1 /*
   2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "code/compiledIC.hpp"
  30 #include "compiler/compileTask.hpp"
  31 #include "compiler/disassembler.hpp"
  32 #include "compiler/oopMap.hpp"
  33 #include "gc/shared/barrierSet.hpp"
  34 #include "gc/shared/barrierSetAssembler.hpp"
  35 #include "gc/shared/cardTableBarrierSet.hpp"
  36 #include "gc/shared/cardTable.hpp"
  37 #include "gc/shared/collectedHeap.hpp"
  38 #include "gc/shared/tlab_globals.hpp"
  39 #include "interpreter/bytecodeHistogram.hpp"
  40 #include "interpreter/interpreter.hpp"
  41 #include "interpreter/interpreterRuntime.hpp"
  42 #include "jvm.h"
  43 #include "memory/resourceArea.hpp"
  44 #include "memory/universe.hpp"
  45 #include "nativeInst_aarch64.hpp"
  46 #include "oops/accessDecorators.hpp"
  47 #include "oops/compressedKlass.inline.hpp"
  48 #include "oops/compressedOops.inline.hpp"
  49 #include "oops/klass.inline.hpp"
  50 #include "runtime/continuation.hpp"
  51 #include "runtime/icache.hpp"
  52 #include "runtime/interfaceSupport.inline.hpp"
  53 #include "runtime/javaThread.hpp"
  54 #include "runtime/jniHandles.inline.hpp"
  55 #include "runtime/sharedRuntime.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "utilities/globalDefinitions.hpp"
  58 #include "utilities/integerCast.hpp"
  59 #include "utilities/powerOfTwo.hpp"
  60 #ifdef COMPILER1
  61 #include "c1/c1_LIRAssembler.hpp"
  62 #endif
  63 #ifdef COMPILER2
  64 #include "oops/oop.hpp"
  65 #include "opto/compile.hpp"
  66 #include "opto/node.hpp"
  67 #include "opto/output.hpp"
  68 #endif
  69 
  70 #include <sys/types.h>
  71 
  72 #ifdef PRODUCT
  73 #define BLOCK_COMMENT(str) /* nothing */
  74 #else
  75 #define BLOCK_COMMENT(str) block_comment(str)
  76 #endif
  77 #define STOP(str) stop(str);
  78 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  79 
  80 #ifdef ASSERT
  81 extern "C" void disnm(intptr_t p);
  82 #endif
  83 // Target-dependent relocation processing
  84 //
  85 // Instruction sequences whose target may need to be retrieved or
  86 // patched are distinguished by their leading instruction, sorting
  87 // them into three main instruction groups and related subgroups.
  88 //
  89 // 1) Branch, Exception and System (insn count = 1)
  90 //    1a) Unconditional branch (immediate):
  91 //      b/bl imm19
  92 //    1b) Compare & branch (immediate):
  93 //      cbz/cbnz Rt imm19
  94 //    1c) Test & branch (immediate):
  95 //      tbz/tbnz Rt imm14
  96 //    1d) Conditional branch (immediate):
  97 //      b.cond imm19
  98 //
  99 // 2) Loads and Stores (insn count = 1)
 100 //    2a) Load register literal:
 101 //      ldr Rt imm19
 102 //
 103 // 3) Data Processing Immediate (insn count = 2 or 3)
 104 //    3a) PC-rel. addressing
 105 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 106 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 107 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 108 //      adr/adrp Rx imm21
 109 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 110 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 111 //      The latter form can only happen when the target is an
 112 //      ExternalAddress, and (by definition) ExternalAddresses don't
 113 //      move. Because of that property, there is never any need to
 114 //      patch the last of the three instructions. However,
 115 //      MacroAssembler::target_addr_for_insn takes all three
 116 //      instructions into account and returns the correct address.
 117 //    3b) Move wide (immediate)
 118 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 119 //
 120 // A switch on a subset of the instruction's bits provides an
 121 // efficient dispatch to these subcases.
 122 //
 123 // insn[28:26] -> main group ('x' == don't care)
 124 //   00x -> UNALLOCATED
 125 //   100 -> Data Processing Immediate
 126 //   101 -> Branch, Exception and System
 127 //   x1x -> Loads and Stores
 128 //
 129 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 130 // n.b. in some cases extra bits need to be checked to verify the
 131 // instruction is as expected
 132 //
 133 // 1) ... xx101x Branch, Exception and System
 134 //   1a)  00___x Unconditional branch (immediate)
 135 //   1b)  01___0 Compare & branch (immediate)
 136 //   1c)  01___1 Test & branch (immediate)
 137 //   1d)  10___0 Conditional branch (immediate)
 138 //        other  Should not happen
 139 //
 140 // 2) ... xxx1x0 Loads and Stores
 141 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 142 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 143 //                strictly should be 64 bit non-FP/SIMD i.e.
 144 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 145 //
 146 // 3) ... xx100x Data Processing Immediate
 147 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 148 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 149 //                 strictly should be 64 bit movz #imm16<<0
 150 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 151 //
 152 
 153 static uint32_t insn_at(address insn_addr, int n) {
 154   return ((uint32_t*)insn_addr)[n];
 155 }
 156 
 157 template<typename T>
 158 class RelocActions : public AllStatic {
 159 
 160 public:
 161 
 162   static int ALWAYSINLINE run(address insn_addr, address &target) {
 163     int instructions = 1;
 164     uint32_t insn = insn_at(insn_addr, 0);
 165 
 166     uint32_t dispatch = Instruction_aarch64::extract(insn, 30, 25);
 167     switch(dispatch) {
 168       case 0b001010:
 169       case 0b001011: {
 170         instructions = T::unconditionalBranch(insn_addr, target);
 171         break;
 172       }
 173       case 0b101010:   // Conditional branch (immediate)
 174       case 0b011010: { // Compare & branch (immediate)
 175         instructions = T::conditionalBranch(insn_addr, target);
 176         break;
 177       }
 178       case 0b011011: {
 179         instructions = T::testAndBranch(insn_addr, target);
 180         break;
 181       }
 182       case 0b001100:
 183       case 0b001110:
 184       case 0b011100:
 185       case 0b011110:
 186       case 0b101100:
 187       case 0b101110:
 188       case 0b111100:
 189       case 0b111110: {
 190         // load/store
 191         if ((Instruction_aarch64::extract(insn, 29, 24) & 0b111011) == 0b011000) {
 192           // Load register (literal)
 193           instructions = T::loadStore(insn_addr, target);
 194           break;
 195         } else {
 196           // nothing to do
 197           assert(target == nullptr, "did not expect to relocate target for polling page load");
 198         }
 199         break;
 200       }
 201       case 0b001000:
 202       case 0b011000:
 203       case 0b101000:
 204       case 0b111000: {
 205         // adr/adrp
 206         assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 207         int shift = Instruction_aarch64::extract(insn, 31, 31);
 208         if (shift) {
 209           uint32_t insn2 = insn_at(insn_addr, 1);
 210           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 211               Instruction_aarch64::extract(insn, 4, 0) ==
 212               Instruction_aarch64::extract(insn2, 9, 5)) {
 213             instructions = T::adrp(insn_addr, target, T::adrpMem);
 214           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 215                      Instruction_aarch64::extract(insn, 4, 0) ==
 216                      Instruction_aarch64::extract(insn2, 4, 0)) {
 217             instructions = T::adrp(insn_addr, target, T::adrpAdd);
 218           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 219                      Instruction_aarch64::extract(insn, 4, 0) ==
 220                      Instruction_aarch64::extract(insn2, 4, 0)) {
 221             instructions = T::adrp(insn_addr, target, T::adrpMovk);
 222           } else {
 223             ShouldNotReachHere();
 224           }
 225         } else {
 226           instructions = T::adr(insn_addr, target);
 227         }
 228         break;
 229       }
 230       case 0b001001:
 231       case 0b011001:
 232       case 0b101001:
 233       case 0b111001: {
 234         instructions = T::immediate(insn_addr, target);
 235         break;
 236       }
 237       default: {
 238         ShouldNotReachHere();
 239       }
 240     }
 241 
 242     T::verify(insn_addr, target);
 243     return instructions * NativeInstruction::instruction_size;
 244   }
 245 };
 246 
 247 class Patcher : public AllStatic {
 248 public:
 249   static int unconditionalBranch(address insn_addr, address &target) {
 250     intptr_t offset = (target - insn_addr) >> 2;
 251     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 252     return 1;
 253   }
 254   static int conditionalBranch(address insn_addr, address &target) {
 255     intptr_t offset = (target - insn_addr) >> 2;
 256     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 257     return 1;
 258   }
 259   static int testAndBranch(address insn_addr, address &target) {
 260     intptr_t offset = (target - insn_addr) >> 2;
 261     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 262     return 1;
 263   }
 264   static int loadStore(address insn_addr, address &target) {
 265     intptr_t offset = (target - insn_addr) >> 2;
 266     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 267     return 1;
 268   }
 269   static int adr(address insn_addr, address &target) {
 270 #ifdef ASSERT
 271     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 272 #endif
 273     // PC-rel. addressing
 274     ptrdiff_t offset = target - insn_addr;
 275     int offset_lo = offset & 3;
 276     offset >>= 2;
 277     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 278     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 279     return 1;
 280   }
 281   template<typename U>
 282   static int adrp(address insn_addr, address &target, U inner) {
 283     int instructions = 1;
 284 #ifdef ASSERT
 285     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 286 #endif
 287     ptrdiff_t offset = target - insn_addr;
 288     instructions = 2;
 289     precond(inner != nullptr);
 290     // Give the inner reloc a chance to modify the target.
 291     address adjusted_target = target;
 292     instructions = inner(insn_addr, adjusted_target);
 293     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 294     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 295     offset = adr_page - pc_page;
 296     int offset_lo = offset & 3;
 297     offset >>= 2;
 298     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 299     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 300     return instructions;
 301   }
 302   static int adrpMem(address insn_addr, address &target) {
 303     uintptr_t dest = (uintptr_t)target;
 304     int offset_lo = dest & 0xfff;
 305     uint32_t insn2 = insn_at(insn_addr, 1);
 306     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 307     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 308     guarantee(((dest >> size) << size) == dest, "misaligned target");
 309     return 2;
 310   }
 311   static int adrpAdd(address insn_addr, address &target) {
 312     uintptr_t dest = (uintptr_t)target;
 313     int offset_lo = dest & 0xfff;
 314     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 315     return 2;
 316   }
 317   static int adrpMovk(address insn_addr, address &target) {
 318     uintptr_t dest = uintptr_t(target);
 319     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 320     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 321     target = address(dest);
 322     return 2;
 323   }
 324   static int immediate(address insn_addr, address &target) {
 325     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010100, "must be");
 326     uint64_t dest = (uint64_t)target;
 327     // Move wide constant
 328     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 329     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 330     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 331     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 332     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 333     return 3;
 334   }
 335   static void verify(address insn_addr, address &target) {
 336 #ifdef ASSERT
 337     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 338     if (!(address_is == target)) {
 339       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 340       disnm((intptr_t)insn_addr);
 341       assert(address_is == target, "should be");
 342     }
 343 #endif
 344   }
 345 };
 346 
 347 // If insn1 and insn2 use the same register to form an address, either
 348 // by an offsetted LDR or a simple ADD, return the offset. If the
 349 // second instruction is an LDR, the offset may be scaled.
 350 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 351   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 352       Instruction_aarch64::extract(insn1, 4, 0) ==
 353       Instruction_aarch64::extract(insn2, 9, 5)) {
 354     // Load/store register (unsigned immediate)
 355     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 356     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 357     byte_offset <<= size;
 358     return true;
 359   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 360              Instruction_aarch64::extract(insn1, 4, 0) ==
 361              Instruction_aarch64::extract(insn2, 4, 0)) {
 362     // add (immediate)
 363     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 364     return true;
 365   }
 366   return false;
 367 }
 368 
 369 class AArch64Decoder : public AllStatic {
 370 public:
 371 
 372   static int loadStore(address insn_addr, address &target) {
 373     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 374     target = insn_addr + (offset << 2);
 375     return 1;
 376   }
 377   static int unconditionalBranch(address insn_addr, address &target) {
 378     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 25, 0);
 379     target = insn_addr + (offset << 2);
 380     return 1;
 381   }
 382   static int conditionalBranch(address insn_addr, address &target) {
 383     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 384     target = address(((uint64_t)insn_addr + (offset << 2)));
 385     return 1;
 386   }
 387   static int testAndBranch(address insn_addr, address &target) {
 388     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 18, 5);
 389     target = address(((uint64_t)insn_addr + (offset << 2)));
 390     return 1;
 391   }
 392   static int adr(address insn_addr, address &target) {
 393     // PC-rel. addressing
 394     uint32_t insn = insn_at(insn_addr, 0);
 395     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 396     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 397     target = address((uint64_t)insn_addr + offset);
 398     return 1;
 399   }
 400   template<typename U>
 401   static int adrp(address insn_addr, address &target, U inner) {
 402     uint32_t insn = insn_at(insn_addr, 0);
 403     assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 404     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 405     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 406     int shift = 12;
 407     offset <<= shift;
 408     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 409     target_page &= ((uint64_t)-1) << shift;
 410     target = address(target_page);
 411     precond(inner != nullptr);
 412     inner(insn_addr, target);
 413     return 2;
 414   }
 415   static int adrpMem(address insn_addr, address &target) {
 416     uint32_t insn2 = insn_at(insn_addr, 1);
 417     // Load/store register (unsigned immediate)
 418     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 419     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 420     byte_offset <<= size;
 421     target += byte_offset;
 422     return 2;
 423   }
 424   static int adrpAdd(address insn_addr, address &target) {
 425     uint32_t insn2 = insn_at(insn_addr, 1);
 426     // add (immediate)
 427     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 428     target += byte_offset;
 429     return 2;
 430   }
 431   static int adrpMovk(address insn_addr, address &target) {
 432     uint32_t insn2 = insn_at(insn_addr, 1);
 433     uint64_t dest = uint64_t(target);
 434     dest = (dest & 0xffff0000ffffffff) |
 435       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 436     target = address(dest);
 437 
 438     // We know the destination 4k page. Maybe we have a third
 439     // instruction.
 440     uint32_t insn = insn_at(insn_addr, 0);
 441     uint32_t insn3 = insn_at(insn_addr, 2);
 442     ptrdiff_t byte_offset;
 443     if (offset_for(insn, insn3, byte_offset)) {
 444       target += byte_offset;
 445       return 3;
 446     } else {
 447       return 2;
 448     }
 449   }
 450   static int immediate(address insn_addr, address &target) {
 451     uint32_t *insns = (uint32_t *)insn_addr;
 452     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 453     // Move wide constant: movz, movk, movk.  See movptr().
 454     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 455     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 456     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 457                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 458                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 459     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 460     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 461     return 3;
 462   }
 463   static void verify(address insn_addr, address &target) {
 464   }
 465 };
 466 
 467 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 468   address target;
 469   RelocActions<AArch64Decoder>::run(insn_addr, target);
 470   return target;
 471 }
 472 
 473 // Patch any kind of instruction; there may be several instructions.
 474 // Return the total length (in bytes) of the instructions.
 475 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 476   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 477   return RelocActions<Patcher>::run(insn_addr, target);
 478 }
 479 
 480 int MacroAssembler::patch_oop(address insn_addr, address o) {
 481   int instructions;
 482   unsigned insn = *(unsigned*)insn_addr;
 483   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 484 
 485   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 486 
 487   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 488   // narrow OOPs by setting the upper 16 bits in the first
 489   // instruction.
 490   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 491     // Move narrow OOP
 492     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 493     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 494     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 495     instructions = 2;
 496   } else {
 497     // Move wide OOP
 498     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 499     uintptr_t dest = (uintptr_t)o;
 500     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 501     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 502     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 503     instructions = 3;
 504   }
 505   return instructions * NativeInstruction::instruction_size;
 506 }
 507 
 508 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 509   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 510   // We encode narrow ones by setting the upper 16 bits in the first
 511   // instruction.
 512   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 513   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 514          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 515 
 516   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 517 
 518   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 519   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 520   return 2 * NativeInstruction::instruction_size;
 521 }
 522 
 523 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 524   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 525   if (at_return) {
 526     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 527     // we may safely use the sp instead to perform the stack watermark check.
 528     cmp(in_nmethod ? sp : rfp, tmp);
 529     br(Assembler::HI, slow_path);
 530   } else {
 531     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 532   }
 533 }
 534 
 535 void MacroAssembler::rt_call(address dest, Register tmp) {
 536   CodeBlob *cb = CodeCache::find_blob(dest);
 537   if (cb) {
 538     far_call(RuntimeAddress(dest));
 539   } else {
 540     lea(tmp, RuntimeAddress(dest));
 541     blr(tmp);
 542   }
 543 }
 544 
 545 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 546   if (!Continuations::enabled()) return;
 547   Label done;
 548   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 549   cmp(sp, rscratch1);
 550   br(Assembler::LS, done);
 551   mov(rscratch1, sp); // we can't use sp as the source in str
 552   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 553   bind(done);
 554 }
 555 
 556 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 557   if (!Continuations::enabled()) return;
 558   Label done;
 559   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 560   cmp(sp, rscratch1);
 561   br(Assembler::LO, done);
 562   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 563   bind(done);
 564 }
 565 
 566 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 567   // we must set sp to zero to clear frame
 568   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 569 
 570   // must clear fp, so that compiled frames are not confused; it is
 571   // possible that we need it only for debugging
 572   if (clear_fp) {
 573     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 574   }
 575 
 576   // Always clear the pc because it could have been set by make_walkable()
 577   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 578 }
 579 
 580 // Calls to C land
 581 //
 582 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 583 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 584 // has to be reset to 0. This is required to allow proper stack traversal.
 585 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 586                                          Register last_java_fp,
 587                                          Register last_java_pc,
 588                                          Register scratch) {
 589 
 590   if (last_java_pc->is_valid()) {
 591       str(last_java_pc, Address(rthread,
 592                                 JavaThread::frame_anchor_offset()
 593                                 + JavaFrameAnchor::last_Java_pc_offset()));
 594     }
 595 
 596   // determine last_java_sp register
 597   if (last_java_sp == sp) {
 598     mov(scratch, sp);
 599     last_java_sp = scratch;
 600   } else if (!last_java_sp->is_valid()) {
 601     last_java_sp = esp;
 602   }
 603 
 604   // last_java_fp is optional
 605   if (last_java_fp->is_valid()) {
 606     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 607   }
 608 
 609   // We must set sp last.
 610   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 611 }
 612 
 613 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 614                                          Register last_java_fp,
 615                                          address  last_java_pc,
 616                                          Register scratch) {
 617   assert(last_java_pc != nullptr, "must provide a valid PC");
 618 
 619   adr(scratch, last_java_pc);
 620   str(scratch, Address(rthread,
 621                        JavaThread::frame_anchor_offset()
 622                        + JavaFrameAnchor::last_Java_pc_offset()));
 623 
 624   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 625 }
 626 
 627 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 628                                          Register last_java_fp,
 629                                          Label &L,
 630                                          Register scratch) {
 631   if (L.is_bound()) {
 632     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 633   } else {
 634     InstructionMark im(this);
 635     L.add_patch_at(code(), locator());
 636     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 637   }
 638 }
 639 
 640 static inline bool target_needs_far_branch(address addr) {
 641   if (AOTCodeCache::is_on_for_dump()) {
 642     return true;
 643   }
 644   // codecache size <= 128M
 645   if (!MacroAssembler::far_branches()) {
 646     return false;
 647   }
 648   // codecache size > 240M
 649   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 650     return true;
 651   }
 652   // codecache size: 128M..240M
 653   return !CodeCache::is_non_nmethod(addr);
 654 }
 655 
 656 void MacroAssembler::far_call(Address entry, Register tmp) {
 657   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 658   assert(CodeCache::find_blob(entry.target()) != nullptr,
 659          "destination of far call not found in code cache");
 660   assert(entry.rspec().type() == relocInfo::external_word_type
 661          || entry.rspec().type() == relocInfo::runtime_call_type
 662          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 663   if (target_needs_far_branch(entry.target())) {
 664     uint64_t offset;
 665     // We can use ADRP here because we know that the total size of
 666     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 667     adrp(tmp, entry, offset);
 668     add(tmp, tmp, offset);
 669     blr(tmp);
 670   } else {
 671     bl(entry);
 672   }
 673 }
 674 
 675 int MacroAssembler::far_jump(Address entry, Register tmp) {
 676   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 677   assert(CodeCache::find_blob(entry.target()) != nullptr,
 678          "destination of far call not found in code cache");
 679   assert(entry.rspec().type() == relocInfo::external_word_type
 680          || entry.rspec().type() == relocInfo::runtime_call_type
 681          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 682   address start = pc();
 683   if (target_needs_far_branch(entry.target())) {
 684     uint64_t offset;
 685     // We can use ADRP here because we know that the total size of
 686     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 687     adrp(tmp, entry, offset);
 688     add(tmp, tmp, offset);
 689     br(tmp);
 690   } else {
 691     b(entry);
 692   }
 693   return pc() - start;
 694 }
 695 
 696 void MacroAssembler::reserved_stack_check() {
 697     // testing if reserved zone needs to be enabled
 698     Label no_reserved_zone_enabling;
 699 
 700     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 701     cmp(sp, rscratch1);
 702     br(Assembler::LO, no_reserved_zone_enabling);
 703 
 704     enter();   // LR and FP are live.
 705     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 706     mov(c_rarg0, rthread);
 707     blr(rscratch1);
 708     leave();
 709 
 710     // We have already removed our own frame.
 711     // throw_delayed_StackOverflowError will think that it's been
 712     // called by our caller.
 713     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 714     br(rscratch1);
 715     should_not_reach_here();
 716 
 717     bind(no_reserved_zone_enabling);
 718 }
 719 
 720 static void pass_arg0(MacroAssembler* masm, Register arg) {
 721   if (c_rarg0 != arg ) {
 722     masm->mov(c_rarg0, arg);
 723   }
 724 }
 725 
 726 static void pass_arg1(MacroAssembler* masm, Register arg) {
 727   if (c_rarg1 != arg ) {
 728     masm->mov(c_rarg1, arg);
 729   }
 730 }
 731 
 732 static void pass_arg2(MacroAssembler* masm, Register arg) {
 733   if (c_rarg2 != arg ) {
 734     masm->mov(c_rarg2, arg);
 735   }
 736 }
 737 
 738 static void pass_arg3(MacroAssembler* masm, Register arg) {
 739   if (c_rarg3 != arg ) {
 740     masm->mov(c_rarg3, arg);
 741   }
 742 }
 743 
 744 void MacroAssembler::call_VM_base(Register oop_result,
 745                                   Register java_thread,
 746                                   Register last_java_sp,
 747                                   Label*   return_pc,
 748                                   address  entry_point,
 749                                   int      number_of_arguments,
 750                                   bool     check_exceptions) {
 751    // determine java_thread register
 752   if (!java_thread->is_valid()) {
 753     java_thread = rthread;
 754   }
 755 
 756   // determine last_java_sp register
 757   if (!last_java_sp->is_valid()) {
 758     last_java_sp = esp;
 759   }
 760 
 761   // debugging support
 762   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 763   assert(java_thread == rthread, "unexpected register");
 764 #ifdef ASSERT
 765   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 766   // if (!TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 767 #endif // ASSERT
 768 
 769   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 770   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 771 
 772   // push java thread (becomes first argument of C function)
 773 
 774   mov(c_rarg0, java_thread);
 775 
 776   // set last Java frame before call
 777   assert(last_java_sp != rfp, "can't use rfp");
 778 
 779   Label l;
 780   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 781 
 782   // do the call, remove parameters
 783   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 784 
 785   // lr could be poisoned with PAC signature during throw_pending_exception
 786   // if it was tail-call optimized by compiler, since lr is not callee-saved
 787   // reload it with proper value
 788   adr(lr, l);
 789 
 790   // reset last Java frame
 791   // Only interpreter should have to clear fp
 792   reset_last_Java_frame(true);
 793 
 794    // C++ interp handles this in the interpreter
 795   check_and_handle_popframe(java_thread);
 796   check_and_handle_earlyret(java_thread);
 797 
 798   if (check_exceptions) {
 799     // check for pending exceptions (java_thread is set upon return)
 800     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 801     Label ok;
 802     cbz(rscratch1, ok);
 803     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 804     br(rscratch1);
 805     bind(ok);
 806   }
 807 
 808   // get oop result if there is one and reset the value in the thread
 809   if (oop_result->is_valid()) {
 810     get_vm_result_oop(oop_result, java_thread);
 811   }
 812 }
 813 
 814 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 815   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 816 }
 817 
 818 // Check the entry target is always reachable from any branch.
 819 static bool is_always_within_branch_range(Address entry) {
 820   if (AOTCodeCache::is_on_for_dump()) {
 821     return false;
 822   }
 823   const address target = entry.target();
 824 
 825   if (!CodeCache::contains(target)) {
 826     // We always use trampolines for callees outside CodeCache.
 827     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 828     return false;
 829   }
 830 
 831   if (!MacroAssembler::far_branches()) {
 832     return true;
 833   }
 834 
 835   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 836     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 837     // Non-compiled methods stay forever in CodeCache.
 838     // We check whether the longest possible branch is within the branch range.
 839     assert(CodeCache::find_blob(target) != nullptr &&
 840           !CodeCache::find_blob(target)->is_nmethod(),
 841           "runtime call of compiled method");
 842     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 843     const address left_longest_branch_start = CodeCache::low_bound();
 844     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 845                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 846     return is_reachable;
 847   }
 848 
 849   return false;
 850 }
 851 
 852 // Maybe emit a call via a trampoline. If the code cache is small
 853 // trampolines won't be emitted.
 854 address MacroAssembler::trampoline_call(Address entry) {
 855   assert(entry.rspec().type() == relocInfo::runtime_call_type
 856          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 857          || entry.rspec().type() == relocInfo::static_call_type
 858          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 859 
 860   address target = entry.target();
 861 
 862   if (!is_always_within_branch_range(entry)) {
 863     if (!in_scratch_emit_size()) {
 864       // We don't want to emit a trampoline if C2 is generating dummy
 865       // code during its branch shortening phase.
 866       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 867         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 868         code()->share_trampoline_for(entry.target(), offset());
 869       } else {
 870         address stub = emit_trampoline_stub(offset(), target);
 871         if (stub == nullptr) {
 872           postcond(pc() == badAddress);
 873           return nullptr; // CodeCache is full
 874         }
 875       }
 876     }
 877     target = pc();
 878   }
 879 
 880   address call_pc = pc();
 881   relocate(entry.rspec());
 882   bl(target);
 883 
 884   postcond(pc() != badAddress);
 885   return call_pc;
 886 }
 887 
 888 // Emit a trampoline stub for a call to a target which is too far away.
 889 //
 890 // code sequences:
 891 //
 892 // call-site:
 893 //   branch-and-link to <destination> or <trampoline stub>
 894 //
 895 // Related trampoline stub for this call site in the stub section:
 896 //   load the call target from the constant pool
 897 //   branch (LR still points to the call site above)
 898 
 899 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 900                                              address dest) {
 901   // Max stub size: alignment nop, TrampolineStub.
 902   address stub = start_a_stub(max_trampoline_stub_size());
 903   if (stub == nullptr) {
 904     return nullptr;  // CodeBuffer::expand failed
 905   }
 906 
 907   // Create a trampoline stub relocation which relates this trampoline stub
 908   // with the call instruction at insts_call_instruction_offset in the
 909   // instructions code-section.
 910   align(wordSize);
 911   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 912                                             + insts_call_instruction_offset));
 913   const int stub_start_offset = offset();
 914 
 915   // Now, create the trampoline stub's code:
 916   // - load the call
 917   // - call
 918   Label target;
 919   ldr(rscratch1, target);
 920   br(rscratch1);
 921   bind(target);
 922   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 923          "should be");
 924   emit_int64((int64_t)dest);
 925 
 926   const address stub_start_addr = addr_at(stub_start_offset);
 927 
 928   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 929 
 930   end_a_stub();
 931   return stub_start_addr;
 932 }
 933 
 934 int MacroAssembler::max_trampoline_stub_size() {
 935   // Max stub size: alignment nop, TrampolineStub.
 936   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 937 }
 938 
 939 void MacroAssembler::emit_static_call_stub() {
 940   // CompiledDirectCall::set_to_interpreted knows the
 941   // exact layout of this stub.
 942 
 943   isb();
 944   mov_metadata(rmethod, nullptr);
 945 
 946   // Jump to the entry point of the c2i stub.
 947   if (codestub_branch_needs_far_jump()) {
 948     movptr(rscratch1, 0);
 949     br(rscratch1);
 950   } else {
 951     b(pc());
 952   }
 953 }
 954 
 955 int MacroAssembler::static_call_stub_size() {
 956   // During AOT production run AOT and JIT compiled code
 957   // are used at the same time. We need this size
 958   // to be the same for both types of code.
 959   if (!codestub_branch_needs_far_jump() && !AOTCodeCache::is_on_for_use()) {
 960     // isb; movk; movz; movz; b
 961     return 5 * NativeInstruction::instruction_size;
 962   }
 963   // isb; movk; movz; movz; movk; movz; movz; br
 964   return 8 * NativeInstruction::instruction_size;
 965 }
 966 
 967 void MacroAssembler::c2bool(Register x) {
 968   // implements x == 0 ? 0 : 1
 969   // note: must only look at least-significant byte of x
 970   //       since C-style booleans are stored in one byte
 971   //       only! (was bug)
 972   tst(x, 0xff);
 973   cset(x, Assembler::NE);
 974 }
 975 
 976 address MacroAssembler::ic_call(address entry, jint method_index) {
 977   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 978   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 979   return trampoline_call(Address(entry, rh));
 980 }
 981 
 982 int MacroAssembler::ic_check_size() {
 983   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 984   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 985     return NativeInstruction::instruction_size * (7 + extra_instructions);
 986   } else {
 987     return NativeInstruction::instruction_size * (5 + extra_instructions);
 988   }
 989 }
 990 
 991 int MacroAssembler::ic_check(int end_alignment) {
 992   Register receiver = j_rarg0;
 993   Register data = rscratch2;
 994   Register tmp1 = rscratch1;
 995   Register tmp2 = r10;
 996 
 997   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
 998   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
 999   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1000   // before the inline cache check here, and not after
1001   align(end_alignment, offset() + ic_check_size());
1002 
1003   int uep_offset = offset();
1004 
1005   if (UseCompactObjectHeaders) {
1006     load_narrow_klass_compact(tmp1, receiver);
1007     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1008     cmpw(tmp1, tmp2);
1009   } else {
1010     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1011     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1012     cmpw(tmp1, tmp2);
1013   }
1014 
1015   Label dont;
1016   br(Assembler::EQ, dont);
1017   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1018   bind(dont);
1019   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1020 
1021   return uep_offset;
1022 }
1023 
1024 // Implementation of call_VM versions
1025 
1026 void MacroAssembler::call_VM(Register oop_result,
1027                              address entry_point,
1028                              bool check_exceptions) {
1029   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1030 }
1031 
1032 void MacroAssembler::call_VM(Register oop_result,
1033                              address entry_point,
1034                              Register arg_1,
1035                              bool check_exceptions) {
1036   pass_arg1(this, arg_1);
1037   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1038 }
1039 
1040 void MacroAssembler::call_VM(Register oop_result,
1041                              address entry_point,
1042                              Register arg_1,
1043                              Register arg_2,
1044                              bool check_exceptions) {
1045   assert_different_registers(arg_1, c_rarg2);
1046   pass_arg2(this, arg_2);
1047   pass_arg1(this, arg_1);
1048   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1049 }
1050 
1051 void MacroAssembler::call_VM(Register oop_result,
1052                              address entry_point,
1053                              Register arg_1,
1054                              Register arg_2,
1055                              Register arg_3,
1056                              bool check_exceptions) {
1057   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1058   assert_different_registers(arg_2, c_rarg3);
1059   pass_arg3(this, arg_3);
1060 
1061   pass_arg2(this, arg_2);
1062 
1063   pass_arg1(this, arg_1);
1064   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1065 }
1066 
1067 void MacroAssembler::call_VM(Register oop_result,
1068                              Register last_java_sp,
1069                              address entry_point,
1070                              int number_of_arguments,
1071                              bool check_exceptions) {
1072   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1073 }
1074 
1075 void MacroAssembler::call_VM(Register oop_result,
1076                              Register last_java_sp,
1077                              address entry_point,
1078                              Register arg_1,
1079                              bool check_exceptions) {
1080   pass_arg1(this, arg_1);
1081   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1082 }
1083 
1084 void MacroAssembler::call_VM(Register oop_result,
1085                              Register last_java_sp,
1086                              address entry_point,
1087                              Register arg_1,
1088                              Register arg_2,
1089                              bool check_exceptions) {
1090 
1091   assert_different_registers(arg_1, c_rarg2);
1092   pass_arg2(this, arg_2);
1093   pass_arg1(this, arg_1);
1094   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1095 }
1096 
1097 void MacroAssembler::call_VM(Register oop_result,
1098                              Register last_java_sp,
1099                              address entry_point,
1100                              Register arg_1,
1101                              Register arg_2,
1102                              Register arg_3,
1103                              bool check_exceptions) {
1104   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1105   assert_different_registers(arg_2, c_rarg3);
1106   pass_arg3(this, arg_3);
1107   pass_arg2(this, arg_2);
1108   pass_arg1(this, arg_1);
1109   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1110 }
1111 
1112 
1113 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1114   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1115   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1116   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1117 }
1118 
1119 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1120   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1121   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1122 }
1123 
1124 void MacroAssembler::align(int modulus) {
1125   align(modulus, offset());
1126 }
1127 
1128 // Ensure that the code at target bytes offset from the current offset() is aligned
1129 // according to modulus.
1130 void MacroAssembler::align(int modulus, int target) {
1131   int delta = target - offset();
1132   while ((offset() + delta) % modulus != 0) nop();
1133 }
1134 
1135 void MacroAssembler::post_call_nop() {
1136   if (!Continuations::enabled()) {
1137     return;
1138   }
1139   InstructionMark im(this);
1140   relocate(post_call_nop_Relocation::spec());
1141   InlineSkippedInstructionsCounter skipCounter(this);
1142   nop();
1143   movk(zr, 0);
1144   movk(zr, 0);
1145 }
1146 
1147 // these are no-ops overridden by InterpreterMacroAssembler
1148 
1149 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1150 
1151 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1152 
1153 // Look up the method for a megamorphic invokeinterface call.
1154 // The target method is determined by <intf_klass, itable_index>.
1155 // The receiver klass is in recv_klass.
1156 // On success, the result will be in method_result, and execution falls through.
1157 // On failure, execution transfers to the given label.
1158 void MacroAssembler::lookup_interface_method(Register recv_klass,
1159                                              Register intf_klass,
1160                                              RegisterOrConstant itable_index,
1161                                              Register method_result,
1162                                              Register scan_temp,
1163                                              Label& L_no_such_interface,
1164                          bool return_method) {
1165   assert_different_registers(recv_klass, intf_klass, scan_temp);
1166   assert_different_registers(method_result, intf_klass, scan_temp);
1167   assert(recv_klass != method_result || !return_method,
1168      "recv_klass can be destroyed when method isn't needed");
1169   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1170          "caller must use same register for non-constant itable index as for method");
1171 
1172   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1173   int vtable_base = in_bytes(Klass::vtable_start_offset());
1174   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1175   int scan_step   = itableOffsetEntry::size() * wordSize;
1176   int vte_size    = vtableEntry::size_in_bytes();
1177   assert(vte_size == wordSize, "else adjust times_vte_scale");
1178 
1179   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1180 
1181   // Could store the aligned, prescaled offset in the klass.
1182   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1183   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1184   add(scan_temp, scan_temp, vtable_base);
1185 
1186   if (return_method) {
1187     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1188     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1189     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1190     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1191     if (itentry_off)
1192       add(recv_klass, recv_klass, itentry_off);
1193   }
1194 
1195   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1196   //   if (scan->interface() == intf) {
1197   //     result = (klass + scan->offset() + itable_index);
1198   //   }
1199   // }
1200   Label search, found_method;
1201 
1202   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1203   cmp(intf_klass, method_result);
1204   br(Assembler::EQ, found_method);
1205   bind(search);
1206   // Check that the previous entry is non-null.  A null entry means that
1207   // the receiver class doesn't implement the interface, and wasn't the
1208   // same as when the caller was compiled.
1209   cbz(method_result, L_no_such_interface);
1210   if (itableOffsetEntry::interface_offset() != 0) {
1211     add(scan_temp, scan_temp, scan_step);
1212     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1213   } else {
1214     ldr(method_result, Address(pre(scan_temp, scan_step)));
1215   }
1216   cmp(intf_klass, method_result);
1217   br(Assembler::NE, search);
1218 
1219   bind(found_method);
1220 
1221   // Got a hit.
1222   if (return_method) {
1223     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1224     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1225   }
1226 }
1227 
1228 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1229 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1230 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1231 // The target method is determined by <holder_klass, itable_index>.
1232 // The receiver klass is in recv_klass.
1233 // On success, the result will be in method_result, and execution falls through.
1234 // On failure, execution transfers to the given label.
1235 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1236                                                   Register holder_klass,
1237                                                   Register resolved_klass,
1238                                                   Register method_result,
1239                                                   Register temp_itbl_klass,
1240                                                   Register scan_temp,
1241                                                   int itable_index,
1242                                                   Label& L_no_such_interface) {
1243   // 'method_result' is only used as output register at the very end of this method.
1244   // Until then we can reuse it as 'holder_offset'.
1245   Register holder_offset = method_result;
1246   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1247 
1248   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1249   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1250   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1251   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1252 
1253   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1254 
1255   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1256   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1257   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1258   // temp_itbl_klass = itable[0]._interface;
1259   int vtblEntrySize = vtableEntry::size_in_bytes();
1260   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1261   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1262   mov(holder_offset, zr);
1263   // scan_temp = &(itable[0]._interface)
1264   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1265 
1266   // Initial checks:
1267   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1268   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1269   //   - if (itable[0] == 0), no such interface
1270   cmp(resolved_klass, holder_klass);
1271   br(Assembler::NE, L_loop_search_resolved_entry);
1272   cmp(holder_klass, temp_itbl_klass);
1273   br(Assembler::EQ, L_holder_found);
1274   cbz(temp_itbl_klass, L_no_such_interface);
1275 
1276   // Loop: Look for holder_klass record in itable
1277   //   do {
1278   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1279   //     if (temp_itbl_klass == holder_klass) {
1280   //       goto L_holder_found; // Found!
1281   //     }
1282   //   } while (temp_itbl_klass != 0);
1283   //   goto L_no_such_interface // Not found.
1284   Label L_search_holder;
1285   bind(L_search_holder);
1286     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1287     cmp(holder_klass, temp_itbl_klass);
1288     br(Assembler::EQ, L_holder_found);
1289     cbnz(temp_itbl_klass, L_search_holder);
1290 
1291   b(L_no_such_interface);
1292 
1293   // Loop: Look for resolved_class record in itable
1294   //   while (true) {
1295   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1296   //     if (temp_itbl_klass == 0) {
1297   //       goto L_no_such_interface;
1298   //     }
1299   //     if (temp_itbl_klass == resolved_klass) {
1300   //        goto L_resolved_found;  // Found!
1301   //     }
1302   //     if (temp_itbl_klass == holder_klass) {
1303   //        holder_offset = scan_temp;
1304   //     }
1305   //   }
1306   //
1307   Label L_loop_search_resolved;
1308   bind(L_loop_search_resolved);
1309     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1310   bind(L_loop_search_resolved_entry);
1311     cbz(temp_itbl_klass, L_no_such_interface);
1312     cmp(resolved_klass, temp_itbl_klass);
1313     br(Assembler::EQ, L_resolved_found);
1314     cmp(holder_klass, temp_itbl_klass);
1315     br(Assembler::NE, L_loop_search_resolved);
1316     mov(holder_offset, scan_temp);
1317     b(L_loop_search_resolved);
1318 
1319   // See if we already have a holder klass. If not, go and scan for it.
1320   bind(L_resolved_found);
1321   cbz(holder_offset, L_search_holder);
1322   mov(scan_temp, holder_offset);
1323 
1324   // Finally, scan_temp contains holder_klass vtable offset
1325   bind(L_holder_found);
1326   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1327   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1328     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1329   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1330 }
1331 
1332 // virtual method calling
1333 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1334                                            RegisterOrConstant vtable_index,
1335                                            Register method_result) {
1336   assert(vtableEntry::size() * wordSize == 8,
1337          "adjust the scaling in the code below");
1338   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1339 
1340   if (vtable_index.is_register()) {
1341     lea(method_result, Address(recv_klass,
1342                                vtable_index.as_register(),
1343                                Address::lsl(LogBytesPerWord)));
1344     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1345   } else {
1346     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1347     ldr(method_result,
1348         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1349   }
1350 }
1351 
1352 void MacroAssembler::check_klass_subtype(Register sub_klass,
1353                            Register super_klass,
1354                            Register temp_reg,
1355                            Label& L_success) {
1356   Label L_failure;
1357   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1358   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1359   bind(L_failure);
1360 }
1361 
1362 
1363 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1364                                                    Register super_klass,
1365                                                    Register temp_reg,
1366                                                    Label* L_success,
1367                                                    Label* L_failure,
1368                                                    Label* L_slow_path,
1369                                                    Register super_check_offset) {
1370   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1371   bool must_load_sco = ! super_check_offset->is_valid();
1372   if (must_load_sco) {
1373     assert(temp_reg != noreg, "supply either a temp or a register offset");
1374   }
1375 
1376   Label L_fallthrough;
1377   int label_nulls = 0;
1378   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1379   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1380   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1381   assert(label_nulls <= 1, "at most one null in the batch");
1382 
1383   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1384   Address super_check_offset_addr(super_klass, sco_offset);
1385 
1386   // Hacked jmp, which may only be used just before L_fallthrough.
1387 #define final_jmp(label)                                                \
1388   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1389   else                            b(label)                /*omit semi*/
1390 
1391   // If the pointers are equal, we are done (e.g., String[] elements).
1392   // This self-check enables sharing of secondary supertype arrays among
1393   // non-primary types such as array-of-interface.  Otherwise, each such
1394   // type would need its own customized SSA.
1395   // We move this check to the front of the fast path because many
1396   // type checks are in fact trivially successful in this manner,
1397   // so we get a nicely predicted branch right at the start of the check.
1398   cmp(sub_klass, super_klass);
1399   br(Assembler::EQ, *L_success);
1400 
1401   // Check the supertype display:
1402   if (must_load_sco) {
1403     ldrw(temp_reg, super_check_offset_addr);
1404     super_check_offset = temp_reg;
1405   }
1406 
1407   Address super_check_addr(sub_klass, super_check_offset);
1408   ldr(rscratch1, super_check_addr);
1409   cmp(super_klass, rscratch1); // load displayed supertype
1410   br(Assembler::EQ, *L_success);
1411 
1412   // This check has worked decisively for primary supers.
1413   // Secondary supers are sought in the super_cache ('super_cache_addr').
1414   // (Secondary supers are interfaces and very deeply nested subtypes.)
1415   // This works in the same check above because of a tricky aliasing
1416   // between the super_cache and the primary super display elements.
1417   // (The 'super_check_addr' can address either, as the case requires.)
1418   // Note that the cache is updated below if it does not help us find
1419   // what we need immediately.
1420   // So if it was a primary super, we can just fail immediately.
1421   // Otherwise, it's the slow path for us (no success at this point).
1422 
1423   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1424   if (L_failure == &L_fallthrough) {
1425     cbz(rscratch1, *L_slow_path);
1426   } else {
1427     cbnz(rscratch1, *L_failure);
1428     final_jmp(*L_slow_path);
1429   }
1430 
1431   bind(L_fallthrough);
1432 
1433 #undef final_jmp
1434 }
1435 
1436 // These two are taken from x86, but they look generally useful
1437 
1438 // scans count pointer sized words at [addr] for occurrence of value,
1439 // generic
1440 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1441                                 Register scratch) {
1442   Label Lloop, Lexit;
1443   cbz(count, Lexit);
1444   bind(Lloop);
1445   ldr(scratch, post(addr, wordSize));
1446   cmp(value, scratch);
1447   br(EQ, Lexit);
1448   sub(count, count, 1);
1449   cbnz(count, Lloop);
1450   bind(Lexit);
1451 }
1452 
1453 // scans count 4 byte words at [addr] for occurrence of value,
1454 // generic
1455 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1456                                 Register scratch) {
1457   Label Lloop, Lexit;
1458   cbz(count, Lexit);
1459   bind(Lloop);
1460   ldrw(scratch, post(addr, wordSize));
1461   cmpw(value, scratch);
1462   br(EQ, Lexit);
1463   sub(count, count, 1);
1464   cbnz(count, Lloop);
1465   bind(Lexit);
1466 }
1467 
1468 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1469                                                           Register super_klass,
1470                                                           Register temp_reg,
1471                                                           Register temp2_reg,
1472                                                           Label* L_success,
1473                                                           Label* L_failure,
1474                                                           bool set_cond_codes) {
1475   // NB! Callers may assume that, when temp2_reg is a valid register,
1476   // this code sets it to a nonzero value.
1477 
1478   assert_different_registers(sub_klass, super_klass, temp_reg);
1479   if (temp2_reg != noreg)
1480     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1481 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1482 
1483   Label L_fallthrough;
1484   int label_nulls = 0;
1485   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1486   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1487   assert(label_nulls <= 1, "at most one null in the batch");
1488 
1489   // a couple of useful fields in sub_klass:
1490   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1491   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1492   Address secondary_supers_addr(sub_klass, ss_offset);
1493   Address super_cache_addr(     sub_klass, sc_offset);
1494 
1495   BLOCK_COMMENT("check_klass_subtype_slow_path");
1496 
1497   // Do a linear scan of the secondary super-klass chain.
1498   // This code is rarely used, so simplicity is a virtue here.
1499   // The repne_scan instruction uses fixed registers, which we must spill.
1500   // Don't worry too much about pre-existing connections with the input regs.
1501 
1502   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1503   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1504 
1505   RegSet pushed_registers;
1506   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1507   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1508 
1509   if (super_klass != r0) {
1510     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1511   }
1512 
1513   push(pushed_registers, sp);
1514 
1515   // Get super_klass value into r0 (even if it was in r5 or r2).
1516   if (super_klass != r0) {
1517     mov(r0, super_klass);
1518   }
1519 
1520 #ifndef PRODUCT
1521   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1522 #endif //PRODUCT
1523 
1524   // We will consult the secondary-super array.
1525   ldr(r5, secondary_supers_addr);
1526   // Load the array length.
1527   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1528   // Skip to start of data.
1529   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1530 
1531   cmp(sp, zr); // Clear Z flag; SP is never zero
1532   // Scan R2 words at [R5] for an occurrence of R0.
1533   // Set NZ/Z based on last compare.
1534   repne_scan(r5, r0, r2, rscratch1);
1535 
1536   // Unspill the temp. registers:
1537   pop(pushed_registers, sp);
1538 
1539   br(Assembler::NE, *L_failure);
1540 
1541   // Success.  Cache the super we found and proceed in triumph.
1542 
1543   if (UseSecondarySupersCache) {
1544     str(super_klass, super_cache_addr);
1545   }
1546 
1547   if (L_success != &L_fallthrough) {
1548     b(*L_success);
1549   }
1550 
1551 #undef IS_A_TEMP
1552 
1553   bind(L_fallthrough);
1554 }
1555 
1556 // If Register r is invalid, remove a new register from
1557 // available_regs, and add new register to regs_to_push.
1558 Register MacroAssembler::allocate_if_noreg(Register r,
1559                                   RegSetIterator<Register> &available_regs,
1560                                   RegSet &regs_to_push) {
1561   if (!r->is_valid()) {
1562     r = *available_regs++;
1563     regs_to_push += r;
1564   }
1565   return r;
1566 }
1567 
1568 // check_klass_subtype_slow_path_table() looks for super_klass in the
1569 // hash table belonging to super_klass, branching to L_success or
1570 // L_failure as appropriate. This is essentially a shim which
1571 // allocates registers as necessary then calls
1572 // lookup_secondary_supers_table() to do the work. Any of the temp
1573 // regs may be noreg, in which case this logic will chooses some
1574 // registers push and pop them from the stack.
1575 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1576                                                          Register super_klass,
1577                                                          Register temp_reg,
1578                                                          Register temp2_reg,
1579                                                          Register temp3_reg,
1580                                                          Register result_reg,
1581                                                          FloatRegister vtemp,
1582                                                          Label* L_success,
1583                                                          Label* L_failure,
1584                                                          bool set_cond_codes) {
1585   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1586 
1587   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1588 
1589   Label L_fallthrough;
1590   int label_nulls = 0;
1591   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1592   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1593   assert(label_nulls <= 1, "at most one null in the batch");
1594 
1595   BLOCK_COMMENT("check_klass_subtype_slow_path");
1596 
1597   RegSetIterator<Register> available_regs
1598     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1599 
1600   RegSet pushed_regs;
1601 
1602   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1603   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1604   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1605   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1606 
1607   push(pushed_regs, sp);
1608 
1609   lookup_secondary_supers_table_var(sub_klass,
1610                                     super_klass,
1611                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1612                                     nullptr);
1613   cmp(result_reg, zr);
1614 
1615   // Unspill the temp. registers:
1616   pop(pushed_regs, sp);
1617 
1618   // NB! Callers may assume that, when set_cond_codes is true, this
1619   // code sets temp2_reg to a nonzero value.
1620   if (set_cond_codes) {
1621     mov(temp2_reg, 1);
1622   }
1623 
1624   br(Assembler::NE, *L_failure);
1625 
1626   if (L_success != &L_fallthrough) {
1627     b(*L_success);
1628   }
1629 
1630   bind(L_fallthrough);
1631 }
1632 
1633 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1634                                                    Register super_klass,
1635                                                    Register temp_reg,
1636                                                    Register temp2_reg,
1637                                                    Label* L_success,
1638                                                    Label* L_failure,
1639                                                    bool set_cond_codes) {
1640   if (UseSecondarySupersTable) {
1641     check_klass_subtype_slow_path_table
1642       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1643        /*vtemp*/fnoreg,
1644        L_success, L_failure, set_cond_codes);
1645   } else {
1646     check_klass_subtype_slow_path_linear
1647       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1648   }
1649 }
1650 
1651 
1652 // Ensure that the inline code and the stub are using the same registers.
1653 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1654 do {                                                               \
1655   assert(r_super_klass  == r0                                   && \
1656          r_array_base   == r1                                   && \
1657          r_array_length == r2                                   && \
1658          (r_array_index == r3        || r_array_index == noreg) && \
1659          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1660          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1661          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1662 } while(0)
1663 
1664 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1665                                                          Register r_super_klass,
1666                                                          Register temp1,
1667                                                          Register temp2,
1668                                                          Register temp3,
1669                                                          FloatRegister vtemp,
1670                                                          Register result,
1671                                                          u1 super_klass_slot,
1672                                                          bool stub_is_near) {
1673   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1674 
1675   Label L_fallthrough;
1676 
1677   BLOCK_COMMENT("lookup_secondary_supers_table {");
1678 
1679   const Register
1680     r_array_base   = temp1, // r1
1681     r_array_length = temp2, // r2
1682     r_array_index  = temp3, // r3
1683     r_bitmap       = rscratch2;
1684 
1685   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1686 
1687   u1 bit = super_klass_slot;
1688 
1689   // Make sure that result is nonzero if the TBZ below misses.
1690   mov(result, 1);
1691 
1692   // We're going to need the bitmap in a vector reg and in a core reg,
1693   // so load both now.
1694   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1695   if (bit != 0) {
1696     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1697   }
1698   // First check the bitmap to see if super_klass might be present. If
1699   // the bit is zero, we are certain that super_klass is not one of
1700   // the secondary supers.
1701   tbz(r_bitmap, bit, L_fallthrough);
1702 
1703   // Get the first array index that can contain super_klass into r_array_index.
1704   if (bit != 0) {
1705     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1706     cnt(vtemp, T8B, vtemp);
1707     addv(vtemp, T8B, vtemp);
1708     fmovd(r_array_index, vtemp);
1709   } else {
1710     mov(r_array_index, (u1)1);
1711   }
1712   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1713 
1714   // We will consult the secondary-super array.
1715   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1716 
1717   // The value i in r_array_index is >= 1, so even though r_array_base
1718   // points to the length, we don't need to adjust it to point to the
1719   // data.
1720   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1721   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1722 
1723   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1724   eor(result, result, r_super_klass);
1725   cbz(result, L_fallthrough); // Found a match
1726 
1727   // Is there another entry to check? Consult the bitmap.
1728   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1729 
1730   // Linear probe.
1731   if (bit != 0) {
1732     ror(r_bitmap, r_bitmap, bit);
1733   }
1734 
1735   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1736   // The next slot to be inspected, by the stub we're about to call,
1737   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1738   // have been checked.
1739   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1740   if (stub_is_near) {
1741     bl(stub);
1742   } else {
1743     address call = trampoline_call(stub);
1744     if (call == nullptr) {
1745       return false; // trampoline allocation failed
1746     }
1747   }
1748 
1749   BLOCK_COMMENT("} lookup_secondary_supers_table");
1750 
1751   bind(L_fallthrough);
1752 
1753   if (VerifySecondarySupers) {
1754     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1755                                   temp1, temp2, result);      // r1, r2, r5
1756   }
1757   return true;
1758 }
1759 
1760 // At runtime, return 0 in result if r_super_klass is a superclass of
1761 // r_sub_klass, otherwise return nonzero. Use this version of
1762 // lookup_secondary_supers_table() if you don't know ahead of time
1763 // which superclass will be searched for. Used by interpreter and
1764 // runtime stubs. It is larger and has somewhat greater latency than
1765 // the version above, which takes a constant super_klass_slot.
1766 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1767                                                        Register r_super_klass,
1768                                                        Register temp1,
1769                                                        Register temp2,
1770                                                        Register temp3,
1771                                                        FloatRegister vtemp,
1772                                                        Register result,
1773                                                        Label *L_success) {
1774   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1775 
1776   Label L_fallthrough;
1777 
1778   BLOCK_COMMENT("lookup_secondary_supers_table {");
1779 
1780   const Register
1781     r_array_index = temp3,
1782     slot          = rscratch1,
1783     r_bitmap      = rscratch2;
1784 
1785   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1786 
1787   // Make sure that result is nonzero if the test below misses.
1788   mov(result, 1);
1789 
1790   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1791 
1792   // First check the bitmap to see if super_klass might be present. If
1793   // the bit is zero, we are certain that super_klass is not one of
1794   // the secondary supers.
1795 
1796   // This next instruction is equivalent to:
1797   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1798   // sub(temp2, tmp_reg, slot);
1799   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1800   lslv(temp2, r_bitmap, temp2);
1801   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1802 
1803   bool must_save_v0 = (vtemp == fnoreg);
1804   if (must_save_v0) {
1805     // temp1 and result are free, so use them to preserve vtemp
1806     vtemp = v0;
1807     mov(temp1,  vtemp, D, 0);
1808     mov(result, vtemp, D, 1);
1809   }
1810 
1811   // Get the first array index that can contain super_klass into r_array_index.
1812   mov(vtemp, D, 0, temp2);
1813   cnt(vtemp, T8B, vtemp);
1814   addv(vtemp, T8B, vtemp);
1815   mov(r_array_index, vtemp, D, 0);
1816 
1817   if (must_save_v0) {
1818     mov(vtemp, D, 0, temp1 );
1819     mov(vtemp, D, 1, result);
1820   }
1821 
1822   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1823 
1824   const Register
1825     r_array_base   = temp1,
1826     r_array_length = temp2;
1827 
1828   // The value i in r_array_index is >= 1, so even though r_array_base
1829   // points to the length, we don't need to adjust it to point to the
1830   // data.
1831   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1832   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1833 
1834   // We will consult the secondary-super array.
1835   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1836 
1837   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1838   eor(result, result, r_super_klass);
1839   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1840 
1841   // Is there another entry to check? Consult the bitmap.
1842   rorv(r_bitmap, r_bitmap, slot);
1843   // rol(r_bitmap, r_bitmap, 1);
1844   tbz(r_bitmap, 1, L_fallthrough);
1845 
1846   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1847   // The next slot to be inspected, by the logic we're about to call,
1848   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1849   // have been checked.
1850   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1851                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1852 
1853   BLOCK_COMMENT("} lookup_secondary_supers_table");
1854 
1855   bind(L_fallthrough);
1856 
1857   if (VerifySecondarySupers) {
1858     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1859                                   temp1, temp2, result);      // r1, r2, r5
1860   }
1861 
1862   if (L_success) {
1863     cbz(result, *L_success);
1864   }
1865 }
1866 
1867 // Called by code generated by check_klass_subtype_slow_path
1868 // above. This is called when there is a collision in the hashed
1869 // lookup in the secondary supers array.
1870 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1871                                                              Register r_array_base,
1872                                                              Register r_array_index,
1873                                                              Register r_bitmap,
1874                                                              Register temp1,
1875                                                              Register result,
1876                                                              bool is_stub) {
1877   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1878 
1879   const Register
1880     r_array_length = temp1,
1881     r_sub_klass    = noreg; // unused
1882 
1883   if (is_stub) {
1884     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1885   }
1886 
1887   Label L_fallthrough, L_huge;
1888 
1889   // Load the array length.
1890   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1891   // And adjust the array base to point to the data.
1892   // NB! Effectively increments current slot index by 1.
1893   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1894   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1895 
1896   // The bitmap is full to bursting.
1897   // Implicit invariant: BITMAP_FULL implies (length > 0)
1898   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1899   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1900   br(GT, L_huge);
1901 
1902   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1903   // current slot (at secondary_supers[r_array_index]) has not yet
1904   // been inspected, and r_array_index may be out of bounds if we
1905   // wrapped around the end of the array.
1906 
1907   { // This is conventional linear probing, but instead of terminating
1908     // when a null entry is found in the table, we maintain a bitmap
1909     // in which a 0 indicates missing entries.
1910     // As long as the bitmap is not completely full,
1911     // array_length == popcount(bitmap). The array_length check above
1912     // guarantees there are 0s in the bitmap, so the loop eventually
1913     // terminates.
1914     Label L_loop;
1915     bind(L_loop);
1916 
1917     // Check for wraparound.
1918     cmp(r_array_index, r_array_length);
1919     csel(r_array_index, zr, r_array_index, GE);
1920 
1921     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1922     eor(result, rscratch1, r_super_klass);
1923     cbz(result, L_fallthrough);
1924 
1925     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1926 
1927     ror(r_bitmap, r_bitmap, 1);
1928     add(r_array_index, r_array_index, 1);
1929     b(L_loop);
1930   }
1931 
1932   { // Degenerate case: more than 64 secondary supers.
1933     // FIXME: We could do something smarter here, maybe a vectorized
1934     // comparison or a binary search, but is that worth any added
1935     // complexity?
1936     bind(L_huge);
1937     cmp(sp, zr); // Clear Z flag; SP is never zero
1938     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1939     cset(result, NE); // result == 0 iff we got a match.
1940   }
1941 
1942   bind(L_fallthrough);
1943 }
1944 
1945 // Make sure that the hashed lookup and a linear scan agree.
1946 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1947                                                    Register r_super_klass,
1948                                                    Register temp1,
1949                                                    Register temp2,
1950                                                    Register result) {
1951   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1952 
1953   const Register
1954     r_array_base   = temp1,
1955     r_array_length = temp2;
1956 
1957   BLOCK_COMMENT("verify_secondary_supers_table {");
1958 
1959   // We will consult the secondary-super array.
1960   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1961 
1962   // Load the array length.
1963   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1964   // And adjust the array base to point to the data.
1965   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1966 
1967   cmp(sp, zr); // Clear Z flag; SP is never zero
1968   // Scan R2 words at [R5] for an occurrence of R0.
1969   // Set NZ/Z based on last compare.
1970   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1971   // rscratch1 == 0 iff we got a match.
1972   cset(rscratch1, NE);
1973 
1974   Label passed;
1975   cmp(result, zr);
1976   cset(result, NE); // normalize result to 0/1 for comparison
1977 
1978   cmp(rscratch1, result);
1979   br(EQ, passed);
1980   {
1981     mov(r0, r_super_klass);         // r0 <- r0
1982     mov(r1, r_sub_klass);           // r1 <- r4
1983     mov(r2, /*expected*/rscratch1); // r2 <- r8
1984     mov(r3, result);                // r3 <- r5
1985     mov(r4, (address)("mismatch")); // r4 <- const
1986     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
1987     should_not_reach_here();
1988   }
1989   bind(passed);
1990 
1991   BLOCK_COMMENT("} verify_secondary_supers_table");
1992 }
1993 
1994 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
1995   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
1996   assert_different_registers(klass, rthread, scratch);
1997 
1998   Label L_fallthrough, L_tmp;
1999   if (L_fast_path == nullptr) {
2000     L_fast_path = &L_fallthrough;
2001   } else if (L_slow_path == nullptr) {
2002     L_slow_path = &L_fallthrough;
2003   }
2004   // Fast path check: class is fully initialized
2005   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2006   ldarb(scratch, scratch);
2007   cmp(scratch, InstanceKlass::fully_initialized);
2008   br(Assembler::EQ, *L_fast_path);
2009 
2010   // Fast path check: current thread is initializer thread
2011   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2012   cmp(rthread, scratch);
2013 
2014   if (L_slow_path == &L_fallthrough) {
2015     br(Assembler::EQ, *L_fast_path);
2016     bind(*L_slow_path);
2017   } else if (L_fast_path == &L_fallthrough) {
2018     br(Assembler::NE, *L_slow_path);
2019     bind(*L_fast_path);
2020   } else {
2021     Unimplemented();
2022   }
2023 }
2024 
2025 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2026   if (!VerifyOops) return;
2027 
2028   // Pass register number to verify_oop_subroutine
2029   const char* b = nullptr;
2030   {
2031     ResourceMark rm;
2032     stringStream ss;
2033     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2034     b = code_string(ss.as_string());
2035   }
2036   BLOCK_COMMENT("verify_oop {");
2037 
2038   strip_return_address(); // This might happen within a stack frame.
2039   protect_return_address();
2040   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2041   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2042 
2043   mov(r0, reg);
2044   movptr(rscratch1, (uintptr_t)(address)b);
2045 
2046   // call indirectly to solve generation ordering problem
2047   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2048   ldr(rscratch2, Address(rscratch2));
2049   blr(rscratch2);
2050 
2051   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2052   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2053   authenticate_return_address();
2054 
2055   BLOCK_COMMENT("} verify_oop");
2056 }
2057 
2058 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2059   if (!VerifyOops) return;
2060 
2061   const char* b = nullptr;
2062   {
2063     ResourceMark rm;
2064     stringStream ss;
2065     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2066     b = code_string(ss.as_string());
2067   }
2068   BLOCK_COMMENT("verify_oop_addr {");
2069 
2070   strip_return_address(); // This might happen within a stack frame.
2071   protect_return_address();
2072   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2073   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2074 
2075   // addr may contain sp so we will have to adjust it based on the
2076   // pushes that we just did.
2077   if (addr.uses(sp)) {
2078     lea(r0, addr);
2079     ldr(r0, Address(r0, 4 * wordSize));
2080   } else {
2081     ldr(r0, addr);
2082   }
2083   movptr(rscratch1, (uintptr_t)(address)b);
2084 
2085   // call indirectly to solve generation ordering problem
2086   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2087   ldr(rscratch2, Address(rscratch2));
2088   blr(rscratch2);
2089 
2090   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2091   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2092   authenticate_return_address();
2093 
2094   BLOCK_COMMENT("} verify_oop_addr");
2095 }
2096 
2097 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2098                                          int extra_slot_offset) {
2099   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2100   int stackElementSize = Interpreter::stackElementSize;
2101   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2102 #ifdef ASSERT
2103   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2104   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2105 #endif
2106   if (arg_slot.is_constant()) {
2107     return Address(esp, arg_slot.as_constant() * stackElementSize
2108                    + offset);
2109   } else {
2110     add(rscratch1, esp, arg_slot.as_register(),
2111         ext::uxtx, exact_log2(stackElementSize));
2112     return Address(rscratch1, offset);
2113   }
2114 }
2115 
2116 // Handle the receiver type profile update given the "recv" klass.
2117 //
2118 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2119 // If there are no matching or claimable receiver entries in RD, updates
2120 // the polymorphic counter.
2121 //
2122 // This code expected to run by either the interpreter or JIT-ed code, without
2123 // extra synchronization. For safety, receiver cells are claimed atomically, which
2124 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2125 // counter updates are not atomic.
2126 //
2127 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2128   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2129 
2130   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2131   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2132   int poly_count_offset      = in_bytes(CounterData::count_offset());
2133   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2134   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2135 
2136   // Adjust for MDP offsets.
2137   base_receiver_offset += mdp_offset;
2138   end_receiver_offset  += mdp_offset;
2139   poly_count_offset    += mdp_offset;
2140 
2141 #ifdef ASSERT
2142   // We are about to walk the MDO slots without asking for offsets.
2143   // Check that our math hits all the right spots.
2144   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2145     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2146     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2147     int offset = base_receiver_offset + receiver_step*c;
2148     int count_offset = offset + receiver_to_count_step;
2149     assert(offset == real_recv_offset, "receiver slot math");
2150     assert(count_offset == real_count_offset, "receiver count math");
2151   }
2152   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2153   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2154 #endif
2155 
2156   // Corner case: no profile table. Increment poly counter and exit.
2157   if (ReceiverTypeData::row_limit() == 0) {
2158     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2159     return;
2160   }
2161 
2162   Register offset = rscratch2;
2163 
2164   Label L_loop_search_receiver, L_loop_search_empty;
2165   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
2166 
2167   // The code here recognizes three major cases:
2168   //   A. Fastest: receiver found in the table
2169   //   B. Fast: no receiver in the table, and the table is full
2170   //   C. Slow: no receiver in the table, free slots in the table
2171   //
2172   // The case A performance is most important, as perfectly-behaved code would end up
2173   // there, especially with larger TypeProfileWidth. The case B performance is
2174   // important as well, this is where bulk of code would land for normally megamorphic
2175   // cases. The case C performance is not essential, its job is to deal with installation
2176   // races, we optimize for code density instead. Case C needs to make sure that receiver
2177   // rows are only claimed once. This makes sure we never overwrite a row for another
2178   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2179   //
2180   // It is very tempting to handle these cases in a single loop, and claim the first slot
2181   // without checking the rest of the table. But, profiling code should tolerate free slots
2182   // in the table, as class unloading can clear them. After such cleanup, the receiver
2183   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2184   // to complete, before trying to install new slots. Splitting the code in several tight
2185   // loops also helpfully optimizes for cases A and B.
2186   //
2187   // This code is effectively:
2188   //
2189   // restart:
2190   //   // Fastest: receiver is already installed
2191   //   for (i = 0; i < receiver_count(); i++) {
2192   //     if (receiver(i) == recv) goto found_recv(i);
2193   //   }
2194   //
2195   //   // Fast: no receiver, but profile is full
2196   //   for (i = 0; i < receiver_count(); i++) {
2197   //     if (receiver(i) == null) goto found_null(i);
2198   //   }
2199   //   goto polymorphic
2200   //
2201   //   // Slow: try to install receiver
2202   // found_null(i):
2203   //   CAS(&receiver(i), null, recv);
2204   //   goto restart
2205   //
2206   // polymorphic:
2207   //   count++;
2208   //   return
2209   //
2210   // found_recv(i):
2211   //   *receiver_count(i)++
2212   //
2213 
2214   bind(L_restart);
2215 
2216   // Fastest: receiver is already installed
2217   mov(offset, base_receiver_offset);
2218   bind(L_loop_search_receiver);
2219     ldr(rscratch1, Address(mdp, offset));
2220     cmp(rscratch1, recv);
2221     br(Assembler::EQ, L_found_recv);
2222   add(offset, offset, receiver_step);
2223   sub(rscratch1, offset, end_receiver_offset);
2224   cbnz(rscratch1, L_loop_search_receiver);
2225 
2226   // Fast: no receiver, but profile is full
2227   mov(offset, base_receiver_offset);
2228   bind(L_loop_search_empty);
2229     ldr(rscratch1, Address(mdp, offset));
2230     cbz(rscratch1, L_found_empty);
2231   add(offset, offset, receiver_step);
2232   sub(rscratch1, offset, end_receiver_offset);
2233   cbnz(rscratch1, L_loop_search_empty);
2234   b(L_polymorphic);
2235 
2236   // Slow: try to install receiver
2237   bind(L_found_empty);
2238 
2239   // Atomically swing receiver slot: null -> recv.
2240   //
2241   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2242   // is used to hold the destination address. This is safe because the
2243   // offset is no longer needed after the address is computed.
2244 
2245   lea(rscratch2, Address(mdp, offset));
2246   cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
2247           /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
2248 
2249   // CAS success means the slot now has the receiver we want. CAS failure means
2250   // something had claimed the slot concurrently: it can be the same receiver we want,
2251   // or something else. Since this is a slow path, we can optimize for code density,
2252   // and just restart the search from the beginning.
2253   b(L_restart);
2254 
2255   // Counter updates:
2256 
2257   // Increment polymorphic counter instead of receiver slot.
2258   bind(L_polymorphic);
2259   mov(offset, poly_count_offset);
2260   b(L_count_update);
2261 
2262   // Found a receiver, convert its slot offset to corresponding count offset.
2263   bind(L_found_recv);
2264   add(offset, offset, receiver_to_count_step);
2265 
2266   bind(L_count_update);
2267   increment(Address(mdp, offset), DataLayout::counter_increment);
2268 }
2269 
2270 
2271 void MacroAssembler::call_VM_leaf_base(address entry_point,
2272                                        int number_of_arguments,
2273                                        Label *retaddr) {
2274   Label E, L;
2275 
2276   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2277 
2278   mov(rscratch1, RuntimeAddress(entry_point));
2279   blr(rscratch1);
2280   if (retaddr)
2281     bind(*retaddr);
2282 
2283   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2284 }
2285 
2286 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2287   call_VM_leaf_base(entry_point, number_of_arguments);
2288 }
2289 
2290 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2291   pass_arg0(this, arg_0);
2292   call_VM_leaf_base(entry_point, 1);
2293 }
2294 
2295 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2296   assert_different_registers(arg_1, c_rarg0);
2297   pass_arg0(this, arg_0);
2298   pass_arg1(this, arg_1);
2299   call_VM_leaf_base(entry_point, 2);
2300 }
2301 
2302 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2303                                   Register arg_1, Register arg_2) {
2304   assert_different_registers(arg_1, c_rarg0);
2305   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2306   pass_arg0(this, arg_0);
2307   pass_arg1(this, arg_1);
2308   pass_arg2(this, arg_2);
2309   call_VM_leaf_base(entry_point, 3);
2310 }
2311 
2312 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2313   pass_arg0(this, arg_0);
2314   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2315 }
2316 
2317 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2318 
2319   assert_different_registers(arg_0, c_rarg1);
2320   pass_arg1(this, arg_1);
2321   pass_arg0(this, arg_0);
2322   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2323 }
2324 
2325 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2326   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2327   assert_different_registers(arg_1, c_rarg2);
2328   pass_arg2(this, arg_2);
2329   pass_arg1(this, arg_1);
2330   pass_arg0(this, arg_0);
2331   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2332 }
2333 
2334 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2335   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2336   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2337   assert_different_registers(arg_2, c_rarg3);
2338   pass_arg3(this, arg_3);
2339   pass_arg2(this, arg_2);
2340   pass_arg1(this, arg_1);
2341   pass_arg0(this, arg_0);
2342   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2343 }
2344 
2345 void MacroAssembler::null_check(Register reg, int offset) {
2346   if (needs_explicit_null_check(offset)) {
2347     // provoke OS null exception if reg is null by
2348     // accessing M[reg] w/o changing any registers
2349     // NOTE: this is plenty to provoke a segv
2350     ldr(zr, Address(reg));
2351   } else {
2352     // nothing to do, (later) access of M[reg + offset]
2353     // will provoke OS null exception if reg is null
2354   }
2355 }
2356 
2357 // MacroAssembler protected routines needed to implement
2358 // public methods
2359 
2360 void MacroAssembler::mov(Register r, Address dest) {
2361   code_section()->relocate(pc(), dest.rspec());
2362   uint64_t imm64 = (uint64_t)dest.target();
2363   movptr(r, imm64);
2364 }
2365 
2366 // Move a constant pointer into r.  In AArch64 mode the virtual
2367 // address space is 48 bits in size, so we only need three
2368 // instructions to create a patchable instruction sequence that can
2369 // reach anywhere.
2370 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2371 #ifndef PRODUCT
2372   {
2373     char buffer[64];
2374     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2375     block_comment(buffer);
2376   }
2377 #endif
2378   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2379   movz(r, imm64 & 0xffff);
2380   imm64 >>= 16;
2381   movk(r, imm64 & 0xffff, 16);
2382   imm64 >>= 16;
2383   movk(r, imm64 & 0xffff, 32);
2384 }
2385 
2386 // Macro to mov replicated immediate to vector register.
2387 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2388 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2389 // Vd will get the following values for different arrangements in T
2390 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2391 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2392 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2393 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2394 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2395 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2396 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2397 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2398 // Clobbers rscratch1
2399 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2400   assert(T != T1Q, "unsupported");
2401   if (T == T1D || T == T2D) {
2402     int imm = operand_valid_for_movi_immediate(imm64, T);
2403     if (-1 != imm) {
2404       movi(Vd, T, imm);
2405     } else {
2406       mov(rscratch1, imm64);
2407       dup(Vd, T, rscratch1);
2408     }
2409     return;
2410   }
2411 
2412 #ifdef ASSERT
2413   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2414   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2415   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2416 #endif
2417   int shift = operand_valid_for_movi_immediate(imm64, T);
2418   uint32_t imm32 = imm64 & 0xffffffffULL;
2419   if (shift >= 0) {
2420     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2421   } else {
2422     movw(rscratch1, imm32);
2423     dup(Vd, T, rscratch1);
2424   }
2425 }
2426 
2427 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2428 {
2429 #ifndef PRODUCT
2430   {
2431     char buffer[64];
2432     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2433     block_comment(buffer);
2434   }
2435 #endif
2436   if (operand_valid_for_logical_immediate(false, imm64)) {
2437     orr(dst, zr, imm64);
2438   } else {
2439     // we can use a combination of MOVZ or MOVN with
2440     // MOVK to build up the constant
2441     uint64_t imm_h[4];
2442     int zero_count = 0;
2443     int neg_count = 0;
2444     int i;
2445     for (i = 0; i < 4; i++) {
2446       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2447       if (imm_h[i] == 0) {
2448         zero_count++;
2449       } else if (imm_h[i] == 0xffffL) {
2450         neg_count++;
2451       }
2452     }
2453     if (zero_count == 4) {
2454       // one MOVZ will do
2455       movz(dst, 0);
2456     } else if (neg_count == 4) {
2457       // one MOVN will do
2458       movn(dst, 0);
2459     } else if (zero_count == 3) {
2460       for (i = 0; i < 4; i++) {
2461         if (imm_h[i] != 0L) {
2462           movz(dst, (uint32_t)imm_h[i], (i << 4));
2463           break;
2464         }
2465       }
2466     } else if (neg_count == 3) {
2467       // one MOVN will do
2468       for (int i = 0; i < 4; i++) {
2469         if (imm_h[i] != 0xffffL) {
2470           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2471           break;
2472         }
2473       }
2474     } else if (zero_count == 2) {
2475       // one MOVZ and one MOVK will do
2476       for (i = 0; i < 3; i++) {
2477         if (imm_h[i] != 0L) {
2478           movz(dst, (uint32_t)imm_h[i], (i << 4));
2479           i++;
2480           break;
2481         }
2482       }
2483       for (;i < 4; i++) {
2484         if (imm_h[i] != 0L) {
2485           movk(dst, (uint32_t)imm_h[i], (i << 4));
2486         }
2487       }
2488     } else if (neg_count == 2) {
2489       // one MOVN and one MOVK will do
2490       for (i = 0; i < 4; i++) {
2491         if (imm_h[i] != 0xffffL) {
2492           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2493           i++;
2494           break;
2495         }
2496       }
2497       for (;i < 4; i++) {
2498         if (imm_h[i] != 0xffffL) {
2499           movk(dst, (uint32_t)imm_h[i], (i << 4));
2500         }
2501       }
2502     } else if (zero_count == 1) {
2503       // one MOVZ and two MOVKs will do
2504       for (i = 0; i < 4; i++) {
2505         if (imm_h[i] != 0L) {
2506           movz(dst, (uint32_t)imm_h[i], (i << 4));
2507           i++;
2508           break;
2509         }
2510       }
2511       for (;i < 4; i++) {
2512         if (imm_h[i] != 0x0L) {
2513           movk(dst, (uint32_t)imm_h[i], (i << 4));
2514         }
2515       }
2516     } else if (neg_count == 1) {
2517       // one MOVN and two MOVKs will do
2518       for (i = 0; i < 4; i++) {
2519         if (imm_h[i] != 0xffffL) {
2520           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2521           i++;
2522           break;
2523         }
2524       }
2525       for (;i < 4; i++) {
2526         if (imm_h[i] != 0xffffL) {
2527           movk(dst, (uint32_t)imm_h[i], (i << 4));
2528         }
2529       }
2530     } else {
2531       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2532       movz(dst, (uint32_t)imm_h[0], 0);
2533       for (i = 1; i < 4; i++) {
2534         movk(dst, (uint32_t)imm_h[i], (i << 4));
2535       }
2536     }
2537   }
2538 }
2539 
2540 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2541 {
2542 #ifndef PRODUCT
2543     {
2544       char buffer[64];
2545       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2546       block_comment(buffer);
2547     }
2548 #endif
2549   if (operand_valid_for_logical_immediate(true, imm32)) {
2550     orrw(dst, zr, imm32);
2551   } else {
2552     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2553     // constant
2554     uint32_t imm_h[2];
2555     imm_h[0] = imm32 & 0xffff;
2556     imm_h[1] = ((imm32 >> 16) & 0xffff);
2557     if (imm_h[0] == 0) {
2558       movzw(dst, imm_h[1], 16);
2559     } else if (imm_h[0] == 0xffff) {
2560       movnw(dst, imm_h[1] ^ 0xffff, 16);
2561     } else if (imm_h[1] == 0) {
2562       movzw(dst, imm_h[0], 0);
2563     } else if (imm_h[1] == 0xffff) {
2564       movnw(dst, imm_h[0] ^ 0xffff, 0);
2565     } else {
2566       // use a MOVZ and MOVK (makes it easier to debug)
2567       movzw(dst, imm_h[0], 0);
2568       movkw(dst, imm_h[1], 16);
2569     }
2570   }
2571 }
2572 
2573 // Form an address from base + offset in Rd.  Rd may or may
2574 // not actually be used: you must use the Address that is returned.
2575 // It is up to you to ensure that the shift provided matches the size
2576 // of your data.
2577 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2578   if (Address::offset_ok_for_immed(byte_offset, shift))
2579     // It fits; no need for any heroics
2580     return Address(base, byte_offset);
2581 
2582   // Don't do anything clever with negative or misaligned offsets
2583   unsigned mask = (1 << shift) - 1;
2584   if (byte_offset < 0 || byte_offset & mask) {
2585     mov(Rd, byte_offset);
2586     add(Rd, base, Rd);
2587     return Address(Rd);
2588   }
2589 
2590   // See if we can do this with two 12-bit offsets
2591   {
2592     uint64_t word_offset = byte_offset >> shift;
2593     uint64_t masked_offset = word_offset & 0xfff000;
2594     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2595         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2596       add(Rd, base, masked_offset << shift);
2597       word_offset -= masked_offset;
2598       return Address(Rd, word_offset << shift);
2599     }
2600   }
2601 
2602   // Do it the hard way
2603   mov(Rd, byte_offset);
2604   add(Rd, base, Rd);
2605   return Address(Rd);
2606 }
2607 
2608 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2609                                     bool want_remainder, Register scratch)
2610 {
2611   // Full implementation of Java idiv and irem.  The function
2612   // returns the (pc) offset of the div instruction - may be needed
2613   // for implicit exceptions.
2614   //
2615   // constraint : ra/rb =/= scratch
2616   //         normal case
2617   //
2618   // input : ra: dividend
2619   //         rb: divisor
2620   //
2621   // result: either
2622   //         quotient  (= ra idiv rb)
2623   //         remainder (= ra irem rb)
2624 
2625   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2626 
2627   int idivl_offset = offset();
2628   if (! want_remainder) {
2629     sdivw(result, ra, rb);
2630   } else {
2631     sdivw(scratch, ra, rb);
2632     Assembler::msubw(result, scratch, rb, ra);
2633   }
2634 
2635   return idivl_offset;
2636 }
2637 
2638 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2639                                     bool want_remainder, Register scratch)
2640 {
2641   // Full implementation of Java ldiv and lrem.  The function
2642   // returns the (pc) offset of the div instruction - may be needed
2643   // for implicit exceptions.
2644   //
2645   // constraint : ra/rb =/= scratch
2646   //         normal case
2647   //
2648   // input : ra: dividend
2649   //         rb: divisor
2650   //
2651   // result: either
2652   //         quotient  (= ra idiv rb)
2653   //         remainder (= ra irem rb)
2654 
2655   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2656 
2657   int idivq_offset = offset();
2658   if (! want_remainder) {
2659     sdiv(result, ra, rb);
2660   } else {
2661     sdiv(scratch, ra, rb);
2662     Assembler::msub(result, scratch, rb, ra);
2663   }
2664 
2665   return idivq_offset;
2666 }
2667 
2668 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2669   address prev = pc() - NativeMembar::instruction_size;
2670   address last = code()->last_insn();
2671   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2672     NativeMembar *bar = NativeMembar_at(prev);
2673     if (AlwaysMergeDMB) {
2674       bar->set_kind(bar->get_kind() | order_constraint);
2675       BLOCK_COMMENT("merged membar(always)");
2676       return;
2677     }
2678     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2679     // doing so would introduce a StoreLoad which the caller did not
2680     // intend
2681     if (bar->get_kind() == order_constraint
2682         || bar->get_kind() == AnyAny
2683         || order_constraint == AnyAny) {
2684       // We are merging two memory barrier instructions.  On AArch64 we
2685       // can do this simply by ORing them together.
2686       bar->set_kind(bar->get_kind() | order_constraint);
2687       BLOCK_COMMENT("merged membar");
2688       return;
2689     } else {
2690       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2691       // We need check the last 2 instructions
2692       address prev2 = prev - NativeMembar::instruction_size;
2693       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2694         NativeMembar *bar2 = NativeMembar_at(prev2);
2695         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2696         BLOCK_COMMENT("merged membar(elided)");
2697         return;
2698       }
2699     }
2700   }
2701   code()->set_last_insn(pc());
2702   dmb(Assembler::barrier(order_constraint));
2703 }
2704 
2705 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2706   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2707     merge_ldst(rt, adr, size_in_bytes, is_store);
2708     code()->clear_last_insn();
2709     return true;
2710   } else {
2711     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2712     const uint64_t mask = size_in_bytes - 1;
2713     if (adr.getMode() == Address::base_plus_offset &&
2714         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2715       code()->set_last_insn(pc());
2716     }
2717     return false;
2718   }
2719 }
2720 
2721 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2722   // We always try to merge two adjacent loads into one ldp.
2723   if (!try_merge_ldst(Rx, adr, 8, false)) {
2724     Assembler::ldr(Rx, adr);
2725   }
2726 }
2727 
2728 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2729   // We always try to merge two adjacent loads into one ldp.
2730   if (!try_merge_ldst(Rw, adr, 4, false)) {
2731     Assembler::ldrw(Rw, adr);
2732   }
2733 }
2734 
2735 void MacroAssembler::str(Register Rx, const Address &adr) {
2736   // We always try to merge two adjacent stores into one stp.
2737   if (!try_merge_ldst(Rx, adr, 8, true)) {
2738     Assembler::str(Rx, adr);
2739   }
2740 }
2741 
2742 void MacroAssembler::strw(Register Rw, const Address &adr) {
2743   // We always try to merge two adjacent stores into one stp.
2744   if (!try_merge_ldst(Rw, adr, 4, true)) {
2745     Assembler::strw(Rw, adr);
2746   }
2747 }
2748 
2749 // MacroAssembler routines found actually to be needed
2750 
2751 void MacroAssembler::push(Register src)
2752 {
2753   str(src, Address(pre(esp, -1 * wordSize)));
2754 }
2755 
2756 void MacroAssembler::pop(Register dst)
2757 {
2758   ldr(dst, Address(post(esp, 1 * wordSize)));
2759 }
2760 
2761 // Note: load_unsigned_short used to be called load_unsigned_word.
2762 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2763   int off = offset();
2764   ldrh(dst, src);
2765   return off;
2766 }
2767 
2768 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2769   int off = offset();
2770   ldrb(dst, src);
2771   return off;
2772 }
2773 
2774 int MacroAssembler::load_signed_short(Register dst, Address src) {
2775   int off = offset();
2776   ldrsh(dst, src);
2777   return off;
2778 }
2779 
2780 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2781   int off = offset();
2782   ldrsb(dst, src);
2783   return off;
2784 }
2785 
2786 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2787   int off = offset();
2788   ldrshw(dst, src);
2789   return off;
2790 }
2791 
2792 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2793   int off = offset();
2794   ldrsbw(dst, src);
2795   return off;
2796 }
2797 
2798 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2799   switch (size_in_bytes) {
2800   case  8:  ldr(dst, src); break;
2801   case  4:  ldrw(dst, src); break;
2802   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2803   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2804   default:  ShouldNotReachHere();
2805   }
2806 }
2807 
2808 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2809   switch (size_in_bytes) {
2810   case  8:  str(src, dst); break;
2811   case  4:  strw(src, dst); break;
2812   case  2:  strh(src, dst); break;
2813   case  1:  strb(src, dst); break;
2814   default:  ShouldNotReachHere();
2815   }
2816 }
2817 
2818 void MacroAssembler::narrow_subword_type(Register reg, BasicType bt) {
2819   assert(is_subword_type(bt), "required");
2820   switch (bt) {
2821   case T_BOOLEAN: andw(reg, reg, 1); break;
2822   case T_BYTE:    sxtbw(reg, reg); break;
2823   case T_CHAR:    uxthw(reg, reg); break;
2824   case T_SHORT:   sxthw(reg, reg); break;
2825   default:        ShouldNotReachHere();
2826   }
2827 }
2828 
2829 void MacroAssembler::decrementw(Register reg, int value)
2830 {
2831   if (value < 0)  { incrementw(reg, -value);      return; }
2832   if (value == 0) {                               return; }
2833   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2834   /* else */ {
2835     guarantee(reg != rscratch2, "invalid dst for register decrement");
2836     movw(rscratch2, (unsigned)value);
2837     subw(reg, reg, rscratch2);
2838   }
2839 }
2840 
2841 void MacroAssembler::decrement(Register reg, int value)
2842 {
2843   if (value < 0)  { increment(reg, -value);      return; }
2844   if (value == 0) {                              return; }
2845   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2846   /* else */ {
2847     assert(reg != rscratch2, "invalid dst for register decrement");
2848     mov(rscratch2, (uint64_t)value);
2849     sub(reg, reg, rscratch2);
2850   }
2851 }
2852 
2853 void MacroAssembler::decrementw(Address dst, int value)
2854 {
2855   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2856   if (dst.getMode() == Address::literal) {
2857     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2858     lea(rscratch2, dst);
2859     dst = Address(rscratch2);
2860   }
2861   ldrw(rscratch1, dst);
2862   decrementw(rscratch1, value);
2863   strw(rscratch1, dst);
2864 }
2865 
2866 void MacroAssembler::decrement(Address dst, int value)
2867 {
2868   assert(!dst.uses(rscratch1), "invalid address for decrement");
2869   if (dst.getMode() == Address::literal) {
2870     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2871     lea(rscratch2, dst);
2872     dst = Address(rscratch2);
2873   }
2874   ldr(rscratch1, dst);
2875   decrement(rscratch1, value);
2876   str(rscratch1, dst);
2877 }
2878 
2879 void MacroAssembler::incrementw(Register reg, int value)
2880 {
2881   if (value < 0)  { decrementw(reg, -value);      return; }
2882   if (value == 0) {                               return; }
2883   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2884   /* else */ {
2885     assert(reg != rscratch2, "invalid dst for register increment");
2886     movw(rscratch2, (unsigned)value);
2887     addw(reg, reg, rscratch2);
2888   }
2889 }
2890 
2891 void MacroAssembler::increment(Register reg, int value)
2892 {
2893   if (value < 0)  { decrement(reg, -value);      return; }
2894   if (value == 0) {                              return; }
2895   if (value < (1 << 12)) { add(reg, reg, value); return; }
2896   /* else */ {
2897     assert(reg != rscratch2, "invalid dst for register increment");
2898     movw(rscratch2, (unsigned)value);
2899     add(reg, reg, rscratch2);
2900   }
2901 }
2902 
2903 void MacroAssembler::incrementw(Address dst, int value)
2904 {
2905   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2906   if (dst.getMode() == Address::literal) {
2907     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2908     lea(rscratch2, dst);
2909     dst = Address(rscratch2);
2910   }
2911   ldrw(rscratch1, dst);
2912   incrementw(rscratch1, value);
2913   strw(rscratch1, dst);
2914 }
2915 
2916 void MacroAssembler::increment(Address dst, int value)
2917 {
2918   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2919   if (dst.getMode() == Address::literal) {
2920     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2921     lea(rscratch2, dst);
2922     dst = Address(rscratch2);
2923   }
2924   ldr(rscratch1, dst);
2925   increment(rscratch1, value);
2926   str(rscratch1, dst);
2927 }
2928 
2929 // Push lots of registers in the bit set supplied.  Don't push sp.
2930 // Return the number of words pushed
2931 int MacroAssembler::push(RegSet regset, Register stack) {
2932   if (regset.bits() == 0) {
2933     return 0;
2934   }
2935   auto bitset = integer_cast<unsigned int>(regset.bits());
2936   int words_pushed = 0;
2937 
2938   // Scan bitset to accumulate register pairs
2939   unsigned char regs[32];
2940   int count = 0;
2941   for (int reg = 0; reg <= 30; reg++) {
2942     if (1 & bitset)
2943       regs[count++] = reg;
2944     bitset >>= 1;
2945   }
2946   regs[count++] = zr->raw_encoding();
2947   count &= ~1;  // Only push an even number of regs
2948 
2949   if (count) {
2950     stp(as_Register(regs[0]), as_Register(regs[1]),
2951        Address(pre(stack, -count * wordSize)));
2952     words_pushed += 2;
2953   }
2954   for (int i = 2; i < count; i += 2) {
2955     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2956        Address(stack, i * wordSize));
2957     words_pushed += 2;
2958   }
2959 
2960   assert(words_pushed == count, "oops, pushed != count");
2961 
2962   return count;
2963 }
2964 
2965 int MacroAssembler::pop(RegSet regset, Register stack) {
2966   if (regset.bits() == 0) {
2967     return 0;
2968   }
2969   auto bitset = integer_cast<unsigned int>(regset.bits());
2970   int words_pushed = 0;
2971 
2972   // Scan bitset to accumulate register pairs
2973   unsigned char regs[32];
2974   int count = 0;
2975   for (int reg = 0; reg <= 30; reg++) {
2976     if (1 & bitset)
2977       regs[count++] = reg;
2978     bitset >>= 1;
2979   }
2980   regs[count++] = zr->raw_encoding();
2981   count &= ~1;
2982 
2983   for (int i = 2; i < count; i += 2) {
2984     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2985        Address(stack, i * wordSize));
2986     words_pushed += 2;
2987   }
2988   if (count) {
2989     ldp(as_Register(regs[0]), as_Register(regs[1]),
2990        Address(post(stack, count * wordSize)));
2991     words_pushed += 2;
2992   }
2993 
2994   assert(words_pushed == count, "oops, pushed != count");
2995 
2996   return count;
2997 }
2998 
2999 // Push lots of registers in the bit set supplied.  Don't push sp.
3000 // Return the number of dwords pushed
3001 int MacroAssembler::push_fp(FloatRegSet regset, Register stack, FpPushPopMode mode) {
3002   if (regset.bits() == 0) {
3003     return 0;
3004   }
3005   auto bitset = integer_cast<unsigned int>(regset.bits());
3006   int words_pushed = 0;
3007   bool use_sve = false;
3008   int sve_vector_size_in_bytes = 0;
3009 
3010 #ifdef COMPILER2
3011   use_sve = Matcher::supports_scalable_vector();
3012   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3013 #endif
3014 
3015   // Scan bitset to accumulate register pairs
3016   unsigned char regs[32];
3017   int count = 0;
3018   for (int reg = 0; reg <= 31; reg++) {
3019     if (1 & bitset)
3020       regs[count++] = reg;
3021     bitset >>= 1;
3022   }
3023 
3024   if (count == 0) {
3025     return 0;
3026   }
3027 
3028   if (mode == PushPopFull) {
3029     if (use_sve && sve_vector_size_in_bytes > 16) {
3030       mode = PushPopSVE;
3031     } else {
3032       mode = PushPopNeon;
3033     }
3034   }
3035 
3036 #ifndef PRODUCT
3037   {
3038     char buffer[48];
3039     if (mode == PushPopSVE) {
3040       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3041     } else if (mode == PushPopNeon) {
3042       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3043     } else {
3044       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3045     }
3046     block_comment(buffer);
3047   }
3048 #endif
3049 
3050   if (mode == PushPopSVE) {
3051     sub(stack, stack, sve_vector_size_in_bytes * count);
3052     for (int i = 0; i < count; i++) {
3053       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3054     }
3055     return count * sve_vector_size_in_bytes / 8;
3056   }
3057 
3058   if (mode == PushPopNeon) {
3059     if (count == 1) {
3060       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3061       return 2;
3062     }
3063 
3064     bool odd = (count & 1) == 1;
3065     int push_slots = count + (odd ? 1 : 0);
3066 
3067     // Always pushing full 128 bit registers.
3068     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3069     words_pushed += 2;
3070 
3071     for (int i = 2; i + 1 < count; i += 2) {
3072       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3073       words_pushed += 2;
3074     }
3075 
3076     if (odd) {
3077       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3078       words_pushed++;
3079     }
3080 
3081     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3082     return count * 2;
3083   }
3084 
3085   if (mode == PushPopFp) {
3086     bool odd = (count & 1) == 1;
3087     int push_slots = count + (odd ? 1 : 0);
3088 
3089     if (count == 1) {
3090       // Stack pointer must be 16 bytes aligned
3091       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3092       return 1;
3093     }
3094 
3095     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3096     words_pushed += 2;
3097 
3098     for (int i = 2; i + 1 < count; i += 2) {
3099       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3100       words_pushed += 2;
3101     }
3102 
3103     if (odd) {
3104       // Stack pointer must be 16 bytes aligned
3105       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3106       words_pushed++;
3107     }
3108 
3109     assert(words_pushed == count, "oops, pushed != count");
3110 
3111     return count;
3112   }
3113 
3114   return 0;
3115 }
3116 
3117 // Return the number of dwords popped
3118 int MacroAssembler::pop_fp(FloatRegSet regset, Register stack, FpPushPopMode mode) {
3119   if (regset.bits() == 0) {
3120     return 0;
3121   }
3122   auto bitset = integer_cast<unsigned int>(regset.bits());
3123   int words_pushed = 0;
3124   bool use_sve = false;
3125   int sve_vector_size_in_bytes = 0;
3126 
3127 #ifdef COMPILER2
3128   use_sve = Matcher::supports_scalable_vector();
3129   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3130 #endif
3131   // Scan bitset to accumulate register pairs
3132   unsigned char regs[32];
3133   int count = 0;
3134   for (int reg = 0; reg <= 31; reg++) {
3135     if (1 & bitset)
3136       regs[count++] = reg;
3137     bitset >>= 1;
3138   }
3139 
3140   if (count == 0) {
3141     return 0;
3142   }
3143 
3144   if (mode == PushPopFull) {
3145     if (use_sve && sve_vector_size_in_bytes > 16) {
3146       mode = PushPopSVE;
3147     } else {
3148       mode = PushPopNeon;
3149     }
3150   }
3151 
3152 #ifndef PRODUCT
3153   {
3154     char buffer[48];
3155     if (mode == PushPopSVE) {
3156       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3157     } else if (mode == PushPopNeon) {
3158       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3159     } else {
3160       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3161     }
3162     block_comment(buffer);
3163   }
3164 #endif
3165 
3166   if (mode == PushPopSVE) {
3167     for (int i = count - 1; i >= 0; i--) {
3168       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3169     }
3170     add(stack, stack, sve_vector_size_in_bytes * count);
3171     return count * sve_vector_size_in_bytes / 8;
3172   }
3173 
3174   if (mode == PushPopNeon) {
3175     if (count == 1) {
3176       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3177       return 2;
3178     }
3179 
3180     bool odd = (count & 1) == 1;
3181     int push_slots = count + (odd ? 1 : 0);
3182 
3183     if (odd) {
3184       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3185       words_pushed++;
3186     }
3187 
3188     for (int i = 2; i + 1 < count; i += 2) {
3189       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3190       words_pushed += 2;
3191     }
3192 
3193     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3194     words_pushed += 2;
3195 
3196     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3197 
3198     return count * 2;
3199   }
3200 
3201   if (mode == PushPopFp) {
3202     bool odd = (count & 1) == 1;
3203     int push_slots = count + (odd ? 1 : 0);
3204 
3205     if (count == 1) {
3206       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3207       return 1;
3208     }
3209 
3210     if (odd) {
3211       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3212       words_pushed++;
3213     }
3214 
3215     for (int i = 2; i + 1 < count; i += 2) {
3216       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3217       words_pushed += 2;
3218     }
3219 
3220     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3221     words_pushed += 2;
3222 
3223     assert(words_pushed == count, "oops, pushed != count");
3224 
3225     return count;
3226   }
3227 
3228   return 0;
3229 }
3230 
3231 // Return the number of dwords pushed
3232 int MacroAssembler::push_p(PRegSet regset, Register stack) {
3233   if (regset.bits() == 0) {
3234     return 0;
3235   }
3236   auto bitset = integer_cast<unsigned int>(regset.bits());
3237   bool use_sve = false;
3238   int sve_predicate_size_in_slots = 0;
3239 
3240 #ifdef COMPILER2
3241   use_sve = Matcher::supports_scalable_vector();
3242   if (use_sve) {
3243     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3244   }
3245 #endif
3246 
3247   if (!use_sve) {
3248     return 0;
3249   }
3250 
3251   unsigned char regs[PRegister::number_of_registers];
3252   int count = 0;
3253   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3254     if (1 & bitset)
3255       regs[count++] = reg;
3256     bitset >>= 1;
3257   }
3258 
3259   if (count == 0) {
3260     return 0;
3261   }
3262 
3263   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3264                                   VMRegImpl::stack_slot_size * count, 16);
3265   sub(stack, stack, total_push_bytes);
3266   for (int i = 0; i < count; i++) {
3267     sve_str(as_PRegister(regs[i]), Address(stack, i));
3268   }
3269   return total_push_bytes / 8;
3270 }
3271 
3272 // Return the number of dwords popped
3273 int MacroAssembler::pop_p(PRegSet regset, Register stack) {
3274   if (regset.bits() == 0) {
3275     return 0;
3276   }
3277   auto bitset = integer_cast<unsigned int>(regset.bits());
3278   bool use_sve = false;
3279   int sve_predicate_size_in_slots = 0;
3280 
3281 #ifdef COMPILER2
3282   use_sve = Matcher::supports_scalable_vector();
3283   if (use_sve) {
3284     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3285   }
3286 #endif
3287 
3288   if (!use_sve) {
3289     return 0;
3290   }
3291 
3292   unsigned char regs[PRegister::number_of_registers];
3293   int count = 0;
3294   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3295     if (1 & bitset)
3296       regs[count++] = reg;
3297     bitset >>= 1;
3298   }
3299 
3300   if (count == 0) {
3301     return 0;
3302   }
3303 
3304   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3305                                  VMRegImpl::stack_slot_size * count, 16);
3306   for (int i = count - 1; i >= 0; i--) {
3307     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3308   }
3309   add(stack, stack, total_pop_bytes);
3310   return total_pop_bytes / 8;
3311 }
3312 
3313 #ifdef ASSERT
3314 void MacroAssembler::verify_heapbase(const char* msg) {
3315 #if 0
3316   assert (Universe::heap() != nullptr, "java heap should be initialized");
3317   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3318     // rheapbase is allocated as general register
3319     return;
3320   }
3321   if (CheckCompressedOops) {
3322     Label ok;
3323     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3324     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3325     br(Assembler::EQ, ok);
3326     stop(msg);
3327     bind(ok);
3328     pop(1 << rscratch1->encoding(), sp);
3329   }
3330 #endif
3331 }
3332 #endif
3333 
3334 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3335   assert_different_registers(value, tmp1, tmp2);
3336   Label done, tagged, weak_tagged;
3337 
3338   cbz(value, done);           // Use null as-is.
3339   tst(value, JNIHandles::tag_mask); // Test for tag.
3340   br(Assembler::NE, tagged);
3341 
3342   // Resolve local handle
3343   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3344   verify_oop(value);
3345   b(done);
3346 
3347   bind(tagged);
3348   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3349   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3350 
3351   // Resolve global handle
3352   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3353   verify_oop(value);
3354   b(done);
3355 
3356   bind(weak_tagged);
3357   // Resolve jweak.
3358   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3359                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3360   verify_oop(value);
3361 
3362   bind(done);
3363 }
3364 
3365 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3366   assert_different_registers(value, tmp1, tmp2);
3367   Label done;
3368 
3369   cbz(value, done);           // Use null as-is.
3370 
3371 #ifdef ASSERT
3372   {
3373     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3374     Label valid_global_tag;
3375     tbnz(value, 1, valid_global_tag); // Test for global tag
3376     stop("non global jobject using resolve_global_jobject");
3377     bind(valid_global_tag);
3378   }
3379 #endif
3380 
3381   // Resolve global handle
3382   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3383   verify_oop(value);
3384 
3385   bind(done);
3386 }
3387 
3388 void MacroAssembler::stop(const char* msg) {
3389   // Skip AOT caching C strings in scratch buffer.
3390   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3391   BLOCK_COMMENT(str);
3392   // load msg into r0 so we can access it from the signal handler
3393   // ExternalAddress enables saving and restoring via the code cache
3394   lea(c_rarg0, ExternalAddress((address) str));
3395   dcps1(0xdeae);
3396 }
3397 
3398 void MacroAssembler::unimplemented(const char* what) {
3399   const char* buf = nullptr;
3400   {
3401     ResourceMark rm;
3402     stringStream ss;
3403     ss.print("unimplemented: %s", what);
3404     buf = code_string(ss.as_string());
3405   }
3406   stop(buf);
3407 }
3408 
3409 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3410 #ifdef ASSERT
3411   Label OK;
3412   br(cc, OK);
3413   stop(msg);
3414   bind(OK);
3415 #endif
3416 }
3417 
3418 // If a constant does not fit in an immediate field, generate some
3419 // number of MOV instructions and then perform the operation.
3420 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3421                                            add_sub_imm_insn insn1,
3422                                            add_sub_reg_insn insn2,
3423                                            bool is32) {
3424   assert(Rd != zr, "Rd = zr and not setting flags?");
3425   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3426   if (fits) {
3427     (this->*insn1)(Rd, Rn, imm);
3428   } else {
3429     if (g_uabs(imm) < (1 << 24)) {
3430        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3431        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3432     } else {
3433        assert_different_registers(Rd, Rn);
3434        mov(Rd, imm);
3435        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3436     }
3437   }
3438 }
3439 
3440 // Separate vsn which sets the flags. Optimisations are more restricted
3441 // because we must set the flags correctly.
3442 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3443                                              add_sub_imm_insn insn1,
3444                                              add_sub_reg_insn insn2,
3445                                              bool is32) {
3446   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3447   if (fits) {
3448     (this->*insn1)(Rd, Rn, imm);
3449   } else {
3450     assert_different_registers(Rd, Rn);
3451     assert(Rd != zr, "overflow in immediate operand");
3452     mov(Rd, imm);
3453     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3454   }
3455 }
3456 
3457 
3458 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3459   if (increment.is_register()) {
3460     add(Rd, Rn, increment.as_register());
3461   } else {
3462     add(Rd, Rn, increment.as_constant());
3463   }
3464 }
3465 
3466 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3467   if (increment.is_register()) {
3468     addw(Rd, Rn, increment.as_register());
3469   } else {
3470     addw(Rd, Rn, increment.as_constant());
3471   }
3472 }
3473 
3474 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3475   if (decrement.is_register()) {
3476     sub(Rd, Rn, decrement.as_register());
3477   } else {
3478     sub(Rd, Rn, decrement.as_constant());
3479   }
3480 }
3481 
3482 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3483   if (decrement.is_register()) {
3484     subw(Rd, Rn, decrement.as_register());
3485   } else {
3486     subw(Rd, Rn, decrement.as_constant());
3487   }
3488 }
3489 
3490 void MacroAssembler::reinit_heapbase()
3491 {
3492   if (UseCompressedOops) {
3493     if (Universe::is_fully_initialized() && !AOTCodeCache::is_on_for_dump()) {
3494       mov(rheapbase, CompressedOops::base());
3495     } else {
3496       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3497       ldr(rheapbase, Address(rheapbase));
3498     }
3499   }
3500 }
3501 
3502 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3503 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3504 // Pass a register for the result, otherwise pass noreg.
3505 
3506 // Clobbers rscratch1
3507 void MacroAssembler::cmpxchg(Register addr, Register expected,
3508                              Register new_val,
3509                              enum operand_size size,
3510                              bool acquire, bool release,
3511                              bool weak,
3512                              Register result) {
3513   if (result == noreg)  result = rscratch1;
3514   BLOCK_COMMENT("cmpxchg {");
3515   if (UseLSE) {
3516     mov(result, expected);
3517     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3518     compare_eq(result, expected, size);
3519 #ifdef ASSERT
3520     // Poison rscratch1 which is written on !UseLSE branch
3521     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3522 #endif
3523   } else {
3524     Label retry_load, done;
3525     prfm(Address(addr), PSTL1STRM);
3526     bind(retry_load);
3527     load_exclusive(result, addr, size, acquire);
3528     compare_eq(result, expected, size);
3529     br(Assembler::NE, done);
3530     store_exclusive(rscratch1, new_val, addr, size, release);
3531     if (weak) {
3532       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3533     } else {
3534       cbnzw(rscratch1, retry_load);
3535     }
3536     bind(done);
3537   }
3538   BLOCK_COMMENT("} cmpxchg");
3539 }
3540 
3541 // A generic comparison. Only compares for equality, clobbers rscratch1.
3542 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3543   if (size == xword) {
3544     cmp(rm, rn);
3545   } else if (size == word) {
3546     cmpw(rm, rn);
3547   } else if (size == halfword) {
3548     eorw(rscratch1, rm, rn);
3549     ands(zr, rscratch1, 0xffff);
3550   } else if (size == byte) {
3551     eorw(rscratch1, rm, rn);
3552     ands(zr, rscratch1, 0xff);
3553   } else {
3554     ShouldNotReachHere();
3555   }
3556 }
3557 
3558 
3559 static bool different(Register a, RegisterOrConstant b, Register c) {
3560   if (b.is_constant())
3561     return a != c;
3562   else
3563     return a != b.as_register() && a != c && b.as_register() != c;
3564 }
3565 
3566 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3567 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3568   if (UseLSE) {                                                         \
3569     prev = prev->is_valid() ? prev : zr;                                \
3570     if (incr.is_register()) {                                           \
3571       AOP(sz, incr.as_register(), prev, addr);                          \
3572     } else {                                                            \
3573       mov(rscratch2, incr.as_constant());                               \
3574       AOP(sz, rscratch2, prev, addr);                                   \
3575     }                                                                   \
3576     return;                                                             \
3577   }                                                                     \
3578   Register result = rscratch2;                                          \
3579   if (prev->is_valid())                                                 \
3580     result = different(prev, incr, addr) ? prev : rscratch2;            \
3581                                                                         \
3582   Label retry_load;                                                     \
3583   prfm(Address(addr), PSTL1STRM);                                       \
3584   bind(retry_load);                                                     \
3585   LDXR(result, addr);                                                   \
3586   OP(rscratch1, result, incr);                                          \
3587   STXR(rscratch2, rscratch1, addr);                                     \
3588   cbnzw(rscratch2, retry_load);                                         \
3589   if (prev->is_valid() && prev != result) {                             \
3590     IOP(prev, rscratch1, incr);                                         \
3591   }                                                                     \
3592 }
3593 
3594 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3595 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3596 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3597 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3598 
3599 #undef ATOMIC_OP
3600 
3601 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3602 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3603   if (UseLSE) {                                                         \
3604     prev = prev->is_valid() ? prev : zr;                                \
3605     AOP(sz, newv, prev, addr);                                          \
3606     return;                                                             \
3607   }                                                                     \
3608   Register result = rscratch2;                                          \
3609   if (prev->is_valid())                                                 \
3610     result = different(prev, newv, addr) ? prev : rscratch2;            \
3611                                                                         \
3612   Label retry_load;                                                     \
3613   prfm(Address(addr), PSTL1STRM);                                       \
3614   bind(retry_load);                                                     \
3615   LDXR(result, addr);                                                   \
3616   STXR(rscratch1, newv, addr);                                          \
3617   cbnzw(rscratch1, retry_load);                                         \
3618   if (prev->is_valid() && prev != result)                               \
3619     mov(prev, result);                                                  \
3620 }
3621 
3622 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3623 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3624 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3625 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3626 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3627 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3628 
3629 #undef ATOMIC_XCHG
3630 
3631 #ifndef PRODUCT
3632 extern "C" void findpc(intptr_t x);
3633 #endif
3634 
3635 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3636 {
3637   // In order to get locks to work, we need to fake a in_VM state
3638   if (ShowMessageBoxOnError) {
3639     JavaThread* thread = JavaThread::current();
3640     thread->set_thread_state(_thread_in_vm);
3641 #ifndef PRODUCT
3642     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3643       ttyLocker ttyl;
3644       BytecodeCounter::print();
3645     }
3646 #endif
3647     if (os::message_box(msg, "Execution stopped, print registers?")) {
3648       ttyLocker ttyl;
3649       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3650 #ifndef PRODUCT
3651       tty->cr();
3652       findpc(pc);
3653       tty->cr();
3654 #endif
3655       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3656       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3657       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3658       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3659       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3660       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3661       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3662       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3663       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3664       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3665       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3666       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3667       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3668       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3669       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3670       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3671       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3672       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3673       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3674       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3675       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3676       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3677       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3678       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3679       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3680       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3681       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3682       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3683       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3684       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3685       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3686       BREAKPOINT;
3687     }
3688   }
3689   fatal("DEBUG MESSAGE: %s", msg);
3690 }
3691 
3692 RegSet MacroAssembler::call_clobbered_gp_registers() {
3693   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3694 #ifndef R18_RESERVED
3695   regs += r18_tls;
3696 #endif
3697   return regs;
3698 }
3699 
3700 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3701   int step = 4 * wordSize;
3702   push(call_clobbered_gp_registers() - exclude, sp);
3703   sub(sp, sp, step);
3704   mov(rscratch1, -step);
3705   // Push v0-v7, v16-v31.
3706   for (int i = 31; i>= 4; i -= 4) {
3707     if (i <= v7->encoding() || i >= v16->encoding())
3708       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3709           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3710   }
3711   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3712       as_FloatRegister(3), T1D, Address(sp));
3713 }
3714 
3715 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3716   for (int i = 0; i < 32; i += 4) {
3717     if (i <= v7->encoding() || i >= v16->encoding())
3718       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3719           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3720   }
3721 
3722   reinitialize_ptrue();
3723 
3724   pop(call_clobbered_gp_registers() - exclude, sp);
3725 }
3726 
3727 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3728                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3729   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3730   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3731     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3732     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3733       sve_str(as_FloatRegister(i), Address(sp, i));
3734     }
3735   } else {
3736     int step = (save_vectors ? 8 : 4) * wordSize;
3737     mov(rscratch1, -step);
3738     sub(sp, sp, step);
3739     for (int i = 28; i >= 4; i -= 4) {
3740       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3741           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3742     }
3743     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3744   }
3745   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3746     sub(sp, sp, total_predicate_in_bytes);
3747     for (int i = 0; i < PRegister::number_of_registers; i++) {
3748       sve_str(as_PRegister(i), Address(sp, i));
3749     }
3750   }
3751 }
3752 
3753 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3754                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3755   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3756     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3757       sve_ldr(as_PRegister(i), Address(sp, i));
3758     }
3759     add(sp, sp, total_predicate_in_bytes);
3760   }
3761   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3762     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3763       sve_ldr(as_FloatRegister(i), Address(sp, i));
3764     }
3765     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3766   } else {
3767     int step = (restore_vectors ? 8 : 4) * wordSize;
3768     for (int i = 0; i <= 28; i += 4)
3769       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3770           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3771   }
3772 
3773   // We may use predicate registers and rely on ptrue with SVE,
3774   // regardless of wide vector (> 8 bytes) used or not.
3775   if (use_sve) {
3776     reinitialize_ptrue();
3777   }
3778 
3779   // integer registers except lr & sp
3780   pop(RegSet::range(r0, r17), sp);
3781 #ifdef R18_RESERVED
3782   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3783   pop(RegSet::range(r20, r29), sp);
3784 #else
3785   pop(RegSet::range(r18_tls, r29), sp);
3786 #endif
3787 }
3788 
3789 /**
3790  * Helpers for multiply_to_len().
3791  */
3792 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3793                                      Register src1, Register src2) {
3794   adds(dest_lo, dest_lo, src1);
3795   adc(dest_hi, dest_hi, zr);
3796   adds(dest_lo, dest_lo, src2);
3797   adc(final_dest_hi, dest_hi, zr);
3798 }
3799 
3800 // Generate an address from (r + r1 extend offset).  "size" is the
3801 // size of the operand.  The result may be in rscratch2.
3802 Address MacroAssembler::offsetted_address(Register r, Register r1,
3803                                           Address::extend ext, int offset, int size) {
3804   if (offset || (ext.shift() % size != 0)) {
3805     lea(rscratch2, Address(r, r1, ext));
3806     return Address(rscratch2, offset);
3807   } else {
3808     return Address(r, r1, ext);
3809   }
3810 }
3811 
3812 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3813 {
3814   assert(offset >= 0, "spill to negative address?");
3815   // Offset reachable ?
3816   //   Not aligned - 9 bits signed offset
3817   //   Aligned - 12 bits unsigned offset shifted
3818   Register base = sp;
3819   if ((offset & (size-1)) && offset >= (1<<8)) {
3820     add(tmp, base, offset & ((1<<12)-1));
3821     base = tmp;
3822     offset &= -1u<<12;
3823   }
3824 
3825   if (offset >= (1<<12) * size) {
3826     add(tmp, base, offset & (((1<<12)-1)<<12));
3827     base = tmp;
3828     offset &= ~(((1<<12)-1)<<12);
3829   }
3830 
3831   return Address(base, offset);
3832 }
3833 
3834 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3835   assert(offset >= 0, "spill to negative address?");
3836 
3837   Register base = sp;
3838 
3839   // An immediate offset in the range 0 to 255 which is multiplied
3840   // by the current vector or predicate register size in bytes.
3841   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3842     return Address(base, offset / sve_reg_size_in_bytes);
3843   }
3844 
3845   add(tmp, base, offset);
3846   return Address(tmp);
3847 }
3848 
3849 // Checks whether offset is aligned.
3850 // Returns true if it is, else false.
3851 bool MacroAssembler::merge_alignment_check(Register base,
3852                                            size_t size,
3853                                            int64_t cur_offset,
3854                                            int64_t prev_offset) const {
3855   if (AvoidUnalignedAccesses) {
3856     if (base == sp) {
3857       // Checks whether low offset if aligned to pair of registers.
3858       int64_t pair_mask = size * 2 - 1;
3859       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3860       return (offset & pair_mask) == 0;
3861     } else { // If base is not sp, we can't guarantee the access is aligned.
3862       return false;
3863     }
3864   } else {
3865     int64_t mask = size - 1;
3866     // Load/store pair instruction only supports element size aligned offset.
3867     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3868   }
3869 }
3870 
3871 // Checks whether current and previous loads/stores can be merged.
3872 // Returns true if it can be merged, else false.
3873 bool MacroAssembler::ldst_can_merge(Register rt,
3874                                     const Address &adr,
3875                                     size_t cur_size_in_bytes,
3876                                     bool is_store) const {
3877   address prev = pc() - NativeInstruction::instruction_size;
3878   address last = code()->last_insn();
3879 
3880   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3881     return false;
3882   }
3883 
3884   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3885     return false;
3886   }
3887 
3888   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3889   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3890 
3891   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3892   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3893 
3894   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3895     return false;
3896   }
3897 
3898   int64_t max_offset = 63 * prev_size_in_bytes;
3899   int64_t min_offset = -64 * prev_size_in_bytes;
3900 
3901   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3902 
3903   // Only same base can be merged.
3904   if (adr.base() != prev_ldst->base()) {
3905     return false;
3906   }
3907 
3908   int64_t cur_offset = adr.offset();
3909   int64_t prev_offset = prev_ldst->offset();
3910   size_t diff = abs(cur_offset - prev_offset);
3911   if (diff != prev_size_in_bytes) {
3912     return false;
3913   }
3914 
3915   // Following cases can not be merged:
3916   // ldr x2, [x2, #8]
3917   // ldr x3, [x2, #16]
3918   // or:
3919   // ldr x2, [x3, #8]
3920   // ldr x2, [x3, #16]
3921   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3922   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3923     return false;
3924   }
3925 
3926   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3927   // Offset range must be in ldp/stp instruction's range.
3928   if (low_offset > max_offset || low_offset < min_offset) {
3929     return false;
3930   }
3931 
3932   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3933     return true;
3934   }
3935 
3936   return false;
3937 }
3938 
3939 // Merge current load/store with previous load/store into ldp/stp.
3940 void MacroAssembler::merge_ldst(Register rt,
3941                                 const Address &adr,
3942                                 size_t cur_size_in_bytes,
3943                                 bool is_store) {
3944 
3945   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3946 
3947   Register rt_low, rt_high;
3948   address prev = pc() - NativeInstruction::instruction_size;
3949   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3950 
3951   int64_t offset;
3952 
3953   if (adr.offset() < prev_ldst->offset()) {
3954     offset = adr.offset();
3955     rt_low = rt;
3956     rt_high = prev_ldst->target();
3957   } else {
3958     offset = prev_ldst->offset();
3959     rt_low = prev_ldst->target();
3960     rt_high = rt;
3961   }
3962 
3963   Address adr_p = Address(prev_ldst->base(), offset);
3964   // Overwrite previous generated binary.
3965   code_section()->set_end(prev);
3966 
3967   const size_t sz = prev_ldst->size_in_bytes();
3968   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3969   if (!is_store) {
3970     BLOCK_COMMENT("merged ldr pair");
3971     if (sz == 8) {
3972       ldp(rt_low, rt_high, adr_p);
3973     } else {
3974       ldpw(rt_low, rt_high, adr_p);
3975     }
3976   } else {
3977     BLOCK_COMMENT("merged str pair");
3978     if (sz == 8) {
3979       stp(rt_low, rt_high, adr_p);
3980     } else {
3981       stpw(rt_low, rt_high, adr_p);
3982     }
3983   }
3984 }
3985 
3986 /**
3987  * Multiply 64 bit by 64 bit first loop.
3988  */
3989 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3990                                            Register y, Register y_idx, Register z,
3991                                            Register carry, Register product,
3992                                            Register idx, Register kdx) {
3993   //
3994   //  jlong carry, x[], y[], z[];
3995   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3996   //    huge_128 product = y[idx] * x[xstart] + carry;
3997   //    z[kdx] = (jlong)product;
3998   //    carry  = (jlong)(product >>> 64);
3999   //  }
4000   //  z[xstart] = carry;
4001   //
4002 
4003   Label L_first_loop, L_first_loop_exit;
4004   Label L_one_x, L_one_y, L_multiply;
4005 
4006   subsw(xstart, xstart, 1);
4007   br(Assembler::MI, L_one_x);
4008 
4009   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
4010   ldr(x_xstart, Address(rscratch1));
4011   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
4012 
4013   bind(L_first_loop);
4014   subsw(idx, idx, 1);
4015   br(Assembler::MI, L_first_loop_exit);
4016   subsw(idx, idx, 1);
4017   br(Assembler::MI, L_one_y);
4018   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4019   ldr(y_idx, Address(rscratch1));
4020   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
4021   bind(L_multiply);
4022 
4023   // AArch64 has a multiply-accumulate instruction that we can't use
4024   // here because it has no way to process carries, so we have to use
4025   // separate add and adc instructions.  Bah.
4026   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
4027   mul(product, x_xstart, y_idx);
4028   adds(product, product, carry);
4029   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
4030 
4031   subw(kdx, kdx, 2);
4032   ror(product, product, 32); // back to big-endian
4033   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4034 
4035   b(L_first_loop);
4036 
4037   bind(L_one_y);
4038   ldrw(y_idx, Address(y,  0));
4039   b(L_multiply);
4040 
4041   bind(L_one_x);
4042   ldrw(x_xstart, Address(x,  0));
4043   b(L_first_loop);
4044 
4045   bind(L_first_loop_exit);
4046 }
4047 
4048 /**
4049  * Multiply 128 bit by 128. Unrolled inner loop.
4050  *
4051  */
4052 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4053                                              Register carry, Register carry2,
4054                                              Register idx, Register jdx,
4055                                              Register yz_idx1, Register yz_idx2,
4056                                              Register tmp, Register tmp3, Register tmp4,
4057                                              Register tmp6, Register product_hi) {
4058 
4059   //   jlong carry, x[], y[], z[];
4060   //   int kdx = ystart+1;
4061   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4062   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4063   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4064   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4065   //     carry  = (jlong)(tmp4 >>> 64);
4066   //     z[kdx+idx+1] = (jlong)tmp3;
4067   //     z[kdx+idx] = (jlong)tmp4;
4068   //   }
4069   //   idx += 2;
4070   //   if (idx > 0) {
4071   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4072   //     z[kdx+idx] = (jlong)yz_idx1;
4073   //     carry  = (jlong)(yz_idx1 >>> 64);
4074   //   }
4075   //
4076 
4077   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4078 
4079   lsrw(jdx, idx, 2);
4080 
4081   bind(L_third_loop);
4082 
4083   subsw(jdx, jdx, 1);
4084   br(Assembler::MI, L_third_loop_exit);
4085   subw(idx, idx, 4);
4086 
4087   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4088 
4089   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4090 
4091   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4092 
4093   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4094   ror(yz_idx2, yz_idx2, 32);
4095 
4096   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4097 
4098   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4099   umulh(tmp4, product_hi, yz_idx1);
4100 
4101   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4102   ror(rscratch2, rscratch2, 32);
4103 
4104   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4105   umulh(carry2, product_hi, yz_idx2);
4106 
4107   // propagate sum of both multiplications into carry:tmp4:tmp3
4108   adds(tmp3, tmp3, carry);
4109   adc(tmp4, tmp4, zr);
4110   adds(tmp3, tmp3, rscratch1);
4111   adcs(tmp4, tmp4, tmp);
4112   adc(carry, carry2, zr);
4113   adds(tmp4, tmp4, rscratch2);
4114   adc(carry, carry, zr);
4115 
4116   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4117   ror(tmp4, tmp4, 32);
4118   stp(tmp4, tmp3, Address(tmp6, 0));
4119 
4120   b(L_third_loop);
4121   bind (L_third_loop_exit);
4122 
4123   andw (idx, idx, 0x3);
4124   cbz(idx, L_post_third_loop_done);
4125 
4126   Label L_check_1;
4127   subsw(idx, idx, 2);
4128   br(Assembler::MI, L_check_1);
4129 
4130   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4131   ldr(yz_idx1, Address(rscratch1, 0));
4132   ror(yz_idx1, yz_idx1, 32);
4133   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4134   umulh(tmp4, product_hi, yz_idx1);
4135   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4136   ldr(yz_idx2, Address(rscratch1, 0));
4137   ror(yz_idx2, yz_idx2, 32);
4138 
4139   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4140 
4141   ror(tmp3, tmp3, 32);
4142   str(tmp3, Address(rscratch1, 0));
4143 
4144   bind (L_check_1);
4145 
4146   andw (idx, idx, 0x1);
4147   subsw(idx, idx, 1);
4148   br(Assembler::MI, L_post_third_loop_done);
4149   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4150   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4151   umulh(carry2, tmp4, product_hi);
4152   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4153 
4154   add2_with_carry(carry2, tmp3, tmp4, carry);
4155 
4156   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4157   extr(carry, carry2, tmp3, 32);
4158 
4159   bind(L_post_third_loop_done);
4160 }
4161 
4162 /**
4163  * Code for BigInteger::multiplyToLen() intrinsic.
4164  *
4165  * r0: x
4166  * r1: xlen
4167  * r2: y
4168  * r3: ylen
4169  * r4:  z
4170  * r5: tmp0
4171  * r10: tmp1
4172  * r11: tmp2
4173  * r12: tmp3
4174  * r13: tmp4
4175  * r14: tmp5
4176  * r15: tmp6
4177  * r16: tmp7
4178  *
4179  */
4180 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4181                                      Register z, Register tmp0,
4182                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4183                                      Register tmp5, Register tmp6, Register product_hi) {
4184 
4185   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4186 
4187   const Register idx = tmp1;
4188   const Register kdx = tmp2;
4189   const Register xstart = tmp3;
4190 
4191   const Register y_idx = tmp4;
4192   const Register carry = tmp5;
4193   const Register product  = xlen;
4194   const Register x_xstart = tmp0;
4195 
4196   // First Loop.
4197   //
4198   //  final static long LONG_MASK = 0xffffffffL;
4199   //  int xstart = xlen - 1;
4200   //  int ystart = ylen - 1;
4201   //  long carry = 0;
4202   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4203   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4204   //    z[kdx] = (int)product;
4205   //    carry = product >>> 32;
4206   //  }
4207   //  z[xstart] = (int)carry;
4208   //
4209 
4210   movw(idx, ylen);       // idx = ylen;
4211   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4212   mov(carry, zr);        // carry = 0;
4213 
4214   Label L_done;
4215 
4216   movw(xstart, xlen);
4217   subsw(xstart, xstart, 1);
4218   br(Assembler::MI, L_done);
4219 
4220   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4221 
4222   Label L_second_loop;
4223   cbzw(kdx, L_second_loop);
4224 
4225   Label L_carry;
4226   subw(kdx, kdx, 1);
4227   cbzw(kdx, L_carry);
4228 
4229   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4230   lsr(carry, carry, 32);
4231   subw(kdx, kdx, 1);
4232 
4233   bind(L_carry);
4234   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4235 
4236   // Second and third (nested) loops.
4237   //
4238   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4239   //   carry = 0;
4240   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4241   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4242   //                    (z[k] & LONG_MASK) + carry;
4243   //     z[k] = (int)product;
4244   //     carry = product >>> 32;
4245   //   }
4246   //   z[i] = (int)carry;
4247   // }
4248   //
4249   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4250 
4251   const Register jdx = tmp1;
4252 
4253   bind(L_second_loop);
4254   mov(carry, zr);                // carry = 0;
4255   movw(jdx, ylen);               // j = ystart+1
4256 
4257   subsw(xstart, xstart, 1);      // i = xstart-1;
4258   br(Assembler::MI, L_done);
4259 
4260   str(z, Address(pre(sp, -4 * wordSize)));
4261 
4262   Label L_last_x;
4263   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4264   subsw(xstart, xstart, 1);       // i = xstart-1;
4265   br(Assembler::MI, L_last_x);
4266 
4267   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4268   ldr(product_hi, Address(rscratch1));
4269   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4270 
4271   Label L_third_loop_prologue;
4272   bind(L_third_loop_prologue);
4273 
4274   str(ylen, Address(sp, wordSize));
4275   stp(x, xstart, Address(sp, 2 * wordSize));
4276   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4277                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4278   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4279   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4280 
4281   addw(tmp3, xlen, 1);
4282   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4283   subsw(tmp3, tmp3, 1);
4284   br(Assembler::MI, L_done);
4285 
4286   lsr(carry, carry, 32);
4287   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4288   b(L_second_loop);
4289 
4290   // Next infrequent code is moved outside loops.
4291   bind(L_last_x);
4292   ldrw(product_hi, Address(x,  0));
4293   b(L_third_loop_prologue);
4294 
4295   bind(L_done);
4296 }
4297 
4298 // Code for BigInteger::mulAdd intrinsic
4299 // out     = r0
4300 // in      = r1
4301 // offset  = r2  (already out.length-offset)
4302 // len     = r3
4303 // k       = r4
4304 //
4305 // pseudo code from java implementation:
4306 // carry = 0;
4307 // offset = out.length-offset - 1;
4308 // for (int j=len-1; j >= 0; j--) {
4309 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4310 //     out[offset--] = (int)product;
4311 //     carry = product >>> 32;
4312 // }
4313 // return (int)carry;
4314 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4315       Register len, Register k) {
4316     Label LOOP, END;
4317     // pre-loop
4318     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4319     csel(out, zr, out, Assembler::EQ);
4320     br(Assembler::EQ, END);
4321     add(in, in, len, LSL, 2); // in[j+1] address
4322     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4323     mov(out, zr); // used to keep carry now
4324     BIND(LOOP);
4325     ldrw(rscratch1, Address(pre(in, -4)));
4326     madd(rscratch1, rscratch1, k, out);
4327     ldrw(rscratch2, Address(pre(offset, -4)));
4328     add(rscratch1, rscratch1, rscratch2);
4329     strw(rscratch1, Address(offset));
4330     lsr(out, rscratch1, 32);
4331     subs(len, len, 1);
4332     br(Assembler::NE, LOOP);
4333     BIND(END);
4334 }
4335 
4336 /**
4337  * Emits code to update CRC-32 with a byte value according to constants in table
4338  *
4339  * @param [in,out]crc   Register containing the crc.
4340  * @param [in]val       Register containing the byte to fold into the CRC.
4341  * @param [in]table     Register containing the table of crc constants.
4342  *
4343  * uint32_t crc;
4344  * val = crc_table[(val ^ crc) & 0xFF];
4345  * crc = val ^ (crc >> 8);
4346  *
4347  */
4348 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4349   eor(val, val, crc);
4350   andr(val, val, 0xff);
4351   ldrw(val, Address(table, val, Address::lsl(2)));
4352   eor(crc, val, crc, Assembler::LSR, 8);
4353 }
4354 
4355 /**
4356  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4357  *
4358  * @param [in,out]crc   Register containing the crc.
4359  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4360  * @param [in]table0    Register containing table 0 of crc constants.
4361  * @param [in]table1    Register containing table 1 of crc constants.
4362  * @param [in]table2    Register containing table 2 of crc constants.
4363  * @param [in]table3    Register containing table 3 of crc constants.
4364  *
4365  * uint32_t crc;
4366  *   v = crc ^ v
4367  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4368  *
4369  */
4370 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4371         Register table0, Register table1, Register table2, Register table3,
4372         bool upper) {
4373   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4374   uxtb(tmp, v);
4375   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4376   ubfx(tmp, v, 8, 8);
4377   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4378   eor(crc, crc, tmp);
4379   ubfx(tmp, v, 16, 8);
4380   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4381   eor(crc, crc, tmp);
4382   ubfx(tmp, v, 24, 8);
4383   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4384   eor(crc, crc, tmp);
4385 }
4386 
4387 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4388         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4389     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4390     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4391 
4392     subs(tmp0, len, 384);
4393     mvnw(crc, crc);
4394     br(Assembler::GE, CRC_by128_pre);
4395   BIND(CRC_less128);
4396     subs(len, len, 32);
4397     br(Assembler::GE, CRC_by32_loop);
4398   BIND(CRC_less32);
4399     adds(len, len, 32 - 4);
4400     br(Assembler::GE, CRC_by4_loop);
4401     adds(len, len, 4);
4402     br(Assembler::GT, CRC_by1_loop);
4403     b(L_exit);
4404 
4405   BIND(CRC_by32_loop);
4406     ldp(tmp0, tmp1, Address(buf));
4407     crc32x(crc, crc, tmp0);
4408     ldp(tmp2, tmp3, Address(buf, 16));
4409     crc32x(crc, crc, tmp1);
4410     add(buf, buf, 32);
4411     crc32x(crc, crc, tmp2);
4412     subs(len, len, 32);
4413     crc32x(crc, crc, tmp3);
4414     br(Assembler::GE, CRC_by32_loop);
4415     cmn(len, (u1)32);
4416     br(Assembler::NE, CRC_less32);
4417     b(L_exit);
4418 
4419   BIND(CRC_by4_loop);
4420     ldrw(tmp0, Address(post(buf, 4)));
4421     subs(len, len, 4);
4422     crc32w(crc, crc, tmp0);
4423     br(Assembler::GE, CRC_by4_loop);
4424     adds(len, len, 4);
4425     br(Assembler::LE, L_exit);
4426   BIND(CRC_by1_loop);
4427     ldrb(tmp0, Address(post(buf, 1)));
4428     subs(len, len, 1);
4429     crc32b(crc, crc, tmp0);
4430     br(Assembler::GT, CRC_by1_loop);
4431     b(L_exit);
4432 
4433   BIND(CRC_by128_pre);
4434     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4435       4*256*sizeof(juint) + 8*sizeof(juint));
4436     mov(crc, 0);
4437     crc32x(crc, crc, tmp0);
4438     crc32x(crc, crc, tmp1);
4439 
4440     cbnz(len, CRC_less128);
4441 
4442   BIND(L_exit);
4443     mvnw(crc, crc);
4444 }
4445 
4446 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4447         Register len, Register tmp0, Register tmp1, Register tmp2,
4448         Register tmp3) {
4449     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4450     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4451 
4452     mvnw(crc, crc);
4453 
4454     subs(len, len, 128);
4455     br(Assembler::GE, CRC_by64_pre);
4456   BIND(CRC_less64);
4457     adds(len, len, 128-32);
4458     br(Assembler::GE, CRC_by32_loop);
4459   BIND(CRC_less32);
4460     adds(len, len, 32-4);
4461     br(Assembler::GE, CRC_by4_loop);
4462     adds(len, len, 4);
4463     br(Assembler::GT, CRC_by1_loop);
4464     b(L_exit);
4465 
4466   BIND(CRC_by32_loop);
4467     ldp(tmp0, tmp1, Address(post(buf, 16)));
4468     subs(len, len, 32);
4469     crc32x(crc, crc, tmp0);
4470     ldr(tmp2, Address(post(buf, 8)));
4471     crc32x(crc, crc, tmp1);
4472     ldr(tmp3, Address(post(buf, 8)));
4473     crc32x(crc, crc, tmp2);
4474     crc32x(crc, crc, tmp3);
4475     br(Assembler::GE, CRC_by32_loop);
4476     cmn(len, (u1)32);
4477     br(Assembler::NE, CRC_less32);
4478     b(L_exit);
4479 
4480   BIND(CRC_by4_loop);
4481     ldrw(tmp0, Address(post(buf, 4)));
4482     subs(len, len, 4);
4483     crc32w(crc, crc, tmp0);
4484     br(Assembler::GE, CRC_by4_loop);
4485     adds(len, len, 4);
4486     br(Assembler::LE, L_exit);
4487   BIND(CRC_by1_loop);
4488     ldrb(tmp0, Address(post(buf, 1)));
4489     subs(len, len, 1);
4490     crc32b(crc, crc, tmp0);
4491     br(Assembler::GT, CRC_by1_loop);
4492     b(L_exit);
4493 
4494   BIND(CRC_by64_pre);
4495     sub(buf, buf, 8);
4496     ldp(tmp0, tmp1, Address(buf, 8));
4497     crc32x(crc, crc, tmp0);
4498     ldr(tmp2, Address(buf, 24));
4499     crc32x(crc, crc, tmp1);
4500     ldr(tmp3, Address(buf, 32));
4501     crc32x(crc, crc, tmp2);
4502     ldr(tmp0, Address(buf, 40));
4503     crc32x(crc, crc, tmp3);
4504     ldr(tmp1, Address(buf, 48));
4505     crc32x(crc, crc, tmp0);
4506     ldr(tmp2, Address(buf, 56));
4507     crc32x(crc, crc, tmp1);
4508     ldr(tmp3, Address(pre(buf, 64)));
4509 
4510     b(CRC_by64_loop);
4511 
4512     align(CodeEntryAlignment);
4513   BIND(CRC_by64_loop);
4514     subs(len, len, 64);
4515     crc32x(crc, crc, tmp2);
4516     ldr(tmp0, Address(buf, 8));
4517     crc32x(crc, crc, tmp3);
4518     ldr(tmp1, Address(buf, 16));
4519     crc32x(crc, crc, tmp0);
4520     ldr(tmp2, Address(buf, 24));
4521     crc32x(crc, crc, tmp1);
4522     ldr(tmp3, Address(buf, 32));
4523     crc32x(crc, crc, tmp2);
4524     ldr(tmp0, Address(buf, 40));
4525     crc32x(crc, crc, tmp3);
4526     ldr(tmp1, Address(buf, 48));
4527     crc32x(crc, crc, tmp0);
4528     ldr(tmp2, Address(buf, 56));
4529     crc32x(crc, crc, tmp1);
4530     ldr(tmp3, Address(pre(buf, 64)));
4531     br(Assembler::GE, CRC_by64_loop);
4532 
4533     // post-loop
4534     crc32x(crc, crc, tmp2);
4535     crc32x(crc, crc, tmp3);
4536 
4537     sub(len, len, 64);
4538     add(buf, buf, 8);
4539     cmn(len, (u1)128);
4540     br(Assembler::NE, CRC_less64);
4541   BIND(L_exit);
4542     mvnw(crc, crc);
4543 }
4544 
4545 /**
4546  * @param crc   register containing existing CRC (32-bit)
4547  * @param buf   register pointing to input byte buffer (byte*)
4548  * @param len   register containing number of bytes
4549  * @param table register that will contain address of CRC table
4550  * @param tmp   scratch register
4551  */
4552 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4553         Register table0, Register table1, Register table2, Register table3,
4554         Register tmp, Register tmp2, Register tmp3) {
4555   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4556 
4557   if (UseCryptoPmullForCRC32) {
4558       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4559       return;
4560   }
4561 
4562   if (UseCRC32) {
4563       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4564       return;
4565   }
4566 
4567     mvnw(crc, crc);
4568 
4569     {
4570       uint64_t offset;
4571       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4572       add(table0, table0, offset);
4573     }
4574     add(table1, table0, 1*256*sizeof(juint));
4575     add(table2, table0, 2*256*sizeof(juint));
4576     add(table3, table0, 3*256*sizeof(juint));
4577 
4578     { // Neon code start
4579       cmp(len, (u1)64);
4580       br(Assembler::LT, L_by16);
4581       eor(v16, T16B, v16, v16);
4582 
4583     Label L_fold;
4584 
4585       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4586 
4587       ld1(v0, v1, T2D, post(buf, 32));
4588       ld1r(v4, T2D, post(tmp, 8));
4589       ld1r(v5, T2D, post(tmp, 8));
4590       ld1r(v6, T2D, post(tmp, 8));
4591       ld1r(v7, T2D, post(tmp, 8));
4592       mov(v16, S, 0, crc);
4593 
4594       eor(v0, T16B, v0, v16);
4595       sub(len, len, 64);
4596 
4597     BIND(L_fold);
4598       pmull(v22, T8H, v0, v5, T8B);
4599       pmull(v20, T8H, v0, v7, T8B);
4600       pmull(v23, T8H, v0, v4, T8B);
4601       pmull(v21, T8H, v0, v6, T8B);
4602 
4603       pmull2(v18, T8H, v0, v5, T16B);
4604       pmull2(v16, T8H, v0, v7, T16B);
4605       pmull2(v19, T8H, v0, v4, T16B);
4606       pmull2(v17, T8H, v0, v6, T16B);
4607 
4608       uzp1(v24, T8H, v20, v22);
4609       uzp2(v25, T8H, v20, v22);
4610       eor(v20, T16B, v24, v25);
4611 
4612       uzp1(v26, T8H, v16, v18);
4613       uzp2(v27, T8H, v16, v18);
4614       eor(v16, T16B, v26, v27);
4615 
4616       ushll2(v22, T4S, v20, T8H, 8);
4617       ushll(v20, T4S, v20, T4H, 8);
4618 
4619       ushll2(v18, T4S, v16, T8H, 8);
4620       ushll(v16, T4S, v16, T4H, 8);
4621 
4622       eor(v22, T16B, v23, v22);
4623       eor(v18, T16B, v19, v18);
4624       eor(v20, T16B, v21, v20);
4625       eor(v16, T16B, v17, v16);
4626 
4627       uzp1(v17, T2D, v16, v20);
4628       uzp2(v21, T2D, v16, v20);
4629       eor(v17, T16B, v17, v21);
4630 
4631       ushll2(v20, T2D, v17, T4S, 16);
4632       ushll(v16, T2D, v17, T2S, 16);
4633 
4634       eor(v20, T16B, v20, v22);
4635       eor(v16, T16B, v16, v18);
4636 
4637       uzp1(v17, T2D, v20, v16);
4638       uzp2(v21, T2D, v20, v16);
4639       eor(v28, T16B, v17, v21);
4640 
4641       pmull(v22, T8H, v1, v5, T8B);
4642       pmull(v20, T8H, v1, v7, T8B);
4643       pmull(v23, T8H, v1, v4, T8B);
4644       pmull(v21, T8H, v1, v6, T8B);
4645 
4646       pmull2(v18, T8H, v1, v5, T16B);
4647       pmull2(v16, T8H, v1, v7, T16B);
4648       pmull2(v19, T8H, v1, v4, T16B);
4649       pmull2(v17, T8H, v1, v6, T16B);
4650 
4651       ld1(v0, v1, T2D, post(buf, 32));
4652 
4653       uzp1(v24, T8H, v20, v22);
4654       uzp2(v25, T8H, v20, v22);
4655       eor(v20, T16B, v24, v25);
4656 
4657       uzp1(v26, T8H, v16, v18);
4658       uzp2(v27, T8H, v16, v18);
4659       eor(v16, T16B, v26, v27);
4660 
4661       ushll2(v22, T4S, v20, T8H, 8);
4662       ushll(v20, T4S, v20, T4H, 8);
4663 
4664       ushll2(v18, T4S, v16, T8H, 8);
4665       ushll(v16, T4S, v16, T4H, 8);
4666 
4667       eor(v22, T16B, v23, v22);
4668       eor(v18, T16B, v19, v18);
4669       eor(v20, T16B, v21, v20);
4670       eor(v16, T16B, v17, v16);
4671 
4672       uzp1(v17, T2D, v16, v20);
4673       uzp2(v21, T2D, v16, v20);
4674       eor(v16, T16B, v17, v21);
4675 
4676       ushll2(v20, T2D, v16, T4S, 16);
4677       ushll(v16, T2D, v16, T2S, 16);
4678 
4679       eor(v20, T16B, v22, v20);
4680       eor(v16, T16B, v16, v18);
4681 
4682       uzp1(v17, T2D, v20, v16);
4683       uzp2(v21, T2D, v20, v16);
4684       eor(v20, T16B, v17, v21);
4685 
4686       shl(v16, T2D, v28, 1);
4687       shl(v17, T2D, v20, 1);
4688 
4689       eor(v0, T16B, v0, v16);
4690       eor(v1, T16B, v1, v17);
4691 
4692       subs(len, len, 32);
4693       br(Assembler::GE, L_fold);
4694 
4695       mov(crc, 0);
4696       mov(tmp, v0, D, 0);
4697       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4698       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4699       mov(tmp, v0, D, 1);
4700       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4701       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4702       mov(tmp, v1, D, 0);
4703       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4704       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4705       mov(tmp, v1, D, 1);
4706       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4707       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4708 
4709       add(len, len, 32);
4710     } // Neon code end
4711 
4712   BIND(L_by16);
4713     subs(len, len, 16);
4714     br(Assembler::GE, L_by16_loop);
4715     adds(len, len, 16-4);
4716     br(Assembler::GE, L_by4_loop);
4717     adds(len, len, 4);
4718     br(Assembler::GT, L_by1_loop);
4719     b(L_exit);
4720 
4721   BIND(L_by4_loop);
4722     ldrw(tmp, Address(post(buf, 4)));
4723     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4724     subs(len, len, 4);
4725     br(Assembler::GE, L_by4_loop);
4726     adds(len, len, 4);
4727     br(Assembler::LE, L_exit);
4728   BIND(L_by1_loop);
4729     subs(len, len, 1);
4730     ldrb(tmp, Address(post(buf, 1)));
4731     update_byte_crc32(crc, tmp, table0);
4732     br(Assembler::GT, L_by1_loop);
4733     b(L_exit);
4734 
4735     align(CodeEntryAlignment);
4736   BIND(L_by16_loop);
4737     subs(len, len, 16);
4738     ldp(tmp, tmp3, Address(post(buf, 16)));
4739     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4740     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4741     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4742     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4743     br(Assembler::GE, L_by16_loop);
4744     adds(len, len, 16-4);
4745     br(Assembler::GE, L_by4_loop);
4746     adds(len, len, 4);
4747     br(Assembler::GT, L_by1_loop);
4748   BIND(L_exit);
4749     mvnw(crc, crc);
4750 }
4751 
4752 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4753         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4754     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4755     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4756 
4757     subs(tmp0, len, 384);
4758     br(Assembler::GE, CRC_by128_pre);
4759   BIND(CRC_less128);
4760     subs(len, len, 32);
4761     br(Assembler::GE, CRC_by32_loop);
4762   BIND(CRC_less32);
4763     adds(len, len, 32 - 4);
4764     br(Assembler::GE, CRC_by4_loop);
4765     adds(len, len, 4);
4766     br(Assembler::GT, CRC_by1_loop);
4767     b(L_exit);
4768 
4769   BIND(CRC_by32_loop);
4770     ldp(tmp0, tmp1, Address(buf));
4771     crc32cx(crc, crc, tmp0);
4772     ldr(tmp2, Address(buf, 16));
4773     crc32cx(crc, crc, tmp1);
4774     ldr(tmp3, Address(buf, 24));
4775     crc32cx(crc, crc, tmp2);
4776     add(buf, buf, 32);
4777     subs(len, len, 32);
4778     crc32cx(crc, crc, tmp3);
4779     br(Assembler::GE, CRC_by32_loop);
4780     cmn(len, (u1)32);
4781     br(Assembler::NE, CRC_less32);
4782     b(L_exit);
4783 
4784   BIND(CRC_by4_loop);
4785     ldrw(tmp0, Address(post(buf, 4)));
4786     subs(len, len, 4);
4787     crc32cw(crc, crc, tmp0);
4788     br(Assembler::GE, CRC_by4_loop);
4789     adds(len, len, 4);
4790     br(Assembler::LE, L_exit);
4791   BIND(CRC_by1_loop);
4792     ldrb(tmp0, Address(post(buf, 1)));
4793     subs(len, len, 1);
4794     crc32cb(crc, crc, tmp0);
4795     br(Assembler::GT, CRC_by1_loop);
4796     b(L_exit);
4797 
4798   BIND(CRC_by128_pre);
4799     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4800       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4801     mov(crc, 0);
4802     crc32cx(crc, crc, tmp0);
4803     crc32cx(crc, crc, tmp1);
4804 
4805     cbnz(len, CRC_less128);
4806 
4807   BIND(L_exit);
4808 }
4809 
4810 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4811         Register len, Register tmp0, Register tmp1, Register tmp2,
4812         Register tmp3) {
4813     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4814     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4815 
4816     subs(len, len, 128);
4817     br(Assembler::GE, CRC_by64_pre);
4818   BIND(CRC_less64);
4819     adds(len, len, 128-32);
4820     br(Assembler::GE, CRC_by32_loop);
4821   BIND(CRC_less32);
4822     adds(len, len, 32-4);
4823     br(Assembler::GE, CRC_by4_loop);
4824     adds(len, len, 4);
4825     br(Assembler::GT, CRC_by1_loop);
4826     b(L_exit);
4827 
4828   BIND(CRC_by32_loop);
4829     ldp(tmp0, tmp1, Address(post(buf, 16)));
4830     subs(len, len, 32);
4831     crc32cx(crc, crc, tmp0);
4832     ldr(tmp2, Address(post(buf, 8)));
4833     crc32cx(crc, crc, tmp1);
4834     ldr(tmp3, Address(post(buf, 8)));
4835     crc32cx(crc, crc, tmp2);
4836     crc32cx(crc, crc, tmp3);
4837     br(Assembler::GE, CRC_by32_loop);
4838     cmn(len, (u1)32);
4839     br(Assembler::NE, CRC_less32);
4840     b(L_exit);
4841 
4842   BIND(CRC_by4_loop);
4843     ldrw(tmp0, Address(post(buf, 4)));
4844     subs(len, len, 4);
4845     crc32cw(crc, crc, tmp0);
4846     br(Assembler::GE, CRC_by4_loop);
4847     adds(len, len, 4);
4848     br(Assembler::LE, L_exit);
4849   BIND(CRC_by1_loop);
4850     ldrb(tmp0, Address(post(buf, 1)));
4851     subs(len, len, 1);
4852     crc32cb(crc, crc, tmp0);
4853     br(Assembler::GT, CRC_by1_loop);
4854     b(L_exit);
4855 
4856   BIND(CRC_by64_pre);
4857     sub(buf, buf, 8);
4858     ldp(tmp0, tmp1, Address(buf, 8));
4859     crc32cx(crc, crc, tmp0);
4860     ldr(tmp2, Address(buf, 24));
4861     crc32cx(crc, crc, tmp1);
4862     ldr(tmp3, Address(buf, 32));
4863     crc32cx(crc, crc, tmp2);
4864     ldr(tmp0, Address(buf, 40));
4865     crc32cx(crc, crc, tmp3);
4866     ldr(tmp1, Address(buf, 48));
4867     crc32cx(crc, crc, tmp0);
4868     ldr(tmp2, Address(buf, 56));
4869     crc32cx(crc, crc, tmp1);
4870     ldr(tmp3, Address(pre(buf, 64)));
4871 
4872     b(CRC_by64_loop);
4873 
4874     align(CodeEntryAlignment);
4875   BIND(CRC_by64_loop);
4876     subs(len, len, 64);
4877     crc32cx(crc, crc, tmp2);
4878     ldr(tmp0, Address(buf, 8));
4879     crc32cx(crc, crc, tmp3);
4880     ldr(tmp1, Address(buf, 16));
4881     crc32cx(crc, crc, tmp0);
4882     ldr(tmp2, Address(buf, 24));
4883     crc32cx(crc, crc, tmp1);
4884     ldr(tmp3, Address(buf, 32));
4885     crc32cx(crc, crc, tmp2);
4886     ldr(tmp0, Address(buf, 40));
4887     crc32cx(crc, crc, tmp3);
4888     ldr(tmp1, Address(buf, 48));
4889     crc32cx(crc, crc, tmp0);
4890     ldr(tmp2, Address(buf, 56));
4891     crc32cx(crc, crc, tmp1);
4892     ldr(tmp3, Address(pre(buf, 64)));
4893     br(Assembler::GE, CRC_by64_loop);
4894 
4895     // post-loop
4896     crc32cx(crc, crc, tmp2);
4897     crc32cx(crc, crc, tmp3);
4898 
4899     sub(len, len, 64);
4900     add(buf, buf, 8);
4901     cmn(len, (u1)128);
4902     br(Assembler::NE, CRC_less64);
4903   BIND(L_exit);
4904 }
4905 
4906 /**
4907  * @param crc   register containing existing CRC (32-bit)
4908  * @param buf   register pointing to input byte buffer (byte*)
4909  * @param len   register containing number of bytes
4910  * @param table register that will contain address of CRC table
4911  * @param tmp   scratch register
4912  */
4913 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4914         Register table0, Register table1, Register table2, Register table3,
4915         Register tmp, Register tmp2, Register tmp3) {
4916   if (UseCryptoPmullForCRC32) {
4917     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4918   } else {
4919     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4920   }
4921 }
4922 
4923 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4924         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4925     Label CRC_by128_loop;
4926     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4927 
4928     sub(len, len, 256);
4929     Register table = tmp0;
4930     {
4931       uint64_t offset;
4932       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4933       add(table, table, offset);
4934     }
4935     add(table, table, table_offset);
4936 
4937     // Registers v0..v7 are used as data registers.
4938     // Registers v16..v31 are used as tmp registers.
4939     sub(buf, buf, 0x10);
4940     ldrq(v0, Address(buf, 0x10));
4941     ldrq(v1, Address(buf, 0x20));
4942     ldrq(v2, Address(buf, 0x30));
4943     ldrq(v3, Address(buf, 0x40));
4944     ldrq(v4, Address(buf, 0x50));
4945     ldrq(v5, Address(buf, 0x60));
4946     ldrq(v6, Address(buf, 0x70));
4947     ldrq(v7, Address(pre(buf, 0x80)));
4948 
4949     movi(v31, T4S, 0);
4950     mov(v31, S, 0, crc);
4951     eor(v0, T16B, v0, v31);
4952 
4953     // Register v16 contains constants from the crc table.
4954     ldrq(v16, Address(table));
4955     b(CRC_by128_loop);
4956 
4957     align(OptoLoopAlignment);
4958   BIND(CRC_by128_loop);
4959     pmull (v17,  T1Q, v0, v16, T1D);
4960     pmull2(v18, T1Q, v0, v16, T2D);
4961     ldrq(v0, Address(buf, 0x10));
4962     eor3(v0, T16B, v17,  v18, v0);
4963 
4964     pmull (v19, T1Q, v1, v16, T1D);
4965     pmull2(v20, T1Q, v1, v16, T2D);
4966     ldrq(v1, Address(buf, 0x20));
4967     eor3(v1, T16B, v19, v20, v1);
4968 
4969     pmull (v21, T1Q, v2, v16, T1D);
4970     pmull2(v22, T1Q, v2, v16, T2D);
4971     ldrq(v2, Address(buf, 0x30));
4972     eor3(v2, T16B, v21, v22, v2);
4973 
4974     pmull (v23, T1Q, v3, v16, T1D);
4975     pmull2(v24, T1Q, v3, v16, T2D);
4976     ldrq(v3, Address(buf, 0x40));
4977     eor3(v3, T16B, v23, v24, v3);
4978 
4979     pmull (v25, T1Q, v4, v16, T1D);
4980     pmull2(v26, T1Q, v4, v16, T2D);
4981     ldrq(v4, Address(buf, 0x50));
4982     eor3(v4, T16B, v25, v26, v4);
4983 
4984     pmull (v27, T1Q, v5, v16, T1D);
4985     pmull2(v28, T1Q, v5, v16, T2D);
4986     ldrq(v5, Address(buf, 0x60));
4987     eor3(v5, T16B, v27, v28, v5);
4988 
4989     pmull (v29, T1Q, v6, v16, T1D);
4990     pmull2(v30, T1Q, v6, v16, T2D);
4991     ldrq(v6, Address(buf, 0x70));
4992     eor3(v6, T16B, v29, v30, v6);
4993 
4994     // Reuse registers v23, v24.
4995     // Using them won't block the first instruction of the next iteration.
4996     pmull (v23, T1Q, v7, v16, T1D);
4997     pmull2(v24, T1Q, v7, v16, T2D);
4998     ldrq(v7, Address(pre(buf, 0x80)));
4999     eor3(v7, T16B, v23, v24, v7);
5000 
5001     subs(len, len, 0x80);
5002     br(Assembler::GE, CRC_by128_loop);
5003 
5004     // fold into 512 bits
5005     // Use v31 for constants because v16 can be still in use.
5006     ldrq(v31, Address(table, 0x10));
5007 
5008     pmull (v17,  T1Q, v0, v31, T1D);
5009     pmull2(v18, T1Q, v0, v31, T2D);
5010     eor3(v0, T16B, v17, v18, v4);
5011 
5012     pmull (v19, T1Q, v1, v31, T1D);
5013     pmull2(v20, T1Q, v1, v31, T2D);
5014     eor3(v1, T16B, v19, v20, v5);
5015 
5016     pmull (v21, T1Q, v2, v31, T1D);
5017     pmull2(v22, T1Q, v2, v31, T2D);
5018     eor3(v2, T16B, v21, v22, v6);
5019 
5020     pmull (v23, T1Q, v3, v31, T1D);
5021     pmull2(v24, T1Q, v3, v31, T2D);
5022     eor3(v3, T16B, v23, v24, v7);
5023 
5024     // fold into 128 bits
5025     // Use v17 for constants because v31 can be still in use.
5026     ldrq(v17, Address(table, 0x20));
5027     pmull (v25, T1Q, v0, v17, T1D);
5028     pmull2(v26, T1Q, v0, v17, T2D);
5029     eor3(v3, T16B, v3, v25, v26);
5030 
5031     // Use v18 for constants because v17 can be still in use.
5032     ldrq(v18, Address(table, 0x30));
5033     pmull (v27, T1Q, v1, v18, T1D);
5034     pmull2(v28, T1Q, v1, v18, T2D);
5035     eor3(v3, T16B, v3, v27, v28);
5036 
5037     // Use v19 for constants because v18 can be still in use.
5038     ldrq(v19, Address(table, 0x40));
5039     pmull (v29, T1Q, v2, v19, T1D);
5040     pmull2(v30, T1Q, v2, v19, T2D);
5041     eor3(v0, T16B, v3, v29, v30);
5042 
5043     add(len, len, 0x80);
5044     add(buf, buf, 0x10);
5045 
5046     mov(tmp0, v0, D, 0);
5047     mov(tmp1, v0, D, 1);
5048 }
5049 
5050 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5051   Address adr;
5052   switch(dst.getMode()) {
5053   case Address::base_plus_offset:
5054     // This is the expected mode, although we allow all the other
5055     // forms below.
5056     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5057     break;
5058   default:
5059     lea(rscratch2, dst);
5060     adr = Address(rscratch2);
5061     break;
5062   }
5063   ldr(rscratch1, adr);
5064   add(rscratch1, rscratch1, src);
5065   str(rscratch1, adr);
5066 }
5067 
5068 void MacroAssembler::cmpptr(Register src1, Address src2) {
5069   uint64_t offset;
5070   adrp(rscratch1, src2, offset);
5071   ldr(rscratch1, Address(rscratch1, offset));
5072   cmp(src1, rscratch1);
5073 }
5074 
5075 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5076   cmp(obj1, obj2);
5077 }
5078 
5079 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5080   load_method_holder(rresult, rmethod);
5081   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5082 }
5083 
5084 void MacroAssembler::load_method_holder(Register holder, Register method) {
5085   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5086   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5087   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5088 }
5089 
5090 // Loads the obj's Klass* into dst.
5091 // Preserves all registers (incl src, rscratch1 and rscratch2).
5092 // Input:
5093 // src - the oop we want to load the klass from.
5094 // dst - output narrow klass.
5095 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5096   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5097   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5098   lsr(dst, dst, markWord::klass_shift);
5099 }
5100 
5101 void MacroAssembler::load_klass(Register dst, Register src) {
5102   if (UseCompactObjectHeaders) {
5103     load_narrow_klass_compact(dst, src);
5104   } else {
5105     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5106   }
5107   decode_klass_not_null(dst);
5108 }
5109 
5110 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5111   if (RestoreMXCSROnJNICalls) {
5112     Label OK;
5113     get_fpcr(tmp1);
5114     mov(tmp2, tmp1);
5115     // Set FPCR to the state we need. We do want Round to Nearest. We
5116     // don't want non-IEEE rounding modes or floating-point traps.
5117     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5118     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5119     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5120     eor(tmp2, tmp1, tmp2);
5121     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5122     set_fpcr(tmp1);
5123     bind(OK);
5124   }
5125 }
5126 
5127 // ((OopHandle)result).resolve();
5128 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5129   // OopHandle::resolve is an indirection.
5130   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5131 }
5132 
5133 // ((WeakHandle)result).resolve();
5134 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5135   assert_different_registers(result, tmp1, tmp2);
5136   Label resolved;
5137 
5138   // A null weak handle resolves to null.
5139   cbz(result, resolved);
5140 
5141   // Only 64 bit platforms support GCs that require a tmp register
5142   // WeakHandle::resolve is an indirection like jweak.
5143   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5144                  result, Address(result), tmp1, tmp2);
5145   bind(resolved);
5146 }
5147 
5148 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5149   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5150   ldr(dst, Address(rmethod, Method::const_offset()));
5151   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5152   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5153   ldr(dst, Address(dst, mirror_offset));
5154   resolve_oop_handle(dst, tmp1, tmp2);
5155 }
5156 
5157 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5158   assert_different_registers(obj, klass, tmp);
5159   if (UseCompactObjectHeaders) {
5160     load_narrow_klass_compact(tmp, obj);
5161   } else {
5162     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5163   }
5164   if (CompressedKlassPointers::base() == nullptr) {
5165     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5166     return;
5167   } else if (!AOTCodeCache::is_on_for_dump() &&
5168              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5169              && CompressedKlassPointers::shift() == 0) {
5170     // Only the bottom 32 bits matter
5171     cmpw(klass, tmp);
5172     return;
5173   }
5174   decode_klass_not_null(tmp);
5175   cmp(klass, tmp);
5176 }
5177 
5178 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5179   if (UseCompactObjectHeaders) {
5180     load_narrow_klass_compact(tmp1, obj1);
5181     load_narrow_klass_compact(tmp2,  obj2);
5182   } else {
5183     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5184     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5185   }
5186   cmpw(tmp1, tmp2);
5187 }
5188 
5189 void MacroAssembler::store_klass(Register dst, Register src) {
5190   // FIXME: Should this be a store release?  concurrent gcs assumes
5191   // klass length is valid if klass field is not null.
5192   assert(!UseCompactObjectHeaders, "not with compact headers");
5193   encode_klass_not_null(src);
5194   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5195 }
5196 
5197 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5198   assert(!UseCompactObjectHeaders, "not with compact headers");
5199   // Store to klass gap in destination
5200   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5201 }
5202 
5203 // Algorithm must match CompressedOops::encode.
5204 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5205 #ifdef ASSERT
5206   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5207 #endif
5208   verify_oop_msg(s, "broken oop in encode_heap_oop");
5209   if (CompressedOops::base() == nullptr) {
5210     if (CompressedOops::shift() != 0) {
5211       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5212       lsr(d, s, LogMinObjAlignmentInBytes);
5213     } else {
5214       mov(d, s);
5215     }
5216   } else {
5217     subs(d, s, rheapbase);
5218     csel(d, d, zr, Assembler::HS);
5219     lsr(d, d, LogMinObjAlignmentInBytes);
5220 
5221     /*  Old algorithm: is this any worse?
5222     Label nonnull;
5223     cbnz(r, nonnull);
5224     sub(r, r, rheapbase);
5225     bind(nonnull);
5226     lsr(r, r, LogMinObjAlignmentInBytes);
5227     */
5228   }
5229 }
5230 
5231 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5232 #ifdef ASSERT
5233   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5234   if (CheckCompressedOops) {
5235     Label ok;
5236     cbnz(r, ok);
5237     stop("null oop passed to encode_heap_oop_not_null");
5238     bind(ok);
5239   }
5240 #endif
5241   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5242   if (CompressedOops::base() != nullptr) {
5243     sub(r, r, rheapbase);
5244   }
5245   if (CompressedOops::shift() != 0) {
5246     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5247     lsr(r, r, LogMinObjAlignmentInBytes);
5248   }
5249 }
5250 
5251 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5252 #ifdef ASSERT
5253   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5254   if (CheckCompressedOops) {
5255     Label ok;
5256     cbnz(src, ok);
5257     stop("null oop passed to encode_heap_oop_not_null2");
5258     bind(ok);
5259   }
5260 #endif
5261   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5262 
5263   Register data = src;
5264   if (CompressedOops::base() != nullptr) {
5265     sub(dst, src, rheapbase);
5266     data = dst;
5267   }
5268   if (CompressedOops::shift() != 0) {
5269     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5270     lsr(dst, data, LogMinObjAlignmentInBytes);
5271     data = dst;
5272   }
5273   if (data == src)
5274     mov(dst, src);
5275 }
5276 
5277 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5278 #ifdef ASSERT
5279   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5280 #endif
5281   if (CompressedOops::base() == nullptr) {
5282     if (CompressedOops::shift() != 0) {
5283       lsl(d, s, CompressedOops::shift());
5284     } else if (d != s) {
5285       mov(d, s);
5286     }
5287   } else {
5288     Label done;
5289     if (d != s)
5290       mov(d, s);
5291     cbz(s, done);
5292     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5293     bind(done);
5294   }
5295   verify_oop_msg(d, "broken oop in decode_heap_oop");
5296 }
5297 
5298 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5299   assert (UseCompressedOops, "should only be used for compressed headers");
5300   assert (Universe::heap() != nullptr, "java heap should be initialized");
5301   // Cannot assert, unverified entry point counts instructions (see .ad file)
5302   // vtableStubs also counts instructions in pd_code_size_limit.
5303   // Also do not verify_oop as this is called by verify_oop.
5304   if (CompressedOops::shift() != 0) {
5305     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5306     if (CompressedOops::base() != nullptr) {
5307       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5308     } else {
5309       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5310     }
5311   } else {
5312     assert (CompressedOops::base() == nullptr, "sanity");
5313   }
5314 }
5315 
5316 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5317   assert (UseCompressedOops, "should only be used for compressed headers");
5318   assert (Universe::heap() != nullptr, "java heap should be initialized");
5319   // Cannot assert, unverified entry point counts instructions (see .ad file)
5320   // vtableStubs also counts instructions in pd_code_size_limit.
5321   // Also do not verify_oop as this is called by verify_oop.
5322   if (CompressedOops::shift() != 0) {
5323     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5324     if (CompressedOops::base() != nullptr) {
5325       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5326     } else {
5327       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5328     }
5329   } else {
5330     assert (CompressedOops::base() == nullptr, "sanity");
5331     if (dst != src) {
5332       mov(dst, src);
5333     }
5334   }
5335 }
5336 
5337 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5338 
5339 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5340   assert(Metaspace::initialized(), "metaspace not initialized yet");
5341   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5342   return _klass_decode_mode;
5343 }
5344 
5345 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5346   // KlassDecodeMode shouldn't be set already.
5347   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5348 
5349   if (base == nullptr) {
5350     return KlassDecodeZero;
5351   }
5352 
5353   if (operand_valid_for_logical_immediate(
5354         /*is32*/false, (uint64_t)base)) {
5355     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5356     if (((uint64_t)base & range_mask) == 0) {
5357       return KlassDecodeXor;
5358     }
5359   }
5360 
5361   const uint64_t shifted_base =
5362     (uint64_t)base >> shift;
5363   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5364     return KlassDecodeMovk;
5365   }
5366 
5367   // No valid encoding.
5368   return KlassDecodeNone;
5369 }
5370 
5371 // Check if one of the above decoding modes will work for given base, shift and range.
5372 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5373   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5374 }
5375 
5376 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5377   _klass_decode_mode = klass_decode_mode(base, shift, range);
5378   return _klass_decode_mode != KlassDecodeNone;
5379 }
5380 
5381 static Register pick_different_tmp(Register dst, Register src) {
5382   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5383   return *tmps.begin();
5384 }
5385 
5386 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5387   // we have to load the klass base from the AOT constants area but
5388   // not the shift because it is not allowed to change
5389   int shift = CompressedKlassPointers::shift();
5390   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5391   if (dst != src) {
5392     // we can load the base into dst, subtract it formthe src and shift down
5393     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5394     ldr(dst, dst);
5395     sub(dst, src, dst);
5396     lsr(dst, dst, 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     sub(dst, src, tmp);
5405     lsr(dst, dst, shift);
5406     pop(regs, sp);
5407   }
5408 }
5409 
5410 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5411   if (CompressedKlassPointers::base() != nullptr && AOTCodeCache::is_on_for_dump()) {
5412     encode_klass_not_null_for_aot(dst, src);
5413     return;
5414   }
5415 
5416   switch (klass_decode_mode()) {
5417   case KlassDecodeZero:
5418     if (CompressedKlassPointers::shift() != 0) {
5419       lsr(dst, src, CompressedKlassPointers::shift());
5420     } else {
5421       if (dst != src) mov(dst, src);
5422     }
5423     break;
5424 
5425   case KlassDecodeXor:
5426     if (CompressedKlassPointers::shift() != 0) {
5427       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5428       lsr(dst, dst, CompressedKlassPointers::shift());
5429     } else {
5430       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5431     }
5432     break;
5433 
5434   case KlassDecodeMovk:
5435     if (CompressedKlassPointers::shift() != 0) {
5436       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5437     } else {
5438       movw(dst, src);
5439     }
5440     break;
5441 
5442   case KlassDecodeNone:
5443     ShouldNotReachHere();
5444     break;
5445   }
5446 }
5447 
5448 void MacroAssembler::encode_klass_not_null(Register r) {
5449   encode_klass_not_null(r, r);
5450 }
5451 
5452 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5453   // we have to load the klass base from the AOT constants area but
5454   // not the shift because it is not allowed to change
5455   int shift = CompressedKlassPointers::shift();
5456   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5457   if (dst != src) {
5458     // we can load the base into dst then add the offset with a suitable shift
5459     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5460     ldr(dst, dst);
5461     add(dst, dst, src, LSL,  shift);
5462   } else {
5463     // we need an extra register in order to load the coop base
5464     Register tmp = pick_different_tmp(dst, src);
5465     RegSet regs = RegSet::of(tmp);
5466     push(regs, sp);
5467     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5468     ldr(tmp, tmp);
5469     add(dst, tmp,  src, LSL,  shift);
5470     pop(regs, sp);
5471   }
5472 }
5473 
5474 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5475   if (AOTCodeCache::is_on_for_dump()) {
5476     decode_klass_not_null_for_aot(dst, src);
5477     return;
5478   }
5479 
5480   switch (klass_decode_mode()) {
5481   case KlassDecodeZero:
5482     if (CompressedKlassPointers::shift() != 0) {
5483       lsl(dst, src, CompressedKlassPointers::shift());
5484     } else {
5485       if (dst != src) mov(dst, src);
5486     }
5487     break;
5488 
5489   case KlassDecodeXor:
5490     if (CompressedKlassPointers::shift() != 0) {
5491       lsl(dst, src, CompressedKlassPointers::shift());
5492       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5493     } else {
5494       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5495     }
5496     break;
5497 
5498   case KlassDecodeMovk: {
5499     const uint64_t shifted_base =
5500       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5501 
5502     if (dst != src) movw(dst, src);
5503     movk(dst, shifted_base >> 32, 32);
5504 
5505     if (CompressedKlassPointers::shift() != 0) {
5506       lsl(dst, dst, CompressedKlassPointers::shift());
5507     }
5508 
5509     break;
5510   }
5511 
5512   case KlassDecodeNone:
5513     ShouldNotReachHere();
5514     break;
5515   }
5516 }
5517 
5518 void  MacroAssembler::decode_klass_not_null(Register r) {
5519   decode_klass_not_null(r, r);
5520 }
5521 
5522 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5523 #ifdef ASSERT
5524   {
5525     ThreadInVMfromUnknown tiv;
5526     assert (UseCompressedOops, "should only be used for compressed oops");
5527     assert (Universe::heap() != nullptr, "java heap should be initialized");
5528     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5529     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5530   }
5531 #endif
5532   int oop_index = oop_recorder()->find_index(obj);
5533   InstructionMark im(this);
5534   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5535   code_section()->relocate(inst_mark(), rspec);
5536   movz(dst, 0xDEAD, 16);
5537   movk(dst, 0xBEEF);
5538 }
5539 
5540 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5541   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5542   int index = oop_recorder()->find_index(k);
5543 
5544   InstructionMark im(this);
5545   RelocationHolder rspec = metadata_Relocation::spec(index);
5546   code_section()->relocate(inst_mark(), rspec);
5547   narrowKlass nk = CompressedKlassPointers::encode(k);
5548   movz(dst, (nk >> 16), 16);
5549   movk(dst, nk & 0xffff);
5550 }
5551 
5552 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5553                                     Register dst, Address src,
5554                                     Register tmp1, Register tmp2) {
5555   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5556   decorators = AccessInternal::decorator_fixup(decorators, type);
5557   bool as_raw = (decorators & AS_RAW) != 0;
5558   if (as_raw) {
5559     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5560   } else {
5561     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5562   }
5563 }
5564 
5565 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5566                                      Address dst, Register val,
5567                                      Register tmp1, Register tmp2, Register tmp3) {
5568   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5569   decorators = AccessInternal::decorator_fixup(decorators, type);
5570   bool as_raw = (decorators & AS_RAW) != 0;
5571   if (as_raw) {
5572     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5573   } else {
5574     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5575   }
5576 }
5577 
5578 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5579                                    Register tmp2, DecoratorSet decorators) {
5580   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5581 }
5582 
5583 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5584                                             Register tmp2, DecoratorSet decorators) {
5585   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5586 }
5587 
5588 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5589                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5590   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5591 }
5592 
5593 // Used for storing nulls.
5594 void MacroAssembler::store_heap_oop_null(Address dst) {
5595   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5596 }
5597 
5598 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5599   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5600   int index = oop_recorder()->allocate_metadata_index(obj);
5601   RelocationHolder rspec = metadata_Relocation::spec(index);
5602   return Address((address)obj, rspec);
5603 }
5604 
5605 // Move an oop into a register.
5606 void MacroAssembler::movoop(Register dst, jobject obj) {
5607   int oop_index;
5608   if (obj == nullptr) {
5609     oop_index = oop_recorder()->allocate_oop_index(obj);
5610   } else {
5611 #ifdef ASSERT
5612     {
5613       ThreadInVMfromUnknown tiv;
5614       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5615     }
5616 #endif
5617     oop_index = oop_recorder()->find_index(obj);
5618   }
5619   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5620 
5621   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5622     mov(dst, Address((address)obj, rspec));
5623   } else {
5624     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5625     ldr(dst, Address(dummy, rspec));
5626   }
5627 }
5628 
5629 // Move a metadata address into a register.
5630 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5631   int oop_index;
5632   if (obj == nullptr) {
5633     oop_index = oop_recorder()->allocate_metadata_index(obj);
5634   } else {
5635     oop_index = oop_recorder()->find_index(obj);
5636   }
5637   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5638   mov(dst, Address((address)obj, rspec));
5639 }
5640 
5641 Address MacroAssembler::constant_oop_address(jobject obj) {
5642 #ifdef ASSERT
5643   {
5644     ThreadInVMfromUnknown tiv;
5645     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5646     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5647   }
5648 #endif
5649   int oop_index = oop_recorder()->find_index(obj);
5650   return Address((address)obj, oop_Relocation::spec(oop_index));
5651 }
5652 
5653 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5654 void MacroAssembler::tlab_allocate(Register obj,
5655                                    Register var_size_in_bytes,
5656                                    int con_size_in_bytes,
5657                                    Register t1,
5658                                    Register t2,
5659                                    Label& slow_case) {
5660   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5661   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5662 }
5663 
5664 void MacroAssembler::verify_tlab() {
5665 #ifdef ASSERT
5666   if (UseTLAB && VerifyOops) {
5667     Label next, ok;
5668 
5669     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5670 
5671     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5672     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5673     cmp(rscratch2, rscratch1);
5674     br(Assembler::HS, next);
5675     STOP("assert(top >= start)");
5676     should_not_reach_here();
5677 
5678     bind(next);
5679     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5680     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5681     cmp(rscratch2, rscratch1);
5682     br(Assembler::HS, ok);
5683     STOP("assert(top <= end)");
5684     should_not_reach_here();
5685 
5686     bind(ok);
5687     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5688   }
5689 #endif
5690 }
5691 
5692 // Writes to stack successive pages until offset reached to check for
5693 // stack overflow + shadow pages.  This clobbers tmp.
5694 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5695   assert_different_registers(tmp, size, rscratch1);
5696   mov(tmp, sp);
5697   // Bang stack for total size given plus shadow page size.
5698   // Bang one page at a time because large size can bang beyond yellow and
5699   // red zones.
5700   Label loop;
5701   mov(rscratch1, (int)os::vm_page_size());
5702   bind(loop);
5703   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5704   subsw(size, size, rscratch1);
5705   str(size, Address(tmp));
5706   br(Assembler::GT, loop);
5707 
5708   // Bang down shadow pages too.
5709   // At this point, (tmp-0) is the last address touched, so don't
5710   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5711   // was post-decremented.)  Skip this address by starting at i=1, and
5712   // touch a few more pages below.  N.B.  It is important to touch all
5713   // the way down to and including i=StackShadowPages.
5714   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5715     // this could be any sized move but this is can be a debugging crumb
5716     // so the bigger the better.
5717     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5718     str(size, Address(tmp));
5719   }
5720 }
5721 
5722 // Move the address of the polling page into dest.
5723 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5724   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5725 }
5726 
5727 // Read the polling page.  The address of the polling page must
5728 // already be in r.
5729 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5730   address mark;
5731   {
5732     InstructionMark im(this);
5733     code_section()->relocate(inst_mark(), rtype);
5734     ldrw(zr, Address(r, 0));
5735     mark = inst_mark();
5736   }
5737   verify_cross_modify_fence_not_required();
5738   return mark;
5739 }
5740 
5741 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5742   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5743   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5744   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5745   int64_t offset_low = dest_page - low_page;
5746   int64_t offset_high = dest_page - high_page;
5747 
5748   assert(is_valid_AArch64_address(dest.target()), "bad address");
5749   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5750 
5751   InstructionMark im(this);
5752   code_section()->relocate(inst_mark(), dest.rspec());
5753   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5754   // the code cache so that if it is relocated we know it will still reach
5755   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5756     _adrp(reg1, dest.target());
5757   } else {
5758     uint64_t target = (uint64_t)dest.target();
5759     uint64_t adrp_target
5760       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5761 
5762     _adrp(reg1, (address)adrp_target);
5763     movk(reg1, target >> 32, 32);
5764   }
5765   byte_offset = (uint64_t)dest.target() & 0xfff;
5766 }
5767 
5768 void MacroAssembler::load_byte_map_base(Register reg) {
5769 #if INCLUDE_CDS
5770   if (AOTCodeCache::is_on_for_dump()) {
5771     address byte_map_base_adr = AOTRuntimeConstants::card_table_base_address();
5772     lea(reg, ExternalAddress(byte_map_base_adr));
5773     ldr(reg, Address(reg));
5774     return;
5775   }
5776 #endif
5777   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
5778 
5779   // Strictly speaking the card table base isn't an address at all, and it might
5780   // even be negative. It is thus materialised as a constant.
5781   mov(reg, (uint64_t)ctbs->card_table_base_const());
5782 }
5783 
5784 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5785 #if INCLUDE_CDS
5786   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5787   if (AOTCodeCache::is_on_for_dump()) {
5788     // all aotrc field addresses should be registered in the AOTCodeCache address table
5789     lea(reg, ExternalAddress(a));
5790   } else {
5791     mov(reg, (uint64_t)a);
5792   }
5793 #else
5794   ShouldNotReachHere();
5795 #endif
5796 }
5797 
5798 void MacroAssembler::build_frame(int framesize) {
5799   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5800   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5801   protect_return_address();
5802   if (framesize < ((1 << 9) + 2 * wordSize)) {
5803     sub(sp, sp, framesize);
5804     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5805     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5806   } else {
5807     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5808     if (PreserveFramePointer) mov(rfp, sp);
5809     if (framesize < ((1 << 12) + 2 * wordSize))
5810       sub(sp, sp, framesize - 2 * wordSize);
5811     else {
5812       mov(rscratch1, framesize - 2 * wordSize);
5813       sub(sp, sp, rscratch1);
5814     }
5815   }
5816   verify_cross_modify_fence_not_required();
5817 }
5818 
5819 void MacroAssembler::remove_frame(int framesize) {
5820   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5821   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5822   if (framesize < ((1 << 9) + 2 * wordSize)) {
5823     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5824     add(sp, sp, framesize);
5825   } else {
5826     if (framesize < ((1 << 12) + 2 * wordSize))
5827       add(sp, sp, framesize - 2 * wordSize);
5828     else {
5829       mov(rscratch1, framesize - 2 * wordSize);
5830       add(sp, sp, rscratch1);
5831     }
5832     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5833   }
5834   authenticate_return_address();
5835 }
5836 
5837 
5838 // This method counts leading positive bytes (highest bit not set) in provided byte array
5839 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5840     // Simple and most common case of aligned small array which is not at the
5841     // end of memory page is placed here. All other cases are in stub.
5842     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5843     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5844     assert_different_registers(ary1, len, result);
5845 
5846     mov(result, len);
5847     cmpw(len, 0);
5848     br(LE, DONE);
5849     cmpw(len, 4 * wordSize);
5850     br(GE, STUB_LONG); // size > 32 then go to stub
5851 
5852     int shift = 64 - exact_log2(os::vm_page_size());
5853     lsl(rscratch1, ary1, shift);
5854     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5855     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5856     br(CS, STUB); // at the end of page then go to stub
5857     subs(len, len, wordSize);
5858     br(LT, END);
5859 
5860   BIND(LOOP);
5861     ldr(rscratch1, Address(post(ary1, wordSize)));
5862     tst(rscratch1, UPPER_BIT_MASK);
5863     br(NE, SET_RESULT);
5864     subs(len, len, wordSize);
5865     br(GE, LOOP);
5866     cmpw(len, -wordSize);
5867     br(EQ, DONE);
5868 
5869   BIND(END);
5870     ldr(rscratch1, Address(ary1));
5871     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5872     lslv(rscratch1, rscratch1, rscratch2);
5873     tst(rscratch1, UPPER_BIT_MASK);
5874     br(NE, SET_RESULT);
5875     b(DONE);
5876 
5877   BIND(STUB);
5878     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5879     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5880     address tpc1 = trampoline_call(count_pos);
5881     if (tpc1 == nullptr) {
5882       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5883       postcond(pc() == badAddress);
5884       return nullptr;
5885     }
5886     b(DONE);
5887 
5888   BIND(STUB_LONG);
5889     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5890     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5891     address tpc2 = trampoline_call(count_pos_long);
5892     if (tpc2 == nullptr) {
5893       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5894       postcond(pc() == badAddress);
5895       return nullptr;
5896     }
5897     b(DONE);
5898 
5899   BIND(SET_RESULT);
5900 
5901     add(len, len, wordSize);
5902     sub(result, result, len);
5903 
5904   BIND(DONE);
5905   postcond(pc() != badAddress);
5906   return pc();
5907 }
5908 
5909 // Clobbers: rscratch1, rscratch2, rflags
5910 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5911 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5912                                       Register tmp4, Register tmp5, Register result,
5913                                       Register cnt1, int elem_size) {
5914   Label DONE, SAME;
5915   Register tmp1 = rscratch1;
5916   Register tmp2 = rscratch2;
5917   int elem_per_word = wordSize/elem_size;
5918   int log_elem_size = exact_log2(elem_size);
5919   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5920   int length_offset = arrayOopDesc::length_offset_in_bytes();
5921   int base_offset
5922     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5923   // When the length offset is not aligned to 8 bytes,
5924   // then we align it down. This is valid because the new
5925   // offset will always be the klass which is the same
5926   // for type arrays.
5927   int start_offset = align_down(length_offset, BytesPerWord);
5928   int extra_length = base_offset - start_offset;
5929   assert(start_offset == length_offset || start_offset == klass_offset,
5930          "start offset must be 8-byte-aligned or be the klass offset");
5931   assert(base_offset != start_offset, "must include the length field");
5932   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5933   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5934 
5935   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5936   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5937 
5938 #ifndef PRODUCT
5939   {
5940     const char kind = (elem_size == 2) ? 'U' : 'L';
5941     char comment[64];
5942     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5943     BLOCK_COMMENT(comment);
5944   }
5945 #endif
5946 
5947   // if (a1 == a2)
5948   //     return true;
5949   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5950   br(EQ, SAME);
5951 
5952   if (UseSimpleArrayEquals) {
5953     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5954     // if (a1 == nullptr || a2 == nullptr)
5955     //     return false;
5956     // a1 & a2 == 0 means (some-pointer is null) or
5957     // (very-rare-or-even-probably-impossible-pointer-values)
5958     // so, we can save one branch in most cases
5959     tst(a1, a2);
5960     mov(result, false);
5961     br(EQ, A_MIGHT_BE_NULL);
5962     // if (a1.length != a2.length)
5963     //      return false;
5964     bind(A_IS_NOT_NULL);
5965     ldrw(cnt1, Address(a1, length_offset));
5966     ldrw(tmp5, Address(a2, length_offset));
5967     cmp(cnt1, tmp5);
5968     br(NE, DONE); // If lengths differ, return false
5969     // Increase loop counter by diff between base- and actual start-offset.
5970     addw(cnt1, cnt1, extra_length);
5971     lea(a1, Address(a1, start_offset));
5972     lea(a2, Address(a2, start_offset));
5973     // Check for short strings, i.e. smaller than wordSize.
5974     subs(cnt1, cnt1, elem_per_word);
5975     br(Assembler::LT, SHORT);
5976     // Main 8 byte comparison loop.
5977     bind(NEXT_WORD); {
5978       ldr(tmp1, Address(post(a1, wordSize)));
5979       ldr(tmp2, Address(post(a2, wordSize)));
5980       subs(cnt1, cnt1, elem_per_word);
5981       eor(tmp5, tmp1, tmp2);
5982       cbnz(tmp5, DONE);
5983     } br(GT, NEXT_WORD);
5984     // Last longword.  In the case where length == 4 we compare the
5985     // same longword twice, but that's still faster than another
5986     // conditional branch.
5987     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5988     // length == 4.
5989     if (log_elem_size > 0)
5990       lsl(cnt1, cnt1, log_elem_size);
5991     ldr(tmp3, Address(a1, cnt1));
5992     ldr(tmp4, Address(a2, cnt1));
5993     eor(tmp5, tmp3, tmp4);
5994     cbnz(tmp5, DONE);
5995     b(SAME);
5996     bind(A_MIGHT_BE_NULL);
5997     // in case both a1 and a2 are not-null, proceed with loads
5998     cbz(a1, DONE);
5999     cbz(a2, DONE);
6000     b(A_IS_NOT_NULL);
6001     bind(SHORT);
6002 
6003     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
6004     {
6005       ldrw(tmp1, Address(post(a1, 4)));
6006       ldrw(tmp2, Address(post(a2, 4)));
6007       eorw(tmp5, tmp1, tmp2);
6008       cbnzw(tmp5, DONE);
6009     }
6010     bind(TAIL03);
6011     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6012     {
6013       ldrh(tmp3, Address(post(a1, 2)));
6014       ldrh(tmp4, Address(post(a2, 2)));
6015       eorw(tmp5, tmp3, tmp4);
6016       cbnzw(tmp5, DONE);
6017     }
6018     bind(TAIL01);
6019     if (elem_size == 1) { // Only needed when comparing byte arrays.
6020       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6021       {
6022         ldrb(tmp1, a1);
6023         ldrb(tmp2, a2);
6024         eorw(tmp5, tmp1, tmp2);
6025         cbnzw(tmp5, DONE);
6026       }
6027     }
6028   } else {
6029     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6030         CSET_EQ, LAST_CHECK;
6031     mov(result, false);
6032     cbz(a1, DONE);
6033     ldrw(cnt1, Address(a1, length_offset));
6034     cbz(a2, DONE);
6035     ldrw(tmp5, Address(a2, length_offset));
6036     cmp(cnt1, tmp5);
6037     br(NE, DONE); // If lengths differ, return false
6038     // Increase loop counter by diff between base- and actual start-offset.
6039     addw(cnt1, cnt1, extra_length);
6040 
6041     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6042     // faster to perform another branch before comparing a1 and a2
6043     cmp(cnt1, (u1)elem_per_word);
6044     br(LE, SHORT); // short or same
6045     ldr(tmp3, Address(pre(a1, start_offset)));
6046     subs(zr, cnt1, stubBytesThreshold);
6047     br(GE, STUB);
6048     ldr(tmp4, Address(pre(a2, start_offset)));
6049     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6050 
6051     // Main 16 byte comparison loop with 2 exits
6052     bind(NEXT_DWORD); {
6053       ldr(tmp1, Address(pre(a1, wordSize)));
6054       ldr(tmp2, Address(pre(a2, wordSize)));
6055       subs(cnt1, cnt1, 2 * elem_per_word);
6056       br(LE, TAIL);
6057       eor(tmp4, tmp3, tmp4);
6058       cbnz(tmp4, DONE);
6059       ldr(tmp3, Address(pre(a1, wordSize)));
6060       ldr(tmp4, Address(pre(a2, wordSize)));
6061       cmp(cnt1, (u1)elem_per_word);
6062       br(LE, TAIL2);
6063       cmp(tmp1, tmp2);
6064     } br(EQ, NEXT_DWORD);
6065     b(DONE);
6066 
6067     bind(TAIL);
6068     eor(tmp4, tmp3, tmp4);
6069     eor(tmp2, tmp1, tmp2);
6070     lslv(tmp2, tmp2, tmp5);
6071     orr(tmp5, tmp4, tmp2);
6072     cmp(tmp5, zr);
6073     b(CSET_EQ);
6074 
6075     bind(TAIL2);
6076     eor(tmp2, tmp1, tmp2);
6077     cbnz(tmp2, DONE);
6078     b(LAST_CHECK);
6079 
6080     bind(STUB);
6081     ldr(tmp4, Address(pre(a2, start_offset)));
6082     if (elem_size == 2) { // convert to byte counter
6083       lsl(cnt1, cnt1, 1);
6084     }
6085     eor(tmp5, tmp3, tmp4);
6086     cbnz(tmp5, DONE);
6087     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6088     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6089     address tpc = trampoline_call(stub);
6090     if (tpc == nullptr) {
6091       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6092       postcond(pc() == badAddress);
6093       return nullptr;
6094     }
6095     b(DONE);
6096 
6097     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6098     // so, if a2 == null => return false(0), else return true, so we can return a2
6099     mov(result, a2);
6100     b(DONE);
6101     bind(SHORT);
6102     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6103     ldr(tmp3, Address(a1, start_offset));
6104     ldr(tmp4, Address(a2, start_offset));
6105     bind(LAST_CHECK);
6106     eor(tmp4, tmp3, tmp4);
6107     lslv(tmp5, tmp4, tmp5);
6108     cmp(tmp5, zr);
6109     bind(CSET_EQ);
6110     cset(result, EQ);
6111     b(DONE);
6112   }
6113 
6114   bind(SAME);
6115   mov(result, true);
6116   // That's it.
6117   bind(DONE);
6118 
6119   BLOCK_COMMENT("} array_equals");
6120   postcond(pc() != badAddress);
6121   return pc();
6122 }
6123 
6124 // Compare Strings
6125 
6126 // For Strings we're passed the address of the first characters in a1
6127 // and a2 and the length in cnt1.
6128 // There are two implementations.  For arrays >= 8 bytes, all
6129 // comparisons (including the final one, which may overlap) are
6130 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6131 // halfword, then a short, and then a byte.
6132 
6133 void MacroAssembler::string_equals(Register a1, Register a2,
6134                                    Register result, Register cnt1)
6135 {
6136   Label SAME, DONE, SHORT, NEXT_WORD;
6137   Register tmp1 = rscratch1;
6138   Register tmp2 = rscratch2;
6139 
6140   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6141 
6142 #ifndef PRODUCT
6143   {
6144     char comment[64];
6145     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6146     BLOCK_COMMENT(comment);
6147   }
6148 #endif
6149 
6150   mov(result, false);
6151 
6152   // Check for short strings, i.e. smaller than wordSize.
6153   subs(cnt1, cnt1, wordSize);
6154   br(Assembler::LT, SHORT);
6155   // Main 8 byte comparison loop.
6156   bind(NEXT_WORD); {
6157     ldr(tmp1, Address(post(a1, wordSize)));
6158     ldr(tmp2, Address(post(a2, wordSize)));
6159     subs(cnt1, cnt1, wordSize);
6160     eor(tmp1, tmp1, tmp2);
6161     cbnz(tmp1, DONE);
6162   } br(GT, NEXT_WORD);
6163   // Last longword.  In the case where length == 4 we compare the
6164   // same longword twice, but that's still faster than another
6165   // conditional branch.
6166   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6167   // length == 4.
6168   ldr(tmp1, Address(a1, cnt1));
6169   ldr(tmp2, Address(a2, cnt1));
6170   eor(tmp2, tmp1, tmp2);
6171   cbnz(tmp2, DONE);
6172   b(SAME);
6173 
6174   bind(SHORT);
6175   Label TAIL03, TAIL01;
6176 
6177   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6178   {
6179     ldrw(tmp1, Address(post(a1, 4)));
6180     ldrw(tmp2, Address(post(a2, 4)));
6181     eorw(tmp1, tmp1, tmp2);
6182     cbnzw(tmp1, DONE);
6183   }
6184   bind(TAIL03);
6185   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6186   {
6187     ldrh(tmp1, Address(post(a1, 2)));
6188     ldrh(tmp2, Address(post(a2, 2)));
6189     eorw(tmp1, tmp1, tmp2);
6190     cbnzw(tmp1, DONE);
6191   }
6192   bind(TAIL01);
6193   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6194     {
6195     ldrb(tmp1, a1);
6196     ldrb(tmp2, a2);
6197     eorw(tmp1, tmp1, tmp2);
6198     cbnzw(tmp1, DONE);
6199   }
6200   // Arrays are equal.
6201   bind(SAME);
6202   mov(result, true);
6203 
6204   // That's it.
6205   bind(DONE);
6206   BLOCK_COMMENT("} string_equals");
6207 }
6208 
6209 
6210 // The size of the blocks erased by the zero_blocks stub.  We must
6211 // handle anything smaller than this ourselves in zero_words().
6212 const int MacroAssembler::zero_words_block_size = 8;
6213 
6214 // zero_words() is used by C2 ClearArray patterns and by
6215 // C1_MacroAssembler.  It is as small as possible, handling small word
6216 // counts locally and delegating anything larger to the zero_blocks
6217 // stub.  It is expanded many times in compiled code, so it is
6218 // important to keep it short.
6219 
6220 // ptr:   Address of a buffer to be zeroed.
6221 // cnt:   Count in HeapWords.
6222 //
6223 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6224 address MacroAssembler::zero_words(Register ptr, Register cnt)
6225 {
6226   assert(is_power_of_2(zero_words_block_size), "adjust this");
6227 
6228   BLOCK_COMMENT("zero_words {");
6229   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6230   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6231   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6232 
6233   subs(rscratch1, cnt, zero_words_block_size);
6234   Label around;
6235   br(LO, around);
6236   {
6237     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6238     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6239     // Make sure this is a C2 compilation. C1 allocates space only for
6240     // trampoline stubs generated by Call LIR ops, and in any case it
6241     // makes sense for a C1 compilation task to proceed as quickly as
6242     // possible.
6243     CompileTask* task;
6244     if (StubRoutines::aarch64::complete()
6245         && Thread::current()->is_Compiler_thread()
6246         && (task = ciEnv::current()->task())
6247         && is_c2_compile(task->comp_level())) {
6248       address tpc = trampoline_call(zero_blocks);
6249       if (tpc == nullptr) {
6250         DEBUG_ONLY(reset_labels(around));
6251         return nullptr;
6252       }
6253     } else {
6254       far_call(zero_blocks);
6255     }
6256   }
6257   bind(around);
6258 
6259   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6260   // for us.
6261   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6262     Label l;
6263     tbz(cnt, exact_log2(i), l);
6264     for (int j = 0; j < i; j += 2) {
6265       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6266     }
6267     bind(l);
6268   }
6269   {
6270     Label l;
6271     tbz(cnt, 0, l);
6272     str(zr, Address(ptr));
6273     bind(l);
6274   }
6275 
6276   BLOCK_COMMENT("} zero_words");
6277   return pc();
6278 }
6279 
6280 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6281 // cnt:          Immediate count in HeapWords.
6282 //
6283 // r10, r11, rscratch1, and rscratch2 are clobbered.
6284 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6285 {
6286   assert(wordSize <= BlockZeroingLowLimit,
6287             "increase BlockZeroingLowLimit");
6288   address result = nullptr;
6289   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6290 #ifndef PRODUCT
6291     {
6292       char buf[64];
6293       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6294       BLOCK_COMMENT(buf);
6295     }
6296 #endif
6297     if (cnt >= 16) {
6298       uint64_t loops = cnt/16;
6299       if (loops > 1) {
6300         mov(rscratch2, loops - 1);
6301       }
6302       {
6303         Label loop;
6304         bind(loop);
6305         for (int i = 0; i < 16; i += 2) {
6306           stp(zr, zr, Address(base, i * BytesPerWord));
6307         }
6308         add(base, base, 16 * BytesPerWord);
6309         if (loops > 1) {
6310           subs(rscratch2, rscratch2, 1);
6311           br(GE, loop);
6312         }
6313       }
6314     }
6315     cnt %= 16;
6316     int i = cnt & 1;  // store any odd word to start
6317     if (i) str(zr, Address(base));
6318     for (; i < (int)cnt; i += 2) {
6319       stp(zr, zr, Address(base, i * wordSize));
6320     }
6321     BLOCK_COMMENT("} zero_words");
6322     result = pc();
6323   } else {
6324     mov(r10, base); mov(r11, cnt);
6325     result = zero_words(r10, r11);
6326   }
6327   return result;
6328 }
6329 
6330 // Zero blocks of memory by using DC ZVA.
6331 //
6332 // Aligns the base address first sufficiently for DC ZVA, then uses
6333 // DC ZVA repeatedly for every full block.  cnt is the size to be
6334 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6335 // in cnt.
6336 //
6337 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6338 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6339 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6340   Register tmp = rscratch1;
6341   Register tmp2 = rscratch2;
6342   int zva_length = VM_Version::zva_length();
6343   Label initial_table_end, loop_zva;
6344   Label fini;
6345 
6346   // Base must be 16 byte aligned. If not just return and let caller handle it
6347   tst(base, 0x0f);
6348   br(Assembler::NE, fini);
6349   // Align base with ZVA length.
6350   neg(tmp, base);
6351   andr(tmp, tmp, zva_length - 1);
6352 
6353   // tmp: the number of bytes to be filled to align the base with ZVA length.
6354   add(base, base, tmp);
6355   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6356   adr(tmp2, initial_table_end);
6357   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6358   br(tmp2);
6359 
6360   for (int i = -zva_length + 16; i < 0; i += 16)
6361     stp(zr, zr, Address(base, i));
6362   bind(initial_table_end);
6363 
6364   sub(cnt, cnt, zva_length >> 3);
6365   bind(loop_zva);
6366   dc(Assembler::ZVA, base);
6367   subs(cnt, cnt, zva_length >> 3);
6368   add(base, base, zva_length);
6369   br(Assembler::GE, loop_zva);
6370   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6371   bind(fini);
6372 }
6373 
6374 // base:   Address of a buffer to be filled, 8 bytes aligned.
6375 // cnt:    Count in 8-byte unit.
6376 // value:  Value to be filled with.
6377 // base will point to the end of the buffer after filling.
6378 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6379 {
6380 //  Algorithm:
6381 //
6382 //    if (cnt == 0) {
6383 //      return;
6384 //    }
6385 //    if ((p & 8) != 0) {
6386 //      *p++ = v;
6387 //    }
6388 //
6389 //    scratch1 = cnt & 14;
6390 //    cnt -= scratch1;
6391 //    p += scratch1;
6392 //    switch (scratch1 / 2) {
6393 //      do {
6394 //        cnt -= 16;
6395 //          p[-16] = v;
6396 //          p[-15] = v;
6397 //        case 7:
6398 //          p[-14] = v;
6399 //          p[-13] = v;
6400 //        case 6:
6401 //          p[-12] = v;
6402 //          p[-11] = v;
6403 //          // ...
6404 //        case 1:
6405 //          p[-2] = v;
6406 //          p[-1] = v;
6407 //        case 0:
6408 //          p += 16;
6409 //      } while (cnt);
6410 //    }
6411 //    if ((cnt & 1) == 1) {
6412 //      *p++ = v;
6413 //    }
6414 
6415   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6416 
6417   Label fini, skip, entry, loop;
6418   const int unroll = 8; // Number of stp instructions we'll unroll
6419 
6420   cbz(cnt, fini);
6421   tbz(base, 3, skip);
6422   str(value, Address(post(base, 8)));
6423   sub(cnt, cnt, 1);
6424   bind(skip);
6425 
6426   andr(rscratch1, cnt, (unroll-1) * 2);
6427   sub(cnt, cnt, rscratch1);
6428   add(base, base, rscratch1, Assembler::LSL, 3);
6429   adr(rscratch2, entry);
6430   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6431   br(rscratch2);
6432 
6433   bind(loop);
6434   add(base, base, unroll * 16);
6435   for (int i = -unroll; i < 0; i++)
6436     stp(value, value, Address(base, i * 16));
6437   bind(entry);
6438   subs(cnt, cnt, unroll * 2);
6439   br(Assembler::GE, loop);
6440 
6441   tbz(cnt, 0, fini);
6442   str(value, Address(post(base, 8)));
6443   bind(fini);
6444 }
6445 
6446 // Intrinsic for
6447 //
6448 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6449 //   Encodes char[] to byte[] in ISO-8859-1
6450 //
6451 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6452 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6453 //
6454 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6455 //   Encodes char[] to byte[] in ASCII
6456 //
6457 // This version always returns the number of characters copied, and does not
6458 // clobber the 'len' register. A successful copy will complete with the post-
6459 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6460 // post-condition: 0 <= 'res' < 'len'.
6461 //
6462 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6463 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6464 //       beyond the acceptable, even though the footprint would be smaller.
6465 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6466 //       avoid additional bloat.
6467 //
6468 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6469 void MacroAssembler::encode_iso_array(Register src, Register dst,
6470                                       Register len, Register res, bool ascii,
6471                                       FloatRegister vtmp0, FloatRegister vtmp1,
6472                                       FloatRegister vtmp2, FloatRegister vtmp3,
6473                                       FloatRegister vtmp4, FloatRegister vtmp5)
6474 {
6475   Register cnt = res;
6476   Register max = rscratch1;
6477   Register chk = rscratch2;
6478 
6479   prfm(Address(src), PLDL1STRM);
6480   movw(cnt, len);
6481 
6482 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6483 
6484   Label LOOP_32, DONE_32, FAIL_32;
6485 
6486   BIND(LOOP_32);
6487   {
6488     cmpw(cnt, 32);
6489     br(LT, DONE_32);
6490     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6491     // Extract lower bytes.
6492     FloatRegister vlo0 = vtmp4;
6493     FloatRegister vlo1 = vtmp5;
6494     uzp1(vlo0, T16B, vtmp0, vtmp1);
6495     uzp1(vlo1, T16B, vtmp2, vtmp3);
6496     // Merge bits...
6497     orr(vtmp0, T16B, vtmp0, vtmp1);
6498     orr(vtmp2, T16B, vtmp2, vtmp3);
6499     // Extract merged upper bytes.
6500     FloatRegister vhix = vtmp0;
6501     uzp2(vhix, T16B, vtmp0, vtmp2);
6502     // ISO-check on hi-parts (all zero).
6503     //                          ASCII-check on lo-parts (no sign).
6504     FloatRegister vlox = vtmp1; // Merge lower bytes.
6505                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6506     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6507     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6508     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6509                                 ASCII(orr(chk, chk, max));
6510     cbnz(chk, FAIL_32);
6511     subw(cnt, cnt, 32);
6512     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6513     b(LOOP_32);
6514   }
6515   BIND(FAIL_32);
6516   sub(src, src, 64);
6517   BIND(DONE_32);
6518 
6519   Label LOOP_8, SKIP_8;
6520 
6521   BIND(LOOP_8);
6522   {
6523     cmpw(cnt, 8);
6524     br(LT, SKIP_8);
6525     FloatRegister vhi = vtmp0;
6526     FloatRegister vlo = vtmp1;
6527     ld1(vtmp3, T8H, src);
6528     uzp1(vlo, T16B, vtmp3, vtmp3);
6529     uzp2(vhi, T16B, vtmp3, vtmp3);
6530     // ISO-check on hi-parts (all zero).
6531     //                          ASCII-check on lo-parts (no sign).
6532                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6533     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6534                                 ASCII(umov(max, vtmp2, B, 0));
6535                                 ASCII(orr(chk, chk, max));
6536     cbnz(chk, SKIP_8);
6537 
6538     strd(vlo, Address(post(dst, 8)));
6539     subw(cnt, cnt, 8);
6540     add(src, src, 16);
6541     b(LOOP_8);
6542   }
6543   BIND(SKIP_8);
6544 
6545 #undef ASCII
6546 
6547   Label LOOP, DONE;
6548 
6549   cbz(cnt, DONE);
6550   BIND(LOOP);
6551   {
6552     Register chr = rscratch1;
6553     ldrh(chr, Address(post(src, 2)));
6554     tst(chr, ascii ? 0xff80 : 0xff00);
6555     br(NE, DONE);
6556     strb(chr, Address(post(dst, 1)));
6557     subs(cnt, cnt, 1);
6558     br(GT, LOOP);
6559   }
6560   BIND(DONE);
6561   // Return index where we stopped.
6562   subw(res, len, cnt);
6563 }
6564 
6565 // Inflate byte[] array to char[].
6566 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6567 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6568                                            FloatRegister vtmp1, FloatRegister vtmp2,
6569                                            FloatRegister vtmp3, Register tmp4) {
6570   Label big, done, after_init, to_stub;
6571 
6572   assert_different_registers(src, dst, len, tmp4, rscratch1);
6573 
6574   fmovd(vtmp1, 0.0);
6575   lsrw(tmp4, len, 3);
6576   bind(after_init);
6577   cbnzw(tmp4, big);
6578   // Short string: less than 8 bytes.
6579   {
6580     Label loop, tiny;
6581 
6582     cmpw(len, 4);
6583     br(LT, tiny);
6584     // Use SIMD to do 4 bytes.
6585     ldrs(vtmp2, post(src, 4));
6586     zip1(vtmp3, T8B, vtmp2, vtmp1);
6587     subw(len, len, 4);
6588     strd(vtmp3, post(dst, 8));
6589 
6590     cbzw(len, done);
6591 
6592     // Do the remaining bytes by steam.
6593     bind(loop);
6594     ldrb(tmp4, post(src, 1));
6595     strh(tmp4, post(dst, 2));
6596     subw(len, len, 1);
6597 
6598     bind(tiny);
6599     cbnz(len, loop);
6600 
6601     b(done);
6602   }
6603 
6604   if (SoftwarePrefetchHintDistance >= 0) {
6605     bind(to_stub);
6606       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6607       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6608       address tpc = trampoline_call(stub);
6609       if (tpc == nullptr) {
6610         DEBUG_ONLY(reset_labels(big, done));
6611         postcond(pc() == badAddress);
6612         return nullptr;
6613       }
6614       b(after_init);
6615   }
6616 
6617   // Unpack the bytes 8 at a time.
6618   bind(big);
6619   {
6620     Label loop, around, loop_last, loop_start;
6621 
6622     if (SoftwarePrefetchHintDistance >= 0) {
6623       const int large_loop_threshold = (64 + 16)/8;
6624       ldrd(vtmp2, post(src, 8));
6625       andw(len, len, 7);
6626       cmp(tmp4, (u1)large_loop_threshold);
6627       br(GE, to_stub);
6628       b(loop_start);
6629 
6630       bind(loop);
6631       ldrd(vtmp2, post(src, 8));
6632       bind(loop_start);
6633       subs(tmp4, tmp4, 1);
6634       br(EQ, loop_last);
6635       zip1(vtmp2, T16B, vtmp2, vtmp1);
6636       ldrd(vtmp3, post(src, 8));
6637       st1(vtmp2, T8H, post(dst, 16));
6638       subs(tmp4, tmp4, 1);
6639       zip1(vtmp3, T16B, vtmp3, vtmp1);
6640       st1(vtmp3, T8H, post(dst, 16));
6641       br(NE, loop);
6642       b(around);
6643       bind(loop_last);
6644       zip1(vtmp2, T16B, vtmp2, vtmp1);
6645       st1(vtmp2, T8H, post(dst, 16));
6646       bind(around);
6647       cbz(len, done);
6648     } else {
6649       andw(len, len, 7);
6650       bind(loop);
6651       ldrd(vtmp2, post(src, 8));
6652       sub(tmp4, tmp4, 1);
6653       zip1(vtmp3, T16B, vtmp2, vtmp1);
6654       st1(vtmp3, T8H, post(dst, 16));
6655       cbnz(tmp4, loop);
6656     }
6657   }
6658 
6659   // Do the tail of up to 8 bytes.
6660   add(src, src, len);
6661   ldrd(vtmp3, Address(src, -8));
6662   add(dst, dst, len, ext::uxtw, 1);
6663   zip1(vtmp3, T16B, vtmp3, vtmp1);
6664   strq(vtmp3, Address(dst, -16));
6665 
6666   bind(done);
6667   postcond(pc() != badAddress);
6668   return pc();
6669 }
6670 
6671 // Compress char[] array to byte[].
6672 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6673 // Return the array length if every element in array can be encoded,
6674 // otherwise, the index of first non-latin1 (> 0xff) character.
6675 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6676                                          Register res,
6677                                          FloatRegister tmp0, FloatRegister tmp1,
6678                                          FloatRegister tmp2, FloatRegister tmp3,
6679                                          FloatRegister tmp4, FloatRegister tmp5) {
6680   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6681 }
6682 
6683 // java.math.round(double a)
6684 // Returns the closest long to the argument, with ties rounding to
6685 // positive infinity.  This requires some fiddling for corner
6686 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6687 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6688                                        FloatRegister ftmp) {
6689   Label DONE;
6690   BLOCK_COMMENT("java_round_double: { ");
6691   fmovd(rscratch1, src);
6692   // Use RoundToNearestTiesAway unless src small and -ve.
6693   fcvtasd(dst, src);
6694   // Test if src >= 0 || abs(src) >= 0x1.0p52
6695   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6696   mov(rscratch2, julong_cast(0x1.0p52));
6697   cmp(rscratch1, rscratch2);
6698   br(HS, DONE); {
6699     // src < 0 && abs(src) < 0x1.0p52
6700     // src may have a fractional part, so add 0.5
6701     fmovd(ftmp, 0.5);
6702     faddd(ftmp, src, ftmp);
6703     // Convert double to jlong, use RoundTowardsNegative
6704     fcvtmsd(dst, ftmp);
6705   }
6706   bind(DONE);
6707   BLOCK_COMMENT("} java_round_double");
6708 }
6709 
6710 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6711                                       FloatRegister ftmp) {
6712   Label DONE;
6713   BLOCK_COMMENT("java_round_float: { ");
6714   fmovs(rscratch1, src);
6715   // Use RoundToNearestTiesAway unless src small and -ve.
6716   fcvtassw(dst, src);
6717   // Test if src >= 0 || abs(src) >= 0x1.0p23
6718   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6719   mov(rscratch2, jint_cast(0x1.0p23f));
6720   cmp(rscratch1, rscratch2);
6721   br(HS, DONE); {
6722     // src < 0 && |src| < 0x1.0p23
6723     // src may have a fractional part, so add 0.5
6724     fmovs(ftmp, 0.5f);
6725     fadds(ftmp, src, ftmp);
6726     // Convert float to jint, use RoundTowardsNegative
6727     fcvtmssw(dst, ftmp);
6728   }
6729   bind(DONE);
6730   BLOCK_COMMENT("} java_round_float");
6731 }
6732 
6733 // get_thread() can be called anywhere inside generated code so we
6734 // need to save whatever non-callee save context might get clobbered
6735 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6736 // the call setup code.
6737 //
6738 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6739 // On other systems, the helper is a usual C function.
6740 //
6741 void MacroAssembler::get_thread(Register dst) {
6742   RegSet saved_regs =
6743     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6744     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6745 
6746   protect_return_address();
6747   push(saved_regs, sp);
6748 
6749   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6750   blr(lr);
6751   if (dst != c_rarg0) {
6752     mov(dst, c_rarg0);
6753   }
6754 
6755   pop(saved_regs, sp);
6756   authenticate_return_address();
6757 }
6758 
6759 void MacroAssembler::cache_wb(Address line) {
6760   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6761   assert(line.index() == noreg, "index should be noreg");
6762   assert(line.offset() == 0, "offset should be 0");
6763   // would like to assert this
6764   // assert(line._ext.shift == 0, "shift should be zero");
6765   if (VM_Version::supports_dcpop()) {
6766     // writeback using clear virtual address to point of persistence
6767     dc(Assembler::CVAP, line.base());
6768   } else {
6769     // no need to generate anything as Unsafe.writebackMemory should
6770     // never invoke this stub
6771   }
6772 }
6773 
6774 void MacroAssembler::cache_wbsync(bool is_pre) {
6775   // we only need a barrier post sync
6776   if (!is_pre) {
6777     membar(Assembler::AnyAny);
6778   }
6779 }
6780 
6781 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6782   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6783     return;
6784   }
6785   // Make sure that native code does not change SVE vector length.
6786   Label verify_ok;
6787   movw(tmp, zr);
6788   sve_inc(tmp, B);
6789   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6790   br(EQ, verify_ok);
6791   stop("Error: SVE vector length has changed since jvm startup");
6792   bind(verify_ok);
6793 }
6794 
6795 void MacroAssembler::verify_ptrue() {
6796   Label verify_ok;
6797   if (!UseSVE) {
6798     return;
6799   }
6800   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6801   sve_dec(rscratch1, B);
6802   cbz(rscratch1, verify_ok);
6803   stop("Error: the preserved predicate register (p7) elements are not all true");
6804   bind(verify_ok);
6805 }
6806 
6807 void MacroAssembler::safepoint_isb() {
6808   isb();
6809 #ifndef PRODUCT
6810   if (VerifyCrossModifyFence) {
6811     // Clear the thread state.
6812     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6813   }
6814 #endif
6815 }
6816 
6817 #ifndef PRODUCT
6818 void MacroAssembler::verify_cross_modify_fence_not_required() {
6819   if (VerifyCrossModifyFence) {
6820     // Check if thread needs a cross modify fence.
6821     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6822     Label fence_not_required;
6823     cbz(rscratch1, fence_not_required);
6824     // If it does then fail.
6825     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6826     mov(c_rarg0, rthread);
6827     blr(rscratch1);
6828     bind(fence_not_required);
6829   }
6830 }
6831 #endif
6832 
6833 void MacroAssembler::spin_wait() {
6834   block_comment("spin_wait {");
6835   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6836     switch (VM_Version::spin_wait_desc().inst()) {
6837       case SpinWait::NOP:
6838         nop();
6839         break;
6840       case SpinWait::ISB:
6841         isb();
6842         break;
6843       case SpinWait::YIELD:
6844         yield();
6845         break;
6846       case SpinWait::SB:
6847         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6848         sb();
6849         break;
6850       case SpinWait::WFET:
6851         spin_wait_wfet(VM_Version::spin_wait_desc().delay());
6852         break;
6853       default:
6854         ShouldNotReachHere();
6855     }
6856   }
6857   block_comment("}");
6858 }
6859 
6860 void MacroAssembler::spin_wait_wfet(int delay_ns) {
6861   // The sequence assumes CNTFRQ_EL0 is fixed to 1GHz. The assumption is valid
6862   // starting from Armv8.6, according to the "D12.1.2 The system counter" of the
6863   // Arm Architecture Reference Manual for A-profile architecture version M.a.a.
6864   // This is sufficient because FEAT_WFXT is introduced from Armv8.6.
6865   Register target = rscratch1;
6866   Register current = rscratch2;
6867   get_cntvctss_el0(current);
6868   add(target, current, delay_ns);
6869 
6870   Label L_wait_loop;
6871   bind(L_wait_loop);
6872 
6873   wfet(target);
6874   get_cntvctss_el0(current);
6875 
6876   cmp(current, target);
6877   br(LT, L_wait_loop);
6878 
6879   sb();
6880 }
6881 
6882 // Stack frame creation/removal
6883 
6884 void MacroAssembler::enter(bool strip_ret_addr) {
6885   if (strip_ret_addr) {
6886     // Addresses can only be signed once. If there are multiple nested frames being created
6887     // in the same function, then the return address needs stripping first.
6888     strip_return_address();
6889   }
6890   protect_return_address();
6891   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6892   mov(rfp, sp);
6893 }
6894 
6895 void MacroAssembler::leave() {
6896   mov(sp, rfp);
6897   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6898   authenticate_return_address();
6899 }
6900 
6901 // ROP Protection
6902 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6903 // destroying stack frames or whenever directly loading/storing the LR to memory.
6904 // If ROP protection is not set then these functions are no-ops.
6905 // For more details on PAC see pauth_aarch64.hpp.
6906 
6907 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6908 // Uses value zero as the modifier.
6909 //
6910 void MacroAssembler::protect_return_address() {
6911   if (VM_Version::use_rop_protection()) {
6912     check_return_address();
6913     paciaz();
6914   }
6915 }
6916 
6917 // Sign the return value in the given register. Use before updating the LR in the existing stack
6918 // frame for the current function.
6919 // Uses value zero as the modifier.
6920 //
6921 void MacroAssembler::protect_return_address(Register return_reg) {
6922   if (VM_Version::use_rop_protection()) {
6923     check_return_address(return_reg);
6924     paciza(return_reg);
6925   }
6926 }
6927 
6928 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6929 // Uses value zero as the modifier.
6930 //
6931 void MacroAssembler::authenticate_return_address() {
6932   if (VM_Version::use_rop_protection()) {
6933     autiaz();
6934     check_return_address();
6935   }
6936 }
6937 
6938 // Authenticate the return value in the given register. Use before updating the LR in the existing
6939 // stack frame for the current function.
6940 // Uses value zero as the modifier.
6941 //
6942 void MacroAssembler::authenticate_return_address(Register return_reg) {
6943   if (VM_Version::use_rop_protection()) {
6944     autiza(return_reg);
6945     check_return_address(return_reg);
6946   }
6947 }
6948 
6949 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6950 // there is no guaranteed way of authenticating the LR.
6951 //
6952 void MacroAssembler::strip_return_address() {
6953   if (VM_Version::use_rop_protection()) {
6954     xpaclri();
6955   }
6956 }
6957 
6958 #ifndef PRODUCT
6959 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6960 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6961 // it is difficult to debug back to the callee function.
6962 // This function simply loads from the address in the given register.
6963 // Use directly after authentication to catch authentication failures.
6964 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6965 //
6966 void MacroAssembler::check_return_address(Register return_reg) {
6967   if (VM_Version::use_rop_protection()) {
6968     ldr(zr, Address(return_reg));
6969   }
6970 }
6971 #endif
6972 
6973 // The java_calling_convention describes stack locations as ideal slots on
6974 // a frame with no abi restrictions. Since we must observe abi restrictions
6975 // (like the placement of the register window) the slots must be biased by
6976 // the following value.
6977 static int reg2offset_in(VMReg r) {
6978   // Account for saved rfp and lr
6979   // This should really be in_preserve_stack_slots
6980   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6981 }
6982 
6983 static int reg2offset_out(VMReg r) {
6984   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6985 }
6986 
6987 // On 64bit we will store integer like items to the stack as
6988 // 64bits items (AArch64 ABI) even though java would only store
6989 // 32bits for a parameter. On 32bit it will simply be 32bits
6990 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6991 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6992   if (src.first()->is_stack()) {
6993     if (dst.first()->is_stack()) {
6994       // stack to stack
6995       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6996       str(tmp, Address(sp, reg2offset_out(dst.first())));
6997     } else {
6998       // stack to reg
6999       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7000     }
7001   } else if (dst.first()->is_stack()) {
7002     // reg to stack
7003     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7004   } else {
7005     if (dst.first() != src.first()) {
7006       sxtw(dst.first()->as_Register(), src.first()->as_Register());
7007     }
7008   }
7009 }
7010 
7011 // An oop arg. Must pass a handle not the oop itself
7012 void MacroAssembler::object_move(
7013                         OopMap* map,
7014                         int oop_handle_offset,
7015                         int framesize_in_slots,
7016                         VMRegPair src,
7017                         VMRegPair dst,
7018                         bool is_receiver,
7019                         int* receiver_offset) {
7020 
7021   // must pass a handle. First figure out the location we use as a handle
7022 
7023   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
7024 
7025   // See if oop is null if it is we need no handle
7026 
7027   if (src.first()->is_stack()) {
7028 
7029     // Oop is already on the stack as an argument
7030     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
7031     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
7032     if (is_receiver) {
7033       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
7034     }
7035 
7036     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7037     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7038     // conditionally move a null
7039     cmp(rscratch1, zr);
7040     csel(rHandle, zr, rHandle, Assembler::EQ);
7041   } else {
7042 
7043     // Oop is in an a register we must store it to the space we reserve
7044     // on the stack for oop_handles and pass a handle if oop is non-null
7045 
7046     const Register rOop = src.first()->as_Register();
7047     int oop_slot;
7048     if (rOop == j_rarg0)
7049       oop_slot = 0;
7050     else if (rOop == j_rarg1)
7051       oop_slot = 1;
7052     else if (rOop == j_rarg2)
7053       oop_slot = 2;
7054     else if (rOop == j_rarg3)
7055       oop_slot = 3;
7056     else if (rOop == j_rarg4)
7057       oop_slot = 4;
7058     else if (rOop == j_rarg5)
7059       oop_slot = 5;
7060     else if (rOop == j_rarg6)
7061       oop_slot = 6;
7062     else {
7063       assert(rOop == j_rarg7, "wrong register");
7064       oop_slot = 7;
7065     }
7066 
7067     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7068     int offset = oop_slot*VMRegImpl::stack_slot_size;
7069 
7070     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7071     // Store oop in handle area, may be null
7072     str(rOop, Address(sp, offset));
7073     if (is_receiver) {
7074       *receiver_offset = offset;
7075     }
7076 
7077     cmp(rOop, zr);
7078     lea(rHandle, Address(sp, offset));
7079     // conditionally move a null
7080     csel(rHandle, zr, rHandle, Assembler::EQ);
7081   }
7082 
7083   // If arg is on the stack then place it otherwise it is already in correct reg.
7084   if (dst.first()->is_stack()) {
7085     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7086   }
7087 }
7088 
7089 // A float arg may have to do float reg int reg conversion
7090 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7091  if (src.first()->is_stack()) {
7092     if (dst.first()->is_stack()) {
7093       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7094       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7095     } else {
7096       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7097     }
7098   } else if (src.first() != dst.first()) {
7099     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7100       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7101     else
7102       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7103   }
7104 }
7105 
7106 // A long move
7107 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7108   if (src.first()->is_stack()) {
7109     if (dst.first()->is_stack()) {
7110       // stack to stack
7111       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7112       str(tmp, Address(sp, reg2offset_out(dst.first())));
7113     } else {
7114       // stack to reg
7115       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7116     }
7117   } else if (dst.first()->is_stack()) {
7118     // reg to stack
7119     // Do we really have to sign extend???
7120     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7121     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7122   } else {
7123     if (dst.first() != src.first()) {
7124       mov(dst.first()->as_Register(), src.first()->as_Register());
7125     }
7126   }
7127 }
7128 
7129 
7130 // A double move
7131 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7132  if (src.first()->is_stack()) {
7133     if (dst.first()->is_stack()) {
7134       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7135       str(tmp, Address(sp, reg2offset_out(dst.first())));
7136     } else {
7137       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7138     }
7139   } else if (src.first() != dst.first()) {
7140     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7141       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7142     else
7143       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7144   }
7145 }
7146 
7147 // Implements fast-locking.
7148 //
7149 //  - obj: the object to be locked
7150 //  - t1, t2, t3: temporary registers, will be destroyed
7151 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7152 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7153   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7154 
7155   Label push;
7156   const Register top = t1;
7157   const Register mark = t2;
7158   const Register t = t3;
7159 
7160   // Preload the markWord. It is important that this is the first
7161   // instruction emitted as it is part of C1's null check semantics.
7162   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7163 
7164   if (UseObjectMonitorTable) {
7165     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7166     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7167   }
7168 
7169   if (DiagnoseSyncOnValueBasedClasses != 0) {
7170     load_klass(t1, obj);
7171     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7172     tst(t1, KlassFlags::_misc_is_value_based_class);
7173     br(Assembler::NE, slow);
7174   }
7175 
7176   // Check if the lock-stack is full.
7177   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7178   cmpw(top, (unsigned)LockStack::end_offset());
7179   br(Assembler::GE, slow);
7180 
7181   // Check for recursion.
7182   subw(t, top, oopSize);
7183   ldr(t, Address(rthread, t));
7184   cmp(obj, t);
7185   br(Assembler::EQ, push);
7186 
7187   // Check header for monitor (0b10).
7188   tst(mark, markWord::monitor_value);
7189   br(Assembler::NE, slow);
7190 
7191   // Try to lock. Transition lock bits 0b01 => 0b00
7192   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7193   orr(mark, mark, markWord::unlocked_value);
7194   eor(t, mark, markWord::unlocked_value);
7195   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7196           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7197   br(Assembler::NE, slow);
7198 
7199   bind(push);
7200   // After successful lock, push object on lock-stack.
7201   str(obj, Address(rthread, top));
7202   addw(top, top, oopSize);
7203   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7204 }
7205 
7206 // Implements fast-unlocking.
7207 //
7208 // - obj: the object to be unlocked
7209 // - t1, t2, t3: temporary registers
7210 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7211 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7212   // cmpxchg clobbers rscratch1.
7213   assert_different_registers(obj, t1, t2, t3, rscratch1);
7214 
7215 #ifdef ASSERT
7216   {
7217     // Check for lock-stack underflow.
7218     Label stack_ok;
7219     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7220     cmpw(t1, (unsigned)LockStack::start_offset());
7221     br(Assembler::GE, stack_ok);
7222     STOP("Lock-stack underflow");
7223     bind(stack_ok);
7224   }
7225 #endif
7226 
7227   Label unlocked, push_and_slow;
7228   const Register top = t1;
7229   const Register mark = t2;
7230   const Register t = t3;
7231 
7232   // Check if obj is top of lock-stack.
7233   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7234   subw(top, top, oopSize);
7235   ldr(t, Address(rthread, top));
7236   cmp(obj, t);
7237   br(Assembler::NE, slow);
7238 
7239   // Pop lock-stack.
7240   DEBUG_ONLY(str(zr, Address(rthread, top));)
7241   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7242 
7243   // Check if recursive.
7244   subw(t, top, oopSize);
7245   ldr(t, Address(rthread, t));
7246   cmp(obj, t);
7247   br(Assembler::EQ, unlocked);
7248 
7249   // Not recursive. Check header for monitor (0b10).
7250   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7251   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7252 
7253 #ifdef ASSERT
7254   // Check header not unlocked (0b01).
7255   Label not_unlocked;
7256   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7257   stop("fast_unlock already unlocked");
7258   bind(not_unlocked);
7259 #endif
7260 
7261   // Try to unlock. Transition lock bits 0b00 => 0b01
7262   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7263   orr(t, mark, markWord::unlocked_value);
7264   cmpxchg(obj, mark, t, Assembler::xword,
7265           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7266   br(Assembler::EQ, unlocked);
7267 
7268   bind(push_and_slow);
7269   // Restore lock-stack and handle the unlock in runtime.
7270   DEBUG_ONLY(str(obj, Address(rthread, top));)
7271   addw(top, top, oopSize);
7272   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7273   b(slow);
7274 
7275   bind(unlocked);
7276 }