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     uint32_t insn2 = insn_at(insn_addr, 1);
 410     target = address(target_page);
 411     precond(inner != nullptr);
 412     inner(insn_addr, target);
 413     return 2;
 414   }
 415   static int adrpMem(address insn_addr, address &target) {
 416     uint32_t insn2 = insn_at(insn_addr, 1);
 417     // Load/store register (unsigned immediate)
 418     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 419     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 420     byte_offset <<= size;
 421     target += byte_offset;
 422     return 2;
 423   }
 424   static int adrpAdd(address insn_addr, address &target) {
 425     uint32_t insn2 = insn_at(insn_addr, 1);
 426     // add (immediate)
 427     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 428     target += byte_offset;
 429     return 2;
 430   }
 431   static int adrpMovk(address insn_addr, address &target) {
 432     uint32_t insn2 = insn_at(insn_addr, 1);
 433     uint64_t dest = uint64_t(target);
 434     dest = (dest & 0xffff0000ffffffff) |
 435       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 436     target = address(dest);
 437 
 438     // We know the destination 4k page. Maybe we have a third
 439     // instruction.
 440     uint32_t insn = insn_at(insn_addr, 0);
 441     uint32_t insn3 = insn_at(insn_addr, 2);
 442     ptrdiff_t byte_offset;
 443     if (offset_for(insn, insn3, byte_offset)) {
 444       target += byte_offset;
 445       return 3;
 446     } else {
 447       return 2;
 448     }
 449   }
 450   static int immediate(address insn_addr, address &target) {
 451     uint32_t *insns = (uint32_t *)insn_addr;
 452     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 453     // Move wide constant: movz, movk, movk.  See movptr().
 454     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 455     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 456     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 457                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 458                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 459     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 460     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 461     return 3;
 462   }
 463   static void verify(address insn_addr, address &target) {
 464   }
 465 };
 466 
 467 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 468   address target;
 469   RelocActions<AArch64Decoder>::run(insn_addr, target);
 470   return target;
 471 }
 472 
 473 // Patch any kind of instruction; there may be several instructions.
 474 // Return the total length (in bytes) of the instructions.
 475 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 476   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 477   return RelocActions<Patcher>::run(insn_addr, target);
 478 }
 479 
 480 int MacroAssembler::patch_oop(address insn_addr, address o) {
 481   int instructions;
 482   unsigned insn = *(unsigned*)insn_addr;
 483   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 484 
 485   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 486 
 487   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 488   // narrow OOPs by setting the upper 16 bits in the first
 489   // instruction.
 490   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 491     // Move narrow OOP
 492     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 493     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 494     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 495     instructions = 2;
 496   } else {
 497     // Move wide OOP
 498     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 499     uintptr_t dest = (uintptr_t)o;
 500     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 501     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 502     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 503     instructions = 3;
 504   }
 505   return instructions * NativeInstruction::instruction_size;
 506 }
 507 
 508 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 509   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 510   // We encode narrow ones by setting the upper 16 bits in the first
 511   // instruction.
 512   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 513   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 514          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 515 
 516   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 517 
 518   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 519   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 520   return 2 * NativeInstruction::instruction_size;
 521 }
 522 
 523 address MacroAssembler::target_addr_for_insn_or_null(address insn_addr) {
 524   if (NativeInstruction::is_ldrw_to_zr(insn_addr)) {
 525     return nullptr;
 526   }
 527   return MacroAssembler::target_addr_for_insn(insn_addr);
 528 }
 529 
 530 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 531   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 532   if (at_return) {
 533     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 534     // we may safely use the sp instead to perform the stack watermark check.
 535     cmp(in_nmethod ? sp : rfp, tmp);
 536     br(Assembler::HI, slow_path);
 537   } else {
 538     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 539   }
 540 }
 541 
 542 void MacroAssembler::rt_call(address dest, Register tmp) {
 543   CodeBlob *cb = CodeCache::find_blob(dest);
 544   if (cb) {
 545     far_call(RuntimeAddress(dest));
 546   } else {
 547     lea(tmp, RuntimeAddress(dest));
 548     blr(tmp);
 549   }
 550 }
 551 
 552 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 553   if (!Continuations::enabled()) return;
 554   Label done;
 555   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 556   cmp(sp, rscratch1);
 557   br(Assembler::LS, done);
 558   mov(rscratch1, sp); // we can't use sp as the source in str
 559   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 560   bind(done);
 561 }
 562 
 563 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 564   if (!Continuations::enabled()) return;
 565   Label done;
 566   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 567   cmp(sp, rscratch1);
 568   br(Assembler::LO, done);
 569   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 570   bind(done);
 571 }
 572 
 573 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 574   // we must set sp to zero to clear frame
 575   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 576 
 577   // must clear fp, so that compiled frames are not confused; it is
 578   // possible that we need it only for debugging
 579   if (clear_fp) {
 580     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 581   }
 582 
 583   // Always clear the pc because it could have been set by make_walkable()
 584   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 585 }
 586 
 587 // Calls to C land
 588 //
 589 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 590 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 591 // has to be reset to 0. This is required to allow proper stack traversal.
 592 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 593                                          Register last_java_fp,
 594                                          Register last_java_pc,
 595                                          Register scratch) {
 596 
 597   if (last_java_pc->is_valid()) {
 598       str(last_java_pc, Address(rthread,
 599                                 JavaThread::frame_anchor_offset()
 600                                 + JavaFrameAnchor::last_Java_pc_offset()));
 601     }
 602 
 603   // determine last_java_sp register
 604   if (last_java_sp == sp) {
 605     mov(scratch, sp);
 606     last_java_sp = scratch;
 607   } else if (!last_java_sp->is_valid()) {
 608     last_java_sp = esp;
 609   }
 610 
 611   // last_java_fp is optional
 612   if (last_java_fp->is_valid()) {
 613     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 614   }
 615 
 616   // We must set sp last.
 617   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 618 }
 619 
 620 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 621                                          Register last_java_fp,
 622                                          address  last_java_pc,
 623                                          Register scratch) {
 624   assert(last_java_pc != nullptr, "must provide a valid PC");
 625 
 626   adr(scratch, last_java_pc);
 627   str(scratch, Address(rthread,
 628                        JavaThread::frame_anchor_offset()
 629                        + JavaFrameAnchor::last_Java_pc_offset()));
 630 
 631   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 632 }
 633 
 634 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 635                                          Register last_java_fp,
 636                                          Label &L,
 637                                          Register scratch) {
 638   if (L.is_bound()) {
 639     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 640   } else {
 641     InstructionMark im(this);
 642     L.add_patch_at(code(), locator());
 643     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 644   }
 645 }
 646 
 647 static inline bool target_needs_far_branch(address addr) {
 648   if (AOTCodeCache::is_on_for_dump()) {
 649     return true;
 650   }
 651   // codecache size <= 128M
 652   if (!MacroAssembler::far_branches()) {
 653     return false;
 654   }
 655   // codecache size > 240M
 656   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 657     return true;
 658   }
 659   // codecache size: 128M..240M
 660   return !CodeCache::is_non_nmethod(addr);
 661 }
 662 
 663 void MacroAssembler::far_call(Address entry, Register tmp) {
 664   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 665   assert(CodeCache::find_blob(entry.target()) != nullptr,
 666          "destination of far call not found in code cache");
 667   assert(entry.rspec().type() == relocInfo::external_word_type
 668          || entry.rspec().type() == relocInfo::runtime_call_type
 669          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 670   if (target_needs_far_branch(entry.target())) {
 671     uint64_t offset;
 672     // We can use ADRP here because we know that the total size of
 673     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 674     adrp(tmp, entry, offset);
 675     add(tmp, tmp, offset);
 676     blr(tmp);
 677   } else {
 678     bl(entry);
 679   }
 680 }
 681 
 682 int MacroAssembler::far_jump(Address entry, Register tmp) {
 683   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 684   assert(CodeCache::find_blob(entry.target()) != nullptr,
 685          "destination of far call not found in code cache");
 686   assert(entry.rspec().type() == relocInfo::external_word_type
 687          || entry.rspec().type() == relocInfo::runtime_call_type
 688          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 689   address start = pc();
 690   if (target_needs_far_branch(entry.target())) {
 691     uint64_t offset;
 692     // We can use ADRP here because we know that the total size of
 693     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 694     adrp(tmp, entry, offset);
 695     add(tmp, tmp, offset);
 696     br(tmp);
 697   } else {
 698     b(entry);
 699   }
 700   return pc() - start;
 701 }
 702 
 703 void MacroAssembler::reserved_stack_check() {
 704     // testing if reserved zone needs to be enabled
 705     Label no_reserved_zone_enabling;
 706 
 707     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 708     cmp(sp, rscratch1);
 709     br(Assembler::LO, no_reserved_zone_enabling);
 710 
 711     enter();   // LR and FP are live.
 712     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 713     mov(c_rarg0, rthread);
 714     blr(rscratch1);
 715     leave();
 716 
 717     // We have already removed our own frame.
 718     // throw_delayed_StackOverflowError will think that it's been
 719     // called by our caller.
 720     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 721     br(rscratch1);
 722     should_not_reach_here();
 723 
 724     bind(no_reserved_zone_enabling);
 725 }
 726 
 727 static void pass_arg0(MacroAssembler* masm, Register arg) {
 728   if (c_rarg0 != arg ) {
 729     masm->mov(c_rarg0, arg);
 730   }
 731 }
 732 
 733 static void pass_arg1(MacroAssembler* masm, Register arg) {
 734   if (c_rarg1 != arg ) {
 735     masm->mov(c_rarg1, arg);
 736   }
 737 }
 738 
 739 static void pass_arg2(MacroAssembler* masm, Register arg) {
 740   if (c_rarg2 != arg ) {
 741     masm->mov(c_rarg2, arg);
 742   }
 743 }
 744 
 745 static void pass_arg3(MacroAssembler* masm, Register arg) {
 746   if (c_rarg3 != arg ) {
 747     masm->mov(c_rarg3, arg);
 748   }
 749 }
 750 
 751 void MacroAssembler::call_VM_base(Register oop_result,
 752                                   Register java_thread,
 753                                   Register last_java_sp,
 754                                   Label*   return_pc,
 755                                   address  entry_point,
 756                                   int      number_of_arguments,
 757                                   bool     check_exceptions) {
 758    // determine java_thread register
 759   if (!java_thread->is_valid()) {
 760     java_thread = rthread;
 761   }
 762 
 763   // determine last_java_sp register
 764   if (!last_java_sp->is_valid()) {
 765     last_java_sp = esp;
 766   }
 767 
 768   // debugging support
 769   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 770   assert(java_thread == rthread, "unexpected register");
 771 #ifdef ASSERT
 772   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 773   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 774 #endif // ASSERT
 775 
 776   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 777   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 778 
 779   // push java thread (becomes first argument of C function)
 780 
 781   mov(c_rarg0, java_thread);
 782 
 783   // set last Java frame before call
 784   assert(last_java_sp != rfp, "can't use rfp");
 785 
 786   Label l;
 787   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 788 
 789   // do the call, remove parameters
 790   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 791 
 792   // lr could be poisoned with PAC signature during throw_pending_exception
 793   // if it was tail-call optimized by compiler, since lr is not callee-saved
 794   // reload it with proper value
 795   adr(lr, l);
 796 
 797   // reset last Java frame
 798   // Only interpreter should have to clear fp
 799   reset_last_Java_frame(true);
 800 
 801    // C++ interp handles this in the interpreter
 802   check_and_handle_popframe(java_thread);
 803   check_and_handle_earlyret(java_thread);
 804 
 805   if (check_exceptions) {
 806     // check for pending exceptions (java_thread is set upon return)
 807     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 808     Label ok;
 809     cbz(rscratch1, ok);
 810     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 811     br(rscratch1);
 812     bind(ok);
 813   }
 814 
 815   // get oop result if there is one and reset the value in the thread
 816   if (oop_result->is_valid()) {
 817     get_vm_result_oop(oop_result, java_thread);
 818   }
 819 }
 820 
 821 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 822   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 823 }
 824 
 825 // Check the entry target is always reachable from any branch.
 826 static bool is_always_within_branch_range(Address entry) {
 827   if (AOTCodeCache::is_on_for_dump()) {
 828     return false;
 829   }
 830   const address target = entry.target();
 831 
 832   if (!CodeCache::contains(target)) {
 833     // We always use trampolines for callees outside CodeCache.
 834     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 835     return false;
 836   }
 837 
 838   if (!MacroAssembler::far_branches()) {
 839     return true;
 840   }
 841 
 842   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 843     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 844     // Non-compiled methods stay forever in CodeCache.
 845     // We check whether the longest possible branch is within the branch range.
 846     assert(CodeCache::find_blob(target) != nullptr &&
 847           !CodeCache::find_blob(target)->is_nmethod(),
 848           "runtime call of compiled method");
 849     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 850     const address left_longest_branch_start = CodeCache::low_bound();
 851     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 852                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 853     return is_reachable;
 854   }
 855 
 856   return false;
 857 }
 858 
 859 // Maybe emit a call via a trampoline. If the code cache is small
 860 // trampolines won't be emitted.
 861 address MacroAssembler::trampoline_call(Address entry) {
 862   assert(entry.rspec().type() == relocInfo::runtime_call_type
 863          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 864          || entry.rspec().type() == relocInfo::static_call_type
 865          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 866 
 867   address target = entry.target();
 868 
 869   if (!is_always_within_branch_range(entry)) {
 870     if (!in_scratch_emit_size()) {
 871       // We don't want to emit a trampoline if C2 is generating dummy
 872       // code during its branch shortening phase.
 873       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 874         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 875         code()->share_trampoline_for(entry.target(), offset());
 876       } else {
 877         address stub = emit_trampoline_stub(offset(), target);
 878         if (stub == nullptr) {
 879           postcond(pc() == badAddress);
 880           return nullptr; // CodeCache is full
 881         }
 882       }
 883     }
 884     target = pc();
 885   }
 886 
 887   address call_pc = pc();
 888   relocate(entry.rspec());
 889   bl(target);
 890 
 891   postcond(pc() != badAddress);
 892   return call_pc;
 893 }
 894 
 895 // Emit a trampoline stub for a call to a target which is too far away.
 896 //
 897 // code sequences:
 898 //
 899 // call-site:
 900 //   branch-and-link to <destination> or <trampoline stub>
 901 //
 902 // Related trampoline stub for this call site in the stub section:
 903 //   load the call target from the constant pool
 904 //   branch (LR still points to the call site above)
 905 
 906 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 907                                              address dest) {
 908   // Max stub size: alignment nop, TrampolineStub.
 909   address stub = start_a_stub(max_trampoline_stub_size());
 910   if (stub == nullptr) {
 911     return nullptr;  // CodeBuffer::expand failed
 912   }
 913 
 914   // Create a trampoline stub relocation which relates this trampoline stub
 915   // with the call instruction at insts_call_instruction_offset in the
 916   // instructions code-section.
 917   align(wordSize);
 918   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 919                                             + insts_call_instruction_offset));
 920   const int stub_start_offset = offset();
 921 
 922   // Now, create the trampoline stub's code:
 923   // - load the call
 924   // - call
 925   Label target;
 926   ldr(rscratch1, target);
 927   br(rscratch1);
 928   bind(target);
 929   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 930          "should be");
 931   emit_int64((int64_t)dest);
 932 
 933   const address stub_start_addr = addr_at(stub_start_offset);
 934 
 935   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 936 
 937   end_a_stub();
 938   return stub_start_addr;
 939 }
 940 
 941 int MacroAssembler::max_trampoline_stub_size() {
 942   // Max stub size: alignment nop, TrampolineStub.
 943   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 944 }
 945 
 946 void MacroAssembler::emit_static_call_stub() {
 947   // CompiledDirectCall::set_to_interpreted knows the
 948   // exact layout of this stub.
 949 
 950   isb();
 951   mov_metadata(rmethod, nullptr);
 952 
 953   // Jump to the entry point of the c2i stub.
 954   if (codestub_branch_needs_far_jump()) {
 955     movptr(rscratch1, 0);
 956     br(rscratch1);
 957   } else {
 958     b(pc());
 959   }
 960 }
 961 
 962 int MacroAssembler::static_call_stub_size() {
 963   if (!codestub_branch_needs_far_jump()) {
 964     // isb; movk; movz; movz; b
 965     return 5 * NativeInstruction::instruction_size;
 966   }
 967   // isb; movk; movz; movz; movk; movz; movz; br
 968   return 8 * NativeInstruction::instruction_size;
 969 }
 970 
 971 void MacroAssembler::c2bool(Register x) {
 972   // implements x == 0 ? 0 : 1
 973   // note: must only look at least-significant byte of x
 974   //       since C-style booleans are stored in one byte
 975   //       only! (was bug)
 976   tst(x, 0xff);
 977   cset(x, Assembler::NE);
 978 }
 979 
 980 address MacroAssembler::ic_call(address entry, jint method_index) {
 981   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 982   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 983   return trampoline_call(Address(entry, rh));
 984 }
 985 
 986 int MacroAssembler::ic_check_size() {
 987   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 988   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 989     return NativeInstruction::instruction_size * (7 + extra_instructions);
 990   } else {
 991     return NativeInstruction::instruction_size * (5 + extra_instructions);
 992   }
 993 }
 994 
 995 int MacroAssembler::ic_check(int end_alignment) {
 996   Register receiver = j_rarg0;
 997   Register data = rscratch2;
 998   Register tmp1 = rscratch1;
 999   Register tmp2 = r10;
1000 
1001   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
1002   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
1003   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1004   // before the inline cache check here, and not after
1005   align(end_alignment, offset() + ic_check_size());
1006 
1007   int uep_offset = offset();
1008 
1009   if (UseCompactObjectHeaders) {
1010     load_narrow_klass_compact(tmp1, receiver);
1011     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1012     cmpw(tmp1, tmp2);
1013   } else if (UseCompressedClassPointers) {
1014     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1015     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1016     cmpw(tmp1, tmp2);
1017   } else {
1018     ldr(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1019     ldr(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1020     cmp(tmp1, tmp2);
1021   }
1022 
1023   Label dont;
1024   br(Assembler::EQ, dont);
1025   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1026   bind(dont);
1027   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1028 
1029   return uep_offset;
1030 }
1031 
1032 // Implementation of call_VM versions
1033 
1034 void MacroAssembler::call_VM(Register oop_result,
1035                              address entry_point,
1036                              bool check_exceptions) {
1037   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1038 }
1039 
1040 void MacroAssembler::call_VM(Register oop_result,
1041                              address entry_point,
1042                              Register arg_1,
1043                              bool check_exceptions) {
1044   pass_arg1(this, arg_1);
1045   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1046 }
1047 
1048 void MacroAssembler::call_VM(Register oop_result,
1049                              address entry_point,
1050                              Register arg_1,
1051                              Register arg_2,
1052                              bool check_exceptions) {
1053   assert_different_registers(arg_1, c_rarg2);
1054   pass_arg2(this, arg_2);
1055   pass_arg1(this, arg_1);
1056   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1057 }
1058 
1059 void MacroAssembler::call_VM(Register oop_result,
1060                              address entry_point,
1061                              Register arg_1,
1062                              Register arg_2,
1063                              Register arg_3,
1064                              bool check_exceptions) {
1065   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1066   assert_different_registers(arg_2, c_rarg3);
1067   pass_arg3(this, arg_3);
1068 
1069   pass_arg2(this, arg_2);
1070 
1071   pass_arg1(this, arg_1);
1072   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1073 }
1074 
1075 void MacroAssembler::call_VM(Register oop_result,
1076                              Register last_java_sp,
1077                              address entry_point,
1078                              int number_of_arguments,
1079                              bool check_exceptions) {
1080   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1081 }
1082 
1083 void MacroAssembler::call_VM(Register oop_result,
1084                              Register last_java_sp,
1085                              address entry_point,
1086                              Register arg_1,
1087                              bool check_exceptions) {
1088   pass_arg1(this, arg_1);
1089   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1090 }
1091 
1092 void MacroAssembler::call_VM(Register oop_result,
1093                              Register last_java_sp,
1094                              address entry_point,
1095                              Register arg_1,
1096                              Register arg_2,
1097                              bool check_exceptions) {
1098 
1099   assert_different_registers(arg_1, c_rarg2);
1100   pass_arg2(this, arg_2);
1101   pass_arg1(this, arg_1);
1102   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1103 }
1104 
1105 void MacroAssembler::call_VM(Register oop_result,
1106                              Register last_java_sp,
1107                              address entry_point,
1108                              Register arg_1,
1109                              Register arg_2,
1110                              Register arg_3,
1111                              bool check_exceptions) {
1112   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1113   assert_different_registers(arg_2, c_rarg3);
1114   pass_arg3(this, arg_3);
1115   pass_arg2(this, arg_2);
1116   pass_arg1(this, arg_1);
1117   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1118 }
1119 
1120 
1121 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1122   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1123   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1124   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1125 }
1126 
1127 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1128   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1129   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1130 }
1131 
1132 void MacroAssembler::align(int modulus) {
1133   align(modulus, offset());
1134 }
1135 
1136 // Ensure that the code at target bytes offset from the current offset() is aligned
1137 // according to modulus.
1138 void MacroAssembler::align(int modulus, int target) {
1139   int delta = target - offset();
1140   while ((offset() + delta) % modulus != 0) nop();
1141 }
1142 
1143 void MacroAssembler::post_call_nop() {
1144   if (!Continuations::enabled()) {
1145     return;
1146   }
1147   InstructionMark im(this);
1148   relocate(post_call_nop_Relocation::spec());
1149   InlineSkippedInstructionsCounter skipCounter(this);
1150   nop();
1151   movk(zr, 0);
1152   movk(zr, 0);
1153 }
1154 
1155 // these are no-ops overridden by InterpreterMacroAssembler
1156 
1157 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1158 
1159 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1160 
1161 // Look up the method for a megamorphic invokeinterface call.
1162 // The target method is determined by <intf_klass, itable_index>.
1163 // The receiver klass is in recv_klass.
1164 // On success, the result will be in method_result, and execution falls through.
1165 // On failure, execution transfers to the given label.
1166 void MacroAssembler::lookup_interface_method(Register recv_klass,
1167                                              Register intf_klass,
1168                                              RegisterOrConstant itable_index,
1169                                              Register method_result,
1170                                              Register scan_temp,
1171                                              Label& L_no_such_interface,
1172                          bool return_method) {
1173   assert_different_registers(recv_klass, intf_klass, scan_temp);
1174   assert_different_registers(method_result, intf_klass, scan_temp);
1175   assert(recv_klass != method_result || !return_method,
1176      "recv_klass can be destroyed when method isn't needed");
1177   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1178          "caller must use same register for non-constant itable index as for method");
1179 
1180   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1181   int vtable_base = in_bytes(Klass::vtable_start_offset());
1182   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1183   int scan_step   = itableOffsetEntry::size() * wordSize;
1184   int vte_size    = vtableEntry::size_in_bytes();
1185   assert(vte_size == wordSize, "else adjust times_vte_scale");
1186 
1187   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1188 
1189   // Could store the aligned, prescaled offset in the klass.
1190   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1191   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1192   add(scan_temp, scan_temp, vtable_base);
1193 
1194   if (return_method) {
1195     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1196     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1197     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1198     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1199     if (itentry_off)
1200       add(recv_klass, recv_klass, itentry_off);
1201   }
1202 
1203   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1204   //   if (scan->interface() == intf) {
1205   //     result = (klass + scan->offset() + itable_index);
1206   //   }
1207   // }
1208   Label search, found_method;
1209 
1210   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1211   cmp(intf_klass, method_result);
1212   br(Assembler::EQ, found_method);
1213   bind(search);
1214   // Check that the previous entry is non-null.  A null entry means that
1215   // the receiver class doesn't implement the interface, and wasn't the
1216   // same as when the caller was compiled.
1217   cbz(method_result, L_no_such_interface);
1218   if (itableOffsetEntry::interface_offset() != 0) {
1219     add(scan_temp, scan_temp, scan_step);
1220     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1221   } else {
1222     ldr(method_result, Address(pre(scan_temp, scan_step)));
1223   }
1224   cmp(intf_klass, method_result);
1225   br(Assembler::NE, search);
1226 
1227   bind(found_method);
1228 
1229   // Got a hit.
1230   if (return_method) {
1231     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1232     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1233   }
1234 }
1235 
1236 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1237 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1238 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1239 // The target method is determined by <holder_klass, itable_index>.
1240 // The receiver klass is in recv_klass.
1241 // On success, the result will be in method_result, and execution falls through.
1242 // On failure, execution transfers to the given label.
1243 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1244                                                   Register holder_klass,
1245                                                   Register resolved_klass,
1246                                                   Register method_result,
1247                                                   Register temp_itbl_klass,
1248                                                   Register scan_temp,
1249                                                   int itable_index,
1250                                                   Label& L_no_such_interface) {
1251   // 'method_result' is only used as output register at the very end of this method.
1252   // Until then we can reuse it as 'holder_offset'.
1253   Register holder_offset = method_result;
1254   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1255 
1256   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1257   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1258   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1259   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1260 
1261   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1262 
1263   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1264   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1265   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1266   // temp_itbl_klass = itable[0]._interface;
1267   int vtblEntrySize = vtableEntry::size_in_bytes();
1268   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1269   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1270   mov(holder_offset, zr);
1271   // scan_temp = &(itable[0]._interface)
1272   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1273 
1274   // Initial checks:
1275   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1276   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1277   //   - if (itable[0] == 0), no such interface
1278   cmp(resolved_klass, holder_klass);
1279   br(Assembler::NE, L_loop_search_resolved_entry);
1280   cmp(holder_klass, temp_itbl_klass);
1281   br(Assembler::EQ, L_holder_found);
1282   cbz(temp_itbl_klass, L_no_such_interface);
1283 
1284   // Loop: Look for holder_klass record in itable
1285   //   do {
1286   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1287   //     if (temp_itbl_klass == holder_klass) {
1288   //       goto L_holder_found; // Found!
1289   //     }
1290   //   } while (temp_itbl_klass != 0);
1291   //   goto L_no_such_interface // Not found.
1292   Label L_search_holder;
1293   bind(L_search_holder);
1294     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1295     cmp(holder_klass, temp_itbl_klass);
1296     br(Assembler::EQ, L_holder_found);
1297     cbnz(temp_itbl_klass, L_search_holder);
1298 
1299   b(L_no_such_interface);
1300 
1301   // Loop: Look for resolved_class record in itable
1302   //   while (true) {
1303   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1304   //     if (temp_itbl_klass == 0) {
1305   //       goto L_no_such_interface;
1306   //     }
1307   //     if (temp_itbl_klass == resolved_klass) {
1308   //        goto L_resolved_found;  // Found!
1309   //     }
1310   //     if (temp_itbl_klass == holder_klass) {
1311   //        holder_offset = scan_temp;
1312   //     }
1313   //   }
1314   //
1315   Label L_loop_search_resolved;
1316   bind(L_loop_search_resolved);
1317     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1318   bind(L_loop_search_resolved_entry);
1319     cbz(temp_itbl_klass, L_no_such_interface);
1320     cmp(resolved_klass, temp_itbl_klass);
1321     br(Assembler::EQ, L_resolved_found);
1322     cmp(holder_klass, temp_itbl_klass);
1323     br(Assembler::NE, L_loop_search_resolved);
1324     mov(holder_offset, scan_temp);
1325     b(L_loop_search_resolved);
1326 
1327   // See if we already have a holder klass. If not, go and scan for it.
1328   bind(L_resolved_found);
1329   cbz(holder_offset, L_search_holder);
1330   mov(scan_temp, holder_offset);
1331 
1332   // Finally, scan_temp contains holder_klass vtable offset
1333   bind(L_holder_found);
1334   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1335   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1336     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1337   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1338 }
1339 
1340 // virtual method calling
1341 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1342                                            RegisterOrConstant vtable_index,
1343                                            Register method_result) {
1344   assert(vtableEntry::size() * wordSize == 8,
1345          "adjust the scaling in the code below");
1346   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1347 
1348   if (vtable_index.is_register()) {
1349     lea(method_result, Address(recv_klass,
1350                                vtable_index.as_register(),
1351                                Address::lsl(LogBytesPerWord)));
1352     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1353   } else {
1354     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1355     ldr(method_result,
1356         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1357   }
1358 }
1359 
1360 void MacroAssembler::check_klass_subtype(Register sub_klass,
1361                            Register super_klass,
1362                            Register temp_reg,
1363                            Label& L_success) {
1364   Label L_failure;
1365   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1366   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1367   bind(L_failure);
1368 }
1369 
1370 
1371 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1372                                                    Register super_klass,
1373                                                    Register temp_reg,
1374                                                    Label* L_success,
1375                                                    Label* L_failure,
1376                                                    Label* L_slow_path,
1377                                                    Register super_check_offset) {
1378   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1379   bool must_load_sco = ! super_check_offset->is_valid();
1380   if (must_load_sco) {
1381     assert(temp_reg != noreg, "supply either a temp or a register offset");
1382   }
1383 
1384   Label L_fallthrough;
1385   int label_nulls = 0;
1386   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1387   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1388   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1389   assert(label_nulls <= 1, "at most one null in the batch");
1390 
1391   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1392   Address super_check_offset_addr(super_klass, sco_offset);
1393 
1394   // Hacked jmp, which may only be used just before L_fallthrough.
1395 #define final_jmp(label)                                                \
1396   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1397   else                            b(label)                /*omit semi*/
1398 
1399   // If the pointers are equal, we are done (e.g., String[] elements).
1400   // This self-check enables sharing of secondary supertype arrays among
1401   // non-primary types such as array-of-interface.  Otherwise, each such
1402   // type would need its own customized SSA.
1403   // We move this check to the front of the fast path because many
1404   // type checks are in fact trivially successful in this manner,
1405   // so we get a nicely predicted branch right at the start of the check.
1406   cmp(sub_klass, super_klass);
1407   br(Assembler::EQ, *L_success);
1408 
1409   // Check the supertype display:
1410   if (must_load_sco) {
1411     ldrw(temp_reg, super_check_offset_addr);
1412     super_check_offset = temp_reg;
1413   }
1414 
1415   Address super_check_addr(sub_klass, super_check_offset);
1416   ldr(rscratch1, super_check_addr);
1417   cmp(super_klass, rscratch1); // load displayed supertype
1418   br(Assembler::EQ, *L_success);
1419 
1420   // This check has worked decisively for primary supers.
1421   // Secondary supers are sought in the super_cache ('super_cache_addr').
1422   // (Secondary supers are interfaces and very deeply nested subtypes.)
1423   // This works in the same check above because of a tricky aliasing
1424   // between the super_cache and the primary super display elements.
1425   // (The 'super_check_addr' can address either, as the case requires.)
1426   // Note that the cache is updated below if it does not help us find
1427   // what we need immediately.
1428   // So if it was a primary super, we can just fail immediately.
1429   // Otherwise, it's the slow path for us (no success at this point).
1430 
1431   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1432   if (L_failure == &L_fallthrough) {
1433     cbz(rscratch1, *L_slow_path);
1434   } else {
1435     cbnz(rscratch1, *L_failure);
1436     final_jmp(*L_slow_path);
1437   }
1438 
1439   bind(L_fallthrough);
1440 
1441 #undef final_jmp
1442 }
1443 
1444 // These two are taken from x86, but they look generally useful
1445 
1446 // scans count pointer sized words at [addr] for occurrence of value,
1447 // generic
1448 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1449                                 Register scratch) {
1450   Label Lloop, Lexit;
1451   cbz(count, Lexit);
1452   bind(Lloop);
1453   ldr(scratch, post(addr, wordSize));
1454   cmp(value, scratch);
1455   br(EQ, Lexit);
1456   sub(count, count, 1);
1457   cbnz(count, Lloop);
1458   bind(Lexit);
1459 }
1460 
1461 // scans count 4 byte words at [addr] for occurrence of value,
1462 // generic
1463 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1464                                 Register scratch) {
1465   Label Lloop, Lexit;
1466   cbz(count, Lexit);
1467   bind(Lloop);
1468   ldrw(scratch, post(addr, wordSize));
1469   cmpw(value, scratch);
1470   br(EQ, Lexit);
1471   sub(count, count, 1);
1472   cbnz(count, Lloop);
1473   bind(Lexit);
1474 }
1475 
1476 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1477                                                           Register super_klass,
1478                                                           Register temp_reg,
1479                                                           Register temp2_reg,
1480                                                           Label* L_success,
1481                                                           Label* L_failure,
1482                                                           bool set_cond_codes) {
1483   // NB! Callers may assume that, when temp2_reg is a valid register,
1484   // this code sets it to a nonzero value.
1485 
1486   assert_different_registers(sub_klass, super_klass, temp_reg);
1487   if (temp2_reg != noreg)
1488     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1489 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1490 
1491   Label L_fallthrough;
1492   int label_nulls = 0;
1493   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1494   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1495   assert(label_nulls <= 1, "at most one null in the batch");
1496 
1497   // a couple of useful fields in sub_klass:
1498   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1499   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1500   Address secondary_supers_addr(sub_klass, ss_offset);
1501   Address super_cache_addr(     sub_klass, sc_offset);
1502 
1503   BLOCK_COMMENT("check_klass_subtype_slow_path");
1504 
1505   // Do a linear scan of the secondary super-klass chain.
1506   // This code is rarely used, so simplicity is a virtue here.
1507   // The repne_scan instruction uses fixed registers, which we must spill.
1508   // Don't worry too much about pre-existing connections with the input regs.
1509 
1510   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1511   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1512 
1513   RegSet pushed_registers;
1514   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1515   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1516 
1517   if (super_klass != r0) {
1518     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1519   }
1520 
1521   push(pushed_registers, sp);
1522 
1523   // Get super_klass value into r0 (even if it was in r5 or r2).
1524   if (super_klass != r0) {
1525     mov(r0, super_klass);
1526   }
1527 
1528 #ifndef PRODUCT
1529   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1530 #endif //PRODUCT
1531 
1532   // We will consult the secondary-super array.
1533   ldr(r5, secondary_supers_addr);
1534   // Load the array length.
1535   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1536   // Skip to start of data.
1537   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1538 
1539   cmp(sp, zr); // Clear Z flag; SP is never zero
1540   // Scan R2 words at [R5] for an occurrence of R0.
1541   // Set NZ/Z based on last compare.
1542   repne_scan(r5, r0, r2, rscratch1);
1543 
1544   // Unspill the temp. registers:
1545   pop(pushed_registers, sp);
1546 
1547   br(Assembler::NE, *L_failure);
1548 
1549   // Success.  Cache the super we found and proceed in triumph.
1550 
1551   if (UseSecondarySupersCache) {
1552     str(super_klass, super_cache_addr);
1553   }
1554 
1555   if (L_success != &L_fallthrough) {
1556     b(*L_success);
1557   }
1558 
1559 #undef IS_A_TEMP
1560 
1561   bind(L_fallthrough);
1562 }
1563 
1564 // If Register r is invalid, remove a new register from
1565 // available_regs, and add new register to regs_to_push.
1566 Register MacroAssembler::allocate_if_noreg(Register r,
1567                                   RegSetIterator<Register> &available_regs,
1568                                   RegSet &regs_to_push) {
1569   if (!r->is_valid()) {
1570     r = *available_regs++;
1571     regs_to_push += r;
1572   }
1573   return r;
1574 }
1575 
1576 // check_klass_subtype_slow_path_table() looks for super_klass in the
1577 // hash table belonging to super_klass, branching to L_success or
1578 // L_failure as appropriate. This is essentially a shim which
1579 // allocates registers as necessary then calls
1580 // lookup_secondary_supers_table() to do the work. Any of the temp
1581 // regs may be noreg, in which case this logic will chooses some
1582 // registers push and pop them from the stack.
1583 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1584                                                          Register super_klass,
1585                                                          Register temp_reg,
1586                                                          Register temp2_reg,
1587                                                          Register temp3_reg,
1588                                                          Register result_reg,
1589                                                          FloatRegister vtemp,
1590                                                          Label* L_success,
1591                                                          Label* L_failure,
1592                                                          bool set_cond_codes) {
1593   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1594 
1595   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1596 
1597   Label L_fallthrough;
1598   int label_nulls = 0;
1599   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1600   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1601   assert(label_nulls <= 1, "at most one null in the batch");
1602 
1603   BLOCK_COMMENT("check_klass_subtype_slow_path");
1604 
1605   RegSetIterator<Register> available_regs
1606     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1607 
1608   RegSet pushed_regs;
1609 
1610   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1611   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1612   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1613   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1614 
1615   push(pushed_regs, sp);
1616 
1617   lookup_secondary_supers_table_var(sub_klass,
1618                                     super_klass,
1619                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1620                                     nullptr);
1621   cmp(result_reg, zr);
1622 
1623   // Unspill the temp. registers:
1624   pop(pushed_regs, sp);
1625 
1626   // NB! Callers may assume that, when set_cond_codes is true, this
1627   // code sets temp2_reg to a nonzero value.
1628   if (set_cond_codes) {
1629     mov(temp2_reg, 1);
1630   }
1631 
1632   br(Assembler::NE, *L_failure);
1633 
1634   if (L_success != &L_fallthrough) {
1635     b(*L_success);
1636   }
1637 
1638   bind(L_fallthrough);
1639 }
1640 
1641 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1642                                                    Register super_klass,
1643                                                    Register temp_reg,
1644                                                    Register temp2_reg,
1645                                                    Label* L_success,
1646                                                    Label* L_failure,
1647                                                    bool set_cond_codes) {
1648   if (UseSecondarySupersTable) {
1649     check_klass_subtype_slow_path_table
1650       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1651        /*vtemp*/fnoreg,
1652        L_success, L_failure, set_cond_codes);
1653   } else {
1654     check_klass_subtype_slow_path_linear
1655       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1656   }
1657 }
1658 
1659 
1660 // Ensure that the inline code and the stub are using the same registers.
1661 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1662 do {                                                               \
1663   assert(r_super_klass  == r0                                   && \
1664          r_array_base   == r1                                   && \
1665          r_array_length == r2                                   && \
1666          (r_array_index == r3        || r_array_index == noreg) && \
1667          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1668          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1669          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1670 } while(0)
1671 
1672 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1673                                                          Register r_super_klass,
1674                                                          Register temp1,
1675                                                          Register temp2,
1676                                                          Register temp3,
1677                                                          FloatRegister vtemp,
1678                                                          Register result,
1679                                                          u1 super_klass_slot,
1680                                                          bool stub_is_near) {
1681   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1682 
1683   Label L_fallthrough;
1684 
1685   BLOCK_COMMENT("lookup_secondary_supers_table {");
1686 
1687   const Register
1688     r_array_base   = temp1, // r1
1689     r_array_length = temp2, // r2
1690     r_array_index  = temp3, // r3
1691     r_bitmap       = rscratch2;
1692 
1693   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1694 
1695   u1 bit = super_klass_slot;
1696 
1697   // Make sure that result is nonzero if the TBZ below misses.
1698   mov(result, 1);
1699 
1700   // We're going to need the bitmap in a vector reg and in a core reg,
1701   // so load both now.
1702   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1703   if (bit != 0) {
1704     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1705   }
1706   // First check the bitmap to see if super_klass might be present. If
1707   // the bit is zero, we are certain that super_klass is not one of
1708   // the secondary supers.
1709   tbz(r_bitmap, bit, L_fallthrough);
1710 
1711   // Get the first array index that can contain super_klass into r_array_index.
1712   if (bit != 0) {
1713     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1714     cnt(vtemp, T8B, vtemp);
1715     addv(vtemp, T8B, vtemp);
1716     fmovd(r_array_index, vtemp);
1717   } else {
1718     mov(r_array_index, (u1)1);
1719   }
1720   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1721 
1722   // We will consult the secondary-super array.
1723   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1724 
1725   // The value i in r_array_index is >= 1, so even though r_array_base
1726   // points to the length, we don't need to adjust it to point to the
1727   // data.
1728   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1729   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1730 
1731   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1732   eor(result, result, r_super_klass);
1733   cbz(result, L_fallthrough); // Found a match
1734 
1735   // Is there another entry to check? Consult the bitmap.
1736   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1737 
1738   // Linear probe.
1739   if (bit != 0) {
1740     ror(r_bitmap, r_bitmap, bit);
1741   }
1742 
1743   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1744   // The next slot to be inspected, by the stub we're about to call,
1745   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1746   // have been checked.
1747   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1748   if (stub_is_near) {
1749     bl(stub);
1750   } else {
1751     address call = trampoline_call(stub);
1752     if (call == nullptr) {
1753       return false; // trampoline allocation failed
1754     }
1755   }
1756 
1757   BLOCK_COMMENT("} lookup_secondary_supers_table");
1758 
1759   bind(L_fallthrough);
1760 
1761   if (VerifySecondarySupers) {
1762     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1763                                   temp1, temp2, result);      // r1, r2, r5
1764   }
1765   return true;
1766 }
1767 
1768 // At runtime, return 0 in result if r_super_klass is a superclass of
1769 // r_sub_klass, otherwise return nonzero. Use this version of
1770 // lookup_secondary_supers_table() if you don't know ahead of time
1771 // which superclass will be searched for. Used by interpreter and
1772 // runtime stubs. It is larger and has somewhat greater latency than
1773 // the version above, which takes a constant super_klass_slot.
1774 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1775                                                        Register r_super_klass,
1776                                                        Register temp1,
1777                                                        Register temp2,
1778                                                        Register temp3,
1779                                                        FloatRegister vtemp,
1780                                                        Register result,
1781                                                        Label *L_success) {
1782   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1783 
1784   Label L_fallthrough;
1785 
1786   BLOCK_COMMENT("lookup_secondary_supers_table {");
1787 
1788   const Register
1789     r_array_index = temp3,
1790     slot          = rscratch1,
1791     r_bitmap      = rscratch2;
1792 
1793   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1794 
1795   // Make sure that result is nonzero if the test below misses.
1796   mov(result, 1);
1797 
1798   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1799 
1800   // First check the bitmap to see if super_klass might be present. If
1801   // the bit is zero, we are certain that super_klass is not one of
1802   // the secondary supers.
1803 
1804   // This next instruction is equivalent to:
1805   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1806   // sub(temp2, tmp_reg, slot);
1807   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1808   lslv(temp2, r_bitmap, temp2);
1809   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1810 
1811   bool must_save_v0 = (vtemp == fnoreg);
1812   if (must_save_v0) {
1813     // temp1 and result are free, so use them to preserve vtemp
1814     vtemp = v0;
1815     mov(temp1,  vtemp, D, 0);
1816     mov(result, vtemp, D, 1);
1817   }
1818 
1819   // Get the first array index that can contain super_klass into r_array_index.
1820   mov(vtemp, D, 0, temp2);
1821   cnt(vtemp, T8B, vtemp);
1822   addv(vtemp, T8B, vtemp);
1823   mov(r_array_index, vtemp, D, 0);
1824 
1825   if (must_save_v0) {
1826     mov(vtemp, D, 0, temp1 );
1827     mov(vtemp, D, 1, result);
1828   }
1829 
1830   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1831 
1832   const Register
1833     r_array_base   = temp1,
1834     r_array_length = temp2;
1835 
1836   // The value i in r_array_index is >= 1, so even though r_array_base
1837   // points to the length, we don't need to adjust it to point to the
1838   // data.
1839   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1840   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1841 
1842   // We will consult the secondary-super array.
1843   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1844 
1845   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1846   eor(result, result, r_super_klass);
1847   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1848 
1849   // Is there another entry to check? Consult the bitmap.
1850   rorv(r_bitmap, r_bitmap, slot);
1851   // rol(r_bitmap, r_bitmap, 1);
1852   tbz(r_bitmap, 1, L_fallthrough);
1853 
1854   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1855   // The next slot to be inspected, by the logic we're about to call,
1856   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1857   // have been checked.
1858   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1859                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1860 
1861   BLOCK_COMMENT("} lookup_secondary_supers_table");
1862 
1863   bind(L_fallthrough);
1864 
1865   if (VerifySecondarySupers) {
1866     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1867                                   temp1, temp2, result);      // r1, r2, r5
1868   }
1869 
1870   if (L_success) {
1871     cbz(result, *L_success);
1872   }
1873 }
1874 
1875 // Called by code generated by check_klass_subtype_slow_path
1876 // above. This is called when there is a collision in the hashed
1877 // lookup in the secondary supers array.
1878 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1879                                                              Register r_array_base,
1880                                                              Register r_array_index,
1881                                                              Register r_bitmap,
1882                                                              Register temp1,
1883                                                              Register result,
1884                                                              bool is_stub) {
1885   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1886 
1887   const Register
1888     r_array_length = temp1,
1889     r_sub_klass    = noreg; // unused
1890 
1891   if (is_stub) {
1892     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1893   }
1894 
1895   Label L_fallthrough, L_huge;
1896 
1897   // Load the array length.
1898   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1899   // And adjust the array base to point to the data.
1900   // NB! Effectively increments current slot index by 1.
1901   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1902   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1903 
1904   // The bitmap is full to bursting.
1905   // Implicit invariant: BITMAP_FULL implies (length > 0)
1906   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1907   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1908   br(GT, L_huge);
1909 
1910   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1911   // current slot (at secondary_supers[r_array_index]) has not yet
1912   // been inspected, and r_array_index may be out of bounds if we
1913   // wrapped around the end of the array.
1914 
1915   { // This is conventional linear probing, but instead of terminating
1916     // when a null entry is found in the table, we maintain a bitmap
1917     // in which a 0 indicates missing entries.
1918     // As long as the bitmap is not completely full,
1919     // array_length == popcount(bitmap). The array_length check above
1920     // guarantees there are 0s in the bitmap, so the loop eventually
1921     // terminates.
1922     Label L_loop;
1923     bind(L_loop);
1924 
1925     // Check for wraparound.
1926     cmp(r_array_index, r_array_length);
1927     csel(r_array_index, zr, r_array_index, GE);
1928 
1929     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1930     eor(result, rscratch1, r_super_klass);
1931     cbz(result, L_fallthrough);
1932 
1933     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1934 
1935     ror(r_bitmap, r_bitmap, 1);
1936     add(r_array_index, r_array_index, 1);
1937     b(L_loop);
1938   }
1939 
1940   { // Degenerate case: more than 64 secondary supers.
1941     // FIXME: We could do something smarter here, maybe a vectorized
1942     // comparison or a binary search, but is that worth any added
1943     // complexity?
1944     bind(L_huge);
1945     cmp(sp, zr); // Clear Z flag; SP is never zero
1946     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1947     cset(result, NE); // result == 0 iff we got a match.
1948   }
1949 
1950   bind(L_fallthrough);
1951 }
1952 
1953 // Make sure that the hashed lookup and a linear scan agree.
1954 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1955                                                    Register r_super_klass,
1956                                                    Register temp1,
1957                                                    Register temp2,
1958                                                    Register result) {
1959   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1960 
1961   const Register
1962     r_array_base   = temp1,
1963     r_array_length = temp2,
1964     r_array_index  = noreg, // unused
1965     r_bitmap       = noreg; // unused
1966 
1967   BLOCK_COMMENT("verify_secondary_supers_table {");
1968 
1969   // We will consult the secondary-super array.
1970   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1971 
1972   // Load the array length.
1973   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1974   // And adjust the array base to point to the data.
1975   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1976 
1977   cmp(sp, zr); // Clear Z flag; SP is never zero
1978   // Scan R2 words at [R5] for an occurrence of R0.
1979   // Set NZ/Z based on last compare.
1980   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1981   // rscratch1 == 0 iff we got a match.
1982   cset(rscratch1, NE);
1983 
1984   Label passed;
1985   cmp(result, zr);
1986   cset(result, NE); // normalize result to 0/1 for comparison
1987 
1988   cmp(rscratch1, result);
1989   br(EQ, passed);
1990   {
1991     mov(r0, r_super_klass);         // r0 <- r0
1992     mov(r1, r_sub_klass);           // r1 <- r4
1993     mov(r2, /*expected*/rscratch1); // r2 <- r8
1994     mov(r3, result);                // r3 <- r5
1995     mov(r4, (address)("mismatch")); // r4 <- const
1996     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
1997     should_not_reach_here();
1998   }
1999   bind(passed);
2000 
2001   BLOCK_COMMENT("} verify_secondary_supers_table");
2002 }
2003 
2004 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
2005   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
2006   assert_different_registers(klass, rthread, scratch);
2007 
2008   Label L_fallthrough, L_tmp;
2009   if (L_fast_path == nullptr) {
2010     L_fast_path = &L_fallthrough;
2011   } else if (L_slow_path == nullptr) {
2012     L_slow_path = &L_fallthrough;
2013   }
2014   // Fast path check: class is fully initialized
2015   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2016   ldarb(scratch, scratch);
2017   cmp(scratch, InstanceKlass::fully_initialized);
2018   br(Assembler::EQ, *L_fast_path);
2019 
2020   // Fast path check: current thread is initializer thread
2021   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2022   cmp(rthread, scratch);
2023 
2024   if (L_slow_path == &L_fallthrough) {
2025     br(Assembler::EQ, *L_fast_path);
2026     bind(*L_slow_path);
2027   } else if (L_fast_path == &L_fallthrough) {
2028     br(Assembler::NE, *L_slow_path);
2029     bind(*L_fast_path);
2030   } else {
2031     Unimplemented();
2032   }
2033 }
2034 
2035 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2036   if (!VerifyOops) return;
2037 
2038   // Pass register number to verify_oop_subroutine
2039   const char* b = nullptr;
2040   {
2041     ResourceMark rm;
2042     stringStream ss;
2043     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2044     b = code_string(ss.as_string());
2045   }
2046   BLOCK_COMMENT("verify_oop {");
2047 
2048   strip_return_address(); // This might happen within a stack frame.
2049   protect_return_address();
2050   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2051   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2052 
2053   mov(r0, reg);
2054   movptr(rscratch1, (uintptr_t)(address)b);
2055 
2056   // call indirectly to solve generation ordering problem
2057   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2058   ldr(rscratch2, Address(rscratch2));
2059   blr(rscratch2);
2060 
2061   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2062   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2063   authenticate_return_address();
2064 
2065   BLOCK_COMMENT("} verify_oop");
2066 }
2067 
2068 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2069   if (!VerifyOops) return;
2070 
2071   const char* b = nullptr;
2072   {
2073     ResourceMark rm;
2074     stringStream ss;
2075     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2076     b = code_string(ss.as_string());
2077   }
2078   BLOCK_COMMENT("verify_oop_addr {");
2079 
2080   strip_return_address(); // This might happen within a stack frame.
2081   protect_return_address();
2082   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2083   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2084 
2085   // addr may contain sp so we will have to adjust it based on the
2086   // pushes that we just did.
2087   if (addr.uses(sp)) {
2088     lea(r0, addr);
2089     ldr(r0, Address(r0, 4 * wordSize));
2090   } else {
2091     ldr(r0, addr);
2092   }
2093   movptr(rscratch1, (uintptr_t)(address)b);
2094 
2095   // call indirectly to solve generation ordering problem
2096   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2097   ldr(rscratch2, Address(rscratch2));
2098   blr(rscratch2);
2099 
2100   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2101   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2102   authenticate_return_address();
2103 
2104   BLOCK_COMMENT("} verify_oop_addr");
2105 }
2106 
2107 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2108                                          int extra_slot_offset) {
2109   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2110   int stackElementSize = Interpreter::stackElementSize;
2111   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2112 #ifdef ASSERT
2113   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2114   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2115 #endif
2116   if (arg_slot.is_constant()) {
2117     return Address(esp, arg_slot.as_constant() * stackElementSize
2118                    + offset);
2119   } else {
2120     add(rscratch1, esp, arg_slot.as_register(),
2121         ext::uxtx, exact_log2(stackElementSize));
2122     return Address(rscratch1, offset);
2123   }
2124 }
2125 
2126 // Handle the receiver type profile update given the "recv" klass.
2127 //
2128 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2129 // If there are no matching or claimable receiver entries in RD, updates
2130 // the polymorphic counter.
2131 //
2132 // This code expected to run by either the interpreter or JIT-ed code, without
2133 // extra synchronization. For safety, receiver cells are claimed atomically, which
2134 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2135 // counter updates are not atomic.
2136 //
2137 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2138   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2139 
2140   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2141   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2142   int poly_count_offset      = in_bytes(CounterData::count_offset());
2143   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2144   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2145 
2146   // Adjust for MDP offsets.
2147   base_receiver_offset += mdp_offset;
2148   end_receiver_offset  += mdp_offset;
2149   poly_count_offset    += mdp_offset;
2150 
2151 #ifdef ASSERT
2152   // We are about to walk the MDO slots without asking for offsets.
2153   // Check that our math hits all the right spots.
2154   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2155     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2156     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2157     int offset = base_receiver_offset + receiver_step*c;
2158     int count_offset = offset + receiver_to_count_step;
2159     assert(offset == real_recv_offset, "receiver slot math");
2160     assert(count_offset == real_count_offset, "receiver count math");
2161   }
2162   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2163   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2164 #endif
2165 
2166   // Corner case: no profile table. Increment poly counter and exit.
2167   if (ReceiverTypeData::row_limit() == 0) {
2168     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2169     return;
2170   }
2171 
2172   Register offset = rscratch2;
2173 
2174   Label L_loop_search_receiver, L_loop_search_empty;
2175   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
2176 
2177   // The code here recognizes three major cases:
2178   //   A. Fastest: receiver found in the table
2179   //   B. Fast: no receiver in the table, and the table is full
2180   //   C. Slow: no receiver in the table, free slots in the table
2181   //
2182   // The case A performance is most important, as perfectly-behaved code would end up
2183   // there, especially with larger TypeProfileWidth. The case B performance is
2184   // important as well, this is where bulk of code would land for normally megamorphic
2185   // cases. The case C performance is not essential, its job is to deal with installation
2186   // races, we optimize for code density instead. Case C needs to make sure that receiver
2187   // rows are only claimed once. This makes sure we never overwrite a row for another
2188   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2189   //
2190   // It is very tempting to handle these cases in a single loop, and claim the first slot
2191   // without checking the rest of the table. But, profiling code should tolerate free slots
2192   // in the table, as class unloading can clear them. After such cleanup, the receiver
2193   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2194   // to complete, before trying to install new slots. Splitting the code in several tight
2195   // loops also helpfully optimizes for cases A and B.
2196   //
2197   // This code is effectively:
2198   //
2199   // restart:
2200   //   // Fastest: receiver is already installed
2201   //   for (i = 0; i < receiver_count(); i++) {
2202   //     if (receiver(i) == recv) goto found_recv(i);
2203   //   }
2204   //
2205   //   // Fast: no receiver, but profile is full
2206   //   for (i = 0; i < receiver_count(); i++) {
2207   //     if (receiver(i) == null) goto found_null(i);
2208   //   }
2209   //   goto polymorphic
2210   //
2211   //   // Slow: try to install receiver
2212   // found_null(i):
2213   //   CAS(&receiver(i), null, recv);
2214   //   goto restart
2215   //
2216   // polymorphic:
2217   //   count++;
2218   //   return
2219   //
2220   // found_recv(i):
2221   //   *receiver_count(i)++
2222   //
2223 
2224   bind(L_restart);
2225 
2226   // Fastest: receiver is already installed
2227   mov(offset, base_receiver_offset);
2228   bind(L_loop_search_receiver);
2229     ldr(rscratch1, Address(mdp, offset));
2230     cmp(rscratch1, recv);
2231     br(Assembler::EQ, L_found_recv);
2232   add(offset, offset, receiver_step);
2233   sub(rscratch1, offset, end_receiver_offset);
2234   cbnz(rscratch1, L_loop_search_receiver);
2235 
2236   // Fast: no receiver, but profile is full
2237   mov(offset, base_receiver_offset);
2238   bind(L_loop_search_empty);
2239     ldr(rscratch1, Address(mdp, offset));
2240     cbz(rscratch1, L_found_empty);
2241   add(offset, offset, receiver_step);
2242   sub(rscratch1, offset, end_receiver_offset);
2243   cbnz(rscratch1, L_loop_search_empty);
2244   b(L_polymorphic);
2245 
2246   // Slow: try to install receiver
2247   bind(L_found_empty);
2248 
2249   // Atomically swing receiver slot: null -> recv.
2250   //
2251   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2252   // is used to hold the destination address. This is safe because the
2253   // offset is no longer needed after the address is computed.
2254 
2255   lea(rscratch2, Address(mdp, offset));
2256   cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
2257           /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
2258 
2259   // CAS success means the slot now has the receiver we want. CAS failure means
2260   // something had claimed the slot concurrently: it can be the same receiver we want,
2261   // or something else. Since this is a slow path, we can optimize for code density,
2262   // and just restart the search from the beginning.
2263   b(L_restart);
2264 
2265   // Counter updates:
2266 
2267   // Increment polymorphic counter instead of receiver slot.
2268   bind(L_polymorphic);
2269   mov(offset, poly_count_offset);
2270   b(L_count_update);
2271 
2272   // Found a receiver, convert its slot offset to corresponding count offset.
2273   bind(L_found_recv);
2274   add(offset, offset, receiver_to_count_step);
2275 
2276   bind(L_count_update);
2277   increment(Address(mdp, offset), DataLayout::counter_increment);
2278 }
2279 
2280 
2281 void MacroAssembler::call_VM_leaf_base(address entry_point,
2282                                        int number_of_arguments,
2283                                        Label *retaddr) {
2284   Label E, L;
2285 
2286   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2287 
2288   mov(rscratch1, RuntimeAddress(entry_point));
2289   blr(rscratch1);
2290   if (retaddr)
2291     bind(*retaddr);
2292 
2293   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2294 }
2295 
2296 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2297   call_VM_leaf_base(entry_point, number_of_arguments);
2298 }
2299 
2300 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2301   pass_arg0(this, arg_0);
2302   call_VM_leaf_base(entry_point, 1);
2303 }
2304 
2305 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2306   assert_different_registers(arg_1, c_rarg0);
2307   pass_arg0(this, arg_0);
2308   pass_arg1(this, arg_1);
2309   call_VM_leaf_base(entry_point, 2);
2310 }
2311 
2312 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2313                                   Register arg_1, Register arg_2) {
2314   assert_different_registers(arg_1, c_rarg0);
2315   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2316   pass_arg0(this, arg_0);
2317   pass_arg1(this, arg_1);
2318   pass_arg2(this, arg_2);
2319   call_VM_leaf_base(entry_point, 3);
2320 }
2321 
2322 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2323   pass_arg0(this, arg_0);
2324   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2325 }
2326 
2327 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2328 
2329   assert_different_registers(arg_0, c_rarg1);
2330   pass_arg1(this, arg_1);
2331   pass_arg0(this, arg_0);
2332   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2333 }
2334 
2335 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2336   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2337   assert_different_registers(arg_1, c_rarg2);
2338   pass_arg2(this, arg_2);
2339   pass_arg1(this, arg_1);
2340   pass_arg0(this, arg_0);
2341   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2342 }
2343 
2344 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2345   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2346   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2347   assert_different_registers(arg_2, c_rarg3);
2348   pass_arg3(this, arg_3);
2349   pass_arg2(this, arg_2);
2350   pass_arg1(this, arg_1);
2351   pass_arg0(this, arg_0);
2352   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2353 }
2354 
2355 void MacroAssembler::null_check(Register reg, int offset) {
2356   if (needs_explicit_null_check(offset)) {
2357     // provoke OS null exception if reg is null by
2358     // accessing M[reg] w/o changing any registers
2359     // NOTE: this is plenty to provoke a segv
2360     ldr(zr, Address(reg));
2361   } else {
2362     // nothing to do, (later) access of M[reg + offset]
2363     // will provoke OS null exception if reg is null
2364   }
2365 }
2366 
2367 // MacroAssembler protected routines needed to implement
2368 // public methods
2369 
2370 void MacroAssembler::mov(Register r, Address dest) {
2371   code_section()->relocate(pc(), dest.rspec());
2372   uint64_t imm64 = (uint64_t)dest.target();
2373   movptr(r, imm64);
2374 }
2375 
2376 // Move a constant pointer into r.  In AArch64 mode the virtual
2377 // address space is 48 bits in size, so we only need three
2378 // instructions to create a patchable instruction sequence that can
2379 // reach anywhere.
2380 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2381 #ifndef PRODUCT
2382   {
2383     char buffer[64];
2384     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2385     block_comment(buffer);
2386   }
2387 #endif
2388   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2389   movz(r, imm64 & 0xffff);
2390   imm64 >>= 16;
2391   movk(r, imm64 & 0xffff, 16);
2392   imm64 >>= 16;
2393   movk(r, imm64 & 0xffff, 32);
2394 }
2395 
2396 // Macro to mov replicated immediate to vector register.
2397 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2398 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2399 // Vd will get the following values for different arrangements in T
2400 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2401 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2402 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2403 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2404 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2405 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2406 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2407 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2408 // Clobbers rscratch1
2409 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2410   assert(T != T1Q, "unsupported");
2411   if (T == T1D || T == T2D) {
2412     int imm = operand_valid_for_movi_immediate(imm64, T);
2413     if (-1 != imm) {
2414       movi(Vd, T, imm);
2415     } else {
2416       mov(rscratch1, imm64);
2417       dup(Vd, T, rscratch1);
2418     }
2419     return;
2420   }
2421 
2422 #ifdef ASSERT
2423   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2424   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2425   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2426 #endif
2427   int shift = operand_valid_for_movi_immediate(imm64, T);
2428   uint32_t imm32 = imm64 & 0xffffffffULL;
2429   if (shift >= 0) {
2430     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2431   } else {
2432     movw(rscratch1, imm32);
2433     dup(Vd, T, rscratch1);
2434   }
2435 }
2436 
2437 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2438 {
2439 #ifndef PRODUCT
2440   {
2441     char buffer[64];
2442     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2443     block_comment(buffer);
2444   }
2445 #endif
2446   if (operand_valid_for_logical_immediate(false, imm64)) {
2447     orr(dst, zr, imm64);
2448   } else {
2449     // we can use a combination of MOVZ or MOVN with
2450     // MOVK to build up the constant
2451     uint64_t imm_h[4];
2452     int zero_count = 0;
2453     int neg_count = 0;
2454     int i;
2455     for (i = 0; i < 4; i++) {
2456       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2457       if (imm_h[i] == 0) {
2458         zero_count++;
2459       } else if (imm_h[i] == 0xffffL) {
2460         neg_count++;
2461       }
2462     }
2463     if (zero_count == 4) {
2464       // one MOVZ will do
2465       movz(dst, 0);
2466     } else if (neg_count == 4) {
2467       // one MOVN will do
2468       movn(dst, 0);
2469     } else if (zero_count == 3) {
2470       for (i = 0; i < 4; i++) {
2471         if (imm_h[i] != 0L) {
2472           movz(dst, (uint32_t)imm_h[i], (i << 4));
2473           break;
2474         }
2475       }
2476     } else if (neg_count == 3) {
2477       // one MOVN will do
2478       for (int i = 0; i < 4; i++) {
2479         if (imm_h[i] != 0xffffL) {
2480           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2481           break;
2482         }
2483       }
2484     } else if (zero_count == 2) {
2485       // one MOVZ and one MOVK will do
2486       for (i = 0; i < 3; i++) {
2487         if (imm_h[i] != 0L) {
2488           movz(dst, (uint32_t)imm_h[i], (i << 4));
2489           i++;
2490           break;
2491         }
2492       }
2493       for (;i < 4; i++) {
2494         if (imm_h[i] != 0L) {
2495           movk(dst, (uint32_t)imm_h[i], (i << 4));
2496         }
2497       }
2498     } else if (neg_count == 2) {
2499       // one MOVN and one MOVK will do
2500       for (i = 0; i < 4; i++) {
2501         if (imm_h[i] != 0xffffL) {
2502           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2503           i++;
2504           break;
2505         }
2506       }
2507       for (;i < 4; i++) {
2508         if (imm_h[i] != 0xffffL) {
2509           movk(dst, (uint32_t)imm_h[i], (i << 4));
2510         }
2511       }
2512     } else if (zero_count == 1) {
2513       // one MOVZ and two MOVKs will do
2514       for (i = 0; i < 4; i++) {
2515         if (imm_h[i] != 0L) {
2516           movz(dst, (uint32_t)imm_h[i], (i << 4));
2517           i++;
2518           break;
2519         }
2520       }
2521       for (;i < 4; i++) {
2522         if (imm_h[i] != 0x0L) {
2523           movk(dst, (uint32_t)imm_h[i], (i << 4));
2524         }
2525       }
2526     } else if (neg_count == 1) {
2527       // one MOVN and two MOVKs will do
2528       for (i = 0; i < 4; i++) {
2529         if (imm_h[i] != 0xffffL) {
2530           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2531           i++;
2532           break;
2533         }
2534       }
2535       for (;i < 4; i++) {
2536         if (imm_h[i] != 0xffffL) {
2537           movk(dst, (uint32_t)imm_h[i], (i << 4));
2538         }
2539       }
2540     } else {
2541       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2542       movz(dst, (uint32_t)imm_h[0], 0);
2543       for (i = 1; i < 4; i++) {
2544         movk(dst, (uint32_t)imm_h[i], (i << 4));
2545       }
2546     }
2547   }
2548 }
2549 
2550 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2551 {
2552 #ifndef PRODUCT
2553     {
2554       char buffer[64];
2555       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2556       block_comment(buffer);
2557     }
2558 #endif
2559   if (operand_valid_for_logical_immediate(true, imm32)) {
2560     orrw(dst, zr, imm32);
2561   } else {
2562     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2563     // constant
2564     uint32_t imm_h[2];
2565     imm_h[0] = imm32 & 0xffff;
2566     imm_h[1] = ((imm32 >> 16) & 0xffff);
2567     if (imm_h[0] == 0) {
2568       movzw(dst, imm_h[1], 16);
2569     } else if (imm_h[0] == 0xffff) {
2570       movnw(dst, imm_h[1] ^ 0xffff, 16);
2571     } else if (imm_h[1] == 0) {
2572       movzw(dst, imm_h[0], 0);
2573     } else if (imm_h[1] == 0xffff) {
2574       movnw(dst, imm_h[0] ^ 0xffff, 0);
2575     } else {
2576       // use a MOVZ and MOVK (makes it easier to debug)
2577       movzw(dst, imm_h[0], 0);
2578       movkw(dst, imm_h[1], 16);
2579     }
2580   }
2581 }
2582 
2583 // Form an address from base + offset in Rd.  Rd may or may
2584 // not actually be used: you must use the Address that is returned.
2585 // It is up to you to ensure that the shift provided matches the size
2586 // of your data.
2587 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2588   if (Address::offset_ok_for_immed(byte_offset, shift))
2589     // It fits; no need for any heroics
2590     return Address(base, byte_offset);
2591 
2592   // Don't do anything clever with negative or misaligned offsets
2593   unsigned mask = (1 << shift) - 1;
2594   if (byte_offset < 0 || byte_offset & mask) {
2595     mov(Rd, byte_offset);
2596     add(Rd, base, Rd);
2597     return Address(Rd);
2598   }
2599 
2600   // See if we can do this with two 12-bit offsets
2601   {
2602     uint64_t word_offset = byte_offset >> shift;
2603     uint64_t masked_offset = word_offset & 0xfff000;
2604     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2605         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2606       add(Rd, base, masked_offset << shift);
2607       word_offset -= masked_offset;
2608       return Address(Rd, word_offset << shift);
2609     }
2610   }
2611 
2612   // Do it the hard way
2613   mov(Rd, byte_offset);
2614   add(Rd, base, Rd);
2615   return Address(Rd);
2616 }
2617 
2618 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2619                                     bool want_remainder, Register scratch)
2620 {
2621   // Full implementation of Java idiv and irem.  The function
2622   // returns the (pc) offset of the div instruction - may be needed
2623   // for implicit exceptions.
2624   //
2625   // constraint : ra/rb =/= scratch
2626   //         normal case
2627   //
2628   // input : ra: dividend
2629   //         rb: divisor
2630   //
2631   // result: either
2632   //         quotient  (= ra idiv rb)
2633   //         remainder (= ra irem rb)
2634 
2635   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2636 
2637   int idivl_offset = offset();
2638   if (! want_remainder) {
2639     sdivw(result, ra, rb);
2640   } else {
2641     sdivw(scratch, ra, rb);
2642     Assembler::msubw(result, scratch, rb, ra);
2643   }
2644 
2645   return idivl_offset;
2646 }
2647 
2648 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2649                                     bool want_remainder, Register scratch)
2650 {
2651   // Full implementation of Java ldiv and lrem.  The function
2652   // returns the (pc) offset of the div instruction - may be needed
2653   // for implicit exceptions.
2654   //
2655   // constraint : ra/rb =/= scratch
2656   //         normal case
2657   //
2658   // input : ra: dividend
2659   //         rb: divisor
2660   //
2661   // result: either
2662   //         quotient  (= ra idiv rb)
2663   //         remainder (= ra irem rb)
2664 
2665   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2666 
2667   int idivq_offset = offset();
2668   if (! want_remainder) {
2669     sdiv(result, ra, rb);
2670   } else {
2671     sdiv(scratch, ra, rb);
2672     Assembler::msub(result, scratch, rb, ra);
2673   }
2674 
2675   return idivq_offset;
2676 }
2677 
2678 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2679   address prev = pc() - NativeMembar::instruction_size;
2680   address last = code()->last_insn();
2681   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2682     NativeMembar *bar = NativeMembar_at(prev);
2683     if (AlwaysMergeDMB) {
2684       bar->set_kind(bar->get_kind() | order_constraint);
2685       BLOCK_COMMENT("merged membar(always)");
2686       return;
2687     }
2688     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2689     // doing so would introduce a StoreLoad which the caller did not
2690     // intend
2691     if (bar->get_kind() == order_constraint
2692         || bar->get_kind() == AnyAny
2693         || order_constraint == AnyAny) {
2694       // We are merging two memory barrier instructions.  On AArch64 we
2695       // can do this simply by ORing them together.
2696       bar->set_kind(bar->get_kind() | order_constraint);
2697       BLOCK_COMMENT("merged membar");
2698       return;
2699     } else {
2700       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2701       // We need check the last 2 instructions
2702       address prev2 = prev - NativeMembar::instruction_size;
2703       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2704         NativeMembar *bar2 = NativeMembar_at(prev2);
2705         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2706         BLOCK_COMMENT("merged membar(elided)");
2707         return;
2708       }
2709     }
2710   }
2711   code()->set_last_insn(pc());
2712   dmb(Assembler::barrier(order_constraint));
2713 }
2714 
2715 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2716   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2717     merge_ldst(rt, adr, size_in_bytes, is_store);
2718     code()->clear_last_insn();
2719     return true;
2720   } else {
2721     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2722     const uint64_t mask = size_in_bytes - 1;
2723     if (adr.getMode() == Address::base_plus_offset &&
2724         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2725       code()->set_last_insn(pc());
2726     }
2727     return false;
2728   }
2729 }
2730 
2731 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2732   // We always try to merge two adjacent loads into one ldp.
2733   if (!try_merge_ldst(Rx, adr, 8, false)) {
2734     Assembler::ldr(Rx, adr);
2735   }
2736 }
2737 
2738 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2739   // We always try to merge two adjacent loads into one ldp.
2740   if (!try_merge_ldst(Rw, adr, 4, false)) {
2741     Assembler::ldrw(Rw, adr);
2742   }
2743 }
2744 
2745 void MacroAssembler::str(Register Rx, const Address &adr) {
2746   // We always try to merge two adjacent stores into one stp.
2747   if (!try_merge_ldst(Rx, adr, 8, true)) {
2748     Assembler::str(Rx, adr);
2749   }
2750 }
2751 
2752 void MacroAssembler::strw(Register Rw, const Address &adr) {
2753   // We always try to merge two adjacent stores into one stp.
2754   if (!try_merge_ldst(Rw, adr, 4, true)) {
2755     Assembler::strw(Rw, adr);
2756   }
2757 }
2758 
2759 // MacroAssembler routines found actually to be needed
2760 
2761 void MacroAssembler::push(Register src)
2762 {
2763   str(src, Address(pre(esp, -1 * wordSize)));
2764 }
2765 
2766 void MacroAssembler::pop(Register dst)
2767 {
2768   ldr(dst, Address(post(esp, 1 * wordSize)));
2769 }
2770 
2771 // Note: load_unsigned_short used to be called load_unsigned_word.
2772 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2773   int off = offset();
2774   ldrh(dst, src);
2775   return off;
2776 }
2777 
2778 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2779   int off = offset();
2780   ldrb(dst, src);
2781   return off;
2782 }
2783 
2784 int MacroAssembler::load_signed_short(Register dst, Address src) {
2785   int off = offset();
2786   ldrsh(dst, src);
2787   return off;
2788 }
2789 
2790 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2791   int off = offset();
2792   ldrsb(dst, src);
2793   return off;
2794 }
2795 
2796 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2797   int off = offset();
2798   ldrshw(dst, src);
2799   return off;
2800 }
2801 
2802 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2803   int off = offset();
2804   ldrsbw(dst, src);
2805   return off;
2806 }
2807 
2808 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2809   switch (size_in_bytes) {
2810   case  8:  ldr(dst, src); break;
2811   case  4:  ldrw(dst, src); break;
2812   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2813   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2814   default:  ShouldNotReachHere();
2815   }
2816 }
2817 
2818 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2819   switch (size_in_bytes) {
2820   case  8:  str(src, dst); break;
2821   case  4:  strw(src, dst); break;
2822   case  2:  strh(src, dst); break;
2823   case  1:  strb(src, dst); break;
2824   default:  ShouldNotReachHere();
2825   }
2826 }
2827 
2828 void MacroAssembler::decrementw(Register reg, int value)
2829 {
2830   if (value < 0)  { incrementw(reg, -value);      return; }
2831   if (value == 0) {                               return; }
2832   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2833   /* else */ {
2834     guarantee(reg != rscratch2, "invalid dst for register decrement");
2835     movw(rscratch2, (unsigned)value);
2836     subw(reg, reg, rscratch2);
2837   }
2838 }
2839 
2840 void MacroAssembler::decrement(Register reg, int value)
2841 {
2842   if (value < 0)  { increment(reg, -value);      return; }
2843   if (value == 0) {                              return; }
2844   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2845   /* else */ {
2846     assert(reg != rscratch2, "invalid dst for register decrement");
2847     mov(rscratch2, (uint64_t)value);
2848     sub(reg, reg, rscratch2);
2849   }
2850 }
2851 
2852 void MacroAssembler::decrementw(Address dst, int value)
2853 {
2854   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2855   if (dst.getMode() == Address::literal) {
2856     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2857     lea(rscratch2, dst);
2858     dst = Address(rscratch2);
2859   }
2860   ldrw(rscratch1, dst);
2861   decrementw(rscratch1, value);
2862   strw(rscratch1, dst);
2863 }
2864 
2865 void MacroAssembler::decrement(Address dst, int value)
2866 {
2867   assert(!dst.uses(rscratch1), "invalid address for decrement");
2868   if (dst.getMode() == Address::literal) {
2869     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2870     lea(rscratch2, dst);
2871     dst = Address(rscratch2);
2872   }
2873   ldr(rscratch1, dst);
2874   decrement(rscratch1, value);
2875   str(rscratch1, dst);
2876 }
2877 
2878 void MacroAssembler::incrementw(Register reg, int value)
2879 {
2880   if (value < 0)  { decrementw(reg, -value);      return; }
2881   if (value == 0) {                               return; }
2882   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2883   /* else */ {
2884     assert(reg != rscratch2, "invalid dst for register increment");
2885     movw(rscratch2, (unsigned)value);
2886     addw(reg, reg, rscratch2);
2887   }
2888 }
2889 
2890 void MacroAssembler::increment(Register reg, int value)
2891 {
2892   if (value < 0)  { decrement(reg, -value);      return; }
2893   if (value == 0) {                              return; }
2894   if (value < (1 << 12)) { add(reg, reg, value); return; }
2895   /* else */ {
2896     assert(reg != rscratch2, "invalid dst for register increment");
2897     movw(rscratch2, (unsigned)value);
2898     add(reg, reg, rscratch2);
2899   }
2900 }
2901 
2902 void MacroAssembler::incrementw(Address dst, int value)
2903 {
2904   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2905   if (dst.getMode() == Address::literal) {
2906     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2907     lea(rscratch2, dst);
2908     dst = Address(rscratch2);
2909   }
2910   ldrw(rscratch1, dst);
2911   incrementw(rscratch1, value);
2912   strw(rscratch1, dst);
2913 }
2914 
2915 void MacroAssembler::increment(Address dst, int value)
2916 {
2917   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2918   if (dst.getMode() == Address::literal) {
2919     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2920     lea(rscratch2, dst);
2921     dst = Address(rscratch2);
2922   }
2923   ldr(rscratch1, dst);
2924   increment(rscratch1, value);
2925   str(rscratch1, dst);
2926 }
2927 
2928 // Push lots of registers in the bit set supplied.  Don't push sp.
2929 // Return the number of words pushed
2930 int MacroAssembler::push(unsigned int bitset, Register stack) {
2931   int words_pushed = 0;
2932 
2933   // Scan bitset to accumulate register pairs
2934   unsigned char regs[32];
2935   int count = 0;
2936   for (int reg = 0; reg <= 30; reg++) {
2937     if (1 & bitset)
2938       regs[count++] = reg;
2939     bitset >>= 1;
2940   }
2941   regs[count++] = zr->raw_encoding();
2942   count &= ~1;  // Only push an even number of regs
2943 
2944   if (count) {
2945     stp(as_Register(regs[0]), as_Register(regs[1]),
2946        Address(pre(stack, -count * wordSize)));
2947     words_pushed += 2;
2948   }
2949   for (int i = 2; i < count; i += 2) {
2950     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2951        Address(stack, i * wordSize));
2952     words_pushed += 2;
2953   }
2954 
2955   assert(words_pushed == count, "oops, pushed != count");
2956 
2957   return count;
2958 }
2959 
2960 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2961   int words_pushed = 0;
2962 
2963   // Scan bitset to accumulate register pairs
2964   unsigned char regs[32];
2965   int count = 0;
2966   for (int reg = 0; reg <= 30; reg++) {
2967     if (1 & bitset)
2968       regs[count++] = reg;
2969     bitset >>= 1;
2970   }
2971   regs[count++] = zr->raw_encoding();
2972   count &= ~1;
2973 
2974   for (int i = 2; i < count; i += 2) {
2975     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2976        Address(stack, i * wordSize));
2977     words_pushed += 2;
2978   }
2979   if (count) {
2980     ldp(as_Register(regs[0]), as_Register(regs[1]),
2981        Address(post(stack, count * wordSize)));
2982     words_pushed += 2;
2983   }
2984 
2985   assert(words_pushed == count, "oops, pushed != count");
2986 
2987   return count;
2988 }
2989 
2990 // Push lots of registers in the bit set supplied.  Don't push sp.
2991 // Return the number of dwords pushed
2992 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2993   int words_pushed = 0;
2994   bool use_sve = false;
2995   int sve_vector_size_in_bytes = 0;
2996 
2997 #ifdef COMPILER2
2998   use_sve = Matcher::supports_scalable_vector();
2999   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3000 #endif
3001 
3002   // Scan bitset to accumulate register pairs
3003   unsigned char regs[32];
3004   int count = 0;
3005   for (int reg = 0; reg <= 31; reg++) {
3006     if (1 & bitset)
3007       regs[count++] = reg;
3008     bitset >>= 1;
3009   }
3010 
3011   if (count == 0) {
3012     return 0;
3013   }
3014 
3015   if (mode == PushPopFull) {
3016     if (use_sve && sve_vector_size_in_bytes > 16) {
3017       mode = PushPopSVE;
3018     } else {
3019       mode = PushPopNeon;
3020     }
3021   }
3022 
3023 #ifndef PRODUCT
3024   {
3025     char buffer[48];
3026     if (mode == PushPopSVE) {
3027       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3028     } else if (mode == PushPopNeon) {
3029       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3030     } else {
3031       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3032     }
3033     block_comment(buffer);
3034   }
3035 #endif
3036 
3037   if (mode == PushPopSVE) {
3038     sub(stack, stack, sve_vector_size_in_bytes * count);
3039     for (int i = 0; i < count; i++) {
3040       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3041     }
3042     return count * sve_vector_size_in_bytes / 8;
3043   }
3044 
3045   if (mode == PushPopNeon) {
3046     if (count == 1) {
3047       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3048       return 2;
3049     }
3050 
3051     bool odd = (count & 1) == 1;
3052     int push_slots = count + (odd ? 1 : 0);
3053 
3054     // Always pushing full 128 bit registers.
3055     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3056     words_pushed += 2;
3057 
3058     for (int i = 2; i + 1 < count; i += 2) {
3059       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3060       words_pushed += 2;
3061     }
3062 
3063     if (odd) {
3064       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3065       words_pushed++;
3066     }
3067 
3068     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3069     return count * 2;
3070   }
3071 
3072   if (mode == PushPopFp) {
3073     bool odd = (count & 1) == 1;
3074     int push_slots = count + (odd ? 1 : 0);
3075 
3076     if (count == 1) {
3077       // Stack pointer must be 16 bytes aligned
3078       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3079       return 1;
3080     }
3081 
3082     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3083     words_pushed += 2;
3084 
3085     for (int i = 2; i + 1 < count; i += 2) {
3086       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3087       words_pushed += 2;
3088     }
3089 
3090     if (odd) {
3091       // Stack pointer must be 16 bytes aligned
3092       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3093       words_pushed++;
3094     }
3095 
3096     assert(words_pushed == count, "oops, pushed != count");
3097 
3098     return count;
3099   }
3100 
3101   return 0;
3102 }
3103 
3104 // Return the number of dwords popped
3105 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3106   int words_pushed = 0;
3107   bool use_sve = false;
3108   int sve_vector_size_in_bytes = 0;
3109 
3110 #ifdef COMPILER2
3111   use_sve = Matcher::supports_scalable_vector();
3112   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3113 #endif
3114   // Scan bitset to accumulate register pairs
3115   unsigned char regs[32];
3116   int count = 0;
3117   for (int reg = 0; reg <= 31; reg++) {
3118     if (1 & bitset)
3119       regs[count++] = reg;
3120     bitset >>= 1;
3121   }
3122 
3123   if (count == 0) {
3124     return 0;
3125   }
3126 
3127   if (mode == PushPopFull) {
3128     if (use_sve && sve_vector_size_in_bytes > 16) {
3129       mode = PushPopSVE;
3130     } else {
3131       mode = PushPopNeon;
3132     }
3133   }
3134 
3135 #ifndef PRODUCT
3136   {
3137     char buffer[48];
3138     if (mode == PushPopSVE) {
3139       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3140     } else if (mode == PushPopNeon) {
3141       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3142     } else {
3143       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3144     }
3145     block_comment(buffer);
3146   }
3147 #endif
3148 
3149   if (mode == PushPopSVE) {
3150     for (int i = count - 1; i >= 0; i--) {
3151       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3152     }
3153     add(stack, stack, sve_vector_size_in_bytes * count);
3154     return count * sve_vector_size_in_bytes / 8;
3155   }
3156 
3157   if (mode == PushPopNeon) {
3158     if (count == 1) {
3159       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3160       return 2;
3161     }
3162 
3163     bool odd = (count & 1) == 1;
3164     int push_slots = count + (odd ? 1 : 0);
3165 
3166     if (odd) {
3167       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3168       words_pushed++;
3169     }
3170 
3171     for (int i = 2; i + 1 < count; i += 2) {
3172       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3173       words_pushed += 2;
3174     }
3175 
3176     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3177     words_pushed += 2;
3178 
3179     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3180 
3181     return count * 2;
3182   }
3183 
3184   if (mode == PushPopFp) {
3185     bool odd = (count & 1) == 1;
3186     int push_slots = count + (odd ? 1 : 0);
3187 
3188     if (count == 1) {
3189       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3190       return 1;
3191     }
3192 
3193     if (odd) {
3194       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3195       words_pushed++;
3196     }
3197 
3198     for (int i = 2; i + 1 < count; i += 2) {
3199       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3200       words_pushed += 2;
3201     }
3202 
3203     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3204     words_pushed += 2;
3205 
3206     assert(words_pushed == count, "oops, pushed != count");
3207 
3208     return count;
3209   }
3210 
3211   return 0;
3212 }
3213 
3214 // Return the number of dwords pushed
3215 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3216   bool use_sve = false;
3217   int sve_predicate_size_in_slots = 0;
3218 
3219 #ifdef COMPILER2
3220   use_sve = Matcher::supports_scalable_vector();
3221   if (use_sve) {
3222     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3223   }
3224 #endif
3225 
3226   if (!use_sve) {
3227     return 0;
3228   }
3229 
3230   unsigned char regs[PRegister::number_of_registers];
3231   int count = 0;
3232   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3233     if (1 & bitset)
3234       regs[count++] = reg;
3235     bitset >>= 1;
3236   }
3237 
3238   if (count == 0) {
3239     return 0;
3240   }
3241 
3242   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3243                                   VMRegImpl::stack_slot_size * count, 16);
3244   sub(stack, stack, total_push_bytes);
3245   for (int i = 0; i < count; i++) {
3246     sve_str(as_PRegister(regs[i]), Address(stack, i));
3247   }
3248   return total_push_bytes / 8;
3249 }
3250 
3251 // Return the number of dwords popped
3252 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3253   bool use_sve = false;
3254   int sve_predicate_size_in_slots = 0;
3255 
3256 #ifdef COMPILER2
3257   use_sve = Matcher::supports_scalable_vector();
3258   if (use_sve) {
3259     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3260   }
3261 #endif
3262 
3263   if (!use_sve) {
3264     return 0;
3265   }
3266 
3267   unsigned char regs[PRegister::number_of_registers];
3268   int count = 0;
3269   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3270     if (1 & bitset)
3271       regs[count++] = reg;
3272     bitset >>= 1;
3273   }
3274 
3275   if (count == 0) {
3276     return 0;
3277   }
3278 
3279   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3280                                  VMRegImpl::stack_slot_size * count, 16);
3281   for (int i = count - 1; i >= 0; i--) {
3282     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3283   }
3284   add(stack, stack, total_pop_bytes);
3285   return total_pop_bytes / 8;
3286 }
3287 
3288 #ifdef ASSERT
3289 void MacroAssembler::verify_heapbase(const char* msg) {
3290 #if 0
3291   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3292   assert (Universe::heap() != nullptr, "java heap should be initialized");
3293   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3294     // rheapbase is allocated as general register
3295     return;
3296   }
3297   if (CheckCompressedOops) {
3298     Label ok;
3299     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3300     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3301     br(Assembler::EQ, ok);
3302     stop(msg);
3303     bind(ok);
3304     pop(1 << rscratch1->encoding(), sp);
3305   }
3306 #endif
3307 }
3308 #endif
3309 
3310 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3311   assert_different_registers(value, tmp1, tmp2);
3312   Label done, tagged, weak_tagged;
3313 
3314   cbz(value, done);           // Use null as-is.
3315   tst(value, JNIHandles::tag_mask); // Test for tag.
3316   br(Assembler::NE, tagged);
3317 
3318   // Resolve local handle
3319   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3320   verify_oop(value);
3321   b(done);
3322 
3323   bind(tagged);
3324   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3325   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3326 
3327   // Resolve global handle
3328   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3329   verify_oop(value);
3330   b(done);
3331 
3332   bind(weak_tagged);
3333   // Resolve jweak.
3334   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3335                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3336   verify_oop(value);
3337 
3338   bind(done);
3339 }
3340 
3341 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3342   assert_different_registers(value, tmp1, tmp2);
3343   Label done;
3344 
3345   cbz(value, done);           // Use null as-is.
3346 
3347 #ifdef ASSERT
3348   {
3349     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3350     Label valid_global_tag;
3351     tbnz(value, 1, valid_global_tag); // Test for global tag
3352     stop("non global jobject using resolve_global_jobject");
3353     bind(valid_global_tag);
3354   }
3355 #endif
3356 
3357   // Resolve global handle
3358   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3359   verify_oop(value);
3360 
3361   bind(done);
3362 }
3363 
3364 void MacroAssembler::stop(const char* msg) {
3365   // Skip AOT caching C strings in scratch buffer.
3366   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3367   BLOCK_COMMENT(str);
3368   // load msg into r0 so we can access it from the signal handler
3369   // ExternalAddress enables saving and restoring via the code cache
3370   lea(c_rarg0, ExternalAddress((address) str));
3371   dcps1(0xdeae);
3372 }
3373 
3374 void MacroAssembler::unimplemented(const char* what) {
3375   const char* buf = nullptr;
3376   {
3377     ResourceMark rm;
3378     stringStream ss;
3379     ss.print("unimplemented: %s", what);
3380     buf = code_string(ss.as_string());
3381   }
3382   stop(buf);
3383 }
3384 
3385 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3386 #ifdef ASSERT
3387   Label OK;
3388   br(cc, OK);
3389   stop(msg);
3390   bind(OK);
3391 #endif
3392 }
3393 
3394 // If a constant does not fit in an immediate field, generate some
3395 // number of MOV instructions and then perform the operation.
3396 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3397                                            add_sub_imm_insn insn1,
3398                                            add_sub_reg_insn insn2,
3399                                            bool is32) {
3400   assert(Rd != zr, "Rd = zr and not setting flags?");
3401   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3402   if (fits) {
3403     (this->*insn1)(Rd, Rn, imm);
3404   } else {
3405     if (g_uabs(imm) < (1 << 24)) {
3406        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3407        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3408     } else {
3409        assert_different_registers(Rd, Rn);
3410        mov(Rd, imm);
3411        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3412     }
3413   }
3414 }
3415 
3416 // Separate vsn which sets the flags. Optimisations are more restricted
3417 // because we must set the flags correctly.
3418 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3419                                              add_sub_imm_insn insn1,
3420                                              add_sub_reg_insn insn2,
3421                                              bool is32) {
3422   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3423   if (fits) {
3424     (this->*insn1)(Rd, Rn, imm);
3425   } else {
3426     assert_different_registers(Rd, Rn);
3427     assert(Rd != zr, "overflow in immediate operand");
3428     mov(Rd, imm);
3429     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3430   }
3431 }
3432 
3433 
3434 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3435   if (increment.is_register()) {
3436     add(Rd, Rn, increment.as_register());
3437   } else {
3438     add(Rd, Rn, increment.as_constant());
3439   }
3440 }
3441 
3442 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3443   if (increment.is_register()) {
3444     addw(Rd, Rn, increment.as_register());
3445   } else {
3446     addw(Rd, Rn, increment.as_constant());
3447   }
3448 }
3449 
3450 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3451   if (decrement.is_register()) {
3452     sub(Rd, Rn, decrement.as_register());
3453   } else {
3454     sub(Rd, Rn, decrement.as_constant());
3455   }
3456 }
3457 
3458 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3459   if (decrement.is_register()) {
3460     subw(Rd, Rn, decrement.as_register());
3461   } else {
3462     subw(Rd, Rn, decrement.as_constant());
3463   }
3464 }
3465 
3466 void MacroAssembler::reinit_heapbase()
3467 {
3468   if (UseCompressedOops) {
3469     if (Universe::is_fully_initialized()) {
3470       mov(rheapbase, CompressedOops::base());
3471     } else {
3472       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3473       ldr(rheapbase, Address(rheapbase));
3474     }
3475   }
3476 }
3477 
3478 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3479 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3480 // Pass a register for the result, otherwise pass noreg.
3481 
3482 // Clobbers rscratch1
3483 void MacroAssembler::cmpxchg(Register addr, Register expected,
3484                              Register new_val,
3485                              enum operand_size size,
3486                              bool acquire, bool release,
3487                              bool weak,
3488                              Register result) {
3489   if (result == noreg)  result = rscratch1;
3490   BLOCK_COMMENT("cmpxchg {");
3491   if (UseLSE) {
3492     mov(result, expected);
3493     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3494     compare_eq(result, expected, size);
3495 #ifdef ASSERT
3496     // Poison rscratch1 which is written on !UseLSE branch
3497     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3498 #endif
3499   } else {
3500     Label retry_load, done;
3501     prfm(Address(addr), PSTL1STRM);
3502     bind(retry_load);
3503     load_exclusive(result, addr, size, acquire);
3504     compare_eq(result, expected, size);
3505     br(Assembler::NE, done);
3506     store_exclusive(rscratch1, new_val, addr, size, release);
3507     if (weak) {
3508       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3509     } else {
3510       cbnzw(rscratch1, retry_load);
3511     }
3512     bind(done);
3513   }
3514   BLOCK_COMMENT("} cmpxchg");
3515 }
3516 
3517 // A generic comparison. Only compares for equality, clobbers rscratch1.
3518 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3519   if (size == xword) {
3520     cmp(rm, rn);
3521   } else if (size == word) {
3522     cmpw(rm, rn);
3523   } else if (size == halfword) {
3524     eorw(rscratch1, rm, rn);
3525     ands(zr, rscratch1, 0xffff);
3526   } else if (size == byte) {
3527     eorw(rscratch1, rm, rn);
3528     ands(zr, rscratch1, 0xff);
3529   } else {
3530     ShouldNotReachHere();
3531   }
3532 }
3533 
3534 
3535 static bool different(Register a, RegisterOrConstant b, Register c) {
3536   if (b.is_constant())
3537     return a != c;
3538   else
3539     return a != b.as_register() && a != c && b.as_register() != c;
3540 }
3541 
3542 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3543 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3544   if (UseLSE) {                                                         \
3545     prev = prev->is_valid() ? prev : zr;                                \
3546     if (incr.is_register()) {                                           \
3547       AOP(sz, incr.as_register(), prev, addr);                          \
3548     } else {                                                            \
3549       mov(rscratch2, incr.as_constant());                               \
3550       AOP(sz, rscratch2, prev, addr);                                   \
3551     }                                                                   \
3552     return;                                                             \
3553   }                                                                     \
3554   Register result = rscratch2;                                          \
3555   if (prev->is_valid())                                                 \
3556     result = different(prev, incr, addr) ? prev : rscratch2;            \
3557                                                                         \
3558   Label retry_load;                                                     \
3559   prfm(Address(addr), PSTL1STRM);                                       \
3560   bind(retry_load);                                                     \
3561   LDXR(result, addr);                                                   \
3562   OP(rscratch1, result, incr);                                          \
3563   STXR(rscratch2, rscratch1, addr);                                     \
3564   cbnzw(rscratch2, retry_load);                                         \
3565   if (prev->is_valid() && prev != result) {                             \
3566     IOP(prev, rscratch1, incr);                                         \
3567   }                                                                     \
3568 }
3569 
3570 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3571 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3572 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3573 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3574 
3575 #undef ATOMIC_OP
3576 
3577 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3578 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3579   if (UseLSE) {                                                         \
3580     prev = prev->is_valid() ? prev : zr;                                \
3581     AOP(sz, newv, prev, addr);                                          \
3582     return;                                                             \
3583   }                                                                     \
3584   Register result = rscratch2;                                          \
3585   if (prev->is_valid())                                                 \
3586     result = different(prev, newv, addr) ? prev : rscratch2;            \
3587                                                                         \
3588   Label retry_load;                                                     \
3589   prfm(Address(addr), PSTL1STRM);                                       \
3590   bind(retry_load);                                                     \
3591   LDXR(result, addr);                                                   \
3592   STXR(rscratch1, newv, addr);                                          \
3593   cbnzw(rscratch1, retry_load);                                         \
3594   if (prev->is_valid() && prev != result)                               \
3595     mov(prev, result);                                                  \
3596 }
3597 
3598 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3599 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3600 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3601 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3602 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3603 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3604 
3605 #undef ATOMIC_XCHG
3606 
3607 #ifndef PRODUCT
3608 extern "C" void findpc(intptr_t x);
3609 #endif
3610 
3611 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3612 {
3613   // In order to get locks to work, we need to fake a in_VM state
3614   if (ShowMessageBoxOnError ) {
3615     JavaThread* thread = JavaThread::current();
3616     JavaThreadState saved_state = thread->thread_state();
3617     thread->set_thread_state(_thread_in_vm);
3618 #ifndef PRODUCT
3619     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3620       ttyLocker ttyl;
3621       BytecodeCounter::print();
3622     }
3623 #endif
3624     if (os::message_box(msg, "Execution stopped, print registers?")) {
3625       ttyLocker ttyl;
3626       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3627 #ifndef PRODUCT
3628       tty->cr();
3629       findpc(pc);
3630       tty->cr();
3631 #endif
3632       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3633       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3634       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3635       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3636       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3637       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3638       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3639       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3640       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3641       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3642       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3643       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3644       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3645       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3646       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3647       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3648       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3649       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3650       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3651       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3652       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3653       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3654       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3655       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3656       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3657       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3658       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3659       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3660       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3661       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3662       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3663       BREAKPOINT;
3664     }
3665   }
3666   fatal("DEBUG MESSAGE: %s", msg);
3667 }
3668 
3669 RegSet MacroAssembler::call_clobbered_gp_registers() {
3670   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3671 #ifndef R18_RESERVED
3672   regs += r18_tls;
3673 #endif
3674   return regs;
3675 }
3676 
3677 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3678   int step = 4 * wordSize;
3679   push(call_clobbered_gp_registers() - exclude, sp);
3680   sub(sp, sp, step);
3681   mov(rscratch1, -step);
3682   // Push v0-v7, v16-v31.
3683   for (int i = 31; i>= 4; i -= 4) {
3684     if (i <= v7->encoding() || i >= v16->encoding())
3685       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3686           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3687   }
3688   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3689       as_FloatRegister(3), T1D, Address(sp));
3690 }
3691 
3692 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3693   for (int i = 0; i < 32; i += 4) {
3694     if (i <= v7->encoding() || i >= v16->encoding())
3695       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3696           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3697   }
3698 
3699   reinitialize_ptrue();
3700 
3701   pop(call_clobbered_gp_registers() - exclude, sp);
3702 }
3703 
3704 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3705                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3706   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3707   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3708     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3709     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3710       sve_str(as_FloatRegister(i), Address(sp, i));
3711     }
3712   } else {
3713     int step = (save_vectors ? 8 : 4) * wordSize;
3714     mov(rscratch1, -step);
3715     sub(sp, sp, step);
3716     for (int i = 28; i >= 4; i -= 4) {
3717       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3718           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3719     }
3720     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3721   }
3722   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3723     sub(sp, sp, total_predicate_in_bytes);
3724     for (int i = 0; i < PRegister::number_of_registers; i++) {
3725       sve_str(as_PRegister(i), Address(sp, i));
3726     }
3727   }
3728 }
3729 
3730 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3731                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3732   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3733     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3734       sve_ldr(as_PRegister(i), Address(sp, i));
3735     }
3736     add(sp, sp, total_predicate_in_bytes);
3737   }
3738   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3739     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3740       sve_ldr(as_FloatRegister(i), Address(sp, i));
3741     }
3742     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3743   } else {
3744     int step = (restore_vectors ? 8 : 4) * wordSize;
3745     for (int i = 0; i <= 28; i += 4)
3746       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3747           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3748   }
3749 
3750   // We may use predicate registers and rely on ptrue with SVE,
3751   // regardless of wide vector (> 8 bytes) used or not.
3752   if (use_sve) {
3753     reinitialize_ptrue();
3754   }
3755 
3756   // integer registers except lr & sp
3757   pop(RegSet::range(r0, r17), sp);
3758 #ifdef R18_RESERVED
3759   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3760   pop(RegSet::range(r20, r29), sp);
3761 #else
3762   pop(RegSet::range(r18_tls, r29), sp);
3763 #endif
3764 }
3765 
3766 /**
3767  * Helpers for multiply_to_len().
3768  */
3769 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3770                                      Register src1, Register src2) {
3771   adds(dest_lo, dest_lo, src1);
3772   adc(dest_hi, dest_hi, zr);
3773   adds(dest_lo, dest_lo, src2);
3774   adc(final_dest_hi, dest_hi, zr);
3775 }
3776 
3777 // Generate an address from (r + r1 extend offset).  "size" is the
3778 // size of the operand.  The result may be in rscratch2.
3779 Address MacroAssembler::offsetted_address(Register r, Register r1,
3780                                           Address::extend ext, int offset, int size) {
3781   if (offset || (ext.shift() % size != 0)) {
3782     lea(rscratch2, Address(r, r1, ext));
3783     return Address(rscratch2, offset);
3784   } else {
3785     return Address(r, r1, ext);
3786   }
3787 }
3788 
3789 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3790 {
3791   assert(offset >= 0, "spill to negative address?");
3792   // Offset reachable ?
3793   //   Not aligned - 9 bits signed offset
3794   //   Aligned - 12 bits unsigned offset shifted
3795   Register base = sp;
3796   if ((offset & (size-1)) && offset >= (1<<8)) {
3797     add(tmp, base, offset & ((1<<12)-1));
3798     base = tmp;
3799     offset &= -1u<<12;
3800   }
3801 
3802   if (offset >= (1<<12) * size) {
3803     add(tmp, base, offset & (((1<<12)-1)<<12));
3804     base = tmp;
3805     offset &= ~(((1<<12)-1)<<12);
3806   }
3807 
3808   return Address(base, offset);
3809 }
3810 
3811 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3812   assert(offset >= 0, "spill to negative address?");
3813 
3814   Register base = sp;
3815 
3816   // An immediate offset in the range 0 to 255 which is multiplied
3817   // by the current vector or predicate register size in bytes.
3818   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3819     return Address(base, offset / sve_reg_size_in_bytes);
3820   }
3821 
3822   add(tmp, base, offset);
3823   return Address(tmp);
3824 }
3825 
3826 // Checks whether offset is aligned.
3827 // Returns true if it is, else false.
3828 bool MacroAssembler::merge_alignment_check(Register base,
3829                                            size_t size,
3830                                            int64_t cur_offset,
3831                                            int64_t prev_offset) const {
3832   if (AvoidUnalignedAccesses) {
3833     if (base == sp) {
3834       // Checks whether low offset if aligned to pair of registers.
3835       int64_t pair_mask = size * 2 - 1;
3836       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3837       return (offset & pair_mask) == 0;
3838     } else { // If base is not sp, we can't guarantee the access is aligned.
3839       return false;
3840     }
3841   } else {
3842     int64_t mask = size - 1;
3843     // Load/store pair instruction only supports element size aligned offset.
3844     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3845   }
3846 }
3847 
3848 // Checks whether current and previous loads/stores can be merged.
3849 // Returns true if it can be merged, else false.
3850 bool MacroAssembler::ldst_can_merge(Register rt,
3851                                     const Address &adr,
3852                                     size_t cur_size_in_bytes,
3853                                     bool is_store) const {
3854   address prev = pc() - NativeInstruction::instruction_size;
3855   address last = code()->last_insn();
3856 
3857   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3858     return false;
3859   }
3860 
3861   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3862     return false;
3863   }
3864 
3865   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3866   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3867 
3868   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3869   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3870 
3871   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3872     return false;
3873   }
3874 
3875   int64_t max_offset = 63 * prev_size_in_bytes;
3876   int64_t min_offset = -64 * prev_size_in_bytes;
3877 
3878   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3879 
3880   // Only same base can be merged.
3881   if (adr.base() != prev_ldst->base()) {
3882     return false;
3883   }
3884 
3885   int64_t cur_offset = adr.offset();
3886   int64_t prev_offset = prev_ldst->offset();
3887   size_t diff = abs(cur_offset - prev_offset);
3888   if (diff != prev_size_in_bytes) {
3889     return false;
3890   }
3891 
3892   // Following cases can not be merged:
3893   // ldr x2, [x2, #8]
3894   // ldr x3, [x2, #16]
3895   // or:
3896   // ldr x2, [x3, #8]
3897   // ldr x2, [x3, #16]
3898   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3899   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3900     return false;
3901   }
3902 
3903   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3904   // Offset range must be in ldp/stp instruction's range.
3905   if (low_offset > max_offset || low_offset < min_offset) {
3906     return false;
3907   }
3908 
3909   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3910     return true;
3911   }
3912 
3913   return false;
3914 }
3915 
3916 // Merge current load/store with previous load/store into ldp/stp.
3917 void MacroAssembler::merge_ldst(Register rt,
3918                                 const Address &adr,
3919                                 size_t cur_size_in_bytes,
3920                                 bool is_store) {
3921 
3922   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3923 
3924   Register rt_low, rt_high;
3925   address prev = pc() - NativeInstruction::instruction_size;
3926   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3927 
3928   int64_t offset;
3929 
3930   if (adr.offset() < prev_ldst->offset()) {
3931     offset = adr.offset();
3932     rt_low = rt;
3933     rt_high = prev_ldst->target();
3934   } else {
3935     offset = prev_ldst->offset();
3936     rt_low = prev_ldst->target();
3937     rt_high = rt;
3938   }
3939 
3940   Address adr_p = Address(prev_ldst->base(), offset);
3941   // Overwrite previous generated binary.
3942   code_section()->set_end(prev);
3943 
3944   const size_t sz = prev_ldst->size_in_bytes();
3945   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3946   if (!is_store) {
3947     BLOCK_COMMENT("merged ldr pair");
3948     if (sz == 8) {
3949       ldp(rt_low, rt_high, adr_p);
3950     } else {
3951       ldpw(rt_low, rt_high, adr_p);
3952     }
3953   } else {
3954     BLOCK_COMMENT("merged str pair");
3955     if (sz == 8) {
3956       stp(rt_low, rt_high, adr_p);
3957     } else {
3958       stpw(rt_low, rt_high, adr_p);
3959     }
3960   }
3961 }
3962 
3963 /**
3964  * Multiply 64 bit by 64 bit first loop.
3965  */
3966 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3967                                            Register y, Register y_idx, Register z,
3968                                            Register carry, Register product,
3969                                            Register idx, Register kdx) {
3970   //
3971   //  jlong carry, x[], y[], z[];
3972   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3973   //    huge_128 product = y[idx] * x[xstart] + carry;
3974   //    z[kdx] = (jlong)product;
3975   //    carry  = (jlong)(product >>> 64);
3976   //  }
3977   //  z[xstart] = carry;
3978   //
3979 
3980   Label L_first_loop, L_first_loop_exit;
3981   Label L_one_x, L_one_y, L_multiply;
3982 
3983   subsw(xstart, xstart, 1);
3984   br(Assembler::MI, L_one_x);
3985 
3986   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3987   ldr(x_xstart, Address(rscratch1));
3988   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3989 
3990   bind(L_first_loop);
3991   subsw(idx, idx, 1);
3992   br(Assembler::MI, L_first_loop_exit);
3993   subsw(idx, idx, 1);
3994   br(Assembler::MI, L_one_y);
3995   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3996   ldr(y_idx, Address(rscratch1));
3997   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
3998   bind(L_multiply);
3999 
4000   // AArch64 has a multiply-accumulate instruction that we can't use
4001   // here because it has no way to process carries, so we have to use
4002   // separate add and adc instructions.  Bah.
4003   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
4004   mul(product, x_xstart, y_idx);
4005   adds(product, product, carry);
4006   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
4007 
4008   subw(kdx, kdx, 2);
4009   ror(product, product, 32); // back to big-endian
4010   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4011 
4012   b(L_first_loop);
4013 
4014   bind(L_one_y);
4015   ldrw(y_idx, Address(y,  0));
4016   b(L_multiply);
4017 
4018   bind(L_one_x);
4019   ldrw(x_xstart, Address(x,  0));
4020   b(L_first_loop);
4021 
4022   bind(L_first_loop_exit);
4023 }
4024 
4025 /**
4026  * Multiply 128 bit by 128. Unrolled inner loop.
4027  *
4028  */
4029 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4030                                              Register carry, Register carry2,
4031                                              Register idx, Register jdx,
4032                                              Register yz_idx1, Register yz_idx2,
4033                                              Register tmp, Register tmp3, Register tmp4,
4034                                              Register tmp6, Register product_hi) {
4035 
4036   //   jlong carry, x[], y[], z[];
4037   //   int kdx = ystart+1;
4038   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4039   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4040   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4041   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4042   //     carry  = (jlong)(tmp4 >>> 64);
4043   //     z[kdx+idx+1] = (jlong)tmp3;
4044   //     z[kdx+idx] = (jlong)tmp4;
4045   //   }
4046   //   idx += 2;
4047   //   if (idx > 0) {
4048   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4049   //     z[kdx+idx] = (jlong)yz_idx1;
4050   //     carry  = (jlong)(yz_idx1 >>> 64);
4051   //   }
4052   //
4053 
4054   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4055 
4056   lsrw(jdx, idx, 2);
4057 
4058   bind(L_third_loop);
4059 
4060   subsw(jdx, jdx, 1);
4061   br(Assembler::MI, L_third_loop_exit);
4062   subw(idx, idx, 4);
4063 
4064   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4065 
4066   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4067 
4068   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4069 
4070   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4071   ror(yz_idx2, yz_idx2, 32);
4072 
4073   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4074 
4075   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4076   umulh(tmp4, product_hi, yz_idx1);
4077 
4078   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4079   ror(rscratch2, rscratch2, 32);
4080 
4081   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4082   umulh(carry2, product_hi, yz_idx2);
4083 
4084   // propagate sum of both multiplications into carry:tmp4:tmp3
4085   adds(tmp3, tmp3, carry);
4086   adc(tmp4, tmp4, zr);
4087   adds(tmp3, tmp3, rscratch1);
4088   adcs(tmp4, tmp4, tmp);
4089   adc(carry, carry2, zr);
4090   adds(tmp4, tmp4, rscratch2);
4091   adc(carry, carry, zr);
4092 
4093   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4094   ror(tmp4, tmp4, 32);
4095   stp(tmp4, tmp3, Address(tmp6, 0));
4096 
4097   b(L_third_loop);
4098   bind (L_third_loop_exit);
4099 
4100   andw (idx, idx, 0x3);
4101   cbz(idx, L_post_third_loop_done);
4102 
4103   Label L_check_1;
4104   subsw(idx, idx, 2);
4105   br(Assembler::MI, L_check_1);
4106 
4107   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4108   ldr(yz_idx1, Address(rscratch1, 0));
4109   ror(yz_idx1, yz_idx1, 32);
4110   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4111   umulh(tmp4, product_hi, yz_idx1);
4112   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4113   ldr(yz_idx2, Address(rscratch1, 0));
4114   ror(yz_idx2, yz_idx2, 32);
4115 
4116   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4117 
4118   ror(tmp3, tmp3, 32);
4119   str(tmp3, Address(rscratch1, 0));
4120 
4121   bind (L_check_1);
4122 
4123   andw (idx, idx, 0x1);
4124   subsw(idx, idx, 1);
4125   br(Assembler::MI, L_post_third_loop_done);
4126   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4127   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4128   umulh(carry2, tmp4, product_hi);
4129   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4130 
4131   add2_with_carry(carry2, tmp3, tmp4, carry);
4132 
4133   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4134   extr(carry, carry2, tmp3, 32);
4135 
4136   bind(L_post_third_loop_done);
4137 }
4138 
4139 /**
4140  * Code for BigInteger::multiplyToLen() intrinsic.
4141  *
4142  * r0: x
4143  * r1: xlen
4144  * r2: y
4145  * r3: ylen
4146  * r4:  z
4147  * r5: tmp0
4148  * r10: tmp1
4149  * r11: tmp2
4150  * r12: tmp3
4151  * r13: tmp4
4152  * r14: tmp5
4153  * r15: tmp6
4154  * r16: tmp7
4155  *
4156  */
4157 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4158                                      Register z, Register tmp0,
4159                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4160                                      Register tmp5, Register tmp6, Register product_hi) {
4161 
4162   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4163 
4164   const Register idx = tmp1;
4165   const Register kdx = tmp2;
4166   const Register xstart = tmp3;
4167 
4168   const Register y_idx = tmp4;
4169   const Register carry = tmp5;
4170   const Register product  = xlen;
4171   const Register x_xstart = tmp0;
4172 
4173   // First Loop.
4174   //
4175   //  final static long LONG_MASK = 0xffffffffL;
4176   //  int xstart = xlen - 1;
4177   //  int ystart = ylen - 1;
4178   //  long carry = 0;
4179   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4180   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4181   //    z[kdx] = (int)product;
4182   //    carry = product >>> 32;
4183   //  }
4184   //  z[xstart] = (int)carry;
4185   //
4186 
4187   movw(idx, ylen);       // idx = ylen;
4188   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4189   mov(carry, zr);        // carry = 0;
4190 
4191   Label L_done;
4192 
4193   movw(xstart, xlen);
4194   subsw(xstart, xstart, 1);
4195   br(Assembler::MI, L_done);
4196 
4197   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4198 
4199   Label L_second_loop;
4200   cbzw(kdx, L_second_loop);
4201 
4202   Label L_carry;
4203   subw(kdx, kdx, 1);
4204   cbzw(kdx, L_carry);
4205 
4206   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4207   lsr(carry, carry, 32);
4208   subw(kdx, kdx, 1);
4209 
4210   bind(L_carry);
4211   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4212 
4213   // Second and third (nested) loops.
4214   //
4215   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4216   //   carry = 0;
4217   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4218   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4219   //                    (z[k] & LONG_MASK) + carry;
4220   //     z[k] = (int)product;
4221   //     carry = product >>> 32;
4222   //   }
4223   //   z[i] = (int)carry;
4224   // }
4225   //
4226   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4227 
4228   const Register jdx = tmp1;
4229 
4230   bind(L_second_loop);
4231   mov(carry, zr);                // carry = 0;
4232   movw(jdx, ylen);               // j = ystart+1
4233 
4234   subsw(xstart, xstart, 1);      // i = xstart-1;
4235   br(Assembler::MI, L_done);
4236 
4237   str(z, Address(pre(sp, -4 * wordSize)));
4238 
4239   Label L_last_x;
4240   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4241   subsw(xstart, xstart, 1);       // i = xstart-1;
4242   br(Assembler::MI, L_last_x);
4243 
4244   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4245   ldr(product_hi, Address(rscratch1));
4246   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4247 
4248   Label L_third_loop_prologue;
4249   bind(L_third_loop_prologue);
4250 
4251   str(ylen, Address(sp, wordSize));
4252   stp(x, xstart, Address(sp, 2 * wordSize));
4253   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4254                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4255   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4256   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4257 
4258   addw(tmp3, xlen, 1);
4259   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4260   subsw(tmp3, tmp3, 1);
4261   br(Assembler::MI, L_done);
4262 
4263   lsr(carry, carry, 32);
4264   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4265   b(L_second_loop);
4266 
4267   // Next infrequent code is moved outside loops.
4268   bind(L_last_x);
4269   ldrw(product_hi, Address(x,  0));
4270   b(L_third_loop_prologue);
4271 
4272   bind(L_done);
4273 }
4274 
4275 // Code for BigInteger::mulAdd intrinsic
4276 // out     = r0
4277 // in      = r1
4278 // offset  = r2  (already out.length-offset)
4279 // len     = r3
4280 // k       = r4
4281 //
4282 // pseudo code from java implementation:
4283 // carry = 0;
4284 // offset = out.length-offset - 1;
4285 // for (int j=len-1; j >= 0; j--) {
4286 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4287 //     out[offset--] = (int)product;
4288 //     carry = product >>> 32;
4289 // }
4290 // return (int)carry;
4291 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4292       Register len, Register k) {
4293     Label LOOP, END;
4294     // pre-loop
4295     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4296     csel(out, zr, out, Assembler::EQ);
4297     br(Assembler::EQ, END);
4298     add(in, in, len, LSL, 2); // in[j+1] address
4299     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4300     mov(out, zr); // used to keep carry now
4301     BIND(LOOP);
4302     ldrw(rscratch1, Address(pre(in, -4)));
4303     madd(rscratch1, rscratch1, k, out);
4304     ldrw(rscratch2, Address(pre(offset, -4)));
4305     add(rscratch1, rscratch1, rscratch2);
4306     strw(rscratch1, Address(offset));
4307     lsr(out, rscratch1, 32);
4308     subs(len, len, 1);
4309     br(Assembler::NE, LOOP);
4310     BIND(END);
4311 }
4312 
4313 /**
4314  * Emits code to update CRC-32 with a byte value according to constants in table
4315  *
4316  * @param [in,out]crc   Register containing the crc.
4317  * @param [in]val       Register containing the byte to fold into the CRC.
4318  * @param [in]table     Register containing the table of crc constants.
4319  *
4320  * uint32_t crc;
4321  * val = crc_table[(val ^ crc) & 0xFF];
4322  * crc = val ^ (crc >> 8);
4323  *
4324  */
4325 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4326   eor(val, val, crc);
4327   andr(val, val, 0xff);
4328   ldrw(val, Address(table, val, Address::lsl(2)));
4329   eor(crc, val, crc, Assembler::LSR, 8);
4330 }
4331 
4332 /**
4333  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4334  *
4335  * @param [in,out]crc   Register containing the crc.
4336  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4337  * @param [in]table0    Register containing table 0 of crc constants.
4338  * @param [in]table1    Register containing table 1 of crc constants.
4339  * @param [in]table2    Register containing table 2 of crc constants.
4340  * @param [in]table3    Register containing table 3 of crc constants.
4341  *
4342  * uint32_t crc;
4343  *   v = crc ^ v
4344  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4345  *
4346  */
4347 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4348         Register table0, Register table1, Register table2, Register table3,
4349         bool upper) {
4350   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4351   uxtb(tmp, v);
4352   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4353   ubfx(tmp, v, 8, 8);
4354   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4355   eor(crc, crc, tmp);
4356   ubfx(tmp, v, 16, 8);
4357   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4358   eor(crc, crc, tmp);
4359   ubfx(tmp, v, 24, 8);
4360   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4361   eor(crc, crc, tmp);
4362 }
4363 
4364 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4365         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4366     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4367     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4368 
4369     subs(tmp0, len, 384);
4370     mvnw(crc, crc);
4371     br(Assembler::GE, CRC_by128_pre);
4372   BIND(CRC_less128);
4373     subs(len, len, 32);
4374     br(Assembler::GE, CRC_by32_loop);
4375   BIND(CRC_less32);
4376     adds(len, len, 32 - 4);
4377     br(Assembler::GE, CRC_by4_loop);
4378     adds(len, len, 4);
4379     br(Assembler::GT, CRC_by1_loop);
4380     b(L_exit);
4381 
4382   BIND(CRC_by32_loop);
4383     ldp(tmp0, tmp1, Address(buf));
4384     crc32x(crc, crc, tmp0);
4385     ldp(tmp2, tmp3, Address(buf, 16));
4386     crc32x(crc, crc, tmp1);
4387     add(buf, buf, 32);
4388     crc32x(crc, crc, tmp2);
4389     subs(len, len, 32);
4390     crc32x(crc, crc, tmp3);
4391     br(Assembler::GE, CRC_by32_loop);
4392     cmn(len, (u1)32);
4393     br(Assembler::NE, CRC_less32);
4394     b(L_exit);
4395 
4396   BIND(CRC_by4_loop);
4397     ldrw(tmp0, Address(post(buf, 4)));
4398     subs(len, len, 4);
4399     crc32w(crc, crc, tmp0);
4400     br(Assembler::GE, CRC_by4_loop);
4401     adds(len, len, 4);
4402     br(Assembler::LE, L_exit);
4403   BIND(CRC_by1_loop);
4404     ldrb(tmp0, Address(post(buf, 1)));
4405     subs(len, len, 1);
4406     crc32b(crc, crc, tmp0);
4407     br(Assembler::GT, CRC_by1_loop);
4408     b(L_exit);
4409 
4410   BIND(CRC_by128_pre);
4411     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4412       4*256*sizeof(juint) + 8*sizeof(juint));
4413     mov(crc, 0);
4414     crc32x(crc, crc, tmp0);
4415     crc32x(crc, crc, tmp1);
4416 
4417     cbnz(len, CRC_less128);
4418 
4419   BIND(L_exit);
4420     mvnw(crc, crc);
4421 }
4422 
4423 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4424         Register len, Register tmp0, Register tmp1, Register tmp2,
4425         Register tmp3) {
4426     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4427     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4428 
4429     mvnw(crc, crc);
4430 
4431     subs(len, len, 128);
4432     br(Assembler::GE, CRC_by64_pre);
4433   BIND(CRC_less64);
4434     adds(len, len, 128-32);
4435     br(Assembler::GE, CRC_by32_loop);
4436   BIND(CRC_less32);
4437     adds(len, len, 32-4);
4438     br(Assembler::GE, CRC_by4_loop);
4439     adds(len, len, 4);
4440     br(Assembler::GT, CRC_by1_loop);
4441     b(L_exit);
4442 
4443   BIND(CRC_by32_loop);
4444     ldp(tmp0, tmp1, Address(post(buf, 16)));
4445     subs(len, len, 32);
4446     crc32x(crc, crc, tmp0);
4447     ldr(tmp2, Address(post(buf, 8)));
4448     crc32x(crc, crc, tmp1);
4449     ldr(tmp3, Address(post(buf, 8)));
4450     crc32x(crc, crc, tmp2);
4451     crc32x(crc, crc, tmp3);
4452     br(Assembler::GE, CRC_by32_loop);
4453     cmn(len, (u1)32);
4454     br(Assembler::NE, CRC_less32);
4455     b(L_exit);
4456 
4457   BIND(CRC_by4_loop);
4458     ldrw(tmp0, Address(post(buf, 4)));
4459     subs(len, len, 4);
4460     crc32w(crc, crc, tmp0);
4461     br(Assembler::GE, CRC_by4_loop);
4462     adds(len, len, 4);
4463     br(Assembler::LE, L_exit);
4464   BIND(CRC_by1_loop);
4465     ldrb(tmp0, Address(post(buf, 1)));
4466     subs(len, len, 1);
4467     crc32b(crc, crc, tmp0);
4468     br(Assembler::GT, CRC_by1_loop);
4469     b(L_exit);
4470 
4471   BIND(CRC_by64_pre);
4472     sub(buf, buf, 8);
4473     ldp(tmp0, tmp1, Address(buf, 8));
4474     crc32x(crc, crc, tmp0);
4475     ldr(tmp2, Address(buf, 24));
4476     crc32x(crc, crc, tmp1);
4477     ldr(tmp3, Address(buf, 32));
4478     crc32x(crc, crc, tmp2);
4479     ldr(tmp0, Address(buf, 40));
4480     crc32x(crc, crc, tmp3);
4481     ldr(tmp1, Address(buf, 48));
4482     crc32x(crc, crc, tmp0);
4483     ldr(tmp2, Address(buf, 56));
4484     crc32x(crc, crc, tmp1);
4485     ldr(tmp3, Address(pre(buf, 64)));
4486 
4487     b(CRC_by64_loop);
4488 
4489     align(CodeEntryAlignment);
4490   BIND(CRC_by64_loop);
4491     subs(len, len, 64);
4492     crc32x(crc, crc, tmp2);
4493     ldr(tmp0, Address(buf, 8));
4494     crc32x(crc, crc, tmp3);
4495     ldr(tmp1, Address(buf, 16));
4496     crc32x(crc, crc, tmp0);
4497     ldr(tmp2, Address(buf, 24));
4498     crc32x(crc, crc, tmp1);
4499     ldr(tmp3, Address(buf, 32));
4500     crc32x(crc, crc, tmp2);
4501     ldr(tmp0, Address(buf, 40));
4502     crc32x(crc, crc, tmp3);
4503     ldr(tmp1, Address(buf, 48));
4504     crc32x(crc, crc, tmp0);
4505     ldr(tmp2, Address(buf, 56));
4506     crc32x(crc, crc, tmp1);
4507     ldr(tmp3, Address(pre(buf, 64)));
4508     br(Assembler::GE, CRC_by64_loop);
4509 
4510     // post-loop
4511     crc32x(crc, crc, tmp2);
4512     crc32x(crc, crc, tmp3);
4513 
4514     sub(len, len, 64);
4515     add(buf, buf, 8);
4516     cmn(len, (u1)128);
4517     br(Assembler::NE, CRC_less64);
4518   BIND(L_exit);
4519     mvnw(crc, crc);
4520 }
4521 
4522 /**
4523  * @param crc   register containing existing CRC (32-bit)
4524  * @param buf   register pointing to input byte buffer (byte*)
4525  * @param len   register containing number of bytes
4526  * @param table register that will contain address of CRC table
4527  * @param tmp   scratch register
4528  */
4529 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4530         Register table0, Register table1, Register table2, Register table3,
4531         Register tmp, Register tmp2, Register tmp3) {
4532   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4533 
4534   if (UseCryptoPmullForCRC32) {
4535       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4536       return;
4537   }
4538 
4539   if (UseCRC32) {
4540       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4541       return;
4542   }
4543 
4544     mvnw(crc, crc);
4545 
4546     {
4547       uint64_t offset;
4548       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4549       add(table0, table0, offset);
4550     }
4551     add(table1, table0, 1*256*sizeof(juint));
4552     add(table2, table0, 2*256*sizeof(juint));
4553     add(table3, table0, 3*256*sizeof(juint));
4554 
4555     { // Neon code start
4556       cmp(len, (u1)64);
4557       br(Assembler::LT, L_by16);
4558       eor(v16, T16B, v16, v16);
4559 
4560     Label L_fold;
4561 
4562       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4563 
4564       ld1(v0, v1, T2D, post(buf, 32));
4565       ld1r(v4, T2D, post(tmp, 8));
4566       ld1r(v5, T2D, post(tmp, 8));
4567       ld1r(v6, T2D, post(tmp, 8));
4568       ld1r(v7, T2D, post(tmp, 8));
4569       mov(v16, S, 0, crc);
4570 
4571       eor(v0, T16B, v0, v16);
4572       sub(len, len, 64);
4573 
4574     BIND(L_fold);
4575       pmull(v22, T8H, v0, v5, T8B);
4576       pmull(v20, T8H, v0, v7, T8B);
4577       pmull(v23, T8H, v0, v4, T8B);
4578       pmull(v21, T8H, v0, v6, T8B);
4579 
4580       pmull2(v18, T8H, v0, v5, T16B);
4581       pmull2(v16, T8H, v0, v7, T16B);
4582       pmull2(v19, T8H, v0, v4, T16B);
4583       pmull2(v17, T8H, v0, v6, T16B);
4584 
4585       uzp1(v24, T8H, v20, v22);
4586       uzp2(v25, T8H, v20, v22);
4587       eor(v20, T16B, v24, v25);
4588 
4589       uzp1(v26, T8H, v16, v18);
4590       uzp2(v27, T8H, v16, v18);
4591       eor(v16, T16B, v26, v27);
4592 
4593       ushll2(v22, T4S, v20, T8H, 8);
4594       ushll(v20, T4S, v20, T4H, 8);
4595 
4596       ushll2(v18, T4S, v16, T8H, 8);
4597       ushll(v16, T4S, v16, T4H, 8);
4598 
4599       eor(v22, T16B, v23, v22);
4600       eor(v18, T16B, v19, v18);
4601       eor(v20, T16B, v21, v20);
4602       eor(v16, T16B, v17, v16);
4603 
4604       uzp1(v17, T2D, v16, v20);
4605       uzp2(v21, T2D, v16, v20);
4606       eor(v17, T16B, v17, v21);
4607 
4608       ushll2(v20, T2D, v17, T4S, 16);
4609       ushll(v16, T2D, v17, T2S, 16);
4610 
4611       eor(v20, T16B, v20, v22);
4612       eor(v16, T16B, v16, v18);
4613 
4614       uzp1(v17, T2D, v20, v16);
4615       uzp2(v21, T2D, v20, v16);
4616       eor(v28, T16B, v17, v21);
4617 
4618       pmull(v22, T8H, v1, v5, T8B);
4619       pmull(v20, T8H, v1, v7, T8B);
4620       pmull(v23, T8H, v1, v4, T8B);
4621       pmull(v21, T8H, v1, v6, T8B);
4622 
4623       pmull2(v18, T8H, v1, v5, T16B);
4624       pmull2(v16, T8H, v1, v7, T16B);
4625       pmull2(v19, T8H, v1, v4, T16B);
4626       pmull2(v17, T8H, v1, v6, T16B);
4627 
4628       ld1(v0, v1, T2D, post(buf, 32));
4629 
4630       uzp1(v24, T8H, v20, v22);
4631       uzp2(v25, T8H, v20, v22);
4632       eor(v20, T16B, v24, v25);
4633 
4634       uzp1(v26, T8H, v16, v18);
4635       uzp2(v27, T8H, v16, v18);
4636       eor(v16, T16B, v26, v27);
4637 
4638       ushll2(v22, T4S, v20, T8H, 8);
4639       ushll(v20, T4S, v20, T4H, 8);
4640 
4641       ushll2(v18, T4S, v16, T8H, 8);
4642       ushll(v16, T4S, v16, T4H, 8);
4643 
4644       eor(v22, T16B, v23, v22);
4645       eor(v18, T16B, v19, v18);
4646       eor(v20, T16B, v21, v20);
4647       eor(v16, T16B, v17, v16);
4648 
4649       uzp1(v17, T2D, v16, v20);
4650       uzp2(v21, T2D, v16, v20);
4651       eor(v16, T16B, v17, v21);
4652 
4653       ushll2(v20, T2D, v16, T4S, 16);
4654       ushll(v16, T2D, v16, T2S, 16);
4655 
4656       eor(v20, T16B, v22, v20);
4657       eor(v16, T16B, v16, v18);
4658 
4659       uzp1(v17, T2D, v20, v16);
4660       uzp2(v21, T2D, v20, v16);
4661       eor(v20, T16B, v17, v21);
4662 
4663       shl(v16, T2D, v28, 1);
4664       shl(v17, T2D, v20, 1);
4665 
4666       eor(v0, T16B, v0, v16);
4667       eor(v1, T16B, v1, v17);
4668 
4669       subs(len, len, 32);
4670       br(Assembler::GE, L_fold);
4671 
4672       mov(crc, 0);
4673       mov(tmp, v0, D, 0);
4674       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4675       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4676       mov(tmp, v0, D, 1);
4677       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4678       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4679       mov(tmp, v1, D, 0);
4680       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4681       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4682       mov(tmp, v1, D, 1);
4683       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4684       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4685 
4686       add(len, len, 32);
4687     } // Neon code end
4688 
4689   BIND(L_by16);
4690     subs(len, len, 16);
4691     br(Assembler::GE, L_by16_loop);
4692     adds(len, len, 16-4);
4693     br(Assembler::GE, L_by4_loop);
4694     adds(len, len, 4);
4695     br(Assembler::GT, L_by1_loop);
4696     b(L_exit);
4697 
4698   BIND(L_by4_loop);
4699     ldrw(tmp, Address(post(buf, 4)));
4700     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4701     subs(len, len, 4);
4702     br(Assembler::GE, L_by4_loop);
4703     adds(len, len, 4);
4704     br(Assembler::LE, L_exit);
4705   BIND(L_by1_loop);
4706     subs(len, len, 1);
4707     ldrb(tmp, Address(post(buf, 1)));
4708     update_byte_crc32(crc, tmp, table0);
4709     br(Assembler::GT, L_by1_loop);
4710     b(L_exit);
4711 
4712     align(CodeEntryAlignment);
4713   BIND(L_by16_loop);
4714     subs(len, len, 16);
4715     ldp(tmp, tmp3, Address(post(buf, 16)));
4716     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4717     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4718     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4719     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4720     br(Assembler::GE, L_by16_loop);
4721     adds(len, len, 16-4);
4722     br(Assembler::GE, L_by4_loop);
4723     adds(len, len, 4);
4724     br(Assembler::GT, L_by1_loop);
4725   BIND(L_exit);
4726     mvnw(crc, crc);
4727 }
4728 
4729 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4730         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4731     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4732     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4733 
4734     subs(tmp0, len, 384);
4735     br(Assembler::GE, CRC_by128_pre);
4736   BIND(CRC_less128);
4737     subs(len, len, 32);
4738     br(Assembler::GE, CRC_by32_loop);
4739   BIND(CRC_less32);
4740     adds(len, len, 32 - 4);
4741     br(Assembler::GE, CRC_by4_loop);
4742     adds(len, len, 4);
4743     br(Assembler::GT, CRC_by1_loop);
4744     b(L_exit);
4745 
4746   BIND(CRC_by32_loop);
4747     ldp(tmp0, tmp1, Address(buf));
4748     crc32cx(crc, crc, tmp0);
4749     ldr(tmp2, Address(buf, 16));
4750     crc32cx(crc, crc, tmp1);
4751     ldr(tmp3, Address(buf, 24));
4752     crc32cx(crc, crc, tmp2);
4753     add(buf, buf, 32);
4754     subs(len, len, 32);
4755     crc32cx(crc, crc, tmp3);
4756     br(Assembler::GE, CRC_by32_loop);
4757     cmn(len, (u1)32);
4758     br(Assembler::NE, CRC_less32);
4759     b(L_exit);
4760 
4761   BIND(CRC_by4_loop);
4762     ldrw(tmp0, Address(post(buf, 4)));
4763     subs(len, len, 4);
4764     crc32cw(crc, crc, tmp0);
4765     br(Assembler::GE, CRC_by4_loop);
4766     adds(len, len, 4);
4767     br(Assembler::LE, L_exit);
4768   BIND(CRC_by1_loop);
4769     ldrb(tmp0, Address(post(buf, 1)));
4770     subs(len, len, 1);
4771     crc32cb(crc, crc, tmp0);
4772     br(Assembler::GT, CRC_by1_loop);
4773     b(L_exit);
4774 
4775   BIND(CRC_by128_pre);
4776     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4777       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4778     mov(crc, 0);
4779     crc32cx(crc, crc, tmp0);
4780     crc32cx(crc, crc, tmp1);
4781 
4782     cbnz(len, CRC_less128);
4783 
4784   BIND(L_exit);
4785 }
4786 
4787 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4788         Register len, Register tmp0, Register tmp1, Register tmp2,
4789         Register tmp3) {
4790     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4791     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4792 
4793     subs(len, len, 128);
4794     br(Assembler::GE, CRC_by64_pre);
4795   BIND(CRC_less64);
4796     adds(len, len, 128-32);
4797     br(Assembler::GE, CRC_by32_loop);
4798   BIND(CRC_less32);
4799     adds(len, len, 32-4);
4800     br(Assembler::GE, CRC_by4_loop);
4801     adds(len, len, 4);
4802     br(Assembler::GT, CRC_by1_loop);
4803     b(L_exit);
4804 
4805   BIND(CRC_by32_loop);
4806     ldp(tmp0, tmp1, Address(post(buf, 16)));
4807     subs(len, len, 32);
4808     crc32cx(crc, crc, tmp0);
4809     ldr(tmp2, Address(post(buf, 8)));
4810     crc32cx(crc, crc, tmp1);
4811     ldr(tmp3, Address(post(buf, 8)));
4812     crc32cx(crc, crc, tmp2);
4813     crc32cx(crc, crc, tmp3);
4814     br(Assembler::GE, CRC_by32_loop);
4815     cmn(len, (u1)32);
4816     br(Assembler::NE, CRC_less32);
4817     b(L_exit);
4818 
4819   BIND(CRC_by4_loop);
4820     ldrw(tmp0, Address(post(buf, 4)));
4821     subs(len, len, 4);
4822     crc32cw(crc, crc, tmp0);
4823     br(Assembler::GE, CRC_by4_loop);
4824     adds(len, len, 4);
4825     br(Assembler::LE, L_exit);
4826   BIND(CRC_by1_loop);
4827     ldrb(tmp0, Address(post(buf, 1)));
4828     subs(len, len, 1);
4829     crc32cb(crc, crc, tmp0);
4830     br(Assembler::GT, CRC_by1_loop);
4831     b(L_exit);
4832 
4833   BIND(CRC_by64_pre);
4834     sub(buf, buf, 8);
4835     ldp(tmp0, tmp1, Address(buf, 8));
4836     crc32cx(crc, crc, tmp0);
4837     ldr(tmp2, Address(buf, 24));
4838     crc32cx(crc, crc, tmp1);
4839     ldr(tmp3, Address(buf, 32));
4840     crc32cx(crc, crc, tmp2);
4841     ldr(tmp0, Address(buf, 40));
4842     crc32cx(crc, crc, tmp3);
4843     ldr(tmp1, Address(buf, 48));
4844     crc32cx(crc, crc, tmp0);
4845     ldr(tmp2, Address(buf, 56));
4846     crc32cx(crc, crc, tmp1);
4847     ldr(tmp3, Address(pre(buf, 64)));
4848 
4849     b(CRC_by64_loop);
4850 
4851     align(CodeEntryAlignment);
4852   BIND(CRC_by64_loop);
4853     subs(len, len, 64);
4854     crc32cx(crc, crc, tmp2);
4855     ldr(tmp0, Address(buf, 8));
4856     crc32cx(crc, crc, tmp3);
4857     ldr(tmp1, Address(buf, 16));
4858     crc32cx(crc, crc, tmp0);
4859     ldr(tmp2, Address(buf, 24));
4860     crc32cx(crc, crc, tmp1);
4861     ldr(tmp3, Address(buf, 32));
4862     crc32cx(crc, crc, tmp2);
4863     ldr(tmp0, Address(buf, 40));
4864     crc32cx(crc, crc, tmp3);
4865     ldr(tmp1, Address(buf, 48));
4866     crc32cx(crc, crc, tmp0);
4867     ldr(tmp2, Address(buf, 56));
4868     crc32cx(crc, crc, tmp1);
4869     ldr(tmp3, Address(pre(buf, 64)));
4870     br(Assembler::GE, CRC_by64_loop);
4871 
4872     // post-loop
4873     crc32cx(crc, crc, tmp2);
4874     crc32cx(crc, crc, tmp3);
4875 
4876     sub(len, len, 64);
4877     add(buf, buf, 8);
4878     cmn(len, (u1)128);
4879     br(Assembler::NE, CRC_less64);
4880   BIND(L_exit);
4881 }
4882 
4883 /**
4884  * @param crc   register containing existing CRC (32-bit)
4885  * @param buf   register pointing to input byte buffer (byte*)
4886  * @param len   register containing number of bytes
4887  * @param table register that will contain address of CRC table
4888  * @param tmp   scratch register
4889  */
4890 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4891         Register table0, Register table1, Register table2, Register table3,
4892         Register tmp, Register tmp2, Register tmp3) {
4893   if (UseCryptoPmullForCRC32) {
4894     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4895   } else {
4896     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4897   }
4898 }
4899 
4900 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4901         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4902     Label CRC_by128_loop;
4903     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4904 
4905     sub(len, len, 256);
4906     Register table = tmp0;
4907     {
4908       uint64_t offset;
4909       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4910       add(table, table, offset);
4911     }
4912     add(table, table, table_offset);
4913 
4914     // Registers v0..v7 are used as data registers.
4915     // Registers v16..v31 are used as tmp registers.
4916     sub(buf, buf, 0x10);
4917     ldrq(v0, Address(buf, 0x10));
4918     ldrq(v1, Address(buf, 0x20));
4919     ldrq(v2, Address(buf, 0x30));
4920     ldrq(v3, Address(buf, 0x40));
4921     ldrq(v4, Address(buf, 0x50));
4922     ldrq(v5, Address(buf, 0x60));
4923     ldrq(v6, Address(buf, 0x70));
4924     ldrq(v7, Address(pre(buf, 0x80)));
4925 
4926     movi(v31, T4S, 0);
4927     mov(v31, S, 0, crc);
4928     eor(v0, T16B, v0, v31);
4929 
4930     // Register v16 contains constants from the crc table.
4931     ldrq(v16, Address(table));
4932     b(CRC_by128_loop);
4933 
4934     align(OptoLoopAlignment);
4935   BIND(CRC_by128_loop);
4936     pmull (v17,  T1Q, v0, v16, T1D);
4937     pmull2(v18, T1Q, v0, v16, T2D);
4938     ldrq(v0, Address(buf, 0x10));
4939     eor3(v0, T16B, v17,  v18, v0);
4940 
4941     pmull (v19, T1Q, v1, v16, T1D);
4942     pmull2(v20, T1Q, v1, v16, T2D);
4943     ldrq(v1, Address(buf, 0x20));
4944     eor3(v1, T16B, v19, v20, v1);
4945 
4946     pmull (v21, T1Q, v2, v16, T1D);
4947     pmull2(v22, T1Q, v2, v16, T2D);
4948     ldrq(v2, Address(buf, 0x30));
4949     eor3(v2, T16B, v21, v22, v2);
4950 
4951     pmull (v23, T1Q, v3, v16, T1D);
4952     pmull2(v24, T1Q, v3, v16, T2D);
4953     ldrq(v3, Address(buf, 0x40));
4954     eor3(v3, T16B, v23, v24, v3);
4955 
4956     pmull (v25, T1Q, v4, v16, T1D);
4957     pmull2(v26, T1Q, v4, v16, T2D);
4958     ldrq(v4, Address(buf, 0x50));
4959     eor3(v4, T16B, v25, v26, v4);
4960 
4961     pmull (v27, T1Q, v5, v16, T1D);
4962     pmull2(v28, T1Q, v5, v16, T2D);
4963     ldrq(v5, Address(buf, 0x60));
4964     eor3(v5, T16B, v27, v28, v5);
4965 
4966     pmull (v29, T1Q, v6, v16, T1D);
4967     pmull2(v30, T1Q, v6, v16, T2D);
4968     ldrq(v6, Address(buf, 0x70));
4969     eor3(v6, T16B, v29, v30, v6);
4970 
4971     // Reuse registers v23, v24.
4972     // Using them won't block the first instruction of the next iteration.
4973     pmull (v23, T1Q, v7, v16, T1D);
4974     pmull2(v24, T1Q, v7, v16, T2D);
4975     ldrq(v7, Address(pre(buf, 0x80)));
4976     eor3(v7, T16B, v23, v24, v7);
4977 
4978     subs(len, len, 0x80);
4979     br(Assembler::GE, CRC_by128_loop);
4980 
4981     // fold into 512 bits
4982     // Use v31 for constants because v16 can be still in use.
4983     ldrq(v31, Address(table, 0x10));
4984 
4985     pmull (v17,  T1Q, v0, v31, T1D);
4986     pmull2(v18, T1Q, v0, v31, T2D);
4987     eor3(v0, T16B, v17, v18, v4);
4988 
4989     pmull (v19, T1Q, v1, v31, T1D);
4990     pmull2(v20, T1Q, v1, v31, T2D);
4991     eor3(v1, T16B, v19, v20, v5);
4992 
4993     pmull (v21, T1Q, v2, v31, T1D);
4994     pmull2(v22, T1Q, v2, v31, T2D);
4995     eor3(v2, T16B, v21, v22, v6);
4996 
4997     pmull (v23, T1Q, v3, v31, T1D);
4998     pmull2(v24, T1Q, v3, v31, T2D);
4999     eor3(v3, T16B, v23, v24, v7);
5000 
5001     // fold into 128 bits
5002     // Use v17 for constants because v31 can be still in use.
5003     ldrq(v17, Address(table, 0x20));
5004     pmull (v25, T1Q, v0, v17, T1D);
5005     pmull2(v26, T1Q, v0, v17, T2D);
5006     eor3(v3, T16B, v3, v25, v26);
5007 
5008     // Use v18 for constants because v17 can be still in use.
5009     ldrq(v18, Address(table, 0x30));
5010     pmull (v27, T1Q, v1, v18, T1D);
5011     pmull2(v28, T1Q, v1, v18, T2D);
5012     eor3(v3, T16B, v3, v27, v28);
5013 
5014     // Use v19 for constants because v18 can be still in use.
5015     ldrq(v19, Address(table, 0x40));
5016     pmull (v29, T1Q, v2, v19, T1D);
5017     pmull2(v30, T1Q, v2, v19, T2D);
5018     eor3(v0, T16B, v3, v29, v30);
5019 
5020     add(len, len, 0x80);
5021     add(buf, buf, 0x10);
5022 
5023     mov(tmp0, v0, D, 0);
5024     mov(tmp1, v0, D, 1);
5025 }
5026 
5027 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5028   Address adr;
5029   switch(dst.getMode()) {
5030   case Address::base_plus_offset:
5031     // This is the expected mode, although we allow all the other
5032     // forms below.
5033     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5034     break;
5035   default:
5036     lea(rscratch2, dst);
5037     adr = Address(rscratch2);
5038     break;
5039   }
5040   ldr(rscratch1, adr);
5041   add(rscratch1, rscratch1, src);
5042   str(rscratch1, adr);
5043 }
5044 
5045 void MacroAssembler::cmpptr(Register src1, Address src2) {
5046   uint64_t offset;
5047   adrp(rscratch1, src2, offset);
5048   ldr(rscratch1, Address(rscratch1, offset));
5049   cmp(src1, rscratch1);
5050 }
5051 
5052 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5053   cmp(obj1, obj2);
5054 }
5055 
5056 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5057   load_method_holder(rresult, rmethod);
5058   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5059 }
5060 
5061 void MacroAssembler::load_method_holder(Register holder, Register method) {
5062   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5063   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5064   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5065 }
5066 
5067 // Loads the obj's Klass* into dst.
5068 // Preserves all registers (incl src, rscratch1 and rscratch2).
5069 // Input:
5070 // src - the oop we want to load the klass from.
5071 // dst - output narrow klass.
5072 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5073   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5074   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5075   lsr(dst, dst, markWord::klass_shift);
5076 }
5077 
5078 void MacroAssembler::load_klass(Register dst, Register src) {
5079   if (UseCompactObjectHeaders) {
5080     load_narrow_klass_compact(dst, src);
5081     decode_klass_not_null(dst);
5082   } else if (UseCompressedClassPointers) {
5083     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5084     decode_klass_not_null(dst);
5085   } else {
5086     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5087   }
5088 }
5089 
5090 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5091   if (RestoreMXCSROnJNICalls) {
5092     Label OK;
5093     get_fpcr(tmp1);
5094     mov(tmp2, tmp1);
5095     // Set FPCR to the state we need. We do want Round to Nearest. We
5096     // don't want non-IEEE rounding modes or floating-point traps.
5097     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5098     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5099     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5100     eor(tmp2, tmp1, tmp2);
5101     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5102     set_fpcr(tmp1);
5103     bind(OK);
5104   }
5105 }
5106 
5107 // ((OopHandle)result).resolve();
5108 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5109   // OopHandle::resolve is an indirection.
5110   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5111 }
5112 
5113 // ((WeakHandle)result).resolve();
5114 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5115   assert_different_registers(result, tmp1, tmp2);
5116   Label resolved;
5117 
5118   // A null weak handle resolves to null.
5119   cbz(result, resolved);
5120 
5121   // Only 64 bit platforms support GCs that require a tmp register
5122   // WeakHandle::resolve is an indirection like jweak.
5123   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5124                  result, Address(result), tmp1, tmp2);
5125   bind(resolved);
5126 }
5127 
5128 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5129   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5130   ldr(dst, Address(rmethod, Method::const_offset()));
5131   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5132   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5133   ldr(dst, Address(dst, mirror_offset));
5134   resolve_oop_handle(dst, tmp1, tmp2);
5135 }
5136 
5137 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5138   assert_different_registers(obj, klass, tmp);
5139   if (UseCompressedClassPointers) {
5140     if (UseCompactObjectHeaders) {
5141       load_narrow_klass_compact(tmp, obj);
5142     } else {
5143       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5144     }
5145     if (CompressedKlassPointers::base() == nullptr) {
5146       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5147       return;
5148     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5149                && CompressedKlassPointers::shift() == 0) {
5150       // Only the bottom 32 bits matter
5151       cmpw(klass, tmp);
5152       return;
5153     }
5154     decode_klass_not_null(tmp);
5155   } else {
5156     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5157   }
5158   cmp(klass, tmp);
5159 }
5160 
5161 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5162   if (UseCompactObjectHeaders) {
5163     load_narrow_klass_compact(tmp1, obj1);
5164     load_narrow_klass_compact(tmp2,  obj2);
5165     cmpw(tmp1, tmp2);
5166   } else if (UseCompressedClassPointers) {
5167     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5168     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5169     cmpw(tmp1, tmp2);
5170   } else {
5171     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5172     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5173     cmp(tmp1, tmp2);
5174   }
5175 }
5176 
5177 void MacroAssembler::store_klass(Register dst, Register src) {
5178   // FIXME: Should this be a store release?  concurrent gcs assumes
5179   // klass length is valid if klass field is not null.
5180   assert(!UseCompactObjectHeaders, "not with compact headers");
5181   if (UseCompressedClassPointers) {
5182     encode_klass_not_null(src);
5183     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5184   } else {
5185     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5186   }
5187 }
5188 
5189 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5190   assert(!UseCompactObjectHeaders, "not with compact headers");
5191   if (UseCompressedClassPointers) {
5192     // Store to klass gap in destination
5193     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5194   }
5195 }
5196 
5197 // Algorithm must match CompressedOops::encode.
5198 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5199 #ifdef ASSERT
5200   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5201 #endif
5202   verify_oop_msg(s, "broken oop in encode_heap_oop");
5203   if (CompressedOops::base() == nullptr) {
5204     if (CompressedOops::shift() != 0) {
5205       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5206       lsr(d, s, LogMinObjAlignmentInBytes);
5207     } else {
5208       mov(d, s);
5209     }
5210   } else {
5211     subs(d, s, rheapbase);
5212     csel(d, d, zr, Assembler::HS);
5213     lsr(d, d, LogMinObjAlignmentInBytes);
5214 
5215     /*  Old algorithm: is this any worse?
5216     Label nonnull;
5217     cbnz(r, nonnull);
5218     sub(r, r, rheapbase);
5219     bind(nonnull);
5220     lsr(r, r, LogMinObjAlignmentInBytes);
5221     */
5222   }
5223 }
5224 
5225 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5226 #ifdef ASSERT
5227   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5228   if (CheckCompressedOops) {
5229     Label ok;
5230     cbnz(r, ok);
5231     stop("null oop passed to encode_heap_oop_not_null");
5232     bind(ok);
5233   }
5234 #endif
5235   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5236   if (CompressedOops::base() != nullptr) {
5237     sub(r, r, rheapbase);
5238   }
5239   if (CompressedOops::shift() != 0) {
5240     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5241     lsr(r, r, LogMinObjAlignmentInBytes);
5242   }
5243 }
5244 
5245 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5246 #ifdef ASSERT
5247   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5248   if (CheckCompressedOops) {
5249     Label ok;
5250     cbnz(src, ok);
5251     stop("null oop passed to encode_heap_oop_not_null2");
5252     bind(ok);
5253   }
5254 #endif
5255   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5256 
5257   Register data = src;
5258   if (CompressedOops::base() != nullptr) {
5259     sub(dst, src, rheapbase);
5260     data = dst;
5261   }
5262   if (CompressedOops::shift() != 0) {
5263     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5264     lsr(dst, data, LogMinObjAlignmentInBytes);
5265     data = dst;
5266   }
5267   if (data == src)
5268     mov(dst, src);
5269 }
5270 
5271 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5272 #ifdef ASSERT
5273   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5274 #endif
5275   if (CompressedOops::base() == nullptr) {
5276     if (CompressedOops::shift() != 0) {
5277       lsl(d, s, CompressedOops::shift());
5278     } else if (d != s) {
5279       mov(d, s);
5280     }
5281   } else {
5282     Label done;
5283     if (d != s)
5284       mov(d, s);
5285     cbz(s, done);
5286     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5287     bind(done);
5288   }
5289   verify_oop_msg(d, "broken oop in decode_heap_oop");
5290 }
5291 
5292 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5293   assert (UseCompressedOops, "should only be used for compressed headers");
5294   assert (Universe::heap() != nullptr, "java heap should be initialized");
5295   // Cannot assert, unverified entry point counts instructions (see .ad file)
5296   // vtableStubs also counts instructions in pd_code_size_limit.
5297   // Also do not verify_oop as this is called by verify_oop.
5298   if (CompressedOops::shift() != 0) {
5299     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5300     if (CompressedOops::base() != nullptr) {
5301       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5302     } else {
5303       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5304     }
5305   } else {
5306     assert (CompressedOops::base() == nullptr, "sanity");
5307   }
5308 }
5309 
5310 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5311   assert (UseCompressedOops, "should only be used for compressed headers");
5312   assert (Universe::heap() != nullptr, "java heap should be initialized");
5313   // Cannot assert, unverified entry point counts instructions (see .ad file)
5314   // vtableStubs also counts instructions in pd_code_size_limit.
5315   // Also do not verify_oop as this is called by verify_oop.
5316   if (CompressedOops::shift() != 0) {
5317     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5318     if (CompressedOops::base() != nullptr) {
5319       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5320     } else {
5321       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5322     }
5323   } else {
5324     assert (CompressedOops::base() == nullptr, "sanity");
5325     if (dst != src) {
5326       mov(dst, src);
5327     }
5328   }
5329 }
5330 
5331 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5332 
5333 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5334   assert(Metaspace::initialized(), "metaspace not initialized yet");
5335   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5336   return _klass_decode_mode;
5337 }
5338 
5339 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5340   assert(UseCompressedClassPointers, "not using compressed class pointers");
5341 
5342   // KlassDecodeMode shouldn't be set already.
5343   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5344 
5345   if (base == nullptr) {
5346     return KlassDecodeZero;
5347   }
5348 
5349   if (operand_valid_for_logical_immediate(
5350         /*is32*/false, (uint64_t)base)) {
5351     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5352     if (((uint64_t)base & range_mask) == 0) {
5353       return KlassDecodeXor;
5354     }
5355   }
5356 
5357   const uint64_t shifted_base =
5358     (uint64_t)base >> shift;
5359   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5360     return KlassDecodeMovk;
5361   }
5362 
5363   // No valid encoding.
5364   return KlassDecodeNone;
5365 }
5366 
5367 // Check if one of the above decoding modes will work for given base, shift and range.
5368 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5369   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5370 }
5371 
5372 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5373   _klass_decode_mode = klass_decode_mode(base, shift, range);
5374   return _klass_decode_mode != KlassDecodeNone;
5375 }
5376 
5377 static Register pick_different_tmp(Register dst, Register src) {
5378   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5379   return *tmps.begin();
5380 }
5381 
5382 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5383   // we have to load the klass base from the AOT constants area but
5384   // not the shift because it is not allowed to change
5385   int shift = CompressedKlassPointers::shift();
5386   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5387   if (dst != src) {
5388     // we can load the base into dst, subtract it formthe src and shift down
5389     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5390     ldr(dst, dst);
5391     sub(dst, src, dst);
5392     lsr(dst, dst, shift);
5393   } else {
5394     // we need an extra register in order to load the coop base
5395     Register tmp = pick_different_tmp(dst, src);
5396     RegSet regs = RegSet::of(tmp);
5397     push(regs, sp);
5398     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5399     ldr(tmp, tmp);
5400     sub(dst, src, tmp);
5401     lsr(dst, dst, shift);
5402     pop(regs, sp);
5403   }
5404 }
5405 
5406 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5407   if (AOTCodeCache::is_on_for_dump()) {
5408     encode_klass_not_null_for_aot(dst, src);
5409     return;
5410   }
5411 
5412   switch (klass_decode_mode()) {
5413   case KlassDecodeZero:
5414     if (CompressedKlassPointers::shift() != 0) {
5415       lsr(dst, src, CompressedKlassPointers::shift());
5416     } else {
5417       if (dst != src) mov(dst, src);
5418     }
5419     break;
5420 
5421   case KlassDecodeXor:
5422     if (CompressedKlassPointers::shift() != 0) {
5423       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5424       lsr(dst, dst, CompressedKlassPointers::shift());
5425     } else {
5426       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5427     }
5428     break;
5429 
5430   case KlassDecodeMovk:
5431     if (CompressedKlassPointers::shift() != 0) {
5432       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5433     } else {
5434       movw(dst, src);
5435     }
5436     break;
5437 
5438   case KlassDecodeNone:
5439     ShouldNotReachHere();
5440     break;
5441   }
5442 }
5443 
5444 void MacroAssembler::encode_klass_not_null(Register r) {
5445   encode_klass_not_null(r, r);
5446 }
5447 
5448 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5449   // we have to load the klass base from the AOT constants area but
5450   // not the shift because it is not allowed to change
5451   int shift = CompressedKlassPointers::shift();
5452   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5453   if (dst != src) {
5454     // we can load the base into dst then add the offset with a suitable shift
5455     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5456     ldr(dst, dst);
5457     add(dst, dst, src, LSL,  shift);
5458   } else {
5459     // we need an extra register in order to load the coop base
5460     Register tmp = pick_different_tmp(dst, src);
5461     RegSet regs = RegSet::of(tmp);
5462     push(regs, sp);
5463     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5464     ldr(tmp, tmp);
5465     add(dst, tmp,  src, LSL,  shift);
5466     pop(regs, sp);
5467   }
5468 }
5469 
5470 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5471   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5472 
5473   if (AOTCodeCache::is_on_for_dump()) {
5474     decode_klass_not_null_for_aot(dst, src);
5475     return;
5476   }
5477 
5478   switch (klass_decode_mode()) {
5479   case KlassDecodeZero:
5480     if (CompressedKlassPointers::shift() != 0) {
5481       lsl(dst, src, CompressedKlassPointers::shift());
5482     } else {
5483       if (dst != src) mov(dst, src);
5484     }
5485     break;
5486 
5487   case KlassDecodeXor:
5488     if (CompressedKlassPointers::shift() != 0) {
5489       lsl(dst, src, CompressedKlassPointers::shift());
5490       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5491     } else {
5492       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5493     }
5494     break;
5495 
5496   case KlassDecodeMovk: {
5497     const uint64_t shifted_base =
5498       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5499 
5500     if (dst != src) movw(dst, src);
5501     movk(dst, shifted_base >> 32, 32);
5502 
5503     if (CompressedKlassPointers::shift() != 0) {
5504       lsl(dst, dst, CompressedKlassPointers::shift());
5505     }
5506 
5507     break;
5508   }
5509 
5510   case KlassDecodeNone:
5511     ShouldNotReachHere();
5512     break;
5513   }
5514 }
5515 
5516 void  MacroAssembler::decode_klass_not_null(Register r) {
5517   decode_klass_not_null(r, r);
5518 }
5519 
5520 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5521 #ifdef ASSERT
5522   {
5523     ThreadInVMfromUnknown tiv;
5524     assert (UseCompressedOops, "should only be used for compressed oops");
5525     assert (Universe::heap() != nullptr, "java heap should be initialized");
5526     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5527     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5528   }
5529 #endif
5530   int oop_index = oop_recorder()->find_index(obj);
5531   InstructionMark im(this);
5532   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5533   code_section()->relocate(inst_mark(), rspec);
5534   movz(dst, 0xDEAD, 16);
5535   movk(dst, 0xBEEF);
5536 }
5537 
5538 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5539   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5540   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5541   int index = oop_recorder()->find_index(k);
5542 
5543   InstructionMark im(this);
5544   RelocationHolder rspec = metadata_Relocation::spec(index);
5545   code_section()->relocate(inst_mark(), rspec);
5546   narrowKlass nk = CompressedKlassPointers::encode(k);
5547   movz(dst, (nk >> 16), 16);
5548   movk(dst, nk & 0xffff);
5549 }
5550 
5551 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5552                                     Register dst, Address src,
5553                                     Register tmp1, Register tmp2) {
5554   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5555   decorators = AccessInternal::decorator_fixup(decorators, type);
5556   bool as_raw = (decorators & AS_RAW) != 0;
5557   if (as_raw) {
5558     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5559   } else {
5560     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5561   }
5562 }
5563 
5564 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5565                                      Address dst, Register val,
5566                                      Register tmp1, Register tmp2, Register tmp3) {
5567   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5568   decorators = AccessInternal::decorator_fixup(decorators, type);
5569   bool as_raw = (decorators & AS_RAW) != 0;
5570   if (as_raw) {
5571     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5572   } else {
5573     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5574   }
5575 }
5576 
5577 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5578                                    Register tmp2, DecoratorSet decorators) {
5579   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5580 }
5581 
5582 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5583                                             Register tmp2, DecoratorSet decorators) {
5584   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5585 }
5586 
5587 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5588                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5589   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5590 }
5591 
5592 // Used for storing nulls.
5593 void MacroAssembler::store_heap_oop_null(Address dst) {
5594   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5595 }
5596 
5597 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5598   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5599   int index = oop_recorder()->allocate_metadata_index(obj);
5600   RelocationHolder rspec = metadata_Relocation::spec(index);
5601   return Address((address)obj, rspec);
5602 }
5603 
5604 // Move an oop into a register.
5605 void MacroAssembler::movoop(Register dst, jobject obj) {
5606   int oop_index;
5607   if (obj == nullptr) {
5608     oop_index = oop_recorder()->allocate_oop_index(obj);
5609   } else {
5610 #ifdef ASSERT
5611     {
5612       ThreadInVMfromUnknown tiv;
5613       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5614     }
5615 #endif
5616     oop_index = oop_recorder()->find_index(obj);
5617   }
5618   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5619 
5620   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5621     mov(dst, Address((address)obj, rspec));
5622   } else {
5623     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5624     ldr(dst, Address(dummy, rspec));
5625   }
5626 }
5627 
5628 // Move a metadata address into a register.
5629 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5630   int oop_index;
5631   if (obj == nullptr) {
5632     oop_index = oop_recorder()->allocate_metadata_index(obj);
5633   } else {
5634     oop_index = oop_recorder()->find_index(obj);
5635   }
5636   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5637   mov(dst, Address((address)obj, rspec));
5638 }
5639 
5640 Address MacroAssembler::constant_oop_address(jobject obj) {
5641 #ifdef ASSERT
5642   {
5643     ThreadInVMfromUnknown tiv;
5644     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5645     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5646   }
5647 #endif
5648   int oop_index = oop_recorder()->find_index(obj);
5649   return Address((address)obj, oop_Relocation::spec(oop_index));
5650 }
5651 
5652 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5653 void MacroAssembler::tlab_allocate(Register obj,
5654                                    Register var_size_in_bytes,
5655                                    int con_size_in_bytes,
5656                                    Register t1,
5657                                    Register t2,
5658                                    Label& slow_case) {
5659   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5660   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5661 }
5662 
5663 void MacroAssembler::verify_tlab() {
5664 #ifdef ASSERT
5665   if (UseTLAB && VerifyOops) {
5666     Label next, ok;
5667 
5668     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5669 
5670     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5671     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5672     cmp(rscratch2, rscratch1);
5673     br(Assembler::HS, next);
5674     STOP("assert(top >= start)");
5675     should_not_reach_here();
5676 
5677     bind(next);
5678     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5679     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5680     cmp(rscratch2, rscratch1);
5681     br(Assembler::HS, ok);
5682     STOP("assert(top <= end)");
5683     should_not_reach_here();
5684 
5685     bind(ok);
5686     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5687   }
5688 #endif
5689 }
5690 
5691 // Writes to stack successive pages until offset reached to check for
5692 // stack overflow + shadow pages.  This clobbers tmp.
5693 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5694   assert_different_registers(tmp, size, rscratch1);
5695   mov(tmp, sp);
5696   // Bang stack for total size given plus shadow page size.
5697   // Bang one page at a time because large size can bang beyond yellow and
5698   // red zones.
5699   Label loop;
5700   mov(rscratch1, (int)os::vm_page_size());
5701   bind(loop);
5702   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5703   subsw(size, size, rscratch1);
5704   str(size, Address(tmp));
5705   br(Assembler::GT, loop);
5706 
5707   // Bang down shadow pages too.
5708   // At this point, (tmp-0) is the last address touched, so don't
5709   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5710   // was post-decremented.)  Skip this address by starting at i=1, and
5711   // touch a few more pages below.  N.B.  It is important to touch all
5712   // the way down to and including i=StackShadowPages.
5713   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5714     // this could be any sized move but this is can be a debugging crumb
5715     // so the bigger the better.
5716     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5717     str(size, Address(tmp));
5718   }
5719 }
5720 
5721 // Move the address of the polling page into dest.
5722 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5723   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5724 }
5725 
5726 // Read the polling page.  The address of the polling page must
5727 // already be in r.
5728 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5729   address mark;
5730   {
5731     InstructionMark im(this);
5732     code_section()->relocate(inst_mark(), rtype);
5733     ldrw(zr, Address(r, 0));
5734     mark = inst_mark();
5735   }
5736   verify_cross_modify_fence_not_required();
5737   return mark;
5738 }
5739 
5740 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5741   relocInfo::relocType rtype = dest.rspec().reloc()->type();
5742   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5743   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5744   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5745   int64_t offset_low = dest_page - low_page;
5746   int64_t offset_high = dest_page - high_page;
5747 
5748   assert(is_valid_AArch64_address(dest.target()), "bad address");
5749   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5750 
5751   InstructionMark im(this);
5752   code_section()->relocate(inst_mark(), dest.rspec());
5753   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5754   // the code cache so that if it is relocated we know it will still reach
5755   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5756     _adrp(reg1, dest.target());
5757   } else {
5758     uint64_t target = (uint64_t)dest.target();
5759     uint64_t adrp_target
5760       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5761 
5762     _adrp(reg1, (address)adrp_target);
5763     movk(reg1, target >> 32, 32);
5764   }
5765   byte_offset = (uint64_t)dest.target() & 0xfff;
5766 }
5767 
5768 void MacroAssembler::load_byte_map_base(Register reg) {
5769   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
5770 
5771   // Strictly speaking the card table base isn't an address at all, and it might
5772   // even be negative. It is thus materialised as a constant.
5773   mov(reg, (uint64_t)ctbs->card_table_base_const());
5774 }
5775 
5776 void MacroAssembler::build_frame(int framesize) {
5777   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5778   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5779   protect_return_address();
5780   if (framesize < ((1 << 9) + 2 * wordSize)) {
5781     sub(sp, sp, framesize);
5782     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5783     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5784   } else {
5785     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5786     if (PreserveFramePointer) mov(rfp, sp);
5787     if (framesize < ((1 << 12) + 2 * wordSize))
5788       sub(sp, sp, framesize - 2 * wordSize);
5789     else {
5790       mov(rscratch1, framesize - 2 * wordSize);
5791       sub(sp, sp, rscratch1);
5792     }
5793   }
5794   verify_cross_modify_fence_not_required();
5795 }
5796 
5797 void MacroAssembler::remove_frame(int framesize) {
5798   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5799   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5800   if (framesize < ((1 << 9) + 2 * wordSize)) {
5801     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5802     add(sp, sp, framesize);
5803   } else {
5804     if (framesize < ((1 << 12) + 2 * wordSize))
5805       add(sp, sp, framesize - 2 * wordSize);
5806     else {
5807       mov(rscratch1, framesize - 2 * wordSize);
5808       add(sp, sp, rscratch1);
5809     }
5810     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5811   }
5812   authenticate_return_address();
5813 }
5814 
5815 
5816 // This method counts leading positive bytes (highest bit not set) in provided byte array
5817 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5818     // Simple and most common case of aligned small array which is not at the
5819     // end of memory page is placed here. All other cases are in stub.
5820     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5821     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5822     assert_different_registers(ary1, len, result);
5823 
5824     mov(result, len);
5825     cmpw(len, 0);
5826     br(LE, DONE);
5827     cmpw(len, 4 * wordSize);
5828     br(GE, STUB_LONG); // size > 32 then go to stub
5829 
5830     int shift = 64 - exact_log2(os::vm_page_size());
5831     lsl(rscratch1, ary1, shift);
5832     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5833     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5834     br(CS, STUB); // at the end of page then go to stub
5835     subs(len, len, wordSize);
5836     br(LT, END);
5837 
5838   BIND(LOOP);
5839     ldr(rscratch1, Address(post(ary1, wordSize)));
5840     tst(rscratch1, UPPER_BIT_MASK);
5841     br(NE, SET_RESULT);
5842     subs(len, len, wordSize);
5843     br(GE, LOOP);
5844     cmpw(len, -wordSize);
5845     br(EQ, DONE);
5846 
5847   BIND(END);
5848     ldr(rscratch1, Address(ary1));
5849     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5850     lslv(rscratch1, rscratch1, rscratch2);
5851     tst(rscratch1, UPPER_BIT_MASK);
5852     br(NE, SET_RESULT);
5853     b(DONE);
5854 
5855   BIND(STUB);
5856     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5857     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5858     address tpc1 = trampoline_call(count_pos);
5859     if (tpc1 == nullptr) {
5860       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5861       postcond(pc() == badAddress);
5862       return nullptr;
5863     }
5864     b(DONE);
5865 
5866   BIND(STUB_LONG);
5867     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5868     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5869     address tpc2 = trampoline_call(count_pos_long);
5870     if (tpc2 == nullptr) {
5871       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5872       postcond(pc() == badAddress);
5873       return nullptr;
5874     }
5875     b(DONE);
5876 
5877   BIND(SET_RESULT);
5878 
5879     add(len, len, wordSize);
5880     sub(result, result, len);
5881 
5882   BIND(DONE);
5883   postcond(pc() != badAddress);
5884   return pc();
5885 }
5886 
5887 // Clobbers: rscratch1, rscratch2, rflags
5888 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5889 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5890                                       Register tmp4, Register tmp5, Register result,
5891                                       Register cnt1, int elem_size) {
5892   Label DONE, SAME;
5893   Register tmp1 = rscratch1;
5894   Register tmp2 = rscratch2;
5895   int elem_per_word = wordSize/elem_size;
5896   int log_elem_size = exact_log2(elem_size);
5897   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5898   int length_offset = arrayOopDesc::length_offset_in_bytes();
5899   int base_offset
5900     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5901   // When the length offset is not aligned to 8 bytes,
5902   // then we align it down. This is valid because the new
5903   // offset will always be the klass which is the same
5904   // for type arrays.
5905   int start_offset = align_down(length_offset, BytesPerWord);
5906   int extra_length = base_offset - start_offset;
5907   assert(start_offset == length_offset || start_offset == klass_offset,
5908          "start offset must be 8-byte-aligned or be the klass offset");
5909   assert(base_offset != start_offset, "must include the length field");
5910   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5911   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5912 
5913   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5914   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5915 
5916 #ifndef PRODUCT
5917   {
5918     const char kind = (elem_size == 2) ? 'U' : 'L';
5919     char comment[64];
5920     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5921     BLOCK_COMMENT(comment);
5922   }
5923 #endif
5924 
5925   // if (a1 == a2)
5926   //     return true;
5927   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5928   br(EQ, SAME);
5929 
5930   if (UseSimpleArrayEquals) {
5931     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5932     // if (a1 == nullptr || a2 == nullptr)
5933     //     return false;
5934     // a1 & a2 == 0 means (some-pointer is null) or
5935     // (very-rare-or-even-probably-impossible-pointer-values)
5936     // so, we can save one branch in most cases
5937     tst(a1, a2);
5938     mov(result, false);
5939     br(EQ, A_MIGHT_BE_NULL);
5940     // if (a1.length != a2.length)
5941     //      return false;
5942     bind(A_IS_NOT_NULL);
5943     ldrw(cnt1, Address(a1, length_offset));
5944     ldrw(tmp5, Address(a2, length_offset));
5945     cmp(cnt1, tmp5);
5946     br(NE, DONE); // If lengths differ, return false
5947     // Increase loop counter by diff between base- and actual start-offset.
5948     addw(cnt1, cnt1, extra_length);
5949     lea(a1, Address(a1, start_offset));
5950     lea(a2, Address(a2, start_offset));
5951     // Check for short strings, i.e. smaller than wordSize.
5952     subs(cnt1, cnt1, elem_per_word);
5953     br(Assembler::LT, SHORT);
5954     // Main 8 byte comparison loop.
5955     bind(NEXT_WORD); {
5956       ldr(tmp1, Address(post(a1, wordSize)));
5957       ldr(tmp2, Address(post(a2, wordSize)));
5958       subs(cnt1, cnt1, elem_per_word);
5959       eor(tmp5, tmp1, tmp2);
5960       cbnz(tmp5, DONE);
5961     } br(GT, NEXT_WORD);
5962     // Last longword.  In the case where length == 4 we compare the
5963     // same longword twice, but that's still faster than another
5964     // conditional branch.
5965     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5966     // length == 4.
5967     if (log_elem_size > 0)
5968       lsl(cnt1, cnt1, log_elem_size);
5969     ldr(tmp3, Address(a1, cnt1));
5970     ldr(tmp4, Address(a2, cnt1));
5971     eor(tmp5, tmp3, tmp4);
5972     cbnz(tmp5, DONE);
5973     b(SAME);
5974     bind(A_MIGHT_BE_NULL);
5975     // in case both a1 and a2 are not-null, proceed with loads
5976     cbz(a1, DONE);
5977     cbz(a2, DONE);
5978     b(A_IS_NOT_NULL);
5979     bind(SHORT);
5980 
5981     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
5982     {
5983       ldrw(tmp1, Address(post(a1, 4)));
5984       ldrw(tmp2, Address(post(a2, 4)));
5985       eorw(tmp5, tmp1, tmp2);
5986       cbnzw(tmp5, DONE);
5987     }
5988     bind(TAIL03);
5989     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
5990     {
5991       ldrh(tmp3, Address(post(a1, 2)));
5992       ldrh(tmp4, Address(post(a2, 2)));
5993       eorw(tmp5, tmp3, tmp4);
5994       cbnzw(tmp5, DONE);
5995     }
5996     bind(TAIL01);
5997     if (elem_size == 1) { // Only needed when comparing byte arrays.
5998       tbz(cnt1, 0, SAME); // 0-1 bytes left.
5999       {
6000         ldrb(tmp1, a1);
6001         ldrb(tmp2, a2);
6002         eorw(tmp5, tmp1, tmp2);
6003         cbnzw(tmp5, DONE);
6004       }
6005     }
6006   } else {
6007     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6008         CSET_EQ, LAST_CHECK;
6009     mov(result, false);
6010     cbz(a1, DONE);
6011     ldrw(cnt1, Address(a1, length_offset));
6012     cbz(a2, DONE);
6013     ldrw(tmp5, Address(a2, length_offset));
6014     cmp(cnt1, tmp5);
6015     br(NE, DONE); // If lengths differ, return false
6016     // Increase loop counter by diff between base- and actual start-offset.
6017     addw(cnt1, cnt1, extra_length);
6018 
6019     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6020     // faster to perform another branch before comparing a1 and a2
6021     cmp(cnt1, (u1)elem_per_word);
6022     br(LE, SHORT); // short or same
6023     ldr(tmp3, Address(pre(a1, start_offset)));
6024     subs(zr, cnt1, stubBytesThreshold);
6025     br(GE, STUB);
6026     ldr(tmp4, Address(pre(a2, start_offset)));
6027     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6028 
6029     // Main 16 byte comparison loop with 2 exits
6030     bind(NEXT_DWORD); {
6031       ldr(tmp1, Address(pre(a1, wordSize)));
6032       ldr(tmp2, Address(pre(a2, wordSize)));
6033       subs(cnt1, cnt1, 2 * elem_per_word);
6034       br(LE, TAIL);
6035       eor(tmp4, tmp3, tmp4);
6036       cbnz(tmp4, DONE);
6037       ldr(tmp3, Address(pre(a1, wordSize)));
6038       ldr(tmp4, Address(pre(a2, wordSize)));
6039       cmp(cnt1, (u1)elem_per_word);
6040       br(LE, TAIL2);
6041       cmp(tmp1, tmp2);
6042     } br(EQ, NEXT_DWORD);
6043     b(DONE);
6044 
6045     bind(TAIL);
6046     eor(tmp4, tmp3, tmp4);
6047     eor(tmp2, tmp1, tmp2);
6048     lslv(tmp2, tmp2, tmp5);
6049     orr(tmp5, tmp4, tmp2);
6050     cmp(tmp5, zr);
6051     b(CSET_EQ);
6052 
6053     bind(TAIL2);
6054     eor(tmp2, tmp1, tmp2);
6055     cbnz(tmp2, DONE);
6056     b(LAST_CHECK);
6057 
6058     bind(STUB);
6059     ldr(tmp4, Address(pre(a2, start_offset)));
6060     if (elem_size == 2) { // convert to byte counter
6061       lsl(cnt1, cnt1, 1);
6062     }
6063     eor(tmp5, tmp3, tmp4);
6064     cbnz(tmp5, DONE);
6065     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6066     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6067     address tpc = trampoline_call(stub);
6068     if (tpc == nullptr) {
6069       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6070       postcond(pc() == badAddress);
6071       return nullptr;
6072     }
6073     b(DONE);
6074 
6075     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6076     // so, if a2 == null => return false(0), else return true, so we can return a2
6077     mov(result, a2);
6078     b(DONE);
6079     bind(SHORT);
6080     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6081     ldr(tmp3, Address(a1, start_offset));
6082     ldr(tmp4, Address(a2, start_offset));
6083     bind(LAST_CHECK);
6084     eor(tmp4, tmp3, tmp4);
6085     lslv(tmp5, tmp4, tmp5);
6086     cmp(tmp5, zr);
6087     bind(CSET_EQ);
6088     cset(result, EQ);
6089     b(DONE);
6090   }
6091 
6092   bind(SAME);
6093   mov(result, true);
6094   // That's it.
6095   bind(DONE);
6096 
6097   BLOCK_COMMENT("} array_equals");
6098   postcond(pc() != badAddress);
6099   return pc();
6100 }
6101 
6102 // Compare Strings
6103 
6104 // For Strings we're passed the address of the first characters in a1
6105 // and a2 and the length in cnt1.
6106 // There are two implementations.  For arrays >= 8 bytes, all
6107 // comparisons (including the final one, which may overlap) are
6108 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6109 // halfword, then a short, and then a byte.
6110 
6111 void MacroAssembler::string_equals(Register a1, Register a2,
6112                                    Register result, Register cnt1)
6113 {
6114   Label SAME, DONE, SHORT, NEXT_WORD;
6115   Register tmp1 = rscratch1;
6116   Register tmp2 = rscratch2;
6117   Register cnt2 = tmp2;  // cnt2 only used in array length compare
6118 
6119   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6120 
6121 #ifndef PRODUCT
6122   {
6123     char comment[64];
6124     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6125     BLOCK_COMMENT(comment);
6126   }
6127 #endif
6128 
6129   mov(result, false);
6130 
6131   // Check for short strings, i.e. smaller than wordSize.
6132   subs(cnt1, cnt1, wordSize);
6133   br(Assembler::LT, SHORT);
6134   // Main 8 byte comparison loop.
6135   bind(NEXT_WORD); {
6136     ldr(tmp1, Address(post(a1, wordSize)));
6137     ldr(tmp2, Address(post(a2, wordSize)));
6138     subs(cnt1, cnt1, wordSize);
6139     eor(tmp1, tmp1, tmp2);
6140     cbnz(tmp1, DONE);
6141   } br(GT, NEXT_WORD);
6142   // Last longword.  In the case where length == 4 we compare the
6143   // same longword twice, but that's still faster than another
6144   // conditional branch.
6145   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6146   // length == 4.
6147   ldr(tmp1, Address(a1, cnt1));
6148   ldr(tmp2, Address(a2, cnt1));
6149   eor(tmp2, tmp1, tmp2);
6150   cbnz(tmp2, DONE);
6151   b(SAME);
6152 
6153   bind(SHORT);
6154   Label TAIL03, TAIL01;
6155 
6156   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6157   {
6158     ldrw(tmp1, Address(post(a1, 4)));
6159     ldrw(tmp2, Address(post(a2, 4)));
6160     eorw(tmp1, tmp1, tmp2);
6161     cbnzw(tmp1, DONE);
6162   }
6163   bind(TAIL03);
6164   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6165   {
6166     ldrh(tmp1, Address(post(a1, 2)));
6167     ldrh(tmp2, Address(post(a2, 2)));
6168     eorw(tmp1, tmp1, tmp2);
6169     cbnzw(tmp1, DONE);
6170   }
6171   bind(TAIL01);
6172   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6173     {
6174     ldrb(tmp1, a1);
6175     ldrb(tmp2, a2);
6176     eorw(tmp1, tmp1, tmp2);
6177     cbnzw(tmp1, DONE);
6178   }
6179   // Arrays are equal.
6180   bind(SAME);
6181   mov(result, true);
6182 
6183   // That's it.
6184   bind(DONE);
6185   BLOCK_COMMENT("} string_equals");
6186 }
6187 
6188 
6189 // The size of the blocks erased by the zero_blocks stub.  We must
6190 // handle anything smaller than this ourselves in zero_words().
6191 const int MacroAssembler::zero_words_block_size = 8;
6192 
6193 // zero_words() is used by C2 ClearArray patterns and by
6194 // C1_MacroAssembler.  It is as small as possible, handling small word
6195 // counts locally and delegating anything larger to the zero_blocks
6196 // stub.  It is expanded many times in compiled code, so it is
6197 // important to keep it short.
6198 
6199 // ptr:   Address of a buffer to be zeroed.
6200 // cnt:   Count in HeapWords.
6201 //
6202 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6203 address MacroAssembler::zero_words(Register ptr, Register cnt)
6204 {
6205   assert(is_power_of_2(zero_words_block_size), "adjust this");
6206 
6207   BLOCK_COMMENT("zero_words {");
6208   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6209   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6210   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6211 
6212   subs(rscratch1, cnt, zero_words_block_size);
6213   Label around;
6214   br(LO, around);
6215   {
6216     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6217     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6218     // Make sure this is a C2 compilation. C1 allocates space only for
6219     // trampoline stubs generated by Call LIR ops, and in any case it
6220     // makes sense for a C1 compilation task to proceed as quickly as
6221     // possible.
6222     CompileTask* task;
6223     if (StubRoutines::aarch64::complete()
6224         && Thread::current()->is_Compiler_thread()
6225         && (task = ciEnv::current()->task())
6226         && is_c2_compile(task->comp_level())) {
6227       address tpc = trampoline_call(zero_blocks);
6228       if (tpc == nullptr) {
6229         DEBUG_ONLY(reset_labels(around));
6230         return nullptr;
6231       }
6232     } else {
6233       far_call(zero_blocks);
6234     }
6235   }
6236   bind(around);
6237 
6238   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6239   // for us.
6240   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6241     Label l;
6242     tbz(cnt, exact_log2(i), l);
6243     for (int j = 0; j < i; j += 2) {
6244       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6245     }
6246     bind(l);
6247   }
6248   {
6249     Label l;
6250     tbz(cnt, 0, l);
6251     str(zr, Address(ptr));
6252     bind(l);
6253   }
6254 
6255   BLOCK_COMMENT("} zero_words");
6256   return pc();
6257 }
6258 
6259 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6260 // cnt:          Immediate count in HeapWords.
6261 //
6262 // r10, r11, rscratch1, and rscratch2 are clobbered.
6263 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6264 {
6265   assert(wordSize <= BlockZeroingLowLimit,
6266             "increase BlockZeroingLowLimit");
6267   address result = nullptr;
6268   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6269 #ifndef PRODUCT
6270     {
6271       char buf[64];
6272       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6273       BLOCK_COMMENT(buf);
6274     }
6275 #endif
6276     if (cnt >= 16) {
6277       uint64_t loops = cnt/16;
6278       if (loops > 1) {
6279         mov(rscratch2, loops - 1);
6280       }
6281       {
6282         Label loop;
6283         bind(loop);
6284         for (int i = 0; i < 16; i += 2) {
6285           stp(zr, zr, Address(base, i * BytesPerWord));
6286         }
6287         add(base, base, 16 * BytesPerWord);
6288         if (loops > 1) {
6289           subs(rscratch2, rscratch2, 1);
6290           br(GE, loop);
6291         }
6292       }
6293     }
6294     cnt %= 16;
6295     int i = cnt & 1;  // store any odd word to start
6296     if (i) str(zr, Address(base));
6297     for (; i < (int)cnt; i += 2) {
6298       stp(zr, zr, Address(base, i * wordSize));
6299     }
6300     BLOCK_COMMENT("} zero_words");
6301     result = pc();
6302   } else {
6303     mov(r10, base); mov(r11, cnt);
6304     result = zero_words(r10, r11);
6305   }
6306   return result;
6307 }
6308 
6309 // Zero blocks of memory by using DC ZVA.
6310 //
6311 // Aligns the base address first sufficiently for DC ZVA, then uses
6312 // DC ZVA repeatedly for every full block.  cnt is the size to be
6313 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6314 // in cnt.
6315 //
6316 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6317 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6318 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6319   Register tmp = rscratch1;
6320   Register tmp2 = rscratch2;
6321   int zva_length = VM_Version::zva_length();
6322   Label initial_table_end, loop_zva;
6323   Label fini;
6324 
6325   // Base must be 16 byte aligned. If not just return and let caller handle it
6326   tst(base, 0x0f);
6327   br(Assembler::NE, fini);
6328   // Align base with ZVA length.
6329   neg(tmp, base);
6330   andr(tmp, tmp, zva_length - 1);
6331 
6332   // tmp: the number of bytes to be filled to align the base with ZVA length.
6333   add(base, base, tmp);
6334   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6335   adr(tmp2, initial_table_end);
6336   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6337   br(tmp2);
6338 
6339   for (int i = -zva_length + 16; i < 0; i += 16)
6340     stp(zr, zr, Address(base, i));
6341   bind(initial_table_end);
6342 
6343   sub(cnt, cnt, zva_length >> 3);
6344   bind(loop_zva);
6345   dc(Assembler::ZVA, base);
6346   subs(cnt, cnt, zva_length >> 3);
6347   add(base, base, zva_length);
6348   br(Assembler::GE, loop_zva);
6349   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6350   bind(fini);
6351 }
6352 
6353 // base:   Address of a buffer to be filled, 8 bytes aligned.
6354 // cnt:    Count in 8-byte unit.
6355 // value:  Value to be filled with.
6356 // base will point to the end of the buffer after filling.
6357 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6358 {
6359 //  Algorithm:
6360 //
6361 //    if (cnt == 0) {
6362 //      return;
6363 //    }
6364 //    if ((p & 8) != 0) {
6365 //      *p++ = v;
6366 //    }
6367 //
6368 //    scratch1 = cnt & 14;
6369 //    cnt -= scratch1;
6370 //    p += scratch1;
6371 //    switch (scratch1 / 2) {
6372 //      do {
6373 //        cnt -= 16;
6374 //          p[-16] = v;
6375 //          p[-15] = v;
6376 //        case 7:
6377 //          p[-14] = v;
6378 //          p[-13] = v;
6379 //        case 6:
6380 //          p[-12] = v;
6381 //          p[-11] = v;
6382 //          // ...
6383 //        case 1:
6384 //          p[-2] = v;
6385 //          p[-1] = v;
6386 //        case 0:
6387 //          p += 16;
6388 //      } while (cnt);
6389 //    }
6390 //    if ((cnt & 1) == 1) {
6391 //      *p++ = v;
6392 //    }
6393 
6394   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6395 
6396   Label fini, skip, entry, loop;
6397   const int unroll = 8; // Number of stp instructions we'll unroll
6398 
6399   cbz(cnt, fini);
6400   tbz(base, 3, skip);
6401   str(value, Address(post(base, 8)));
6402   sub(cnt, cnt, 1);
6403   bind(skip);
6404 
6405   andr(rscratch1, cnt, (unroll-1) * 2);
6406   sub(cnt, cnt, rscratch1);
6407   add(base, base, rscratch1, Assembler::LSL, 3);
6408   adr(rscratch2, entry);
6409   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6410   br(rscratch2);
6411 
6412   bind(loop);
6413   add(base, base, unroll * 16);
6414   for (int i = -unroll; i < 0; i++)
6415     stp(value, value, Address(base, i * 16));
6416   bind(entry);
6417   subs(cnt, cnt, unroll * 2);
6418   br(Assembler::GE, loop);
6419 
6420   tbz(cnt, 0, fini);
6421   str(value, Address(post(base, 8)));
6422   bind(fini);
6423 }
6424 
6425 // Intrinsic for
6426 //
6427 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6428 //   Encodes char[] to byte[] in ISO-8859-1
6429 //
6430 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6431 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6432 //
6433 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6434 //   Encodes char[] to byte[] in ASCII
6435 //
6436 // This version always returns the number of characters copied, and does not
6437 // clobber the 'len' register. A successful copy will complete with the post-
6438 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6439 // post-condition: 0 <= 'res' < 'len'.
6440 //
6441 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6442 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6443 //       beyond the acceptable, even though the footprint would be smaller.
6444 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6445 //       avoid additional bloat.
6446 //
6447 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6448 void MacroAssembler::encode_iso_array(Register src, Register dst,
6449                                       Register len, Register res, bool ascii,
6450                                       FloatRegister vtmp0, FloatRegister vtmp1,
6451                                       FloatRegister vtmp2, FloatRegister vtmp3,
6452                                       FloatRegister vtmp4, FloatRegister vtmp5)
6453 {
6454   Register cnt = res;
6455   Register max = rscratch1;
6456   Register chk = rscratch2;
6457 
6458   prfm(Address(src), PLDL1STRM);
6459   movw(cnt, len);
6460 
6461 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6462 
6463   Label LOOP_32, DONE_32, FAIL_32;
6464 
6465   BIND(LOOP_32);
6466   {
6467     cmpw(cnt, 32);
6468     br(LT, DONE_32);
6469     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6470     // Extract lower bytes.
6471     FloatRegister vlo0 = vtmp4;
6472     FloatRegister vlo1 = vtmp5;
6473     uzp1(vlo0, T16B, vtmp0, vtmp1);
6474     uzp1(vlo1, T16B, vtmp2, vtmp3);
6475     // Merge bits...
6476     orr(vtmp0, T16B, vtmp0, vtmp1);
6477     orr(vtmp2, T16B, vtmp2, vtmp3);
6478     // Extract merged upper bytes.
6479     FloatRegister vhix = vtmp0;
6480     uzp2(vhix, T16B, vtmp0, vtmp2);
6481     // ISO-check on hi-parts (all zero).
6482     //                          ASCII-check on lo-parts (no sign).
6483     FloatRegister vlox = vtmp1; // Merge lower bytes.
6484                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6485     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6486     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6487     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6488                                 ASCII(orr(chk, chk, max));
6489     cbnz(chk, FAIL_32);
6490     subw(cnt, cnt, 32);
6491     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6492     b(LOOP_32);
6493   }
6494   BIND(FAIL_32);
6495   sub(src, src, 64);
6496   BIND(DONE_32);
6497 
6498   Label LOOP_8, SKIP_8;
6499 
6500   BIND(LOOP_8);
6501   {
6502     cmpw(cnt, 8);
6503     br(LT, SKIP_8);
6504     FloatRegister vhi = vtmp0;
6505     FloatRegister vlo = vtmp1;
6506     ld1(vtmp3, T8H, src);
6507     uzp1(vlo, T16B, vtmp3, vtmp3);
6508     uzp2(vhi, T16B, vtmp3, vtmp3);
6509     // ISO-check on hi-parts (all zero).
6510     //                          ASCII-check on lo-parts (no sign).
6511                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6512     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6513                                 ASCII(umov(max, vtmp2, B, 0));
6514                                 ASCII(orr(chk, chk, max));
6515     cbnz(chk, SKIP_8);
6516 
6517     strd(vlo, Address(post(dst, 8)));
6518     subw(cnt, cnt, 8);
6519     add(src, src, 16);
6520     b(LOOP_8);
6521   }
6522   BIND(SKIP_8);
6523 
6524 #undef ASCII
6525 
6526   Label LOOP, DONE;
6527 
6528   cbz(cnt, DONE);
6529   BIND(LOOP);
6530   {
6531     Register chr = rscratch1;
6532     ldrh(chr, Address(post(src, 2)));
6533     tst(chr, ascii ? 0xff80 : 0xff00);
6534     br(NE, DONE);
6535     strb(chr, Address(post(dst, 1)));
6536     subs(cnt, cnt, 1);
6537     br(GT, LOOP);
6538   }
6539   BIND(DONE);
6540   // Return index where we stopped.
6541   subw(res, len, cnt);
6542 }
6543 
6544 // Inflate byte[] array to char[].
6545 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6546 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6547                                            FloatRegister vtmp1, FloatRegister vtmp2,
6548                                            FloatRegister vtmp3, Register tmp4) {
6549   Label big, done, after_init, to_stub;
6550 
6551   assert_different_registers(src, dst, len, tmp4, rscratch1);
6552 
6553   fmovd(vtmp1, 0.0);
6554   lsrw(tmp4, len, 3);
6555   bind(after_init);
6556   cbnzw(tmp4, big);
6557   // Short string: less than 8 bytes.
6558   {
6559     Label loop, tiny;
6560 
6561     cmpw(len, 4);
6562     br(LT, tiny);
6563     // Use SIMD to do 4 bytes.
6564     ldrs(vtmp2, post(src, 4));
6565     zip1(vtmp3, T8B, vtmp2, vtmp1);
6566     subw(len, len, 4);
6567     strd(vtmp3, post(dst, 8));
6568 
6569     cbzw(len, done);
6570 
6571     // Do the remaining bytes by steam.
6572     bind(loop);
6573     ldrb(tmp4, post(src, 1));
6574     strh(tmp4, post(dst, 2));
6575     subw(len, len, 1);
6576 
6577     bind(tiny);
6578     cbnz(len, loop);
6579 
6580     b(done);
6581   }
6582 
6583   if (SoftwarePrefetchHintDistance >= 0) {
6584     bind(to_stub);
6585       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6586       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6587       address tpc = trampoline_call(stub);
6588       if (tpc == nullptr) {
6589         DEBUG_ONLY(reset_labels(big, done));
6590         postcond(pc() == badAddress);
6591         return nullptr;
6592       }
6593       b(after_init);
6594   }
6595 
6596   // Unpack the bytes 8 at a time.
6597   bind(big);
6598   {
6599     Label loop, around, loop_last, loop_start;
6600 
6601     if (SoftwarePrefetchHintDistance >= 0) {
6602       const int large_loop_threshold = (64 + 16)/8;
6603       ldrd(vtmp2, post(src, 8));
6604       andw(len, len, 7);
6605       cmp(tmp4, (u1)large_loop_threshold);
6606       br(GE, to_stub);
6607       b(loop_start);
6608 
6609       bind(loop);
6610       ldrd(vtmp2, post(src, 8));
6611       bind(loop_start);
6612       subs(tmp4, tmp4, 1);
6613       br(EQ, loop_last);
6614       zip1(vtmp2, T16B, vtmp2, vtmp1);
6615       ldrd(vtmp3, post(src, 8));
6616       st1(vtmp2, T8H, post(dst, 16));
6617       subs(tmp4, tmp4, 1);
6618       zip1(vtmp3, T16B, vtmp3, vtmp1);
6619       st1(vtmp3, T8H, post(dst, 16));
6620       br(NE, loop);
6621       b(around);
6622       bind(loop_last);
6623       zip1(vtmp2, T16B, vtmp2, vtmp1);
6624       st1(vtmp2, T8H, post(dst, 16));
6625       bind(around);
6626       cbz(len, done);
6627     } else {
6628       andw(len, len, 7);
6629       bind(loop);
6630       ldrd(vtmp2, post(src, 8));
6631       sub(tmp4, tmp4, 1);
6632       zip1(vtmp3, T16B, vtmp2, vtmp1);
6633       st1(vtmp3, T8H, post(dst, 16));
6634       cbnz(tmp4, loop);
6635     }
6636   }
6637 
6638   // Do the tail of up to 8 bytes.
6639   add(src, src, len);
6640   ldrd(vtmp3, Address(src, -8));
6641   add(dst, dst, len, ext::uxtw, 1);
6642   zip1(vtmp3, T16B, vtmp3, vtmp1);
6643   strq(vtmp3, Address(dst, -16));
6644 
6645   bind(done);
6646   postcond(pc() != badAddress);
6647   return pc();
6648 }
6649 
6650 // Compress char[] array to byte[].
6651 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6652 // Return the array length if every element in array can be encoded,
6653 // otherwise, the index of first non-latin1 (> 0xff) character.
6654 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6655                                          Register res,
6656                                          FloatRegister tmp0, FloatRegister tmp1,
6657                                          FloatRegister tmp2, FloatRegister tmp3,
6658                                          FloatRegister tmp4, FloatRegister tmp5) {
6659   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6660 }
6661 
6662 // java.math.round(double a)
6663 // Returns the closest long to the argument, with ties rounding to
6664 // positive infinity.  This requires some fiddling for corner
6665 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6666 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6667                                        FloatRegister ftmp) {
6668   Label DONE;
6669   BLOCK_COMMENT("java_round_double: { ");
6670   fmovd(rscratch1, src);
6671   // Use RoundToNearestTiesAway unless src small and -ve.
6672   fcvtasd(dst, src);
6673   // Test if src >= 0 || abs(src) >= 0x1.0p52
6674   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6675   mov(rscratch2, julong_cast(0x1.0p52));
6676   cmp(rscratch1, rscratch2);
6677   br(HS, DONE); {
6678     // src < 0 && abs(src) < 0x1.0p52
6679     // src may have a fractional part, so add 0.5
6680     fmovd(ftmp, 0.5);
6681     faddd(ftmp, src, ftmp);
6682     // Convert double to jlong, use RoundTowardsNegative
6683     fcvtmsd(dst, ftmp);
6684   }
6685   bind(DONE);
6686   BLOCK_COMMENT("} java_round_double");
6687 }
6688 
6689 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6690                                       FloatRegister ftmp) {
6691   Label DONE;
6692   BLOCK_COMMENT("java_round_float: { ");
6693   fmovs(rscratch1, src);
6694   // Use RoundToNearestTiesAway unless src small and -ve.
6695   fcvtassw(dst, src);
6696   // Test if src >= 0 || abs(src) >= 0x1.0p23
6697   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6698   mov(rscratch2, jint_cast(0x1.0p23f));
6699   cmp(rscratch1, rscratch2);
6700   br(HS, DONE); {
6701     // src < 0 && |src| < 0x1.0p23
6702     // src may have a fractional part, so add 0.5
6703     fmovs(ftmp, 0.5f);
6704     fadds(ftmp, src, ftmp);
6705     // Convert float to jint, use RoundTowardsNegative
6706     fcvtmssw(dst, ftmp);
6707   }
6708   bind(DONE);
6709   BLOCK_COMMENT("} java_round_float");
6710 }
6711 
6712 // get_thread() can be called anywhere inside generated code so we
6713 // need to save whatever non-callee save context might get clobbered
6714 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6715 // the call setup code.
6716 //
6717 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6718 // On other systems, the helper is a usual C function.
6719 //
6720 void MacroAssembler::get_thread(Register dst) {
6721   RegSet saved_regs =
6722     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6723     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6724 
6725   protect_return_address();
6726   push(saved_regs, sp);
6727 
6728   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6729   blr(lr);
6730   if (dst != c_rarg0) {
6731     mov(dst, c_rarg0);
6732   }
6733 
6734   pop(saved_regs, sp);
6735   authenticate_return_address();
6736 }
6737 
6738 void MacroAssembler::cache_wb(Address line) {
6739   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6740   assert(line.index() == noreg, "index should be noreg");
6741   assert(line.offset() == 0, "offset should be 0");
6742   // would like to assert this
6743   // assert(line._ext.shift == 0, "shift should be zero");
6744   if (VM_Version::supports_dcpop()) {
6745     // writeback using clear virtual address to point of persistence
6746     dc(Assembler::CVAP, line.base());
6747   } else {
6748     // no need to generate anything as Unsafe.writebackMemory should
6749     // never invoke this stub
6750   }
6751 }
6752 
6753 void MacroAssembler::cache_wbsync(bool is_pre) {
6754   // we only need a barrier post sync
6755   if (!is_pre) {
6756     membar(Assembler::AnyAny);
6757   }
6758 }
6759 
6760 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6761   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6762     return;
6763   }
6764   // Make sure that native code does not change SVE vector length.
6765   Label verify_ok;
6766   movw(tmp, zr);
6767   sve_inc(tmp, B);
6768   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6769   br(EQ, verify_ok);
6770   stop("Error: SVE vector length has changed since jvm startup");
6771   bind(verify_ok);
6772 }
6773 
6774 void MacroAssembler::verify_ptrue() {
6775   Label verify_ok;
6776   if (!UseSVE) {
6777     return;
6778   }
6779   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6780   sve_dec(rscratch1, B);
6781   cbz(rscratch1, verify_ok);
6782   stop("Error: the preserved predicate register (p7) elements are not all true");
6783   bind(verify_ok);
6784 }
6785 
6786 void MacroAssembler::safepoint_isb() {
6787   isb();
6788 #ifndef PRODUCT
6789   if (VerifyCrossModifyFence) {
6790     // Clear the thread state.
6791     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6792   }
6793 #endif
6794 }
6795 
6796 #ifndef PRODUCT
6797 void MacroAssembler::verify_cross_modify_fence_not_required() {
6798   if (VerifyCrossModifyFence) {
6799     // Check if thread needs a cross modify fence.
6800     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6801     Label fence_not_required;
6802     cbz(rscratch1, fence_not_required);
6803     // If it does then fail.
6804     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6805     mov(c_rarg0, rthread);
6806     blr(rscratch1);
6807     bind(fence_not_required);
6808   }
6809 }
6810 #endif
6811 
6812 void MacroAssembler::spin_wait() {
6813   block_comment("spin_wait {");
6814   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6815     switch (VM_Version::spin_wait_desc().inst()) {
6816       case SpinWait::NOP:
6817         nop();
6818         break;
6819       case SpinWait::ISB:
6820         isb();
6821         break;
6822       case SpinWait::YIELD:
6823         yield();
6824         break;
6825       case SpinWait::SB:
6826         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6827         sb();
6828         break;
6829       default:
6830         ShouldNotReachHere();
6831     }
6832   }
6833   block_comment("}");
6834 }
6835 
6836 // Stack frame creation/removal
6837 
6838 void MacroAssembler::enter(bool strip_ret_addr) {
6839   if (strip_ret_addr) {
6840     // Addresses can only be signed once. If there are multiple nested frames being created
6841     // in the same function, then the return address needs stripping first.
6842     strip_return_address();
6843   }
6844   protect_return_address();
6845   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6846   mov(rfp, sp);
6847 }
6848 
6849 void MacroAssembler::leave() {
6850   mov(sp, rfp);
6851   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6852   authenticate_return_address();
6853 }
6854 
6855 // ROP Protection
6856 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6857 // destroying stack frames or whenever directly loading/storing the LR to memory.
6858 // If ROP protection is not set then these functions are no-ops.
6859 // For more details on PAC see pauth_aarch64.hpp.
6860 
6861 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6862 // Uses value zero as the modifier.
6863 //
6864 void MacroAssembler::protect_return_address() {
6865   if (VM_Version::use_rop_protection()) {
6866     check_return_address();
6867     paciaz();
6868   }
6869 }
6870 
6871 // Sign the return value in the given register. Use before updating the LR in the existing stack
6872 // frame for the current function.
6873 // Uses value zero as the modifier.
6874 //
6875 void MacroAssembler::protect_return_address(Register return_reg) {
6876   if (VM_Version::use_rop_protection()) {
6877     check_return_address(return_reg);
6878     paciza(return_reg);
6879   }
6880 }
6881 
6882 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6883 // Uses value zero as the modifier.
6884 //
6885 void MacroAssembler::authenticate_return_address() {
6886   if (VM_Version::use_rop_protection()) {
6887     autiaz();
6888     check_return_address();
6889   }
6890 }
6891 
6892 // Authenticate the return value in the given register. Use before updating the LR in the existing
6893 // stack frame for the current function.
6894 // Uses value zero as the modifier.
6895 //
6896 void MacroAssembler::authenticate_return_address(Register return_reg) {
6897   if (VM_Version::use_rop_protection()) {
6898     autiza(return_reg);
6899     check_return_address(return_reg);
6900   }
6901 }
6902 
6903 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6904 // there is no guaranteed way of authenticating the LR.
6905 //
6906 void MacroAssembler::strip_return_address() {
6907   if (VM_Version::use_rop_protection()) {
6908     xpaclri();
6909   }
6910 }
6911 
6912 #ifndef PRODUCT
6913 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6914 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6915 // it is difficult to debug back to the callee function.
6916 // This function simply loads from the address in the given register.
6917 // Use directly after authentication to catch authentication failures.
6918 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6919 //
6920 void MacroAssembler::check_return_address(Register return_reg) {
6921   if (VM_Version::use_rop_protection()) {
6922     ldr(zr, Address(return_reg));
6923   }
6924 }
6925 #endif
6926 
6927 // The java_calling_convention describes stack locations as ideal slots on
6928 // a frame with no abi restrictions. Since we must observe abi restrictions
6929 // (like the placement of the register window) the slots must be biased by
6930 // the following value.
6931 static int reg2offset_in(VMReg r) {
6932   // Account for saved rfp and lr
6933   // This should really be in_preserve_stack_slots
6934   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6935 }
6936 
6937 static int reg2offset_out(VMReg r) {
6938   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6939 }
6940 
6941 // On 64bit we will store integer like items to the stack as
6942 // 64bits items (AArch64 ABI) even though java would only store
6943 // 32bits for a parameter. On 32bit it will simply be 32bits
6944 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6945 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6946   if (src.first()->is_stack()) {
6947     if (dst.first()->is_stack()) {
6948       // stack to stack
6949       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6950       str(tmp, Address(sp, reg2offset_out(dst.first())));
6951     } else {
6952       // stack to reg
6953       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6954     }
6955   } else if (dst.first()->is_stack()) {
6956     // reg to stack
6957     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6958   } else {
6959     if (dst.first() != src.first()) {
6960       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6961     }
6962   }
6963 }
6964 
6965 // An oop arg. Must pass a handle not the oop itself
6966 void MacroAssembler::object_move(
6967                         OopMap* map,
6968                         int oop_handle_offset,
6969                         int framesize_in_slots,
6970                         VMRegPair src,
6971                         VMRegPair dst,
6972                         bool is_receiver,
6973                         int* receiver_offset) {
6974 
6975   // must pass a handle. First figure out the location we use as a handle
6976 
6977   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
6978 
6979   // See if oop is null if it is we need no handle
6980 
6981   if (src.first()->is_stack()) {
6982 
6983     // Oop is already on the stack as an argument
6984     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
6985     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
6986     if (is_receiver) {
6987       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
6988     }
6989 
6990     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
6991     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
6992     // conditionally move a null
6993     cmp(rscratch1, zr);
6994     csel(rHandle, zr, rHandle, Assembler::EQ);
6995   } else {
6996 
6997     // Oop is in an a register we must store it to the space we reserve
6998     // on the stack for oop_handles and pass a handle if oop is non-null
6999 
7000     const Register rOop = src.first()->as_Register();
7001     int oop_slot;
7002     if (rOop == j_rarg0)
7003       oop_slot = 0;
7004     else if (rOop == j_rarg1)
7005       oop_slot = 1;
7006     else if (rOop == j_rarg2)
7007       oop_slot = 2;
7008     else if (rOop == j_rarg3)
7009       oop_slot = 3;
7010     else if (rOop == j_rarg4)
7011       oop_slot = 4;
7012     else if (rOop == j_rarg5)
7013       oop_slot = 5;
7014     else if (rOop == j_rarg6)
7015       oop_slot = 6;
7016     else {
7017       assert(rOop == j_rarg7, "wrong register");
7018       oop_slot = 7;
7019     }
7020 
7021     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7022     int offset = oop_slot*VMRegImpl::stack_slot_size;
7023 
7024     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7025     // Store oop in handle area, may be null
7026     str(rOop, Address(sp, offset));
7027     if (is_receiver) {
7028       *receiver_offset = offset;
7029     }
7030 
7031     cmp(rOop, zr);
7032     lea(rHandle, Address(sp, offset));
7033     // conditionally move a null
7034     csel(rHandle, zr, rHandle, Assembler::EQ);
7035   }
7036 
7037   // If arg is on the stack then place it otherwise it is already in correct reg.
7038   if (dst.first()->is_stack()) {
7039     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7040   }
7041 }
7042 
7043 // A float arg may have to do float reg int reg conversion
7044 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7045  if (src.first()->is_stack()) {
7046     if (dst.first()->is_stack()) {
7047       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7048       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7049     } else {
7050       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7051     }
7052   } else if (src.first() != dst.first()) {
7053     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7054       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7055     else
7056       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7057   }
7058 }
7059 
7060 // A long move
7061 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7062   if (src.first()->is_stack()) {
7063     if (dst.first()->is_stack()) {
7064       // stack to stack
7065       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7066       str(tmp, Address(sp, reg2offset_out(dst.first())));
7067     } else {
7068       // stack to reg
7069       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7070     }
7071   } else if (dst.first()->is_stack()) {
7072     // reg to stack
7073     // Do we really have to sign extend???
7074     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7075     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7076   } else {
7077     if (dst.first() != src.first()) {
7078       mov(dst.first()->as_Register(), src.first()->as_Register());
7079     }
7080   }
7081 }
7082 
7083 
7084 // A double move
7085 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7086  if (src.first()->is_stack()) {
7087     if (dst.first()->is_stack()) {
7088       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7089       str(tmp, Address(sp, reg2offset_out(dst.first())));
7090     } else {
7091       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7092     }
7093   } else if (src.first() != dst.first()) {
7094     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7095       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7096     else
7097       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7098   }
7099 }
7100 
7101 // Implements fast-locking.
7102 //
7103 //  - obj: the object to be locked
7104 //  - t1, t2, t3: temporary registers, will be destroyed
7105 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7106 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7107   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7108 
7109   Label push;
7110   const Register top = t1;
7111   const Register mark = t2;
7112   const Register t = t3;
7113 
7114   // Preload the markWord. It is important that this is the first
7115   // instruction emitted as it is part of C1's null check semantics.
7116   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7117 
7118   if (UseObjectMonitorTable) {
7119     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7120     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7121   }
7122 
7123   if (DiagnoseSyncOnValueBasedClasses != 0) {
7124     load_klass(t1, obj);
7125     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7126     tst(t1, KlassFlags::_misc_is_value_based_class);
7127     br(Assembler::NE, slow);
7128   }
7129 
7130   // Check if the lock-stack is full.
7131   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7132   cmpw(top, (unsigned)LockStack::end_offset());
7133   br(Assembler::GE, slow);
7134 
7135   // Check for recursion.
7136   subw(t, top, oopSize);
7137   ldr(t, Address(rthread, t));
7138   cmp(obj, t);
7139   br(Assembler::EQ, push);
7140 
7141   // Check header for monitor (0b10).
7142   tst(mark, markWord::monitor_value);
7143   br(Assembler::NE, slow);
7144 
7145   // Try to lock. Transition lock bits 0b01 => 0b00
7146   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7147   orr(mark, mark, markWord::unlocked_value);
7148   eor(t, mark, markWord::unlocked_value);
7149   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7150           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7151   br(Assembler::NE, slow);
7152 
7153   bind(push);
7154   // After successful lock, push object on lock-stack.
7155   str(obj, Address(rthread, top));
7156   addw(top, top, oopSize);
7157   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7158 }
7159 
7160 // Implements fast-unlocking.
7161 //
7162 // - obj: the object to be unlocked
7163 // - t1, t2, t3: temporary registers
7164 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7165 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7166   // cmpxchg clobbers rscratch1.
7167   assert_different_registers(obj, t1, t2, t3, rscratch1);
7168 
7169 #ifdef ASSERT
7170   {
7171     // Check for lock-stack underflow.
7172     Label stack_ok;
7173     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7174     cmpw(t1, (unsigned)LockStack::start_offset());
7175     br(Assembler::GE, stack_ok);
7176     STOP("Lock-stack underflow");
7177     bind(stack_ok);
7178   }
7179 #endif
7180 
7181   Label unlocked, push_and_slow;
7182   const Register top = t1;
7183   const Register mark = t2;
7184   const Register t = t3;
7185 
7186   // Check if obj is top of lock-stack.
7187   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7188   subw(top, top, oopSize);
7189   ldr(t, Address(rthread, top));
7190   cmp(obj, t);
7191   br(Assembler::NE, slow);
7192 
7193   // Pop lock-stack.
7194   DEBUG_ONLY(str(zr, Address(rthread, top));)
7195   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7196 
7197   // Check if recursive.
7198   subw(t, top, oopSize);
7199   ldr(t, Address(rthread, t));
7200   cmp(obj, t);
7201   br(Assembler::EQ, unlocked);
7202 
7203   // Not recursive. Check header for monitor (0b10).
7204   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7205   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7206 
7207 #ifdef ASSERT
7208   // Check header not unlocked (0b01).
7209   Label not_unlocked;
7210   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7211   stop("fast_unlock already unlocked");
7212   bind(not_unlocked);
7213 #endif
7214 
7215   // Try to unlock. Transition lock bits 0b00 => 0b01
7216   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7217   orr(t, mark, markWord::unlocked_value);
7218   cmpxchg(obj, mark, t, Assembler::xword,
7219           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7220   br(Assembler::EQ, unlocked);
7221 
7222   bind(push_and_slow);
7223   // Restore lock-stack and handle the unlock in runtime.
7224   DEBUG_ONLY(str(obj, Address(rthread, top));)
7225   addw(top, top, oopSize);
7226   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7227   b(slow);
7228 
7229   bind(unlocked);
7230 }