
 303     int offset_lo = dest & 0xfff;
 304     uint32_t insn2 = insn_at(insn_addr, 1);
 305     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 306     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 307     guarantee(((dest >> size) << size) == dest, "misaligned target");
 308     return 2;
 309   }
 310   static int adrpAdd(address insn_addr, address &target) {
 311     uintptr_t dest = (uintptr_t)target;
 312     int offset_lo = dest & 0xfff;
 313     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 314     return 2;
 315   }
 316   static int adrpMovk(address insn_addr, address &target) {
 317     uintptr_t dest = uintptr_t(target);
 318     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 319     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 320     target = address(dest);
 321     return 2;
 322   }
 323   static int immediate(address insn_addr, address &target) {
 324     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010100, "must be");
 325     uint64_t dest = (uint64_t)target;
 326     // Move wide constant
 327     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 328     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 329     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 330     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 331     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 332     return 3;
 333   }
 334   static void verify(address insn_addr, address &target) {
 335 #ifdef ASSERT
 336     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 337     if (!(address_is == target)) {
 338       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 339       disnm((intptr_t)insn_addr);
 340       assert(address_is == target, "should be");
 341     }
 342 #endif
 343   }
 344 };
 345 
 346 // If insn1 and insn2 use the same register to form an address, either
 347 // by an offsetted LDR or a simple ADD, return the offset. If the
 348 // second instruction is an LDR, the offset may be scaled.
 349 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 350   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 351       Instruction_aarch64::extract(insn1, 4, 0) ==
 352       Instruction_aarch64::extract(insn2, 9, 5)) {
 353     // Load/store register (unsigned immediate)
 354     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 355     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 356     byte_offset <<= size;
 357     return true;
 358   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 359              Instruction_aarch64::extract(insn1, 4, 0) ==
 360              Instruction_aarch64::extract(insn2, 4, 0)) {
 361     // add (immediate)
 362     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 363     return true;
 364   }
 365   return false;
 366 }
 367 
 368 class AArch64Decoder : public AllStatic {
 369 public:
 370 
 371   static int loadStore(address insn_addr, address &target) {
 372     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 373     target = insn_addr + (offset << 2);
 374     return 1;
 375   }
 376   static int unconditionalBranch(address insn_addr, address &target) {
 377     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 25, 0);
 378     target = insn_addr + (offset << 2);
 379     return 1;
 380   }
 381   static int conditionalBranch(address insn_addr, address &target) {
 382     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 383     target = address(((uint64_t)insn_addr + (offset << 2)));
 384     return 1;
 385   }
 386   static int testAndBranch(address insn_addr, address &target) {
 387     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 18, 5);
 388     target = address(((uint64_t)insn_addr + (offset << 2)));
 389     return 1;
 390   }
 391   static int adr(address insn_addr, address &target) {
 392     // PC-rel. addressing
 393     uint32_t insn = insn_at(insn_addr, 0);
 394     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 395     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 396     target = address((uint64_t)insn_addr + offset);
 397     return 1;
 398   }
 399   template<typename U>
 400   static int adrp(address insn_addr, address &target, U inner) {
 401     uint32_t insn = insn_at(insn_addr, 0);
 402     assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 403     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 404     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 405     int shift = 12;
 406     offset <<= shift;
 407     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 408     target_page &= ((uint64_t)-1) << shift;
 409     target = address(target_page);
 410     precond(inner != nullptr);
 411     inner(insn_addr, target);
 412     return 2;
 413   }
 414   static int adrpMem(address insn_addr, address &target) {
 415     uint32_t insn2 = insn_at(insn_addr, 1);
 416     // Load/store register (unsigned immediate)
 417     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 418     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 419     byte_offset <<= size;
 420     target += byte_offset;
 421     return 2;
 422   }
 423   static int adrpAdd(address insn_addr, address &target) {
 424     uint32_t insn2 = insn_at(insn_addr, 1);
 425     // add (immediate)
 426     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 427     target += byte_offset;
 428     return 2;
 429   }
 430   static int adrpMovk(address insn_addr, address &target) {
 431     uint32_t insn2 = insn_at(insn_addr, 1);
 432     uint64_t dest = uint64_t(target);
 433     dest = (dest & 0xffff0000ffffffff) |
 434       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 435     target = address(dest);
 436 
 437     // We know the destination 4k page. Maybe we have a third
 438     // instruction.
 439     uint32_t insn = insn_at(insn_addr, 0);
 440     uint32_t insn3 = insn_at(insn_addr, 2);
 441     ptrdiff_t byte_offset;
 442     if (offset_for(insn, insn3, byte_offset)) {
 443       target += byte_offset;
 444       return 3;
 445     } else {
 446       return 2;
 447     }
 448   }
 449   static int immediate(address insn_addr, address &target) {
 450     uint32_t *insns = (uint32_t *)insn_addr;
 451     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 452     // Move wide constant: movz, movk, movk.  See movptr().
 453     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 454     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 455     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 456                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 457                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 458     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 459     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 460     return 3;
 461   }
 462   static void verify(address insn_addr, address &target) {
 463   }
 464 };
 465 
 466 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 467   address target;
 468   RelocActions<AArch64Decoder>::run(insn_addr, target);
 469   return target;
 470 }
 471 
 472 // Patch any kind of instruction; there may be several instructions.
 473 // Return the total length (in bytes) of the instructions.
 474 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 475   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 476   return RelocActions<Patcher>::run(insn_addr, target);
 477 }
 478 
 479 int MacroAssembler::patch_oop(address insn_addr, address o) {
 480   int instructions;
 481   unsigned insn = *(unsigned*)insn_addr;
 482   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 483 
 484   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 485 
 486   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 487   // narrow OOPs by setting the upper 16 bits in the first
 488   // instruction.
 489   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 490     // Move narrow OOP
 491     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 492     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 493     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 494     instructions = 2;
 495   } else {
 496     // Move wide OOP
 497     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 498     uintptr_t dest = (uintptr_t)o;
 499     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 500     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 501     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 502     instructions = 3;
 503   }
 504   return instructions * NativeInstruction::instruction_size;
 505 }
 506 
 507 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 508   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 509   // We encode narrow ones by setting the upper 16 bits in the first
 510   // instruction.
 511   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 512   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 513          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 514 
 515   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 516 
 517   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 518   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 519   return 2 * NativeInstruction::instruction_size;
 520 }
 521 
 522 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 523   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 524   if (at_return) {
 525     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 526     // we may safely use the sp instead to perform the stack watermark check.
 527     cmp(in_nmethod ? sp : rfp, tmp);
 528     br(Assembler::HI, slow_path);
 529   } else {
 530     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 531   }
 532 }
 533 
 534 void MacroAssembler::rt_call(address dest, Register tmp) {
 535   CodeBlob *cb = CodeCache::find_blob(dest);
 536   if (cb) {
 537     far_call(RuntimeAddress(dest));
 538   } else {
 539     lea(tmp, RuntimeAddress(dest));
 540     blr(tmp);
 541   }
 542 }
 543 
 544 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 545   if (!Continuations::enabled()) return;
 546   Label done;
 547   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 548   cmp(sp, rscratch1);
 549   br(Assembler::LS, done);
 550   mov(rscratch1, sp); // we can't use sp as the source in str
 551   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 552   bind(done);
 553 }
 554 
 555 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 556   if (!Continuations::enabled()) return;
 557   Label done;
 558   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 559   cmp(sp, rscratch1);
 560   br(Assembler::LO, done);
 561   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 562   bind(done);
 563 }
 564 
 565 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 566   // we must set sp to zero to clear frame
 567   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 568 
 569   // must clear fp, so that compiled frames are not confused; it is
 570   // possible that we need it only for debugging
 571   if (clear_fp) {
 572     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 573   }
 574 
 575   // Always clear the pc because it could have been set by make_walkable()
 576   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 577 }
 578 
 579 // Calls to C land
 580 //
 581 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 582 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 583 // has to be reset to 0. This is required to allow proper stack traversal.
 584 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 585                                          Register last_java_fp,
 586                                          Register last_java_pc,
 587                                          Register scratch) {
 588 
 589   if (last_java_pc->is_valid()) {
 590       str(last_java_pc, Address(rthread,
 591                                 JavaThread::frame_anchor_offset()
 592                                 + JavaFrameAnchor::last_Java_pc_offset()));
 593     }
 594 
 595   // determine last_java_sp register
 596   if (last_java_sp == sp) {
 597     mov(scratch, sp);
 598     last_java_sp = scratch;
 599   } else if (!last_java_sp->is_valid()) {
 600     last_java_sp = esp;
 601   }
 602 
 603   // last_java_fp is optional
 604   if (last_java_fp->is_valid()) {
 605     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 606   }
 607 
 608   // We must set sp last.
 609   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 610 }
 611 
 612 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 613                                          Register last_java_fp,
 614                                          address  last_java_pc,
 615                                          Register scratch) {
 616   assert(last_java_pc != nullptr, "must provide a valid PC");
 617 
 618   adr(scratch, last_java_pc);
 619   str(scratch, Address(rthread,
 620                        JavaThread::frame_anchor_offset()
 621                        + JavaFrameAnchor::last_Java_pc_offset()));
 622 
 623   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 624 }
 625 
 626 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 627                                          Register last_java_fp,
 628                                          Label &L,
 629                                          Register scratch) {
 630   if (L.is_bound()) {
 631     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 632   } else {
 633     InstructionMark im(this);
 634     L.add_patch_at(code(), locator());
 635     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 636   }
 637 }
 638 
 639 static inline bool target_needs_far_branch(address addr) {
 640   if (AOTCodeCache::is_on_for_dump()) {
 641     return true;
 642   }
 643   // codecache size <= 128M
 644   if (!MacroAssembler::far_branches()) {
 645     return false;
 646   }
 647   // codecache size > 240M
 648   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 649     return true;
 650   }
 651   // codecache size: 128M..240M
 652   return !CodeCache::is_non_nmethod(addr);
 653 }
 654 
 655 void MacroAssembler::far_call(Address entry, Register tmp) {
 656   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 657   assert(CodeCache::find_blob(entry.target()) != nullptr,
 658          "destination of far call not found in code cache");
 659   assert(entry.rspec().type() == relocInfo::external_word_type
 660          || entry.rspec().type() == relocInfo::runtime_call_type
 661          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 662   if (target_needs_far_branch(entry.target())) {
 663     uint64_t offset;
 664     // We can use ADRP here because we know that the total size of
 665     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 666     adrp(tmp, entry, offset);
 667     add(tmp, tmp, offset);
 668     blr(tmp);
 669   } else {
 670     bl(entry);
 671   }
 672 }
 673 
 674 int MacroAssembler::far_jump(Address entry, Register tmp) {
 675   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 676   assert(CodeCache::find_blob(entry.target()) != nullptr,
 677          "destination of far call not found in code cache");
 678   assert(entry.rspec().type() == relocInfo::external_word_type
 679          || entry.rspec().type() == relocInfo::runtime_call_type
 680          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 681   address start = pc();
 682   if (target_needs_far_branch(entry.target())) {
 683     uint64_t offset;
 684     // We can use ADRP here because we know that the total size of
 685     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 686     adrp(tmp, entry, offset);
 687     add(tmp, tmp, offset);
 688     br(tmp);
 689   } else {
 690     b(entry);
 691   }
 692   return pc() - start;
 693 }
 694 
 695 void MacroAssembler::reserved_stack_check() {
 696     // testing if reserved zone needs to be enabled
 697     Label no_reserved_zone_enabling;
 698 
 699     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 700     cmp(sp, rscratch1);
 701     br(Assembler::LO, no_reserved_zone_enabling);
 702 
 703     enter();   // LR and FP are live.
 704     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 705     mov(c_rarg0, rthread);
 706     blr(rscratch1);
 707     leave();
 708 
 709     // We have already removed our own frame.
 710     // throw_delayed_StackOverflowError will think that it's been
 711     // called by our caller.
 712     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 713     br(rscratch1);
 714     should_not_reach_here();
 715 
 716     bind(no_reserved_zone_enabling);
 717 }
 718 
 719 static void pass_arg0(MacroAssembler* masm, Register arg) {
 720   if (c_rarg0 != arg ) {
 721     masm->mov(c_rarg0, arg);
 722   }
 723 }
 724 
 725 static void pass_arg1(MacroAssembler* masm, Register arg) {
 726   if (c_rarg1 != arg ) {
 727     masm->mov(c_rarg1, arg);
 728   }
 729 }
 730 
 731 static void pass_arg2(MacroAssembler* masm, Register arg) {
 732   if (c_rarg2 != arg ) {
 733     masm->mov(c_rarg2, arg);
 734   }
 735 }
 736 
 737 static void pass_arg3(MacroAssembler* masm, Register arg) {
 738   if (c_rarg3 != arg ) {
 739     masm->mov(c_rarg3, arg);
 740   }
 741 }
 742 
 743 void MacroAssembler::call_VM_base(Register oop_result,
 744                                   Register java_thread,
 745                                   Register last_java_sp,
 746                                   Label*   return_pc,
 747                                   address  entry_point,
 748                                   int      number_of_arguments,
 749                                   bool     check_exceptions) {
 750    // determine java_thread register
 751   if (!java_thread->is_valid()) {
 752     java_thread = rthread;
 753   }
 754 
 755   // determine last_java_sp register
 756   if (!last_java_sp->is_valid()) {
 757     last_java_sp = esp;
 758   }
 759 
 760   // debugging support
 761   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 762   assert(java_thread == rthread, "unexpected register");
 763 #ifdef ASSERT
 764   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 765   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 766 #endif // ASSERT
 767 
 768   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 769   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 770 
 771   // push java thread (becomes first argument of C function)
 772 
 773   mov(c_rarg0, java_thread);
 774 
 775   // set last Java frame before call
 776   assert(last_java_sp != rfp, "can't use rfp");
 777 
 778   Label l;
 779   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 780 
 781   // do the call, remove parameters
 782   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 783 
 784   // lr could be poisoned with PAC signature during throw_pending_exception
 785   // if it was tail-call optimized by compiler, since lr is not callee-saved
 786   // reload it with proper value
 787   adr(lr, l);
 788 
 789   // reset last Java frame
 790   // Only interpreter should have to clear fp
 791   reset_last_Java_frame(true);
 792 
 793    // C++ interp handles this in the interpreter
 794   check_and_handle_popframe(java_thread);
 795   check_and_handle_earlyret(java_thread);
 796 
 797   if (check_exceptions) {
 798     // check for pending exceptions (java_thread is set upon return)
 799     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 800     Label ok;
 801     cbz(rscratch1, ok);
 802     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 803     br(rscratch1);
 804     bind(ok);
 805   }
 806 
 807   // get oop result if there is one and reset the value in the thread
 808   if (oop_result->is_valid()) {
 809     get_vm_result_oop(oop_result, java_thread);
 810   }
 811 }
 812 
 813 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 814   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 815 }
 816 
 817 // Check the entry target is always reachable from any branch.
 818 static bool is_always_within_branch_range(Address entry) {
 819   if (AOTCodeCache::is_on_for_dump()) {
 820     return false;
 821   }
 822   const address target = entry.target();
 823 
 824   if (!CodeCache::contains(target)) {
 825     // We always use trampolines for callees outside CodeCache.
 826     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 827     return false;
 828   }
 829 
 830   if (!MacroAssembler::far_branches()) {
 831     return true;
 832   }
 833 
 834   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 835     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 836     // Non-compiled methods stay forever in CodeCache.
 837     // We check whether the longest possible branch is within the branch range.
 838     assert(CodeCache::find_blob(target) != nullptr &&
 839           !CodeCache::find_blob(target)->is_nmethod(),
 840           "runtime call of compiled method");
 841     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 842     const address left_longest_branch_start = CodeCache::low_bound();
 843     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 844                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 845     return is_reachable;
 846   }
 847 
 848   return false;
 849 }
 850 
 851 // Maybe emit a call via a trampoline. If the code cache is small
 852 // trampolines won't be emitted.
 853 address MacroAssembler::trampoline_call(Address entry) {
 854   assert(entry.rspec().type() == relocInfo::runtime_call_type
 855          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 856          || entry.rspec().type() == relocInfo::static_call_type
 857          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 858 
 859   address target = entry.target();
 860 
 861   if (!is_always_within_branch_range(entry)) {
 862     if (!in_scratch_emit_size()) {
 863       // We don't want to emit a trampoline if C2 is generating dummy
 864       // code during its branch shortening phase.
 865       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 866         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 867         code()->share_trampoline_for(entry.target(), offset());
 868       } else {
 869         address stub = emit_trampoline_stub(offset(), target);
 870         if (stub == nullptr) {
 871           postcond(pc() == badAddress);
 872           return nullptr; // CodeCache is full
 873         }
 874       }
 875     }
 876     target = pc();
 877   }
 878 
 879   address call_pc = pc();
 880   relocate(entry.rspec());
 881   bl(target);
 882 
 883   postcond(pc() != badAddress);
 884   return call_pc;
 885 }
 886 
 887 // Emit a trampoline stub for a call to a target which is too far away.
 888 //
 889 // code sequences:
 890 //
 891 // call-site:
 892 //   branch-and-link to <destination> or <trampoline stub>
 893 //
 894 // Related trampoline stub for this call site in the stub section:
 895 //   load the call target from the constant pool
 896 //   branch (LR still points to the call site above)
 897 
 898 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 899                                              address dest) {
 900   // Max stub size: alignment nop, TrampolineStub.
 901   address stub = start_a_stub(max_trampoline_stub_size());
 902   if (stub == nullptr) {
 903     return nullptr;  // CodeBuffer::expand failed
 904   }
 905 
 906   // Create a trampoline stub relocation which relates this trampoline stub
 907   // with the call instruction at insts_call_instruction_offset in the
 908   // instructions code-section.
 909   align(wordSize);
 910   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 911                                             + insts_call_instruction_offset));
 912   const int stub_start_offset = offset();
 913 
 914   // Now, create the trampoline stub's code:
 915   // - load the call
 916   // - call
 917   Label target;
 918   ldr(rscratch1, target);
 919   br(rscratch1);
 920   bind(target);
 921   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 922          "should be");
 923   emit_int64((int64_t)dest);
 924 
 925   const address stub_start_addr = addr_at(stub_start_offset);
 926 
 927   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 928 
 929   end_a_stub();
 930   return stub_start_addr;
 931 }
 932 
 933 int MacroAssembler::max_trampoline_stub_size() {
 934   // Max stub size: alignment nop, TrampolineStub.
 935   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 936 }
 937 
 938 void MacroAssembler::emit_static_call_stub() {
 939   // CompiledDirectCall::set_to_interpreted knows the
 940   // exact layout of this stub.
 941 
 942   isb();
 943   mov_metadata(rmethod, nullptr);
 944 
 945   // Jump to the entry point of the c2i stub.
 946   if (codestub_branch_needs_far_jump()) {
 947     movptr(rscratch1, 0);
 948     br(rscratch1);
 949   } else {
 950     b(pc());
 951   }
 952 }
 953 
 954 int MacroAssembler::static_call_stub_size() {
 955   if (!codestub_branch_needs_far_jump()) {
 956     // isb; movk; movz; movz; b
 957     return 5 * NativeInstruction::instruction_size;
 958   }
 959   // isb; movk; movz; movz; movk; movz; movz; br
 960   return 8 * NativeInstruction::instruction_size;
 961 }
 962 
 963 void MacroAssembler::c2bool(Register x) {
 964   // implements x == 0 ? 0 : 1
 965   // note: must only look at least-significant byte of x
 966   //       since C-style booleans are stored in one byte
 967   //       only! (was bug)
 968   tst(x, 0xff);
 969   cset(x, Assembler::NE);
 970 }
 971 
 972 address MacroAssembler::ic_call(address entry, jint method_index) {
 973   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 974   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 975   return trampoline_call(Address(entry, rh));
 976 }
 977 
 978 int MacroAssembler::ic_check_size() {
 979   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 980   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 981     return NativeInstruction::instruction_size * (7 + extra_instructions);
 982   } else {
 983     return NativeInstruction::instruction_size * (5 + extra_instructions);
 984   }
 985 }
 986 
 987 int MacroAssembler::ic_check(int end_alignment) {
 988   Register receiver = j_rarg0;
 989   Register data = rscratch2;
 990   Register tmp1 = rscratch1;
 991   Register tmp2 = r10;
 992 
 993   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
 994   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
 995   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
 996   // before the inline cache check here, and not after
 997   align(end_alignment, offset() + ic_check_size());
 998 
 999   int uep_offset = offset();
1000 
1001   if (UseCompactObjectHeaders) {
1002     load_narrow_klass_compact(tmp1, receiver);
1003     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1004     cmpw(tmp1, tmp2);
1005   } else if (UseCompressedClassPointers) {
1006     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1007     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1008     cmpw(tmp1, tmp2);
1009   } else {
1010     ldr(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1011     ldr(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1012     cmp(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 check the last 2 instructions
2692       address prev2 = prev - NativeMembar::instruction_size;
2693       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2694         NativeMembar *bar2 = NativeMembar_at(prev2);
2695         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2696         BLOCK_COMMENT("merged membar(elided)");
2697         return;
2698       }
2699     }
2700   }
2701   code()->set_last_insn(pc());
2702   dmb(Assembler::barrier(order_constraint));
2703 }
2704 
2705 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2706   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2707     merge_ldst(rt, adr, size_in_bytes, is_store);
2708     code()->clear_last_insn();
2709     return true;
2710   } else {
2711     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2712     const uint64_t mask = size_in_bytes - 1;
2713     if (adr.getMode() == Address::base_plus_offset &&
2714         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2715       code()->set_last_insn(pc());
2716     }
2717     return false;
2718   }
2719 }
2720 
2721 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2722   // We always try to merge two adjacent loads into one ldp.
2723   if (!try_merge_ldst(Rx, adr, 8, false)) {
2724     Assembler::ldr(Rx, adr);
2725   }
2726 }
2727 
2728 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2729   // We always try to merge two adjacent loads into one ldp.
2730   if (!try_merge_ldst(Rw, adr, 4, false)) {
2731     Assembler::ldrw(Rw, adr);
2732   }
2733 }
2734 
2735 void MacroAssembler::str(Register Rx, const Address &adr) {
2736   // We always try to merge two adjacent stores into one stp.
2737   if (!try_merge_ldst(Rx, adr, 8, true)) {
2738     Assembler::str(Rx, adr);
2739   }
2740 }
2741 
2742 void MacroAssembler::strw(Register Rw, const Address &adr) {
2743   // We always try to merge two adjacent stores into one stp.
2744   if (!try_merge_ldst(Rw, adr, 4, true)) {
2745     Assembler::strw(Rw, adr);
2746   }
2747 }
2748 
2749 // MacroAssembler routines found actually to be needed
2750 
2751 void MacroAssembler::push(Register src)
2752 {
2753   str(src, Address(pre(esp, -1 * wordSize)));
2754 }
2755 
2756 void MacroAssembler::pop(Register dst)
2757 {
2758   ldr(dst, Address(post(esp, 1 * wordSize)));
2759 }
2760 
2761 // Note: load_unsigned_short used to be called load_unsigned_word.
2762 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2763   int off = offset();
2764   ldrh(dst, src);
2765   return off;
2766 }
2767 
2768 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2769   int off = offset();
2770   ldrb(dst, src);
2771   return off;
2772 }
2773 
2774 int MacroAssembler::load_signed_short(Register dst, Address src) {
2775   int off = offset();
2776   ldrsh(dst, src);
2777   return off;
2778 }
2779 
2780 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2781   int off = offset();
2782   ldrsb(dst, src);
2783   return off;
2784 }
2785 
2786 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2787   int off = offset();
2788   ldrshw(dst, src);
2789   return off;
2790 }
2791 
2792 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2793   int off = offset();
2794   ldrsbw(dst, src);
2795   return off;
2796 }
2797 
2798 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2799   switch (size_in_bytes) {
2800   case  8:  ldr(dst, src); break;
2801   case  4:  ldrw(dst, src); break;
2802   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2803   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2804   default:  ShouldNotReachHere();
2805   }
2806 }
2807 
2808 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2809   switch (size_in_bytes) {
2810   case  8:  str(src, dst); break;
2811   case  4:  strw(src, dst); break;
2812   case  2:  strh(src, dst); break;
2813   case  1:  strb(src, dst); break;
2814   default:  ShouldNotReachHere();
2815   }
2816 }
2817 
2818 void MacroAssembler::decrementw(Register reg, int value)
2819 {
2820   if (value < 0)  { incrementw(reg, -value);      return; }
2821   if (value == 0) {                               return; }
2822   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2823   /* else */ {
2824     guarantee(reg != rscratch2, "invalid dst for register decrement");
2825     movw(rscratch2, (unsigned)value);
2826     subw(reg, reg, rscratch2);
2827   }
2828 }
2829 
2830 void MacroAssembler::decrement(Register reg, int value)
2831 {
2832   if (value < 0)  { increment(reg, -value);      return; }
2833   if (value == 0) {                              return; }
2834   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2835   /* else */ {
2836     assert(reg != rscratch2, "invalid dst for register decrement");
2837     mov(rscratch2, (uint64_t)value);
2838     sub(reg, reg, rscratch2);
2839   }
2840 }
2841 
2842 void MacroAssembler::decrementw(Address dst, int value)
2843 {
2844   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2845   if (dst.getMode() == Address::literal) {
2846     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2847     lea(rscratch2, dst);
2848     dst = Address(rscratch2);
2849   }
2850   ldrw(rscratch1, dst);
2851   decrementw(rscratch1, value);
2852   strw(rscratch1, dst);
2853 }
2854 
2855 void MacroAssembler::decrement(Address dst, int value)
2856 {
2857   assert(!dst.uses(rscratch1), "invalid address for decrement");
2858   if (dst.getMode() == Address::literal) {
2859     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2860     lea(rscratch2, dst);
2861     dst = Address(rscratch2);
2862   }
2863   ldr(rscratch1, dst);
2864   decrement(rscratch1, value);
2865   str(rscratch1, dst);
2866 }
2867 
2868 void MacroAssembler::incrementw(Register reg, int value)
2869 {
2870   if (value < 0)  { decrementw(reg, -value);      return; }
2871   if (value == 0) {                               return; }
2872   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2873   /* else */ {
2874     assert(reg != rscratch2, "invalid dst for register increment");
2875     movw(rscratch2, (unsigned)value);
2876     addw(reg, reg, rscratch2);
2877   }
2878 }
2879 
2880 void MacroAssembler::increment(Register reg, int value)
2881 {
2882   if (value < 0)  { decrement(reg, -value);      return; }
2883   if (value == 0) {                              return; }
2884   if (value < (1 << 12)) { add(reg, reg, value); return; }
2885   /* else */ {
2886     assert(reg != rscratch2, "invalid dst for register increment");
2887     movw(rscratch2, (unsigned)value);
2888     add(reg, reg, rscratch2);
2889   }
2890 }
2891 
2892 void MacroAssembler::incrementw(Address dst, int value)
2893 {
2894   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2895   if (dst.getMode() == Address::literal) {
2896     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2897     lea(rscratch2, dst);
2898     dst = Address(rscratch2);
2899   }
2900   ldrw(rscratch1, dst);
2901   incrementw(rscratch1, value);
2902   strw(rscratch1, dst);
2903 }
2904 
2905 void MacroAssembler::increment(Address dst, int value)
2906 {
2907   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2908   if (dst.getMode() == Address::literal) {
2909     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2910     lea(rscratch2, dst);
2911     dst = Address(rscratch2);
2912   }
2913   ldr(rscratch1, dst);
2914   increment(rscratch1, value);
2915   str(rscratch1, dst);
2916 }
2917 
2918 // Push lots of registers in the bit set supplied.  Don't push sp.
2919 // Return the number of words pushed
2920 int MacroAssembler::push(unsigned int bitset, Register stack) {
2921   int words_pushed = 0;
2922 
2923   // Scan bitset to accumulate register pairs
2924   unsigned char regs[32];
2925   int count = 0;
2926   for (int reg = 0; reg <= 30; reg++) {
2927     if (1 & bitset)
2928       regs[count++] = reg;
2929     bitset >>= 1;
2930   }
2931   regs[count++] = zr->raw_encoding();
2932   count &= ~1;  // Only push an even number of regs
2933 
2934   if (count) {
2935     stp(as_Register(regs[0]), as_Register(regs[1]),
2936        Address(pre(stack, -count * wordSize)));
2937     words_pushed += 2;
2938   }
2939   for (int i = 2; i < count; i += 2) {
2940     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2941        Address(stack, i * wordSize));
2942     words_pushed += 2;
2943   }
2944 
2945   assert(words_pushed == count, "oops, pushed != count");
2946 
2947   return count;
2948 }
2949 
2950 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2951   int words_pushed = 0;
2952 
2953   // Scan bitset to accumulate register pairs
2954   unsigned char regs[32];
2955   int count = 0;
2956   for (int reg = 0; reg <= 30; reg++) {
2957     if (1 & bitset)
2958       regs[count++] = reg;
2959     bitset >>= 1;
2960   }
2961   regs[count++] = zr->raw_encoding();
2962   count &= ~1;
2963 
2964   for (int i = 2; i < count; i += 2) {
2965     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2966        Address(stack, i * wordSize));
2967     words_pushed += 2;
2968   }
2969   if (count) {
2970     ldp(as_Register(regs[0]), as_Register(regs[1]),
2971        Address(post(stack, count * wordSize)));
2972     words_pushed += 2;
2973   }
2974 
2975   assert(words_pushed == count, "oops, pushed != count");
2976 
2977   return count;
2978 }
2979 
2980 // Push lots of registers in the bit set supplied.  Don't push sp.
2981 // Return the number of dwords pushed
2982 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2983   int words_pushed = 0;
2984   bool use_sve = false;
2985   int sve_vector_size_in_bytes = 0;
2986 
2987 #ifdef COMPILER2
2988   use_sve = Matcher::supports_scalable_vector();
2989   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2990 #endif
2991 
2992   // Scan bitset to accumulate register pairs
2993   unsigned char regs[32];
2994   int count = 0;
2995   for (int reg = 0; reg <= 31; reg++) {
2996     if (1 & bitset)
2997       regs[count++] = reg;
2998     bitset >>= 1;
2999   }
3000 
3001   if (count == 0) {
3002     return 0;
3003   }
3004 
3005   if (mode == PushPopFull) {
3006     if (use_sve && sve_vector_size_in_bytes > 16) {
3007       mode = PushPopSVE;
3008     } else {
3009       mode = PushPopNeon;
3010     }
3011   }
3012 
3013 #ifndef PRODUCT
3014   {
3015     char buffer[48];
3016     if (mode == PushPopSVE) {
3017       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3018     } else if (mode == PushPopNeon) {
3019       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3020     } else {
3021       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3022     }
3023     block_comment(buffer);
3024   }
3025 #endif
3026 
3027   if (mode == PushPopSVE) {
3028     sub(stack, stack, sve_vector_size_in_bytes * count);
3029     for (int i = 0; i < count; i++) {
3030       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3031     }
3032     return count * sve_vector_size_in_bytes / 8;
3033   }
3034 
3035   if (mode == PushPopNeon) {
3036     if (count == 1) {
3037       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3038       return 2;
3039     }
3040 
3041     bool odd = (count & 1) == 1;
3042     int push_slots = count + (odd ? 1 : 0);
3043 
3044     // Always pushing full 128 bit registers.
3045     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3046     words_pushed += 2;
3047 
3048     for (int i = 2; i + 1 < count; i += 2) {
3049       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3050       words_pushed += 2;
3051     }
3052 
3053     if (odd) {
3054       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3055       words_pushed++;
3056     }
3057 
3058     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3059     return count * 2;
3060   }
3061 
3062   if (mode == PushPopFp) {
3063     bool odd = (count & 1) == 1;
3064     int push_slots = count + (odd ? 1 : 0);
3065 
3066     if (count == 1) {
3067       // Stack pointer must be 16 bytes aligned
3068       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3069       return 1;
3070     }
3071 
3072     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3073     words_pushed += 2;
3074 
3075     for (int i = 2; i + 1 < count; i += 2) {
3076       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3077       words_pushed += 2;
3078     }
3079 
3080     if (odd) {
3081       // Stack pointer must be 16 bytes aligned
3082       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3083       words_pushed++;
3084     }
3085 
3086     assert(words_pushed == count, "oops, pushed != count");
3087 
3088     return count;
3089   }
3090 
3091   return 0;
3092 }
3093 
3094 // Return the number of dwords popped
3095 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3096   int words_pushed = 0;
3097   bool use_sve = false;
3098   int sve_vector_size_in_bytes = 0;
3099 
3100 #ifdef COMPILER2
3101   use_sve = Matcher::supports_scalable_vector();
3102   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3103 #endif
3104   // Scan bitset to accumulate register pairs
3105   unsigned char regs[32];
3106   int count = 0;
3107   for (int reg = 0; reg <= 31; reg++) {
3108     if (1 & bitset)
3109       regs[count++] = reg;
3110     bitset >>= 1;
3111   }
3112 
3113   if (count == 0) {
3114     return 0;
3115   }
3116 
3117   if (mode == PushPopFull) {
3118     if (use_sve && sve_vector_size_in_bytes > 16) {
3119       mode = PushPopSVE;
3120     } else {
3121       mode = PushPopNeon;
3122     }
3123   }
3124 
3125 #ifndef PRODUCT
3126   {
3127     char buffer[48];
3128     if (mode == PushPopSVE) {
3129       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3130     } else if (mode == PushPopNeon) {
3131       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3132     } else {
3133       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3134     }
3135     block_comment(buffer);
3136   }
3137 #endif
3138 
3139   if (mode == PushPopSVE) {
3140     for (int i = count - 1; i >= 0; i--) {
3141       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3142     }
3143     add(stack, stack, sve_vector_size_in_bytes * count);
3144     return count * sve_vector_size_in_bytes / 8;
3145   }
3146 
3147   if (mode == PushPopNeon) {
3148     if (count == 1) {
3149       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3150       return 2;
3151     }
3152 
3153     bool odd = (count & 1) == 1;
3154     int push_slots = count + (odd ? 1 : 0);
3155 
3156     if (odd) {
3157       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3158       words_pushed++;
3159     }
3160 
3161     for (int i = 2; i + 1 < count; i += 2) {
3162       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3163       words_pushed += 2;
3164     }
3165 
3166     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3167     words_pushed += 2;
3168 
3169     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3170 
3171     return count * 2;
3172   }
3173 
3174   if (mode == PushPopFp) {
3175     bool odd = (count & 1) == 1;
3176     int push_slots = count + (odd ? 1 : 0);
3177 
3178     if (count == 1) {
3179       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3180       return 1;
3181     }
3182 
3183     if (odd) {
3184       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3185       words_pushed++;
3186     }
3187 
3188     for (int i = 2; i + 1 < count; i += 2) {
3189       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3190       words_pushed += 2;
3191     }
3192 
3193     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3194     words_pushed += 2;
3195 
3196     assert(words_pushed == count, "oops, pushed != count");
3197 
3198     return count;
3199   }
3200 
3201   return 0;
3202 }
3203 
3204 // Return the number of dwords pushed
3205 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3206   bool use_sve = false;
3207   int sve_predicate_size_in_slots = 0;
3208 
3209 #ifdef COMPILER2
3210   use_sve = Matcher::supports_scalable_vector();
3211   if (use_sve) {
3212     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3213   }
3214 #endif
3215 
3216   if (!use_sve) {
3217     return 0;
3218   }
3219 
3220   unsigned char regs[PRegister::number_of_registers];
3221   int count = 0;
3222   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3223     if (1 & bitset)
3224       regs[count++] = reg;
3225     bitset >>= 1;
3226   }
3227 
3228   if (count == 0) {
3229     return 0;
3230   }
3231 
3232   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3233                                   VMRegImpl::stack_slot_size * count, 16);
3234   sub(stack, stack, total_push_bytes);
3235   for (int i = 0; i < count; i++) {
3236     sve_str(as_PRegister(regs[i]), Address(stack, i));
3237   }
3238   return total_push_bytes / 8;
3239 }
3240 
3241 // Return the number of dwords popped
3242 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3243   bool use_sve = false;
3244   int sve_predicate_size_in_slots = 0;
3245 
3246 #ifdef COMPILER2
3247   use_sve = Matcher::supports_scalable_vector();
3248   if (use_sve) {
3249     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3250   }
3251 #endif
3252 
3253   if (!use_sve) {
3254     return 0;
3255   }
3256 
3257   unsigned char regs[PRegister::number_of_registers];
3258   int count = 0;
3259   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3260     if (1 & bitset)
3261       regs[count++] = reg;
3262     bitset >>= 1;
3263   }
3264 
3265   if (count == 0) {
3266     return 0;
3267   }
3268 
3269   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3270                                  VMRegImpl::stack_slot_size * count, 16);
3271   for (int i = count - 1; i >= 0; i--) {
3272     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3273   }
3274   add(stack, stack, total_pop_bytes);
3275   return total_pop_bytes / 8;
3276 }
3277 
3278 #ifdef ASSERT
3279 void MacroAssembler::verify_heapbase(const char* msg) {
3280 #if 0
3281   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3282   assert (Universe::heap() != nullptr, "java heap should be initialized");
3283   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3284     // rheapbase is allocated as general register
3285     return;
3286   }
3287   if (CheckCompressedOops) {
3288     Label ok;
3289     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3290     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3291     br(Assembler::EQ, ok);
3292     stop(msg);
3293     bind(ok);
3294     pop(1 << rscratch1->encoding(), sp);
3295   }
3296 #endif
3297 }
3298 #endif
3299 
3300 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3301   assert_different_registers(value, tmp1, tmp2);
3302   Label done, tagged, weak_tagged;
3303 
3304   cbz(value, done);           // Use null as-is.
3305   tst(value, JNIHandles::tag_mask); // Test for tag.
3306   br(Assembler::NE, tagged);
3307 
3308   // Resolve local handle
3309   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3310   verify_oop(value);
3311   b(done);
3312 
3313   bind(tagged);
3314   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3315   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3316 
3317   // Resolve global handle
3318   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3319   verify_oop(value);
3320   b(done);
3321 
3322   bind(weak_tagged);
3323   // Resolve jweak.
3324   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3325                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3326   verify_oop(value);
3327 
3328   bind(done);
3329 }
3330 
3331 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3332   assert_different_registers(value, tmp1, tmp2);
3333   Label done;
3334 
3335   cbz(value, done);           // Use null as-is.
3336 
3337 #ifdef ASSERT
3338   {
3339     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3340     Label valid_global_tag;
3341     tbnz(value, 1, valid_global_tag); // Test for global tag
3342     stop("non global jobject using resolve_global_jobject");
3343     bind(valid_global_tag);
3344   }
3345 #endif
3346 
3347   // Resolve global handle
3348   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3349   verify_oop(value);
3350 
3351   bind(done);
3352 }
3353 
3354 void MacroAssembler::stop(const char* msg) {
3355   // Skip AOT caching C strings in scratch buffer.
3356   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3357   BLOCK_COMMENT(str);
3358   // load msg into r0 so we can access it from the signal handler
3359   // ExternalAddress enables saving and restoring via the code cache
3360   lea(c_rarg0, ExternalAddress((address) str));
3361   dcps1(0xdeae);
3362 }
3363 
3364 void MacroAssembler::unimplemented(const char* what) {
3365   const char* buf = nullptr;
3366   {
3367     ResourceMark rm;
3368     stringStream ss;
3369     ss.print("unimplemented: %s", what);
3370     buf = code_string(ss.as_string());
3371   }
3372   stop(buf);
3373 }
3374 
3375 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3376 #ifdef ASSERT
3377   Label OK;
3378   br(cc, OK);
3379   stop(msg);
3380   bind(OK);
3381 #endif
3382 }
3383 
3384 // If a constant does not fit in an immediate field, generate some
3385 // number of MOV instructions and then perform the operation.
3386 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3387                                            add_sub_imm_insn insn1,
3388                                            add_sub_reg_insn insn2,
3389                                            bool is32) {
3390   assert(Rd != zr, "Rd = zr and not setting flags?");
3391   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3392   if (fits) {
3393     (this->*insn1)(Rd, Rn, imm);
3394   } else {
3395     if (g_uabs(imm) < (1 << 24)) {
3396        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3397        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3398     } else {
3399        assert_different_registers(Rd, Rn);
3400        mov(Rd, imm);
3401        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3402     }
3403   }
3404 }
3405 
3406 // Separate vsn which sets the flags. Optimisations are more restricted
3407 // because we must set the flags correctly.
3408 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3409                                              add_sub_imm_insn insn1,
3410                                              add_sub_reg_insn insn2,
3411                                              bool is32) {
3412   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3413   if (fits) {
3414     (this->*insn1)(Rd, Rn, imm);
3415   } else {
3416     assert_different_registers(Rd, Rn);
3417     assert(Rd != zr, "overflow in immediate operand");
3418     mov(Rd, imm);
3419     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3420   }
3421 }
3422 
3423 
3424 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3425   if (increment.is_register()) {
3426     add(Rd, Rn, increment.as_register());
3427   } else {
3428     add(Rd, Rn, increment.as_constant());
3429   }
3430 }
3431 
3432 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3433   if (increment.is_register()) {
3434     addw(Rd, Rn, increment.as_register());
3435   } else {
3436     addw(Rd, Rn, increment.as_constant());
3437   }
3438 }
3439 
3440 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3441   if (decrement.is_register()) {
3442     sub(Rd, Rn, decrement.as_register());
3443   } else {
3444     sub(Rd, Rn, decrement.as_constant());
3445   }
3446 }
3447 
3448 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3449   if (decrement.is_register()) {
3450     subw(Rd, Rn, decrement.as_register());
3451   } else {
3452     subw(Rd, Rn, decrement.as_constant());
3453   }
3454 }
3455 
3456 void MacroAssembler::reinit_heapbase()
3457 {
3458   if (UseCompressedOops) {
3459     if (Universe::is_fully_initialized()) {
3460       mov(rheapbase, CompressedOops::base());
3461     } else {
3462       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3463       ldr(rheapbase, Address(rheapbase));
3464     }
3465   }
3466 }
3467 
3468 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3469 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3470 // Pass a register for the result, otherwise pass noreg.
3471 
3472 // Clobbers rscratch1
3473 void MacroAssembler::cmpxchg(Register addr, Register expected,
3474                              Register new_val,
3475                              enum operand_size size,
3476                              bool acquire, bool release,
3477                              bool weak,
3478                              Register result) {
3479   if (result == noreg)  result = rscratch1;
3480   BLOCK_COMMENT("cmpxchg {");
3481   if (UseLSE) {
3482     mov(result, expected);
3483     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3484     compare_eq(result, expected, size);
3485 #ifdef ASSERT
3486     // Poison rscratch1 which is written on !UseLSE branch
3487     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3488 #endif
3489   } else {
3490     Label retry_load, done;
3491     prfm(Address(addr), PSTL1STRM);
3492     bind(retry_load);
3493     load_exclusive(result, addr, size, acquire);
3494     compare_eq(result, expected, size);
3495     br(Assembler::NE, done);
3496     store_exclusive(rscratch1, new_val, addr, size, release);
3497     if (weak) {
3498       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3499     } else {
3500       cbnzw(rscratch1, retry_load);
3501     }
3502     bind(done);
3503   }
3504   BLOCK_COMMENT("} cmpxchg");
3505 }
3506 
3507 // A generic comparison. Only compares for equality, clobbers rscratch1.
3508 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3509   if (size == xword) {
3510     cmp(rm, rn);
3511   } else if (size == word) {
3512     cmpw(rm, rn);
3513   } else if (size == halfword) {
3514     eorw(rscratch1, rm, rn);
3515     ands(zr, rscratch1, 0xffff);
3516   } else if (size == byte) {
3517     eorw(rscratch1, rm, rn);
3518     ands(zr, rscratch1, 0xff);
3519   } else {
3520     ShouldNotReachHere();
3521   }
3522 }
3523 
3524 
3525 static bool different(Register a, RegisterOrConstant b, Register c) {
3526   if (b.is_constant())
3527     return a != c;
3528   else
3529     return a != b.as_register() && a != c && b.as_register() != c;
3530 }
3531 
3532 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3533 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3534   if (UseLSE) {                                                         \
3535     prev = prev->is_valid() ? prev : zr;                                \
3536     if (incr.is_register()) {                                           \
3537       AOP(sz, incr.as_register(), prev, addr);                          \
3538     } else {                                                            \
3539       mov(rscratch2, incr.as_constant());                               \
3540       AOP(sz, rscratch2, prev, addr);                                   \
3541     }                                                                   \
3542     return;                                                             \
3543   }                                                                     \
3544   Register result = rscratch2;                                          \
3545   if (prev->is_valid())                                                 \
3546     result = different(prev, incr, addr) ? prev : rscratch2;            \
3547                                                                         \
3548   Label retry_load;                                                     \
3549   prfm(Address(addr), PSTL1STRM);                                       \
3550   bind(retry_load);                                                     \
3551   LDXR(result, addr);                                                   \
3552   OP(rscratch1, result, incr);                                          \
3553   STXR(rscratch2, rscratch1, addr);                                     \
3554   cbnzw(rscratch2, retry_load);                                         \
3555   if (prev->is_valid() && prev != result) {                             \
3556     IOP(prev, rscratch1, incr);                                         \
3557   }                                                                     \
3558 }
3559 
3560 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3561 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3562 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3563 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3564 
3565 #undef ATOMIC_OP
3566 
3567 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3568 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3569   if (UseLSE) {                                                         \
3570     prev = prev->is_valid() ? prev : zr;                                \
3571     AOP(sz, newv, prev, addr);                                          \
3572     return;                                                             \
3573   }                                                                     \
3574   Register result = rscratch2;                                          \
3575   if (prev->is_valid())                                                 \
3576     result = different(prev, newv, addr) ? prev : rscratch2;            \
3577                                                                         \
3578   Label retry_load;                                                     \
3579   prfm(Address(addr), PSTL1STRM);                                       \
3580   bind(retry_load);                                                     \
3581   LDXR(result, addr);                                                   \
3582   STXR(rscratch1, newv, addr);                                          \
3583   cbnzw(rscratch1, retry_load);                                         \
3584   if (prev->is_valid() && prev != result)                               \
3585     mov(prev, result);                                                  \
3586 }
3587 
3588 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3589 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3590 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3591 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3592 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3593 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3594 
3595 #undef ATOMIC_XCHG
3596 
3597 #ifndef PRODUCT
3598 extern "C" void findpc(intptr_t x);
3599 #endif
3600 
3601 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3602 {
3603   // In order to get locks to work, we need to fake a in_VM state
3604   if (ShowMessageBoxOnError) {
3605     JavaThread* thread = JavaThread::current();
3606     thread->set_thread_state(_thread_in_vm);
3607 #ifndef PRODUCT
3608     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3609       ttyLocker ttyl;
3610       BytecodeCounter::print();
3611     }
3612 #endif
3613     if (os::message_box(msg, "Execution stopped, print registers?")) {
3614       ttyLocker ttyl;
3615       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3616 #ifndef PRODUCT
3617       tty->cr();
3618       findpc(pc);
3619       tty->cr();
3620 #endif
3621       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3622       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3623       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3624       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3625       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3626       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3627       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3628       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3629       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3630       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3631       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3632       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3633       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3634       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3635       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3636       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3637       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3638       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3639       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3640       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3641       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3642       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3643       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3644       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3645       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3646       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3647       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3648       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3649       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3650       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3651       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3652       BREAKPOINT;
3653     }
3654   }
3655   fatal("DEBUG MESSAGE: %s", msg);
3656 }
3657 
3658 RegSet MacroAssembler::call_clobbered_gp_registers() {
3659   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3660 #ifndef R18_RESERVED
3661   regs += r18_tls;
3662 #endif
3663   return regs;
3664 }
3665 
3666 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3667   int step = 4 * wordSize;
3668   push(call_clobbered_gp_registers() - exclude, sp);
3669   sub(sp, sp, step);
3670   mov(rscratch1, -step);
3671   // Push v0-v7, v16-v31.
3672   for (int i = 31; i>= 4; i -= 4) {
3673     if (i <= v7->encoding() || i >= v16->encoding())
3674       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3675           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3676   }
3677   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3678       as_FloatRegister(3), T1D, Address(sp));
3679 }
3680 
3681 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3682   for (int i = 0; i < 32; i += 4) {
3683     if (i <= v7->encoding() || i >= v16->encoding())
3684       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3685           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3686   }
3687 
3688   reinitialize_ptrue();
3689 
3690   pop(call_clobbered_gp_registers() - exclude, sp);
3691 }
3692 
3693 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3694                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3695   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3696   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3697     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3698     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3699       sve_str(as_FloatRegister(i), Address(sp, i));
3700     }
3701   } else {
3702     int step = (save_vectors ? 8 : 4) * wordSize;
3703     mov(rscratch1, -step);
3704     sub(sp, sp, step);
3705     for (int i = 28; i >= 4; i -= 4) {
3706       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3707           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3708     }
3709     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3710   }
3711   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3712     sub(sp, sp, total_predicate_in_bytes);
3713     for (int i = 0; i < PRegister::number_of_registers; i++) {
3714       sve_str(as_PRegister(i), Address(sp, i));
3715     }
3716   }
3717 }
3718 
3719 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3720                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3721   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3722     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3723       sve_ldr(as_PRegister(i), Address(sp, i));
3724     }
3725     add(sp, sp, total_predicate_in_bytes);
3726   }
3727   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3728     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3729       sve_ldr(as_FloatRegister(i), Address(sp, i));
3730     }
3731     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3732   } else {
3733     int step = (restore_vectors ? 8 : 4) * wordSize;
3734     for (int i = 0; i <= 28; i += 4)
3735       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3736           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3737   }
3738 
3739   // We may use predicate registers and rely on ptrue with SVE,
3740   // regardless of wide vector (> 8 bytes) used or not.
3741   if (use_sve) {
3742     reinitialize_ptrue();
3743   }
3744 
3745   // integer registers except lr & sp
3746   pop(RegSet::range(r0, r17), sp);
3747 #ifdef R18_RESERVED
3748   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3749   pop(RegSet::range(r20, r29), sp);
3750 #else
3751   pop(RegSet::range(r18_tls, r29), sp);
3752 #endif
3753 }
3754 
3755 /**
3756  * Helpers for multiply_to_len().
3757  */
3758 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3759                                      Register src1, Register src2) {
3760   adds(dest_lo, dest_lo, src1);
3761   adc(dest_hi, dest_hi, zr);
3762   adds(dest_lo, dest_lo, src2);
3763   adc(final_dest_hi, dest_hi, zr);
3764 }
3765 
3766 // Generate an address from (r + r1 extend offset).  "size" is the
3767 // size of the operand.  The result may be in rscratch2.
3768 Address MacroAssembler::offsetted_address(Register r, Register r1,
3769                                           Address::extend ext, int offset, int size) {
3770   if (offset || (ext.shift() % size != 0)) {
3771     lea(rscratch2, Address(r, r1, ext));
3772     return Address(rscratch2, offset);
3773   } else {
3774     return Address(r, r1, ext);
3775   }
3776 }
3777 
3778 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3779 {
3780   assert(offset >= 0, "spill to negative address?");
3781   // Offset reachable ?
3782   //   Not aligned - 9 bits signed offset
3783   //   Aligned - 12 bits unsigned offset shifted
3784   Register base = sp;
3785   if ((offset & (size-1)) && offset >= (1<<8)) {
3786     add(tmp, base, offset & ((1<<12)-1));
3787     base = tmp;
3788     offset &= -1u<<12;
3789   }
3790 
3791   if (offset >= (1<<12) * size) {
3792     add(tmp, base, offset & (((1<<12)-1)<<12));
3793     base = tmp;
3794     offset &= ~(((1<<12)-1)<<12);
3795   }
3796 
3797   return Address(base, offset);
3798 }
3799 
3800 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3801   assert(offset >= 0, "spill to negative address?");
3802 
3803   Register base = sp;
3804 
3805   // An immediate offset in the range 0 to 255 which is multiplied
3806   // by the current vector or predicate register size in bytes.
3807   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3808     return Address(base, offset / sve_reg_size_in_bytes);
3809   }
3810 
3811   add(tmp, base, offset);
3812   return Address(tmp);
3813 }
3814 
3815 // Checks whether offset is aligned.
3816 // Returns true if it is, else false.
3817 bool MacroAssembler::merge_alignment_check(Register base,
3818                                            size_t size,
3819                                            int64_t cur_offset,
3820                                            int64_t prev_offset) const {
3821   if (AvoidUnalignedAccesses) {
3822     if (base == sp) {
3823       // Checks whether low offset if aligned to pair of registers.
3824       int64_t pair_mask = size * 2 - 1;
3825       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3826       return (offset & pair_mask) == 0;
3827     } else { // If base is not sp, we can't guarantee the access is aligned.
3828       return false;
3829     }
3830   } else {
3831     int64_t mask = size - 1;
3832     // Load/store pair instruction only supports element size aligned offset.
3833     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3834   }
3835 }
3836 
3837 // Checks whether current and previous loads/stores can be merged.
3838 // Returns true if it can be merged, else false.
3839 bool MacroAssembler::ldst_can_merge(Register rt,
3840                                     const Address &adr,
3841                                     size_t cur_size_in_bytes,
3842                                     bool is_store) const {
3843   address prev = pc() - NativeInstruction::instruction_size;
3844   address last = code()->last_insn();
3845 
3846   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3847     return false;
3848   }
3849 
3850   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3851     return false;
3852   }
3853 
3854   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3855   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3856 
3857   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3858   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3859 
3860   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3861     return false;
3862   }
3863 
3864   int64_t max_offset = 63 * prev_size_in_bytes;
3865   int64_t min_offset = -64 * prev_size_in_bytes;
3866 
3867   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3868 
3869   // Only same base can be merged.
3870   if (adr.base() != prev_ldst->base()) {
3871     return false;
3872   }
3873 
3874   int64_t cur_offset = adr.offset();
3875   int64_t prev_offset = prev_ldst->offset();
3876   size_t diff = abs(cur_offset - prev_offset);
3877   if (diff != prev_size_in_bytes) {
3878     return false;
3879   }
3880 
3881   // Following cases can not be merged:
3882   // ldr x2, [x2, #8]
3883   // ldr x3, [x2, #16]
3884   // or:
3885   // ldr x2, [x3, #8]
3886   // ldr x2, [x3, #16]
3887   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3888   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3889     return false;
3890   }
3891 
3892   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3893   // Offset range must be in ldp/stp instruction's range.
3894   if (low_offset > max_offset || low_offset < min_offset) {
3895     return false;
3896   }
3897 
3898   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3899     return true;
3900   }
3901 
3902   return false;
3903 }
3904 
3905 // Merge current load/store with previous load/store into ldp/stp.
3906 void MacroAssembler::merge_ldst(Register rt,
3907                                 const Address &adr,
3908                                 size_t cur_size_in_bytes,
3909                                 bool is_store) {
3910 
3911   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3912 
3913   Register rt_low, rt_high;
3914   address prev = pc() - NativeInstruction::instruction_size;
3915   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3916 
3917   int64_t offset;
3918 
3919   if (adr.offset() < prev_ldst->offset()) {
3920     offset = adr.offset();
3921     rt_low = rt;
3922     rt_high = prev_ldst->target();
3923   } else {
3924     offset = prev_ldst->offset();
3925     rt_low = prev_ldst->target();
3926     rt_high = rt;
3927   }
3928 
3929   Address adr_p = Address(prev_ldst->base(), offset);
3930   // Overwrite previous generated binary.
3931   code_section()->set_end(prev);
3932 
3933   const size_t sz = prev_ldst->size_in_bytes();
3934   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3935   if (!is_store) {
3936     BLOCK_COMMENT("merged ldr pair");
3937     if (sz == 8) {
3938       ldp(rt_low, rt_high, adr_p);
3939     } else {
3940       ldpw(rt_low, rt_high, adr_p);
3941     }
3942   } else {
3943     BLOCK_COMMENT("merged str pair");
3944     if (sz == 8) {
3945       stp(rt_low, rt_high, adr_p);
3946     } else {
3947       stpw(rt_low, rt_high, adr_p);
3948     }
3949   }
3950 }
3951 
3952 /**
3953  * Multiply 64 bit by 64 bit first loop.
3954  */
3955 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3956                                            Register y, Register y_idx, Register z,
3957                                            Register carry, Register product,
3958                                            Register idx, Register kdx) {
3959   //
3960   //  jlong carry, x[], y[], z[];
3961   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3962   //    huge_128 product = y[idx] * x[xstart] + carry;
3963   //    z[kdx] = (jlong)product;
3964   //    carry  = (jlong)(product >>> 64);
3965   //  }
3966   //  z[xstart] = carry;
3967   //
3968 
3969   Label L_first_loop, L_first_loop_exit;
3970   Label L_one_x, L_one_y, L_multiply;
3971 
3972   subsw(xstart, xstart, 1);
3973   br(Assembler::MI, L_one_x);
3974 
3975   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3976   ldr(x_xstart, Address(rscratch1));
3977   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3978 
3979   bind(L_first_loop);
3980   subsw(idx, idx, 1);
3981   br(Assembler::MI, L_first_loop_exit);
3982   subsw(idx, idx, 1);
3983   br(Assembler::MI, L_one_y);
3984   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3985   ldr(y_idx, Address(rscratch1));
3986   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
3987   bind(L_multiply);
3988 
3989   // AArch64 has a multiply-accumulate instruction that we can't use
3990   // here because it has no way to process carries, so we have to use
3991   // separate add and adc instructions.  Bah.
3992   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
3993   mul(product, x_xstart, y_idx);
3994   adds(product, product, carry);
3995   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
3996 
3997   subw(kdx, kdx, 2);
3998   ror(product, product, 32); // back to big-endian
3999   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4000 
4001   b(L_first_loop);
4002 
4003   bind(L_one_y);
4004   ldrw(y_idx, Address(y,  0));
4005   b(L_multiply);
4006 
4007   bind(L_one_x);
4008   ldrw(x_xstart, Address(x,  0));
4009   b(L_first_loop);
4010 
4011   bind(L_first_loop_exit);
4012 }
4013 
4014 /**
4015  * Multiply 128 bit by 128. Unrolled inner loop.
4016  *
4017  */
4018 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4019                                              Register carry, Register carry2,
4020                                              Register idx, Register jdx,
4021                                              Register yz_idx1, Register yz_idx2,
4022                                              Register tmp, Register tmp3, Register tmp4,
4023                                              Register tmp6, Register product_hi) {
4024 
4025   //   jlong carry, x[], y[], z[];
4026   //   int kdx = ystart+1;
4027   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4028   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4029   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4030   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4031   //     carry  = (jlong)(tmp4 >>> 64);
4032   //     z[kdx+idx+1] = (jlong)tmp3;
4033   //     z[kdx+idx] = (jlong)tmp4;
4034   //   }
4035   //   idx += 2;
4036   //   if (idx > 0) {
4037   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4038   //     z[kdx+idx] = (jlong)yz_idx1;
4039   //     carry  = (jlong)(yz_idx1 >>> 64);
4040   //   }
4041   //
4042 
4043   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4044 
4045   lsrw(jdx, idx, 2);
4046 
4047   bind(L_third_loop);
4048 
4049   subsw(jdx, jdx, 1);
4050   br(Assembler::MI, L_third_loop_exit);
4051   subw(idx, idx, 4);
4052 
4053   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4054 
4055   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4056 
4057   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4058 
4059   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4060   ror(yz_idx2, yz_idx2, 32);
4061 
4062   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4063 
4064   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4065   umulh(tmp4, product_hi, yz_idx1);
4066 
4067   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4068   ror(rscratch2, rscratch2, 32);
4069 
4070   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4071   umulh(carry2, product_hi, yz_idx2);
4072 
4073   // propagate sum of both multiplications into carry:tmp4:tmp3
4074   adds(tmp3, tmp3, carry);
4075   adc(tmp4, tmp4, zr);
4076   adds(tmp3, tmp3, rscratch1);
4077   adcs(tmp4, tmp4, tmp);
4078   adc(carry, carry2, zr);
4079   adds(tmp4, tmp4, rscratch2);
4080   adc(carry, carry, zr);
4081 
4082   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4083   ror(tmp4, tmp4, 32);
4084   stp(tmp4, tmp3, Address(tmp6, 0));
4085 
4086   b(L_third_loop);
4087   bind (L_third_loop_exit);
4088 
4089   andw (idx, idx, 0x3);
4090   cbz(idx, L_post_third_loop_done);
4091 
4092   Label L_check_1;
4093   subsw(idx, idx, 2);
4094   br(Assembler::MI, L_check_1);
4095 
4096   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4097   ldr(yz_idx1, Address(rscratch1, 0));
4098   ror(yz_idx1, yz_idx1, 32);
4099   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4100   umulh(tmp4, product_hi, yz_idx1);
4101   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4102   ldr(yz_idx2, Address(rscratch1, 0));
4103   ror(yz_idx2, yz_idx2, 32);
4104 
4105   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4106 
4107   ror(tmp3, tmp3, 32);
4108   str(tmp3, Address(rscratch1, 0));
4109 
4110   bind (L_check_1);
4111 
4112   andw (idx, idx, 0x1);
4113   subsw(idx, idx, 1);
4114   br(Assembler::MI, L_post_third_loop_done);
4115   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4116   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4117   umulh(carry2, tmp4, product_hi);
4118   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4119 
4120   add2_with_carry(carry2, tmp3, tmp4, carry);
4121 
4122   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4123   extr(carry, carry2, tmp3, 32);
4124 
4125   bind(L_post_third_loop_done);
4126 }
4127 
4128 /**
4129  * Code for BigInteger::multiplyToLen() intrinsic.
4130  *
4131  * r0: x
4132  * r1: xlen
4133  * r2: y
4134  * r3: ylen
4135  * r4:  z
4136  * r5: tmp0
4137  * r10: tmp1
4138  * r11: tmp2
4139  * r12: tmp3
4140  * r13: tmp4
4141  * r14: tmp5
4142  * r15: tmp6
4143  * r16: tmp7
4144  *
4145  */
4146 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4147                                      Register z, Register tmp0,
4148                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4149                                      Register tmp5, Register tmp6, Register product_hi) {
4150 
4151   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4152 
4153   const Register idx = tmp1;
4154   const Register kdx = tmp2;
4155   const Register xstart = tmp3;
4156 
4157   const Register y_idx = tmp4;
4158   const Register carry = tmp5;
4159   const Register product  = xlen;
4160   const Register x_xstart = tmp0;
4161 
4162   // First Loop.
4163   //
4164   //  final static long LONG_MASK = 0xffffffffL;
4165   //  int xstart = xlen - 1;
4166   //  int ystart = ylen - 1;
4167   //  long carry = 0;
4168   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4169   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4170   //    z[kdx] = (int)product;
4171   //    carry = product >>> 32;
4172   //  }
4173   //  z[xstart] = (int)carry;
4174   //
4175 
4176   movw(idx, ylen);       // idx = ylen;
4177   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4178   mov(carry, zr);        // carry = 0;
4179 
4180   Label L_done;
4181 
4182   movw(xstart, xlen);
4183   subsw(xstart, xstart, 1);
4184   br(Assembler::MI, L_done);
4185 
4186   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4187 
4188   Label L_second_loop;
4189   cbzw(kdx, L_second_loop);
4190 
4191   Label L_carry;
4192   subw(kdx, kdx, 1);
4193   cbzw(kdx, L_carry);
4194 
4195   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4196   lsr(carry, carry, 32);
4197   subw(kdx, kdx, 1);
4198 
4199   bind(L_carry);
4200   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4201 
4202   // Second and third (nested) loops.
4203   //
4204   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4205   //   carry = 0;
4206   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4207   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4208   //                    (z[k] & LONG_MASK) + carry;
4209   //     z[k] = (int)product;
4210   //     carry = product >>> 32;
4211   //   }
4212   //   z[i] = (int)carry;
4213   // }
4214   //
4215   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4216 
4217   const Register jdx = tmp1;
4218 
4219   bind(L_second_loop);
4220   mov(carry, zr);                // carry = 0;
4221   movw(jdx, ylen);               // j = ystart+1
4222 
4223   subsw(xstart, xstart, 1);      // i = xstart-1;
4224   br(Assembler::MI, L_done);
4225 
4226   str(z, Address(pre(sp, -4 * wordSize)));
4227 
4228   Label L_last_x;
4229   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4230   subsw(xstart, xstart, 1);       // i = xstart-1;
4231   br(Assembler::MI, L_last_x);
4232 
4233   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4234   ldr(product_hi, Address(rscratch1));
4235   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4236 
4237   Label L_third_loop_prologue;
4238   bind(L_third_loop_prologue);
4239 
4240   str(ylen, Address(sp, wordSize));
4241   stp(x, xstart, Address(sp, 2 * wordSize));
4242   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4243                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4244   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4245   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4246 
4247   addw(tmp3, xlen, 1);
4248   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4249   subsw(tmp3, tmp3, 1);
4250   br(Assembler::MI, L_done);
4251 
4252   lsr(carry, carry, 32);
4253   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4254   b(L_second_loop);
4255 
4256   // Next infrequent code is moved outside loops.
4257   bind(L_last_x);
4258   ldrw(product_hi, Address(x,  0));
4259   b(L_third_loop_prologue);
4260 
4261   bind(L_done);
4262 }
4263 
4264 // Code for BigInteger::mulAdd intrinsic
4265 // out     = r0
4266 // in      = r1
4267 // offset  = r2  (already out.length-offset)
4268 // len     = r3
4269 // k       = r4
4270 //
4271 // pseudo code from java implementation:
4272 // carry = 0;
4273 // offset = out.length-offset - 1;
4274 // for (int j=len-1; j >= 0; j--) {
4275 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4276 //     out[offset--] = (int)product;
4277 //     carry = product >>> 32;
4278 // }
4279 // return (int)carry;
4280 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4281       Register len, Register k) {
4282     Label LOOP, END;
4283     // pre-loop
4284     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4285     csel(out, zr, out, Assembler::EQ);
4286     br(Assembler::EQ, END);
4287     add(in, in, len, LSL, 2); // in[j+1] address
4288     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4289     mov(out, zr); // used to keep carry now
4290     BIND(LOOP);
4291     ldrw(rscratch1, Address(pre(in, -4)));
4292     madd(rscratch1, rscratch1, k, out);
4293     ldrw(rscratch2, Address(pre(offset, -4)));
4294     add(rscratch1, rscratch1, rscratch2);
4295     strw(rscratch1, Address(offset));
4296     lsr(out, rscratch1, 32);
4297     subs(len, len, 1);
4298     br(Assembler::NE, LOOP);
4299     BIND(END);
4300 }
4301 
4302 /**
4303  * Emits code to update CRC-32 with a byte value according to constants in table
4304  *
4305  * @param [in,out]crc   Register containing the crc.
4306  * @param [in]val       Register containing the byte to fold into the CRC.
4307  * @param [in]table     Register containing the table of crc constants.
4308  *
4309  * uint32_t crc;
4310  * val = crc_table[(val ^ crc) & 0xFF];
4311  * crc = val ^ (crc >> 8);
4312  *
4313  */
4314 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4315   eor(val, val, crc);
4316   andr(val, val, 0xff);
4317   ldrw(val, Address(table, val, Address::lsl(2)));
4318   eor(crc, val, crc, Assembler::LSR, 8);
4319 }
4320 
4321 /**
4322  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4323  *
4324  * @param [in,out]crc   Register containing the crc.
4325  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4326  * @param [in]table0    Register containing table 0 of crc constants.
4327  * @param [in]table1    Register containing table 1 of crc constants.
4328  * @param [in]table2    Register containing table 2 of crc constants.
4329  * @param [in]table3    Register containing table 3 of crc constants.
4330  *
4331  * uint32_t crc;
4332  *   v = crc ^ v
4333  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4334  *
4335  */
4336 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4337         Register table0, Register table1, Register table2, Register table3,
4338         bool upper) {
4339   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4340   uxtb(tmp, v);
4341   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4342   ubfx(tmp, v, 8, 8);
4343   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4344   eor(crc, crc, tmp);
4345   ubfx(tmp, v, 16, 8);
4346   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4347   eor(crc, crc, tmp);
4348   ubfx(tmp, v, 24, 8);
4349   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4350   eor(crc, crc, tmp);
4351 }
4352 
4353 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4354         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4355     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4356     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4357 
4358     subs(tmp0, len, 384);
4359     mvnw(crc, crc);
4360     br(Assembler::GE, CRC_by128_pre);
4361   BIND(CRC_less128);
4362     subs(len, len, 32);
4363     br(Assembler::GE, CRC_by32_loop);
4364   BIND(CRC_less32);
4365     adds(len, len, 32 - 4);
4366     br(Assembler::GE, CRC_by4_loop);
4367     adds(len, len, 4);
4368     br(Assembler::GT, CRC_by1_loop);
4369     b(L_exit);
4370 
4371   BIND(CRC_by32_loop);
4372     ldp(tmp0, tmp1, Address(buf));
4373     crc32x(crc, crc, tmp0);
4374     ldp(tmp2, tmp3, Address(buf, 16));
4375     crc32x(crc, crc, tmp1);
4376     add(buf, buf, 32);
4377     crc32x(crc, crc, tmp2);
4378     subs(len, len, 32);
4379     crc32x(crc, crc, tmp3);
4380     br(Assembler::GE, CRC_by32_loop);
4381     cmn(len, (u1)32);
4382     br(Assembler::NE, CRC_less32);
4383     b(L_exit);
4384 
4385   BIND(CRC_by4_loop);
4386     ldrw(tmp0, Address(post(buf, 4)));
4387     subs(len, len, 4);
4388     crc32w(crc, crc, tmp0);
4389     br(Assembler::GE, CRC_by4_loop);
4390     adds(len, len, 4);
4391     br(Assembler::LE, L_exit);
4392   BIND(CRC_by1_loop);
4393     ldrb(tmp0, Address(post(buf, 1)));
4394     subs(len, len, 1);
4395     crc32b(crc, crc, tmp0);
4396     br(Assembler::GT, CRC_by1_loop);
4397     b(L_exit);
4398 
4399   BIND(CRC_by128_pre);
4400     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4401       4*256*sizeof(juint) + 8*sizeof(juint));
4402     mov(crc, 0);
4403     crc32x(crc, crc, tmp0);
4404     crc32x(crc, crc, tmp1);
4405 
4406     cbnz(len, CRC_less128);
4407 
4408   BIND(L_exit);
4409     mvnw(crc, crc);
4410 }
4411 
4412 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4413         Register len, Register tmp0, Register tmp1, Register tmp2,
4414         Register tmp3) {
4415     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4416     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4417 
4418     mvnw(crc, crc);
4419 
4420     subs(len, len, 128);
4421     br(Assembler::GE, CRC_by64_pre);
4422   BIND(CRC_less64);
4423     adds(len, len, 128-32);
4424     br(Assembler::GE, CRC_by32_loop);
4425   BIND(CRC_less32);
4426     adds(len, len, 32-4);
4427     br(Assembler::GE, CRC_by4_loop);
4428     adds(len, len, 4);
4429     br(Assembler::GT, CRC_by1_loop);
4430     b(L_exit);
4431 
4432   BIND(CRC_by32_loop);
4433     ldp(tmp0, tmp1, Address(post(buf, 16)));
4434     subs(len, len, 32);
4435     crc32x(crc, crc, tmp0);
4436     ldr(tmp2, Address(post(buf, 8)));
4437     crc32x(crc, crc, tmp1);
4438     ldr(tmp3, Address(post(buf, 8)));
4439     crc32x(crc, crc, tmp2);
4440     crc32x(crc, crc, tmp3);
4441     br(Assembler::GE, CRC_by32_loop);
4442     cmn(len, (u1)32);
4443     br(Assembler::NE, CRC_less32);
4444     b(L_exit);
4445 
4446   BIND(CRC_by4_loop);
4447     ldrw(tmp0, Address(post(buf, 4)));
4448     subs(len, len, 4);
4449     crc32w(crc, crc, tmp0);
4450     br(Assembler::GE, CRC_by4_loop);
4451     adds(len, len, 4);
4452     br(Assembler::LE, L_exit);
4453   BIND(CRC_by1_loop);
4454     ldrb(tmp0, Address(post(buf, 1)));
4455     subs(len, len, 1);
4456     crc32b(crc, crc, tmp0);
4457     br(Assembler::GT, CRC_by1_loop);
4458     b(L_exit);
4459 
4460   BIND(CRC_by64_pre);
4461     sub(buf, buf, 8);
4462     ldp(tmp0, tmp1, Address(buf, 8));
4463     crc32x(crc, crc, tmp0);
4464     ldr(tmp2, Address(buf, 24));
4465     crc32x(crc, crc, tmp1);
4466     ldr(tmp3, Address(buf, 32));
4467     crc32x(crc, crc, tmp2);
4468     ldr(tmp0, Address(buf, 40));
4469     crc32x(crc, crc, tmp3);
4470     ldr(tmp1, Address(buf, 48));
4471     crc32x(crc, crc, tmp0);
4472     ldr(tmp2, Address(buf, 56));
4473     crc32x(crc, crc, tmp1);
4474     ldr(tmp3, Address(pre(buf, 64)));
4475 
4476     b(CRC_by64_loop);
4477 
4478     align(CodeEntryAlignment);
4479   BIND(CRC_by64_loop);
4480     subs(len, len, 64);
4481     crc32x(crc, crc, tmp2);
4482     ldr(tmp0, Address(buf, 8));
4483     crc32x(crc, crc, tmp3);
4484     ldr(tmp1, Address(buf, 16));
4485     crc32x(crc, crc, tmp0);
4486     ldr(tmp2, Address(buf, 24));
4487     crc32x(crc, crc, tmp1);
4488     ldr(tmp3, Address(buf, 32));
4489     crc32x(crc, crc, tmp2);
4490     ldr(tmp0, Address(buf, 40));
4491     crc32x(crc, crc, tmp3);
4492     ldr(tmp1, Address(buf, 48));
4493     crc32x(crc, crc, tmp0);
4494     ldr(tmp2, Address(buf, 56));
4495     crc32x(crc, crc, tmp1);
4496     ldr(tmp3, Address(pre(buf, 64)));
4497     br(Assembler::GE, CRC_by64_loop);
4498 
4499     // post-loop
4500     crc32x(crc, crc, tmp2);
4501     crc32x(crc, crc, tmp3);
4502 
4503     sub(len, len, 64);
4504     add(buf, buf, 8);
4505     cmn(len, (u1)128);
4506     br(Assembler::NE, CRC_less64);
4507   BIND(L_exit);
4508     mvnw(crc, crc);
4509 }
4510 
4511 /**
4512  * @param crc   register containing existing CRC (32-bit)
4513  * @param buf   register pointing to input byte buffer (byte*)
4514  * @param len   register containing number of bytes
4515  * @param table register that will contain address of CRC table
4516  * @param tmp   scratch register
4517  */
4518 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4519         Register table0, Register table1, Register table2, Register table3,
4520         Register tmp, Register tmp2, Register tmp3) {
4521   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4522 
4523   if (UseCryptoPmullForCRC32) {
4524       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4525       return;
4526   }
4527 
4528   if (UseCRC32) {
4529       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4530       return;
4531   }
4532 
4533     mvnw(crc, crc);
4534 
4535     {
4536       uint64_t offset;
4537       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4538       add(table0, table0, offset);
4539     }
4540     add(table1, table0, 1*256*sizeof(juint));
4541     add(table2, table0, 2*256*sizeof(juint));
4542     add(table3, table0, 3*256*sizeof(juint));
4543 
4544     { // Neon code start
4545       cmp(len, (u1)64);
4546       br(Assembler::LT, L_by16);
4547       eor(v16, T16B, v16, v16);
4548 
4549     Label L_fold;
4550 
4551       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4552 
4553       ld1(v0, v1, T2D, post(buf, 32));
4554       ld1r(v4, T2D, post(tmp, 8));
4555       ld1r(v5, T2D, post(tmp, 8));
4556       ld1r(v6, T2D, post(tmp, 8));
4557       ld1r(v7, T2D, post(tmp, 8));
4558       mov(v16, S, 0, crc);
4559 
4560       eor(v0, T16B, v0, v16);
4561       sub(len, len, 64);
4562 
4563     BIND(L_fold);
4564       pmull(v22, T8H, v0, v5, T8B);
4565       pmull(v20, T8H, v0, v7, T8B);
4566       pmull(v23, T8H, v0, v4, T8B);
4567       pmull(v21, T8H, v0, v6, T8B);
4568 
4569       pmull2(v18, T8H, v0, v5, T16B);
4570       pmull2(v16, T8H, v0, v7, T16B);
4571       pmull2(v19, T8H, v0, v4, T16B);
4572       pmull2(v17, T8H, v0, v6, T16B);
4573 
4574       uzp1(v24, T8H, v20, v22);
4575       uzp2(v25, T8H, v20, v22);
4576       eor(v20, T16B, v24, v25);
4577 
4578       uzp1(v26, T8H, v16, v18);
4579       uzp2(v27, T8H, v16, v18);
4580       eor(v16, T16B, v26, v27);
4581 
4582       ushll2(v22, T4S, v20, T8H, 8);
4583       ushll(v20, T4S, v20, T4H, 8);
4584 
4585       ushll2(v18, T4S, v16, T8H, 8);
4586       ushll(v16, T4S, v16, T4H, 8);
4587 
4588       eor(v22, T16B, v23, v22);
4589       eor(v18, T16B, v19, v18);
4590       eor(v20, T16B, v21, v20);
4591       eor(v16, T16B, v17, v16);
4592 
4593       uzp1(v17, T2D, v16, v20);
4594       uzp2(v21, T2D, v16, v20);
4595       eor(v17, T16B, v17, v21);
4596 
4597       ushll2(v20, T2D, v17, T4S, 16);
4598       ushll(v16, T2D, v17, T2S, 16);
4599 
4600       eor(v20, T16B, v20, v22);
4601       eor(v16, T16B, v16, v18);
4602 
4603       uzp1(v17, T2D, v20, v16);
4604       uzp2(v21, T2D, v20, v16);
4605       eor(v28, T16B, v17, v21);
4606 
4607       pmull(v22, T8H, v1, v5, T8B);
4608       pmull(v20, T8H, v1, v7, T8B);
4609       pmull(v23, T8H, v1, v4, T8B);
4610       pmull(v21, T8H, v1, v6, T8B);
4611 
4612       pmull2(v18, T8H, v1, v5, T16B);
4613       pmull2(v16, T8H, v1, v7, T16B);
4614       pmull2(v19, T8H, v1, v4, T16B);
4615       pmull2(v17, T8H, v1, v6, T16B);
4616 
4617       ld1(v0, v1, T2D, post(buf, 32));
4618 
4619       uzp1(v24, T8H, v20, v22);
4620       uzp2(v25, T8H, v20, v22);
4621       eor(v20, T16B, v24, v25);
4622 
4623       uzp1(v26, T8H, v16, v18);
4624       uzp2(v27, T8H, v16, v18);
4625       eor(v16, T16B, v26, v27);
4626 
4627       ushll2(v22, T4S, v20, T8H, 8);
4628       ushll(v20, T4S, v20, T4H, 8);
4629 
4630       ushll2(v18, T4S, v16, T8H, 8);
4631       ushll(v16, T4S, v16, T4H, 8);
4632 
4633       eor(v22, T16B, v23, v22);
4634       eor(v18, T16B, v19, v18);
4635       eor(v20, T16B, v21, v20);
4636       eor(v16, T16B, v17, v16);
4637 
4638       uzp1(v17, T2D, v16, v20);
4639       uzp2(v21, T2D, v16, v20);
4640       eor(v16, T16B, v17, v21);
4641 
4642       ushll2(v20, T2D, v16, T4S, 16);
4643       ushll(v16, T2D, v16, T2S, 16);
4644 
4645       eor(v20, T16B, v22, v20);
4646       eor(v16, T16B, v16, v18);
4647 
4648       uzp1(v17, T2D, v20, v16);
4649       uzp2(v21, T2D, v20, v16);
4650       eor(v20, T16B, v17, v21);
4651 
4652       shl(v16, T2D, v28, 1);
4653       shl(v17, T2D, v20, 1);
4654 
4655       eor(v0, T16B, v0, v16);
4656       eor(v1, T16B, v1, v17);
4657 
4658       subs(len, len, 32);
4659       br(Assembler::GE, L_fold);
4660 
4661       mov(crc, 0);
4662       mov(tmp, v0, D, 0);
4663       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4664       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4665       mov(tmp, v0, D, 1);
4666       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4667       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4668       mov(tmp, v1, D, 0);
4669       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4670       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4671       mov(tmp, v1, D, 1);
4672       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4673       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4674 
4675       add(len, len, 32);
4676     } // Neon code end
4677 
4678   BIND(L_by16);
4679     subs(len, len, 16);
4680     br(Assembler::GE, L_by16_loop);
4681     adds(len, len, 16-4);
4682     br(Assembler::GE, L_by4_loop);
4683     adds(len, len, 4);
4684     br(Assembler::GT, L_by1_loop);
4685     b(L_exit);
4686 
4687   BIND(L_by4_loop);
4688     ldrw(tmp, Address(post(buf, 4)));
4689     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4690     subs(len, len, 4);
4691     br(Assembler::GE, L_by4_loop);
4692     adds(len, len, 4);
4693     br(Assembler::LE, L_exit);
4694   BIND(L_by1_loop);
4695     subs(len, len, 1);
4696     ldrb(tmp, Address(post(buf, 1)));
4697     update_byte_crc32(crc, tmp, table0);
4698     br(Assembler::GT, L_by1_loop);
4699     b(L_exit);
4700 
4701     align(CodeEntryAlignment);
4702   BIND(L_by16_loop);
4703     subs(len, len, 16);
4704     ldp(tmp, tmp3, Address(post(buf, 16)));
4705     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4706     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4707     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4708     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4709     br(Assembler::GE, L_by16_loop);
4710     adds(len, len, 16-4);
4711     br(Assembler::GE, L_by4_loop);
4712     adds(len, len, 4);
4713     br(Assembler::GT, L_by1_loop);
4714   BIND(L_exit);
4715     mvnw(crc, crc);
4716 }
4717 
4718 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4719         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4720     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4721     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4722 
4723     subs(tmp0, len, 384);
4724     br(Assembler::GE, CRC_by128_pre);
4725   BIND(CRC_less128);
4726     subs(len, len, 32);
4727     br(Assembler::GE, CRC_by32_loop);
4728   BIND(CRC_less32);
4729     adds(len, len, 32 - 4);
4730     br(Assembler::GE, CRC_by4_loop);
4731     adds(len, len, 4);
4732     br(Assembler::GT, CRC_by1_loop);
4733     b(L_exit);
4734 
4735   BIND(CRC_by32_loop);
4736     ldp(tmp0, tmp1, Address(buf));
4737     crc32cx(crc, crc, tmp0);
4738     ldr(tmp2, Address(buf, 16));
4739     crc32cx(crc, crc, tmp1);
4740     ldr(tmp3, Address(buf, 24));
4741     crc32cx(crc, crc, tmp2);
4742     add(buf, buf, 32);
4743     subs(len, len, 32);
4744     crc32cx(crc, crc, tmp3);
4745     br(Assembler::GE, CRC_by32_loop);
4746     cmn(len, (u1)32);
4747     br(Assembler::NE, CRC_less32);
4748     b(L_exit);
4749 
4750   BIND(CRC_by4_loop);
4751     ldrw(tmp0, Address(post(buf, 4)));
4752     subs(len, len, 4);
4753     crc32cw(crc, crc, tmp0);
4754     br(Assembler::GE, CRC_by4_loop);
4755     adds(len, len, 4);
4756     br(Assembler::LE, L_exit);
4757   BIND(CRC_by1_loop);
4758     ldrb(tmp0, Address(post(buf, 1)));
4759     subs(len, len, 1);
4760     crc32cb(crc, crc, tmp0);
4761     br(Assembler::GT, CRC_by1_loop);
4762     b(L_exit);
4763 
4764   BIND(CRC_by128_pre);
4765     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4766       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4767     mov(crc, 0);
4768     crc32cx(crc, crc, tmp0);
4769     crc32cx(crc, crc, tmp1);
4770 
4771     cbnz(len, CRC_less128);
4772 
4773   BIND(L_exit);
4774 }
4775 
4776 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4777         Register len, Register tmp0, Register tmp1, Register tmp2,
4778         Register tmp3) {
4779     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4780     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4781 
4782     subs(len, len, 128);
4783     br(Assembler::GE, CRC_by64_pre);
4784   BIND(CRC_less64);
4785     adds(len, len, 128-32);
4786     br(Assembler::GE, CRC_by32_loop);
4787   BIND(CRC_less32);
4788     adds(len, len, 32-4);
4789     br(Assembler::GE, CRC_by4_loop);
4790     adds(len, len, 4);
4791     br(Assembler::GT, CRC_by1_loop);
4792     b(L_exit);
4793 
4794   BIND(CRC_by32_loop);
4795     ldp(tmp0, tmp1, Address(post(buf, 16)));
4796     subs(len, len, 32);
4797     crc32cx(crc, crc, tmp0);
4798     ldr(tmp2, Address(post(buf, 8)));
4799     crc32cx(crc, crc, tmp1);
4800     ldr(tmp3, Address(post(buf, 8)));
4801     crc32cx(crc, crc, tmp2);
4802     crc32cx(crc, crc, tmp3);
4803     br(Assembler::GE, CRC_by32_loop);
4804     cmn(len, (u1)32);
4805     br(Assembler::NE, CRC_less32);
4806     b(L_exit);
4807 
4808   BIND(CRC_by4_loop);
4809     ldrw(tmp0, Address(post(buf, 4)));
4810     subs(len, len, 4);
4811     crc32cw(crc, crc, tmp0);
4812     br(Assembler::GE, CRC_by4_loop);
4813     adds(len, len, 4);
4814     br(Assembler::LE, L_exit);
4815   BIND(CRC_by1_loop);
4816     ldrb(tmp0, Address(post(buf, 1)));
4817     subs(len, len, 1);
4818     crc32cb(crc, crc, tmp0);
4819     br(Assembler::GT, CRC_by1_loop);
4820     b(L_exit);
4821 
4822   BIND(CRC_by64_pre);
4823     sub(buf, buf, 8);
4824     ldp(tmp0, tmp1, Address(buf, 8));
4825     crc32cx(crc, crc, tmp0);
4826     ldr(tmp2, Address(buf, 24));
4827     crc32cx(crc, crc, tmp1);
4828     ldr(tmp3, Address(buf, 32));
4829     crc32cx(crc, crc, tmp2);
4830     ldr(tmp0, Address(buf, 40));
4831     crc32cx(crc, crc, tmp3);
4832     ldr(tmp1, Address(buf, 48));
4833     crc32cx(crc, crc, tmp0);
4834     ldr(tmp2, Address(buf, 56));
4835     crc32cx(crc, crc, tmp1);
4836     ldr(tmp3, Address(pre(buf, 64)));
4837 
4838     b(CRC_by64_loop);
4839 
4840     align(CodeEntryAlignment);
4841   BIND(CRC_by64_loop);
4842     subs(len, len, 64);
4843     crc32cx(crc, crc, tmp2);
4844     ldr(tmp0, Address(buf, 8));
4845     crc32cx(crc, crc, tmp3);
4846     ldr(tmp1, Address(buf, 16));
4847     crc32cx(crc, crc, tmp0);
4848     ldr(tmp2, Address(buf, 24));
4849     crc32cx(crc, crc, tmp1);
4850     ldr(tmp3, Address(buf, 32));
4851     crc32cx(crc, crc, tmp2);
4852     ldr(tmp0, Address(buf, 40));
4853     crc32cx(crc, crc, tmp3);
4854     ldr(tmp1, Address(buf, 48));
4855     crc32cx(crc, crc, tmp0);
4856     ldr(tmp2, Address(buf, 56));
4857     crc32cx(crc, crc, tmp1);
4858     ldr(tmp3, Address(pre(buf, 64)));
4859     br(Assembler::GE, CRC_by64_loop);
4860 
4861     // post-loop
4862     crc32cx(crc, crc, tmp2);
4863     crc32cx(crc, crc, tmp3);
4864 
4865     sub(len, len, 64);
4866     add(buf, buf, 8);
4867     cmn(len, (u1)128);
4868     br(Assembler::NE, CRC_less64);
4869   BIND(L_exit);
4870 }
4871 
4872 /**
4873  * @param crc   register containing existing CRC (32-bit)
4874  * @param buf   register pointing to input byte buffer (byte*)
4875  * @param len   register containing number of bytes
4876  * @param table register that will contain address of CRC table
4877  * @param tmp   scratch register
4878  */
4879 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4880         Register table0, Register table1, Register table2, Register table3,
4881         Register tmp, Register tmp2, Register tmp3) {
4882   if (UseCryptoPmullForCRC32) {
4883     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4884   } else {
4885     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4886   }
4887 }
4888 
4889 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4890         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4891     Label CRC_by128_loop;
4892     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4893 
4894     sub(len, len, 256);
4895     Register table = tmp0;
4896     {
4897       uint64_t offset;
4898       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4899       add(table, table, offset);
4900     }
4901     add(table, table, table_offset);
4902 
4903     // Registers v0..v7 are used as data registers.
4904     // Registers v16..v31 are used as tmp registers.
4905     sub(buf, buf, 0x10);
4906     ldrq(v0, Address(buf, 0x10));
4907     ldrq(v1, Address(buf, 0x20));
4908     ldrq(v2, Address(buf, 0x30));
4909     ldrq(v3, Address(buf, 0x40));
4910     ldrq(v4, Address(buf, 0x50));
4911     ldrq(v5, Address(buf, 0x60));
4912     ldrq(v6, Address(buf, 0x70));
4913     ldrq(v7, Address(pre(buf, 0x80)));
4914 
4915     movi(v31, T4S, 0);
4916     mov(v31, S, 0, crc);
4917     eor(v0, T16B, v0, v31);
4918 
4919     // Register v16 contains constants from the crc table.
4920     ldrq(v16, Address(table));
4921     b(CRC_by128_loop);
4922 
4923     align(OptoLoopAlignment);
4924   BIND(CRC_by128_loop);
4925     pmull (v17,  T1Q, v0, v16, T1D);
4926     pmull2(v18, T1Q, v0, v16, T2D);
4927     ldrq(v0, Address(buf, 0x10));
4928     eor3(v0, T16B, v17,  v18, v0);
4929 
4930     pmull (v19, T1Q, v1, v16, T1D);
4931     pmull2(v20, T1Q, v1, v16, T2D);
4932     ldrq(v1, Address(buf, 0x20));
4933     eor3(v1, T16B, v19, v20, v1);
4934 
4935     pmull (v21, T1Q, v2, v16, T1D);
4936     pmull2(v22, T1Q, v2, v16, T2D);
4937     ldrq(v2, Address(buf, 0x30));
4938     eor3(v2, T16B, v21, v22, v2);
4939 
4940     pmull (v23, T1Q, v3, v16, T1D);
4941     pmull2(v24, T1Q, v3, v16, T2D);
4942     ldrq(v3, Address(buf, 0x40));
4943     eor3(v3, T16B, v23, v24, v3);
4944 
4945     pmull (v25, T1Q, v4, v16, T1D);
4946     pmull2(v26, T1Q, v4, v16, T2D);
4947     ldrq(v4, Address(buf, 0x50));
4948     eor3(v4, T16B, v25, v26, v4);
4949 
4950     pmull (v27, T1Q, v5, v16, T1D);
4951     pmull2(v28, T1Q, v5, v16, T2D);
4952     ldrq(v5, Address(buf, 0x60));
4953     eor3(v5, T16B, v27, v28, v5);
4954 
4955     pmull (v29, T1Q, v6, v16, T1D);
4956     pmull2(v30, T1Q, v6, v16, T2D);
4957     ldrq(v6, Address(buf, 0x70));
4958     eor3(v6, T16B, v29, v30, v6);
4959 
4960     // Reuse registers v23, v24.
4961     // Using them won't block the first instruction of the next iteration.
4962     pmull (v23, T1Q, v7, v16, T1D);
4963     pmull2(v24, T1Q, v7, v16, T2D);
4964     ldrq(v7, Address(pre(buf, 0x80)));
4965     eor3(v7, T16B, v23, v24, v7);
4966 
4967     subs(len, len, 0x80);
4968     br(Assembler::GE, CRC_by128_loop);
4969 
4970     // fold into 512 bits
4971     // Use v31 for constants because v16 can be still in use.
4972     ldrq(v31, Address(table, 0x10));
4973 
4974     pmull (v17,  T1Q, v0, v31, T1D);
4975     pmull2(v18, T1Q, v0, v31, T2D);
4976     eor3(v0, T16B, v17, v18, v4);
4977 
4978     pmull (v19, T1Q, v1, v31, T1D);
4979     pmull2(v20, T1Q, v1, v31, T2D);
4980     eor3(v1, T16B, v19, v20, v5);
4981 
4982     pmull (v21, T1Q, v2, v31, T1D);
4983     pmull2(v22, T1Q, v2, v31, T2D);
4984     eor3(v2, T16B, v21, v22, v6);
4985 
4986     pmull (v23, T1Q, v3, v31, T1D);
4987     pmull2(v24, T1Q, v3, v31, T2D);
4988     eor3(v3, T16B, v23, v24, v7);
4989 
4990     // fold into 128 bits
4991     // Use v17 for constants because v31 can be still in use.
4992     ldrq(v17, Address(table, 0x20));
4993     pmull (v25, T1Q, v0, v17, T1D);
4994     pmull2(v26, T1Q, v0, v17, T2D);
4995     eor3(v3, T16B, v3, v25, v26);
4996 
4997     // Use v18 for constants because v17 can be still in use.
4998     ldrq(v18, Address(table, 0x30));
4999     pmull (v27, T1Q, v1, v18, T1D);
5000     pmull2(v28, T1Q, v1, v18, T2D);
5001     eor3(v3, T16B, v3, v27, v28);
5002 
5003     // Use v19 for constants because v18 can be still in use.
5004     ldrq(v19, Address(table, 0x40));
5005     pmull (v29, T1Q, v2, v19, T1D);
5006     pmull2(v30, T1Q, v2, v19, T2D);
5007     eor3(v0, T16B, v3, v29, v30);
5008 
5009     add(len, len, 0x80);
5010     add(buf, buf, 0x10);
5011 
5012     mov(tmp0, v0, D, 0);
5013     mov(tmp1, v0, D, 1);
5014 }
5015 
5016 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5017   Address adr;
5018   switch(dst.getMode()) {
5019   case Address::base_plus_offset:
5020     // This is the expected mode, although we allow all the other
5021     // forms below.
5022     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5023     break;
5024   default:
5025     lea(rscratch2, dst);
5026     adr = Address(rscratch2);
5027     break;
5028   }
5029   ldr(rscratch1, adr);
5030   add(rscratch1, rscratch1, src);
5031   str(rscratch1, adr);
5032 }
5033 
5034 void MacroAssembler::cmpptr(Register src1, Address src2) {
5035   uint64_t offset;
5036   adrp(rscratch1, src2, offset);
5037   ldr(rscratch1, Address(rscratch1, offset));
5038   cmp(src1, rscratch1);
5039 }
5040 
5041 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5042   cmp(obj1, obj2);
5043 }
5044 
5045 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5046   load_method_holder(rresult, rmethod);
5047   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5048 }
5049 
5050 void MacroAssembler::load_method_holder(Register holder, Register method) {
5051   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5052   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5053   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5054 }
5055 
5056 // Loads the obj's Klass* into dst.
5057 // Preserves all registers (incl src, rscratch1 and rscratch2).
5058 // Input:
5059 // src - the oop we want to load the klass from.
5060 // dst - output narrow klass.
5061 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5062   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5063   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5064   lsr(dst, dst, markWord::klass_shift);
5065 }
5066 
5067 void MacroAssembler::load_klass(Register dst, Register src) {
5068   if (UseCompactObjectHeaders) {
5069     load_narrow_klass_compact(dst, src);
5070     decode_klass_not_null(dst);
5071   } else if (UseCompressedClassPointers) {
5072     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5073     decode_klass_not_null(dst);
5074   } else {
5075     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5076   }
5077 }
5078 
5079 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5080   if (RestoreMXCSROnJNICalls) {
5081     Label OK;
5082     get_fpcr(tmp1);
5083     mov(tmp2, tmp1);
5084     // Set FPCR to the state we need. We do want Round to Nearest. We
5085     // don't want non-IEEE rounding modes or floating-point traps.
5086     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5087     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5088     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5089     eor(tmp2, tmp1, tmp2);
5090     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5091     set_fpcr(tmp1);
5092     bind(OK);
5093   }
5094 }
5095 
5096 // ((OopHandle)result).resolve();
5097 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5098   // OopHandle::resolve is an indirection.
5099   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5100 }
5101 
5102 // ((WeakHandle)result).resolve();
5103 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5104   assert_different_registers(result, tmp1, tmp2);
5105   Label resolved;
5106 
5107   // A null weak handle resolves to null.
5108   cbz(result, resolved);
5109 
5110   // Only 64 bit platforms support GCs that require a tmp register
5111   // WeakHandle::resolve is an indirection like jweak.
5112   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5113                  result, Address(result), tmp1, tmp2);
5114   bind(resolved);
5115 }
5116 
5117 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5118   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5119   ldr(dst, Address(rmethod, Method::const_offset()));
5120   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5121   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5122   ldr(dst, Address(dst, mirror_offset));
5123   resolve_oop_handle(dst, tmp1, tmp2);
5124 }
5125 
5126 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5127   assert_different_registers(obj, klass, tmp);
5128   if (UseCompressedClassPointers) {
5129     if (UseCompactObjectHeaders) {
5130       load_narrow_klass_compact(tmp, obj);
5131     } else {
5132       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5133     }
5134     if (CompressedKlassPointers::base() == nullptr) {
5135       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5136       return;
5137     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5138                && CompressedKlassPointers::shift() == 0) {
5139       // Only the bottom 32 bits matter
5140       cmpw(klass, tmp);
5141       return;
5142     }
5143     decode_klass_not_null(tmp);
5144   } else {
5145     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5146   }
5147   cmp(klass, tmp);
5148 }
5149 
5150 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5151   if (UseCompactObjectHeaders) {
5152     load_narrow_klass_compact(tmp1, obj1);
5153     load_narrow_klass_compact(tmp2,  obj2);
5154     cmpw(tmp1, tmp2);
5155   } else if (UseCompressedClassPointers) {
5156     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5157     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5158     cmpw(tmp1, tmp2);
5159   } else {
5160     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5161     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5162     cmp(tmp1, tmp2);
5163   }
5164 }
5165 
5166 void MacroAssembler::store_klass(Register dst, Register src) {
5167   // FIXME: Should this be a store release?  concurrent gcs assumes
5168   // klass length is valid if klass field is not null.
5169   assert(!UseCompactObjectHeaders, "not with compact headers");
5170   if (UseCompressedClassPointers) {
5171     encode_klass_not_null(src);
5172     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5173   } else {
5174     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5175   }
5176 }
5177 
5178 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5179   assert(!UseCompactObjectHeaders, "not with compact headers");
5180   if (UseCompressedClassPointers) {
5181     // Store to klass gap in destination
5182     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5183   }
5184 }
5185 
5186 // Algorithm must match CompressedOops::encode.
5187 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5188 #ifdef ASSERT
5189   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5190 #endif
5191   verify_oop_msg(s, "broken oop in encode_heap_oop");
5192   if (CompressedOops::base() == nullptr) {
5193     if (CompressedOops::shift() != 0) {
5194       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5195       lsr(d, s, LogMinObjAlignmentInBytes);
5196     } else {
5197       mov(d, s);
5198     }
5199   } else {
5200     subs(d, s, rheapbase);
5201     csel(d, d, zr, Assembler::HS);
5202     lsr(d, d, LogMinObjAlignmentInBytes);
5203 
5204     /*  Old algorithm: is this any worse?
5205     Label nonnull;
5206     cbnz(r, nonnull);
5207     sub(r, r, rheapbase);
5208     bind(nonnull);
5209     lsr(r, r, LogMinObjAlignmentInBytes);
5210     */
5211   }
5212 }
5213 
5214 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5215 #ifdef ASSERT
5216   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5217   if (CheckCompressedOops) {
5218     Label ok;
5219     cbnz(r, ok);
5220     stop("null oop passed to encode_heap_oop_not_null");
5221     bind(ok);
5222   }
5223 #endif
5224   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5225   if (CompressedOops::base() != nullptr) {
5226     sub(r, r, rheapbase);
5227   }
5228   if (CompressedOops::shift() != 0) {
5229     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5230     lsr(r, r, LogMinObjAlignmentInBytes);
5231   }
5232 }
5233 
5234 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5235 #ifdef ASSERT
5236   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5237   if (CheckCompressedOops) {
5238     Label ok;
5239     cbnz(src, ok);
5240     stop("null oop passed to encode_heap_oop_not_null2");
5241     bind(ok);
5242   }
5243 #endif
5244   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5245 
5246   Register data = src;
5247   if (CompressedOops::base() != nullptr) {
5248     sub(dst, src, rheapbase);
5249     data = dst;
5250   }
5251   if (CompressedOops::shift() != 0) {
5252     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5253     lsr(dst, data, LogMinObjAlignmentInBytes);
5254     data = dst;
5255   }
5256   if (data == src)
5257     mov(dst, src);
5258 }
5259 
5260 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5261 #ifdef ASSERT
5262   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5263 #endif
5264   if (CompressedOops::base() == nullptr) {
5265     if (CompressedOops::shift() != 0) {
5266       lsl(d, s, CompressedOops::shift());
5267     } else if (d != s) {
5268       mov(d, s);
5269     }
5270   } else {
5271     Label done;
5272     if (d != s)
5273       mov(d, s);
5274     cbz(s, done);
5275     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5276     bind(done);
5277   }
5278   verify_oop_msg(d, "broken oop in decode_heap_oop");
5279 }
5280 
5281 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5282   assert (UseCompressedOops, "should only be used for compressed headers");
5283   assert (Universe::heap() != nullptr, "java heap should be initialized");
5284   // Cannot assert, unverified entry point counts instructions (see .ad file)
5285   // vtableStubs also counts instructions in pd_code_size_limit.
5286   // Also do not verify_oop as this is called by verify_oop.
5287   if (CompressedOops::shift() != 0) {
5288     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5289     if (CompressedOops::base() != nullptr) {
5290       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5291     } else {
5292       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5293     }
5294   } else {
5295     assert (CompressedOops::base() == nullptr, "sanity");
5296   }
5297 }
5298 
5299 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
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(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5309     } else {
5310       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5311     }
5312   } else {
5313     assert (CompressedOops::base() == nullptr, "sanity");
5314     if (dst != src) {
5315       mov(dst, src);
5316     }
5317   }
5318 }
5319 
5320 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5321 
5322 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5323   assert(Metaspace::initialized(), "metaspace not initialized yet");
5324   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5325   return _klass_decode_mode;
5326 }
5327 
5328 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5329   assert(UseCompressedClassPointers, "not using compressed class pointers");
5330 
5331   // KlassDecodeMode shouldn't be set already.
5332   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5333 
5334   if (base == nullptr) {
5335     return KlassDecodeZero;
5336   }
5337 
5338   if (operand_valid_for_logical_immediate(
5339         /*is32*/false, (uint64_t)base)) {
5340     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5341     if (((uint64_t)base & range_mask) == 0) {
5342       return KlassDecodeXor;
5343     }
5344   }
5345 
5346   const uint64_t shifted_base =
5347     (uint64_t)base >> shift;
5348   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5349     return KlassDecodeMovk;
5350   }
5351 
5352   // No valid encoding.
5353   return KlassDecodeNone;
5354 }
5355 
5356 // Check if one of the above decoding modes will work for given base, shift and range.
5357 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5358   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5359 }
5360 
5361 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5362   _klass_decode_mode = klass_decode_mode(base, shift, range);
5363   return _klass_decode_mode != KlassDecodeNone;
5364 }
5365 
5366 static Register pick_different_tmp(Register dst, Register src) {
5367   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5368   return *tmps.begin();
5369 }
5370 
5371 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5372   // we have to load the klass base from the AOT constants area but
5373   // not the shift because it is not allowed to change
5374   int shift = CompressedKlassPointers::shift();
5375   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5376   if (dst != src) {
5377     // we can load the base into dst, subtract it formthe src and shift down
5378     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5379     ldr(dst, dst);
5380     sub(dst, src, dst);
5381     lsr(dst, dst, shift);
5382   } else {
5383     // we need an extra register in order to load the coop base
5384     Register tmp = pick_different_tmp(dst, src);
5385     RegSet regs = RegSet::of(tmp);
5386     push(regs, sp);
5387     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5388     ldr(tmp, tmp);
5389     sub(dst, src, tmp);
5390     lsr(dst, dst, shift);
5391     pop(regs, sp);
5392   }
5393 }
5394 
5395 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5396   if (AOTCodeCache::is_on_for_dump()) {
5397     encode_klass_not_null_for_aot(dst, src);
5398     return;
5399   }
5400 
5401   switch (klass_decode_mode()) {
5402   case KlassDecodeZero:
5403     if (CompressedKlassPointers::shift() != 0) {
5404       lsr(dst, src, CompressedKlassPointers::shift());
5405     } else {
5406       if (dst != src) mov(dst, src);
5407     }
5408     break;
5409 
5410   case KlassDecodeXor:
5411     if (CompressedKlassPointers::shift() != 0) {
5412       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5413       lsr(dst, dst, CompressedKlassPointers::shift());
5414     } else {
5415       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5416     }
5417     break;
5418 
5419   case KlassDecodeMovk:
5420     if (CompressedKlassPointers::shift() != 0) {
5421       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5422     } else {
5423       movw(dst, src);
5424     }
5425     break;
5426 
5427   case KlassDecodeNone:
5428     ShouldNotReachHere();
5429     break;
5430   }
5431 }
5432 
5433 void MacroAssembler::encode_klass_not_null(Register r) {
5434   encode_klass_not_null(r, r);
5435 }
5436 
5437 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5438   // we have to load the klass base from the AOT constants area but
5439   // not the shift because it is not allowed to change
5440   int shift = CompressedKlassPointers::shift();
5441   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5442   if (dst != src) {
5443     // we can load the base into dst then add the offset with a suitable shift
5444     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5445     ldr(dst, dst);
5446     add(dst, dst, src, LSL,  shift);
5447   } else {
5448     // we need an extra register in order to load the coop base
5449     Register tmp = pick_different_tmp(dst, src);
5450     RegSet regs = RegSet::of(tmp);
5451     push(regs, sp);
5452     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5453     ldr(tmp, tmp);
5454     add(dst, tmp,  src, LSL,  shift);
5455     pop(regs, sp);
5456   }
5457 }
5458 
5459 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5460   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5461 
5462   if (AOTCodeCache::is_on_for_dump()) {
5463     decode_klass_not_null_for_aot(dst, src);
5464     return;
5465   }
5466 
5467   switch (klass_decode_mode()) {
5468   case KlassDecodeZero:
5469     if (CompressedKlassPointers::shift() != 0) {
5470       lsl(dst, src, CompressedKlassPointers::shift());
5471     } else {
5472       if (dst != src) mov(dst, src);
5473     }
5474     break;
5475 
5476   case KlassDecodeXor:
5477     if (CompressedKlassPointers::shift() != 0) {
5478       lsl(dst, src, CompressedKlassPointers::shift());
5479       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5480     } else {
5481       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5482     }
5483     break;
5484 
5485   case KlassDecodeMovk: {
5486     const uint64_t shifted_base =
5487       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5488 
5489     if (dst != src) movw(dst, src);
5490     movk(dst, shifted_base >> 32, 32);
5491 
5492     if (CompressedKlassPointers::shift() != 0) {
5493       lsl(dst, dst, CompressedKlassPointers::shift());
5494     }
5495 
5496     break;
5497   }
5498 
5499   case KlassDecodeNone:
5500     ShouldNotReachHere();
5501     break;
5502   }
5503 }
5504 
5505 void  MacroAssembler::decode_klass_not_null(Register r) {
5506   decode_klass_not_null(r, r);
5507 }
5508 
5509 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5510 #ifdef ASSERT
5511   {
5512     ThreadInVMfromUnknown tiv;
5513     assert (UseCompressedOops, "should only be used for compressed oops");
5514     assert (Universe::heap() != nullptr, "java heap should be initialized");
5515     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5516     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5517   }
5518 #endif
5519   int oop_index = oop_recorder()->find_index(obj);
5520   InstructionMark im(this);
5521   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5522   code_section()->relocate(inst_mark(), rspec);
5523   movz(dst, 0xDEAD, 16);
5524   movk(dst, 0xBEEF);
5525 }
5526 
5527 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5528   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5529   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5530   int index = oop_recorder()->find_index(k);
5531 
5532   InstructionMark im(this);
5533   RelocationHolder rspec = metadata_Relocation::spec(index);
5534   code_section()->relocate(inst_mark(), rspec);
5535   narrowKlass nk = CompressedKlassPointers::encode(k);
5536   movz(dst, (nk >> 16), 16);
5537   movk(dst, nk & 0xffff);
5538 }
5539 
5540 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5541                                     Register dst, Address src,
5542                                     Register tmp1, Register tmp2) {
5543   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5544   decorators = AccessInternal::decorator_fixup(decorators, type);
5545   bool as_raw = (decorators & AS_RAW) != 0;
5546   if (as_raw) {
5547     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5548   } else {
5549     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5550   }
5551 }
5552 
5553 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5554                                      Address dst, Register val,
5555                                      Register tmp1, Register tmp2, Register tmp3) {
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::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5561   } else {
5562     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5563   }
5564 }
5565 
5566 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5567                                    Register tmp2, DecoratorSet decorators) {
5568   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5569 }
5570 
5571 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5572                                             Register tmp2, DecoratorSet decorators) {
5573   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5574 }
5575 
5576 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5577                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5578   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5579 }
5580 
5581 // Used for storing nulls.
5582 void MacroAssembler::store_heap_oop_null(Address dst) {
5583   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5584 }
5585 
5586 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5587   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5588   int index = oop_recorder()->allocate_metadata_index(obj);
5589   RelocationHolder rspec = metadata_Relocation::spec(index);
5590   return Address((address)obj, rspec);
5591 }
5592 
5593 // Move an oop into a register.
5594 void MacroAssembler::movoop(Register dst, jobject obj) {
5595   int oop_index;
5596   if (obj == nullptr) {
5597     oop_index = oop_recorder()->allocate_oop_index(obj);
5598   } else {
5599 #ifdef ASSERT
5600     {
5601       ThreadInVMfromUnknown tiv;
5602       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5603     }
5604 #endif
5605     oop_index = oop_recorder()->find_index(obj);
5606   }
5607   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5608 
5609   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5610     mov(dst, Address((address)obj, rspec));
5611   } else {
5612     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5613     ldr(dst, Address(dummy, rspec));
5614   }
5615 }
5616 
5617 // Move a metadata address into a register.
5618 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5619   int oop_index;
5620   if (obj == nullptr) {
5621     oop_index = oop_recorder()->allocate_metadata_index(obj);
5622   } else {
5623     oop_index = oop_recorder()->find_index(obj);
5624   }
5625   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5626   mov(dst, Address((address)obj, rspec));
5627 }
5628 
5629 Address MacroAssembler::constant_oop_address(jobject obj) {
5630 #ifdef ASSERT
5631   {
5632     ThreadInVMfromUnknown tiv;
5633     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5634     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5635   }
5636 #endif
5637   int oop_index = oop_recorder()->find_index(obj);
5638   return Address((address)obj, oop_Relocation::spec(oop_index));
5639 }
5640 
5641 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5642 void MacroAssembler::tlab_allocate(Register obj,
5643                                    Register var_size_in_bytes,
5644                                    int con_size_in_bytes,
5645                                    Register t1,
5646                                    Register t2,
5647                                    Label& slow_case) {
5648   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5649   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5650 }
5651 
5652 void MacroAssembler::verify_tlab() {
5653 #ifdef ASSERT
5654   if (UseTLAB && VerifyOops) {
5655     Label next, ok;
5656 
5657     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5658 
5659     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5660     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5661     cmp(rscratch2, rscratch1);
5662     br(Assembler::HS, next);
5663     STOP("assert(top >= start)");
5664     should_not_reach_here();
5665 
5666     bind(next);
5667     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5668     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5669     cmp(rscratch2, rscratch1);
5670     br(Assembler::HS, ok);
5671     STOP("assert(top <= end)");
5672     should_not_reach_here();
5673 
5674     bind(ok);
5675     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5676   }
5677 #endif
5678 }
5679 
5680 // Writes to stack successive pages until offset reached to check for
5681 // stack overflow + shadow pages.  This clobbers tmp.
5682 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5683   assert_different_registers(tmp, size, rscratch1);
5684   mov(tmp, sp);
5685   // Bang stack for total size given plus shadow page size.
5686   // Bang one page at a time because large size can bang beyond yellow and
5687   // red zones.
5688   Label loop;
5689   mov(rscratch1, (int)os::vm_page_size());
5690   bind(loop);
5691   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5692   subsw(size, size, rscratch1);
5693   str(size, Address(tmp));
5694   br(Assembler::GT, loop);
5695 
5696   // Bang down shadow pages too.
5697   // At this point, (tmp-0) is the last address touched, so don't
5698   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5699   // was post-decremented.)  Skip this address by starting at i=1, and
5700   // touch a few more pages below.  N.B.  It is important to touch all
5701   // the way down to and including i=StackShadowPages.
5702   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5703     // this could be any sized move but this is can be a debugging crumb
5704     // so the bigger the better.
5705     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5706     str(size, Address(tmp));
5707   }
5708 }
5709 
5710 // Move the address of the polling page into dest.
5711 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5712   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5713 }
5714 
5715 // Read the polling page.  The address of the polling page must
5716 // already be in r.
5717 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5718   address mark;
5719   {
5720     InstructionMark im(this);
5721     code_section()->relocate(inst_mark(), rtype);
5722     ldrw(zr, Address(r, 0));
5723     mark = inst_mark();
5724   }
5725   verify_cross_modify_fence_not_required();
5726   return mark;
5727 }
5728 
5729 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5730   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5731   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5732   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5733   int64_t offset_low = dest_page - low_page;
5734   int64_t offset_high = dest_page - high_page;
5735 
5736   assert(is_valid_AArch64_address(dest.target()), "bad address");
5737   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5738 
5739   InstructionMark im(this);
5740   code_section()->relocate(inst_mark(), dest.rspec());
5741   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5742   // the code cache so that if it is relocated we know it will still reach
5743   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5744     _adrp(reg1, dest.target());
5745   } else {
5746     uint64_t target = (uint64_t)dest.target();
5747     uint64_t adrp_target
5748       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5749 
5750     _adrp(reg1, (address)adrp_target);
5751     movk(reg1, target >> 32, 32);
5752   }
5753   byte_offset = (uint64_t)dest.target() & 0xfff;
5754 }
5755 
5756 void MacroAssembler::load_byte_map_base(Register reg) {
5757 #if INCLUDE_CDS
5758   if (AOTCodeCache::is_on_for_dump()) {
5759     address byte_map_base_adr = AOTRuntimeConstants::card_table_base_address();
5760     lea(reg, ExternalAddress(byte_map_base_adr));
5761     ldr(reg, Address(reg));
5762     return;
5763   }
5764 #endif
5765   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
5766 
5767   // Strictly speaking the card table base isn't an address at all, and it might
5768   // even be negative. It is thus materialised as a constant.
5769   mov(reg, (uint64_t)ctbs->card_table_base_const());
5770 }
5771 
5772 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5773 #if INCLUDE_CDS
5774   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5775   if (AOTCodeCache::is_on_for_dump()) {
5776     // all aotrc field addresses should be registered in the AOTCodeCache address table
5777     lea(reg, ExternalAddress(a));
5778   } else {
5779     mov(reg, (uint64_t)a);
5780   }
5781 #else
5782   ShouldNotReachHere();
5783 #endif
5784 }
5785 
5786 void MacroAssembler::build_frame(int framesize) {
5787   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5788   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5789   protect_return_address();
5790   if (framesize < ((1 << 9) + 2 * wordSize)) {
5791     sub(sp, sp, framesize);
5792     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5793     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5794   } else {
5795     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5796     if (PreserveFramePointer) mov(rfp, sp);
5797     if (framesize < ((1 << 12) + 2 * wordSize))
5798       sub(sp, sp, framesize - 2 * wordSize);
5799     else {
5800       mov(rscratch1, framesize - 2 * wordSize);
5801       sub(sp, sp, rscratch1);
5802     }
5803   }
5804   verify_cross_modify_fence_not_required();
5805 }
5806 
5807 void MacroAssembler::remove_frame(int framesize) {
5808   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5809   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5810   if (framesize < ((1 << 9) + 2 * wordSize)) {
5811     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5812     add(sp, sp, framesize);
5813   } else {
5814     if (framesize < ((1 << 12) + 2 * wordSize))
5815       add(sp, sp, framesize - 2 * wordSize);
5816     else {
5817       mov(rscratch1, framesize - 2 * wordSize);
5818       add(sp, sp, rscratch1);
5819     }
5820     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5821   }
5822   authenticate_return_address();
5823 }
5824 
5825 
5826 // This method counts leading positive bytes (highest bit not set) in provided byte array
5827 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5828     // Simple and most common case of aligned small array which is not at the
5829     // end of memory page is placed here. All other cases are in stub.
5830     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5831     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5832     assert_different_registers(ary1, len, result);
5833 
5834     mov(result, len);
5835     cmpw(len, 0);
5836     br(LE, DONE);
5837     cmpw(len, 4 * wordSize);
5838     br(GE, STUB_LONG); // size > 32 then go to stub
5839 
5840     int shift = 64 - exact_log2(os::vm_page_size());
5841     lsl(rscratch1, ary1, shift);
5842     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5843     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5844     br(CS, STUB); // at the end of page then go to stub
5845     subs(len, len, wordSize);
5846     br(LT, END);
5847 
5848   BIND(LOOP);
5849     ldr(rscratch1, Address(post(ary1, wordSize)));
5850     tst(rscratch1, UPPER_BIT_MASK);
5851     br(NE, SET_RESULT);
5852     subs(len, len, wordSize);
5853     br(GE, LOOP);
5854     cmpw(len, -wordSize);
5855     br(EQ, DONE);
5856 
5857   BIND(END);
5858     ldr(rscratch1, Address(ary1));
5859     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5860     lslv(rscratch1, rscratch1, rscratch2);
5861     tst(rscratch1, UPPER_BIT_MASK);
5862     br(NE, SET_RESULT);
5863     b(DONE);
5864 
5865   BIND(STUB);
5866     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5867     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5868     address tpc1 = trampoline_call(count_pos);
5869     if (tpc1 == nullptr) {
5870       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5871       postcond(pc() == badAddress);
5872       return nullptr;
5873     }
5874     b(DONE);
5875 
5876   BIND(STUB_LONG);
5877     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5878     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5879     address tpc2 = trampoline_call(count_pos_long);
5880     if (tpc2 == nullptr) {
5881       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5882       postcond(pc() == badAddress);
5883       return nullptr;
5884     }
5885     b(DONE);
5886 
5887   BIND(SET_RESULT);
5888 
5889     add(len, len, wordSize);
5890     sub(result, result, len);
5891 
5892   BIND(DONE);
5893   postcond(pc() != badAddress);
5894   return pc();
5895 }
5896 
5897 // Clobbers: rscratch1, rscratch2, rflags
5898 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5899 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5900                                       Register tmp4, Register tmp5, Register result,
5901                                       Register cnt1, int elem_size) {
5902   Label DONE, SAME;
5903   Register tmp1 = rscratch1;
5904   Register tmp2 = rscratch2;
5905   int elem_per_word = wordSize/elem_size;
5906   int log_elem_size = exact_log2(elem_size);
5907   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5908   int length_offset = arrayOopDesc::length_offset_in_bytes();
5909   int base_offset
5910     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5911   // When the length offset is not aligned to 8 bytes,
5912   // then we align it down. This is valid because the new
5913   // offset will always be the klass which is the same
5914   // for type arrays.
5915   int start_offset = align_down(length_offset, BytesPerWord);
5916   int extra_length = base_offset - start_offset;
5917   assert(start_offset == length_offset || start_offset == klass_offset,
5918          "start offset must be 8-byte-aligned or be the klass offset");
5919   assert(base_offset != start_offset, "must include the length field");
5920   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5921   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5922 
5923   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5924   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5925 
5926 #ifndef PRODUCT
5927   {
5928     const char kind = (elem_size == 2) ? 'U' : 'L';
5929     char comment[64];
5930     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5931     BLOCK_COMMENT(comment);
5932   }
5933 #endif
5934 
5935   // if (a1 == a2)
5936   //     return true;
5937   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5938   br(EQ, SAME);
5939 
5940   if (UseSimpleArrayEquals) {
5941     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5942     // if (a1 == nullptr || a2 == nullptr)
5943     //     return false;
5944     // a1 & a2 == 0 means (some-pointer is null) or
5945     // (very-rare-or-even-probably-impossible-pointer-values)
5946     // so, we can save one branch in most cases
5947     tst(a1, a2);
5948     mov(result, false);
5949     br(EQ, A_MIGHT_BE_NULL);
5950     // if (a1.length != a2.length)
5951     //      return false;
5952     bind(A_IS_NOT_NULL);
5953     ldrw(cnt1, Address(a1, length_offset));
5954     ldrw(tmp5, Address(a2, length_offset));
5955     cmp(cnt1, tmp5);
5956     br(NE, DONE); // If lengths differ, return false
5957     // Increase loop counter by diff between base- and actual start-offset.
5958     addw(cnt1, cnt1, extra_length);
5959     lea(a1, Address(a1, start_offset));
5960     lea(a2, Address(a2, start_offset));
5961     // Check for short strings, i.e. smaller than wordSize.
5962     subs(cnt1, cnt1, elem_per_word);
5963     br(Assembler::LT, SHORT);
5964     // Main 8 byte comparison loop.
5965     bind(NEXT_WORD); {
5966       ldr(tmp1, Address(post(a1, wordSize)));
5967       ldr(tmp2, Address(post(a2, wordSize)));
5968       subs(cnt1, cnt1, elem_per_word);
5969       eor(tmp5, tmp1, tmp2);
5970       cbnz(tmp5, DONE);
5971     } br(GT, NEXT_WORD);
5972     // Last longword.  In the case where length == 4 we compare the
5973     // same longword twice, but that's still faster than another
5974     // conditional branch.
5975     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5976     // length == 4.
5977     if (log_elem_size > 0)
5978       lsl(cnt1, cnt1, log_elem_size);
5979     ldr(tmp3, Address(a1, cnt1));
5980     ldr(tmp4, Address(a2, cnt1));
5981     eor(tmp5, tmp3, tmp4);
5982     cbnz(tmp5, DONE);
5983     b(SAME);
5984     bind(A_MIGHT_BE_NULL);
5985     // in case both a1 and a2 are not-null, proceed with loads
5986     cbz(a1, DONE);
5987     cbz(a2, DONE);
5988     b(A_IS_NOT_NULL);
5989     bind(SHORT);
5990 
5991     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
5992     {
5993       ldrw(tmp1, Address(post(a1, 4)));
5994       ldrw(tmp2, Address(post(a2, 4)));
5995       eorw(tmp5, tmp1, tmp2);
5996       cbnzw(tmp5, DONE);
5997     }
5998     bind(TAIL03);
5999     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6000     {
6001       ldrh(tmp3, Address(post(a1, 2)));
6002       ldrh(tmp4, Address(post(a2, 2)));
6003       eorw(tmp5, tmp3, tmp4);
6004       cbnzw(tmp5, DONE);
6005     }
6006     bind(TAIL01);
6007     if (elem_size == 1) { // Only needed when comparing byte arrays.
6008       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6009       {
6010         ldrb(tmp1, a1);
6011         ldrb(tmp2, a2);
6012         eorw(tmp5, tmp1, tmp2);
6013         cbnzw(tmp5, DONE);
6014       }
6015     }
6016   } else {
6017     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6018         CSET_EQ, LAST_CHECK;
6019     mov(result, false);
6020     cbz(a1, DONE);
6021     ldrw(cnt1, Address(a1, length_offset));
6022     cbz(a2, DONE);
6023     ldrw(tmp5, Address(a2, length_offset));
6024     cmp(cnt1, tmp5);
6025     br(NE, DONE); // If lengths differ, return false
6026     // Increase loop counter by diff between base- and actual start-offset.
6027     addw(cnt1, cnt1, extra_length);
6028 
6029     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6030     // faster to perform another branch before comparing a1 and a2
6031     cmp(cnt1, (u1)elem_per_word);
6032     br(LE, SHORT); // short or same
6033     ldr(tmp3, Address(pre(a1, start_offset)));
6034     subs(zr, cnt1, stubBytesThreshold);
6035     br(GE, STUB);
6036     ldr(tmp4, Address(pre(a2, start_offset)));
6037     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6038 
6039     // Main 16 byte comparison loop with 2 exits
6040     bind(NEXT_DWORD); {
6041       ldr(tmp1, Address(pre(a1, wordSize)));
6042       ldr(tmp2, Address(pre(a2, wordSize)));
6043       subs(cnt1, cnt1, 2 * elem_per_word);
6044       br(LE, TAIL);
6045       eor(tmp4, tmp3, tmp4);
6046       cbnz(tmp4, DONE);
6047       ldr(tmp3, Address(pre(a1, wordSize)));
6048       ldr(tmp4, Address(pre(a2, wordSize)));
6049       cmp(cnt1, (u1)elem_per_word);
6050       br(LE, TAIL2);
6051       cmp(tmp1, tmp2);
6052     } br(EQ, NEXT_DWORD);
6053     b(DONE);
6054 
6055     bind(TAIL);
6056     eor(tmp4, tmp3, tmp4);
6057     eor(tmp2, tmp1, tmp2);
6058     lslv(tmp2, tmp2, tmp5);
6059     orr(tmp5, tmp4, tmp2);
6060     cmp(tmp5, zr);
6061     b(CSET_EQ);
6062 
6063     bind(TAIL2);
6064     eor(tmp2, tmp1, tmp2);
6065     cbnz(tmp2, DONE);
6066     b(LAST_CHECK);
6067 
6068     bind(STUB);
6069     ldr(tmp4, Address(pre(a2, start_offset)));
6070     if (elem_size == 2) { // convert to byte counter
6071       lsl(cnt1, cnt1, 1);
6072     }
6073     eor(tmp5, tmp3, tmp4);
6074     cbnz(tmp5, DONE);
6075     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6076     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6077     address tpc = trampoline_call(stub);
6078     if (tpc == nullptr) {
6079       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6080       postcond(pc() == badAddress);
6081       return nullptr;
6082     }
6083     b(DONE);
6084 
6085     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6086     // so, if a2 == null => return false(0), else return true, so we can return a2
6087     mov(result, a2);
6088     b(DONE);
6089     bind(SHORT);
6090     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6091     ldr(tmp3, Address(a1, start_offset));
6092     ldr(tmp4, Address(a2, start_offset));
6093     bind(LAST_CHECK);
6094     eor(tmp4, tmp3, tmp4);
6095     lslv(tmp5, tmp4, tmp5);
6096     cmp(tmp5, zr);
6097     bind(CSET_EQ);
6098     cset(result, EQ);
6099     b(DONE);
6100   }
6101 
6102   bind(SAME);
6103   mov(result, true);
6104   // That's it.
6105   bind(DONE);
6106 
6107   BLOCK_COMMENT("} array_equals");
6108   postcond(pc() != badAddress);
6109   return pc();
6110 }
6111 
6112 // Compare Strings
6113 
6114 // For Strings we're passed the address of the first characters in a1
6115 // and a2 and the length in cnt1.
6116 // There are two implementations.  For arrays >= 8 bytes, all
6117 // comparisons (including the final one, which may overlap) are
6118 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6119 // halfword, then a short, and then a byte.
6120 
6121 void MacroAssembler::string_equals(Register a1, Register a2,
6122                                    Register result, Register cnt1)
6123 {
6124   Label SAME, DONE, SHORT, NEXT_WORD;
6125   Register tmp1 = rscratch1;
6126   Register tmp2 = rscratch2;
6127 
6128   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6129 
6130 #ifndef PRODUCT
6131   {
6132     char comment[64];
6133     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6134     BLOCK_COMMENT(comment);
6135   }
6136 #endif
6137 
6138   mov(result, false);
6139 
6140   // Check for short strings, i.e. smaller than wordSize.
6141   subs(cnt1, cnt1, wordSize);
6142   br(Assembler::LT, SHORT);
6143   // Main 8 byte comparison loop.
6144   bind(NEXT_WORD); {
6145     ldr(tmp1, Address(post(a1, wordSize)));
6146     ldr(tmp2, Address(post(a2, wordSize)));
6147     subs(cnt1, cnt1, wordSize);
6148     eor(tmp1, tmp1, tmp2);
6149     cbnz(tmp1, DONE);
6150   } br(GT, NEXT_WORD);
6151   // Last longword.  In the case where length == 4 we compare the
6152   // same longword twice, but that's still faster than another
6153   // conditional branch.
6154   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6155   // length == 4.
6156   ldr(tmp1, Address(a1, cnt1));
6157   ldr(tmp2, Address(a2, cnt1));
6158   eor(tmp2, tmp1, tmp2);
6159   cbnz(tmp2, DONE);
6160   b(SAME);
6161 
6162   bind(SHORT);
6163   Label TAIL03, TAIL01;
6164 
6165   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6166   {
6167     ldrw(tmp1, Address(post(a1, 4)));
6168     ldrw(tmp2, Address(post(a2, 4)));
6169     eorw(tmp1, tmp1, tmp2);
6170     cbnzw(tmp1, DONE);
6171   }
6172   bind(TAIL03);
6173   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6174   {
6175     ldrh(tmp1, Address(post(a1, 2)));
6176     ldrh(tmp2, Address(post(a2, 2)));
6177     eorw(tmp1, tmp1, tmp2);
6178     cbnzw(tmp1, DONE);
6179   }
6180   bind(TAIL01);
6181   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6182     {
6183     ldrb(tmp1, a1);
6184     ldrb(tmp2, a2);
6185     eorw(tmp1, tmp1, tmp2);
6186     cbnzw(tmp1, DONE);
6187   }
6188   // Arrays are equal.
6189   bind(SAME);
6190   mov(result, true);
6191 
6192   // That's it.
6193   bind(DONE);
6194   BLOCK_COMMENT("} string_equals");
6195 }
6196 
6197 
6198 // The size of the blocks erased by the zero_blocks stub.  We must
6199 // handle anything smaller than this ourselves in zero_words().
6200 const int MacroAssembler::zero_words_block_size = 8;
6201 
6202 // zero_words() is used by C2 ClearArray patterns and by
6203 // C1_MacroAssembler.  It is as small as possible, handling small word
6204 // counts locally and delegating anything larger to the zero_blocks
6205 // stub.  It is expanded many times in compiled code, so it is
6206 // important to keep it short.
6207 
6208 // ptr:   Address of a buffer to be zeroed.
6209 // cnt:   Count in HeapWords.
6210 //
6211 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6212 address MacroAssembler::zero_words(Register ptr, Register cnt)
6213 {
6214   assert(is_power_of_2(zero_words_block_size), "adjust this");
6215 
6216   BLOCK_COMMENT("zero_words {");
6217   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6218   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6219   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6220 
6221   subs(rscratch1, cnt, zero_words_block_size);
6222   Label around;
6223   br(LO, around);
6224   {
6225     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6226     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6227     // Make sure this is a C2 compilation. C1 allocates space only for
6228     // trampoline stubs generated by Call LIR ops, and in any case it
6229     // makes sense for a C1 compilation task to proceed as quickly as
6230     // possible.
6231     CompileTask* task;
6232     if (StubRoutines::aarch64::complete()
6233         && Thread::current()->is_Compiler_thread()
6234         && (task = ciEnv::current()->task())
6235         && is_c2_compile(task->comp_level())) {
6236       address tpc = trampoline_call(zero_blocks);
6237       if (tpc == nullptr) {
6238         DEBUG_ONLY(reset_labels(around));
6239         return nullptr;
6240       }
6241     } else {
6242       far_call(zero_blocks);
6243     }
6244   }
6245   bind(around);
6246 
6247   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6248   // for us.
6249   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6250     Label l;
6251     tbz(cnt, exact_log2(i), l);
6252     for (int j = 0; j < i; j += 2) {
6253       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6254     }
6255     bind(l);
6256   }
6257   {
6258     Label l;
6259     tbz(cnt, 0, l);
6260     str(zr, Address(ptr));
6261     bind(l);
6262   }
6263 
6264   BLOCK_COMMENT("} zero_words");
6265   return pc();
6266 }
6267 
6268 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6269 // cnt:          Immediate count in HeapWords.
6270 //
6271 // r10, r11, rscratch1, and rscratch2 are clobbered.
6272 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6273 {
6274   assert(wordSize <= BlockZeroingLowLimit,
6275             "increase BlockZeroingLowLimit");
6276   address result = nullptr;
6277   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6278 #ifndef PRODUCT
6279     {
6280       char buf[64];
6281       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6282       BLOCK_COMMENT(buf);
6283     }
6284 #endif
6285     if (cnt >= 16) {
6286       uint64_t loops = cnt/16;
6287       if (loops > 1) {
6288         mov(rscratch2, loops - 1);
6289       }
6290       {
6291         Label loop;
6292         bind(loop);
6293         for (int i = 0; i < 16; i += 2) {
6294           stp(zr, zr, Address(base, i * BytesPerWord));
6295         }
6296         add(base, base, 16 * BytesPerWord);
6297         if (loops > 1) {
6298           subs(rscratch2, rscratch2, 1);
6299           br(GE, loop);
6300         }
6301       }
6302     }
6303     cnt %= 16;
6304     int i = cnt & 1;  // store any odd word to start
6305     if (i) str(zr, Address(base));
6306     for (; i < (int)cnt; i += 2) {
6307       stp(zr, zr, Address(base, i * wordSize));
6308     }
6309     BLOCK_COMMENT("} zero_words");
6310     result = pc();
6311   } else {
6312     mov(r10, base); mov(r11, cnt);
6313     result = zero_words(r10, r11);
6314   }
6315   return result;
6316 }
6317 
6318 // Zero blocks of memory by using DC ZVA.
6319 //
6320 // Aligns the base address first sufficiently for DC ZVA, then uses
6321 // DC ZVA repeatedly for every full block.  cnt is the size to be
6322 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6323 // in cnt.
6324 //
6325 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6326 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6327 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6328   Register tmp = rscratch1;
6329   Register tmp2 = rscratch2;
6330   int zva_length = VM_Version::zva_length();
6331   Label initial_table_end, loop_zva;
6332   Label fini;
6333 
6334   // Base must be 16 byte aligned. If not just return and let caller handle it
6335   tst(base, 0x0f);
6336   br(Assembler::NE, fini);
6337   // Align base with ZVA length.
6338   neg(tmp, base);
6339   andr(tmp, tmp, zva_length - 1);
6340 
6341   // tmp: the number of bytes to be filled to align the base with ZVA length.
6342   add(base, base, tmp);
6343   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6344   adr(tmp2, initial_table_end);
6345   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6346   br(tmp2);
6347 
6348   for (int i = -zva_length + 16; i < 0; i += 16)
6349     stp(zr, zr, Address(base, i));
6350   bind(initial_table_end);
6351 
6352   sub(cnt, cnt, zva_length >> 3);
6353   bind(loop_zva);
6354   dc(Assembler::ZVA, base);
6355   subs(cnt, cnt, zva_length >> 3);
6356   add(base, base, zva_length);
6357   br(Assembler::GE, loop_zva);
6358   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6359   bind(fini);
6360 }
6361 
6362 // base:   Address of a buffer to be filled, 8 bytes aligned.
6363 // cnt:    Count in 8-byte unit.
6364 // value:  Value to be filled with.
6365 // base will point to the end of the buffer after filling.
6366 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6367 {
6368 //  Algorithm:
6369 //
6370 //    if (cnt == 0) {
6371 //      return;
6372 //    }
6373 //    if ((p & 8) != 0) {
6374 //      *p++ = v;
6375 //    }
6376 //
6377 //    scratch1 = cnt & 14;
6378 //    cnt -= scratch1;
6379 //    p += scratch1;
6380 //    switch (scratch1 / 2) {
6381 //      do {
6382 //        cnt -= 16;
6383 //          p[-16] = v;
6384 //          p[-15] = v;
6385 //        case 7:
6386 //          p[-14] = v;
6387 //          p[-13] = v;
6388 //        case 6:
6389 //          p[-12] = v;
6390 //          p[-11] = v;
6391 //          // ...
6392 //        case 1:
6393 //          p[-2] = v;
6394 //          p[-1] = v;
6395 //        case 0:
6396 //          p += 16;
6397 //      } while (cnt);
6398 //    }
6399 //    if ((cnt & 1) == 1) {
6400 //      *p++ = v;
6401 //    }
6402 
6403   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6404 
6405   Label fini, skip, entry, loop;
6406   const int unroll = 8; // Number of stp instructions we'll unroll
6407 
6408   cbz(cnt, fini);
6409   tbz(base, 3, skip);
6410   str(value, Address(post(base, 8)));
6411   sub(cnt, cnt, 1);
6412   bind(skip);
6413 
6414   andr(rscratch1, cnt, (unroll-1) * 2);
6415   sub(cnt, cnt, rscratch1);
6416   add(base, base, rscratch1, Assembler::LSL, 3);
6417   adr(rscratch2, entry);
6418   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6419   br(rscratch2);
6420 
6421   bind(loop);
6422   add(base, base, unroll * 16);
6423   for (int i = -unroll; i < 0; i++)
6424     stp(value, value, Address(base, i * 16));
6425   bind(entry);
6426   subs(cnt, cnt, unroll * 2);
6427   br(Assembler::GE, loop);
6428 
6429   tbz(cnt, 0, fini);
6430   str(value, Address(post(base, 8)));
6431   bind(fini);
6432 }
6433 
6434 // Intrinsic for
6435 //
6436 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6437 //   Encodes char[] to byte[] in ISO-8859-1
6438 //
6439 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6440 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6441 //
6442 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6443 //   Encodes char[] to byte[] in ASCII
6444 //
6445 // This version always returns the number of characters copied, and does not
6446 // clobber the 'len' register. A successful copy will complete with the post-
6447 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6448 // post-condition: 0 <= 'res' < 'len'.
6449 //
6450 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6451 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6452 //       beyond the acceptable, even though the footprint would be smaller.
6453 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6454 //       avoid additional bloat.
6455 //
6456 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6457 void MacroAssembler::encode_iso_array(Register src, Register dst,
6458                                       Register len, Register res, bool ascii,
6459                                       FloatRegister vtmp0, FloatRegister vtmp1,
6460                                       FloatRegister vtmp2, FloatRegister vtmp3,
6461                                       FloatRegister vtmp4, FloatRegister vtmp5)
6462 {
6463   Register cnt = res;
6464   Register max = rscratch1;
6465   Register chk = rscratch2;
6466 
6467   prfm(Address(src), PLDL1STRM);
6468   movw(cnt, len);
6469 
6470 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6471 
6472   Label LOOP_32, DONE_32, FAIL_32;
6473 
6474   BIND(LOOP_32);
6475   {
6476     cmpw(cnt, 32);
6477     br(LT, DONE_32);
6478     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6479     // Extract lower bytes.
6480     FloatRegister vlo0 = vtmp4;
6481     FloatRegister vlo1 = vtmp5;
6482     uzp1(vlo0, T16B, vtmp0, vtmp1);
6483     uzp1(vlo1, T16B, vtmp2, vtmp3);
6484     // Merge bits...
6485     orr(vtmp0, T16B, vtmp0, vtmp1);
6486     orr(vtmp2, T16B, vtmp2, vtmp3);
6487     // Extract merged upper bytes.
6488     FloatRegister vhix = vtmp0;
6489     uzp2(vhix, T16B, vtmp0, vtmp2);
6490     // ISO-check on hi-parts (all zero).
6491     //                          ASCII-check on lo-parts (no sign).
6492     FloatRegister vlox = vtmp1; // Merge lower bytes.
6493                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6494     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6495     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6496     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6497                                 ASCII(orr(chk, chk, max));
6498     cbnz(chk, FAIL_32);
6499     subw(cnt, cnt, 32);
6500     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6501     b(LOOP_32);
6502   }
6503   BIND(FAIL_32);
6504   sub(src, src, 64);
6505   BIND(DONE_32);
6506 
6507   Label LOOP_8, SKIP_8;
6508 
6509   BIND(LOOP_8);
6510   {
6511     cmpw(cnt, 8);
6512     br(LT, SKIP_8);
6513     FloatRegister vhi = vtmp0;
6514     FloatRegister vlo = vtmp1;
6515     ld1(vtmp3, T8H, src);
6516     uzp1(vlo, T16B, vtmp3, vtmp3);
6517     uzp2(vhi, T16B, vtmp3, vtmp3);
6518     // ISO-check on hi-parts (all zero).
6519     //                          ASCII-check on lo-parts (no sign).
6520                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6521     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6522                                 ASCII(umov(max, vtmp2, B, 0));
6523                                 ASCII(orr(chk, chk, max));
6524     cbnz(chk, SKIP_8);
6525 
6526     strd(vlo, Address(post(dst, 8)));
6527     subw(cnt, cnt, 8);
6528     add(src, src, 16);
6529     b(LOOP_8);
6530   }
6531   BIND(SKIP_8);
6532 
6533 #undef ASCII
6534 
6535   Label LOOP, DONE;
6536 
6537   cbz(cnt, DONE);
6538   BIND(LOOP);
6539   {
6540     Register chr = rscratch1;
6541     ldrh(chr, Address(post(src, 2)));
6542     tst(chr, ascii ? 0xff80 : 0xff00);
6543     br(NE, DONE);
6544     strb(chr, Address(post(dst, 1)));
6545     subs(cnt, cnt, 1);
6546     br(GT, LOOP);
6547   }
6548   BIND(DONE);
6549   // Return index where we stopped.
6550   subw(res, len, cnt);
6551 }
6552 
6553 // Inflate byte[] array to char[].
6554 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6555 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6556                                            FloatRegister vtmp1, FloatRegister vtmp2,
6557                                            FloatRegister vtmp3, Register tmp4) {
6558   Label big, done, after_init, to_stub;
6559 
6560   assert_different_registers(src, dst, len, tmp4, rscratch1);
6561 
6562   fmovd(vtmp1, 0.0);
6563   lsrw(tmp4, len, 3);
6564   bind(after_init);
6565   cbnzw(tmp4, big);
6566   // Short string: less than 8 bytes.
6567   {
6568     Label loop, tiny;
6569 
6570     cmpw(len, 4);
6571     br(LT, tiny);
6572     // Use SIMD to do 4 bytes.
6573     ldrs(vtmp2, post(src, 4));
6574     zip1(vtmp3, T8B, vtmp2, vtmp1);
6575     subw(len, len, 4);
6576     strd(vtmp3, post(dst, 8));
6577 
6578     cbzw(len, done);
6579 
6580     // Do the remaining bytes by steam.
6581     bind(loop);
6582     ldrb(tmp4, post(src, 1));
6583     strh(tmp4, post(dst, 2));
6584     subw(len, len, 1);
6585 
6586     bind(tiny);
6587     cbnz(len, loop);
6588 
6589     b(done);
6590   }
6591 
6592   if (SoftwarePrefetchHintDistance >= 0) {
6593     bind(to_stub);
6594       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6595       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6596       address tpc = trampoline_call(stub);
6597       if (tpc == nullptr) {
6598         DEBUG_ONLY(reset_labels(big, done));
6599         postcond(pc() == badAddress);
6600         return nullptr;
6601       }
6602       b(after_init);
6603   }
6604 
6605   // Unpack the bytes 8 at a time.
6606   bind(big);
6607   {
6608     Label loop, around, loop_last, loop_start;
6609 
6610     if (SoftwarePrefetchHintDistance >= 0) {
6611       const int large_loop_threshold = (64 + 16)/8;
6612       ldrd(vtmp2, post(src, 8));
6613       andw(len, len, 7);
6614       cmp(tmp4, (u1)large_loop_threshold);
6615       br(GE, to_stub);
6616       b(loop_start);
6617 
6618       bind(loop);
6619       ldrd(vtmp2, post(src, 8));
6620       bind(loop_start);
6621       subs(tmp4, tmp4, 1);
6622       br(EQ, loop_last);
6623       zip1(vtmp2, T16B, vtmp2, vtmp1);
6624       ldrd(vtmp3, post(src, 8));
6625       st1(vtmp2, T8H, post(dst, 16));
6626       subs(tmp4, tmp4, 1);
6627       zip1(vtmp3, T16B, vtmp3, vtmp1);
6628       st1(vtmp3, T8H, post(dst, 16));
6629       br(NE, loop);
6630       b(around);
6631       bind(loop_last);
6632       zip1(vtmp2, T16B, vtmp2, vtmp1);
6633       st1(vtmp2, T8H, post(dst, 16));
6634       bind(around);
6635       cbz(len, done);
6636     } else {
6637       andw(len, len, 7);
6638       bind(loop);
6639       ldrd(vtmp2, post(src, 8));
6640       sub(tmp4, tmp4, 1);
6641       zip1(vtmp3, T16B, vtmp2, vtmp1);
6642       st1(vtmp3, T8H, post(dst, 16));
6643       cbnz(tmp4, loop);
6644     }
6645   }
6646 
6647   // Do the tail of up to 8 bytes.
6648   add(src, src, len);
6649   ldrd(vtmp3, Address(src, -8));
6650   add(dst, dst, len, ext::uxtw, 1);
6651   zip1(vtmp3, T16B, vtmp3, vtmp1);
6652   strq(vtmp3, Address(dst, -16));
6653 
6654   bind(done);
6655   postcond(pc() != badAddress);
6656   return pc();
6657 }
6658 
6659 // Compress char[] array to byte[].
6660 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6661 // Return the array length if every element in array can be encoded,
6662 // otherwise, the index of first non-latin1 (> 0xff) character.
6663 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6664                                          Register res,
6665                                          FloatRegister tmp0, FloatRegister tmp1,
6666                                          FloatRegister tmp2, FloatRegister tmp3,
6667                                          FloatRegister tmp4, FloatRegister tmp5) {
6668   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6669 }
6670 
6671 // java.math.round(double a)
6672 // Returns the closest long to the argument, with ties rounding to
6673 // positive infinity.  This requires some fiddling for corner
6674 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6675 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6676                                        FloatRegister ftmp) {
6677   Label DONE;
6678   BLOCK_COMMENT("java_round_double: { ");
6679   fmovd(rscratch1, src);
6680   // Use RoundToNearestTiesAway unless src small and -ve.
6681   fcvtasd(dst, src);
6682   // Test if src >= 0 || abs(src) >= 0x1.0p52
6683   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6684   mov(rscratch2, julong_cast(0x1.0p52));
6685   cmp(rscratch1, rscratch2);
6686   br(HS, DONE); {
6687     // src < 0 && abs(src) < 0x1.0p52
6688     // src may have a fractional part, so add 0.5
6689     fmovd(ftmp, 0.5);
6690     faddd(ftmp, src, ftmp);
6691     // Convert double to jlong, use RoundTowardsNegative
6692     fcvtmsd(dst, ftmp);
6693   }
6694   bind(DONE);
6695   BLOCK_COMMENT("} java_round_double");
6696 }
6697 
6698 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6699                                       FloatRegister ftmp) {
6700   Label DONE;
6701   BLOCK_COMMENT("java_round_float: { ");
6702   fmovs(rscratch1, src);
6703   // Use RoundToNearestTiesAway unless src small and -ve.
6704   fcvtassw(dst, src);
6705   // Test if src >= 0 || abs(src) >= 0x1.0p23
6706   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6707   mov(rscratch2, jint_cast(0x1.0p23f));
6708   cmp(rscratch1, rscratch2);
6709   br(HS, DONE); {
6710     // src < 0 && |src| < 0x1.0p23
6711     // src may have a fractional part, so add 0.5
6712     fmovs(ftmp, 0.5f);
6713     fadds(ftmp, src, ftmp);
6714     // Convert float to jint, use RoundTowardsNegative
6715     fcvtmssw(dst, ftmp);
6716   }
6717   bind(DONE);
6718   BLOCK_COMMENT("} java_round_float");
6719 }
6720 
6721 // get_thread() can be called anywhere inside generated code so we
6722 // need to save whatever non-callee save context might get clobbered
6723 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6724 // the call setup code.
6725 //
6726 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6727 // On other systems, the helper is a usual C function.
6728 //
6729 void MacroAssembler::get_thread(Register dst) {
6730   RegSet saved_regs =
6731     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6732     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6733 
6734   protect_return_address();
6735   push(saved_regs, sp);
6736 
6737   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6738   blr(lr);
6739   if (dst != c_rarg0) {
6740     mov(dst, c_rarg0);
6741   }
6742 
6743   pop(saved_regs, sp);
6744   authenticate_return_address();
6745 }
6746 
6747 void MacroAssembler::cache_wb(Address line) {
6748   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6749   assert(line.index() == noreg, "index should be noreg");
6750   assert(line.offset() == 0, "offset should be 0");
6751   // would like to assert this
6752   // assert(line._ext.shift == 0, "shift should be zero");
6753   if (VM_Version::supports_dcpop()) {
6754     // writeback using clear virtual address to point of persistence
6755     dc(Assembler::CVAP, line.base());
6756   } else {
6757     // no need to generate anything as Unsafe.writebackMemory should
6758     // never invoke this stub
6759   }
6760 }
6761 
6762 void MacroAssembler::cache_wbsync(bool is_pre) {
6763   // we only need a barrier post sync
6764   if (!is_pre) {
6765     membar(Assembler::AnyAny);
6766   }
6767 }
6768 
6769 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6770   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6771     return;
6772   }
6773   // Make sure that native code does not change SVE vector length.
6774   Label verify_ok;
6775   movw(tmp, zr);
6776   sve_inc(tmp, B);
6777   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6778   br(EQ, verify_ok);
6779   stop("Error: SVE vector length has changed since jvm startup");
6780   bind(verify_ok);
6781 }
6782 
6783 void MacroAssembler::verify_ptrue() {
6784   Label verify_ok;
6785   if (!UseSVE) {
6786     return;
6787   }
6788   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6789   sve_dec(rscratch1, B);
6790   cbz(rscratch1, verify_ok);
6791   stop("Error: the preserved predicate register (p7) elements are not all true");
6792   bind(verify_ok);
6793 }
6794 
6795 void MacroAssembler::safepoint_isb() {
6796   isb();
6797 #ifndef PRODUCT
6798   if (VerifyCrossModifyFence) {
6799     // Clear the thread state.
6800     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6801   }
6802 #endif
6803 }
6804 
6805 #ifndef PRODUCT
6806 void MacroAssembler::verify_cross_modify_fence_not_required() {
6807   if (VerifyCrossModifyFence) {
6808     // Check if thread needs a cross modify fence.
6809     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6810     Label fence_not_required;
6811     cbz(rscratch1, fence_not_required);
6812     // If it does then fail.
6813     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6814     mov(c_rarg0, rthread);
6815     blr(rscratch1);
6816     bind(fence_not_required);
6817   }
6818 }
6819 #endif
6820 
6821 void MacroAssembler::spin_wait() {
6822   block_comment("spin_wait {");
6823   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6824     switch (VM_Version::spin_wait_desc().inst()) {
6825       case SpinWait::NOP:
6826         nop();
6827         break;
6828       case SpinWait::ISB:
6829         isb();
6830         break;
6831       case SpinWait::YIELD:
6832         yield();
6833         break;
6834       case SpinWait::SB:
6835         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6836         sb();
6837         break;
6838       default:
6839         ShouldNotReachHere();
6840     }
6841   }
6842   block_comment("}");
6843 }
6844 
6845 // Stack frame creation/removal
6846 
6847 void MacroAssembler::enter(bool strip_ret_addr) {
6848   if (strip_ret_addr) {
6849     // Addresses can only be signed once. If there are multiple nested frames being created
6850     // in the same function, then the return address needs stripping first.
6851     strip_return_address();
6852   }
6853   protect_return_address();
6854   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6855   mov(rfp, sp);
6856 }
6857 
6858 void MacroAssembler::leave() {
6859   mov(sp, rfp);
6860   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6861   authenticate_return_address();
6862 }
6863 
6864 // ROP Protection
6865 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6866 // destroying stack frames or whenever directly loading/storing the LR to memory.
6867 // If ROP protection is not set then these functions are no-ops.
6868 // For more details on PAC see pauth_aarch64.hpp.
6869 
6870 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6871 // Uses value zero as the modifier.
6872 //
6873 void MacroAssembler::protect_return_address() {
6874   if (VM_Version::use_rop_protection()) {
6875     check_return_address();
6876     paciaz();
6877   }
6878 }
6879 
6880 // Sign the return value in the given register. Use before updating the LR in the existing stack
6881 // frame for the current function.
6882 // Uses value zero as the modifier.
6883 //
6884 void MacroAssembler::protect_return_address(Register return_reg) {
6885   if (VM_Version::use_rop_protection()) {
6886     check_return_address(return_reg);
6887     paciza(return_reg);
6888   }
6889 }
6890 
6891 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6892 // Uses value zero as the modifier.
6893 //
6894 void MacroAssembler::authenticate_return_address() {
6895   if (VM_Version::use_rop_protection()) {
6896     autiaz();
6897     check_return_address();
6898   }
6899 }
6900 
6901 // Authenticate the return value in the given register. Use before updating the LR in the existing
6902 // stack frame for the current function.
6903 // Uses value zero as the modifier.
6904 //
6905 void MacroAssembler::authenticate_return_address(Register return_reg) {
6906   if (VM_Version::use_rop_protection()) {
6907     autiza(return_reg);
6908     check_return_address(return_reg);
6909   }
6910 }
6911 
6912 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6913 // there is no guaranteed way of authenticating the LR.
6914 //
6915 void MacroAssembler::strip_return_address() {
6916   if (VM_Version::use_rop_protection()) {
6917     xpaclri();
6918   }
6919 }
6920 
6921 #ifndef PRODUCT
6922 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6923 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6924 // it is difficult to debug back to the callee function.
6925 // This function simply loads from the address in the given register.
6926 // Use directly after authentication to catch authentication failures.
6927 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6928 //
6929 void MacroAssembler::check_return_address(Register return_reg) {
6930   if (VM_Version::use_rop_protection()) {
6931     ldr(zr, Address(return_reg));
6932   }
6933 }
6934 #endif
6935 
6936 // The java_calling_convention describes stack locations as ideal slots on
6937 // a frame with no abi restrictions. Since we must observe abi restrictions
6938 // (like the placement of the register window) the slots must be biased by
6939 // the following value.
6940 static int reg2offset_in(VMReg r) {
6941   // Account for saved rfp and lr
6942   // This should really be in_preserve_stack_slots
6943   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6944 }
6945 
6946 static int reg2offset_out(VMReg r) {
6947   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6948 }
6949 
6950 // On 64bit we will store integer like items to the stack as
6951 // 64bits items (AArch64 ABI) even though java would only store
6952 // 32bits for a parameter. On 32bit it will simply be 32bits
6953 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6954 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6955   if (src.first()->is_stack()) {
6956     if (dst.first()->is_stack()) {
6957       // stack to stack
6958       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6959       str(tmp, Address(sp, reg2offset_out(dst.first())));
6960     } else {
6961       // stack to reg
6962       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6963     }
6964   } else if (dst.first()->is_stack()) {
6965     // reg to stack
6966     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6967   } else {
6968     if (dst.first() != src.first()) {
6969       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6970     }
6971   }
6972 }
6973 
6974 // An oop arg. Must pass a handle not the oop itself
6975 void MacroAssembler::object_move(
6976                         OopMap* map,
6977                         int oop_handle_offset,
6978                         int framesize_in_slots,
6979                         VMRegPair src,
6980                         VMRegPair dst,
6981                         bool is_receiver,
6982                         int* receiver_offset) {
6983 
6984   // must pass a handle. First figure out the location we use as a handle
6985 
6986   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
6987 
6988   // See if oop is null if it is we need no handle
6989 
6990   if (src.first()->is_stack()) {
6991 
6992     // Oop is already on the stack as an argument
6993     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
6994     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
6995     if (is_receiver) {
6996       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
6997     }
6998 
6999     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7000     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7001     // conditionally move a null
7002     cmp(rscratch1, zr);
7003     csel(rHandle, zr, rHandle, Assembler::EQ);
7004   } else {
7005 
7006     // Oop is in an a register we must store it to the space we reserve
7007     // on the stack for oop_handles and pass a handle if oop is non-null
7008 
7009     const Register rOop = src.first()->as_Register();
7010     int oop_slot;
7011     if (rOop == j_rarg0)
7012       oop_slot = 0;
7013     else if (rOop == j_rarg1)
7014       oop_slot = 1;
7015     else if (rOop == j_rarg2)
7016       oop_slot = 2;
7017     else if (rOop == j_rarg3)
7018       oop_slot = 3;
7019     else if (rOop == j_rarg4)
7020       oop_slot = 4;
7021     else if (rOop == j_rarg5)
7022       oop_slot = 5;
7023     else if (rOop == j_rarg6)
7024       oop_slot = 6;
7025     else {
7026       assert(rOop == j_rarg7, "wrong register");
7027       oop_slot = 7;
7028     }
7029 
7030     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7031     int offset = oop_slot*VMRegImpl::stack_slot_size;
7032 
7033     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7034     // Store oop in handle area, may be null
7035     str(rOop, Address(sp, offset));
7036     if (is_receiver) {
7037       *receiver_offset = offset;
7038     }
7039 
7040     cmp(rOop, zr);
7041     lea(rHandle, Address(sp, offset));
7042     // conditionally move a null
7043     csel(rHandle, zr, rHandle, Assembler::EQ);
7044   }
7045 
7046   // If arg is on the stack then place it otherwise it is already in correct reg.
7047   if (dst.first()->is_stack()) {
7048     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7049   }
7050 }
7051 
7052 // A float arg may have to do float reg int reg conversion
7053 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7054  if (src.first()->is_stack()) {
7055     if (dst.first()->is_stack()) {
7056       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7057       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7058     } else {
7059       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7060     }
7061   } else if (src.first() != dst.first()) {
7062     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7063       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7064     else
7065       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7066   }
7067 }
7068 
7069 // A long move
7070 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7071   if (src.first()->is_stack()) {
7072     if (dst.first()->is_stack()) {
7073       // stack to stack
7074       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7075       str(tmp, Address(sp, reg2offset_out(dst.first())));
7076     } else {
7077       // stack to reg
7078       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7079     }
7080   } else if (dst.first()->is_stack()) {
7081     // reg to stack
7082     // Do we really have to sign extend???
7083     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7084     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7085   } else {
7086     if (dst.first() != src.first()) {
7087       mov(dst.first()->as_Register(), src.first()->as_Register());
7088     }
7089   }
7090 }
7091 
7092 
7093 // A double move
7094 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7095  if (src.first()->is_stack()) {
7096     if (dst.first()->is_stack()) {
7097       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7098       str(tmp, Address(sp, reg2offset_out(dst.first())));
7099     } else {
7100       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7101     }
7102   } else if (src.first() != dst.first()) {
7103     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7104       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7105     else
7106       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7107   }
7108 }
7109 
7110 // Implements fast-locking.
7111 //
7112 //  - obj: the object to be locked
7113 //  - t1, t2, t3: temporary registers, will be destroyed
7114 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7115 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7116   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7117 
7118   Label push;
7119   const Register top = t1;
7120   const Register mark = t2;
7121   const Register t = t3;
7122 
7123   // Preload the markWord. It is important that this is the first
7124   // instruction emitted as it is part of C1's null check semantics.
7125   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7126 
7127   if (UseObjectMonitorTable) {
7128     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7129     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7130   }
7131 
7132   if (DiagnoseSyncOnValueBasedClasses != 0) {
7133     load_klass(t1, obj);
7134     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7135     tst(t1, KlassFlags::_misc_is_value_based_class);
7136     br(Assembler::NE, slow);
7137   }
7138 
7139   // Check if the lock-stack is full.
7140   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7141   cmpw(top, (unsigned)LockStack::end_offset());
7142   br(Assembler::GE, slow);
7143 
7144   // Check for recursion.
7145   subw(t, top, oopSize);
7146   ldr(t, Address(rthread, t));
7147   cmp(obj, t);
7148   br(Assembler::EQ, push);
7149 
7150   // Check header for monitor (0b10).
7151   tst(mark, markWord::monitor_value);
7152   br(Assembler::NE, slow);
7153 
7154   // Try to lock. Transition lock bits 0b01 => 0b00
7155   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7156   orr(mark, mark, markWord::unlocked_value);
7157   eor(t, mark, markWord::unlocked_value);
7158   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7159           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7160   br(Assembler::NE, slow);
7161 
7162   bind(push);
7163   // After successful lock, push object on lock-stack.
7164   str(obj, Address(rthread, top));
7165   addw(top, top, oopSize);
7166   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7167 }
7168 
7169 // Implements fast-unlocking.
7170 //
7171 // - obj: the object to be unlocked
7172 // - t1, t2, t3: temporary registers
7173 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7174 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7175   // cmpxchg clobbers rscratch1.
7176   assert_different_registers(obj, t1, t2, t3, rscratch1);
7177 
7178 #ifdef ASSERT
7179   {
7180     // Check for lock-stack underflow.
7181     Label stack_ok;
7182     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7183     cmpw(t1, (unsigned)LockStack::start_offset());
7184     br(Assembler::GE, stack_ok);
7185     STOP("Lock-stack underflow");
7186     bind(stack_ok);
7187   }
7188 #endif
7189 
7190   Label unlocked, push_and_slow;
7191   const Register top = t1;
7192   const Register mark = t2;
7193   const Register t = t3;
7194 
7195   // Check if obj is top of lock-stack.
7196   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7197   subw(top, top, oopSize);
7198   ldr(t, Address(rthread, top));
7199   cmp(obj, t);
7200   br(Assembler::NE, slow);
7201 
7202   // Pop lock-stack.
7203   DEBUG_ONLY(str(zr, Address(rthread, top));)
7204   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7205 
7206   // Check if recursive.
7207   subw(t, top, oopSize);
7208   ldr(t, Address(rthread, t));
7209   cmp(obj, t);
7210   br(Assembler::EQ, unlocked);
7211 
7212   // Not recursive. Check header for monitor (0b10).
7213   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7214   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7215 
7216 #ifdef ASSERT
7217   // Check header not unlocked (0b01).
7218   Label not_unlocked;
7219   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7220   stop("fast_unlock already unlocked");
7221   bind(not_unlocked);
7222 #endif
7223 
7224   // Try to unlock. Transition lock bits 0b00 => 0b01
7225   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7226   orr(t, mark, markWord::unlocked_value);
7227   cmpxchg(obj, mark, t, Assembler::xword,
7228           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7229   br(Assembler::EQ, unlocked);
7230 
7231   bind(push_and_slow);
7232   // Restore lock-stack and handle the unlock in runtime.
7233   DEBUG_ONLY(str(obj, Address(rthread, top));)
7234   addw(top, top, oopSize);
7235   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7236   b(slow);
7237 
7238   bind(unlocked);
7239 }