1 /*
   2  * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "ci/ciInlineKlass.hpp"
  30 #include "code/compiledIC.hpp"
  31 #include "compiler/compileTask.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/barrierSet.hpp"
  35 #include "gc/shared/barrierSetAssembler.hpp"
  36 #include "gc/shared/cardTableBarrierSet.hpp"
  37 #include "gc/shared/cardTable.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/tlab_globals.hpp"
  40 #include "interpreter/bytecodeHistogram.hpp"
  41 #include "interpreter/interpreter.hpp"
  42 #include "interpreter/interpreterRuntime.hpp"
  43 #include "jvm.h"
  44 #include "memory/resourceArea.hpp"
  45 #include "memory/universe.hpp"
  46 #include "nativeInst_aarch64.hpp"
  47 #include "oops/accessDecorators.hpp"
  48 #include "oops/compressedKlass.inline.hpp"
  49 #include "oops/compressedOops.inline.hpp"
  50 #include "oops/klass.inline.hpp"
  51 #include "oops/resolvedFieldEntry.hpp"
  52 #include "runtime/arguments.hpp"
  53 #include "runtime/continuation.hpp"
  54 #include "runtime/globals.hpp"
  55 #include "runtime/icache.hpp"
  56 #include "runtime/interfaceSupport.inline.hpp"
  57 #include "runtime/javaThread.hpp"
  58 #include "runtime/jniHandles.inline.hpp"
  59 #include "runtime/sharedRuntime.hpp"
  60 #include "runtime/signature_cc.hpp"
  61 #include "runtime/stubRoutines.hpp"
  62 #include "utilities/globalDefinitions.hpp"
  63 #include "utilities/powerOfTwo.hpp"
  64 #include "vmreg_aarch64.inline.hpp"
  65 #ifdef COMPILER1
  66 #include "c1/c1_LIRAssembler.hpp"
  67 #endif
  68 #ifdef COMPILER2
  69 #include "oops/oop.hpp"
  70 #include "opto/compile.hpp"
  71 #include "opto/node.hpp"
  72 #include "opto/output.hpp"
  73 #endif
  74 
  75 #include <sys/types.h>
  76 
  77 #ifdef PRODUCT
  78 #define BLOCK_COMMENT(str) /* nothing */
  79 #else
  80 #define BLOCK_COMMENT(str) block_comment(str)
  81 #endif
  82 #define STOP(str) stop(str);
  83 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  84 
  85 #ifdef ASSERT
  86 extern "C" void disnm(intptr_t p);
  87 #endif
  88 // Target-dependent relocation processing
  89 //
  90 // Instruction sequences whose target may need to be retrieved or
  91 // patched are distinguished by their leading instruction, sorting
  92 // them into three main instruction groups and related subgroups.
  93 //
  94 // 1) Branch, Exception and System (insn count = 1)
  95 //    1a) Unconditional branch (immediate):
  96 //      b/bl imm19
  97 //    1b) Compare & branch (immediate):
  98 //      cbz/cbnz Rt imm19
  99 //    1c) Test & branch (immediate):
 100 //      tbz/tbnz Rt imm14
 101 //    1d) Conditional branch (immediate):
 102 //      b.cond imm19
 103 //
 104 // 2) Loads and Stores (insn count = 1)
 105 //    2a) Load register literal:
 106 //      ldr Rt imm19
 107 //
 108 // 3) Data Processing Immediate (insn count = 2 or 3)
 109 //    3a) PC-rel. addressing
 110 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 111 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 112 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 113 //      adr/adrp Rx imm21
 114 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 115 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 116 //      The latter form can only happen when the target is an
 117 //      ExternalAddress, and (by definition) ExternalAddresses don't
 118 //      move. Because of that property, there is never any need to
 119 //      patch the last of the three instructions. However,
 120 //      MacroAssembler::target_addr_for_insn takes all three
 121 //      instructions into account and returns the correct address.
 122 //    3b) Move wide (immediate)
 123 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 124 //
 125 // A switch on a subset of the instruction's bits provides an
 126 // efficient dispatch to these subcases.
 127 //
 128 // insn[28:26] -> main group ('x' == don't care)
 129 //   00x -> UNALLOCATED
 130 //   100 -> Data Processing Immediate
 131 //   101 -> Branch, Exception and System
 132 //   x1x -> Loads and Stores
 133 //
 134 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 135 // n.b. in some cases extra bits need to be checked to verify the
 136 // instruction is as expected
 137 //
 138 // 1) ... xx101x Branch, Exception and System
 139 //   1a)  00___x Unconditional branch (immediate)
 140 //   1b)  01___0 Compare & branch (immediate)
 141 //   1c)  01___1 Test & branch (immediate)
 142 //   1d)  10___0 Conditional branch (immediate)
 143 //        other  Should not happen
 144 //
 145 // 2) ... xxx1x0 Loads and Stores
 146 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 147 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 148 //                strictly should be 64 bit non-FP/SIMD i.e.
 149 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 150 //
 151 // 3) ... xx100x Data Processing Immediate
 152 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 153 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 154 //                 strictly should be 64 bit movz #imm16<<0
 155 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 156 //
 157 
 158 static uint32_t insn_at(address insn_addr, int n) {
 159   return ((uint32_t*)insn_addr)[n];
 160 }
 161 
 162 template<typename T>
 163 class RelocActions : public AllStatic {
 164 
 165 public:
 166 
 167   static int ALWAYSINLINE run(address insn_addr, address &target) {
 168     int instructions = 1;
 169     uint32_t insn = insn_at(insn_addr, 0);
 170 
 171     uint32_t dispatch = Instruction_aarch64::extract(insn, 30, 25);
 172     switch(dispatch) {
 173       case 0b001010:
 174       case 0b001011: {
 175         instructions = T::unconditionalBranch(insn_addr, target);
 176         break;
 177       }
 178       case 0b101010:   // Conditional branch (immediate)
 179       case 0b011010: { // Compare & branch (immediate)
 180         instructions = T::conditionalBranch(insn_addr, target);
 181         break;
 182       }
 183       case 0b011011: {
 184         instructions = T::testAndBranch(insn_addr, target);
 185         break;
 186       }
 187       case 0b001100:
 188       case 0b001110:
 189       case 0b011100:
 190       case 0b011110:
 191       case 0b101100:
 192       case 0b101110:
 193       case 0b111100:
 194       case 0b111110: {
 195         // load/store
 196         if ((Instruction_aarch64::extract(insn, 29, 24) & 0b111011) == 0b011000) {
 197           // Load register (literal)
 198           instructions = T::loadStore(insn_addr, target);
 199           break;
 200         } else {
 201           // nothing to do
 202           assert(target == nullptr, "did not expect to relocate target for polling page load");
 203         }
 204         break;
 205       }
 206       case 0b001000:
 207       case 0b011000:
 208       case 0b101000:
 209       case 0b111000: {
 210         // adr/adrp
 211         assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 212         int shift = Instruction_aarch64::extract(insn, 31, 31);
 213         if (shift) {
 214           uint32_t insn2 = insn_at(insn_addr, 1);
 215           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 216               Instruction_aarch64::extract(insn, 4, 0) ==
 217               Instruction_aarch64::extract(insn2, 9, 5)) {
 218             instructions = T::adrp(insn_addr, target, T::adrpMem);
 219           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 220                      Instruction_aarch64::extract(insn, 4, 0) ==
 221                      Instruction_aarch64::extract(insn2, 4, 0)) {
 222             instructions = T::adrp(insn_addr, target, T::adrpAdd);
 223           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 224                      Instruction_aarch64::extract(insn, 4, 0) ==
 225                      Instruction_aarch64::extract(insn2, 4, 0)) {
 226             instructions = T::adrp(insn_addr, target, T::adrpMovk);
 227           } else {
 228             ShouldNotReachHere();
 229           }
 230         } else {
 231           instructions = T::adr(insn_addr, target);
 232         }
 233         break;
 234       }
 235       case 0b001001:
 236       case 0b011001:
 237       case 0b101001:
 238       case 0b111001: {
 239         instructions = T::immediate(insn_addr, target);
 240         break;
 241       }
 242       default: {
 243         ShouldNotReachHere();
 244       }
 245     }
 246 
 247     T::verify(insn_addr, target);
 248     return instructions * NativeInstruction::instruction_size;
 249   }
 250 };
 251 
 252 class Patcher : public AllStatic {
 253 public:
 254   static int unconditionalBranch(address insn_addr, address &target) {
 255     intptr_t offset = (target - insn_addr) >> 2;
 256     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 257     return 1;
 258   }
 259   static int conditionalBranch(address insn_addr, address &target) {
 260     intptr_t offset = (target - insn_addr) >> 2;
 261     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 262     return 1;
 263   }
 264   static int testAndBranch(address insn_addr, address &target) {
 265     intptr_t offset = (target - insn_addr) >> 2;
 266     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 267     return 1;
 268   }
 269   static int loadStore(address insn_addr, address &target) {
 270     intptr_t offset = (target - insn_addr) >> 2;
 271     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 272     return 1;
 273   }
 274   static int adr(address insn_addr, address &target) {
 275 #ifdef ASSERT
 276     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 277 #endif
 278     // PC-rel. addressing
 279     ptrdiff_t offset = target - insn_addr;
 280     int offset_lo = offset & 3;
 281     offset >>= 2;
 282     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 283     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 284     return 1;
 285   }
 286   template<typename U>
 287   static int adrp(address insn_addr, address &target, U inner) {
 288     int instructions = 1;
 289 #ifdef ASSERT
 290     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 291 #endif
 292     ptrdiff_t offset = target - insn_addr;
 293     instructions = 2;
 294     precond(inner != nullptr);
 295     // Give the inner reloc a chance to modify the target.
 296     address adjusted_target = target;
 297     instructions = inner(insn_addr, adjusted_target);
 298     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 299     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 300     offset = adr_page - pc_page;
 301     int offset_lo = offset & 3;
 302     offset >>= 2;
 303     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 304     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 305     return instructions;
 306   }
 307   static int adrpMem(address insn_addr, address &target) {
 308     uintptr_t dest = (uintptr_t)target;
 309     int offset_lo = dest & 0xfff;
 310     uint32_t insn2 = insn_at(insn_addr, 1);
 311     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 312     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 313     guarantee(((dest >> size) << size) == dest, "misaligned target");
 314     return 2;
 315   }
 316   static int adrpAdd(address insn_addr, address &target) {
 317     uintptr_t dest = (uintptr_t)target;
 318     int offset_lo = dest & 0xfff;
 319     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 320     return 2;
 321   }
 322   static int adrpMovk(address insn_addr, address &target) {
 323     uintptr_t dest = uintptr_t(target);
 324     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 325     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 326     target = address(dest);
 327     return 2;
 328   }
 329   static int immediate(address insn_addr, address &target) {
 330     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010100, "must be");
 331     uint64_t dest = (uint64_t)target;
 332     // Move wide constant
 333     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 334     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 335     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 336     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 337     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 338     return 3;
 339   }
 340   static void verify(address insn_addr, address &target) {
 341 #ifdef ASSERT
 342     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 343     if (!(address_is == target)) {
 344       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 345       disnm((intptr_t)insn_addr);
 346       assert(address_is == target, "should be");
 347     }
 348 #endif
 349   }
 350 };
 351 
 352 // If insn1 and insn2 use the same register to form an address, either
 353 // by an offsetted LDR or a simple ADD, return the offset. If the
 354 // second instruction is an LDR, the offset may be scaled.
 355 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 356   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 357       Instruction_aarch64::extract(insn1, 4, 0) ==
 358       Instruction_aarch64::extract(insn2, 9, 5)) {
 359     // Load/store register (unsigned immediate)
 360     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 361     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 362     byte_offset <<= size;
 363     return true;
 364   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 365              Instruction_aarch64::extract(insn1, 4, 0) ==
 366              Instruction_aarch64::extract(insn2, 4, 0)) {
 367     // add (immediate)
 368     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 369     return true;
 370   }
 371   return false;
 372 }
 373 
 374 class AArch64Decoder : public AllStatic {
 375 public:
 376 
 377   static int loadStore(address insn_addr, address &target) {
 378     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 379     target = insn_addr + (offset << 2);
 380     return 1;
 381   }
 382   static int unconditionalBranch(address insn_addr, address &target) {
 383     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 25, 0);
 384     target = insn_addr + (offset << 2);
 385     return 1;
 386   }
 387   static int conditionalBranch(address insn_addr, address &target) {
 388     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 389     target = address(((uint64_t)insn_addr + (offset << 2)));
 390     return 1;
 391   }
 392   static int testAndBranch(address insn_addr, address &target) {
 393     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 18, 5);
 394     target = address(((uint64_t)insn_addr + (offset << 2)));
 395     return 1;
 396   }
 397   static int adr(address insn_addr, address &target) {
 398     // PC-rel. addressing
 399     uint32_t insn = insn_at(insn_addr, 0);
 400     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 401     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 402     target = address((uint64_t)insn_addr + offset);
 403     return 1;
 404   }
 405   template<typename U>
 406   static int adrp(address insn_addr, address &target, U inner) {
 407     uint32_t insn = insn_at(insn_addr, 0);
 408     assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 409     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 410     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 411     int shift = 12;
 412     offset <<= shift;
 413     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 414     target_page &= ((uint64_t)-1) << shift;
 415     uint32_t insn2 = insn_at(insn_addr, 1);
 416     target = address(target_page);
 417     precond(inner != nullptr);
 418     inner(insn_addr, target);
 419     return 2;
 420   }
 421   static int adrpMem(address insn_addr, address &target) {
 422     uint32_t insn2 = insn_at(insn_addr, 1);
 423     // Load/store register (unsigned immediate)
 424     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 425     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 426     byte_offset <<= size;
 427     target += byte_offset;
 428     return 2;
 429   }
 430   static int adrpAdd(address insn_addr, address &target) {
 431     uint32_t insn2 = insn_at(insn_addr, 1);
 432     // add (immediate)
 433     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 434     target += byte_offset;
 435     return 2;
 436   }
 437   static int adrpMovk(address insn_addr, address &target) {
 438     uint32_t insn2 = insn_at(insn_addr, 1);
 439     uint64_t dest = uint64_t(target);
 440     dest = (dest & 0xffff0000ffffffff) |
 441       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 442     target = address(dest);
 443 
 444     // We know the destination 4k page. Maybe we have a third
 445     // instruction.
 446     uint32_t insn = insn_at(insn_addr, 0);
 447     uint32_t insn3 = insn_at(insn_addr, 2);
 448     ptrdiff_t byte_offset;
 449     if (offset_for(insn, insn3, byte_offset)) {
 450       target += byte_offset;
 451       return 3;
 452     } else {
 453       return 2;
 454     }
 455   }
 456   static int immediate(address insn_addr, address &target) {
 457     uint32_t *insns = (uint32_t *)insn_addr;
 458     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 459     // Move wide constant: movz, movk, movk.  See movptr().
 460     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 461     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 462     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 463                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 464                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 465     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 466     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 467     return 3;
 468   }
 469   static void verify(address insn_addr, address &target) {
 470   }
 471 };
 472 
 473 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 474   address target;
 475   RelocActions<AArch64Decoder>::run(insn_addr, target);
 476   return target;
 477 }
 478 
 479 // Patch any kind of instruction; there may be several instructions.
 480 // Return the total length (in bytes) of the instructions.
 481 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 482   return RelocActions<Patcher>::run(insn_addr, target);
 483 }
 484 
 485 int MacroAssembler::patch_oop(address insn_addr, address o) {
 486   int instructions;
 487   unsigned insn = *(unsigned*)insn_addr;
 488   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 489 
 490   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 491   // narrow OOPs by setting the upper 16 bits in the first
 492   // instruction.
 493   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 494     // Move narrow OOP
 495     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 496     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 497     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 498     instructions = 2;
 499   } else {
 500     // Move wide OOP
 501     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 502     uintptr_t dest = (uintptr_t)o;
 503     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 504     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 505     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 506     instructions = 3;
 507   }
 508   return instructions * NativeInstruction::instruction_size;
 509 }
 510 
 511 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 512   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 513   // We encode narrow ones by setting the upper 16 bits in the first
 514   // instruction.
 515   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 516   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 517          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 518 
 519   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 520   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 521   return 2 * NativeInstruction::instruction_size;
 522 }
 523 
 524 address MacroAssembler::target_addr_for_insn_or_null(address insn_addr) {
 525   if (NativeInstruction::is_ldrw_to_zr(insn_addr)) {
 526     return nullptr;
 527   }
 528   return MacroAssembler::target_addr_for_insn(insn_addr);
 529 }
 530 
 531 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 532   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 533   if (at_return) {
 534     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 535     // we may safely use the sp instead to perform the stack watermark check.
 536     cmp(in_nmethod ? sp : rfp, tmp);
 537     br(Assembler::HI, slow_path);
 538   } else {
 539     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 540   }
 541 }
 542 
 543 void MacroAssembler::rt_call(address dest, Register tmp) {
 544   CodeBlob *cb = CodeCache::find_blob(dest);
 545   if (cb) {
 546     far_call(RuntimeAddress(dest));
 547   } else {
 548     lea(tmp, RuntimeAddress(dest));
 549     blr(tmp);
 550   }
 551 }
 552 
 553 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 554   if (!Continuations::enabled()) return;
 555   Label done;
 556   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 557   cmp(sp, rscratch1);
 558   br(Assembler::LS, done);
 559   mov(rscratch1, sp); // we can't use sp as the source in str
 560   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 561   bind(done);
 562 }
 563 
 564 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 565   if (!Continuations::enabled()) return;
 566   Label done;
 567   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 568   cmp(sp, rscratch1);
 569   br(Assembler::LO, done);
 570   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 571   bind(done);
 572 }
 573 
 574 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 575   // we must set sp to zero to clear frame
 576   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 577 
 578   // must clear fp, so that compiled frames are not confused; it is
 579   // possible that we need it only for debugging
 580   if (clear_fp) {
 581     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 582   }
 583 
 584   // Always clear the pc because it could have been set by make_walkable()
 585   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 586 }
 587 
 588 // Calls to C land
 589 //
 590 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 591 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 592 // has to be reset to 0. This is required to allow proper stack traversal.
 593 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 594                                          Register last_java_fp,
 595                                          Register last_java_pc,
 596                                          Register scratch) {
 597 
 598   if (last_java_pc->is_valid()) {
 599       str(last_java_pc, Address(rthread,
 600                                 JavaThread::frame_anchor_offset()
 601                                 + JavaFrameAnchor::last_Java_pc_offset()));
 602     }
 603 
 604   // determine last_java_sp register
 605   if (last_java_sp == sp) {
 606     mov(scratch, sp);
 607     last_java_sp = scratch;
 608   } else if (!last_java_sp->is_valid()) {
 609     last_java_sp = esp;
 610   }
 611 
 612   // last_java_fp is optional
 613   if (last_java_fp->is_valid()) {
 614     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 615   }
 616 
 617   // We must set sp last.
 618   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 619 }
 620 
 621 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 622                                          Register last_java_fp,
 623                                          address  last_java_pc,
 624                                          Register scratch) {
 625   assert(last_java_pc != nullptr, "must provide a valid PC");
 626 
 627   adr(scratch, last_java_pc);
 628   str(scratch, Address(rthread,
 629                        JavaThread::frame_anchor_offset()
 630                        + JavaFrameAnchor::last_Java_pc_offset()));
 631 
 632   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 633 }
 634 
 635 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 636                                          Register last_java_fp,
 637                                          Label &L,
 638                                          Register scratch) {
 639   if (L.is_bound()) {
 640     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 641   } else {
 642     InstructionMark im(this);
 643     L.add_patch_at(code(), locator());
 644     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 645   }
 646 }
 647 
 648 static inline bool target_needs_far_branch(address addr) {
 649   if (AOTCodeCache::is_on_for_dump()) {
 650     return true;
 651   }
 652   // codecache size <= 128M
 653   if (!MacroAssembler::far_branches()) {
 654     return false;
 655   }
 656   // codecache size > 240M
 657   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 658     return true;
 659   }
 660   // codecache size: 128M..240M
 661   return !CodeCache::is_non_nmethod(addr);
 662 }
 663 
 664 void MacroAssembler::far_call(Address entry, Register tmp) {
 665   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 666   assert(CodeCache::find_blob(entry.target()) != nullptr,
 667          "destination of far call not found in code cache");
 668   assert(entry.rspec().type() == relocInfo::external_word_type
 669          || entry.rspec().type() == relocInfo::runtime_call_type
 670          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 671   if (target_needs_far_branch(entry.target())) {
 672     uint64_t offset;
 673     // We can use ADRP here because we know that the total size of
 674     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 675     adrp(tmp, entry, offset);
 676     add(tmp, tmp, offset);
 677     blr(tmp);
 678   } else {
 679     bl(entry);
 680   }
 681 }
 682 
 683 int MacroAssembler::far_jump(Address entry, Register tmp) {
 684   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 685   assert(CodeCache::find_blob(entry.target()) != nullptr,
 686          "destination of far call not found in code cache");
 687   assert(entry.rspec().type() == relocInfo::external_word_type
 688          || entry.rspec().type() == relocInfo::runtime_call_type
 689          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 690   address start = pc();
 691   if (target_needs_far_branch(entry.target())) {
 692     uint64_t offset;
 693     // We can use ADRP here because we know that the total size of
 694     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 695     adrp(tmp, entry, offset);
 696     add(tmp, tmp, offset);
 697     br(tmp);
 698   } else {
 699     b(entry);
 700   }
 701   return pc() - start;
 702 }
 703 
 704 void MacroAssembler::reserved_stack_check() {
 705     // testing if reserved zone needs to be enabled
 706     Label no_reserved_zone_enabling;
 707 
 708     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 709     cmp(sp, rscratch1);
 710     br(Assembler::LO, no_reserved_zone_enabling);
 711 
 712     enter();   // LR and FP are live.
 713     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 714     mov(c_rarg0, rthread);
 715     blr(rscratch1);
 716     leave();
 717 
 718     // We have already removed our own frame.
 719     // throw_delayed_StackOverflowError will think that it's been
 720     // called by our caller.
 721     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 722     br(rscratch1);
 723     should_not_reach_here();
 724 
 725     bind(no_reserved_zone_enabling);
 726 }
 727 
 728 static void pass_arg0(MacroAssembler* masm, Register arg) {
 729   if (c_rarg0 != arg ) {
 730     masm->mov(c_rarg0, arg);
 731   }
 732 }
 733 
 734 static void pass_arg1(MacroAssembler* masm, Register arg) {
 735   if (c_rarg1 != arg ) {
 736     masm->mov(c_rarg1, arg);
 737   }
 738 }
 739 
 740 static void pass_arg2(MacroAssembler* masm, Register arg) {
 741   if (c_rarg2 != arg ) {
 742     masm->mov(c_rarg2, arg);
 743   }
 744 }
 745 
 746 static void pass_arg3(MacroAssembler* masm, Register arg) {
 747   if (c_rarg3 != arg ) {
 748     masm->mov(c_rarg3, arg);
 749   }
 750 }
 751 
 752 void MacroAssembler::call_VM_base(Register oop_result,
 753                                   Register java_thread,
 754                                   Register last_java_sp,
 755                                   Label*   return_pc,
 756                                   address  entry_point,
 757                                   int      number_of_arguments,
 758                                   bool     check_exceptions) {
 759    // determine java_thread register
 760   if (!java_thread->is_valid()) {
 761     java_thread = rthread;
 762   }
 763 
 764   // determine last_java_sp register
 765   if (!last_java_sp->is_valid()) {
 766     last_java_sp = esp;
 767   }
 768 
 769   // debugging support
 770   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 771   assert(java_thread == rthread, "unexpected register");
 772 #ifdef ASSERT
 773   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 774   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 775 #endif // ASSERT
 776 
 777   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 778   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 779 
 780   // push java thread (becomes first argument of C function)
 781 
 782   mov(c_rarg0, java_thread);
 783 
 784   // set last Java frame before call
 785   assert(last_java_sp != rfp, "can't use rfp");
 786 
 787   Label l;
 788   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 789 
 790   // do the call, remove parameters
 791   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 792 
 793   // lr could be poisoned with PAC signature during throw_pending_exception
 794   // if it was tail-call optimized by compiler, since lr is not callee-saved
 795   // reload it with proper value
 796   adr(lr, l);
 797 
 798   // reset last Java frame
 799   // Only interpreter should have to clear fp
 800   reset_last_Java_frame(true);
 801 
 802    // C++ interp handles this in the interpreter
 803   check_and_handle_popframe(java_thread);
 804   check_and_handle_earlyret(java_thread);
 805 
 806   if (check_exceptions) {
 807     // check for pending exceptions (java_thread is set upon return)
 808     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 809     Label ok;
 810     cbz(rscratch1, ok);
 811     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 812     br(rscratch1);
 813     bind(ok);
 814   }
 815 
 816   // get oop result if there is one and reset the value in the thread
 817   if (oop_result->is_valid()) {
 818     get_vm_result_oop(oop_result, java_thread);
 819   }
 820 }
 821 
 822 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 823   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 824 }
 825 
 826 // Check the entry target is always reachable from any branch.
 827 static bool is_always_within_branch_range(Address entry) {
 828   if (AOTCodeCache::is_on_for_dump()) {
 829     return false;
 830   }
 831   const address target = entry.target();
 832 
 833   if (!CodeCache::contains(target)) {
 834     // We always use trampolines for callees outside CodeCache.
 835     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 836     return false;
 837   }
 838 
 839   if (!MacroAssembler::far_branches()) {
 840     return true;
 841   }
 842 
 843   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 844     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 845     // Non-compiled methods stay forever in CodeCache.
 846     // We check whether the longest possible branch is within the branch range.
 847     assert(CodeCache::find_blob(target) != nullptr &&
 848           !CodeCache::find_blob(target)->is_nmethod(),
 849           "runtime call of compiled method");
 850     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 851     const address left_longest_branch_start = CodeCache::low_bound();
 852     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 853                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 854     return is_reachable;
 855   }
 856 
 857   return false;
 858 }
 859 
 860 // Maybe emit a call via a trampoline. If the code cache is small
 861 // trampolines won't be emitted.
 862 address MacroAssembler::trampoline_call(Address entry) {
 863   assert(entry.rspec().type() == relocInfo::runtime_call_type
 864          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 865          || entry.rspec().type() == relocInfo::static_call_type
 866          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 867 
 868   address target = entry.target();
 869 
 870   if (!is_always_within_branch_range(entry)) {
 871     if (!in_scratch_emit_size()) {
 872       // We don't want to emit a trampoline if C2 is generating dummy
 873       // code during its branch shortening phase.
 874       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 875         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 876         code()->share_trampoline_for(entry.target(), offset());
 877       } else {
 878         address stub = emit_trampoline_stub(offset(), target);
 879         if (stub == nullptr) {
 880           postcond(pc() == badAddress);
 881           return nullptr; // CodeCache is full
 882         }
 883       }
 884     }
 885     target = pc();
 886   }
 887 
 888   address call_pc = pc();
 889   relocate(entry.rspec());
 890   bl(target);
 891 
 892   postcond(pc() != badAddress);
 893   return call_pc;
 894 }
 895 
 896 // Emit a trampoline stub for a call to a target which is too far away.
 897 //
 898 // code sequences:
 899 //
 900 // call-site:
 901 //   branch-and-link to <destination> or <trampoline stub>
 902 //
 903 // Related trampoline stub for this call site in the stub section:
 904 //   load the call target from the constant pool
 905 //   branch (LR still points to the call site above)
 906 
 907 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 908                                              address dest) {
 909   // Max stub size: alignment nop, TrampolineStub.
 910   address stub = start_a_stub(max_trampoline_stub_size());
 911   if (stub == nullptr) {
 912     return nullptr;  // CodeBuffer::expand failed
 913   }
 914 
 915   // Create a trampoline stub relocation which relates this trampoline stub
 916   // with the call instruction at insts_call_instruction_offset in the
 917   // instructions code-section.
 918   align(wordSize);
 919   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 920                                             + insts_call_instruction_offset));
 921   const int stub_start_offset = offset();
 922 
 923   // Now, create the trampoline stub's code:
 924   // - load the call
 925   // - call
 926   Label target;
 927   ldr(rscratch1, target);
 928   br(rscratch1);
 929   bind(target);
 930   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 931          "should be");
 932   emit_int64((int64_t)dest);
 933 
 934   const address stub_start_addr = addr_at(stub_start_offset);
 935 
 936   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 937 
 938   end_a_stub();
 939   return stub_start_addr;
 940 }
 941 
 942 int MacroAssembler::max_trampoline_stub_size() {
 943   // Max stub size: alignment nop, TrampolineStub.
 944   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 945 }
 946 
 947 void MacroAssembler::emit_static_call_stub() {
 948   // CompiledDirectCall::set_to_interpreted knows the
 949   // exact layout of this stub.
 950 
 951   isb();
 952   mov_metadata(rmethod, nullptr);
 953 
 954   // Jump to the entry point of the c2i stub.
 955   if (codestub_branch_needs_far_jump()) {
 956     movptr(rscratch1, 0);
 957     br(rscratch1);
 958   } else {
 959     b(pc());
 960   }
 961 }
 962 
 963 int MacroAssembler::static_call_stub_size() {
 964   if (!codestub_branch_needs_far_jump()) {
 965     // isb; movk; movz; movz; b
 966     return 5 * NativeInstruction::instruction_size;
 967   }
 968   // isb; movk; movz; movz; movk; movz; movz; br
 969   return 8 * NativeInstruction::instruction_size;
 970 }
 971 
 972 void MacroAssembler::c2bool(Register x) {
 973   // implements x == 0 ? 0 : 1
 974   // note: must only look at least-significant byte of x
 975   //       since C-style booleans are stored in one byte
 976   //       only! (was bug)
 977   tst(x, 0xff);
 978   cset(x, Assembler::NE);
 979 }
 980 
 981 address MacroAssembler::ic_call(address entry, jint method_index) {
 982   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 983   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 984   return trampoline_call(Address(entry, rh));
 985 }
 986 
 987 int MacroAssembler::ic_check_size() {
 988   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 989   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 990     return NativeInstruction::instruction_size * (7 + extra_instructions);
 991   } else {
 992     return NativeInstruction::instruction_size * (5 + extra_instructions);
 993   }
 994 }
 995 
 996 int MacroAssembler::ic_check(int end_alignment) {
 997   Register receiver = j_rarg0;
 998   Register data = rscratch2;
 999   Register tmp1 = rscratch1;
1000   Register tmp2 = r10;
1001 
1002   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
1003   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
1004   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1005   // before the inline cache check here, and not after
1006   align(end_alignment, offset() + ic_check_size());
1007 
1008   int uep_offset = offset();
1009 
1010   if (UseCompactObjectHeaders) {
1011     load_narrow_klass_compact(tmp1, receiver);
1012     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1013     cmpw(tmp1, tmp2);
1014   } else if (UseCompressedClassPointers) {
1015     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1016     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1017     cmpw(tmp1, tmp2);
1018   } else {
1019     ldr(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1020     ldr(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1021     cmp(tmp1, tmp2);
1022   }
1023 
1024   Label dont;
1025   br(Assembler::EQ, dont);
1026   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1027   bind(dont);
1028   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1029 
1030   return uep_offset;
1031 }
1032 
1033 // Implementation of call_VM versions
1034 
1035 void MacroAssembler::call_VM(Register oop_result,
1036                              address entry_point,
1037                              bool check_exceptions) {
1038   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1039 }
1040 
1041 void MacroAssembler::call_VM(Register oop_result,
1042                              address entry_point,
1043                              Register arg_1,
1044                              bool check_exceptions) {
1045   pass_arg1(this, arg_1);
1046   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1047 }
1048 
1049 void MacroAssembler::call_VM(Register oop_result,
1050                              address entry_point,
1051                              Register arg_1,
1052                              Register arg_2,
1053                              bool check_exceptions) {
1054   assert_different_registers(arg_1, c_rarg2);
1055   pass_arg2(this, arg_2);
1056   pass_arg1(this, arg_1);
1057   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1058 }
1059 
1060 void MacroAssembler::call_VM(Register oop_result,
1061                              address entry_point,
1062                              Register arg_1,
1063                              Register arg_2,
1064                              Register arg_3,
1065                              bool check_exceptions) {
1066   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1067   assert_different_registers(arg_2, c_rarg3);
1068   pass_arg3(this, arg_3);
1069 
1070   pass_arg2(this, arg_2);
1071 
1072   pass_arg1(this, arg_1);
1073   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1074 }
1075 
1076 void MacroAssembler::call_VM(Register oop_result,
1077                              Register last_java_sp,
1078                              address entry_point,
1079                              int number_of_arguments,
1080                              bool check_exceptions) {
1081   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, 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                              bool check_exceptions) {
1089   pass_arg1(this, arg_1);
1090   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1091 }
1092 
1093 void MacroAssembler::call_VM(Register oop_result,
1094                              Register last_java_sp,
1095                              address entry_point,
1096                              Register arg_1,
1097                              Register arg_2,
1098                              bool check_exceptions) {
1099 
1100   assert_different_registers(arg_1, c_rarg2);
1101   pass_arg2(this, arg_2);
1102   pass_arg1(this, arg_1);
1103   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1104 }
1105 
1106 void MacroAssembler::call_VM(Register oop_result,
1107                              Register last_java_sp,
1108                              address entry_point,
1109                              Register arg_1,
1110                              Register arg_2,
1111                              Register arg_3,
1112                              bool check_exceptions) {
1113   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1114   assert_different_registers(arg_2, c_rarg3);
1115   pass_arg3(this, arg_3);
1116   pass_arg2(this, arg_2);
1117   pass_arg1(this, arg_1);
1118   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1119 }
1120 
1121 
1122 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1123   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1124   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1125   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1126 }
1127 
1128 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1129   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1130   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1131 }
1132 
1133 void MacroAssembler::align(int modulus) {
1134   align(modulus, offset());
1135 }
1136 
1137 // Ensure that the code at target bytes offset from the current offset() is aligned
1138 // according to modulus.
1139 void MacroAssembler::align(int modulus, int target) {
1140   int delta = target - offset();
1141   while ((offset() + delta) % modulus != 0) nop();
1142 }
1143 
1144 void MacroAssembler::post_call_nop() {
1145   if (!Continuations::enabled()) {
1146     return;
1147   }
1148   InstructionMark im(this);
1149   relocate(post_call_nop_Relocation::spec());
1150   InlineSkippedInstructionsCounter skipCounter(this);
1151   nop();
1152   movk(zr, 0);
1153   movk(zr, 0);
1154 }
1155 
1156 // these are no-ops overridden by InterpreterMacroAssembler
1157 
1158 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1159 
1160 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1161 
1162 // Look up the method for a megamorphic invokeinterface call.
1163 // The target method is determined by <intf_klass, itable_index>.
1164 // The receiver klass is in recv_klass.
1165 // On success, the result will be in method_result, and execution falls through.
1166 // On failure, execution transfers to the given label.
1167 void MacroAssembler::lookup_interface_method(Register recv_klass,
1168                                              Register intf_klass,
1169                                              RegisterOrConstant itable_index,
1170                                              Register method_result,
1171                                              Register scan_temp,
1172                                              Label& L_no_such_interface,
1173                          bool return_method) {
1174   assert_different_registers(recv_klass, intf_klass, scan_temp);
1175   assert_different_registers(method_result, intf_klass, scan_temp);
1176   assert(recv_klass != method_result || !return_method,
1177      "recv_klass can be destroyed when method isn't needed");
1178   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1179          "caller must use same register for non-constant itable index as for method");
1180 
1181   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1182   int vtable_base = in_bytes(Klass::vtable_start_offset());
1183   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1184   int scan_step   = itableOffsetEntry::size() * wordSize;
1185   int vte_size    = vtableEntry::size_in_bytes();
1186   assert(vte_size == wordSize, "else adjust times_vte_scale");
1187 
1188   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1189 
1190   // Could store the aligned, prescaled offset in the klass.
1191   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1192   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1193   add(scan_temp, scan_temp, vtable_base);
1194 
1195   if (return_method) {
1196     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1197     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1198     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1199     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1200     if (itentry_off)
1201       add(recv_klass, recv_klass, itentry_off);
1202   }
1203 
1204   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1205   //   if (scan->interface() == intf) {
1206   //     result = (klass + scan->offset() + itable_index);
1207   //   }
1208   // }
1209   Label search, found_method;
1210 
1211   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1212   cmp(intf_klass, method_result);
1213   br(Assembler::EQ, found_method);
1214   bind(search);
1215   // Check that the previous entry is non-null.  A null entry means that
1216   // the receiver class doesn't implement the interface, and wasn't the
1217   // same as when the caller was compiled.
1218   cbz(method_result, L_no_such_interface);
1219   if (itableOffsetEntry::interface_offset() != 0) {
1220     add(scan_temp, scan_temp, scan_step);
1221     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1222   } else {
1223     ldr(method_result, Address(pre(scan_temp, scan_step)));
1224   }
1225   cmp(intf_klass, method_result);
1226   br(Assembler::NE, search);
1227 
1228   bind(found_method);
1229 
1230   // Got a hit.
1231   if (return_method) {
1232     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1233     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1234   }
1235 }
1236 
1237 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1238 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1239 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1240 // The target method is determined by <holder_klass, itable_index>.
1241 // The receiver klass is in recv_klass.
1242 // On success, the result will be in method_result, and execution falls through.
1243 // On failure, execution transfers to the given label.
1244 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1245                                                   Register holder_klass,
1246                                                   Register resolved_klass,
1247                                                   Register method_result,
1248                                                   Register temp_itbl_klass,
1249                                                   Register scan_temp,
1250                                                   int itable_index,
1251                                                   Label& L_no_such_interface) {
1252   // 'method_result' is only used as output register at the very end of this method.
1253   // Until then we can reuse it as 'holder_offset'.
1254   Register holder_offset = method_result;
1255   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1256 
1257   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1258   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1259   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1260   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1261 
1262   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1263 
1264   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1265   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1266   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1267   // temp_itbl_klass = itable[0]._interface;
1268   int vtblEntrySize = vtableEntry::size_in_bytes();
1269   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1270   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1271   mov(holder_offset, zr);
1272   // scan_temp = &(itable[0]._interface)
1273   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1274 
1275   // Initial checks:
1276   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1277   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1278   //   - if (itable[0] == 0), no such interface
1279   cmp(resolved_klass, holder_klass);
1280   br(Assembler::NE, L_loop_search_resolved_entry);
1281   cmp(holder_klass, temp_itbl_klass);
1282   br(Assembler::EQ, L_holder_found);
1283   cbz(temp_itbl_klass, L_no_such_interface);
1284 
1285   // Loop: Look for holder_klass record in itable
1286   //   do {
1287   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1288   //     if (temp_itbl_klass == holder_klass) {
1289   //       goto L_holder_found; // Found!
1290   //     }
1291   //   } while (temp_itbl_klass != 0);
1292   //   goto L_no_such_interface // Not found.
1293   Label L_search_holder;
1294   bind(L_search_holder);
1295     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1296     cmp(holder_klass, temp_itbl_klass);
1297     br(Assembler::EQ, L_holder_found);
1298     cbnz(temp_itbl_klass, L_search_holder);
1299 
1300   b(L_no_such_interface);
1301 
1302   // Loop: Look for resolved_class record in itable
1303   //   while (true) {
1304   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1305   //     if (temp_itbl_klass == 0) {
1306   //       goto L_no_such_interface;
1307   //     }
1308   //     if (temp_itbl_klass == resolved_klass) {
1309   //        goto L_resolved_found;  // Found!
1310   //     }
1311   //     if (temp_itbl_klass == holder_klass) {
1312   //        holder_offset = scan_temp;
1313   //     }
1314   //   }
1315   //
1316   Label L_loop_search_resolved;
1317   bind(L_loop_search_resolved);
1318     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1319   bind(L_loop_search_resolved_entry);
1320     cbz(temp_itbl_klass, L_no_such_interface);
1321     cmp(resolved_klass, temp_itbl_klass);
1322     br(Assembler::EQ, L_resolved_found);
1323     cmp(holder_klass, temp_itbl_klass);
1324     br(Assembler::NE, L_loop_search_resolved);
1325     mov(holder_offset, scan_temp);
1326     b(L_loop_search_resolved);
1327 
1328   // See if we already have a holder klass. If not, go and scan for it.
1329   bind(L_resolved_found);
1330   cbz(holder_offset, L_search_holder);
1331   mov(scan_temp, holder_offset);
1332 
1333   // Finally, scan_temp contains holder_klass vtable offset
1334   bind(L_holder_found);
1335   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1336   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1337     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1338   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1339 }
1340 
1341 // virtual method calling
1342 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1343                                            RegisterOrConstant vtable_index,
1344                                            Register method_result) {
1345   assert(vtableEntry::size() * wordSize == 8,
1346          "adjust the scaling in the code below");
1347   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1348 
1349   if (vtable_index.is_register()) {
1350     lea(method_result, Address(recv_klass,
1351                                vtable_index.as_register(),
1352                                Address::lsl(LogBytesPerWord)));
1353     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1354   } else {
1355     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1356     ldr(method_result,
1357         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1358   }
1359 }
1360 
1361 void MacroAssembler::check_klass_subtype(Register sub_klass,
1362                            Register super_klass,
1363                            Register temp_reg,
1364                            Label& L_success) {
1365   Label L_failure;
1366   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1367   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1368   bind(L_failure);
1369 }
1370 
1371 
1372 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1373                                                    Register super_klass,
1374                                                    Register temp_reg,
1375                                                    Label* L_success,
1376                                                    Label* L_failure,
1377                                                    Label* L_slow_path,
1378                                                    Register super_check_offset) {
1379   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1380   bool must_load_sco = ! super_check_offset->is_valid();
1381   if (must_load_sco) {
1382     assert(temp_reg != noreg, "supply either a temp or a register offset");
1383   }
1384 
1385   Label L_fallthrough;
1386   int label_nulls = 0;
1387   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1388   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1389   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1390   assert(label_nulls <= 1, "at most one null in the batch");
1391 
1392   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1393   Address super_check_offset_addr(super_klass, sco_offset);
1394 
1395   // Hacked jmp, which may only be used just before L_fallthrough.
1396 #define final_jmp(label)                                                \
1397   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1398   else                            b(label)                /*omit semi*/
1399 
1400   // If the pointers are equal, we are done (e.g., String[] elements).
1401   // This self-check enables sharing of secondary supertype arrays among
1402   // non-primary types such as array-of-interface.  Otherwise, each such
1403   // type would need its own customized SSA.
1404   // We move this check to the front of the fast path because many
1405   // type checks are in fact trivially successful in this manner,
1406   // so we get a nicely predicted branch right at the start of the check.
1407   cmp(sub_klass, super_klass);
1408   br(Assembler::EQ, *L_success);
1409 
1410   // Check the supertype display:
1411   if (must_load_sco) {
1412     ldrw(temp_reg, super_check_offset_addr);
1413     super_check_offset = temp_reg;
1414   }
1415 
1416   Address super_check_addr(sub_klass, super_check_offset);
1417   ldr(rscratch1, super_check_addr);
1418   cmp(super_klass, rscratch1); // load displayed supertype
1419   br(Assembler::EQ, *L_success);
1420 
1421   // This check has worked decisively for primary supers.
1422   // Secondary supers are sought in the super_cache ('super_cache_addr').
1423   // (Secondary supers are interfaces and very deeply nested subtypes.)
1424   // This works in the same check above because of a tricky aliasing
1425   // between the super_cache and the primary super display elements.
1426   // (The 'super_check_addr' can address either, as the case requires.)
1427   // Note that the cache is updated below if it does not help us find
1428   // what we need immediately.
1429   // So if it was a primary super, we can just fail immediately.
1430   // Otherwise, it's the slow path for us (no success at this point).
1431 
1432   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1433   if (L_failure == &L_fallthrough) {
1434     cbz(rscratch1, *L_slow_path);
1435   } else {
1436     cbnz(rscratch1, *L_failure);
1437     final_jmp(*L_slow_path);
1438   }
1439 
1440   bind(L_fallthrough);
1441 
1442 #undef final_jmp
1443 }
1444 
1445 // These two are taken from x86, but they look generally useful
1446 
1447 // scans count pointer sized words at [addr] for occurrence of value,
1448 // generic
1449 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1450                                 Register scratch) {
1451   Label Lloop, Lexit;
1452   cbz(count, Lexit);
1453   bind(Lloop);
1454   ldr(scratch, post(addr, wordSize));
1455   cmp(value, scratch);
1456   br(EQ, Lexit);
1457   sub(count, count, 1);
1458   cbnz(count, Lloop);
1459   bind(Lexit);
1460 }
1461 
1462 // scans count 4 byte words at [addr] for occurrence of value,
1463 // generic
1464 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1465                                 Register scratch) {
1466   Label Lloop, Lexit;
1467   cbz(count, Lexit);
1468   bind(Lloop);
1469   ldrw(scratch, post(addr, wordSize));
1470   cmpw(value, scratch);
1471   br(EQ, Lexit);
1472   sub(count, count, 1);
1473   cbnz(count, Lloop);
1474   bind(Lexit);
1475 }
1476 
1477 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1478                                                           Register super_klass,
1479                                                           Register temp_reg,
1480                                                           Register temp2_reg,
1481                                                           Label* L_success,
1482                                                           Label* L_failure,
1483                                                           bool set_cond_codes) {
1484   // NB! Callers may assume that, when temp2_reg is a valid register,
1485   // this code sets it to a nonzero value.
1486 
1487   assert_different_registers(sub_klass, super_klass, temp_reg);
1488   if (temp2_reg != noreg)
1489     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1490 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1491 
1492   Label L_fallthrough;
1493   int label_nulls = 0;
1494   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1495   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1496   assert(label_nulls <= 1, "at most one null in the batch");
1497 
1498   // a couple of useful fields in sub_klass:
1499   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1500   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1501   Address secondary_supers_addr(sub_klass, ss_offset);
1502   Address super_cache_addr(     sub_klass, sc_offset);
1503 
1504   BLOCK_COMMENT("check_klass_subtype_slow_path");
1505 
1506   // Do a linear scan of the secondary super-klass chain.
1507   // This code is rarely used, so simplicity is a virtue here.
1508   // The repne_scan instruction uses fixed registers, which we must spill.
1509   // Don't worry too much about pre-existing connections with the input regs.
1510 
1511   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1512   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1513 
1514   RegSet pushed_registers;
1515   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1516   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1517 
1518   if (super_klass != r0) {
1519     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1520   }
1521 
1522   push(pushed_registers, sp);
1523 
1524   // Get super_klass value into r0 (even if it was in r5 or r2).
1525   if (super_klass != r0) {
1526     mov(r0, super_klass);
1527   }
1528 
1529 #ifndef PRODUCT
1530   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1531 #endif //PRODUCT
1532 
1533   // We will consult the secondary-super array.
1534   ldr(r5, secondary_supers_addr);
1535   // Load the array length.
1536   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1537   // Skip to start of data.
1538   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1539 
1540   cmp(sp, zr); // Clear Z flag; SP is never zero
1541   // Scan R2 words at [R5] for an occurrence of R0.
1542   // Set NZ/Z based on last compare.
1543   repne_scan(r5, r0, r2, rscratch1);
1544 
1545   // Unspill the temp. registers:
1546   pop(pushed_registers, sp);
1547 
1548   br(Assembler::NE, *L_failure);
1549 
1550   // Success.  Cache the super we found and proceed in triumph.
1551 
1552   if (UseSecondarySupersCache) {
1553     str(super_klass, super_cache_addr);
1554   }
1555 
1556   if (L_success != &L_fallthrough) {
1557     b(*L_success);
1558   }
1559 
1560 #undef IS_A_TEMP
1561 
1562   bind(L_fallthrough);
1563 }
1564 
1565 // If Register r is invalid, remove a new register from
1566 // available_regs, and add new register to regs_to_push.
1567 Register MacroAssembler::allocate_if_noreg(Register r,
1568                                   RegSetIterator<Register> &available_regs,
1569                                   RegSet &regs_to_push) {
1570   if (!r->is_valid()) {
1571     r = *available_regs++;
1572     regs_to_push += r;
1573   }
1574   return r;
1575 }
1576 
1577 // check_klass_subtype_slow_path_table() looks for super_klass in the
1578 // hash table belonging to super_klass, branching to L_success or
1579 // L_failure as appropriate. This is essentially a shim which
1580 // allocates registers as necessary then calls
1581 // lookup_secondary_supers_table() to do the work. Any of the temp
1582 // regs may be noreg, in which case this logic will chooses some
1583 // registers push and pop them from the stack.
1584 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1585                                                          Register super_klass,
1586                                                          Register temp_reg,
1587                                                          Register temp2_reg,
1588                                                          Register temp3_reg,
1589                                                          Register result_reg,
1590                                                          FloatRegister vtemp,
1591                                                          Label* L_success,
1592                                                          Label* L_failure,
1593                                                          bool set_cond_codes) {
1594   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1595 
1596   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1597 
1598   Label L_fallthrough;
1599   int label_nulls = 0;
1600   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1601   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1602   assert(label_nulls <= 1, "at most one null in the batch");
1603 
1604   BLOCK_COMMENT("check_klass_subtype_slow_path");
1605 
1606   RegSetIterator<Register> available_regs
1607     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1608 
1609   RegSet pushed_regs;
1610 
1611   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1612   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1613   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1614   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1615 
1616   push(pushed_regs, sp);
1617 
1618   lookup_secondary_supers_table_var(sub_klass,
1619                                     super_klass,
1620                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1621                                     nullptr);
1622   cmp(result_reg, zr);
1623 
1624   // Unspill the temp. registers:
1625   pop(pushed_regs, sp);
1626 
1627   // NB! Callers may assume that, when set_cond_codes is true, this
1628   // code sets temp2_reg to a nonzero value.
1629   if (set_cond_codes) {
1630     mov(temp2_reg, 1);
1631   }
1632 
1633   br(Assembler::NE, *L_failure);
1634 
1635   if (L_success != &L_fallthrough) {
1636     b(*L_success);
1637   }
1638 
1639   bind(L_fallthrough);
1640 }
1641 
1642 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1643                                                    Register super_klass,
1644                                                    Register temp_reg,
1645                                                    Register temp2_reg,
1646                                                    Label* L_success,
1647                                                    Label* L_failure,
1648                                                    bool set_cond_codes) {
1649   if (UseSecondarySupersTable) {
1650     check_klass_subtype_slow_path_table
1651       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1652        /*vtemp*/fnoreg,
1653        L_success, L_failure, set_cond_codes);
1654   } else {
1655     check_klass_subtype_slow_path_linear
1656       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1657   }
1658 }
1659 
1660 
1661 // Ensure that the inline code and the stub are using the same registers.
1662 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1663 do {                                                               \
1664   assert(r_super_klass  == r0                                   && \
1665          r_array_base   == r1                                   && \
1666          r_array_length == r2                                   && \
1667          (r_array_index == r3        || r_array_index == noreg) && \
1668          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1669          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1670          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1671 } while(0)
1672 
1673 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1674                                                          Register r_super_klass,
1675                                                          Register temp1,
1676                                                          Register temp2,
1677                                                          Register temp3,
1678                                                          FloatRegister vtemp,
1679                                                          Register result,
1680                                                          u1 super_klass_slot,
1681                                                          bool stub_is_near) {
1682   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1683 
1684   Label L_fallthrough;
1685 
1686   BLOCK_COMMENT("lookup_secondary_supers_table {");
1687 
1688   const Register
1689     r_array_base   = temp1, // r1
1690     r_array_length = temp2, // r2
1691     r_array_index  = temp3, // r3
1692     r_bitmap       = rscratch2;
1693 
1694   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1695 
1696   u1 bit = super_klass_slot;
1697 
1698   // Make sure that result is nonzero if the TBZ below misses.
1699   mov(result, 1);
1700 
1701   // We're going to need the bitmap in a vector reg and in a core reg,
1702   // so load both now.
1703   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1704   if (bit != 0) {
1705     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1706   }
1707   // First check the bitmap to see if super_klass might be present. If
1708   // the bit is zero, we are certain that super_klass is not one of
1709   // the secondary supers.
1710   tbz(r_bitmap, bit, L_fallthrough);
1711 
1712   // Get the first array index that can contain super_klass into r_array_index.
1713   if (bit != 0) {
1714     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1715     cnt(vtemp, T8B, vtemp);
1716     addv(vtemp, T8B, vtemp);
1717     fmovd(r_array_index, vtemp);
1718   } else {
1719     mov(r_array_index, (u1)1);
1720   }
1721   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1722 
1723   // We will consult the secondary-super array.
1724   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1725 
1726   // The value i in r_array_index is >= 1, so even though r_array_base
1727   // points to the length, we don't need to adjust it to point to the
1728   // data.
1729   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1730   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1731 
1732   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1733   eor(result, result, r_super_klass);
1734   cbz(result, L_fallthrough); // Found a match
1735 
1736   // Is there another entry to check? Consult the bitmap.
1737   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1738 
1739   // Linear probe.
1740   if (bit != 0) {
1741     ror(r_bitmap, r_bitmap, bit);
1742   }
1743 
1744   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1745   // The next slot to be inspected, by the stub we're about to call,
1746   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1747   // have been checked.
1748   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1749   if (stub_is_near) {
1750     bl(stub);
1751   } else {
1752     address call = trampoline_call(stub);
1753     if (call == nullptr) {
1754       return false; // trampoline allocation failed
1755     }
1756   }
1757 
1758   BLOCK_COMMENT("} lookup_secondary_supers_table");
1759 
1760   bind(L_fallthrough);
1761 
1762   if (VerifySecondarySupers) {
1763     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1764                                   temp1, temp2, result);      // r1, r2, r5
1765   }
1766   return true;
1767 }
1768 
1769 // At runtime, return 0 in result if r_super_klass is a superclass of
1770 // r_sub_klass, otherwise return nonzero. Use this version of
1771 // lookup_secondary_supers_table() if you don't know ahead of time
1772 // which superclass will be searched for. Used by interpreter and
1773 // runtime stubs. It is larger and has somewhat greater latency than
1774 // the version above, which takes a constant super_klass_slot.
1775 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1776                                                        Register r_super_klass,
1777                                                        Register temp1,
1778                                                        Register temp2,
1779                                                        Register temp3,
1780                                                        FloatRegister vtemp,
1781                                                        Register result,
1782                                                        Label *L_success) {
1783   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1784 
1785   Label L_fallthrough;
1786 
1787   BLOCK_COMMENT("lookup_secondary_supers_table {");
1788 
1789   const Register
1790     r_array_index = temp3,
1791     slot          = rscratch1,
1792     r_bitmap      = rscratch2;
1793 
1794   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1795 
1796   // Make sure that result is nonzero if the test below misses.
1797   mov(result, 1);
1798 
1799   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1800 
1801   // First check the bitmap to see if super_klass might be present. If
1802   // the bit is zero, we are certain that super_klass is not one of
1803   // the secondary supers.
1804 
1805   // This next instruction is equivalent to:
1806   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1807   // sub(temp2, tmp_reg, slot);
1808   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1809   lslv(temp2, r_bitmap, temp2);
1810   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1811 
1812   bool must_save_v0 = (vtemp == fnoreg);
1813   if (must_save_v0) {
1814     // temp1 and result are free, so use them to preserve vtemp
1815     vtemp = v0;
1816     mov(temp1,  vtemp, D, 0);
1817     mov(result, vtemp, D, 1);
1818   }
1819 
1820   // Get the first array index that can contain super_klass into r_array_index.
1821   mov(vtemp, D, 0, temp2);
1822   cnt(vtemp, T8B, vtemp);
1823   addv(vtemp, T8B, vtemp);
1824   mov(r_array_index, vtemp, D, 0);
1825 
1826   if (must_save_v0) {
1827     mov(vtemp, D, 0, temp1 );
1828     mov(vtemp, D, 1, result);
1829   }
1830 
1831   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1832 
1833   const Register
1834     r_array_base   = temp1,
1835     r_array_length = temp2;
1836 
1837   // The value i in r_array_index is >= 1, so even though r_array_base
1838   // points to the length, we don't need to adjust it to point to the
1839   // data.
1840   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1841   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1842 
1843   // We will consult the secondary-super array.
1844   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1845 
1846   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1847   eor(result, result, r_super_klass);
1848   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1849 
1850   // Is there another entry to check? Consult the bitmap.
1851   rorv(r_bitmap, r_bitmap, slot);
1852   // rol(r_bitmap, r_bitmap, 1);
1853   tbz(r_bitmap, 1, L_fallthrough);
1854 
1855   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1856   // The next slot to be inspected, by the logic we're about to call,
1857   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1858   // have been checked.
1859   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1860                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1861 
1862   BLOCK_COMMENT("} lookup_secondary_supers_table");
1863 
1864   bind(L_fallthrough);
1865 
1866   if (VerifySecondarySupers) {
1867     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1868                                   temp1, temp2, result);      // r1, r2, r5
1869   }
1870 
1871   if (L_success) {
1872     cbz(result, *L_success);
1873   }
1874 }
1875 
1876 // Called by code generated by check_klass_subtype_slow_path
1877 // above. This is called when there is a collision in the hashed
1878 // lookup in the secondary supers array.
1879 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1880                                                              Register r_array_base,
1881                                                              Register r_array_index,
1882                                                              Register r_bitmap,
1883                                                              Register temp1,
1884                                                              Register result,
1885                                                              bool is_stub) {
1886   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1887 
1888   const Register
1889     r_array_length = temp1,
1890     r_sub_klass    = noreg; // unused
1891 
1892   if (is_stub) {
1893     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1894   }
1895 
1896   Label L_fallthrough, L_huge;
1897 
1898   // Load the array length.
1899   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1900   // And adjust the array base to point to the data.
1901   // NB! Effectively increments current slot index by 1.
1902   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1903   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1904 
1905   // The bitmap is full to bursting.
1906   // Implicit invariant: BITMAP_FULL implies (length > 0)
1907   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1908   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1909   br(GT, L_huge);
1910 
1911   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1912   // current slot (at secondary_supers[r_array_index]) has not yet
1913   // been inspected, and r_array_index may be out of bounds if we
1914   // wrapped around the end of the array.
1915 
1916   { // This is conventional linear probing, but instead of terminating
1917     // when a null entry is found in the table, we maintain a bitmap
1918     // in which a 0 indicates missing entries.
1919     // As long as the bitmap is not completely full,
1920     // array_length == popcount(bitmap). The array_length check above
1921     // guarantees there are 0s in the bitmap, so the loop eventually
1922     // terminates.
1923     Label L_loop;
1924     bind(L_loop);
1925 
1926     // Check for wraparound.
1927     cmp(r_array_index, r_array_length);
1928     csel(r_array_index, zr, r_array_index, GE);
1929 
1930     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1931     eor(result, rscratch1, r_super_klass);
1932     cbz(result, L_fallthrough);
1933 
1934     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1935 
1936     ror(r_bitmap, r_bitmap, 1);
1937     add(r_array_index, r_array_index, 1);
1938     b(L_loop);
1939   }
1940 
1941   { // Degenerate case: more than 64 secondary supers.
1942     // FIXME: We could do something smarter here, maybe a vectorized
1943     // comparison or a binary search, but is that worth any added
1944     // complexity?
1945     bind(L_huge);
1946     cmp(sp, zr); // Clear Z flag; SP is never zero
1947     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1948     cset(result, NE); // result == 0 iff we got a match.
1949   }
1950 
1951   bind(L_fallthrough);
1952 }
1953 
1954 // Make sure that the hashed lookup and a linear scan agree.
1955 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1956                                                    Register r_super_klass,
1957                                                    Register temp1,
1958                                                    Register temp2,
1959                                                    Register result) {
1960   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1961 
1962   const Register
1963     r_array_base   = temp1,
1964     r_array_length = temp2,
1965     r_array_index  = noreg, // unused
1966     r_bitmap       = noreg; // unused
1967 
1968   BLOCK_COMMENT("verify_secondary_supers_table {");
1969 
1970   // We will consult the secondary-super array.
1971   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1972 
1973   // Load the array length.
1974   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1975   // And adjust the array base to point to the data.
1976   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1977 
1978   cmp(sp, zr); // Clear Z flag; SP is never zero
1979   // Scan R2 words at [R5] for an occurrence of R0.
1980   // Set NZ/Z based on last compare.
1981   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1982   // rscratch1 == 0 iff we got a match.
1983   cset(rscratch1, NE);
1984 
1985   Label passed;
1986   cmp(result, zr);
1987   cset(result, NE); // normalize result to 0/1 for comparison
1988 
1989   cmp(rscratch1, result);
1990   br(EQ, passed);
1991   {
1992     mov(r0, r_super_klass);         // r0 <- r0
1993     mov(r1, r_sub_klass);           // r1 <- r4
1994     mov(r2, /*expected*/rscratch1); // r2 <- r8
1995     mov(r3, result);                // r3 <- r5
1996     mov(r4, (address)("mismatch")); // r4 <- const
1997     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
1998     should_not_reach_here();
1999   }
2000   bind(passed);
2001 
2002   BLOCK_COMMENT("} verify_secondary_supers_table");
2003 }
2004 
2005 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
2006   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
2007   assert_different_registers(klass, rthread, scratch);
2008 
2009   Label L_fallthrough, L_tmp;
2010   if (L_fast_path == nullptr) {
2011     L_fast_path = &L_fallthrough;
2012   } else if (L_slow_path == nullptr) {
2013     L_slow_path = &L_fallthrough;
2014   }
2015   // Fast path check: class is fully initialized
2016   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2017   ldarb(scratch, scratch);
2018   cmp(scratch, InstanceKlass::fully_initialized);
2019   br(Assembler::EQ, *L_fast_path);
2020 
2021   // Fast path check: current thread is initializer thread
2022   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2023   cmp(rthread, scratch);
2024 
2025   if (L_slow_path == &L_fallthrough) {
2026     br(Assembler::EQ, *L_fast_path);
2027     bind(*L_slow_path);
2028   } else if (L_fast_path == &L_fallthrough) {
2029     br(Assembler::NE, *L_slow_path);
2030     bind(*L_fast_path);
2031   } else {
2032     Unimplemented();
2033   }
2034 }
2035 
2036 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2037   if (!VerifyOops || VerifyAdapterSharing) {
2038     // Below address of the code string confuses VerifyAdapterSharing
2039     // because it may differ between otherwise equivalent adapters.
2040     return;
2041   }
2042 
2043   // Pass register number to verify_oop_subroutine
2044   const char* b = nullptr;
2045   {
2046     ResourceMark rm;
2047     stringStream ss;
2048     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2049     b = code_string(ss.as_string());
2050   }
2051   BLOCK_COMMENT("verify_oop {");
2052 
2053   strip_return_address(); // This might happen within a stack frame.
2054   protect_return_address();
2055   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2056   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2057 
2058   mov(r0, reg);
2059   movptr(rscratch1, (uintptr_t)(address)b);
2060 
2061   // call indirectly to solve generation ordering problem
2062   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2063   ldr(rscratch2, Address(rscratch2));
2064   blr(rscratch2);
2065 
2066   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2067   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2068   authenticate_return_address();
2069 
2070   BLOCK_COMMENT("} verify_oop");
2071 }
2072 
2073 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2074   if (!VerifyOops || VerifyAdapterSharing) {
2075     // Below address of the code string confuses VerifyAdapterSharing
2076     // because it may differ between otherwise equivalent adapters.
2077     return;
2078   }
2079 
2080   const char* b = nullptr;
2081   {
2082     ResourceMark rm;
2083     stringStream ss;
2084     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2085     b = code_string(ss.as_string());
2086   }
2087   BLOCK_COMMENT("verify_oop_addr {");
2088 
2089   strip_return_address(); // This might happen within a stack frame.
2090   protect_return_address();
2091   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2092   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2093 
2094   // addr may contain sp so we will have to adjust it based on the
2095   // pushes that we just did.
2096   if (addr.uses(sp)) {
2097     lea(r0, addr);
2098     ldr(r0, Address(r0, 4 * wordSize));
2099   } else {
2100     ldr(r0, addr);
2101   }
2102   movptr(rscratch1, (uintptr_t)(address)b);
2103 
2104   // call indirectly to solve generation ordering problem
2105   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2106   ldr(rscratch2, Address(rscratch2));
2107   blr(rscratch2);
2108 
2109   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2110   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2111   authenticate_return_address();
2112 
2113   BLOCK_COMMENT("} verify_oop_addr");
2114 }
2115 
2116 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2117                                          int extra_slot_offset) {
2118   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2119   int stackElementSize = Interpreter::stackElementSize;
2120   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2121 #ifdef ASSERT
2122   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2123   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2124 #endif
2125   if (arg_slot.is_constant()) {
2126     return Address(esp, arg_slot.as_constant() * stackElementSize
2127                    + offset);
2128   } else {
2129     add(rscratch1, esp, arg_slot.as_register(),
2130         ext::uxtx, exact_log2(stackElementSize));
2131     return Address(rscratch1, offset);
2132   }
2133 }
2134 
2135 void MacroAssembler::call_VM_leaf_base(address entry_point,
2136                                        int number_of_arguments,
2137                                        Label *retaddr) {
2138   Label E, L;
2139 
2140   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2141 
2142   mov(rscratch1, RuntimeAddress(entry_point));
2143   blr(rscratch1);
2144   if (retaddr)
2145     bind(*retaddr);
2146 
2147   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2148 }
2149 
2150 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2151   call_VM_leaf_base(entry_point, number_of_arguments);
2152 }
2153 
2154 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2155   pass_arg0(this, arg_0);
2156   call_VM_leaf_base(entry_point, 1);
2157 }
2158 
2159 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2160   assert_different_registers(arg_1, c_rarg0);
2161   pass_arg0(this, arg_0);
2162   pass_arg1(this, arg_1);
2163   call_VM_leaf_base(entry_point, 2);
2164 }
2165 
2166 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2167                                   Register arg_1, Register arg_2) {
2168   assert_different_registers(arg_1, c_rarg0);
2169   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2170   pass_arg0(this, arg_0);
2171   pass_arg1(this, arg_1);
2172   pass_arg2(this, arg_2);
2173   call_VM_leaf_base(entry_point, 3);
2174 }
2175 
2176 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2177   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2178 }
2179 
2180 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2181   pass_arg0(this, arg_0);
2182   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2183 }
2184 
2185 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2186 
2187   assert_different_registers(arg_0, c_rarg1);
2188   pass_arg1(this, arg_1);
2189   pass_arg0(this, arg_0);
2190   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2191 }
2192 
2193 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2194   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2195   assert_different_registers(arg_1, c_rarg2);
2196   pass_arg2(this, arg_2);
2197   pass_arg1(this, arg_1);
2198   pass_arg0(this, arg_0);
2199   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2200 }
2201 
2202 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2203   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2204   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2205   assert_different_registers(arg_2, c_rarg3);
2206   pass_arg3(this, arg_3);
2207   pass_arg2(this, arg_2);
2208   pass_arg1(this, arg_1);
2209   pass_arg0(this, arg_0);
2210   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2211 }
2212 
2213 void MacroAssembler::null_check(Register reg, int offset) {
2214   if (needs_explicit_null_check(offset)) {
2215     // provoke OS null exception if reg is null by
2216     // accessing M[reg] w/o changing any registers
2217     // NOTE: this is plenty to provoke a segv
2218     ldr(zr, Address(reg));
2219   } else {
2220     // nothing to do, (later) access of M[reg + offset]
2221     // will provoke OS null exception if reg is null
2222   }
2223 }
2224 
2225 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2226   assert_different_registers(markword, rscratch2);
2227   mov(rscratch2, markWord::inline_type_mask_in_place);
2228   andr(markword, markword, rscratch2);
2229   mov(rscratch2, markWord::inline_type_pattern);
2230   cmp(markword, rscratch2);
2231   br(Assembler::EQ, is_inline_type);
2232 }
2233 
2234 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2235   assert_different_registers(tmp, rscratch1);
2236   if (can_be_null) {
2237     cbz(object, not_inline_type);
2238   }
2239   const int is_inline_type_mask = markWord::inline_type_pattern;
2240   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2241   mov(rscratch1, is_inline_type_mask);
2242   andr(tmp, tmp, rscratch1);
2243   cmp(tmp, rscratch1);
2244   br(Assembler::NE, not_inline_type);
2245 }
2246 
2247 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2248   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2249   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2250 }
2251 
2252 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2253   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2254   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2255 }
2256 
2257 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2258   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2259   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2260 }
2261 
2262 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
2263   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2264   tbnz(flags, ResolvedFieldEntry::has_null_marker_shift, has_null_marker);
2265 }
2266 
2267 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2268   Label test_mark_word;
2269   // load mark word
2270   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2271   // check displaced
2272   tst(temp_reg, markWord::unlocked_value);
2273   br(Assembler::NE, test_mark_word);
2274   // slow path use klass prototype
2275   load_prototype_header(temp_reg, oop);
2276 
2277   bind(test_mark_word);
2278   andr(temp_reg, temp_reg, test_bit);
2279   if (jmp_set) {
2280     cbnz(temp_reg, jmp_label);
2281   } else {
2282     cbz(temp_reg, jmp_label);
2283   }
2284 }
2285 
2286 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2287   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2288 }
2289 
2290 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2291                                                   Label&is_non_flat_array) {
2292   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2293 }
2294 
2295 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2296   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2297 }
2298 
2299 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2300   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2301 }
2302 
2303 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2304   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2305   br(Assembler::NE, is_flat_array);
2306 }
2307 
2308 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
2309   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2310   br(Assembler::EQ, is_non_flat_array);
2311 }
2312 
2313 // MacroAssembler protected routines needed to implement
2314 // public methods
2315 
2316 void MacroAssembler::mov(Register r, Address dest) {
2317   code_section()->relocate(pc(), dest.rspec());
2318   uint64_t imm64 = (uint64_t)dest.target();
2319   movptr(r, imm64);
2320 }
2321 
2322 // Move a constant pointer into r.  In AArch64 mode the virtual
2323 // address space is 48 bits in size, so we only need three
2324 // instructions to create a patchable instruction sequence that can
2325 // reach anywhere.
2326 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2327 #ifndef PRODUCT
2328   {
2329     char buffer[64];
2330     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2331     block_comment(buffer);
2332   }
2333 #endif
2334   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2335   movz(r, imm64 & 0xffff);
2336   imm64 >>= 16;
2337   movk(r, imm64 & 0xffff, 16);
2338   imm64 >>= 16;
2339   movk(r, imm64 & 0xffff, 32);
2340 }
2341 
2342 // Macro to mov replicated immediate to vector register.
2343 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2344 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2345 // Vd will get the following values for different arrangements in T
2346 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2347 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2348 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2349 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2350 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2351 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2352 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2353 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2354 // Clobbers rscratch1
2355 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2356   assert(T != T1Q, "unsupported");
2357   if (T == T1D || T == T2D) {
2358     int imm = operand_valid_for_movi_immediate(imm64, T);
2359     if (-1 != imm) {
2360       movi(Vd, T, imm);
2361     } else {
2362       mov(rscratch1, imm64);
2363       dup(Vd, T, rscratch1);
2364     }
2365     return;
2366   }
2367 
2368 #ifdef ASSERT
2369   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2370   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2371   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2372 #endif
2373   int shift = operand_valid_for_movi_immediate(imm64, T);
2374   uint32_t imm32 = imm64 & 0xffffffffULL;
2375   if (shift >= 0) {
2376     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2377   } else {
2378     movw(rscratch1, imm32);
2379     dup(Vd, T, rscratch1);
2380   }
2381 }
2382 
2383 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2384 {
2385 #ifndef PRODUCT
2386   {
2387     char buffer[64];
2388     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2389     block_comment(buffer);
2390   }
2391 #endif
2392   if (operand_valid_for_logical_immediate(false, imm64)) {
2393     orr(dst, zr, imm64);
2394   } else {
2395     // we can use a combination of MOVZ or MOVN with
2396     // MOVK to build up the constant
2397     uint64_t imm_h[4];
2398     int zero_count = 0;
2399     int neg_count = 0;
2400     int i;
2401     for (i = 0; i < 4; i++) {
2402       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2403       if (imm_h[i] == 0) {
2404         zero_count++;
2405       } else if (imm_h[i] == 0xffffL) {
2406         neg_count++;
2407       }
2408     }
2409     if (zero_count == 4) {
2410       // one MOVZ will do
2411       movz(dst, 0);
2412     } else if (neg_count == 4) {
2413       // one MOVN will do
2414       movn(dst, 0);
2415     } else if (zero_count == 3) {
2416       for (i = 0; i < 4; i++) {
2417         if (imm_h[i] != 0L) {
2418           movz(dst, (uint32_t)imm_h[i], (i << 4));
2419           break;
2420         }
2421       }
2422     } else if (neg_count == 3) {
2423       // one MOVN will do
2424       for (int i = 0; i < 4; i++) {
2425         if (imm_h[i] != 0xffffL) {
2426           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2427           break;
2428         }
2429       }
2430     } else if (zero_count == 2) {
2431       // one MOVZ and one MOVK will do
2432       for (i = 0; i < 3; i++) {
2433         if (imm_h[i] != 0L) {
2434           movz(dst, (uint32_t)imm_h[i], (i << 4));
2435           i++;
2436           break;
2437         }
2438       }
2439       for (;i < 4; i++) {
2440         if (imm_h[i] != 0L) {
2441           movk(dst, (uint32_t)imm_h[i], (i << 4));
2442         }
2443       }
2444     } else if (neg_count == 2) {
2445       // one MOVN and one MOVK will do
2446       for (i = 0; i < 4; i++) {
2447         if (imm_h[i] != 0xffffL) {
2448           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2449           i++;
2450           break;
2451         }
2452       }
2453       for (;i < 4; i++) {
2454         if (imm_h[i] != 0xffffL) {
2455           movk(dst, (uint32_t)imm_h[i], (i << 4));
2456         }
2457       }
2458     } else if (zero_count == 1) {
2459       // one MOVZ and two MOVKs will do
2460       for (i = 0; i < 4; i++) {
2461         if (imm_h[i] != 0L) {
2462           movz(dst, (uint32_t)imm_h[i], (i << 4));
2463           i++;
2464           break;
2465         }
2466       }
2467       for (;i < 4; i++) {
2468         if (imm_h[i] != 0x0L) {
2469           movk(dst, (uint32_t)imm_h[i], (i << 4));
2470         }
2471       }
2472     } else if (neg_count == 1) {
2473       // one MOVN and two MOVKs will do
2474       for (i = 0; i < 4; i++) {
2475         if (imm_h[i] != 0xffffL) {
2476           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2477           i++;
2478           break;
2479         }
2480       }
2481       for (;i < 4; i++) {
2482         if (imm_h[i] != 0xffffL) {
2483           movk(dst, (uint32_t)imm_h[i], (i << 4));
2484         }
2485       }
2486     } else {
2487       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2488       movz(dst, (uint32_t)imm_h[0], 0);
2489       for (i = 1; i < 4; i++) {
2490         movk(dst, (uint32_t)imm_h[i], (i << 4));
2491       }
2492     }
2493   }
2494 }
2495 
2496 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2497 {
2498 #ifndef PRODUCT
2499     {
2500       char buffer[64];
2501       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2502       block_comment(buffer);
2503     }
2504 #endif
2505   if (operand_valid_for_logical_immediate(true, imm32)) {
2506     orrw(dst, zr, imm32);
2507   } else {
2508     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2509     // constant
2510     uint32_t imm_h[2];
2511     imm_h[0] = imm32 & 0xffff;
2512     imm_h[1] = ((imm32 >> 16) & 0xffff);
2513     if (imm_h[0] == 0) {
2514       movzw(dst, imm_h[1], 16);
2515     } else if (imm_h[0] == 0xffff) {
2516       movnw(dst, imm_h[1] ^ 0xffff, 16);
2517     } else if (imm_h[1] == 0) {
2518       movzw(dst, imm_h[0], 0);
2519     } else if (imm_h[1] == 0xffff) {
2520       movnw(dst, imm_h[0] ^ 0xffff, 0);
2521     } else {
2522       // use a MOVZ and MOVK (makes it easier to debug)
2523       movzw(dst, imm_h[0], 0);
2524       movkw(dst, imm_h[1], 16);
2525     }
2526   }
2527 }
2528 
2529 // Form an address from base + offset in Rd.  Rd may or may
2530 // not actually be used: you must use the Address that is returned.
2531 // It is up to you to ensure that the shift provided matches the size
2532 // of your data.
2533 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2534   if (Address::offset_ok_for_immed(byte_offset, shift))
2535     // It fits; no need for any heroics
2536     return Address(base, byte_offset);
2537 
2538   // Don't do anything clever with negative or misaligned offsets
2539   unsigned mask = (1 << shift) - 1;
2540   if (byte_offset < 0 || byte_offset & mask) {
2541     mov(Rd, byte_offset);
2542     add(Rd, base, Rd);
2543     return Address(Rd);
2544   }
2545 
2546   // See if we can do this with two 12-bit offsets
2547   {
2548     uint64_t word_offset = byte_offset >> shift;
2549     uint64_t masked_offset = word_offset & 0xfff000;
2550     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2551         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2552       add(Rd, base, masked_offset << shift);
2553       word_offset -= masked_offset;
2554       return Address(Rd, word_offset << shift);
2555     }
2556   }
2557 
2558   // Do it the hard way
2559   mov(Rd, byte_offset);
2560   add(Rd, base, Rd);
2561   return Address(Rd);
2562 }
2563 
2564 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2565                                     bool want_remainder, Register scratch)
2566 {
2567   // Full implementation of Java idiv and irem.  The function
2568   // returns the (pc) offset of the div instruction - may be needed
2569   // for implicit exceptions.
2570   //
2571   // constraint : ra/rb =/= scratch
2572   //         normal case
2573   //
2574   // input : ra: dividend
2575   //         rb: divisor
2576   //
2577   // result: either
2578   //         quotient  (= ra idiv rb)
2579   //         remainder (= ra irem rb)
2580 
2581   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2582 
2583   int idivl_offset = offset();
2584   if (! want_remainder) {
2585     sdivw(result, ra, rb);
2586   } else {
2587     sdivw(scratch, ra, rb);
2588     Assembler::msubw(result, scratch, rb, ra);
2589   }
2590 
2591   return idivl_offset;
2592 }
2593 
2594 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2595                                     bool want_remainder, Register scratch)
2596 {
2597   // Full implementation of Java ldiv and lrem.  The function
2598   // returns the (pc) offset of the div instruction - may be needed
2599   // for implicit exceptions.
2600   //
2601   // constraint : ra/rb =/= scratch
2602   //         normal case
2603   //
2604   // input : ra: dividend
2605   //         rb: divisor
2606   //
2607   // result: either
2608   //         quotient  (= ra idiv rb)
2609   //         remainder (= ra irem rb)
2610 
2611   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2612 
2613   int idivq_offset = offset();
2614   if (! want_remainder) {
2615     sdiv(result, ra, rb);
2616   } else {
2617     sdiv(scratch, ra, rb);
2618     Assembler::msub(result, scratch, rb, ra);
2619   }
2620 
2621   return idivq_offset;
2622 }
2623 
2624 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2625   address prev = pc() - NativeMembar::instruction_size;
2626   address last = code()->last_insn();
2627   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2628     NativeMembar *bar = NativeMembar_at(prev);
2629     if (AlwaysMergeDMB) {
2630       bar->set_kind(bar->get_kind() | order_constraint);
2631       BLOCK_COMMENT("merged membar(always)");
2632       return;
2633     }
2634     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2635     // doing so would introduce a StoreLoad which the caller did not
2636     // intend
2637     if (bar->get_kind() == order_constraint
2638         || bar->get_kind() == AnyAny
2639         || order_constraint == AnyAny) {
2640       // We are merging two memory barrier instructions.  On AArch64 we
2641       // can do this simply by ORing them together.
2642       bar->set_kind(bar->get_kind() | order_constraint);
2643       BLOCK_COMMENT("merged membar");
2644       return;
2645     } else {
2646       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2647       // We need check the last 2 instructions
2648       address prev2 = prev - NativeMembar::instruction_size;
2649       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2650         NativeMembar *bar2 = NativeMembar_at(prev2);
2651         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2652         BLOCK_COMMENT("merged membar(elided)");
2653         return;
2654       }
2655     }
2656   }
2657   code()->set_last_insn(pc());
2658   dmb(Assembler::barrier(order_constraint));
2659 }
2660 
2661 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2662   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2663     merge_ldst(rt, adr, size_in_bytes, is_store);
2664     code()->clear_last_insn();
2665     return true;
2666   } else {
2667     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2668     const uint64_t mask = size_in_bytes - 1;
2669     if (adr.getMode() == Address::base_plus_offset &&
2670         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2671       code()->set_last_insn(pc());
2672     }
2673     return false;
2674   }
2675 }
2676 
2677 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2678   // We always try to merge two adjacent loads into one ldp.
2679   if (!try_merge_ldst(Rx, adr, 8, false)) {
2680     Assembler::ldr(Rx, adr);
2681   }
2682 }
2683 
2684 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2685   // We always try to merge two adjacent loads into one ldp.
2686   if (!try_merge_ldst(Rw, adr, 4, false)) {
2687     Assembler::ldrw(Rw, adr);
2688   }
2689 }
2690 
2691 void MacroAssembler::str(Register Rx, const Address &adr) {
2692   // We always try to merge two adjacent stores into one stp.
2693   if (!try_merge_ldst(Rx, adr, 8, true)) {
2694     Assembler::str(Rx, adr);
2695   }
2696 }
2697 
2698 void MacroAssembler::strw(Register Rw, const Address &adr) {
2699   // We always try to merge two adjacent stores into one stp.
2700   if (!try_merge_ldst(Rw, adr, 4, true)) {
2701     Assembler::strw(Rw, adr);
2702   }
2703 }
2704 
2705 // MacroAssembler routines found actually to be needed
2706 
2707 void MacroAssembler::push(Register src)
2708 {
2709   str(src, Address(pre(esp, -1 * wordSize)));
2710 }
2711 
2712 void MacroAssembler::pop(Register dst)
2713 {
2714   ldr(dst, Address(post(esp, 1 * wordSize)));
2715 }
2716 
2717 // Note: load_unsigned_short used to be called load_unsigned_word.
2718 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2719   int off = offset();
2720   ldrh(dst, src);
2721   return off;
2722 }
2723 
2724 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2725   int off = offset();
2726   ldrb(dst, src);
2727   return off;
2728 }
2729 
2730 int MacroAssembler::load_signed_short(Register dst, Address src) {
2731   int off = offset();
2732   ldrsh(dst, src);
2733   return off;
2734 }
2735 
2736 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2737   int off = offset();
2738   ldrsb(dst, src);
2739   return off;
2740 }
2741 
2742 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2743   int off = offset();
2744   ldrshw(dst, src);
2745   return off;
2746 }
2747 
2748 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2749   int off = offset();
2750   ldrsbw(dst, src);
2751   return off;
2752 }
2753 
2754 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2755   switch (size_in_bytes) {
2756   case  8:  ldr(dst, src); break;
2757   case  4:  ldrw(dst, src); break;
2758   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2759   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2760   default:  ShouldNotReachHere();
2761   }
2762 }
2763 
2764 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2765   switch (size_in_bytes) {
2766   case  8:  str(src, dst); break;
2767   case  4:  strw(src, dst); break;
2768   case  2:  strh(src, dst); break;
2769   case  1:  strb(src, dst); break;
2770   default:  ShouldNotReachHere();
2771   }
2772 }
2773 
2774 void MacroAssembler::decrementw(Register reg, int value)
2775 {
2776   if (value < 0)  { incrementw(reg, -value);      return; }
2777   if (value == 0) {                               return; }
2778   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2779   /* else */ {
2780     guarantee(reg != rscratch2, "invalid dst for register decrement");
2781     movw(rscratch2, (unsigned)value);
2782     subw(reg, reg, rscratch2);
2783   }
2784 }
2785 
2786 void MacroAssembler::decrement(Register reg, int value)
2787 {
2788   if (value < 0)  { increment(reg, -value);      return; }
2789   if (value == 0) {                              return; }
2790   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2791   /* else */ {
2792     assert(reg != rscratch2, "invalid dst for register decrement");
2793     mov(rscratch2, (uint64_t)value);
2794     sub(reg, reg, rscratch2);
2795   }
2796 }
2797 
2798 void MacroAssembler::decrementw(Address dst, int value)
2799 {
2800   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2801   if (dst.getMode() == Address::literal) {
2802     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2803     lea(rscratch2, dst);
2804     dst = Address(rscratch2);
2805   }
2806   ldrw(rscratch1, dst);
2807   decrementw(rscratch1, value);
2808   strw(rscratch1, dst);
2809 }
2810 
2811 void MacroAssembler::decrement(Address dst, int value)
2812 {
2813   assert(!dst.uses(rscratch1), "invalid address for decrement");
2814   if (dst.getMode() == Address::literal) {
2815     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2816     lea(rscratch2, dst);
2817     dst = Address(rscratch2);
2818   }
2819   ldr(rscratch1, dst);
2820   decrement(rscratch1, value);
2821   str(rscratch1, dst);
2822 }
2823 
2824 void MacroAssembler::incrementw(Register reg, int value)
2825 {
2826   if (value < 0)  { decrementw(reg, -value);      return; }
2827   if (value == 0) {                               return; }
2828   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2829   /* else */ {
2830     assert(reg != rscratch2, "invalid dst for register increment");
2831     movw(rscratch2, (unsigned)value);
2832     addw(reg, reg, rscratch2);
2833   }
2834 }
2835 
2836 void MacroAssembler::increment(Register reg, int value)
2837 {
2838   if (value < 0)  { decrement(reg, -value);      return; }
2839   if (value == 0) {                              return; }
2840   if (value < (1 << 12)) { add(reg, reg, value); return; }
2841   /* else */ {
2842     assert(reg != rscratch2, "invalid dst for register increment");
2843     movw(rscratch2, (unsigned)value);
2844     add(reg, reg, rscratch2);
2845   }
2846 }
2847 
2848 void MacroAssembler::incrementw(Address dst, int value)
2849 {
2850   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2851   if (dst.getMode() == Address::literal) {
2852     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2853     lea(rscratch2, dst);
2854     dst = Address(rscratch2);
2855   }
2856   ldrw(rscratch1, dst);
2857   incrementw(rscratch1, value);
2858   strw(rscratch1, dst);
2859 }
2860 
2861 void MacroAssembler::increment(Address dst, int value)
2862 {
2863   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2864   if (dst.getMode() == Address::literal) {
2865     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2866     lea(rscratch2, dst);
2867     dst = Address(rscratch2);
2868   }
2869   ldr(rscratch1, dst);
2870   increment(rscratch1, value);
2871   str(rscratch1, dst);
2872 }
2873 
2874 // Push lots of registers in the bit set supplied.  Don't push sp.
2875 // Return the number of words pushed
2876 int MacroAssembler::push(unsigned int bitset, Register stack) {
2877   int words_pushed = 0;
2878 
2879   // Scan bitset to accumulate register pairs
2880   unsigned char regs[32];
2881   int count = 0;
2882   for (int reg = 0; reg <= 30; reg++) {
2883     if (1 & bitset)
2884       regs[count++] = reg;
2885     bitset >>= 1;
2886   }
2887   regs[count++] = zr->raw_encoding();
2888   count &= ~1;  // Only push an even number of regs
2889 
2890   if (count) {
2891     stp(as_Register(regs[0]), as_Register(regs[1]),
2892        Address(pre(stack, -count * wordSize)));
2893     words_pushed += 2;
2894   }
2895   for (int i = 2; i < count; i += 2) {
2896     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2897        Address(stack, i * wordSize));
2898     words_pushed += 2;
2899   }
2900 
2901   assert(words_pushed == count, "oops, pushed != count");
2902 
2903   return count;
2904 }
2905 
2906 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2907   int words_pushed = 0;
2908 
2909   // Scan bitset to accumulate register pairs
2910   unsigned char regs[32];
2911   int count = 0;
2912   for (int reg = 0; reg <= 30; reg++) {
2913     if (1 & bitset)
2914       regs[count++] = reg;
2915     bitset >>= 1;
2916   }
2917   regs[count++] = zr->raw_encoding();
2918   count &= ~1;
2919 
2920   for (int i = 2; i < count; i += 2) {
2921     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2922        Address(stack, i * wordSize));
2923     words_pushed += 2;
2924   }
2925   if (count) {
2926     ldp(as_Register(regs[0]), as_Register(regs[1]),
2927        Address(post(stack, count * wordSize)));
2928     words_pushed += 2;
2929   }
2930 
2931   assert(words_pushed == count, "oops, pushed != count");
2932 
2933   return count;
2934 }
2935 
2936 // Push lots of registers in the bit set supplied.  Don't push sp.
2937 // Return the number of dwords pushed
2938 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2939   int words_pushed = 0;
2940   bool use_sve = false;
2941   int sve_vector_size_in_bytes = 0;
2942 
2943 #ifdef COMPILER2
2944   use_sve = Matcher::supports_scalable_vector();
2945   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2946 #endif
2947 
2948   // Scan bitset to accumulate register pairs
2949   unsigned char regs[32];
2950   int count = 0;
2951   for (int reg = 0; reg <= 31; reg++) {
2952     if (1 & bitset)
2953       regs[count++] = reg;
2954     bitset >>= 1;
2955   }
2956 
2957   if (count == 0) {
2958     return 0;
2959   }
2960 
2961   if (mode == PushPopFull) {
2962     if (use_sve && sve_vector_size_in_bytes > 16) {
2963       mode = PushPopSVE;
2964     } else {
2965       mode = PushPopNeon;
2966     }
2967   }
2968 
2969 #ifndef PRODUCT
2970   {
2971     char buffer[48];
2972     if (mode == PushPopSVE) {
2973       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
2974     } else if (mode == PushPopNeon) {
2975       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
2976     } else {
2977       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
2978     }
2979     block_comment(buffer);
2980   }
2981 #endif
2982 
2983   if (mode == PushPopSVE) {
2984     sub(stack, stack, sve_vector_size_in_bytes * count);
2985     for (int i = 0; i < count; i++) {
2986       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
2987     }
2988     return count * sve_vector_size_in_bytes / 8;
2989   }
2990 
2991   if (mode == PushPopNeon) {
2992     if (count == 1) {
2993       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
2994       return 2;
2995     }
2996 
2997     bool odd = (count & 1) == 1;
2998     int push_slots = count + (odd ? 1 : 0);
2999 
3000     // Always pushing full 128 bit registers.
3001     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3002     words_pushed += 2;
3003 
3004     for (int i = 2; i + 1 < count; i += 2) {
3005       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3006       words_pushed += 2;
3007     }
3008 
3009     if (odd) {
3010       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3011       words_pushed++;
3012     }
3013 
3014     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3015     return count * 2;
3016   }
3017 
3018   if (mode == PushPopFp) {
3019     bool odd = (count & 1) == 1;
3020     int push_slots = count + (odd ? 1 : 0);
3021 
3022     if (count == 1) {
3023       // Stack pointer must be 16 bytes aligned
3024       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3025       return 1;
3026     }
3027 
3028     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3029     words_pushed += 2;
3030 
3031     for (int i = 2; i + 1 < count; i += 2) {
3032       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3033       words_pushed += 2;
3034     }
3035 
3036     if (odd) {
3037       // Stack pointer must be 16 bytes aligned
3038       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3039       words_pushed++;
3040     }
3041 
3042     assert(words_pushed == count, "oops, pushed != count");
3043 
3044     return count;
3045   }
3046 
3047   return 0;
3048 }
3049 
3050 // Return the number of dwords popped
3051 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3052   int words_pushed = 0;
3053   bool use_sve = false;
3054   int sve_vector_size_in_bytes = 0;
3055 
3056 #ifdef COMPILER2
3057   use_sve = Matcher::supports_scalable_vector();
3058   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3059 #endif
3060   // Scan bitset to accumulate register pairs
3061   unsigned char regs[32];
3062   int count = 0;
3063   for (int reg = 0; reg <= 31; reg++) {
3064     if (1 & bitset)
3065       regs[count++] = reg;
3066     bitset >>= 1;
3067   }
3068 
3069   if (count == 0) {
3070     return 0;
3071   }
3072 
3073   if (mode == PushPopFull) {
3074     if (use_sve && sve_vector_size_in_bytes > 16) {
3075       mode = PushPopSVE;
3076     } else {
3077       mode = PushPopNeon;
3078     }
3079   }
3080 
3081 #ifndef PRODUCT
3082   {
3083     char buffer[48];
3084     if (mode == PushPopSVE) {
3085       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3086     } else if (mode == PushPopNeon) {
3087       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3088     } else {
3089       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3090     }
3091     block_comment(buffer);
3092   }
3093 #endif
3094 
3095   if (mode == PushPopSVE) {
3096     for (int i = count - 1; i >= 0; i--) {
3097       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3098     }
3099     add(stack, stack, sve_vector_size_in_bytes * count);
3100     return count * sve_vector_size_in_bytes / 8;
3101   }
3102 
3103   if (mode == PushPopNeon) {
3104     if (count == 1) {
3105       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3106       return 2;
3107     }
3108 
3109     bool odd = (count & 1) == 1;
3110     int push_slots = count + (odd ? 1 : 0);
3111 
3112     if (odd) {
3113       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3114       words_pushed++;
3115     }
3116 
3117     for (int i = 2; i + 1 < count; i += 2) {
3118       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3119       words_pushed += 2;
3120     }
3121 
3122     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3123     words_pushed += 2;
3124 
3125     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3126 
3127     return count * 2;
3128   }
3129 
3130   if (mode == PushPopFp) {
3131     bool odd = (count & 1) == 1;
3132     int push_slots = count + (odd ? 1 : 0);
3133 
3134     if (count == 1) {
3135       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3136       return 1;
3137     }
3138 
3139     if (odd) {
3140       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3141       words_pushed++;
3142     }
3143 
3144     for (int i = 2; i + 1 < count; i += 2) {
3145       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3146       words_pushed += 2;
3147     }
3148 
3149     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3150     words_pushed += 2;
3151 
3152     assert(words_pushed == count, "oops, pushed != count");
3153 
3154     return count;
3155   }
3156 
3157   return 0;
3158 }
3159 
3160 // Return the number of dwords pushed
3161 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3162   bool use_sve = false;
3163   int sve_predicate_size_in_slots = 0;
3164 
3165 #ifdef COMPILER2
3166   use_sve = Matcher::supports_scalable_vector();
3167   if (use_sve) {
3168     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3169   }
3170 #endif
3171 
3172   if (!use_sve) {
3173     return 0;
3174   }
3175 
3176   unsigned char regs[PRegister::number_of_registers];
3177   int count = 0;
3178   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3179     if (1 & bitset)
3180       regs[count++] = reg;
3181     bitset >>= 1;
3182   }
3183 
3184   if (count == 0) {
3185     return 0;
3186   }
3187 
3188   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3189                                   VMRegImpl::stack_slot_size * count, 16);
3190   sub(stack, stack, total_push_bytes);
3191   for (int i = 0; i < count; i++) {
3192     sve_str(as_PRegister(regs[i]), Address(stack, i));
3193   }
3194   return total_push_bytes / 8;
3195 }
3196 
3197 // Return the number of dwords popped
3198 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3199   bool use_sve = false;
3200   int sve_predicate_size_in_slots = 0;
3201 
3202 #ifdef COMPILER2
3203   use_sve = Matcher::supports_scalable_vector();
3204   if (use_sve) {
3205     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3206   }
3207 #endif
3208 
3209   if (!use_sve) {
3210     return 0;
3211   }
3212 
3213   unsigned char regs[PRegister::number_of_registers];
3214   int count = 0;
3215   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3216     if (1 & bitset)
3217       regs[count++] = reg;
3218     bitset >>= 1;
3219   }
3220 
3221   if (count == 0) {
3222     return 0;
3223   }
3224 
3225   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3226                                  VMRegImpl::stack_slot_size * count, 16);
3227   for (int i = count - 1; i >= 0; i--) {
3228     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3229   }
3230   add(stack, stack, total_pop_bytes);
3231   return total_pop_bytes / 8;
3232 }
3233 
3234 #ifdef ASSERT
3235 void MacroAssembler::verify_heapbase(const char* msg) {
3236 #if 0
3237   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3238   assert (Universe::heap() != nullptr, "java heap should be initialized");
3239   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3240     // rheapbase is allocated as general register
3241     return;
3242   }
3243   if (CheckCompressedOops) {
3244     Label ok;
3245     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3246     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3247     br(Assembler::EQ, ok);
3248     stop(msg);
3249     bind(ok);
3250     pop(1 << rscratch1->encoding(), sp);
3251   }
3252 #endif
3253 }
3254 #endif
3255 
3256 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3257   assert_different_registers(value, tmp1, tmp2);
3258   Label done, tagged, weak_tagged;
3259 
3260   cbz(value, done);           // Use null as-is.
3261   tst(value, JNIHandles::tag_mask); // Test for tag.
3262   br(Assembler::NE, tagged);
3263 
3264   // Resolve local handle
3265   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3266   verify_oop(value);
3267   b(done);
3268 
3269   bind(tagged);
3270   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3271   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3272 
3273   // Resolve global handle
3274   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3275   verify_oop(value);
3276   b(done);
3277 
3278   bind(weak_tagged);
3279   // Resolve jweak.
3280   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3281                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3282   verify_oop(value);
3283 
3284   bind(done);
3285 }
3286 
3287 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3288   assert_different_registers(value, tmp1, tmp2);
3289   Label done;
3290 
3291   cbz(value, done);           // Use null as-is.
3292 
3293 #ifdef ASSERT
3294   {
3295     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3296     Label valid_global_tag;
3297     tbnz(value, 1, valid_global_tag); // Test for global tag
3298     stop("non global jobject using resolve_global_jobject");
3299     bind(valid_global_tag);
3300   }
3301 #endif
3302 
3303   // Resolve global handle
3304   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3305   verify_oop(value);
3306 
3307   bind(done);
3308 }
3309 
3310 void MacroAssembler::stop(const char* msg) {
3311   // Skip AOT caching C strings in scratch buffer.
3312   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3313   BLOCK_COMMENT(str);
3314   // load msg into r0 so we can access it from the signal handler
3315   // ExternalAddress enables saving and restoring via the code cache
3316   lea(c_rarg0, ExternalAddress((address) str));
3317   dcps1(0xdeae);
3318 }
3319 
3320 void MacroAssembler::unimplemented(const char* what) {
3321   const char* buf = nullptr;
3322   {
3323     ResourceMark rm;
3324     stringStream ss;
3325     ss.print("unimplemented: %s", what);
3326     buf = code_string(ss.as_string());
3327   }
3328   stop(buf);
3329 }
3330 
3331 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3332 #ifdef ASSERT
3333   Label OK;
3334   br(cc, OK);
3335   stop(msg);
3336   bind(OK);
3337 #endif
3338 }
3339 
3340 // If a constant does not fit in an immediate field, generate some
3341 // number of MOV instructions and then perform the operation.
3342 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3343                                            add_sub_imm_insn insn1,
3344                                            add_sub_reg_insn insn2,
3345                                            bool is32) {
3346   assert(Rd != zr, "Rd = zr and not setting flags?");
3347   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3348   if (fits) {
3349     (this->*insn1)(Rd, Rn, imm);
3350   } else {
3351     if (g_uabs(imm) < (1 << 24)) {
3352        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3353        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3354     } else {
3355        assert_different_registers(Rd, Rn);
3356        mov(Rd, imm);
3357        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3358     }
3359   }
3360 }
3361 
3362 // Separate vsn which sets the flags. Optimisations are more restricted
3363 // because we must set the flags correctly.
3364 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3365                                              add_sub_imm_insn insn1,
3366                                              add_sub_reg_insn insn2,
3367                                              bool is32) {
3368   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3369   if (fits) {
3370     (this->*insn1)(Rd, Rn, imm);
3371   } else {
3372     assert_different_registers(Rd, Rn);
3373     assert(Rd != zr, "overflow in immediate operand");
3374     mov(Rd, imm);
3375     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3376   }
3377 }
3378 
3379 
3380 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3381   if (increment.is_register()) {
3382     add(Rd, Rn, increment.as_register());
3383   } else {
3384     add(Rd, Rn, increment.as_constant());
3385   }
3386 }
3387 
3388 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3389   if (increment.is_register()) {
3390     addw(Rd, Rn, increment.as_register());
3391   } else {
3392     addw(Rd, Rn, increment.as_constant());
3393   }
3394 }
3395 
3396 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3397   if (decrement.is_register()) {
3398     sub(Rd, Rn, decrement.as_register());
3399   } else {
3400     sub(Rd, Rn, decrement.as_constant());
3401   }
3402 }
3403 
3404 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3405   if (decrement.is_register()) {
3406     subw(Rd, Rn, decrement.as_register());
3407   } else {
3408     subw(Rd, Rn, decrement.as_constant());
3409   }
3410 }
3411 
3412 void MacroAssembler::reinit_heapbase()
3413 {
3414   if (UseCompressedOops) {
3415     if (Universe::is_fully_initialized()) {
3416       mov(rheapbase, CompressedOops::base());
3417     } else {
3418       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3419       ldr(rheapbase, Address(rheapbase));
3420     }
3421   }
3422 }
3423 
3424 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3425 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3426 // Pass a register for the result, otherwise pass noreg.
3427 
3428 // Clobbers rscratch1
3429 void MacroAssembler::cmpxchg(Register addr, Register expected,
3430                              Register new_val,
3431                              enum operand_size size,
3432                              bool acquire, bool release,
3433                              bool weak,
3434                              Register result) {
3435   if (result == noreg)  result = rscratch1;
3436   BLOCK_COMMENT("cmpxchg {");
3437   if (UseLSE) {
3438     mov(result, expected);
3439     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3440     compare_eq(result, expected, size);
3441 #ifdef ASSERT
3442     // Poison rscratch1 which is written on !UseLSE branch
3443     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3444 #endif
3445   } else {
3446     Label retry_load, done;
3447     prfm(Address(addr), PSTL1STRM);
3448     bind(retry_load);
3449     load_exclusive(result, addr, size, acquire);
3450     compare_eq(result, expected, size);
3451     br(Assembler::NE, done);
3452     store_exclusive(rscratch1, new_val, addr, size, release);
3453     if (weak) {
3454       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3455     } else {
3456       cbnzw(rscratch1, retry_load);
3457     }
3458     bind(done);
3459   }
3460   BLOCK_COMMENT("} cmpxchg");
3461 }
3462 
3463 // A generic comparison. Only compares for equality, clobbers rscratch1.
3464 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3465   if (size == xword) {
3466     cmp(rm, rn);
3467   } else if (size == word) {
3468     cmpw(rm, rn);
3469   } else if (size == halfword) {
3470     eorw(rscratch1, rm, rn);
3471     ands(zr, rscratch1, 0xffff);
3472   } else if (size == byte) {
3473     eorw(rscratch1, rm, rn);
3474     ands(zr, rscratch1, 0xff);
3475   } else {
3476     ShouldNotReachHere();
3477   }
3478 }
3479 
3480 
3481 static bool different(Register a, RegisterOrConstant b, Register c) {
3482   if (b.is_constant())
3483     return a != c;
3484   else
3485     return a != b.as_register() && a != c && b.as_register() != c;
3486 }
3487 
3488 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3489 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3490   if (UseLSE) {                                                         \
3491     prev = prev->is_valid() ? prev : zr;                                \
3492     if (incr.is_register()) {                                           \
3493       AOP(sz, incr.as_register(), prev, addr);                          \
3494     } else {                                                            \
3495       mov(rscratch2, incr.as_constant());                               \
3496       AOP(sz, rscratch2, prev, addr);                                   \
3497     }                                                                   \
3498     return;                                                             \
3499   }                                                                     \
3500   Register result = rscratch2;                                          \
3501   if (prev->is_valid())                                                 \
3502     result = different(prev, incr, addr) ? prev : rscratch2;            \
3503                                                                         \
3504   Label retry_load;                                                     \
3505   prfm(Address(addr), PSTL1STRM);                                       \
3506   bind(retry_load);                                                     \
3507   LDXR(result, addr);                                                   \
3508   OP(rscratch1, result, incr);                                          \
3509   STXR(rscratch2, rscratch1, addr);                                     \
3510   cbnzw(rscratch2, retry_load);                                         \
3511   if (prev->is_valid() && prev != result) {                             \
3512     IOP(prev, rscratch1, incr);                                         \
3513   }                                                                     \
3514 }
3515 
3516 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3517 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3518 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3519 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3520 
3521 #undef ATOMIC_OP
3522 
3523 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3524 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3525   if (UseLSE) {                                                         \
3526     prev = prev->is_valid() ? prev : zr;                                \
3527     AOP(sz, newv, prev, addr);                                          \
3528     return;                                                             \
3529   }                                                                     \
3530   Register result = rscratch2;                                          \
3531   if (prev->is_valid())                                                 \
3532     result = different(prev, newv, addr) ? prev : rscratch2;            \
3533                                                                         \
3534   Label retry_load;                                                     \
3535   prfm(Address(addr), PSTL1STRM);                                       \
3536   bind(retry_load);                                                     \
3537   LDXR(result, addr);                                                   \
3538   STXR(rscratch1, newv, addr);                                          \
3539   cbnzw(rscratch1, retry_load);                                         \
3540   if (prev->is_valid() && prev != result)                               \
3541     mov(prev, result);                                                  \
3542 }
3543 
3544 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3545 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3546 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3547 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3548 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3549 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3550 
3551 #undef ATOMIC_XCHG
3552 
3553 #ifndef PRODUCT
3554 extern "C" void findpc(intptr_t x);
3555 #endif
3556 
3557 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3558 {
3559   // In order to get locks to work, we need to fake a in_VM state
3560   if (ShowMessageBoxOnError ) {
3561     JavaThread* thread = JavaThread::current();
3562     JavaThreadState saved_state = thread->thread_state();
3563     thread->set_thread_state(_thread_in_vm);
3564 #ifndef PRODUCT
3565     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3566       ttyLocker ttyl;
3567       BytecodeCounter::print();
3568     }
3569 #endif
3570     if (os::message_box(msg, "Execution stopped, print registers?")) {
3571       ttyLocker ttyl;
3572       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3573 #ifndef PRODUCT
3574       tty->cr();
3575       findpc(pc);
3576       tty->cr();
3577 #endif
3578       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3579       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3580       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3581       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3582       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3583       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3584       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3585       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3586       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3587       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3588       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3589       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3590       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3591       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3592       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3593       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3594       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3595       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3596       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3597       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3598       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3599       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3600       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3601       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3602       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3603       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3604       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3605       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3606       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3607       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3608       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3609       BREAKPOINT;
3610     }
3611   }
3612   fatal("DEBUG MESSAGE: %s", msg);
3613 }
3614 
3615 RegSet MacroAssembler::call_clobbered_gp_registers() {
3616   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3617 #ifndef R18_RESERVED
3618   regs += r18_tls;
3619 #endif
3620   return regs;
3621 }
3622 
3623 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3624   int step = 4 * wordSize;
3625   push(call_clobbered_gp_registers() - exclude, sp);
3626   sub(sp, sp, step);
3627   mov(rscratch1, -step);
3628   // Push v0-v7, v16-v31.
3629   for (int i = 31; i>= 4; i -= 4) {
3630     if (i <= v7->encoding() || i >= v16->encoding())
3631       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3632           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3633   }
3634   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3635       as_FloatRegister(3), T1D, Address(sp));
3636 }
3637 
3638 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3639   for (int i = 0; i < 32; i += 4) {
3640     if (i <= v7->encoding() || i >= v16->encoding())
3641       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3642           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3643   }
3644 
3645   reinitialize_ptrue();
3646 
3647   pop(call_clobbered_gp_registers() - exclude, sp);
3648 }
3649 
3650 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3651                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3652   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3653   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3654     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3655     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3656       sve_str(as_FloatRegister(i), Address(sp, i));
3657     }
3658   } else {
3659     int step = (save_vectors ? 8 : 4) * wordSize;
3660     mov(rscratch1, -step);
3661     sub(sp, sp, step);
3662     for (int i = 28; i >= 4; i -= 4) {
3663       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3664           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3665     }
3666     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3667   }
3668   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3669     sub(sp, sp, total_predicate_in_bytes);
3670     for (int i = 0; i < PRegister::number_of_registers; i++) {
3671       sve_str(as_PRegister(i), Address(sp, i));
3672     }
3673   }
3674 }
3675 
3676 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3677                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3678   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3679     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3680       sve_ldr(as_PRegister(i), Address(sp, i));
3681     }
3682     add(sp, sp, total_predicate_in_bytes);
3683   }
3684   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3685     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3686       sve_ldr(as_FloatRegister(i), Address(sp, i));
3687     }
3688     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3689   } else {
3690     int step = (restore_vectors ? 8 : 4) * wordSize;
3691     for (int i = 0; i <= 28; i += 4)
3692       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3693           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3694   }
3695 
3696   // We may use predicate registers and rely on ptrue with SVE,
3697   // regardless of wide vector (> 8 bytes) used or not.
3698   if (use_sve) {
3699     reinitialize_ptrue();
3700   }
3701 
3702   // integer registers except lr & sp
3703   pop(RegSet::range(r0, r17), sp);
3704 #ifdef R18_RESERVED
3705   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3706   pop(RegSet::range(r20, r29), sp);
3707 #else
3708   pop(RegSet::range(r18_tls, r29), sp);
3709 #endif
3710 }
3711 
3712 /**
3713  * Helpers for multiply_to_len().
3714  */
3715 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3716                                      Register src1, Register src2) {
3717   adds(dest_lo, dest_lo, src1);
3718   adc(dest_hi, dest_hi, zr);
3719   adds(dest_lo, dest_lo, src2);
3720   adc(final_dest_hi, dest_hi, zr);
3721 }
3722 
3723 // Generate an address from (r + r1 extend offset).  "size" is the
3724 // size of the operand.  The result may be in rscratch2.
3725 Address MacroAssembler::offsetted_address(Register r, Register r1,
3726                                           Address::extend ext, int offset, int size) {
3727   if (offset || (ext.shift() % size != 0)) {
3728     lea(rscratch2, Address(r, r1, ext));
3729     return Address(rscratch2, offset);
3730   } else {
3731     return Address(r, r1, ext);
3732   }
3733 }
3734 
3735 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3736 {
3737   assert(offset >= 0, "spill to negative address?");
3738   // Offset reachable ?
3739   //   Not aligned - 9 bits signed offset
3740   //   Aligned - 12 bits unsigned offset shifted
3741   Register base = sp;
3742   if ((offset & (size-1)) && offset >= (1<<8)) {
3743     add(tmp, base, offset & ((1<<12)-1));
3744     base = tmp;
3745     offset &= -1u<<12;
3746   }
3747 
3748   if (offset >= (1<<12) * size) {
3749     add(tmp, base, offset & (((1<<12)-1)<<12));
3750     base = tmp;
3751     offset &= ~(((1<<12)-1)<<12);
3752   }
3753 
3754   return Address(base, offset);
3755 }
3756 
3757 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3758   assert(offset >= 0, "spill to negative address?");
3759 
3760   Register base = sp;
3761 
3762   // An immediate offset in the range 0 to 255 which is multiplied
3763   // by the current vector or predicate register size in bytes.
3764   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3765     return Address(base, offset / sve_reg_size_in_bytes);
3766   }
3767 
3768   add(tmp, base, offset);
3769   return Address(tmp);
3770 }
3771 
3772 // Checks whether offset is aligned.
3773 // Returns true if it is, else false.
3774 bool MacroAssembler::merge_alignment_check(Register base,
3775                                            size_t size,
3776                                            int64_t cur_offset,
3777                                            int64_t prev_offset) const {
3778   if (AvoidUnalignedAccesses) {
3779     if (base == sp) {
3780       // Checks whether low offset if aligned to pair of registers.
3781       int64_t pair_mask = size * 2 - 1;
3782       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3783       return (offset & pair_mask) == 0;
3784     } else { // If base is not sp, we can't guarantee the access is aligned.
3785       return false;
3786     }
3787   } else {
3788     int64_t mask = size - 1;
3789     // Load/store pair instruction only supports element size aligned offset.
3790     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3791   }
3792 }
3793 
3794 // Checks whether current and previous loads/stores can be merged.
3795 // Returns true if it can be merged, else false.
3796 bool MacroAssembler::ldst_can_merge(Register rt,
3797                                     const Address &adr,
3798                                     size_t cur_size_in_bytes,
3799                                     bool is_store) const {
3800   address prev = pc() - NativeInstruction::instruction_size;
3801   address last = code()->last_insn();
3802 
3803   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3804     return false;
3805   }
3806 
3807   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3808     return false;
3809   }
3810 
3811   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3812   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3813 
3814   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3815   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3816 
3817   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3818     return false;
3819   }
3820 
3821   int64_t max_offset = 63 * prev_size_in_bytes;
3822   int64_t min_offset = -64 * prev_size_in_bytes;
3823 
3824   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3825 
3826   // Only same base can be merged.
3827   if (adr.base() != prev_ldst->base()) {
3828     return false;
3829   }
3830 
3831   int64_t cur_offset = adr.offset();
3832   int64_t prev_offset = prev_ldst->offset();
3833   size_t diff = abs(cur_offset - prev_offset);
3834   if (diff != prev_size_in_bytes) {
3835     return false;
3836   }
3837 
3838   // Following cases can not be merged:
3839   // ldr x2, [x2, #8]
3840   // ldr x3, [x2, #16]
3841   // or:
3842   // ldr x2, [x3, #8]
3843   // ldr x2, [x3, #16]
3844   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3845   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3846     return false;
3847   }
3848 
3849   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3850   // Offset range must be in ldp/stp instruction's range.
3851   if (low_offset > max_offset || low_offset < min_offset) {
3852     return false;
3853   }
3854 
3855   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3856     return true;
3857   }
3858 
3859   return false;
3860 }
3861 
3862 // Merge current load/store with previous load/store into ldp/stp.
3863 void MacroAssembler::merge_ldst(Register rt,
3864                                 const Address &adr,
3865                                 size_t cur_size_in_bytes,
3866                                 bool is_store) {
3867 
3868   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3869 
3870   Register rt_low, rt_high;
3871   address prev = pc() - NativeInstruction::instruction_size;
3872   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3873 
3874   int64_t offset;
3875 
3876   if (adr.offset() < prev_ldst->offset()) {
3877     offset = adr.offset();
3878     rt_low = rt;
3879     rt_high = prev_ldst->target();
3880   } else {
3881     offset = prev_ldst->offset();
3882     rt_low = prev_ldst->target();
3883     rt_high = rt;
3884   }
3885 
3886   Address adr_p = Address(prev_ldst->base(), offset);
3887   // Overwrite previous generated binary.
3888   code_section()->set_end(prev);
3889 
3890   const size_t sz = prev_ldst->size_in_bytes();
3891   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3892   if (!is_store) {
3893     BLOCK_COMMENT("merged ldr pair");
3894     if (sz == 8) {
3895       ldp(rt_low, rt_high, adr_p);
3896     } else {
3897       ldpw(rt_low, rt_high, adr_p);
3898     }
3899   } else {
3900     BLOCK_COMMENT("merged str pair");
3901     if (sz == 8) {
3902       stp(rt_low, rt_high, adr_p);
3903     } else {
3904       stpw(rt_low, rt_high, adr_p);
3905     }
3906   }
3907 }
3908 
3909 /**
3910  * Multiply 64 bit by 64 bit first loop.
3911  */
3912 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3913                                            Register y, Register y_idx, Register z,
3914                                            Register carry, Register product,
3915                                            Register idx, Register kdx) {
3916   //
3917   //  jlong carry, x[], y[], z[];
3918   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3919   //    huge_128 product = y[idx] * x[xstart] + carry;
3920   //    z[kdx] = (jlong)product;
3921   //    carry  = (jlong)(product >>> 64);
3922   //  }
3923   //  z[xstart] = carry;
3924   //
3925 
3926   Label L_first_loop, L_first_loop_exit;
3927   Label L_one_x, L_one_y, L_multiply;
3928 
3929   subsw(xstart, xstart, 1);
3930   br(Assembler::MI, L_one_x);
3931 
3932   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3933   ldr(x_xstart, Address(rscratch1));
3934   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3935 
3936   bind(L_first_loop);
3937   subsw(idx, idx, 1);
3938   br(Assembler::MI, L_first_loop_exit);
3939   subsw(idx, idx, 1);
3940   br(Assembler::MI, L_one_y);
3941   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3942   ldr(y_idx, Address(rscratch1));
3943   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
3944   bind(L_multiply);
3945 
3946   // AArch64 has a multiply-accumulate instruction that we can't use
3947   // here because it has no way to process carries, so we have to use
3948   // separate add and adc instructions.  Bah.
3949   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
3950   mul(product, x_xstart, y_idx);
3951   adds(product, product, carry);
3952   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
3953 
3954   subw(kdx, kdx, 2);
3955   ror(product, product, 32); // back to big-endian
3956   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
3957 
3958   b(L_first_loop);
3959 
3960   bind(L_one_y);
3961   ldrw(y_idx, Address(y,  0));
3962   b(L_multiply);
3963 
3964   bind(L_one_x);
3965   ldrw(x_xstart, Address(x,  0));
3966   b(L_first_loop);
3967 
3968   bind(L_first_loop_exit);
3969 }
3970 
3971 /**
3972  * Multiply 128 bit by 128. Unrolled inner loop.
3973  *
3974  */
3975 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
3976                                              Register carry, Register carry2,
3977                                              Register idx, Register jdx,
3978                                              Register yz_idx1, Register yz_idx2,
3979                                              Register tmp, Register tmp3, Register tmp4,
3980                                              Register tmp6, Register product_hi) {
3981 
3982   //   jlong carry, x[], y[], z[];
3983   //   int kdx = ystart+1;
3984   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
3985   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
3986   //     jlong carry2  = (jlong)(tmp3 >>> 64);
3987   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
3988   //     carry  = (jlong)(tmp4 >>> 64);
3989   //     z[kdx+idx+1] = (jlong)tmp3;
3990   //     z[kdx+idx] = (jlong)tmp4;
3991   //   }
3992   //   idx += 2;
3993   //   if (idx > 0) {
3994   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
3995   //     z[kdx+idx] = (jlong)yz_idx1;
3996   //     carry  = (jlong)(yz_idx1 >>> 64);
3997   //   }
3998   //
3999 
4000   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4001 
4002   lsrw(jdx, idx, 2);
4003 
4004   bind(L_third_loop);
4005 
4006   subsw(jdx, jdx, 1);
4007   br(Assembler::MI, L_third_loop_exit);
4008   subw(idx, idx, 4);
4009 
4010   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4011 
4012   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4013 
4014   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4015 
4016   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4017   ror(yz_idx2, yz_idx2, 32);
4018 
4019   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4020 
4021   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4022   umulh(tmp4, product_hi, yz_idx1);
4023 
4024   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4025   ror(rscratch2, rscratch2, 32);
4026 
4027   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4028   umulh(carry2, product_hi, yz_idx2);
4029 
4030   // propagate sum of both multiplications into carry:tmp4:tmp3
4031   adds(tmp3, tmp3, carry);
4032   adc(tmp4, tmp4, zr);
4033   adds(tmp3, tmp3, rscratch1);
4034   adcs(tmp4, tmp4, tmp);
4035   adc(carry, carry2, zr);
4036   adds(tmp4, tmp4, rscratch2);
4037   adc(carry, carry, zr);
4038 
4039   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4040   ror(tmp4, tmp4, 32);
4041   stp(tmp4, tmp3, Address(tmp6, 0));
4042 
4043   b(L_third_loop);
4044   bind (L_third_loop_exit);
4045 
4046   andw (idx, idx, 0x3);
4047   cbz(idx, L_post_third_loop_done);
4048 
4049   Label L_check_1;
4050   subsw(idx, idx, 2);
4051   br(Assembler::MI, L_check_1);
4052 
4053   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4054   ldr(yz_idx1, Address(rscratch1, 0));
4055   ror(yz_idx1, yz_idx1, 32);
4056   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4057   umulh(tmp4, product_hi, yz_idx1);
4058   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4059   ldr(yz_idx2, Address(rscratch1, 0));
4060   ror(yz_idx2, yz_idx2, 32);
4061 
4062   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4063 
4064   ror(tmp3, tmp3, 32);
4065   str(tmp3, Address(rscratch1, 0));
4066 
4067   bind (L_check_1);
4068 
4069   andw (idx, idx, 0x1);
4070   subsw(idx, idx, 1);
4071   br(Assembler::MI, L_post_third_loop_done);
4072   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4073   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4074   umulh(carry2, tmp4, product_hi);
4075   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4076 
4077   add2_with_carry(carry2, tmp3, tmp4, carry);
4078 
4079   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4080   extr(carry, carry2, tmp3, 32);
4081 
4082   bind(L_post_third_loop_done);
4083 }
4084 
4085 /**
4086  * Code for BigInteger::multiplyToLen() intrinsic.
4087  *
4088  * r0: x
4089  * r1: xlen
4090  * r2: y
4091  * r3: ylen
4092  * r4:  z
4093  * r5: tmp0
4094  * r10: tmp1
4095  * r11: tmp2
4096  * r12: tmp3
4097  * r13: tmp4
4098  * r14: tmp5
4099  * r15: tmp6
4100  * r16: tmp7
4101  *
4102  */
4103 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4104                                      Register z, Register tmp0,
4105                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4106                                      Register tmp5, Register tmp6, Register product_hi) {
4107 
4108   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4109 
4110   const Register idx = tmp1;
4111   const Register kdx = tmp2;
4112   const Register xstart = tmp3;
4113 
4114   const Register y_idx = tmp4;
4115   const Register carry = tmp5;
4116   const Register product  = xlen;
4117   const Register x_xstart = tmp0;
4118 
4119   // First Loop.
4120   //
4121   //  final static long LONG_MASK = 0xffffffffL;
4122   //  int xstart = xlen - 1;
4123   //  int ystart = ylen - 1;
4124   //  long carry = 0;
4125   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4126   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4127   //    z[kdx] = (int)product;
4128   //    carry = product >>> 32;
4129   //  }
4130   //  z[xstart] = (int)carry;
4131   //
4132 
4133   movw(idx, ylen);       // idx = ylen;
4134   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4135   mov(carry, zr);        // carry = 0;
4136 
4137   Label L_done;
4138 
4139   movw(xstart, xlen);
4140   subsw(xstart, xstart, 1);
4141   br(Assembler::MI, L_done);
4142 
4143   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4144 
4145   Label L_second_loop;
4146   cbzw(kdx, L_second_loop);
4147 
4148   Label L_carry;
4149   subw(kdx, kdx, 1);
4150   cbzw(kdx, L_carry);
4151 
4152   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4153   lsr(carry, carry, 32);
4154   subw(kdx, kdx, 1);
4155 
4156   bind(L_carry);
4157   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4158 
4159   // Second and third (nested) loops.
4160   //
4161   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4162   //   carry = 0;
4163   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4164   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4165   //                    (z[k] & LONG_MASK) + carry;
4166   //     z[k] = (int)product;
4167   //     carry = product >>> 32;
4168   //   }
4169   //   z[i] = (int)carry;
4170   // }
4171   //
4172   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4173 
4174   const Register jdx = tmp1;
4175 
4176   bind(L_second_loop);
4177   mov(carry, zr);                // carry = 0;
4178   movw(jdx, ylen);               // j = ystart+1
4179 
4180   subsw(xstart, xstart, 1);      // i = xstart-1;
4181   br(Assembler::MI, L_done);
4182 
4183   str(z, Address(pre(sp, -4 * wordSize)));
4184 
4185   Label L_last_x;
4186   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4187   subsw(xstart, xstart, 1);       // i = xstart-1;
4188   br(Assembler::MI, L_last_x);
4189 
4190   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4191   ldr(product_hi, Address(rscratch1));
4192   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4193 
4194   Label L_third_loop_prologue;
4195   bind(L_third_loop_prologue);
4196 
4197   str(ylen, Address(sp, wordSize));
4198   stp(x, xstart, Address(sp, 2 * wordSize));
4199   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4200                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4201   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4202   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4203 
4204   addw(tmp3, xlen, 1);
4205   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4206   subsw(tmp3, tmp3, 1);
4207   br(Assembler::MI, L_done);
4208 
4209   lsr(carry, carry, 32);
4210   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4211   b(L_second_loop);
4212 
4213   // Next infrequent code is moved outside loops.
4214   bind(L_last_x);
4215   ldrw(product_hi, Address(x,  0));
4216   b(L_third_loop_prologue);
4217 
4218   bind(L_done);
4219 }
4220 
4221 // Code for BigInteger::mulAdd intrinsic
4222 // out     = r0
4223 // in      = r1
4224 // offset  = r2  (already out.length-offset)
4225 // len     = r3
4226 // k       = r4
4227 //
4228 // pseudo code from java implementation:
4229 // carry = 0;
4230 // offset = out.length-offset - 1;
4231 // for (int j=len-1; j >= 0; j--) {
4232 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4233 //     out[offset--] = (int)product;
4234 //     carry = product >>> 32;
4235 // }
4236 // return (int)carry;
4237 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4238       Register len, Register k) {
4239     Label LOOP, END;
4240     // pre-loop
4241     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4242     csel(out, zr, out, Assembler::EQ);
4243     br(Assembler::EQ, END);
4244     add(in, in, len, LSL, 2); // in[j+1] address
4245     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4246     mov(out, zr); // used to keep carry now
4247     BIND(LOOP);
4248     ldrw(rscratch1, Address(pre(in, -4)));
4249     madd(rscratch1, rscratch1, k, out);
4250     ldrw(rscratch2, Address(pre(offset, -4)));
4251     add(rscratch1, rscratch1, rscratch2);
4252     strw(rscratch1, Address(offset));
4253     lsr(out, rscratch1, 32);
4254     subs(len, len, 1);
4255     br(Assembler::NE, LOOP);
4256     BIND(END);
4257 }
4258 
4259 /**
4260  * Emits code to update CRC-32 with a byte value according to constants in table
4261  *
4262  * @param [in,out]crc   Register containing the crc.
4263  * @param [in]val       Register containing the byte to fold into the CRC.
4264  * @param [in]table     Register containing the table of crc constants.
4265  *
4266  * uint32_t crc;
4267  * val = crc_table[(val ^ crc) & 0xFF];
4268  * crc = val ^ (crc >> 8);
4269  *
4270  */
4271 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4272   eor(val, val, crc);
4273   andr(val, val, 0xff);
4274   ldrw(val, Address(table, val, Address::lsl(2)));
4275   eor(crc, val, crc, Assembler::LSR, 8);
4276 }
4277 
4278 /**
4279  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4280  *
4281  * @param [in,out]crc   Register containing the crc.
4282  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4283  * @param [in]table0    Register containing table 0 of crc constants.
4284  * @param [in]table1    Register containing table 1 of crc constants.
4285  * @param [in]table2    Register containing table 2 of crc constants.
4286  * @param [in]table3    Register containing table 3 of crc constants.
4287  *
4288  * uint32_t crc;
4289  *   v = crc ^ v
4290  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4291  *
4292  */
4293 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4294         Register table0, Register table1, Register table2, Register table3,
4295         bool upper) {
4296   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4297   uxtb(tmp, v);
4298   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4299   ubfx(tmp, v, 8, 8);
4300   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4301   eor(crc, crc, tmp);
4302   ubfx(tmp, v, 16, 8);
4303   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4304   eor(crc, crc, tmp);
4305   ubfx(tmp, v, 24, 8);
4306   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4307   eor(crc, crc, tmp);
4308 }
4309 
4310 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4311         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4312     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4313     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4314 
4315     subs(tmp0, len, 384);
4316     mvnw(crc, crc);
4317     br(Assembler::GE, CRC_by128_pre);
4318   BIND(CRC_less128);
4319     subs(len, len, 32);
4320     br(Assembler::GE, CRC_by32_loop);
4321   BIND(CRC_less32);
4322     adds(len, len, 32 - 4);
4323     br(Assembler::GE, CRC_by4_loop);
4324     adds(len, len, 4);
4325     br(Assembler::GT, CRC_by1_loop);
4326     b(L_exit);
4327 
4328   BIND(CRC_by32_loop);
4329     ldp(tmp0, tmp1, Address(buf));
4330     crc32x(crc, crc, tmp0);
4331     ldp(tmp2, tmp3, Address(buf, 16));
4332     crc32x(crc, crc, tmp1);
4333     add(buf, buf, 32);
4334     crc32x(crc, crc, tmp2);
4335     subs(len, len, 32);
4336     crc32x(crc, crc, tmp3);
4337     br(Assembler::GE, CRC_by32_loop);
4338     cmn(len, (u1)32);
4339     br(Assembler::NE, CRC_less32);
4340     b(L_exit);
4341 
4342   BIND(CRC_by4_loop);
4343     ldrw(tmp0, Address(post(buf, 4)));
4344     subs(len, len, 4);
4345     crc32w(crc, crc, tmp0);
4346     br(Assembler::GE, CRC_by4_loop);
4347     adds(len, len, 4);
4348     br(Assembler::LE, L_exit);
4349   BIND(CRC_by1_loop);
4350     ldrb(tmp0, Address(post(buf, 1)));
4351     subs(len, len, 1);
4352     crc32b(crc, crc, tmp0);
4353     br(Assembler::GT, CRC_by1_loop);
4354     b(L_exit);
4355 
4356   BIND(CRC_by128_pre);
4357     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4358       4*256*sizeof(juint) + 8*sizeof(juint));
4359     mov(crc, 0);
4360     crc32x(crc, crc, tmp0);
4361     crc32x(crc, crc, tmp1);
4362 
4363     cbnz(len, CRC_less128);
4364 
4365   BIND(L_exit);
4366     mvnw(crc, crc);
4367 }
4368 
4369 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4370         Register len, Register tmp0, Register tmp1, Register tmp2,
4371         Register tmp3) {
4372     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4373     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4374 
4375     mvnw(crc, crc);
4376 
4377     subs(len, len, 128);
4378     br(Assembler::GE, CRC_by64_pre);
4379   BIND(CRC_less64);
4380     adds(len, len, 128-32);
4381     br(Assembler::GE, CRC_by32_loop);
4382   BIND(CRC_less32);
4383     adds(len, len, 32-4);
4384     br(Assembler::GE, CRC_by4_loop);
4385     adds(len, len, 4);
4386     br(Assembler::GT, CRC_by1_loop);
4387     b(L_exit);
4388 
4389   BIND(CRC_by32_loop);
4390     ldp(tmp0, tmp1, Address(post(buf, 16)));
4391     subs(len, len, 32);
4392     crc32x(crc, crc, tmp0);
4393     ldr(tmp2, Address(post(buf, 8)));
4394     crc32x(crc, crc, tmp1);
4395     ldr(tmp3, Address(post(buf, 8)));
4396     crc32x(crc, crc, tmp2);
4397     crc32x(crc, crc, tmp3);
4398     br(Assembler::GE, CRC_by32_loop);
4399     cmn(len, (u1)32);
4400     br(Assembler::NE, CRC_less32);
4401     b(L_exit);
4402 
4403   BIND(CRC_by4_loop);
4404     ldrw(tmp0, Address(post(buf, 4)));
4405     subs(len, len, 4);
4406     crc32w(crc, crc, tmp0);
4407     br(Assembler::GE, CRC_by4_loop);
4408     adds(len, len, 4);
4409     br(Assembler::LE, L_exit);
4410   BIND(CRC_by1_loop);
4411     ldrb(tmp0, Address(post(buf, 1)));
4412     subs(len, len, 1);
4413     crc32b(crc, crc, tmp0);
4414     br(Assembler::GT, CRC_by1_loop);
4415     b(L_exit);
4416 
4417   BIND(CRC_by64_pre);
4418     sub(buf, buf, 8);
4419     ldp(tmp0, tmp1, Address(buf, 8));
4420     crc32x(crc, crc, tmp0);
4421     ldr(tmp2, Address(buf, 24));
4422     crc32x(crc, crc, tmp1);
4423     ldr(tmp3, Address(buf, 32));
4424     crc32x(crc, crc, tmp2);
4425     ldr(tmp0, Address(buf, 40));
4426     crc32x(crc, crc, tmp3);
4427     ldr(tmp1, Address(buf, 48));
4428     crc32x(crc, crc, tmp0);
4429     ldr(tmp2, Address(buf, 56));
4430     crc32x(crc, crc, tmp1);
4431     ldr(tmp3, Address(pre(buf, 64)));
4432 
4433     b(CRC_by64_loop);
4434 
4435     align(CodeEntryAlignment);
4436   BIND(CRC_by64_loop);
4437     subs(len, len, 64);
4438     crc32x(crc, crc, tmp2);
4439     ldr(tmp0, Address(buf, 8));
4440     crc32x(crc, crc, tmp3);
4441     ldr(tmp1, Address(buf, 16));
4442     crc32x(crc, crc, tmp0);
4443     ldr(tmp2, Address(buf, 24));
4444     crc32x(crc, crc, tmp1);
4445     ldr(tmp3, Address(buf, 32));
4446     crc32x(crc, crc, tmp2);
4447     ldr(tmp0, Address(buf, 40));
4448     crc32x(crc, crc, tmp3);
4449     ldr(tmp1, Address(buf, 48));
4450     crc32x(crc, crc, tmp0);
4451     ldr(tmp2, Address(buf, 56));
4452     crc32x(crc, crc, tmp1);
4453     ldr(tmp3, Address(pre(buf, 64)));
4454     br(Assembler::GE, CRC_by64_loop);
4455 
4456     // post-loop
4457     crc32x(crc, crc, tmp2);
4458     crc32x(crc, crc, tmp3);
4459 
4460     sub(len, len, 64);
4461     add(buf, buf, 8);
4462     cmn(len, (u1)128);
4463     br(Assembler::NE, CRC_less64);
4464   BIND(L_exit);
4465     mvnw(crc, crc);
4466 }
4467 
4468 /**
4469  * @param crc   register containing existing CRC (32-bit)
4470  * @param buf   register pointing to input byte buffer (byte*)
4471  * @param len   register containing number of bytes
4472  * @param table register that will contain address of CRC table
4473  * @param tmp   scratch register
4474  */
4475 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4476         Register table0, Register table1, Register table2, Register table3,
4477         Register tmp, Register tmp2, Register tmp3) {
4478   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4479 
4480   if (UseCryptoPmullForCRC32) {
4481       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4482       return;
4483   }
4484 
4485   if (UseCRC32) {
4486       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4487       return;
4488   }
4489 
4490     mvnw(crc, crc);
4491 
4492     {
4493       uint64_t offset;
4494       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4495       add(table0, table0, offset);
4496     }
4497     add(table1, table0, 1*256*sizeof(juint));
4498     add(table2, table0, 2*256*sizeof(juint));
4499     add(table3, table0, 3*256*sizeof(juint));
4500 
4501     { // Neon code start
4502       cmp(len, (u1)64);
4503       br(Assembler::LT, L_by16);
4504       eor(v16, T16B, v16, v16);
4505 
4506     Label L_fold;
4507 
4508       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4509 
4510       ld1(v0, v1, T2D, post(buf, 32));
4511       ld1r(v4, T2D, post(tmp, 8));
4512       ld1r(v5, T2D, post(tmp, 8));
4513       ld1r(v6, T2D, post(tmp, 8));
4514       ld1r(v7, T2D, post(tmp, 8));
4515       mov(v16, S, 0, crc);
4516 
4517       eor(v0, T16B, v0, v16);
4518       sub(len, len, 64);
4519 
4520     BIND(L_fold);
4521       pmull(v22, T8H, v0, v5, T8B);
4522       pmull(v20, T8H, v0, v7, T8B);
4523       pmull(v23, T8H, v0, v4, T8B);
4524       pmull(v21, T8H, v0, v6, T8B);
4525 
4526       pmull2(v18, T8H, v0, v5, T16B);
4527       pmull2(v16, T8H, v0, v7, T16B);
4528       pmull2(v19, T8H, v0, v4, T16B);
4529       pmull2(v17, T8H, v0, v6, T16B);
4530 
4531       uzp1(v24, T8H, v20, v22);
4532       uzp2(v25, T8H, v20, v22);
4533       eor(v20, T16B, v24, v25);
4534 
4535       uzp1(v26, T8H, v16, v18);
4536       uzp2(v27, T8H, v16, v18);
4537       eor(v16, T16B, v26, v27);
4538 
4539       ushll2(v22, T4S, v20, T8H, 8);
4540       ushll(v20, T4S, v20, T4H, 8);
4541 
4542       ushll2(v18, T4S, v16, T8H, 8);
4543       ushll(v16, T4S, v16, T4H, 8);
4544 
4545       eor(v22, T16B, v23, v22);
4546       eor(v18, T16B, v19, v18);
4547       eor(v20, T16B, v21, v20);
4548       eor(v16, T16B, v17, v16);
4549 
4550       uzp1(v17, T2D, v16, v20);
4551       uzp2(v21, T2D, v16, v20);
4552       eor(v17, T16B, v17, v21);
4553 
4554       ushll2(v20, T2D, v17, T4S, 16);
4555       ushll(v16, T2D, v17, T2S, 16);
4556 
4557       eor(v20, T16B, v20, v22);
4558       eor(v16, T16B, v16, v18);
4559 
4560       uzp1(v17, T2D, v20, v16);
4561       uzp2(v21, T2D, v20, v16);
4562       eor(v28, T16B, v17, v21);
4563 
4564       pmull(v22, T8H, v1, v5, T8B);
4565       pmull(v20, T8H, v1, v7, T8B);
4566       pmull(v23, T8H, v1, v4, T8B);
4567       pmull(v21, T8H, v1, v6, T8B);
4568 
4569       pmull2(v18, T8H, v1, v5, T16B);
4570       pmull2(v16, T8H, v1, v7, T16B);
4571       pmull2(v19, T8H, v1, v4, T16B);
4572       pmull2(v17, T8H, v1, v6, T16B);
4573 
4574       ld1(v0, v1, T2D, post(buf, 32));
4575 
4576       uzp1(v24, T8H, v20, v22);
4577       uzp2(v25, T8H, v20, v22);
4578       eor(v20, T16B, v24, v25);
4579 
4580       uzp1(v26, T8H, v16, v18);
4581       uzp2(v27, T8H, v16, v18);
4582       eor(v16, T16B, v26, v27);
4583 
4584       ushll2(v22, T4S, v20, T8H, 8);
4585       ushll(v20, T4S, v20, T4H, 8);
4586 
4587       ushll2(v18, T4S, v16, T8H, 8);
4588       ushll(v16, T4S, v16, T4H, 8);
4589 
4590       eor(v22, T16B, v23, v22);
4591       eor(v18, T16B, v19, v18);
4592       eor(v20, T16B, v21, v20);
4593       eor(v16, T16B, v17, v16);
4594 
4595       uzp1(v17, T2D, v16, v20);
4596       uzp2(v21, T2D, v16, v20);
4597       eor(v16, T16B, v17, v21);
4598 
4599       ushll2(v20, T2D, v16, T4S, 16);
4600       ushll(v16, T2D, v16, T2S, 16);
4601 
4602       eor(v20, T16B, v22, v20);
4603       eor(v16, T16B, v16, v18);
4604 
4605       uzp1(v17, T2D, v20, v16);
4606       uzp2(v21, T2D, v20, v16);
4607       eor(v20, T16B, v17, v21);
4608 
4609       shl(v16, T2D, v28, 1);
4610       shl(v17, T2D, v20, 1);
4611 
4612       eor(v0, T16B, v0, v16);
4613       eor(v1, T16B, v1, v17);
4614 
4615       subs(len, len, 32);
4616       br(Assembler::GE, L_fold);
4617 
4618       mov(crc, 0);
4619       mov(tmp, v0, D, 0);
4620       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4621       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4622       mov(tmp, v0, D, 1);
4623       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4624       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4625       mov(tmp, v1, D, 0);
4626       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4627       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4628       mov(tmp, v1, D, 1);
4629       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4630       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4631 
4632       add(len, len, 32);
4633     } // Neon code end
4634 
4635   BIND(L_by16);
4636     subs(len, len, 16);
4637     br(Assembler::GE, L_by16_loop);
4638     adds(len, len, 16-4);
4639     br(Assembler::GE, L_by4_loop);
4640     adds(len, len, 4);
4641     br(Assembler::GT, L_by1_loop);
4642     b(L_exit);
4643 
4644   BIND(L_by4_loop);
4645     ldrw(tmp, Address(post(buf, 4)));
4646     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4647     subs(len, len, 4);
4648     br(Assembler::GE, L_by4_loop);
4649     adds(len, len, 4);
4650     br(Assembler::LE, L_exit);
4651   BIND(L_by1_loop);
4652     subs(len, len, 1);
4653     ldrb(tmp, Address(post(buf, 1)));
4654     update_byte_crc32(crc, tmp, table0);
4655     br(Assembler::GT, L_by1_loop);
4656     b(L_exit);
4657 
4658     align(CodeEntryAlignment);
4659   BIND(L_by16_loop);
4660     subs(len, len, 16);
4661     ldp(tmp, tmp3, Address(post(buf, 16)));
4662     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4663     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4664     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4665     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4666     br(Assembler::GE, L_by16_loop);
4667     adds(len, len, 16-4);
4668     br(Assembler::GE, L_by4_loop);
4669     adds(len, len, 4);
4670     br(Assembler::GT, L_by1_loop);
4671   BIND(L_exit);
4672     mvnw(crc, crc);
4673 }
4674 
4675 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4676         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4677     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4678     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4679 
4680     subs(tmp0, len, 384);
4681     br(Assembler::GE, CRC_by128_pre);
4682   BIND(CRC_less128);
4683     subs(len, len, 32);
4684     br(Assembler::GE, CRC_by32_loop);
4685   BIND(CRC_less32);
4686     adds(len, len, 32 - 4);
4687     br(Assembler::GE, CRC_by4_loop);
4688     adds(len, len, 4);
4689     br(Assembler::GT, CRC_by1_loop);
4690     b(L_exit);
4691 
4692   BIND(CRC_by32_loop);
4693     ldp(tmp0, tmp1, Address(buf));
4694     crc32cx(crc, crc, tmp0);
4695     ldr(tmp2, Address(buf, 16));
4696     crc32cx(crc, crc, tmp1);
4697     ldr(tmp3, Address(buf, 24));
4698     crc32cx(crc, crc, tmp2);
4699     add(buf, buf, 32);
4700     subs(len, len, 32);
4701     crc32cx(crc, crc, tmp3);
4702     br(Assembler::GE, CRC_by32_loop);
4703     cmn(len, (u1)32);
4704     br(Assembler::NE, CRC_less32);
4705     b(L_exit);
4706 
4707   BIND(CRC_by4_loop);
4708     ldrw(tmp0, Address(post(buf, 4)));
4709     subs(len, len, 4);
4710     crc32cw(crc, crc, tmp0);
4711     br(Assembler::GE, CRC_by4_loop);
4712     adds(len, len, 4);
4713     br(Assembler::LE, L_exit);
4714   BIND(CRC_by1_loop);
4715     ldrb(tmp0, Address(post(buf, 1)));
4716     subs(len, len, 1);
4717     crc32cb(crc, crc, tmp0);
4718     br(Assembler::GT, CRC_by1_loop);
4719     b(L_exit);
4720 
4721   BIND(CRC_by128_pre);
4722     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4723       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4724     mov(crc, 0);
4725     crc32cx(crc, crc, tmp0);
4726     crc32cx(crc, crc, tmp1);
4727 
4728     cbnz(len, CRC_less128);
4729 
4730   BIND(L_exit);
4731 }
4732 
4733 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4734         Register len, Register tmp0, Register tmp1, Register tmp2,
4735         Register tmp3) {
4736     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4737     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4738 
4739     subs(len, len, 128);
4740     br(Assembler::GE, CRC_by64_pre);
4741   BIND(CRC_less64);
4742     adds(len, len, 128-32);
4743     br(Assembler::GE, CRC_by32_loop);
4744   BIND(CRC_less32);
4745     adds(len, len, 32-4);
4746     br(Assembler::GE, CRC_by4_loop);
4747     adds(len, len, 4);
4748     br(Assembler::GT, CRC_by1_loop);
4749     b(L_exit);
4750 
4751   BIND(CRC_by32_loop);
4752     ldp(tmp0, tmp1, Address(post(buf, 16)));
4753     subs(len, len, 32);
4754     crc32cx(crc, crc, tmp0);
4755     ldr(tmp2, Address(post(buf, 8)));
4756     crc32cx(crc, crc, tmp1);
4757     ldr(tmp3, Address(post(buf, 8)));
4758     crc32cx(crc, crc, tmp2);
4759     crc32cx(crc, crc, tmp3);
4760     br(Assembler::GE, CRC_by32_loop);
4761     cmn(len, (u1)32);
4762     br(Assembler::NE, CRC_less32);
4763     b(L_exit);
4764 
4765   BIND(CRC_by4_loop);
4766     ldrw(tmp0, Address(post(buf, 4)));
4767     subs(len, len, 4);
4768     crc32cw(crc, crc, tmp0);
4769     br(Assembler::GE, CRC_by4_loop);
4770     adds(len, len, 4);
4771     br(Assembler::LE, L_exit);
4772   BIND(CRC_by1_loop);
4773     ldrb(tmp0, Address(post(buf, 1)));
4774     subs(len, len, 1);
4775     crc32cb(crc, crc, tmp0);
4776     br(Assembler::GT, CRC_by1_loop);
4777     b(L_exit);
4778 
4779   BIND(CRC_by64_pre);
4780     sub(buf, buf, 8);
4781     ldp(tmp0, tmp1, Address(buf, 8));
4782     crc32cx(crc, crc, tmp0);
4783     ldr(tmp2, Address(buf, 24));
4784     crc32cx(crc, crc, tmp1);
4785     ldr(tmp3, Address(buf, 32));
4786     crc32cx(crc, crc, tmp2);
4787     ldr(tmp0, Address(buf, 40));
4788     crc32cx(crc, crc, tmp3);
4789     ldr(tmp1, Address(buf, 48));
4790     crc32cx(crc, crc, tmp0);
4791     ldr(tmp2, Address(buf, 56));
4792     crc32cx(crc, crc, tmp1);
4793     ldr(tmp3, Address(pre(buf, 64)));
4794 
4795     b(CRC_by64_loop);
4796 
4797     align(CodeEntryAlignment);
4798   BIND(CRC_by64_loop);
4799     subs(len, len, 64);
4800     crc32cx(crc, crc, tmp2);
4801     ldr(tmp0, Address(buf, 8));
4802     crc32cx(crc, crc, tmp3);
4803     ldr(tmp1, Address(buf, 16));
4804     crc32cx(crc, crc, tmp0);
4805     ldr(tmp2, Address(buf, 24));
4806     crc32cx(crc, crc, tmp1);
4807     ldr(tmp3, Address(buf, 32));
4808     crc32cx(crc, crc, tmp2);
4809     ldr(tmp0, Address(buf, 40));
4810     crc32cx(crc, crc, tmp3);
4811     ldr(tmp1, Address(buf, 48));
4812     crc32cx(crc, crc, tmp0);
4813     ldr(tmp2, Address(buf, 56));
4814     crc32cx(crc, crc, tmp1);
4815     ldr(tmp3, Address(pre(buf, 64)));
4816     br(Assembler::GE, CRC_by64_loop);
4817 
4818     // post-loop
4819     crc32cx(crc, crc, tmp2);
4820     crc32cx(crc, crc, tmp3);
4821 
4822     sub(len, len, 64);
4823     add(buf, buf, 8);
4824     cmn(len, (u1)128);
4825     br(Assembler::NE, CRC_less64);
4826   BIND(L_exit);
4827 }
4828 
4829 /**
4830  * @param crc   register containing existing CRC (32-bit)
4831  * @param buf   register pointing to input byte buffer (byte*)
4832  * @param len   register containing number of bytes
4833  * @param table register that will contain address of CRC table
4834  * @param tmp   scratch register
4835  */
4836 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4837         Register table0, Register table1, Register table2, Register table3,
4838         Register tmp, Register tmp2, Register tmp3) {
4839   if (UseCryptoPmullForCRC32) {
4840     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4841   } else {
4842     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4843   }
4844 }
4845 
4846 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4847         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4848     Label CRC_by128_loop;
4849     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4850 
4851     sub(len, len, 256);
4852     Register table = tmp0;
4853     {
4854       uint64_t offset;
4855       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4856       add(table, table, offset);
4857     }
4858     add(table, table, table_offset);
4859 
4860     // Registers v0..v7 are used as data registers.
4861     // Registers v16..v31 are used as tmp registers.
4862     sub(buf, buf, 0x10);
4863     ldrq(v0, Address(buf, 0x10));
4864     ldrq(v1, Address(buf, 0x20));
4865     ldrq(v2, Address(buf, 0x30));
4866     ldrq(v3, Address(buf, 0x40));
4867     ldrq(v4, Address(buf, 0x50));
4868     ldrq(v5, Address(buf, 0x60));
4869     ldrq(v6, Address(buf, 0x70));
4870     ldrq(v7, Address(pre(buf, 0x80)));
4871 
4872     movi(v31, T4S, 0);
4873     mov(v31, S, 0, crc);
4874     eor(v0, T16B, v0, v31);
4875 
4876     // Register v16 contains constants from the crc table.
4877     ldrq(v16, Address(table));
4878     b(CRC_by128_loop);
4879 
4880     align(OptoLoopAlignment);
4881   BIND(CRC_by128_loop);
4882     pmull (v17,  T1Q, v0, v16, T1D);
4883     pmull2(v18, T1Q, v0, v16, T2D);
4884     ldrq(v0, Address(buf, 0x10));
4885     eor3(v0, T16B, v17,  v18, v0);
4886 
4887     pmull (v19, T1Q, v1, v16, T1D);
4888     pmull2(v20, T1Q, v1, v16, T2D);
4889     ldrq(v1, Address(buf, 0x20));
4890     eor3(v1, T16B, v19, v20, v1);
4891 
4892     pmull (v21, T1Q, v2, v16, T1D);
4893     pmull2(v22, T1Q, v2, v16, T2D);
4894     ldrq(v2, Address(buf, 0x30));
4895     eor3(v2, T16B, v21, v22, v2);
4896 
4897     pmull (v23, T1Q, v3, v16, T1D);
4898     pmull2(v24, T1Q, v3, v16, T2D);
4899     ldrq(v3, Address(buf, 0x40));
4900     eor3(v3, T16B, v23, v24, v3);
4901 
4902     pmull (v25, T1Q, v4, v16, T1D);
4903     pmull2(v26, T1Q, v4, v16, T2D);
4904     ldrq(v4, Address(buf, 0x50));
4905     eor3(v4, T16B, v25, v26, v4);
4906 
4907     pmull (v27, T1Q, v5, v16, T1D);
4908     pmull2(v28, T1Q, v5, v16, T2D);
4909     ldrq(v5, Address(buf, 0x60));
4910     eor3(v5, T16B, v27, v28, v5);
4911 
4912     pmull (v29, T1Q, v6, v16, T1D);
4913     pmull2(v30, T1Q, v6, v16, T2D);
4914     ldrq(v6, Address(buf, 0x70));
4915     eor3(v6, T16B, v29, v30, v6);
4916 
4917     // Reuse registers v23, v24.
4918     // Using them won't block the first instruction of the next iteration.
4919     pmull (v23, T1Q, v7, v16, T1D);
4920     pmull2(v24, T1Q, v7, v16, T2D);
4921     ldrq(v7, Address(pre(buf, 0x80)));
4922     eor3(v7, T16B, v23, v24, v7);
4923 
4924     subs(len, len, 0x80);
4925     br(Assembler::GE, CRC_by128_loop);
4926 
4927     // fold into 512 bits
4928     // Use v31 for constants because v16 can be still in use.
4929     ldrq(v31, Address(table, 0x10));
4930 
4931     pmull (v17,  T1Q, v0, v31, T1D);
4932     pmull2(v18, T1Q, v0, v31, T2D);
4933     eor3(v0, T16B, v17, v18, v4);
4934 
4935     pmull (v19, T1Q, v1, v31, T1D);
4936     pmull2(v20, T1Q, v1, v31, T2D);
4937     eor3(v1, T16B, v19, v20, v5);
4938 
4939     pmull (v21, T1Q, v2, v31, T1D);
4940     pmull2(v22, T1Q, v2, v31, T2D);
4941     eor3(v2, T16B, v21, v22, v6);
4942 
4943     pmull (v23, T1Q, v3, v31, T1D);
4944     pmull2(v24, T1Q, v3, v31, T2D);
4945     eor3(v3, T16B, v23, v24, v7);
4946 
4947     // fold into 128 bits
4948     // Use v17 for constants because v31 can be still in use.
4949     ldrq(v17, Address(table, 0x20));
4950     pmull (v25, T1Q, v0, v17, T1D);
4951     pmull2(v26, T1Q, v0, v17, T2D);
4952     eor3(v3, T16B, v3, v25, v26);
4953 
4954     // Use v18 for constants because v17 can be still in use.
4955     ldrq(v18, Address(table, 0x30));
4956     pmull (v27, T1Q, v1, v18, T1D);
4957     pmull2(v28, T1Q, v1, v18, T2D);
4958     eor3(v3, T16B, v3, v27, v28);
4959 
4960     // Use v19 for constants because v18 can be still in use.
4961     ldrq(v19, Address(table, 0x40));
4962     pmull (v29, T1Q, v2, v19, T1D);
4963     pmull2(v30, T1Q, v2, v19, T2D);
4964     eor3(v0, T16B, v3, v29, v30);
4965 
4966     add(len, len, 0x80);
4967     add(buf, buf, 0x10);
4968 
4969     mov(tmp0, v0, D, 0);
4970     mov(tmp1, v0, D, 1);
4971 }
4972 
4973 void MacroAssembler::addptr(const Address &dst, int32_t src) {
4974   Address adr;
4975   switch(dst.getMode()) {
4976   case Address::base_plus_offset:
4977     // This is the expected mode, although we allow all the other
4978     // forms below.
4979     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
4980     break;
4981   default:
4982     lea(rscratch2, dst);
4983     adr = Address(rscratch2);
4984     break;
4985   }
4986   ldr(rscratch1, adr);
4987   add(rscratch1, rscratch1, src);
4988   str(rscratch1, adr);
4989 }
4990 
4991 void MacroAssembler::cmpptr(Register src1, Address src2) {
4992   uint64_t offset;
4993   adrp(rscratch1, src2, offset);
4994   ldr(rscratch1, Address(rscratch1, offset));
4995   cmp(src1, rscratch1);
4996 }
4997 
4998 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
4999   cmp(obj1, obj2);
5000 }
5001 
5002 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5003   load_method_holder(rresult, rmethod);
5004   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5005 }
5006 
5007 void MacroAssembler::load_method_holder(Register holder, Register method) {
5008   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5009   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5010   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5011 }
5012 
5013 void MacroAssembler::load_metadata(Register dst, Register src) {
5014   if (UseCompactObjectHeaders) {
5015     load_narrow_klass_compact(dst, src);
5016   } else if (UseCompressedClassPointers) {
5017     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5018   } else {
5019     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5020   }
5021 }
5022 
5023 // Loads the obj's Klass* into dst.
5024 // Preserves all registers (incl src, rscratch1 and rscratch2).
5025 // Input:
5026 // src - the oop we want to load the klass from.
5027 // dst - output narrow klass.
5028 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5029   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5030   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5031   lsr(dst, dst, markWord::klass_shift);
5032 }
5033 
5034 void MacroAssembler::load_klass(Register dst, Register src) {
5035   if (UseCompactObjectHeaders) {
5036     load_narrow_klass_compact(dst, src);
5037     decode_klass_not_null(dst);
5038   } else if (UseCompressedClassPointers) {
5039     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5040     decode_klass_not_null(dst);
5041   } else {
5042     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5043   }
5044 }
5045 
5046 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5047   if (RestoreMXCSROnJNICalls) {
5048     Label OK;
5049     get_fpcr(tmp1);
5050     mov(tmp2, tmp1);
5051     // Set FPCR to the state we need. We do want Round to Nearest. We
5052     // don't want non-IEEE rounding modes or floating-point traps.
5053     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5054     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5055     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5056     eor(tmp2, tmp1, tmp2);
5057     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5058     set_fpcr(tmp1);
5059     bind(OK);
5060   }
5061 }
5062 
5063 // ((OopHandle)result).resolve();
5064 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5065   // OopHandle::resolve is an indirection.
5066   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5067 }
5068 
5069 // ((WeakHandle)result).resolve();
5070 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5071   assert_different_registers(result, tmp1, tmp2);
5072   Label resolved;
5073 
5074   // A null weak handle resolves to null.
5075   cbz(result, resolved);
5076 
5077   // Only 64 bit platforms support GCs that require a tmp register
5078   // WeakHandle::resolve is an indirection like jweak.
5079   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5080                  result, Address(result), tmp1, tmp2);
5081   bind(resolved);
5082 }
5083 
5084 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5085   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5086   ldr(dst, Address(rmethod, Method::const_offset()));
5087   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5088   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5089   ldr(dst, Address(dst, mirror_offset));
5090   resolve_oop_handle(dst, tmp1, tmp2);
5091 }
5092 
5093 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5094   assert_different_registers(obj, klass, tmp);
5095   if (UseCompressedClassPointers) {
5096     if (UseCompactObjectHeaders) {
5097       load_narrow_klass_compact(tmp, obj);
5098     } else {
5099       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5100     }
5101     if (CompressedKlassPointers::base() == nullptr) {
5102       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5103       return;
5104     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5105                && CompressedKlassPointers::shift() == 0) {
5106       // Only the bottom 32 bits matter
5107       cmpw(klass, tmp);
5108       return;
5109     }
5110     decode_klass_not_null(tmp);
5111   } else {
5112     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5113   }
5114   cmp(klass, tmp);
5115 }
5116 
5117 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5118   if (UseCompactObjectHeaders) {
5119     load_narrow_klass_compact(tmp1, obj1);
5120     load_narrow_klass_compact(tmp2,  obj2);
5121     cmpw(tmp1, tmp2);
5122   } else if (UseCompressedClassPointers) {
5123     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5124     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5125     cmpw(tmp1, tmp2);
5126   } else {
5127     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5128     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5129     cmp(tmp1, tmp2);
5130   }
5131 }
5132 
5133 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5134   load_klass(dst, src);
5135   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5136 }
5137 
5138 void MacroAssembler::store_klass(Register dst, Register src) {
5139   // FIXME: Should this be a store release?  concurrent gcs assumes
5140   // klass length is valid if klass field is not null.
5141   assert(!UseCompactObjectHeaders, "not with compact headers");
5142   if (UseCompressedClassPointers) {
5143     encode_klass_not_null(src);
5144     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5145   } else {
5146     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5147   }
5148 }
5149 
5150 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5151   assert(!UseCompactObjectHeaders, "not with compact headers");
5152   if (UseCompressedClassPointers) {
5153     // Store to klass gap in destination
5154     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5155   }
5156 }
5157 
5158 // Algorithm must match CompressedOops::encode.
5159 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5160 #ifdef ASSERT
5161   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5162 #endif
5163   verify_oop_msg(s, "broken oop in encode_heap_oop");
5164   if (CompressedOops::base() == nullptr) {
5165     if (CompressedOops::shift() != 0) {
5166       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5167       lsr(d, s, LogMinObjAlignmentInBytes);
5168     } else {
5169       mov(d, s);
5170     }
5171   } else {
5172     subs(d, s, rheapbase);
5173     csel(d, d, zr, Assembler::HS);
5174     lsr(d, d, LogMinObjAlignmentInBytes);
5175 
5176     /*  Old algorithm: is this any worse?
5177     Label nonnull;
5178     cbnz(r, nonnull);
5179     sub(r, r, rheapbase);
5180     bind(nonnull);
5181     lsr(r, r, LogMinObjAlignmentInBytes);
5182     */
5183   }
5184 }
5185 
5186 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5187 #ifdef ASSERT
5188   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5189   if (CheckCompressedOops) {
5190     Label ok;
5191     cbnz(r, ok);
5192     stop("null oop passed to encode_heap_oop_not_null");
5193     bind(ok);
5194   }
5195 #endif
5196   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5197   if (CompressedOops::base() != nullptr) {
5198     sub(r, r, rheapbase);
5199   }
5200   if (CompressedOops::shift() != 0) {
5201     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5202     lsr(r, r, LogMinObjAlignmentInBytes);
5203   }
5204 }
5205 
5206 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5207 #ifdef ASSERT
5208   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5209   if (CheckCompressedOops) {
5210     Label ok;
5211     cbnz(src, ok);
5212     stop("null oop passed to encode_heap_oop_not_null2");
5213     bind(ok);
5214   }
5215 #endif
5216   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5217 
5218   Register data = src;
5219   if (CompressedOops::base() != nullptr) {
5220     sub(dst, src, rheapbase);
5221     data = dst;
5222   }
5223   if (CompressedOops::shift() != 0) {
5224     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5225     lsr(dst, data, LogMinObjAlignmentInBytes);
5226     data = dst;
5227   }
5228   if (data == src)
5229     mov(dst, src);
5230 }
5231 
5232 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5233 #ifdef ASSERT
5234   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5235 #endif
5236   if (CompressedOops::base() == nullptr) {
5237     if (CompressedOops::shift() != 0) {
5238       lsl(d, s, CompressedOops::shift());
5239     } else if (d != s) {
5240       mov(d, s);
5241     }
5242   } else {
5243     Label done;
5244     if (d != s)
5245       mov(d, s);
5246     cbz(s, done);
5247     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5248     bind(done);
5249   }
5250   verify_oop_msg(d, "broken oop in decode_heap_oop");
5251 }
5252 
5253 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5254   assert (UseCompressedOops, "should only be used for compressed headers");
5255   assert (Universe::heap() != nullptr, "java heap should be initialized");
5256   // Cannot assert, unverified entry point counts instructions (see .ad file)
5257   // vtableStubs also counts instructions in pd_code_size_limit.
5258   // Also do not verify_oop as this is called by verify_oop.
5259   if (CompressedOops::shift() != 0) {
5260     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5261     if (CompressedOops::base() != nullptr) {
5262       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5263     } else {
5264       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5265     }
5266   } else {
5267     assert (CompressedOops::base() == nullptr, "sanity");
5268   }
5269 }
5270 
5271 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5272   assert (UseCompressedOops, "should only be used for compressed headers");
5273   assert (Universe::heap() != nullptr, "java heap should be initialized");
5274   // Cannot assert, unverified entry point counts instructions (see .ad file)
5275   // vtableStubs also counts instructions in pd_code_size_limit.
5276   // Also do not verify_oop as this is called by verify_oop.
5277   if (CompressedOops::shift() != 0) {
5278     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5279     if (CompressedOops::base() != nullptr) {
5280       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5281     } else {
5282       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5283     }
5284   } else {
5285     assert (CompressedOops::base() == nullptr, "sanity");
5286     if (dst != src) {
5287       mov(dst, src);
5288     }
5289   }
5290 }
5291 
5292 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5293 
5294 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5295   assert(Metaspace::initialized(), "metaspace not initialized yet");
5296   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5297   return _klass_decode_mode;
5298 }
5299 
5300 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5301   assert(UseCompressedClassPointers, "not using compressed class pointers");
5302 
5303   // KlassDecodeMode shouldn't be set already.
5304   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5305 
5306   if (base == nullptr) {
5307     return KlassDecodeZero;
5308   }
5309 
5310   if (operand_valid_for_logical_immediate(
5311         /*is32*/false, (uint64_t)base)) {
5312     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5313     if (((uint64_t)base & range_mask) == 0) {
5314       return KlassDecodeXor;
5315     }
5316   }
5317 
5318   const uint64_t shifted_base =
5319     (uint64_t)base >> shift;
5320   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5321     return KlassDecodeMovk;
5322   }
5323 
5324   // No valid encoding.
5325   return KlassDecodeNone;
5326 }
5327 
5328 // Check if one of the above decoding modes will work for given base, shift and range.
5329 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5330   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5331 }
5332 
5333 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5334   _klass_decode_mode = klass_decode_mode(base, shift, range);
5335   return _klass_decode_mode != KlassDecodeNone;
5336 }
5337 
5338 static Register pick_different_tmp(Register dst, Register src) {
5339   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5340   return *tmps.begin();
5341 }
5342 
5343 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5344   // we have to load the klass base from the AOT constants area but
5345   // not the shift because it is not allowed to change
5346   int shift = CompressedKlassPointers::shift();
5347   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5348   if (dst != src) {
5349     // we can load the base into dst, subtract it formthe src and shift down
5350     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5351     ldr(dst, dst);
5352     sub(dst, src, dst);
5353     lsr(dst, dst, shift);
5354   } else {
5355     // we need an extra register in order to load the coop base
5356     Register tmp = pick_different_tmp(dst, src);
5357     RegSet regs = RegSet::of(tmp);
5358     push(regs, sp);
5359     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5360     ldr(tmp, tmp);
5361     sub(dst, src, tmp);
5362     lsr(dst, dst, shift);
5363     pop(regs, sp);
5364   }
5365 }
5366 
5367 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5368   if (AOTCodeCache::is_on_for_dump()) {
5369     encode_klass_not_null_for_aot(dst, src);
5370     return;
5371   }
5372 
5373   switch (klass_decode_mode()) {
5374   case KlassDecodeZero:
5375     if (CompressedKlassPointers::shift() != 0) {
5376       lsr(dst, src, CompressedKlassPointers::shift());
5377     } else {
5378       if (dst != src) mov(dst, src);
5379     }
5380     break;
5381 
5382   case KlassDecodeXor:
5383     if (CompressedKlassPointers::shift() != 0) {
5384       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5385       lsr(dst, dst, CompressedKlassPointers::shift());
5386     } else {
5387       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5388     }
5389     break;
5390 
5391   case KlassDecodeMovk:
5392     if (CompressedKlassPointers::shift() != 0) {
5393       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5394     } else {
5395       movw(dst, src);
5396     }
5397     break;
5398 
5399   case KlassDecodeNone:
5400     ShouldNotReachHere();
5401     break;
5402   }
5403 }
5404 
5405 void MacroAssembler::encode_klass_not_null(Register r) {
5406   encode_klass_not_null(r, r);
5407 }
5408 
5409 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5410   // we have to load the klass base from the AOT constants area but
5411   // not the shift because it is not allowed to change
5412   int shift = CompressedKlassPointers::shift();
5413   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5414   if (dst != src) {
5415     // we can load the base into dst then add the offset with a suitable shift
5416     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5417     ldr(dst, dst);
5418     add(dst, dst, src, LSL,  shift);
5419   } else {
5420     // we need an extra register in order to load the coop base
5421     Register tmp = pick_different_tmp(dst, src);
5422     RegSet regs = RegSet::of(tmp);
5423     push(regs, sp);
5424     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5425     ldr(tmp, tmp);
5426     add(dst, tmp,  src, LSL,  shift);
5427     pop(regs, sp);
5428   }
5429 }
5430 
5431 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5432   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5433 
5434   if (AOTCodeCache::is_on_for_dump()) {
5435     decode_klass_not_null_for_aot(dst, src);
5436     return;
5437   }
5438 
5439   switch (klass_decode_mode()) {
5440   case KlassDecodeZero:
5441     if (CompressedKlassPointers::shift() != 0) {
5442       lsl(dst, src, CompressedKlassPointers::shift());
5443     } else {
5444       if (dst != src) mov(dst, src);
5445     }
5446     break;
5447 
5448   case KlassDecodeXor:
5449     if (CompressedKlassPointers::shift() != 0) {
5450       lsl(dst, src, CompressedKlassPointers::shift());
5451       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5452     } else {
5453       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5454     }
5455     break;
5456 
5457   case KlassDecodeMovk: {
5458     const uint64_t shifted_base =
5459       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5460 
5461     if (dst != src) movw(dst, src);
5462     movk(dst, shifted_base >> 32, 32);
5463 
5464     if (CompressedKlassPointers::shift() != 0) {
5465       lsl(dst, dst, CompressedKlassPointers::shift());
5466     }
5467 
5468     break;
5469   }
5470 
5471   case KlassDecodeNone:
5472     ShouldNotReachHere();
5473     break;
5474   }
5475 }
5476 
5477 void  MacroAssembler::decode_klass_not_null(Register r) {
5478   decode_klass_not_null(r, r);
5479 }
5480 
5481 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5482 #ifdef ASSERT
5483   {
5484     ThreadInVMfromUnknown tiv;
5485     assert (UseCompressedOops, "should only be used for compressed oops");
5486     assert (Universe::heap() != nullptr, "java heap should be initialized");
5487     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5488     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5489   }
5490 #endif
5491   int oop_index = oop_recorder()->find_index(obj);
5492   InstructionMark im(this);
5493   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5494   code_section()->relocate(inst_mark(), rspec);
5495   movz(dst, 0xDEAD, 16);
5496   movk(dst, 0xBEEF);
5497 }
5498 
5499 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5500   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5501   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5502   int index = oop_recorder()->find_index(k);
5503 
5504   InstructionMark im(this);
5505   RelocationHolder rspec = metadata_Relocation::spec(index);
5506   code_section()->relocate(inst_mark(), rspec);
5507   narrowKlass nk = CompressedKlassPointers::encode(k);
5508   movz(dst, (nk >> 16), 16);
5509   movk(dst, nk & 0xffff);
5510 }
5511 
5512 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5513                                     Register dst, Address src,
5514                                     Register tmp1, Register tmp2) {
5515   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5516   decorators = AccessInternal::decorator_fixup(decorators, type);
5517   bool as_raw = (decorators & AS_RAW) != 0;
5518   if (as_raw) {
5519     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5520   } else {
5521     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5522   }
5523 }
5524 
5525 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5526                                      Address dst, Register val,
5527                                      Register tmp1, Register tmp2, Register tmp3) {
5528   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5529   decorators = AccessInternal::decorator_fixup(decorators, type);
5530   bool as_raw = (decorators & AS_RAW) != 0;
5531   if (as_raw) {
5532     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5533   } else {
5534     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5535   }
5536 }
5537 
5538 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5539                                      Register inline_layout_info) {
5540   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5541   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5542 }
5543 
5544 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5545   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5546   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5547 }
5548 
5549 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5550   // ((address) (void*) o) + vk->payload_offset();
5551   Register offset = (data == oop) ? rscratch1 : data;
5552   payload_offset(inline_klass, offset);
5553   if (data == oop) {
5554     add(data, data, offset);
5555   } else {
5556     lea(data, Address(oop, offset));
5557   }
5558 }
5559 
5560 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
5561                                                 Register index, Register data) {
5562   assert_different_registers(array, array_klass, index);
5563   assert_different_registers(rscratch1, array, index);
5564 
5565   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
5566   ldrw(rscratch1, Address(array_klass, Klass::layout_helper_offset()));
5567 
5568   // Klass::layout_helper_log2_element_size(lh)
5569   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
5570   lsr(rscratch1, rscratch1, Klass::_lh_log2_element_size_shift);
5571   andr(rscratch1, rscratch1, Klass::_lh_log2_element_size_mask);
5572   lslv(index, index, rscratch1);
5573 
5574   add(data, array, index);
5575   add(data, data, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT));
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 // Object / value buffer allocation...
5654 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
5655                                        Register t1, Register t2,
5656                                        bool clear_fields, Label& alloc_failed)
5657 {
5658   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
5659   Register layout_size = t1;
5660   assert(new_obj == r0, "needs to be r0");
5661   assert_different_registers(klass, new_obj, t1, t2);
5662 
5663   // get instance_size in InstanceKlass (scaled to a count of bytes)
5664   ldrw(layout_size, Address(klass, Klass::layout_helper_offset()));
5665   // test to see if it is malformed in some way
5666   tst(layout_size, Klass::_lh_instance_slow_path_bit);
5667   br(Assembler::NE, slow_case_no_pop);
5668 
5669   // Allocate the instance:
5670   //  If TLAB is enabled:
5671   //    Try to allocate in the TLAB.
5672   //    If fails, go to the slow path.
5673   //    Initialize the allocation.
5674   //    Exit.
5675   //
5676   //  Go to slow path.
5677 
5678   if (UseTLAB) {
5679     push(klass);
5680     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
5681     if (ZeroTLAB || (!clear_fields)) {
5682       // the fields have been already cleared
5683       b(initialize_header);
5684     } else {
5685       // initialize both the header and fields
5686       b(initialize_object);
5687     }
5688 
5689     if (clear_fields) {
5690       // The object is initialized before the header.  If the object size is
5691       // zero, go directly to the header initialization.
5692       bind(initialize_object);
5693       int header_size = oopDesc::header_size() * HeapWordSize;
5694       assert(is_aligned(header_size, BytesPerLong), "oop header size must be 8-byte-aligned");
5695       subs(layout_size, layout_size, header_size);
5696       br(Assembler::EQ, initialize_header);
5697 
5698       // Initialize topmost object field, divide size by 8, check if odd and
5699       // test if zero.
5700 
5701   #ifdef ASSERT
5702       // make sure instance_size was multiple of 8
5703       Label L;
5704       tst(layout_size, 7);
5705       br(Assembler::EQ, L);
5706       stop("object size is not multiple of 8 - adjust this code");
5707       bind(L);
5708       // must be > 0, no extra check needed here
5709   #endif
5710 
5711       lsr(layout_size, layout_size, LogBytesPerLong);
5712 
5713       // initialize remaining object fields: instance_size was a multiple of 8
5714       {
5715         Label loop;
5716         Register base = t2;
5717 
5718         bind(loop);
5719         add(rscratch1, new_obj, layout_size, Assembler::LSL, LogBytesPerLong);
5720         str(zr, Address(rscratch1, header_size - 1*oopSize));
5721         subs(layout_size, layout_size, 1);
5722         br(Assembler::NE, loop);
5723       }
5724     } // clear_fields
5725 
5726     // initialize object header only.
5727     bind(initialize_header);
5728     pop(klass);
5729     Register mark_word = t2;
5730     if (UseCompactObjectHeaders || Arguments::is_valhalla_enabled()) {
5731       ldr(mark_word, Address(klass, Klass::prototype_header_offset()));
5732       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5733     } else {
5734       mov(mark_word, (intptr_t)markWord::prototype().value());
5735       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5736     }
5737     if (!UseCompactObjectHeaders) {
5738       store_klass_gap(new_obj, zr);  // zero klass gap for compressed oops
5739       mov(t2, klass);                // preserve klass
5740       store_klass(new_obj, t2);      // src klass reg is potentially compressed
5741     }
5742     b(done);
5743   }
5744 
5745   if (UseTLAB) {
5746     bind(slow_case);
5747     pop(klass);
5748   }
5749   bind(slow_case_no_pop);
5750   b(alloc_failed);
5751 
5752   bind(done);
5753 }
5754 
5755 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5756 void MacroAssembler::tlab_allocate(Register obj,
5757                                    Register var_size_in_bytes,
5758                                    int con_size_in_bytes,
5759                                    Register t1,
5760                                    Register t2,
5761                                    Label& slow_case) {
5762   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5763   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5764 }
5765 
5766 void MacroAssembler::verify_tlab() {
5767 #ifdef ASSERT
5768   if (UseTLAB && VerifyOops) {
5769     Label next, ok;
5770 
5771     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5772 
5773     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5774     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5775     cmp(rscratch2, rscratch1);
5776     br(Assembler::HS, next);
5777     STOP("assert(top >= start)");
5778     should_not_reach_here();
5779 
5780     bind(next);
5781     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5782     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5783     cmp(rscratch2, rscratch1);
5784     br(Assembler::HS, ok);
5785     STOP("assert(top <= end)");
5786     should_not_reach_here();
5787 
5788     bind(ok);
5789     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5790   }
5791 #endif
5792 }
5793 
5794 void MacroAssembler::get_inline_type_field_klass(Register holder_klass, Register index, Register inline_klass) {
5795   inline_layout_info(holder_klass, index, inline_klass);
5796   ldr(inline_klass, Address(inline_klass, InlineLayoutInfo::klass_offset()));
5797 }
5798 
5799 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5800   assert_different_registers(holder_klass, index, layout_info);
5801   InlineLayoutInfo array[2];
5802   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5803   if (is_power_of_2(size)) {
5804     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5805   } else {
5806     mov(layout_info, size);
5807     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5808   }
5809   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5810   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5811   lea(layout_info, Address(layout_info, index));
5812 }
5813 
5814 // Writes to stack successive pages until offset reached to check for
5815 // stack overflow + shadow pages.  This clobbers tmp.
5816 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5817   assert_different_registers(tmp, size, rscratch1);
5818   mov(tmp, sp);
5819   // Bang stack for total size given plus shadow page size.
5820   // Bang one page at a time because large size can bang beyond yellow and
5821   // red zones.
5822   Label loop;
5823   mov(rscratch1, (int)os::vm_page_size());
5824   bind(loop);
5825   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5826   subsw(size, size, rscratch1);
5827   str(size, Address(tmp));
5828   br(Assembler::GT, loop);
5829 
5830   // Bang down shadow pages too.
5831   // At this point, (tmp-0) is the last address touched, so don't
5832   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5833   // was post-decremented.)  Skip this address by starting at i=1, and
5834   // touch a few more pages below.  N.B.  It is important to touch all
5835   // the way down to and including i=StackShadowPages.
5836   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5837     // this could be any sized move but this is can be a debugging crumb
5838     // so the bigger the better.
5839     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5840     str(size, Address(tmp));
5841   }
5842 }
5843 
5844 // Move the address of the polling page into dest.
5845 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5846   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5847 }
5848 
5849 // Read the polling page.  The address of the polling page must
5850 // already be in r.
5851 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5852   address mark;
5853   {
5854     InstructionMark im(this);
5855     code_section()->relocate(inst_mark(), rtype);
5856     ldrw(zr, Address(r, 0));
5857     mark = inst_mark();
5858   }
5859   verify_cross_modify_fence_not_required();
5860   return mark;
5861 }
5862 
5863 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5864   relocInfo::relocType rtype = dest.rspec().reloc()->type();
5865   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5866   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5867   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5868   int64_t offset_low = dest_page - low_page;
5869   int64_t offset_high = dest_page - high_page;
5870 
5871   assert(is_valid_AArch64_address(dest.target()), "bad address");
5872   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5873 
5874   InstructionMark im(this);
5875   code_section()->relocate(inst_mark(), dest.rspec());
5876   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5877   // the code cache so that if it is relocated we know it will still reach
5878   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5879     _adrp(reg1, dest.target());
5880   } else {
5881     uint64_t target = (uint64_t)dest.target();
5882     uint64_t adrp_target
5883       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5884 
5885     _adrp(reg1, (address)adrp_target);
5886     movk(reg1, target >> 32, 32);
5887   }
5888   byte_offset = (uint64_t)dest.target() & 0xfff;
5889 }
5890 
5891 void MacroAssembler::load_byte_map_base(Register reg) {
5892   CardTable::CardValue* byte_map_base =
5893     ((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base();
5894 
5895   // Strictly speaking the byte_map_base isn't an address at all, and it might
5896   // even be negative. It is thus materialised as a constant.
5897   mov(reg, (uint64_t)byte_map_base);
5898 }
5899 
5900 #ifdef ASSERT
5901 void MacroAssembler::build_frame(int framesize) {
5902   build_frame(framesize, false);
5903 }
5904 #endif
5905 
5906 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
5907   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5908   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5909   protect_return_address();
5910   if (framesize < ((1 << 9) + 2 * wordSize)) {
5911     sub(sp, sp, framesize);
5912     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5913       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5914       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
5915     } else {
5916       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5917     }
5918     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5919   } else {
5920     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5921       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5922       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
5923     } else {
5924       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5925     }
5926     if (PreserveFramePointer) mov(rfp, sp);
5927     if (framesize < ((1 << 12) + 2 * wordSize))
5928       sub(sp, sp, framesize - 2 * wordSize);
5929     else {
5930       mov(rscratch1, framesize - 2 * wordSize);
5931       sub(sp, sp, rscratch1);
5932     }
5933   }
5934   verify_cross_modify_fence_not_required();
5935 }
5936 
5937 void MacroAssembler::remove_frame(int framesize) {
5938   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5939   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5940   if (framesize < ((1 << 9) + 2 * wordSize)) {
5941     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5942     add(sp, sp, framesize);
5943   } else {
5944     if (framesize < ((1 << 12) + 2 * wordSize))
5945       add(sp, sp, framesize - 2 * wordSize);
5946     else {
5947       mov(rscratch1, framesize - 2 * wordSize);
5948       add(sp, sp, rscratch1);
5949     }
5950     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5951   }
5952   authenticate_return_address();
5953 }
5954 
5955 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
5956   if (needs_stack_repair) {
5957     // Remove the extension of the caller's frame used for inline type unpacking
5958     //
5959     // Right now the stack looks like this:
5960     //
5961     // | Arguments from caller     |
5962     // |---------------------------|  <-- caller's SP
5963     // | Saved LR #1               |
5964     // | Saved FP #1               |
5965     // |---------------------------|
5966     // | Extension space for       |
5967     // |   inline arg (un)packing  |
5968     // |---------------------------|  <-- start of this method's frame
5969     // | Saved LR #2               |
5970     // | Saved FP #2               |
5971     // |---------------------------|  <-- FP
5972     // | sp_inc                    |
5973     // | method locals             |
5974     // |---------------------------|  <-- SP
5975     //
5976     // There are two copies of FP and LR on the stack. They will be identical at
5977     // first, but that can change.
5978     // If the caller has been deoptimized, LR #1 will be patched to point at the
5979     // deopt blob, and LR #2 will still point into the old method.
5980     // If the saved FP (x29) was not used as the frame pointer, but to store an
5981     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
5982     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
5983     // needed to add space between the extension space and the locals, as there
5984     // would be between the real arguments and the locals if we don't need to
5985     // do unpacking.
5986     //
5987     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
5988     // while keeping in mind that from the scalarized entry point, there will be
5989     // only one copy of each.
5990     //
5991     // The sp_inc stack slot holds the total size of the frame including the
5992     // extension space minus two words for the saved FP and LR. That is how to
5993     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
5994     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
5995     //
5996     // TODO 8371993 store fake values instead of LR/FP#2
5997 
5998     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
5999 
6000     ldr(rscratch1, Address(sp, sp_inc_offset));
6001     add(sp, sp, rscratch1);
6002     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6003   } else {
6004     remove_frame(initial_framesize);
6005   }
6006 }
6007 
6008 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6009   int real_frame_size = frame_size + sp_inc;
6010   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6011   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6012   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6013 
6014   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6015 
6016   // Subtract two words for the saved FP and LR as these will be popped
6017   // separately. See remove_frame above.
6018   mov(rscratch1, real_frame_size - 2*wordSize);
6019   str(rscratch1, Address(sp, sp_inc_offset));
6020 }
6021 
6022 // This method counts leading positive bytes (highest bit not set) in provided byte array
6023 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6024     // Simple and most common case of aligned small array which is not at the
6025     // end of memory page is placed here. All other cases are in stub.
6026     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6027     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6028     assert_different_registers(ary1, len, result);
6029 
6030     mov(result, len);
6031     cmpw(len, 0);
6032     br(LE, DONE);
6033     cmpw(len, 4 * wordSize);
6034     br(GE, STUB_LONG); // size > 32 then go to stub
6035 
6036     int shift = 64 - exact_log2(os::vm_page_size());
6037     lsl(rscratch1, ary1, shift);
6038     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6039     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6040     br(CS, STUB); // at the end of page then go to stub
6041     subs(len, len, wordSize);
6042     br(LT, END);
6043 
6044   BIND(LOOP);
6045     ldr(rscratch1, Address(post(ary1, wordSize)));
6046     tst(rscratch1, UPPER_BIT_MASK);
6047     br(NE, SET_RESULT);
6048     subs(len, len, wordSize);
6049     br(GE, LOOP);
6050     cmpw(len, -wordSize);
6051     br(EQ, DONE);
6052 
6053   BIND(END);
6054     ldr(rscratch1, Address(ary1));
6055     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
6056     lslv(rscratch1, rscratch1, rscratch2);
6057     tst(rscratch1, UPPER_BIT_MASK);
6058     br(NE, SET_RESULT);
6059     b(DONE);
6060 
6061   BIND(STUB);
6062     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
6063     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
6064     address tpc1 = trampoline_call(count_pos);
6065     if (tpc1 == nullptr) {
6066       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
6067       postcond(pc() == badAddress);
6068       return nullptr;
6069     }
6070     b(DONE);
6071 
6072   BIND(STUB_LONG);
6073     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
6074     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
6075     address tpc2 = trampoline_call(count_pos_long);
6076     if (tpc2 == nullptr) {
6077       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
6078       postcond(pc() == badAddress);
6079       return nullptr;
6080     }
6081     b(DONE);
6082 
6083   BIND(SET_RESULT);
6084 
6085     add(len, len, wordSize);
6086     sub(result, result, len);
6087 
6088   BIND(DONE);
6089   postcond(pc() != badAddress);
6090   return pc();
6091 }
6092 
6093 // Clobbers: rscratch1, rscratch2, rflags
6094 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
6095 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
6096                                       Register tmp4, Register tmp5, Register result,
6097                                       Register cnt1, int elem_size) {
6098   Label DONE, SAME;
6099   Register tmp1 = rscratch1;
6100   Register tmp2 = rscratch2;
6101   int elem_per_word = wordSize/elem_size;
6102   int log_elem_size = exact_log2(elem_size);
6103   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
6104   int length_offset = arrayOopDesc::length_offset_in_bytes();
6105   int base_offset
6106     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
6107   // When the length offset is not aligned to 8 bytes,
6108   // then we align it down. This is valid because the new
6109   // offset will always be the klass which is the same
6110   // for type arrays.
6111   int start_offset = align_down(length_offset, BytesPerWord);
6112   int extra_length = base_offset - start_offset;
6113   assert(start_offset == length_offset || start_offset == klass_offset,
6114          "start offset must be 8-byte-aligned or be the klass offset");
6115   assert(base_offset != start_offset, "must include the length field");
6116   extra_length = extra_length / elem_size; // We count in elements, not bytes.
6117   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
6118 
6119   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
6120   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6121 
6122 #ifndef PRODUCT
6123   {
6124     const char kind = (elem_size == 2) ? 'U' : 'L';
6125     char comment[64];
6126     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
6127     BLOCK_COMMENT(comment);
6128   }
6129 #endif
6130 
6131   // if (a1 == a2)
6132   //     return true;
6133   cmpoop(a1, a2); // May have read barriers for a1 and a2.
6134   br(EQ, SAME);
6135 
6136   if (UseSimpleArrayEquals) {
6137     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
6138     // if (a1 == nullptr || a2 == nullptr)
6139     //     return false;
6140     // a1 & a2 == 0 means (some-pointer is null) or
6141     // (very-rare-or-even-probably-impossible-pointer-values)
6142     // so, we can save one branch in most cases
6143     tst(a1, a2);
6144     mov(result, false);
6145     br(EQ, A_MIGHT_BE_NULL);
6146     // if (a1.length != a2.length)
6147     //      return false;
6148     bind(A_IS_NOT_NULL);
6149     ldrw(cnt1, Address(a1, length_offset));
6150     // Increase loop counter by diff between base- and actual start-offset.
6151     addw(cnt1, cnt1, extra_length);
6152     lea(a1, Address(a1, start_offset));
6153     lea(a2, Address(a2, start_offset));
6154     // Check for short strings, i.e. smaller than wordSize.
6155     subs(cnt1, cnt1, elem_per_word);
6156     br(Assembler::LT, SHORT);
6157     // Main 8 byte comparison loop.
6158     bind(NEXT_WORD); {
6159       ldr(tmp1, Address(post(a1, wordSize)));
6160       ldr(tmp2, Address(post(a2, wordSize)));
6161       subs(cnt1, cnt1, elem_per_word);
6162       eor(tmp5, tmp1, tmp2);
6163       cbnz(tmp5, DONE);
6164     } br(GT, NEXT_WORD);
6165     // Last longword.  In the case where length == 4 we compare the
6166     // same longword twice, but that's still faster than another
6167     // conditional branch.
6168     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6169     // length == 4.
6170     if (log_elem_size > 0)
6171       lsl(cnt1, cnt1, log_elem_size);
6172     ldr(tmp3, Address(a1, cnt1));
6173     ldr(tmp4, Address(a2, cnt1));
6174     eor(tmp5, tmp3, tmp4);
6175     cbnz(tmp5, DONE);
6176     b(SAME);
6177     bind(A_MIGHT_BE_NULL);
6178     // in case both a1 and a2 are not-null, proceed with loads
6179     cbz(a1, DONE);
6180     cbz(a2, DONE);
6181     b(A_IS_NOT_NULL);
6182     bind(SHORT);
6183 
6184     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
6185     {
6186       ldrw(tmp1, Address(post(a1, 4)));
6187       ldrw(tmp2, Address(post(a2, 4)));
6188       eorw(tmp5, tmp1, tmp2);
6189       cbnzw(tmp5, DONE);
6190     }
6191     bind(TAIL03);
6192     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6193     {
6194       ldrh(tmp3, Address(post(a1, 2)));
6195       ldrh(tmp4, Address(post(a2, 2)));
6196       eorw(tmp5, tmp3, tmp4);
6197       cbnzw(tmp5, DONE);
6198     }
6199     bind(TAIL01);
6200     if (elem_size == 1) { // Only needed when comparing byte arrays.
6201       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6202       {
6203         ldrb(tmp1, a1);
6204         ldrb(tmp2, a2);
6205         eorw(tmp5, tmp1, tmp2);
6206         cbnzw(tmp5, DONE);
6207       }
6208     }
6209   } else {
6210     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6211         CSET_EQ, LAST_CHECK;
6212     mov(result, false);
6213     cbz(a1, DONE);
6214     ldrw(cnt1, Address(a1, length_offset));
6215     cbz(a2, DONE);
6216     // Increase loop counter by diff between base- and actual start-offset.
6217     addw(cnt1, cnt1, extra_length);
6218 
6219     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6220     // faster to perform another branch before comparing a1 and a2
6221     cmp(cnt1, (u1)elem_per_word);
6222     br(LE, SHORT); // short or same
6223     ldr(tmp3, Address(pre(a1, start_offset)));
6224     subs(zr, cnt1, stubBytesThreshold);
6225     br(GE, STUB);
6226     ldr(tmp4, Address(pre(a2, start_offset)));
6227     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6228 
6229     // Main 16 byte comparison loop with 2 exits
6230     bind(NEXT_DWORD); {
6231       ldr(tmp1, Address(pre(a1, wordSize)));
6232       ldr(tmp2, Address(pre(a2, wordSize)));
6233       subs(cnt1, cnt1, 2 * elem_per_word);
6234       br(LE, TAIL);
6235       eor(tmp4, tmp3, tmp4);
6236       cbnz(tmp4, DONE);
6237       ldr(tmp3, Address(pre(a1, wordSize)));
6238       ldr(tmp4, Address(pre(a2, wordSize)));
6239       cmp(cnt1, (u1)elem_per_word);
6240       br(LE, TAIL2);
6241       cmp(tmp1, tmp2);
6242     } br(EQ, NEXT_DWORD);
6243     b(DONE);
6244 
6245     bind(TAIL);
6246     eor(tmp4, tmp3, tmp4);
6247     eor(tmp2, tmp1, tmp2);
6248     lslv(tmp2, tmp2, tmp5);
6249     orr(tmp5, tmp4, tmp2);
6250     cmp(tmp5, zr);
6251     b(CSET_EQ);
6252 
6253     bind(TAIL2);
6254     eor(tmp2, tmp1, tmp2);
6255     cbnz(tmp2, DONE);
6256     b(LAST_CHECK);
6257 
6258     bind(STUB);
6259     ldr(tmp4, Address(pre(a2, start_offset)));
6260     if (elem_size == 2) { // convert to byte counter
6261       lsl(cnt1, cnt1, 1);
6262     }
6263     eor(tmp5, tmp3, tmp4);
6264     cbnz(tmp5, DONE);
6265     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6266     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6267     address tpc = trampoline_call(stub);
6268     if (tpc == nullptr) {
6269       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6270       postcond(pc() == badAddress);
6271       return nullptr;
6272     }
6273     b(DONE);
6274 
6275     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6276     // so, if a2 == null => return false(0), else return true, so we can return a2
6277     mov(result, a2);
6278     b(DONE);
6279     bind(SHORT);
6280     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6281     ldr(tmp3, Address(a1, start_offset));
6282     ldr(tmp4, Address(a2, start_offset));
6283     bind(LAST_CHECK);
6284     eor(tmp4, tmp3, tmp4);
6285     lslv(tmp5, tmp4, tmp5);
6286     cmp(tmp5, zr);
6287     bind(CSET_EQ);
6288     cset(result, EQ);
6289     b(DONE);
6290   }
6291 
6292   bind(SAME);
6293   mov(result, true);
6294   // That's it.
6295   bind(DONE);
6296 
6297   BLOCK_COMMENT("} array_equals");
6298   postcond(pc() != badAddress);
6299   return pc();
6300 }
6301 
6302 // Compare Strings
6303 
6304 // For Strings we're passed the address of the first characters in a1
6305 // and a2 and the length in cnt1.
6306 // There are two implementations.  For arrays >= 8 bytes, all
6307 // comparisons (including the final one, which may overlap) are
6308 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6309 // halfword, then a short, and then a byte.
6310 
6311 void MacroAssembler::string_equals(Register a1, Register a2,
6312                                    Register result, Register cnt1)
6313 {
6314   Label SAME, DONE, SHORT, NEXT_WORD;
6315   Register tmp1 = rscratch1;
6316   Register tmp2 = rscratch2;
6317   Register cnt2 = tmp2;  // cnt2 only used in array length compare
6318 
6319   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6320 
6321 #ifndef PRODUCT
6322   {
6323     char comment[64];
6324     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6325     BLOCK_COMMENT(comment);
6326   }
6327 #endif
6328 
6329   mov(result, false);
6330 
6331   // Check for short strings, i.e. smaller than wordSize.
6332   subs(cnt1, cnt1, wordSize);
6333   br(Assembler::LT, SHORT);
6334   // Main 8 byte comparison loop.
6335   bind(NEXT_WORD); {
6336     ldr(tmp1, Address(post(a1, wordSize)));
6337     ldr(tmp2, Address(post(a2, wordSize)));
6338     subs(cnt1, cnt1, wordSize);
6339     eor(tmp1, tmp1, tmp2);
6340     cbnz(tmp1, DONE);
6341   } br(GT, NEXT_WORD);
6342   // Last longword.  In the case where length == 4 we compare the
6343   // same longword twice, but that's still faster than another
6344   // conditional branch.
6345   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6346   // length == 4.
6347   ldr(tmp1, Address(a1, cnt1));
6348   ldr(tmp2, Address(a2, cnt1));
6349   eor(tmp2, tmp1, tmp2);
6350   cbnz(tmp2, DONE);
6351   b(SAME);
6352 
6353   bind(SHORT);
6354   Label TAIL03, TAIL01;
6355 
6356   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6357   {
6358     ldrw(tmp1, Address(post(a1, 4)));
6359     ldrw(tmp2, Address(post(a2, 4)));
6360     eorw(tmp1, tmp1, tmp2);
6361     cbnzw(tmp1, DONE);
6362   }
6363   bind(TAIL03);
6364   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6365   {
6366     ldrh(tmp1, Address(post(a1, 2)));
6367     ldrh(tmp2, Address(post(a2, 2)));
6368     eorw(tmp1, tmp1, tmp2);
6369     cbnzw(tmp1, DONE);
6370   }
6371   bind(TAIL01);
6372   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6373     {
6374     ldrb(tmp1, a1);
6375     ldrb(tmp2, a2);
6376     eorw(tmp1, tmp1, tmp2);
6377     cbnzw(tmp1, DONE);
6378   }
6379   // Arrays are equal.
6380   bind(SAME);
6381   mov(result, true);
6382 
6383   // That's it.
6384   bind(DONE);
6385   BLOCK_COMMENT("} string_equals");
6386 }
6387 
6388 
6389 // The size of the blocks erased by the zero_blocks stub.  We must
6390 // handle anything smaller than this ourselves in zero_words().
6391 const int MacroAssembler::zero_words_block_size = 8;
6392 
6393 // zero_words() is used by C2 ClearArray patterns and by
6394 // C1_MacroAssembler.  It is as small as possible, handling small word
6395 // counts locally and delegating anything larger to the zero_blocks
6396 // stub.  It is expanded many times in compiled code, so it is
6397 // important to keep it short.
6398 
6399 // ptr:   Address of a buffer to be zeroed.
6400 // cnt:   Count in HeapWords.
6401 //
6402 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6403 address MacroAssembler::zero_words(Register ptr, Register cnt)
6404 {
6405   assert(is_power_of_2(zero_words_block_size), "adjust this");
6406 
6407   BLOCK_COMMENT("zero_words {");
6408   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6409   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6410   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6411 
6412   subs(rscratch1, cnt, zero_words_block_size);
6413   Label around;
6414   br(LO, around);
6415   {
6416     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6417     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6418     // Make sure this is a C2 compilation. C1 allocates space only for
6419     // trampoline stubs generated by Call LIR ops, and in any case it
6420     // makes sense for a C1 compilation task to proceed as quickly as
6421     // possible.
6422     CompileTask* task;
6423     if (StubRoutines::aarch64::complete()
6424         && Thread::current()->is_Compiler_thread()
6425         && (task = ciEnv::current()->task())
6426         && is_c2_compile(task->comp_level())) {
6427       address tpc = trampoline_call(zero_blocks);
6428       if (tpc == nullptr) {
6429         DEBUG_ONLY(reset_labels(around));
6430         return nullptr;
6431       }
6432     } else {
6433       far_call(zero_blocks);
6434     }
6435   }
6436   bind(around);
6437 
6438   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6439   // for us.
6440   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6441     Label l;
6442     tbz(cnt, exact_log2(i), l);
6443     for (int j = 0; j < i; j += 2) {
6444       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6445     }
6446     bind(l);
6447   }
6448   {
6449     Label l;
6450     tbz(cnt, 0, l);
6451     str(zr, Address(ptr));
6452     bind(l);
6453   }
6454 
6455   BLOCK_COMMENT("} zero_words");
6456   return pc();
6457 }
6458 
6459 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6460 // cnt:          Immediate count in HeapWords.
6461 //
6462 // r10, r11, rscratch1, and rscratch2 are clobbered.
6463 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6464 {
6465   assert(wordSize <= BlockZeroingLowLimit,
6466             "increase BlockZeroingLowLimit");
6467   address result = nullptr;
6468   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6469 #ifndef PRODUCT
6470     {
6471       char buf[64];
6472       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6473       BLOCK_COMMENT(buf);
6474     }
6475 #endif
6476     if (cnt >= 16) {
6477       uint64_t loops = cnt/16;
6478       if (loops > 1) {
6479         mov(rscratch2, loops - 1);
6480       }
6481       {
6482         Label loop;
6483         bind(loop);
6484         for (int i = 0; i < 16; i += 2) {
6485           stp(zr, zr, Address(base, i * BytesPerWord));
6486         }
6487         add(base, base, 16 * BytesPerWord);
6488         if (loops > 1) {
6489           subs(rscratch2, rscratch2, 1);
6490           br(GE, loop);
6491         }
6492       }
6493     }
6494     cnt %= 16;
6495     int i = cnt & 1;  // store any odd word to start
6496     if (i) str(zr, Address(base));
6497     for (; i < (int)cnt; i += 2) {
6498       stp(zr, zr, Address(base, i * wordSize));
6499     }
6500     BLOCK_COMMENT("} zero_words");
6501     result = pc();
6502   } else {
6503     mov(r10, base); mov(r11, cnt);
6504     result = zero_words(r10, r11);
6505   }
6506   return result;
6507 }
6508 
6509 // Zero blocks of memory by using DC ZVA.
6510 //
6511 // Aligns the base address first sufficiently for DC ZVA, then uses
6512 // DC ZVA repeatedly for every full block.  cnt is the size to be
6513 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6514 // in cnt.
6515 //
6516 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6517 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6518 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6519   Register tmp = rscratch1;
6520   Register tmp2 = rscratch2;
6521   int zva_length = VM_Version::zva_length();
6522   Label initial_table_end, loop_zva;
6523   Label fini;
6524 
6525   // Base must be 16 byte aligned. If not just return and let caller handle it
6526   tst(base, 0x0f);
6527   br(Assembler::NE, fini);
6528   // Align base with ZVA length.
6529   neg(tmp, base);
6530   andr(tmp, tmp, zva_length - 1);
6531 
6532   // tmp: the number of bytes to be filled to align the base with ZVA length.
6533   add(base, base, tmp);
6534   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6535   adr(tmp2, initial_table_end);
6536   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6537   br(tmp2);
6538 
6539   for (int i = -zva_length + 16; i < 0; i += 16)
6540     stp(zr, zr, Address(base, i));
6541   bind(initial_table_end);
6542 
6543   sub(cnt, cnt, zva_length >> 3);
6544   bind(loop_zva);
6545   dc(Assembler::ZVA, base);
6546   subs(cnt, cnt, zva_length >> 3);
6547   add(base, base, zva_length);
6548   br(Assembler::GE, loop_zva);
6549   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6550   bind(fini);
6551 }
6552 
6553 // base:   Address of a buffer to be filled, 8 bytes aligned.
6554 // cnt:    Count in 8-byte unit.
6555 // value:  Value to be filled with.
6556 // base will point to the end of the buffer after filling.
6557 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6558 {
6559 //  Algorithm:
6560 //
6561 //    if (cnt == 0) {
6562 //      return;
6563 //    }
6564 //    if ((p & 8) != 0) {
6565 //      *p++ = v;
6566 //    }
6567 //
6568 //    scratch1 = cnt & 14;
6569 //    cnt -= scratch1;
6570 //    p += scratch1;
6571 //    switch (scratch1 / 2) {
6572 //      do {
6573 //        cnt -= 16;
6574 //          p[-16] = v;
6575 //          p[-15] = v;
6576 //        case 7:
6577 //          p[-14] = v;
6578 //          p[-13] = v;
6579 //        case 6:
6580 //          p[-12] = v;
6581 //          p[-11] = v;
6582 //          // ...
6583 //        case 1:
6584 //          p[-2] = v;
6585 //          p[-1] = v;
6586 //        case 0:
6587 //          p += 16;
6588 //      } while (cnt);
6589 //    }
6590 //    if ((cnt & 1) == 1) {
6591 //      *p++ = v;
6592 //    }
6593 
6594   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6595 
6596   Label fini, skip, entry, loop;
6597   const int unroll = 8; // Number of stp instructions we'll unroll
6598 
6599   cbz(cnt, fini);
6600   tbz(base, 3, skip);
6601   str(value, Address(post(base, 8)));
6602   sub(cnt, cnt, 1);
6603   bind(skip);
6604 
6605   andr(rscratch1, cnt, (unroll-1) * 2);
6606   sub(cnt, cnt, rscratch1);
6607   add(base, base, rscratch1, Assembler::LSL, 3);
6608   adr(rscratch2, entry);
6609   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6610   br(rscratch2);
6611 
6612   bind(loop);
6613   add(base, base, unroll * 16);
6614   for (int i = -unroll; i < 0; i++)
6615     stp(value, value, Address(base, i * 16));
6616   bind(entry);
6617   subs(cnt, cnt, unroll * 2);
6618   br(Assembler::GE, loop);
6619 
6620   tbz(cnt, 0, fini);
6621   str(value, Address(post(base, 8)));
6622   bind(fini);
6623 }
6624 
6625 // Intrinsic for
6626 //
6627 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6628 //   Encodes char[] to byte[] in ISO-8859-1
6629 //
6630 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6631 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6632 //
6633 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6634 //   Encodes char[] to byte[] in ASCII
6635 //
6636 // This version always returns the number of characters copied, and does not
6637 // clobber the 'len' register. A successful copy will complete with the post-
6638 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6639 // post-condition: 0 <= 'res' < 'len'.
6640 //
6641 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6642 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6643 //       beyond the acceptable, even though the footprint would be smaller.
6644 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6645 //       avoid additional bloat.
6646 //
6647 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6648 void MacroAssembler::encode_iso_array(Register src, Register dst,
6649                                       Register len, Register res, bool ascii,
6650                                       FloatRegister vtmp0, FloatRegister vtmp1,
6651                                       FloatRegister vtmp2, FloatRegister vtmp3,
6652                                       FloatRegister vtmp4, FloatRegister vtmp5)
6653 {
6654   Register cnt = res;
6655   Register max = rscratch1;
6656   Register chk = rscratch2;
6657 
6658   prfm(Address(src), PLDL1STRM);
6659   movw(cnt, len);
6660 
6661 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6662 
6663   Label LOOP_32, DONE_32, FAIL_32;
6664 
6665   BIND(LOOP_32);
6666   {
6667     cmpw(cnt, 32);
6668     br(LT, DONE_32);
6669     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6670     // Extract lower bytes.
6671     FloatRegister vlo0 = vtmp4;
6672     FloatRegister vlo1 = vtmp5;
6673     uzp1(vlo0, T16B, vtmp0, vtmp1);
6674     uzp1(vlo1, T16B, vtmp2, vtmp3);
6675     // Merge bits...
6676     orr(vtmp0, T16B, vtmp0, vtmp1);
6677     orr(vtmp2, T16B, vtmp2, vtmp3);
6678     // Extract merged upper bytes.
6679     FloatRegister vhix = vtmp0;
6680     uzp2(vhix, T16B, vtmp0, vtmp2);
6681     // ISO-check on hi-parts (all zero).
6682     //                          ASCII-check on lo-parts (no sign).
6683     FloatRegister vlox = vtmp1; // Merge lower bytes.
6684                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6685     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6686     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6687     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6688                                 ASCII(orr(chk, chk, max));
6689     cbnz(chk, FAIL_32);
6690     subw(cnt, cnt, 32);
6691     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6692     b(LOOP_32);
6693   }
6694   BIND(FAIL_32);
6695   sub(src, src, 64);
6696   BIND(DONE_32);
6697 
6698   Label LOOP_8, SKIP_8;
6699 
6700   BIND(LOOP_8);
6701   {
6702     cmpw(cnt, 8);
6703     br(LT, SKIP_8);
6704     FloatRegister vhi = vtmp0;
6705     FloatRegister vlo = vtmp1;
6706     ld1(vtmp3, T8H, src);
6707     uzp1(vlo, T16B, vtmp3, vtmp3);
6708     uzp2(vhi, T16B, vtmp3, vtmp3);
6709     // ISO-check on hi-parts (all zero).
6710     //                          ASCII-check on lo-parts (no sign).
6711                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6712     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6713                                 ASCII(umov(max, vtmp2, B, 0));
6714                                 ASCII(orr(chk, chk, max));
6715     cbnz(chk, SKIP_8);
6716 
6717     strd(vlo, Address(post(dst, 8)));
6718     subw(cnt, cnt, 8);
6719     add(src, src, 16);
6720     b(LOOP_8);
6721   }
6722   BIND(SKIP_8);
6723 
6724 #undef ASCII
6725 
6726   Label LOOP, DONE;
6727 
6728   cbz(cnt, DONE);
6729   BIND(LOOP);
6730   {
6731     Register chr = rscratch1;
6732     ldrh(chr, Address(post(src, 2)));
6733     tst(chr, ascii ? 0xff80 : 0xff00);
6734     br(NE, DONE);
6735     strb(chr, Address(post(dst, 1)));
6736     subs(cnt, cnt, 1);
6737     br(GT, LOOP);
6738   }
6739   BIND(DONE);
6740   // Return index where we stopped.
6741   subw(res, len, cnt);
6742 }
6743 
6744 // Inflate byte[] array to char[].
6745 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6746 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6747                                            FloatRegister vtmp1, FloatRegister vtmp2,
6748                                            FloatRegister vtmp3, Register tmp4) {
6749   Label big, done, after_init, to_stub;
6750 
6751   assert_different_registers(src, dst, len, tmp4, rscratch1);
6752 
6753   fmovd(vtmp1, 0.0);
6754   lsrw(tmp4, len, 3);
6755   bind(after_init);
6756   cbnzw(tmp4, big);
6757   // Short string: less than 8 bytes.
6758   {
6759     Label loop, tiny;
6760 
6761     cmpw(len, 4);
6762     br(LT, tiny);
6763     // Use SIMD to do 4 bytes.
6764     ldrs(vtmp2, post(src, 4));
6765     zip1(vtmp3, T8B, vtmp2, vtmp1);
6766     subw(len, len, 4);
6767     strd(vtmp3, post(dst, 8));
6768 
6769     cbzw(len, done);
6770 
6771     // Do the remaining bytes by steam.
6772     bind(loop);
6773     ldrb(tmp4, post(src, 1));
6774     strh(tmp4, post(dst, 2));
6775     subw(len, len, 1);
6776 
6777     bind(tiny);
6778     cbnz(len, loop);
6779 
6780     b(done);
6781   }
6782 
6783   if (SoftwarePrefetchHintDistance >= 0) {
6784     bind(to_stub);
6785       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6786       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6787       address tpc = trampoline_call(stub);
6788       if (tpc == nullptr) {
6789         DEBUG_ONLY(reset_labels(big, done));
6790         postcond(pc() == badAddress);
6791         return nullptr;
6792       }
6793       b(after_init);
6794   }
6795 
6796   // Unpack the bytes 8 at a time.
6797   bind(big);
6798   {
6799     Label loop, around, loop_last, loop_start;
6800 
6801     if (SoftwarePrefetchHintDistance >= 0) {
6802       const int large_loop_threshold = (64 + 16)/8;
6803       ldrd(vtmp2, post(src, 8));
6804       andw(len, len, 7);
6805       cmp(tmp4, (u1)large_loop_threshold);
6806       br(GE, to_stub);
6807       b(loop_start);
6808 
6809       bind(loop);
6810       ldrd(vtmp2, post(src, 8));
6811       bind(loop_start);
6812       subs(tmp4, tmp4, 1);
6813       br(EQ, loop_last);
6814       zip1(vtmp2, T16B, vtmp2, vtmp1);
6815       ldrd(vtmp3, post(src, 8));
6816       st1(vtmp2, T8H, post(dst, 16));
6817       subs(tmp4, tmp4, 1);
6818       zip1(vtmp3, T16B, vtmp3, vtmp1);
6819       st1(vtmp3, T8H, post(dst, 16));
6820       br(NE, loop);
6821       b(around);
6822       bind(loop_last);
6823       zip1(vtmp2, T16B, vtmp2, vtmp1);
6824       st1(vtmp2, T8H, post(dst, 16));
6825       bind(around);
6826       cbz(len, done);
6827     } else {
6828       andw(len, len, 7);
6829       bind(loop);
6830       ldrd(vtmp2, post(src, 8));
6831       sub(tmp4, tmp4, 1);
6832       zip1(vtmp3, T16B, vtmp2, vtmp1);
6833       st1(vtmp3, T8H, post(dst, 16));
6834       cbnz(tmp4, loop);
6835     }
6836   }
6837 
6838   // Do the tail of up to 8 bytes.
6839   add(src, src, len);
6840   ldrd(vtmp3, Address(src, -8));
6841   add(dst, dst, len, ext::uxtw, 1);
6842   zip1(vtmp3, T16B, vtmp3, vtmp1);
6843   strq(vtmp3, Address(dst, -16));
6844 
6845   bind(done);
6846   postcond(pc() != badAddress);
6847   return pc();
6848 }
6849 
6850 // Compress char[] array to byte[].
6851 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6852 // Return the array length if every element in array can be encoded,
6853 // otherwise, the index of first non-latin1 (> 0xff) character.
6854 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6855                                          Register res,
6856                                          FloatRegister tmp0, FloatRegister tmp1,
6857                                          FloatRegister tmp2, FloatRegister tmp3,
6858                                          FloatRegister tmp4, FloatRegister tmp5) {
6859   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6860 }
6861 
6862 // java.math.round(double a)
6863 // Returns the closest long to the argument, with ties rounding to
6864 // positive infinity.  This requires some fiddling for corner
6865 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6866 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6867                                        FloatRegister ftmp) {
6868   Label DONE;
6869   BLOCK_COMMENT("java_round_double: { ");
6870   fmovd(rscratch1, src);
6871   // Use RoundToNearestTiesAway unless src small and -ve.
6872   fcvtasd(dst, src);
6873   // Test if src >= 0 || abs(src) >= 0x1.0p52
6874   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6875   mov(rscratch2, julong_cast(0x1.0p52));
6876   cmp(rscratch1, rscratch2);
6877   br(HS, DONE); {
6878     // src < 0 && abs(src) < 0x1.0p52
6879     // src may have a fractional part, so add 0.5
6880     fmovd(ftmp, 0.5);
6881     faddd(ftmp, src, ftmp);
6882     // Convert double to jlong, use RoundTowardsNegative
6883     fcvtmsd(dst, ftmp);
6884   }
6885   bind(DONE);
6886   BLOCK_COMMENT("} java_round_double");
6887 }
6888 
6889 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6890                                       FloatRegister ftmp) {
6891   Label DONE;
6892   BLOCK_COMMENT("java_round_float: { ");
6893   fmovs(rscratch1, src);
6894   // Use RoundToNearestTiesAway unless src small and -ve.
6895   fcvtassw(dst, src);
6896   // Test if src >= 0 || abs(src) >= 0x1.0p23
6897   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6898   mov(rscratch2, jint_cast(0x1.0p23f));
6899   cmp(rscratch1, rscratch2);
6900   br(HS, DONE); {
6901     // src < 0 && |src| < 0x1.0p23
6902     // src may have a fractional part, so add 0.5
6903     fmovs(ftmp, 0.5f);
6904     fadds(ftmp, src, ftmp);
6905     // Convert float to jint, use RoundTowardsNegative
6906     fcvtmssw(dst, ftmp);
6907   }
6908   bind(DONE);
6909   BLOCK_COMMENT("} java_round_float");
6910 }
6911 
6912 // get_thread() can be called anywhere inside generated code so we
6913 // need to save whatever non-callee save context might get clobbered
6914 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6915 // the call setup code.
6916 //
6917 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6918 // On other systems, the helper is a usual C function.
6919 //
6920 void MacroAssembler::get_thread(Register dst) {
6921   RegSet saved_regs =
6922     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6923     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6924 
6925   protect_return_address();
6926   push(saved_regs, sp);
6927 
6928   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6929   blr(lr);
6930   if (dst != c_rarg0) {
6931     mov(dst, c_rarg0);
6932   }
6933 
6934   pop(saved_regs, sp);
6935   authenticate_return_address();
6936 }
6937 
6938 #ifdef COMPILER2
6939 // C2 compiled method's prolog code
6940 // Moved here from aarch64.ad to support Valhalla code belows
6941 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
6942   if (C->clinit_barrier_on_entry()) {
6943     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
6944 
6945     Label L_skip_barrier;
6946 
6947     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
6948     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
6949     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
6950     bind(L_skip_barrier);
6951   }
6952 
6953   if (C->max_vector_size() > 0) {
6954     reinitialize_ptrue();
6955   }
6956 
6957   int bangsize = C->output()->bang_size_in_bytes();
6958   if (C->output()->need_stack_bang(bangsize))
6959     generate_stack_overflow_check(bangsize);
6960 
6961   // n.b. frame size includes space for return pc and rfp
6962   const long framesize = C->output()->frame_size_in_bytes();
6963   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
6964 
6965   if (C->needs_stack_repair()) {
6966     save_stack_increment(sp_inc, framesize);
6967   }
6968 
6969   if (VerifyStackAtCalls) {
6970     Unimplemented();
6971   }
6972 }
6973 #endif // COMPILER2
6974 
6975 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
6976   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
6977   // An inline type might be returned. If fields are in registers we
6978   // need to allocate an inline type instance and initialize it with
6979   // the value of the fields.
6980   Label skip;
6981   // We only need a new buffered inline type if a new one is not returned
6982   tbz(r0, 0, skip);
6983   int call_offset = -1;
6984 
6985   // Be careful not to clobber r1-7 which hold returned fields
6986   // Also do not use callee-saved registers as these may be live in the interpreter
6987   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
6988 
6989   // The following code is similar to allocate_instance but has some slight differences,
6990   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
6991   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
6992   Label slow_case;
6993   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
6994   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
6995 
6996   if (vk != nullptr) {
6997     // Called from C1, where the return type is statically known.
6998     movptr(klass, (intptr_t)vk->get_InlineKlass());
6999     jint lh = vk->layout_helper();
7000     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7001     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7002       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7003     } else {
7004       b(slow_case);
7005     }
7006   } else {
7007     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7008     andr(klass, r0, -2);
7009     if (UseTLAB) {
7010       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7011       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7012       br(Assembler::NE, slow_case);
7013       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7014     } else {
7015       b(slow_case);
7016     }
7017   }
7018   if (UseTLAB) {
7019     // 2. Initialize buffered inline instance header
7020     Register buffer_obj = r0;
7021     if (UseCompactObjectHeaders) {
7022       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7023       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7024     } else {
7025       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7026       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7027       store_klass_gap(buffer_obj, zr);
7028       if (vk == nullptr) {
7029         // store_klass corrupts klass, so save it for later use (interpreter case only).
7030         mov(tmp1, klass);
7031       }
7032       store_klass(buffer_obj, klass);
7033       klass = tmp1;
7034     }
7035     // 3. Initialize its fields with an inline class specific handler
7036     if (vk != nullptr) {
7037       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7038     } else {
7039       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7040       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7041       blr(tmp1);
7042     }
7043 
7044     membar(Assembler::StoreStore);
7045     b(skip);
7046   } else {
7047     // Must have already branched to slow_case above.
7048     DEBUG_ONLY(should_not_reach_here());
7049   }
7050   bind(slow_case);
7051   // We failed to allocate a new inline type, fall back to a runtime
7052   // call. Some oop field may be live in some registers but we can't
7053   // tell. That runtime call will take care of preserving them
7054   // across a GC if there's one.
7055   mov(r0, r0_preserved);
7056 
7057   if (from_interpreter) {
7058     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7059   } else {
7060     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7061     call_offset = offset();
7062   }
7063   membar(Assembler::StoreStore);
7064 
7065   bind(skip);
7066   return call_offset;
7067 }
7068 
7069 // Move a value between registers/stack slots and update the reg_state
7070 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7071   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7072   if (reg_state[to->value()] == reg_written) {
7073     return true; // Already written
7074   }
7075 
7076   if (from != to && bt != T_VOID) {
7077     if (reg_state[to->value()] == reg_readonly) {
7078       return false; // Not yet writable
7079     }
7080     if (from->is_reg()) {
7081       if (to->is_reg()) {
7082         if (from->is_Register() && to->is_Register()) {
7083           mov(to->as_Register(), from->as_Register());
7084         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7085           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7086         } else {
7087           ShouldNotReachHere();
7088         }
7089       } else {
7090         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7091         Address to_addr = Address(sp, st_off);
7092         if (from->is_FloatRegister()) {
7093           if (bt == T_DOUBLE) {
7094              strd(from->as_FloatRegister(), to_addr);
7095           } else {
7096              assert(bt == T_FLOAT, "must be float");
7097              strs(from->as_FloatRegister(), to_addr);
7098           }
7099         } else {
7100           str(from->as_Register(), to_addr);
7101         }
7102       }
7103     } else {
7104       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7105       if (to->is_reg()) {
7106         if (to->is_FloatRegister()) {
7107           if (bt == T_DOUBLE) {
7108             ldrd(to->as_FloatRegister(), from_addr);
7109           } else {
7110             assert(bt == T_FLOAT, "must be float");
7111             ldrs(to->as_FloatRegister(), from_addr);
7112           }
7113         } else {
7114           ldr(to->as_Register(), from_addr);
7115         }
7116       } else {
7117         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7118         ldr(rscratch1, from_addr);
7119         str(rscratch1, Address(sp, st_off));
7120       }
7121     }
7122   }
7123 
7124   // Update register states
7125   reg_state[from->value()] = reg_writable;
7126   reg_state[to->value()] = reg_written;
7127   return true;
7128 }
7129 
7130 // Calculate the extra stack space required for packing or unpacking inline
7131 // args and adjust the stack pointer
7132 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7133   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7134   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7135   assert(sp_inc > 0, "sanity");
7136 
7137   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7138   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7139 
7140   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7141   if (sp_inc < (1 << 9)) {
7142     sub(sp, sp, sp_inc);   // Fits in an immediate
7143   } else {
7144     mov(rscratch1, sp_inc);
7145     sub(sp, sp, rscratch1);
7146   }
7147 
7148   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7149 }
7150 
7151 // Read all fields from an inline type oop and store the values in registers/stack slots
7152 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7153                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7154                                           RegState reg_state[]) {
7155   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7156   assert(from->is_valid(), "source must be valid");
7157   bool progress = false;
7158 #ifdef ASSERT
7159   const int start_offset = offset();
7160 #endif
7161 
7162   Label L_null, L_notNull;
7163   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7164   // TODO 8366717 We need to make sure that r14 (and potentially other long-life regs) are kept live in slowpath runtime calls in GC barriers
7165   Register tmp1 = r10;
7166   Register tmp2 = r11;
7167   Register fromReg = noreg;
7168   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, -1);
7169   bool done = true;
7170   bool mark_done = true;
7171   VMReg toReg;
7172   BasicType bt;
7173   // Check if argument requires a null check
7174   bool null_check = false;
7175   VMReg nullCheckReg;
7176   while (stream.next(nullCheckReg, bt)) {
7177     if (sig->at(stream.sig_index())._offset == -1) {
7178       null_check = true;
7179       break;
7180     }
7181   }
7182   stream.reset(sig_index, to_index);
7183   while (stream.next(toReg, bt)) {
7184     assert(toReg->is_valid(), "destination must be valid");
7185     int idx = (int)toReg->value();
7186     if (reg_state[idx] == reg_readonly) {
7187       if (idx != from->value()) {
7188         mark_done = false;
7189       }
7190       done = false;
7191       continue;
7192     } else if (reg_state[idx] == reg_written) {
7193       continue;
7194     }
7195     assert(reg_state[idx] == reg_writable, "must be writable");
7196     reg_state[idx] = reg_written;
7197     progress = true;
7198 
7199     if (fromReg == noreg) {
7200       if (from->is_reg()) {
7201         fromReg = from->as_Register();
7202       } else {
7203         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7204         ldr(tmp1, Address(sp, st_off));
7205         fromReg = tmp1;
7206       }
7207       if (null_check) {
7208         // Nullable inline type argument, emit null check
7209         cbz(fromReg, L_null);
7210       }
7211     }
7212     int off = sig->at(stream.sig_index())._offset;
7213     if (off == -1) {
7214       assert(null_check, "Missing null check at");
7215       if (toReg->is_stack()) {
7216         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7217         mov(tmp2, 1);
7218         str(tmp2, Address(sp, st_off));
7219       } else {
7220         mov(toReg->as_Register(), 1);
7221       }
7222       continue;
7223     }
7224     assert(off > 0, "offset in object should be positive");
7225     Address fromAddr = Address(fromReg, off);
7226     if (!toReg->is_FloatRegister()) {
7227       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7228       if (is_reference_type(bt)) {
7229         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7230       } else {
7231         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7232         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7233       }
7234       if (toReg->is_stack()) {
7235         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7236         str(dst, Address(sp, st_off));
7237       }
7238     } else if (bt == T_DOUBLE) {
7239       ldrd(toReg->as_FloatRegister(), fromAddr);
7240     } else {
7241       assert(bt == T_FLOAT, "must be float");
7242       ldrs(toReg->as_FloatRegister(), fromAddr);
7243     }
7244   }
7245   if (progress && null_check) {
7246     if (done) {
7247       b(L_notNull);
7248       bind(L_null);
7249       // Set null marker to zero to signal that the argument is null.
7250       // Also set all oop fields to zero to make the GC happy.
7251       stream.reset(sig_index, to_index);
7252       while (stream.next(toReg, bt)) {
7253         if (sig->at(stream.sig_index())._offset == -1 ||
7254             bt == T_OBJECT || bt == T_ARRAY) {
7255           if (toReg->is_stack()) {
7256             int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7257             str(zr, Address(sp, st_off));
7258           } else {
7259             mov(toReg->as_Register(), zr);
7260           }
7261         }
7262       }
7263       bind(L_notNull);
7264     } else {
7265       bind(L_null);
7266     }
7267   }
7268 
7269   // TODO 8366717 This is probably okay but looks fishy because stream is reset in the "Set null marker to zero" case just above. Same on x64.
7270   sig_index = stream.sig_index();
7271   to_index = stream.regs_index();
7272 
7273   if (mark_done && reg_state[from->value()] != reg_written) {
7274     // This is okay because no one else will write to that slot
7275     reg_state[from->value()] = reg_writable;
7276   }
7277   from_index--;
7278   assert(progress || (start_offset == offset()), "should not emit code");
7279   return done;
7280 }
7281 
7282 // Pack fields back into an inline type oop
7283 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7284                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7285                                         RegState reg_state[], Register val_array) {
7286   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7287   assert(to->is_valid(), "destination must be valid");
7288 
7289   if (reg_state[to->value()] == reg_written) {
7290     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7291     return true; // Already written
7292   }
7293 
7294   // The GC barrier expanded by store_heap_oop below may call into the
7295   // runtime so use callee-saved registers for any values that need to be
7296   // preserved. The GC barrier assembler should take care of saving the
7297   // Java argument registers.
7298   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
7299   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7300   Register val_obj_tmp = r21;
7301   Register from_reg_tmp = r22;
7302   Register tmp1 = r14;
7303   Register tmp2 = r13;
7304   Register tmp3 = r12;
7305   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7306 
7307   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7308 
7309   if (reg_state[to->value()] == reg_readonly) {
7310     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7311       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7312       return false; // Not yet writable
7313     }
7314     val_obj = val_obj_tmp;
7315   }
7316 
7317   int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7318   load_heap_oop(val_obj, Address(val_array, index), tmp1, tmp2);
7319 
7320   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7321   VMReg fromReg;
7322   BasicType bt;
7323   Label L_null;
7324   while (stream.next(fromReg, bt)) {
7325     assert(fromReg->is_valid(), "source must be valid");
7326     reg_state[fromReg->value()] = reg_writable;
7327 
7328     int off = sig->at(stream.sig_index())._offset;
7329     if (off == -1) {
7330       // Nullable inline type argument, emit null check
7331       Label L_notNull;
7332       if (fromReg->is_stack()) {
7333         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7334         ldrb(tmp2, Address(sp, ld_off));
7335         cbnz(tmp2, L_notNull);
7336       } else {
7337         cbnz(fromReg->as_Register(), L_notNull);
7338       }
7339       mov(val_obj, 0);
7340       b(L_null);
7341       bind(L_notNull);
7342       continue;
7343     }
7344 
7345     assert(off > 0, "offset in object should be positive");
7346     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7347 
7348     // Pack the scalarized field into the value object.
7349     Address dst(val_obj, off);
7350     if (!fromReg->is_FloatRegister()) {
7351       Register src;
7352       if (fromReg->is_stack()) {
7353         src = from_reg_tmp;
7354         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7355         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7356       } else {
7357         src = fromReg->as_Register();
7358       }
7359       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7360       if (is_reference_type(bt)) {
7361         store_heap_oop(dst, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7362       } else {
7363         store_sized_value(dst, src, size_in_bytes);
7364       }
7365     } else if (bt == T_DOUBLE) {
7366       strd(fromReg->as_FloatRegister(), dst);
7367     } else {
7368       assert(bt == T_FLOAT, "must be float");
7369       strs(fromReg->as_FloatRegister(), dst);
7370     }
7371   }
7372   bind(L_null);
7373   sig_index = stream.sig_index();
7374   from_index = stream.regs_index();
7375 
7376   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7377   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7378   assert(success, "to register must be writeable");
7379   return true;
7380 }
7381 
7382 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7383   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7384 }
7385 
7386 void MacroAssembler::cache_wb(Address line) {
7387   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7388   assert(line.index() == noreg, "index should be noreg");
7389   assert(line.offset() == 0, "offset should be 0");
7390   // would like to assert this
7391   // assert(line._ext.shift == 0, "shift should be zero");
7392   if (VM_Version::supports_dcpop()) {
7393     // writeback using clear virtual address to point of persistence
7394     dc(Assembler::CVAP, line.base());
7395   } else {
7396     // no need to generate anything as Unsafe.writebackMemory should
7397     // never invoke this stub
7398   }
7399 }
7400 
7401 void MacroAssembler::cache_wbsync(bool is_pre) {
7402   // we only need a barrier post sync
7403   if (!is_pre) {
7404     membar(Assembler::AnyAny);
7405   }
7406 }
7407 
7408 void MacroAssembler::verify_sve_vector_length(Register tmp) {
7409   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
7410     return;
7411   }
7412   // Make sure that native code does not change SVE vector length.
7413   Label verify_ok;
7414   movw(tmp, zr);
7415   sve_inc(tmp, B);
7416   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
7417   br(EQ, verify_ok);
7418   stop("Error: SVE vector length has changed since jvm startup");
7419   bind(verify_ok);
7420 }
7421 
7422 void MacroAssembler::verify_ptrue() {
7423   Label verify_ok;
7424   if (!UseSVE) {
7425     return;
7426   }
7427   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
7428   sve_dec(rscratch1, B);
7429   cbz(rscratch1, verify_ok);
7430   stop("Error: the preserved predicate register (p7) elements are not all true");
7431   bind(verify_ok);
7432 }
7433 
7434 void MacroAssembler::safepoint_isb() {
7435   isb();
7436 #ifndef PRODUCT
7437   if (VerifyCrossModifyFence) {
7438     // Clear the thread state.
7439     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7440   }
7441 #endif
7442 }
7443 
7444 #ifndef PRODUCT
7445 void MacroAssembler::verify_cross_modify_fence_not_required() {
7446   if (VerifyCrossModifyFence) {
7447     // Check if thread needs a cross modify fence.
7448     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7449     Label fence_not_required;
7450     cbz(rscratch1, fence_not_required);
7451     // If it does then fail.
7452     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
7453     mov(c_rarg0, rthread);
7454     blr(rscratch1);
7455     bind(fence_not_required);
7456   }
7457 }
7458 #endif
7459 
7460 void MacroAssembler::spin_wait() {
7461   block_comment("spin_wait {");
7462   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
7463     switch (VM_Version::spin_wait_desc().inst()) {
7464       case SpinWait::NOP:
7465         nop();
7466         break;
7467       case SpinWait::ISB:
7468         isb();
7469         break;
7470       case SpinWait::YIELD:
7471         yield();
7472         break;
7473       case SpinWait::SB:
7474         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
7475         sb();
7476         break;
7477       default:
7478         ShouldNotReachHere();
7479     }
7480   }
7481   block_comment("}");
7482 }
7483 
7484 // Stack frame creation/removal
7485 
7486 void MacroAssembler::enter(bool strip_ret_addr) {
7487   if (strip_ret_addr) {
7488     // Addresses can only be signed once. If there are multiple nested frames being created
7489     // in the same function, then the return address needs stripping first.
7490     strip_return_address();
7491   }
7492   protect_return_address();
7493   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7494   mov(rfp, sp);
7495 }
7496 
7497 void MacroAssembler::leave() {
7498   mov(sp, rfp);
7499   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
7500   authenticate_return_address();
7501 }
7502 
7503 // ROP Protection
7504 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
7505 // destroying stack frames or whenever directly loading/storing the LR to memory.
7506 // If ROP protection is not set then these functions are no-ops.
7507 // For more details on PAC see pauth_aarch64.hpp.
7508 
7509 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
7510 // Uses value zero as the modifier.
7511 //
7512 void MacroAssembler::protect_return_address() {
7513   if (VM_Version::use_rop_protection()) {
7514     check_return_address();
7515     paciaz();
7516   }
7517 }
7518 
7519 // Sign the return value in the given register. Use before updating the LR in the existing stack
7520 // frame for the current function.
7521 // Uses value zero as the modifier.
7522 //
7523 void MacroAssembler::protect_return_address(Register return_reg) {
7524   if (VM_Version::use_rop_protection()) {
7525     check_return_address(return_reg);
7526     paciza(return_reg);
7527   }
7528 }
7529 
7530 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
7531 // Uses value zero as the modifier.
7532 //
7533 void MacroAssembler::authenticate_return_address() {
7534   if (VM_Version::use_rop_protection()) {
7535     autiaz();
7536     check_return_address();
7537   }
7538 }
7539 
7540 // Authenticate the return value in the given register. Use before updating the LR in the existing
7541 // stack frame for the current function.
7542 // Uses value zero as the modifier.
7543 //
7544 void MacroAssembler::authenticate_return_address(Register return_reg) {
7545   if (VM_Version::use_rop_protection()) {
7546     autiza(return_reg);
7547     check_return_address(return_reg);
7548   }
7549 }
7550 
7551 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
7552 // there is no guaranteed way of authenticating the LR.
7553 //
7554 void MacroAssembler::strip_return_address() {
7555   if (VM_Version::use_rop_protection()) {
7556     xpaclri();
7557   }
7558 }
7559 
7560 #ifndef PRODUCT
7561 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
7562 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
7563 // it is difficult to debug back to the callee function.
7564 // This function simply loads from the address in the given register.
7565 // Use directly after authentication to catch authentication failures.
7566 // Also use before signing to check that the pointer is valid and hasn't already been signed.
7567 //
7568 void MacroAssembler::check_return_address(Register return_reg) {
7569   if (VM_Version::use_rop_protection()) {
7570     ldr(zr, Address(return_reg));
7571   }
7572 }
7573 #endif
7574 
7575 // The java_calling_convention describes stack locations as ideal slots on
7576 // a frame with no abi restrictions. Since we must observe abi restrictions
7577 // (like the placement of the register window) the slots must be biased by
7578 // the following value.
7579 static int reg2offset_in(VMReg r) {
7580   // Account for saved rfp and lr
7581   // This should really be in_preserve_stack_slots
7582   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
7583 }
7584 
7585 static int reg2offset_out(VMReg r) {
7586   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
7587 }
7588 
7589 // On 64bit we will store integer like items to the stack as
7590 // 64bits items (AArch64 ABI) even though java would only store
7591 // 32bits for a parameter. On 32bit it will simply be 32bits
7592 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
7593 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
7594   if (src.first()->is_stack()) {
7595     if (dst.first()->is_stack()) {
7596       // stack to stack
7597       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7598       str(tmp, Address(sp, reg2offset_out(dst.first())));
7599     } else {
7600       // stack to reg
7601       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7602     }
7603   } else if (dst.first()->is_stack()) {
7604     // reg to stack
7605     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7606   } else {
7607     if (dst.first() != src.first()) {
7608       sxtw(dst.first()->as_Register(), src.first()->as_Register());
7609     }
7610   }
7611 }
7612 
7613 // An oop arg. Must pass a handle not the oop itself
7614 void MacroAssembler::object_move(
7615                         OopMap* map,
7616                         int oop_handle_offset,
7617                         int framesize_in_slots,
7618                         VMRegPair src,
7619                         VMRegPair dst,
7620                         bool is_receiver,
7621                         int* receiver_offset) {
7622 
7623   // must pass a handle. First figure out the location we use as a handle
7624 
7625   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
7626 
7627   // See if oop is null if it is we need no handle
7628 
7629   if (src.first()->is_stack()) {
7630 
7631     // Oop is already on the stack as an argument
7632     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
7633     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
7634     if (is_receiver) {
7635       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
7636     }
7637 
7638     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7639     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7640     // conditionally move a null
7641     cmp(rscratch1, zr);
7642     csel(rHandle, zr, rHandle, Assembler::EQ);
7643   } else {
7644 
7645     // Oop is in an a register we must store it to the space we reserve
7646     // on the stack for oop_handles and pass a handle if oop is non-null
7647 
7648     const Register rOop = src.first()->as_Register();
7649     int oop_slot;
7650     if (rOop == j_rarg0)
7651       oop_slot = 0;
7652     else if (rOop == j_rarg1)
7653       oop_slot = 1;
7654     else if (rOop == j_rarg2)
7655       oop_slot = 2;
7656     else if (rOop == j_rarg3)
7657       oop_slot = 3;
7658     else if (rOop == j_rarg4)
7659       oop_slot = 4;
7660     else if (rOop == j_rarg5)
7661       oop_slot = 5;
7662     else if (rOop == j_rarg6)
7663       oop_slot = 6;
7664     else {
7665       assert(rOop == j_rarg7, "wrong register");
7666       oop_slot = 7;
7667     }
7668 
7669     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7670     int offset = oop_slot*VMRegImpl::stack_slot_size;
7671 
7672     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7673     // Store oop in handle area, may be null
7674     str(rOop, Address(sp, offset));
7675     if (is_receiver) {
7676       *receiver_offset = offset;
7677     }
7678 
7679     cmp(rOop, zr);
7680     lea(rHandle, Address(sp, offset));
7681     // conditionally move a null
7682     csel(rHandle, zr, rHandle, Assembler::EQ);
7683   }
7684 
7685   // If arg is on the stack then place it otherwise it is already in correct reg.
7686   if (dst.first()->is_stack()) {
7687     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7688   }
7689 }
7690 
7691 // A float arg may have to do float reg int reg conversion
7692 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7693  if (src.first()->is_stack()) {
7694     if (dst.first()->is_stack()) {
7695       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7696       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7697     } else {
7698       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7699     }
7700   } else if (src.first() != dst.first()) {
7701     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7702       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7703     else
7704       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7705   }
7706 }
7707 
7708 // A long move
7709 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7710   if (src.first()->is_stack()) {
7711     if (dst.first()->is_stack()) {
7712       // stack to stack
7713       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7714       str(tmp, Address(sp, reg2offset_out(dst.first())));
7715     } else {
7716       // stack to reg
7717       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7718     }
7719   } else if (dst.first()->is_stack()) {
7720     // reg to stack
7721     // Do we really have to sign extend???
7722     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7723     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7724   } else {
7725     if (dst.first() != src.first()) {
7726       mov(dst.first()->as_Register(), src.first()->as_Register());
7727     }
7728   }
7729 }
7730 
7731 
7732 // A double move
7733 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7734  if (src.first()->is_stack()) {
7735     if (dst.first()->is_stack()) {
7736       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7737       str(tmp, Address(sp, reg2offset_out(dst.first())));
7738     } else {
7739       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7740     }
7741   } else if (src.first() != dst.first()) {
7742     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7743       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7744     else
7745       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7746   }
7747 }
7748 
7749 // Implements fast-locking.
7750 //
7751 //  - obj: the object to be locked
7752 //  - t1, t2, t3: temporary registers, will be destroyed
7753 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7754 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7755   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7756 
7757   Label push;
7758   const Register top = t1;
7759   const Register mark = t2;
7760   const Register t = t3;
7761 
7762   // Preload the markWord. It is important that this is the first
7763   // instruction emitted as it is part of C1's null check semantics.
7764   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7765 
7766   if (UseObjectMonitorTable) {
7767     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7768     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7769   }
7770 
7771   if (DiagnoseSyncOnValueBasedClasses != 0) {
7772     load_klass(t1, obj);
7773     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7774     tst(t1, KlassFlags::_misc_is_value_based_class);
7775     br(Assembler::NE, slow);
7776   }
7777 
7778   // Check if the lock-stack is full.
7779   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7780   cmpw(top, (unsigned)LockStack::end_offset());
7781   br(Assembler::GE, slow);
7782 
7783   // Check for recursion.
7784   subw(t, top, oopSize);
7785   ldr(t, Address(rthread, t));
7786   cmp(obj, t);
7787   br(Assembler::EQ, push);
7788 
7789   // Check header for monitor (0b10).
7790   tst(mark, markWord::monitor_value);
7791   br(Assembler::NE, slow);
7792 
7793   // Try to lock. Transition lock bits 0b01 => 0b00
7794   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7795   orr(mark, mark, markWord::unlocked_value);
7796   // Mask inline_type bit such that we go to the slow path if object is an inline type
7797   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7798 
7799   eor(t, mark, markWord::unlocked_value);
7800   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7801           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7802   br(Assembler::NE, slow);
7803 
7804   bind(push);
7805   // After successful lock, push object on lock-stack.
7806   str(obj, Address(rthread, top));
7807   addw(top, top, oopSize);
7808   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7809 }
7810 
7811 // Implements fast-unlocking.
7812 //
7813 // - obj: the object to be unlocked
7814 // - t1, t2, t3: temporary registers
7815 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7816 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7817   // cmpxchg clobbers rscratch1.
7818   assert_different_registers(obj, t1, t2, t3, rscratch1);
7819 
7820 #ifdef ASSERT
7821   {
7822     // Check for lock-stack underflow.
7823     Label stack_ok;
7824     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7825     cmpw(t1, (unsigned)LockStack::start_offset());
7826     br(Assembler::GE, stack_ok);
7827     STOP("Lock-stack underflow");
7828     bind(stack_ok);
7829   }
7830 #endif
7831 
7832   Label unlocked, push_and_slow;
7833   const Register top = t1;
7834   const Register mark = t2;
7835   const Register t = t3;
7836 
7837   // Check if obj is top of lock-stack.
7838   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7839   subw(top, top, oopSize);
7840   ldr(t, Address(rthread, top));
7841   cmp(obj, t);
7842   br(Assembler::NE, slow);
7843 
7844   // Pop lock-stack.
7845   DEBUG_ONLY(str(zr, Address(rthread, top));)
7846   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7847 
7848   // Check if recursive.
7849   subw(t, top, oopSize);
7850   ldr(t, Address(rthread, t));
7851   cmp(obj, t);
7852   br(Assembler::EQ, unlocked);
7853 
7854   // Not recursive. Check header for monitor (0b10).
7855   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7856   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7857 
7858 #ifdef ASSERT
7859   // Check header not unlocked (0b01).
7860   Label not_unlocked;
7861   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7862   stop("fast_unlock already unlocked");
7863   bind(not_unlocked);
7864 #endif
7865 
7866   // Try to unlock. Transition lock bits 0b00 => 0b01
7867   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7868   orr(t, mark, markWord::unlocked_value);
7869   cmpxchg(obj, mark, t, Assembler::xword,
7870           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7871   br(Assembler::EQ, unlocked);
7872 
7873   bind(push_and_slow);
7874   // Restore lock-stack and handle the unlock in runtime.
7875   DEBUG_ONLY(str(obj, Address(rthread, top));)
7876   addw(top, top, oopSize);
7877   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7878   b(slow);
7879 
7880   bind(unlocked);
7881 }
--- EOF ---