1 /*
   2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "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     target = address(target_page);
 416     precond(inner != nullptr);
 417     inner(insn_addr, target);
 418     return 2;
 419   }
 420   static int adrpMem(address insn_addr, address &target) {
 421     uint32_t insn2 = insn_at(insn_addr, 1);
 422     // Load/store register (unsigned immediate)
 423     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 424     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 425     byte_offset <<= size;
 426     target += byte_offset;
 427     return 2;
 428   }
 429   static int adrpAdd(address insn_addr, address &target) {
 430     uint32_t insn2 = insn_at(insn_addr, 1);
 431     // add (immediate)
 432     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 433     target += byte_offset;
 434     return 2;
 435   }
 436   static int adrpMovk(address insn_addr, address &target) {
 437     uint32_t insn2 = insn_at(insn_addr, 1);
 438     uint64_t dest = uint64_t(target);
 439     dest = (dest & 0xffff0000ffffffff) |
 440       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 441     target = address(dest);
 442 
 443     // We know the destination 4k page. Maybe we have a third
 444     // instruction.
 445     uint32_t insn = insn_at(insn_addr, 0);
 446     uint32_t insn3 = insn_at(insn_addr, 2);
 447     ptrdiff_t byte_offset;
 448     if (offset_for(insn, insn3, byte_offset)) {
 449       target += byte_offset;
 450       return 3;
 451     } else {
 452       return 2;
 453     }
 454   }
 455   static int immediate(address insn_addr, address &target) {
 456     uint32_t *insns = (uint32_t *)insn_addr;
 457     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 458     // Move wide constant: movz, movk, movk.  See movptr().
 459     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 460     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 461     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 462                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 463                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 464     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 465     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 466     return 3;
 467   }
 468   static void verify(address insn_addr, address &target) {
 469   }
 470 };
 471 
 472 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 473   address target;
 474   RelocActions<AArch64Decoder>::run(insn_addr, target);
 475   return target;
 476 }
 477 
 478 // Patch any kind of instruction; there may be several instructions.
 479 // Return the total length (in bytes) of the instructions.
 480 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 481   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 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   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 491 
 492   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 493   // narrow OOPs by setting the upper 16 bits in the first
 494   // instruction.
 495   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 496     // Move narrow OOP
 497     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 498     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 499     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 500     instructions = 2;
 501   } else {
 502     // Move wide OOP
 503     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 504     uintptr_t dest = (uintptr_t)o;
 505     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 506     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 507     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 508     instructions = 3;
 509   }
 510   return instructions * NativeInstruction::instruction_size;
 511 }
 512 
 513 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 514   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 515   // We encode narrow ones by setting the upper 16 bits in the first
 516   // instruction.
 517   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 518   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 519          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 520 
 521   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 522 
 523   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 524   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 525   return 2 * NativeInstruction::instruction_size;
 526 }
 527 
 528 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 529   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 530   if (at_return) {
 531     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 532     // we may safely use the sp instead to perform the stack watermark check.
 533     cmp(in_nmethod ? sp : rfp, tmp);
 534     br(Assembler::HI, slow_path);
 535   } else {
 536     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 537   }
 538 }
 539 
 540 void MacroAssembler::rt_call(address dest, Register tmp) {
 541   CodeBlob *cb = CodeCache::find_blob(dest);
 542   if (cb) {
 543     far_call(RuntimeAddress(dest));
 544   } else {
 545     lea(tmp, RuntimeAddress(dest));
 546     blr(tmp);
 547   }
 548 }
 549 
 550 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 551   if (!Continuations::enabled()) return;
 552   Label done;
 553   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 554   cmp(sp, rscratch1);
 555   br(Assembler::LS, done);
 556   mov(rscratch1, sp); // we can't use sp as the source in str
 557   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 558   bind(done);
 559 }
 560 
 561 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 562   if (!Continuations::enabled()) return;
 563   Label done;
 564   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 565   cmp(sp, rscratch1);
 566   br(Assembler::LO, done);
 567   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 568   bind(done);
 569 }
 570 
 571 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 572   // we must set sp to zero to clear frame
 573   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 574 
 575   // must clear fp, so that compiled frames are not confused; it is
 576   // possible that we need it only for debugging
 577   if (clear_fp) {
 578     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 579   }
 580 
 581   // Always clear the pc because it could have been set by make_walkable()
 582   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 583 }
 584 
 585 // Calls to C land
 586 //
 587 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 588 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 589 // has to be reset to 0. This is required to allow proper stack traversal.
 590 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 591                                          Register last_java_fp,
 592                                          Register last_java_pc,
 593                                          Register scratch) {
 594 
 595   if (last_java_pc->is_valid()) {
 596       str(last_java_pc, Address(rthread,
 597                                 JavaThread::frame_anchor_offset()
 598                                 + JavaFrameAnchor::last_Java_pc_offset()));
 599     }
 600 
 601   // determine last_java_sp register
 602   if (last_java_sp == sp) {
 603     mov(scratch, sp);
 604     last_java_sp = scratch;
 605   } else if (!last_java_sp->is_valid()) {
 606     last_java_sp = esp;
 607   }
 608 
 609   // last_java_fp is optional
 610   if (last_java_fp->is_valid()) {
 611     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 612   }
 613 
 614   // We must set sp last.
 615   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 616 }
 617 
 618 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 619                                          Register last_java_fp,
 620                                          address  last_java_pc,
 621                                          Register scratch) {
 622   assert(last_java_pc != nullptr, "must provide a valid PC");
 623 
 624   adr(scratch, last_java_pc);
 625   str(scratch, Address(rthread,
 626                        JavaThread::frame_anchor_offset()
 627                        + JavaFrameAnchor::last_Java_pc_offset()));
 628 
 629   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 630 }
 631 
 632 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 633                                          Register last_java_fp,
 634                                          Label &L,
 635                                          Register scratch) {
 636   if (L.is_bound()) {
 637     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 638   } else {
 639     InstructionMark im(this);
 640     L.add_patch_at(code(), locator());
 641     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 642   }
 643 }
 644 
 645 static inline bool target_needs_far_branch(address addr) {
 646   if (AOTCodeCache::is_on_for_dump()) {
 647     return true;
 648   }
 649   // codecache size <= 128M
 650   if (!MacroAssembler::far_branches()) {
 651     return false;
 652   }
 653   // codecache size > 240M
 654   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 655     return true;
 656   }
 657   // codecache size: 128M..240M
 658   return !CodeCache::is_non_nmethod(addr);
 659 }
 660 
 661 void MacroAssembler::far_call(Address entry, Register tmp) {
 662   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 663   assert(CodeCache::find_blob(entry.target()) != nullptr,
 664          "destination of far call not found in code cache");
 665   assert(entry.rspec().type() == relocInfo::external_word_type
 666          || entry.rspec().type() == relocInfo::runtime_call_type
 667          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 668   if (target_needs_far_branch(entry.target())) {
 669     uint64_t offset;
 670     // We can use ADRP here because we know that the total size of
 671     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 672     adrp(tmp, entry, offset);
 673     add(tmp, tmp, offset);
 674     blr(tmp);
 675   } else {
 676     bl(entry);
 677   }
 678 }
 679 
 680 int MacroAssembler::far_jump(Address entry, Register tmp) {
 681   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 682   assert(CodeCache::find_blob(entry.target()) != nullptr,
 683          "destination of far call not found in code cache");
 684   assert(entry.rspec().type() == relocInfo::external_word_type
 685          || entry.rspec().type() == relocInfo::runtime_call_type
 686          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 687   address start = pc();
 688   if (target_needs_far_branch(entry.target())) {
 689     uint64_t offset;
 690     // We can use ADRP here because we know that the total size of
 691     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 692     adrp(tmp, entry, offset);
 693     add(tmp, tmp, offset);
 694     br(tmp);
 695   } else {
 696     b(entry);
 697   }
 698   return pc() - start;
 699 }
 700 
 701 void MacroAssembler::reserved_stack_check() {
 702     // testing if reserved zone needs to be enabled
 703     Label no_reserved_zone_enabling;
 704 
 705     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 706     cmp(sp, rscratch1);
 707     br(Assembler::LO, no_reserved_zone_enabling);
 708 
 709     enter();   // LR and FP are live.
 710     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 711     mov(c_rarg0, rthread);
 712     blr(rscratch1);
 713     leave();
 714 
 715     // We have already removed our own frame.
 716     // throw_delayed_StackOverflowError will think that it's been
 717     // called by our caller.
 718     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 719     br(rscratch1);
 720     should_not_reach_here();
 721 
 722     bind(no_reserved_zone_enabling);
 723 }
 724 
 725 static void pass_arg0(MacroAssembler* masm, Register arg) {
 726   if (c_rarg0 != arg ) {
 727     masm->mov(c_rarg0, arg);
 728   }
 729 }
 730 
 731 static void pass_arg1(MacroAssembler* masm, Register arg) {
 732   if (c_rarg1 != arg ) {
 733     masm->mov(c_rarg1, arg);
 734   }
 735 }
 736 
 737 static void pass_arg2(MacroAssembler* masm, Register arg) {
 738   if (c_rarg2 != arg ) {
 739     masm->mov(c_rarg2, arg);
 740   }
 741 }
 742 
 743 static void pass_arg3(MacroAssembler* masm, Register arg) {
 744   if (c_rarg3 != arg ) {
 745     masm->mov(c_rarg3, arg);
 746   }
 747 }
 748 
 749 void MacroAssembler::call_VM_base(Register oop_result,
 750                                   Register java_thread,
 751                                   Register last_java_sp,
 752                                   Label*   return_pc,
 753                                   address  entry_point,
 754                                   int      number_of_arguments,
 755                                   bool     check_exceptions) {
 756    // determine java_thread register
 757   if (!java_thread->is_valid()) {
 758     java_thread = rthread;
 759   }
 760 
 761   // determine last_java_sp register
 762   if (!last_java_sp->is_valid()) {
 763     last_java_sp = esp;
 764   }
 765 
 766   // debugging support
 767   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 768   assert(java_thread == rthread, "unexpected register");
 769 #ifdef ASSERT
 770   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 771   // if (!TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 772 #endif // ASSERT
 773 
 774   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 775   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 776 
 777   // push java thread (becomes first argument of C function)
 778 
 779   mov(c_rarg0, java_thread);
 780 
 781   // set last Java frame before call
 782   assert(last_java_sp != rfp, "can't use rfp");
 783 
 784   Label l;
 785   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 786 
 787   // do the call, remove parameters
 788   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 789 
 790   // lr could be poisoned with PAC signature during throw_pending_exception
 791   // if it was tail-call optimized by compiler, since lr is not callee-saved
 792   // reload it with proper value
 793   adr(lr, l);
 794 
 795   // reset last Java frame
 796   // Only interpreter should have to clear fp
 797   reset_last_Java_frame(true);
 798 
 799    // C++ interp handles this in the interpreter
 800   check_and_handle_popframe(java_thread);
 801   check_and_handle_earlyret(java_thread);
 802 
 803   if (check_exceptions) {
 804     // check for pending exceptions (java_thread is set upon return)
 805     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 806     Label ok;
 807     cbz(rscratch1, ok);
 808     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 809     br(rscratch1);
 810     bind(ok);
 811   }
 812 
 813   // get oop result if there is one and reset the value in the thread
 814   if (oop_result->is_valid()) {
 815     get_vm_result_oop(oop_result, java_thread);
 816   }
 817 }
 818 
 819 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 820   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 821 }
 822 
 823 // Check the entry target is always reachable from any branch.
 824 static bool is_always_within_branch_range(Address entry) {
 825   if (AOTCodeCache::is_on_for_dump()) {
 826     return false;
 827   }
 828   const address target = entry.target();
 829 
 830   if (!CodeCache::contains(target)) {
 831     // We always use trampolines for callees outside CodeCache.
 832     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 833     return false;
 834   }
 835 
 836   if (!MacroAssembler::far_branches()) {
 837     return true;
 838   }
 839 
 840   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 841     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 842     // Non-compiled methods stay forever in CodeCache.
 843     // We check whether the longest possible branch is within the branch range.
 844     assert(CodeCache::find_blob(target) != nullptr &&
 845           !CodeCache::find_blob(target)->is_nmethod(),
 846           "runtime call of compiled method");
 847     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 848     const address left_longest_branch_start = CodeCache::low_bound();
 849     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 850                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 851     return is_reachable;
 852   }
 853 
 854   return false;
 855 }
 856 
 857 // Maybe emit a call via a trampoline. If the code cache is small
 858 // trampolines won't be emitted.
 859 address MacroAssembler::trampoline_call(Address entry) {
 860   assert(entry.rspec().type() == relocInfo::runtime_call_type
 861          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 862          || entry.rspec().type() == relocInfo::static_call_type
 863          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 864 
 865   address target = entry.target();
 866 
 867   if (!is_always_within_branch_range(entry)) {
 868     if (!in_scratch_emit_size()) {
 869       // We don't want to emit a trampoline if C2 is generating dummy
 870       // code during its branch shortening phase.
 871       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 872         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 873         code()->share_trampoline_for(entry.target(), offset());
 874       } else {
 875         address stub = emit_trampoline_stub(offset(), target);
 876         if (stub == nullptr) {
 877           postcond(pc() == badAddress);
 878           return nullptr; // CodeCache is full
 879         }
 880       }
 881     }
 882     target = pc();
 883   }
 884 
 885   address call_pc = pc();
 886   relocate(entry.rspec());
 887   bl(target);
 888 
 889   postcond(pc() != badAddress);
 890   return call_pc;
 891 }
 892 
 893 // Emit a trampoline stub for a call to a target which is too far away.
 894 //
 895 // code sequences:
 896 //
 897 // call-site:
 898 //   branch-and-link to <destination> or <trampoline stub>
 899 //
 900 // Related trampoline stub for this call site in the stub section:
 901 //   load the call target from the constant pool
 902 //   branch (LR still points to the call site above)
 903 
 904 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 905                                              address dest) {
 906   // Max stub size: alignment nop, TrampolineStub.
 907   address stub = start_a_stub(max_trampoline_stub_size());
 908   if (stub == nullptr) {
 909     return nullptr;  // CodeBuffer::expand failed
 910   }
 911 
 912   // Create a trampoline stub relocation which relates this trampoline stub
 913   // with the call instruction at insts_call_instruction_offset in the
 914   // instructions code-section.
 915   align(wordSize);
 916   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 917                                             + insts_call_instruction_offset));
 918   const int stub_start_offset = offset();
 919 
 920   // Now, create the trampoline stub's code:
 921   // - load the call
 922   // - call
 923   Label target;
 924   ldr(rscratch1, target);
 925   br(rscratch1);
 926   bind(target);
 927   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 928          "should be");
 929   emit_int64((int64_t)dest);
 930 
 931   const address stub_start_addr = addr_at(stub_start_offset);
 932 
 933   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 934 
 935   end_a_stub();
 936   return stub_start_addr;
 937 }
 938 
 939 int MacroAssembler::max_trampoline_stub_size() {
 940   // Max stub size: alignment nop, TrampolineStub.
 941   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 942 }
 943 
 944 void MacroAssembler::emit_static_call_stub() {
 945   // CompiledDirectCall::set_to_interpreted knows the
 946   // exact layout of this stub.
 947 
 948   isb();
 949   mov_metadata(rmethod, nullptr);
 950 
 951   // Jump to the entry point of the c2i stub.
 952   if (codestub_branch_needs_far_jump()) {
 953     movptr(rscratch1, 0);
 954     br(rscratch1);
 955   } else {
 956     b(pc());
 957   }
 958 }
 959 
 960 int MacroAssembler::static_call_stub_size() {
 961   if (!codestub_branch_needs_far_jump()) {
 962     // isb; movk; movz; movz; b
 963     return 5 * NativeInstruction::instruction_size;
 964   }
 965   // isb; movk; movz; movz; movk; movz; movz; br
 966   return 8 * NativeInstruction::instruction_size;
 967 }
 968 
 969 void MacroAssembler::c2bool(Register x) {
 970   // implements x == 0 ? 0 : 1
 971   // note: must only look at least-significant byte of x
 972   //       since C-style booleans are stored in one byte
 973   //       only! (was bug)
 974   tst(x, 0xff);
 975   cset(x, Assembler::NE);
 976 }
 977 
 978 address MacroAssembler::ic_call(address entry, jint method_index) {
 979   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 980   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 981   return trampoline_call(Address(entry, rh));
 982 }
 983 
 984 int MacroAssembler::ic_check_size() {
 985   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 986   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 987     return NativeInstruction::instruction_size * (7 + extra_instructions);
 988   } else {
 989     return NativeInstruction::instruction_size * (5 + extra_instructions);
 990   }
 991 }
 992 
 993 int MacroAssembler::ic_check(int end_alignment) {
 994   Register receiver = j_rarg0;
 995   Register data = rscratch2;
 996   Register tmp1 = rscratch1;
 997   Register tmp2 = r10;
 998 
 999   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
1000   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
1001   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1002   // before the inline cache check here, and not after
1003   align(end_alignment, offset() + ic_check_size());
1004 
1005   int uep_offset = offset();
1006 
1007   if (UseCompactObjectHeaders) {
1008     load_narrow_klass_compact(tmp1, receiver);
1009     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1010     cmpw(tmp1, tmp2);
1011   } else {
1012     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1013     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1014     cmpw(tmp1, tmp2);
1015   }
1016 
1017   Label dont;
1018   br(Assembler::EQ, dont);
1019   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1020   bind(dont);
1021   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1022 
1023   return uep_offset;
1024 }
1025 
1026 // Implementation of call_VM versions
1027 
1028 void MacroAssembler::call_VM(Register oop_result,
1029                              address entry_point,
1030                              bool check_exceptions) {
1031   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1032 }
1033 
1034 void MacroAssembler::call_VM(Register oop_result,
1035                              address entry_point,
1036                              Register arg_1,
1037                              bool check_exceptions) {
1038   pass_arg1(this, arg_1);
1039   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1040 }
1041 
1042 void MacroAssembler::call_VM(Register oop_result,
1043                              address entry_point,
1044                              Register arg_1,
1045                              Register arg_2,
1046                              bool check_exceptions) {
1047   assert_different_registers(arg_1, c_rarg2);
1048   pass_arg2(this, arg_2);
1049   pass_arg1(this, arg_1);
1050   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1051 }
1052 
1053 void MacroAssembler::call_VM(Register oop_result,
1054                              address entry_point,
1055                              Register arg_1,
1056                              Register arg_2,
1057                              Register arg_3,
1058                              bool check_exceptions) {
1059   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1060   assert_different_registers(arg_2, c_rarg3);
1061   pass_arg3(this, arg_3);
1062 
1063   pass_arg2(this, arg_2);
1064 
1065   pass_arg1(this, arg_1);
1066   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1067 }
1068 
1069 void MacroAssembler::call_VM(Register oop_result,
1070                              Register last_java_sp,
1071                              address entry_point,
1072                              int number_of_arguments,
1073                              bool check_exceptions) {
1074   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1075 }
1076 
1077 void MacroAssembler::call_VM(Register oop_result,
1078                              Register last_java_sp,
1079                              address entry_point,
1080                              Register arg_1,
1081                              bool check_exceptions) {
1082   pass_arg1(this, arg_1);
1083   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1084 }
1085 
1086 void MacroAssembler::call_VM(Register oop_result,
1087                              Register last_java_sp,
1088                              address entry_point,
1089                              Register arg_1,
1090                              Register arg_2,
1091                              bool check_exceptions) {
1092 
1093   assert_different_registers(arg_1, c_rarg2);
1094   pass_arg2(this, arg_2);
1095   pass_arg1(this, arg_1);
1096   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1097 }
1098 
1099 void MacroAssembler::call_VM(Register oop_result,
1100                              Register last_java_sp,
1101                              address entry_point,
1102                              Register arg_1,
1103                              Register arg_2,
1104                              Register arg_3,
1105                              bool check_exceptions) {
1106   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1107   assert_different_registers(arg_2, c_rarg3);
1108   pass_arg3(this, arg_3);
1109   pass_arg2(this, arg_2);
1110   pass_arg1(this, arg_1);
1111   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1112 }
1113 
1114 
1115 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1116   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1117   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1118   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1119 }
1120 
1121 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1122   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1123   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1124 }
1125 
1126 void MacroAssembler::align(int modulus) {
1127   align(modulus, offset());
1128 }
1129 
1130 // Ensure that the code at target bytes offset from the current offset() is aligned
1131 // according to modulus.
1132 void MacroAssembler::align(int modulus, int target) {
1133   int delta = target - offset();
1134   while ((offset() + delta) % modulus != 0) nop();
1135 }
1136 
1137 void MacroAssembler::post_call_nop() {
1138   if (!Continuations::enabled()) {
1139     return;
1140   }
1141   InstructionMark im(this);
1142   relocate(post_call_nop_Relocation::spec());
1143   InlineSkippedInstructionsCounter skipCounter(this);
1144   nop();
1145   movk(zr, 0);
1146   movk(zr, 0);
1147 }
1148 
1149 // these are no-ops overridden by InterpreterMacroAssembler
1150 
1151 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1152 
1153 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1154 
1155 // Look up the method for a megamorphic invokeinterface call.
1156 // The target method is determined by <intf_klass, itable_index>.
1157 // The receiver klass is in recv_klass.
1158 // On success, the result will be in method_result, and execution falls through.
1159 // On failure, execution transfers to the given label.
1160 void MacroAssembler::lookup_interface_method(Register recv_klass,
1161                                              Register intf_klass,
1162                                              RegisterOrConstant itable_index,
1163                                              Register method_result,
1164                                              Register scan_temp,
1165                                              Label& L_no_such_interface,
1166                          bool return_method) {
1167   assert_different_registers(recv_klass, intf_klass, scan_temp);
1168   assert_different_registers(method_result, intf_klass, scan_temp);
1169   assert(recv_klass != method_result || !return_method,
1170      "recv_klass can be destroyed when method isn't needed");
1171   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1172          "caller must use same register for non-constant itable index as for method");
1173 
1174   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1175   int vtable_base = in_bytes(Klass::vtable_start_offset());
1176   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1177   int scan_step   = itableOffsetEntry::size() * wordSize;
1178   int vte_size    = vtableEntry::size_in_bytes();
1179   assert(vte_size == wordSize, "else adjust times_vte_scale");
1180 
1181   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1182 
1183   // Could store the aligned, prescaled offset in the klass.
1184   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1185   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1186   add(scan_temp, scan_temp, vtable_base);
1187 
1188   if (return_method) {
1189     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1190     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1191     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1192     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1193     if (itentry_off)
1194       add(recv_klass, recv_klass, itentry_off);
1195   }
1196 
1197   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1198   //   if (scan->interface() == intf) {
1199   //     result = (klass + scan->offset() + itable_index);
1200   //   }
1201   // }
1202   Label search, found_method;
1203 
1204   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1205   cmp(intf_klass, method_result);
1206   br(Assembler::EQ, found_method);
1207   bind(search);
1208   // Check that the previous entry is non-null.  A null entry means that
1209   // the receiver class doesn't implement the interface, and wasn't the
1210   // same as when the caller was compiled.
1211   cbz(method_result, L_no_such_interface);
1212   if (itableOffsetEntry::interface_offset() != 0) {
1213     add(scan_temp, scan_temp, scan_step);
1214     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1215   } else {
1216     ldr(method_result, Address(pre(scan_temp, scan_step)));
1217   }
1218   cmp(intf_klass, method_result);
1219   br(Assembler::NE, search);
1220 
1221   bind(found_method);
1222 
1223   // Got a hit.
1224   if (return_method) {
1225     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1226     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1227   }
1228 }
1229 
1230 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1231 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1232 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1233 // The target method is determined by <holder_klass, itable_index>.
1234 // The receiver klass is in recv_klass.
1235 // On success, the result will be in method_result, and execution falls through.
1236 // On failure, execution transfers to the given label.
1237 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1238                                                   Register holder_klass,
1239                                                   Register resolved_klass,
1240                                                   Register method_result,
1241                                                   Register temp_itbl_klass,
1242                                                   Register scan_temp,
1243                                                   int itable_index,
1244                                                   Label& L_no_such_interface) {
1245   // 'method_result' is only used as output register at the very end of this method.
1246   // Until then we can reuse it as 'holder_offset'.
1247   Register holder_offset = method_result;
1248   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1249 
1250   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1251   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1252   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1253   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1254 
1255   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1256 
1257   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1258   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1259   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1260   // temp_itbl_klass = itable[0]._interface;
1261   int vtblEntrySize = vtableEntry::size_in_bytes();
1262   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1263   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1264   mov(holder_offset, zr);
1265   // scan_temp = &(itable[0]._interface)
1266   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1267 
1268   // Initial checks:
1269   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1270   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1271   //   - if (itable[0] == 0), no such interface
1272   cmp(resolved_klass, holder_klass);
1273   br(Assembler::NE, L_loop_search_resolved_entry);
1274   cmp(holder_klass, temp_itbl_klass);
1275   br(Assembler::EQ, L_holder_found);
1276   cbz(temp_itbl_klass, L_no_such_interface);
1277 
1278   // Loop: Look for holder_klass record in itable
1279   //   do {
1280   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1281   //     if (temp_itbl_klass == holder_klass) {
1282   //       goto L_holder_found; // Found!
1283   //     }
1284   //   } while (temp_itbl_klass != 0);
1285   //   goto L_no_such_interface // Not found.
1286   Label L_search_holder;
1287   bind(L_search_holder);
1288     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1289     cmp(holder_klass, temp_itbl_klass);
1290     br(Assembler::EQ, L_holder_found);
1291     cbnz(temp_itbl_klass, L_search_holder);
1292 
1293   b(L_no_such_interface);
1294 
1295   // Loop: Look for resolved_class record in itable
1296   //   while (true) {
1297   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1298   //     if (temp_itbl_klass == 0) {
1299   //       goto L_no_such_interface;
1300   //     }
1301   //     if (temp_itbl_klass == resolved_klass) {
1302   //        goto L_resolved_found;  // Found!
1303   //     }
1304   //     if (temp_itbl_klass == holder_klass) {
1305   //        holder_offset = scan_temp;
1306   //     }
1307   //   }
1308   //
1309   Label L_loop_search_resolved;
1310   bind(L_loop_search_resolved);
1311     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1312   bind(L_loop_search_resolved_entry);
1313     cbz(temp_itbl_klass, L_no_such_interface);
1314     cmp(resolved_klass, temp_itbl_klass);
1315     br(Assembler::EQ, L_resolved_found);
1316     cmp(holder_klass, temp_itbl_klass);
1317     br(Assembler::NE, L_loop_search_resolved);
1318     mov(holder_offset, scan_temp);
1319     b(L_loop_search_resolved);
1320 
1321   // See if we already have a holder klass. If not, go and scan for it.
1322   bind(L_resolved_found);
1323   cbz(holder_offset, L_search_holder);
1324   mov(scan_temp, holder_offset);
1325 
1326   // Finally, scan_temp contains holder_klass vtable offset
1327   bind(L_holder_found);
1328   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1329   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1330     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1331   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1332 }
1333 
1334 // virtual method calling
1335 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1336                                            RegisterOrConstant vtable_index,
1337                                            Register method_result) {
1338   assert(vtableEntry::size() * wordSize == 8,
1339          "adjust the scaling in the code below");
1340   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1341 
1342   if (vtable_index.is_register()) {
1343     lea(method_result, Address(recv_klass,
1344                                vtable_index.as_register(),
1345                                Address::lsl(LogBytesPerWord)));
1346     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1347   } else {
1348     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1349     ldr(method_result,
1350         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1351   }
1352 }
1353 
1354 void MacroAssembler::check_klass_subtype(Register sub_klass,
1355                            Register super_klass,
1356                            Register temp_reg,
1357                            Label& L_success) {
1358   Label L_failure;
1359   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1360   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1361   bind(L_failure);
1362 }
1363 
1364 
1365 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1366                                                    Register super_klass,
1367                                                    Register temp_reg,
1368                                                    Label* L_success,
1369                                                    Label* L_failure,
1370                                                    Label* L_slow_path,
1371                                                    Register super_check_offset) {
1372   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1373   bool must_load_sco = ! super_check_offset->is_valid();
1374   if (must_load_sco) {
1375     assert(temp_reg != noreg, "supply either a temp or a register offset");
1376   }
1377 
1378   Label L_fallthrough;
1379   int label_nulls = 0;
1380   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1381   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1382   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1383   assert(label_nulls <= 1, "at most one null in the batch");
1384 
1385   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1386   Address super_check_offset_addr(super_klass, sco_offset);
1387 
1388   // Hacked jmp, which may only be used just before L_fallthrough.
1389 #define final_jmp(label)                                                \
1390   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1391   else                            b(label)                /*omit semi*/
1392 
1393   // If the pointers are equal, we are done (e.g., String[] elements).
1394   // This self-check enables sharing of secondary supertype arrays among
1395   // non-primary types such as array-of-interface.  Otherwise, each such
1396   // type would need its own customized SSA.
1397   // We move this check to the front of the fast path because many
1398   // type checks are in fact trivially successful in this manner,
1399   // so we get a nicely predicted branch right at the start of the check.
1400   cmp(sub_klass, super_klass);
1401   br(Assembler::EQ, *L_success);
1402 
1403   // Check the supertype display:
1404   if (must_load_sco) {
1405     ldrw(temp_reg, super_check_offset_addr);
1406     super_check_offset = temp_reg;
1407   }
1408 
1409   Address super_check_addr(sub_klass, super_check_offset);
1410   ldr(rscratch1, super_check_addr);
1411   cmp(super_klass, rscratch1); // load displayed supertype
1412   br(Assembler::EQ, *L_success);
1413 
1414   // This check has worked decisively for primary supers.
1415   // Secondary supers are sought in the super_cache ('super_cache_addr').
1416   // (Secondary supers are interfaces and very deeply nested subtypes.)
1417   // This works in the same check above because of a tricky aliasing
1418   // between the super_cache and the primary super display elements.
1419   // (The 'super_check_addr' can address either, as the case requires.)
1420   // Note that the cache is updated below if it does not help us find
1421   // what we need immediately.
1422   // So if it was a primary super, we can just fail immediately.
1423   // Otherwise, it's the slow path for us (no success at this point).
1424 
1425   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1426   if (L_failure == &L_fallthrough) {
1427     cbz(rscratch1, *L_slow_path);
1428   } else {
1429     cbnz(rscratch1, *L_failure);
1430     final_jmp(*L_slow_path);
1431   }
1432 
1433   bind(L_fallthrough);
1434 
1435 #undef final_jmp
1436 }
1437 
1438 // These two are taken from x86, but they look generally useful
1439 
1440 // scans count pointer sized words at [addr] for occurrence of value,
1441 // generic
1442 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1443                                 Register scratch) {
1444   Label Lloop, Lexit;
1445   cbz(count, Lexit);
1446   bind(Lloop);
1447   ldr(scratch, post(addr, wordSize));
1448   cmp(value, scratch);
1449   br(EQ, Lexit);
1450   sub(count, count, 1);
1451   cbnz(count, Lloop);
1452   bind(Lexit);
1453 }
1454 
1455 // scans count 4 byte words at [addr] for occurrence of value,
1456 // generic
1457 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1458                                 Register scratch) {
1459   Label Lloop, Lexit;
1460   cbz(count, Lexit);
1461   bind(Lloop);
1462   ldrw(scratch, post(addr, wordSize));
1463   cmpw(value, scratch);
1464   br(EQ, Lexit);
1465   sub(count, count, 1);
1466   cbnz(count, Lloop);
1467   bind(Lexit);
1468 }
1469 
1470 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1471                                                           Register super_klass,
1472                                                           Register temp_reg,
1473                                                           Register temp2_reg,
1474                                                           Label* L_success,
1475                                                           Label* L_failure,
1476                                                           bool set_cond_codes) {
1477   // NB! Callers may assume that, when temp2_reg is a valid register,
1478   // this code sets it to a nonzero value.
1479 
1480   assert_different_registers(sub_klass, super_klass, temp_reg);
1481   if (temp2_reg != noreg)
1482     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1483 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1484 
1485   Label L_fallthrough;
1486   int label_nulls = 0;
1487   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1488   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1489   assert(label_nulls <= 1, "at most one null in the batch");
1490 
1491   // a couple of useful fields in sub_klass:
1492   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1493   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1494   Address secondary_supers_addr(sub_klass, ss_offset);
1495   Address super_cache_addr(     sub_klass, sc_offset);
1496 
1497   BLOCK_COMMENT("check_klass_subtype_slow_path");
1498 
1499   // Do a linear scan of the secondary super-klass chain.
1500   // This code is rarely used, so simplicity is a virtue here.
1501   // The repne_scan instruction uses fixed registers, which we must spill.
1502   // Don't worry too much about pre-existing connections with the input regs.
1503 
1504   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1505   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1506 
1507   RegSet pushed_registers;
1508   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1509   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1510 
1511   if (super_klass != r0) {
1512     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1513   }
1514 
1515   push(pushed_registers, sp);
1516 
1517   // Get super_klass value into r0 (even if it was in r5 or r2).
1518   if (super_klass != r0) {
1519     mov(r0, super_klass);
1520   }
1521 
1522 #ifndef PRODUCT
1523   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1524 #endif //PRODUCT
1525 
1526   // We will consult the secondary-super array.
1527   ldr(r5, secondary_supers_addr);
1528   // Load the array length.
1529   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1530   // Skip to start of data.
1531   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1532 
1533   cmp(sp, zr); // Clear Z flag; SP is never zero
1534   // Scan R2 words at [R5] for an occurrence of R0.
1535   // Set NZ/Z based on last compare.
1536   repne_scan(r5, r0, r2, rscratch1);
1537 
1538   // Unspill the temp. registers:
1539   pop(pushed_registers, sp);
1540 
1541   br(Assembler::NE, *L_failure);
1542 
1543   // Success.  Cache the super we found and proceed in triumph.
1544 
1545   if (UseSecondarySupersCache) {
1546     str(super_klass, super_cache_addr);
1547   }
1548 
1549   if (L_success != &L_fallthrough) {
1550     b(*L_success);
1551   }
1552 
1553 #undef IS_A_TEMP
1554 
1555   bind(L_fallthrough);
1556 }
1557 
1558 // If Register r is invalid, remove a new register from
1559 // available_regs, and add new register to regs_to_push.
1560 Register MacroAssembler::allocate_if_noreg(Register r,
1561                                   RegSetIterator<Register> &available_regs,
1562                                   RegSet &regs_to_push) {
1563   if (!r->is_valid()) {
1564     r = *available_regs++;
1565     regs_to_push += r;
1566   }
1567   return r;
1568 }
1569 
1570 // check_klass_subtype_slow_path_table() looks for super_klass in the
1571 // hash table belonging to super_klass, branching to L_success or
1572 // L_failure as appropriate. This is essentially a shim which
1573 // allocates registers as necessary then calls
1574 // lookup_secondary_supers_table() to do the work. Any of the temp
1575 // regs may be noreg, in which case this logic will chooses some
1576 // registers push and pop them from the stack.
1577 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1578                                                          Register super_klass,
1579                                                          Register temp_reg,
1580                                                          Register temp2_reg,
1581                                                          Register temp3_reg,
1582                                                          Register result_reg,
1583                                                          FloatRegister vtemp,
1584                                                          Label* L_success,
1585                                                          Label* L_failure,
1586                                                          bool set_cond_codes) {
1587   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1588 
1589   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1590 
1591   Label L_fallthrough;
1592   int label_nulls = 0;
1593   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1594   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1595   assert(label_nulls <= 1, "at most one null in the batch");
1596 
1597   BLOCK_COMMENT("check_klass_subtype_slow_path");
1598 
1599   RegSetIterator<Register> available_regs
1600     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1601 
1602   RegSet pushed_regs;
1603 
1604   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1605   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1606   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1607   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1608 
1609   push(pushed_regs, sp);
1610 
1611   lookup_secondary_supers_table_var(sub_klass,
1612                                     super_klass,
1613                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1614                                     nullptr);
1615   cmp(result_reg, zr);
1616 
1617   // Unspill the temp. registers:
1618   pop(pushed_regs, sp);
1619 
1620   // NB! Callers may assume that, when set_cond_codes is true, this
1621   // code sets temp2_reg to a nonzero value.
1622   if (set_cond_codes) {
1623     mov(temp2_reg, 1);
1624   }
1625 
1626   br(Assembler::NE, *L_failure);
1627 
1628   if (L_success != &L_fallthrough) {
1629     b(*L_success);
1630   }
1631 
1632   bind(L_fallthrough);
1633 }
1634 
1635 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1636                                                    Register super_klass,
1637                                                    Register temp_reg,
1638                                                    Register temp2_reg,
1639                                                    Label* L_success,
1640                                                    Label* L_failure,
1641                                                    bool set_cond_codes) {
1642   if (UseSecondarySupersTable) {
1643     check_klass_subtype_slow_path_table
1644       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1645        /*vtemp*/fnoreg,
1646        L_success, L_failure, set_cond_codes);
1647   } else {
1648     check_klass_subtype_slow_path_linear
1649       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1650   }
1651 }
1652 
1653 
1654 // Ensure that the inline code and the stub are using the same registers.
1655 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1656 do {                                                               \
1657   assert(r_super_klass  == r0                                   && \
1658          r_array_base   == r1                                   && \
1659          r_array_length == r2                                   && \
1660          (r_array_index == r3        || r_array_index == noreg) && \
1661          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1662          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1663          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1664 } while(0)
1665 
1666 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1667                                                          Register r_super_klass,
1668                                                          Register temp1,
1669                                                          Register temp2,
1670                                                          Register temp3,
1671                                                          FloatRegister vtemp,
1672                                                          Register result,
1673                                                          u1 super_klass_slot,
1674                                                          bool stub_is_near) {
1675   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1676 
1677   Label L_fallthrough;
1678 
1679   BLOCK_COMMENT("lookup_secondary_supers_table {");
1680 
1681   const Register
1682     r_array_base   = temp1, // r1
1683     r_array_length = temp2, // r2
1684     r_array_index  = temp3, // r3
1685     r_bitmap       = rscratch2;
1686 
1687   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1688 
1689   u1 bit = super_klass_slot;
1690 
1691   // Make sure that result is nonzero if the TBZ below misses.
1692   mov(result, 1);
1693 
1694   // We're going to need the bitmap in a vector reg and in a core reg,
1695   // so load both now.
1696   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1697   if (bit != 0) {
1698     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1699   }
1700   // First check the bitmap to see if super_klass might be present. If
1701   // the bit is zero, we are certain that super_klass is not one of
1702   // the secondary supers.
1703   tbz(r_bitmap, bit, L_fallthrough);
1704 
1705   // Get the first array index that can contain super_klass into r_array_index.
1706   if (bit != 0) {
1707     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1708     cnt(vtemp, T8B, vtemp);
1709     addv(vtemp, T8B, vtemp);
1710     fmovd(r_array_index, vtemp);
1711   } else {
1712     mov(r_array_index, (u1)1);
1713   }
1714   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1715 
1716   // We will consult the secondary-super array.
1717   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1718 
1719   // The value i in r_array_index is >= 1, so even though r_array_base
1720   // points to the length, we don't need to adjust it to point to the
1721   // data.
1722   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1723   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1724 
1725   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1726   eor(result, result, r_super_klass);
1727   cbz(result, L_fallthrough); // Found a match
1728 
1729   // Is there another entry to check? Consult the bitmap.
1730   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1731 
1732   // Linear probe.
1733   if (bit != 0) {
1734     ror(r_bitmap, r_bitmap, bit);
1735   }
1736 
1737   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1738   // The next slot to be inspected, by the stub we're about to call,
1739   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1740   // have been checked.
1741   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1742   if (stub_is_near) {
1743     bl(stub);
1744   } else {
1745     address call = trampoline_call(stub);
1746     if (call == nullptr) {
1747       return false; // trampoline allocation failed
1748     }
1749   }
1750 
1751   BLOCK_COMMENT("} lookup_secondary_supers_table");
1752 
1753   bind(L_fallthrough);
1754 
1755   if (VerifySecondarySupers) {
1756     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1757                                   temp1, temp2, result);      // r1, r2, r5
1758   }
1759   return true;
1760 }
1761 
1762 // At runtime, return 0 in result if r_super_klass is a superclass of
1763 // r_sub_klass, otherwise return nonzero. Use this version of
1764 // lookup_secondary_supers_table() if you don't know ahead of time
1765 // which superclass will be searched for. Used by interpreter and
1766 // runtime stubs. It is larger and has somewhat greater latency than
1767 // the version above, which takes a constant super_klass_slot.
1768 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1769                                                        Register r_super_klass,
1770                                                        Register temp1,
1771                                                        Register temp2,
1772                                                        Register temp3,
1773                                                        FloatRegister vtemp,
1774                                                        Register result,
1775                                                        Label *L_success) {
1776   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1777 
1778   Label L_fallthrough;
1779 
1780   BLOCK_COMMENT("lookup_secondary_supers_table {");
1781 
1782   const Register
1783     r_array_index = temp3,
1784     slot          = rscratch1,
1785     r_bitmap      = rscratch2;
1786 
1787   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1788 
1789   // Make sure that result is nonzero if the test below misses.
1790   mov(result, 1);
1791 
1792   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1793 
1794   // First check the bitmap to see if super_klass might be present. If
1795   // the bit is zero, we are certain that super_klass is not one of
1796   // the secondary supers.
1797 
1798   // This next instruction is equivalent to:
1799   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1800   // sub(temp2, tmp_reg, slot);
1801   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1802   lslv(temp2, r_bitmap, temp2);
1803   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1804 
1805   bool must_save_v0 = (vtemp == fnoreg);
1806   if (must_save_v0) {
1807     // temp1 and result are free, so use them to preserve vtemp
1808     vtemp = v0;
1809     mov(temp1,  vtemp, D, 0);
1810     mov(result, vtemp, D, 1);
1811   }
1812 
1813   // Get the first array index that can contain super_klass into r_array_index.
1814   mov(vtemp, D, 0, temp2);
1815   cnt(vtemp, T8B, vtemp);
1816   addv(vtemp, T8B, vtemp);
1817   mov(r_array_index, vtemp, D, 0);
1818 
1819   if (must_save_v0) {
1820     mov(vtemp, D, 0, temp1 );
1821     mov(vtemp, D, 1, result);
1822   }
1823 
1824   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1825 
1826   const Register
1827     r_array_base   = temp1,
1828     r_array_length = temp2;
1829 
1830   // The value i in r_array_index is >= 1, so even though r_array_base
1831   // points to the length, we don't need to adjust it to point to the
1832   // data.
1833   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1834   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1835 
1836   // We will consult the secondary-super array.
1837   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1838 
1839   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1840   eor(result, result, r_super_klass);
1841   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1842 
1843   // Is there another entry to check? Consult the bitmap.
1844   rorv(r_bitmap, r_bitmap, slot);
1845   // rol(r_bitmap, r_bitmap, 1);
1846   tbz(r_bitmap, 1, L_fallthrough);
1847 
1848   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1849   // The next slot to be inspected, by the logic we're about to call,
1850   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1851   // have been checked.
1852   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1853                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1854 
1855   BLOCK_COMMENT("} lookup_secondary_supers_table");
1856 
1857   bind(L_fallthrough);
1858 
1859   if (VerifySecondarySupers) {
1860     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1861                                   temp1, temp2, result);      // r1, r2, r5
1862   }
1863 
1864   if (L_success) {
1865     cbz(result, *L_success);
1866   }
1867 }
1868 
1869 // Called by code generated by check_klass_subtype_slow_path
1870 // above. This is called when there is a collision in the hashed
1871 // lookup in the secondary supers array.
1872 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1873                                                              Register r_array_base,
1874                                                              Register r_array_index,
1875                                                              Register r_bitmap,
1876                                                              Register temp1,
1877                                                              Register result,
1878                                                              bool is_stub) {
1879   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1880 
1881   const Register
1882     r_array_length = temp1,
1883     r_sub_klass    = noreg; // unused
1884 
1885   if (is_stub) {
1886     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1887   }
1888 
1889   Label L_fallthrough, L_huge;
1890 
1891   // Load the array length.
1892   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1893   // And adjust the array base to point to the data.
1894   // NB! Effectively increments current slot index by 1.
1895   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1896   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1897 
1898   // The bitmap is full to bursting.
1899   // Implicit invariant: BITMAP_FULL implies (length > 0)
1900   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1901   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1902   br(GT, L_huge);
1903 
1904   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1905   // current slot (at secondary_supers[r_array_index]) has not yet
1906   // been inspected, and r_array_index may be out of bounds if we
1907   // wrapped around the end of the array.
1908 
1909   { // This is conventional linear probing, but instead of terminating
1910     // when a null entry is found in the table, we maintain a bitmap
1911     // in which a 0 indicates missing entries.
1912     // As long as the bitmap is not completely full,
1913     // array_length == popcount(bitmap). The array_length check above
1914     // guarantees there are 0s in the bitmap, so the loop eventually
1915     // terminates.
1916     Label L_loop;
1917     bind(L_loop);
1918 
1919     // Check for wraparound.
1920     cmp(r_array_index, r_array_length);
1921     csel(r_array_index, zr, r_array_index, GE);
1922 
1923     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1924     eor(result, rscratch1, r_super_klass);
1925     cbz(result, L_fallthrough);
1926 
1927     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1928 
1929     ror(r_bitmap, r_bitmap, 1);
1930     add(r_array_index, r_array_index, 1);
1931     b(L_loop);
1932   }
1933 
1934   { // Degenerate case: more than 64 secondary supers.
1935     // FIXME: We could do something smarter here, maybe a vectorized
1936     // comparison or a binary search, but is that worth any added
1937     // complexity?
1938     bind(L_huge);
1939     cmp(sp, zr); // Clear Z flag; SP is never zero
1940     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1941     cset(result, NE); // result == 0 iff we got a match.
1942   }
1943 
1944   bind(L_fallthrough);
1945 }
1946 
1947 // Make sure that the hashed lookup and a linear scan agree.
1948 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1949                                                    Register r_super_klass,
1950                                                    Register temp1,
1951                                                    Register temp2,
1952                                                    Register result) {
1953   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1954 
1955   const Register
1956     r_array_base   = temp1,
1957     r_array_length = temp2;
1958 
1959   BLOCK_COMMENT("verify_secondary_supers_table {");
1960 
1961   // We will consult the secondary-super array.
1962   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1963 
1964   // Load the array length.
1965   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1966   // And adjust the array base to point to the data.
1967   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1968 
1969   cmp(sp, zr); // Clear Z flag; SP is never zero
1970   // Scan R2 words at [R5] for an occurrence of R0.
1971   // Set NZ/Z based on last compare.
1972   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1973   // rscratch1 == 0 iff we got a match.
1974   cset(rscratch1, NE);
1975 
1976   Label passed;
1977   cmp(result, zr);
1978   cset(result, NE); // normalize result to 0/1 for comparison
1979 
1980   cmp(rscratch1, result);
1981   br(EQ, passed);
1982   {
1983     mov(r0, r_super_klass);         // r0 <- r0
1984     mov(r1, r_sub_klass);           // r1 <- r4
1985     mov(r2, /*expected*/rscratch1); // r2 <- r8
1986     mov(r3, result);                // r3 <- r5
1987     mov(r4, (address)("mismatch")); // r4 <- const
1988     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
1989     should_not_reach_here();
1990   }
1991   bind(passed);
1992 
1993   BLOCK_COMMENT("} verify_secondary_supers_table");
1994 }
1995 
1996 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
1997   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
1998   assert_different_registers(klass, rthread, scratch);
1999 
2000   Label L_fallthrough, L_tmp;
2001   if (L_fast_path == nullptr) {
2002     L_fast_path = &L_fallthrough;
2003   } else if (L_slow_path == nullptr) {
2004     L_slow_path = &L_fallthrough;
2005   }
2006   // Fast path check: class is fully initialized
2007   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2008   ldarb(scratch, scratch);
2009   cmp(scratch, InstanceKlass::fully_initialized);
2010   br(Assembler::EQ, *L_fast_path);
2011 
2012   // Fast path check: current thread is initializer thread
2013   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2014   cmp(rthread, scratch);
2015 
2016   if (L_slow_path == &L_fallthrough) {
2017     br(Assembler::EQ, *L_fast_path);
2018     bind(*L_slow_path);
2019   } else if (L_fast_path == &L_fallthrough) {
2020     br(Assembler::NE, *L_slow_path);
2021     bind(*L_fast_path);
2022   } else {
2023     Unimplemented();
2024   }
2025 }
2026 
2027 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2028   if (!VerifyOops || VerifyAdapterSharing) {
2029     // Below address of the code string confuses VerifyAdapterSharing
2030     // because it may differ between otherwise equivalent adapters.
2031     return;
2032   }
2033 
2034   // Pass register number to verify_oop_subroutine
2035   const char* b = nullptr;
2036   {
2037     ResourceMark rm;
2038     stringStream ss;
2039     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2040     b = code_string(ss.as_string());
2041   }
2042   BLOCK_COMMENT("verify_oop {");
2043 
2044   strip_return_address(); // This might happen within a stack frame.
2045   protect_return_address();
2046   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2047   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2048 
2049   mov(r0, reg);
2050   movptr(rscratch1, (uintptr_t)(address)b);
2051 
2052   // call indirectly to solve generation ordering problem
2053   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2054   ldr(rscratch2, Address(rscratch2));
2055   blr(rscratch2);
2056 
2057   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2058   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2059   authenticate_return_address();
2060 
2061   BLOCK_COMMENT("} verify_oop");
2062 }
2063 
2064 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2065   if (!VerifyOops || VerifyAdapterSharing) {
2066     // Below address of the code string confuses VerifyAdapterSharing
2067     // because it may differ between otherwise equivalent adapters.
2068     return;
2069   }
2070 
2071   const char* b = nullptr;
2072   {
2073     ResourceMark rm;
2074     stringStream ss;
2075     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2076     b = code_string(ss.as_string());
2077   }
2078   BLOCK_COMMENT("verify_oop_addr {");
2079 
2080   strip_return_address(); // This might happen within a stack frame.
2081   protect_return_address();
2082   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2083   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2084 
2085   // addr may contain sp so we will have to adjust it based on the
2086   // pushes that we just did.
2087   if (addr.uses(sp)) {
2088     lea(r0, addr);
2089     ldr(r0, Address(r0, 4 * wordSize));
2090   } else {
2091     ldr(r0, addr);
2092   }
2093   movptr(rscratch1, (uintptr_t)(address)b);
2094 
2095   // call indirectly to solve generation ordering problem
2096   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2097   ldr(rscratch2, Address(rscratch2));
2098   blr(rscratch2);
2099 
2100   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2101   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2102   authenticate_return_address();
2103 
2104   BLOCK_COMMENT("} verify_oop_addr");
2105 }
2106 
2107 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2108                                          int extra_slot_offset) {
2109   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2110   int stackElementSize = Interpreter::stackElementSize;
2111   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2112 #ifdef ASSERT
2113   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2114   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2115 #endif
2116   if (arg_slot.is_constant()) {
2117     return Address(esp, arg_slot.as_constant() * stackElementSize
2118                    + offset);
2119   } else {
2120     add(rscratch1, esp, arg_slot.as_register(),
2121         ext::uxtx, exact_log2(stackElementSize));
2122     return Address(rscratch1, offset);
2123   }
2124 }
2125 
2126 // Handle the receiver type profile update given the "recv" klass.
2127 //
2128 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2129 // If there are no matching or claimable receiver entries in RD, updates
2130 // the polymorphic counter.
2131 //
2132 // This code expected to run by either the interpreter or JIT-ed code, without
2133 // extra synchronization. For safety, receiver cells are claimed atomically, which
2134 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2135 // counter updates are not atomic.
2136 //
2137 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2138   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2139 
2140   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2141   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2142   int poly_count_offset      = in_bytes(CounterData::count_offset());
2143   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2144   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2145 
2146   // Adjust for MDP offsets.
2147   base_receiver_offset += mdp_offset;
2148   end_receiver_offset  += mdp_offset;
2149   poly_count_offset    += mdp_offset;
2150 
2151 #ifdef ASSERT
2152   // We are about to walk the MDO slots without asking for offsets.
2153   // Check that our math hits all the right spots.
2154   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2155     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2156     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2157     int offset = base_receiver_offset + receiver_step*c;
2158     int count_offset = offset + receiver_to_count_step;
2159     assert(offset == real_recv_offset, "receiver slot math");
2160     assert(count_offset == real_count_offset, "receiver count math");
2161   }
2162   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2163   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2164 #endif
2165 
2166   // Corner case: no profile table. Increment poly counter and exit.
2167   if (ReceiverTypeData::row_limit() == 0) {
2168     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2169     return;
2170   }
2171 
2172   Register offset = rscratch2;
2173 
2174   Label L_loop_search_receiver, L_loop_search_empty;
2175   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
2176 
2177   // The code here recognizes three major cases:
2178   //   A. Fastest: receiver found in the table
2179   //   B. Fast: no receiver in the table, and the table is full
2180   //   C. Slow: no receiver in the table, free slots in the table
2181   //
2182   // The case A performance is most important, as perfectly-behaved code would end up
2183   // there, especially with larger TypeProfileWidth. The case B performance is
2184   // important as well, this is where bulk of code would land for normally megamorphic
2185   // cases. The case C performance is not essential, its job is to deal with installation
2186   // races, we optimize for code density instead. Case C needs to make sure that receiver
2187   // rows are only claimed once. This makes sure we never overwrite a row for another
2188   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2189   //
2190   // It is very tempting to handle these cases in a single loop, and claim the first slot
2191   // without checking the rest of the table. But, profiling code should tolerate free slots
2192   // in the table, as class unloading can clear them. After such cleanup, the receiver
2193   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2194   // to complete, before trying to install new slots. Splitting the code in several tight
2195   // loops also helpfully optimizes for cases A and B.
2196   //
2197   // This code is effectively:
2198   //
2199   // restart:
2200   //   // Fastest: receiver is already installed
2201   //   for (i = 0; i < receiver_count(); i++) {
2202   //     if (receiver(i) == recv) goto found_recv(i);
2203   //   }
2204   //
2205   //   // Fast: no receiver, but profile is full
2206   //   for (i = 0; i < receiver_count(); i++) {
2207   //     if (receiver(i) == null) goto found_null(i);
2208   //   }
2209   //   goto polymorphic
2210   //
2211   //   // Slow: try to install receiver
2212   // found_null(i):
2213   //   CAS(&receiver(i), null, recv);
2214   //   goto restart
2215   //
2216   // polymorphic:
2217   //   count++;
2218   //   return
2219   //
2220   // found_recv(i):
2221   //   *receiver_count(i)++
2222   //
2223 
2224   bind(L_restart);
2225 
2226   // Fastest: receiver is already installed
2227   mov(offset, base_receiver_offset);
2228   bind(L_loop_search_receiver);
2229     ldr(rscratch1, Address(mdp, offset));
2230     cmp(rscratch1, recv);
2231     br(Assembler::EQ, L_found_recv);
2232   add(offset, offset, receiver_step);
2233   sub(rscratch1, offset, end_receiver_offset);
2234   cbnz(rscratch1, L_loop_search_receiver);
2235 
2236   // Fast: no receiver, but profile is full
2237   mov(offset, base_receiver_offset);
2238   bind(L_loop_search_empty);
2239     ldr(rscratch1, Address(mdp, offset));
2240     cbz(rscratch1, L_found_empty);
2241   add(offset, offset, receiver_step);
2242   sub(rscratch1, offset, end_receiver_offset);
2243   cbnz(rscratch1, L_loop_search_empty);
2244   b(L_polymorphic);
2245 
2246   // Slow: try to install receiver
2247   bind(L_found_empty);
2248 
2249   // Atomically swing receiver slot: null -> recv.
2250   //
2251   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2252   // is used to hold the destination address. This is safe because the
2253   // offset is no longer needed after the address is computed.
2254 
2255   lea(rscratch2, Address(mdp, offset));
2256   cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
2257           /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
2258 
2259   // CAS success means the slot now has the receiver we want. CAS failure means
2260   // something had claimed the slot concurrently: it can be the same receiver we want,
2261   // or something else. Since this is a slow path, we can optimize for code density,
2262   // and just restart the search from the beginning.
2263   b(L_restart);
2264 
2265   // Counter updates:
2266 
2267   // Increment polymorphic counter instead of receiver slot.
2268   bind(L_polymorphic);
2269   mov(offset, poly_count_offset);
2270   b(L_count_update);
2271 
2272   // Found a receiver, convert its slot offset to corresponding count offset.
2273   bind(L_found_recv);
2274   add(offset, offset, receiver_to_count_step);
2275 
2276   bind(L_count_update);
2277   increment(Address(mdp, offset), DataLayout::counter_increment);
2278 }
2279 
2280 
2281 void MacroAssembler::call_VM_leaf_base(address entry_point,
2282                                        int number_of_arguments,
2283                                        Label *retaddr) {
2284   Label E, L;
2285 
2286   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2287 
2288   mov(rscratch1, RuntimeAddress(entry_point));
2289   blr(rscratch1);
2290   if (retaddr)
2291     bind(*retaddr);
2292 
2293   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2294 }
2295 
2296 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2297   call_VM_leaf_base(entry_point, number_of_arguments);
2298 }
2299 
2300 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2301   pass_arg0(this, arg_0);
2302   call_VM_leaf_base(entry_point, 1);
2303 }
2304 
2305 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2306   assert_different_registers(arg_1, c_rarg0);
2307   pass_arg0(this, arg_0);
2308   pass_arg1(this, arg_1);
2309   call_VM_leaf_base(entry_point, 2);
2310 }
2311 
2312 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2313                                   Register arg_1, Register arg_2) {
2314   assert_different_registers(arg_1, c_rarg0);
2315   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2316   pass_arg0(this, arg_0);
2317   pass_arg1(this, arg_1);
2318   pass_arg2(this, arg_2);
2319   call_VM_leaf_base(entry_point, 3);
2320 }
2321 
2322 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2323   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2324 }
2325 
2326 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2327   pass_arg0(this, arg_0);
2328   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2329 }
2330 
2331 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2332 
2333   assert_different_registers(arg_0, c_rarg1);
2334   pass_arg1(this, arg_1);
2335   pass_arg0(this, arg_0);
2336   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2337 }
2338 
2339 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2340   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2341   assert_different_registers(arg_1, c_rarg2);
2342   pass_arg2(this, arg_2);
2343   pass_arg1(this, arg_1);
2344   pass_arg0(this, arg_0);
2345   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2346 }
2347 
2348 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2349   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2350   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2351   assert_different_registers(arg_2, c_rarg3);
2352   pass_arg3(this, arg_3);
2353   pass_arg2(this, arg_2);
2354   pass_arg1(this, arg_1);
2355   pass_arg0(this, arg_0);
2356   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2357 }
2358 
2359 void MacroAssembler::null_check(Register reg, int offset) {
2360   if (needs_explicit_null_check(offset)) {
2361     // provoke OS null exception if reg is null by
2362     // accessing M[reg] w/o changing any registers
2363     // NOTE: this is plenty to provoke a segv
2364     ldr(zr, Address(reg));
2365   } else {
2366     // nothing to do, (later) access of M[reg + offset]
2367     // will provoke OS null exception if reg is null
2368   }
2369 }
2370 
2371 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2372   assert_different_registers(markword, rscratch2);
2373   mov(rscratch2, markWord::inline_type_pattern_mask);
2374   andr(markword, markword, rscratch2);
2375   mov(rscratch2, markWord::inline_type_pattern);
2376   cmp(markword, rscratch2);
2377   br(Assembler::EQ, is_inline_type);
2378 }
2379 
2380 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2381   assert_different_registers(tmp, rscratch1);
2382   if (can_be_null) {
2383     cbz(object, not_inline_type);
2384   }
2385   const int is_inline_type_mask = markWord::inline_type_pattern;
2386   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2387   mov(rscratch1, is_inline_type_mask);
2388   andr(tmp, tmp, rscratch1);
2389   cmp(tmp, rscratch1);
2390   br(Assembler::NE, not_inline_type);
2391 }
2392 
2393 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2394   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2395   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2396 }
2397 
2398 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2399   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2400   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2401 }
2402 
2403 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2404   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2405   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2406 }
2407 
2408 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
2409   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2410   tbnz(flags, ResolvedFieldEntry::has_null_marker_shift, has_null_marker);
2411 }
2412 
2413 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2414   Label test_mark_word;
2415   // load mark word
2416   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2417   // check displaced
2418   tst(temp_reg, markWord::unlocked_value);
2419   br(Assembler::NE, test_mark_word);
2420   // slow path use klass prototype
2421   load_prototype_header(temp_reg, oop);
2422 
2423   bind(test_mark_word);
2424   andr(temp_reg, temp_reg, test_bit);
2425   if (jmp_set) {
2426     cbnz(temp_reg, jmp_label);
2427   } else {
2428     cbz(temp_reg, jmp_label);
2429   }
2430 }
2431 
2432 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2433   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2434 }
2435 
2436 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2437                                                   Label&is_non_flat_array) {
2438   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2439 }
2440 
2441 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2442   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2443 }
2444 
2445 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2446   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2447 }
2448 
2449 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2450   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2451   br(Assembler::NE, is_flat_array);
2452 }
2453 
2454 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
2455   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2456   br(Assembler::EQ, is_non_flat_array);
2457 }
2458 
2459 // MacroAssembler protected routines needed to implement
2460 // public methods
2461 
2462 void MacroAssembler::mov(Register r, Address dest) {
2463   code_section()->relocate(pc(), dest.rspec());
2464   uint64_t imm64 = (uint64_t)dest.target();
2465   movptr(r, imm64);
2466 }
2467 
2468 // Move a constant pointer into r.  In AArch64 mode the virtual
2469 // address space is 48 bits in size, so we only need three
2470 // instructions to create a patchable instruction sequence that can
2471 // reach anywhere.
2472 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2473 #ifndef PRODUCT
2474   {
2475     char buffer[64];
2476     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2477     block_comment(buffer);
2478   }
2479 #endif
2480   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2481   movz(r, imm64 & 0xffff);
2482   imm64 >>= 16;
2483   movk(r, imm64 & 0xffff, 16);
2484   imm64 >>= 16;
2485   movk(r, imm64 & 0xffff, 32);
2486 }
2487 
2488 // Macro to mov replicated immediate to vector register.
2489 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2490 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2491 // Vd will get the following values for different arrangements in T
2492 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2493 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2494 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2495 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2496 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2497 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2498 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2499 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2500 // Clobbers rscratch1
2501 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2502   assert(T != T1Q, "unsupported");
2503   if (T == T1D || T == T2D) {
2504     int imm = operand_valid_for_movi_immediate(imm64, T);
2505     if (-1 != imm) {
2506       movi(Vd, T, imm);
2507     } else {
2508       mov(rscratch1, imm64);
2509       dup(Vd, T, rscratch1);
2510     }
2511     return;
2512   }
2513 
2514 #ifdef ASSERT
2515   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2516   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2517   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2518 #endif
2519   int shift = operand_valid_for_movi_immediate(imm64, T);
2520   uint32_t imm32 = imm64 & 0xffffffffULL;
2521   if (shift >= 0) {
2522     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2523   } else {
2524     movw(rscratch1, imm32);
2525     dup(Vd, T, rscratch1);
2526   }
2527 }
2528 
2529 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2530 {
2531 #ifndef PRODUCT
2532   {
2533     char buffer[64];
2534     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2535     block_comment(buffer);
2536   }
2537 #endif
2538   if (operand_valid_for_logical_immediate(false, imm64)) {
2539     orr(dst, zr, imm64);
2540   } else {
2541     // we can use a combination of MOVZ or MOVN with
2542     // MOVK to build up the constant
2543     uint64_t imm_h[4];
2544     int zero_count = 0;
2545     int neg_count = 0;
2546     int i;
2547     for (i = 0; i < 4; i++) {
2548       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2549       if (imm_h[i] == 0) {
2550         zero_count++;
2551       } else if (imm_h[i] == 0xffffL) {
2552         neg_count++;
2553       }
2554     }
2555     if (zero_count == 4) {
2556       // one MOVZ will do
2557       movz(dst, 0);
2558     } else if (neg_count == 4) {
2559       // one MOVN will do
2560       movn(dst, 0);
2561     } else if (zero_count == 3) {
2562       for (i = 0; i < 4; i++) {
2563         if (imm_h[i] != 0L) {
2564           movz(dst, (uint32_t)imm_h[i], (i << 4));
2565           break;
2566         }
2567       }
2568     } else if (neg_count == 3) {
2569       // one MOVN will do
2570       for (int i = 0; i < 4; i++) {
2571         if (imm_h[i] != 0xffffL) {
2572           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2573           break;
2574         }
2575       }
2576     } else if (zero_count == 2) {
2577       // one MOVZ and one MOVK will do
2578       for (i = 0; i < 3; i++) {
2579         if (imm_h[i] != 0L) {
2580           movz(dst, (uint32_t)imm_h[i], (i << 4));
2581           i++;
2582           break;
2583         }
2584       }
2585       for (;i < 4; i++) {
2586         if (imm_h[i] != 0L) {
2587           movk(dst, (uint32_t)imm_h[i], (i << 4));
2588         }
2589       }
2590     } else if (neg_count == 2) {
2591       // one MOVN and one MOVK will do
2592       for (i = 0; i < 4; i++) {
2593         if (imm_h[i] != 0xffffL) {
2594           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2595           i++;
2596           break;
2597         }
2598       }
2599       for (;i < 4; i++) {
2600         if (imm_h[i] != 0xffffL) {
2601           movk(dst, (uint32_t)imm_h[i], (i << 4));
2602         }
2603       }
2604     } else if (zero_count == 1) {
2605       // one MOVZ and two MOVKs will do
2606       for (i = 0; i < 4; i++) {
2607         if (imm_h[i] != 0L) {
2608           movz(dst, (uint32_t)imm_h[i], (i << 4));
2609           i++;
2610           break;
2611         }
2612       }
2613       for (;i < 4; i++) {
2614         if (imm_h[i] != 0x0L) {
2615           movk(dst, (uint32_t)imm_h[i], (i << 4));
2616         }
2617       }
2618     } else if (neg_count == 1) {
2619       // one MOVN and two MOVKs will do
2620       for (i = 0; i < 4; i++) {
2621         if (imm_h[i] != 0xffffL) {
2622           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2623           i++;
2624           break;
2625         }
2626       }
2627       for (;i < 4; i++) {
2628         if (imm_h[i] != 0xffffL) {
2629           movk(dst, (uint32_t)imm_h[i], (i << 4));
2630         }
2631       }
2632     } else {
2633       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2634       movz(dst, (uint32_t)imm_h[0], 0);
2635       for (i = 1; i < 4; i++) {
2636         movk(dst, (uint32_t)imm_h[i], (i << 4));
2637       }
2638     }
2639   }
2640 }
2641 
2642 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2643 {
2644 #ifndef PRODUCT
2645     {
2646       char buffer[64];
2647       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2648       block_comment(buffer);
2649     }
2650 #endif
2651   if (operand_valid_for_logical_immediate(true, imm32)) {
2652     orrw(dst, zr, imm32);
2653   } else {
2654     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2655     // constant
2656     uint32_t imm_h[2];
2657     imm_h[0] = imm32 & 0xffff;
2658     imm_h[1] = ((imm32 >> 16) & 0xffff);
2659     if (imm_h[0] == 0) {
2660       movzw(dst, imm_h[1], 16);
2661     } else if (imm_h[0] == 0xffff) {
2662       movnw(dst, imm_h[1] ^ 0xffff, 16);
2663     } else if (imm_h[1] == 0) {
2664       movzw(dst, imm_h[0], 0);
2665     } else if (imm_h[1] == 0xffff) {
2666       movnw(dst, imm_h[0] ^ 0xffff, 0);
2667     } else {
2668       // use a MOVZ and MOVK (makes it easier to debug)
2669       movzw(dst, imm_h[0], 0);
2670       movkw(dst, imm_h[1], 16);
2671     }
2672   }
2673 }
2674 
2675 // Form an address from base + offset in Rd.  Rd may or may
2676 // not actually be used: you must use the Address that is returned.
2677 // It is up to you to ensure that the shift provided matches the size
2678 // of your data.
2679 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2680   if (Address::offset_ok_for_immed(byte_offset, shift))
2681     // It fits; no need for any heroics
2682     return Address(base, byte_offset);
2683 
2684   // Don't do anything clever with negative or misaligned offsets
2685   unsigned mask = (1 << shift) - 1;
2686   if (byte_offset < 0 || byte_offset & mask) {
2687     mov(Rd, byte_offset);
2688     add(Rd, base, Rd);
2689     return Address(Rd);
2690   }
2691 
2692   // See if we can do this with two 12-bit offsets
2693   {
2694     uint64_t word_offset = byte_offset >> shift;
2695     uint64_t masked_offset = word_offset & 0xfff000;
2696     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2697         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2698       add(Rd, base, masked_offset << shift);
2699       word_offset -= masked_offset;
2700       return Address(Rd, word_offset << shift);
2701     }
2702   }
2703 
2704   // Do it the hard way
2705   mov(Rd, byte_offset);
2706   add(Rd, base, Rd);
2707   return Address(Rd);
2708 }
2709 
2710 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2711                                     bool want_remainder, Register scratch)
2712 {
2713   // Full implementation of Java idiv and irem.  The function
2714   // returns the (pc) offset of the div instruction - may be needed
2715   // for implicit exceptions.
2716   //
2717   // constraint : ra/rb =/= scratch
2718   //         normal case
2719   //
2720   // input : ra: dividend
2721   //         rb: divisor
2722   //
2723   // result: either
2724   //         quotient  (= ra idiv rb)
2725   //         remainder (= ra irem rb)
2726 
2727   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2728 
2729   int idivl_offset = offset();
2730   if (! want_remainder) {
2731     sdivw(result, ra, rb);
2732   } else {
2733     sdivw(scratch, ra, rb);
2734     Assembler::msubw(result, scratch, rb, ra);
2735   }
2736 
2737   return idivl_offset;
2738 }
2739 
2740 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2741                                     bool want_remainder, Register scratch)
2742 {
2743   // Full implementation of Java ldiv and lrem.  The function
2744   // returns the (pc) offset of the div instruction - may be needed
2745   // for implicit exceptions.
2746   //
2747   // constraint : ra/rb =/= scratch
2748   //         normal case
2749   //
2750   // input : ra: dividend
2751   //         rb: divisor
2752   //
2753   // result: either
2754   //         quotient  (= ra idiv rb)
2755   //         remainder (= ra irem rb)
2756 
2757   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2758 
2759   int idivq_offset = offset();
2760   if (! want_remainder) {
2761     sdiv(result, ra, rb);
2762   } else {
2763     sdiv(scratch, ra, rb);
2764     Assembler::msub(result, scratch, rb, ra);
2765   }
2766 
2767   return idivq_offset;
2768 }
2769 
2770 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2771   address prev = pc() - NativeMembar::instruction_size;
2772   address last = code()->last_insn();
2773   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2774     NativeMembar *bar = NativeMembar_at(prev);
2775     if (AlwaysMergeDMB) {
2776       bar->set_kind(bar->get_kind() | order_constraint);
2777       BLOCK_COMMENT("merged membar(always)");
2778       return;
2779     }
2780     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2781     // doing so would introduce a StoreLoad which the caller did not
2782     // intend
2783     if (bar->get_kind() == order_constraint
2784         || bar->get_kind() == AnyAny
2785         || order_constraint == AnyAny) {
2786       // We are merging two memory barrier instructions.  On AArch64 we
2787       // can do this simply by ORing them together.
2788       bar->set_kind(bar->get_kind() | order_constraint);
2789       BLOCK_COMMENT("merged membar");
2790       return;
2791     } else {
2792       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2793       // We need check the last 2 instructions
2794       address prev2 = prev - NativeMembar::instruction_size;
2795       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2796         NativeMembar *bar2 = NativeMembar_at(prev2);
2797         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2798         BLOCK_COMMENT("merged membar(elided)");
2799         return;
2800       }
2801     }
2802   }
2803   code()->set_last_insn(pc());
2804   dmb(Assembler::barrier(order_constraint));
2805 }
2806 
2807 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2808   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2809     merge_ldst(rt, adr, size_in_bytes, is_store);
2810     code()->clear_last_insn();
2811     return true;
2812   } else {
2813     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2814     const uint64_t mask = size_in_bytes - 1;
2815     if (adr.getMode() == Address::base_plus_offset &&
2816         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2817       code()->set_last_insn(pc());
2818     }
2819     return false;
2820   }
2821 }
2822 
2823 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2824   // We always try to merge two adjacent loads into one ldp.
2825   if (!try_merge_ldst(Rx, adr, 8, false)) {
2826     Assembler::ldr(Rx, adr);
2827   }
2828 }
2829 
2830 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2831   // We always try to merge two adjacent loads into one ldp.
2832   if (!try_merge_ldst(Rw, adr, 4, false)) {
2833     Assembler::ldrw(Rw, adr);
2834   }
2835 }
2836 
2837 void MacroAssembler::str(Register Rx, const Address &adr) {
2838   // We always try to merge two adjacent stores into one stp.
2839   if (!try_merge_ldst(Rx, adr, 8, true)) {
2840     Assembler::str(Rx, adr);
2841   }
2842 }
2843 
2844 void MacroAssembler::strw(Register Rw, const Address &adr) {
2845   // We always try to merge two adjacent stores into one stp.
2846   if (!try_merge_ldst(Rw, adr, 4, true)) {
2847     Assembler::strw(Rw, adr);
2848   }
2849 }
2850 
2851 // MacroAssembler routines found actually to be needed
2852 
2853 void MacroAssembler::push(Register src)
2854 {
2855   str(src, Address(pre(esp, -1 * wordSize)));
2856 }
2857 
2858 void MacroAssembler::pop(Register dst)
2859 {
2860   ldr(dst, Address(post(esp, 1 * wordSize)));
2861 }
2862 
2863 // Note: load_unsigned_short used to be called load_unsigned_word.
2864 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2865   int off = offset();
2866   ldrh(dst, src);
2867   return off;
2868 }
2869 
2870 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2871   int off = offset();
2872   ldrb(dst, src);
2873   return off;
2874 }
2875 
2876 int MacroAssembler::load_signed_short(Register dst, Address src) {
2877   int off = offset();
2878   ldrsh(dst, src);
2879   return off;
2880 }
2881 
2882 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2883   int off = offset();
2884   ldrsb(dst, src);
2885   return off;
2886 }
2887 
2888 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2889   int off = offset();
2890   ldrshw(dst, src);
2891   return off;
2892 }
2893 
2894 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2895   int off = offset();
2896   ldrsbw(dst, src);
2897   return off;
2898 }
2899 
2900 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2901   switch (size_in_bytes) {
2902   case  8:  ldr(dst, src); break;
2903   case  4:  ldrw(dst, src); break;
2904   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2905   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2906   default:  ShouldNotReachHere();
2907   }
2908 }
2909 
2910 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2911   switch (size_in_bytes) {
2912   case  8:  str(src, dst); break;
2913   case  4:  strw(src, dst); break;
2914   case  2:  strh(src, dst); break;
2915   case  1:  strb(src, dst); break;
2916   default:  ShouldNotReachHere();
2917   }
2918 }
2919 
2920 void MacroAssembler::decrementw(Register reg, int value)
2921 {
2922   if (value < 0)  { incrementw(reg, -value);      return; }
2923   if (value == 0) {                               return; }
2924   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2925   /* else */ {
2926     guarantee(reg != rscratch2, "invalid dst for register decrement");
2927     movw(rscratch2, (unsigned)value);
2928     subw(reg, reg, rscratch2);
2929   }
2930 }
2931 
2932 void MacroAssembler::decrement(Register reg, int value)
2933 {
2934   if (value < 0)  { increment(reg, -value);      return; }
2935   if (value == 0) {                              return; }
2936   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2937   /* else */ {
2938     assert(reg != rscratch2, "invalid dst for register decrement");
2939     mov(rscratch2, (uint64_t)value);
2940     sub(reg, reg, rscratch2);
2941   }
2942 }
2943 
2944 void MacroAssembler::decrementw(Address dst, int value)
2945 {
2946   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2947   if (dst.getMode() == Address::literal) {
2948     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2949     lea(rscratch2, dst);
2950     dst = Address(rscratch2);
2951   }
2952   ldrw(rscratch1, dst);
2953   decrementw(rscratch1, value);
2954   strw(rscratch1, dst);
2955 }
2956 
2957 void MacroAssembler::decrement(Address dst, int value)
2958 {
2959   assert(!dst.uses(rscratch1), "invalid address for decrement");
2960   if (dst.getMode() == Address::literal) {
2961     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2962     lea(rscratch2, dst);
2963     dst = Address(rscratch2);
2964   }
2965   ldr(rscratch1, dst);
2966   decrement(rscratch1, value);
2967   str(rscratch1, dst);
2968 }
2969 
2970 void MacroAssembler::incrementw(Register reg, int value)
2971 {
2972   if (value < 0)  { decrementw(reg, -value);      return; }
2973   if (value == 0) {                               return; }
2974   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2975   /* else */ {
2976     assert(reg != rscratch2, "invalid dst for register increment");
2977     movw(rscratch2, (unsigned)value);
2978     addw(reg, reg, rscratch2);
2979   }
2980 }
2981 
2982 void MacroAssembler::increment(Register reg, int value)
2983 {
2984   if (value < 0)  { decrement(reg, -value);      return; }
2985   if (value == 0) {                              return; }
2986   if (value < (1 << 12)) { add(reg, reg, value); return; }
2987   /* else */ {
2988     assert(reg != rscratch2, "invalid dst for register increment");
2989     movw(rscratch2, (unsigned)value);
2990     add(reg, reg, rscratch2);
2991   }
2992 }
2993 
2994 void MacroAssembler::incrementw(Address dst, int value)
2995 {
2996   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2997   if (dst.getMode() == Address::literal) {
2998     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2999     lea(rscratch2, dst);
3000     dst = Address(rscratch2);
3001   }
3002   ldrw(rscratch1, dst);
3003   incrementw(rscratch1, value);
3004   strw(rscratch1, dst);
3005 }
3006 
3007 void MacroAssembler::increment(Address dst, int value)
3008 {
3009   assert(!dst.uses(rscratch1), "invalid dst for address increment");
3010   if (dst.getMode() == Address::literal) {
3011     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
3012     lea(rscratch2, dst);
3013     dst = Address(rscratch2);
3014   }
3015   ldr(rscratch1, dst);
3016   increment(rscratch1, value);
3017   str(rscratch1, dst);
3018 }
3019 
3020 // Push lots of registers in the bit set supplied.  Don't push sp.
3021 // Return the number of words pushed
3022 int MacroAssembler::push(unsigned int bitset, Register stack) {
3023   int words_pushed = 0;
3024 
3025   // Scan bitset to accumulate register pairs
3026   unsigned char regs[32];
3027   int count = 0;
3028   for (int reg = 0; reg <= 30; reg++) {
3029     if (1 & bitset)
3030       regs[count++] = reg;
3031     bitset >>= 1;
3032   }
3033   regs[count++] = zr->raw_encoding();
3034   count &= ~1;  // Only push an even number of regs
3035 
3036   if (count) {
3037     stp(as_Register(regs[0]), as_Register(regs[1]),
3038        Address(pre(stack, -count * wordSize)));
3039     words_pushed += 2;
3040   }
3041   for (int i = 2; i < count; i += 2) {
3042     stp(as_Register(regs[i]), as_Register(regs[i+1]),
3043        Address(stack, i * wordSize));
3044     words_pushed += 2;
3045   }
3046 
3047   assert(words_pushed == count, "oops, pushed != count");
3048 
3049   return count;
3050 }
3051 
3052 int MacroAssembler::pop(unsigned int bitset, Register stack) {
3053   int words_pushed = 0;
3054 
3055   // Scan bitset to accumulate register pairs
3056   unsigned char regs[32];
3057   int count = 0;
3058   for (int reg = 0; reg <= 30; reg++) {
3059     if (1 & bitset)
3060       regs[count++] = reg;
3061     bitset >>= 1;
3062   }
3063   regs[count++] = zr->raw_encoding();
3064   count &= ~1;
3065 
3066   for (int i = 2; i < count; i += 2) {
3067     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
3068        Address(stack, i * wordSize));
3069     words_pushed += 2;
3070   }
3071   if (count) {
3072     ldp(as_Register(regs[0]), as_Register(regs[1]),
3073        Address(post(stack, count * wordSize)));
3074     words_pushed += 2;
3075   }
3076 
3077   assert(words_pushed == count, "oops, pushed != count");
3078 
3079   return count;
3080 }
3081 
3082 // Push lots of registers in the bit set supplied.  Don't push sp.
3083 // Return the number of dwords pushed
3084 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3085   int words_pushed = 0;
3086   bool use_sve = false;
3087   int sve_vector_size_in_bytes = 0;
3088 
3089 #ifdef COMPILER2
3090   use_sve = Matcher::supports_scalable_vector();
3091   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3092 #endif
3093 
3094   // Scan bitset to accumulate register pairs
3095   unsigned char regs[32];
3096   int count = 0;
3097   for (int reg = 0; reg <= 31; reg++) {
3098     if (1 & bitset)
3099       regs[count++] = reg;
3100     bitset >>= 1;
3101   }
3102 
3103   if (count == 0) {
3104     return 0;
3105   }
3106 
3107   if (mode == PushPopFull) {
3108     if (use_sve && sve_vector_size_in_bytes > 16) {
3109       mode = PushPopSVE;
3110     } else {
3111       mode = PushPopNeon;
3112     }
3113   }
3114 
3115 #ifndef PRODUCT
3116   {
3117     char buffer[48];
3118     if (mode == PushPopSVE) {
3119       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3120     } else if (mode == PushPopNeon) {
3121       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3122     } else {
3123       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3124     }
3125     block_comment(buffer);
3126   }
3127 #endif
3128 
3129   if (mode == PushPopSVE) {
3130     sub(stack, stack, sve_vector_size_in_bytes * count);
3131     for (int i = 0; i < count; i++) {
3132       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3133     }
3134     return count * sve_vector_size_in_bytes / 8;
3135   }
3136 
3137   if (mode == PushPopNeon) {
3138     if (count == 1) {
3139       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3140       return 2;
3141     }
3142 
3143     bool odd = (count & 1) == 1;
3144     int push_slots = count + (odd ? 1 : 0);
3145 
3146     // Always pushing full 128 bit registers.
3147     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3148     words_pushed += 2;
3149 
3150     for (int i = 2; i + 1 < count; i += 2) {
3151       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3152       words_pushed += 2;
3153     }
3154 
3155     if (odd) {
3156       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3157       words_pushed++;
3158     }
3159 
3160     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3161     return count * 2;
3162   }
3163 
3164   if (mode == PushPopFp) {
3165     bool odd = (count & 1) == 1;
3166     int push_slots = count + (odd ? 1 : 0);
3167 
3168     if (count == 1) {
3169       // Stack pointer must be 16 bytes aligned
3170       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3171       return 1;
3172     }
3173 
3174     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3175     words_pushed += 2;
3176 
3177     for (int i = 2; i + 1 < count; i += 2) {
3178       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3179       words_pushed += 2;
3180     }
3181 
3182     if (odd) {
3183       // Stack pointer must be 16 bytes aligned
3184       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3185       words_pushed++;
3186     }
3187 
3188     assert(words_pushed == count, "oops, pushed != count");
3189 
3190     return count;
3191   }
3192 
3193   return 0;
3194 }
3195 
3196 // Return the number of dwords popped
3197 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3198   int words_pushed = 0;
3199   bool use_sve = false;
3200   int sve_vector_size_in_bytes = 0;
3201 
3202 #ifdef COMPILER2
3203   use_sve = Matcher::supports_scalable_vector();
3204   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3205 #endif
3206   // Scan bitset to accumulate register pairs
3207   unsigned char regs[32];
3208   int count = 0;
3209   for (int reg = 0; reg <= 31; reg++) {
3210     if (1 & bitset)
3211       regs[count++] = reg;
3212     bitset >>= 1;
3213   }
3214 
3215   if (count == 0) {
3216     return 0;
3217   }
3218 
3219   if (mode == PushPopFull) {
3220     if (use_sve && sve_vector_size_in_bytes > 16) {
3221       mode = PushPopSVE;
3222     } else {
3223       mode = PushPopNeon;
3224     }
3225   }
3226 
3227 #ifndef PRODUCT
3228   {
3229     char buffer[48];
3230     if (mode == PushPopSVE) {
3231       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3232     } else if (mode == PushPopNeon) {
3233       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3234     } else {
3235       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3236     }
3237     block_comment(buffer);
3238   }
3239 #endif
3240 
3241   if (mode == PushPopSVE) {
3242     for (int i = count - 1; i >= 0; i--) {
3243       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3244     }
3245     add(stack, stack, sve_vector_size_in_bytes * count);
3246     return count * sve_vector_size_in_bytes / 8;
3247   }
3248 
3249   if (mode == PushPopNeon) {
3250     if (count == 1) {
3251       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3252       return 2;
3253     }
3254 
3255     bool odd = (count & 1) == 1;
3256     int push_slots = count + (odd ? 1 : 0);
3257 
3258     if (odd) {
3259       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3260       words_pushed++;
3261     }
3262 
3263     for (int i = 2; i + 1 < count; i += 2) {
3264       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3265       words_pushed += 2;
3266     }
3267 
3268     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3269     words_pushed += 2;
3270 
3271     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3272 
3273     return count * 2;
3274   }
3275 
3276   if (mode == PushPopFp) {
3277     bool odd = (count & 1) == 1;
3278     int push_slots = count + (odd ? 1 : 0);
3279 
3280     if (count == 1) {
3281       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3282       return 1;
3283     }
3284 
3285     if (odd) {
3286       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3287       words_pushed++;
3288     }
3289 
3290     for (int i = 2; i + 1 < count; i += 2) {
3291       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3292       words_pushed += 2;
3293     }
3294 
3295     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3296     words_pushed += 2;
3297 
3298     assert(words_pushed == count, "oops, pushed != count");
3299 
3300     return count;
3301   }
3302 
3303   return 0;
3304 }
3305 
3306 // Return the number of dwords pushed
3307 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3308   bool use_sve = false;
3309   int sve_predicate_size_in_slots = 0;
3310 
3311 #ifdef COMPILER2
3312   use_sve = Matcher::supports_scalable_vector();
3313   if (use_sve) {
3314     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3315   }
3316 #endif
3317 
3318   if (!use_sve) {
3319     return 0;
3320   }
3321 
3322   unsigned char regs[PRegister::number_of_registers];
3323   int count = 0;
3324   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3325     if (1 & bitset)
3326       regs[count++] = reg;
3327     bitset >>= 1;
3328   }
3329 
3330   if (count == 0) {
3331     return 0;
3332   }
3333 
3334   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3335                                   VMRegImpl::stack_slot_size * count, 16);
3336   sub(stack, stack, total_push_bytes);
3337   for (int i = 0; i < count; i++) {
3338     sve_str(as_PRegister(regs[i]), Address(stack, i));
3339   }
3340   return total_push_bytes / 8;
3341 }
3342 
3343 // Return the number of dwords popped
3344 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3345   bool use_sve = false;
3346   int sve_predicate_size_in_slots = 0;
3347 
3348 #ifdef COMPILER2
3349   use_sve = Matcher::supports_scalable_vector();
3350   if (use_sve) {
3351     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3352   }
3353 #endif
3354 
3355   if (!use_sve) {
3356     return 0;
3357   }
3358 
3359   unsigned char regs[PRegister::number_of_registers];
3360   int count = 0;
3361   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3362     if (1 & bitset)
3363       regs[count++] = reg;
3364     bitset >>= 1;
3365   }
3366 
3367   if (count == 0) {
3368     return 0;
3369   }
3370 
3371   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3372                                  VMRegImpl::stack_slot_size * count, 16);
3373   for (int i = count - 1; i >= 0; i--) {
3374     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3375   }
3376   add(stack, stack, total_pop_bytes);
3377   return total_pop_bytes / 8;
3378 }
3379 
3380 #ifdef ASSERT
3381 void MacroAssembler::verify_heapbase(const char* msg) {
3382 #if 0
3383   assert (Universe::heap() != nullptr, "java heap should be initialized");
3384   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3385     // rheapbase is allocated as general register
3386     return;
3387   }
3388   if (CheckCompressedOops) {
3389     Label ok;
3390     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3391     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3392     br(Assembler::EQ, ok);
3393     stop(msg);
3394     bind(ok);
3395     pop(1 << rscratch1->encoding(), sp);
3396   }
3397 #endif
3398 }
3399 #endif
3400 
3401 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3402   assert_different_registers(value, tmp1, tmp2);
3403   Label done, tagged, weak_tagged;
3404 
3405   cbz(value, done);           // Use null as-is.
3406   tst(value, JNIHandles::tag_mask); // Test for tag.
3407   br(Assembler::NE, tagged);
3408 
3409   // Resolve local handle
3410   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3411   verify_oop(value);
3412   b(done);
3413 
3414   bind(tagged);
3415   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3416   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3417 
3418   // Resolve global handle
3419   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3420   verify_oop(value);
3421   b(done);
3422 
3423   bind(weak_tagged);
3424   // Resolve jweak.
3425   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3426                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3427   verify_oop(value);
3428 
3429   bind(done);
3430 }
3431 
3432 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3433   assert_different_registers(value, tmp1, tmp2);
3434   Label done;
3435 
3436   cbz(value, done);           // Use null as-is.
3437 
3438 #ifdef ASSERT
3439   {
3440     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3441     Label valid_global_tag;
3442     tbnz(value, 1, valid_global_tag); // Test for global tag
3443     stop("non global jobject using resolve_global_jobject");
3444     bind(valid_global_tag);
3445   }
3446 #endif
3447 
3448   // Resolve global handle
3449   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3450   verify_oop(value);
3451 
3452   bind(done);
3453 }
3454 
3455 void MacroAssembler::stop(const char* msg) {
3456   // Skip AOT caching C strings in scratch buffer.
3457   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3458   BLOCK_COMMENT(str);
3459   // load msg into r0 so we can access it from the signal handler
3460   // ExternalAddress enables saving and restoring via the code cache
3461   lea(c_rarg0, ExternalAddress((address) str));
3462   dcps1(0xdeae);
3463 }
3464 
3465 void MacroAssembler::unimplemented(const char* what) {
3466   const char* buf = nullptr;
3467   {
3468     ResourceMark rm;
3469     stringStream ss;
3470     ss.print("unimplemented: %s", what);
3471     buf = code_string(ss.as_string());
3472   }
3473   stop(buf);
3474 }
3475 
3476 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3477 #ifdef ASSERT
3478   Label OK;
3479   br(cc, OK);
3480   stop(msg);
3481   bind(OK);
3482 #endif
3483 }
3484 
3485 // If a constant does not fit in an immediate field, generate some
3486 // number of MOV instructions and then perform the operation.
3487 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3488                                            add_sub_imm_insn insn1,
3489                                            add_sub_reg_insn insn2,
3490                                            bool is32) {
3491   assert(Rd != zr, "Rd = zr and not setting flags?");
3492   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3493   if (fits) {
3494     (this->*insn1)(Rd, Rn, imm);
3495   } else {
3496     if (g_uabs(imm) < (1 << 24)) {
3497        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3498        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3499     } else {
3500        assert_different_registers(Rd, Rn);
3501        mov(Rd, imm);
3502        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3503     }
3504   }
3505 }
3506 
3507 // Separate vsn which sets the flags. Optimisations are more restricted
3508 // because we must set the flags correctly.
3509 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3510                                              add_sub_imm_insn insn1,
3511                                              add_sub_reg_insn insn2,
3512                                              bool is32) {
3513   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3514   if (fits) {
3515     (this->*insn1)(Rd, Rn, imm);
3516   } else {
3517     assert_different_registers(Rd, Rn);
3518     assert(Rd != zr, "overflow in immediate operand");
3519     mov(Rd, imm);
3520     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3521   }
3522 }
3523 
3524 
3525 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3526   if (increment.is_register()) {
3527     add(Rd, Rn, increment.as_register());
3528   } else {
3529     add(Rd, Rn, increment.as_constant());
3530   }
3531 }
3532 
3533 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3534   if (increment.is_register()) {
3535     addw(Rd, Rn, increment.as_register());
3536   } else {
3537     addw(Rd, Rn, increment.as_constant());
3538   }
3539 }
3540 
3541 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3542   if (decrement.is_register()) {
3543     sub(Rd, Rn, decrement.as_register());
3544   } else {
3545     sub(Rd, Rn, decrement.as_constant());
3546   }
3547 }
3548 
3549 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3550   if (decrement.is_register()) {
3551     subw(Rd, Rn, decrement.as_register());
3552   } else {
3553     subw(Rd, Rn, decrement.as_constant());
3554   }
3555 }
3556 
3557 void MacroAssembler::reinit_heapbase()
3558 {
3559   if (UseCompressedOops) {
3560     if (Universe::is_fully_initialized()) {
3561       mov(rheapbase, CompressedOops::base());
3562     } else {
3563       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3564       ldr(rheapbase, Address(rheapbase));
3565     }
3566   }
3567 }
3568 
3569 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3570 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3571 // Pass a register for the result, otherwise pass noreg.
3572 
3573 // Clobbers rscratch1
3574 void MacroAssembler::cmpxchg(Register addr, Register expected,
3575                              Register new_val,
3576                              enum operand_size size,
3577                              bool acquire, bool release,
3578                              bool weak,
3579                              Register result) {
3580   if (result == noreg)  result = rscratch1;
3581   BLOCK_COMMENT("cmpxchg {");
3582   if (UseLSE) {
3583     mov(result, expected);
3584     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3585     compare_eq(result, expected, size);
3586 #ifdef ASSERT
3587     // Poison rscratch1 which is written on !UseLSE branch
3588     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3589 #endif
3590   } else {
3591     Label retry_load, done;
3592     prfm(Address(addr), PSTL1STRM);
3593     bind(retry_load);
3594     load_exclusive(result, addr, size, acquire);
3595     compare_eq(result, expected, size);
3596     br(Assembler::NE, done);
3597     store_exclusive(rscratch1, new_val, addr, size, release);
3598     if (weak) {
3599       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3600     } else {
3601       cbnzw(rscratch1, retry_load);
3602     }
3603     bind(done);
3604   }
3605   BLOCK_COMMENT("} cmpxchg");
3606 }
3607 
3608 // A generic comparison. Only compares for equality, clobbers rscratch1.
3609 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3610   if (size == xword) {
3611     cmp(rm, rn);
3612   } else if (size == word) {
3613     cmpw(rm, rn);
3614   } else if (size == halfword) {
3615     eorw(rscratch1, rm, rn);
3616     ands(zr, rscratch1, 0xffff);
3617   } else if (size == byte) {
3618     eorw(rscratch1, rm, rn);
3619     ands(zr, rscratch1, 0xff);
3620   } else {
3621     ShouldNotReachHere();
3622   }
3623 }
3624 
3625 
3626 static bool different(Register a, RegisterOrConstant b, Register c) {
3627   if (b.is_constant())
3628     return a != c;
3629   else
3630     return a != b.as_register() && a != c && b.as_register() != c;
3631 }
3632 
3633 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3634 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3635   if (UseLSE) {                                                         \
3636     prev = prev->is_valid() ? prev : zr;                                \
3637     if (incr.is_register()) {                                           \
3638       AOP(sz, incr.as_register(), prev, addr);                          \
3639     } else {                                                            \
3640       mov(rscratch2, incr.as_constant());                               \
3641       AOP(sz, rscratch2, prev, addr);                                   \
3642     }                                                                   \
3643     return;                                                             \
3644   }                                                                     \
3645   Register result = rscratch2;                                          \
3646   if (prev->is_valid())                                                 \
3647     result = different(prev, incr, addr) ? prev : rscratch2;            \
3648                                                                         \
3649   Label retry_load;                                                     \
3650   prfm(Address(addr), PSTL1STRM);                                       \
3651   bind(retry_load);                                                     \
3652   LDXR(result, addr);                                                   \
3653   OP(rscratch1, result, incr);                                          \
3654   STXR(rscratch2, rscratch1, addr);                                     \
3655   cbnzw(rscratch2, retry_load);                                         \
3656   if (prev->is_valid() && prev != result) {                             \
3657     IOP(prev, rscratch1, incr);                                         \
3658   }                                                                     \
3659 }
3660 
3661 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3662 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3663 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3664 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3665 
3666 #undef ATOMIC_OP
3667 
3668 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3669 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3670   if (UseLSE) {                                                         \
3671     prev = prev->is_valid() ? prev : zr;                                \
3672     AOP(sz, newv, prev, addr);                                          \
3673     return;                                                             \
3674   }                                                                     \
3675   Register result = rscratch2;                                          \
3676   if (prev->is_valid())                                                 \
3677     result = different(prev, newv, addr) ? prev : rscratch2;            \
3678                                                                         \
3679   Label retry_load;                                                     \
3680   prfm(Address(addr), PSTL1STRM);                                       \
3681   bind(retry_load);                                                     \
3682   LDXR(result, addr);                                                   \
3683   STXR(rscratch1, newv, addr);                                          \
3684   cbnzw(rscratch1, retry_load);                                         \
3685   if (prev->is_valid() && prev != result)                               \
3686     mov(prev, result);                                                  \
3687 }
3688 
3689 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3690 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3691 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3692 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3693 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3694 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3695 
3696 #undef ATOMIC_XCHG
3697 
3698 #ifndef PRODUCT
3699 extern "C" void findpc(intptr_t x);
3700 #endif
3701 
3702 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3703 {
3704   // In order to get locks to work, we need to fake a in_VM state
3705   if (ShowMessageBoxOnError) {
3706     JavaThread* thread = JavaThread::current();
3707     thread->set_thread_state(_thread_in_vm);
3708 #ifndef PRODUCT
3709     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3710       ttyLocker ttyl;
3711       BytecodeCounter::print();
3712     }
3713 #endif
3714     if (os::message_box(msg, "Execution stopped, print registers?")) {
3715       ttyLocker ttyl;
3716       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3717 #ifndef PRODUCT
3718       tty->cr();
3719       findpc(pc);
3720       tty->cr();
3721 #endif
3722       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3723       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3724       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3725       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3726       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3727       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3728       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3729       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3730       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3731       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3732       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3733       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3734       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3735       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3736       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3737       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3738       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3739       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3740       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3741       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3742       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3743       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3744       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3745       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3746       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3747       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3748       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3749       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3750       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3751       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3752       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3753       BREAKPOINT;
3754     }
3755   }
3756   fatal("DEBUG MESSAGE: %s", msg);
3757 }
3758 
3759 RegSet MacroAssembler::call_clobbered_gp_registers() {
3760   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3761 #ifndef R18_RESERVED
3762   regs += r18_tls;
3763 #endif
3764   return regs;
3765 }
3766 
3767 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3768   int step = 4 * wordSize;
3769   push(call_clobbered_gp_registers() - exclude, sp);
3770   sub(sp, sp, step);
3771   mov(rscratch1, -step);
3772   // Push v0-v7, v16-v31.
3773   for (int i = 31; i>= 4; i -= 4) {
3774     if (i <= v7->encoding() || i >= v16->encoding())
3775       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3776           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3777   }
3778   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3779       as_FloatRegister(3), T1D, Address(sp));
3780 }
3781 
3782 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3783   for (int i = 0; i < 32; i += 4) {
3784     if (i <= v7->encoding() || i >= v16->encoding())
3785       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3786           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3787   }
3788 
3789   reinitialize_ptrue();
3790 
3791   pop(call_clobbered_gp_registers() - exclude, sp);
3792 }
3793 
3794 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3795                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3796   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3797   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3798     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3799     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3800       sve_str(as_FloatRegister(i), Address(sp, i));
3801     }
3802   } else {
3803     int step = (save_vectors ? 8 : 4) * wordSize;
3804     mov(rscratch1, -step);
3805     sub(sp, sp, step);
3806     for (int i = 28; i >= 4; i -= 4) {
3807       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3808           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3809     }
3810     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3811   }
3812   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3813     sub(sp, sp, total_predicate_in_bytes);
3814     for (int i = 0; i < PRegister::number_of_registers; i++) {
3815       sve_str(as_PRegister(i), Address(sp, i));
3816     }
3817   }
3818 }
3819 
3820 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3821                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3822   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3823     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3824       sve_ldr(as_PRegister(i), Address(sp, i));
3825     }
3826     add(sp, sp, total_predicate_in_bytes);
3827   }
3828   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3829     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3830       sve_ldr(as_FloatRegister(i), Address(sp, i));
3831     }
3832     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3833   } else {
3834     int step = (restore_vectors ? 8 : 4) * wordSize;
3835     for (int i = 0; i <= 28; i += 4)
3836       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3837           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3838   }
3839 
3840   // We may use predicate registers and rely on ptrue with SVE,
3841   // regardless of wide vector (> 8 bytes) used or not.
3842   if (use_sve) {
3843     reinitialize_ptrue();
3844   }
3845 
3846   // integer registers except lr & sp
3847   pop(RegSet::range(r0, r17), sp);
3848 #ifdef R18_RESERVED
3849   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3850   pop(RegSet::range(r20, r29), sp);
3851 #else
3852   pop(RegSet::range(r18_tls, r29), sp);
3853 #endif
3854 }
3855 
3856 /**
3857  * Helpers for multiply_to_len().
3858  */
3859 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3860                                      Register src1, Register src2) {
3861   adds(dest_lo, dest_lo, src1);
3862   adc(dest_hi, dest_hi, zr);
3863   adds(dest_lo, dest_lo, src2);
3864   adc(final_dest_hi, dest_hi, zr);
3865 }
3866 
3867 // Generate an address from (r + r1 extend offset).  "size" is the
3868 // size of the operand.  The result may be in rscratch2.
3869 Address MacroAssembler::offsetted_address(Register r, Register r1,
3870                                           Address::extend ext, int offset, int size) {
3871   if (offset || (ext.shift() % size != 0)) {
3872     lea(rscratch2, Address(r, r1, ext));
3873     return Address(rscratch2, offset);
3874   } else {
3875     return Address(r, r1, ext);
3876   }
3877 }
3878 
3879 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3880 {
3881   assert(offset >= 0, "spill to negative address?");
3882   // Offset reachable ?
3883   //   Not aligned - 9 bits signed offset
3884   //   Aligned - 12 bits unsigned offset shifted
3885   Register base = sp;
3886   if ((offset & (size-1)) && offset >= (1<<8)) {
3887     add(tmp, base, offset & ((1<<12)-1));
3888     base = tmp;
3889     offset &= -1u<<12;
3890   }
3891 
3892   if (offset >= (1<<12) * size) {
3893     add(tmp, base, offset & (((1<<12)-1)<<12));
3894     base = tmp;
3895     offset &= ~(((1<<12)-1)<<12);
3896   }
3897 
3898   return Address(base, offset);
3899 }
3900 
3901 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3902   assert(offset >= 0, "spill to negative address?");
3903 
3904   Register base = sp;
3905 
3906   // An immediate offset in the range 0 to 255 which is multiplied
3907   // by the current vector or predicate register size in bytes.
3908   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3909     return Address(base, offset / sve_reg_size_in_bytes);
3910   }
3911 
3912   add(tmp, base, offset);
3913   return Address(tmp);
3914 }
3915 
3916 // Checks whether offset is aligned.
3917 // Returns true if it is, else false.
3918 bool MacroAssembler::merge_alignment_check(Register base,
3919                                            size_t size,
3920                                            int64_t cur_offset,
3921                                            int64_t prev_offset) const {
3922   if (AvoidUnalignedAccesses) {
3923     if (base == sp) {
3924       // Checks whether low offset if aligned to pair of registers.
3925       int64_t pair_mask = size * 2 - 1;
3926       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3927       return (offset & pair_mask) == 0;
3928     } else { // If base is not sp, we can't guarantee the access is aligned.
3929       return false;
3930     }
3931   } else {
3932     int64_t mask = size - 1;
3933     // Load/store pair instruction only supports element size aligned offset.
3934     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3935   }
3936 }
3937 
3938 // Checks whether current and previous loads/stores can be merged.
3939 // Returns true if it can be merged, else false.
3940 bool MacroAssembler::ldst_can_merge(Register rt,
3941                                     const Address &adr,
3942                                     size_t cur_size_in_bytes,
3943                                     bool is_store) const {
3944   address prev = pc() - NativeInstruction::instruction_size;
3945   address last = code()->last_insn();
3946 
3947   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3948     return false;
3949   }
3950 
3951   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3952     return false;
3953   }
3954 
3955   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3956   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3957 
3958   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3959   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3960 
3961   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3962     return false;
3963   }
3964 
3965   int64_t max_offset = 63 * prev_size_in_bytes;
3966   int64_t min_offset = -64 * prev_size_in_bytes;
3967 
3968   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3969 
3970   // Only same base can be merged.
3971   if (adr.base() != prev_ldst->base()) {
3972     return false;
3973   }
3974 
3975   int64_t cur_offset = adr.offset();
3976   int64_t prev_offset = prev_ldst->offset();
3977   size_t diff = abs(cur_offset - prev_offset);
3978   if (diff != prev_size_in_bytes) {
3979     return false;
3980   }
3981 
3982   // Following cases can not be merged:
3983   // ldr x2, [x2, #8]
3984   // ldr x3, [x2, #16]
3985   // or:
3986   // ldr x2, [x3, #8]
3987   // ldr x2, [x3, #16]
3988   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3989   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3990     return false;
3991   }
3992 
3993   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3994   // Offset range must be in ldp/stp instruction's range.
3995   if (low_offset > max_offset || low_offset < min_offset) {
3996     return false;
3997   }
3998 
3999   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
4000     return true;
4001   }
4002 
4003   return false;
4004 }
4005 
4006 // Merge current load/store with previous load/store into ldp/stp.
4007 void MacroAssembler::merge_ldst(Register rt,
4008                                 const Address &adr,
4009                                 size_t cur_size_in_bytes,
4010                                 bool is_store) {
4011 
4012   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
4013 
4014   Register rt_low, rt_high;
4015   address prev = pc() - NativeInstruction::instruction_size;
4016   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
4017 
4018   int64_t offset;
4019 
4020   if (adr.offset() < prev_ldst->offset()) {
4021     offset = adr.offset();
4022     rt_low = rt;
4023     rt_high = prev_ldst->target();
4024   } else {
4025     offset = prev_ldst->offset();
4026     rt_low = prev_ldst->target();
4027     rt_high = rt;
4028   }
4029 
4030   Address adr_p = Address(prev_ldst->base(), offset);
4031   // Overwrite previous generated binary.
4032   code_section()->set_end(prev);
4033 
4034   const size_t sz = prev_ldst->size_in_bytes();
4035   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
4036   if (!is_store) {
4037     BLOCK_COMMENT("merged ldr pair");
4038     if (sz == 8) {
4039       ldp(rt_low, rt_high, adr_p);
4040     } else {
4041       ldpw(rt_low, rt_high, adr_p);
4042     }
4043   } else {
4044     BLOCK_COMMENT("merged str pair");
4045     if (sz == 8) {
4046       stp(rt_low, rt_high, adr_p);
4047     } else {
4048       stpw(rt_low, rt_high, adr_p);
4049     }
4050   }
4051 }
4052 
4053 /**
4054  * Multiply 64 bit by 64 bit first loop.
4055  */
4056 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
4057                                            Register y, Register y_idx, Register z,
4058                                            Register carry, Register product,
4059                                            Register idx, Register kdx) {
4060   //
4061   //  jlong carry, x[], y[], z[];
4062   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4063   //    huge_128 product = y[idx] * x[xstart] + carry;
4064   //    z[kdx] = (jlong)product;
4065   //    carry  = (jlong)(product >>> 64);
4066   //  }
4067   //  z[xstart] = carry;
4068   //
4069 
4070   Label L_first_loop, L_first_loop_exit;
4071   Label L_one_x, L_one_y, L_multiply;
4072 
4073   subsw(xstart, xstart, 1);
4074   br(Assembler::MI, L_one_x);
4075 
4076   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
4077   ldr(x_xstart, Address(rscratch1));
4078   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
4079 
4080   bind(L_first_loop);
4081   subsw(idx, idx, 1);
4082   br(Assembler::MI, L_first_loop_exit);
4083   subsw(idx, idx, 1);
4084   br(Assembler::MI, L_one_y);
4085   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4086   ldr(y_idx, Address(rscratch1));
4087   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
4088   bind(L_multiply);
4089 
4090   // AArch64 has a multiply-accumulate instruction that we can't use
4091   // here because it has no way to process carries, so we have to use
4092   // separate add and adc instructions.  Bah.
4093   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
4094   mul(product, x_xstart, y_idx);
4095   adds(product, product, carry);
4096   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
4097 
4098   subw(kdx, kdx, 2);
4099   ror(product, product, 32); // back to big-endian
4100   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4101 
4102   b(L_first_loop);
4103 
4104   bind(L_one_y);
4105   ldrw(y_idx, Address(y,  0));
4106   b(L_multiply);
4107 
4108   bind(L_one_x);
4109   ldrw(x_xstart, Address(x,  0));
4110   b(L_first_loop);
4111 
4112   bind(L_first_loop_exit);
4113 }
4114 
4115 /**
4116  * Multiply 128 bit by 128. Unrolled inner loop.
4117  *
4118  */
4119 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4120                                              Register carry, Register carry2,
4121                                              Register idx, Register jdx,
4122                                              Register yz_idx1, Register yz_idx2,
4123                                              Register tmp, Register tmp3, Register tmp4,
4124                                              Register tmp6, Register product_hi) {
4125 
4126   //   jlong carry, x[], y[], z[];
4127   //   int kdx = ystart+1;
4128   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4129   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4130   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4131   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4132   //     carry  = (jlong)(tmp4 >>> 64);
4133   //     z[kdx+idx+1] = (jlong)tmp3;
4134   //     z[kdx+idx] = (jlong)tmp4;
4135   //   }
4136   //   idx += 2;
4137   //   if (idx > 0) {
4138   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4139   //     z[kdx+idx] = (jlong)yz_idx1;
4140   //     carry  = (jlong)(yz_idx1 >>> 64);
4141   //   }
4142   //
4143 
4144   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4145 
4146   lsrw(jdx, idx, 2);
4147 
4148   bind(L_third_loop);
4149 
4150   subsw(jdx, jdx, 1);
4151   br(Assembler::MI, L_third_loop_exit);
4152   subw(idx, idx, 4);
4153 
4154   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4155 
4156   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4157 
4158   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4159 
4160   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4161   ror(yz_idx2, yz_idx2, 32);
4162 
4163   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4164 
4165   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4166   umulh(tmp4, product_hi, yz_idx1);
4167 
4168   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4169   ror(rscratch2, rscratch2, 32);
4170 
4171   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4172   umulh(carry2, product_hi, yz_idx2);
4173 
4174   // propagate sum of both multiplications into carry:tmp4:tmp3
4175   adds(tmp3, tmp3, carry);
4176   adc(tmp4, tmp4, zr);
4177   adds(tmp3, tmp3, rscratch1);
4178   adcs(tmp4, tmp4, tmp);
4179   adc(carry, carry2, zr);
4180   adds(tmp4, tmp4, rscratch2);
4181   adc(carry, carry, zr);
4182 
4183   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4184   ror(tmp4, tmp4, 32);
4185   stp(tmp4, tmp3, Address(tmp6, 0));
4186 
4187   b(L_third_loop);
4188   bind (L_third_loop_exit);
4189 
4190   andw (idx, idx, 0x3);
4191   cbz(idx, L_post_third_loop_done);
4192 
4193   Label L_check_1;
4194   subsw(idx, idx, 2);
4195   br(Assembler::MI, L_check_1);
4196 
4197   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4198   ldr(yz_idx1, Address(rscratch1, 0));
4199   ror(yz_idx1, yz_idx1, 32);
4200   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4201   umulh(tmp4, product_hi, yz_idx1);
4202   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4203   ldr(yz_idx2, Address(rscratch1, 0));
4204   ror(yz_idx2, yz_idx2, 32);
4205 
4206   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4207 
4208   ror(tmp3, tmp3, 32);
4209   str(tmp3, Address(rscratch1, 0));
4210 
4211   bind (L_check_1);
4212 
4213   andw (idx, idx, 0x1);
4214   subsw(idx, idx, 1);
4215   br(Assembler::MI, L_post_third_loop_done);
4216   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4217   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4218   umulh(carry2, tmp4, product_hi);
4219   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4220 
4221   add2_with_carry(carry2, tmp3, tmp4, carry);
4222 
4223   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4224   extr(carry, carry2, tmp3, 32);
4225 
4226   bind(L_post_third_loop_done);
4227 }
4228 
4229 /**
4230  * Code for BigInteger::multiplyToLen() intrinsic.
4231  *
4232  * r0: x
4233  * r1: xlen
4234  * r2: y
4235  * r3: ylen
4236  * r4:  z
4237  * r5: tmp0
4238  * r10: tmp1
4239  * r11: tmp2
4240  * r12: tmp3
4241  * r13: tmp4
4242  * r14: tmp5
4243  * r15: tmp6
4244  * r16: tmp7
4245  *
4246  */
4247 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4248                                      Register z, Register tmp0,
4249                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4250                                      Register tmp5, Register tmp6, Register product_hi) {
4251 
4252   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4253 
4254   const Register idx = tmp1;
4255   const Register kdx = tmp2;
4256   const Register xstart = tmp3;
4257 
4258   const Register y_idx = tmp4;
4259   const Register carry = tmp5;
4260   const Register product  = xlen;
4261   const Register x_xstart = tmp0;
4262 
4263   // First Loop.
4264   //
4265   //  final static long LONG_MASK = 0xffffffffL;
4266   //  int xstart = xlen - 1;
4267   //  int ystart = ylen - 1;
4268   //  long carry = 0;
4269   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4270   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4271   //    z[kdx] = (int)product;
4272   //    carry = product >>> 32;
4273   //  }
4274   //  z[xstart] = (int)carry;
4275   //
4276 
4277   movw(idx, ylen);       // idx = ylen;
4278   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4279   mov(carry, zr);        // carry = 0;
4280 
4281   Label L_done;
4282 
4283   movw(xstart, xlen);
4284   subsw(xstart, xstart, 1);
4285   br(Assembler::MI, L_done);
4286 
4287   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4288 
4289   Label L_second_loop;
4290   cbzw(kdx, L_second_loop);
4291 
4292   Label L_carry;
4293   subw(kdx, kdx, 1);
4294   cbzw(kdx, L_carry);
4295 
4296   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4297   lsr(carry, carry, 32);
4298   subw(kdx, kdx, 1);
4299 
4300   bind(L_carry);
4301   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4302 
4303   // Second and third (nested) loops.
4304   //
4305   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4306   //   carry = 0;
4307   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4308   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4309   //                    (z[k] & LONG_MASK) + carry;
4310   //     z[k] = (int)product;
4311   //     carry = product >>> 32;
4312   //   }
4313   //   z[i] = (int)carry;
4314   // }
4315   //
4316   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4317 
4318   const Register jdx = tmp1;
4319 
4320   bind(L_second_loop);
4321   mov(carry, zr);                // carry = 0;
4322   movw(jdx, ylen);               // j = ystart+1
4323 
4324   subsw(xstart, xstart, 1);      // i = xstart-1;
4325   br(Assembler::MI, L_done);
4326 
4327   str(z, Address(pre(sp, -4 * wordSize)));
4328 
4329   Label L_last_x;
4330   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4331   subsw(xstart, xstart, 1);       // i = xstart-1;
4332   br(Assembler::MI, L_last_x);
4333 
4334   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4335   ldr(product_hi, Address(rscratch1));
4336   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4337 
4338   Label L_third_loop_prologue;
4339   bind(L_third_loop_prologue);
4340 
4341   str(ylen, Address(sp, wordSize));
4342   stp(x, xstart, Address(sp, 2 * wordSize));
4343   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4344                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4345   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4346   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4347 
4348   addw(tmp3, xlen, 1);
4349   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4350   subsw(tmp3, tmp3, 1);
4351   br(Assembler::MI, L_done);
4352 
4353   lsr(carry, carry, 32);
4354   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4355   b(L_second_loop);
4356 
4357   // Next infrequent code is moved outside loops.
4358   bind(L_last_x);
4359   ldrw(product_hi, Address(x,  0));
4360   b(L_third_loop_prologue);
4361 
4362   bind(L_done);
4363 }
4364 
4365 // Code for BigInteger::mulAdd intrinsic
4366 // out     = r0
4367 // in      = r1
4368 // offset  = r2  (already out.length-offset)
4369 // len     = r3
4370 // k       = r4
4371 //
4372 // pseudo code from java implementation:
4373 // carry = 0;
4374 // offset = out.length-offset - 1;
4375 // for (int j=len-1; j >= 0; j--) {
4376 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4377 //     out[offset--] = (int)product;
4378 //     carry = product >>> 32;
4379 // }
4380 // return (int)carry;
4381 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4382       Register len, Register k) {
4383     Label LOOP, END;
4384     // pre-loop
4385     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4386     csel(out, zr, out, Assembler::EQ);
4387     br(Assembler::EQ, END);
4388     add(in, in, len, LSL, 2); // in[j+1] address
4389     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4390     mov(out, zr); // used to keep carry now
4391     BIND(LOOP);
4392     ldrw(rscratch1, Address(pre(in, -4)));
4393     madd(rscratch1, rscratch1, k, out);
4394     ldrw(rscratch2, Address(pre(offset, -4)));
4395     add(rscratch1, rscratch1, rscratch2);
4396     strw(rscratch1, Address(offset));
4397     lsr(out, rscratch1, 32);
4398     subs(len, len, 1);
4399     br(Assembler::NE, LOOP);
4400     BIND(END);
4401 }
4402 
4403 /**
4404  * Emits code to update CRC-32 with a byte value according to constants in table
4405  *
4406  * @param [in,out]crc   Register containing the crc.
4407  * @param [in]val       Register containing the byte to fold into the CRC.
4408  * @param [in]table     Register containing the table of crc constants.
4409  *
4410  * uint32_t crc;
4411  * val = crc_table[(val ^ crc) & 0xFF];
4412  * crc = val ^ (crc >> 8);
4413  *
4414  */
4415 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4416   eor(val, val, crc);
4417   andr(val, val, 0xff);
4418   ldrw(val, Address(table, val, Address::lsl(2)));
4419   eor(crc, val, crc, Assembler::LSR, 8);
4420 }
4421 
4422 /**
4423  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4424  *
4425  * @param [in,out]crc   Register containing the crc.
4426  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4427  * @param [in]table0    Register containing table 0 of crc constants.
4428  * @param [in]table1    Register containing table 1 of crc constants.
4429  * @param [in]table2    Register containing table 2 of crc constants.
4430  * @param [in]table3    Register containing table 3 of crc constants.
4431  *
4432  * uint32_t crc;
4433  *   v = crc ^ v
4434  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4435  *
4436  */
4437 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4438         Register table0, Register table1, Register table2, Register table3,
4439         bool upper) {
4440   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4441   uxtb(tmp, v);
4442   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4443   ubfx(tmp, v, 8, 8);
4444   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4445   eor(crc, crc, tmp);
4446   ubfx(tmp, v, 16, 8);
4447   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4448   eor(crc, crc, tmp);
4449   ubfx(tmp, v, 24, 8);
4450   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4451   eor(crc, crc, tmp);
4452 }
4453 
4454 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4455         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4456     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4457     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4458 
4459     subs(tmp0, len, 384);
4460     mvnw(crc, crc);
4461     br(Assembler::GE, CRC_by128_pre);
4462   BIND(CRC_less128);
4463     subs(len, len, 32);
4464     br(Assembler::GE, CRC_by32_loop);
4465   BIND(CRC_less32);
4466     adds(len, len, 32 - 4);
4467     br(Assembler::GE, CRC_by4_loop);
4468     adds(len, len, 4);
4469     br(Assembler::GT, CRC_by1_loop);
4470     b(L_exit);
4471 
4472   BIND(CRC_by32_loop);
4473     ldp(tmp0, tmp1, Address(buf));
4474     crc32x(crc, crc, tmp0);
4475     ldp(tmp2, tmp3, Address(buf, 16));
4476     crc32x(crc, crc, tmp1);
4477     add(buf, buf, 32);
4478     crc32x(crc, crc, tmp2);
4479     subs(len, len, 32);
4480     crc32x(crc, crc, tmp3);
4481     br(Assembler::GE, CRC_by32_loop);
4482     cmn(len, (u1)32);
4483     br(Assembler::NE, CRC_less32);
4484     b(L_exit);
4485 
4486   BIND(CRC_by4_loop);
4487     ldrw(tmp0, Address(post(buf, 4)));
4488     subs(len, len, 4);
4489     crc32w(crc, crc, tmp0);
4490     br(Assembler::GE, CRC_by4_loop);
4491     adds(len, len, 4);
4492     br(Assembler::LE, L_exit);
4493   BIND(CRC_by1_loop);
4494     ldrb(tmp0, Address(post(buf, 1)));
4495     subs(len, len, 1);
4496     crc32b(crc, crc, tmp0);
4497     br(Assembler::GT, CRC_by1_loop);
4498     b(L_exit);
4499 
4500   BIND(CRC_by128_pre);
4501     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4502       4*256*sizeof(juint) + 8*sizeof(juint));
4503     mov(crc, 0);
4504     crc32x(crc, crc, tmp0);
4505     crc32x(crc, crc, tmp1);
4506 
4507     cbnz(len, CRC_less128);
4508 
4509   BIND(L_exit);
4510     mvnw(crc, crc);
4511 }
4512 
4513 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4514         Register len, Register tmp0, Register tmp1, Register tmp2,
4515         Register tmp3) {
4516     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4517     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4518 
4519     mvnw(crc, crc);
4520 
4521     subs(len, len, 128);
4522     br(Assembler::GE, CRC_by64_pre);
4523   BIND(CRC_less64);
4524     adds(len, len, 128-32);
4525     br(Assembler::GE, CRC_by32_loop);
4526   BIND(CRC_less32);
4527     adds(len, len, 32-4);
4528     br(Assembler::GE, CRC_by4_loop);
4529     adds(len, len, 4);
4530     br(Assembler::GT, CRC_by1_loop);
4531     b(L_exit);
4532 
4533   BIND(CRC_by32_loop);
4534     ldp(tmp0, tmp1, Address(post(buf, 16)));
4535     subs(len, len, 32);
4536     crc32x(crc, crc, tmp0);
4537     ldr(tmp2, Address(post(buf, 8)));
4538     crc32x(crc, crc, tmp1);
4539     ldr(tmp3, Address(post(buf, 8)));
4540     crc32x(crc, crc, tmp2);
4541     crc32x(crc, crc, tmp3);
4542     br(Assembler::GE, CRC_by32_loop);
4543     cmn(len, (u1)32);
4544     br(Assembler::NE, CRC_less32);
4545     b(L_exit);
4546 
4547   BIND(CRC_by4_loop);
4548     ldrw(tmp0, Address(post(buf, 4)));
4549     subs(len, len, 4);
4550     crc32w(crc, crc, tmp0);
4551     br(Assembler::GE, CRC_by4_loop);
4552     adds(len, len, 4);
4553     br(Assembler::LE, L_exit);
4554   BIND(CRC_by1_loop);
4555     ldrb(tmp0, Address(post(buf, 1)));
4556     subs(len, len, 1);
4557     crc32b(crc, crc, tmp0);
4558     br(Assembler::GT, CRC_by1_loop);
4559     b(L_exit);
4560 
4561   BIND(CRC_by64_pre);
4562     sub(buf, buf, 8);
4563     ldp(tmp0, tmp1, Address(buf, 8));
4564     crc32x(crc, crc, tmp0);
4565     ldr(tmp2, Address(buf, 24));
4566     crc32x(crc, crc, tmp1);
4567     ldr(tmp3, Address(buf, 32));
4568     crc32x(crc, crc, tmp2);
4569     ldr(tmp0, Address(buf, 40));
4570     crc32x(crc, crc, tmp3);
4571     ldr(tmp1, Address(buf, 48));
4572     crc32x(crc, crc, tmp0);
4573     ldr(tmp2, Address(buf, 56));
4574     crc32x(crc, crc, tmp1);
4575     ldr(tmp3, Address(pre(buf, 64)));
4576 
4577     b(CRC_by64_loop);
4578 
4579     align(CodeEntryAlignment);
4580   BIND(CRC_by64_loop);
4581     subs(len, len, 64);
4582     crc32x(crc, crc, tmp2);
4583     ldr(tmp0, Address(buf, 8));
4584     crc32x(crc, crc, tmp3);
4585     ldr(tmp1, Address(buf, 16));
4586     crc32x(crc, crc, tmp0);
4587     ldr(tmp2, Address(buf, 24));
4588     crc32x(crc, crc, tmp1);
4589     ldr(tmp3, Address(buf, 32));
4590     crc32x(crc, crc, tmp2);
4591     ldr(tmp0, Address(buf, 40));
4592     crc32x(crc, crc, tmp3);
4593     ldr(tmp1, Address(buf, 48));
4594     crc32x(crc, crc, tmp0);
4595     ldr(tmp2, Address(buf, 56));
4596     crc32x(crc, crc, tmp1);
4597     ldr(tmp3, Address(pre(buf, 64)));
4598     br(Assembler::GE, CRC_by64_loop);
4599 
4600     // post-loop
4601     crc32x(crc, crc, tmp2);
4602     crc32x(crc, crc, tmp3);
4603 
4604     sub(len, len, 64);
4605     add(buf, buf, 8);
4606     cmn(len, (u1)128);
4607     br(Assembler::NE, CRC_less64);
4608   BIND(L_exit);
4609     mvnw(crc, crc);
4610 }
4611 
4612 /**
4613  * @param crc   register containing existing CRC (32-bit)
4614  * @param buf   register pointing to input byte buffer (byte*)
4615  * @param len   register containing number of bytes
4616  * @param table register that will contain address of CRC table
4617  * @param tmp   scratch register
4618  */
4619 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4620         Register table0, Register table1, Register table2, Register table3,
4621         Register tmp, Register tmp2, Register tmp3) {
4622   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4623 
4624   if (UseCryptoPmullForCRC32) {
4625       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4626       return;
4627   }
4628 
4629   if (UseCRC32) {
4630       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4631       return;
4632   }
4633 
4634     mvnw(crc, crc);
4635 
4636     {
4637       uint64_t offset;
4638       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4639       add(table0, table0, offset);
4640     }
4641     add(table1, table0, 1*256*sizeof(juint));
4642     add(table2, table0, 2*256*sizeof(juint));
4643     add(table3, table0, 3*256*sizeof(juint));
4644 
4645     { // Neon code start
4646       cmp(len, (u1)64);
4647       br(Assembler::LT, L_by16);
4648       eor(v16, T16B, v16, v16);
4649 
4650     Label L_fold;
4651 
4652       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4653 
4654       ld1(v0, v1, T2D, post(buf, 32));
4655       ld1r(v4, T2D, post(tmp, 8));
4656       ld1r(v5, T2D, post(tmp, 8));
4657       ld1r(v6, T2D, post(tmp, 8));
4658       ld1r(v7, T2D, post(tmp, 8));
4659       mov(v16, S, 0, crc);
4660 
4661       eor(v0, T16B, v0, v16);
4662       sub(len, len, 64);
4663 
4664     BIND(L_fold);
4665       pmull(v22, T8H, v0, v5, T8B);
4666       pmull(v20, T8H, v0, v7, T8B);
4667       pmull(v23, T8H, v0, v4, T8B);
4668       pmull(v21, T8H, v0, v6, T8B);
4669 
4670       pmull2(v18, T8H, v0, v5, T16B);
4671       pmull2(v16, T8H, v0, v7, T16B);
4672       pmull2(v19, T8H, v0, v4, T16B);
4673       pmull2(v17, T8H, v0, v6, T16B);
4674 
4675       uzp1(v24, T8H, v20, v22);
4676       uzp2(v25, T8H, v20, v22);
4677       eor(v20, T16B, v24, v25);
4678 
4679       uzp1(v26, T8H, v16, v18);
4680       uzp2(v27, T8H, v16, v18);
4681       eor(v16, T16B, v26, v27);
4682 
4683       ushll2(v22, T4S, v20, T8H, 8);
4684       ushll(v20, T4S, v20, T4H, 8);
4685 
4686       ushll2(v18, T4S, v16, T8H, 8);
4687       ushll(v16, T4S, v16, T4H, 8);
4688 
4689       eor(v22, T16B, v23, v22);
4690       eor(v18, T16B, v19, v18);
4691       eor(v20, T16B, v21, v20);
4692       eor(v16, T16B, v17, v16);
4693 
4694       uzp1(v17, T2D, v16, v20);
4695       uzp2(v21, T2D, v16, v20);
4696       eor(v17, T16B, v17, v21);
4697 
4698       ushll2(v20, T2D, v17, T4S, 16);
4699       ushll(v16, T2D, v17, T2S, 16);
4700 
4701       eor(v20, T16B, v20, v22);
4702       eor(v16, T16B, v16, v18);
4703 
4704       uzp1(v17, T2D, v20, v16);
4705       uzp2(v21, T2D, v20, v16);
4706       eor(v28, T16B, v17, v21);
4707 
4708       pmull(v22, T8H, v1, v5, T8B);
4709       pmull(v20, T8H, v1, v7, T8B);
4710       pmull(v23, T8H, v1, v4, T8B);
4711       pmull(v21, T8H, v1, v6, T8B);
4712 
4713       pmull2(v18, T8H, v1, v5, T16B);
4714       pmull2(v16, T8H, v1, v7, T16B);
4715       pmull2(v19, T8H, v1, v4, T16B);
4716       pmull2(v17, T8H, v1, v6, T16B);
4717 
4718       ld1(v0, v1, T2D, post(buf, 32));
4719 
4720       uzp1(v24, T8H, v20, v22);
4721       uzp2(v25, T8H, v20, v22);
4722       eor(v20, T16B, v24, v25);
4723 
4724       uzp1(v26, T8H, v16, v18);
4725       uzp2(v27, T8H, v16, v18);
4726       eor(v16, T16B, v26, v27);
4727 
4728       ushll2(v22, T4S, v20, T8H, 8);
4729       ushll(v20, T4S, v20, T4H, 8);
4730 
4731       ushll2(v18, T4S, v16, T8H, 8);
4732       ushll(v16, T4S, v16, T4H, 8);
4733 
4734       eor(v22, T16B, v23, v22);
4735       eor(v18, T16B, v19, v18);
4736       eor(v20, T16B, v21, v20);
4737       eor(v16, T16B, v17, v16);
4738 
4739       uzp1(v17, T2D, v16, v20);
4740       uzp2(v21, T2D, v16, v20);
4741       eor(v16, T16B, v17, v21);
4742 
4743       ushll2(v20, T2D, v16, T4S, 16);
4744       ushll(v16, T2D, v16, T2S, 16);
4745 
4746       eor(v20, T16B, v22, v20);
4747       eor(v16, T16B, v16, v18);
4748 
4749       uzp1(v17, T2D, v20, v16);
4750       uzp2(v21, T2D, v20, v16);
4751       eor(v20, T16B, v17, v21);
4752 
4753       shl(v16, T2D, v28, 1);
4754       shl(v17, T2D, v20, 1);
4755 
4756       eor(v0, T16B, v0, v16);
4757       eor(v1, T16B, v1, v17);
4758 
4759       subs(len, len, 32);
4760       br(Assembler::GE, L_fold);
4761 
4762       mov(crc, 0);
4763       mov(tmp, v0, D, 0);
4764       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4765       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4766       mov(tmp, v0, D, 1);
4767       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4768       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4769       mov(tmp, v1, D, 0);
4770       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4771       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4772       mov(tmp, v1, D, 1);
4773       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4774       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4775 
4776       add(len, len, 32);
4777     } // Neon code end
4778 
4779   BIND(L_by16);
4780     subs(len, len, 16);
4781     br(Assembler::GE, L_by16_loop);
4782     adds(len, len, 16-4);
4783     br(Assembler::GE, L_by4_loop);
4784     adds(len, len, 4);
4785     br(Assembler::GT, L_by1_loop);
4786     b(L_exit);
4787 
4788   BIND(L_by4_loop);
4789     ldrw(tmp, Address(post(buf, 4)));
4790     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4791     subs(len, len, 4);
4792     br(Assembler::GE, L_by4_loop);
4793     adds(len, len, 4);
4794     br(Assembler::LE, L_exit);
4795   BIND(L_by1_loop);
4796     subs(len, len, 1);
4797     ldrb(tmp, Address(post(buf, 1)));
4798     update_byte_crc32(crc, tmp, table0);
4799     br(Assembler::GT, L_by1_loop);
4800     b(L_exit);
4801 
4802     align(CodeEntryAlignment);
4803   BIND(L_by16_loop);
4804     subs(len, len, 16);
4805     ldp(tmp, tmp3, Address(post(buf, 16)));
4806     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4807     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4808     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4809     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4810     br(Assembler::GE, L_by16_loop);
4811     adds(len, len, 16-4);
4812     br(Assembler::GE, L_by4_loop);
4813     adds(len, len, 4);
4814     br(Assembler::GT, L_by1_loop);
4815   BIND(L_exit);
4816     mvnw(crc, crc);
4817 }
4818 
4819 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4820         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4821     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4822     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4823 
4824     subs(tmp0, len, 384);
4825     br(Assembler::GE, CRC_by128_pre);
4826   BIND(CRC_less128);
4827     subs(len, len, 32);
4828     br(Assembler::GE, CRC_by32_loop);
4829   BIND(CRC_less32);
4830     adds(len, len, 32 - 4);
4831     br(Assembler::GE, CRC_by4_loop);
4832     adds(len, len, 4);
4833     br(Assembler::GT, CRC_by1_loop);
4834     b(L_exit);
4835 
4836   BIND(CRC_by32_loop);
4837     ldp(tmp0, tmp1, Address(buf));
4838     crc32cx(crc, crc, tmp0);
4839     ldr(tmp2, Address(buf, 16));
4840     crc32cx(crc, crc, tmp1);
4841     ldr(tmp3, Address(buf, 24));
4842     crc32cx(crc, crc, tmp2);
4843     add(buf, buf, 32);
4844     subs(len, len, 32);
4845     crc32cx(crc, crc, tmp3);
4846     br(Assembler::GE, CRC_by32_loop);
4847     cmn(len, (u1)32);
4848     br(Assembler::NE, CRC_less32);
4849     b(L_exit);
4850 
4851   BIND(CRC_by4_loop);
4852     ldrw(tmp0, Address(post(buf, 4)));
4853     subs(len, len, 4);
4854     crc32cw(crc, crc, tmp0);
4855     br(Assembler::GE, CRC_by4_loop);
4856     adds(len, len, 4);
4857     br(Assembler::LE, L_exit);
4858   BIND(CRC_by1_loop);
4859     ldrb(tmp0, Address(post(buf, 1)));
4860     subs(len, len, 1);
4861     crc32cb(crc, crc, tmp0);
4862     br(Assembler::GT, CRC_by1_loop);
4863     b(L_exit);
4864 
4865   BIND(CRC_by128_pre);
4866     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4867       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4868     mov(crc, 0);
4869     crc32cx(crc, crc, tmp0);
4870     crc32cx(crc, crc, tmp1);
4871 
4872     cbnz(len, CRC_less128);
4873 
4874   BIND(L_exit);
4875 }
4876 
4877 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4878         Register len, Register tmp0, Register tmp1, Register tmp2,
4879         Register tmp3) {
4880     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4881     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4882 
4883     subs(len, len, 128);
4884     br(Assembler::GE, CRC_by64_pre);
4885   BIND(CRC_less64);
4886     adds(len, len, 128-32);
4887     br(Assembler::GE, CRC_by32_loop);
4888   BIND(CRC_less32);
4889     adds(len, len, 32-4);
4890     br(Assembler::GE, CRC_by4_loop);
4891     adds(len, len, 4);
4892     br(Assembler::GT, CRC_by1_loop);
4893     b(L_exit);
4894 
4895   BIND(CRC_by32_loop);
4896     ldp(tmp0, tmp1, Address(post(buf, 16)));
4897     subs(len, len, 32);
4898     crc32cx(crc, crc, tmp0);
4899     ldr(tmp2, Address(post(buf, 8)));
4900     crc32cx(crc, crc, tmp1);
4901     ldr(tmp3, Address(post(buf, 8)));
4902     crc32cx(crc, crc, tmp2);
4903     crc32cx(crc, crc, tmp3);
4904     br(Assembler::GE, CRC_by32_loop);
4905     cmn(len, (u1)32);
4906     br(Assembler::NE, CRC_less32);
4907     b(L_exit);
4908 
4909   BIND(CRC_by4_loop);
4910     ldrw(tmp0, Address(post(buf, 4)));
4911     subs(len, len, 4);
4912     crc32cw(crc, crc, tmp0);
4913     br(Assembler::GE, CRC_by4_loop);
4914     adds(len, len, 4);
4915     br(Assembler::LE, L_exit);
4916   BIND(CRC_by1_loop);
4917     ldrb(tmp0, Address(post(buf, 1)));
4918     subs(len, len, 1);
4919     crc32cb(crc, crc, tmp0);
4920     br(Assembler::GT, CRC_by1_loop);
4921     b(L_exit);
4922 
4923   BIND(CRC_by64_pre);
4924     sub(buf, buf, 8);
4925     ldp(tmp0, tmp1, Address(buf, 8));
4926     crc32cx(crc, crc, tmp0);
4927     ldr(tmp2, Address(buf, 24));
4928     crc32cx(crc, crc, tmp1);
4929     ldr(tmp3, Address(buf, 32));
4930     crc32cx(crc, crc, tmp2);
4931     ldr(tmp0, Address(buf, 40));
4932     crc32cx(crc, crc, tmp3);
4933     ldr(tmp1, Address(buf, 48));
4934     crc32cx(crc, crc, tmp0);
4935     ldr(tmp2, Address(buf, 56));
4936     crc32cx(crc, crc, tmp1);
4937     ldr(tmp3, Address(pre(buf, 64)));
4938 
4939     b(CRC_by64_loop);
4940 
4941     align(CodeEntryAlignment);
4942   BIND(CRC_by64_loop);
4943     subs(len, len, 64);
4944     crc32cx(crc, crc, tmp2);
4945     ldr(tmp0, Address(buf, 8));
4946     crc32cx(crc, crc, tmp3);
4947     ldr(tmp1, Address(buf, 16));
4948     crc32cx(crc, crc, tmp0);
4949     ldr(tmp2, Address(buf, 24));
4950     crc32cx(crc, crc, tmp1);
4951     ldr(tmp3, Address(buf, 32));
4952     crc32cx(crc, crc, tmp2);
4953     ldr(tmp0, Address(buf, 40));
4954     crc32cx(crc, crc, tmp3);
4955     ldr(tmp1, Address(buf, 48));
4956     crc32cx(crc, crc, tmp0);
4957     ldr(tmp2, Address(buf, 56));
4958     crc32cx(crc, crc, tmp1);
4959     ldr(tmp3, Address(pre(buf, 64)));
4960     br(Assembler::GE, CRC_by64_loop);
4961 
4962     // post-loop
4963     crc32cx(crc, crc, tmp2);
4964     crc32cx(crc, crc, tmp3);
4965 
4966     sub(len, len, 64);
4967     add(buf, buf, 8);
4968     cmn(len, (u1)128);
4969     br(Assembler::NE, CRC_less64);
4970   BIND(L_exit);
4971 }
4972 
4973 /**
4974  * @param crc   register containing existing CRC (32-bit)
4975  * @param buf   register pointing to input byte buffer (byte*)
4976  * @param len   register containing number of bytes
4977  * @param table register that will contain address of CRC table
4978  * @param tmp   scratch register
4979  */
4980 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4981         Register table0, Register table1, Register table2, Register table3,
4982         Register tmp, Register tmp2, Register tmp3) {
4983   if (UseCryptoPmullForCRC32) {
4984     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4985   } else {
4986     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4987   }
4988 }
4989 
4990 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4991         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4992     Label CRC_by128_loop;
4993     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4994 
4995     sub(len, len, 256);
4996     Register table = tmp0;
4997     {
4998       uint64_t offset;
4999       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
5000       add(table, table, offset);
5001     }
5002     add(table, table, table_offset);
5003 
5004     // Registers v0..v7 are used as data registers.
5005     // Registers v16..v31 are used as tmp registers.
5006     sub(buf, buf, 0x10);
5007     ldrq(v0, Address(buf, 0x10));
5008     ldrq(v1, Address(buf, 0x20));
5009     ldrq(v2, Address(buf, 0x30));
5010     ldrq(v3, Address(buf, 0x40));
5011     ldrq(v4, Address(buf, 0x50));
5012     ldrq(v5, Address(buf, 0x60));
5013     ldrq(v6, Address(buf, 0x70));
5014     ldrq(v7, Address(pre(buf, 0x80)));
5015 
5016     movi(v31, T4S, 0);
5017     mov(v31, S, 0, crc);
5018     eor(v0, T16B, v0, v31);
5019 
5020     // Register v16 contains constants from the crc table.
5021     ldrq(v16, Address(table));
5022     b(CRC_by128_loop);
5023 
5024     align(OptoLoopAlignment);
5025   BIND(CRC_by128_loop);
5026     pmull (v17,  T1Q, v0, v16, T1D);
5027     pmull2(v18, T1Q, v0, v16, T2D);
5028     ldrq(v0, Address(buf, 0x10));
5029     eor3(v0, T16B, v17,  v18, v0);
5030 
5031     pmull (v19, T1Q, v1, v16, T1D);
5032     pmull2(v20, T1Q, v1, v16, T2D);
5033     ldrq(v1, Address(buf, 0x20));
5034     eor3(v1, T16B, v19, v20, v1);
5035 
5036     pmull (v21, T1Q, v2, v16, T1D);
5037     pmull2(v22, T1Q, v2, v16, T2D);
5038     ldrq(v2, Address(buf, 0x30));
5039     eor3(v2, T16B, v21, v22, v2);
5040 
5041     pmull (v23, T1Q, v3, v16, T1D);
5042     pmull2(v24, T1Q, v3, v16, T2D);
5043     ldrq(v3, Address(buf, 0x40));
5044     eor3(v3, T16B, v23, v24, v3);
5045 
5046     pmull (v25, T1Q, v4, v16, T1D);
5047     pmull2(v26, T1Q, v4, v16, T2D);
5048     ldrq(v4, Address(buf, 0x50));
5049     eor3(v4, T16B, v25, v26, v4);
5050 
5051     pmull (v27, T1Q, v5, v16, T1D);
5052     pmull2(v28, T1Q, v5, v16, T2D);
5053     ldrq(v5, Address(buf, 0x60));
5054     eor3(v5, T16B, v27, v28, v5);
5055 
5056     pmull (v29, T1Q, v6, v16, T1D);
5057     pmull2(v30, T1Q, v6, v16, T2D);
5058     ldrq(v6, Address(buf, 0x70));
5059     eor3(v6, T16B, v29, v30, v6);
5060 
5061     // Reuse registers v23, v24.
5062     // Using them won't block the first instruction of the next iteration.
5063     pmull (v23, T1Q, v7, v16, T1D);
5064     pmull2(v24, T1Q, v7, v16, T2D);
5065     ldrq(v7, Address(pre(buf, 0x80)));
5066     eor3(v7, T16B, v23, v24, v7);
5067 
5068     subs(len, len, 0x80);
5069     br(Assembler::GE, CRC_by128_loop);
5070 
5071     // fold into 512 bits
5072     // Use v31 for constants because v16 can be still in use.
5073     ldrq(v31, Address(table, 0x10));
5074 
5075     pmull (v17,  T1Q, v0, v31, T1D);
5076     pmull2(v18, T1Q, v0, v31, T2D);
5077     eor3(v0, T16B, v17, v18, v4);
5078 
5079     pmull (v19, T1Q, v1, v31, T1D);
5080     pmull2(v20, T1Q, v1, v31, T2D);
5081     eor3(v1, T16B, v19, v20, v5);
5082 
5083     pmull (v21, T1Q, v2, v31, T1D);
5084     pmull2(v22, T1Q, v2, v31, T2D);
5085     eor3(v2, T16B, v21, v22, v6);
5086 
5087     pmull (v23, T1Q, v3, v31, T1D);
5088     pmull2(v24, T1Q, v3, v31, T2D);
5089     eor3(v3, T16B, v23, v24, v7);
5090 
5091     // fold into 128 bits
5092     // Use v17 for constants because v31 can be still in use.
5093     ldrq(v17, Address(table, 0x20));
5094     pmull (v25, T1Q, v0, v17, T1D);
5095     pmull2(v26, T1Q, v0, v17, T2D);
5096     eor3(v3, T16B, v3, v25, v26);
5097 
5098     // Use v18 for constants because v17 can be still in use.
5099     ldrq(v18, Address(table, 0x30));
5100     pmull (v27, T1Q, v1, v18, T1D);
5101     pmull2(v28, T1Q, v1, v18, T2D);
5102     eor3(v3, T16B, v3, v27, v28);
5103 
5104     // Use v19 for constants because v18 can be still in use.
5105     ldrq(v19, Address(table, 0x40));
5106     pmull (v29, T1Q, v2, v19, T1D);
5107     pmull2(v30, T1Q, v2, v19, T2D);
5108     eor3(v0, T16B, v3, v29, v30);
5109 
5110     add(len, len, 0x80);
5111     add(buf, buf, 0x10);
5112 
5113     mov(tmp0, v0, D, 0);
5114     mov(tmp1, v0, D, 1);
5115 }
5116 
5117 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5118   Address adr;
5119   switch(dst.getMode()) {
5120   case Address::base_plus_offset:
5121     // This is the expected mode, although we allow all the other
5122     // forms below.
5123     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5124     break;
5125   default:
5126     lea(rscratch2, dst);
5127     adr = Address(rscratch2);
5128     break;
5129   }
5130   ldr(rscratch1, adr);
5131   add(rscratch1, rscratch1, src);
5132   str(rscratch1, adr);
5133 }
5134 
5135 void MacroAssembler::cmpptr(Register src1, Address src2) {
5136   uint64_t offset;
5137   adrp(rscratch1, src2, offset);
5138   ldr(rscratch1, Address(rscratch1, offset));
5139   cmp(src1, rscratch1);
5140 }
5141 
5142 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5143   cmp(obj1, obj2);
5144 }
5145 
5146 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5147   load_method_holder(rresult, rmethod);
5148   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5149 }
5150 
5151 void MacroAssembler::load_method_holder(Register holder, Register method) {
5152   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5153   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5154   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5155 }
5156 
5157 void MacroAssembler::load_metadata(Register dst, Register src) {
5158   if (UseCompactObjectHeaders) {
5159     load_narrow_klass_compact(dst, src);
5160   } else {
5161     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5162   }
5163 }
5164 
5165 // Loads the obj's Klass* into dst.
5166 // Preserves all registers (incl src, rscratch1 and rscratch2).
5167 // Input:
5168 // src - the oop we want to load the klass from.
5169 // dst - output narrow klass.
5170 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5171   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5172   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5173   lsr(dst, dst, markWord::klass_shift);
5174 }
5175 
5176 void MacroAssembler::load_klass(Register dst, Register src) {
5177   if (UseCompactObjectHeaders) {
5178     load_narrow_klass_compact(dst, src);
5179   } else {
5180     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5181   }
5182   decode_klass_not_null(dst);
5183 }
5184 
5185 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5186   if (RestoreMXCSROnJNICalls) {
5187     Label OK;
5188     get_fpcr(tmp1);
5189     mov(tmp2, tmp1);
5190     // Set FPCR to the state we need. We do want Round to Nearest. We
5191     // don't want non-IEEE rounding modes or floating-point traps.
5192     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5193     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5194     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5195     eor(tmp2, tmp1, tmp2);
5196     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5197     set_fpcr(tmp1);
5198     bind(OK);
5199   }
5200 }
5201 
5202 // ((OopHandle)result).resolve();
5203 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5204   // OopHandle::resolve is an indirection.
5205   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5206 }
5207 
5208 // ((WeakHandle)result).resolve();
5209 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5210   assert_different_registers(result, tmp1, tmp2);
5211   Label resolved;
5212 
5213   // A null weak handle resolves to null.
5214   cbz(result, resolved);
5215 
5216   // Only 64 bit platforms support GCs that require a tmp register
5217   // WeakHandle::resolve is an indirection like jweak.
5218   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5219                  result, Address(result), tmp1, tmp2);
5220   bind(resolved);
5221 }
5222 
5223 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5224   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5225   ldr(dst, Address(rmethod, Method::const_offset()));
5226   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5227   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5228   ldr(dst, Address(dst, mirror_offset));
5229   resolve_oop_handle(dst, tmp1, tmp2);
5230 }
5231 
5232 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5233   assert_different_registers(obj, klass, tmp);
5234   if (UseCompactObjectHeaders) {
5235     load_narrow_klass_compact(tmp, obj);
5236   } else {
5237     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5238   }
5239   if (CompressedKlassPointers::base() == nullptr) {
5240     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5241     return;
5242   } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5243              && CompressedKlassPointers::shift() == 0) {
5244     // Only the bottom 32 bits matter
5245     cmpw(klass, tmp);
5246     return;
5247   }
5248   decode_klass_not_null(tmp);
5249   cmp(klass, tmp);
5250 }
5251 
5252 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5253   if (UseCompactObjectHeaders) {
5254     load_narrow_klass_compact(tmp1, obj1);
5255     load_narrow_klass_compact(tmp2, obj2);
5256   } else {
5257     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5258     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5259   }
5260   cmpw(tmp1, tmp2);
5261 }
5262 
5263 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5264   load_klass(dst, src);
5265   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5266 }
5267 
5268 void MacroAssembler::store_klass(Register dst, Register src) {
5269   // FIXME: Should this be a store release?  concurrent gcs assumes
5270   // klass length is valid if klass field is not null.
5271   assert(!UseCompactObjectHeaders, "not with compact headers");
5272   encode_klass_not_null(src);
5273   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5274 }
5275 
5276 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5277   assert(!UseCompactObjectHeaders, "not with compact headers");
5278   // Store to klass gap in destination
5279   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5280 }
5281 
5282 // Algorithm must match CompressedOops::encode.
5283 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5284 #ifdef ASSERT
5285   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5286 #endif
5287   verify_oop_msg(s, "broken oop in encode_heap_oop");
5288   if (CompressedOops::base() == nullptr) {
5289     if (CompressedOops::shift() != 0) {
5290       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5291       lsr(d, s, LogMinObjAlignmentInBytes);
5292     } else {
5293       mov(d, s);
5294     }
5295   } else {
5296     subs(d, s, rheapbase);
5297     csel(d, d, zr, Assembler::HS);
5298     lsr(d, d, LogMinObjAlignmentInBytes);
5299 
5300     /*  Old algorithm: is this any worse?
5301     Label nonnull;
5302     cbnz(r, nonnull);
5303     sub(r, r, rheapbase);
5304     bind(nonnull);
5305     lsr(r, r, LogMinObjAlignmentInBytes);
5306     */
5307   }
5308 }
5309 
5310 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5311 #ifdef ASSERT
5312   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5313   if (CheckCompressedOops) {
5314     Label ok;
5315     cbnz(r, ok);
5316     stop("null oop passed to encode_heap_oop_not_null");
5317     bind(ok);
5318   }
5319 #endif
5320   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5321   if (CompressedOops::base() != nullptr) {
5322     sub(r, r, rheapbase);
5323   }
5324   if (CompressedOops::shift() != 0) {
5325     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5326     lsr(r, r, LogMinObjAlignmentInBytes);
5327   }
5328 }
5329 
5330 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5331 #ifdef ASSERT
5332   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5333   if (CheckCompressedOops) {
5334     Label ok;
5335     cbnz(src, ok);
5336     stop("null oop passed to encode_heap_oop_not_null2");
5337     bind(ok);
5338   }
5339 #endif
5340   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5341 
5342   Register data = src;
5343   if (CompressedOops::base() != nullptr) {
5344     sub(dst, src, rheapbase);
5345     data = dst;
5346   }
5347   if (CompressedOops::shift() != 0) {
5348     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5349     lsr(dst, data, LogMinObjAlignmentInBytes);
5350     data = dst;
5351   }
5352   if (data == src)
5353     mov(dst, src);
5354 }
5355 
5356 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5357 #ifdef ASSERT
5358   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5359 #endif
5360   if (CompressedOops::base() == nullptr) {
5361     if (CompressedOops::shift() != 0) {
5362       lsl(d, s, CompressedOops::shift());
5363     } else if (d != s) {
5364       mov(d, s);
5365     }
5366   } else {
5367     Label done;
5368     if (d != s)
5369       mov(d, s);
5370     cbz(s, done);
5371     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5372     bind(done);
5373   }
5374   verify_oop_msg(d, "broken oop in decode_heap_oop");
5375 }
5376 
5377 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5378   assert (UseCompressedOops, "should only be used for compressed headers");
5379   assert (Universe::heap() != nullptr, "java heap should be initialized");
5380   // Cannot assert, unverified entry point counts instructions (see .ad file)
5381   // vtableStubs also counts instructions in pd_code_size_limit.
5382   // Also do not verify_oop as this is called by verify_oop.
5383   if (CompressedOops::shift() != 0) {
5384     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5385     if (CompressedOops::base() != nullptr) {
5386       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5387     } else {
5388       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5389     }
5390   } else {
5391     assert (CompressedOops::base() == nullptr, "sanity");
5392   }
5393 }
5394 
5395 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5396   assert (UseCompressedOops, "should only be used for compressed headers");
5397   assert (Universe::heap() != nullptr, "java heap should be initialized");
5398   // Cannot assert, unverified entry point counts instructions (see .ad file)
5399   // vtableStubs also counts instructions in pd_code_size_limit.
5400   // Also do not verify_oop as this is called by verify_oop.
5401   if (CompressedOops::shift() != 0) {
5402     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5403     if (CompressedOops::base() != nullptr) {
5404       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5405     } else {
5406       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5407     }
5408   } else {
5409     assert (CompressedOops::base() == nullptr, "sanity");
5410     if (dst != src) {
5411       mov(dst, src);
5412     }
5413   }
5414 }
5415 
5416 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5417 
5418 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5419   assert(Metaspace::initialized(), "metaspace not initialized yet");
5420   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5421   return _klass_decode_mode;
5422 }
5423 
5424 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5425   // KlassDecodeMode shouldn't be set already.
5426   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5427 
5428   if (base == nullptr) {
5429     return KlassDecodeZero;
5430   }
5431 
5432   if (operand_valid_for_logical_immediate(
5433         /*is32*/false, (uint64_t)base)) {
5434     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5435     if (((uint64_t)base & range_mask) == 0) {
5436       return KlassDecodeXor;
5437     }
5438   }
5439 
5440   const uint64_t shifted_base =
5441     (uint64_t)base >> shift;
5442   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5443     return KlassDecodeMovk;
5444   }
5445 
5446   // No valid encoding.
5447   return KlassDecodeNone;
5448 }
5449 
5450 // Check if one of the above decoding modes will work for given base, shift and range.
5451 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5452   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5453 }
5454 
5455 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5456   _klass_decode_mode = klass_decode_mode(base, shift, range);
5457   return _klass_decode_mode != KlassDecodeNone;
5458 }
5459 
5460 static Register pick_different_tmp(Register dst, Register src) {
5461   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5462   return *tmps.begin();
5463 }
5464 
5465 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5466   // we have to load the klass base from the AOT constants area but
5467   // not the shift because it is not allowed to change
5468   int shift = CompressedKlassPointers::shift();
5469   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5470   if (dst != src) {
5471     // we can load the base into dst, subtract it formthe src and shift down
5472     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5473     ldr(dst, dst);
5474     sub(dst, src, dst);
5475     lsr(dst, dst, shift);
5476   } else {
5477     // we need an extra register in order to load the coop base
5478     Register tmp = pick_different_tmp(dst, src);
5479     RegSet regs = RegSet::of(tmp);
5480     push(regs, sp);
5481     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5482     ldr(tmp, tmp);
5483     sub(dst, src, tmp);
5484     lsr(dst, dst, shift);
5485     pop(regs, sp);
5486   }
5487 }
5488 
5489 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5490   if (AOTCodeCache::is_on_for_dump()) {
5491     encode_klass_not_null_for_aot(dst, src);
5492     return;
5493   }
5494 
5495   switch (klass_decode_mode()) {
5496   case KlassDecodeZero:
5497     if (CompressedKlassPointers::shift() != 0) {
5498       lsr(dst, src, CompressedKlassPointers::shift());
5499     } else {
5500       if (dst != src) mov(dst, src);
5501     }
5502     break;
5503 
5504   case KlassDecodeXor:
5505     if (CompressedKlassPointers::shift() != 0) {
5506       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5507       lsr(dst, dst, CompressedKlassPointers::shift());
5508     } else {
5509       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5510     }
5511     break;
5512 
5513   case KlassDecodeMovk:
5514     if (CompressedKlassPointers::shift() != 0) {
5515       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5516     } else {
5517       movw(dst, src);
5518     }
5519     break;
5520 
5521   case KlassDecodeNone:
5522     ShouldNotReachHere();
5523     break;
5524   }
5525 }
5526 
5527 void MacroAssembler::encode_klass_not_null(Register r) {
5528   encode_klass_not_null(r, r);
5529 }
5530 
5531 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5532   // we have to load the klass base from the AOT constants area but
5533   // not the shift because it is not allowed to change
5534   int shift = CompressedKlassPointers::shift();
5535   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5536   if (dst != src) {
5537     // we can load the base into dst then add the offset with a suitable shift
5538     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5539     ldr(dst, dst);
5540     add(dst, dst, src, LSL,  shift);
5541   } else {
5542     // we need an extra register in order to load the coop base
5543     Register tmp = pick_different_tmp(dst, src);
5544     RegSet regs = RegSet::of(tmp);
5545     push(regs, sp);
5546     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5547     ldr(tmp, tmp);
5548     add(dst, tmp,  src, LSL,  shift);
5549     pop(regs, sp);
5550   }
5551 }
5552 
5553 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5554   if (AOTCodeCache::is_on_for_dump()) {
5555     decode_klass_not_null_for_aot(dst, src);
5556     return;
5557   }
5558 
5559   switch (klass_decode_mode()) {
5560   case KlassDecodeZero:
5561     if (CompressedKlassPointers::shift() != 0) {
5562       lsl(dst, src, CompressedKlassPointers::shift());
5563     } else {
5564       if (dst != src) mov(dst, src);
5565     }
5566     break;
5567 
5568   case KlassDecodeXor:
5569     if (CompressedKlassPointers::shift() != 0) {
5570       lsl(dst, src, CompressedKlassPointers::shift());
5571       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5572     } else {
5573       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5574     }
5575     break;
5576 
5577   case KlassDecodeMovk: {
5578     const uint64_t shifted_base =
5579       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5580 
5581     if (dst != src) movw(dst, src);
5582     movk(dst, shifted_base >> 32, 32);
5583 
5584     if (CompressedKlassPointers::shift() != 0) {
5585       lsl(dst, dst, CompressedKlassPointers::shift());
5586     }
5587 
5588     break;
5589   }
5590 
5591   case KlassDecodeNone:
5592     ShouldNotReachHere();
5593     break;
5594   }
5595 }
5596 
5597 void  MacroAssembler::decode_klass_not_null(Register r) {
5598   decode_klass_not_null(r, r);
5599 }
5600 
5601 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5602 #ifdef ASSERT
5603   {
5604     ThreadInVMfromUnknown tiv;
5605     assert (UseCompressedOops, "should only be used for compressed oops");
5606     assert (Universe::heap() != nullptr, "java heap should be initialized");
5607     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5608     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5609   }
5610 #endif
5611   int oop_index = oop_recorder()->find_index(obj);
5612   InstructionMark im(this);
5613   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5614   code_section()->relocate(inst_mark(), rspec);
5615   movz(dst, 0xDEAD, 16);
5616   movk(dst, 0xBEEF);
5617 }
5618 
5619 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5620   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5621   int index = oop_recorder()->find_index(k);
5622 
5623   InstructionMark im(this);
5624   RelocationHolder rspec = metadata_Relocation::spec(index);
5625   code_section()->relocate(inst_mark(), rspec);
5626   narrowKlass nk = CompressedKlassPointers::encode(k);
5627   movz(dst, (nk >> 16), 16);
5628   movk(dst, nk & 0xffff);
5629 }
5630 
5631 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5632                                     Register dst, Address src,
5633                                     Register tmp1, Register tmp2) {
5634   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5635   decorators = AccessInternal::decorator_fixup(decorators, type);
5636   bool as_raw = (decorators & AS_RAW) != 0;
5637   if (as_raw) {
5638     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5639   } else {
5640     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5641   }
5642 }
5643 
5644 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5645                                      Address dst, Register val,
5646                                      Register tmp1, Register tmp2, Register tmp3) {
5647   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5648   decorators = AccessInternal::decorator_fixup(decorators, type);
5649   bool as_raw = (decorators & AS_RAW) != 0;
5650   if (as_raw) {
5651     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5652   } else {
5653     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5654   }
5655 }
5656 
5657 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5658                                      Register inline_layout_info) {
5659   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5660   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5661 }
5662 
5663 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5664   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5665   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5666 }
5667 
5668 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5669   // ((address) (void*) o) + vk->payload_offset();
5670   Register offset = (data == oop) ? rscratch1 : data;
5671   payload_offset(inline_klass, offset);
5672   if (data == oop) {
5673     add(data, data, offset);
5674   } else {
5675     lea(data, Address(oop, offset));
5676   }
5677 }
5678 
5679 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
5680                                                 Register index, Register data) {
5681   assert_different_registers(array, array_klass, index);
5682   assert_different_registers(rscratch1, array, index);
5683 
5684   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
5685   ldrw(rscratch1, Address(array_klass, Klass::layout_helper_offset()));
5686 
5687   // Klass::layout_helper_log2_element_size(lh)
5688   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
5689   lsr(rscratch1, rscratch1, Klass::_lh_log2_element_size_shift);
5690   andr(rscratch1, rscratch1, Klass::_lh_log2_element_size_mask);
5691   lslv(index, index, rscratch1);
5692 
5693   add(data, array, index);
5694   add(data, data, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT));
5695 }
5696 
5697 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5698                                    Register tmp2, DecoratorSet decorators) {
5699   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5700 }
5701 
5702 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5703                                             Register tmp2, DecoratorSet decorators) {
5704   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5705 }
5706 
5707 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5708                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5709   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5710 }
5711 
5712 // Used for storing nulls.
5713 void MacroAssembler::store_heap_oop_null(Address dst) {
5714   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5715 }
5716 
5717 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5718   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5719   int index = oop_recorder()->allocate_metadata_index(obj);
5720   RelocationHolder rspec = metadata_Relocation::spec(index);
5721   return Address((address)obj, rspec);
5722 }
5723 
5724 // Move an oop into a register.
5725 void MacroAssembler::movoop(Register dst, jobject obj) {
5726   int oop_index;
5727   if (obj == nullptr) {
5728     oop_index = oop_recorder()->allocate_oop_index(obj);
5729   } else {
5730 #ifdef ASSERT
5731     {
5732       ThreadInVMfromUnknown tiv;
5733       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5734     }
5735 #endif
5736     oop_index = oop_recorder()->find_index(obj);
5737   }
5738   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5739 
5740   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5741     mov(dst, Address((address)obj, rspec));
5742   } else {
5743     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5744     ldr(dst, Address(dummy, rspec));
5745   }
5746 }
5747 
5748 // Move a metadata address into a register.
5749 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5750   int oop_index;
5751   if (obj == nullptr) {
5752     oop_index = oop_recorder()->allocate_metadata_index(obj);
5753   } else {
5754     oop_index = oop_recorder()->find_index(obj);
5755   }
5756   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5757   mov(dst, Address((address)obj, rspec));
5758 }
5759 
5760 Address MacroAssembler::constant_oop_address(jobject obj) {
5761 #ifdef ASSERT
5762   {
5763     ThreadInVMfromUnknown tiv;
5764     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5765     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5766   }
5767 #endif
5768   int oop_index = oop_recorder()->find_index(obj);
5769   return Address((address)obj, oop_Relocation::spec(oop_index));
5770 }
5771 
5772 // Object / value buffer allocation...
5773 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
5774                                        Register t1, Register t2,
5775                                        bool clear_fields, Label& alloc_failed)
5776 {
5777   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
5778   Register layout_size = t1;
5779   assert(new_obj == r0, "needs to be r0");
5780   assert_different_registers(klass, new_obj, t1, t2);
5781 
5782   // get instance_size in InstanceKlass (scaled to a count of bytes)
5783   ldrw(layout_size, Address(klass, Klass::layout_helper_offset()));
5784   // test to see if it is malformed in some way
5785   tst(layout_size, Klass::_lh_instance_slow_path_bit);
5786   br(Assembler::NE, slow_case_no_pop);
5787 
5788   // Allocate the instance:
5789   //  If TLAB is enabled:
5790   //    Try to allocate in the TLAB.
5791   //    If fails, go to the slow path.
5792   //    Initialize the allocation.
5793   //    Exit.
5794   //
5795   //  Go to slow path.
5796 
5797   if (UseTLAB) {
5798     push(klass);
5799     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
5800     if (ZeroTLAB || (!clear_fields)) {
5801       // the fields have been already cleared
5802       b(initialize_header);
5803     } else {
5804       // initialize both the header and fields
5805       b(initialize_object);
5806     }
5807 
5808     if (clear_fields) {
5809       // The object is initialized before the header.  If the object size is
5810       // zero, go directly to the header initialization.
5811       bind(initialize_object);
5812       int header_size = oopDesc::header_size() * HeapWordSize;
5813       assert(is_aligned(header_size, BytesPerLong), "oop header size must be 8-byte-aligned");
5814       subs(layout_size, layout_size, header_size);
5815       br(Assembler::EQ, initialize_header);
5816 
5817       // Initialize topmost object field, divide size by 8, check if odd and
5818       // test if zero.
5819 
5820   #ifdef ASSERT
5821       // make sure instance_size was multiple of 8
5822       Label L;
5823       tst(layout_size, 7);
5824       br(Assembler::EQ, L);
5825       stop("object size is not multiple of 8 - adjust this code");
5826       bind(L);
5827       // must be > 0, no extra check needed here
5828   #endif
5829 
5830       lsr(layout_size, layout_size, LogBytesPerLong);
5831 
5832       // initialize remaining object fields: instance_size was a multiple of 8
5833       {
5834         Label loop;
5835         Register base = t2;
5836 
5837         bind(loop);
5838         add(rscratch1, new_obj, layout_size, Assembler::LSL, LogBytesPerLong);
5839         str(zr, Address(rscratch1, header_size - 1*oopSize));
5840         subs(layout_size, layout_size, 1);
5841         br(Assembler::NE, loop);
5842       }
5843     } // clear_fields
5844 
5845     // initialize object header only.
5846     bind(initialize_header);
5847     pop(klass);
5848     Register mark_word = t2;
5849     if (UseCompactObjectHeaders || Arguments::is_valhalla_enabled()) {
5850       ldr(mark_word, Address(klass, Klass::prototype_header_offset()));
5851       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5852     } else {
5853       mov(mark_word, (intptr_t)markWord::prototype().value());
5854       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5855     }
5856     if (!UseCompactObjectHeaders) {
5857       store_klass_gap(new_obj, zr);  // zero klass gap for compressed oops
5858       mov(t2, klass);                // preserve klass
5859       store_klass(new_obj, t2);      // src klass reg is potentially compressed
5860     }
5861     b(done);
5862   }
5863 
5864   if (UseTLAB) {
5865     bind(slow_case);
5866     pop(klass);
5867   }
5868   bind(slow_case_no_pop);
5869   b(alloc_failed);
5870 
5871   bind(done);
5872 }
5873 
5874 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5875 void MacroAssembler::tlab_allocate(Register obj,
5876                                    Register var_size_in_bytes,
5877                                    int con_size_in_bytes,
5878                                    Register t1,
5879                                    Register t2,
5880                                    Label& slow_case) {
5881   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5882   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5883 }
5884 
5885 void MacroAssembler::verify_tlab() {
5886 #ifdef ASSERT
5887   if (UseTLAB && VerifyOops) {
5888     Label next, ok;
5889 
5890     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5891 
5892     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5893     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5894     cmp(rscratch2, rscratch1);
5895     br(Assembler::HS, next);
5896     STOP("assert(top >= start)");
5897     should_not_reach_here();
5898 
5899     bind(next);
5900     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5901     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5902     cmp(rscratch2, rscratch1);
5903     br(Assembler::HS, ok);
5904     STOP("assert(top <= end)");
5905     should_not_reach_here();
5906 
5907     bind(ok);
5908     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5909   }
5910 #endif
5911 }
5912 
5913 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5914   assert_different_registers(holder_klass, index, layout_info);
5915   InlineLayoutInfo array[2];
5916   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5917   if (is_power_of_2(size)) {
5918     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5919   } else {
5920     mov(layout_info, size);
5921     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5922   }
5923   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5924   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5925   lea(layout_info, Address(layout_info, index));
5926 }
5927 
5928 // Writes to stack successive pages until offset reached to check for
5929 // stack overflow + shadow pages.  This clobbers tmp.
5930 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5931   assert_different_registers(tmp, size, rscratch1);
5932   mov(tmp, sp);
5933   // Bang stack for total size given plus shadow page size.
5934   // Bang one page at a time because large size can bang beyond yellow and
5935   // red zones.
5936   Label loop;
5937   mov(rscratch1, (int)os::vm_page_size());
5938   bind(loop);
5939   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5940   subsw(size, size, rscratch1);
5941   str(size, Address(tmp));
5942   br(Assembler::GT, loop);
5943 
5944   // Bang down shadow pages too.
5945   // At this point, (tmp-0) is the last address touched, so don't
5946   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5947   // was post-decremented.)  Skip this address by starting at i=1, and
5948   // touch a few more pages below.  N.B.  It is important to touch all
5949   // the way down to and including i=StackShadowPages.
5950   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5951     // this could be any sized move but this is can be a debugging crumb
5952     // so the bigger the better.
5953     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5954     str(size, Address(tmp));
5955   }
5956 }
5957 
5958 // Move the address of the polling page into dest.
5959 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5960   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5961 }
5962 
5963 // Read the polling page.  The address of the polling page must
5964 // already be in r.
5965 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5966   address mark;
5967   {
5968     InstructionMark im(this);
5969     code_section()->relocate(inst_mark(), rtype);
5970     ldrw(zr, Address(r, 0));
5971     mark = inst_mark();
5972   }
5973   verify_cross_modify_fence_not_required();
5974   return mark;
5975 }
5976 
5977 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5978   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5979   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5980   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5981   int64_t offset_low = dest_page - low_page;
5982   int64_t offset_high = dest_page - high_page;
5983 
5984   assert(is_valid_AArch64_address(dest.target()), "bad address");
5985   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5986 
5987   InstructionMark im(this);
5988   code_section()->relocate(inst_mark(), dest.rspec());
5989   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5990   // the code cache so that if it is relocated we know it will still reach
5991   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5992     _adrp(reg1, dest.target());
5993   } else {
5994     uint64_t target = (uint64_t)dest.target();
5995     uint64_t adrp_target
5996       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5997 
5998     _adrp(reg1, (address)adrp_target);
5999     movk(reg1, target >> 32, 32);
6000   }
6001   byte_offset = (uint64_t)dest.target() & 0xfff;
6002 }
6003 
6004 void MacroAssembler::load_byte_map_base(Register reg) {
6005 #if INCLUDE_CDS
6006   if (AOTCodeCache::is_on_for_dump()) {
6007     address byte_map_base_adr = AOTRuntimeConstants::card_table_base_address();
6008     lea(reg, ExternalAddress(byte_map_base_adr));
6009     ldr(reg, Address(reg));
6010     return;
6011   }
6012 #endif
6013   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
6014 
6015   // Strictly speaking the card table base isn't an address at all, and it might
6016   // even be negative. It is thus materialised as a constant.
6017   mov(reg, (uint64_t)ctbs->card_table_base_const());
6018 }
6019 
6020 void MacroAssembler::load_aotrc_address(Register reg, address a) {
6021 #if INCLUDE_CDS
6022   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
6023   if (AOTCodeCache::is_on_for_dump()) {
6024     // all aotrc field addresses should be registered in the AOTCodeCache address table
6025     lea(reg, ExternalAddress(a));
6026   } else {
6027     mov(reg, (uint64_t)a);
6028   }
6029 #else
6030   ShouldNotReachHere();
6031 #endif
6032 }
6033 
6034 #ifdef ASSERT
6035 void MacroAssembler::build_frame(int framesize) {
6036   build_frame(framesize, false);
6037 }
6038 #endif
6039 
6040 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
6041   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6042   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6043   protect_return_address();
6044   if (framesize < ((1 << 9) + 2 * wordSize)) {
6045     sub(sp, sp, framesize);
6046     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
6047       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
6048       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
6049     } else {
6050       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6051     }
6052     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
6053   } else {
6054     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
6055       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
6056       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
6057     } else {
6058       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6059     }
6060     if (PreserveFramePointer) mov(rfp, sp);
6061     if (framesize < ((1 << 12) + 2 * wordSize))
6062       sub(sp, sp, framesize - 2 * wordSize);
6063     else {
6064       mov(rscratch1, framesize - 2 * wordSize);
6065       sub(sp, sp, rscratch1);
6066     }
6067   }
6068   verify_cross_modify_fence_not_required();
6069 }
6070 
6071 void MacroAssembler::remove_frame(int framesize) {
6072   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6073   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6074   if (framesize < ((1 << 9) + 2 * wordSize)) {
6075     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6076     add(sp, sp, framesize);
6077   } else {
6078     if (framesize < ((1 << 12) + 2 * wordSize))
6079       add(sp, sp, framesize - 2 * wordSize);
6080     else {
6081       mov(rscratch1, framesize - 2 * wordSize);
6082       add(sp, sp, rscratch1);
6083     }
6084     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6085   }
6086   authenticate_return_address();
6087 }
6088 
6089 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
6090   if (needs_stack_repair) {
6091     // The method has a scalarized entry point (where fields of value object arguments
6092     // are passed through registers and stack), and a non-scalarized entry point (where
6093     // value object arguments are given as oops). The non-scalarized entry point will
6094     // first load each field of value object arguments and store them in registers and on
6095     // the stack in a way compatible with the scalarized entry point. To do so, some extra
6096     // stack space might be reserved (if argument registers are not enough). On leaving the
6097     // method, this space must be freed.
6098     //
6099     // In case we used the non-scalarized entry point the stack looks like this:
6100     //
6101     // | Arguments from caller     |
6102     // |---------------------------|  <-- caller's SP
6103     // | Saved LR #1               |
6104     // | Saved FP #1               |
6105     // |---------------------------|
6106     // | Extension space for       |
6107     // |   inline arg (un)packing  |
6108     // |---------------------------|  <-- start of this method's frame
6109     // | Saved LR #2               |
6110     // | Saved FP #2               |
6111     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6112     // | sp_inc                    |
6113     // | method locals             |
6114     // |---------------------------|  <-- SP
6115     //
6116     // There are two copies of FP and LR on the stack. They will be identical at
6117     // first, but that can change.
6118     // If the caller has been deoptimized, LR #1 will be patched to point at the
6119     // deopt blob, and LR #2 will still point into the old method.
6120     // If the saved FP (x29) was not used as the frame pointer, but to store an
6121     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
6122     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
6123     // needed to add space between the extension space and the locals, as there
6124     // would be between the real arguments and the locals if we don't need to
6125     // do unpacking (from the scalarized entry point).
6126     //
6127     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
6128     // while keeping in mind that from the scalarized entry point, there will be
6129     // only one copy of each. Indeed, in the case we used the scalarized calling
6130     // convention, the stack looks like this:
6131     //
6132     // | Arguments from caller     |
6133     // |---------------------------|  <-- caller's SP / start of this method's frame
6134     // | Saved LR                  |
6135     // | Saved FP                  |
6136     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6137     // | sp_inc                    |
6138     // | method locals             |
6139     // |---------------------------|  <-- SP
6140     //
6141     // The sp_inc stack slot holds the total size of the frame including the
6142     // extension space minus two words for the saved FP and LR. That is how to
6143     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
6144     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
6145     //
6146     // One can find sp_inc since the start the method's frame is SP + initial_framesize.
6147 
6148     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6149 
6150     ldr(rscratch1, Address(sp, sp_inc_offset));
6151     add(sp, sp, rscratch1);
6152     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6153   } else {
6154     remove_frame(initial_framesize);
6155   }
6156 }
6157 
6158 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6159   int real_frame_size = frame_size + sp_inc;
6160   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6161   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6162   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6163 
6164   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6165 
6166   // Subtract two words for the saved FP and LR as these will be popped
6167   // separately. See remove_frame above.
6168   mov(rscratch1, real_frame_size - 2*wordSize);
6169   str(rscratch1, Address(sp, sp_inc_offset));
6170 }
6171 
6172 // This method counts leading positive bytes (highest bit not set) in provided byte array
6173 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6174     // Simple and most common case of aligned small array which is not at the
6175     // end of memory page is placed here. All other cases are in stub.
6176     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6177     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6178     assert_different_registers(ary1, len, result);
6179 
6180     mov(result, len);
6181     cmpw(len, 0);
6182     br(LE, DONE);
6183     cmpw(len, 4 * wordSize);
6184     br(GE, STUB_LONG); // size > 32 then go to stub
6185 
6186     int shift = 64 - exact_log2(os::vm_page_size());
6187     lsl(rscratch1, ary1, shift);
6188     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6189     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6190     br(CS, STUB); // at the end of page then go to stub
6191     subs(len, len, wordSize);
6192     br(LT, END);
6193 
6194   BIND(LOOP);
6195     ldr(rscratch1, Address(post(ary1, wordSize)));
6196     tst(rscratch1, UPPER_BIT_MASK);
6197     br(NE, SET_RESULT);
6198     subs(len, len, wordSize);
6199     br(GE, LOOP);
6200     cmpw(len, -wordSize);
6201     br(EQ, DONE);
6202 
6203   BIND(END);
6204     ldr(rscratch1, Address(ary1));
6205     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
6206     lslv(rscratch1, rscratch1, rscratch2);
6207     tst(rscratch1, UPPER_BIT_MASK);
6208     br(NE, SET_RESULT);
6209     b(DONE);
6210 
6211   BIND(STUB);
6212     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
6213     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
6214     address tpc1 = trampoline_call(count_pos);
6215     if (tpc1 == nullptr) {
6216       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
6217       postcond(pc() == badAddress);
6218       return nullptr;
6219     }
6220     b(DONE);
6221 
6222   BIND(STUB_LONG);
6223     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
6224     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
6225     address tpc2 = trampoline_call(count_pos_long);
6226     if (tpc2 == nullptr) {
6227       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
6228       postcond(pc() == badAddress);
6229       return nullptr;
6230     }
6231     b(DONE);
6232 
6233   BIND(SET_RESULT);
6234 
6235     add(len, len, wordSize);
6236     sub(result, result, len);
6237 
6238   BIND(DONE);
6239   postcond(pc() != badAddress);
6240   return pc();
6241 }
6242 
6243 // Clobbers: rscratch1, rscratch2, rflags
6244 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
6245 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
6246                                       Register tmp4, Register tmp5, Register result,
6247                                       Register cnt1, int elem_size) {
6248   Label DONE, SAME;
6249   Register tmp1 = rscratch1;
6250   Register tmp2 = rscratch2;
6251   int elem_per_word = wordSize/elem_size;
6252   int log_elem_size = exact_log2(elem_size);
6253   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
6254   int length_offset = arrayOopDesc::length_offset_in_bytes();
6255   int base_offset
6256     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
6257   // When the length offset is not aligned to 8 bytes,
6258   // then we align it down. This is valid because the new
6259   // offset will always be the klass which is the same
6260   // for type arrays.
6261   int start_offset = align_down(length_offset, BytesPerWord);
6262   int extra_length = base_offset - start_offset;
6263   assert(start_offset == length_offset || start_offset == klass_offset,
6264          "start offset must be 8-byte-aligned or be the klass offset");
6265   assert(base_offset != start_offset, "must include the length field");
6266   extra_length = extra_length / elem_size; // We count in elements, not bytes.
6267   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
6268 
6269   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
6270   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6271 
6272 #ifndef PRODUCT
6273   {
6274     const char kind = (elem_size == 2) ? 'U' : 'L';
6275     char comment[64];
6276     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
6277     BLOCK_COMMENT(comment);
6278   }
6279 #endif
6280 
6281   // if (a1 == a2)
6282   //     return true;
6283   cmpoop(a1, a2); // May have read barriers for a1 and a2.
6284   br(EQ, SAME);
6285 
6286   if (UseSimpleArrayEquals) {
6287     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
6288     // if (a1 == nullptr || a2 == nullptr)
6289     //     return false;
6290     // a1 & a2 == 0 means (some-pointer is null) or
6291     // (very-rare-or-even-probably-impossible-pointer-values)
6292     // so, we can save one branch in most cases
6293     tst(a1, a2);
6294     mov(result, false);
6295     br(EQ, A_MIGHT_BE_NULL);
6296     // if (a1.length != a2.length)
6297     //      return false;
6298     bind(A_IS_NOT_NULL);
6299     ldrw(cnt1, Address(a1, length_offset));
6300     ldrw(tmp5, Address(a2, length_offset));
6301     cmp(cnt1, tmp5);
6302     br(NE, DONE); // If lengths differ, return false
6303     // Increase loop counter by diff between base- and actual start-offset.
6304     addw(cnt1, cnt1, extra_length);
6305     lea(a1, Address(a1, start_offset));
6306     lea(a2, Address(a2, start_offset));
6307     // Check for short strings, i.e. smaller than wordSize.
6308     subs(cnt1, cnt1, elem_per_word);
6309     br(Assembler::LT, SHORT);
6310     // Main 8 byte comparison loop.
6311     bind(NEXT_WORD); {
6312       ldr(tmp1, Address(post(a1, wordSize)));
6313       ldr(tmp2, Address(post(a2, wordSize)));
6314       subs(cnt1, cnt1, elem_per_word);
6315       eor(tmp5, tmp1, tmp2);
6316       cbnz(tmp5, DONE);
6317     } br(GT, NEXT_WORD);
6318     // Last longword.  In the case where length == 4 we compare the
6319     // same longword twice, but that's still faster than another
6320     // conditional branch.
6321     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6322     // length == 4.
6323     if (log_elem_size > 0)
6324       lsl(cnt1, cnt1, log_elem_size);
6325     ldr(tmp3, Address(a1, cnt1));
6326     ldr(tmp4, Address(a2, cnt1));
6327     eor(tmp5, tmp3, tmp4);
6328     cbnz(tmp5, DONE);
6329     b(SAME);
6330     bind(A_MIGHT_BE_NULL);
6331     // in case both a1 and a2 are not-null, proceed with loads
6332     cbz(a1, DONE);
6333     cbz(a2, DONE);
6334     b(A_IS_NOT_NULL);
6335     bind(SHORT);
6336 
6337     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
6338     {
6339       ldrw(tmp1, Address(post(a1, 4)));
6340       ldrw(tmp2, Address(post(a2, 4)));
6341       eorw(tmp5, tmp1, tmp2);
6342       cbnzw(tmp5, DONE);
6343     }
6344     bind(TAIL03);
6345     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6346     {
6347       ldrh(tmp3, Address(post(a1, 2)));
6348       ldrh(tmp4, Address(post(a2, 2)));
6349       eorw(tmp5, tmp3, tmp4);
6350       cbnzw(tmp5, DONE);
6351     }
6352     bind(TAIL01);
6353     if (elem_size == 1) { // Only needed when comparing byte arrays.
6354       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6355       {
6356         ldrb(tmp1, a1);
6357         ldrb(tmp2, a2);
6358         eorw(tmp5, tmp1, tmp2);
6359         cbnzw(tmp5, DONE);
6360       }
6361     }
6362   } else {
6363     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6364         CSET_EQ, LAST_CHECK;
6365     mov(result, false);
6366     cbz(a1, DONE);
6367     ldrw(cnt1, Address(a1, length_offset));
6368     cbz(a2, DONE);
6369     ldrw(tmp5, Address(a2, length_offset));
6370     cmp(cnt1, tmp5);
6371     br(NE, DONE); // If lengths differ, return false
6372     // Increase loop counter by diff between base- and actual start-offset.
6373     addw(cnt1, cnt1, extra_length);
6374 
6375     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6376     // faster to perform another branch before comparing a1 and a2
6377     cmp(cnt1, (u1)elem_per_word);
6378     br(LE, SHORT); // short or same
6379     ldr(tmp3, Address(pre(a1, start_offset)));
6380     subs(zr, cnt1, stubBytesThreshold);
6381     br(GE, STUB);
6382     ldr(tmp4, Address(pre(a2, start_offset)));
6383     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6384 
6385     // Main 16 byte comparison loop with 2 exits
6386     bind(NEXT_DWORD); {
6387       ldr(tmp1, Address(pre(a1, wordSize)));
6388       ldr(tmp2, Address(pre(a2, wordSize)));
6389       subs(cnt1, cnt1, 2 * elem_per_word);
6390       br(LE, TAIL);
6391       eor(tmp4, tmp3, tmp4);
6392       cbnz(tmp4, DONE);
6393       ldr(tmp3, Address(pre(a1, wordSize)));
6394       ldr(tmp4, Address(pre(a2, wordSize)));
6395       cmp(cnt1, (u1)elem_per_word);
6396       br(LE, TAIL2);
6397       cmp(tmp1, tmp2);
6398     } br(EQ, NEXT_DWORD);
6399     b(DONE);
6400 
6401     bind(TAIL);
6402     eor(tmp4, tmp3, tmp4);
6403     eor(tmp2, tmp1, tmp2);
6404     lslv(tmp2, tmp2, tmp5);
6405     orr(tmp5, tmp4, tmp2);
6406     cmp(tmp5, zr);
6407     b(CSET_EQ);
6408 
6409     bind(TAIL2);
6410     eor(tmp2, tmp1, tmp2);
6411     cbnz(tmp2, DONE);
6412     b(LAST_CHECK);
6413 
6414     bind(STUB);
6415     ldr(tmp4, Address(pre(a2, start_offset)));
6416     if (elem_size == 2) { // convert to byte counter
6417       lsl(cnt1, cnt1, 1);
6418     }
6419     eor(tmp5, tmp3, tmp4);
6420     cbnz(tmp5, DONE);
6421     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6422     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6423     address tpc = trampoline_call(stub);
6424     if (tpc == nullptr) {
6425       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6426       postcond(pc() == badAddress);
6427       return nullptr;
6428     }
6429     b(DONE);
6430 
6431     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6432     // so, if a2 == null => return false(0), else return true, so we can return a2
6433     mov(result, a2);
6434     b(DONE);
6435     bind(SHORT);
6436     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6437     ldr(tmp3, Address(a1, start_offset));
6438     ldr(tmp4, Address(a2, start_offset));
6439     bind(LAST_CHECK);
6440     eor(tmp4, tmp3, tmp4);
6441     lslv(tmp5, tmp4, tmp5);
6442     cmp(tmp5, zr);
6443     bind(CSET_EQ);
6444     cset(result, EQ);
6445     b(DONE);
6446   }
6447 
6448   bind(SAME);
6449   mov(result, true);
6450   // That's it.
6451   bind(DONE);
6452 
6453   BLOCK_COMMENT("} array_equals");
6454   postcond(pc() != badAddress);
6455   return pc();
6456 }
6457 
6458 // Compare Strings
6459 
6460 // For Strings we're passed the address of the first characters in a1
6461 // and a2 and the length in cnt1.
6462 // There are two implementations.  For arrays >= 8 bytes, all
6463 // comparisons (including the final one, which may overlap) are
6464 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6465 // halfword, then a short, and then a byte.
6466 
6467 void MacroAssembler::string_equals(Register a1, Register a2,
6468                                    Register result, Register cnt1)
6469 {
6470   Label SAME, DONE, SHORT, NEXT_WORD;
6471   Register tmp1 = rscratch1;
6472   Register tmp2 = rscratch2;
6473 
6474   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6475 
6476 #ifndef PRODUCT
6477   {
6478     char comment[64];
6479     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6480     BLOCK_COMMENT(comment);
6481   }
6482 #endif
6483 
6484   mov(result, false);
6485 
6486   // Check for short strings, i.e. smaller than wordSize.
6487   subs(cnt1, cnt1, wordSize);
6488   br(Assembler::LT, SHORT);
6489   // Main 8 byte comparison loop.
6490   bind(NEXT_WORD); {
6491     ldr(tmp1, Address(post(a1, wordSize)));
6492     ldr(tmp2, Address(post(a2, wordSize)));
6493     subs(cnt1, cnt1, wordSize);
6494     eor(tmp1, tmp1, tmp2);
6495     cbnz(tmp1, DONE);
6496   } br(GT, NEXT_WORD);
6497   // Last longword.  In the case where length == 4 we compare the
6498   // same longword twice, but that's still faster than another
6499   // conditional branch.
6500   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6501   // length == 4.
6502   ldr(tmp1, Address(a1, cnt1));
6503   ldr(tmp2, Address(a2, cnt1));
6504   eor(tmp2, tmp1, tmp2);
6505   cbnz(tmp2, DONE);
6506   b(SAME);
6507 
6508   bind(SHORT);
6509   Label TAIL03, TAIL01;
6510 
6511   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6512   {
6513     ldrw(tmp1, Address(post(a1, 4)));
6514     ldrw(tmp2, Address(post(a2, 4)));
6515     eorw(tmp1, tmp1, tmp2);
6516     cbnzw(tmp1, DONE);
6517   }
6518   bind(TAIL03);
6519   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6520   {
6521     ldrh(tmp1, Address(post(a1, 2)));
6522     ldrh(tmp2, Address(post(a2, 2)));
6523     eorw(tmp1, tmp1, tmp2);
6524     cbnzw(tmp1, DONE);
6525   }
6526   bind(TAIL01);
6527   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6528     {
6529     ldrb(tmp1, a1);
6530     ldrb(tmp2, a2);
6531     eorw(tmp1, tmp1, tmp2);
6532     cbnzw(tmp1, DONE);
6533   }
6534   // Arrays are equal.
6535   bind(SAME);
6536   mov(result, true);
6537 
6538   // That's it.
6539   bind(DONE);
6540   BLOCK_COMMENT("} string_equals");
6541 }
6542 
6543 
6544 // The size of the blocks erased by the zero_blocks stub.  We must
6545 // handle anything smaller than this ourselves in zero_words().
6546 const int MacroAssembler::zero_words_block_size = 8;
6547 
6548 // zero_words() is used by C2 ClearArray patterns and by
6549 // C1_MacroAssembler.  It is as small as possible, handling small word
6550 // counts locally and delegating anything larger to the zero_blocks
6551 // stub.  It is expanded many times in compiled code, so it is
6552 // important to keep it short.
6553 
6554 // ptr:   Address of a buffer to be zeroed.
6555 // cnt:   Count in HeapWords.
6556 //
6557 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6558 address MacroAssembler::zero_words(Register ptr, Register cnt)
6559 {
6560   assert(is_power_of_2(zero_words_block_size), "adjust this");
6561 
6562   BLOCK_COMMENT("zero_words {");
6563   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6564   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6565   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6566 
6567   subs(rscratch1, cnt, zero_words_block_size);
6568   Label around;
6569   br(LO, around);
6570   {
6571     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6572     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6573     // Make sure this is a C2 compilation. C1 allocates space only for
6574     // trampoline stubs generated by Call LIR ops, and in any case it
6575     // makes sense for a C1 compilation task to proceed as quickly as
6576     // possible.
6577     CompileTask* task;
6578     if (StubRoutines::aarch64::complete()
6579         && Thread::current()->is_Compiler_thread()
6580         && (task = ciEnv::current()->task())
6581         && is_c2_compile(task->comp_level())) {
6582       address tpc = trampoline_call(zero_blocks);
6583       if (tpc == nullptr) {
6584         DEBUG_ONLY(reset_labels(around));
6585         return nullptr;
6586       }
6587     } else {
6588       far_call(zero_blocks);
6589     }
6590   }
6591   bind(around);
6592 
6593   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6594   // for us.
6595   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6596     Label l;
6597     tbz(cnt, exact_log2(i), l);
6598     for (int j = 0; j < i; j += 2) {
6599       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6600     }
6601     bind(l);
6602   }
6603   {
6604     Label l;
6605     tbz(cnt, 0, l);
6606     str(zr, Address(ptr));
6607     bind(l);
6608   }
6609 
6610   BLOCK_COMMENT("} zero_words");
6611   return pc();
6612 }
6613 
6614 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6615 // cnt:          Immediate count in HeapWords.
6616 //
6617 // r10, r11, rscratch1, and rscratch2 are clobbered.
6618 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6619 {
6620   assert(wordSize <= BlockZeroingLowLimit,
6621             "increase BlockZeroingLowLimit");
6622   address result = nullptr;
6623   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6624 #ifndef PRODUCT
6625     {
6626       char buf[64];
6627       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6628       BLOCK_COMMENT(buf);
6629     }
6630 #endif
6631     if (cnt >= 16) {
6632       uint64_t loops = cnt/16;
6633       if (loops > 1) {
6634         mov(rscratch2, loops - 1);
6635       }
6636       {
6637         Label loop;
6638         bind(loop);
6639         for (int i = 0; i < 16; i += 2) {
6640           stp(zr, zr, Address(base, i * BytesPerWord));
6641         }
6642         add(base, base, 16 * BytesPerWord);
6643         if (loops > 1) {
6644           subs(rscratch2, rscratch2, 1);
6645           br(GE, loop);
6646         }
6647       }
6648     }
6649     cnt %= 16;
6650     int i = cnt & 1;  // store any odd word to start
6651     if (i) str(zr, Address(base));
6652     for (; i < (int)cnt; i += 2) {
6653       stp(zr, zr, Address(base, i * wordSize));
6654     }
6655     BLOCK_COMMENT("} zero_words");
6656     result = pc();
6657   } else {
6658     mov(r10, base); mov(r11, cnt);
6659     result = zero_words(r10, r11);
6660   }
6661   return result;
6662 }
6663 
6664 // Zero blocks of memory by using DC ZVA.
6665 //
6666 // Aligns the base address first sufficiently for DC ZVA, then uses
6667 // DC ZVA repeatedly for every full block.  cnt is the size to be
6668 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6669 // in cnt.
6670 //
6671 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6672 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6673 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6674   Register tmp = rscratch1;
6675   Register tmp2 = rscratch2;
6676   int zva_length = VM_Version::zva_length();
6677   Label initial_table_end, loop_zva;
6678   Label fini;
6679 
6680   // Base must be 16 byte aligned. If not just return and let caller handle it
6681   tst(base, 0x0f);
6682   br(Assembler::NE, fini);
6683   // Align base with ZVA length.
6684   neg(tmp, base);
6685   andr(tmp, tmp, zva_length - 1);
6686 
6687   // tmp: the number of bytes to be filled to align the base with ZVA length.
6688   add(base, base, tmp);
6689   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6690   adr(tmp2, initial_table_end);
6691   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6692   br(tmp2);
6693 
6694   for (int i = -zva_length + 16; i < 0; i += 16)
6695     stp(zr, zr, Address(base, i));
6696   bind(initial_table_end);
6697 
6698   sub(cnt, cnt, zva_length >> 3);
6699   bind(loop_zva);
6700   dc(Assembler::ZVA, base);
6701   subs(cnt, cnt, zva_length >> 3);
6702   add(base, base, zva_length);
6703   br(Assembler::GE, loop_zva);
6704   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6705   bind(fini);
6706 }
6707 
6708 // base:   Address of a buffer to be filled, 8 bytes aligned.
6709 // cnt:    Count in 8-byte unit.
6710 // value:  Value to be filled with.
6711 // base will point to the end of the buffer after filling.
6712 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6713 {
6714 //  Algorithm:
6715 //
6716 //    if (cnt == 0) {
6717 //      return;
6718 //    }
6719 //    if ((p & 8) != 0) {
6720 //      *p++ = v;
6721 //    }
6722 //
6723 //    scratch1 = cnt & 14;
6724 //    cnt -= scratch1;
6725 //    p += scratch1;
6726 //    switch (scratch1 / 2) {
6727 //      do {
6728 //        cnt -= 16;
6729 //          p[-16] = v;
6730 //          p[-15] = v;
6731 //        case 7:
6732 //          p[-14] = v;
6733 //          p[-13] = v;
6734 //        case 6:
6735 //          p[-12] = v;
6736 //          p[-11] = v;
6737 //          // ...
6738 //        case 1:
6739 //          p[-2] = v;
6740 //          p[-1] = v;
6741 //        case 0:
6742 //          p += 16;
6743 //      } while (cnt);
6744 //    }
6745 //    if ((cnt & 1) == 1) {
6746 //      *p++ = v;
6747 //    }
6748 
6749   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6750 
6751   Label fini, skip, entry, loop;
6752   const int unroll = 8; // Number of stp instructions we'll unroll
6753 
6754   cbz(cnt, fini);
6755   tbz(base, 3, skip);
6756   str(value, Address(post(base, 8)));
6757   sub(cnt, cnt, 1);
6758   bind(skip);
6759 
6760   andr(rscratch1, cnt, (unroll-1) * 2);
6761   sub(cnt, cnt, rscratch1);
6762   add(base, base, rscratch1, Assembler::LSL, 3);
6763   adr(rscratch2, entry);
6764   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6765   br(rscratch2);
6766 
6767   bind(loop);
6768   add(base, base, unroll * 16);
6769   for (int i = -unroll; i < 0; i++)
6770     stp(value, value, Address(base, i * 16));
6771   bind(entry);
6772   subs(cnt, cnt, unroll * 2);
6773   br(Assembler::GE, loop);
6774 
6775   tbz(cnt, 0, fini);
6776   str(value, Address(post(base, 8)));
6777   bind(fini);
6778 }
6779 
6780 // Intrinsic for
6781 //
6782 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6783 //   Encodes char[] to byte[] in ISO-8859-1
6784 //
6785 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6786 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6787 //
6788 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6789 //   Encodes char[] to byte[] in ASCII
6790 //
6791 // This version always returns the number of characters copied, and does not
6792 // clobber the 'len' register. A successful copy will complete with the post-
6793 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6794 // post-condition: 0 <= 'res' < 'len'.
6795 //
6796 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6797 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6798 //       beyond the acceptable, even though the footprint would be smaller.
6799 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6800 //       avoid additional bloat.
6801 //
6802 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6803 void MacroAssembler::encode_iso_array(Register src, Register dst,
6804                                       Register len, Register res, bool ascii,
6805                                       FloatRegister vtmp0, FloatRegister vtmp1,
6806                                       FloatRegister vtmp2, FloatRegister vtmp3,
6807                                       FloatRegister vtmp4, FloatRegister vtmp5)
6808 {
6809   Register cnt = res;
6810   Register max = rscratch1;
6811   Register chk = rscratch2;
6812 
6813   prfm(Address(src), PLDL1STRM);
6814   movw(cnt, len);
6815 
6816 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6817 
6818   Label LOOP_32, DONE_32, FAIL_32;
6819 
6820   BIND(LOOP_32);
6821   {
6822     cmpw(cnt, 32);
6823     br(LT, DONE_32);
6824     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6825     // Extract lower bytes.
6826     FloatRegister vlo0 = vtmp4;
6827     FloatRegister vlo1 = vtmp5;
6828     uzp1(vlo0, T16B, vtmp0, vtmp1);
6829     uzp1(vlo1, T16B, vtmp2, vtmp3);
6830     // Merge bits...
6831     orr(vtmp0, T16B, vtmp0, vtmp1);
6832     orr(vtmp2, T16B, vtmp2, vtmp3);
6833     // Extract merged upper bytes.
6834     FloatRegister vhix = vtmp0;
6835     uzp2(vhix, T16B, vtmp0, vtmp2);
6836     // ISO-check on hi-parts (all zero).
6837     //                          ASCII-check on lo-parts (no sign).
6838     FloatRegister vlox = vtmp1; // Merge lower bytes.
6839                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6840     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6841     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6842     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6843                                 ASCII(orr(chk, chk, max));
6844     cbnz(chk, FAIL_32);
6845     subw(cnt, cnt, 32);
6846     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6847     b(LOOP_32);
6848   }
6849   BIND(FAIL_32);
6850   sub(src, src, 64);
6851   BIND(DONE_32);
6852 
6853   Label LOOP_8, SKIP_8;
6854 
6855   BIND(LOOP_8);
6856   {
6857     cmpw(cnt, 8);
6858     br(LT, SKIP_8);
6859     FloatRegister vhi = vtmp0;
6860     FloatRegister vlo = vtmp1;
6861     ld1(vtmp3, T8H, src);
6862     uzp1(vlo, T16B, vtmp3, vtmp3);
6863     uzp2(vhi, T16B, vtmp3, vtmp3);
6864     // ISO-check on hi-parts (all zero).
6865     //                          ASCII-check on lo-parts (no sign).
6866                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6867     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6868                                 ASCII(umov(max, vtmp2, B, 0));
6869                                 ASCII(orr(chk, chk, max));
6870     cbnz(chk, SKIP_8);
6871 
6872     strd(vlo, Address(post(dst, 8)));
6873     subw(cnt, cnt, 8);
6874     add(src, src, 16);
6875     b(LOOP_8);
6876   }
6877   BIND(SKIP_8);
6878 
6879 #undef ASCII
6880 
6881   Label LOOP, DONE;
6882 
6883   cbz(cnt, DONE);
6884   BIND(LOOP);
6885   {
6886     Register chr = rscratch1;
6887     ldrh(chr, Address(post(src, 2)));
6888     tst(chr, ascii ? 0xff80 : 0xff00);
6889     br(NE, DONE);
6890     strb(chr, Address(post(dst, 1)));
6891     subs(cnt, cnt, 1);
6892     br(GT, LOOP);
6893   }
6894   BIND(DONE);
6895   // Return index where we stopped.
6896   subw(res, len, cnt);
6897 }
6898 
6899 // Inflate byte[] array to char[].
6900 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6901 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6902                                            FloatRegister vtmp1, FloatRegister vtmp2,
6903                                            FloatRegister vtmp3, Register tmp4) {
6904   Label big, done, after_init, to_stub;
6905 
6906   assert_different_registers(src, dst, len, tmp4, rscratch1);
6907 
6908   fmovd(vtmp1, 0.0);
6909   lsrw(tmp4, len, 3);
6910   bind(after_init);
6911   cbnzw(tmp4, big);
6912   // Short string: less than 8 bytes.
6913   {
6914     Label loop, tiny;
6915 
6916     cmpw(len, 4);
6917     br(LT, tiny);
6918     // Use SIMD to do 4 bytes.
6919     ldrs(vtmp2, post(src, 4));
6920     zip1(vtmp3, T8B, vtmp2, vtmp1);
6921     subw(len, len, 4);
6922     strd(vtmp3, post(dst, 8));
6923 
6924     cbzw(len, done);
6925 
6926     // Do the remaining bytes by steam.
6927     bind(loop);
6928     ldrb(tmp4, post(src, 1));
6929     strh(tmp4, post(dst, 2));
6930     subw(len, len, 1);
6931 
6932     bind(tiny);
6933     cbnz(len, loop);
6934 
6935     b(done);
6936   }
6937 
6938   if (SoftwarePrefetchHintDistance >= 0) {
6939     bind(to_stub);
6940       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6941       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6942       address tpc = trampoline_call(stub);
6943       if (tpc == nullptr) {
6944         DEBUG_ONLY(reset_labels(big, done));
6945         postcond(pc() == badAddress);
6946         return nullptr;
6947       }
6948       b(after_init);
6949   }
6950 
6951   // Unpack the bytes 8 at a time.
6952   bind(big);
6953   {
6954     Label loop, around, loop_last, loop_start;
6955 
6956     if (SoftwarePrefetchHintDistance >= 0) {
6957       const int large_loop_threshold = (64 + 16)/8;
6958       ldrd(vtmp2, post(src, 8));
6959       andw(len, len, 7);
6960       cmp(tmp4, (u1)large_loop_threshold);
6961       br(GE, to_stub);
6962       b(loop_start);
6963 
6964       bind(loop);
6965       ldrd(vtmp2, post(src, 8));
6966       bind(loop_start);
6967       subs(tmp4, tmp4, 1);
6968       br(EQ, loop_last);
6969       zip1(vtmp2, T16B, vtmp2, vtmp1);
6970       ldrd(vtmp3, post(src, 8));
6971       st1(vtmp2, T8H, post(dst, 16));
6972       subs(tmp4, tmp4, 1);
6973       zip1(vtmp3, T16B, vtmp3, vtmp1);
6974       st1(vtmp3, T8H, post(dst, 16));
6975       br(NE, loop);
6976       b(around);
6977       bind(loop_last);
6978       zip1(vtmp2, T16B, vtmp2, vtmp1);
6979       st1(vtmp2, T8H, post(dst, 16));
6980       bind(around);
6981       cbz(len, done);
6982     } else {
6983       andw(len, len, 7);
6984       bind(loop);
6985       ldrd(vtmp2, post(src, 8));
6986       sub(tmp4, tmp4, 1);
6987       zip1(vtmp3, T16B, vtmp2, vtmp1);
6988       st1(vtmp3, T8H, post(dst, 16));
6989       cbnz(tmp4, loop);
6990     }
6991   }
6992 
6993   // Do the tail of up to 8 bytes.
6994   add(src, src, len);
6995   ldrd(vtmp3, Address(src, -8));
6996   add(dst, dst, len, ext::uxtw, 1);
6997   zip1(vtmp3, T16B, vtmp3, vtmp1);
6998   strq(vtmp3, Address(dst, -16));
6999 
7000   bind(done);
7001   postcond(pc() != badAddress);
7002   return pc();
7003 }
7004 
7005 // Compress char[] array to byte[].
7006 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
7007 // Return the array length if every element in array can be encoded,
7008 // otherwise, the index of first non-latin1 (> 0xff) character.
7009 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
7010                                          Register res,
7011                                          FloatRegister tmp0, FloatRegister tmp1,
7012                                          FloatRegister tmp2, FloatRegister tmp3,
7013                                          FloatRegister tmp4, FloatRegister tmp5) {
7014   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
7015 }
7016 
7017 // java.math.round(double a)
7018 // Returns the closest long to the argument, with ties rounding to
7019 // positive infinity.  This requires some fiddling for corner
7020 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
7021 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
7022                                        FloatRegister ftmp) {
7023   Label DONE;
7024   BLOCK_COMMENT("java_round_double: { ");
7025   fmovd(rscratch1, src);
7026   // Use RoundToNearestTiesAway unless src small and -ve.
7027   fcvtasd(dst, src);
7028   // Test if src >= 0 || abs(src) >= 0x1.0p52
7029   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
7030   mov(rscratch2, julong_cast(0x1.0p52));
7031   cmp(rscratch1, rscratch2);
7032   br(HS, DONE); {
7033     // src < 0 && abs(src) < 0x1.0p52
7034     // src may have a fractional part, so add 0.5
7035     fmovd(ftmp, 0.5);
7036     faddd(ftmp, src, ftmp);
7037     // Convert double to jlong, use RoundTowardsNegative
7038     fcvtmsd(dst, ftmp);
7039   }
7040   bind(DONE);
7041   BLOCK_COMMENT("} java_round_double");
7042 }
7043 
7044 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
7045                                       FloatRegister ftmp) {
7046   Label DONE;
7047   BLOCK_COMMENT("java_round_float: { ");
7048   fmovs(rscratch1, src);
7049   // Use RoundToNearestTiesAway unless src small and -ve.
7050   fcvtassw(dst, src);
7051   // Test if src >= 0 || abs(src) >= 0x1.0p23
7052   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
7053   mov(rscratch2, jint_cast(0x1.0p23f));
7054   cmp(rscratch1, rscratch2);
7055   br(HS, DONE); {
7056     // src < 0 && |src| < 0x1.0p23
7057     // src may have a fractional part, so add 0.5
7058     fmovs(ftmp, 0.5f);
7059     fadds(ftmp, src, ftmp);
7060     // Convert float to jint, use RoundTowardsNegative
7061     fcvtmssw(dst, ftmp);
7062   }
7063   bind(DONE);
7064   BLOCK_COMMENT("} java_round_float");
7065 }
7066 
7067 // get_thread() can be called anywhere inside generated code so we
7068 // need to save whatever non-callee save context might get clobbered
7069 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
7070 // the call setup code.
7071 //
7072 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
7073 // On other systems, the helper is a usual C function.
7074 //
7075 void MacroAssembler::get_thread(Register dst) {
7076   RegSet saved_regs =
7077     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
7078     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
7079 
7080   protect_return_address();
7081   push(saved_regs, sp);
7082 
7083   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
7084   blr(lr);
7085   if (dst != c_rarg0) {
7086     mov(dst, c_rarg0);
7087   }
7088 
7089   pop(saved_regs, sp);
7090   authenticate_return_address();
7091 }
7092 
7093 #ifdef COMPILER2
7094 // C2 compiled method's prolog code
7095 // Moved here from aarch64.ad to support Valhalla code below
7096 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
7097   if (C->clinit_barrier_on_entry()) {
7098     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
7099 
7100     Label L_skip_barrier;
7101 
7102     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7103     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7104     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7105     bind(L_skip_barrier);
7106   }
7107 
7108   if (C->max_vector_size() > 0) {
7109     reinitialize_ptrue();
7110   }
7111 
7112   int bangsize = C->output()->bang_size_in_bytes();
7113   if (C->output()->need_stack_bang(bangsize))
7114     generate_stack_overflow_check(bangsize);
7115 
7116   // n.b. frame size includes space for return pc and rfp
7117   const long framesize = C->output()->frame_size_in_bytes();
7118   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
7119 
7120   if (C->needs_stack_repair()) {
7121     save_stack_increment(sp_inc, framesize);
7122   }
7123 
7124   if (VerifyStackAtCalls) {
7125     Unimplemented();
7126   }
7127 }
7128 #endif // COMPILER2
7129 
7130 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7131   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7132   // An inline type might be returned. If fields are in registers we
7133   // need to allocate an inline type instance and initialize it with
7134   // the value of the fields.
7135   Label skip;
7136   // We only need a new buffered inline type if a new one is not returned
7137   tbz(r0, 0, skip);
7138   int call_offset = -1;
7139 
7140   // Be careful not to clobber r1-7 which hold returned fields
7141   // Also do not use callee-saved registers as these may be live in the interpreter
7142   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7143 
7144   // The following code is similar to allocate_instance but has some slight differences,
7145   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7146   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
7147   Label slow_case;
7148   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7149   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7150 
7151   if (vk != nullptr) {
7152     // Called from C1, where the return type is statically known.
7153     movptr(klass, (intptr_t)vk->get_InlineKlass());
7154     jint lh = vk->layout_helper();
7155     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7156     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7157       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7158     } else {
7159       b(slow_case);
7160     }
7161   } else {
7162     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7163     andr(klass, r0, -2);
7164     if (UseTLAB) {
7165       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7166       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7167       br(Assembler::NE, slow_case);
7168       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7169     } else {
7170       b(slow_case);
7171     }
7172   }
7173   if (UseTLAB) {
7174     // 2. Initialize buffered inline instance header
7175     Register buffer_obj = r0;
7176     if (UseCompactObjectHeaders) {
7177       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7178       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7179     } else {
7180       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7181       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7182       store_klass_gap(buffer_obj, zr);
7183       if (vk == nullptr) {
7184         // store_klass corrupts klass, so save it for later use (interpreter case only).
7185         mov(tmp1, klass);
7186       }
7187       store_klass(buffer_obj, klass);
7188       klass = tmp1;
7189     }
7190     // 3. Initialize its fields with an inline class specific handler
7191     if (vk != nullptr) {
7192       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7193     } else {
7194       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7195       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7196       blr(tmp1);
7197     }
7198 
7199     membar(Assembler::StoreStore);
7200     b(skip);
7201   } else {
7202     // Must have already branched to slow_case above.
7203     DEBUG_ONLY(should_not_reach_here());
7204   }
7205   bind(slow_case);
7206   // We failed to allocate a new inline type, fall back to a runtime
7207   // call. Some oop field may be live in some registers but we can't
7208   // tell. That runtime call will take care of preserving them
7209   // across a GC if there's one.
7210   mov(r0, r0_preserved);
7211 
7212   if (from_interpreter) {
7213     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7214   } else {
7215     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7216     call_offset = offset();
7217   }
7218   membar(Assembler::StoreStore);
7219 
7220   bind(skip);
7221   return call_offset;
7222 }
7223 
7224 // Move a value between registers/stack slots and update the reg_state
7225 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7226   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7227   if (reg_state[to->value()] == reg_written) {
7228     return true; // Already written
7229   }
7230 
7231   if (from != to && bt != T_VOID) {
7232     if (reg_state[to->value()] == reg_readonly) {
7233       return false; // Not yet writable
7234     }
7235     if (from->is_reg()) {
7236       if (to->is_reg()) {
7237         if (from->is_Register() && to->is_Register()) {
7238           mov(to->as_Register(), from->as_Register());
7239         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7240           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7241         } else {
7242           ShouldNotReachHere();
7243         }
7244       } else {
7245         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7246         Address to_addr = Address(sp, st_off);
7247         if (from->is_FloatRegister()) {
7248           if (bt == T_DOUBLE) {
7249              strd(from->as_FloatRegister(), to_addr);
7250           } else {
7251              assert(bt == T_FLOAT, "must be float");
7252              strs(from->as_FloatRegister(), to_addr);
7253           }
7254         } else {
7255           str(from->as_Register(), to_addr);
7256         }
7257       }
7258     } else {
7259       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7260       if (to->is_reg()) {
7261         if (to->is_FloatRegister()) {
7262           if (bt == T_DOUBLE) {
7263             ldrd(to->as_FloatRegister(), from_addr);
7264           } else {
7265             assert(bt == T_FLOAT, "must be float");
7266             ldrs(to->as_FloatRegister(), from_addr);
7267           }
7268         } else {
7269           ldr(to->as_Register(), from_addr);
7270         }
7271       } else {
7272         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7273         ldr(rscratch1, from_addr);
7274         str(rscratch1, Address(sp, st_off));
7275       }
7276     }
7277   }
7278 
7279   // Update register states
7280   reg_state[from->value()] = reg_writable;
7281   reg_state[to->value()] = reg_written;
7282   return true;
7283 }
7284 
7285 // Calculate the extra stack space required for packing or unpacking inline
7286 // args and adjust the stack pointer
7287 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7288   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7289   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7290   assert(sp_inc > 0, "sanity");
7291 
7292   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7293   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7294 
7295   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7296   if (sp_inc < (1 << 9)) {
7297     sub(sp, sp, sp_inc);   // Fits in an immediate
7298   } else {
7299     mov(rscratch1, sp_inc);
7300     sub(sp, sp, rscratch1);
7301   }
7302 
7303   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7304 }
7305 
7306 // Read all fields from an inline type oop and store the values in registers/stack slots
7307 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7308                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7309                                           RegState reg_state[]) {
7310   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7311   assert(from->is_valid(), "source must be valid");
7312   bool progress = false;
7313 #ifdef ASSERT
7314   const int start_offset = offset();
7315 #endif
7316 
7317   Label L_null, L_notNull;
7318   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7319   Register tmp1 = r10;
7320   Register tmp2 = r11;
7321 
7322 #ifndef ASSERT
7323   RegSet clobbered_gp_regs = MacroAssembler::call_clobbered_gp_registers();
7324   assert(clobbered_gp_regs.contains(tmp1), "tmp1 must be saved explicitly if it's not a clobber");
7325   assert(clobbered_gp_regs.contains(tmp2), "tmp2 must be saved explicitly if it's not a clobber");
7326   assert(clobbered_gp_regs.contains(r14), "r14 must be saved explicitly if it's not a clobber");
7327 #endif
7328 
7329   Register fromReg = noreg;
7330   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
7331   bool done = true;
7332   bool mark_done = true;
7333   VMReg toReg;
7334   BasicType bt;
7335   // Check if argument requires a null check
7336   bool null_check = false;
7337   VMReg nullCheckReg;
7338   while (stream.next(nullCheckReg, bt)) {
7339     if (sig->at(stream.sig_index())._offset == -1) {
7340       null_check = true;
7341       break;
7342     }
7343   }
7344   stream.reset(sig_index, to_index);
7345   while (stream.next(toReg, bt)) {
7346     assert(toReg->is_valid(), "destination must be valid");
7347     int idx = (int)toReg->value();
7348     if (reg_state[idx] == reg_readonly) {
7349       if (idx != from->value()) {
7350         mark_done = false;
7351       }
7352       done = false;
7353       continue;
7354     } else if (reg_state[idx] == reg_written) {
7355       continue;
7356     }
7357     assert(reg_state[idx] == reg_writable, "must be writable");
7358     reg_state[idx] = reg_written;
7359     progress = true;
7360 
7361     if (fromReg == noreg) {
7362       if (from->is_reg()) {
7363         fromReg = from->as_Register();
7364       } else {
7365         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7366         ldr(tmp1, Address(sp, st_off));
7367         fromReg = tmp1;
7368       }
7369       if (null_check) {
7370         // Nullable inline type argument, emit null check
7371         cbz(fromReg, L_null);
7372       }
7373     }
7374     int off = sig->at(stream.sig_index())._offset;
7375     if (off == -1) {
7376       assert(null_check, "Missing null check at");
7377       if (toReg->is_stack()) {
7378         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7379         mov(tmp2, 1);
7380         str(tmp2, Address(sp, st_off));
7381       } else {
7382         mov(toReg->as_Register(), 1);
7383       }
7384       continue;
7385     }
7386     if (sig->at(stream.sig_index())._vt_oop) {
7387       if (toReg->is_stack()) {
7388         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7389         str(fromReg, Address(sp, st_off));
7390       } else {
7391         mov(toReg->as_Register(), fromReg);
7392       }
7393       continue;
7394     }
7395     assert(off > 0, "offset in object should be positive");
7396     Address fromAddr = Address(fromReg, off);
7397     if (!toReg->is_FloatRegister()) {
7398       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7399       if (is_reference_type(bt)) {
7400         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7401       } else {
7402         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7403         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7404       }
7405       if (toReg->is_stack()) {
7406         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7407         str(dst, Address(sp, st_off));
7408       }
7409     } else if (bt == T_DOUBLE) {
7410       ldrd(toReg->as_FloatRegister(), fromAddr);
7411     } else {
7412       assert(bt == T_FLOAT, "must be float");
7413       ldrs(toReg->as_FloatRegister(), fromAddr);
7414     }
7415   }
7416   if (progress && null_check) {
7417     if (done) {
7418       b(L_notNull);
7419       bind(L_null);
7420       // Set null marker to zero to signal that the argument is null.
7421       // Also set all fields to zero since the runtime requires a canonical
7422       // representation of a flat null.
7423       stream.reset(sig_index, to_index);
7424       while (stream.next(toReg, bt)) {
7425         if (toReg->is_stack()) {
7426           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7427           str(zr, Address(sp, st_off));
7428         } else if (toReg->is_FloatRegister()) {
7429           mov(toReg->as_FloatRegister(), T2S, 0);
7430         } else {
7431           mov(toReg->as_Register(), zr);
7432         }
7433       }
7434       bind(L_notNull);
7435     } else {
7436       bind(L_null);
7437     }
7438   }
7439 
7440   sig_index = stream.sig_index();
7441   to_index = stream.regs_index();
7442 
7443   if (mark_done && reg_state[from->value()] != reg_written) {
7444     // This is okay because no one else will write to that slot
7445     reg_state[from->value()] = reg_writable;
7446   }
7447   from_index--;
7448   assert(progress || (start_offset == offset()), "should not emit code");
7449   return done;
7450 }
7451 
7452 // Pack fields back into an inline type oop
7453 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7454                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7455                                         RegState reg_state[], Register val_array) {
7456   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7457   assert(to->is_valid(), "destination must be valid");
7458 
7459   if (reg_state[to->value()] == reg_written) {
7460     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7461     return true; // Already written
7462   }
7463 
7464   // The GC barrier expanded by store_heap_oop below may call into the
7465   // runtime so use callee-saved registers for any values that need to be
7466   // preserved. The GC barrier assembler should take care of saving the
7467   // Java argument registers.
7468   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
7469   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7470   Register val_obj_tmp = r21;
7471   Register from_reg_tmp = r22;
7472   Register tmp1 = r14;
7473   Register tmp2 = r13;
7474   Register tmp3 = r12;
7475   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7476 
7477   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7478 
7479   if (reg_state[to->value()] == reg_readonly) {
7480     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7481       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7482       return false; // Not yet writable
7483     }
7484     val_obj = val_obj_tmp;
7485   }
7486 
7487   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7488   VMReg fromReg;
7489   BasicType bt;
7490   Label L_null;
7491   while (stream.next(fromReg, bt)) {
7492     assert(fromReg->is_valid(), "source must be valid");
7493     reg_state[fromReg->value()] = reg_writable;
7494 
7495     int off = sig->at(stream.sig_index())._offset;
7496     if (off == -1) {
7497       // Nullable inline type argument, emit null check
7498       Label L_notNull;
7499       if (fromReg->is_stack()) {
7500         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7501         ldrb(tmp2, Address(sp, ld_off));
7502         cbnz(tmp2, L_notNull);
7503       } else {
7504         cbnz(fromReg->as_Register(), L_notNull);
7505       }
7506       mov(val_obj, 0);
7507       b(L_null);
7508       bind(L_notNull);
7509       continue;
7510     }
7511     if (sig->at(stream.sig_index())._vt_oop) {
7512       if (fromReg->is_stack()) {
7513         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7514         ldr(val_obj, Address(sp, ld_off));
7515       } else {
7516         mov(val_obj, fromReg->as_Register());
7517       }
7518       cbnz(val_obj, L_null);
7519       // get the buffer from the just allocated pool of buffers
7520       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7521       load_heap_oop(val_obj, Address(val_array, index), rscratch1, rscratch2);
7522       continue;
7523     }
7524 
7525     assert(off > 0, "offset in object should be positive");
7526     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7527 
7528     // Pack the scalarized field into the value object.
7529     Address dst(val_obj, off);
7530     if (!fromReg->is_FloatRegister()) {
7531       Register src;
7532       if (fromReg->is_stack()) {
7533         src = from_reg_tmp;
7534         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7535         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7536       } else {
7537         src = fromReg->as_Register();
7538       }
7539       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7540       if (is_reference_type(bt)) {
7541         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
7542         mov(tmp3, val_obj);
7543         Address dst_with_tmp3(tmp3, off);
7544         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7545       } else {
7546         store_sized_value(dst, src, size_in_bytes);
7547       }
7548     } else if (bt == T_DOUBLE) {
7549       strd(fromReg->as_FloatRegister(), dst);
7550     } else {
7551       assert(bt == T_FLOAT, "must be float");
7552       strs(fromReg->as_FloatRegister(), dst);
7553     }
7554   }
7555   bind(L_null);
7556   sig_index = stream.sig_index();
7557   from_index = stream.regs_index();
7558 
7559   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7560   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7561   assert(success, "to register must be writeable");
7562   return true;
7563 }
7564 
7565 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7566   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7567 }
7568 
7569 void MacroAssembler::cache_wb(Address line) {
7570   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7571   assert(line.index() == noreg, "index should be noreg");
7572   assert(line.offset() == 0, "offset should be 0");
7573   // would like to assert this
7574   // assert(line._ext.shift == 0, "shift should be zero");
7575   if (VM_Version::supports_dcpop()) {
7576     // writeback using clear virtual address to point of persistence
7577     dc(Assembler::CVAP, line.base());
7578   } else {
7579     // no need to generate anything as Unsafe.writebackMemory should
7580     // never invoke this stub
7581   }
7582 }
7583 
7584 void MacroAssembler::cache_wbsync(bool is_pre) {
7585   // we only need a barrier post sync
7586   if (!is_pre) {
7587     membar(Assembler::AnyAny);
7588   }
7589 }
7590 
7591 void MacroAssembler::verify_sve_vector_length(Register tmp) {
7592   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
7593     return;
7594   }
7595   // Make sure that native code does not change SVE vector length.
7596   Label verify_ok;
7597   movw(tmp, zr);
7598   sve_inc(tmp, B);
7599   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
7600   br(EQ, verify_ok);
7601   stop("Error: SVE vector length has changed since jvm startup");
7602   bind(verify_ok);
7603 }
7604 
7605 void MacroAssembler::verify_ptrue() {
7606   Label verify_ok;
7607   if (!UseSVE) {
7608     return;
7609   }
7610   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
7611   sve_dec(rscratch1, B);
7612   cbz(rscratch1, verify_ok);
7613   stop("Error: the preserved predicate register (p7) elements are not all true");
7614   bind(verify_ok);
7615 }
7616 
7617 void MacroAssembler::safepoint_isb() {
7618   isb();
7619 #ifndef PRODUCT
7620   if (VerifyCrossModifyFence) {
7621     // Clear the thread state.
7622     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7623   }
7624 #endif
7625 }
7626 
7627 #ifndef PRODUCT
7628 void MacroAssembler::verify_cross_modify_fence_not_required() {
7629   if (VerifyCrossModifyFence) {
7630     // Check if thread needs a cross modify fence.
7631     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7632     Label fence_not_required;
7633     cbz(rscratch1, fence_not_required);
7634     // If it does then fail.
7635     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
7636     mov(c_rarg0, rthread);
7637     blr(rscratch1);
7638     bind(fence_not_required);
7639   }
7640 }
7641 #endif
7642 
7643 void MacroAssembler::spin_wait() {
7644   block_comment("spin_wait {");
7645   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
7646     switch (VM_Version::spin_wait_desc().inst()) {
7647       case SpinWait::NOP:
7648         nop();
7649         break;
7650       case SpinWait::ISB:
7651         isb();
7652         break;
7653       case SpinWait::YIELD:
7654         yield();
7655         break;
7656       case SpinWait::SB:
7657         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
7658         sb();
7659         break;
7660       case SpinWait::WFET:
7661         spin_wait_wfet(VM_Version::spin_wait_desc().delay());
7662         break;
7663       default:
7664         ShouldNotReachHere();
7665     }
7666   }
7667   block_comment("}");
7668 }
7669 
7670 void MacroAssembler::spin_wait_wfet(int delay_ns) {
7671   // The sequence assumes CNTFRQ_EL0 is fixed to 1GHz. The assumption is valid
7672   // starting from Armv8.6, according to the "D12.1.2 The system counter" of the
7673   // Arm Architecture Reference Manual for A-profile architecture version M.a.a.
7674   // This is sufficient because FEAT_WFXT is introduced from Armv8.6.
7675   Register target = rscratch1;
7676   Register current = rscratch2;
7677   get_cntvctss_el0(current);
7678   add(target, current, delay_ns);
7679 
7680   Label L_wait_loop;
7681   bind(L_wait_loop);
7682 
7683   wfet(target);
7684   get_cntvctss_el0(current);
7685 
7686   cmp(current, target);
7687   br(LT, L_wait_loop);
7688 
7689   sb();
7690 }
7691 
7692 // Stack frame creation/removal
7693 
7694 void MacroAssembler::enter(bool strip_ret_addr) {
7695   if (strip_ret_addr) {
7696     // Addresses can only be signed once. If there are multiple nested frames being created
7697     // in the same function, then the return address needs stripping first.
7698     strip_return_address();
7699   }
7700   protect_return_address();
7701   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7702   mov(rfp, sp);
7703 }
7704 
7705 void MacroAssembler::leave() {
7706   mov(sp, rfp);
7707   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
7708   authenticate_return_address();
7709 }
7710 
7711 // ROP Protection
7712 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
7713 // destroying stack frames or whenever directly loading/storing the LR to memory.
7714 // If ROP protection is not set then these functions are no-ops.
7715 // For more details on PAC see pauth_aarch64.hpp.
7716 
7717 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
7718 // Uses value zero as the modifier.
7719 //
7720 void MacroAssembler::protect_return_address() {
7721   if (VM_Version::use_rop_protection()) {
7722     check_return_address();
7723     paciaz();
7724   }
7725 }
7726 
7727 // Sign the return value in the given register. Use before updating the LR in the existing stack
7728 // frame for the current function.
7729 // Uses value zero as the modifier.
7730 //
7731 void MacroAssembler::protect_return_address(Register return_reg) {
7732   if (VM_Version::use_rop_protection()) {
7733     check_return_address(return_reg);
7734     paciza(return_reg);
7735   }
7736 }
7737 
7738 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
7739 // Uses value zero as the modifier.
7740 //
7741 void MacroAssembler::authenticate_return_address() {
7742   if (VM_Version::use_rop_protection()) {
7743     autiaz();
7744     check_return_address();
7745   }
7746 }
7747 
7748 // Authenticate the return value in the given register. Use before updating the LR in the existing
7749 // stack frame for the current function.
7750 // Uses value zero as the modifier.
7751 //
7752 void MacroAssembler::authenticate_return_address(Register return_reg) {
7753   if (VM_Version::use_rop_protection()) {
7754     autiza(return_reg);
7755     check_return_address(return_reg);
7756   }
7757 }
7758 
7759 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
7760 // there is no guaranteed way of authenticating the LR.
7761 //
7762 void MacroAssembler::strip_return_address() {
7763   if (VM_Version::use_rop_protection()) {
7764     xpaclri();
7765   }
7766 }
7767 
7768 #ifndef PRODUCT
7769 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
7770 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
7771 // it is difficult to debug back to the callee function.
7772 // This function simply loads from the address in the given register.
7773 // Use directly after authentication to catch authentication failures.
7774 // Also use before signing to check that the pointer is valid and hasn't already been signed.
7775 //
7776 void MacroAssembler::check_return_address(Register return_reg) {
7777   if (VM_Version::use_rop_protection()) {
7778     ldr(zr, Address(return_reg));
7779   }
7780 }
7781 #endif
7782 
7783 // The java_calling_convention describes stack locations as ideal slots on
7784 // a frame with no abi restrictions. Since we must observe abi restrictions
7785 // (like the placement of the register window) the slots must be biased by
7786 // the following value.
7787 static int reg2offset_in(VMReg r) {
7788   // Account for saved rfp and lr
7789   // This should really be in_preserve_stack_slots
7790   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
7791 }
7792 
7793 static int reg2offset_out(VMReg r) {
7794   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
7795 }
7796 
7797 // On 64bit we will store integer like items to the stack as
7798 // 64bits items (AArch64 ABI) even though java would only store
7799 // 32bits for a parameter. On 32bit it will simply be 32bits
7800 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
7801 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
7802   if (src.first()->is_stack()) {
7803     if (dst.first()->is_stack()) {
7804       // stack to stack
7805       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7806       str(tmp, Address(sp, reg2offset_out(dst.first())));
7807     } else {
7808       // stack to reg
7809       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7810     }
7811   } else if (dst.first()->is_stack()) {
7812     // reg to stack
7813     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7814   } else {
7815     if (dst.first() != src.first()) {
7816       sxtw(dst.first()->as_Register(), src.first()->as_Register());
7817     }
7818   }
7819 }
7820 
7821 // An oop arg. Must pass a handle not the oop itself
7822 void MacroAssembler::object_move(
7823                         OopMap* map,
7824                         int oop_handle_offset,
7825                         int framesize_in_slots,
7826                         VMRegPair src,
7827                         VMRegPair dst,
7828                         bool is_receiver,
7829                         int* receiver_offset) {
7830 
7831   // must pass a handle. First figure out the location we use as a handle
7832 
7833   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
7834 
7835   // See if oop is null if it is we need no handle
7836 
7837   if (src.first()->is_stack()) {
7838 
7839     // Oop is already on the stack as an argument
7840     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
7841     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
7842     if (is_receiver) {
7843       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
7844     }
7845 
7846     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7847     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7848     // conditionally move a null
7849     cmp(rscratch1, zr);
7850     csel(rHandle, zr, rHandle, Assembler::EQ);
7851   } else {
7852 
7853     // Oop is in an a register we must store it to the space we reserve
7854     // on the stack for oop_handles and pass a handle if oop is non-null
7855 
7856     const Register rOop = src.first()->as_Register();
7857     int oop_slot;
7858     if (rOop == j_rarg0)
7859       oop_slot = 0;
7860     else if (rOop == j_rarg1)
7861       oop_slot = 1;
7862     else if (rOop == j_rarg2)
7863       oop_slot = 2;
7864     else if (rOop == j_rarg3)
7865       oop_slot = 3;
7866     else if (rOop == j_rarg4)
7867       oop_slot = 4;
7868     else if (rOop == j_rarg5)
7869       oop_slot = 5;
7870     else if (rOop == j_rarg6)
7871       oop_slot = 6;
7872     else {
7873       assert(rOop == j_rarg7, "wrong register");
7874       oop_slot = 7;
7875     }
7876 
7877     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7878     int offset = oop_slot*VMRegImpl::stack_slot_size;
7879 
7880     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7881     // Store oop in handle area, may be null
7882     str(rOop, Address(sp, offset));
7883     if (is_receiver) {
7884       *receiver_offset = offset;
7885     }
7886 
7887     cmp(rOop, zr);
7888     lea(rHandle, Address(sp, offset));
7889     // conditionally move a null
7890     csel(rHandle, zr, rHandle, Assembler::EQ);
7891   }
7892 
7893   // If arg is on the stack then place it otherwise it is already in correct reg.
7894   if (dst.first()->is_stack()) {
7895     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7896   }
7897 }
7898 
7899 // A float arg may have to do float reg int reg conversion
7900 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7901  if (src.first()->is_stack()) {
7902     if (dst.first()->is_stack()) {
7903       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7904       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7905     } else {
7906       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7907     }
7908   } else if (src.first() != dst.first()) {
7909     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7910       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7911     else
7912       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7913   }
7914 }
7915 
7916 // A long move
7917 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7918   if (src.first()->is_stack()) {
7919     if (dst.first()->is_stack()) {
7920       // stack to stack
7921       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7922       str(tmp, Address(sp, reg2offset_out(dst.first())));
7923     } else {
7924       // stack to reg
7925       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7926     }
7927   } else if (dst.first()->is_stack()) {
7928     // reg to stack
7929     // Do we really have to sign extend???
7930     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7931     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7932   } else {
7933     if (dst.first() != src.first()) {
7934       mov(dst.first()->as_Register(), src.first()->as_Register());
7935     }
7936   }
7937 }
7938 
7939 
7940 // A double move
7941 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7942  if (src.first()->is_stack()) {
7943     if (dst.first()->is_stack()) {
7944       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7945       str(tmp, Address(sp, reg2offset_out(dst.first())));
7946     } else {
7947       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7948     }
7949   } else if (src.first() != dst.first()) {
7950     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7951       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7952     else
7953       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7954   }
7955 }
7956 
7957 // Implements fast-locking.
7958 //
7959 //  - obj: the object to be locked
7960 //  - t1, t2, t3: temporary registers, will be destroyed
7961 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7962 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7963   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7964 
7965   Label push;
7966   const Register top = t1;
7967   const Register mark = t2;
7968   const Register t = t3;
7969 
7970   // Preload the markWord. It is important that this is the first
7971   // instruction emitted as it is part of C1's null check semantics.
7972   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7973 
7974   if (UseObjectMonitorTable) {
7975     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7976     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7977   }
7978 
7979   if (DiagnoseSyncOnValueBasedClasses != 0) {
7980     load_klass(t1, obj);
7981     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7982     tst(t1, KlassFlags::_misc_is_value_based_class);
7983     br(Assembler::NE, slow);
7984   }
7985 
7986   // Check if the lock-stack is full.
7987   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7988   cmpw(top, (unsigned)LockStack::end_offset());
7989   br(Assembler::GE, slow);
7990 
7991   // Check for recursion.
7992   subw(t, top, oopSize);
7993   ldr(t, Address(rthread, t));
7994   cmp(obj, t);
7995   br(Assembler::EQ, push);
7996 
7997   // Check header for monitor (0b10).
7998   tst(mark, markWord::monitor_value);
7999   br(Assembler::NE, slow);
8000 
8001   // Try to lock. Transition lock bits 0b01 => 0b00
8002   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
8003   orr(mark, mark, markWord::unlocked_value);
8004   // Mask inline_type bit such that we go to the slow path if object is an inline type
8005   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
8006 
8007   eor(t, mark, markWord::unlocked_value);
8008   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
8009           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
8010   br(Assembler::NE, slow);
8011 
8012   bind(push);
8013   // After successful lock, push object on lock-stack.
8014   str(obj, Address(rthread, top));
8015   addw(top, top, oopSize);
8016   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8017 }
8018 
8019 // Implements fast-unlocking.
8020 //
8021 // - obj: the object to be unlocked
8022 // - t1, t2, t3: temporary registers
8023 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
8024 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
8025   // cmpxchg clobbers rscratch1.
8026   assert_different_registers(obj, t1, t2, t3, rscratch1);
8027 
8028 #ifdef ASSERT
8029   {
8030     // Check for lock-stack underflow.
8031     Label stack_ok;
8032     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
8033     cmpw(t1, (unsigned)LockStack::start_offset());
8034     br(Assembler::GE, stack_ok);
8035     STOP("Lock-stack underflow");
8036     bind(stack_ok);
8037   }
8038 #endif
8039 
8040   Label unlocked, push_and_slow;
8041   const Register top = t1;
8042   const Register mark = t2;
8043   const Register t = t3;
8044 
8045   // Check if obj is top of lock-stack.
8046   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8047   subw(top, top, oopSize);
8048   ldr(t, Address(rthread, top));
8049   cmp(obj, t);
8050   br(Assembler::NE, slow);
8051 
8052   // Pop lock-stack.
8053   DEBUG_ONLY(str(zr, Address(rthread, top));)
8054   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8055 
8056   // Check if recursive.
8057   subw(t, top, oopSize);
8058   ldr(t, Address(rthread, t));
8059   cmp(obj, t);
8060   br(Assembler::EQ, unlocked);
8061 
8062   // Not recursive. Check header for monitor (0b10).
8063   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
8064   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
8065 
8066 #ifdef ASSERT
8067   // Check header not unlocked (0b01).
8068   Label not_unlocked;
8069   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
8070   stop("fast_unlock already unlocked");
8071   bind(not_unlocked);
8072 #endif
8073 
8074   // Try to unlock. Transition lock bits 0b00 => 0b01
8075   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
8076   orr(t, mark, markWord::unlocked_value);
8077   cmpxchg(obj, mark, t, Assembler::xword,
8078           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
8079   br(Assembler::EQ, unlocked);
8080 
8081   bind(push_and_slow);
8082   // Restore lock-stack and handle the unlock in runtime.
8083   DEBUG_ONLY(str(obj, Address(rthread, top));)
8084   addw(top, top, oopSize);
8085   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8086   b(slow);
8087 
8088   bind(unlocked);
8089 }