1 /*
   2  * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "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   return RelocActions<Patcher>::run(insn_addr, target);
 477 }
 478 
 479 int MacroAssembler::patch_oop(address insn_addr, address o) {
 480   int instructions;
 481   unsigned insn = *(unsigned*)insn_addr;
 482   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 483 
 484   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 485   // narrow OOPs by setting the upper 16 bits in the first
 486   // instruction.
 487   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 488     // Move narrow OOP
 489     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 490     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 491     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 492     instructions = 2;
 493   } else {
 494     // Move wide OOP
 495     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 496     uintptr_t dest = (uintptr_t)o;
 497     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 498     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 499     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 500     instructions = 3;
 501   }
 502   return instructions * NativeInstruction::instruction_size;
 503 }
 504 
 505 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 506   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 507   // We encode narrow ones by setting the upper 16 bits in the first
 508   // instruction.
 509   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 510   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 511          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 512 
 513   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 514   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 515   return 2 * NativeInstruction::instruction_size;
 516 }
 517 
 518 address MacroAssembler::target_addr_for_insn_or_null(address insn_addr) {
 519   if (NativeInstruction::is_ldrw_to_zr(insn_addr)) {
 520     return nullptr;
 521   }
 522   return MacroAssembler::target_addr_for_insn(insn_addr);
 523 }
 524 
 525 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 526   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 527   if (at_return) {
 528     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 529     // we may safely use the sp instead to perform the stack watermark check.
 530     cmp(in_nmethod ? sp : rfp, tmp);
 531     br(Assembler::HI, slow_path);
 532   } else {
 533     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 534   }
 535 }
 536 
 537 void MacroAssembler::rt_call(address dest, Register tmp) {
 538   CodeBlob *cb = CodeCache::find_blob(dest);
 539   if (cb) {
 540     far_call(RuntimeAddress(dest));
 541   } else {
 542     lea(tmp, RuntimeAddress(dest));
 543     blr(tmp);
 544   }
 545 }
 546 
 547 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 548   if (!Continuations::enabled()) return;
 549   Label done;
 550   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 551   cmp(sp, rscratch1);
 552   br(Assembler::LS, done);
 553   mov(rscratch1, sp); // we can't use sp as the source in str
 554   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 555   bind(done);
 556 }
 557 
 558 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 559   if (!Continuations::enabled()) return;
 560   Label done;
 561   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 562   cmp(sp, rscratch1);
 563   br(Assembler::LO, done);
 564   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 565   bind(done);
 566 }
 567 
 568 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 569   // we must set sp to zero to clear frame
 570   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 571 
 572   // must clear fp, so that compiled frames are not confused; it is
 573   // possible that we need it only for debugging
 574   if (clear_fp) {
 575     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 576   }
 577 
 578   // Always clear the pc because it could have been set by make_walkable()
 579   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 580 }
 581 
 582 // Calls to C land
 583 //
 584 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 585 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 586 // has to be reset to 0. This is required to allow proper stack traversal.
 587 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 588                                          Register last_java_fp,
 589                                          Register last_java_pc,
 590                                          Register scratch) {
 591 
 592   if (last_java_pc->is_valid()) {
 593       str(last_java_pc, Address(rthread,
 594                                 JavaThread::frame_anchor_offset()
 595                                 + JavaFrameAnchor::last_Java_pc_offset()));
 596     }
 597 
 598   // determine last_java_sp register
 599   if (last_java_sp == sp) {
 600     mov(scratch, sp);
 601     last_java_sp = scratch;
 602   } else if (!last_java_sp->is_valid()) {
 603     last_java_sp = esp;
 604   }
 605 
 606   // last_java_fp is optional
 607   if (last_java_fp->is_valid()) {
 608     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 609   }
 610 
 611   // We must set sp last.
 612   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 613 }
 614 
 615 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 616                                          Register last_java_fp,
 617                                          address  last_java_pc,
 618                                          Register scratch) {
 619   assert(last_java_pc != nullptr, "must provide a valid PC");
 620 
 621   adr(scratch, last_java_pc);
 622   str(scratch, Address(rthread,
 623                        JavaThread::frame_anchor_offset()
 624                        + JavaFrameAnchor::last_Java_pc_offset()));
 625 
 626   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 627 }
 628 
 629 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 630                                          Register last_java_fp,
 631                                          Label &L,
 632                                          Register scratch) {
 633   if (L.is_bound()) {
 634     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 635   } else {
 636     InstructionMark im(this);
 637     L.add_patch_at(code(), locator());
 638     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 639   }
 640 }
 641 
 642 static inline bool target_needs_far_branch(address addr) {
 643   if (AOTCodeCache::is_on_for_dump()) {
 644     return true;
 645   }
 646   // codecache size <= 128M
 647   if (!MacroAssembler::far_branches()) {
 648     return false;
 649   }
 650   // codecache size > 240M
 651   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 652     return true;
 653   }
 654   // codecache size: 128M..240M
 655   return !CodeCache::is_non_nmethod(addr);
 656 }
 657 
 658 void MacroAssembler::far_call(Address entry, Register tmp) {
 659   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 660   assert(CodeCache::find_blob(entry.target()) != nullptr,
 661          "destination of far call not found in code cache");
 662   assert(entry.rspec().type() == relocInfo::external_word_type
 663          || entry.rspec().type() == relocInfo::runtime_call_type
 664          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 665   if (target_needs_far_branch(entry.target())) {
 666     uint64_t offset;
 667     // We can use ADRP here because we know that the total size of
 668     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 669     adrp(tmp, entry, offset);
 670     add(tmp, tmp, offset);
 671     blr(tmp);
 672   } else {
 673     bl(entry);
 674   }
 675 }
 676 
 677 int MacroAssembler::far_jump(Address entry, Register tmp) {
 678   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 679   assert(CodeCache::find_blob(entry.target()) != nullptr,
 680          "destination of far call not found in code cache");
 681   assert(entry.rspec().type() == relocInfo::external_word_type
 682          || entry.rspec().type() == relocInfo::runtime_call_type
 683          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 684   address start = pc();
 685   if (target_needs_far_branch(entry.target())) {
 686     uint64_t offset;
 687     // We can use ADRP here because we know that the total size of
 688     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 689     adrp(tmp, entry, offset);
 690     add(tmp, tmp, offset);
 691     br(tmp);
 692   } else {
 693     b(entry);
 694   }
 695   return pc() - start;
 696 }
 697 
 698 void MacroAssembler::reserved_stack_check() {
 699     // testing if reserved zone needs to be enabled
 700     Label no_reserved_zone_enabling;
 701 
 702     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 703     cmp(sp, rscratch1);
 704     br(Assembler::LO, no_reserved_zone_enabling);
 705 
 706     enter();   // LR and FP are live.
 707     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 708     mov(c_rarg0, rthread);
 709     blr(rscratch1);
 710     leave();
 711 
 712     // We have already removed our own frame.
 713     // throw_delayed_StackOverflowError will think that it's been
 714     // called by our caller.
 715     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 716     br(rscratch1);
 717     should_not_reach_here();
 718 
 719     bind(no_reserved_zone_enabling);
 720 }
 721 
 722 static void pass_arg0(MacroAssembler* masm, Register arg) {
 723   if (c_rarg0 != arg ) {
 724     masm->mov(c_rarg0, arg);
 725   }
 726 }
 727 
 728 static void pass_arg1(MacroAssembler* masm, Register arg) {
 729   if (c_rarg1 != arg ) {
 730     masm->mov(c_rarg1, arg);
 731   }
 732 }
 733 
 734 static void pass_arg2(MacroAssembler* masm, Register arg) {
 735   if (c_rarg2 != arg ) {
 736     masm->mov(c_rarg2, arg);
 737   }
 738 }
 739 
 740 static void pass_arg3(MacroAssembler* masm, Register arg) {
 741   if (c_rarg3 != arg ) {
 742     masm->mov(c_rarg3, arg);
 743   }
 744 }
 745 
 746 static bool is_preemptable(address entry_point) {
 747   return entry_point == CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter);
 748 }
 749 
 750 void MacroAssembler::call_VM_base(Register oop_result,
 751                                   Register java_thread,
 752                                   Register last_java_sp,
 753                                   address  entry_point,
 754                                   int      number_of_arguments,
 755                                   bool     check_exceptions) {
 756    // determine java_thread register
 757   if (!java_thread->is_valid()) {
 758     java_thread = rthread;
 759   }
 760 
 761   // determine last_java_sp register
 762   if (!last_java_sp->is_valid()) {
 763     last_java_sp = esp;
 764   }
 765 
 766   // debugging support
 767   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 768   assert(java_thread == rthread, "unexpected register");
 769 #ifdef ASSERT
 770   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 771   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 772 #endif // ASSERT
 773 
 774   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 775   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 776 
 777   // push java thread (becomes first argument of C function)
 778 
 779   mov(c_rarg0, java_thread);
 780 
 781   // set last Java frame before call
 782   assert(last_java_sp != rfp, "can't use rfp");
 783 
 784   Label l;
 785   if (is_preemptable(entry_point)) {
 786     // skip setting last_pc since we already set it to desired value.
 787     set_last_Java_frame(last_java_sp, rfp, noreg, rscratch1);
 788   } else {
 789     set_last_Java_frame(last_java_sp, rfp, l, rscratch1);
 790   }
 791 
 792   // do the call, remove parameters
 793   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 794 
 795   // lr could be poisoned with PAC signature during throw_pending_exception
 796   // if it was tail-call optimized by compiler, since lr is not callee-saved
 797   // reload it with proper value
 798   adr(lr, l);
 799 
 800   // reset last Java frame
 801   // Only interpreter should have to clear fp
 802   reset_last_Java_frame(true);
 803 
 804    // C++ interp handles this in the interpreter
 805   check_and_handle_popframe(java_thread);
 806   check_and_handle_earlyret(java_thread);
 807 
 808   if (check_exceptions) {
 809     // check for pending exceptions (java_thread is set upon return)
 810     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 811     Label ok;
 812     cbz(rscratch1, ok);
 813     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 814     br(rscratch1);
 815     bind(ok);
 816   }
 817 
 818   // get oop result if there is one and reset the value in the thread
 819   if (oop_result->is_valid()) {
 820     get_vm_result_oop(oop_result, java_thread);
 821   }
 822 }
 823 
 824 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 825   call_VM_base(oop_result, noreg, noreg, entry_point, number_of_arguments, check_exceptions);
 826 }
 827 
 828 // Check the entry target is always reachable from any branch.
 829 static bool is_always_within_branch_range(Address entry) {
 830   if (AOTCodeCache::is_on_for_dump()) {
 831     return false;
 832   }
 833   const address target = entry.target();
 834 
 835   if (!CodeCache::contains(target)) {
 836     // We always use trampolines for callees outside CodeCache.
 837     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 838     return false;
 839   }
 840 
 841   if (!MacroAssembler::far_branches()) {
 842     return true;
 843   }
 844 
 845   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 846     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 847     // Non-compiled methods stay forever in CodeCache.
 848     // We check whether the longest possible branch is within the branch range.
 849     assert(CodeCache::find_blob(target) != nullptr &&
 850           !CodeCache::find_blob(target)->is_nmethod(),
 851           "runtime call of compiled method");
 852     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 853     const address left_longest_branch_start = CodeCache::low_bound();
 854     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 855                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 856     return is_reachable;
 857   }
 858 
 859   return false;
 860 }
 861 
 862 // Maybe emit a call via a trampoline. If the code cache is small
 863 // trampolines won't be emitted.
 864 address MacroAssembler::trampoline_call(Address entry) {
 865   assert(entry.rspec().type() == relocInfo::runtime_call_type
 866          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 867          || entry.rspec().type() == relocInfo::static_call_type
 868          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 869 
 870   address target = entry.target();
 871 
 872   if (!is_always_within_branch_range(entry)) {
 873     if (!in_scratch_emit_size()) {
 874       // We don't want to emit a trampoline if C2 is generating dummy
 875       // code during its branch shortening phase.
 876       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 877         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 878         code()->share_trampoline_for(entry.target(), offset());
 879       } else {
 880         address stub = emit_trampoline_stub(offset(), target);
 881         if (stub == nullptr) {
 882           postcond(pc() == badAddress);
 883           return nullptr; // CodeCache is full
 884         }
 885       }
 886     }
 887     target = pc();
 888   }
 889 
 890   address call_pc = pc();
 891   relocate(entry.rspec());
 892   bl(target);
 893 
 894   postcond(pc() != badAddress);
 895   return call_pc;
 896 }
 897 
 898 // Emit a trampoline stub for a call to a target which is too far away.
 899 //
 900 // code sequences:
 901 //
 902 // call-site:
 903 //   branch-and-link to <destination> or <trampoline stub>
 904 //
 905 // Related trampoline stub for this call site in the stub section:
 906 //   load the call target from the constant pool
 907 //   branch (LR still points to the call site above)
 908 
 909 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 910                                              address dest) {
 911   // Max stub size: alignment nop, TrampolineStub.
 912   address stub = start_a_stub(max_trampoline_stub_size());
 913   if (stub == nullptr) {
 914     return nullptr;  // CodeBuffer::expand failed
 915   }
 916 
 917   // Create a trampoline stub relocation which relates this trampoline stub
 918   // with the call instruction at insts_call_instruction_offset in the
 919   // instructions code-section.
 920   align(wordSize);
 921   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 922                                             + insts_call_instruction_offset));
 923   const int stub_start_offset = offset();
 924 
 925   // Now, create the trampoline stub's code:
 926   // - load the call
 927   // - call
 928   Label target;
 929   ldr(rscratch1, target);
 930   br(rscratch1);
 931   bind(target);
 932   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 933          "should be");
 934   emit_int64((int64_t)dest);
 935 
 936   const address stub_start_addr = addr_at(stub_start_offset);
 937 
 938   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 939 
 940   end_a_stub();
 941   return stub_start_addr;
 942 }
 943 
 944 int MacroAssembler::max_trampoline_stub_size() {
 945   // Max stub size: alignment nop, TrampolineStub.
 946   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 947 }
 948 
 949 void MacroAssembler::emit_static_call_stub() {
 950   // CompiledDirectCall::set_to_interpreted knows the
 951   // exact layout of this stub.
 952 
 953   isb();
 954   mov_metadata(rmethod, nullptr);
 955 
 956   // Jump to the entry point of the c2i stub.
 957   if (codestub_branch_needs_far_jump()) {
 958     movptr(rscratch1, 0);
 959     br(rscratch1);
 960   } else {
 961     b(pc());
 962   }
 963 }
 964 
 965 int MacroAssembler::static_call_stub_size() {
 966   if (!codestub_branch_needs_far_jump()) {
 967     // isb; movk; movz; movz; b
 968     return 5 * NativeInstruction::instruction_size;
 969   }
 970   // isb; movk; movz; movz; movk; movz; movz; br
 971   return 8 * NativeInstruction::instruction_size;
 972 }
 973 
 974 void MacroAssembler::c2bool(Register x) {
 975   // implements x == 0 ? 0 : 1
 976   // note: must only look at least-significant byte of x
 977   //       since C-style booleans are stored in one byte
 978   //       only! (was bug)
 979   tst(x, 0xff);
 980   cset(x, Assembler::NE);
 981 }
 982 
 983 address MacroAssembler::ic_call(address entry, jint method_index) {
 984   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 985   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 986   return trampoline_call(Address(entry, rh));
 987 }
 988 
 989 int MacroAssembler::ic_check_size() {
 990   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 991   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 992     return NativeInstruction::instruction_size * (7 + extra_instructions);
 993   } else {
 994     return NativeInstruction::instruction_size * (5 + extra_instructions);
 995   }
 996 }
 997 
 998 int MacroAssembler::ic_check(int end_alignment) {
 999   Register receiver = j_rarg0;
1000   Register data = rscratch2;
1001   Register tmp1 = rscratch1;
1002   Register tmp2 = r10;
1003 
1004   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
1005   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
1006   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1007   // before the inline cache check here, and not after
1008   align(end_alignment, offset() + ic_check_size());
1009 
1010   int uep_offset = offset();
1011 
1012   if (UseCompactObjectHeaders) {
1013     load_narrow_klass_compact(tmp1, receiver);
1014     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1015     cmpw(tmp1, tmp2);
1016   } else if (UseCompressedClassPointers) {
1017     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1018     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1019     cmpw(tmp1, tmp2);
1020   } else {
1021     ldr(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1022     ldr(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1023     cmp(tmp1, tmp2);
1024   }
1025 
1026   Label dont;
1027   br(Assembler::EQ, dont);
1028   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1029   bind(dont);
1030   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1031 
1032   return uep_offset;
1033 }
1034 
1035 // Implementation of call_VM versions
1036 
1037 void MacroAssembler::call_VM(Register oop_result,
1038                              address entry_point,
1039                              bool check_exceptions) {
1040   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1041 }
1042 
1043 void MacroAssembler::call_VM(Register oop_result,
1044                              address entry_point,
1045                              Register arg_1,
1046                              bool check_exceptions) {
1047   pass_arg1(this, arg_1);
1048   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1049 }
1050 
1051 void MacroAssembler::call_VM(Register oop_result,
1052                              address entry_point,
1053                              Register arg_1,
1054                              Register arg_2,
1055                              bool check_exceptions) {
1056   assert_different_registers(arg_1, c_rarg2);
1057   pass_arg2(this, arg_2);
1058   pass_arg1(this, arg_1);
1059   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1060 }
1061 
1062 void MacroAssembler::call_VM(Register oop_result,
1063                              address entry_point,
1064                              Register arg_1,
1065                              Register arg_2,
1066                              Register arg_3,
1067                              bool check_exceptions) {
1068   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1069   assert_different_registers(arg_2, c_rarg3);
1070   pass_arg3(this, arg_3);
1071 
1072   pass_arg2(this, arg_2);
1073 
1074   pass_arg1(this, arg_1);
1075   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1076 }
1077 
1078 void MacroAssembler::call_VM(Register oop_result,
1079                              Register last_java_sp,
1080                              address entry_point,
1081                              int number_of_arguments,
1082                              bool check_exceptions) {
1083   call_VM_base(oop_result, rthread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1084 }
1085 
1086 void MacroAssembler::call_VM(Register oop_result,
1087                              Register last_java_sp,
1088                              address entry_point,
1089                              Register arg_1,
1090                              bool check_exceptions) {
1091   pass_arg1(this, arg_1);
1092   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1093 }
1094 
1095 void MacroAssembler::call_VM(Register oop_result,
1096                              Register last_java_sp,
1097                              address entry_point,
1098                              Register arg_1,
1099                              Register arg_2,
1100                              bool check_exceptions) {
1101 
1102   assert_different_registers(arg_1, c_rarg2);
1103   pass_arg2(this, arg_2);
1104   pass_arg1(this, arg_1);
1105   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1106 }
1107 
1108 void MacroAssembler::call_VM(Register oop_result,
1109                              Register last_java_sp,
1110                              address entry_point,
1111                              Register arg_1,
1112                              Register arg_2,
1113                              Register arg_3,
1114                              bool check_exceptions) {
1115   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1116   assert_different_registers(arg_2, c_rarg3);
1117   pass_arg3(this, arg_3);
1118   pass_arg2(this, arg_2);
1119   pass_arg1(this, arg_1);
1120   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1121 }
1122 
1123 
1124 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1125   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1126   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1127   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1128 }
1129 
1130 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1131   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1132   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1133 }
1134 
1135 void MacroAssembler::align(int modulus) {
1136   align(modulus, offset());
1137 }
1138 
1139 // Ensure that the code at target bytes offset from the current offset() is aligned
1140 // according to modulus.
1141 void MacroAssembler::align(int modulus, int target) {
1142   int delta = target - offset();
1143   while ((offset() + delta) % modulus != 0) nop();
1144 }
1145 
1146 void MacroAssembler::post_call_nop() {
1147   if (!Continuations::enabled()) {
1148     return;
1149   }
1150   InstructionMark im(this);
1151   relocate(post_call_nop_Relocation::spec());
1152   InlineSkippedInstructionsCounter skipCounter(this);
1153   nop();
1154   movk(zr, 0);
1155   movk(zr, 0);
1156 }
1157 
1158 // these are no-ops overridden by InterpreterMacroAssembler
1159 
1160 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1161 
1162 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1163 
1164 // Look up the method for a megamorphic invokeinterface call.
1165 // The target method is determined by <intf_klass, itable_index>.
1166 // The receiver klass is in recv_klass.
1167 // On success, the result will be in method_result, and execution falls through.
1168 // On failure, execution transfers to the given label.
1169 void MacroAssembler::lookup_interface_method(Register recv_klass,
1170                                              Register intf_klass,
1171                                              RegisterOrConstant itable_index,
1172                                              Register method_result,
1173                                              Register scan_temp,
1174                                              Label& L_no_such_interface,
1175                          bool return_method) {
1176   assert_different_registers(recv_klass, intf_klass, scan_temp);
1177   assert_different_registers(method_result, intf_klass, scan_temp);
1178   assert(recv_klass != method_result || !return_method,
1179      "recv_klass can be destroyed when method isn't needed");
1180   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1181          "caller must use same register for non-constant itable index as for method");
1182 
1183   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1184   int vtable_base = in_bytes(Klass::vtable_start_offset());
1185   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1186   int scan_step   = itableOffsetEntry::size() * wordSize;
1187   int vte_size    = vtableEntry::size_in_bytes();
1188   assert(vte_size == wordSize, "else adjust times_vte_scale");
1189 
1190   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1191 
1192   // Could store the aligned, prescaled offset in the klass.
1193   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1194   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1195   add(scan_temp, scan_temp, vtable_base);
1196 
1197   if (return_method) {
1198     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1199     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1200     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1201     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1202     if (itentry_off)
1203       add(recv_klass, recv_klass, itentry_off);
1204   }
1205 
1206   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1207   //   if (scan->interface() == intf) {
1208   //     result = (klass + scan->offset() + itable_index);
1209   //   }
1210   // }
1211   Label search, found_method;
1212 
1213   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1214   cmp(intf_klass, method_result);
1215   br(Assembler::EQ, found_method);
1216   bind(search);
1217   // Check that the previous entry is non-null.  A null entry means that
1218   // the receiver class doesn't implement the interface, and wasn't the
1219   // same as when the caller was compiled.
1220   cbz(method_result, L_no_such_interface);
1221   if (itableOffsetEntry::interface_offset() != 0) {
1222     add(scan_temp, scan_temp, scan_step);
1223     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1224   } else {
1225     ldr(method_result, Address(pre(scan_temp, scan_step)));
1226   }
1227   cmp(intf_klass, method_result);
1228   br(Assembler::NE, search);
1229 
1230   bind(found_method);
1231 
1232   // Got a hit.
1233   if (return_method) {
1234     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1235     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1236   }
1237 }
1238 
1239 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1240 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1241 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1242 // The target method is determined by <holder_klass, itable_index>.
1243 // The receiver klass is in recv_klass.
1244 // On success, the result will be in method_result, and execution falls through.
1245 // On failure, execution transfers to the given label.
1246 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1247                                                   Register holder_klass,
1248                                                   Register resolved_klass,
1249                                                   Register method_result,
1250                                                   Register temp_itbl_klass,
1251                                                   Register scan_temp,
1252                                                   int itable_index,
1253                                                   Label& L_no_such_interface) {
1254   // 'method_result' is only used as output register at the very end of this method.
1255   // Until then we can reuse it as 'holder_offset'.
1256   Register holder_offset = method_result;
1257   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1258 
1259   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1260   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1261   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1262   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1263 
1264   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1265 
1266   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1267   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1268   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1269   // temp_itbl_klass = itable[0]._interface;
1270   int vtblEntrySize = vtableEntry::size_in_bytes();
1271   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1272   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1273   mov(holder_offset, zr);
1274   // scan_temp = &(itable[0]._interface)
1275   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1276 
1277   // Initial checks:
1278   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1279   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1280   //   - if (itable[0] == 0), no such interface
1281   cmp(resolved_klass, holder_klass);
1282   br(Assembler::NE, L_loop_search_resolved_entry);
1283   cmp(holder_klass, temp_itbl_klass);
1284   br(Assembler::EQ, L_holder_found);
1285   cbz(temp_itbl_klass, L_no_such_interface);
1286 
1287   // Loop: Look for holder_klass record in itable
1288   //   do {
1289   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1290   //     if (temp_itbl_klass == holder_klass) {
1291   //       goto L_holder_found; // Found!
1292   //     }
1293   //   } while (temp_itbl_klass != 0);
1294   //   goto L_no_such_interface // Not found.
1295   Label L_search_holder;
1296   bind(L_search_holder);
1297     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1298     cmp(holder_klass, temp_itbl_klass);
1299     br(Assembler::EQ, L_holder_found);
1300     cbnz(temp_itbl_klass, L_search_holder);
1301 
1302   b(L_no_such_interface);
1303 
1304   // Loop: Look for resolved_class record in itable
1305   //   while (true) {
1306   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1307   //     if (temp_itbl_klass == 0) {
1308   //       goto L_no_such_interface;
1309   //     }
1310   //     if (temp_itbl_klass == resolved_klass) {
1311   //        goto L_resolved_found;  // Found!
1312   //     }
1313   //     if (temp_itbl_klass == holder_klass) {
1314   //        holder_offset = scan_temp;
1315   //     }
1316   //   }
1317   //
1318   Label L_loop_search_resolved;
1319   bind(L_loop_search_resolved);
1320     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1321   bind(L_loop_search_resolved_entry);
1322     cbz(temp_itbl_klass, L_no_such_interface);
1323     cmp(resolved_klass, temp_itbl_klass);
1324     br(Assembler::EQ, L_resolved_found);
1325     cmp(holder_klass, temp_itbl_klass);
1326     br(Assembler::NE, L_loop_search_resolved);
1327     mov(holder_offset, scan_temp);
1328     b(L_loop_search_resolved);
1329 
1330   // See if we already have a holder klass. If not, go and scan for it.
1331   bind(L_resolved_found);
1332   cbz(holder_offset, L_search_holder);
1333   mov(scan_temp, holder_offset);
1334 
1335   // Finally, scan_temp contains holder_klass vtable offset
1336   bind(L_holder_found);
1337   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1338   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1339     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1340   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1341 }
1342 
1343 // virtual method calling
1344 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1345                                            RegisterOrConstant vtable_index,
1346                                            Register method_result) {
1347   assert(vtableEntry::size() * wordSize == 8,
1348          "adjust the scaling in the code below");
1349   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1350 
1351   if (vtable_index.is_register()) {
1352     lea(method_result, Address(recv_klass,
1353                                vtable_index.as_register(),
1354                                Address::lsl(LogBytesPerWord)));
1355     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1356   } else {
1357     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1358     ldr(method_result,
1359         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1360   }
1361 }
1362 
1363 void MacroAssembler::check_klass_subtype(Register sub_klass,
1364                            Register super_klass,
1365                            Register temp_reg,
1366                            Label& L_success) {
1367   Label L_failure;
1368   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1369   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1370   bind(L_failure);
1371 }
1372 
1373 
1374 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1375                                                    Register super_klass,
1376                                                    Register temp_reg,
1377                                                    Label* L_success,
1378                                                    Label* L_failure,
1379                                                    Label* L_slow_path,
1380                                                    Register super_check_offset) {
1381   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1382   bool must_load_sco = ! super_check_offset->is_valid();
1383   if (must_load_sco) {
1384     assert(temp_reg != noreg, "supply either a temp or a register offset");
1385   }
1386 
1387   Label L_fallthrough;
1388   int label_nulls = 0;
1389   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1390   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1391   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1392   assert(label_nulls <= 1, "at most one null in the batch");
1393 
1394   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1395   Address super_check_offset_addr(super_klass, sco_offset);
1396 
1397   // Hacked jmp, which may only be used just before L_fallthrough.
1398 #define final_jmp(label)                                                \
1399   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1400   else                            b(label)                /*omit semi*/
1401 
1402   // If the pointers are equal, we are done (e.g., String[] elements).
1403   // This self-check enables sharing of secondary supertype arrays among
1404   // non-primary types such as array-of-interface.  Otherwise, each such
1405   // type would need its own customized SSA.
1406   // We move this check to the front of the fast path because many
1407   // type checks are in fact trivially successful in this manner,
1408   // so we get a nicely predicted branch right at the start of the check.
1409   cmp(sub_klass, super_klass);
1410   br(Assembler::EQ, *L_success);
1411 
1412   // Check the supertype display:
1413   if (must_load_sco) {
1414     ldrw(temp_reg, super_check_offset_addr);
1415     super_check_offset = temp_reg;
1416   }
1417 
1418   Address super_check_addr(sub_klass, super_check_offset);
1419   ldr(rscratch1, super_check_addr);
1420   cmp(super_klass, rscratch1); // load displayed supertype
1421   br(Assembler::EQ, *L_success);
1422 
1423   // This check has worked decisively for primary supers.
1424   // Secondary supers are sought in the super_cache ('super_cache_addr').
1425   // (Secondary supers are interfaces and very deeply nested subtypes.)
1426   // This works in the same check above because of a tricky aliasing
1427   // between the super_cache and the primary super display elements.
1428   // (The 'super_check_addr' can address either, as the case requires.)
1429   // Note that the cache is updated below if it does not help us find
1430   // what we need immediately.
1431   // So if it was a primary super, we can just fail immediately.
1432   // Otherwise, it's the slow path for us (no success at this point).
1433 
1434   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1435   if (L_failure == &L_fallthrough) {
1436     cbz(rscratch1, *L_slow_path);
1437   } else {
1438     cbnz(rscratch1, *L_failure);
1439     final_jmp(*L_slow_path);
1440   }
1441 
1442   bind(L_fallthrough);
1443 
1444 #undef final_jmp
1445 }
1446 
1447 // These two are taken from x86, but they look generally useful
1448 
1449 // scans count pointer sized words at [addr] for occurrence of value,
1450 // generic
1451 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1452                                 Register scratch) {
1453   Label Lloop, Lexit;
1454   cbz(count, Lexit);
1455   bind(Lloop);
1456   ldr(scratch, post(addr, wordSize));
1457   cmp(value, scratch);
1458   br(EQ, Lexit);
1459   sub(count, count, 1);
1460   cbnz(count, Lloop);
1461   bind(Lexit);
1462 }
1463 
1464 // scans count 4 byte words at [addr] for occurrence of value,
1465 // generic
1466 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1467                                 Register scratch) {
1468   Label Lloop, Lexit;
1469   cbz(count, Lexit);
1470   bind(Lloop);
1471   ldrw(scratch, post(addr, wordSize));
1472   cmpw(value, scratch);
1473   br(EQ, Lexit);
1474   sub(count, count, 1);
1475   cbnz(count, Lloop);
1476   bind(Lexit);
1477 }
1478 
1479 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1480                                                           Register super_klass,
1481                                                           Register temp_reg,
1482                                                           Register temp2_reg,
1483                                                           Label* L_success,
1484                                                           Label* L_failure,
1485                                                           bool set_cond_codes) {
1486   // NB! Callers may assume that, when temp2_reg is a valid register,
1487   // this code sets it to a nonzero value.
1488 
1489   assert_different_registers(sub_klass, super_klass, temp_reg);
1490   if (temp2_reg != noreg)
1491     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1492 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1493 
1494   Label L_fallthrough;
1495   int label_nulls = 0;
1496   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1497   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1498   assert(label_nulls <= 1, "at most one null in the batch");
1499 
1500   // a couple of useful fields in sub_klass:
1501   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1502   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1503   Address secondary_supers_addr(sub_klass, ss_offset);
1504   Address super_cache_addr(     sub_klass, sc_offset);
1505 
1506   BLOCK_COMMENT("check_klass_subtype_slow_path");
1507 
1508   // Do a linear scan of the secondary super-klass chain.
1509   // This code is rarely used, so simplicity is a virtue here.
1510   // The repne_scan instruction uses fixed registers, which we must spill.
1511   // Don't worry too much about pre-existing connections with the input regs.
1512 
1513   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1514   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1515 
1516   RegSet pushed_registers;
1517   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1518   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1519 
1520   if (super_klass != r0) {
1521     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1522   }
1523 
1524   push(pushed_registers, sp);
1525 
1526   // Get super_klass value into r0 (even if it was in r5 or r2).
1527   if (super_klass != r0) {
1528     mov(r0, super_klass);
1529   }
1530 
1531 #ifndef PRODUCT
1532   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1533 #endif //PRODUCT
1534 
1535   // We will consult the secondary-super array.
1536   ldr(r5, secondary_supers_addr);
1537   // Load the array length.
1538   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1539   // Skip to start of data.
1540   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1541 
1542   cmp(sp, zr); // Clear Z flag; SP is never zero
1543   // Scan R2 words at [R5] for an occurrence of R0.
1544   // Set NZ/Z based on last compare.
1545   repne_scan(r5, r0, r2, rscratch1);
1546 
1547   // Unspill the temp. registers:
1548   pop(pushed_registers, sp);
1549 
1550   br(Assembler::NE, *L_failure);
1551 
1552   // Success.  Cache the super we found and proceed in triumph.
1553 
1554   if (UseSecondarySupersCache) {
1555     str(super_klass, super_cache_addr);
1556   }
1557 
1558   if (L_success != &L_fallthrough) {
1559     b(*L_success);
1560   }
1561 
1562 #undef IS_A_TEMP
1563 
1564   bind(L_fallthrough);
1565 }
1566 
1567 // If Register r is invalid, remove a new register from
1568 // available_regs, and add new register to regs_to_push.
1569 Register MacroAssembler::allocate_if_noreg(Register r,
1570                                   RegSetIterator<Register> &available_regs,
1571                                   RegSet &regs_to_push) {
1572   if (!r->is_valid()) {
1573     r = *available_regs++;
1574     regs_to_push += r;
1575   }
1576   return r;
1577 }
1578 
1579 // check_klass_subtype_slow_path_table() looks for super_klass in the
1580 // hash table belonging to super_klass, branching to L_success or
1581 // L_failure as appropriate. This is essentially a shim which
1582 // allocates registers as necessary then calls
1583 // lookup_secondary_supers_table() to do the work. Any of the temp
1584 // regs may be noreg, in which case this logic will chooses some
1585 // registers push and pop them from the stack.
1586 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1587                                                          Register super_klass,
1588                                                          Register temp_reg,
1589                                                          Register temp2_reg,
1590                                                          Register temp3_reg,
1591                                                          Register result_reg,
1592                                                          FloatRegister vtemp,
1593                                                          Label* L_success,
1594                                                          Label* L_failure,
1595                                                          bool set_cond_codes) {
1596   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1597 
1598   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1599 
1600   Label L_fallthrough;
1601   int label_nulls = 0;
1602   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1603   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1604   assert(label_nulls <= 1, "at most one null in the batch");
1605 
1606   BLOCK_COMMENT("check_klass_subtype_slow_path");
1607 
1608   RegSetIterator<Register> available_regs
1609     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1610 
1611   RegSet pushed_regs;
1612 
1613   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1614   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1615   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1616   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1617 
1618   push(pushed_regs, sp);
1619 
1620   lookup_secondary_supers_table_var(sub_klass,
1621                                     super_klass,
1622                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1623                                     nullptr);
1624   cmp(result_reg, zr);
1625 
1626   // Unspill the temp. registers:
1627   pop(pushed_regs, sp);
1628 
1629   // NB! Callers may assume that, when set_cond_codes is true, this
1630   // code sets temp2_reg to a nonzero value.
1631   if (set_cond_codes) {
1632     mov(temp2_reg, 1);
1633   }
1634 
1635   br(Assembler::NE, *L_failure);
1636 
1637   if (L_success != &L_fallthrough) {
1638     b(*L_success);
1639   }
1640 
1641   bind(L_fallthrough);
1642 }
1643 
1644 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1645                                                    Register super_klass,
1646                                                    Register temp_reg,
1647                                                    Register temp2_reg,
1648                                                    Label* L_success,
1649                                                    Label* L_failure,
1650                                                    bool set_cond_codes) {
1651   if (UseSecondarySupersTable) {
1652     check_klass_subtype_slow_path_table
1653       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1654        /*vtemp*/fnoreg,
1655        L_success, L_failure, set_cond_codes);
1656   } else {
1657     check_klass_subtype_slow_path_linear
1658       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1659   }
1660 }
1661 
1662 
1663 // Ensure that the inline code and the stub are using the same registers.
1664 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1665 do {                                                               \
1666   assert(r_super_klass  == r0                                   && \
1667          r_array_base   == r1                                   && \
1668          r_array_length == r2                                   && \
1669          (r_array_index == r3        || r_array_index == noreg) && \
1670          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1671          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1672          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1673 } while(0)
1674 
1675 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1676                                                          Register r_super_klass,
1677                                                          Register temp1,
1678                                                          Register temp2,
1679                                                          Register temp3,
1680                                                          FloatRegister vtemp,
1681                                                          Register result,
1682                                                          u1 super_klass_slot,
1683                                                          bool stub_is_near) {
1684   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1685 
1686   Label L_fallthrough;
1687 
1688   BLOCK_COMMENT("lookup_secondary_supers_table {");
1689 
1690   const Register
1691     r_array_base   = temp1, // r1
1692     r_array_length = temp2, // r2
1693     r_array_index  = temp3, // r3
1694     r_bitmap       = rscratch2;
1695 
1696   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1697 
1698   u1 bit = super_klass_slot;
1699 
1700   // Make sure that result is nonzero if the TBZ below misses.
1701   mov(result, 1);
1702 
1703   // We're going to need the bitmap in a vector reg and in a core reg,
1704   // so load both now.
1705   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1706   if (bit != 0) {
1707     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1708   }
1709   // First check the bitmap to see if super_klass might be present. If
1710   // the bit is zero, we are certain that super_klass is not one of
1711   // the secondary supers.
1712   tbz(r_bitmap, bit, L_fallthrough);
1713 
1714   // Get the first array index that can contain super_klass into r_array_index.
1715   if (bit != 0) {
1716     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1717     cnt(vtemp, T8B, vtemp);
1718     addv(vtemp, T8B, vtemp);
1719     fmovd(r_array_index, vtemp);
1720   } else {
1721     mov(r_array_index, (u1)1);
1722   }
1723   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1724 
1725   // We will consult the secondary-super array.
1726   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1727 
1728   // The value i in r_array_index is >= 1, so even though r_array_base
1729   // points to the length, we don't need to adjust it to point to the
1730   // data.
1731   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1732   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1733 
1734   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1735   eor(result, result, r_super_klass);
1736   cbz(result, L_fallthrough); // Found a match
1737 
1738   // Is there another entry to check? Consult the bitmap.
1739   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1740 
1741   // Linear probe.
1742   if (bit != 0) {
1743     ror(r_bitmap, r_bitmap, bit);
1744   }
1745 
1746   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1747   // The next slot to be inspected, by the stub we're about to call,
1748   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1749   // have been checked.
1750   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1751   if (stub_is_near) {
1752     bl(stub);
1753   } else {
1754     address call = trampoline_call(stub);
1755     if (call == nullptr) {
1756       return false; // trampoline allocation failed
1757     }
1758   }
1759 
1760   BLOCK_COMMENT("} lookup_secondary_supers_table");
1761 
1762   bind(L_fallthrough);
1763 
1764   if (VerifySecondarySupers) {
1765     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1766                                   temp1, temp2, result);      // r1, r2, r5
1767   }
1768   return true;
1769 }
1770 
1771 // At runtime, return 0 in result if r_super_klass is a superclass of
1772 // r_sub_klass, otherwise return nonzero. Use this version of
1773 // lookup_secondary_supers_table() if you don't know ahead of time
1774 // which superclass will be searched for. Used by interpreter and
1775 // runtime stubs. It is larger and has somewhat greater latency than
1776 // the version above, which takes a constant super_klass_slot.
1777 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1778                                                        Register r_super_klass,
1779                                                        Register temp1,
1780                                                        Register temp2,
1781                                                        Register temp3,
1782                                                        FloatRegister vtemp,
1783                                                        Register result,
1784                                                        Label *L_success) {
1785   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1786 
1787   Label L_fallthrough;
1788 
1789   BLOCK_COMMENT("lookup_secondary_supers_table {");
1790 
1791   const Register
1792     r_array_index = temp3,
1793     slot          = rscratch1,
1794     r_bitmap      = rscratch2;
1795 
1796   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1797 
1798   // Make sure that result is nonzero if the test below misses.
1799   mov(result, 1);
1800 
1801   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1802 
1803   // First check the bitmap to see if super_klass might be present. If
1804   // the bit is zero, we are certain that super_klass is not one of
1805   // the secondary supers.
1806 
1807   // This next instruction is equivalent to:
1808   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1809   // sub(temp2, tmp_reg, slot);
1810   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1811   lslv(temp2, r_bitmap, temp2);
1812   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1813 
1814   bool must_save_v0 = (vtemp == fnoreg);
1815   if (must_save_v0) {
1816     // temp1 and result are free, so use them to preserve vtemp
1817     vtemp = v0;
1818     mov(temp1,  vtemp, D, 0);
1819     mov(result, vtemp, D, 1);
1820   }
1821 
1822   // Get the first array index that can contain super_klass into r_array_index.
1823   mov(vtemp, D, 0, temp2);
1824   cnt(vtemp, T8B, vtemp);
1825   addv(vtemp, T8B, vtemp);
1826   mov(r_array_index, vtemp, D, 0);
1827 
1828   if (must_save_v0) {
1829     mov(vtemp, D, 0, temp1 );
1830     mov(vtemp, D, 1, result);
1831   }
1832 
1833   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1834 
1835   const Register
1836     r_array_base   = temp1,
1837     r_array_length = temp2;
1838 
1839   // The value i in r_array_index is >= 1, so even though r_array_base
1840   // points to the length, we don't need to adjust it to point to the
1841   // data.
1842   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1843   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1844 
1845   // We will consult the secondary-super array.
1846   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1847 
1848   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1849   eor(result, result, r_super_klass);
1850   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1851 
1852   // Is there another entry to check? Consult the bitmap.
1853   rorv(r_bitmap, r_bitmap, slot);
1854   // rol(r_bitmap, r_bitmap, 1);
1855   tbz(r_bitmap, 1, L_fallthrough);
1856 
1857   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1858   // The next slot to be inspected, by the logic we're about to call,
1859   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1860   // have been checked.
1861   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1862                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1863 
1864   BLOCK_COMMENT("} lookup_secondary_supers_table");
1865 
1866   bind(L_fallthrough);
1867 
1868   if (VerifySecondarySupers) {
1869     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1870                                   temp1, temp2, result);      // r1, r2, r5
1871   }
1872 
1873   if (L_success) {
1874     cbz(result, *L_success);
1875   }
1876 }
1877 
1878 // Called by code generated by check_klass_subtype_slow_path
1879 // above. This is called when there is a collision in the hashed
1880 // lookup in the secondary supers array.
1881 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1882                                                              Register r_array_base,
1883                                                              Register r_array_index,
1884                                                              Register r_bitmap,
1885                                                              Register temp1,
1886                                                              Register result,
1887                                                              bool is_stub) {
1888   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1889 
1890   const Register
1891     r_array_length = temp1,
1892     r_sub_klass    = noreg; // unused
1893 
1894   if (is_stub) {
1895     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1896   }
1897 
1898   Label L_fallthrough, L_huge;
1899 
1900   // Load the array length.
1901   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1902   // And adjust the array base to point to the data.
1903   // NB! Effectively increments current slot index by 1.
1904   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1905   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1906 
1907   // The bitmap is full to bursting.
1908   // Implicit invariant: BITMAP_FULL implies (length > 0)
1909   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1910   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1911   br(GT, L_huge);
1912 
1913   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1914   // current slot (at secondary_supers[r_array_index]) has not yet
1915   // been inspected, and r_array_index may be out of bounds if we
1916   // wrapped around the end of the array.
1917 
1918   { // This is conventional linear probing, but instead of terminating
1919     // when a null entry is found in the table, we maintain a bitmap
1920     // in which a 0 indicates missing entries.
1921     // As long as the bitmap is not completely full,
1922     // array_length == popcount(bitmap). The array_length check above
1923     // guarantees there are 0s in the bitmap, so the loop eventually
1924     // terminates.
1925     Label L_loop;
1926     bind(L_loop);
1927 
1928     // Check for wraparound.
1929     cmp(r_array_index, r_array_length);
1930     csel(r_array_index, zr, r_array_index, GE);
1931 
1932     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1933     eor(result, rscratch1, r_super_klass);
1934     cbz(result, L_fallthrough);
1935 
1936     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1937 
1938     ror(r_bitmap, r_bitmap, 1);
1939     add(r_array_index, r_array_index, 1);
1940     b(L_loop);
1941   }
1942 
1943   { // Degenerate case: more than 64 secondary supers.
1944     // FIXME: We could do something smarter here, maybe a vectorized
1945     // comparison or a binary search, but is that worth any added
1946     // complexity?
1947     bind(L_huge);
1948     cmp(sp, zr); // Clear Z flag; SP is never zero
1949     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1950     cset(result, NE); // result == 0 iff we got a match.
1951   }
1952 
1953   bind(L_fallthrough);
1954 }
1955 
1956 // Make sure that the hashed lookup and a linear scan agree.
1957 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1958                                                    Register r_super_klass,
1959                                                    Register temp1,
1960                                                    Register temp2,
1961                                                    Register result) {
1962   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1963 
1964   const Register
1965     r_array_base   = temp1,
1966     r_array_length = temp2,
1967     r_array_index  = noreg, // unused
1968     r_bitmap       = noreg; // unused
1969 
1970   BLOCK_COMMENT("verify_secondary_supers_table {");
1971 
1972   // We will consult the secondary-super array.
1973   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1974 
1975   // Load the array length.
1976   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1977   // And adjust the array base to point to the data.
1978   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1979 
1980   cmp(sp, zr); // Clear Z flag; SP is never zero
1981   // Scan R2 words at [R5] for an occurrence of R0.
1982   // Set NZ/Z based on last compare.
1983   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1984   // rscratch1 == 0 iff we got a match.
1985   cset(rscratch1, NE);
1986 
1987   Label passed;
1988   cmp(result, zr);
1989   cset(result, NE); // normalize result to 0/1 for comparison
1990 
1991   cmp(rscratch1, result);
1992   br(EQ, passed);
1993   {
1994     mov(r0, r_super_klass);         // r0 <- r0
1995     mov(r1, r_sub_klass);           // r1 <- r4
1996     mov(r2, /*expected*/rscratch1); // r2 <- r8
1997     mov(r3, result);                // r3 <- r5
1998     mov(r4, (address)("mismatch")); // r4 <- const
1999     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
2000     should_not_reach_here();
2001   }
2002   bind(passed);
2003 
2004   BLOCK_COMMENT("} verify_secondary_supers_table");
2005 }
2006 
2007 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
2008   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
2009   assert_different_registers(klass, rthread, scratch);
2010 
2011   Label L_fallthrough, L_tmp;
2012   if (L_fast_path == nullptr) {
2013     L_fast_path = &L_fallthrough;
2014   } else if (L_slow_path == nullptr) {
2015     L_slow_path = &L_fallthrough;
2016   }
2017   // Fast path check: class is fully initialized
2018   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2019   ldarb(scratch, scratch);
2020   cmp(scratch, InstanceKlass::fully_initialized);
2021   br(Assembler::EQ, *L_fast_path);
2022 
2023   // Fast path check: current thread is initializer thread
2024   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2025   cmp(rthread, scratch);
2026 
2027   if (L_slow_path == &L_fallthrough) {
2028     br(Assembler::EQ, *L_fast_path);
2029     bind(*L_slow_path);
2030   } else if (L_fast_path == &L_fallthrough) {
2031     br(Assembler::NE, *L_slow_path);
2032     bind(*L_fast_path);
2033   } else {
2034     Unimplemented();
2035   }
2036 }
2037 
2038 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2039   if (!VerifyOops) return;
2040 
2041   // Pass register number to verify_oop_subroutine
2042   const char* b = nullptr;
2043   {
2044     ResourceMark rm;
2045     stringStream ss;
2046     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2047     b = code_string(ss.as_string());
2048   }
2049   BLOCK_COMMENT("verify_oop {");
2050 
2051   strip_return_address(); // This might happen within a stack frame.
2052   protect_return_address();
2053   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2054   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2055 
2056   mov(r0, reg);
2057   movptr(rscratch1, (uintptr_t)(address)b);
2058 
2059   // call indirectly to solve generation ordering problem
2060   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2061   ldr(rscratch2, Address(rscratch2));
2062   blr(rscratch2);
2063 
2064   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2065   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2066   authenticate_return_address();
2067 
2068   BLOCK_COMMENT("} verify_oop");
2069 }
2070 
2071 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2072   if (!VerifyOops) return;
2073 
2074   const char* b = nullptr;
2075   {
2076     ResourceMark rm;
2077     stringStream ss;
2078     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2079     b = code_string(ss.as_string());
2080   }
2081   BLOCK_COMMENT("verify_oop_addr {");
2082 
2083   strip_return_address(); // This might happen within a stack frame.
2084   protect_return_address();
2085   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2086   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2087 
2088   // addr may contain sp so we will have to adjust it based on the
2089   // pushes that we just did.
2090   if (addr.uses(sp)) {
2091     lea(r0, addr);
2092     ldr(r0, Address(r0, 4 * wordSize));
2093   } else {
2094     ldr(r0, addr);
2095   }
2096   movptr(rscratch1, (uintptr_t)(address)b);
2097 
2098   // call indirectly to solve generation ordering problem
2099   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2100   ldr(rscratch2, Address(rscratch2));
2101   blr(rscratch2);
2102 
2103   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2104   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2105   authenticate_return_address();
2106 
2107   BLOCK_COMMENT("} verify_oop_addr");
2108 }
2109 
2110 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2111                                          int extra_slot_offset) {
2112   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2113   int stackElementSize = Interpreter::stackElementSize;
2114   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2115 #ifdef ASSERT
2116   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2117   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2118 #endif
2119   if (arg_slot.is_constant()) {
2120     return Address(esp, arg_slot.as_constant() * stackElementSize
2121                    + offset);
2122   } else {
2123     add(rscratch1, esp, arg_slot.as_register(),
2124         ext::uxtx, exact_log2(stackElementSize));
2125     return Address(rscratch1, offset);
2126   }
2127 }
2128 
2129 void MacroAssembler::call_VM_leaf_base(address entry_point,
2130                                        int number_of_arguments,
2131                                        Label *retaddr) {
2132   Label E, L;
2133 
2134   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2135 
2136   mov(rscratch1, RuntimeAddress(entry_point));
2137   blr(rscratch1);
2138   if (retaddr)
2139     bind(*retaddr);
2140 
2141   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2142 }
2143 
2144 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2145   call_VM_leaf_base(entry_point, number_of_arguments);
2146 }
2147 
2148 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2149   pass_arg0(this, arg_0);
2150   call_VM_leaf_base(entry_point, 1);
2151 }
2152 
2153 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2154   assert_different_registers(arg_1, c_rarg0);
2155   pass_arg0(this, arg_0);
2156   pass_arg1(this, arg_1);
2157   call_VM_leaf_base(entry_point, 2);
2158 }
2159 
2160 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2161                                   Register arg_1, Register arg_2) {
2162   assert_different_registers(arg_1, c_rarg0);
2163   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2164   pass_arg0(this, arg_0);
2165   pass_arg1(this, arg_1);
2166   pass_arg2(this, arg_2);
2167   call_VM_leaf_base(entry_point, 3);
2168 }
2169 
2170 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2171   pass_arg0(this, arg_0);
2172   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2173 }
2174 
2175 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2176 
2177   assert_different_registers(arg_0, c_rarg1);
2178   pass_arg1(this, arg_1);
2179   pass_arg0(this, arg_0);
2180   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2181 }
2182 
2183 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2184   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2185   assert_different_registers(arg_1, c_rarg2);
2186   pass_arg2(this, arg_2);
2187   pass_arg1(this, arg_1);
2188   pass_arg0(this, arg_0);
2189   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2190 }
2191 
2192 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2193   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2194   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2195   assert_different_registers(arg_2, c_rarg3);
2196   pass_arg3(this, arg_3);
2197   pass_arg2(this, arg_2);
2198   pass_arg1(this, arg_1);
2199   pass_arg0(this, arg_0);
2200   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2201 }
2202 
2203 void MacroAssembler::null_check(Register reg, int offset) {
2204   if (needs_explicit_null_check(offset)) {
2205     // provoke OS null exception if reg is null by
2206     // accessing M[reg] w/o changing any registers
2207     // NOTE: this is plenty to provoke a segv
2208     ldr(zr, Address(reg));
2209   } else {
2210     // nothing to do, (later) access of M[reg + offset]
2211     // will provoke OS null exception if reg is null
2212   }
2213 }
2214 
2215 // MacroAssembler protected routines needed to implement
2216 // public methods
2217 
2218 void MacroAssembler::mov(Register r, Address dest) {
2219   code_section()->relocate(pc(), dest.rspec());
2220   uint64_t imm64 = (uint64_t)dest.target();
2221   movptr(r, imm64);
2222 }
2223 
2224 // Move a constant pointer into r.  In AArch64 mode the virtual
2225 // address space is 48 bits in size, so we only need three
2226 // instructions to create a patchable instruction sequence that can
2227 // reach anywhere.
2228 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2229 #ifndef PRODUCT
2230   {
2231     char buffer[64];
2232     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2233     block_comment(buffer);
2234   }
2235 #endif
2236   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2237   movz(r, imm64 & 0xffff);
2238   imm64 >>= 16;
2239   movk(r, imm64 & 0xffff, 16);
2240   imm64 >>= 16;
2241   movk(r, imm64 & 0xffff, 32);
2242 }
2243 
2244 // Macro to mov replicated immediate to vector register.
2245 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2246 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2247 // Vd will get the following values for different arrangements in T
2248 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2249 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2250 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2251 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2252 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2253 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2254 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2255 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2256 // Clobbers rscratch1
2257 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2258   assert(T != T1Q, "unsupported");
2259   if (T == T1D || T == T2D) {
2260     int imm = operand_valid_for_movi_immediate(imm64, T);
2261     if (-1 != imm) {
2262       movi(Vd, T, imm);
2263     } else {
2264       mov(rscratch1, imm64);
2265       dup(Vd, T, rscratch1);
2266     }
2267     return;
2268   }
2269 
2270 #ifdef ASSERT
2271   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2272   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2273   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2274 #endif
2275   int shift = operand_valid_for_movi_immediate(imm64, T);
2276   uint32_t imm32 = imm64 & 0xffffffffULL;
2277   if (shift >= 0) {
2278     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2279   } else {
2280     movw(rscratch1, imm32);
2281     dup(Vd, T, rscratch1);
2282   }
2283 }
2284 
2285 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2286 {
2287 #ifndef PRODUCT
2288   {
2289     char buffer[64];
2290     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2291     block_comment(buffer);
2292   }
2293 #endif
2294   if (operand_valid_for_logical_immediate(false, imm64)) {
2295     orr(dst, zr, imm64);
2296   } else {
2297     // we can use a combination of MOVZ or MOVN with
2298     // MOVK to build up the constant
2299     uint64_t imm_h[4];
2300     int zero_count = 0;
2301     int neg_count = 0;
2302     int i;
2303     for (i = 0; i < 4; i++) {
2304       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2305       if (imm_h[i] == 0) {
2306         zero_count++;
2307       } else if (imm_h[i] == 0xffffL) {
2308         neg_count++;
2309       }
2310     }
2311     if (zero_count == 4) {
2312       // one MOVZ will do
2313       movz(dst, 0);
2314     } else if (neg_count == 4) {
2315       // one MOVN will do
2316       movn(dst, 0);
2317     } else if (zero_count == 3) {
2318       for (i = 0; i < 4; i++) {
2319         if (imm_h[i] != 0L) {
2320           movz(dst, (uint32_t)imm_h[i], (i << 4));
2321           break;
2322         }
2323       }
2324     } else if (neg_count == 3) {
2325       // one MOVN will do
2326       for (int i = 0; i < 4; i++) {
2327         if (imm_h[i] != 0xffffL) {
2328           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2329           break;
2330         }
2331       }
2332     } else if (zero_count == 2) {
2333       // one MOVZ and one MOVK will do
2334       for (i = 0; i < 3; i++) {
2335         if (imm_h[i] != 0L) {
2336           movz(dst, (uint32_t)imm_h[i], (i << 4));
2337           i++;
2338           break;
2339         }
2340       }
2341       for (;i < 4; i++) {
2342         if (imm_h[i] != 0L) {
2343           movk(dst, (uint32_t)imm_h[i], (i << 4));
2344         }
2345       }
2346     } else if (neg_count == 2) {
2347       // one MOVN and one MOVK will do
2348       for (i = 0; i < 4; i++) {
2349         if (imm_h[i] != 0xffffL) {
2350           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2351           i++;
2352           break;
2353         }
2354       }
2355       for (;i < 4; i++) {
2356         if (imm_h[i] != 0xffffL) {
2357           movk(dst, (uint32_t)imm_h[i], (i << 4));
2358         }
2359       }
2360     } else if (zero_count == 1) {
2361       // one MOVZ and two MOVKs will do
2362       for (i = 0; i < 4; i++) {
2363         if (imm_h[i] != 0L) {
2364           movz(dst, (uint32_t)imm_h[i], (i << 4));
2365           i++;
2366           break;
2367         }
2368       }
2369       for (;i < 4; i++) {
2370         if (imm_h[i] != 0x0L) {
2371           movk(dst, (uint32_t)imm_h[i], (i << 4));
2372         }
2373       }
2374     } else if (neg_count == 1) {
2375       // one MOVN and two MOVKs will do
2376       for (i = 0; i < 4; i++) {
2377         if (imm_h[i] != 0xffffL) {
2378           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2379           i++;
2380           break;
2381         }
2382       }
2383       for (;i < 4; i++) {
2384         if (imm_h[i] != 0xffffL) {
2385           movk(dst, (uint32_t)imm_h[i], (i << 4));
2386         }
2387       }
2388     } else {
2389       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2390       movz(dst, (uint32_t)imm_h[0], 0);
2391       for (i = 1; i < 4; i++) {
2392         movk(dst, (uint32_t)imm_h[i], (i << 4));
2393       }
2394     }
2395   }
2396 }
2397 
2398 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2399 {
2400 #ifndef PRODUCT
2401     {
2402       char buffer[64];
2403       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2404       block_comment(buffer);
2405     }
2406 #endif
2407   if (operand_valid_for_logical_immediate(true, imm32)) {
2408     orrw(dst, zr, imm32);
2409   } else {
2410     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2411     // constant
2412     uint32_t imm_h[2];
2413     imm_h[0] = imm32 & 0xffff;
2414     imm_h[1] = ((imm32 >> 16) & 0xffff);
2415     if (imm_h[0] == 0) {
2416       movzw(dst, imm_h[1], 16);
2417     } else if (imm_h[0] == 0xffff) {
2418       movnw(dst, imm_h[1] ^ 0xffff, 16);
2419     } else if (imm_h[1] == 0) {
2420       movzw(dst, imm_h[0], 0);
2421     } else if (imm_h[1] == 0xffff) {
2422       movnw(dst, imm_h[0] ^ 0xffff, 0);
2423     } else {
2424       // use a MOVZ and MOVK (makes it easier to debug)
2425       movzw(dst, imm_h[0], 0);
2426       movkw(dst, imm_h[1], 16);
2427     }
2428   }
2429 }
2430 
2431 // Form an address from base + offset in Rd.  Rd may or may
2432 // not actually be used: you must use the Address that is returned.
2433 // It is up to you to ensure that the shift provided matches the size
2434 // of your data.
2435 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2436   if (Address::offset_ok_for_immed(byte_offset, shift))
2437     // It fits; no need for any heroics
2438     return Address(base, byte_offset);
2439 
2440   // Don't do anything clever with negative or misaligned offsets
2441   unsigned mask = (1 << shift) - 1;
2442   if (byte_offset < 0 || byte_offset & mask) {
2443     mov(Rd, byte_offset);
2444     add(Rd, base, Rd);
2445     return Address(Rd);
2446   }
2447 
2448   // See if we can do this with two 12-bit offsets
2449   {
2450     uint64_t word_offset = byte_offset >> shift;
2451     uint64_t masked_offset = word_offset & 0xfff000;
2452     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2453         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2454       add(Rd, base, masked_offset << shift);
2455       word_offset -= masked_offset;
2456       return Address(Rd, word_offset << shift);
2457     }
2458   }
2459 
2460   // Do it the hard way
2461   mov(Rd, byte_offset);
2462   add(Rd, base, Rd);
2463   return Address(Rd);
2464 }
2465 
2466 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2467                                     bool want_remainder, Register scratch)
2468 {
2469   // Full implementation of Java idiv and irem.  The function
2470   // returns the (pc) offset of the div instruction - may be needed
2471   // for implicit exceptions.
2472   //
2473   // constraint : ra/rb =/= scratch
2474   //         normal case
2475   //
2476   // input : ra: dividend
2477   //         rb: divisor
2478   //
2479   // result: either
2480   //         quotient  (= ra idiv rb)
2481   //         remainder (= ra irem rb)
2482 
2483   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2484 
2485   int idivl_offset = offset();
2486   if (! want_remainder) {
2487     sdivw(result, ra, rb);
2488   } else {
2489     sdivw(scratch, ra, rb);
2490     Assembler::msubw(result, scratch, rb, ra);
2491   }
2492 
2493   return idivl_offset;
2494 }
2495 
2496 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2497                                     bool want_remainder, Register scratch)
2498 {
2499   // Full implementation of Java ldiv and lrem.  The function
2500   // returns the (pc) offset of the div instruction - may be needed
2501   // for implicit exceptions.
2502   //
2503   // constraint : ra/rb =/= scratch
2504   //         normal case
2505   //
2506   // input : ra: dividend
2507   //         rb: divisor
2508   //
2509   // result: either
2510   //         quotient  (= ra idiv rb)
2511   //         remainder (= ra irem rb)
2512 
2513   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2514 
2515   int idivq_offset = offset();
2516   if (! want_remainder) {
2517     sdiv(result, ra, rb);
2518   } else {
2519     sdiv(scratch, ra, rb);
2520     Assembler::msub(result, scratch, rb, ra);
2521   }
2522 
2523   return idivq_offset;
2524 }
2525 
2526 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2527   address prev = pc() - NativeMembar::instruction_size;
2528   address last = code()->last_insn();
2529   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2530     NativeMembar *bar = NativeMembar_at(prev);
2531     if (AlwaysMergeDMB) {
2532       bar->set_kind(bar->get_kind() | order_constraint);
2533       BLOCK_COMMENT("merged membar(always)");
2534       return;
2535     }
2536     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2537     // doing so would introduce a StoreLoad which the caller did not
2538     // intend
2539     if (bar->get_kind() == order_constraint
2540         || bar->get_kind() == AnyAny
2541         || order_constraint == AnyAny) {
2542       // We are merging two memory barrier instructions.  On AArch64 we
2543       // can do this simply by ORing them together.
2544       bar->set_kind(bar->get_kind() | order_constraint);
2545       BLOCK_COMMENT("merged membar");
2546       return;
2547     } else {
2548       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2549       // We need check the last 2 instructions
2550       address prev2 = prev - NativeMembar::instruction_size;
2551       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2552         NativeMembar *bar2 = NativeMembar_at(prev2);
2553         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2554         BLOCK_COMMENT("merged membar(elided)");
2555         return;
2556       }
2557     }
2558   }
2559   code()->set_last_insn(pc());
2560   dmb(Assembler::barrier(order_constraint));
2561 }
2562 
2563 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2564   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2565     merge_ldst(rt, adr, size_in_bytes, is_store);
2566     code()->clear_last_insn();
2567     return true;
2568   } else {
2569     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2570     const uint64_t mask = size_in_bytes - 1;
2571     if (adr.getMode() == Address::base_plus_offset &&
2572         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2573       code()->set_last_insn(pc());
2574     }
2575     return false;
2576   }
2577 }
2578 
2579 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2580   // We always try to merge two adjacent loads into one ldp.
2581   if (!try_merge_ldst(Rx, adr, 8, false)) {
2582     Assembler::ldr(Rx, adr);
2583   }
2584 }
2585 
2586 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2587   // We always try to merge two adjacent loads into one ldp.
2588   if (!try_merge_ldst(Rw, adr, 4, false)) {
2589     Assembler::ldrw(Rw, adr);
2590   }
2591 }
2592 
2593 void MacroAssembler::str(Register Rx, const Address &adr) {
2594   // We always try to merge two adjacent stores into one stp.
2595   if (!try_merge_ldst(Rx, adr, 8, true)) {
2596     Assembler::str(Rx, adr);
2597   }
2598 }
2599 
2600 void MacroAssembler::strw(Register Rw, const Address &adr) {
2601   // We always try to merge two adjacent stores into one stp.
2602   if (!try_merge_ldst(Rw, adr, 4, true)) {
2603     Assembler::strw(Rw, adr);
2604   }
2605 }
2606 
2607 // MacroAssembler routines found actually to be needed
2608 
2609 void MacroAssembler::push(Register src)
2610 {
2611   str(src, Address(pre(esp, -1 * wordSize)));
2612 }
2613 
2614 void MacroAssembler::pop(Register dst)
2615 {
2616   ldr(dst, Address(post(esp, 1 * wordSize)));
2617 }
2618 
2619 // Note: load_unsigned_short used to be called load_unsigned_word.
2620 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2621   int off = offset();
2622   ldrh(dst, src);
2623   return off;
2624 }
2625 
2626 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2627   int off = offset();
2628   ldrb(dst, src);
2629   return off;
2630 }
2631 
2632 int MacroAssembler::load_signed_short(Register dst, Address src) {
2633   int off = offset();
2634   ldrsh(dst, src);
2635   return off;
2636 }
2637 
2638 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2639   int off = offset();
2640   ldrsb(dst, src);
2641   return off;
2642 }
2643 
2644 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2645   int off = offset();
2646   ldrshw(dst, src);
2647   return off;
2648 }
2649 
2650 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2651   int off = offset();
2652   ldrsbw(dst, src);
2653   return off;
2654 }
2655 
2656 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2657   switch (size_in_bytes) {
2658   case  8:  ldr(dst, src); break;
2659   case  4:  ldrw(dst, src); break;
2660   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2661   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2662   default:  ShouldNotReachHere();
2663   }
2664 }
2665 
2666 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2667   switch (size_in_bytes) {
2668   case  8:  str(src, dst); break;
2669   case  4:  strw(src, dst); break;
2670   case  2:  strh(src, dst); break;
2671   case  1:  strb(src, dst); break;
2672   default:  ShouldNotReachHere();
2673   }
2674 }
2675 
2676 void MacroAssembler::decrementw(Register reg, int value)
2677 {
2678   if (value < 0)  { incrementw(reg, -value);      return; }
2679   if (value == 0) {                               return; }
2680   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2681   /* else */ {
2682     guarantee(reg != rscratch2, "invalid dst for register decrement");
2683     movw(rscratch2, (unsigned)value);
2684     subw(reg, reg, rscratch2);
2685   }
2686 }
2687 
2688 void MacroAssembler::decrement(Register reg, int value)
2689 {
2690   if (value < 0)  { increment(reg, -value);      return; }
2691   if (value == 0) {                              return; }
2692   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2693   /* else */ {
2694     assert(reg != rscratch2, "invalid dst for register decrement");
2695     mov(rscratch2, (uint64_t)value);
2696     sub(reg, reg, rscratch2);
2697   }
2698 }
2699 
2700 void MacroAssembler::decrementw(Address dst, int value)
2701 {
2702   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2703   if (dst.getMode() == Address::literal) {
2704     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2705     lea(rscratch2, dst);
2706     dst = Address(rscratch2);
2707   }
2708   ldrw(rscratch1, dst);
2709   decrementw(rscratch1, value);
2710   strw(rscratch1, dst);
2711 }
2712 
2713 void MacroAssembler::decrement(Address dst, int value)
2714 {
2715   assert(!dst.uses(rscratch1), "invalid address for decrement");
2716   if (dst.getMode() == Address::literal) {
2717     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2718     lea(rscratch2, dst);
2719     dst = Address(rscratch2);
2720   }
2721   ldr(rscratch1, dst);
2722   decrement(rscratch1, value);
2723   str(rscratch1, dst);
2724 }
2725 
2726 void MacroAssembler::incrementw(Register reg, int value)
2727 {
2728   if (value < 0)  { decrementw(reg, -value);      return; }
2729   if (value == 0) {                               return; }
2730   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2731   /* else */ {
2732     assert(reg != rscratch2, "invalid dst for register increment");
2733     movw(rscratch2, (unsigned)value);
2734     addw(reg, reg, rscratch2);
2735   }
2736 }
2737 
2738 void MacroAssembler::increment(Register reg, int value)
2739 {
2740   if (value < 0)  { decrement(reg, -value);      return; }
2741   if (value == 0) {                              return; }
2742   if (value < (1 << 12)) { add(reg, reg, value); return; }
2743   /* else */ {
2744     assert(reg != rscratch2, "invalid dst for register increment");
2745     movw(rscratch2, (unsigned)value);
2746     add(reg, reg, rscratch2);
2747   }
2748 }
2749 
2750 void MacroAssembler::incrementw(Address dst, int value)
2751 {
2752   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2753   if (dst.getMode() == Address::literal) {
2754     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2755     lea(rscratch2, dst);
2756     dst = Address(rscratch2);
2757   }
2758   ldrw(rscratch1, dst);
2759   incrementw(rscratch1, value);
2760   strw(rscratch1, dst);
2761 }
2762 
2763 void MacroAssembler::increment(Address dst, int value)
2764 {
2765   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2766   if (dst.getMode() == Address::literal) {
2767     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2768     lea(rscratch2, dst);
2769     dst = Address(rscratch2);
2770   }
2771   ldr(rscratch1, dst);
2772   increment(rscratch1, value);
2773   str(rscratch1, dst);
2774 }
2775 
2776 // Push lots of registers in the bit set supplied.  Don't push sp.
2777 // Return the number of words pushed
2778 int MacroAssembler::push(unsigned int bitset, Register stack) {
2779   int words_pushed = 0;
2780 
2781   // Scan bitset to accumulate register pairs
2782   unsigned char regs[32];
2783   int count = 0;
2784   for (int reg = 0; reg <= 30; reg++) {
2785     if (1 & bitset)
2786       regs[count++] = reg;
2787     bitset >>= 1;
2788   }
2789   regs[count++] = zr->raw_encoding();
2790   count &= ~1;  // Only push an even number of regs
2791 
2792   if (count) {
2793     stp(as_Register(regs[0]), as_Register(regs[1]),
2794        Address(pre(stack, -count * wordSize)));
2795     words_pushed += 2;
2796   }
2797   for (int i = 2; i < count; i += 2) {
2798     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2799        Address(stack, i * wordSize));
2800     words_pushed += 2;
2801   }
2802 
2803   assert(words_pushed == count, "oops, pushed != count");
2804 
2805   return count;
2806 }
2807 
2808 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2809   int words_pushed = 0;
2810 
2811   // Scan bitset to accumulate register pairs
2812   unsigned char regs[32];
2813   int count = 0;
2814   for (int reg = 0; reg <= 30; reg++) {
2815     if (1 & bitset)
2816       regs[count++] = reg;
2817     bitset >>= 1;
2818   }
2819   regs[count++] = zr->raw_encoding();
2820   count &= ~1;
2821 
2822   for (int i = 2; i < count; i += 2) {
2823     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2824        Address(stack, i * wordSize));
2825     words_pushed += 2;
2826   }
2827   if (count) {
2828     ldp(as_Register(regs[0]), as_Register(regs[1]),
2829        Address(post(stack, count * wordSize)));
2830     words_pushed += 2;
2831   }
2832 
2833   assert(words_pushed == count, "oops, pushed != count");
2834 
2835   return count;
2836 }
2837 
2838 // Push lots of registers in the bit set supplied.  Don't push sp.
2839 // Return the number of dwords pushed
2840 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2841   int words_pushed = 0;
2842   bool use_sve = false;
2843   int sve_vector_size_in_bytes = 0;
2844 
2845 #ifdef COMPILER2
2846   use_sve = Matcher::supports_scalable_vector();
2847   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2848 #endif
2849 
2850   // Scan bitset to accumulate register pairs
2851   unsigned char regs[32];
2852   int count = 0;
2853   for (int reg = 0; reg <= 31; reg++) {
2854     if (1 & bitset)
2855       regs[count++] = reg;
2856     bitset >>= 1;
2857   }
2858 
2859   if (count == 0) {
2860     return 0;
2861   }
2862 
2863   if (mode == PushPopFull) {
2864     if (use_sve && sve_vector_size_in_bytes > 16) {
2865       mode = PushPopSVE;
2866     } else {
2867       mode = PushPopNeon;
2868     }
2869   }
2870 
2871 #ifndef PRODUCT
2872   {
2873     char buffer[48];
2874     if (mode == PushPopSVE) {
2875       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
2876     } else if (mode == PushPopNeon) {
2877       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
2878     } else {
2879       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
2880     }
2881     block_comment(buffer);
2882   }
2883 #endif
2884 
2885   if (mode == PushPopSVE) {
2886     sub(stack, stack, sve_vector_size_in_bytes * count);
2887     for (int i = 0; i < count; i++) {
2888       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
2889     }
2890     return count * sve_vector_size_in_bytes / 8;
2891   }
2892 
2893   if (mode == PushPopNeon) {
2894     if (count == 1) {
2895       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
2896       return 2;
2897     }
2898 
2899     bool odd = (count & 1) == 1;
2900     int push_slots = count + (odd ? 1 : 0);
2901 
2902     // Always pushing full 128 bit registers.
2903     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
2904     words_pushed += 2;
2905 
2906     for (int i = 2; i + 1 < count; i += 2) {
2907       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
2908       words_pushed += 2;
2909     }
2910 
2911     if (odd) {
2912       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
2913       words_pushed++;
2914     }
2915 
2916     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
2917     return count * 2;
2918   }
2919 
2920   if (mode == PushPopFp) {
2921     bool odd = (count & 1) == 1;
2922     int push_slots = count + (odd ? 1 : 0);
2923 
2924     if (count == 1) {
2925       // Stack pointer must be 16 bytes aligned
2926       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
2927       return 1;
2928     }
2929 
2930     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
2931     words_pushed += 2;
2932 
2933     for (int i = 2; i + 1 < count; i += 2) {
2934       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
2935       words_pushed += 2;
2936     }
2937 
2938     if (odd) {
2939       // Stack pointer must be 16 bytes aligned
2940       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
2941       words_pushed++;
2942     }
2943 
2944     assert(words_pushed == count, "oops, pushed != count");
2945 
2946     return count;
2947   }
2948 
2949   return 0;
2950 }
2951 
2952 // Return the number of dwords popped
2953 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2954   int words_pushed = 0;
2955   bool use_sve = false;
2956   int sve_vector_size_in_bytes = 0;
2957 
2958 #ifdef COMPILER2
2959   use_sve = Matcher::supports_scalable_vector();
2960   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2961 #endif
2962   // Scan bitset to accumulate register pairs
2963   unsigned char regs[32];
2964   int count = 0;
2965   for (int reg = 0; reg <= 31; reg++) {
2966     if (1 & bitset)
2967       regs[count++] = reg;
2968     bitset >>= 1;
2969   }
2970 
2971   if (count == 0) {
2972     return 0;
2973   }
2974 
2975   if (mode == PushPopFull) {
2976     if (use_sve && sve_vector_size_in_bytes > 16) {
2977       mode = PushPopSVE;
2978     } else {
2979       mode = PushPopNeon;
2980     }
2981   }
2982 
2983 #ifndef PRODUCT
2984   {
2985     char buffer[48];
2986     if (mode == PushPopSVE) {
2987       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
2988     } else if (mode == PushPopNeon) {
2989       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
2990     } else {
2991       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
2992     }
2993     block_comment(buffer);
2994   }
2995 #endif
2996 
2997   if (mode == PushPopSVE) {
2998     for (int i = count - 1; i >= 0; i--) {
2999       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3000     }
3001     add(stack, stack, sve_vector_size_in_bytes * count);
3002     return count * sve_vector_size_in_bytes / 8;
3003   }
3004 
3005   if (mode == PushPopNeon) {
3006     if (count == 1) {
3007       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3008       return 2;
3009     }
3010 
3011     bool odd = (count & 1) == 1;
3012     int push_slots = count + (odd ? 1 : 0);
3013 
3014     if (odd) {
3015       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3016       words_pushed++;
3017     }
3018 
3019     for (int i = 2; i + 1 < count; i += 2) {
3020       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3021       words_pushed += 2;
3022     }
3023 
3024     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3025     words_pushed += 2;
3026 
3027     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3028 
3029     return count * 2;
3030   }
3031 
3032   if (mode == PushPopFp) {
3033     bool odd = (count & 1) == 1;
3034     int push_slots = count + (odd ? 1 : 0);
3035 
3036     if (count == 1) {
3037       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3038       return 1;
3039     }
3040 
3041     if (odd) {
3042       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3043       words_pushed++;
3044     }
3045 
3046     for (int i = 2; i + 1 < count; i += 2) {
3047       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3048       words_pushed += 2;
3049     }
3050 
3051     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3052     words_pushed += 2;
3053 
3054     assert(words_pushed == count, "oops, pushed != count");
3055 
3056     return count;
3057   }
3058 
3059   return 0;
3060 }
3061 
3062 // Return the number of dwords pushed
3063 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3064   bool use_sve = false;
3065   int sve_predicate_size_in_slots = 0;
3066 
3067 #ifdef COMPILER2
3068   use_sve = Matcher::supports_scalable_vector();
3069   if (use_sve) {
3070     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3071   }
3072 #endif
3073 
3074   if (!use_sve) {
3075     return 0;
3076   }
3077 
3078   unsigned char regs[PRegister::number_of_registers];
3079   int count = 0;
3080   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3081     if (1 & bitset)
3082       regs[count++] = reg;
3083     bitset >>= 1;
3084   }
3085 
3086   if (count == 0) {
3087     return 0;
3088   }
3089 
3090   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3091                                   VMRegImpl::stack_slot_size * count, 16);
3092   sub(stack, stack, total_push_bytes);
3093   for (int i = 0; i < count; i++) {
3094     sve_str(as_PRegister(regs[i]), Address(stack, i));
3095   }
3096   return total_push_bytes / 8;
3097 }
3098 
3099 // Return the number of dwords popped
3100 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3101   bool use_sve = false;
3102   int sve_predicate_size_in_slots = 0;
3103 
3104 #ifdef COMPILER2
3105   use_sve = Matcher::supports_scalable_vector();
3106   if (use_sve) {
3107     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3108   }
3109 #endif
3110 
3111   if (!use_sve) {
3112     return 0;
3113   }
3114 
3115   unsigned char regs[PRegister::number_of_registers];
3116   int count = 0;
3117   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3118     if (1 & bitset)
3119       regs[count++] = reg;
3120     bitset >>= 1;
3121   }
3122 
3123   if (count == 0) {
3124     return 0;
3125   }
3126 
3127   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3128                                  VMRegImpl::stack_slot_size * count, 16);
3129   for (int i = count - 1; i >= 0; i--) {
3130     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3131   }
3132   add(stack, stack, total_pop_bytes);
3133   return total_pop_bytes / 8;
3134 }
3135 
3136 #ifdef ASSERT
3137 void MacroAssembler::verify_heapbase(const char* msg) {
3138 #if 0
3139   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3140   assert (Universe::heap() != nullptr, "java heap should be initialized");
3141   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3142     // rheapbase is allocated as general register
3143     return;
3144   }
3145   if (CheckCompressedOops) {
3146     Label ok;
3147     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3148     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3149     br(Assembler::EQ, ok);
3150     stop(msg);
3151     bind(ok);
3152     pop(1 << rscratch1->encoding(), sp);
3153   }
3154 #endif
3155 }
3156 #endif
3157 
3158 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3159   assert_different_registers(value, tmp1, tmp2);
3160   Label done, tagged, weak_tagged;
3161 
3162   cbz(value, done);           // Use null as-is.
3163   tst(value, JNIHandles::tag_mask); // Test for tag.
3164   br(Assembler::NE, tagged);
3165 
3166   // Resolve local handle
3167   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3168   verify_oop(value);
3169   b(done);
3170 
3171   bind(tagged);
3172   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3173   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3174 
3175   // Resolve global handle
3176   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3177   verify_oop(value);
3178   b(done);
3179 
3180   bind(weak_tagged);
3181   // Resolve jweak.
3182   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3183                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3184   verify_oop(value);
3185 
3186   bind(done);
3187 }
3188 
3189 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3190   assert_different_registers(value, tmp1, tmp2);
3191   Label done;
3192 
3193   cbz(value, done);           // Use null as-is.
3194 
3195 #ifdef ASSERT
3196   {
3197     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3198     Label valid_global_tag;
3199     tbnz(value, 1, valid_global_tag); // Test for global tag
3200     stop("non global jobject using resolve_global_jobject");
3201     bind(valid_global_tag);
3202   }
3203 #endif
3204 
3205   // Resolve global handle
3206   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3207   verify_oop(value);
3208 
3209   bind(done);
3210 }
3211 
3212 void MacroAssembler::stop(const char* msg) {
3213   // Skip AOT caching C strings in scratch buffer.
3214   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3215   BLOCK_COMMENT(str);
3216   // load msg into r0 so we can access it from the signal handler
3217   // ExternalAddress enables saving and restoring via the code cache
3218   lea(c_rarg0, ExternalAddress((address) str));
3219   dcps1(0xdeae);
3220 }
3221 
3222 void MacroAssembler::unimplemented(const char* what) {
3223   const char* buf = nullptr;
3224   {
3225     ResourceMark rm;
3226     stringStream ss;
3227     ss.print("unimplemented: %s", what);
3228     buf = code_string(ss.as_string());
3229   }
3230   stop(buf);
3231 }
3232 
3233 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3234 #ifdef ASSERT
3235   Label OK;
3236   br(cc, OK);
3237   stop(msg);
3238   bind(OK);
3239 #endif
3240 }
3241 
3242 // If a constant does not fit in an immediate field, generate some
3243 // number of MOV instructions and then perform the operation.
3244 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3245                                            add_sub_imm_insn insn1,
3246                                            add_sub_reg_insn insn2,
3247                                            bool is32) {
3248   assert(Rd != zr, "Rd = zr and not setting flags?");
3249   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3250   if (fits) {
3251     (this->*insn1)(Rd, Rn, imm);
3252   } else {
3253     if (g_uabs(imm) < (1 << 24)) {
3254        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3255        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3256     } else {
3257        assert_different_registers(Rd, Rn);
3258        mov(Rd, imm);
3259        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3260     }
3261   }
3262 }
3263 
3264 // Separate vsn which sets the flags. Optimisations are more restricted
3265 // because we must set the flags correctly.
3266 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3267                                              add_sub_imm_insn insn1,
3268                                              add_sub_reg_insn insn2,
3269                                              bool is32) {
3270   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3271   if (fits) {
3272     (this->*insn1)(Rd, Rn, imm);
3273   } else {
3274     assert_different_registers(Rd, Rn);
3275     assert(Rd != zr, "overflow in immediate operand");
3276     mov(Rd, imm);
3277     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3278   }
3279 }
3280 
3281 
3282 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3283   if (increment.is_register()) {
3284     add(Rd, Rn, increment.as_register());
3285   } else {
3286     add(Rd, Rn, increment.as_constant());
3287   }
3288 }
3289 
3290 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3291   if (increment.is_register()) {
3292     addw(Rd, Rn, increment.as_register());
3293   } else {
3294     addw(Rd, Rn, increment.as_constant());
3295   }
3296 }
3297 
3298 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3299   if (decrement.is_register()) {
3300     sub(Rd, Rn, decrement.as_register());
3301   } else {
3302     sub(Rd, Rn, decrement.as_constant());
3303   }
3304 }
3305 
3306 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3307   if (decrement.is_register()) {
3308     subw(Rd, Rn, decrement.as_register());
3309   } else {
3310     subw(Rd, Rn, decrement.as_constant());
3311   }
3312 }
3313 
3314 void MacroAssembler::reinit_heapbase()
3315 {
3316   if (UseCompressedOops) {
3317     if (Universe::is_fully_initialized()) {
3318       mov(rheapbase, CompressedOops::base());
3319     } else {
3320       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3321       ldr(rheapbase, Address(rheapbase));
3322     }
3323   }
3324 }
3325 
3326 // this simulates the behaviour of the x86 cmpxchg instruction using a
3327 // load linked/store conditional pair. we use the acquire/release
3328 // versions of these instructions so that we flush pending writes as
3329 // per Java semantics.
3330 
3331 // n.b the x86 version assumes the old value to be compared against is
3332 // in rax and updates rax with the value located in memory if the
3333 // cmpxchg fails. we supply a register for the old value explicitly
3334 
3335 // the aarch64 load linked/store conditional instructions do not
3336 // accept an offset. so, unlike x86, we must provide a plain register
3337 // to identify the memory word to be compared/exchanged rather than a
3338 // register+offset Address.
3339 
3340 void MacroAssembler::cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
3341                                 Label &succeed, Label *fail) {
3342   // oldv holds comparison value
3343   // newv holds value to write in exchange
3344   // addr identifies memory word to compare against/update
3345   if (UseLSE) {
3346     mov(tmp, oldv);
3347     casal(Assembler::xword, oldv, newv, addr);
3348     cmp(tmp, oldv);
3349     br(Assembler::EQ, succeed);
3350     membar(AnyAny);
3351   } else {
3352     Label retry_load, nope;
3353     prfm(Address(addr), PSTL1STRM);
3354     bind(retry_load);
3355     // flush and load exclusive from the memory location
3356     // and fail if it is not what we expect
3357     ldaxr(tmp, addr);
3358     cmp(tmp, oldv);
3359     br(Assembler::NE, nope);
3360     // if we store+flush with no intervening write tmp will be zero
3361     stlxr(tmp, newv, addr);
3362     cbzw(tmp, succeed);
3363     // retry so we only ever return after a load fails to compare
3364     // ensures we don't return a stale value after a failed write.
3365     b(retry_load);
3366     // if the memory word differs we return it in oldv and signal a fail
3367     bind(nope);
3368     membar(AnyAny);
3369     mov(oldv, tmp);
3370   }
3371   if (fail)
3372     b(*fail);
3373 }
3374 
3375 void MacroAssembler::cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp,
3376                                         Label &succeed, Label *fail) {
3377   assert(oopDesc::mark_offset_in_bytes() == 0, "assumption");
3378   cmpxchgptr(oldv, newv, obj, tmp, succeed, fail);
3379 }
3380 
3381 void MacroAssembler::cmpxchgw(Register oldv, Register newv, Register addr, Register tmp,
3382                                 Label &succeed, Label *fail) {
3383   // oldv holds comparison value
3384   // newv holds value to write in exchange
3385   // addr identifies memory word to compare against/update
3386   // tmp returns 0/1 for success/failure
3387   if (UseLSE) {
3388     mov(tmp, oldv);
3389     casal(Assembler::word, oldv, newv, addr);
3390     cmp(tmp, oldv);
3391     br(Assembler::EQ, succeed);
3392     membar(AnyAny);
3393   } else {
3394     Label retry_load, nope;
3395     prfm(Address(addr), PSTL1STRM);
3396     bind(retry_load);
3397     // flush and load exclusive from the memory location
3398     // and fail if it is not what we expect
3399     ldaxrw(tmp, addr);
3400     cmp(tmp, oldv);
3401     br(Assembler::NE, nope);
3402     // if we store+flush with no intervening write tmp will be zero
3403     stlxrw(tmp, newv, addr);
3404     cbzw(tmp, succeed);
3405     // retry so we only ever return after a load fails to compare
3406     // ensures we don't return a stale value after a failed write.
3407     b(retry_load);
3408     // if the memory word differs we return it in oldv and signal a fail
3409     bind(nope);
3410     membar(AnyAny);
3411     mov(oldv, tmp);
3412   }
3413   if (fail)
3414     b(*fail);
3415 }
3416 
3417 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3418 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3419 // Pass a register for the result, otherwise pass noreg.
3420 
3421 // Clobbers rscratch1
3422 void MacroAssembler::cmpxchg(Register addr, Register expected,
3423                              Register new_val,
3424                              enum operand_size size,
3425                              bool acquire, bool release,
3426                              bool weak,
3427                              Register result) {
3428   if (result == noreg)  result = rscratch1;
3429   BLOCK_COMMENT("cmpxchg {");
3430   if (UseLSE) {
3431     mov(result, expected);
3432     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3433     compare_eq(result, expected, size);
3434 #ifdef ASSERT
3435     // Poison rscratch1 which is written on !UseLSE branch
3436     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3437 #endif
3438   } else {
3439     Label retry_load, done;
3440     prfm(Address(addr), PSTL1STRM);
3441     bind(retry_load);
3442     load_exclusive(result, addr, size, acquire);
3443     compare_eq(result, expected, size);
3444     br(Assembler::NE, done);
3445     store_exclusive(rscratch1, new_val, addr, size, release);
3446     if (weak) {
3447       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3448     } else {
3449       cbnzw(rscratch1, retry_load);
3450     }
3451     bind(done);
3452   }
3453   BLOCK_COMMENT("} cmpxchg");
3454 }
3455 
3456 // A generic comparison. Only compares for equality, clobbers rscratch1.
3457 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3458   if (size == xword) {
3459     cmp(rm, rn);
3460   } else if (size == word) {
3461     cmpw(rm, rn);
3462   } else if (size == halfword) {
3463     eorw(rscratch1, rm, rn);
3464     ands(zr, rscratch1, 0xffff);
3465   } else if (size == byte) {
3466     eorw(rscratch1, rm, rn);
3467     ands(zr, rscratch1, 0xff);
3468   } else {
3469     ShouldNotReachHere();
3470   }
3471 }
3472 
3473 
3474 static bool different(Register a, RegisterOrConstant b, Register c) {
3475   if (b.is_constant())
3476     return a != c;
3477   else
3478     return a != b.as_register() && a != c && b.as_register() != c;
3479 }
3480 
3481 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3482 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3483   if (UseLSE) {                                                         \
3484     prev = prev->is_valid() ? prev : zr;                                \
3485     if (incr.is_register()) {                                           \
3486       AOP(sz, incr.as_register(), prev, addr);                          \
3487     } else {                                                            \
3488       mov(rscratch2, incr.as_constant());                               \
3489       AOP(sz, rscratch2, prev, addr);                                   \
3490     }                                                                   \
3491     return;                                                             \
3492   }                                                                     \
3493   Register result = rscratch2;                                          \
3494   if (prev->is_valid())                                                 \
3495     result = different(prev, incr, addr) ? prev : rscratch2;            \
3496                                                                         \
3497   Label retry_load;                                                     \
3498   prfm(Address(addr), PSTL1STRM);                                       \
3499   bind(retry_load);                                                     \
3500   LDXR(result, addr);                                                   \
3501   OP(rscratch1, result, incr);                                          \
3502   STXR(rscratch2, rscratch1, addr);                                     \
3503   cbnzw(rscratch2, retry_load);                                         \
3504   if (prev->is_valid() && prev != result) {                             \
3505     IOP(prev, rscratch1, incr);                                         \
3506   }                                                                     \
3507 }
3508 
3509 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3510 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3511 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3512 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3513 
3514 #undef ATOMIC_OP
3515 
3516 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3517 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3518   if (UseLSE) {                                                         \
3519     prev = prev->is_valid() ? prev : zr;                                \
3520     AOP(sz, newv, prev, addr);                                          \
3521     return;                                                             \
3522   }                                                                     \
3523   Register result = rscratch2;                                          \
3524   if (prev->is_valid())                                                 \
3525     result = different(prev, newv, addr) ? prev : rscratch2;            \
3526                                                                         \
3527   Label retry_load;                                                     \
3528   prfm(Address(addr), PSTL1STRM);                                       \
3529   bind(retry_load);                                                     \
3530   LDXR(result, addr);                                                   \
3531   STXR(rscratch1, newv, addr);                                          \
3532   cbnzw(rscratch1, retry_load);                                         \
3533   if (prev->is_valid() && prev != result)                               \
3534     mov(prev, result);                                                  \
3535 }
3536 
3537 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3538 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3539 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3540 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3541 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3542 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3543 
3544 #undef ATOMIC_XCHG
3545 
3546 #ifndef PRODUCT
3547 extern "C" void findpc(intptr_t x);
3548 #endif
3549 
3550 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3551 {
3552   // In order to get locks to work, we need to fake a in_VM state
3553   if (ShowMessageBoxOnError ) {
3554     JavaThread* thread = JavaThread::current();
3555     JavaThreadState saved_state = thread->thread_state();
3556     thread->set_thread_state(_thread_in_vm);
3557 #ifndef PRODUCT
3558     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3559       ttyLocker ttyl;
3560       BytecodeCounter::print();
3561     }
3562 #endif
3563     if (os::message_box(msg, "Execution stopped, print registers?")) {
3564       ttyLocker ttyl;
3565       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3566 #ifndef PRODUCT
3567       tty->cr();
3568       findpc(pc);
3569       tty->cr();
3570 #endif
3571       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3572       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3573       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3574       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3575       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3576       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3577       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3578       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3579       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3580       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3581       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3582       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3583       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3584       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3585       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3586       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3587       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3588       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3589       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3590       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3591       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3592       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3593       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3594       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3595       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3596       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3597       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3598       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3599       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3600       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3601       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3602       BREAKPOINT;
3603     }
3604   }
3605   fatal("DEBUG MESSAGE: %s", msg);
3606 }
3607 
3608 RegSet MacroAssembler::call_clobbered_gp_registers() {
3609   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3610 #ifndef R18_RESERVED
3611   regs += r18_tls;
3612 #endif
3613   return regs;
3614 }
3615 
3616 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3617   int step = 4 * wordSize;
3618   push(call_clobbered_gp_registers() - exclude, sp);
3619   sub(sp, sp, step);
3620   mov(rscratch1, -step);
3621   // Push v0-v7, v16-v31.
3622   for (int i = 31; i>= 4; i -= 4) {
3623     if (i <= v7->encoding() || i >= v16->encoding())
3624       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3625           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3626   }
3627   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3628       as_FloatRegister(3), T1D, Address(sp));
3629 }
3630 
3631 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3632   for (int i = 0; i < 32; i += 4) {
3633     if (i <= v7->encoding() || i >= v16->encoding())
3634       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3635           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3636   }
3637 
3638   reinitialize_ptrue();
3639 
3640   pop(call_clobbered_gp_registers() - exclude, sp);
3641 }
3642 
3643 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3644                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3645   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3646   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3647     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3648     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3649       sve_str(as_FloatRegister(i), Address(sp, i));
3650     }
3651   } else {
3652     int step = (save_vectors ? 8 : 4) * wordSize;
3653     mov(rscratch1, -step);
3654     sub(sp, sp, step);
3655     for (int i = 28; i >= 4; i -= 4) {
3656       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3657           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3658     }
3659     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3660   }
3661   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3662     sub(sp, sp, total_predicate_in_bytes);
3663     for (int i = 0; i < PRegister::number_of_registers; i++) {
3664       sve_str(as_PRegister(i), Address(sp, i));
3665     }
3666   }
3667 }
3668 
3669 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3670                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3671   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3672     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3673       sve_ldr(as_PRegister(i), Address(sp, i));
3674     }
3675     add(sp, sp, total_predicate_in_bytes);
3676   }
3677   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3678     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3679       sve_ldr(as_FloatRegister(i), Address(sp, i));
3680     }
3681     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3682   } else {
3683     int step = (restore_vectors ? 8 : 4) * wordSize;
3684     for (int i = 0; i <= 28; i += 4)
3685       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3686           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3687   }
3688 
3689   // We may use predicate registers and rely on ptrue with SVE,
3690   // regardless of wide vector (> 8 bytes) used or not.
3691   if (use_sve) {
3692     reinitialize_ptrue();
3693   }
3694 
3695   // integer registers except lr & sp
3696   pop(RegSet::range(r0, r17), sp);
3697 #ifdef R18_RESERVED
3698   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3699   pop(RegSet::range(r20, r29), sp);
3700 #else
3701   pop(RegSet::range(r18_tls, r29), sp);
3702 #endif
3703 }
3704 
3705 /**
3706  * Helpers for multiply_to_len().
3707  */
3708 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3709                                      Register src1, Register src2) {
3710   adds(dest_lo, dest_lo, src1);
3711   adc(dest_hi, dest_hi, zr);
3712   adds(dest_lo, dest_lo, src2);
3713   adc(final_dest_hi, dest_hi, zr);
3714 }
3715 
3716 // Generate an address from (r + r1 extend offset).  "size" is the
3717 // size of the operand.  The result may be in rscratch2.
3718 Address MacroAssembler::offsetted_address(Register r, Register r1,
3719                                           Address::extend ext, int offset, int size) {
3720   if (offset || (ext.shift() % size != 0)) {
3721     lea(rscratch2, Address(r, r1, ext));
3722     return Address(rscratch2, offset);
3723   } else {
3724     return Address(r, r1, ext);
3725   }
3726 }
3727 
3728 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3729 {
3730   assert(offset >= 0, "spill to negative address?");
3731   // Offset reachable ?
3732   //   Not aligned - 9 bits signed offset
3733   //   Aligned - 12 bits unsigned offset shifted
3734   Register base = sp;
3735   if ((offset & (size-1)) && offset >= (1<<8)) {
3736     add(tmp, base, offset & ((1<<12)-1));
3737     base = tmp;
3738     offset &= -1u<<12;
3739   }
3740 
3741   if (offset >= (1<<12) * size) {
3742     add(tmp, base, offset & (((1<<12)-1)<<12));
3743     base = tmp;
3744     offset &= ~(((1<<12)-1)<<12);
3745   }
3746 
3747   return Address(base, offset);
3748 }
3749 
3750 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3751   assert(offset >= 0, "spill to negative address?");
3752 
3753   Register base = sp;
3754 
3755   // An immediate offset in the range 0 to 255 which is multiplied
3756   // by the current vector or predicate register size in bytes.
3757   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3758     return Address(base, offset / sve_reg_size_in_bytes);
3759   }
3760 
3761   add(tmp, base, offset);
3762   return Address(tmp);
3763 }
3764 
3765 // Checks whether offset is aligned.
3766 // Returns true if it is, else false.
3767 bool MacroAssembler::merge_alignment_check(Register base,
3768                                            size_t size,
3769                                            int64_t cur_offset,
3770                                            int64_t prev_offset) const {
3771   if (AvoidUnalignedAccesses) {
3772     if (base == sp) {
3773       // Checks whether low offset if aligned to pair of registers.
3774       int64_t pair_mask = size * 2 - 1;
3775       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3776       return (offset & pair_mask) == 0;
3777     } else { // If base is not sp, we can't guarantee the access is aligned.
3778       return false;
3779     }
3780   } else {
3781     int64_t mask = size - 1;
3782     // Load/store pair instruction only supports element size aligned offset.
3783     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3784   }
3785 }
3786 
3787 // Checks whether current and previous loads/stores can be merged.
3788 // Returns true if it can be merged, else false.
3789 bool MacroAssembler::ldst_can_merge(Register rt,
3790                                     const Address &adr,
3791                                     size_t cur_size_in_bytes,
3792                                     bool is_store) const {
3793   address prev = pc() - NativeInstruction::instruction_size;
3794   address last = code()->last_insn();
3795 
3796   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3797     return false;
3798   }
3799 
3800   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3801     return false;
3802   }
3803 
3804   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3805   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3806 
3807   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3808   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3809 
3810   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3811     return false;
3812   }
3813 
3814   int64_t max_offset = 63 * prev_size_in_bytes;
3815   int64_t min_offset = -64 * prev_size_in_bytes;
3816 
3817   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3818 
3819   // Only same base can be merged.
3820   if (adr.base() != prev_ldst->base()) {
3821     return false;
3822   }
3823 
3824   int64_t cur_offset = adr.offset();
3825   int64_t prev_offset = prev_ldst->offset();
3826   size_t diff = abs(cur_offset - prev_offset);
3827   if (diff != prev_size_in_bytes) {
3828     return false;
3829   }
3830 
3831   // Following cases can not be merged:
3832   // ldr x2, [x2, #8]
3833   // ldr x3, [x2, #16]
3834   // or:
3835   // ldr x2, [x3, #8]
3836   // ldr x2, [x3, #16]
3837   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3838   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3839     return false;
3840   }
3841 
3842   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3843   // Offset range must be in ldp/stp instruction's range.
3844   if (low_offset > max_offset || low_offset < min_offset) {
3845     return false;
3846   }
3847 
3848   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3849     return true;
3850   }
3851 
3852   return false;
3853 }
3854 
3855 // Merge current load/store with previous load/store into ldp/stp.
3856 void MacroAssembler::merge_ldst(Register rt,
3857                                 const Address &adr,
3858                                 size_t cur_size_in_bytes,
3859                                 bool is_store) {
3860 
3861   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3862 
3863   Register rt_low, rt_high;
3864   address prev = pc() - NativeInstruction::instruction_size;
3865   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3866 
3867   int64_t offset;
3868 
3869   if (adr.offset() < prev_ldst->offset()) {
3870     offset = adr.offset();
3871     rt_low = rt;
3872     rt_high = prev_ldst->target();
3873   } else {
3874     offset = prev_ldst->offset();
3875     rt_low = prev_ldst->target();
3876     rt_high = rt;
3877   }
3878 
3879   Address adr_p = Address(prev_ldst->base(), offset);
3880   // Overwrite previous generated binary.
3881   code_section()->set_end(prev);
3882 
3883   const size_t sz = prev_ldst->size_in_bytes();
3884   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3885   if (!is_store) {
3886     BLOCK_COMMENT("merged ldr pair");
3887     if (sz == 8) {
3888       ldp(rt_low, rt_high, adr_p);
3889     } else {
3890       ldpw(rt_low, rt_high, adr_p);
3891     }
3892   } else {
3893     BLOCK_COMMENT("merged str pair");
3894     if (sz == 8) {
3895       stp(rt_low, rt_high, adr_p);
3896     } else {
3897       stpw(rt_low, rt_high, adr_p);
3898     }
3899   }
3900 }
3901 
3902 /**
3903  * Multiply 64 bit by 64 bit first loop.
3904  */
3905 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3906                                            Register y, Register y_idx, Register z,
3907                                            Register carry, Register product,
3908                                            Register idx, Register kdx) {
3909   //
3910   //  jlong carry, x[], y[], z[];
3911   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3912   //    huge_128 product = y[idx] * x[xstart] + carry;
3913   //    z[kdx] = (jlong)product;
3914   //    carry  = (jlong)(product >>> 64);
3915   //  }
3916   //  z[xstart] = carry;
3917   //
3918 
3919   Label L_first_loop, L_first_loop_exit;
3920   Label L_one_x, L_one_y, L_multiply;
3921 
3922   subsw(xstart, xstart, 1);
3923   br(Assembler::MI, L_one_x);
3924 
3925   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3926   ldr(x_xstart, Address(rscratch1));
3927   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3928 
3929   bind(L_first_loop);
3930   subsw(idx, idx, 1);
3931   br(Assembler::MI, L_first_loop_exit);
3932   subsw(idx, idx, 1);
3933   br(Assembler::MI, L_one_y);
3934   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3935   ldr(y_idx, Address(rscratch1));
3936   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
3937   bind(L_multiply);
3938 
3939   // AArch64 has a multiply-accumulate instruction that we can't use
3940   // here because it has no way to process carries, so we have to use
3941   // separate add and adc instructions.  Bah.
3942   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
3943   mul(product, x_xstart, y_idx);
3944   adds(product, product, carry);
3945   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
3946 
3947   subw(kdx, kdx, 2);
3948   ror(product, product, 32); // back to big-endian
3949   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
3950 
3951   b(L_first_loop);
3952 
3953   bind(L_one_y);
3954   ldrw(y_idx, Address(y,  0));
3955   b(L_multiply);
3956 
3957   bind(L_one_x);
3958   ldrw(x_xstart, Address(x,  0));
3959   b(L_first_loop);
3960 
3961   bind(L_first_loop_exit);
3962 }
3963 
3964 /**
3965  * Multiply 128 bit by 128. Unrolled inner loop.
3966  *
3967  */
3968 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
3969                                              Register carry, Register carry2,
3970                                              Register idx, Register jdx,
3971                                              Register yz_idx1, Register yz_idx2,
3972                                              Register tmp, Register tmp3, Register tmp4,
3973                                              Register tmp6, Register product_hi) {
3974 
3975   //   jlong carry, x[], y[], z[];
3976   //   int kdx = ystart+1;
3977   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
3978   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
3979   //     jlong carry2  = (jlong)(tmp3 >>> 64);
3980   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
3981   //     carry  = (jlong)(tmp4 >>> 64);
3982   //     z[kdx+idx+1] = (jlong)tmp3;
3983   //     z[kdx+idx] = (jlong)tmp4;
3984   //   }
3985   //   idx += 2;
3986   //   if (idx > 0) {
3987   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
3988   //     z[kdx+idx] = (jlong)yz_idx1;
3989   //     carry  = (jlong)(yz_idx1 >>> 64);
3990   //   }
3991   //
3992 
3993   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
3994 
3995   lsrw(jdx, idx, 2);
3996 
3997   bind(L_third_loop);
3998 
3999   subsw(jdx, jdx, 1);
4000   br(Assembler::MI, L_third_loop_exit);
4001   subw(idx, idx, 4);
4002 
4003   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4004 
4005   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4006 
4007   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4008 
4009   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4010   ror(yz_idx2, yz_idx2, 32);
4011 
4012   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4013 
4014   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4015   umulh(tmp4, product_hi, yz_idx1);
4016 
4017   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4018   ror(rscratch2, rscratch2, 32);
4019 
4020   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4021   umulh(carry2, product_hi, yz_idx2);
4022 
4023   // propagate sum of both multiplications into carry:tmp4:tmp3
4024   adds(tmp3, tmp3, carry);
4025   adc(tmp4, tmp4, zr);
4026   adds(tmp3, tmp3, rscratch1);
4027   adcs(tmp4, tmp4, tmp);
4028   adc(carry, carry2, zr);
4029   adds(tmp4, tmp4, rscratch2);
4030   adc(carry, carry, zr);
4031 
4032   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4033   ror(tmp4, tmp4, 32);
4034   stp(tmp4, tmp3, Address(tmp6, 0));
4035 
4036   b(L_third_loop);
4037   bind (L_third_loop_exit);
4038 
4039   andw (idx, idx, 0x3);
4040   cbz(idx, L_post_third_loop_done);
4041 
4042   Label L_check_1;
4043   subsw(idx, idx, 2);
4044   br(Assembler::MI, L_check_1);
4045 
4046   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4047   ldr(yz_idx1, Address(rscratch1, 0));
4048   ror(yz_idx1, yz_idx1, 32);
4049   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4050   umulh(tmp4, product_hi, yz_idx1);
4051   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4052   ldr(yz_idx2, Address(rscratch1, 0));
4053   ror(yz_idx2, yz_idx2, 32);
4054 
4055   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4056 
4057   ror(tmp3, tmp3, 32);
4058   str(tmp3, Address(rscratch1, 0));
4059 
4060   bind (L_check_1);
4061 
4062   andw (idx, idx, 0x1);
4063   subsw(idx, idx, 1);
4064   br(Assembler::MI, L_post_third_loop_done);
4065   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4066   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4067   umulh(carry2, tmp4, product_hi);
4068   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4069 
4070   add2_with_carry(carry2, tmp3, tmp4, carry);
4071 
4072   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4073   extr(carry, carry2, tmp3, 32);
4074 
4075   bind(L_post_third_loop_done);
4076 }
4077 
4078 /**
4079  * Code for BigInteger::multiplyToLen() intrinsic.
4080  *
4081  * r0: x
4082  * r1: xlen
4083  * r2: y
4084  * r3: ylen
4085  * r4:  z
4086  * r5: tmp0
4087  * r10: tmp1
4088  * r11: tmp2
4089  * r12: tmp3
4090  * r13: tmp4
4091  * r14: tmp5
4092  * r15: tmp6
4093  * r16: tmp7
4094  *
4095  */
4096 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4097                                      Register z, Register tmp0,
4098                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4099                                      Register tmp5, Register tmp6, Register product_hi) {
4100 
4101   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4102 
4103   const Register idx = tmp1;
4104   const Register kdx = tmp2;
4105   const Register xstart = tmp3;
4106 
4107   const Register y_idx = tmp4;
4108   const Register carry = tmp5;
4109   const Register product  = xlen;
4110   const Register x_xstart = tmp0;
4111 
4112   // First Loop.
4113   //
4114   //  final static long LONG_MASK = 0xffffffffL;
4115   //  int xstart = xlen - 1;
4116   //  int ystart = ylen - 1;
4117   //  long carry = 0;
4118   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4119   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4120   //    z[kdx] = (int)product;
4121   //    carry = product >>> 32;
4122   //  }
4123   //  z[xstart] = (int)carry;
4124   //
4125 
4126   movw(idx, ylen);       // idx = ylen;
4127   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4128   mov(carry, zr);        // carry = 0;
4129 
4130   Label L_done;
4131 
4132   movw(xstart, xlen);
4133   subsw(xstart, xstart, 1);
4134   br(Assembler::MI, L_done);
4135 
4136   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4137 
4138   Label L_second_loop;
4139   cbzw(kdx, L_second_loop);
4140 
4141   Label L_carry;
4142   subw(kdx, kdx, 1);
4143   cbzw(kdx, L_carry);
4144 
4145   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4146   lsr(carry, carry, 32);
4147   subw(kdx, kdx, 1);
4148 
4149   bind(L_carry);
4150   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4151 
4152   // Second and third (nested) loops.
4153   //
4154   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4155   //   carry = 0;
4156   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4157   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4158   //                    (z[k] & LONG_MASK) + carry;
4159   //     z[k] = (int)product;
4160   //     carry = product >>> 32;
4161   //   }
4162   //   z[i] = (int)carry;
4163   // }
4164   //
4165   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4166 
4167   const Register jdx = tmp1;
4168 
4169   bind(L_second_loop);
4170   mov(carry, zr);                // carry = 0;
4171   movw(jdx, ylen);               // j = ystart+1
4172 
4173   subsw(xstart, xstart, 1);      // i = xstart-1;
4174   br(Assembler::MI, L_done);
4175 
4176   str(z, Address(pre(sp, -4 * wordSize)));
4177 
4178   Label L_last_x;
4179   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4180   subsw(xstart, xstart, 1);       // i = xstart-1;
4181   br(Assembler::MI, L_last_x);
4182 
4183   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4184   ldr(product_hi, Address(rscratch1));
4185   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4186 
4187   Label L_third_loop_prologue;
4188   bind(L_third_loop_prologue);
4189 
4190   str(ylen, Address(sp, wordSize));
4191   stp(x, xstart, Address(sp, 2 * wordSize));
4192   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4193                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4194   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4195   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4196 
4197   addw(tmp3, xlen, 1);
4198   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4199   subsw(tmp3, tmp3, 1);
4200   br(Assembler::MI, L_done);
4201 
4202   lsr(carry, carry, 32);
4203   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4204   b(L_second_loop);
4205 
4206   // Next infrequent code is moved outside loops.
4207   bind(L_last_x);
4208   ldrw(product_hi, Address(x,  0));
4209   b(L_third_loop_prologue);
4210 
4211   bind(L_done);
4212 }
4213 
4214 // Code for BigInteger::mulAdd intrinsic
4215 // out     = r0
4216 // in      = r1
4217 // offset  = r2  (already out.length-offset)
4218 // len     = r3
4219 // k       = r4
4220 //
4221 // pseudo code from java implementation:
4222 // carry = 0;
4223 // offset = out.length-offset - 1;
4224 // for (int j=len-1; j >= 0; j--) {
4225 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4226 //     out[offset--] = (int)product;
4227 //     carry = product >>> 32;
4228 // }
4229 // return (int)carry;
4230 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4231       Register len, Register k) {
4232     Label LOOP, END;
4233     // pre-loop
4234     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4235     csel(out, zr, out, Assembler::EQ);
4236     br(Assembler::EQ, END);
4237     add(in, in, len, LSL, 2); // in[j+1] address
4238     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4239     mov(out, zr); // used to keep carry now
4240     BIND(LOOP);
4241     ldrw(rscratch1, Address(pre(in, -4)));
4242     madd(rscratch1, rscratch1, k, out);
4243     ldrw(rscratch2, Address(pre(offset, -4)));
4244     add(rscratch1, rscratch1, rscratch2);
4245     strw(rscratch1, Address(offset));
4246     lsr(out, rscratch1, 32);
4247     subs(len, len, 1);
4248     br(Assembler::NE, LOOP);
4249     BIND(END);
4250 }
4251 
4252 /**
4253  * Emits code to update CRC-32 with a byte value according to constants in table
4254  *
4255  * @param [in,out]crc   Register containing the crc.
4256  * @param [in]val       Register containing the byte to fold into the CRC.
4257  * @param [in]table     Register containing the table of crc constants.
4258  *
4259  * uint32_t crc;
4260  * val = crc_table[(val ^ crc) & 0xFF];
4261  * crc = val ^ (crc >> 8);
4262  *
4263  */
4264 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4265   eor(val, val, crc);
4266   andr(val, val, 0xff);
4267   ldrw(val, Address(table, val, Address::lsl(2)));
4268   eor(crc, val, crc, Assembler::LSR, 8);
4269 }
4270 
4271 /**
4272  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4273  *
4274  * @param [in,out]crc   Register containing the crc.
4275  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4276  * @param [in]table0    Register containing table 0 of crc constants.
4277  * @param [in]table1    Register containing table 1 of crc constants.
4278  * @param [in]table2    Register containing table 2 of crc constants.
4279  * @param [in]table3    Register containing table 3 of crc constants.
4280  *
4281  * uint32_t crc;
4282  *   v = crc ^ v
4283  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4284  *
4285  */
4286 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4287         Register table0, Register table1, Register table2, Register table3,
4288         bool upper) {
4289   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4290   uxtb(tmp, v);
4291   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4292   ubfx(tmp, v, 8, 8);
4293   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4294   eor(crc, crc, tmp);
4295   ubfx(tmp, v, 16, 8);
4296   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4297   eor(crc, crc, tmp);
4298   ubfx(tmp, v, 24, 8);
4299   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4300   eor(crc, crc, tmp);
4301 }
4302 
4303 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4304         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4305     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4306     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4307 
4308     subs(tmp0, len, 384);
4309     mvnw(crc, crc);
4310     br(Assembler::GE, CRC_by128_pre);
4311   BIND(CRC_less128);
4312     subs(len, len, 32);
4313     br(Assembler::GE, CRC_by32_loop);
4314   BIND(CRC_less32);
4315     adds(len, len, 32 - 4);
4316     br(Assembler::GE, CRC_by4_loop);
4317     adds(len, len, 4);
4318     br(Assembler::GT, CRC_by1_loop);
4319     b(L_exit);
4320 
4321   BIND(CRC_by32_loop);
4322     ldp(tmp0, tmp1, Address(buf));
4323     crc32x(crc, crc, tmp0);
4324     ldp(tmp2, tmp3, Address(buf, 16));
4325     crc32x(crc, crc, tmp1);
4326     add(buf, buf, 32);
4327     crc32x(crc, crc, tmp2);
4328     subs(len, len, 32);
4329     crc32x(crc, crc, tmp3);
4330     br(Assembler::GE, CRC_by32_loop);
4331     cmn(len, (u1)32);
4332     br(Assembler::NE, CRC_less32);
4333     b(L_exit);
4334 
4335   BIND(CRC_by4_loop);
4336     ldrw(tmp0, Address(post(buf, 4)));
4337     subs(len, len, 4);
4338     crc32w(crc, crc, tmp0);
4339     br(Assembler::GE, CRC_by4_loop);
4340     adds(len, len, 4);
4341     br(Assembler::LE, L_exit);
4342   BIND(CRC_by1_loop);
4343     ldrb(tmp0, Address(post(buf, 1)));
4344     subs(len, len, 1);
4345     crc32b(crc, crc, tmp0);
4346     br(Assembler::GT, CRC_by1_loop);
4347     b(L_exit);
4348 
4349   BIND(CRC_by128_pre);
4350     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4351       4*256*sizeof(juint) + 8*sizeof(juint));
4352     mov(crc, 0);
4353     crc32x(crc, crc, tmp0);
4354     crc32x(crc, crc, tmp1);
4355 
4356     cbnz(len, CRC_less128);
4357 
4358   BIND(L_exit);
4359     mvnw(crc, crc);
4360 }
4361 
4362 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4363         Register len, Register tmp0, Register tmp1, Register tmp2,
4364         Register tmp3) {
4365     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4366     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4367 
4368     mvnw(crc, crc);
4369 
4370     subs(len, len, 128);
4371     br(Assembler::GE, CRC_by64_pre);
4372   BIND(CRC_less64);
4373     adds(len, len, 128-32);
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(post(buf, 16)));
4384     subs(len, len, 32);
4385     crc32x(crc, crc, tmp0);
4386     ldr(tmp2, Address(post(buf, 8)));
4387     crc32x(crc, crc, tmp1);
4388     ldr(tmp3, Address(post(buf, 8)));
4389     crc32x(crc, crc, tmp2);
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_by64_pre);
4411     sub(buf, buf, 8);
4412     ldp(tmp0, tmp1, Address(buf, 8));
4413     crc32x(crc, crc, tmp0);
4414     ldr(tmp2, Address(buf, 24));
4415     crc32x(crc, crc, tmp1);
4416     ldr(tmp3, Address(buf, 32));
4417     crc32x(crc, crc, tmp2);
4418     ldr(tmp0, Address(buf, 40));
4419     crc32x(crc, crc, tmp3);
4420     ldr(tmp1, Address(buf, 48));
4421     crc32x(crc, crc, tmp0);
4422     ldr(tmp2, Address(buf, 56));
4423     crc32x(crc, crc, tmp1);
4424     ldr(tmp3, Address(pre(buf, 64)));
4425 
4426     b(CRC_by64_loop);
4427 
4428     align(CodeEntryAlignment);
4429   BIND(CRC_by64_loop);
4430     subs(len, len, 64);
4431     crc32x(crc, crc, tmp2);
4432     ldr(tmp0, Address(buf, 8));
4433     crc32x(crc, crc, tmp3);
4434     ldr(tmp1, Address(buf, 16));
4435     crc32x(crc, crc, tmp0);
4436     ldr(tmp2, Address(buf, 24));
4437     crc32x(crc, crc, tmp1);
4438     ldr(tmp3, Address(buf, 32));
4439     crc32x(crc, crc, tmp2);
4440     ldr(tmp0, Address(buf, 40));
4441     crc32x(crc, crc, tmp3);
4442     ldr(tmp1, Address(buf, 48));
4443     crc32x(crc, crc, tmp0);
4444     ldr(tmp2, Address(buf, 56));
4445     crc32x(crc, crc, tmp1);
4446     ldr(tmp3, Address(pre(buf, 64)));
4447     br(Assembler::GE, CRC_by64_loop);
4448 
4449     // post-loop
4450     crc32x(crc, crc, tmp2);
4451     crc32x(crc, crc, tmp3);
4452 
4453     sub(len, len, 64);
4454     add(buf, buf, 8);
4455     cmn(len, (u1)128);
4456     br(Assembler::NE, CRC_less64);
4457   BIND(L_exit);
4458     mvnw(crc, crc);
4459 }
4460 
4461 /**
4462  * @param crc   register containing existing CRC (32-bit)
4463  * @param buf   register pointing to input byte buffer (byte*)
4464  * @param len   register containing number of bytes
4465  * @param table register that will contain address of CRC table
4466  * @param tmp   scratch register
4467  */
4468 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4469         Register table0, Register table1, Register table2, Register table3,
4470         Register tmp, Register tmp2, Register tmp3) {
4471   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4472 
4473   if (UseCryptoPmullForCRC32) {
4474       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4475       return;
4476   }
4477 
4478   if (UseCRC32) {
4479       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4480       return;
4481   }
4482 
4483     mvnw(crc, crc);
4484 
4485     {
4486       uint64_t offset;
4487       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4488       add(table0, table0, offset);
4489     }
4490     add(table1, table0, 1*256*sizeof(juint));
4491     add(table2, table0, 2*256*sizeof(juint));
4492     add(table3, table0, 3*256*sizeof(juint));
4493 
4494     { // Neon code start
4495       cmp(len, (u1)64);
4496       br(Assembler::LT, L_by16);
4497       eor(v16, T16B, v16, v16);
4498 
4499     Label L_fold;
4500 
4501       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4502 
4503       ld1(v0, v1, T2D, post(buf, 32));
4504       ld1r(v4, T2D, post(tmp, 8));
4505       ld1r(v5, T2D, post(tmp, 8));
4506       ld1r(v6, T2D, post(tmp, 8));
4507       ld1r(v7, T2D, post(tmp, 8));
4508       mov(v16, S, 0, crc);
4509 
4510       eor(v0, T16B, v0, v16);
4511       sub(len, len, 64);
4512 
4513     BIND(L_fold);
4514       pmull(v22, T8H, v0, v5, T8B);
4515       pmull(v20, T8H, v0, v7, T8B);
4516       pmull(v23, T8H, v0, v4, T8B);
4517       pmull(v21, T8H, v0, v6, T8B);
4518 
4519       pmull2(v18, T8H, v0, v5, T16B);
4520       pmull2(v16, T8H, v0, v7, T16B);
4521       pmull2(v19, T8H, v0, v4, T16B);
4522       pmull2(v17, T8H, v0, v6, T16B);
4523 
4524       uzp1(v24, T8H, v20, v22);
4525       uzp2(v25, T8H, v20, v22);
4526       eor(v20, T16B, v24, v25);
4527 
4528       uzp1(v26, T8H, v16, v18);
4529       uzp2(v27, T8H, v16, v18);
4530       eor(v16, T16B, v26, v27);
4531 
4532       ushll2(v22, T4S, v20, T8H, 8);
4533       ushll(v20, T4S, v20, T4H, 8);
4534 
4535       ushll2(v18, T4S, v16, T8H, 8);
4536       ushll(v16, T4S, v16, T4H, 8);
4537 
4538       eor(v22, T16B, v23, v22);
4539       eor(v18, T16B, v19, v18);
4540       eor(v20, T16B, v21, v20);
4541       eor(v16, T16B, v17, v16);
4542 
4543       uzp1(v17, T2D, v16, v20);
4544       uzp2(v21, T2D, v16, v20);
4545       eor(v17, T16B, v17, v21);
4546 
4547       ushll2(v20, T2D, v17, T4S, 16);
4548       ushll(v16, T2D, v17, T2S, 16);
4549 
4550       eor(v20, T16B, v20, v22);
4551       eor(v16, T16B, v16, v18);
4552 
4553       uzp1(v17, T2D, v20, v16);
4554       uzp2(v21, T2D, v20, v16);
4555       eor(v28, T16B, v17, v21);
4556 
4557       pmull(v22, T8H, v1, v5, T8B);
4558       pmull(v20, T8H, v1, v7, T8B);
4559       pmull(v23, T8H, v1, v4, T8B);
4560       pmull(v21, T8H, v1, v6, T8B);
4561 
4562       pmull2(v18, T8H, v1, v5, T16B);
4563       pmull2(v16, T8H, v1, v7, T16B);
4564       pmull2(v19, T8H, v1, v4, T16B);
4565       pmull2(v17, T8H, v1, v6, T16B);
4566 
4567       ld1(v0, v1, T2D, post(buf, 32));
4568 
4569       uzp1(v24, T8H, v20, v22);
4570       uzp2(v25, T8H, v20, v22);
4571       eor(v20, T16B, v24, v25);
4572 
4573       uzp1(v26, T8H, v16, v18);
4574       uzp2(v27, T8H, v16, v18);
4575       eor(v16, T16B, v26, v27);
4576 
4577       ushll2(v22, T4S, v20, T8H, 8);
4578       ushll(v20, T4S, v20, T4H, 8);
4579 
4580       ushll2(v18, T4S, v16, T8H, 8);
4581       ushll(v16, T4S, v16, T4H, 8);
4582 
4583       eor(v22, T16B, v23, v22);
4584       eor(v18, T16B, v19, v18);
4585       eor(v20, T16B, v21, v20);
4586       eor(v16, T16B, v17, v16);
4587 
4588       uzp1(v17, T2D, v16, v20);
4589       uzp2(v21, T2D, v16, v20);
4590       eor(v16, T16B, v17, v21);
4591 
4592       ushll2(v20, T2D, v16, T4S, 16);
4593       ushll(v16, T2D, v16, T2S, 16);
4594 
4595       eor(v20, T16B, v22, v20);
4596       eor(v16, T16B, v16, v18);
4597 
4598       uzp1(v17, T2D, v20, v16);
4599       uzp2(v21, T2D, v20, v16);
4600       eor(v20, T16B, v17, v21);
4601 
4602       shl(v16, T2D, v28, 1);
4603       shl(v17, T2D, v20, 1);
4604 
4605       eor(v0, T16B, v0, v16);
4606       eor(v1, T16B, v1, v17);
4607 
4608       subs(len, len, 32);
4609       br(Assembler::GE, L_fold);
4610 
4611       mov(crc, 0);
4612       mov(tmp, v0, D, 0);
4613       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4614       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4615       mov(tmp, v0, D, 1);
4616       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4617       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4618       mov(tmp, v1, D, 0);
4619       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4620       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4621       mov(tmp, v1, D, 1);
4622       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4623       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4624 
4625       add(len, len, 32);
4626     } // Neon code end
4627 
4628   BIND(L_by16);
4629     subs(len, len, 16);
4630     br(Assembler::GE, L_by16_loop);
4631     adds(len, len, 16-4);
4632     br(Assembler::GE, L_by4_loop);
4633     adds(len, len, 4);
4634     br(Assembler::GT, L_by1_loop);
4635     b(L_exit);
4636 
4637   BIND(L_by4_loop);
4638     ldrw(tmp, Address(post(buf, 4)));
4639     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4640     subs(len, len, 4);
4641     br(Assembler::GE, L_by4_loop);
4642     adds(len, len, 4);
4643     br(Assembler::LE, L_exit);
4644   BIND(L_by1_loop);
4645     subs(len, len, 1);
4646     ldrb(tmp, Address(post(buf, 1)));
4647     update_byte_crc32(crc, tmp, table0);
4648     br(Assembler::GT, L_by1_loop);
4649     b(L_exit);
4650 
4651     align(CodeEntryAlignment);
4652   BIND(L_by16_loop);
4653     subs(len, len, 16);
4654     ldp(tmp, tmp3, Address(post(buf, 16)));
4655     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4656     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4657     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4658     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4659     br(Assembler::GE, L_by16_loop);
4660     adds(len, len, 16-4);
4661     br(Assembler::GE, L_by4_loop);
4662     adds(len, len, 4);
4663     br(Assembler::GT, L_by1_loop);
4664   BIND(L_exit);
4665     mvnw(crc, crc);
4666 }
4667 
4668 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4669         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4670     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4671     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4672 
4673     subs(tmp0, len, 384);
4674     br(Assembler::GE, CRC_by128_pre);
4675   BIND(CRC_less128);
4676     subs(len, len, 32);
4677     br(Assembler::GE, CRC_by32_loop);
4678   BIND(CRC_less32);
4679     adds(len, len, 32 - 4);
4680     br(Assembler::GE, CRC_by4_loop);
4681     adds(len, len, 4);
4682     br(Assembler::GT, CRC_by1_loop);
4683     b(L_exit);
4684 
4685   BIND(CRC_by32_loop);
4686     ldp(tmp0, tmp1, Address(buf));
4687     crc32cx(crc, crc, tmp0);
4688     ldr(tmp2, Address(buf, 16));
4689     crc32cx(crc, crc, tmp1);
4690     ldr(tmp3, Address(buf, 24));
4691     crc32cx(crc, crc, tmp2);
4692     add(buf, buf, 32);
4693     subs(len, len, 32);
4694     crc32cx(crc, crc, tmp3);
4695     br(Assembler::GE, CRC_by32_loop);
4696     cmn(len, (u1)32);
4697     br(Assembler::NE, CRC_less32);
4698     b(L_exit);
4699 
4700   BIND(CRC_by4_loop);
4701     ldrw(tmp0, Address(post(buf, 4)));
4702     subs(len, len, 4);
4703     crc32cw(crc, crc, tmp0);
4704     br(Assembler::GE, CRC_by4_loop);
4705     adds(len, len, 4);
4706     br(Assembler::LE, L_exit);
4707   BIND(CRC_by1_loop);
4708     ldrb(tmp0, Address(post(buf, 1)));
4709     subs(len, len, 1);
4710     crc32cb(crc, crc, tmp0);
4711     br(Assembler::GT, CRC_by1_loop);
4712     b(L_exit);
4713 
4714   BIND(CRC_by128_pre);
4715     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4716       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4717     mov(crc, 0);
4718     crc32cx(crc, crc, tmp0);
4719     crc32cx(crc, crc, tmp1);
4720 
4721     cbnz(len, CRC_less128);
4722 
4723   BIND(L_exit);
4724 }
4725 
4726 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4727         Register len, Register tmp0, Register tmp1, Register tmp2,
4728         Register tmp3) {
4729     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4730     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4731 
4732     subs(len, len, 128);
4733     br(Assembler::GE, CRC_by64_pre);
4734   BIND(CRC_less64);
4735     adds(len, len, 128-32);
4736     br(Assembler::GE, CRC_by32_loop);
4737   BIND(CRC_less32);
4738     adds(len, len, 32-4);
4739     br(Assembler::GE, CRC_by4_loop);
4740     adds(len, len, 4);
4741     br(Assembler::GT, CRC_by1_loop);
4742     b(L_exit);
4743 
4744   BIND(CRC_by32_loop);
4745     ldp(tmp0, tmp1, Address(post(buf, 16)));
4746     subs(len, len, 32);
4747     crc32cx(crc, crc, tmp0);
4748     ldr(tmp2, Address(post(buf, 8)));
4749     crc32cx(crc, crc, tmp1);
4750     ldr(tmp3, Address(post(buf, 8)));
4751     crc32cx(crc, crc, tmp2);
4752     crc32cx(crc, crc, tmp3);
4753     br(Assembler::GE, CRC_by32_loop);
4754     cmn(len, (u1)32);
4755     br(Assembler::NE, CRC_less32);
4756     b(L_exit);
4757 
4758   BIND(CRC_by4_loop);
4759     ldrw(tmp0, Address(post(buf, 4)));
4760     subs(len, len, 4);
4761     crc32cw(crc, crc, tmp0);
4762     br(Assembler::GE, CRC_by4_loop);
4763     adds(len, len, 4);
4764     br(Assembler::LE, L_exit);
4765   BIND(CRC_by1_loop);
4766     ldrb(tmp0, Address(post(buf, 1)));
4767     subs(len, len, 1);
4768     crc32cb(crc, crc, tmp0);
4769     br(Assembler::GT, CRC_by1_loop);
4770     b(L_exit);
4771 
4772   BIND(CRC_by64_pre);
4773     sub(buf, buf, 8);
4774     ldp(tmp0, tmp1, Address(buf, 8));
4775     crc32cx(crc, crc, tmp0);
4776     ldr(tmp2, Address(buf, 24));
4777     crc32cx(crc, crc, tmp1);
4778     ldr(tmp3, Address(buf, 32));
4779     crc32cx(crc, crc, tmp2);
4780     ldr(tmp0, Address(buf, 40));
4781     crc32cx(crc, crc, tmp3);
4782     ldr(tmp1, Address(buf, 48));
4783     crc32cx(crc, crc, tmp0);
4784     ldr(tmp2, Address(buf, 56));
4785     crc32cx(crc, crc, tmp1);
4786     ldr(tmp3, Address(pre(buf, 64)));
4787 
4788     b(CRC_by64_loop);
4789 
4790     align(CodeEntryAlignment);
4791   BIND(CRC_by64_loop);
4792     subs(len, len, 64);
4793     crc32cx(crc, crc, tmp2);
4794     ldr(tmp0, Address(buf, 8));
4795     crc32cx(crc, crc, tmp3);
4796     ldr(tmp1, Address(buf, 16));
4797     crc32cx(crc, crc, tmp0);
4798     ldr(tmp2, Address(buf, 24));
4799     crc32cx(crc, crc, tmp1);
4800     ldr(tmp3, Address(buf, 32));
4801     crc32cx(crc, crc, tmp2);
4802     ldr(tmp0, Address(buf, 40));
4803     crc32cx(crc, crc, tmp3);
4804     ldr(tmp1, Address(buf, 48));
4805     crc32cx(crc, crc, tmp0);
4806     ldr(tmp2, Address(buf, 56));
4807     crc32cx(crc, crc, tmp1);
4808     ldr(tmp3, Address(pre(buf, 64)));
4809     br(Assembler::GE, CRC_by64_loop);
4810 
4811     // post-loop
4812     crc32cx(crc, crc, tmp2);
4813     crc32cx(crc, crc, tmp3);
4814 
4815     sub(len, len, 64);
4816     add(buf, buf, 8);
4817     cmn(len, (u1)128);
4818     br(Assembler::NE, CRC_less64);
4819   BIND(L_exit);
4820 }
4821 
4822 /**
4823  * @param crc   register containing existing CRC (32-bit)
4824  * @param buf   register pointing to input byte buffer (byte*)
4825  * @param len   register containing number of bytes
4826  * @param table register that will contain address of CRC table
4827  * @param tmp   scratch register
4828  */
4829 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4830         Register table0, Register table1, Register table2, Register table3,
4831         Register tmp, Register tmp2, Register tmp3) {
4832   if (UseCryptoPmullForCRC32) {
4833     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4834   } else {
4835     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4836   }
4837 }
4838 
4839 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4840         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4841     Label CRC_by128_loop;
4842     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4843 
4844     sub(len, len, 256);
4845     Register table = tmp0;
4846     {
4847       uint64_t offset;
4848       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4849       add(table, table, offset);
4850     }
4851     add(table, table, table_offset);
4852 
4853     // Registers v0..v7 are used as data registers.
4854     // Registers v16..v31 are used as tmp registers.
4855     sub(buf, buf, 0x10);
4856     ldrq(v0, Address(buf, 0x10));
4857     ldrq(v1, Address(buf, 0x20));
4858     ldrq(v2, Address(buf, 0x30));
4859     ldrq(v3, Address(buf, 0x40));
4860     ldrq(v4, Address(buf, 0x50));
4861     ldrq(v5, Address(buf, 0x60));
4862     ldrq(v6, Address(buf, 0x70));
4863     ldrq(v7, Address(pre(buf, 0x80)));
4864 
4865     movi(v31, T4S, 0);
4866     mov(v31, S, 0, crc);
4867     eor(v0, T16B, v0, v31);
4868 
4869     // Register v16 contains constants from the crc table.
4870     ldrq(v16, Address(table));
4871     b(CRC_by128_loop);
4872 
4873     align(OptoLoopAlignment);
4874   BIND(CRC_by128_loop);
4875     pmull (v17,  T1Q, v0, v16, T1D);
4876     pmull2(v18, T1Q, v0, v16, T2D);
4877     ldrq(v0, Address(buf, 0x10));
4878     eor3(v0, T16B, v17,  v18, v0);
4879 
4880     pmull (v19, T1Q, v1, v16, T1D);
4881     pmull2(v20, T1Q, v1, v16, T2D);
4882     ldrq(v1, Address(buf, 0x20));
4883     eor3(v1, T16B, v19, v20, v1);
4884 
4885     pmull (v21, T1Q, v2, v16, T1D);
4886     pmull2(v22, T1Q, v2, v16, T2D);
4887     ldrq(v2, Address(buf, 0x30));
4888     eor3(v2, T16B, v21, v22, v2);
4889 
4890     pmull (v23, T1Q, v3, v16, T1D);
4891     pmull2(v24, T1Q, v3, v16, T2D);
4892     ldrq(v3, Address(buf, 0x40));
4893     eor3(v3, T16B, v23, v24, v3);
4894 
4895     pmull (v25, T1Q, v4, v16, T1D);
4896     pmull2(v26, T1Q, v4, v16, T2D);
4897     ldrq(v4, Address(buf, 0x50));
4898     eor3(v4, T16B, v25, v26, v4);
4899 
4900     pmull (v27, T1Q, v5, v16, T1D);
4901     pmull2(v28, T1Q, v5, v16, T2D);
4902     ldrq(v5, Address(buf, 0x60));
4903     eor3(v5, T16B, v27, v28, v5);
4904 
4905     pmull (v29, T1Q, v6, v16, T1D);
4906     pmull2(v30, T1Q, v6, v16, T2D);
4907     ldrq(v6, Address(buf, 0x70));
4908     eor3(v6, T16B, v29, v30, v6);
4909 
4910     // Reuse registers v23, v24.
4911     // Using them won't block the first instruction of the next iteration.
4912     pmull (v23, T1Q, v7, v16, T1D);
4913     pmull2(v24, T1Q, v7, v16, T2D);
4914     ldrq(v7, Address(pre(buf, 0x80)));
4915     eor3(v7, T16B, v23, v24, v7);
4916 
4917     subs(len, len, 0x80);
4918     br(Assembler::GE, CRC_by128_loop);
4919 
4920     // fold into 512 bits
4921     // Use v31 for constants because v16 can be still in use.
4922     ldrq(v31, Address(table, 0x10));
4923 
4924     pmull (v17,  T1Q, v0, v31, T1D);
4925     pmull2(v18, T1Q, v0, v31, T2D);
4926     eor3(v0, T16B, v17, v18, v4);
4927 
4928     pmull (v19, T1Q, v1, v31, T1D);
4929     pmull2(v20, T1Q, v1, v31, T2D);
4930     eor3(v1, T16B, v19, v20, v5);
4931 
4932     pmull (v21, T1Q, v2, v31, T1D);
4933     pmull2(v22, T1Q, v2, v31, T2D);
4934     eor3(v2, T16B, v21, v22, v6);
4935 
4936     pmull (v23, T1Q, v3, v31, T1D);
4937     pmull2(v24, T1Q, v3, v31, T2D);
4938     eor3(v3, T16B, v23, v24, v7);
4939 
4940     // fold into 128 bits
4941     // Use v17 for constants because v31 can be still in use.
4942     ldrq(v17, Address(table, 0x20));
4943     pmull (v25, T1Q, v0, v17, T1D);
4944     pmull2(v26, T1Q, v0, v17, T2D);
4945     eor3(v3, T16B, v3, v25, v26);
4946 
4947     // Use v18 for constants because v17 can be still in use.
4948     ldrq(v18, Address(table, 0x30));
4949     pmull (v27, T1Q, v1, v18, T1D);
4950     pmull2(v28, T1Q, v1, v18, T2D);
4951     eor3(v3, T16B, v3, v27, v28);
4952 
4953     // Use v19 for constants because v18 can be still in use.
4954     ldrq(v19, Address(table, 0x40));
4955     pmull (v29, T1Q, v2, v19, T1D);
4956     pmull2(v30, T1Q, v2, v19, T2D);
4957     eor3(v0, T16B, v3, v29, v30);
4958 
4959     add(len, len, 0x80);
4960     add(buf, buf, 0x10);
4961 
4962     mov(tmp0, v0, D, 0);
4963     mov(tmp1, v0, D, 1);
4964 }
4965 
4966 void MacroAssembler::addptr(const Address &dst, int32_t src) {
4967   Address adr;
4968   switch(dst.getMode()) {
4969   case Address::base_plus_offset:
4970     // This is the expected mode, although we allow all the other
4971     // forms below.
4972     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
4973     break;
4974   default:
4975     lea(rscratch2, dst);
4976     adr = Address(rscratch2);
4977     break;
4978   }
4979   ldr(rscratch1, adr);
4980   add(rscratch1, rscratch1, src);
4981   str(rscratch1, adr);
4982 }
4983 
4984 void MacroAssembler::cmpptr(Register src1, Address src2) {
4985   uint64_t offset;
4986   adrp(rscratch1, src2, offset);
4987   ldr(rscratch1, Address(rscratch1, offset));
4988   cmp(src1, rscratch1);
4989 }
4990 
4991 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
4992   cmp(obj1, obj2);
4993 }
4994 
4995 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
4996   load_method_holder(rresult, rmethod);
4997   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
4998 }
4999 
5000 void MacroAssembler::load_method_holder(Register holder, Register method) {
5001   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5002   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5003   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5004 }
5005 
5006 // Loads the obj's Klass* into dst.
5007 // Preserves all registers (incl src, rscratch1 and rscratch2).
5008 // Input:
5009 // src - the oop we want to load the klass from.
5010 // dst - output narrow klass.
5011 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5012   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5013   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5014   lsr(dst, dst, markWord::klass_shift);
5015 }
5016 
5017 void MacroAssembler::load_klass(Register dst, Register src) {
5018   if (UseCompactObjectHeaders) {
5019     load_narrow_klass_compact(dst, src);
5020     decode_klass_not_null(dst);
5021   } else if (UseCompressedClassPointers) {
5022     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5023     decode_klass_not_null(dst);
5024   } else {
5025     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5026   }
5027 }
5028 
5029 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5030   if (RestoreMXCSROnJNICalls) {
5031     Label OK;
5032     get_fpcr(tmp1);
5033     mov(tmp2, tmp1);
5034     // Set FPCR to the state we need. We do want Round to Nearest. We
5035     // don't want non-IEEE rounding modes or floating-point traps.
5036     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5037     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5038     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5039     eor(tmp2, tmp1, tmp2);
5040     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5041     set_fpcr(tmp1);
5042     bind(OK);
5043   }
5044 }
5045 
5046 // ((OopHandle)result).resolve();
5047 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5048   // OopHandle::resolve is an indirection.
5049   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5050 }
5051 
5052 // ((WeakHandle)result).resolve();
5053 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5054   assert_different_registers(result, tmp1, tmp2);
5055   Label resolved;
5056 
5057   // A null weak handle resolves to null.
5058   cbz(result, resolved);
5059 
5060   // Only 64 bit platforms support GCs that require a tmp register
5061   // WeakHandle::resolve is an indirection like jweak.
5062   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5063                  result, Address(result), tmp1, tmp2);
5064   bind(resolved);
5065 }
5066 
5067 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5068   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5069   ldr(dst, Address(rmethod, Method::const_offset()));
5070   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5071   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5072   ldr(dst, Address(dst, mirror_offset));
5073   resolve_oop_handle(dst, tmp1, tmp2);
5074 }
5075 
5076 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5077   assert_different_registers(obj, klass, tmp);
5078   if (UseCompressedClassPointers) {
5079     if (UseCompactObjectHeaders) {
5080       load_narrow_klass_compact(tmp, obj);
5081     } else {
5082       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5083     }
5084     if (CompressedKlassPointers::base() == nullptr) {
5085       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5086       return;
5087     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5088                && CompressedKlassPointers::shift() == 0) {
5089       // Only the bottom 32 bits matter
5090       cmpw(klass, tmp);
5091       return;
5092     }
5093     decode_klass_not_null(tmp);
5094   } else {
5095     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5096   }
5097   cmp(klass, tmp);
5098 }
5099 
5100 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5101   if (UseCompactObjectHeaders) {
5102     load_narrow_klass_compact(tmp1, obj1);
5103     load_narrow_klass_compact(tmp2,  obj2);
5104     cmpw(tmp1, tmp2);
5105   } else if (UseCompressedClassPointers) {
5106     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5107     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5108     cmpw(tmp1, tmp2);
5109   } else {
5110     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5111     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5112     cmp(tmp1, tmp2);
5113   }
5114 }
5115 
5116 void MacroAssembler::store_klass(Register dst, Register src) {
5117   // FIXME: Should this be a store release?  concurrent gcs assumes
5118   // klass length is valid if klass field is not null.
5119   assert(!UseCompactObjectHeaders, "not with compact headers");
5120   if (UseCompressedClassPointers) {
5121     encode_klass_not_null(src);
5122     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5123   } else {
5124     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5125   }
5126 }
5127 
5128 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5129   assert(!UseCompactObjectHeaders, "not with compact headers");
5130   if (UseCompressedClassPointers) {
5131     // Store to klass gap in destination
5132     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5133   }
5134 }
5135 
5136 // Algorithm must match CompressedOops::encode.
5137 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5138 #ifdef ASSERT
5139   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5140 #endif
5141   verify_oop_msg(s, "broken oop in encode_heap_oop");
5142   if (CompressedOops::base() == nullptr) {
5143     if (CompressedOops::shift() != 0) {
5144       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5145       lsr(d, s, LogMinObjAlignmentInBytes);
5146     } else {
5147       mov(d, s);
5148     }
5149   } else {
5150     subs(d, s, rheapbase);
5151     csel(d, d, zr, Assembler::HS);
5152     lsr(d, d, LogMinObjAlignmentInBytes);
5153 
5154     /*  Old algorithm: is this any worse?
5155     Label nonnull;
5156     cbnz(r, nonnull);
5157     sub(r, r, rheapbase);
5158     bind(nonnull);
5159     lsr(r, r, LogMinObjAlignmentInBytes);
5160     */
5161   }
5162 }
5163 
5164 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5165 #ifdef ASSERT
5166   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5167   if (CheckCompressedOops) {
5168     Label ok;
5169     cbnz(r, ok);
5170     stop("null oop passed to encode_heap_oop_not_null");
5171     bind(ok);
5172   }
5173 #endif
5174   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5175   if (CompressedOops::base() != nullptr) {
5176     sub(r, r, rheapbase);
5177   }
5178   if (CompressedOops::shift() != 0) {
5179     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5180     lsr(r, r, LogMinObjAlignmentInBytes);
5181   }
5182 }
5183 
5184 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5185 #ifdef ASSERT
5186   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5187   if (CheckCompressedOops) {
5188     Label ok;
5189     cbnz(src, ok);
5190     stop("null oop passed to encode_heap_oop_not_null2");
5191     bind(ok);
5192   }
5193 #endif
5194   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5195 
5196   Register data = src;
5197   if (CompressedOops::base() != nullptr) {
5198     sub(dst, src, rheapbase);
5199     data = dst;
5200   }
5201   if (CompressedOops::shift() != 0) {
5202     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5203     lsr(dst, data, LogMinObjAlignmentInBytes);
5204     data = dst;
5205   }
5206   if (data == src)
5207     mov(dst, src);
5208 }
5209 
5210 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5211 #ifdef ASSERT
5212   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5213 #endif
5214   if (CompressedOops::base() == nullptr) {
5215     if (CompressedOops::shift() != 0) {
5216       lsl(d, s, CompressedOops::shift());
5217     } else if (d != s) {
5218       mov(d, s);
5219     }
5220   } else {
5221     Label done;
5222     if (d != s)
5223       mov(d, s);
5224     cbz(s, done);
5225     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5226     bind(done);
5227   }
5228   verify_oop_msg(d, "broken oop in decode_heap_oop");
5229 }
5230 
5231 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5232   assert (UseCompressedOops, "should only be used for compressed headers");
5233   assert (Universe::heap() != nullptr, "java heap should be initialized");
5234   // Cannot assert, unverified entry point counts instructions (see .ad file)
5235   // vtableStubs also counts instructions in pd_code_size_limit.
5236   // Also do not verify_oop as this is called by verify_oop.
5237   if (CompressedOops::shift() != 0) {
5238     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5239     if (CompressedOops::base() != nullptr) {
5240       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5241     } else {
5242       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5243     }
5244   } else {
5245     assert (CompressedOops::base() == nullptr, "sanity");
5246   }
5247 }
5248 
5249 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5250   assert (UseCompressedOops, "should only be used for compressed headers");
5251   assert (Universe::heap() != nullptr, "java heap should be initialized");
5252   // Cannot assert, unverified entry point counts instructions (see .ad file)
5253   // vtableStubs also counts instructions in pd_code_size_limit.
5254   // Also do not verify_oop as this is called by verify_oop.
5255   if (CompressedOops::shift() != 0) {
5256     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5257     if (CompressedOops::base() != nullptr) {
5258       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5259     } else {
5260       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5261     }
5262   } else {
5263     assert (CompressedOops::base() == nullptr, "sanity");
5264     if (dst != src) {
5265       mov(dst, src);
5266     }
5267   }
5268 }
5269 
5270 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5271 
5272 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5273   assert(Metaspace::initialized(), "metaspace not initialized yet");
5274   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5275   return _klass_decode_mode;
5276 }
5277 
5278 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5279   assert(UseCompressedClassPointers, "not using compressed class pointers");
5280 
5281   // KlassDecodeMode shouldn't be set already.
5282   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5283 
5284   if (base == nullptr) {
5285     return KlassDecodeZero;
5286   }
5287 
5288   if (operand_valid_for_logical_immediate(
5289         /*is32*/false, (uint64_t)base)) {
5290     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5291     if (((uint64_t)base & range_mask) == 0) {
5292       return KlassDecodeXor;
5293     }
5294   }
5295 
5296   const uint64_t shifted_base =
5297     (uint64_t)base >> shift;
5298   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5299     return KlassDecodeMovk;
5300   }
5301 
5302   // No valid encoding.
5303   return KlassDecodeNone;
5304 }
5305 
5306 // Check if one of the above decoding modes will work for given base, shift and range.
5307 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5308   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5309 }
5310 
5311 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5312   _klass_decode_mode = klass_decode_mode(base, shift, range);
5313   return _klass_decode_mode != KlassDecodeNone;
5314 }
5315 
5316 static Register pick_different_tmp(Register dst, Register src) {
5317   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5318   return *tmps.begin();
5319 }
5320 
5321 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5322   // we have to load the klass base from the AOT constants area but
5323   // not the shift because it is not allowed to change
5324   int shift = CompressedKlassPointers::shift();
5325   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5326   if (dst != src) {
5327     // we can load the base into dst, subtract it formthe src and shift down
5328     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5329     ldr(dst, dst);
5330     sub(dst, src, dst);
5331     lsr(dst, dst, shift);
5332   } else {
5333     // we need an extra register in order to load the coop base
5334     Register tmp = pick_different_tmp(dst, src);
5335     RegSet regs = RegSet::of(tmp);
5336     push(regs, sp);
5337     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5338     ldr(tmp, tmp);
5339     sub(dst, src, tmp);
5340     lsr(dst, dst, shift);
5341     pop(regs, sp);
5342   }
5343 }
5344 
5345 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5346   if (AOTCodeCache::is_on_for_dump()) {
5347     encode_klass_not_null_for_aot(dst, src);
5348     return;
5349   }
5350 
5351   switch (klass_decode_mode()) {
5352   case KlassDecodeZero:
5353     if (CompressedKlassPointers::shift() != 0) {
5354       lsr(dst, src, CompressedKlassPointers::shift());
5355     } else {
5356       if (dst != src) mov(dst, src);
5357     }
5358     break;
5359 
5360   case KlassDecodeXor:
5361     if (CompressedKlassPointers::shift() != 0) {
5362       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5363       lsr(dst, dst, CompressedKlassPointers::shift());
5364     } else {
5365       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5366     }
5367     break;
5368 
5369   case KlassDecodeMovk:
5370     if (CompressedKlassPointers::shift() != 0) {
5371       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5372     } else {
5373       movw(dst, src);
5374     }
5375     break;
5376 
5377   case KlassDecodeNone:
5378     ShouldNotReachHere();
5379     break;
5380   }
5381 }
5382 
5383 void MacroAssembler::encode_klass_not_null(Register r) {
5384   encode_klass_not_null(r, r);
5385 }
5386 
5387 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5388   // we have to load the klass base from the AOT constants area but
5389   // not the shift because it is not allowed to change
5390   int shift = CompressedKlassPointers::shift();
5391   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5392   if (dst != src) {
5393     // we can load the base into dst then add the offset with a suitable shift
5394     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5395     ldr(dst, dst);
5396     add(dst, dst, src, LSL,  shift);
5397   } else {
5398     // we need an extra register in order to load the coop base
5399     Register tmp = pick_different_tmp(dst, src);
5400     RegSet regs = RegSet::of(tmp);
5401     push(regs, sp);
5402     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5403     ldr(tmp, tmp);
5404     add(dst, tmp,  src, LSL,  shift);
5405     pop(regs, sp);
5406   }
5407 }
5408 
5409 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5410   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5411 
5412   if (AOTCodeCache::is_on_for_dump()) {
5413     decode_klass_not_null_for_aot(dst, src);
5414     return;
5415   }
5416 
5417   switch (klass_decode_mode()) {
5418   case KlassDecodeZero:
5419     if (CompressedKlassPointers::shift() != 0) {
5420       lsl(dst, src, CompressedKlassPointers::shift());
5421     } else {
5422       if (dst != src) mov(dst, src);
5423     }
5424     break;
5425 
5426   case KlassDecodeXor:
5427     if (CompressedKlassPointers::shift() != 0) {
5428       lsl(dst, src, CompressedKlassPointers::shift());
5429       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5430     } else {
5431       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5432     }
5433     break;
5434 
5435   case KlassDecodeMovk: {
5436     const uint64_t shifted_base =
5437       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5438 
5439     if (dst != src) movw(dst, src);
5440     movk(dst, shifted_base >> 32, 32);
5441 
5442     if (CompressedKlassPointers::shift() != 0) {
5443       lsl(dst, dst, CompressedKlassPointers::shift());
5444     }
5445 
5446     break;
5447   }
5448 
5449   case KlassDecodeNone:
5450     ShouldNotReachHere();
5451     break;
5452   }
5453 }
5454 
5455 void  MacroAssembler::decode_klass_not_null(Register r) {
5456   decode_klass_not_null(r, r);
5457 }
5458 
5459 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5460 #ifdef ASSERT
5461   {
5462     ThreadInVMfromUnknown tiv;
5463     assert (UseCompressedOops, "should only be used for compressed oops");
5464     assert (Universe::heap() != nullptr, "java heap should be initialized");
5465     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5466     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5467   }
5468 #endif
5469   int oop_index = oop_recorder()->find_index(obj);
5470   InstructionMark im(this);
5471   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5472   code_section()->relocate(inst_mark(), rspec);
5473   movz(dst, 0xDEAD, 16);
5474   movk(dst, 0xBEEF);
5475 }
5476 
5477 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5478   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5479   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5480   int index = oop_recorder()->find_index(k);
5481   assert(! Universe::heap()->is_in(k), "should not be an oop");
5482 
5483   InstructionMark im(this);
5484   RelocationHolder rspec = metadata_Relocation::spec(index);
5485   code_section()->relocate(inst_mark(), rspec);
5486   narrowKlass nk = CompressedKlassPointers::encode(k);
5487   movz(dst, (nk >> 16), 16);
5488   movk(dst, nk & 0xffff);
5489 }
5490 
5491 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5492                                     Register dst, Address src,
5493                                     Register tmp1, Register tmp2) {
5494   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5495   decorators = AccessInternal::decorator_fixup(decorators, type);
5496   bool as_raw = (decorators & AS_RAW) != 0;
5497   if (as_raw) {
5498     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5499   } else {
5500     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5501   }
5502 }
5503 
5504 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5505                                      Address dst, Register val,
5506                                      Register tmp1, Register tmp2, Register tmp3) {
5507   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5508   decorators = AccessInternal::decorator_fixup(decorators, type);
5509   bool as_raw = (decorators & AS_RAW) != 0;
5510   if (as_raw) {
5511     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5512   } else {
5513     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5514   }
5515 }
5516 
5517 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5518                                    Register tmp2, DecoratorSet decorators) {
5519   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5520 }
5521 
5522 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5523                                             Register tmp2, DecoratorSet decorators) {
5524   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5525 }
5526 
5527 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5528                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5529   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5530 }
5531 
5532 // Used for storing nulls.
5533 void MacroAssembler::store_heap_oop_null(Address dst) {
5534   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5535 }
5536 
5537 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5538   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5539   int index = oop_recorder()->allocate_metadata_index(obj);
5540   RelocationHolder rspec = metadata_Relocation::spec(index);
5541   return Address((address)obj, rspec);
5542 }
5543 
5544 // Move an oop into a register.
5545 void MacroAssembler::movoop(Register dst, jobject obj) {
5546   int oop_index;
5547   if (obj == nullptr) {
5548     oop_index = oop_recorder()->allocate_oop_index(obj);
5549   } else {
5550 #ifdef ASSERT
5551     {
5552       ThreadInVMfromUnknown tiv;
5553       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5554     }
5555 #endif
5556     oop_index = oop_recorder()->find_index(obj);
5557   }
5558   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5559 
5560   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5561     mov(dst, Address((address)obj, rspec));
5562   } else {
5563     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5564     ldr(dst, Address(dummy, rspec));
5565   }
5566 }
5567 
5568 // Move a metadata address into a register.
5569 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5570   int oop_index;
5571   if (obj == nullptr) {
5572     oop_index = oop_recorder()->allocate_metadata_index(obj);
5573   } else {
5574     oop_index = oop_recorder()->find_index(obj);
5575   }
5576   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5577   mov(dst, Address((address)obj, rspec));
5578 }
5579 
5580 Address MacroAssembler::constant_oop_address(jobject obj) {
5581 #ifdef ASSERT
5582   {
5583     ThreadInVMfromUnknown tiv;
5584     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5585     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5586   }
5587 #endif
5588   int oop_index = oop_recorder()->find_index(obj);
5589   return Address((address)obj, oop_Relocation::spec(oop_index));
5590 }
5591 
5592 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5593 void MacroAssembler::tlab_allocate(Register obj,
5594                                    Register var_size_in_bytes,
5595                                    int con_size_in_bytes,
5596                                    Register t1,
5597                                    Register t2,
5598                                    Label& slow_case) {
5599   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5600   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5601 }
5602 
5603 void MacroAssembler::verify_tlab() {
5604 #ifdef ASSERT
5605   if (UseTLAB && VerifyOops) {
5606     Label next, ok;
5607 
5608     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5609 
5610     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5611     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5612     cmp(rscratch2, rscratch1);
5613     br(Assembler::HS, next);
5614     STOP("assert(top >= start)");
5615     should_not_reach_here();
5616 
5617     bind(next);
5618     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5619     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5620     cmp(rscratch2, rscratch1);
5621     br(Assembler::HS, ok);
5622     STOP("assert(top <= end)");
5623     should_not_reach_here();
5624 
5625     bind(ok);
5626     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5627   }
5628 #endif
5629 }
5630 
5631 // Writes to stack successive pages until offset reached to check for
5632 // stack overflow + shadow pages.  This clobbers tmp.
5633 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5634   assert_different_registers(tmp, size, rscratch1);
5635   mov(tmp, sp);
5636   // Bang stack for total size given plus shadow page size.
5637   // Bang one page at a time because large size can bang beyond yellow and
5638   // red zones.
5639   Label loop;
5640   mov(rscratch1, (int)os::vm_page_size());
5641   bind(loop);
5642   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5643   subsw(size, size, rscratch1);
5644   str(size, Address(tmp));
5645   br(Assembler::GT, loop);
5646 
5647   // Bang down shadow pages too.
5648   // At this point, (tmp-0) is the last address touched, so don't
5649   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5650   // was post-decremented.)  Skip this address by starting at i=1, and
5651   // touch a few more pages below.  N.B.  It is important to touch all
5652   // the way down to and including i=StackShadowPages.
5653   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5654     // this could be any sized move but this is can be a debugging crumb
5655     // so the bigger the better.
5656     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5657     str(size, Address(tmp));
5658   }
5659 }
5660 
5661 // Move the address of the polling page into dest.
5662 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5663   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5664 }
5665 
5666 // Read the polling page.  The address of the polling page must
5667 // already be in r.
5668 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5669   address mark;
5670   {
5671     InstructionMark im(this);
5672     code_section()->relocate(inst_mark(), rtype);
5673     ldrw(zr, Address(r, 0));
5674     mark = inst_mark();
5675   }
5676   verify_cross_modify_fence_not_required();
5677   return mark;
5678 }
5679 
5680 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5681   relocInfo::relocType rtype = dest.rspec().reloc()->type();
5682   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5683   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5684   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5685   int64_t offset_low = dest_page - low_page;
5686   int64_t offset_high = dest_page - high_page;
5687 
5688   assert(is_valid_AArch64_address(dest.target()), "bad address");
5689   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5690 
5691   InstructionMark im(this);
5692   code_section()->relocate(inst_mark(), dest.rspec());
5693   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5694   // the code cache so that if it is relocated we know it will still reach
5695   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5696     _adrp(reg1, dest.target());
5697   } else {
5698     uint64_t target = (uint64_t)dest.target();
5699     uint64_t adrp_target
5700       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5701 
5702     _adrp(reg1, (address)adrp_target);
5703     movk(reg1, target >> 32, 32);
5704   }
5705   byte_offset = (uint64_t)dest.target() & 0xfff;
5706 }
5707 
5708 void MacroAssembler::load_byte_map_base(Register reg) {
5709   CardTable::CardValue* byte_map_base =
5710     ((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base();
5711 
5712   // Strictly speaking the byte_map_base isn't an address at all, and it might
5713   // even be negative. It is thus materialised as a constant.
5714   mov(reg, (uint64_t)byte_map_base);
5715 }
5716 
5717 void MacroAssembler::build_frame(int framesize) {
5718   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5719   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5720   protect_return_address();
5721   if (framesize < ((1 << 9) + 2 * wordSize)) {
5722     sub(sp, sp, framesize);
5723     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5724     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5725   } else {
5726     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5727     if (PreserveFramePointer) mov(rfp, sp);
5728     if (framesize < ((1 << 12) + 2 * wordSize))
5729       sub(sp, sp, framesize - 2 * wordSize);
5730     else {
5731       mov(rscratch1, framesize - 2 * wordSize);
5732       sub(sp, sp, rscratch1);
5733     }
5734   }
5735   verify_cross_modify_fence_not_required();
5736 }
5737 
5738 void MacroAssembler::remove_frame(int framesize) {
5739   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5740   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5741   if (framesize < ((1 << 9) + 2 * wordSize)) {
5742     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5743     add(sp, sp, framesize);
5744   } else {
5745     if (framesize < ((1 << 12) + 2 * wordSize))
5746       add(sp, sp, framesize - 2 * wordSize);
5747     else {
5748       mov(rscratch1, framesize - 2 * wordSize);
5749       add(sp, sp, rscratch1);
5750     }
5751     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5752   }
5753   authenticate_return_address();
5754 }
5755 
5756 
5757 // This method counts leading positive bytes (highest bit not set) in provided byte array
5758 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5759     // Simple and most common case of aligned small array which is not at the
5760     // end of memory page is placed here. All other cases are in stub.
5761     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5762     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5763     assert_different_registers(ary1, len, result);
5764 
5765     mov(result, len);
5766     cmpw(len, 0);
5767     br(LE, DONE);
5768     cmpw(len, 4 * wordSize);
5769     br(GE, STUB_LONG); // size > 32 then go to stub
5770 
5771     int shift = 64 - exact_log2(os::vm_page_size());
5772     lsl(rscratch1, ary1, shift);
5773     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5774     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5775     br(CS, STUB); // at the end of page then go to stub
5776     subs(len, len, wordSize);
5777     br(LT, END);
5778 
5779   BIND(LOOP);
5780     ldr(rscratch1, Address(post(ary1, wordSize)));
5781     tst(rscratch1, UPPER_BIT_MASK);
5782     br(NE, SET_RESULT);
5783     subs(len, len, wordSize);
5784     br(GE, LOOP);
5785     cmpw(len, -wordSize);
5786     br(EQ, DONE);
5787 
5788   BIND(END);
5789     ldr(rscratch1, Address(ary1));
5790     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5791     lslv(rscratch1, rscratch1, rscratch2);
5792     tst(rscratch1, UPPER_BIT_MASK);
5793     br(NE, SET_RESULT);
5794     b(DONE);
5795 
5796   BIND(STUB);
5797     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5798     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5799     address tpc1 = trampoline_call(count_pos);
5800     if (tpc1 == nullptr) {
5801       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5802       postcond(pc() == badAddress);
5803       return nullptr;
5804     }
5805     b(DONE);
5806 
5807   BIND(STUB_LONG);
5808     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5809     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5810     address tpc2 = trampoline_call(count_pos_long);
5811     if (tpc2 == nullptr) {
5812       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5813       postcond(pc() == badAddress);
5814       return nullptr;
5815     }
5816     b(DONE);
5817 
5818   BIND(SET_RESULT);
5819 
5820     add(len, len, wordSize);
5821     sub(result, result, len);
5822 
5823   BIND(DONE);
5824   postcond(pc() != badAddress);
5825   return pc();
5826 }
5827 
5828 // Clobbers: rscratch1, rscratch2, rflags
5829 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5830 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5831                                       Register tmp4, Register tmp5, Register result,
5832                                       Register cnt1, int elem_size) {
5833   Label DONE, SAME;
5834   Register tmp1 = rscratch1;
5835   Register tmp2 = rscratch2;
5836   int elem_per_word = wordSize/elem_size;
5837   int log_elem_size = exact_log2(elem_size);
5838   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5839   int length_offset = arrayOopDesc::length_offset_in_bytes();
5840   int base_offset
5841     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5842   // When the length offset is not aligned to 8 bytes,
5843   // then we align it down. This is valid because the new
5844   // offset will always be the klass which is the same
5845   // for type arrays.
5846   int start_offset = align_down(length_offset, BytesPerWord);
5847   int extra_length = base_offset - start_offset;
5848   assert(start_offset == length_offset || start_offset == klass_offset,
5849          "start offset must be 8-byte-aligned or be the klass offset");
5850   assert(base_offset != start_offset, "must include the length field");
5851   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5852   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5853 
5854   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5855   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5856 
5857 #ifndef PRODUCT
5858   {
5859     const char kind = (elem_size == 2) ? 'U' : 'L';
5860     char comment[64];
5861     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5862     BLOCK_COMMENT(comment);
5863   }
5864 #endif
5865 
5866   // if (a1 == a2)
5867   //     return true;
5868   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5869   br(EQ, SAME);
5870 
5871   if (UseSimpleArrayEquals) {
5872     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5873     // if (a1 == nullptr || a2 == nullptr)
5874     //     return false;
5875     // a1 & a2 == 0 means (some-pointer is null) or
5876     // (very-rare-or-even-probably-impossible-pointer-values)
5877     // so, we can save one branch in most cases
5878     tst(a1, a2);
5879     mov(result, false);
5880     br(EQ, A_MIGHT_BE_NULL);
5881     // if (a1.length != a2.length)
5882     //      return false;
5883     bind(A_IS_NOT_NULL);
5884     ldrw(cnt1, Address(a1, length_offset));
5885     // Increase loop counter by diff between base- and actual start-offset.
5886     addw(cnt1, cnt1, extra_length);
5887     lea(a1, Address(a1, start_offset));
5888     lea(a2, Address(a2, start_offset));
5889     // Check for short strings, i.e. smaller than wordSize.
5890     subs(cnt1, cnt1, elem_per_word);
5891     br(Assembler::LT, SHORT);
5892     // Main 8 byte comparison loop.
5893     bind(NEXT_WORD); {
5894       ldr(tmp1, Address(post(a1, wordSize)));
5895       ldr(tmp2, Address(post(a2, wordSize)));
5896       subs(cnt1, cnt1, elem_per_word);
5897       eor(tmp5, tmp1, tmp2);
5898       cbnz(tmp5, DONE);
5899     } br(GT, NEXT_WORD);
5900     // Last longword.  In the case where length == 4 we compare the
5901     // same longword twice, but that's still faster than another
5902     // conditional branch.
5903     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5904     // length == 4.
5905     if (log_elem_size > 0)
5906       lsl(cnt1, cnt1, log_elem_size);
5907     ldr(tmp3, Address(a1, cnt1));
5908     ldr(tmp4, Address(a2, cnt1));
5909     eor(tmp5, tmp3, tmp4);
5910     cbnz(tmp5, DONE);
5911     b(SAME);
5912     bind(A_MIGHT_BE_NULL);
5913     // in case both a1 and a2 are not-null, proceed with loads
5914     cbz(a1, DONE);
5915     cbz(a2, DONE);
5916     b(A_IS_NOT_NULL);
5917     bind(SHORT);
5918 
5919     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
5920     {
5921       ldrw(tmp1, Address(post(a1, 4)));
5922       ldrw(tmp2, Address(post(a2, 4)));
5923       eorw(tmp5, tmp1, tmp2);
5924       cbnzw(tmp5, DONE);
5925     }
5926     bind(TAIL03);
5927     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
5928     {
5929       ldrh(tmp3, Address(post(a1, 2)));
5930       ldrh(tmp4, Address(post(a2, 2)));
5931       eorw(tmp5, tmp3, tmp4);
5932       cbnzw(tmp5, DONE);
5933     }
5934     bind(TAIL01);
5935     if (elem_size == 1) { // Only needed when comparing byte arrays.
5936       tbz(cnt1, 0, SAME); // 0-1 bytes left.
5937       {
5938         ldrb(tmp1, a1);
5939         ldrb(tmp2, a2);
5940         eorw(tmp5, tmp1, tmp2);
5941         cbnzw(tmp5, DONE);
5942       }
5943     }
5944   } else {
5945     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
5946         CSET_EQ, LAST_CHECK;
5947     mov(result, false);
5948     cbz(a1, DONE);
5949     ldrw(cnt1, Address(a1, length_offset));
5950     cbz(a2, DONE);
5951     // Increase loop counter by diff between base- and actual start-offset.
5952     addw(cnt1, cnt1, extra_length);
5953 
5954     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
5955     // faster to perform another branch before comparing a1 and a2
5956     cmp(cnt1, (u1)elem_per_word);
5957     br(LE, SHORT); // short or same
5958     ldr(tmp3, Address(pre(a1, start_offset)));
5959     subs(zr, cnt1, stubBytesThreshold);
5960     br(GE, STUB);
5961     ldr(tmp4, Address(pre(a2, start_offset)));
5962     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
5963 
5964     // Main 16 byte comparison loop with 2 exits
5965     bind(NEXT_DWORD); {
5966       ldr(tmp1, Address(pre(a1, wordSize)));
5967       ldr(tmp2, Address(pre(a2, wordSize)));
5968       subs(cnt1, cnt1, 2 * elem_per_word);
5969       br(LE, TAIL);
5970       eor(tmp4, tmp3, tmp4);
5971       cbnz(tmp4, DONE);
5972       ldr(tmp3, Address(pre(a1, wordSize)));
5973       ldr(tmp4, Address(pre(a2, wordSize)));
5974       cmp(cnt1, (u1)elem_per_word);
5975       br(LE, TAIL2);
5976       cmp(tmp1, tmp2);
5977     } br(EQ, NEXT_DWORD);
5978     b(DONE);
5979 
5980     bind(TAIL);
5981     eor(tmp4, tmp3, tmp4);
5982     eor(tmp2, tmp1, tmp2);
5983     lslv(tmp2, tmp2, tmp5);
5984     orr(tmp5, tmp4, tmp2);
5985     cmp(tmp5, zr);
5986     b(CSET_EQ);
5987 
5988     bind(TAIL2);
5989     eor(tmp2, tmp1, tmp2);
5990     cbnz(tmp2, DONE);
5991     b(LAST_CHECK);
5992 
5993     bind(STUB);
5994     ldr(tmp4, Address(pre(a2, start_offset)));
5995     if (elem_size == 2) { // convert to byte counter
5996       lsl(cnt1, cnt1, 1);
5997     }
5998     eor(tmp5, tmp3, tmp4);
5999     cbnz(tmp5, DONE);
6000     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6001     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6002     address tpc = trampoline_call(stub);
6003     if (tpc == nullptr) {
6004       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6005       postcond(pc() == badAddress);
6006       return nullptr;
6007     }
6008     b(DONE);
6009 
6010     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6011     // so, if a2 == null => return false(0), else return true, so we can return a2
6012     mov(result, a2);
6013     b(DONE);
6014     bind(SHORT);
6015     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6016     ldr(tmp3, Address(a1, start_offset));
6017     ldr(tmp4, Address(a2, start_offset));
6018     bind(LAST_CHECK);
6019     eor(tmp4, tmp3, tmp4);
6020     lslv(tmp5, tmp4, tmp5);
6021     cmp(tmp5, zr);
6022     bind(CSET_EQ);
6023     cset(result, EQ);
6024     b(DONE);
6025   }
6026 
6027   bind(SAME);
6028   mov(result, true);
6029   // That's it.
6030   bind(DONE);
6031 
6032   BLOCK_COMMENT("} array_equals");
6033   postcond(pc() != badAddress);
6034   return pc();
6035 }
6036 
6037 // Compare Strings
6038 
6039 // For Strings we're passed the address of the first characters in a1
6040 // and a2 and the length in cnt1.
6041 // There are two implementations.  For arrays >= 8 bytes, all
6042 // comparisons (including the final one, which may overlap) are
6043 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6044 // halfword, then a short, and then a byte.
6045 
6046 void MacroAssembler::string_equals(Register a1, Register a2,
6047                                    Register result, Register cnt1)
6048 {
6049   Label SAME, DONE, SHORT, NEXT_WORD;
6050   Register tmp1 = rscratch1;
6051   Register tmp2 = rscratch2;
6052   Register cnt2 = tmp2;  // cnt2 only used in array length compare
6053 
6054   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6055 
6056 #ifndef PRODUCT
6057   {
6058     char comment[64];
6059     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6060     BLOCK_COMMENT(comment);
6061   }
6062 #endif
6063 
6064   mov(result, false);
6065 
6066   // Check for short strings, i.e. smaller than wordSize.
6067   subs(cnt1, cnt1, wordSize);
6068   br(Assembler::LT, SHORT);
6069   // Main 8 byte comparison loop.
6070   bind(NEXT_WORD); {
6071     ldr(tmp1, Address(post(a1, wordSize)));
6072     ldr(tmp2, Address(post(a2, wordSize)));
6073     subs(cnt1, cnt1, wordSize);
6074     eor(tmp1, tmp1, tmp2);
6075     cbnz(tmp1, DONE);
6076   } br(GT, NEXT_WORD);
6077   // Last longword.  In the case where length == 4 we compare the
6078   // same longword twice, but that's still faster than another
6079   // conditional branch.
6080   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6081   // length == 4.
6082   ldr(tmp1, Address(a1, cnt1));
6083   ldr(tmp2, Address(a2, cnt1));
6084   eor(tmp2, tmp1, tmp2);
6085   cbnz(tmp2, DONE);
6086   b(SAME);
6087 
6088   bind(SHORT);
6089   Label TAIL03, TAIL01;
6090 
6091   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6092   {
6093     ldrw(tmp1, Address(post(a1, 4)));
6094     ldrw(tmp2, Address(post(a2, 4)));
6095     eorw(tmp1, tmp1, tmp2);
6096     cbnzw(tmp1, DONE);
6097   }
6098   bind(TAIL03);
6099   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6100   {
6101     ldrh(tmp1, Address(post(a1, 2)));
6102     ldrh(tmp2, Address(post(a2, 2)));
6103     eorw(tmp1, tmp1, tmp2);
6104     cbnzw(tmp1, DONE);
6105   }
6106   bind(TAIL01);
6107   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6108     {
6109     ldrb(tmp1, a1);
6110     ldrb(tmp2, a2);
6111     eorw(tmp1, tmp1, tmp2);
6112     cbnzw(tmp1, DONE);
6113   }
6114   // Arrays are equal.
6115   bind(SAME);
6116   mov(result, true);
6117 
6118   // That's it.
6119   bind(DONE);
6120   BLOCK_COMMENT("} string_equals");
6121 }
6122 
6123 
6124 // The size of the blocks erased by the zero_blocks stub.  We must
6125 // handle anything smaller than this ourselves in zero_words().
6126 const int MacroAssembler::zero_words_block_size = 8;
6127 
6128 // zero_words() is used by C2 ClearArray patterns and by
6129 // C1_MacroAssembler.  It is as small as possible, handling small word
6130 // counts locally and delegating anything larger to the zero_blocks
6131 // stub.  It is expanded many times in compiled code, so it is
6132 // important to keep it short.
6133 
6134 // ptr:   Address of a buffer to be zeroed.
6135 // cnt:   Count in HeapWords.
6136 //
6137 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6138 address MacroAssembler::zero_words(Register ptr, Register cnt)
6139 {
6140   assert(is_power_of_2(zero_words_block_size), "adjust this");
6141 
6142   BLOCK_COMMENT("zero_words {");
6143   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6144   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6145   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6146 
6147   subs(rscratch1, cnt, zero_words_block_size);
6148   Label around;
6149   br(LO, around);
6150   {
6151     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6152     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6153     // Make sure this is a C2 compilation. C1 allocates space only for
6154     // trampoline stubs generated by Call LIR ops, and in any case it
6155     // makes sense for a C1 compilation task to proceed as quickly as
6156     // possible.
6157     CompileTask* task;
6158     if (StubRoutines::aarch64::complete()
6159         && Thread::current()->is_Compiler_thread()
6160         && (task = ciEnv::current()->task())
6161         && is_c2_compile(task->comp_level())) {
6162       address tpc = trampoline_call(zero_blocks);
6163       if (tpc == nullptr) {
6164         DEBUG_ONLY(reset_labels(around));
6165         return nullptr;
6166       }
6167     } else {
6168       far_call(zero_blocks);
6169     }
6170   }
6171   bind(around);
6172 
6173   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6174   // for us.
6175   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6176     Label l;
6177     tbz(cnt, exact_log2(i), l);
6178     for (int j = 0; j < i; j += 2) {
6179       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6180     }
6181     bind(l);
6182   }
6183   {
6184     Label l;
6185     tbz(cnt, 0, l);
6186     str(zr, Address(ptr));
6187     bind(l);
6188   }
6189 
6190   BLOCK_COMMENT("} zero_words");
6191   return pc();
6192 }
6193 
6194 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6195 // cnt:          Immediate count in HeapWords.
6196 //
6197 // r10, r11, rscratch1, and rscratch2 are clobbered.
6198 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6199 {
6200   assert(wordSize <= BlockZeroingLowLimit,
6201             "increase BlockZeroingLowLimit");
6202   address result = nullptr;
6203   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6204 #ifndef PRODUCT
6205     {
6206       char buf[64];
6207       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6208       BLOCK_COMMENT(buf);
6209     }
6210 #endif
6211     if (cnt >= 16) {
6212       uint64_t loops = cnt/16;
6213       if (loops > 1) {
6214         mov(rscratch2, loops - 1);
6215       }
6216       {
6217         Label loop;
6218         bind(loop);
6219         for (int i = 0; i < 16; i += 2) {
6220           stp(zr, zr, Address(base, i * BytesPerWord));
6221         }
6222         add(base, base, 16 * BytesPerWord);
6223         if (loops > 1) {
6224           subs(rscratch2, rscratch2, 1);
6225           br(GE, loop);
6226         }
6227       }
6228     }
6229     cnt %= 16;
6230     int i = cnt & 1;  // store any odd word to start
6231     if (i) str(zr, Address(base));
6232     for (; i < (int)cnt; i += 2) {
6233       stp(zr, zr, Address(base, i * wordSize));
6234     }
6235     BLOCK_COMMENT("} zero_words");
6236     result = pc();
6237   } else {
6238     mov(r10, base); mov(r11, cnt);
6239     result = zero_words(r10, r11);
6240   }
6241   return result;
6242 }
6243 
6244 // Zero blocks of memory by using DC ZVA.
6245 //
6246 // Aligns the base address first sufficiently for DC ZVA, then uses
6247 // DC ZVA repeatedly for every full block.  cnt is the size to be
6248 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6249 // in cnt.
6250 //
6251 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6252 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6253 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6254   Register tmp = rscratch1;
6255   Register tmp2 = rscratch2;
6256   int zva_length = VM_Version::zva_length();
6257   Label initial_table_end, loop_zva;
6258   Label fini;
6259 
6260   // Base must be 16 byte aligned. If not just return and let caller handle it
6261   tst(base, 0x0f);
6262   br(Assembler::NE, fini);
6263   // Align base with ZVA length.
6264   neg(tmp, base);
6265   andr(tmp, tmp, zva_length - 1);
6266 
6267   // tmp: the number of bytes to be filled to align the base with ZVA length.
6268   add(base, base, tmp);
6269   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6270   adr(tmp2, initial_table_end);
6271   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6272   br(tmp2);
6273 
6274   for (int i = -zva_length + 16; i < 0; i += 16)
6275     stp(zr, zr, Address(base, i));
6276   bind(initial_table_end);
6277 
6278   sub(cnt, cnt, zva_length >> 3);
6279   bind(loop_zva);
6280   dc(Assembler::ZVA, base);
6281   subs(cnt, cnt, zva_length >> 3);
6282   add(base, base, zva_length);
6283   br(Assembler::GE, loop_zva);
6284   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6285   bind(fini);
6286 }
6287 
6288 // base:   Address of a buffer to be filled, 8 bytes aligned.
6289 // cnt:    Count in 8-byte unit.
6290 // value:  Value to be filled with.
6291 // base will point to the end of the buffer after filling.
6292 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6293 {
6294 //  Algorithm:
6295 //
6296 //    if (cnt == 0) {
6297 //      return;
6298 //    }
6299 //    if ((p & 8) != 0) {
6300 //      *p++ = v;
6301 //    }
6302 //
6303 //    scratch1 = cnt & 14;
6304 //    cnt -= scratch1;
6305 //    p += scratch1;
6306 //    switch (scratch1 / 2) {
6307 //      do {
6308 //        cnt -= 16;
6309 //          p[-16] = v;
6310 //          p[-15] = v;
6311 //        case 7:
6312 //          p[-14] = v;
6313 //          p[-13] = v;
6314 //        case 6:
6315 //          p[-12] = v;
6316 //          p[-11] = v;
6317 //          // ...
6318 //        case 1:
6319 //          p[-2] = v;
6320 //          p[-1] = v;
6321 //        case 0:
6322 //          p += 16;
6323 //      } while (cnt);
6324 //    }
6325 //    if ((cnt & 1) == 1) {
6326 //      *p++ = v;
6327 //    }
6328 
6329   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6330 
6331   Label fini, skip, entry, loop;
6332   const int unroll = 8; // Number of stp instructions we'll unroll
6333 
6334   cbz(cnt, fini);
6335   tbz(base, 3, skip);
6336   str(value, Address(post(base, 8)));
6337   sub(cnt, cnt, 1);
6338   bind(skip);
6339 
6340   andr(rscratch1, cnt, (unroll-1) * 2);
6341   sub(cnt, cnt, rscratch1);
6342   add(base, base, rscratch1, Assembler::LSL, 3);
6343   adr(rscratch2, entry);
6344   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6345   br(rscratch2);
6346 
6347   bind(loop);
6348   add(base, base, unroll * 16);
6349   for (int i = -unroll; i < 0; i++)
6350     stp(value, value, Address(base, i * 16));
6351   bind(entry);
6352   subs(cnt, cnt, unroll * 2);
6353   br(Assembler::GE, loop);
6354 
6355   tbz(cnt, 0, fini);
6356   str(value, Address(post(base, 8)));
6357   bind(fini);
6358 }
6359 
6360 // Intrinsic for
6361 //
6362 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6363 //   Encodes char[] to byte[] in ISO-8859-1
6364 //
6365 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6366 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6367 //
6368 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6369 //   Encodes char[] to byte[] in ASCII
6370 //
6371 // This version always returns the number of characters copied, and does not
6372 // clobber the 'len' register. A successful copy will complete with the post-
6373 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6374 // post-condition: 0 <= 'res' < 'len'.
6375 //
6376 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6377 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6378 //       beyond the acceptable, even though the footprint would be smaller.
6379 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6380 //       avoid additional bloat.
6381 //
6382 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6383 void MacroAssembler::encode_iso_array(Register src, Register dst,
6384                                       Register len, Register res, bool ascii,
6385                                       FloatRegister vtmp0, FloatRegister vtmp1,
6386                                       FloatRegister vtmp2, FloatRegister vtmp3,
6387                                       FloatRegister vtmp4, FloatRegister vtmp5)
6388 {
6389   Register cnt = res;
6390   Register max = rscratch1;
6391   Register chk = rscratch2;
6392 
6393   prfm(Address(src), PLDL1STRM);
6394   movw(cnt, len);
6395 
6396 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6397 
6398   Label LOOP_32, DONE_32, FAIL_32;
6399 
6400   BIND(LOOP_32);
6401   {
6402     cmpw(cnt, 32);
6403     br(LT, DONE_32);
6404     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6405     // Extract lower bytes.
6406     FloatRegister vlo0 = vtmp4;
6407     FloatRegister vlo1 = vtmp5;
6408     uzp1(vlo0, T16B, vtmp0, vtmp1);
6409     uzp1(vlo1, T16B, vtmp2, vtmp3);
6410     // Merge bits...
6411     orr(vtmp0, T16B, vtmp0, vtmp1);
6412     orr(vtmp2, T16B, vtmp2, vtmp3);
6413     // Extract merged upper bytes.
6414     FloatRegister vhix = vtmp0;
6415     uzp2(vhix, T16B, vtmp0, vtmp2);
6416     // ISO-check on hi-parts (all zero).
6417     //                          ASCII-check on lo-parts (no sign).
6418     FloatRegister vlox = vtmp1; // Merge lower bytes.
6419                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6420     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6421     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6422     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6423                                 ASCII(orr(chk, chk, max));
6424     cbnz(chk, FAIL_32);
6425     subw(cnt, cnt, 32);
6426     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6427     b(LOOP_32);
6428   }
6429   BIND(FAIL_32);
6430   sub(src, src, 64);
6431   BIND(DONE_32);
6432 
6433   Label LOOP_8, SKIP_8;
6434 
6435   BIND(LOOP_8);
6436   {
6437     cmpw(cnt, 8);
6438     br(LT, SKIP_8);
6439     FloatRegister vhi = vtmp0;
6440     FloatRegister vlo = vtmp1;
6441     ld1(vtmp3, T8H, src);
6442     uzp1(vlo, T16B, vtmp3, vtmp3);
6443     uzp2(vhi, T16B, vtmp3, vtmp3);
6444     // ISO-check on hi-parts (all zero).
6445     //                          ASCII-check on lo-parts (no sign).
6446                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6447     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6448                                 ASCII(umov(max, vtmp2, B, 0));
6449                                 ASCII(orr(chk, chk, max));
6450     cbnz(chk, SKIP_8);
6451 
6452     strd(vlo, Address(post(dst, 8)));
6453     subw(cnt, cnt, 8);
6454     add(src, src, 16);
6455     b(LOOP_8);
6456   }
6457   BIND(SKIP_8);
6458 
6459 #undef ASCII
6460 
6461   Label LOOP, DONE;
6462 
6463   cbz(cnt, DONE);
6464   BIND(LOOP);
6465   {
6466     Register chr = rscratch1;
6467     ldrh(chr, Address(post(src, 2)));
6468     tst(chr, ascii ? 0xff80 : 0xff00);
6469     br(NE, DONE);
6470     strb(chr, Address(post(dst, 1)));
6471     subs(cnt, cnt, 1);
6472     br(GT, LOOP);
6473   }
6474   BIND(DONE);
6475   // Return index where we stopped.
6476   subw(res, len, cnt);
6477 }
6478 
6479 // Inflate byte[] array to char[].
6480 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6481 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6482                                            FloatRegister vtmp1, FloatRegister vtmp2,
6483                                            FloatRegister vtmp3, Register tmp4) {
6484   Label big, done, after_init, to_stub;
6485 
6486   assert_different_registers(src, dst, len, tmp4, rscratch1);
6487 
6488   fmovd(vtmp1, 0.0);
6489   lsrw(tmp4, len, 3);
6490   bind(after_init);
6491   cbnzw(tmp4, big);
6492   // Short string: less than 8 bytes.
6493   {
6494     Label loop, tiny;
6495 
6496     cmpw(len, 4);
6497     br(LT, tiny);
6498     // Use SIMD to do 4 bytes.
6499     ldrs(vtmp2, post(src, 4));
6500     zip1(vtmp3, T8B, vtmp2, vtmp1);
6501     subw(len, len, 4);
6502     strd(vtmp3, post(dst, 8));
6503 
6504     cbzw(len, done);
6505 
6506     // Do the remaining bytes by steam.
6507     bind(loop);
6508     ldrb(tmp4, post(src, 1));
6509     strh(tmp4, post(dst, 2));
6510     subw(len, len, 1);
6511 
6512     bind(tiny);
6513     cbnz(len, loop);
6514 
6515     b(done);
6516   }
6517 
6518   if (SoftwarePrefetchHintDistance >= 0) {
6519     bind(to_stub);
6520       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6521       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6522       address tpc = trampoline_call(stub);
6523       if (tpc == nullptr) {
6524         DEBUG_ONLY(reset_labels(big, done));
6525         postcond(pc() == badAddress);
6526         return nullptr;
6527       }
6528       b(after_init);
6529   }
6530 
6531   // Unpack the bytes 8 at a time.
6532   bind(big);
6533   {
6534     Label loop, around, loop_last, loop_start;
6535 
6536     if (SoftwarePrefetchHintDistance >= 0) {
6537       const int large_loop_threshold = (64 + 16)/8;
6538       ldrd(vtmp2, post(src, 8));
6539       andw(len, len, 7);
6540       cmp(tmp4, (u1)large_loop_threshold);
6541       br(GE, to_stub);
6542       b(loop_start);
6543 
6544       bind(loop);
6545       ldrd(vtmp2, post(src, 8));
6546       bind(loop_start);
6547       subs(tmp4, tmp4, 1);
6548       br(EQ, loop_last);
6549       zip1(vtmp2, T16B, vtmp2, vtmp1);
6550       ldrd(vtmp3, post(src, 8));
6551       st1(vtmp2, T8H, post(dst, 16));
6552       subs(tmp4, tmp4, 1);
6553       zip1(vtmp3, T16B, vtmp3, vtmp1);
6554       st1(vtmp3, T8H, post(dst, 16));
6555       br(NE, loop);
6556       b(around);
6557       bind(loop_last);
6558       zip1(vtmp2, T16B, vtmp2, vtmp1);
6559       st1(vtmp2, T8H, post(dst, 16));
6560       bind(around);
6561       cbz(len, done);
6562     } else {
6563       andw(len, len, 7);
6564       bind(loop);
6565       ldrd(vtmp2, post(src, 8));
6566       sub(tmp4, tmp4, 1);
6567       zip1(vtmp3, T16B, vtmp2, vtmp1);
6568       st1(vtmp3, T8H, post(dst, 16));
6569       cbnz(tmp4, loop);
6570     }
6571   }
6572 
6573   // Do the tail of up to 8 bytes.
6574   add(src, src, len);
6575   ldrd(vtmp3, Address(src, -8));
6576   add(dst, dst, len, ext::uxtw, 1);
6577   zip1(vtmp3, T16B, vtmp3, vtmp1);
6578   strq(vtmp3, Address(dst, -16));
6579 
6580   bind(done);
6581   postcond(pc() != badAddress);
6582   return pc();
6583 }
6584 
6585 // Compress char[] array to byte[].
6586 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6587 // Return the array length if every element in array can be encoded,
6588 // otherwise, the index of first non-latin1 (> 0xff) character.
6589 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6590                                          Register res,
6591                                          FloatRegister tmp0, FloatRegister tmp1,
6592                                          FloatRegister tmp2, FloatRegister tmp3,
6593                                          FloatRegister tmp4, FloatRegister tmp5) {
6594   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6595 }
6596 
6597 // java.math.round(double a)
6598 // Returns the closest long to the argument, with ties rounding to
6599 // positive infinity.  This requires some fiddling for corner
6600 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6601 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6602                                        FloatRegister ftmp) {
6603   Label DONE;
6604   BLOCK_COMMENT("java_round_double: { ");
6605   fmovd(rscratch1, src);
6606   // Use RoundToNearestTiesAway unless src small and -ve.
6607   fcvtasd(dst, src);
6608   // Test if src >= 0 || abs(src) >= 0x1.0p52
6609   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6610   mov(rscratch2, julong_cast(0x1.0p52));
6611   cmp(rscratch1, rscratch2);
6612   br(HS, DONE); {
6613     // src < 0 && abs(src) < 0x1.0p52
6614     // src may have a fractional part, so add 0.5
6615     fmovd(ftmp, 0.5);
6616     faddd(ftmp, src, ftmp);
6617     // Convert double to jlong, use RoundTowardsNegative
6618     fcvtmsd(dst, ftmp);
6619   }
6620   bind(DONE);
6621   BLOCK_COMMENT("} java_round_double");
6622 }
6623 
6624 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6625                                       FloatRegister ftmp) {
6626   Label DONE;
6627   BLOCK_COMMENT("java_round_float: { ");
6628   fmovs(rscratch1, src);
6629   // Use RoundToNearestTiesAway unless src small and -ve.
6630   fcvtassw(dst, src);
6631   // Test if src >= 0 || abs(src) >= 0x1.0p23
6632   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6633   mov(rscratch2, jint_cast(0x1.0p23f));
6634   cmp(rscratch1, rscratch2);
6635   br(HS, DONE); {
6636     // src < 0 && |src| < 0x1.0p23
6637     // src may have a fractional part, so add 0.5
6638     fmovs(ftmp, 0.5f);
6639     fadds(ftmp, src, ftmp);
6640     // Convert float to jint, use RoundTowardsNegative
6641     fcvtmssw(dst, ftmp);
6642   }
6643   bind(DONE);
6644   BLOCK_COMMENT("} java_round_float");
6645 }
6646 
6647 // get_thread() can be called anywhere inside generated code so we
6648 // need to save whatever non-callee save context might get clobbered
6649 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6650 // the call setup code.
6651 //
6652 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6653 // On other systems, the helper is a usual C function.
6654 //
6655 void MacroAssembler::get_thread(Register dst) {
6656   RegSet saved_regs =
6657     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6658     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6659 
6660   protect_return_address();
6661   push(saved_regs, sp);
6662 
6663   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6664   blr(lr);
6665   if (dst != c_rarg0) {
6666     mov(dst, c_rarg0);
6667   }
6668 
6669   pop(saved_regs, sp);
6670   authenticate_return_address();
6671 }
6672 
6673 void MacroAssembler::cache_wb(Address line) {
6674   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6675   assert(line.index() == noreg, "index should be noreg");
6676   assert(line.offset() == 0, "offset should be 0");
6677   // would like to assert this
6678   // assert(line._ext.shift == 0, "shift should be zero");
6679   if (VM_Version::supports_dcpop()) {
6680     // writeback using clear virtual address to point of persistence
6681     dc(Assembler::CVAP, line.base());
6682   } else {
6683     // no need to generate anything as Unsafe.writebackMemory should
6684     // never invoke this stub
6685   }
6686 }
6687 
6688 void MacroAssembler::cache_wbsync(bool is_pre) {
6689   // we only need a barrier post sync
6690   if (!is_pre) {
6691     membar(Assembler::AnyAny);
6692   }
6693 }
6694 
6695 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6696   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6697     return;
6698   }
6699   // Make sure that native code does not change SVE vector length.
6700   Label verify_ok;
6701   movw(tmp, zr);
6702   sve_inc(tmp, B);
6703   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6704   br(EQ, verify_ok);
6705   stop("Error: SVE vector length has changed since jvm startup");
6706   bind(verify_ok);
6707 }
6708 
6709 void MacroAssembler::verify_ptrue() {
6710   Label verify_ok;
6711   if (!UseSVE) {
6712     return;
6713   }
6714   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6715   sve_dec(rscratch1, B);
6716   cbz(rscratch1, verify_ok);
6717   stop("Error: the preserved predicate register (p7) elements are not all true");
6718   bind(verify_ok);
6719 }
6720 
6721 void MacroAssembler::safepoint_isb() {
6722   isb();
6723 #ifndef PRODUCT
6724   if (VerifyCrossModifyFence) {
6725     // Clear the thread state.
6726     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6727   }
6728 #endif
6729 }
6730 
6731 #ifndef PRODUCT
6732 void MacroAssembler::verify_cross_modify_fence_not_required() {
6733   if (VerifyCrossModifyFence) {
6734     // Check if thread needs a cross modify fence.
6735     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6736     Label fence_not_required;
6737     cbz(rscratch1, fence_not_required);
6738     // If it does then fail.
6739     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6740     mov(c_rarg0, rthread);
6741     blr(rscratch1);
6742     bind(fence_not_required);
6743   }
6744 }
6745 #endif
6746 
6747 void MacroAssembler::spin_wait() {
6748   block_comment("spin_wait {");
6749   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6750     switch (VM_Version::spin_wait_desc().inst()) {
6751       case SpinWait::NOP:
6752         nop();
6753         break;
6754       case SpinWait::ISB:
6755         isb();
6756         break;
6757       case SpinWait::YIELD:
6758         yield();
6759         break;
6760       case SpinWait::SB:
6761         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6762         sb();
6763         break;
6764       default:
6765         ShouldNotReachHere();
6766     }
6767   }
6768   block_comment("}");
6769 }
6770 
6771 // Stack frame creation/removal
6772 
6773 void MacroAssembler::enter(bool strip_ret_addr) {
6774   if (strip_ret_addr) {
6775     // Addresses can only be signed once. If there are multiple nested frames being created
6776     // in the same function, then the return address needs stripping first.
6777     strip_return_address();
6778   }
6779   protect_return_address();
6780   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6781   mov(rfp, sp);
6782 }
6783 
6784 void MacroAssembler::leave() {
6785   mov(sp, rfp);
6786   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6787   authenticate_return_address();
6788 }
6789 
6790 // ROP Protection
6791 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6792 // destroying stack frames or whenever directly loading/storing the LR to memory.
6793 // If ROP protection is not set then these functions are no-ops.
6794 // For more details on PAC see pauth_aarch64.hpp.
6795 
6796 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6797 // Uses value zero as the modifier.
6798 //
6799 void MacroAssembler::protect_return_address() {
6800   if (VM_Version::use_rop_protection()) {
6801     check_return_address();
6802     paciaz();
6803   }
6804 }
6805 
6806 // Sign the return value in the given register. Use before updating the LR in the existing stack
6807 // frame for the current function.
6808 // Uses value zero as the modifier.
6809 //
6810 void MacroAssembler::protect_return_address(Register return_reg) {
6811   if (VM_Version::use_rop_protection()) {
6812     check_return_address(return_reg);
6813     paciza(return_reg);
6814   }
6815 }
6816 
6817 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6818 // Uses value zero as the modifier.
6819 //
6820 void MacroAssembler::authenticate_return_address() {
6821   if (VM_Version::use_rop_protection()) {
6822     autiaz();
6823     check_return_address();
6824   }
6825 }
6826 
6827 // Authenticate the return value in the given register. Use before updating the LR in the existing
6828 // stack frame for the current function.
6829 // Uses value zero as the modifier.
6830 //
6831 void MacroAssembler::authenticate_return_address(Register return_reg) {
6832   if (VM_Version::use_rop_protection()) {
6833     autiza(return_reg);
6834     check_return_address(return_reg);
6835   }
6836 }
6837 
6838 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6839 // there is no guaranteed way of authenticating the LR.
6840 //
6841 void MacroAssembler::strip_return_address() {
6842   if (VM_Version::use_rop_protection()) {
6843     xpaclri();
6844   }
6845 }
6846 
6847 #ifndef PRODUCT
6848 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6849 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6850 // it is difficult to debug back to the callee function.
6851 // This function simply loads from the address in the given register.
6852 // Use directly after authentication to catch authentication failures.
6853 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6854 //
6855 void MacroAssembler::check_return_address(Register return_reg) {
6856   if (VM_Version::use_rop_protection()) {
6857     ldr(zr, Address(return_reg));
6858   }
6859 }
6860 #endif
6861 
6862 // The java_calling_convention describes stack locations as ideal slots on
6863 // a frame with no abi restrictions. Since we must observe abi restrictions
6864 // (like the placement of the register window) the slots must be biased by
6865 // the following value.
6866 static int reg2offset_in(VMReg r) {
6867   // Account for saved rfp and lr
6868   // This should really be in_preserve_stack_slots
6869   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6870 }
6871 
6872 static int reg2offset_out(VMReg r) {
6873   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6874 }
6875 
6876 // On 64bit we will store integer like items to the stack as
6877 // 64bits items (AArch64 ABI) even though java would only store
6878 // 32bits for a parameter. On 32bit it will simply be 32bits
6879 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6880 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6881   if (src.first()->is_stack()) {
6882     if (dst.first()->is_stack()) {
6883       // stack to stack
6884       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6885       str(tmp, Address(sp, reg2offset_out(dst.first())));
6886     } else {
6887       // stack to reg
6888       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6889     }
6890   } else if (dst.first()->is_stack()) {
6891     // reg to stack
6892     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6893   } else {
6894     if (dst.first() != src.first()) {
6895       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6896     }
6897   }
6898 }
6899 
6900 // An oop arg. Must pass a handle not the oop itself
6901 void MacroAssembler::object_move(
6902                         OopMap* map,
6903                         int oop_handle_offset,
6904                         int framesize_in_slots,
6905                         VMRegPair src,
6906                         VMRegPair dst,
6907                         bool is_receiver,
6908                         int* receiver_offset) {
6909 
6910   // must pass a handle. First figure out the location we use as a handle
6911 
6912   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
6913 
6914   // See if oop is null if it is we need no handle
6915 
6916   if (src.first()->is_stack()) {
6917 
6918     // Oop is already on the stack as an argument
6919     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
6920     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
6921     if (is_receiver) {
6922       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
6923     }
6924 
6925     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
6926     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
6927     // conditionally move a null
6928     cmp(rscratch1, zr);
6929     csel(rHandle, zr, rHandle, Assembler::EQ);
6930   } else {
6931 
6932     // Oop is in an a register we must store it to the space we reserve
6933     // on the stack for oop_handles and pass a handle if oop is non-null
6934 
6935     const Register rOop = src.first()->as_Register();
6936     int oop_slot;
6937     if (rOop == j_rarg0)
6938       oop_slot = 0;
6939     else if (rOop == j_rarg1)
6940       oop_slot = 1;
6941     else if (rOop == j_rarg2)
6942       oop_slot = 2;
6943     else if (rOop == j_rarg3)
6944       oop_slot = 3;
6945     else if (rOop == j_rarg4)
6946       oop_slot = 4;
6947     else if (rOop == j_rarg5)
6948       oop_slot = 5;
6949     else if (rOop == j_rarg6)
6950       oop_slot = 6;
6951     else {
6952       assert(rOop == j_rarg7, "wrong register");
6953       oop_slot = 7;
6954     }
6955 
6956     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
6957     int offset = oop_slot*VMRegImpl::stack_slot_size;
6958 
6959     map->set_oop(VMRegImpl::stack2reg(oop_slot));
6960     // Store oop in handle area, may be null
6961     str(rOop, Address(sp, offset));
6962     if (is_receiver) {
6963       *receiver_offset = offset;
6964     }
6965 
6966     cmp(rOop, zr);
6967     lea(rHandle, Address(sp, offset));
6968     // conditionally move a null
6969     csel(rHandle, zr, rHandle, Assembler::EQ);
6970   }
6971 
6972   // If arg is on the stack then place it otherwise it is already in correct reg.
6973   if (dst.first()->is_stack()) {
6974     str(rHandle, Address(sp, reg2offset_out(dst.first())));
6975   }
6976 }
6977 
6978 // A float arg may have to do float reg int reg conversion
6979 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
6980  if (src.first()->is_stack()) {
6981     if (dst.first()->is_stack()) {
6982       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
6983       strw(tmp, Address(sp, reg2offset_out(dst.first())));
6984     } else {
6985       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
6986     }
6987   } else if (src.first() != dst.first()) {
6988     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
6989       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
6990     else
6991       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
6992   }
6993 }
6994 
6995 // A long move
6996 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
6997   if (src.first()->is_stack()) {
6998     if (dst.first()->is_stack()) {
6999       // stack to stack
7000       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7001       str(tmp, Address(sp, reg2offset_out(dst.first())));
7002     } else {
7003       // stack to reg
7004       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7005     }
7006   } else if (dst.first()->is_stack()) {
7007     // reg to stack
7008     // Do we really have to sign extend???
7009     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7010     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7011   } else {
7012     if (dst.first() != src.first()) {
7013       mov(dst.first()->as_Register(), src.first()->as_Register());
7014     }
7015   }
7016 }
7017 
7018 
7019 // A double move
7020 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7021  if (src.first()->is_stack()) {
7022     if (dst.first()->is_stack()) {
7023       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7024       str(tmp, Address(sp, reg2offset_out(dst.first())));
7025     } else {
7026       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7027     }
7028   } else if (src.first() != dst.first()) {
7029     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7030       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7031     else
7032       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7033   }
7034 }
7035 
7036 // Implements lightweight-locking.
7037 //
7038 //  - obj: the object to be locked
7039 //  - t1, t2, t3: temporary registers, will be destroyed
7040 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7041 void MacroAssembler::lightweight_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7042   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7043 
7044   Label push;
7045   const Register top = t1;
7046   const Register mark = t2;
7047   const Register t = t3;
7048 
7049   // Preload the markWord. It is important that this is the first
7050   // instruction emitted as it is part of C1's null check semantics.
7051   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7052 
7053   if (UseObjectMonitorTable) {
7054     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7055     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7056   }
7057 
7058   if (DiagnoseSyncOnValueBasedClasses != 0) {
7059     load_klass(t1, obj);
7060     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7061     tst(t1, KlassFlags::_misc_is_value_based_class);
7062     br(Assembler::NE, slow);
7063   }
7064 
7065   // Check if the lock-stack is full.
7066   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7067   cmpw(top, (unsigned)LockStack::end_offset());
7068   br(Assembler::GE, slow);
7069 
7070   // Check for recursion.
7071   subw(t, top, oopSize);
7072   ldr(t, Address(rthread, t));
7073   cmp(obj, t);
7074   br(Assembler::EQ, push);
7075 
7076   // Check header for monitor (0b10).
7077   tst(mark, markWord::monitor_value);
7078   br(Assembler::NE, slow);
7079 
7080   // Try to lock. Transition lock bits 0b01 => 0b00
7081   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7082   orr(mark, mark, markWord::unlocked_value);
7083   eor(t, mark, markWord::unlocked_value);
7084   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7085           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7086   br(Assembler::NE, slow);
7087 
7088   bind(push);
7089   // After successful lock, push object on lock-stack.
7090   str(obj, Address(rthread, top));
7091   addw(top, top, oopSize);
7092   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7093 }
7094 
7095 // Implements lightweight-unlocking.
7096 //
7097 // - obj: the object to be unlocked
7098 // - t1, t2, t3: temporary registers
7099 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7100 void MacroAssembler::lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7101   // cmpxchg clobbers rscratch1.
7102   assert_different_registers(obj, t1, t2, t3, rscratch1);
7103 
7104 #ifdef ASSERT
7105   {
7106     // Check for lock-stack underflow.
7107     Label stack_ok;
7108     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7109     cmpw(t1, (unsigned)LockStack::start_offset());
7110     br(Assembler::GE, stack_ok);
7111     STOP("Lock-stack underflow");
7112     bind(stack_ok);
7113   }
7114 #endif
7115 
7116   Label unlocked, push_and_slow;
7117   const Register top = t1;
7118   const Register mark = t2;
7119   const Register t = t3;
7120 
7121   // Check if obj is top of lock-stack.
7122   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7123   subw(top, top, oopSize);
7124   ldr(t, Address(rthread, top));
7125   cmp(obj, t);
7126   br(Assembler::NE, slow);
7127 
7128   // Pop lock-stack.
7129   DEBUG_ONLY(str(zr, Address(rthread, top));)
7130   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7131 
7132   // Check if recursive.
7133   subw(t, top, oopSize);
7134   ldr(t, Address(rthread, t));
7135   cmp(obj, t);
7136   br(Assembler::EQ, unlocked);
7137 
7138   // Not recursive. Check header for monitor (0b10).
7139   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7140   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7141 
7142 #ifdef ASSERT
7143   // Check header not unlocked (0b01).
7144   Label not_unlocked;
7145   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7146   stop("lightweight_unlock already unlocked");
7147   bind(not_unlocked);
7148 #endif
7149 
7150   // Try to unlock. Transition lock bits 0b00 => 0b01
7151   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7152   orr(t, mark, markWord::unlocked_value);
7153   cmpxchg(obj, mark, t, Assembler::xword,
7154           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7155   br(Assembler::EQ, unlocked);
7156 
7157   bind(push_and_slow);
7158   // Restore lock-stack and handle the unlock in runtime.
7159   DEBUG_ONLY(str(obj, Address(rthread, top));)
7160   addw(top, top, oopSize);
7161   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7162   b(slow);
7163 
7164   bind(unlocked);
7165 }