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 (!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 {
1006     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1007     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1008     cmpw(tmp1, tmp2);
1009   }
1010 
1011   Label dont;
1012   br(Assembler::EQ, dont);
1013   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1014   bind(dont);
1015   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1016 
1017   return uep_offset;
1018 }
1019 
1020 // Implementation of call_VM versions
1021 
1022 void MacroAssembler::call_VM(Register oop_result,
1023                              address entry_point,
1024                              bool check_exceptions) {
1025   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1026 }
1027 
1028 void MacroAssembler::call_VM(Register oop_result,
1029                              address entry_point,
1030                              Register arg_1,
1031                              bool check_exceptions) {
1032   pass_arg1(this, arg_1);
1033   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1034 }
1035 
1036 void MacroAssembler::call_VM(Register oop_result,
1037                              address entry_point,
1038                              Register arg_1,
1039                              Register arg_2,
1040                              bool check_exceptions) {
1041   assert_different_registers(arg_1, c_rarg2);
1042   pass_arg2(this, arg_2);
1043   pass_arg1(this, arg_1);
1044   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1045 }
1046 
1047 void MacroAssembler::call_VM(Register oop_result,
1048                              address entry_point,
1049                              Register arg_1,
1050                              Register arg_2,
1051                              Register arg_3,
1052                              bool check_exceptions) {
1053   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1054   assert_different_registers(arg_2, c_rarg3);
1055   pass_arg3(this, arg_3);
1056 
1057   pass_arg2(this, arg_2);
1058 
1059   pass_arg1(this, arg_1);
1060   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1061 }
1062 
1063 void MacroAssembler::call_VM(Register oop_result,
1064                              Register last_java_sp,
1065                              address entry_point,
1066                              int number_of_arguments,
1067                              bool check_exceptions) {
1068   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1069 }
1070 
1071 void MacroAssembler::call_VM(Register oop_result,
1072                              Register last_java_sp,
1073                              address entry_point,
1074                              Register arg_1,
1075                              bool check_exceptions) {
1076   pass_arg1(this, arg_1);
1077   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1078 }
1079 
1080 void MacroAssembler::call_VM(Register oop_result,
1081                              Register last_java_sp,
1082                              address entry_point,
1083                              Register arg_1,
1084                              Register arg_2,
1085                              bool check_exceptions) {
1086 
1087   assert_different_registers(arg_1, c_rarg2);
1088   pass_arg2(this, arg_2);
1089   pass_arg1(this, arg_1);
1090   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1091 }
1092 
1093 void MacroAssembler::call_VM(Register oop_result,
1094                              Register last_java_sp,
1095                              address entry_point,
1096                              Register arg_1,
1097                              Register arg_2,
1098                              Register arg_3,
1099                              bool check_exceptions) {
1100   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1101   assert_different_registers(arg_2, c_rarg3);
1102   pass_arg3(this, arg_3);
1103   pass_arg2(this, arg_2);
1104   pass_arg1(this, arg_1);
1105   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1106 }
1107 
1108 
1109 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1110   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1111   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1112   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1113 }
1114 
1115 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1116   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1117   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1118 }
1119 
1120 void MacroAssembler::align(int modulus) {
1121   align(modulus, offset());
1122 }
1123 
1124 // Ensure that the code at target bytes offset from the current offset() is aligned
1125 // according to modulus.
1126 void MacroAssembler::align(int modulus, int target) {
1127   int delta = target - offset();
1128   while ((offset() + delta) % modulus != 0) nop();
1129 }
1130 
1131 void MacroAssembler::post_call_nop() {
1132   if (!Continuations::enabled()) {
1133     return;
1134   }
1135   InstructionMark im(this);
1136   relocate(post_call_nop_Relocation::spec());
1137   InlineSkippedInstructionsCounter skipCounter(this);
1138   nop();
1139   movk(zr, 0);
1140   movk(zr, 0);
1141 }
1142 
1143 // these are no-ops overridden by InterpreterMacroAssembler
1144 
1145 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1146 
1147 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1148 
1149 // Look up the method for a megamorphic invokeinterface call.
1150 // The target method is determined by <intf_klass, itable_index>.
1151 // The receiver klass is in recv_klass.
1152 // On success, the result will be in method_result, and execution falls through.
1153 // On failure, execution transfers to the given label.
1154 void MacroAssembler::lookup_interface_method(Register recv_klass,
1155                                              Register intf_klass,
1156                                              RegisterOrConstant itable_index,
1157                                              Register method_result,
1158                                              Register scan_temp,
1159                                              Label& L_no_such_interface,
1160                          bool return_method) {
1161   assert_different_registers(recv_klass, intf_klass, scan_temp);
1162   assert_different_registers(method_result, intf_klass, scan_temp);
1163   assert(recv_klass != method_result || !return_method,
1164      "recv_klass can be destroyed when method isn't needed");
1165   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1166          "caller must use same register for non-constant itable index as for method");
1167 
1168   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1169   int vtable_base = in_bytes(Klass::vtable_start_offset());
1170   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1171   int scan_step   = itableOffsetEntry::size() * wordSize;
1172   int vte_size    = vtableEntry::size_in_bytes();
1173   assert(vte_size == wordSize, "else adjust times_vte_scale");
1174 
1175   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1176 
1177   // Could store the aligned, prescaled offset in the klass.
1178   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1179   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1180   add(scan_temp, scan_temp, vtable_base);
1181 
1182   if (return_method) {
1183     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1184     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1185     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1186     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1187     if (itentry_off)
1188       add(recv_klass, recv_klass, itentry_off);
1189   }
1190 
1191   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1192   //   if (scan->interface() == intf) {
1193   //     result = (klass + scan->offset() + itable_index);
1194   //   }
1195   // }
1196   Label search, found_method;
1197 
1198   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1199   cmp(intf_klass, method_result);
1200   br(Assembler::EQ, found_method);
1201   bind(search);
1202   // Check that the previous entry is non-null.  A null entry means that
1203   // the receiver class doesn't implement the interface, and wasn't the
1204   // same as when the caller was compiled.
1205   cbz(method_result, L_no_such_interface);
1206   if (itableOffsetEntry::interface_offset() != 0) {
1207     add(scan_temp, scan_temp, scan_step);
1208     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1209   } else {
1210     ldr(method_result, Address(pre(scan_temp, scan_step)));
1211   }
1212   cmp(intf_klass, method_result);
1213   br(Assembler::NE, search);
1214 
1215   bind(found_method);
1216 
1217   // Got a hit.
1218   if (return_method) {
1219     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1220     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1221   }
1222 }
1223 
1224 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1225 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1226 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1227 // The target method is determined by <holder_klass, itable_index>.
1228 // The receiver klass is in recv_klass.
1229 // On success, the result will be in method_result, and execution falls through.
1230 // On failure, execution transfers to the given label.
1231 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1232                                                   Register holder_klass,
1233                                                   Register resolved_klass,
1234                                                   Register method_result,
1235                                                   Register temp_itbl_klass,
1236                                                   Register scan_temp,
1237                                                   int itable_index,
1238                                                   Label& L_no_such_interface) {
1239   // 'method_result' is only used as output register at the very end of this method.
1240   // Until then we can reuse it as 'holder_offset'.
1241   Register holder_offset = method_result;
1242   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1243 
1244   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1245   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1246   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1247   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1248 
1249   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1250 
1251   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1252   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1253   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1254   // temp_itbl_klass = itable[0]._interface;
1255   int vtblEntrySize = vtableEntry::size_in_bytes();
1256   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1257   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1258   mov(holder_offset, zr);
1259   // scan_temp = &(itable[0]._interface)
1260   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1261 
1262   // Initial checks:
1263   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1264   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1265   //   - if (itable[0] == 0), no such interface
1266   cmp(resolved_klass, holder_klass);
1267   br(Assembler::NE, L_loop_search_resolved_entry);
1268   cmp(holder_klass, temp_itbl_klass);
1269   br(Assembler::EQ, L_holder_found);
1270   cbz(temp_itbl_klass, L_no_such_interface);
1271 
1272   // Loop: Look for holder_klass record in itable
1273   //   do {
1274   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1275   //     if (temp_itbl_klass == holder_klass) {
1276   //       goto L_holder_found; // Found!
1277   //     }
1278   //   } while (temp_itbl_klass != 0);
1279   //   goto L_no_such_interface // Not found.
1280   Label L_search_holder;
1281   bind(L_search_holder);
1282     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1283     cmp(holder_klass, temp_itbl_klass);
1284     br(Assembler::EQ, L_holder_found);
1285     cbnz(temp_itbl_klass, L_search_holder);
1286 
1287   b(L_no_such_interface);
1288 
1289   // Loop: Look for resolved_class record in itable
1290   //   while (true) {
1291   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1292   //     if (temp_itbl_klass == 0) {
1293   //       goto L_no_such_interface;
1294   //     }
1295   //     if (temp_itbl_klass == resolved_klass) {
1296   //        goto L_resolved_found;  // Found!
1297   //     }
1298   //     if (temp_itbl_klass == holder_klass) {
1299   //        holder_offset = scan_temp;
1300   //     }
1301   //   }
1302   //
1303   Label L_loop_search_resolved;
1304   bind(L_loop_search_resolved);
1305     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1306   bind(L_loop_search_resolved_entry);
1307     cbz(temp_itbl_klass, L_no_such_interface);
1308     cmp(resolved_klass, temp_itbl_klass);
1309     br(Assembler::EQ, L_resolved_found);
1310     cmp(holder_klass, temp_itbl_klass);
1311     br(Assembler::NE, L_loop_search_resolved);
1312     mov(holder_offset, scan_temp);
1313     b(L_loop_search_resolved);
1314 
1315   // See if we already have a holder klass. If not, go and scan for it.
1316   bind(L_resolved_found);
1317   cbz(holder_offset, L_search_holder);
1318   mov(scan_temp, holder_offset);
1319 
1320   // Finally, scan_temp contains holder_klass vtable offset
1321   bind(L_holder_found);
1322   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1323   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1324     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1325   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1326 }
1327 
1328 // virtual method calling
1329 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1330                                            RegisterOrConstant vtable_index,
1331                                            Register method_result) {
1332   assert(vtableEntry::size() * wordSize == 8,
1333          "adjust the scaling in the code below");
1334   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1335 
1336   if (vtable_index.is_register()) {
1337     lea(method_result, Address(recv_klass,
1338                                vtable_index.as_register(),
1339                                Address::lsl(LogBytesPerWord)));
1340     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1341   } else {
1342     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1343     ldr(method_result,
1344         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1345   }
1346 }
1347 
1348 void MacroAssembler::check_klass_subtype(Register sub_klass,
1349                            Register super_klass,
1350                            Register temp_reg,
1351                            Label& L_success) {
1352   Label L_failure;
1353   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1354   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1355   bind(L_failure);
1356 }
1357 
1358 
1359 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1360                                                    Register super_klass,
1361                                                    Register temp_reg,
1362                                                    Label* L_success,
1363                                                    Label* L_failure,
1364                                                    Label* L_slow_path,
1365                                                    Register super_check_offset) {
1366   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1367   bool must_load_sco = ! super_check_offset->is_valid();
1368   if (must_load_sco) {
1369     assert(temp_reg != noreg, "supply either a temp or a register offset");
1370   }
1371 
1372   Label L_fallthrough;
1373   int label_nulls = 0;
1374   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1375   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1376   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1377   assert(label_nulls <= 1, "at most one null in the batch");
1378 
1379   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1380   Address super_check_offset_addr(super_klass, sco_offset);
1381 
1382   // Hacked jmp, which may only be used just before L_fallthrough.
1383 #define final_jmp(label)                                                \
1384   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1385   else                            b(label)                /*omit semi*/
1386 
1387   // If the pointers are equal, we are done (e.g., String[] elements).
1388   // This self-check enables sharing of secondary supertype arrays among
1389   // non-primary types such as array-of-interface.  Otherwise, each such
1390   // type would need its own customized SSA.
1391   // We move this check to the front of the fast path because many
1392   // type checks are in fact trivially successful in this manner,
1393   // so we get a nicely predicted branch right at the start of the check.
1394   cmp(sub_klass, super_klass);
1395   br(Assembler::EQ, *L_success);
1396 
1397   // Check the supertype display:
1398   if (must_load_sco) {
1399     ldrw(temp_reg, super_check_offset_addr);
1400     super_check_offset = temp_reg;
1401   }
1402 
1403   Address super_check_addr(sub_klass, super_check_offset);
1404   ldr(rscratch1, super_check_addr);
1405   cmp(super_klass, rscratch1); // load displayed supertype
1406   br(Assembler::EQ, *L_success);
1407 
1408   // This check has worked decisively for primary supers.
1409   // Secondary supers are sought in the super_cache ('super_cache_addr').
1410   // (Secondary supers are interfaces and very deeply nested subtypes.)
1411   // This works in the same check above because of a tricky aliasing
1412   // between the super_cache and the primary super display elements.
1413   // (The 'super_check_addr' can address either, as the case requires.)
1414   // Note that the cache is updated below if it does not help us find
1415   // what we need immediately.
1416   // So if it was a primary super, we can just fail immediately.
1417   // Otherwise, it's the slow path for us (no success at this point).
1418 
1419   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1420   if (L_failure == &L_fallthrough) {
1421     cbz(rscratch1, *L_slow_path);
1422   } else {
1423     cbnz(rscratch1, *L_failure);
1424     final_jmp(*L_slow_path);
1425   }
1426 
1427   bind(L_fallthrough);
1428 
1429 #undef final_jmp
1430 }
1431 
1432 // These two are taken from x86, but they look generally useful
1433 
1434 // scans count pointer sized words at [addr] for occurrence of value,
1435 // generic
1436 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1437                                 Register scratch) {
1438   Label Lloop, Lexit;
1439   cbz(count, Lexit);
1440   bind(Lloop);
1441   ldr(scratch, post(addr, wordSize));
1442   cmp(value, scratch);
1443   br(EQ, Lexit);
1444   sub(count, count, 1);
1445   cbnz(count, Lloop);
1446   bind(Lexit);
1447 }
1448 
1449 // scans count 4 byte words at [addr] for occurrence of value,
1450 // generic
1451 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1452                                 Register scratch) {
1453   Label Lloop, Lexit;
1454   cbz(count, Lexit);
1455   bind(Lloop);
1456   ldrw(scratch, post(addr, wordSize));
1457   cmpw(value, scratch);
1458   br(EQ, Lexit);
1459   sub(count, count, 1);
1460   cbnz(count, Lloop);
1461   bind(Lexit);
1462 }
1463 
1464 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1465                                                           Register super_klass,
1466                                                           Register temp_reg,
1467                                                           Register temp2_reg,
1468                                                           Label* L_success,
1469                                                           Label* L_failure,
1470                                                           bool set_cond_codes) {
1471   // NB! Callers may assume that, when temp2_reg is a valid register,
1472   // this code sets it to a nonzero value.
1473 
1474   assert_different_registers(sub_klass, super_klass, temp_reg);
1475   if (temp2_reg != noreg)
1476     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1477 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1478 
1479   Label L_fallthrough;
1480   int label_nulls = 0;
1481   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1482   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1483   assert(label_nulls <= 1, "at most one null in the batch");
1484 
1485   // a couple of useful fields in sub_klass:
1486   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1487   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1488   Address secondary_supers_addr(sub_klass, ss_offset);
1489   Address super_cache_addr(     sub_klass, sc_offset);
1490 
1491   BLOCK_COMMENT("check_klass_subtype_slow_path");
1492 
1493   // Do a linear scan of the secondary super-klass chain.
1494   // This code is rarely used, so simplicity is a virtue here.
1495   // The repne_scan instruction uses fixed registers, which we must spill.
1496   // Don't worry too much about pre-existing connections with the input regs.
1497 
1498   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1499   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1500 
1501   RegSet pushed_registers;
1502   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1503   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1504 
1505   if (super_klass != r0) {
1506     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1507   }
1508 
1509   push(pushed_registers, sp);
1510 
1511   // Get super_klass value into r0 (even if it was in r5 or r2).
1512   if (super_klass != r0) {
1513     mov(r0, super_klass);
1514   }
1515 
1516 #ifndef PRODUCT
1517   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1518 #endif //PRODUCT
1519 
1520   // We will consult the secondary-super array.
1521   ldr(r5, secondary_supers_addr);
1522   // Load the array length.
1523   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1524   // Skip to start of data.
1525   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1526 
1527   cmp(sp, zr); // Clear Z flag; SP is never zero
1528   // Scan R2 words at [R5] for an occurrence of R0.
1529   // Set NZ/Z based on last compare.
1530   repne_scan(r5, r0, r2, rscratch1);
1531 
1532   // Unspill the temp. registers:
1533   pop(pushed_registers, sp);
1534 
1535   br(Assembler::NE, *L_failure);
1536 
1537   // Success.  Cache the super we found and proceed in triumph.
1538 
1539   if (UseSecondarySupersCache) {
1540     str(super_klass, super_cache_addr);
1541   }
1542 
1543   if (L_success != &L_fallthrough) {
1544     b(*L_success);
1545   }
1546 
1547 #undef IS_A_TEMP
1548 
1549   bind(L_fallthrough);
1550 }
1551 
1552 // If Register r is invalid, remove a new register from
1553 // available_regs, and add new register to regs_to_push.
1554 Register MacroAssembler::allocate_if_noreg(Register r,
1555                                   RegSetIterator<Register> &available_regs,
1556                                   RegSet &regs_to_push) {
1557   if (!r->is_valid()) {
1558     r = *available_regs++;
1559     regs_to_push += r;
1560   }
1561   return r;
1562 }
1563 
1564 // check_klass_subtype_slow_path_table() looks for super_klass in the
1565 // hash table belonging to super_klass, branching to L_success or
1566 // L_failure as appropriate. This is essentially a shim which
1567 // allocates registers as necessary then calls
1568 // lookup_secondary_supers_table() to do the work. Any of the temp
1569 // regs may be noreg, in which case this logic will chooses some
1570 // registers push and pop them from the stack.
1571 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1572                                                          Register super_klass,
1573                                                          Register temp_reg,
1574                                                          Register temp2_reg,
1575                                                          Register temp3_reg,
1576                                                          Register result_reg,
1577                                                          FloatRegister vtemp,
1578                                                          Label* L_success,
1579                                                          Label* L_failure,
1580                                                          bool set_cond_codes) {
1581   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1582 
1583   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1584 
1585   Label L_fallthrough;
1586   int label_nulls = 0;
1587   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1588   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1589   assert(label_nulls <= 1, "at most one null in the batch");
1590 
1591   BLOCK_COMMENT("check_klass_subtype_slow_path");
1592 
1593   RegSetIterator<Register> available_regs
1594     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1595 
1596   RegSet pushed_regs;
1597 
1598   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1599   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1600   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1601   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1602 
1603   push(pushed_regs, sp);
1604 
1605   lookup_secondary_supers_table_var(sub_klass,
1606                                     super_klass,
1607                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1608                                     nullptr);
1609   cmp(result_reg, zr);
1610 
1611   // Unspill the temp. registers:
1612   pop(pushed_regs, sp);
1613 
1614   // NB! Callers may assume that, when set_cond_codes is true, this
1615   // code sets temp2_reg to a nonzero value.
1616   if (set_cond_codes) {
1617     mov(temp2_reg, 1);
1618   }
1619 
1620   br(Assembler::NE, *L_failure);
1621 
1622   if (L_success != &L_fallthrough) {
1623     b(*L_success);
1624   }
1625 
1626   bind(L_fallthrough);
1627 }
1628 
1629 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1630                                                    Register super_klass,
1631                                                    Register temp_reg,
1632                                                    Register temp2_reg,
1633                                                    Label* L_success,
1634                                                    Label* L_failure,
1635                                                    bool set_cond_codes) {
1636   if (UseSecondarySupersTable) {
1637     check_klass_subtype_slow_path_table
1638       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1639        /*vtemp*/fnoreg,
1640        L_success, L_failure, set_cond_codes);
1641   } else {
1642     check_klass_subtype_slow_path_linear
1643       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1644   }
1645 }
1646 
1647 
1648 // Ensure that the inline code and the stub are using the same registers.
1649 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1650 do {                                                               \
1651   assert(r_super_klass  == r0                                   && \
1652          r_array_base   == r1                                   && \
1653          r_array_length == r2                                   && \
1654          (r_array_index == r3        || r_array_index == noreg) && \
1655          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1656          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1657          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1658 } while(0)
1659 
1660 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1661                                                          Register r_super_klass,
1662                                                          Register temp1,
1663                                                          Register temp2,
1664                                                          Register temp3,
1665                                                          FloatRegister vtemp,
1666                                                          Register result,
1667                                                          u1 super_klass_slot,
1668                                                          bool stub_is_near) {
1669   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1670 
1671   Label L_fallthrough;
1672 
1673   BLOCK_COMMENT("lookup_secondary_supers_table {");
1674 
1675   const Register
1676     r_array_base   = temp1, // r1
1677     r_array_length = temp2, // r2
1678     r_array_index  = temp3, // r3
1679     r_bitmap       = rscratch2;
1680 
1681   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1682 
1683   u1 bit = super_klass_slot;
1684 
1685   // Make sure that result is nonzero if the TBZ below misses.
1686   mov(result, 1);
1687 
1688   // We're going to need the bitmap in a vector reg and in a core reg,
1689   // so load both now.
1690   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1691   if (bit != 0) {
1692     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1693   }
1694   // First check the bitmap to see if super_klass might be present. If
1695   // the bit is zero, we are certain that super_klass is not one of
1696   // the secondary supers.
1697   tbz(r_bitmap, bit, L_fallthrough);
1698 
1699   // Get the first array index that can contain super_klass into r_array_index.
1700   if (bit != 0) {
1701     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1702     cnt(vtemp, T8B, vtemp);
1703     addv(vtemp, T8B, vtemp);
1704     fmovd(r_array_index, vtemp);
1705   } else {
1706     mov(r_array_index, (u1)1);
1707   }
1708   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1709 
1710   // We will consult the secondary-super array.
1711   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1712 
1713   // The value i in r_array_index is >= 1, so even though r_array_base
1714   // points to the length, we don't need to adjust it to point to the
1715   // data.
1716   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1717   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1718 
1719   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1720   eor(result, result, r_super_klass);
1721   cbz(result, L_fallthrough); // Found a match
1722 
1723   // Is there another entry to check? Consult the bitmap.
1724   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1725 
1726   // Linear probe.
1727   if (bit != 0) {
1728     ror(r_bitmap, r_bitmap, bit);
1729   }
1730 
1731   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1732   // The next slot to be inspected, by the stub we're about to call,
1733   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1734   // have been checked.
1735   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1736   if (stub_is_near) {
1737     bl(stub);
1738   } else {
1739     address call = trampoline_call(stub);
1740     if (call == nullptr) {
1741       return false; // trampoline allocation failed
1742     }
1743   }
1744 
1745   BLOCK_COMMENT("} lookup_secondary_supers_table");
1746 
1747   bind(L_fallthrough);
1748 
1749   if (VerifySecondarySupers) {
1750     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1751                                   temp1, temp2, result);      // r1, r2, r5
1752   }
1753   return true;
1754 }
1755 
1756 // At runtime, return 0 in result if r_super_klass is a superclass of
1757 // r_sub_klass, otherwise return nonzero. Use this version of
1758 // lookup_secondary_supers_table() if you don't know ahead of time
1759 // which superclass will be searched for. Used by interpreter and
1760 // runtime stubs. It is larger and has somewhat greater latency than
1761 // the version above, which takes a constant super_klass_slot.
1762 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1763                                                        Register r_super_klass,
1764                                                        Register temp1,
1765                                                        Register temp2,
1766                                                        Register temp3,
1767                                                        FloatRegister vtemp,
1768                                                        Register result,
1769                                                        Label *L_success) {
1770   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1771 
1772   Label L_fallthrough;
1773 
1774   BLOCK_COMMENT("lookup_secondary_supers_table {");
1775 
1776   const Register
1777     r_array_index = temp3,
1778     slot          = rscratch1,
1779     r_bitmap      = rscratch2;
1780 
1781   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1782 
1783   // Make sure that result is nonzero if the test below misses.
1784   mov(result, 1);
1785 
1786   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1787 
1788   // First check the bitmap to see if super_klass might be present. If
1789   // the bit is zero, we are certain that super_klass is not one of
1790   // the secondary supers.
1791 
1792   // This next instruction is equivalent to:
1793   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1794   // sub(temp2, tmp_reg, slot);
1795   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1796   lslv(temp2, r_bitmap, temp2);
1797   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1798 
1799   bool must_save_v0 = (vtemp == fnoreg);
1800   if (must_save_v0) {
1801     // temp1 and result are free, so use them to preserve vtemp
1802     vtemp = v0;
1803     mov(temp1,  vtemp, D, 0);
1804     mov(result, vtemp, D, 1);
1805   }
1806 
1807   // Get the first array index that can contain super_klass into r_array_index.
1808   mov(vtemp, D, 0, temp2);
1809   cnt(vtemp, T8B, vtemp);
1810   addv(vtemp, T8B, vtemp);
1811   mov(r_array_index, vtemp, D, 0);
1812 
1813   if (must_save_v0) {
1814     mov(vtemp, D, 0, temp1 );
1815     mov(vtemp, D, 1, result);
1816   }
1817 
1818   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1819 
1820   const Register
1821     r_array_base   = temp1,
1822     r_array_length = temp2;
1823 
1824   // The value i in r_array_index is >= 1, so even though r_array_base
1825   // points to the length, we don't need to adjust it to point to the
1826   // data.
1827   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1828   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1829 
1830   // We will consult the secondary-super array.
1831   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1832 
1833   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1834   eor(result, result, r_super_klass);
1835   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1836 
1837   // Is there another entry to check? Consult the bitmap.
1838   rorv(r_bitmap, r_bitmap, slot);
1839   // rol(r_bitmap, r_bitmap, 1);
1840   tbz(r_bitmap, 1, L_fallthrough);
1841 
1842   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1843   // The next slot to be inspected, by the logic we're about to call,
1844   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1845   // have been checked.
1846   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1847                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1848 
1849   BLOCK_COMMENT("} lookup_secondary_supers_table");
1850 
1851   bind(L_fallthrough);
1852 
1853   if (VerifySecondarySupers) {
1854     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1855                                   temp1, temp2, result);      // r1, r2, r5
1856   }
1857 
1858   if (L_success) {
1859     cbz(result, *L_success);
1860   }
1861 }
1862 
1863 // Called by code generated by check_klass_subtype_slow_path
1864 // above. This is called when there is a collision in the hashed
1865 // lookup in the secondary supers array.
1866 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1867                                                              Register r_array_base,
1868                                                              Register r_array_index,
1869                                                              Register r_bitmap,
1870                                                              Register temp1,
1871                                                              Register result,
1872                                                              bool is_stub) {
1873   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1874 
1875   const Register
1876     r_array_length = temp1,
1877     r_sub_klass    = noreg; // unused
1878 
1879   if (is_stub) {
1880     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1881   }
1882 
1883   Label L_fallthrough, L_huge;
1884 
1885   // Load the array length.
1886   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1887   // And adjust the array base to point to the data.
1888   // NB! Effectively increments current slot index by 1.
1889   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1890   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1891 
1892   // The bitmap is full to bursting.
1893   // Implicit invariant: BITMAP_FULL implies (length > 0)
1894   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1895   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1896   br(GT, L_huge);
1897 
1898   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1899   // current slot (at secondary_supers[r_array_index]) has not yet
1900   // been inspected, and r_array_index may be out of bounds if we
1901   // wrapped around the end of the array.
1902 
1903   { // This is conventional linear probing, but instead of terminating
1904     // when a null entry is found in the table, we maintain a bitmap
1905     // in which a 0 indicates missing entries.
1906     // As long as the bitmap is not completely full,
1907     // array_length == popcount(bitmap). The array_length check above
1908     // guarantees there are 0s in the bitmap, so the loop eventually
1909     // terminates.
1910     Label L_loop;
1911     bind(L_loop);
1912 
1913     // Check for wraparound.
1914     cmp(r_array_index, r_array_length);
1915     csel(r_array_index, zr, r_array_index, GE);
1916 
1917     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1918     eor(result, rscratch1, r_super_klass);
1919     cbz(result, L_fallthrough);
1920 
1921     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1922 
1923     ror(r_bitmap, r_bitmap, 1);
1924     add(r_array_index, r_array_index, 1);
1925     b(L_loop);
1926   }
1927 
1928   { // Degenerate case: more than 64 secondary supers.
1929     // FIXME: We could do something smarter here, maybe a vectorized
1930     // comparison or a binary search, but is that worth any added
1931     // complexity?
1932     bind(L_huge);
1933     cmp(sp, zr); // Clear Z flag; SP is never zero
1934     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1935     cset(result, NE); // result == 0 iff we got a match.
1936   }
1937 
1938   bind(L_fallthrough);
1939 }
1940 
1941 // Make sure that the hashed lookup and a linear scan agree.
1942 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1943                                                    Register r_super_klass,
1944                                                    Register temp1,
1945                                                    Register temp2,
1946                                                    Register result) {
1947   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1948 
1949   const Register
1950     r_array_base   = temp1,
1951     r_array_length = temp2;
1952 
1953   BLOCK_COMMENT("verify_secondary_supers_table {");
1954 
1955   // We will consult the secondary-super array.
1956   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1957 
1958   // Load the array length.
1959   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1960   // And adjust the array base to point to the data.
1961   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1962 
1963   cmp(sp, zr); // Clear Z flag; SP is never zero
1964   // Scan R2 words at [R5] for an occurrence of R0.
1965   // Set NZ/Z based on last compare.
1966   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1967   // rscratch1 == 0 iff we got a match.
1968   cset(rscratch1, NE);
1969 
1970   Label passed;
1971   cmp(result, zr);
1972   cset(result, NE); // normalize result to 0/1 for comparison
1973 
1974   cmp(rscratch1, result);
1975   br(EQ, passed);
1976   {
1977     mov(r0, r_super_klass);         // r0 <- r0
1978     mov(r1, r_sub_klass);           // r1 <- r4
1979     mov(r2, /*expected*/rscratch1); // r2 <- r8
1980     mov(r3, result);                // r3 <- r5
1981     mov(r4, (address)("mismatch")); // r4 <- const
1982     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
1983     should_not_reach_here();
1984   }
1985   bind(passed);
1986 
1987   BLOCK_COMMENT("} verify_secondary_supers_table");
1988 }
1989 
1990 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
1991   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
1992   assert_different_registers(klass, rthread, scratch);
1993 
1994   Label L_fallthrough, L_tmp;
1995   if (L_fast_path == nullptr) {
1996     L_fast_path = &L_fallthrough;
1997   } else if (L_slow_path == nullptr) {
1998     L_slow_path = &L_fallthrough;
1999   }
2000   // Fast path check: class is fully initialized
2001   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2002   ldarb(scratch, scratch);
2003   cmp(scratch, InstanceKlass::fully_initialized);
2004   br(Assembler::EQ, *L_fast_path);
2005 
2006   // Fast path check: current thread is initializer thread
2007   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2008   cmp(rthread, scratch);
2009 
2010   if (L_slow_path == &L_fallthrough) {
2011     br(Assembler::EQ, *L_fast_path);
2012     bind(*L_slow_path);
2013   } else if (L_fast_path == &L_fallthrough) {
2014     br(Assembler::NE, *L_slow_path);
2015     bind(*L_fast_path);
2016   } else {
2017     Unimplemented();
2018   }
2019 }
2020 
2021 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2022   if (!VerifyOops) return;
2023 
2024   // Pass register number to verify_oop_subroutine
2025   const char* b = nullptr;
2026   {
2027     ResourceMark rm;
2028     stringStream ss;
2029     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2030     b = code_string(ss.as_string());
2031   }
2032   BLOCK_COMMENT("verify_oop {");
2033 
2034   strip_return_address(); // This might happen within a stack frame.
2035   protect_return_address();
2036   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2037   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2038 
2039   mov(r0, reg);
2040   movptr(rscratch1, (uintptr_t)(address)b);
2041 
2042   // call indirectly to solve generation ordering problem
2043   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2044   ldr(rscratch2, Address(rscratch2));
2045   blr(rscratch2);
2046 
2047   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2048   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2049   authenticate_return_address();
2050 
2051   BLOCK_COMMENT("} verify_oop");
2052 }
2053 
2054 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2055   if (!VerifyOops) return;
2056 
2057   const char* b = nullptr;
2058   {
2059     ResourceMark rm;
2060     stringStream ss;
2061     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2062     b = code_string(ss.as_string());
2063   }
2064   BLOCK_COMMENT("verify_oop_addr {");
2065 
2066   strip_return_address(); // This might happen within a stack frame.
2067   protect_return_address();
2068   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2069   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2070 
2071   // addr may contain sp so we will have to adjust it based on the
2072   // pushes that we just did.
2073   if (addr.uses(sp)) {
2074     lea(r0, addr);
2075     ldr(r0, Address(r0, 4 * wordSize));
2076   } else {
2077     ldr(r0, addr);
2078   }
2079   movptr(rscratch1, (uintptr_t)(address)b);
2080 
2081   // call indirectly to solve generation ordering problem
2082   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2083   ldr(rscratch2, Address(rscratch2));
2084   blr(rscratch2);
2085 
2086   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2087   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2088   authenticate_return_address();
2089 
2090   BLOCK_COMMENT("} verify_oop_addr");
2091 }
2092 
2093 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2094                                          int extra_slot_offset) {
2095   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2096   int stackElementSize = Interpreter::stackElementSize;
2097   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2098 #ifdef ASSERT
2099   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2100   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2101 #endif
2102   if (arg_slot.is_constant()) {
2103     return Address(esp, arg_slot.as_constant() * stackElementSize
2104                    + offset);
2105   } else {
2106     add(rscratch1, esp, arg_slot.as_register(),
2107         ext::uxtx, exact_log2(stackElementSize));
2108     return Address(rscratch1, offset);
2109   }
2110 }
2111 
2112 // Handle the receiver type profile update given the "recv" klass.
2113 //
2114 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2115 // If there are no matching or claimable receiver entries in RD, updates
2116 // the polymorphic counter.
2117 //
2118 // This code expected to run by either the interpreter or JIT-ed code, without
2119 // extra synchronization. For safety, receiver cells are claimed atomically, which
2120 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2121 // counter updates are not atomic.
2122 //
2123 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2124   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2125 
2126   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2127   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2128   int poly_count_offset      = in_bytes(CounterData::count_offset());
2129   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2130   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2131 
2132   // Adjust for MDP offsets.
2133   base_receiver_offset += mdp_offset;
2134   end_receiver_offset  += mdp_offset;
2135   poly_count_offset    += mdp_offset;
2136 
2137 #ifdef ASSERT
2138   // We are about to walk the MDO slots without asking for offsets.
2139   // Check that our math hits all the right spots.
2140   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2141     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2142     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2143     int offset = base_receiver_offset + receiver_step*c;
2144     int count_offset = offset + receiver_to_count_step;
2145     assert(offset == real_recv_offset, "receiver slot math");
2146     assert(count_offset == real_count_offset, "receiver count math");
2147   }
2148   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2149   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2150 #endif
2151 
2152   // Corner case: no profile table. Increment poly counter and exit.
2153   if (ReceiverTypeData::row_limit() == 0) {
2154     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2155     return;
2156   }
2157 
2158   Register offset = rscratch2;
2159 
2160   Label L_loop_search_receiver, L_loop_search_empty;
2161   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
2162 
2163   // The code here recognizes three major cases:
2164   //   A. Fastest: receiver found in the table
2165   //   B. Fast: no receiver in the table, and the table is full
2166   //   C. Slow: no receiver in the table, free slots in the table
2167   //
2168   // The case A performance is most important, as perfectly-behaved code would end up
2169   // there, especially with larger TypeProfileWidth. The case B performance is
2170   // important as well, this is where bulk of code would land for normally megamorphic
2171   // cases. The case C performance is not essential, its job is to deal with installation
2172   // races, we optimize for code density instead. Case C needs to make sure that receiver
2173   // rows are only claimed once. This makes sure we never overwrite a row for another
2174   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2175   //
2176   // It is very tempting to handle these cases in a single loop, and claim the first slot
2177   // without checking the rest of the table. But, profiling code should tolerate free slots
2178   // in the table, as class unloading can clear them. After such cleanup, the receiver
2179   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2180   // to complete, before trying to install new slots. Splitting the code in several tight
2181   // loops also helpfully optimizes for cases A and B.
2182   //
2183   // This code is effectively:
2184   //
2185   // restart:
2186   //   // Fastest: receiver is already installed
2187   //   for (i = 0; i < receiver_count(); i++) {
2188   //     if (receiver(i) == recv) goto found_recv(i);
2189   //   }
2190   //
2191   //   // Fast: no receiver, but profile is full
2192   //   for (i = 0; i < receiver_count(); i++) {
2193   //     if (receiver(i) == null) goto found_null(i);
2194   //   }
2195   //   goto polymorphic
2196   //
2197   //   // Slow: try to install receiver
2198   // found_null(i):
2199   //   CAS(&receiver(i), null, recv);
2200   //   goto restart
2201   //
2202   // polymorphic:
2203   //   count++;
2204   //   return
2205   //
2206   // found_recv(i):
2207   //   *receiver_count(i)++
2208   //
2209 
2210   bind(L_restart);
2211 
2212   // Fastest: receiver is already installed
2213   mov(offset, base_receiver_offset);
2214   bind(L_loop_search_receiver);
2215     ldr(rscratch1, Address(mdp, offset));
2216     cmp(rscratch1, recv);
2217     br(Assembler::EQ, L_found_recv);
2218   add(offset, offset, receiver_step);
2219   sub(rscratch1, offset, end_receiver_offset);
2220   cbnz(rscratch1, L_loop_search_receiver);
2221 
2222   // Fast: no receiver, but profile is full
2223   mov(offset, base_receiver_offset);
2224   bind(L_loop_search_empty);
2225     ldr(rscratch1, Address(mdp, offset));
2226     cbz(rscratch1, L_found_empty);
2227   add(offset, offset, receiver_step);
2228   sub(rscratch1, offset, end_receiver_offset);
2229   cbnz(rscratch1, L_loop_search_empty);
2230   b(L_polymorphic);
2231 
2232   // Slow: try to install receiver
2233   bind(L_found_empty);
2234 
2235   // Atomically swing receiver slot: null -> recv.
2236   //
2237   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2238   // is used to hold the destination address. This is safe because the
2239   // offset is no longer needed after the address is computed.
2240 
2241   lea(rscratch2, Address(mdp, offset));
2242   cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
2243           /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
2244 
2245   // CAS success means the slot now has the receiver we want. CAS failure means
2246   // something had claimed the slot concurrently: it can be the same receiver we want,
2247   // or something else. Since this is a slow path, we can optimize for code density,
2248   // and just restart the search from the beginning.
2249   b(L_restart);
2250 
2251   // Counter updates:
2252 
2253   // Increment polymorphic counter instead of receiver slot.
2254   bind(L_polymorphic);
2255   mov(offset, poly_count_offset);
2256   b(L_count_update);
2257 
2258   // Found a receiver, convert its slot offset to corresponding count offset.
2259   bind(L_found_recv);
2260   add(offset, offset, receiver_to_count_step);
2261 
2262   bind(L_count_update);
2263   increment(Address(mdp, offset), DataLayout::counter_increment);
2264 }
2265 
2266 
2267 void MacroAssembler::call_VM_leaf_base(address entry_point,
2268                                        int number_of_arguments,
2269                                        Label *retaddr) {
2270   Label E, L;
2271 
2272   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2273 
2274   mov(rscratch1, RuntimeAddress(entry_point));
2275   blr(rscratch1);
2276   if (retaddr)
2277     bind(*retaddr);
2278 
2279   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2280 }
2281 
2282 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2283   call_VM_leaf_base(entry_point, number_of_arguments);
2284 }
2285 
2286 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2287   pass_arg0(this, arg_0);
2288   call_VM_leaf_base(entry_point, 1);
2289 }
2290 
2291 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2292   assert_different_registers(arg_1, c_rarg0);
2293   pass_arg0(this, arg_0);
2294   pass_arg1(this, arg_1);
2295   call_VM_leaf_base(entry_point, 2);
2296 }
2297 
2298 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2299                                   Register arg_1, Register arg_2) {
2300   assert_different_registers(arg_1, c_rarg0);
2301   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2302   pass_arg0(this, arg_0);
2303   pass_arg1(this, arg_1);
2304   pass_arg2(this, arg_2);
2305   call_VM_leaf_base(entry_point, 3);
2306 }
2307 
2308 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2309   pass_arg0(this, arg_0);
2310   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2311 }
2312 
2313 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2314 
2315   assert_different_registers(arg_0, c_rarg1);
2316   pass_arg1(this, arg_1);
2317   pass_arg0(this, arg_0);
2318   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2319 }
2320 
2321 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2322   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2323   assert_different_registers(arg_1, c_rarg2);
2324   pass_arg2(this, arg_2);
2325   pass_arg1(this, arg_1);
2326   pass_arg0(this, arg_0);
2327   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2328 }
2329 
2330 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2331   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2332   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2333   assert_different_registers(arg_2, c_rarg3);
2334   pass_arg3(this, arg_3);
2335   pass_arg2(this, arg_2);
2336   pass_arg1(this, arg_1);
2337   pass_arg0(this, arg_0);
2338   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2339 }
2340 
2341 void MacroAssembler::null_check(Register reg, int offset) {
2342   if (needs_explicit_null_check(offset)) {
2343     // provoke OS null exception if reg is null by
2344     // accessing M[reg] w/o changing any registers
2345     // NOTE: this is plenty to provoke a segv
2346     ldr(zr, Address(reg));
2347   } else {
2348     // nothing to do, (later) access of M[reg + offset]
2349     // will provoke OS null exception if reg is null
2350   }
2351 }
2352 
2353 // MacroAssembler protected routines needed to implement
2354 // public methods
2355 
2356 void MacroAssembler::mov(Register r, Address dest) {
2357   code_section()->relocate(pc(), dest.rspec());
2358   uint64_t imm64 = (uint64_t)dest.target();
2359   movptr(r, imm64);
2360 }
2361 
2362 // Move a constant pointer into r.  In AArch64 mode the virtual
2363 // address space is 48 bits in size, so we only need three
2364 // instructions to create a patchable instruction sequence that can
2365 // reach anywhere.
2366 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2367 #ifndef PRODUCT
2368   {
2369     char buffer[64];
2370     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2371     block_comment(buffer);
2372   }
2373 #endif
2374   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2375   movz(r, imm64 & 0xffff);
2376   imm64 >>= 16;
2377   movk(r, imm64 & 0xffff, 16);
2378   imm64 >>= 16;
2379   movk(r, imm64 & 0xffff, 32);
2380 }
2381 
2382 // Macro to mov replicated immediate to vector register.
2383 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2384 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2385 // Vd will get the following values for different arrangements in T
2386 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2387 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2388 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2389 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2390 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2391 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2392 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2393 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2394 // Clobbers rscratch1
2395 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2396   assert(T != T1Q, "unsupported");
2397   if (T == T1D || T == T2D) {
2398     int imm = operand_valid_for_movi_immediate(imm64, T);
2399     if (-1 != imm) {
2400       movi(Vd, T, imm);
2401     } else {
2402       mov(rscratch1, imm64);
2403       dup(Vd, T, rscratch1);
2404     }
2405     return;
2406   }
2407 
2408 #ifdef ASSERT
2409   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2410   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2411   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2412 #endif
2413   int shift = operand_valid_for_movi_immediate(imm64, T);
2414   uint32_t imm32 = imm64 & 0xffffffffULL;
2415   if (shift >= 0) {
2416     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2417   } else {
2418     movw(rscratch1, imm32);
2419     dup(Vd, T, rscratch1);
2420   }
2421 }
2422 
2423 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2424 {
2425 #ifndef PRODUCT
2426   {
2427     char buffer[64];
2428     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2429     block_comment(buffer);
2430   }
2431 #endif
2432   if (operand_valid_for_logical_immediate(false, imm64)) {
2433     orr(dst, zr, imm64);
2434   } else {
2435     // we can use a combination of MOVZ or MOVN with
2436     // MOVK to build up the constant
2437     uint64_t imm_h[4];
2438     int zero_count = 0;
2439     int neg_count = 0;
2440     int i;
2441     for (i = 0; i < 4; i++) {
2442       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2443       if (imm_h[i] == 0) {
2444         zero_count++;
2445       } else if (imm_h[i] == 0xffffL) {
2446         neg_count++;
2447       }
2448     }
2449     if (zero_count == 4) {
2450       // one MOVZ will do
2451       movz(dst, 0);
2452     } else if (neg_count == 4) {
2453       // one MOVN will do
2454       movn(dst, 0);
2455     } else if (zero_count == 3) {
2456       for (i = 0; i < 4; i++) {
2457         if (imm_h[i] != 0L) {
2458           movz(dst, (uint32_t)imm_h[i], (i << 4));
2459           break;
2460         }
2461       }
2462     } else if (neg_count == 3) {
2463       // one MOVN will do
2464       for (int i = 0; i < 4; i++) {
2465         if (imm_h[i] != 0xffffL) {
2466           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2467           break;
2468         }
2469       }
2470     } else if (zero_count == 2) {
2471       // one MOVZ and one MOVK will do
2472       for (i = 0; i < 3; i++) {
2473         if (imm_h[i] != 0L) {
2474           movz(dst, (uint32_t)imm_h[i], (i << 4));
2475           i++;
2476           break;
2477         }
2478       }
2479       for (;i < 4; i++) {
2480         if (imm_h[i] != 0L) {
2481           movk(dst, (uint32_t)imm_h[i], (i << 4));
2482         }
2483       }
2484     } else if (neg_count == 2) {
2485       // one MOVN and one MOVK will do
2486       for (i = 0; i < 4; i++) {
2487         if (imm_h[i] != 0xffffL) {
2488           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2489           i++;
2490           break;
2491         }
2492       }
2493       for (;i < 4; i++) {
2494         if (imm_h[i] != 0xffffL) {
2495           movk(dst, (uint32_t)imm_h[i], (i << 4));
2496         }
2497       }
2498     } else if (zero_count == 1) {
2499       // one MOVZ and two MOVKs will do
2500       for (i = 0; i < 4; i++) {
2501         if (imm_h[i] != 0L) {
2502           movz(dst, (uint32_t)imm_h[i], (i << 4));
2503           i++;
2504           break;
2505         }
2506       }
2507       for (;i < 4; i++) {
2508         if (imm_h[i] != 0x0L) {
2509           movk(dst, (uint32_t)imm_h[i], (i << 4));
2510         }
2511       }
2512     } else if (neg_count == 1) {
2513       // one MOVN and two MOVKs will do
2514       for (i = 0; i < 4; i++) {
2515         if (imm_h[i] != 0xffffL) {
2516           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2517           i++;
2518           break;
2519         }
2520       }
2521       for (;i < 4; i++) {
2522         if (imm_h[i] != 0xffffL) {
2523           movk(dst, (uint32_t)imm_h[i], (i << 4));
2524         }
2525       }
2526     } else {
2527       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2528       movz(dst, (uint32_t)imm_h[0], 0);
2529       for (i = 1; i < 4; i++) {
2530         movk(dst, (uint32_t)imm_h[i], (i << 4));
2531       }
2532     }
2533   }
2534 }
2535 
2536 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2537 {
2538 #ifndef PRODUCT
2539     {
2540       char buffer[64];
2541       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2542       block_comment(buffer);
2543     }
2544 #endif
2545   if (operand_valid_for_logical_immediate(true, imm32)) {
2546     orrw(dst, zr, imm32);
2547   } else {
2548     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2549     // constant
2550     uint32_t imm_h[2];
2551     imm_h[0] = imm32 & 0xffff;
2552     imm_h[1] = ((imm32 >> 16) & 0xffff);
2553     if (imm_h[0] == 0) {
2554       movzw(dst, imm_h[1], 16);
2555     } else if (imm_h[0] == 0xffff) {
2556       movnw(dst, imm_h[1] ^ 0xffff, 16);
2557     } else if (imm_h[1] == 0) {
2558       movzw(dst, imm_h[0], 0);
2559     } else if (imm_h[1] == 0xffff) {
2560       movnw(dst, imm_h[0] ^ 0xffff, 0);
2561     } else {
2562       // use a MOVZ and MOVK (makes it easier to debug)
2563       movzw(dst, imm_h[0], 0);
2564       movkw(dst, imm_h[1], 16);
2565     }
2566   }
2567 }
2568 
2569 // Form an address from base + offset in Rd.  Rd may or may
2570 // not actually be used: you must use the Address that is returned.
2571 // It is up to you to ensure that the shift provided matches the size
2572 // of your data.
2573 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2574   if (Address::offset_ok_for_immed(byte_offset, shift))
2575     // It fits; no need for any heroics
2576     return Address(base, byte_offset);
2577 
2578   // Don't do anything clever with negative or misaligned offsets
2579   unsigned mask = (1 << shift) - 1;
2580   if (byte_offset < 0 || byte_offset & mask) {
2581     mov(Rd, byte_offset);
2582     add(Rd, base, Rd);
2583     return Address(Rd);
2584   }
2585 
2586   // See if we can do this with two 12-bit offsets
2587   {
2588     uint64_t word_offset = byte_offset >> shift;
2589     uint64_t masked_offset = word_offset & 0xfff000;
2590     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2591         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2592       add(Rd, base, masked_offset << shift);
2593       word_offset -= masked_offset;
2594       return Address(Rd, word_offset << shift);
2595     }
2596   }
2597 
2598   // Do it the hard way
2599   mov(Rd, byte_offset);
2600   add(Rd, base, Rd);
2601   return Address(Rd);
2602 }
2603 
2604 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2605                                     bool want_remainder, Register scratch)
2606 {
2607   // Full implementation of Java idiv and irem.  The function
2608   // returns the (pc) offset of the div instruction - may be needed
2609   // for implicit exceptions.
2610   //
2611   // constraint : ra/rb =/= scratch
2612   //         normal case
2613   //
2614   // input : ra: dividend
2615   //         rb: divisor
2616   //
2617   // result: either
2618   //         quotient  (= ra idiv rb)
2619   //         remainder (= ra irem rb)
2620 
2621   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2622 
2623   int idivl_offset = offset();
2624   if (! want_remainder) {
2625     sdivw(result, ra, rb);
2626   } else {
2627     sdivw(scratch, ra, rb);
2628     Assembler::msubw(result, scratch, rb, ra);
2629   }
2630 
2631   return idivl_offset;
2632 }
2633 
2634 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2635                                     bool want_remainder, Register scratch)
2636 {
2637   // Full implementation of Java ldiv and lrem.  The function
2638   // returns the (pc) offset of the div instruction - may be needed
2639   // for implicit exceptions.
2640   //
2641   // constraint : ra/rb =/= scratch
2642   //         normal case
2643   //
2644   // input : ra: dividend
2645   //         rb: divisor
2646   //
2647   // result: either
2648   //         quotient  (= ra idiv rb)
2649   //         remainder (= ra irem rb)
2650 
2651   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2652 
2653   int idivq_offset = offset();
2654   if (! want_remainder) {
2655     sdiv(result, ra, rb);
2656   } else {
2657     sdiv(scratch, ra, rb);
2658     Assembler::msub(result, scratch, rb, ra);
2659   }
2660 
2661   return idivq_offset;
2662 }
2663 
2664 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2665   address prev = pc() - NativeMembar::instruction_size;
2666   address last = code()->last_insn();
2667   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2668     NativeMembar *bar = NativeMembar_at(prev);
2669     if (AlwaysMergeDMB) {
2670       bar->set_kind(bar->get_kind() | order_constraint);
2671       BLOCK_COMMENT("merged membar(always)");
2672       return;
2673     }
2674     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2675     // doing so would introduce a StoreLoad which the caller did not
2676     // intend
2677     if (bar->get_kind() == order_constraint
2678         || bar->get_kind() == AnyAny
2679         || order_constraint == AnyAny) {
2680       // We are merging two memory barrier instructions.  On AArch64 we
2681       // can do this simply by ORing them together.
2682       bar->set_kind(bar->get_kind() | order_constraint);
2683       BLOCK_COMMENT("merged membar");
2684       return;
2685     } else {
2686       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2687       // We need check the last 2 instructions
2688       address prev2 = prev - NativeMembar::instruction_size;
2689       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2690         NativeMembar *bar2 = NativeMembar_at(prev2);
2691         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2692         BLOCK_COMMENT("merged membar(elided)");
2693         return;
2694       }
2695     }
2696   }
2697   code()->set_last_insn(pc());
2698   dmb(Assembler::barrier(order_constraint));
2699 }
2700 
2701 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2702   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2703     merge_ldst(rt, adr, size_in_bytes, is_store);
2704     code()->clear_last_insn();
2705     return true;
2706   } else {
2707     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2708     const uint64_t mask = size_in_bytes - 1;
2709     if (adr.getMode() == Address::base_plus_offset &&
2710         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2711       code()->set_last_insn(pc());
2712     }
2713     return false;
2714   }
2715 }
2716 
2717 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2718   // We always try to merge two adjacent loads into one ldp.
2719   if (!try_merge_ldst(Rx, adr, 8, false)) {
2720     Assembler::ldr(Rx, adr);
2721   }
2722 }
2723 
2724 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2725   // We always try to merge two adjacent loads into one ldp.
2726   if (!try_merge_ldst(Rw, adr, 4, false)) {
2727     Assembler::ldrw(Rw, adr);
2728   }
2729 }
2730 
2731 void MacroAssembler::str(Register Rx, const Address &adr) {
2732   // We always try to merge two adjacent stores into one stp.
2733   if (!try_merge_ldst(Rx, adr, 8, true)) {
2734     Assembler::str(Rx, adr);
2735   }
2736 }
2737 
2738 void MacroAssembler::strw(Register Rw, const Address &adr) {
2739   // We always try to merge two adjacent stores into one stp.
2740   if (!try_merge_ldst(Rw, adr, 4, true)) {
2741     Assembler::strw(Rw, adr);
2742   }
2743 }
2744 
2745 // MacroAssembler routines found actually to be needed
2746 
2747 void MacroAssembler::push(Register src)
2748 {
2749   str(src, Address(pre(esp, -1 * wordSize)));
2750 }
2751 
2752 void MacroAssembler::pop(Register dst)
2753 {
2754   ldr(dst, Address(post(esp, 1 * wordSize)));
2755 }
2756 
2757 // Note: load_unsigned_short used to be called load_unsigned_word.
2758 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2759   int off = offset();
2760   ldrh(dst, src);
2761   return off;
2762 }
2763 
2764 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2765   int off = offset();
2766   ldrb(dst, src);
2767   return off;
2768 }
2769 
2770 int MacroAssembler::load_signed_short(Register dst, Address src) {
2771   int off = offset();
2772   ldrsh(dst, src);
2773   return off;
2774 }
2775 
2776 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2777   int off = offset();
2778   ldrsb(dst, src);
2779   return off;
2780 }
2781 
2782 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2783   int off = offset();
2784   ldrshw(dst, src);
2785   return off;
2786 }
2787 
2788 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2789   int off = offset();
2790   ldrsbw(dst, src);
2791   return off;
2792 }
2793 
2794 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2795   switch (size_in_bytes) {
2796   case  8:  ldr(dst, src); break;
2797   case  4:  ldrw(dst, src); break;
2798   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2799   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2800   default:  ShouldNotReachHere();
2801   }
2802 }
2803 
2804 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2805   switch (size_in_bytes) {
2806   case  8:  str(src, dst); break;
2807   case  4:  strw(src, dst); break;
2808   case  2:  strh(src, dst); break;
2809   case  1:  strb(src, dst); break;
2810   default:  ShouldNotReachHere();
2811   }
2812 }
2813 
2814 void MacroAssembler::decrementw(Register reg, int value)
2815 {
2816   if (value < 0)  { incrementw(reg, -value);      return; }
2817   if (value == 0) {                               return; }
2818   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2819   /* else */ {
2820     guarantee(reg != rscratch2, "invalid dst for register decrement");
2821     movw(rscratch2, (unsigned)value);
2822     subw(reg, reg, rscratch2);
2823   }
2824 }
2825 
2826 void MacroAssembler::decrement(Register reg, int value)
2827 {
2828   if (value < 0)  { increment(reg, -value);      return; }
2829   if (value == 0) {                              return; }
2830   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2831   /* else */ {
2832     assert(reg != rscratch2, "invalid dst for register decrement");
2833     mov(rscratch2, (uint64_t)value);
2834     sub(reg, reg, rscratch2);
2835   }
2836 }
2837 
2838 void MacroAssembler::decrementw(Address dst, int value)
2839 {
2840   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2841   if (dst.getMode() == Address::literal) {
2842     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2843     lea(rscratch2, dst);
2844     dst = Address(rscratch2);
2845   }
2846   ldrw(rscratch1, dst);
2847   decrementw(rscratch1, value);
2848   strw(rscratch1, dst);
2849 }
2850 
2851 void MacroAssembler::decrement(Address dst, int value)
2852 {
2853   assert(!dst.uses(rscratch1), "invalid address for decrement");
2854   if (dst.getMode() == Address::literal) {
2855     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2856     lea(rscratch2, dst);
2857     dst = Address(rscratch2);
2858   }
2859   ldr(rscratch1, dst);
2860   decrement(rscratch1, value);
2861   str(rscratch1, dst);
2862 }
2863 
2864 void MacroAssembler::incrementw(Register reg, int value)
2865 {
2866   if (value < 0)  { decrementw(reg, -value);      return; }
2867   if (value == 0) {                               return; }
2868   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2869   /* else */ {
2870     assert(reg != rscratch2, "invalid dst for register increment");
2871     movw(rscratch2, (unsigned)value);
2872     addw(reg, reg, rscratch2);
2873   }
2874 }
2875 
2876 void MacroAssembler::increment(Register reg, int value)
2877 {
2878   if (value < 0)  { decrement(reg, -value);      return; }
2879   if (value == 0) {                              return; }
2880   if (value < (1 << 12)) { add(reg, reg, value); return; }
2881   /* else */ {
2882     assert(reg != rscratch2, "invalid dst for register increment");
2883     movw(rscratch2, (unsigned)value);
2884     add(reg, reg, rscratch2);
2885   }
2886 }
2887 
2888 void MacroAssembler::incrementw(Address dst, int value)
2889 {
2890   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2891   if (dst.getMode() == Address::literal) {
2892     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2893     lea(rscratch2, dst);
2894     dst = Address(rscratch2);
2895   }
2896   ldrw(rscratch1, dst);
2897   incrementw(rscratch1, value);
2898   strw(rscratch1, dst);
2899 }
2900 
2901 void MacroAssembler::increment(Address dst, int value)
2902 {
2903   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2904   if (dst.getMode() == Address::literal) {
2905     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2906     lea(rscratch2, dst);
2907     dst = Address(rscratch2);
2908   }
2909   ldr(rscratch1, dst);
2910   increment(rscratch1, value);
2911   str(rscratch1, dst);
2912 }
2913 
2914 // Push lots of registers in the bit set supplied.  Don't push sp.
2915 // Return the number of words pushed
2916 int MacroAssembler::push(unsigned int bitset, Register stack) {
2917   int words_pushed = 0;
2918 
2919   // Scan bitset to accumulate register pairs
2920   unsigned char regs[32];
2921   int count = 0;
2922   for (int reg = 0; reg <= 30; reg++) {
2923     if (1 & bitset)
2924       regs[count++] = reg;
2925     bitset >>= 1;
2926   }
2927   regs[count++] = zr->raw_encoding();
2928   count &= ~1;  // Only push an even number of regs
2929 
2930   if (count) {
2931     stp(as_Register(regs[0]), as_Register(regs[1]),
2932        Address(pre(stack, -count * wordSize)));
2933     words_pushed += 2;
2934   }
2935   for (int i = 2; i < count; i += 2) {
2936     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2937        Address(stack, i * wordSize));
2938     words_pushed += 2;
2939   }
2940 
2941   assert(words_pushed == count, "oops, pushed != count");
2942 
2943   return count;
2944 }
2945 
2946 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2947   int words_pushed = 0;
2948 
2949   // Scan bitset to accumulate register pairs
2950   unsigned char regs[32];
2951   int count = 0;
2952   for (int reg = 0; reg <= 30; reg++) {
2953     if (1 & bitset)
2954       regs[count++] = reg;
2955     bitset >>= 1;
2956   }
2957   regs[count++] = zr->raw_encoding();
2958   count &= ~1;
2959 
2960   for (int i = 2; i < count; i += 2) {
2961     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2962        Address(stack, i * wordSize));
2963     words_pushed += 2;
2964   }
2965   if (count) {
2966     ldp(as_Register(regs[0]), as_Register(regs[1]),
2967        Address(post(stack, count * wordSize)));
2968     words_pushed += 2;
2969   }
2970 
2971   assert(words_pushed == count, "oops, pushed != count");
2972 
2973   return count;
2974 }
2975 
2976 // Push lots of registers in the bit set supplied.  Don't push sp.
2977 // Return the number of dwords pushed
2978 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2979   int words_pushed = 0;
2980   bool use_sve = false;
2981   int sve_vector_size_in_bytes = 0;
2982 
2983 #ifdef COMPILER2
2984   use_sve = Matcher::supports_scalable_vector();
2985   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2986 #endif
2987 
2988   // Scan bitset to accumulate register pairs
2989   unsigned char regs[32];
2990   int count = 0;
2991   for (int reg = 0; reg <= 31; reg++) {
2992     if (1 & bitset)
2993       regs[count++] = reg;
2994     bitset >>= 1;
2995   }
2996 
2997   if (count == 0) {
2998     return 0;
2999   }
3000 
3001   if (mode == PushPopFull) {
3002     if (use_sve && sve_vector_size_in_bytes > 16) {
3003       mode = PushPopSVE;
3004     } else {
3005       mode = PushPopNeon;
3006     }
3007   }
3008 
3009 #ifndef PRODUCT
3010   {
3011     char buffer[48];
3012     if (mode == PushPopSVE) {
3013       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3014     } else if (mode == PushPopNeon) {
3015       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3016     } else {
3017       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3018     }
3019     block_comment(buffer);
3020   }
3021 #endif
3022 
3023   if (mode == PushPopSVE) {
3024     sub(stack, stack, sve_vector_size_in_bytes * count);
3025     for (int i = 0; i < count; i++) {
3026       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3027     }
3028     return count * sve_vector_size_in_bytes / 8;
3029   }
3030 
3031   if (mode == PushPopNeon) {
3032     if (count == 1) {
3033       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3034       return 2;
3035     }
3036 
3037     bool odd = (count & 1) == 1;
3038     int push_slots = count + (odd ? 1 : 0);
3039 
3040     // Always pushing full 128 bit registers.
3041     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3042     words_pushed += 2;
3043 
3044     for (int i = 2; i + 1 < count; i += 2) {
3045       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3046       words_pushed += 2;
3047     }
3048 
3049     if (odd) {
3050       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3051       words_pushed++;
3052     }
3053 
3054     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3055     return count * 2;
3056   }
3057 
3058   if (mode == PushPopFp) {
3059     bool odd = (count & 1) == 1;
3060     int push_slots = count + (odd ? 1 : 0);
3061 
3062     if (count == 1) {
3063       // Stack pointer must be 16 bytes aligned
3064       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3065       return 1;
3066     }
3067 
3068     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3069     words_pushed += 2;
3070 
3071     for (int i = 2; i + 1 < count; i += 2) {
3072       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3073       words_pushed += 2;
3074     }
3075 
3076     if (odd) {
3077       // Stack pointer must be 16 bytes aligned
3078       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3079       words_pushed++;
3080     }
3081 
3082     assert(words_pushed == count, "oops, pushed != count");
3083 
3084     return count;
3085   }
3086 
3087   return 0;
3088 }
3089 
3090 // Return the number of dwords popped
3091 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3092   int words_pushed = 0;
3093   bool use_sve = false;
3094   int sve_vector_size_in_bytes = 0;
3095 
3096 #ifdef COMPILER2
3097   use_sve = Matcher::supports_scalable_vector();
3098   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3099 #endif
3100   // Scan bitset to accumulate register pairs
3101   unsigned char regs[32];
3102   int count = 0;
3103   for (int reg = 0; reg <= 31; reg++) {
3104     if (1 & bitset)
3105       regs[count++] = reg;
3106     bitset >>= 1;
3107   }
3108 
3109   if (count == 0) {
3110     return 0;
3111   }
3112 
3113   if (mode == PushPopFull) {
3114     if (use_sve && sve_vector_size_in_bytes > 16) {
3115       mode = PushPopSVE;
3116     } else {
3117       mode = PushPopNeon;
3118     }
3119   }
3120 
3121 #ifndef PRODUCT
3122   {
3123     char buffer[48];
3124     if (mode == PushPopSVE) {
3125       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3126     } else if (mode == PushPopNeon) {
3127       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3128     } else {
3129       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3130     }
3131     block_comment(buffer);
3132   }
3133 #endif
3134 
3135   if (mode == PushPopSVE) {
3136     for (int i = count - 1; i >= 0; i--) {
3137       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3138     }
3139     add(stack, stack, sve_vector_size_in_bytes * count);
3140     return count * sve_vector_size_in_bytes / 8;
3141   }
3142 
3143   if (mode == PushPopNeon) {
3144     if (count == 1) {
3145       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3146       return 2;
3147     }
3148 
3149     bool odd = (count & 1) == 1;
3150     int push_slots = count + (odd ? 1 : 0);
3151 
3152     if (odd) {
3153       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3154       words_pushed++;
3155     }
3156 
3157     for (int i = 2; i + 1 < count; i += 2) {
3158       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3159       words_pushed += 2;
3160     }
3161 
3162     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3163     words_pushed += 2;
3164 
3165     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3166 
3167     return count * 2;
3168   }
3169 
3170   if (mode == PushPopFp) {
3171     bool odd = (count & 1) == 1;
3172     int push_slots = count + (odd ? 1 : 0);
3173 
3174     if (count == 1) {
3175       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3176       return 1;
3177     }
3178 
3179     if (odd) {
3180       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3181       words_pushed++;
3182     }
3183 
3184     for (int i = 2; i + 1 < count; i += 2) {
3185       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3186       words_pushed += 2;
3187     }
3188 
3189     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3190     words_pushed += 2;
3191 
3192     assert(words_pushed == count, "oops, pushed != count");
3193 
3194     return count;
3195   }
3196 
3197   return 0;
3198 }
3199 
3200 // Return the number of dwords pushed
3201 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3202   bool use_sve = false;
3203   int sve_predicate_size_in_slots = 0;
3204 
3205 #ifdef COMPILER2
3206   use_sve = Matcher::supports_scalable_vector();
3207   if (use_sve) {
3208     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3209   }
3210 #endif
3211 
3212   if (!use_sve) {
3213     return 0;
3214   }
3215 
3216   unsigned char regs[PRegister::number_of_registers];
3217   int count = 0;
3218   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3219     if (1 & bitset)
3220       regs[count++] = reg;
3221     bitset >>= 1;
3222   }
3223 
3224   if (count == 0) {
3225     return 0;
3226   }
3227 
3228   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3229                                   VMRegImpl::stack_slot_size * count, 16);
3230   sub(stack, stack, total_push_bytes);
3231   for (int i = 0; i < count; i++) {
3232     sve_str(as_PRegister(regs[i]), Address(stack, i));
3233   }
3234   return total_push_bytes / 8;
3235 }
3236 
3237 // Return the number of dwords popped
3238 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3239   bool use_sve = false;
3240   int sve_predicate_size_in_slots = 0;
3241 
3242 #ifdef COMPILER2
3243   use_sve = Matcher::supports_scalable_vector();
3244   if (use_sve) {
3245     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3246   }
3247 #endif
3248 
3249   if (!use_sve) {
3250     return 0;
3251   }
3252 
3253   unsigned char regs[PRegister::number_of_registers];
3254   int count = 0;
3255   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3256     if (1 & bitset)
3257       regs[count++] = reg;
3258     bitset >>= 1;
3259   }
3260 
3261   if (count == 0) {
3262     return 0;
3263   }
3264 
3265   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3266                                  VMRegImpl::stack_slot_size * count, 16);
3267   for (int i = count - 1; i >= 0; i--) {
3268     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3269   }
3270   add(stack, stack, total_pop_bytes);
3271   return total_pop_bytes / 8;
3272 }
3273 
3274 #ifdef ASSERT
3275 void MacroAssembler::verify_heapbase(const char* msg) {
3276 #if 0
3277   assert (Universe::heap() != nullptr, "java heap should be initialized");
3278   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3279     // rheapbase is allocated as general register
3280     return;
3281   }
3282   if (CheckCompressedOops) {
3283     Label ok;
3284     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3285     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3286     br(Assembler::EQ, ok);
3287     stop(msg);
3288     bind(ok);
3289     pop(1 << rscratch1->encoding(), sp);
3290   }
3291 #endif
3292 }
3293 #endif
3294 
3295 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3296   assert_different_registers(value, tmp1, tmp2);
3297   Label done, tagged, weak_tagged;
3298 
3299   cbz(value, done);           // Use null as-is.
3300   tst(value, JNIHandles::tag_mask); // Test for tag.
3301   br(Assembler::NE, tagged);
3302 
3303   // Resolve local handle
3304   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3305   verify_oop(value);
3306   b(done);
3307 
3308   bind(tagged);
3309   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3310   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3311 
3312   // Resolve global handle
3313   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3314   verify_oop(value);
3315   b(done);
3316 
3317   bind(weak_tagged);
3318   // Resolve jweak.
3319   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3320                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3321   verify_oop(value);
3322 
3323   bind(done);
3324 }
3325 
3326 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3327   assert_different_registers(value, tmp1, tmp2);
3328   Label done;
3329 
3330   cbz(value, done);           // Use null as-is.
3331 
3332 #ifdef ASSERT
3333   {
3334     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3335     Label valid_global_tag;
3336     tbnz(value, 1, valid_global_tag); // Test for global tag
3337     stop("non global jobject using resolve_global_jobject");
3338     bind(valid_global_tag);
3339   }
3340 #endif
3341 
3342   // Resolve global handle
3343   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3344   verify_oop(value);
3345 
3346   bind(done);
3347 }
3348 
3349 void MacroAssembler::stop(const char* msg) {
3350   // Skip AOT caching C strings in scratch buffer.
3351   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3352   BLOCK_COMMENT(str);
3353   // load msg into r0 so we can access it from the signal handler
3354   // ExternalAddress enables saving and restoring via the code cache
3355   lea(c_rarg0, ExternalAddress((address) str));
3356   dcps1(0xdeae);
3357 }
3358 
3359 void MacroAssembler::unimplemented(const char* what) {
3360   const char* buf = nullptr;
3361   {
3362     ResourceMark rm;
3363     stringStream ss;
3364     ss.print("unimplemented: %s", what);
3365     buf = code_string(ss.as_string());
3366   }
3367   stop(buf);
3368 }
3369 
3370 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3371 #ifdef ASSERT
3372   Label OK;
3373   br(cc, OK);
3374   stop(msg);
3375   bind(OK);
3376 #endif
3377 }
3378 
3379 // If a constant does not fit in an immediate field, generate some
3380 // number of MOV instructions and then perform the operation.
3381 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3382                                            add_sub_imm_insn insn1,
3383                                            add_sub_reg_insn insn2,
3384                                            bool is32) {
3385   assert(Rd != zr, "Rd = zr and not setting flags?");
3386   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3387   if (fits) {
3388     (this->*insn1)(Rd, Rn, imm);
3389   } else {
3390     if (g_uabs(imm) < (1 << 24)) {
3391        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3392        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3393     } else {
3394        assert_different_registers(Rd, Rn);
3395        mov(Rd, imm);
3396        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3397     }
3398   }
3399 }
3400 
3401 // Separate vsn which sets the flags. Optimisations are more restricted
3402 // because we must set the flags correctly.
3403 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3404                                              add_sub_imm_insn insn1,
3405                                              add_sub_reg_insn insn2,
3406                                              bool is32) {
3407   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3408   if (fits) {
3409     (this->*insn1)(Rd, Rn, imm);
3410   } else {
3411     assert_different_registers(Rd, Rn);
3412     assert(Rd != zr, "overflow in immediate operand");
3413     mov(Rd, imm);
3414     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3415   }
3416 }
3417 
3418 
3419 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3420   if (increment.is_register()) {
3421     add(Rd, Rn, increment.as_register());
3422   } else {
3423     add(Rd, Rn, increment.as_constant());
3424   }
3425 }
3426 
3427 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3428   if (increment.is_register()) {
3429     addw(Rd, Rn, increment.as_register());
3430   } else {
3431     addw(Rd, Rn, increment.as_constant());
3432   }
3433 }
3434 
3435 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3436   if (decrement.is_register()) {
3437     sub(Rd, Rn, decrement.as_register());
3438   } else {
3439     sub(Rd, Rn, decrement.as_constant());
3440   }
3441 }
3442 
3443 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3444   if (decrement.is_register()) {
3445     subw(Rd, Rn, decrement.as_register());
3446   } else {
3447     subw(Rd, Rn, decrement.as_constant());
3448   }
3449 }
3450 
3451 void MacroAssembler::reinit_heapbase()
3452 {
3453   if (UseCompressedOops) {
3454     if (Universe::is_fully_initialized()) {
3455       mov(rheapbase, CompressedOops::base());
3456     } else {
3457       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3458       ldr(rheapbase, Address(rheapbase));
3459     }
3460   }
3461 }
3462 
3463 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3464 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3465 // Pass a register for the result, otherwise pass noreg.
3466 
3467 // Clobbers rscratch1
3468 void MacroAssembler::cmpxchg(Register addr, Register expected,
3469                              Register new_val,
3470                              enum operand_size size,
3471                              bool acquire, bool release,
3472                              bool weak,
3473                              Register result) {
3474   if (result == noreg)  result = rscratch1;
3475   BLOCK_COMMENT("cmpxchg {");
3476   if (UseLSE) {
3477     mov(result, expected);
3478     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3479     compare_eq(result, expected, size);
3480 #ifdef ASSERT
3481     // Poison rscratch1 which is written on !UseLSE branch
3482     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3483 #endif
3484   } else {
3485     Label retry_load, done;
3486     prfm(Address(addr), PSTL1STRM);
3487     bind(retry_load);
3488     load_exclusive(result, addr, size, acquire);
3489     compare_eq(result, expected, size);
3490     br(Assembler::NE, done);
3491     store_exclusive(rscratch1, new_val, addr, size, release);
3492     if (weak) {
3493       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3494     } else {
3495       cbnzw(rscratch1, retry_load);
3496     }
3497     bind(done);
3498   }
3499   BLOCK_COMMENT("} cmpxchg");
3500 }
3501 
3502 // A generic comparison. Only compares for equality, clobbers rscratch1.
3503 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3504   if (size == xword) {
3505     cmp(rm, rn);
3506   } else if (size == word) {
3507     cmpw(rm, rn);
3508   } else if (size == halfword) {
3509     eorw(rscratch1, rm, rn);
3510     ands(zr, rscratch1, 0xffff);
3511   } else if (size == byte) {
3512     eorw(rscratch1, rm, rn);
3513     ands(zr, rscratch1, 0xff);
3514   } else {
3515     ShouldNotReachHere();
3516   }
3517 }
3518 
3519 
3520 static bool different(Register a, RegisterOrConstant b, Register c) {
3521   if (b.is_constant())
3522     return a != c;
3523   else
3524     return a != b.as_register() && a != c && b.as_register() != c;
3525 }
3526 
3527 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3528 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3529   if (UseLSE) {                                                         \
3530     prev = prev->is_valid() ? prev : zr;                                \
3531     if (incr.is_register()) {                                           \
3532       AOP(sz, incr.as_register(), prev, addr);                          \
3533     } else {                                                            \
3534       mov(rscratch2, incr.as_constant());                               \
3535       AOP(sz, rscratch2, prev, addr);                                   \
3536     }                                                                   \
3537     return;                                                             \
3538   }                                                                     \
3539   Register result = rscratch2;                                          \
3540   if (prev->is_valid())                                                 \
3541     result = different(prev, incr, addr) ? prev : rscratch2;            \
3542                                                                         \
3543   Label retry_load;                                                     \
3544   prfm(Address(addr), PSTL1STRM);                                       \
3545   bind(retry_load);                                                     \
3546   LDXR(result, addr);                                                   \
3547   OP(rscratch1, result, incr);                                          \
3548   STXR(rscratch2, rscratch1, addr);                                     \
3549   cbnzw(rscratch2, retry_load);                                         \
3550   if (prev->is_valid() && prev != result) {                             \
3551     IOP(prev, rscratch1, incr);                                         \
3552   }                                                                     \
3553 }
3554 
3555 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3556 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3557 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3558 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3559 
3560 #undef ATOMIC_OP
3561 
3562 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3563 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3564   if (UseLSE) {                                                         \
3565     prev = prev->is_valid() ? prev : zr;                                \
3566     AOP(sz, newv, prev, addr);                                          \
3567     return;                                                             \
3568   }                                                                     \
3569   Register result = rscratch2;                                          \
3570   if (prev->is_valid())                                                 \
3571     result = different(prev, newv, addr) ? prev : rscratch2;            \
3572                                                                         \
3573   Label retry_load;                                                     \
3574   prfm(Address(addr), PSTL1STRM);                                       \
3575   bind(retry_load);                                                     \
3576   LDXR(result, addr);                                                   \
3577   STXR(rscratch1, newv, addr);                                          \
3578   cbnzw(rscratch1, retry_load);                                         \
3579   if (prev->is_valid() && prev != result)                               \
3580     mov(prev, result);                                                  \
3581 }
3582 
3583 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3584 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3585 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3586 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3587 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3588 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3589 
3590 #undef ATOMIC_XCHG
3591 
3592 #ifndef PRODUCT
3593 extern "C" void findpc(intptr_t x);
3594 #endif
3595 
3596 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3597 {
3598   // In order to get locks to work, we need to fake a in_VM state
3599   if (ShowMessageBoxOnError) {
3600     JavaThread* thread = JavaThread::current();
3601     thread->set_thread_state(_thread_in_vm);
3602 #ifndef PRODUCT
3603     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3604       ttyLocker ttyl;
3605       BytecodeCounter::print();
3606     }
3607 #endif
3608     if (os::message_box(msg, "Execution stopped, print registers?")) {
3609       ttyLocker ttyl;
3610       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3611 #ifndef PRODUCT
3612       tty->cr();
3613       findpc(pc);
3614       tty->cr();
3615 #endif
3616       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3617       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3618       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3619       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3620       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3621       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3622       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3623       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3624       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3625       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3626       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3627       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3628       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3629       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3630       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3631       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3632       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3633       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3634       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3635       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3636       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3637       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3638       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3639       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3640       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3641       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3642       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3643       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3644       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3645       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3646       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3647       BREAKPOINT;
3648     }
3649   }
3650   fatal("DEBUG MESSAGE: %s", msg);
3651 }
3652 
3653 RegSet MacroAssembler::call_clobbered_gp_registers() {
3654   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3655 #ifndef R18_RESERVED
3656   regs += r18_tls;
3657 #endif
3658   return regs;
3659 }
3660 
3661 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3662   int step = 4 * wordSize;
3663   push(call_clobbered_gp_registers() - exclude, sp);
3664   sub(sp, sp, step);
3665   mov(rscratch1, -step);
3666   // Push v0-v7, v16-v31.
3667   for (int i = 31; i>= 4; i -= 4) {
3668     if (i <= v7->encoding() || i >= v16->encoding())
3669       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3670           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3671   }
3672   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3673       as_FloatRegister(3), T1D, Address(sp));
3674 }
3675 
3676 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3677   for (int i = 0; i < 32; i += 4) {
3678     if (i <= v7->encoding() || i >= v16->encoding())
3679       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3680           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3681   }
3682 
3683   reinitialize_ptrue();
3684 
3685   pop(call_clobbered_gp_registers() - exclude, sp);
3686 }
3687 
3688 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3689                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3690   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3691   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3692     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3693     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3694       sve_str(as_FloatRegister(i), Address(sp, i));
3695     }
3696   } else {
3697     int step = (save_vectors ? 8 : 4) * wordSize;
3698     mov(rscratch1, -step);
3699     sub(sp, sp, step);
3700     for (int i = 28; i >= 4; i -= 4) {
3701       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3702           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3703     }
3704     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3705   }
3706   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3707     sub(sp, sp, total_predicate_in_bytes);
3708     for (int i = 0; i < PRegister::number_of_registers; i++) {
3709       sve_str(as_PRegister(i), Address(sp, i));
3710     }
3711   }
3712 }
3713 
3714 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3715                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3716   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3717     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3718       sve_ldr(as_PRegister(i), Address(sp, i));
3719     }
3720     add(sp, sp, total_predicate_in_bytes);
3721   }
3722   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3723     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3724       sve_ldr(as_FloatRegister(i), Address(sp, i));
3725     }
3726     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3727   } else {
3728     int step = (restore_vectors ? 8 : 4) * wordSize;
3729     for (int i = 0; i <= 28; i += 4)
3730       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3731           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3732   }
3733 
3734   // We may use predicate registers and rely on ptrue with SVE,
3735   // regardless of wide vector (> 8 bytes) used or not.
3736   if (use_sve) {
3737     reinitialize_ptrue();
3738   }
3739 
3740   // integer registers except lr & sp
3741   pop(RegSet::range(r0, r17), sp);
3742 #ifdef R18_RESERVED
3743   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3744   pop(RegSet::range(r20, r29), sp);
3745 #else
3746   pop(RegSet::range(r18_tls, r29), sp);
3747 #endif
3748 }
3749 
3750 /**
3751  * Helpers for multiply_to_len().
3752  */
3753 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3754                                      Register src1, Register src2) {
3755   adds(dest_lo, dest_lo, src1);
3756   adc(dest_hi, dest_hi, zr);
3757   adds(dest_lo, dest_lo, src2);
3758   adc(final_dest_hi, dest_hi, zr);
3759 }
3760 
3761 // Generate an address from (r + r1 extend offset).  "size" is the
3762 // size of the operand.  The result may be in rscratch2.
3763 Address MacroAssembler::offsetted_address(Register r, Register r1,
3764                                           Address::extend ext, int offset, int size) {
3765   if (offset || (ext.shift() % size != 0)) {
3766     lea(rscratch2, Address(r, r1, ext));
3767     return Address(rscratch2, offset);
3768   } else {
3769     return Address(r, r1, ext);
3770   }
3771 }
3772 
3773 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3774 {
3775   assert(offset >= 0, "spill to negative address?");
3776   // Offset reachable ?
3777   //   Not aligned - 9 bits signed offset
3778   //   Aligned - 12 bits unsigned offset shifted
3779   Register base = sp;
3780   if ((offset & (size-1)) && offset >= (1<<8)) {
3781     add(tmp, base, offset & ((1<<12)-1));
3782     base = tmp;
3783     offset &= -1u<<12;
3784   }
3785 
3786   if (offset >= (1<<12) * size) {
3787     add(tmp, base, offset & (((1<<12)-1)<<12));
3788     base = tmp;
3789     offset &= ~(((1<<12)-1)<<12);
3790   }
3791 
3792   return Address(base, offset);
3793 }
3794 
3795 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3796   assert(offset >= 0, "spill to negative address?");
3797 
3798   Register base = sp;
3799 
3800   // An immediate offset in the range 0 to 255 which is multiplied
3801   // by the current vector or predicate register size in bytes.
3802   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3803     return Address(base, offset / sve_reg_size_in_bytes);
3804   }
3805 
3806   add(tmp, base, offset);
3807   return Address(tmp);
3808 }
3809 
3810 // Checks whether offset is aligned.
3811 // Returns true if it is, else false.
3812 bool MacroAssembler::merge_alignment_check(Register base,
3813                                            size_t size,
3814                                            int64_t cur_offset,
3815                                            int64_t prev_offset) const {
3816   if (AvoidUnalignedAccesses) {
3817     if (base == sp) {
3818       // Checks whether low offset if aligned to pair of registers.
3819       int64_t pair_mask = size * 2 - 1;
3820       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3821       return (offset & pair_mask) == 0;
3822     } else { // If base is not sp, we can't guarantee the access is aligned.
3823       return false;
3824     }
3825   } else {
3826     int64_t mask = size - 1;
3827     // Load/store pair instruction only supports element size aligned offset.
3828     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3829   }
3830 }
3831 
3832 // Checks whether current and previous loads/stores can be merged.
3833 // Returns true if it can be merged, else false.
3834 bool MacroAssembler::ldst_can_merge(Register rt,
3835                                     const Address &adr,
3836                                     size_t cur_size_in_bytes,
3837                                     bool is_store) const {
3838   address prev = pc() - NativeInstruction::instruction_size;
3839   address last = code()->last_insn();
3840 
3841   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3842     return false;
3843   }
3844 
3845   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3846     return false;
3847   }
3848 
3849   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3850   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3851 
3852   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3853   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3854 
3855   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3856     return false;
3857   }
3858 
3859   int64_t max_offset = 63 * prev_size_in_bytes;
3860   int64_t min_offset = -64 * prev_size_in_bytes;
3861 
3862   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3863 
3864   // Only same base can be merged.
3865   if (adr.base() != prev_ldst->base()) {
3866     return false;
3867   }
3868 
3869   int64_t cur_offset = adr.offset();
3870   int64_t prev_offset = prev_ldst->offset();
3871   size_t diff = abs(cur_offset - prev_offset);
3872   if (diff != prev_size_in_bytes) {
3873     return false;
3874   }
3875 
3876   // Following cases can not be merged:
3877   // ldr x2, [x2, #8]
3878   // ldr x3, [x2, #16]
3879   // or:
3880   // ldr x2, [x3, #8]
3881   // ldr x2, [x3, #16]
3882   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3883   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3884     return false;
3885   }
3886 
3887   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3888   // Offset range must be in ldp/stp instruction's range.
3889   if (low_offset > max_offset || low_offset < min_offset) {
3890     return false;
3891   }
3892 
3893   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3894     return true;
3895   }
3896 
3897   return false;
3898 }
3899 
3900 // Merge current load/store with previous load/store into ldp/stp.
3901 void MacroAssembler::merge_ldst(Register rt,
3902                                 const Address &adr,
3903                                 size_t cur_size_in_bytes,
3904                                 bool is_store) {
3905 
3906   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3907 
3908   Register rt_low, rt_high;
3909   address prev = pc() - NativeInstruction::instruction_size;
3910   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3911 
3912   int64_t offset;
3913 
3914   if (adr.offset() < prev_ldst->offset()) {
3915     offset = adr.offset();
3916     rt_low = rt;
3917     rt_high = prev_ldst->target();
3918   } else {
3919     offset = prev_ldst->offset();
3920     rt_low = prev_ldst->target();
3921     rt_high = rt;
3922   }
3923 
3924   Address adr_p = Address(prev_ldst->base(), offset);
3925   // Overwrite previous generated binary.
3926   code_section()->set_end(prev);
3927 
3928   const size_t sz = prev_ldst->size_in_bytes();
3929   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3930   if (!is_store) {
3931     BLOCK_COMMENT("merged ldr pair");
3932     if (sz == 8) {
3933       ldp(rt_low, rt_high, adr_p);
3934     } else {
3935       ldpw(rt_low, rt_high, adr_p);
3936     }
3937   } else {
3938     BLOCK_COMMENT("merged str pair");
3939     if (sz == 8) {
3940       stp(rt_low, rt_high, adr_p);
3941     } else {
3942       stpw(rt_low, rt_high, adr_p);
3943     }
3944   }
3945 }
3946 
3947 /**
3948  * Multiply 64 bit by 64 bit first loop.
3949  */
3950 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3951                                            Register y, Register y_idx, Register z,
3952                                            Register carry, Register product,
3953                                            Register idx, Register kdx) {
3954   //
3955   //  jlong carry, x[], y[], z[];
3956   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3957   //    huge_128 product = y[idx] * x[xstart] + carry;
3958   //    z[kdx] = (jlong)product;
3959   //    carry  = (jlong)(product >>> 64);
3960   //  }
3961   //  z[xstart] = carry;
3962   //
3963 
3964   Label L_first_loop, L_first_loop_exit;
3965   Label L_one_x, L_one_y, L_multiply;
3966 
3967   subsw(xstart, xstart, 1);
3968   br(Assembler::MI, L_one_x);
3969 
3970   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3971   ldr(x_xstart, Address(rscratch1));
3972   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3973 
3974   bind(L_first_loop);
3975   subsw(idx, idx, 1);
3976   br(Assembler::MI, L_first_loop_exit);
3977   subsw(idx, idx, 1);
3978   br(Assembler::MI, L_one_y);
3979   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3980   ldr(y_idx, Address(rscratch1));
3981   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
3982   bind(L_multiply);
3983 
3984   // AArch64 has a multiply-accumulate instruction that we can't use
3985   // here because it has no way to process carries, so we have to use
3986   // separate add and adc instructions.  Bah.
3987   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
3988   mul(product, x_xstart, y_idx);
3989   adds(product, product, carry);
3990   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
3991 
3992   subw(kdx, kdx, 2);
3993   ror(product, product, 32); // back to big-endian
3994   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
3995 
3996   b(L_first_loop);
3997 
3998   bind(L_one_y);
3999   ldrw(y_idx, Address(y,  0));
4000   b(L_multiply);
4001 
4002   bind(L_one_x);
4003   ldrw(x_xstart, Address(x,  0));
4004   b(L_first_loop);
4005 
4006   bind(L_first_loop_exit);
4007 }
4008 
4009 /**
4010  * Multiply 128 bit by 128. Unrolled inner loop.
4011  *
4012  */
4013 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4014                                              Register carry, Register carry2,
4015                                              Register idx, Register jdx,
4016                                              Register yz_idx1, Register yz_idx2,
4017                                              Register tmp, Register tmp3, Register tmp4,
4018                                              Register tmp6, Register product_hi) {
4019 
4020   //   jlong carry, x[], y[], z[];
4021   //   int kdx = ystart+1;
4022   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4023   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4024   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4025   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4026   //     carry  = (jlong)(tmp4 >>> 64);
4027   //     z[kdx+idx+1] = (jlong)tmp3;
4028   //     z[kdx+idx] = (jlong)tmp4;
4029   //   }
4030   //   idx += 2;
4031   //   if (idx > 0) {
4032   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4033   //     z[kdx+idx] = (jlong)yz_idx1;
4034   //     carry  = (jlong)(yz_idx1 >>> 64);
4035   //   }
4036   //
4037 
4038   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4039 
4040   lsrw(jdx, idx, 2);
4041 
4042   bind(L_third_loop);
4043 
4044   subsw(jdx, jdx, 1);
4045   br(Assembler::MI, L_third_loop_exit);
4046   subw(idx, idx, 4);
4047 
4048   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4049 
4050   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4051 
4052   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4053 
4054   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4055   ror(yz_idx2, yz_idx2, 32);
4056 
4057   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4058 
4059   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4060   umulh(tmp4, product_hi, yz_idx1);
4061 
4062   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4063   ror(rscratch2, rscratch2, 32);
4064 
4065   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4066   umulh(carry2, product_hi, yz_idx2);
4067 
4068   // propagate sum of both multiplications into carry:tmp4:tmp3
4069   adds(tmp3, tmp3, carry);
4070   adc(tmp4, tmp4, zr);
4071   adds(tmp3, tmp3, rscratch1);
4072   adcs(tmp4, tmp4, tmp);
4073   adc(carry, carry2, zr);
4074   adds(tmp4, tmp4, rscratch2);
4075   adc(carry, carry, zr);
4076 
4077   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4078   ror(tmp4, tmp4, 32);
4079   stp(tmp4, tmp3, Address(tmp6, 0));
4080 
4081   b(L_third_loop);
4082   bind (L_third_loop_exit);
4083 
4084   andw (idx, idx, 0x3);
4085   cbz(idx, L_post_third_loop_done);
4086 
4087   Label L_check_1;
4088   subsw(idx, idx, 2);
4089   br(Assembler::MI, L_check_1);
4090 
4091   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4092   ldr(yz_idx1, Address(rscratch1, 0));
4093   ror(yz_idx1, yz_idx1, 32);
4094   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4095   umulh(tmp4, product_hi, yz_idx1);
4096   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4097   ldr(yz_idx2, Address(rscratch1, 0));
4098   ror(yz_idx2, yz_idx2, 32);
4099 
4100   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4101 
4102   ror(tmp3, tmp3, 32);
4103   str(tmp3, Address(rscratch1, 0));
4104 
4105   bind (L_check_1);
4106 
4107   andw (idx, idx, 0x1);
4108   subsw(idx, idx, 1);
4109   br(Assembler::MI, L_post_third_loop_done);
4110   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4111   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4112   umulh(carry2, tmp4, product_hi);
4113   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4114 
4115   add2_with_carry(carry2, tmp3, tmp4, carry);
4116 
4117   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4118   extr(carry, carry2, tmp3, 32);
4119 
4120   bind(L_post_third_loop_done);
4121 }
4122 
4123 /**
4124  * Code for BigInteger::multiplyToLen() intrinsic.
4125  *
4126  * r0: x
4127  * r1: xlen
4128  * r2: y
4129  * r3: ylen
4130  * r4:  z
4131  * r5: tmp0
4132  * r10: tmp1
4133  * r11: tmp2
4134  * r12: tmp3
4135  * r13: tmp4
4136  * r14: tmp5
4137  * r15: tmp6
4138  * r16: tmp7
4139  *
4140  */
4141 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4142                                      Register z, Register tmp0,
4143                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4144                                      Register tmp5, Register tmp6, Register product_hi) {
4145 
4146   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4147 
4148   const Register idx = tmp1;
4149   const Register kdx = tmp2;
4150   const Register xstart = tmp3;
4151 
4152   const Register y_idx = tmp4;
4153   const Register carry = tmp5;
4154   const Register product  = xlen;
4155   const Register x_xstart = tmp0;
4156 
4157   // First Loop.
4158   //
4159   //  final static long LONG_MASK = 0xffffffffL;
4160   //  int xstart = xlen - 1;
4161   //  int ystart = ylen - 1;
4162   //  long carry = 0;
4163   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4164   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4165   //    z[kdx] = (int)product;
4166   //    carry = product >>> 32;
4167   //  }
4168   //  z[xstart] = (int)carry;
4169   //
4170 
4171   movw(idx, ylen);       // idx = ylen;
4172   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4173   mov(carry, zr);        // carry = 0;
4174 
4175   Label L_done;
4176 
4177   movw(xstart, xlen);
4178   subsw(xstart, xstart, 1);
4179   br(Assembler::MI, L_done);
4180 
4181   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4182 
4183   Label L_second_loop;
4184   cbzw(kdx, L_second_loop);
4185 
4186   Label L_carry;
4187   subw(kdx, kdx, 1);
4188   cbzw(kdx, L_carry);
4189 
4190   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4191   lsr(carry, carry, 32);
4192   subw(kdx, kdx, 1);
4193 
4194   bind(L_carry);
4195   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4196 
4197   // Second and third (nested) loops.
4198   //
4199   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4200   //   carry = 0;
4201   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4202   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4203   //                    (z[k] & LONG_MASK) + carry;
4204   //     z[k] = (int)product;
4205   //     carry = product >>> 32;
4206   //   }
4207   //   z[i] = (int)carry;
4208   // }
4209   //
4210   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4211 
4212   const Register jdx = tmp1;
4213 
4214   bind(L_second_loop);
4215   mov(carry, zr);                // carry = 0;
4216   movw(jdx, ylen);               // j = ystart+1
4217 
4218   subsw(xstart, xstart, 1);      // i = xstart-1;
4219   br(Assembler::MI, L_done);
4220 
4221   str(z, Address(pre(sp, -4 * wordSize)));
4222 
4223   Label L_last_x;
4224   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4225   subsw(xstart, xstart, 1);       // i = xstart-1;
4226   br(Assembler::MI, L_last_x);
4227 
4228   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4229   ldr(product_hi, Address(rscratch1));
4230   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4231 
4232   Label L_third_loop_prologue;
4233   bind(L_third_loop_prologue);
4234 
4235   str(ylen, Address(sp, wordSize));
4236   stp(x, xstart, Address(sp, 2 * wordSize));
4237   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4238                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4239   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4240   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4241 
4242   addw(tmp3, xlen, 1);
4243   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4244   subsw(tmp3, tmp3, 1);
4245   br(Assembler::MI, L_done);
4246 
4247   lsr(carry, carry, 32);
4248   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4249   b(L_second_loop);
4250 
4251   // Next infrequent code is moved outside loops.
4252   bind(L_last_x);
4253   ldrw(product_hi, Address(x,  0));
4254   b(L_third_loop_prologue);
4255 
4256   bind(L_done);
4257 }
4258 
4259 // Code for BigInteger::mulAdd intrinsic
4260 // out     = r0
4261 // in      = r1
4262 // offset  = r2  (already out.length-offset)
4263 // len     = r3
4264 // k       = r4
4265 //
4266 // pseudo code from java implementation:
4267 // carry = 0;
4268 // offset = out.length-offset - 1;
4269 // for (int j=len-1; j >= 0; j--) {
4270 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4271 //     out[offset--] = (int)product;
4272 //     carry = product >>> 32;
4273 // }
4274 // return (int)carry;
4275 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4276       Register len, Register k) {
4277     Label LOOP, END;
4278     // pre-loop
4279     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4280     csel(out, zr, out, Assembler::EQ);
4281     br(Assembler::EQ, END);
4282     add(in, in, len, LSL, 2); // in[j+1] address
4283     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4284     mov(out, zr); // used to keep carry now
4285     BIND(LOOP);
4286     ldrw(rscratch1, Address(pre(in, -4)));
4287     madd(rscratch1, rscratch1, k, out);
4288     ldrw(rscratch2, Address(pre(offset, -4)));
4289     add(rscratch1, rscratch1, rscratch2);
4290     strw(rscratch1, Address(offset));
4291     lsr(out, rscratch1, 32);
4292     subs(len, len, 1);
4293     br(Assembler::NE, LOOP);
4294     BIND(END);
4295 }
4296 
4297 /**
4298  * Emits code to update CRC-32 with a byte value according to constants in table
4299  *
4300  * @param [in,out]crc   Register containing the crc.
4301  * @param [in]val       Register containing the byte to fold into the CRC.
4302  * @param [in]table     Register containing the table of crc constants.
4303  *
4304  * uint32_t crc;
4305  * val = crc_table[(val ^ crc) & 0xFF];
4306  * crc = val ^ (crc >> 8);
4307  *
4308  */
4309 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4310   eor(val, val, crc);
4311   andr(val, val, 0xff);
4312   ldrw(val, Address(table, val, Address::lsl(2)));
4313   eor(crc, val, crc, Assembler::LSR, 8);
4314 }
4315 
4316 /**
4317  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4318  *
4319  * @param [in,out]crc   Register containing the crc.
4320  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4321  * @param [in]table0    Register containing table 0 of crc constants.
4322  * @param [in]table1    Register containing table 1 of crc constants.
4323  * @param [in]table2    Register containing table 2 of crc constants.
4324  * @param [in]table3    Register containing table 3 of crc constants.
4325  *
4326  * uint32_t crc;
4327  *   v = crc ^ v
4328  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4329  *
4330  */
4331 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4332         Register table0, Register table1, Register table2, Register table3,
4333         bool upper) {
4334   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4335   uxtb(tmp, v);
4336   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4337   ubfx(tmp, v, 8, 8);
4338   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4339   eor(crc, crc, tmp);
4340   ubfx(tmp, v, 16, 8);
4341   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4342   eor(crc, crc, tmp);
4343   ubfx(tmp, v, 24, 8);
4344   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4345   eor(crc, crc, tmp);
4346 }
4347 
4348 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4349         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4350     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4351     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4352 
4353     subs(tmp0, len, 384);
4354     mvnw(crc, crc);
4355     br(Assembler::GE, CRC_by128_pre);
4356   BIND(CRC_less128);
4357     subs(len, len, 32);
4358     br(Assembler::GE, CRC_by32_loop);
4359   BIND(CRC_less32);
4360     adds(len, len, 32 - 4);
4361     br(Assembler::GE, CRC_by4_loop);
4362     adds(len, len, 4);
4363     br(Assembler::GT, CRC_by1_loop);
4364     b(L_exit);
4365 
4366   BIND(CRC_by32_loop);
4367     ldp(tmp0, tmp1, Address(buf));
4368     crc32x(crc, crc, tmp0);
4369     ldp(tmp2, tmp3, Address(buf, 16));
4370     crc32x(crc, crc, tmp1);
4371     add(buf, buf, 32);
4372     crc32x(crc, crc, tmp2);
4373     subs(len, len, 32);
4374     crc32x(crc, crc, tmp3);
4375     br(Assembler::GE, CRC_by32_loop);
4376     cmn(len, (u1)32);
4377     br(Assembler::NE, CRC_less32);
4378     b(L_exit);
4379 
4380   BIND(CRC_by4_loop);
4381     ldrw(tmp0, Address(post(buf, 4)));
4382     subs(len, len, 4);
4383     crc32w(crc, crc, tmp0);
4384     br(Assembler::GE, CRC_by4_loop);
4385     adds(len, len, 4);
4386     br(Assembler::LE, L_exit);
4387   BIND(CRC_by1_loop);
4388     ldrb(tmp0, Address(post(buf, 1)));
4389     subs(len, len, 1);
4390     crc32b(crc, crc, tmp0);
4391     br(Assembler::GT, CRC_by1_loop);
4392     b(L_exit);
4393 
4394   BIND(CRC_by128_pre);
4395     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4396       4*256*sizeof(juint) + 8*sizeof(juint));
4397     mov(crc, 0);
4398     crc32x(crc, crc, tmp0);
4399     crc32x(crc, crc, tmp1);
4400 
4401     cbnz(len, CRC_less128);
4402 
4403   BIND(L_exit);
4404     mvnw(crc, crc);
4405 }
4406 
4407 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4408         Register len, Register tmp0, Register tmp1, Register tmp2,
4409         Register tmp3) {
4410     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4411     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4412 
4413     mvnw(crc, crc);
4414 
4415     subs(len, len, 128);
4416     br(Assembler::GE, CRC_by64_pre);
4417   BIND(CRC_less64);
4418     adds(len, len, 128-32);
4419     br(Assembler::GE, CRC_by32_loop);
4420   BIND(CRC_less32);
4421     adds(len, len, 32-4);
4422     br(Assembler::GE, CRC_by4_loop);
4423     adds(len, len, 4);
4424     br(Assembler::GT, CRC_by1_loop);
4425     b(L_exit);
4426 
4427   BIND(CRC_by32_loop);
4428     ldp(tmp0, tmp1, Address(post(buf, 16)));
4429     subs(len, len, 32);
4430     crc32x(crc, crc, tmp0);
4431     ldr(tmp2, Address(post(buf, 8)));
4432     crc32x(crc, crc, tmp1);
4433     ldr(tmp3, Address(post(buf, 8)));
4434     crc32x(crc, crc, tmp2);
4435     crc32x(crc, crc, tmp3);
4436     br(Assembler::GE, CRC_by32_loop);
4437     cmn(len, (u1)32);
4438     br(Assembler::NE, CRC_less32);
4439     b(L_exit);
4440 
4441   BIND(CRC_by4_loop);
4442     ldrw(tmp0, Address(post(buf, 4)));
4443     subs(len, len, 4);
4444     crc32w(crc, crc, tmp0);
4445     br(Assembler::GE, CRC_by4_loop);
4446     adds(len, len, 4);
4447     br(Assembler::LE, L_exit);
4448   BIND(CRC_by1_loop);
4449     ldrb(tmp0, Address(post(buf, 1)));
4450     subs(len, len, 1);
4451     crc32b(crc, crc, tmp0);
4452     br(Assembler::GT, CRC_by1_loop);
4453     b(L_exit);
4454 
4455   BIND(CRC_by64_pre);
4456     sub(buf, buf, 8);
4457     ldp(tmp0, tmp1, Address(buf, 8));
4458     crc32x(crc, crc, tmp0);
4459     ldr(tmp2, Address(buf, 24));
4460     crc32x(crc, crc, tmp1);
4461     ldr(tmp3, Address(buf, 32));
4462     crc32x(crc, crc, tmp2);
4463     ldr(tmp0, Address(buf, 40));
4464     crc32x(crc, crc, tmp3);
4465     ldr(tmp1, Address(buf, 48));
4466     crc32x(crc, crc, tmp0);
4467     ldr(tmp2, Address(buf, 56));
4468     crc32x(crc, crc, tmp1);
4469     ldr(tmp3, Address(pre(buf, 64)));
4470 
4471     b(CRC_by64_loop);
4472 
4473     align(CodeEntryAlignment);
4474   BIND(CRC_by64_loop);
4475     subs(len, len, 64);
4476     crc32x(crc, crc, tmp2);
4477     ldr(tmp0, Address(buf, 8));
4478     crc32x(crc, crc, tmp3);
4479     ldr(tmp1, Address(buf, 16));
4480     crc32x(crc, crc, tmp0);
4481     ldr(tmp2, Address(buf, 24));
4482     crc32x(crc, crc, tmp1);
4483     ldr(tmp3, Address(buf, 32));
4484     crc32x(crc, crc, tmp2);
4485     ldr(tmp0, Address(buf, 40));
4486     crc32x(crc, crc, tmp3);
4487     ldr(tmp1, Address(buf, 48));
4488     crc32x(crc, crc, tmp0);
4489     ldr(tmp2, Address(buf, 56));
4490     crc32x(crc, crc, tmp1);
4491     ldr(tmp3, Address(pre(buf, 64)));
4492     br(Assembler::GE, CRC_by64_loop);
4493 
4494     // post-loop
4495     crc32x(crc, crc, tmp2);
4496     crc32x(crc, crc, tmp3);
4497 
4498     sub(len, len, 64);
4499     add(buf, buf, 8);
4500     cmn(len, (u1)128);
4501     br(Assembler::NE, CRC_less64);
4502   BIND(L_exit);
4503     mvnw(crc, crc);
4504 }
4505 
4506 /**
4507  * @param crc   register containing existing CRC (32-bit)
4508  * @param buf   register pointing to input byte buffer (byte*)
4509  * @param len   register containing number of bytes
4510  * @param table register that will contain address of CRC table
4511  * @param tmp   scratch register
4512  */
4513 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4514         Register table0, Register table1, Register table2, Register table3,
4515         Register tmp, Register tmp2, Register tmp3) {
4516   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4517 
4518   if (UseCryptoPmullForCRC32) {
4519       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4520       return;
4521   }
4522 
4523   if (UseCRC32) {
4524       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4525       return;
4526   }
4527 
4528     mvnw(crc, crc);
4529 
4530     {
4531       uint64_t offset;
4532       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4533       add(table0, table0, offset);
4534     }
4535     add(table1, table0, 1*256*sizeof(juint));
4536     add(table2, table0, 2*256*sizeof(juint));
4537     add(table3, table0, 3*256*sizeof(juint));
4538 
4539     { // Neon code start
4540       cmp(len, (u1)64);
4541       br(Assembler::LT, L_by16);
4542       eor(v16, T16B, v16, v16);
4543 
4544     Label L_fold;
4545 
4546       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4547 
4548       ld1(v0, v1, T2D, post(buf, 32));
4549       ld1r(v4, T2D, post(tmp, 8));
4550       ld1r(v5, T2D, post(tmp, 8));
4551       ld1r(v6, T2D, post(tmp, 8));
4552       ld1r(v7, T2D, post(tmp, 8));
4553       mov(v16, S, 0, crc);
4554 
4555       eor(v0, T16B, v0, v16);
4556       sub(len, len, 64);
4557 
4558     BIND(L_fold);
4559       pmull(v22, T8H, v0, v5, T8B);
4560       pmull(v20, T8H, v0, v7, T8B);
4561       pmull(v23, T8H, v0, v4, T8B);
4562       pmull(v21, T8H, v0, v6, T8B);
4563 
4564       pmull2(v18, T8H, v0, v5, T16B);
4565       pmull2(v16, T8H, v0, v7, T16B);
4566       pmull2(v19, T8H, v0, v4, T16B);
4567       pmull2(v17, T8H, v0, v6, T16B);
4568 
4569       uzp1(v24, T8H, v20, v22);
4570       uzp2(v25, T8H, v20, v22);
4571       eor(v20, T16B, v24, v25);
4572 
4573       uzp1(v26, T8H, v16, v18);
4574       uzp2(v27, T8H, v16, v18);
4575       eor(v16, T16B, v26, v27);
4576 
4577       ushll2(v22, T4S, v20, T8H, 8);
4578       ushll(v20, T4S, v20, T4H, 8);
4579 
4580       ushll2(v18, T4S, v16, T8H, 8);
4581       ushll(v16, T4S, v16, T4H, 8);
4582 
4583       eor(v22, T16B, v23, v22);
4584       eor(v18, T16B, v19, v18);
4585       eor(v20, T16B, v21, v20);
4586       eor(v16, T16B, v17, v16);
4587 
4588       uzp1(v17, T2D, v16, v20);
4589       uzp2(v21, T2D, v16, v20);
4590       eor(v17, T16B, v17, v21);
4591 
4592       ushll2(v20, T2D, v17, T4S, 16);
4593       ushll(v16, T2D, v17, T2S, 16);
4594 
4595       eor(v20, T16B, v20, v22);
4596       eor(v16, T16B, v16, v18);
4597 
4598       uzp1(v17, T2D, v20, v16);
4599       uzp2(v21, T2D, v20, v16);
4600       eor(v28, T16B, v17, v21);
4601 
4602       pmull(v22, T8H, v1, v5, T8B);
4603       pmull(v20, T8H, v1, v7, T8B);
4604       pmull(v23, T8H, v1, v4, T8B);
4605       pmull(v21, T8H, v1, v6, T8B);
4606 
4607       pmull2(v18, T8H, v1, v5, T16B);
4608       pmull2(v16, T8H, v1, v7, T16B);
4609       pmull2(v19, T8H, v1, v4, T16B);
4610       pmull2(v17, T8H, v1, v6, T16B);
4611 
4612       ld1(v0, v1, T2D, post(buf, 32));
4613 
4614       uzp1(v24, T8H, v20, v22);
4615       uzp2(v25, T8H, v20, v22);
4616       eor(v20, T16B, v24, v25);
4617 
4618       uzp1(v26, T8H, v16, v18);
4619       uzp2(v27, T8H, v16, v18);
4620       eor(v16, T16B, v26, v27);
4621 
4622       ushll2(v22, T4S, v20, T8H, 8);
4623       ushll(v20, T4S, v20, T4H, 8);
4624 
4625       ushll2(v18, T4S, v16, T8H, 8);
4626       ushll(v16, T4S, v16, T4H, 8);
4627 
4628       eor(v22, T16B, v23, v22);
4629       eor(v18, T16B, v19, v18);
4630       eor(v20, T16B, v21, v20);
4631       eor(v16, T16B, v17, v16);
4632 
4633       uzp1(v17, T2D, v16, v20);
4634       uzp2(v21, T2D, v16, v20);
4635       eor(v16, T16B, v17, v21);
4636 
4637       ushll2(v20, T2D, v16, T4S, 16);
4638       ushll(v16, T2D, v16, T2S, 16);
4639 
4640       eor(v20, T16B, v22, v20);
4641       eor(v16, T16B, v16, v18);
4642 
4643       uzp1(v17, T2D, v20, v16);
4644       uzp2(v21, T2D, v20, v16);
4645       eor(v20, T16B, v17, v21);
4646 
4647       shl(v16, T2D, v28, 1);
4648       shl(v17, T2D, v20, 1);
4649 
4650       eor(v0, T16B, v0, v16);
4651       eor(v1, T16B, v1, v17);
4652 
4653       subs(len, len, 32);
4654       br(Assembler::GE, L_fold);
4655 
4656       mov(crc, 0);
4657       mov(tmp, v0, D, 0);
4658       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4659       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4660       mov(tmp, v0, D, 1);
4661       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4662       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4663       mov(tmp, v1, D, 0);
4664       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4665       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4666       mov(tmp, v1, D, 1);
4667       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4668       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4669 
4670       add(len, len, 32);
4671     } // Neon code end
4672 
4673   BIND(L_by16);
4674     subs(len, len, 16);
4675     br(Assembler::GE, L_by16_loop);
4676     adds(len, len, 16-4);
4677     br(Assembler::GE, L_by4_loop);
4678     adds(len, len, 4);
4679     br(Assembler::GT, L_by1_loop);
4680     b(L_exit);
4681 
4682   BIND(L_by4_loop);
4683     ldrw(tmp, Address(post(buf, 4)));
4684     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4685     subs(len, len, 4);
4686     br(Assembler::GE, L_by4_loop);
4687     adds(len, len, 4);
4688     br(Assembler::LE, L_exit);
4689   BIND(L_by1_loop);
4690     subs(len, len, 1);
4691     ldrb(tmp, Address(post(buf, 1)));
4692     update_byte_crc32(crc, tmp, table0);
4693     br(Assembler::GT, L_by1_loop);
4694     b(L_exit);
4695 
4696     align(CodeEntryAlignment);
4697   BIND(L_by16_loop);
4698     subs(len, len, 16);
4699     ldp(tmp, tmp3, Address(post(buf, 16)));
4700     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4701     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4702     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4703     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4704     br(Assembler::GE, L_by16_loop);
4705     adds(len, len, 16-4);
4706     br(Assembler::GE, L_by4_loop);
4707     adds(len, len, 4);
4708     br(Assembler::GT, L_by1_loop);
4709   BIND(L_exit);
4710     mvnw(crc, crc);
4711 }
4712 
4713 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4714         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4715     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4716     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4717 
4718     subs(tmp0, len, 384);
4719     br(Assembler::GE, CRC_by128_pre);
4720   BIND(CRC_less128);
4721     subs(len, len, 32);
4722     br(Assembler::GE, CRC_by32_loop);
4723   BIND(CRC_less32);
4724     adds(len, len, 32 - 4);
4725     br(Assembler::GE, CRC_by4_loop);
4726     adds(len, len, 4);
4727     br(Assembler::GT, CRC_by1_loop);
4728     b(L_exit);
4729 
4730   BIND(CRC_by32_loop);
4731     ldp(tmp0, tmp1, Address(buf));
4732     crc32cx(crc, crc, tmp0);
4733     ldr(tmp2, Address(buf, 16));
4734     crc32cx(crc, crc, tmp1);
4735     ldr(tmp3, Address(buf, 24));
4736     crc32cx(crc, crc, tmp2);
4737     add(buf, buf, 32);
4738     subs(len, len, 32);
4739     crc32cx(crc, crc, tmp3);
4740     br(Assembler::GE, CRC_by32_loop);
4741     cmn(len, (u1)32);
4742     br(Assembler::NE, CRC_less32);
4743     b(L_exit);
4744 
4745   BIND(CRC_by4_loop);
4746     ldrw(tmp0, Address(post(buf, 4)));
4747     subs(len, len, 4);
4748     crc32cw(crc, crc, tmp0);
4749     br(Assembler::GE, CRC_by4_loop);
4750     adds(len, len, 4);
4751     br(Assembler::LE, L_exit);
4752   BIND(CRC_by1_loop);
4753     ldrb(tmp0, Address(post(buf, 1)));
4754     subs(len, len, 1);
4755     crc32cb(crc, crc, tmp0);
4756     br(Assembler::GT, CRC_by1_loop);
4757     b(L_exit);
4758 
4759   BIND(CRC_by128_pre);
4760     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4761       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4762     mov(crc, 0);
4763     crc32cx(crc, crc, tmp0);
4764     crc32cx(crc, crc, tmp1);
4765 
4766     cbnz(len, CRC_less128);
4767 
4768   BIND(L_exit);
4769 }
4770 
4771 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4772         Register len, Register tmp0, Register tmp1, Register tmp2,
4773         Register tmp3) {
4774     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4775     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4776 
4777     subs(len, len, 128);
4778     br(Assembler::GE, CRC_by64_pre);
4779   BIND(CRC_less64);
4780     adds(len, len, 128-32);
4781     br(Assembler::GE, CRC_by32_loop);
4782   BIND(CRC_less32);
4783     adds(len, len, 32-4);
4784     br(Assembler::GE, CRC_by4_loop);
4785     adds(len, len, 4);
4786     br(Assembler::GT, CRC_by1_loop);
4787     b(L_exit);
4788 
4789   BIND(CRC_by32_loop);
4790     ldp(tmp0, tmp1, Address(post(buf, 16)));
4791     subs(len, len, 32);
4792     crc32cx(crc, crc, tmp0);
4793     ldr(tmp2, Address(post(buf, 8)));
4794     crc32cx(crc, crc, tmp1);
4795     ldr(tmp3, Address(post(buf, 8)));
4796     crc32cx(crc, crc, tmp2);
4797     crc32cx(crc, crc, tmp3);
4798     br(Assembler::GE, CRC_by32_loop);
4799     cmn(len, (u1)32);
4800     br(Assembler::NE, CRC_less32);
4801     b(L_exit);
4802 
4803   BIND(CRC_by4_loop);
4804     ldrw(tmp0, Address(post(buf, 4)));
4805     subs(len, len, 4);
4806     crc32cw(crc, crc, tmp0);
4807     br(Assembler::GE, CRC_by4_loop);
4808     adds(len, len, 4);
4809     br(Assembler::LE, L_exit);
4810   BIND(CRC_by1_loop);
4811     ldrb(tmp0, Address(post(buf, 1)));
4812     subs(len, len, 1);
4813     crc32cb(crc, crc, tmp0);
4814     br(Assembler::GT, CRC_by1_loop);
4815     b(L_exit);
4816 
4817   BIND(CRC_by64_pre);
4818     sub(buf, buf, 8);
4819     ldp(tmp0, tmp1, Address(buf, 8));
4820     crc32cx(crc, crc, tmp0);
4821     ldr(tmp2, Address(buf, 24));
4822     crc32cx(crc, crc, tmp1);
4823     ldr(tmp3, Address(buf, 32));
4824     crc32cx(crc, crc, tmp2);
4825     ldr(tmp0, Address(buf, 40));
4826     crc32cx(crc, crc, tmp3);
4827     ldr(tmp1, Address(buf, 48));
4828     crc32cx(crc, crc, tmp0);
4829     ldr(tmp2, Address(buf, 56));
4830     crc32cx(crc, crc, tmp1);
4831     ldr(tmp3, Address(pre(buf, 64)));
4832 
4833     b(CRC_by64_loop);
4834 
4835     align(CodeEntryAlignment);
4836   BIND(CRC_by64_loop);
4837     subs(len, len, 64);
4838     crc32cx(crc, crc, tmp2);
4839     ldr(tmp0, Address(buf, 8));
4840     crc32cx(crc, crc, tmp3);
4841     ldr(tmp1, Address(buf, 16));
4842     crc32cx(crc, crc, tmp0);
4843     ldr(tmp2, Address(buf, 24));
4844     crc32cx(crc, crc, tmp1);
4845     ldr(tmp3, Address(buf, 32));
4846     crc32cx(crc, crc, tmp2);
4847     ldr(tmp0, Address(buf, 40));
4848     crc32cx(crc, crc, tmp3);
4849     ldr(tmp1, Address(buf, 48));
4850     crc32cx(crc, crc, tmp0);
4851     ldr(tmp2, Address(buf, 56));
4852     crc32cx(crc, crc, tmp1);
4853     ldr(tmp3, Address(pre(buf, 64)));
4854     br(Assembler::GE, CRC_by64_loop);
4855 
4856     // post-loop
4857     crc32cx(crc, crc, tmp2);
4858     crc32cx(crc, crc, tmp3);
4859 
4860     sub(len, len, 64);
4861     add(buf, buf, 8);
4862     cmn(len, (u1)128);
4863     br(Assembler::NE, CRC_less64);
4864   BIND(L_exit);
4865 }
4866 
4867 /**
4868  * @param crc   register containing existing CRC (32-bit)
4869  * @param buf   register pointing to input byte buffer (byte*)
4870  * @param len   register containing number of bytes
4871  * @param table register that will contain address of CRC table
4872  * @param tmp   scratch register
4873  */
4874 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4875         Register table0, Register table1, Register table2, Register table3,
4876         Register tmp, Register tmp2, Register tmp3) {
4877   if (UseCryptoPmullForCRC32) {
4878     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4879   } else {
4880     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4881   }
4882 }
4883 
4884 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4885         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4886     Label CRC_by128_loop;
4887     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4888 
4889     sub(len, len, 256);
4890     Register table = tmp0;
4891     {
4892       uint64_t offset;
4893       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4894       add(table, table, offset);
4895     }
4896     add(table, table, table_offset);
4897 
4898     // Registers v0..v7 are used as data registers.
4899     // Registers v16..v31 are used as tmp registers.
4900     sub(buf, buf, 0x10);
4901     ldrq(v0, Address(buf, 0x10));
4902     ldrq(v1, Address(buf, 0x20));
4903     ldrq(v2, Address(buf, 0x30));
4904     ldrq(v3, Address(buf, 0x40));
4905     ldrq(v4, Address(buf, 0x50));
4906     ldrq(v5, Address(buf, 0x60));
4907     ldrq(v6, Address(buf, 0x70));
4908     ldrq(v7, Address(pre(buf, 0x80)));
4909 
4910     movi(v31, T4S, 0);
4911     mov(v31, S, 0, crc);
4912     eor(v0, T16B, v0, v31);
4913 
4914     // Register v16 contains constants from the crc table.
4915     ldrq(v16, Address(table));
4916     b(CRC_by128_loop);
4917 
4918     align(OptoLoopAlignment);
4919   BIND(CRC_by128_loop);
4920     pmull (v17,  T1Q, v0, v16, T1D);
4921     pmull2(v18, T1Q, v0, v16, T2D);
4922     ldrq(v0, Address(buf, 0x10));
4923     eor3(v0, T16B, v17,  v18, v0);
4924 
4925     pmull (v19, T1Q, v1, v16, T1D);
4926     pmull2(v20, T1Q, v1, v16, T2D);
4927     ldrq(v1, Address(buf, 0x20));
4928     eor3(v1, T16B, v19, v20, v1);
4929 
4930     pmull (v21, T1Q, v2, v16, T1D);
4931     pmull2(v22, T1Q, v2, v16, T2D);
4932     ldrq(v2, Address(buf, 0x30));
4933     eor3(v2, T16B, v21, v22, v2);
4934 
4935     pmull (v23, T1Q, v3, v16, T1D);
4936     pmull2(v24, T1Q, v3, v16, T2D);
4937     ldrq(v3, Address(buf, 0x40));
4938     eor3(v3, T16B, v23, v24, v3);
4939 
4940     pmull (v25, T1Q, v4, v16, T1D);
4941     pmull2(v26, T1Q, v4, v16, T2D);
4942     ldrq(v4, Address(buf, 0x50));
4943     eor3(v4, T16B, v25, v26, v4);
4944 
4945     pmull (v27, T1Q, v5, v16, T1D);
4946     pmull2(v28, T1Q, v5, v16, T2D);
4947     ldrq(v5, Address(buf, 0x60));
4948     eor3(v5, T16B, v27, v28, v5);
4949 
4950     pmull (v29, T1Q, v6, v16, T1D);
4951     pmull2(v30, T1Q, v6, v16, T2D);
4952     ldrq(v6, Address(buf, 0x70));
4953     eor3(v6, T16B, v29, v30, v6);
4954 
4955     // Reuse registers v23, v24.
4956     // Using them won't block the first instruction of the next iteration.
4957     pmull (v23, T1Q, v7, v16, T1D);
4958     pmull2(v24, T1Q, v7, v16, T2D);
4959     ldrq(v7, Address(pre(buf, 0x80)));
4960     eor3(v7, T16B, v23, v24, v7);
4961 
4962     subs(len, len, 0x80);
4963     br(Assembler::GE, CRC_by128_loop);
4964 
4965     // fold into 512 bits
4966     // Use v31 for constants because v16 can be still in use.
4967     ldrq(v31, Address(table, 0x10));
4968 
4969     pmull (v17,  T1Q, v0, v31, T1D);
4970     pmull2(v18, T1Q, v0, v31, T2D);
4971     eor3(v0, T16B, v17, v18, v4);
4972 
4973     pmull (v19, T1Q, v1, v31, T1D);
4974     pmull2(v20, T1Q, v1, v31, T2D);
4975     eor3(v1, T16B, v19, v20, v5);
4976 
4977     pmull (v21, T1Q, v2, v31, T1D);
4978     pmull2(v22, T1Q, v2, v31, T2D);
4979     eor3(v2, T16B, v21, v22, v6);
4980 
4981     pmull (v23, T1Q, v3, v31, T1D);
4982     pmull2(v24, T1Q, v3, v31, T2D);
4983     eor3(v3, T16B, v23, v24, v7);
4984 
4985     // fold into 128 bits
4986     // Use v17 for constants because v31 can be still in use.
4987     ldrq(v17, Address(table, 0x20));
4988     pmull (v25, T1Q, v0, v17, T1D);
4989     pmull2(v26, T1Q, v0, v17, T2D);
4990     eor3(v3, T16B, v3, v25, v26);
4991 
4992     // Use v18 for constants because v17 can be still in use.
4993     ldrq(v18, Address(table, 0x30));
4994     pmull (v27, T1Q, v1, v18, T1D);
4995     pmull2(v28, T1Q, v1, v18, T2D);
4996     eor3(v3, T16B, v3, v27, v28);
4997 
4998     // Use v19 for constants because v18 can be still in use.
4999     ldrq(v19, Address(table, 0x40));
5000     pmull (v29, T1Q, v2, v19, T1D);
5001     pmull2(v30, T1Q, v2, v19, T2D);
5002     eor3(v0, T16B, v3, v29, v30);
5003 
5004     add(len, len, 0x80);
5005     add(buf, buf, 0x10);
5006 
5007     mov(tmp0, v0, D, 0);
5008     mov(tmp1, v0, D, 1);
5009 }
5010 
5011 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5012   Address adr;
5013   switch(dst.getMode()) {
5014   case Address::base_plus_offset:
5015     // This is the expected mode, although we allow all the other
5016     // forms below.
5017     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5018     break;
5019   default:
5020     lea(rscratch2, dst);
5021     adr = Address(rscratch2);
5022     break;
5023   }
5024   ldr(rscratch1, adr);
5025   add(rscratch1, rscratch1, src);
5026   str(rscratch1, adr);
5027 }
5028 
5029 void MacroAssembler::cmpptr(Register src1, Address src2) {
5030   uint64_t offset;
5031   adrp(rscratch1, src2, offset);
5032   ldr(rscratch1, Address(rscratch1, offset));
5033   cmp(src1, rscratch1);
5034 }
5035 
5036 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5037   cmp(obj1, obj2);
5038 }
5039 
5040 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5041   load_method_holder(rresult, rmethod);
5042   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5043 }
5044 
5045 void MacroAssembler::load_method_holder(Register holder, Register method) {
5046   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5047   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5048   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5049 }
5050 
5051 // Loads the obj's Klass* into dst.
5052 // Preserves all registers (incl src, rscratch1 and rscratch2).
5053 // Input:
5054 // src - the oop we want to load the klass from.
5055 // dst - output narrow klass.
5056 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5057   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5058   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5059   lsr(dst, dst, markWord::klass_shift);
5060 }
5061 
5062 void MacroAssembler::load_klass(Register dst, Register src) {
5063   if (UseCompactObjectHeaders) {
5064     load_narrow_klass_compact(dst, src);
5065   } else {
5066     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5067   }
5068   decode_klass_not_null(dst);
5069 }
5070 
5071 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5072   if (RestoreMXCSROnJNICalls) {
5073     Label OK;
5074     get_fpcr(tmp1);
5075     mov(tmp2, tmp1);
5076     // Set FPCR to the state we need. We do want Round to Nearest. We
5077     // don't want non-IEEE rounding modes or floating-point traps.
5078     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5079     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5080     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5081     eor(tmp2, tmp1, tmp2);
5082     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5083     set_fpcr(tmp1);
5084     bind(OK);
5085   }
5086 }
5087 
5088 // ((OopHandle)result).resolve();
5089 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5090   // OopHandle::resolve is an indirection.
5091   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5092 }
5093 
5094 // ((WeakHandle)result).resolve();
5095 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5096   assert_different_registers(result, tmp1, tmp2);
5097   Label resolved;
5098 
5099   // A null weak handle resolves to null.
5100   cbz(result, resolved);
5101 
5102   // Only 64 bit platforms support GCs that require a tmp register
5103   // WeakHandle::resolve is an indirection like jweak.
5104   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5105                  result, Address(result), tmp1, tmp2);
5106   bind(resolved);
5107 }
5108 
5109 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5110   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5111   ldr(dst, Address(rmethod, Method::const_offset()));
5112   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5113   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5114   ldr(dst, Address(dst, mirror_offset));
5115   resolve_oop_handle(dst, tmp1, tmp2);
5116 }
5117 
5118 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5119   assert_different_registers(obj, klass, tmp);
5120   if (UseCompactObjectHeaders) {
5121     load_narrow_klass_compact(tmp, obj);
5122   } else {
5123     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5124   }
5125   if (CompressedKlassPointers::base() == nullptr) {
5126     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5127     return;
5128   } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5129              && CompressedKlassPointers::shift() == 0) {
5130     // Only the bottom 32 bits matter
5131     cmpw(klass, tmp);
5132     return;
5133   }
5134   decode_klass_not_null(tmp);
5135   cmp(klass, tmp);
5136 }
5137 
5138 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5139   if (UseCompactObjectHeaders) {
5140     load_narrow_klass_compact(tmp1, obj1);
5141     load_narrow_klass_compact(tmp2,  obj2);
5142   } else {
5143     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5144     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5145   }
5146   cmpw(tmp1, tmp2);
5147 }
5148 
5149 void MacroAssembler::store_klass(Register dst, Register src) {
5150   // FIXME: Should this be a store release?  concurrent gcs assumes
5151   // klass length is valid if klass field is not null.
5152   assert(!UseCompactObjectHeaders, "not with compact headers");
5153   encode_klass_not_null(src);
5154   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5155 }
5156 
5157 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5158   assert(!UseCompactObjectHeaders, "not with compact headers");
5159   // Store to klass gap in destination
5160   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5161 }
5162 
5163 // Algorithm must match CompressedOops::encode.
5164 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5165 #ifdef ASSERT
5166   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5167 #endif
5168   verify_oop_msg(s, "broken oop in encode_heap_oop");
5169   if (CompressedOops::base() == nullptr) {
5170     if (CompressedOops::shift() != 0) {
5171       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5172       lsr(d, s, LogMinObjAlignmentInBytes);
5173     } else {
5174       mov(d, s);
5175     }
5176   } else {
5177     subs(d, s, rheapbase);
5178     csel(d, d, zr, Assembler::HS);
5179     lsr(d, d, LogMinObjAlignmentInBytes);
5180 
5181     /*  Old algorithm: is this any worse?
5182     Label nonnull;
5183     cbnz(r, nonnull);
5184     sub(r, r, rheapbase);
5185     bind(nonnull);
5186     lsr(r, r, LogMinObjAlignmentInBytes);
5187     */
5188   }
5189 }
5190 
5191 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5192 #ifdef ASSERT
5193   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5194   if (CheckCompressedOops) {
5195     Label ok;
5196     cbnz(r, ok);
5197     stop("null oop passed to encode_heap_oop_not_null");
5198     bind(ok);
5199   }
5200 #endif
5201   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5202   if (CompressedOops::base() != nullptr) {
5203     sub(r, r, rheapbase);
5204   }
5205   if (CompressedOops::shift() != 0) {
5206     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5207     lsr(r, r, LogMinObjAlignmentInBytes);
5208   }
5209 }
5210 
5211 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5212 #ifdef ASSERT
5213   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5214   if (CheckCompressedOops) {
5215     Label ok;
5216     cbnz(src, ok);
5217     stop("null oop passed to encode_heap_oop_not_null2");
5218     bind(ok);
5219   }
5220 #endif
5221   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5222 
5223   Register data = src;
5224   if (CompressedOops::base() != nullptr) {
5225     sub(dst, src, rheapbase);
5226     data = dst;
5227   }
5228   if (CompressedOops::shift() != 0) {
5229     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5230     lsr(dst, data, LogMinObjAlignmentInBytes);
5231     data = dst;
5232   }
5233   if (data == src)
5234     mov(dst, src);
5235 }
5236 
5237 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5238 #ifdef ASSERT
5239   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5240 #endif
5241   if (CompressedOops::base() == nullptr) {
5242     if (CompressedOops::shift() != 0) {
5243       lsl(d, s, CompressedOops::shift());
5244     } else if (d != s) {
5245       mov(d, s);
5246     }
5247   } else {
5248     Label done;
5249     if (d != s)
5250       mov(d, s);
5251     cbz(s, done);
5252     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5253     bind(done);
5254   }
5255   verify_oop_msg(d, "broken oop in decode_heap_oop");
5256 }
5257 
5258 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5259   assert (UseCompressedOops, "should only be used for compressed headers");
5260   assert (Universe::heap() != nullptr, "java heap should be initialized");
5261   // Cannot assert, unverified entry point counts instructions (see .ad file)
5262   // vtableStubs also counts instructions in pd_code_size_limit.
5263   // Also do not verify_oop as this is called by verify_oop.
5264   if (CompressedOops::shift() != 0) {
5265     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5266     if (CompressedOops::base() != nullptr) {
5267       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5268     } else {
5269       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5270     }
5271   } else {
5272     assert (CompressedOops::base() == nullptr, "sanity");
5273   }
5274 }
5275 
5276 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5277   assert (UseCompressedOops, "should only be used for compressed headers");
5278   assert (Universe::heap() != nullptr, "java heap should be initialized");
5279   // Cannot assert, unverified entry point counts instructions (see .ad file)
5280   // vtableStubs also counts instructions in pd_code_size_limit.
5281   // Also do not verify_oop as this is called by verify_oop.
5282   if (CompressedOops::shift() != 0) {
5283     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5284     if (CompressedOops::base() != nullptr) {
5285       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5286     } else {
5287       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5288     }
5289   } else {
5290     assert (CompressedOops::base() == nullptr, "sanity");
5291     if (dst != src) {
5292       mov(dst, src);
5293     }
5294   }
5295 }
5296 
5297 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5298 
5299 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5300   assert(Metaspace::initialized(), "metaspace not initialized yet");
5301   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5302   return _klass_decode_mode;
5303 }
5304 
5305 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5306   // KlassDecodeMode shouldn't be set already.
5307   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5308 
5309   if (base == nullptr) {
5310     return KlassDecodeZero;
5311   }
5312 
5313   if (operand_valid_for_logical_immediate(
5314         /*is32*/false, (uint64_t)base)) {
5315     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5316     if (((uint64_t)base & range_mask) == 0) {
5317       return KlassDecodeXor;
5318     }
5319   }
5320 
5321   const uint64_t shifted_base =
5322     (uint64_t)base >> shift;
5323   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5324     return KlassDecodeMovk;
5325   }
5326 
5327   // No valid encoding.
5328   return KlassDecodeNone;
5329 }
5330 
5331 // Check if one of the above decoding modes will work for given base, shift and range.
5332 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5333   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5334 }
5335 
5336 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5337   _klass_decode_mode = klass_decode_mode(base, shift, range);
5338   return _klass_decode_mode != KlassDecodeNone;
5339 }
5340 
5341 static Register pick_different_tmp(Register dst, Register src) {
5342   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5343   return *tmps.begin();
5344 }
5345 
5346 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5347   // we have to load the klass base from the AOT constants area but
5348   // not the shift because it is not allowed to change
5349   int shift = CompressedKlassPointers::shift();
5350   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5351   if (dst != src) {
5352     // we can load the base into dst, subtract it formthe src and shift down
5353     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5354     ldr(dst, dst);
5355     sub(dst, src, dst);
5356     lsr(dst, dst, shift);
5357   } else {
5358     // we need an extra register in order to load the coop base
5359     Register tmp = pick_different_tmp(dst, src);
5360     RegSet regs = RegSet::of(tmp);
5361     push(regs, sp);
5362     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5363     ldr(tmp, tmp);
5364     sub(dst, src, tmp);
5365     lsr(dst, dst, shift);
5366     pop(regs, sp);
5367   }
5368 }
5369 
5370 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5371   if (AOTCodeCache::is_on_for_dump()) {
5372     encode_klass_not_null_for_aot(dst, src);
5373     return;
5374   }
5375 
5376   switch (klass_decode_mode()) {
5377   case KlassDecodeZero:
5378     if (CompressedKlassPointers::shift() != 0) {
5379       lsr(dst, src, CompressedKlassPointers::shift());
5380     } else {
5381       if (dst != src) mov(dst, src);
5382     }
5383     break;
5384 
5385   case KlassDecodeXor:
5386     if (CompressedKlassPointers::shift() != 0) {
5387       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5388       lsr(dst, dst, CompressedKlassPointers::shift());
5389     } else {
5390       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5391     }
5392     break;
5393 
5394   case KlassDecodeMovk:
5395     if (CompressedKlassPointers::shift() != 0) {
5396       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5397     } else {
5398       movw(dst, src);
5399     }
5400     break;
5401 
5402   case KlassDecodeNone:
5403     ShouldNotReachHere();
5404     break;
5405   }
5406 }
5407 
5408 void MacroAssembler::encode_klass_not_null(Register r) {
5409   encode_klass_not_null(r, r);
5410 }
5411 
5412 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5413   // we have to load the klass base from the AOT constants area but
5414   // not the shift because it is not allowed to change
5415   int shift = CompressedKlassPointers::shift();
5416   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5417   if (dst != src) {
5418     // we can load the base into dst then add the offset with a suitable shift
5419     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5420     ldr(dst, dst);
5421     add(dst, dst, src, LSL,  shift);
5422   } else {
5423     // we need an extra register in order to load the coop base
5424     Register tmp = pick_different_tmp(dst, src);
5425     RegSet regs = RegSet::of(tmp);
5426     push(regs, sp);
5427     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5428     ldr(tmp, tmp);
5429     add(dst, tmp,  src, LSL,  shift);
5430     pop(regs, sp);
5431   }
5432 }
5433 
5434 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5435   if (AOTCodeCache::is_on_for_dump()) {
5436     decode_klass_not_null_for_aot(dst, src);
5437     return;
5438   }
5439 
5440   switch (klass_decode_mode()) {
5441   case KlassDecodeZero:
5442     if (CompressedKlassPointers::shift() != 0) {
5443       lsl(dst, src, CompressedKlassPointers::shift());
5444     } else {
5445       if (dst != src) mov(dst, src);
5446     }
5447     break;
5448 
5449   case KlassDecodeXor:
5450     if (CompressedKlassPointers::shift() != 0) {
5451       lsl(dst, src, CompressedKlassPointers::shift());
5452       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5453     } else {
5454       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5455     }
5456     break;
5457 
5458   case KlassDecodeMovk: {
5459     const uint64_t shifted_base =
5460       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5461 
5462     if (dst != src) movw(dst, src);
5463     movk(dst, shifted_base >> 32, 32);
5464 
5465     if (CompressedKlassPointers::shift() != 0) {
5466       lsl(dst, dst, CompressedKlassPointers::shift());
5467     }
5468 
5469     break;
5470   }
5471 
5472   case KlassDecodeNone:
5473     ShouldNotReachHere();
5474     break;
5475   }
5476 }
5477 
5478 void  MacroAssembler::decode_klass_not_null(Register r) {
5479   decode_klass_not_null(r, r);
5480 }
5481 
5482 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5483 #ifdef ASSERT
5484   {
5485     ThreadInVMfromUnknown tiv;
5486     assert (UseCompressedOops, "should only be used for compressed oops");
5487     assert (Universe::heap() != nullptr, "java heap should be initialized");
5488     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5489     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5490   }
5491 #endif
5492   int oop_index = oop_recorder()->find_index(obj);
5493   InstructionMark im(this);
5494   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5495   code_section()->relocate(inst_mark(), rspec);
5496   movz(dst, 0xDEAD, 16);
5497   movk(dst, 0xBEEF);
5498 }
5499 
5500 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5501   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5502   int index = oop_recorder()->find_index(k);
5503 
5504   InstructionMark im(this);
5505   RelocationHolder rspec = metadata_Relocation::spec(index);
5506   code_section()->relocate(inst_mark(), rspec);
5507   narrowKlass nk = CompressedKlassPointers::encode(k);
5508   movz(dst, (nk >> 16), 16);
5509   movk(dst, nk & 0xffff);
5510 }
5511 
5512 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5513                                     Register dst, Address src,
5514                                     Register tmp1, Register tmp2) {
5515   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5516   decorators = AccessInternal::decorator_fixup(decorators, type);
5517   bool as_raw = (decorators & AS_RAW) != 0;
5518   if (as_raw) {
5519     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5520   } else {
5521     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5522   }
5523 }
5524 
5525 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5526                                      Address dst, Register val,
5527                                      Register tmp1, Register tmp2, Register tmp3) {
5528   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5529   decorators = AccessInternal::decorator_fixup(decorators, type);
5530   bool as_raw = (decorators & AS_RAW) != 0;
5531   if (as_raw) {
5532     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5533   } else {
5534     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5535   }
5536 }
5537 
5538 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5539                                    Register tmp2, DecoratorSet decorators) {
5540   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5541 }
5542 
5543 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5544                                             Register tmp2, DecoratorSet decorators) {
5545   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5546 }
5547 
5548 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5549                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5550   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5551 }
5552 
5553 // Used for storing nulls.
5554 void MacroAssembler::store_heap_oop_null(Address dst) {
5555   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5556 }
5557 
5558 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5559   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5560   int index = oop_recorder()->allocate_metadata_index(obj);
5561   RelocationHolder rspec = metadata_Relocation::spec(index);
5562   return Address((address)obj, rspec);
5563 }
5564 
5565 // Move an oop into a register.
5566 void MacroAssembler::movoop(Register dst, jobject obj) {
5567   int oop_index;
5568   if (obj == nullptr) {
5569     oop_index = oop_recorder()->allocate_oop_index(obj);
5570   } else {
5571 #ifdef ASSERT
5572     {
5573       ThreadInVMfromUnknown tiv;
5574       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5575     }
5576 #endif
5577     oop_index = oop_recorder()->find_index(obj);
5578   }
5579   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5580 
5581   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5582     mov(dst, Address((address)obj, rspec));
5583   } else {
5584     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5585     ldr(dst, Address(dummy, rspec));
5586   }
5587 }
5588 
5589 // Move a metadata address into a register.
5590 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5591   int oop_index;
5592   if (obj == nullptr) {
5593     oop_index = oop_recorder()->allocate_metadata_index(obj);
5594   } else {
5595     oop_index = oop_recorder()->find_index(obj);
5596   }
5597   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5598   mov(dst, Address((address)obj, rspec));
5599 }
5600 
5601 Address MacroAssembler::constant_oop_address(jobject obj) {
5602 #ifdef ASSERT
5603   {
5604     ThreadInVMfromUnknown tiv;
5605     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5606     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5607   }
5608 #endif
5609   int oop_index = oop_recorder()->find_index(obj);
5610   return Address((address)obj, oop_Relocation::spec(oop_index));
5611 }
5612 
5613 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5614 void MacroAssembler::tlab_allocate(Register obj,
5615                                    Register var_size_in_bytes,
5616                                    int con_size_in_bytes,
5617                                    Register t1,
5618                                    Register t2,
5619                                    Label& slow_case) {
5620   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5621   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5622 }
5623 
5624 void MacroAssembler::verify_tlab() {
5625 #ifdef ASSERT
5626   if (UseTLAB && VerifyOops) {
5627     Label next, ok;
5628 
5629     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5630 
5631     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5632     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5633     cmp(rscratch2, rscratch1);
5634     br(Assembler::HS, next);
5635     STOP("assert(top >= start)");
5636     should_not_reach_here();
5637 
5638     bind(next);
5639     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5640     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5641     cmp(rscratch2, rscratch1);
5642     br(Assembler::HS, ok);
5643     STOP("assert(top <= end)");
5644     should_not_reach_here();
5645 
5646     bind(ok);
5647     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5648   }
5649 #endif
5650 }
5651 
5652 // Writes to stack successive pages until offset reached to check for
5653 // stack overflow + shadow pages.  This clobbers tmp.
5654 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5655   assert_different_registers(tmp, size, rscratch1);
5656   mov(tmp, sp);
5657   // Bang stack for total size given plus shadow page size.
5658   // Bang one page at a time because large size can bang beyond yellow and
5659   // red zones.
5660   Label loop;
5661   mov(rscratch1, (int)os::vm_page_size());
5662   bind(loop);
5663   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5664   subsw(size, size, rscratch1);
5665   str(size, Address(tmp));
5666   br(Assembler::GT, loop);
5667 
5668   // Bang down shadow pages too.
5669   // At this point, (tmp-0) is the last address touched, so don't
5670   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5671   // was post-decremented.)  Skip this address by starting at i=1, and
5672   // touch a few more pages below.  N.B.  It is important to touch all
5673   // the way down to and including i=StackShadowPages.
5674   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5675     // this could be any sized move but this is can be a debugging crumb
5676     // so the bigger the better.
5677     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5678     str(size, Address(tmp));
5679   }
5680 }
5681 
5682 // Move the address of the polling page into dest.
5683 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5684   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5685 }
5686 
5687 // Read the polling page.  The address of the polling page must
5688 // already be in r.
5689 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5690   address mark;
5691   {
5692     InstructionMark im(this);
5693     code_section()->relocate(inst_mark(), rtype);
5694     ldrw(zr, Address(r, 0));
5695     mark = inst_mark();
5696   }
5697   verify_cross_modify_fence_not_required();
5698   return mark;
5699 }
5700 
5701 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5702   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5703   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5704   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5705   int64_t offset_low = dest_page - low_page;
5706   int64_t offset_high = dest_page - high_page;
5707 
5708   assert(is_valid_AArch64_address(dest.target()), "bad address");
5709   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5710 
5711   InstructionMark im(this);
5712   code_section()->relocate(inst_mark(), dest.rspec());
5713   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5714   // the code cache so that if it is relocated we know it will still reach
5715   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5716     _adrp(reg1, dest.target());
5717   } else {
5718     uint64_t target = (uint64_t)dest.target();
5719     uint64_t adrp_target
5720       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5721 
5722     _adrp(reg1, (address)adrp_target);
5723     movk(reg1, target >> 32, 32);
5724   }
5725   byte_offset = (uint64_t)dest.target() & 0xfff;
5726 }
5727 
5728 void MacroAssembler::load_byte_map_base(Register reg) {
5729 #if INCLUDE_CDS
5730   if (AOTCodeCache::is_on_for_dump()) {
5731     address byte_map_base_adr = AOTRuntimeConstants::card_table_base_address();
5732     lea(reg, ExternalAddress(byte_map_base_adr));
5733     ldr(reg, Address(reg));
5734     return;
5735   }
5736 #endif
5737   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
5738 
5739   // Strictly speaking the card table base isn't an address at all, and it might
5740   // even be negative. It is thus materialised as a constant.
5741   mov(reg, (uint64_t)ctbs->card_table_base_const());
5742 }
5743 
5744 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5745 #if INCLUDE_CDS
5746   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5747   if (AOTCodeCache::is_on_for_dump()) {
5748     // all aotrc field addresses should be registered in the AOTCodeCache address table
5749     lea(reg, ExternalAddress(a));
5750   } else {
5751     mov(reg, (uint64_t)a);
5752   }
5753 #else
5754   ShouldNotReachHere();
5755 #endif
5756 }
5757 
5758 void MacroAssembler::build_frame(int framesize) {
5759   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5760   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5761   protect_return_address();
5762   if (framesize < ((1 << 9) + 2 * wordSize)) {
5763     sub(sp, sp, framesize);
5764     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5765     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5766   } else {
5767     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5768     if (PreserveFramePointer) mov(rfp, sp);
5769     if (framesize < ((1 << 12) + 2 * wordSize))
5770       sub(sp, sp, framesize - 2 * wordSize);
5771     else {
5772       mov(rscratch1, framesize - 2 * wordSize);
5773       sub(sp, sp, rscratch1);
5774     }
5775   }
5776   verify_cross_modify_fence_not_required();
5777 }
5778 
5779 void MacroAssembler::remove_frame(int framesize) {
5780   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5781   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5782   if (framesize < ((1 << 9) + 2 * wordSize)) {
5783     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5784     add(sp, sp, framesize);
5785   } else {
5786     if (framesize < ((1 << 12) + 2 * wordSize))
5787       add(sp, sp, framesize - 2 * wordSize);
5788     else {
5789       mov(rscratch1, framesize - 2 * wordSize);
5790       add(sp, sp, rscratch1);
5791     }
5792     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5793   }
5794   authenticate_return_address();
5795 }
5796 
5797 
5798 // This method counts leading positive bytes (highest bit not set) in provided byte array
5799 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5800     // Simple and most common case of aligned small array which is not at the
5801     // end of memory page is placed here. All other cases are in stub.
5802     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5803     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5804     assert_different_registers(ary1, len, result);
5805 
5806     mov(result, len);
5807     cmpw(len, 0);
5808     br(LE, DONE);
5809     cmpw(len, 4 * wordSize);
5810     br(GE, STUB_LONG); // size > 32 then go to stub
5811 
5812     int shift = 64 - exact_log2(os::vm_page_size());
5813     lsl(rscratch1, ary1, shift);
5814     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5815     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5816     br(CS, STUB); // at the end of page then go to stub
5817     subs(len, len, wordSize);
5818     br(LT, END);
5819 
5820   BIND(LOOP);
5821     ldr(rscratch1, Address(post(ary1, wordSize)));
5822     tst(rscratch1, UPPER_BIT_MASK);
5823     br(NE, SET_RESULT);
5824     subs(len, len, wordSize);
5825     br(GE, LOOP);
5826     cmpw(len, -wordSize);
5827     br(EQ, DONE);
5828 
5829   BIND(END);
5830     ldr(rscratch1, Address(ary1));
5831     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5832     lslv(rscratch1, rscratch1, rscratch2);
5833     tst(rscratch1, UPPER_BIT_MASK);
5834     br(NE, SET_RESULT);
5835     b(DONE);
5836 
5837   BIND(STUB);
5838     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5839     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5840     address tpc1 = trampoline_call(count_pos);
5841     if (tpc1 == nullptr) {
5842       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5843       postcond(pc() == badAddress);
5844       return nullptr;
5845     }
5846     b(DONE);
5847 
5848   BIND(STUB_LONG);
5849     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5850     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5851     address tpc2 = trampoline_call(count_pos_long);
5852     if (tpc2 == nullptr) {
5853       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5854       postcond(pc() == badAddress);
5855       return nullptr;
5856     }
5857     b(DONE);
5858 
5859   BIND(SET_RESULT);
5860 
5861     add(len, len, wordSize);
5862     sub(result, result, len);
5863 
5864   BIND(DONE);
5865   postcond(pc() != badAddress);
5866   return pc();
5867 }
5868 
5869 // Clobbers: rscratch1, rscratch2, rflags
5870 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5871 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5872                                       Register tmp4, Register tmp5, Register result,
5873                                       Register cnt1, int elem_size) {
5874   Label DONE, SAME;
5875   Register tmp1 = rscratch1;
5876   Register tmp2 = rscratch2;
5877   int elem_per_word = wordSize/elem_size;
5878   int log_elem_size = exact_log2(elem_size);
5879   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5880   int length_offset = arrayOopDesc::length_offset_in_bytes();
5881   int base_offset
5882     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5883   // When the length offset is not aligned to 8 bytes,
5884   // then we align it down. This is valid because the new
5885   // offset will always be the klass which is the same
5886   // for type arrays.
5887   int start_offset = align_down(length_offset, BytesPerWord);
5888   int extra_length = base_offset - start_offset;
5889   assert(start_offset == length_offset || start_offset == klass_offset,
5890          "start offset must be 8-byte-aligned or be the klass offset");
5891   assert(base_offset != start_offset, "must include the length field");
5892   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5893   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5894 
5895   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5896   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5897 
5898 #ifndef PRODUCT
5899   {
5900     const char kind = (elem_size == 2) ? 'U' : 'L';
5901     char comment[64];
5902     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5903     BLOCK_COMMENT(comment);
5904   }
5905 #endif
5906 
5907   // if (a1 == a2)
5908   //     return true;
5909   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5910   br(EQ, SAME);
5911 
5912   if (UseSimpleArrayEquals) {
5913     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5914     // if (a1 == nullptr || a2 == nullptr)
5915     //     return false;
5916     // a1 & a2 == 0 means (some-pointer is null) or
5917     // (very-rare-or-even-probably-impossible-pointer-values)
5918     // so, we can save one branch in most cases
5919     tst(a1, a2);
5920     mov(result, false);
5921     br(EQ, A_MIGHT_BE_NULL);
5922     // if (a1.length != a2.length)
5923     //      return false;
5924     bind(A_IS_NOT_NULL);
5925     ldrw(cnt1, Address(a1, length_offset));
5926     ldrw(tmp5, Address(a2, length_offset));
5927     cmp(cnt1, tmp5);
5928     br(NE, DONE); // If lengths differ, return false
5929     // Increase loop counter by diff between base- and actual start-offset.
5930     addw(cnt1, cnt1, extra_length);
5931     lea(a1, Address(a1, start_offset));
5932     lea(a2, Address(a2, start_offset));
5933     // Check for short strings, i.e. smaller than wordSize.
5934     subs(cnt1, cnt1, elem_per_word);
5935     br(Assembler::LT, SHORT);
5936     // Main 8 byte comparison loop.
5937     bind(NEXT_WORD); {
5938       ldr(tmp1, Address(post(a1, wordSize)));
5939       ldr(tmp2, Address(post(a2, wordSize)));
5940       subs(cnt1, cnt1, elem_per_word);
5941       eor(tmp5, tmp1, tmp2);
5942       cbnz(tmp5, DONE);
5943     } br(GT, NEXT_WORD);
5944     // Last longword.  In the case where length == 4 we compare the
5945     // same longword twice, but that's still faster than another
5946     // conditional branch.
5947     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5948     // length == 4.
5949     if (log_elem_size > 0)
5950       lsl(cnt1, cnt1, log_elem_size);
5951     ldr(tmp3, Address(a1, cnt1));
5952     ldr(tmp4, Address(a2, cnt1));
5953     eor(tmp5, tmp3, tmp4);
5954     cbnz(tmp5, DONE);
5955     b(SAME);
5956     bind(A_MIGHT_BE_NULL);
5957     // in case both a1 and a2 are not-null, proceed with loads
5958     cbz(a1, DONE);
5959     cbz(a2, DONE);
5960     b(A_IS_NOT_NULL);
5961     bind(SHORT);
5962 
5963     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
5964     {
5965       ldrw(tmp1, Address(post(a1, 4)));
5966       ldrw(tmp2, Address(post(a2, 4)));
5967       eorw(tmp5, tmp1, tmp2);
5968       cbnzw(tmp5, DONE);
5969     }
5970     bind(TAIL03);
5971     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
5972     {
5973       ldrh(tmp3, Address(post(a1, 2)));
5974       ldrh(tmp4, Address(post(a2, 2)));
5975       eorw(tmp5, tmp3, tmp4);
5976       cbnzw(tmp5, DONE);
5977     }
5978     bind(TAIL01);
5979     if (elem_size == 1) { // Only needed when comparing byte arrays.
5980       tbz(cnt1, 0, SAME); // 0-1 bytes left.
5981       {
5982         ldrb(tmp1, a1);
5983         ldrb(tmp2, a2);
5984         eorw(tmp5, tmp1, tmp2);
5985         cbnzw(tmp5, DONE);
5986       }
5987     }
5988   } else {
5989     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
5990         CSET_EQ, LAST_CHECK;
5991     mov(result, false);
5992     cbz(a1, DONE);
5993     ldrw(cnt1, Address(a1, length_offset));
5994     cbz(a2, DONE);
5995     ldrw(tmp5, Address(a2, length_offset));
5996     cmp(cnt1, tmp5);
5997     br(NE, DONE); // If lengths differ, return false
5998     // Increase loop counter by diff between base- and actual start-offset.
5999     addw(cnt1, cnt1, extra_length);
6000 
6001     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6002     // faster to perform another branch before comparing a1 and a2
6003     cmp(cnt1, (u1)elem_per_word);
6004     br(LE, SHORT); // short or same
6005     ldr(tmp3, Address(pre(a1, start_offset)));
6006     subs(zr, cnt1, stubBytesThreshold);
6007     br(GE, STUB);
6008     ldr(tmp4, Address(pre(a2, start_offset)));
6009     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6010 
6011     // Main 16 byte comparison loop with 2 exits
6012     bind(NEXT_DWORD); {
6013       ldr(tmp1, Address(pre(a1, wordSize)));
6014       ldr(tmp2, Address(pre(a2, wordSize)));
6015       subs(cnt1, cnt1, 2 * elem_per_word);
6016       br(LE, TAIL);
6017       eor(tmp4, tmp3, tmp4);
6018       cbnz(tmp4, DONE);
6019       ldr(tmp3, Address(pre(a1, wordSize)));
6020       ldr(tmp4, Address(pre(a2, wordSize)));
6021       cmp(cnt1, (u1)elem_per_word);
6022       br(LE, TAIL2);
6023       cmp(tmp1, tmp2);
6024     } br(EQ, NEXT_DWORD);
6025     b(DONE);
6026 
6027     bind(TAIL);
6028     eor(tmp4, tmp3, tmp4);
6029     eor(tmp2, tmp1, tmp2);
6030     lslv(tmp2, tmp2, tmp5);
6031     orr(tmp5, tmp4, tmp2);
6032     cmp(tmp5, zr);
6033     b(CSET_EQ);
6034 
6035     bind(TAIL2);
6036     eor(tmp2, tmp1, tmp2);
6037     cbnz(tmp2, DONE);
6038     b(LAST_CHECK);
6039 
6040     bind(STUB);
6041     ldr(tmp4, Address(pre(a2, start_offset)));
6042     if (elem_size == 2) { // convert to byte counter
6043       lsl(cnt1, cnt1, 1);
6044     }
6045     eor(tmp5, tmp3, tmp4);
6046     cbnz(tmp5, DONE);
6047     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6048     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6049     address tpc = trampoline_call(stub);
6050     if (tpc == nullptr) {
6051       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6052       postcond(pc() == badAddress);
6053       return nullptr;
6054     }
6055     b(DONE);
6056 
6057     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6058     // so, if a2 == null => return false(0), else return true, so we can return a2
6059     mov(result, a2);
6060     b(DONE);
6061     bind(SHORT);
6062     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6063     ldr(tmp3, Address(a1, start_offset));
6064     ldr(tmp4, Address(a2, start_offset));
6065     bind(LAST_CHECK);
6066     eor(tmp4, tmp3, tmp4);
6067     lslv(tmp5, tmp4, tmp5);
6068     cmp(tmp5, zr);
6069     bind(CSET_EQ);
6070     cset(result, EQ);
6071     b(DONE);
6072   }
6073 
6074   bind(SAME);
6075   mov(result, true);
6076   // That's it.
6077   bind(DONE);
6078 
6079   BLOCK_COMMENT("} array_equals");
6080   postcond(pc() != badAddress);
6081   return pc();
6082 }
6083 
6084 // Compare Strings
6085 
6086 // For Strings we're passed the address of the first characters in a1
6087 // and a2 and the length in cnt1.
6088 // There are two implementations.  For arrays >= 8 bytes, all
6089 // comparisons (including the final one, which may overlap) are
6090 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6091 // halfword, then a short, and then a byte.
6092 
6093 void MacroAssembler::string_equals(Register a1, Register a2,
6094                                    Register result, Register cnt1)
6095 {
6096   Label SAME, DONE, SHORT, NEXT_WORD;
6097   Register tmp1 = rscratch1;
6098   Register tmp2 = rscratch2;
6099 
6100   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6101 
6102 #ifndef PRODUCT
6103   {
6104     char comment[64];
6105     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6106     BLOCK_COMMENT(comment);
6107   }
6108 #endif
6109 
6110   mov(result, false);
6111 
6112   // Check for short strings, i.e. smaller than wordSize.
6113   subs(cnt1, cnt1, wordSize);
6114   br(Assembler::LT, SHORT);
6115   // Main 8 byte comparison loop.
6116   bind(NEXT_WORD); {
6117     ldr(tmp1, Address(post(a1, wordSize)));
6118     ldr(tmp2, Address(post(a2, wordSize)));
6119     subs(cnt1, cnt1, wordSize);
6120     eor(tmp1, tmp1, tmp2);
6121     cbnz(tmp1, DONE);
6122   } br(GT, NEXT_WORD);
6123   // Last longword.  In the case where length == 4 we compare the
6124   // same longword twice, but that's still faster than another
6125   // conditional branch.
6126   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6127   // length == 4.
6128   ldr(tmp1, Address(a1, cnt1));
6129   ldr(tmp2, Address(a2, cnt1));
6130   eor(tmp2, tmp1, tmp2);
6131   cbnz(tmp2, DONE);
6132   b(SAME);
6133 
6134   bind(SHORT);
6135   Label TAIL03, TAIL01;
6136 
6137   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6138   {
6139     ldrw(tmp1, Address(post(a1, 4)));
6140     ldrw(tmp2, Address(post(a2, 4)));
6141     eorw(tmp1, tmp1, tmp2);
6142     cbnzw(tmp1, DONE);
6143   }
6144   bind(TAIL03);
6145   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6146   {
6147     ldrh(tmp1, Address(post(a1, 2)));
6148     ldrh(tmp2, Address(post(a2, 2)));
6149     eorw(tmp1, tmp1, tmp2);
6150     cbnzw(tmp1, DONE);
6151   }
6152   bind(TAIL01);
6153   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6154     {
6155     ldrb(tmp1, a1);
6156     ldrb(tmp2, a2);
6157     eorw(tmp1, tmp1, tmp2);
6158     cbnzw(tmp1, DONE);
6159   }
6160   // Arrays are equal.
6161   bind(SAME);
6162   mov(result, true);
6163 
6164   // That's it.
6165   bind(DONE);
6166   BLOCK_COMMENT("} string_equals");
6167 }
6168 
6169 
6170 // The size of the blocks erased by the zero_blocks stub.  We must
6171 // handle anything smaller than this ourselves in zero_words().
6172 const int MacroAssembler::zero_words_block_size = 8;
6173 
6174 // zero_words() is used by C2 ClearArray patterns and by
6175 // C1_MacroAssembler.  It is as small as possible, handling small word
6176 // counts locally and delegating anything larger to the zero_blocks
6177 // stub.  It is expanded many times in compiled code, so it is
6178 // important to keep it short.
6179 
6180 // ptr:   Address of a buffer to be zeroed.
6181 // cnt:   Count in HeapWords.
6182 //
6183 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6184 address MacroAssembler::zero_words(Register ptr, Register cnt)
6185 {
6186   assert(is_power_of_2(zero_words_block_size), "adjust this");
6187 
6188   BLOCK_COMMENT("zero_words {");
6189   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6190   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6191   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6192 
6193   subs(rscratch1, cnt, zero_words_block_size);
6194   Label around;
6195   br(LO, around);
6196   {
6197     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6198     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6199     // Make sure this is a C2 compilation. C1 allocates space only for
6200     // trampoline stubs generated by Call LIR ops, and in any case it
6201     // makes sense for a C1 compilation task to proceed as quickly as
6202     // possible.
6203     CompileTask* task;
6204     if (StubRoutines::aarch64::complete()
6205         && Thread::current()->is_Compiler_thread()
6206         && (task = ciEnv::current()->task())
6207         && is_c2_compile(task->comp_level())) {
6208       address tpc = trampoline_call(zero_blocks);
6209       if (tpc == nullptr) {
6210         DEBUG_ONLY(reset_labels(around));
6211         return nullptr;
6212       }
6213     } else {
6214       far_call(zero_blocks);
6215     }
6216   }
6217   bind(around);
6218 
6219   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6220   // for us.
6221   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6222     Label l;
6223     tbz(cnt, exact_log2(i), l);
6224     for (int j = 0; j < i; j += 2) {
6225       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6226     }
6227     bind(l);
6228   }
6229   {
6230     Label l;
6231     tbz(cnt, 0, l);
6232     str(zr, Address(ptr));
6233     bind(l);
6234   }
6235 
6236   BLOCK_COMMENT("} zero_words");
6237   return pc();
6238 }
6239 
6240 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6241 // cnt:          Immediate count in HeapWords.
6242 //
6243 // r10, r11, rscratch1, and rscratch2 are clobbered.
6244 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6245 {
6246   assert(wordSize <= BlockZeroingLowLimit,
6247             "increase BlockZeroingLowLimit");
6248   address result = nullptr;
6249   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6250 #ifndef PRODUCT
6251     {
6252       char buf[64];
6253       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6254       BLOCK_COMMENT(buf);
6255     }
6256 #endif
6257     if (cnt >= 16) {
6258       uint64_t loops = cnt/16;
6259       if (loops > 1) {
6260         mov(rscratch2, loops - 1);
6261       }
6262       {
6263         Label loop;
6264         bind(loop);
6265         for (int i = 0; i < 16; i += 2) {
6266           stp(zr, zr, Address(base, i * BytesPerWord));
6267         }
6268         add(base, base, 16 * BytesPerWord);
6269         if (loops > 1) {
6270           subs(rscratch2, rscratch2, 1);
6271           br(GE, loop);
6272         }
6273       }
6274     }
6275     cnt %= 16;
6276     int i = cnt & 1;  // store any odd word to start
6277     if (i) str(zr, Address(base));
6278     for (; i < (int)cnt; i += 2) {
6279       stp(zr, zr, Address(base, i * wordSize));
6280     }
6281     BLOCK_COMMENT("} zero_words");
6282     result = pc();
6283   } else {
6284     mov(r10, base); mov(r11, cnt);
6285     result = zero_words(r10, r11);
6286   }
6287   return result;
6288 }
6289 
6290 // Zero blocks of memory by using DC ZVA.
6291 //
6292 // Aligns the base address first sufficiently for DC ZVA, then uses
6293 // DC ZVA repeatedly for every full block.  cnt is the size to be
6294 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6295 // in cnt.
6296 //
6297 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6298 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6299 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6300   Register tmp = rscratch1;
6301   Register tmp2 = rscratch2;
6302   int zva_length = VM_Version::zva_length();
6303   Label initial_table_end, loop_zva;
6304   Label fini;
6305 
6306   // Base must be 16 byte aligned. If not just return and let caller handle it
6307   tst(base, 0x0f);
6308   br(Assembler::NE, fini);
6309   // Align base with ZVA length.
6310   neg(tmp, base);
6311   andr(tmp, tmp, zva_length - 1);
6312 
6313   // tmp: the number of bytes to be filled to align the base with ZVA length.
6314   add(base, base, tmp);
6315   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6316   adr(tmp2, initial_table_end);
6317   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6318   br(tmp2);
6319 
6320   for (int i = -zva_length + 16; i < 0; i += 16)
6321     stp(zr, zr, Address(base, i));
6322   bind(initial_table_end);
6323 
6324   sub(cnt, cnt, zva_length >> 3);
6325   bind(loop_zva);
6326   dc(Assembler::ZVA, base);
6327   subs(cnt, cnt, zva_length >> 3);
6328   add(base, base, zva_length);
6329   br(Assembler::GE, loop_zva);
6330   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6331   bind(fini);
6332 }
6333 
6334 // base:   Address of a buffer to be filled, 8 bytes aligned.
6335 // cnt:    Count in 8-byte unit.
6336 // value:  Value to be filled with.
6337 // base will point to the end of the buffer after filling.
6338 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6339 {
6340 //  Algorithm:
6341 //
6342 //    if (cnt == 0) {
6343 //      return;
6344 //    }
6345 //    if ((p & 8) != 0) {
6346 //      *p++ = v;
6347 //    }
6348 //
6349 //    scratch1 = cnt & 14;
6350 //    cnt -= scratch1;
6351 //    p += scratch1;
6352 //    switch (scratch1 / 2) {
6353 //      do {
6354 //        cnt -= 16;
6355 //          p[-16] = v;
6356 //          p[-15] = v;
6357 //        case 7:
6358 //          p[-14] = v;
6359 //          p[-13] = v;
6360 //        case 6:
6361 //          p[-12] = v;
6362 //          p[-11] = v;
6363 //          // ...
6364 //        case 1:
6365 //          p[-2] = v;
6366 //          p[-1] = v;
6367 //        case 0:
6368 //          p += 16;
6369 //      } while (cnt);
6370 //    }
6371 //    if ((cnt & 1) == 1) {
6372 //      *p++ = v;
6373 //    }
6374 
6375   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6376 
6377   Label fini, skip, entry, loop;
6378   const int unroll = 8; // Number of stp instructions we'll unroll
6379 
6380   cbz(cnt, fini);
6381   tbz(base, 3, skip);
6382   str(value, Address(post(base, 8)));
6383   sub(cnt, cnt, 1);
6384   bind(skip);
6385 
6386   andr(rscratch1, cnt, (unroll-1) * 2);
6387   sub(cnt, cnt, rscratch1);
6388   add(base, base, rscratch1, Assembler::LSL, 3);
6389   adr(rscratch2, entry);
6390   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6391   br(rscratch2);
6392 
6393   bind(loop);
6394   add(base, base, unroll * 16);
6395   for (int i = -unroll; i < 0; i++)
6396     stp(value, value, Address(base, i * 16));
6397   bind(entry);
6398   subs(cnt, cnt, unroll * 2);
6399   br(Assembler::GE, loop);
6400 
6401   tbz(cnt, 0, fini);
6402   str(value, Address(post(base, 8)));
6403   bind(fini);
6404 }
6405 
6406 // Intrinsic for
6407 //
6408 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6409 //   Encodes char[] to byte[] in ISO-8859-1
6410 //
6411 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6412 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6413 //
6414 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6415 //   Encodes char[] to byte[] in ASCII
6416 //
6417 // This version always returns the number of characters copied, and does not
6418 // clobber the 'len' register. A successful copy will complete with the post-
6419 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6420 // post-condition: 0 <= 'res' < 'len'.
6421 //
6422 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6423 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6424 //       beyond the acceptable, even though the footprint would be smaller.
6425 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6426 //       avoid additional bloat.
6427 //
6428 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6429 void MacroAssembler::encode_iso_array(Register src, Register dst,
6430                                       Register len, Register res, bool ascii,
6431                                       FloatRegister vtmp0, FloatRegister vtmp1,
6432                                       FloatRegister vtmp2, FloatRegister vtmp3,
6433                                       FloatRegister vtmp4, FloatRegister vtmp5)
6434 {
6435   Register cnt = res;
6436   Register max = rscratch1;
6437   Register chk = rscratch2;
6438 
6439   prfm(Address(src), PLDL1STRM);
6440   movw(cnt, len);
6441 
6442 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6443 
6444   Label LOOP_32, DONE_32, FAIL_32;
6445 
6446   BIND(LOOP_32);
6447   {
6448     cmpw(cnt, 32);
6449     br(LT, DONE_32);
6450     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6451     // Extract lower bytes.
6452     FloatRegister vlo0 = vtmp4;
6453     FloatRegister vlo1 = vtmp5;
6454     uzp1(vlo0, T16B, vtmp0, vtmp1);
6455     uzp1(vlo1, T16B, vtmp2, vtmp3);
6456     // Merge bits...
6457     orr(vtmp0, T16B, vtmp0, vtmp1);
6458     orr(vtmp2, T16B, vtmp2, vtmp3);
6459     // Extract merged upper bytes.
6460     FloatRegister vhix = vtmp0;
6461     uzp2(vhix, T16B, vtmp0, vtmp2);
6462     // ISO-check on hi-parts (all zero).
6463     //                          ASCII-check on lo-parts (no sign).
6464     FloatRegister vlox = vtmp1; // Merge lower bytes.
6465                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6466     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6467     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6468     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6469                                 ASCII(orr(chk, chk, max));
6470     cbnz(chk, FAIL_32);
6471     subw(cnt, cnt, 32);
6472     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6473     b(LOOP_32);
6474   }
6475   BIND(FAIL_32);
6476   sub(src, src, 64);
6477   BIND(DONE_32);
6478 
6479   Label LOOP_8, SKIP_8;
6480 
6481   BIND(LOOP_8);
6482   {
6483     cmpw(cnt, 8);
6484     br(LT, SKIP_8);
6485     FloatRegister vhi = vtmp0;
6486     FloatRegister vlo = vtmp1;
6487     ld1(vtmp3, T8H, src);
6488     uzp1(vlo, T16B, vtmp3, vtmp3);
6489     uzp2(vhi, T16B, vtmp3, vtmp3);
6490     // ISO-check on hi-parts (all zero).
6491     //                          ASCII-check on lo-parts (no sign).
6492                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6493     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6494                                 ASCII(umov(max, vtmp2, B, 0));
6495                                 ASCII(orr(chk, chk, max));
6496     cbnz(chk, SKIP_8);
6497 
6498     strd(vlo, Address(post(dst, 8)));
6499     subw(cnt, cnt, 8);
6500     add(src, src, 16);
6501     b(LOOP_8);
6502   }
6503   BIND(SKIP_8);
6504 
6505 #undef ASCII
6506 
6507   Label LOOP, DONE;
6508 
6509   cbz(cnt, DONE);
6510   BIND(LOOP);
6511   {
6512     Register chr = rscratch1;
6513     ldrh(chr, Address(post(src, 2)));
6514     tst(chr, ascii ? 0xff80 : 0xff00);
6515     br(NE, DONE);
6516     strb(chr, Address(post(dst, 1)));
6517     subs(cnt, cnt, 1);
6518     br(GT, LOOP);
6519   }
6520   BIND(DONE);
6521   // Return index where we stopped.
6522   subw(res, len, cnt);
6523 }
6524 
6525 // Inflate byte[] array to char[].
6526 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6527 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6528                                            FloatRegister vtmp1, FloatRegister vtmp2,
6529                                            FloatRegister vtmp3, Register tmp4) {
6530   Label big, done, after_init, to_stub;
6531 
6532   assert_different_registers(src, dst, len, tmp4, rscratch1);
6533 
6534   fmovd(vtmp1, 0.0);
6535   lsrw(tmp4, len, 3);
6536   bind(after_init);
6537   cbnzw(tmp4, big);
6538   // Short string: less than 8 bytes.
6539   {
6540     Label loop, tiny;
6541 
6542     cmpw(len, 4);
6543     br(LT, tiny);
6544     // Use SIMD to do 4 bytes.
6545     ldrs(vtmp2, post(src, 4));
6546     zip1(vtmp3, T8B, vtmp2, vtmp1);
6547     subw(len, len, 4);
6548     strd(vtmp3, post(dst, 8));
6549 
6550     cbzw(len, done);
6551 
6552     // Do the remaining bytes by steam.
6553     bind(loop);
6554     ldrb(tmp4, post(src, 1));
6555     strh(tmp4, post(dst, 2));
6556     subw(len, len, 1);
6557 
6558     bind(tiny);
6559     cbnz(len, loop);
6560 
6561     b(done);
6562   }
6563 
6564   if (SoftwarePrefetchHintDistance >= 0) {
6565     bind(to_stub);
6566       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6567       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6568       address tpc = trampoline_call(stub);
6569       if (tpc == nullptr) {
6570         DEBUG_ONLY(reset_labels(big, done));
6571         postcond(pc() == badAddress);
6572         return nullptr;
6573       }
6574       b(after_init);
6575   }
6576 
6577   // Unpack the bytes 8 at a time.
6578   bind(big);
6579   {
6580     Label loop, around, loop_last, loop_start;
6581 
6582     if (SoftwarePrefetchHintDistance >= 0) {
6583       const int large_loop_threshold = (64 + 16)/8;
6584       ldrd(vtmp2, post(src, 8));
6585       andw(len, len, 7);
6586       cmp(tmp4, (u1)large_loop_threshold);
6587       br(GE, to_stub);
6588       b(loop_start);
6589 
6590       bind(loop);
6591       ldrd(vtmp2, post(src, 8));
6592       bind(loop_start);
6593       subs(tmp4, tmp4, 1);
6594       br(EQ, loop_last);
6595       zip1(vtmp2, T16B, vtmp2, vtmp1);
6596       ldrd(vtmp3, post(src, 8));
6597       st1(vtmp2, T8H, post(dst, 16));
6598       subs(tmp4, tmp4, 1);
6599       zip1(vtmp3, T16B, vtmp3, vtmp1);
6600       st1(vtmp3, T8H, post(dst, 16));
6601       br(NE, loop);
6602       b(around);
6603       bind(loop_last);
6604       zip1(vtmp2, T16B, vtmp2, vtmp1);
6605       st1(vtmp2, T8H, post(dst, 16));
6606       bind(around);
6607       cbz(len, done);
6608     } else {
6609       andw(len, len, 7);
6610       bind(loop);
6611       ldrd(vtmp2, post(src, 8));
6612       sub(tmp4, tmp4, 1);
6613       zip1(vtmp3, T16B, vtmp2, vtmp1);
6614       st1(vtmp3, T8H, post(dst, 16));
6615       cbnz(tmp4, loop);
6616     }
6617   }
6618 
6619   // Do the tail of up to 8 bytes.
6620   add(src, src, len);
6621   ldrd(vtmp3, Address(src, -8));
6622   add(dst, dst, len, ext::uxtw, 1);
6623   zip1(vtmp3, T16B, vtmp3, vtmp1);
6624   strq(vtmp3, Address(dst, -16));
6625 
6626   bind(done);
6627   postcond(pc() != badAddress);
6628   return pc();
6629 }
6630 
6631 // Compress char[] array to byte[].
6632 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6633 // Return the array length if every element in array can be encoded,
6634 // otherwise, the index of first non-latin1 (> 0xff) character.
6635 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6636                                          Register res,
6637                                          FloatRegister tmp0, FloatRegister tmp1,
6638                                          FloatRegister tmp2, FloatRegister tmp3,
6639                                          FloatRegister tmp4, FloatRegister tmp5) {
6640   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6641 }
6642 
6643 // java.math.round(double a)
6644 // Returns the closest long to the argument, with ties rounding to
6645 // positive infinity.  This requires some fiddling for corner
6646 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6647 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6648                                        FloatRegister ftmp) {
6649   Label DONE;
6650   BLOCK_COMMENT("java_round_double: { ");
6651   fmovd(rscratch1, src);
6652   // Use RoundToNearestTiesAway unless src small and -ve.
6653   fcvtasd(dst, src);
6654   // Test if src >= 0 || abs(src) >= 0x1.0p52
6655   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6656   mov(rscratch2, julong_cast(0x1.0p52));
6657   cmp(rscratch1, rscratch2);
6658   br(HS, DONE); {
6659     // src < 0 && abs(src) < 0x1.0p52
6660     // src may have a fractional part, so add 0.5
6661     fmovd(ftmp, 0.5);
6662     faddd(ftmp, src, ftmp);
6663     // Convert double to jlong, use RoundTowardsNegative
6664     fcvtmsd(dst, ftmp);
6665   }
6666   bind(DONE);
6667   BLOCK_COMMENT("} java_round_double");
6668 }
6669 
6670 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6671                                       FloatRegister ftmp) {
6672   Label DONE;
6673   BLOCK_COMMENT("java_round_float: { ");
6674   fmovs(rscratch1, src);
6675   // Use RoundToNearestTiesAway unless src small and -ve.
6676   fcvtassw(dst, src);
6677   // Test if src >= 0 || abs(src) >= 0x1.0p23
6678   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6679   mov(rscratch2, jint_cast(0x1.0p23f));
6680   cmp(rscratch1, rscratch2);
6681   br(HS, DONE); {
6682     // src < 0 && |src| < 0x1.0p23
6683     // src may have a fractional part, so add 0.5
6684     fmovs(ftmp, 0.5f);
6685     fadds(ftmp, src, ftmp);
6686     // Convert float to jint, use RoundTowardsNegative
6687     fcvtmssw(dst, ftmp);
6688   }
6689   bind(DONE);
6690   BLOCK_COMMENT("} java_round_float");
6691 }
6692 
6693 // get_thread() can be called anywhere inside generated code so we
6694 // need to save whatever non-callee save context might get clobbered
6695 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6696 // the call setup code.
6697 //
6698 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6699 // On other systems, the helper is a usual C function.
6700 //
6701 void MacroAssembler::get_thread(Register dst) {
6702   RegSet saved_regs =
6703     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6704     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6705 
6706   protect_return_address();
6707   push(saved_regs, sp);
6708 
6709   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6710   blr(lr);
6711   if (dst != c_rarg0) {
6712     mov(dst, c_rarg0);
6713   }
6714 
6715   pop(saved_regs, sp);
6716   authenticate_return_address();
6717 }
6718 
6719 void MacroAssembler::cache_wb(Address line) {
6720   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6721   assert(line.index() == noreg, "index should be noreg");
6722   assert(line.offset() == 0, "offset should be 0");
6723   // would like to assert this
6724   // assert(line._ext.shift == 0, "shift should be zero");
6725   if (VM_Version::supports_dcpop()) {
6726     // writeback using clear virtual address to point of persistence
6727     dc(Assembler::CVAP, line.base());
6728   } else {
6729     // no need to generate anything as Unsafe.writebackMemory should
6730     // never invoke this stub
6731   }
6732 }
6733 
6734 void MacroAssembler::cache_wbsync(bool is_pre) {
6735   // we only need a barrier post sync
6736   if (!is_pre) {
6737     membar(Assembler::AnyAny);
6738   }
6739 }
6740 
6741 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6742   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6743     return;
6744   }
6745   // Make sure that native code does not change SVE vector length.
6746   Label verify_ok;
6747   movw(tmp, zr);
6748   sve_inc(tmp, B);
6749   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6750   br(EQ, verify_ok);
6751   stop("Error: SVE vector length has changed since jvm startup");
6752   bind(verify_ok);
6753 }
6754 
6755 void MacroAssembler::verify_ptrue() {
6756   Label verify_ok;
6757   if (!UseSVE) {
6758     return;
6759   }
6760   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6761   sve_dec(rscratch1, B);
6762   cbz(rscratch1, verify_ok);
6763   stop("Error: the preserved predicate register (p7) elements are not all true");
6764   bind(verify_ok);
6765 }
6766 
6767 void MacroAssembler::safepoint_isb() {
6768   isb();
6769 #ifndef PRODUCT
6770   if (VerifyCrossModifyFence) {
6771     // Clear the thread state.
6772     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6773   }
6774 #endif
6775 }
6776 
6777 #ifndef PRODUCT
6778 void MacroAssembler::verify_cross_modify_fence_not_required() {
6779   if (VerifyCrossModifyFence) {
6780     // Check if thread needs a cross modify fence.
6781     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6782     Label fence_not_required;
6783     cbz(rscratch1, fence_not_required);
6784     // If it does then fail.
6785     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6786     mov(c_rarg0, rthread);
6787     blr(rscratch1);
6788     bind(fence_not_required);
6789   }
6790 }
6791 #endif
6792 
6793 void MacroAssembler::spin_wait() {
6794   block_comment("spin_wait {");
6795   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6796     switch (VM_Version::spin_wait_desc().inst()) {
6797       case SpinWait::NOP:
6798         nop();
6799         break;
6800       case SpinWait::ISB:
6801         isb();
6802         break;
6803       case SpinWait::YIELD:
6804         yield();
6805         break;
6806       case SpinWait::SB:
6807         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6808         sb();
6809         break;
6810       case SpinWait::WFET:
6811         spin_wait_wfet(VM_Version::spin_wait_desc().delay());
6812         break;
6813       default:
6814         ShouldNotReachHere();
6815     }
6816   }
6817   block_comment("}");
6818 }
6819 
6820 void MacroAssembler::spin_wait_wfet(int delay_ns) {
6821   // The sequence assumes CNTFRQ_EL0 is fixed to 1GHz. The assumption is valid
6822   // starting from Armv8.6, according to the "D12.1.2 The system counter" of the
6823   // Arm Architecture Reference Manual for A-profile architecture version M.a.a.
6824   // This is sufficient because FEAT_WFXT is introduced from Armv8.6.
6825   Register target = rscratch1;
6826   Register current = rscratch2;
6827   get_cntvctss_el0(current);
6828   add(target, current, delay_ns);
6829 
6830   Label L_wait_loop;
6831   bind(L_wait_loop);
6832 
6833   wfet(target);
6834   get_cntvctss_el0(current);
6835 
6836   cmp(current, target);
6837   br(LT, L_wait_loop);
6838 
6839   sb();
6840 }
6841 
6842 // Stack frame creation/removal
6843 
6844 void MacroAssembler::enter(bool strip_ret_addr) {
6845   if (strip_ret_addr) {
6846     // Addresses can only be signed once. If there are multiple nested frames being created
6847     // in the same function, then the return address needs stripping first.
6848     strip_return_address();
6849   }
6850   protect_return_address();
6851   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6852   mov(rfp, sp);
6853 }
6854 
6855 void MacroAssembler::leave() {
6856   mov(sp, rfp);
6857   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6858   authenticate_return_address();
6859 }
6860 
6861 // ROP Protection
6862 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6863 // destroying stack frames or whenever directly loading/storing the LR to memory.
6864 // If ROP protection is not set then these functions are no-ops.
6865 // For more details on PAC see pauth_aarch64.hpp.
6866 
6867 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6868 // Uses value zero as the modifier.
6869 //
6870 void MacroAssembler::protect_return_address() {
6871   if (VM_Version::use_rop_protection()) {
6872     check_return_address();
6873     paciaz();
6874   }
6875 }
6876 
6877 // Sign the return value in the given register. Use before updating the LR in the existing stack
6878 // frame for the current function.
6879 // Uses value zero as the modifier.
6880 //
6881 void MacroAssembler::protect_return_address(Register return_reg) {
6882   if (VM_Version::use_rop_protection()) {
6883     check_return_address(return_reg);
6884     paciza(return_reg);
6885   }
6886 }
6887 
6888 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6889 // Uses value zero as the modifier.
6890 //
6891 void MacroAssembler::authenticate_return_address() {
6892   if (VM_Version::use_rop_protection()) {
6893     autiaz();
6894     check_return_address();
6895   }
6896 }
6897 
6898 // Authenticate the return value in the given register. Use before updating the LR in the existing
6899 // stack frame for the current function.
6900 // Uses value zero as the modifier.
6901 //
6902 void MacroAssembler::authenticate_return_address(Register return_reg) {
6903   if (VM_Version::use_rop_protection()) {
6904     autiza(return_reg);
6905     check_return_address(return_reg);
6906   }
6907 }
6908 
6909 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6910 // there is no guaranteed way of authenticating the LR.
6911 //
6912 void MacroAssembler::strip_return_address() {
6913   if (VM_Version::use_rop_protection()) {
6914     xpaclri();
6915   }
6916 }
6917 
6918 #ifndef PRODUCT
6919 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6920 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6921 // it is difficult to debug back to the callee function.
6922 // This function simply loads from the address in the given register.
6923 // Use directly after authentication to catch authentication failures.
6924 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6925 //
6926 void MacroAssembler::check_return_address(Register return_reg) {
6927   if (VM_Version::use_rop_protection()) {
6928     ldr(zr, Address(return_reg));
6929   }
6930 }
6931 #endif
6932 
6933 // The java_calling_convention describes stack locations as ideal slots on
6934 // a frame with no abi restrictions. Since we must observe abi restrictions
6935 // (like the placement of the register window) the slots must be biased by
6936 // the following value.
6937 static int reg2offset_in(VMReg r) {
6938   // Account for saved rfp and lr
6939   // This should really be in_preserve_stack_slots
6940   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6941 }
6942 
6943 static int reg2offset_out(VMReg r) {
6944   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6945 }
6946 
6947 // On 64bit we will store integer like items to the stack as
6948 // 64bits items (AArch64 ABI) even though java would only store
6949 // 32bits for a parameter. On 32bit it will simply be 32bits
6950 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6951 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6952   if (src.first()->is_stack()) {
6953     if (dst.first()->is_stack()) {
6954       // stack to stack
6955       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6956       str(tmp, Address(sp, reg2offset_out(dst.first())));
6957     } else {
6958       // stack to reg
6959       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6960     }
6961   } else if (dst.first()->is_stack()) {
6962     // reg to stack
6963     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6964   } else {
6965     if (dst.first() != src.first()) {
6966       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6967     }
6968   }
6969 }
6970 
6971 // An oop arg. Must pass a handle not the oop itself
6972 void MacroAssembler::object_move(
6973                         OopMap* map,
6974                         int oop_handle_offset,
6975                         int framesize_in_slots,
6976                         VMRegPair src,
6977                         VMRegPair dst,
6978                         bool is_receiver,
6979                         int* receiver_offset) {
6980 
6981   // must pass a handle. First figure out the location we use as a handle
6982 
6983   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
6984 
6985   // See if oop is null if it is we need no handle
6986 
6987   if (src.first()->is_stack()) {
6988 
6989     // Oop is already on the stack as an argument
6990     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
6991     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
6992     if (is_receiver) {
6993       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
6994     }
6995 
6996     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
6997     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
6998     // conditionally move a null
6999     cmp(rscratch1, zr);
7000     csel(rHandle, zr, rHandle, Assembler::EQ);
7001   } else {
7002 
7003     // Oop is in an a register we must store it to the space we reserve
7004     // on the stack for oop_handles and pass a handle if oop is non-null
7005 
7006     const Register rOop = src.first()->as_Register();
7007     int oop_slot;
7008     if (rOop == j_rarg0)
7009       oop_slot = 0;
7010     else if (rOop == j_rarg1)
7011       oop_slot = 1;
7012     else if (rOop == j_rarg2)
7013       oop_slot = 2;
7014     else if (rOop == j_rarg3)
7015       oop_slot = 3;
7016     else if (rOop == j_rarg4)
7017       oop_slot = 4;
7018     else if (rOop == j_rarg5)
7019       oop_slot = 5;
7020     else if (rOop == j_rarg6)
7021       oop_slot = 6;
7022     else {
7023       assert(rOop == j_rarg7, "wrong register");
7024       oop_slot = 7;
7025     }
7026 
7027     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7028     int offset = oop_slot*VMRegImpl::stack_slot_size;
7029 
7030     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7031     // Store oop in handle area, may be null
7032     str(rOop, Address(sp, offset));
7033     if (is_receiver) {
7034       *receiver_offset = offset;
7035     }
7036 
7037     cmp(rOop, zr);
7038     lea(rHandle, Address(sp, offset));
7039     // conditionally move a null
7040     csel(rHandle, zr, rHandle, Assembler::EQ);
7041   }
7042 
7043   // If arg is on the stack then place it otherwise it is already in correct reg.
7044   if (dst.first()->is_stack()) {
7045     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7046   }
7047 }
7048 
7049 // A float arg may have to do float reg int reg conversion
7050 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7051  if (src.first()->is_stack()) {
7052     if (dst.first()->is_stack()) {
7053       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7054       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7055     } else {
7056       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7057     }
7058   } else if (src.first() != dst.first()) {
7059     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7060       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7061     else
7062       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7063   }
7064 }
7065 
7066 // A long move
7067 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7068   if (src.first()->is_stack()) {
7069     if (dst.first()->is_stack()) {
7070       // stack to stack
7071       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7072       str(tmp, Address(sp, reg2offset_out(dst.first())));
7073     } else {
7074       // stack to reg
7075       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7076     }
7077   } else if (dst.first()->is_stack()) {
7078     // reg to stack
7079     // Do we really have to sign extend???
7080     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7081     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7082   } else {
7083     if (dst.first() != src.first()) {
7084       mov(dst.first()->as_Register(), src.first()->as_Register());
7085     }
7086   }
7087 }
7088 
7089 
7090 // A double move
7091 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7092  if (src.first()->is_stack()) {
7093     if (dst.first()->is_stack()) {
7094       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7095       str(tmp, Address(sp, reg2offset_out(dst.first())));
7096     } else {
7097       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7098     }
7099   } else if (src.first() != dst.first()) {
7100     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7101       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7102     else
7103       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7104   }
7105 }
7106 
7107 // Implements fast-locking.
7108 //
7109 //  - obj: the object to be locked
7110 //  - t1, t2, t3: temporary registers, will be destroyed
7111 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7112 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7113   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7114 
7115   Label push;
7116   const Register top = t1;
7117   const Register mark = t2;
7118   const Register t = t3;
7119 
7120   // Preload the markWord. It is important that this is the first
7121   // instruction emitted as it is part of C1's null check semantics.
7122   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7123 
7124   if (UseObjectMonitorTable) {
7125     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7126     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7127   }
7128 
7129   if (DiagnoseSyncOnValueBasedClasses != 0) {
7130     load_klass(t1, obj);
7131     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7132     tst(t1, KlassFlags::_misc_is_value_based_class);
7133     br(Assembler::NE, slow);
7134   }
7135 
7136   // Check if the lock-stack is full.
7137   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7138   cmpw(top, (unsigned)LockStack::end_offset());
7139   br(Assembler::GE, slow);
7140 
7141   // Check for recursion.
7142   subw(t, top, oopSize);
7143   ldr(t, Address(rthread, t));
7144   cmp(obj, t);
7145   br(Assembler::EQ, push);
7146 
7147   // Check header for monitor (0b10).
7148   tst(mark, markWord::monitor_value);
7149   br(Assembler::NE, slow);
7150 
7151   // Try to lock. Transition lock bits 0b01 => 0b00
7152   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7153   orr(mark, mark, markWord::unlocked_value);
7154   eor(t, mark, markWord::unlocked_value);
7155   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7156           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7157   br(Assembler::NE, slow);
7158 
7159   bind(push);
7160   // After successful lock, push object on lock-stack.
7161   str(obj, Address(rthread, top));
7162   addw(top, top, oopSize);
7163   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7164 }
7165 
7166 // Implements fast-unlocking.
7167 //
7168 // - obj: the object to be unlocked
7169 // - t1, t2, t3: temporary registers
7170 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7171 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7172   // cmpxchg clobbers rscratch1.
7173   assert_different_registers(obj, t1, t2, t3, rscratch1);
7174 
7175 #ifdef ASSERT
7176   {
7177     // Check for lock-stack underflow.
7178     Label stack_ok;
7179     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7180     cmpw(t1, (unsigned)LockStack::start_offset());
7181     br(Assembler::GE, stack_ok);
7182     STOP("Lock-stack underflow");
7183     bind(stack_ok);
7184   }
7185 #endif
7186 
7187   Label unlocked, push_and_slow;
7188   const Register top = t1;
7189   const Register mark = t2;
7190   const Register t = t3;
7191 
7192   // Check if obj is top of lock-stack.
7193   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7194   subw(top, top, oopSize);
7195   ldr(t, Address(rthread, top));
7196   cmp(obj, t);
7197   br(Assembler::NE, slow);
7198 
7199   // Pop lock-stack.
7200   DEBUG_ONLY(str(zr, Address(rthread, top));)
7201   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7202 
7203   // Check if recursive.
7204   subw(t, top, oopSize);
7205   ldr(t, Address(rthread, t));
7206   cmp(obj, t);
7207   br(Assembler::EQ, unlocked);
7208 
7209   // Not recursive. Check header for monitor (0b10).
7210   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7211   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7212 
7213 #ifdef ASSERT
7214   // Check header not unlocked (0b01).
7215   Label not_unlocked;
7216   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7217   stop("fast_unlock already unlocked");
7218   bind(not_unlocked);
7219 #endif
7220 
7221   // Try to unlock. Transition lock bits 0b00 => 0b01
7222   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7223   orr(t, mark, markWord::unlocked_value);
7224   cmpxchg(obj, mark, t, Assembler::xword,
7225           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7226   br(Assembler::EQ, unlocked);
7227 
7228   bind(push_and_slow);
7229   // Restore lock-stack and handle the unlock in runtime.
7230   DEBUG_ONLY(str(obj, Address(rthread, top));)
7231   addw(top, top, oopSize);
7232   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7233   b(slow);
7234 
7235   bind(unlocked);
7236 }