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 "cds/archiveBuilder.hpp"
  29 #include "ci/ciEnv.hpp"
  30 #include "ci/ciInlineKlass.hpp"
  31 #include "code/compiledIC.hpp"
  32 #include "compiler/compileTask.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "compiler/oopMap.hpp"
  35 #include "gc/shared/barrierSet.hpp"
  36 #include "gc/shared/barrierSetAssembler.hpp"
  37 #include "gc/shared/cardTableBarrierSet.hpp"
  38 #include "gc/shared/cardTable.hpp"
  39 #include "gc/shared/collectedHeap.hpp"
  40 #include "gc/shared/tlab_globals.hpp"
  41 #include "interpreter/bytecodeHistogram.hpp"
  42 #include "interpreter/interpreter.hpp"
  43 #include "interpreter/interpreterRuntime.hpp"
  44 #include "jvm.h"
  45 #include "memory/resourceArea.hpp"
  46 #include "memory/universe.hpp"
  47 #include "nativeInst_aarch64.hpp"
  48 #include "oops/accessDecorators.hpp"
  49 #include "oops/compressedKlass.inline.hpp"
  50 #include "oops/compressedOops.inline.hpp"
  51 #include "oops/klass.inline.hpp"
  52 #include "oops/resolvedFieldEntry.hpp"
  53 #include "runtime/arguments.hpp"
  54 #include "runtime/continuation.hpp"
  55 #include "runtime/globals.hpp"
  56 #include "runtime/icache.hpp"
  57 #include "runtime/interfaceSupport.inline.hpp"
  58 #include "runtime/javaThread.hpp"
  59 #include "runtime/jniHandles.inline.hpp"
  60 #include "runtime/sharedRuntime.hpp"
  61 #include "runtime/signature_cc.hpp"
  62 #include "runtime/stubRoutines.hpp"
  63 #include "utilities/globalDefinitions.hpp"
  64 #include "utilities/integerCast.hpp"
  65 #include "utilities/powerOfTwo.hpp"
  66 #include "vmreg_aarch64.inline.hpp"
  67 #ifdef COMPILER1
  68 #include "c1/c1_LIRAssembler.hpp"
  69 #endif
  70 #ifdef COMPILER2
  71 #include "oops/oop.hpp"
  72 #include "opto/compile.hpp"
  73 #include "opto/node.hpp"
  74 #include "opto/output.hpp"
  75 #endif
  76 
  77 #include <sys/types.h>
  78 
  79 #ifdef PRODUCT
  80 #define BLOCK_COMMENT(str) /* nothing */
  81 #else
  82 #define BLOCK_COMMENT(str) block_comment(str)
  83 #endif
  84 #define STOP(str) stop(str);
  85 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  86 
  87 #ifdef ASSERT
  88 extern "C" void disnm(intptr_t p);
  89 #endif
  90 // Target-dependent relocation processing
  91 //
  92 // Instruction sequences whose target may need to be retrieved or
  93 // patched are distinguished by their leading instruction, sorting
  94 // them into three main instruction groups and related subgroups.
  95 //
  96 // 1) Branch, Exception and System (insn count = 1)
  97 //    1a) Unconditional branch (immediate):
  98 //      b/bl imm19
  99 //    1b) Compare & branch (immediate):
 100 //      cbz/cbnz Rt imm19
 101 //    1c) Test & branch (immediate):
 102 //      tbz/tbnz Rt imm14
 103 //    1d) Conditional branch (immediate):
 104 //      b.cond imm19
 105 //
 106 // 2) Loads and Stores (insn count = 1)
 107 //    2a) Load register literal:
 108 //      ldr Rt imm19
 109 //
 110 // 3) Data Processing Immediate (insn count = 2 or 3)
 111 //    3a) PC-rel. addressing
 112 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 113 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 114 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 115 //      adr/adrp Rx imm21
 116 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 117 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 118 //      The latter form can only happen when the target is an
 119 //      ExternalAddress, and (by definition) ExternalAddresses don't
 120 //      move. Because of that property, there is never any need to
 121 //      patch the last of the three instructions. However,
 122 //      MacroAssembler::target_addr_for_insn takes all three
 123 //      instructions into account and returns the correct address.
 124 //    3b) Move wide (immediate)
 125 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 126 //
 127 // A switch on a subset of the instruction's bits provides an
 128 // efficient dispatch to these subcases.
 129 //
 130 // insn[28:26] -> main group ('x' == don't care)
 131 //   00x -> UNALLOCATED
 132 //   100 -> Data Processing Immediate
 133 //   101 -> Branch, Exception and System
 134 //   x1x -> Loads and Stores
 135 //
 136 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 137 // n.b. in some cases extra bits need to be checked to verify the
 138 // instruction is as expected
 139 //
 140 // 1) ... xx101x Branch, Exception and System
 141 //   1a)  00___x Unconditional branch (immediate)
 142 //   1b)  01___0 Compare & branch (immediate)
 143 //   1c)  01___1 Test & branch (immediate)
 144 //   1d)  10___0 Conditional branch (immediate)
 145 //        other  Should not happen
 146 //
 147 // 2) ... xxx1x0 Loads and Stores
 148 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 149 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 150 //                strictly should be 64 bit non-FP/SIMD i.e.
 151 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 152 //
 153 // 3) ... xx100x Data Processing Immediate
 154 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 155 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 156 //                 strictly should be 64 bit movz #imm16<<0
 157 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 158 //
 159 
 160 static uint32_t insn_at(address insn_addr, int n) {
 161   return ((uint32_t*)insn_addr)[n];
 162 }
 163 
 164 template<typename T>
 165 class RelocActions : public AllStatic {
 166 
 167 public:
 168 
 169   static int ALWAYSINLINE run(address insn_addr, address &target) {
 170     int instructions = 1;
 171     uint32_t insn = insn_at(insn_addr, 0);
 172 
 173     uint32_t dispatch = Instruction_aarch64::extract(insn, 30, 25);
 174     switch(dispatch) {
 175       case 0b001010:
 176       case 0b001011: {
 177         instructions = T::unconditionalBranch(insn_addr, target);
 178         break;
 179       }
 180       case 0b101010:   // Conditional branch (immediate)
 181       case 0b011010: { // Compare & branch (immediate)
 182         instructions = T::conditionalBranch(insn_addr, target);
 183         break;
 184       }
 185       case 0b011011: {
 186         instructions = T::testAndBranch(insn_addr, target);
 187         break;
 188       }
 189       case 0b001100:
 190       case 0b001110:
 191       case 0b011100:
 192       case 0b011110:
 193       case 0b101100:
 194       case 0b101110:
 195       case 0b111100:
 196       case 0b111110: {
 197         // load/store
 198         if ((Instruction_aarch64::extract(insn, 29, 24) & 0b111011) == 0b011000) {
 199           // Load register (literal)
 200           instructions = T::loadStore(insn_addr, target);
 201           break;
 202         } else {
 203           // nothing to do
 204           assert(target == nullptr, "did not expect to relocate target for polling page load");
 205         }
 206         break;
 207       }
 208       case 0b001000:
 209       case 0b011000:
 210       case 0b101000:
 211       case 0b111000: {
 212         // adr/adrp
 213         assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 214         int shift = Instruction_aarch64::extract(insn, 31, 31);
 215         if (shift) {
 216           uint32_t insn2 = insn_at(insn_addr, 1);
 217           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 218               Instruction_aarch64::extract(insn, 4, 0) ==
 219               Instruction_aarch64::extract(insn2, 9, 5)) {
 220             instructions = T::adrp(insn_addr, target, T::adrpMem);
 221           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 222                      Instruction_aarch64::extract(insn, 4, 0) ==
 223                      Instruction_aarch64::extract(insn2, 4, 0)) {
 224             instructions = T::adrp(insn_addr, target, T::adrpAdd);
 225           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 226                      Instruction_aarch64::extract(insn, 4, 0) ==
 227                      Instruction_aarch64::extract(insn2, 4, 0)) {
 228             instructions = T::adrp(insn_addr, target, T::adrpMovk);
 229           } else {
 230             ShouldNotReachHere();
 231           }
 232         } else {
 233           instructions = T::adr(insn_addr, target);
 234         }
 235         break;
 236       }
 237       case 0b001001:
 238       case 0b011001:
 239       case 0b101001:
 240       case 0b111001: {
 241         instructions = T::immediate(insn_addr, target);
 242         break;
 243       }
 244       default: {
 245         ShouldNotReachHere();
 246       }
 247     }
 248 
 249     T::verify(insn_addr, target);
 250     return instructions * NativeInstruction::instruction_size;
 251   }
 252 };
 253 
 254 class Patcher : public AllStatic {
 255 public:
 256   static int unconditionalBranch(address insn_addr, address &target) {
 257     intptr_t offset = (target - insn_addr) >> 2;
 258     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 259     return 1;
 260   }
 261   static int conditionalBranch(address insn_addr, address &target) {
 262     intptr_t offset = (target - insn_addr) >> 2;
 263     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 264     return 1;
 265   }
 266   static int testAndBranch(address insn_addr, address &target) {
 267     intptr_t offset = (target - insn_addr) >> 2;
 268     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 269     return 1;
 270   }
 271   static int loadStore(address insn_addr, address &target) {
 272     intptr_t offset = (target - insn_addr) >> 2;
 273     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 274     return 1;
 275   }
 276   static int adr(address insn_addr, address &target) {
 277 #ifdef ASSERT
 278     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 279 #endif
 280     // PC-rel. addressing
 281     ptrdiff_t offset = target - insn_addr;
 282     int offset_lo = offset & 3;
 283     offset >>= 2;
 284     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 285     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 286     return 1;
 287   }
 288   template<typename U>
 289   static int adrp(address insn_addr, address &target, U inner) {
 290     int instructions = 1;
 291 #ifdef ASSERT
 292     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 293 #endif
 294     ptrdiff_t offset = target - insn_addr;
 295     instructions = 2;
 296     precond(inner != nullptr);
 297     // Give the inner reloc a chance to modify the target.
 298     address adjusted_target = target;
 299     instructions = inner(insn_addr, adjusted_target);
 300     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 301     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 302     offset = adr_page - pc_page;
 303     int offset_lo = offset & 3;
 304     offset >>= 2;
 305     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 306     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 307     return instructions;
 308   }
 309   static int adrpMem(address insn_addr, address &target) {
 310     uintptr_t dest = (uintptr_t)target;
 311     int offset_lo = dest & 0xfff;
 312     uint32_t insn2 = insn_at(insn_addr, 1);
 313     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 314     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 315     guarantee(((dest >> size) << size) == dest, "misaligned target");
 316     return 2;
 317   }
 318   static int adrpAdd(address insn_addr, address &target) {
 319     uintptr_t dest = (uintptr_t)target;
 320     int offset_lo = dest & 0xfff;
 321     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 322     return 2;
 323   }
 324   static int adrpMovk(address insn_addr, address &target) {
 325     uintptr_t dest = uintptr_t(target);
 326     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 327     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 328     target = address(dest);
 329     return 2;
 330   }
 331   static int immediate(address insn_addr, address &target) {
 332     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010100, "must be");
 333     uint64_t dest = (uint64_t)target;
 334     // Move wide constant
 335     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 336     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 337     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 338     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 339     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 340     return 3;
 341   }
 342   static void verify(address insn_addr, address &target) {
 343 #ifdef ASSERT
 344     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 345     if (!(address_is == target)) {
 346       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 347       disnm((intptr_t)insn_addr);
 348       assert(address_is == target, "should be");
 349     }
 350 #endif
 351   }
 352 };
 353 
 354 // If insn1 and insn2 use the same register to form an address, either
 355 // by an offsetted LDR or a simple ADD, return the offset. If the
 356 // second instruction is an LDR, the offset may be scaled.
 357 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 358   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 359       Instruction_aarch64::extract(insn1, 4, 0) ==
 360       Instruction_aarch64::extract(insn2, 9, 5)) {
 361     // Load/store register (unsigned immediate)
 362     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 363     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 364     byte_offset <<= size;
 365     return true;
 366   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 367              Instruction_aarch64::extract(insn1, 4, 0) ==
 368              Instruction_aarch64::extract(insn2, 4, 0)) {
 369     // add (immediate)
 370     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 371     return true;
 372   }
 373   return false;
 374 }
 375 
 376 class AArch64Decoder : public AllStatic {
 377 public:
 378 
 379   static int loadStore(address insn_addr, address &target) {
 380     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 381     target = insn_addr + (offset << 2);
 382     return 1;
 383   }
 384   static int unconditionalBranch(address insn_addr, address &target) {
 385     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 25, 0);
 386     target = insn_addr + (offset << 2);
 387     return 1;
 388   }
 389   static int conditionalBranch(address insn_addr, address &target) {
 390     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 391     target = address(((uint64_t)insn_addr + (offset << 2)));
 392     return 1;
 393   }
 394   static int testAndBranch(address insn_addr, address &target) {
 395     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 18, 5);
 396     target = address(((uint64_t)insn_addr + (offset << 2)));
 397     return 1;
 398   }
 399   static int adr(address insn_addr, address &target) {
 400     // PC-rel. addressing
 401     uint32_t insn = insn_at(insn_addr, 0);
 402     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 403     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 404     target = address((uint64_t)insn_addr + offset);
 405     return 1;
 406   }
 407   template<typename U>
 408   static int adrp(address insn_addr, address &target, U inner) {
 409     uint32_t insn = insn_at(insn_addr, 0);
 410     assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 411     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 412     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 413     int shift = 12;
 414     offset <<= shift;
 415     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 416     target_page &= ((uint64_t)-1) << shift;
 417     target = address(target_page);
 418     precond(inner != nullptr);
 419     inner(insn_addr, target);
 420     return 2;
 421   }
 422   static int adrpMem(address insn_addr, address &target) {
 423     uint32_t insn2 = insn_at(insn_addr, 1);
 424     // Load/store register (unsigned immediate)
 425     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 426     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 427     byte_offset <<= size;
 428     target += byte_offset;
 429     return 2;
 430   }
 431   static int adrpAdd(address insn_addr, address &target) {
 432     uint32_t insn2 = insn_at(insn_addr, 1);
 433     // add (immediate)
 434     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 435     target += byte_offset;
 436     return 2;
 437   }
 438   static int adrpMovk(address insn_addr, address &target) {
 439     uint32_t insn2 = insn_at(insn_addr, 1);
 440     uint64_t dest = uint64_t(target);
 441     dest = (dest & 0xffff0000ffffffff) |
 442       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 443     target = address(dest);
 444 
 445     // We know the destination 4k page. Maybe we have a third
 446     // instruction.
 447     uint32_t insn = insn_at(insn_addr, 0);
 448     uint32_t insn3 = insn_at(insn_addr, 2);
 449     ptrdiff_t byte_offset;
 450     if (offset_for(insn, insn3, byte_offset)) {
 451       target += byte_offset;
 452       return 3;
 453     } else {
 454       return 2;
 455     }
 456   }
 457   static int immediate(address insn_addr, address &target) {
 458     uint32_t *insns = (uint32_t *)insn_addr;
 459     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 460     // Move wide constant: movz, movk, movk.  See movptr().
 461     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 462     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 463     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 464                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 465                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 466     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 467     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 468     return 3;
 469   }
 470   static void verify(address insn_addr, address &target) {
 471   }
 472 };
 473 
 474 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 475   address target;
 476   RelocActions<AArch64Decoder>::run(insn_addr, target);
 477   return target;
 478 }
 479 
 480 // Patch any kind of instruction; there may be several instructions.
 481 // Return the total length (in bytes) of the instructions.
 482 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 483   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 484   return RelocActions<Patcher>::run(insn_addr, target);
 485 }
 486 
 487 int MacroAssembler::patch_oop(address insn_addr, address o) {
 488   int instructions;
 489   unsigned insn = *(unsigned*)insn_addr;
 490   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 491 
 492   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 493 
 494   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 495   // narrow OOPs by setting the upper 16 bits in the first
 496   // instruction.
 497   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 498     // Move narrow OOP
 499     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 500     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 501     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 502     instructions = 2;
 503   } else {
 504     // Move wide OOP
 505     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 506     uintptr_t dest = (uintptr_t)o;
 507     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 508     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 509     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 510     instructions = 3;
 511   }
 512   return instructions * NativeInstruction::instruction_size;
 513 }
 514 
 515 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 516   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 517   if (at_return) {
 518     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 519     // we may safely use the sp instead to perform the stack watermark check.
 520     cmp(in_nmethod ? sp : rfp, tmp);
 521     br(Assembler::HI, slow_path);
 522   } else {
 523     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 524   }
 525 }
 526 
 527 void MacroAssembler::rt_call(address dest, Register tmp) {
 528   CodeBlob *cb = CodeCache::find_blob(dest);
 529   if (cb) {
 530     far_call(RuntimeAddress(dest));
 531   } else {
 532     lea(tmp, RuntimeAddress(dest));
 533     blr(tmp);
 534   }
 535 }
 536 
 537 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 538   if (!Continuations::enabled()) return;
 539   Label done;
 540   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 541   cmp(sp, rscratch1);
 542   br(Assembler::LS, done);
 543   mov(rscratch1, sp); // we can't use sp as the source in str
 544   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 545   bind(done);
 546 }
 547 
 548 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 549   if (!Continuations::enabled()) return;
 550   Label done;
 551   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 552   cmp(sp, rscratch1);
 553   br(Assembler::LO, done);
 554   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 555   bind(done);
 556 }
 557 
 558 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 559   // we must set sp to zero to clear frame
 560   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 561 
 562   // must clear fp, so that compiled frames are not confused; it is
 563   // possible that we need it only for debugging
 564   if (clear_fp) {
 565     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 566   }
 567 
 568   // Always clear the pc because it could have been set by make_walkable()
 569   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 570 }
 571 
 572 // Calls to C land
 573 //
 574 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 575 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 576 // has to be reset to 0. This is required to allow proper stack traversal.
 577 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 578                                          Register last_java_fp,
 579                                          Register last_java_pc,
 580                                          Register scratch) {
 581 
 582   if (last_java_pc->is_valid()) {
 583       str(last_java_pc, Address(rthread,
 584                                 JavaThread::frame_anchor_offset()
 585                                 + JavaFrameAnchor::last_Java_pc_offset()));
 586     }
 587 
 588   // determine last_java_sp register
 589   if (last_java_sp == sp) {
 590     mov(scratch, sp);
 591     last_java_sp = scratch;
 592   } else if (!last_java_sp->is_valid()) {
 593     last_java_sp = esp;
 594   }
 595 
 596   // last_java_fp is optional
 597   if (last_java_fp->is_valid()) {
 598     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 599   }
 600 
 601   // We must set sp last.
 602   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 603 }
 604 
 605 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 606                                          Register last_java_fp,
 607                                          address  last_java_pc,
 608                                          Register scratch) {
 609   assert(last_java_pc != nullptr, "must provide a valid PC");
 610 
 611   adr(scratch, last_java_pc);
 612   str(scratch, Address(rthread,
 613                        JavaThread::frame_anchor_offset()
 614                        + JavaFrameAnchor::last_Java_pc_offset()));
 615 
 616   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 617 }
 618 
 619 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 620                                          Register last_java_fp,
 621                                          Label &L,
 622                                          Register scratch) {
 623   if (L.is_bound()) {
 624     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 625   } else {
 626     InstructionMark im(this);
 627     L.add_patch_at(code(), locator());
 628     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 629   }
 630 }
 631 
 632 bool MacroAssembler::target_needs_far_branch(address addr) {
 633   if (AOTCodeCache::is_on_for_dump()) {
 634     return true;
 635   }
 636   if (!far_branches()) {
 637     return false;
 638   }
 639   if (CodeCache::is_non_nmethod(addr) &&
 640       CodeCache::max_distance_to_non_nmethod() <= branch_range) {
 641     return false;
 642   }
 643   return true;
 644 }
 645 
 646 void MacroAssembler::far_call(Address entry, Register tmp) {
 647   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 648   assert(CodeCache::find_blob(entry.target()) != nullptr,
 649          "destination of far call not found in code cache");
 650   assert(entry.rspec().type() == relocInfo::external_word_type
 651          || entry.rspec().type() == relocInfo::runtime_call_type
 652          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 653   if (target_needs_far_branch(entry.target())) {
 654     uint64_t offset;
 655     // We can use ADRP here because we know that the total size of
 656     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 657     adrp(tmp, entry, offset);
 658     add(tmp, tmp, offset);
 659     blr(tmp);
 660   } else {
 661     bl(entry);
 662   }
 663 }
 664 
 665 int MacroAssembler::far_jump(Address entry, Register tmp) {
 666   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 667   assert(CodeCache::find_blob(entry.target()) != nullptr,
 668          "destination of far call not found in code cache");
 669   assert(entry.rspec().type() == relocInfo::external_word_type
 670          || entry.rspec().type() == relocInfo::runtime_call_type
 671          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 672   address start = pc();
 673   if (target_needs_far_branch(entry.target())) {
 674     uint64_t offset;
 675     // We can use ADRP here because we know that the total size of
 676     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 677     adrp(tmp, entry, offset);
 678     add(tmp, tmp, offset);
 679     br(tmp);
 680   } else {
 681     b(entry);
 682   }
 683   return pc() - start;
 684 }
 685 
 686 void MacroAssembler::reserved_stack_check() {
 687     // testing if reserved zone needs to be enabled
 688     Label no_reserved_zone_enabling;
 689 
 690     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 691     cmp(sp, rscratch1);
 692     br(Assembler::LO, no_reserved_zone_enabling);
 693 
 694     enter();   // LR and FP are live.
 695     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 696     mov(c_rarg0, rthread);
 697     blr(rscratch1);
 698     leave();
 699 
 700     // We have already removed our own frame.
 701     // throw_delayed_StackOverflowError will think that it's been
 702     // called by our caller.
 703     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 704     br(rscratch1);
 705     should_not_reach_here();
 706 
 707     bind(no_reserved_zone_enabling);
 708 }
 709 
 710 static void pass_arg0(MacroAssembler* masm, Register arg) {
 711   if (c_rarg0 != arg ) {
 712     masm->mov(c_rarg0, arg);
 713   }
 714 }
 715 
 716 static void pass_arg1(MacroAssembler* masm, Register arg) {
 717   if (c_rarg1 != arg ) {
 718     masm->mov(c_rarg1, arg);
 719   }
 720 }
 721 
 722 static void pass_arg2(MacroAssembler* masm, Register arg) {
 723   if (c_rarg2 != arg ) {
 724     masm->mov(c_rarg2, arg);
 725   }
 726 }
 727 
 728 static void pass_arg3(MacroAssembler* masm, Register arg) {
 729   if (c_rarg3 != arg ) {
 730     masm->mov(c_rarg3, arg);
 731   }
 732 }
 733 
 734 void MacroAssembler::call_VM_base(Register oop_result,
 735                                   Register java_thread,
 736                                   Register last_java_sp,
 737                                   Label*   return_pc,
 738                                   address  entry_point,
 739                                   int      number_of_arguments,
 740                                   bool     check_exceptions) {
 741    // determine java_thread register
 742   if (!java_thread->is_valid()) {
 743     java_thread = rthread;
 744   }
 745 
 746   // determine last_java_sp register
 747   if (!last_java_sp->is_valid()) {
 748     last_java_sp = esp;
 749   }
 750 
 751   // debugging support
 752   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 753   assert(java_thread == rthread, "unexpected register");
 754 #ifdef ASSERT
 755   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 756   // if (!TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 757 #endif // ASSERT
 758 
 759   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 760   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 761 
 762   // push java thread (becomes first argument of C function)
 763 
 764   mov(c_rarg0, java_thread);
 765 
 766   // set last Java frame before call
 767   assert(last_java_sp != rfp, "can't use rfp");
 768 
 769   Label l;
 770   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 771 
 772   // do the call, remove parameters
 773   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 774 
 775   // lr could be poisoned with PAC signature during throw_pending_exception
 776   // if it was tail-call optimized by compiler, since lr is not callee-saved
 777   // reload it with proper value
 778   adr(lr, l);
 779 
 780   // reset last Java frame
 781   // Only interpreter should have to clear fp
 782   reset_last_Java_frame(true);
 783 
 784    // C++ interp handles this in the interpreter
 785   check_and_handle_popframe(java_thread);
 786   check_and_handle_earlyret(java_thread);
 787 
 788   if (check_exceptions) {
 789     // check for pending exceptions (java_thread is set upon return)
 790     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 791     Label ok;
 792     cbz(rscratch1, ok);
 793     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 794     br(rscratch1);
 795     bind(ok);
 796   }
 797 
 798   // get oop result if there is one and reset the value in the thread
 799   if (oop_result->is_valid()) {
 800     get_vm_result_oop(oop_result, java_thread);
 801   }
 802 }
 803 
 804 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 805   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 806 }
 807 
 808 // Check the entry target is always reachable from any branch.
 809 static bool is_always_within_branch_range(Address entry) {
 810   if (AOTCodeCache::is_on_for_dump()) {
 811     return false;
 812   }
 813   const address target = entry.target();
 814 
 815   if (!CodeCache::contains(target)) {
 816     // We always use trampolines for callees outside CodeCache.
 817     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 818     return false;
 819   }
 820 
 821   if (!MacroAssembler::far_branches()) {
 822     return true;
 823   }
 824 
 825   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 826     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 827     // Non-compiled methods stay forever in CodeCache.
 828     // We check whether the longest possible branch is within the branch range.
 829     assert(CodeCache::find_blob(target) != nullptr &&
 830           !CodeCache::find_blob(target)->is_nmethod(),
 831           "runtime call of compiled method");
 832     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 833     const address left_longest_branch_start = CodeCache::low_bound();
 834     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 835                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 836     return is_reachable;
 837   }
 838 
 839   return false;
 840 }
 841 
 842 // Maybe emit a call via a trampoline. If the code cache is small
 843 // trampolines won't be emitted.
 844 address MacroAssembler::trampoline_call(Address entry) {
 845   assert(entry.rspec().type() == relocInfo::runtime_call_type
 846          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 847          || entry.rspec().type() == relocInfo::static_call_type
 848          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 849 
 850   address target = entry.target();
 851 
 852   if (!is_always_within_branch_range(entry)) {
 853     if (!in_scratch_emit_size()) {
 854       // We don't want to emit a trampoline if C2 is generating dummy
 855       // code during its branch shortening phase.
 856       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 857         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 858         code()->share_trampoline_for(entry.target(), offset());
 859       } else {
 860         address stub = emit_trampoline_stub(offset(), target);
 861         if (stub == nullptr) {
 862           postcond(pc() == badAddress);
 863           return nullptr; // CodeCache is full
 864         }
 865       }
 866     }
 867     target = pc();
 868   }
 869 
 870   address call_pc = pc();
 871   relocate(entry.rspec());
 872   bl(target);
 873 
 874   postcond(pc() != badAddress);
 875   return call_pc;
 876 }
 877 
 878 // Emit a trampoline stub for a call to a target which is too far away.
 879 //
 880 // code sequences:
 881 //
 882 // call-site:
 883 //   branch-and-link to <destination> or <trampoline stub>
 884 //
 885 // Related trampoline stub for this call site in the stub section:
 886 //   load the call target from the constant pool
 887 //   branch (LR still points to the call site above)
 888 
 889 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 890                                              address dest) {
 891   // Max stub size: alignment nop, TrampolineStub.
 892   address stub = start_a_stub(max_trampoline_stub_size());
 893   if (stub == nullptr) {
 894     return nullptr;  // CodeBuffer::expand failed
 895   }
 896 
 897   // Create a trampoline stub relocation which relates this trampoline stub
 898   // with the call instruction at insts_call_instruction_offset in the
 899   // instructions code-section.
 900   align(wordSize);
 901   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 902                                             + insts_call_instruction_offset));
 903   const int stub_start_offset = offset();
 904 
 905   // Now, create the trampoline stub's code:
 906   // - load the call
 907   // - call
 908   Label target;
 909   ldr(rscratch1, target);
 910   br(rscratch1);
 911   bind(target);
 912   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 913          "should be");
 914   emit_int64((int64_t)dest);
 915 
 916   const address stub_start_addr = addr_at(stub_start_offset);
 917 
 918   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 919 
 920   end_a_stub();
 921   return stub_start_addr;
 922 }
 923 
 924 int MacroAssembler::max_trampoline_stub_size() {
 925   // Max stub size: alignment nop, TrampolineStub.
 926   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 927 }
 928 
 929 void MacroAssembler::emit_static_call_stub() {
 930   // CompiledDirectCall::set_to_interpreted knows the
 931   // exact layout of this stub.
 932 
 933   isb();
 934   mov_metadata(rmethod, nullptr);
 935 
 936   // Jump to the entry point of the c2i stub.
 937   if (codestub_branch_needs_far_jump()) {
 938     movptr(rscratch1, 0);
 939     br(rscratch1);
 940   } else {
 941     b(pc());
 942   }
 943 }
 944 
 945 int MacroAssembler::static_call_stub_size() {
 946   // During AOT production run AOT and JIT compiled code
 947   // are used at the same time. We need this size
 948   // to be the same for both types of code.
 949   if (!codestub_branch_needs_far_jump() && !AOTCodeCache::is_on_for_use()) {
 950     // isb; movk; movz; movz; b
 951     return 5 * NativeInstruction::instruction_size;
 952   }
 953   // isb; movk; movz; movz; movk; movz; movz; br
 954   return 8 * NativeInstruction::instruction_size;
 955 }
 956 
 957 void MacroAssembler::c2bool(Register x) {
 958   // implements x == 0 ? 0 : 1
 959   // note: must only look at least-significant byte of x
 960   //       since C-style booleans are stored in one byte
 961   //       only! (was bug)
 962   tst(x, 0xff);
 963   cset(x, Assembler::NE);
 964 }
 965 
 966 address MacroAssembler::ic_call(address entry, jint method_index) {
 967   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 968   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 969   return trampoline_call(Address(entry, rh));
 970 }
 971 
 972 int MacroAssembler::ic_check_size() {
 973   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 974   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 975     return NativeInstruction::instruction_size * (7 + extra_instructions);
 976   } else {
 977     return NativeInstruction::instruction_size * (5 + extra_instructions);
 978   }
 979 }
 980 
 981 int MacroAssembler::ic_check(int end_alignment) {
 982   Register receiver = j_rarg0;
 983   Register data = rscratch2;
 984   Register tmp1 = rscratch1;
 985   Register tmp2 = r10;
 986 
 987   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
 988   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
 989   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
 990   // before the inline cache check here, and not after
 991   align(end_alignment, offset() + ic_check_size());
 992 
 993   int uep_offset = offset();
 994 
 995   if (UseCompactObjectHeaders) {
 996     load_narrow_klass_compact(tmp1, receiver);
 997     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
 998     cmpw(tmp1, tmp2);
 999   } else {
1000     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1001     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1002     cmpw(tmp1, tmp2);
1003   }
1004 
1005   Label dont;
1006   br(Assembler::EQ, dont);
1007   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1008   bind(dont);
1009   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1010 
1011   return uep_offset;
1012 }
1013 
1014 // Implementation of call_VM versions
1015 
1016 void MacroAssembler::call_VM(Register oop_result,
1017                              address entry_point,
1018                              bool check_exceptions) {
1019   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1020 }
1021 
1022 void MacroAssembler::call_VM(Register oop_result,
1023                              address entry_point,
1024                              Register arg_1,
1025                              bool check_exceptions) {
1026   pass_arg1(this, arg_1);
1027   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1028 }
1029 
1030 void MacroAssembler::call_VM(Register oop_result,
1031                              address entry_point,
1032                              Register arg_1,
1033                              Register arg_2,
1034                              bool check_exceptions) {
1035   assert_different_registers(arg_1, c_rarg2);
1036   pass_arg2(this, arg_2);
1037   pass_arg1(this, arg_1);
1038   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1039 }
1040 
1041 void MacroAssembler::call_VM(Register oop_result,
1042                              address entry_point,
1043                              Register arg_1,
1044                              Register arg_2,
1045                              Register arg_3,
1046                              bool check_exceptions) {
1047   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1048   assert_different_registers(arg_2, c_rarg3);
1049   pass_arg3(this, arg_3);
1050 
1051   pass_arg2(this, arg_2);
1052 
1053   pass_arg1(this, arg_1);
1054   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1055 }
1056 
1057 void MacroAssembler::call_VM(Register oop_result,
1058                              Register last_java_sp,
1059                              address entry_point,
1060                              int number_of_arguments,
1061                              bool check_exceptions) {
1062   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1063 }
1064 
1065 void MacroAssembler::call_VM(Register oop_result,
1066                              Register last_java_sp,
1067                              address entry_point,
1068                              Register arg_1,
1069                              bool check_exceptions) {
1070   pass_arg1(this, arg_1);
1071   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1072 }
1073 
1074 void MacroAssembler::call_VM(Register oop_result,
1075                              Register last_java_sp,
1076                              address entry_point,
1077                              Register arg_1,
1078                              Register arg_2,
1079                              bool check_exceptions) {
1080 
1081   assert_different_registers(arg_1, c_rarg2);
1082   pass_arg2(this, arg_2);
1083   pass_arg1(this, arg_1);
1084   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1085 }
1086 
1087 void MacroAssembler::call_VM(Register oop_result,
1088                              Register last_java_sp,
1089                              address entry_point,
1090                              Register arg_1,
1091                              Register arg_2,
1092                              Register arg_3,
1093                              bool check_exceptions) {
1094   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1095   assert_different_registers(arg_2, c_rarg3);
1096   pass_arg3(this, arg_3);
1097   pass_arg2(this, arg_2);
1098   pass_arg1(this, arg_1);
1099   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1100 }
1101 
1102 
1103 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1104   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1105   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1106   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1107 }
1108 
1109 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1110   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1111   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1112 }
1113 
1114 void MacroAssembler::align(int modulus) {
1115   align(modulus, offset());
1116 }
1117 
1118 // Ensure that the code at target bytes offset from the current offset() is aligned
1119 // according to modulus.
1120 void MacroAssembler::align(int modulus, int target) {
1121   int delta = target - offset();
1122   while ((offset() + delta) % modulus != 0) nop();
1123 }
1124 
1125 void MacroAssembler::post_call_nop() {
1126   if (!Continuations::enabled()) {
1127     return;
1128   }
1129   InstructionMark im(this);
1130   relocate(post_call_nop_Relocation::spec());
1131   InlineSkippedInstructionsCounter skipCounter(this);
1132   nop();
1133   movk(zr, 0);
1134   movk(zr, 0);
1135 }
1136 
1137 // these are no-ops overridden by InterpreterMacroAssembler
1138 
1139 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1140 
1141 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1142 
1143 // Look up the method for a megamorphic invokeinterface call.
1144 // The target method is determined by <intf_klass, itable_index>.
1145 // The receiver klass is in recv_klass.
1146 // On success, the result will be in method_result, and execution falls through.
1147 // On failure, execution transfers to the given label.
1148 void MacroAssembler::lookup_interface_method(Register recv_klass,
1149                                              Register intf_klass,
1150                                              RegisterOrConstant itable_index,
1151                                              Register method_result,
1152                                              Register scan_temp,
1153                                              Label& L_no_such_interface,
1154                          bool return_method) {
1155   assert_different_registers(recv_klass, intf_klass, scan_temp);
1156   assert_different_registers(method_result, intf_klass, scan_temp);
1157   assert(recv_klass != method_result || !return_method,
1158      "recv_klass can be destroyed when method isn't needed");
1159   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1160          "caller must use same register for non-constant itable index as for method");
1161 
1162   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1163   int vtable_base = in_bytes(Klass::vtable_start_offset());
1164   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1165   int scan_step   = itableOffsetEntry::size() * wordSize;
1166   int vte_size    = vtableEntry::size_in_bytes();
1167   assert(vte_size == wordSize, "else adjust times_vte_scale");
1168 
1169   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1170 
1171   // Could store the aligned, prescaled offset in the klass.
1172   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1173   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1174   add(scan_temp, scan_temp, vtable_base);
1175 
1176   if (return_method) {
1177     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1178     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1179     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1180     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1181     if (itentry_off)
1182       add(recv_klass, recv_klass, itentry_off);
1183   }
1184 
1185   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1186   //   if (scan->interface() == intf) {
1187   //     result = (klass + scan->offset() + itable_index);
1188   //   }
1189   // }
1190   Label search, found_method;
1191 
1192   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1193   cmp(intf_klass, method_result);
1194   br(Assembler::EQ, found_method);
1195   bind(search);
1196   // Check that the previous entry is non-null.  A null entry means that
1197   // the receiver class doesn't implement the interface, and wasn't the
1198   // same as when the caller was compiled.
1199   cbz(method_result, L_no_such_interface);
1200   if (itableOffsetEntry::interface_offset() != 0) {
1201     add(scan_temp, scan_temp, scan_step);
1202     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1203   } else {
1204     ldr(method_result, Address(pre(scan_temp, scan_step)));
1205   }
1206   cmp(intf_klass, method_result);
1207   br(Assembler::NE, search);
1208 
1209   bind(found_method);
1210 
1211   // Got a hit.
1212   if (return_method) {
1213     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1214     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1215   }
1216 }
1217 
1218 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1219 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1220 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1221 // The target method is determined by <holder_klass, itable_index>.
1222 // The receiver klass is in recv_klass.
1223 // On success, the result will be in method_result, and execution falls through.
1224 // On failure, execution transfers to the given label.
1225 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1226                                                   Register holder_klass,
1227                                                   Register resolved_klass,
1228                                                   Register method_result,
1229                                                   Register temp_itbl_klass,
1230                                                   Register scan_temp,
1231                                                   int itable_index,
1232                                                   Label& L_no_such_interface) {
1233   // 'method_result' is only used as output register at the very end of this method.
1234   // Until then we can reuse it as 'holder_offset'.
1235   Register holder_offset = method_result;
1236   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1237 
1238   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1239   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1240   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1241   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1242 
1243   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1244 
1245   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1246   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1247   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1248   // temp_itbl_klass = itable[0]._interface;
1249   int vtblEntrySize = vtableEntry::size_in_bytes();
1250   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1251   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1252   mov(holder_offset, zr);
1253   // scan_temp = &(itable[0]._interface)
1254   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1255 
1256   // Initial checks:
1257   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1258   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1259   //   - if (itable[0] == 0), no such interface
1260   cmp(resolved_klass, holder_klass);
1261   br(Assembler::NE, L_loop_search_resolved_entry);
1262   cmp(holder_klass, temp_itbl_klass);
1263   br(Assembler::EQ, L_holder_found);
1264   cbz(temp_itbl_klass, L_no_such_interface);
1265 
1266   // Loop: Look for holder_klass record in itable
1267   //   do {
1268   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1269   //     if (temp_itbl_klass == holder_klass) {
1270   //       goto L_holder_found; // Found!
1271   //     }
1272   //   } while (temp_itbl_klass != 0);
1273   //   goto L_no_such_interface // Not found.
1274   Label L_search_holder;
1275   bind(L_search_holder);
1276     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1277     cmp(holder_klass, temp_itbl_klass);
1278     br(Assembler::EQ, L_holder_found);
1279     cbnz(temp_itbl_klass, L_search_holder);
1280 
1281   b(L_no_such_interface);
1282 
1283   // Loop: Look for resolved_class record in itable
1284   //   while (true) {
1285   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1286   //     if (temp_itbl_klass == 0) {
1287   //       goto L_no_such_interface;
1288   //     }
1289   //     if (temp_itbl_klass == resolved_klass) {
1290   //        goto L_resolved_found;  // Found!
1291   //     }
1292   //     if (temp_itbl_klass == holder_klass) {
1293   //        holder_offset = scan_temp;
1294   //     }
1295   //   }
1296   //
1297   Label L_loop_search_resolved;
1298   bind(L_loop_search_resolved);
1299     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1300   bind(L_loop_search_resolved_entry);
1301     cbz(temp_itbl_klass, L_no_such_interface);
1302     cmp(resolved_klass, temp_itbl_klass);
1303     br(Assembler::EQ, L_resolved_found);
1304     cmp(holder_klass, temp_itbl_klass);
1305     br(Assembler::NE, L_loop_search_resolved);
1306     mov(holder_offset, scan_temp);
1307     b(L_loop_search_resolved);
1308 
1309   // See if we already have a holder klass. If not, go and scan for it.
1310   bind(L_resolved_found);
1311   cbz(holder_offset, L_search_holder);
1312   mov(scan_temp, holder_offset);
1313 
1314   // Finally, scan_temp contains holder_klass vtable offset
1315   bind(L_holder_found);
1316   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1317   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1318     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1319   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1320 }
1321 
1322 // virtual method calling
1323 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1324                                            RegisterOrConstant vtable_index,
1325                                            Register method_result) {
1326   assert(vtableEntry::size() * wordSize == 8,
1327          "adjust the scaling in the code below");
1328   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1329 
1330   if (vtable_index.is_register()) {
1331     lea(method_result, Address(recv_klass,
1332                                vtable_index.as_register(),
1333                                Address::lsl(LogBytesPerWord)));
1334     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1335   } else {
1336     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1337     ldr(method_result,
1338         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1339   }
1340 }
1341 
1342 void MacroAssembler::check_klass_subtype(Register sub_klass,
1343                            Register super_klass,
1344                            Register temp_reg,
1345                            Label& L_success) {
1346   Label L_failure;
1347   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1348   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1349   bind(L_failure);
1350 }
1351 
1352 
1353 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1354                                                    Register super_klass,
1355                                                    Register temp_reg,
1356                                                    Label* L_success,
1357                                                    Label* L_failure,
1358                                                    Label* L_slow_path,
1359                                                    Register super_check_offset) {
1360   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1361   bool must_load_sco = ! super_check_offset->is_valid();
1362   if (must_load_sco) {
1363     assert(temp_reg != noreg, "supply either a temp or a register offset");
1364   }
1365 
1366   Label L_fallthrough;
1367   int label_nulls = 0;
1368   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1369   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1370   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1371   assert(label_nulls <= 1, "at most one null in the batch");
1372 
1373   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1374   Address super_check_offset_addr(super_klass, sco_offset);
1375 
1376   // Hacked jmp, which may only be used just before L_fallthrough.
1377 #define final_jmp(label)                                                \
1378   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1379   else                            b(label)                /*omit semi*/
1380 
1381   // If the pointers are equal, we are done (e.g., String[] elements).
1382   // This self-check enables sharing of secondary supertype arrays among
1383   // non-primary types such as array-of-interface.  Otherwise, each such
1384   // type would need its own customized SSA.
1385   // We move this check to the front of the fast path because many
1386   // type checks are in fact trivially successful in this manner,
1387   // so we get a nicely predicted branch right at the start of the check.
1388   cmp(sub_klass, super_klass);
1389   br(Assembler::EQ, *L_success);
1390 
1391   // Check the supertype display:
1392   if (must_load_sco) {
1393     ldrw(temp_reg, super_check_offset_addr);
1394     super_check_offset = temp_reg;
1395   }
1396 
1397   Address super_check_addr(sub_klass, super_check_offset);
1398   ldr(rscratch1, super_check_addr);
1399   cmp(super_klass, rscratch1); // load displayed supertype
1400   br(Assembler::EQ, *L_success);
1401 
1402   // This check has worked decisively for primary supers.
1403   // Secondary supers are sought in the super_cache ('super_cache_addr').
1404   // (Secondary supers are interfaces and very deeply nested subtypes.)
1405   // This works in the same check above because of a tricky aliasing
1406   // between the super_cache and the primary super display elements.
1407   // (The 'super_check_addr' can address either, as the case requires.)
1408   // Note that the cache is updated below if it does not help us find
1409   // what we need immediately.
1410   // So if it was a primary super, we can just fail immediately.
1411   // Otherwise, it's the slow path for us (no success at this point).
1412 
1413   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1414   if (L_failure == &L_fallthrough) {
1415     cbz(rscratch1, *L_slow_path);
1416   } else {
1417     cbnz(rscratch1, *L_failure);
1418     final_jmp(*L_slow_path);
1419   }
1420 
1421   bind(L_fallthrough);
1422 
1423 #undef final_jmp
1424 }
1425 
1426 // These two are taken from x86, but they look generally useful
1427 
1428 // scans count pointer sized words at [addr] for occurrence of value,
1429 // generic
1430 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1431                                 Register scratch) {
1432   Label Lloop, Lexit;
1433   cbz(count, Lexit);
1434   bind(Lloop);
1435   ldr(scratch, post(addr, wordSize));
1436   cmp(value, scratch);
1437   br(EQ, Lexit);
1438   sub(count, count, 1);
1439   cbnz(count, Lloop);
1440   bind(Lexit);
1441 }
1442 
1443 // scans count 4 byte words at [addr] for occurrence of value,
1444 // generic
1445 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1446                                 Register scratch) {
1447   Label Lloop, Lexit;
1448   cbz(count, Lexit);
1449   bind(Lloop);
1450   ldrw(scratch, post(addr, wordSize));
1451   cmpw(value, scratch);
1452   br(EQ, Lexit);
1453   sub(count, count, 1);
1454   cbnz(count, Lloop);
1455   bind(Lexit);
1456 }
1457 
1458 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1459                                                           Register super_klass,
1460                                                           Register temp_reg,
1461                                                           Register temp2_reg,
1462                                                           Label* L_success,
1463                                                           Label* L_failure,
1464                                                           bool set_cond_codes) {
1465   // NB! Callers may assume that, when temp2_reg is a valid register,
1466   // this code sets it to a nonzero value.
1467 
1468   assert_different_registers(sub_klass, super_klass, temp_reg);
1469   if (temp2_reg != noreg)
1470     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1471 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1472 
1473   Label L_fallthrough;
1474   int label_nulls = 0;
1475   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1476   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1477   assert(label_nulls <= 1, "at most one null in the batch");
1478 
1479   // a couple of useful fields in sub_klass:
1480   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1481   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1482   Address secondary_supers_addr(sub_klass, ss_offset);
1483   Address super_cache_addr(     sub_klass, sc_offset);
1484 
1485   BLOCK_COMMENT("check_klass_subtype_slow_path");
1486 
1487   // Do a linear scan of the secondary super-klass chain.
1488   // This code is rarely used, so simplicity is a virtue here.
1489   // The repne_scan instruction uses fixed registers, which we must spill.
1490   // Don't worry too much about pre-existing connections with the input regs.
1491 
1492   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1493   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1494 
1495   RegSet pushed_registers;
1496   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1497   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1498 
1499   if (super_klass != r0) {
1500     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1501   }
1502 
1503   push(pushed_registers, sp);
1504 
1505   // Get super_klass value into r0 (even if it was in r5 or r2).
1506   if (super_klass != r0) {
1507     mov(r0, super_klass);
1508   }
1509 
1510 #ifndef PRODUCT
1511   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1512 #endif //PRODUCT
1513 
1514   // We will consult the secondary-super array.
1515   ldr(r5, secondary_supers_addr);
1516   // Load the array length.
1517   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1518   // Skip to start of data.
1519   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1520 
1521   cmp(sp, zr); // Clear Z flag; SP is never zero
1522   // Scan R2 words at [R5] for an occurrence of R0.
1523   // Set NZ/Z based on last compare.
1524   repne_scan(r5, r0, r2, rscratch1);
1525 
1526   // Unspill the temp. registers:
1527   pop(pushed_registers, sp);
1528 
1529   br(Assembler::NE, *L_failure);
1530 
1531   // Success.  Cache the super we found and proceed in triumph.
1532 
1533   if (UseSecondarySupersCache) {
1534     str(super_klass, super_cache_addr);
1535   }
1536 
1537   if (L_success != &L_fallthrough) {
1538     b(*L_success);
1539   }
1540 
1541 #undef IS_A_TEMP
1542 
1543   bind(L_fallthrough);
1544 }
1545 
1546 // If Register r is invalid, remove a new register from
1547 // available_regs, and add new register to regs_to_push.
1548 Register MacroAssembler::allocate_if_noreg(Register r,
1549                                   RegSetIterator<Register> &available_regs,
1550                                   RegSet &regs_to_push) {
1551   if (!r->is_valid()) {
1552     r = *available_regs++;
1553     regs_to_push += r;
1554   }
1555   return r;
1556 }
1557 
1558 // check_klass_subtype_slow_path_table() looks for super_klass in the
1559 // hash table belonging to super_klass, branching to L_success or
1560 // L_failure as appropriate. This is essentially a shim which
1561 // allocates registers as necessary then calls
1562 // lookup_secondary_supers_table() to do the work. Any of the temp
1563 // regs may be noreg, in which case this logic will chooses some
1564 // registers push and pop them from the stack.
1565 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1566                                                          Register super_klass,
1567                                                          Register temp_reg,
1568                                                          Register temp2_reg,
1569                                                          Register temp3_reg,
1570                                                          Register result_reg,
1571                                                          FloatRegister vtemp,
1572                                                          Label* L_success,
1573                                                          Label* L_failure,
1574                                                          bool set_cond_codes) {
1575   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1576 
1577   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1578 
1579   Label L_fallthrough;
1580   int label_nulls = 0;
1581   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1582   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1583   assert(label_nulls <= 1, "at most one null in the batch");
1584 
1585   BLOCK_COMMENT("check_klass_subtype_slow_path");
1586 
1587   RegSetIterator<Register> available_regs
1588     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1589 
1590   RegSet pushed_regs;
1591 
1592   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1593   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1594   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1595   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1596 
1597   push(pushed_regs, sp);
1598 
1599   lookup_secondary_supers_table_var(sub_klass,
1600                                     super_klass,
1601                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1602                                     nullptr);
1603   cmp(result_reg, zr);
1604 
1605   // Unspill the temp. registers:
1606   pop(pushed_regs, sp);
1607 
1608   // NB! Callers may assume that, when set_cond_codes is true, this
1609   // code sets temp2_reg to a nonzero value.
1610   if (set_cond_codes) {
1611     mov(temp2_reg, 1);
1612   }
1613 
1614   br(Assembler::NE, *L_failure);
1615 
1616   if (L_success != &L_fallthrough) {
1617     b(*L_success);
1618   }
1619 
1620   bind(L_fallthrough);
1621 }
1622 
1623 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1624                                                    Register super_klass,
1625                                                    Register temp_reg,
1626                                                    Register temp2_reg,
1627                                                    Label* L_success,
1628                                                    Label* L_failure,
1629                                                    bool set_cond_codes) {
1630   if (UseSecondarySupersTable) {
1631     check_klass_subtype_slow_path_table
1632       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1633        /*vtemp*/fnoreg,
1634        L_success, L_failure, set_cond_codes);
1635   } else {
1636     check_klass_subtype_slow_path_linear
1637       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1638   }
1639 }
1640 
1641 
1642 // Ensure that the inline code and the stub are using the same registers.
1643 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1644 do {                                                               \
1645   assert(r_super_klass  == r0                                   && \
1646          r_array_base   == r1                                   && \
1647          r_array_length == r2                                   && \
1648          (r_array_index == r3        || r_array_index == noreg) && \
1649          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1650          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1651          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1652 } while(0)
1653 
1654 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1655                                                          Register r_super_klass,
1656                                                          Register temp1,
1657                                                          Register temp2,
1658                                                          Register temp3,
1659                                                          FloatRegister vtemp,
1660                                                          Register result,
1661                                                          u1 super_klass_slot,
1662                                                          bool stub_is_near) {
1663   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1664 
1665   Label L_fallthrough;
1666 
1667   BLOCK_COMMENT("lookup_secondary_supers_table {");
1668 
1669   const Register
1670     r_array_base   = temp1, // r1
1671     r_array_length = temp2, // r2
1672     r_array_index  = temp3, // r3
1673     r_bitmap       = rscratch2;
1674 
1675   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1676 
1677   u1 bit = super_klass_slot;
1678 
1679   // Make sure that result is nonzero if the TBZ below misses.
1680   mov(result, 1);
1681 
1682   // We're going to need the bitmap in a vector reg and in a core reg,
1683   // so load both now.
1684   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1685   if (bit != 0) {
1686     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1687   }
1688   // First check the bitmap to see if super_klass might be present. If
1689   // the bit is zero, we are certain that super_klass is not one of
1690   // the secondary supers.
1691   tbz(r_bitmap, bit, L_fallthrough);
1692 
1693   // Get the first array index that can contain super_klass into r_array_index.
1694   if (bit != 0) {
1695     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1696     cnt(vtemp, T8B, vtemp);
1697     addv(vtemp, T8B, vtemp);
1698     fmovd(r_array_index, vtemp);
1699   } else {
1700     mov(r_array_index, (u1)1);
1701   }
1702   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1703 
1704   // We will consult the secondary-super array.
1705   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1706 
1707   // The value i in r_array_index is >= 1, so even though r_array_base
1708   // points to the length, we don't need to adjust it to point to the
1709   // data.
1710   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1711   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1712 
1713   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1714   eor(result, result, r_super_klass);
1715   cbz(result, L_fallthrough); // Found a match
1716 
1717   // Is there another entry to check? Consult the bitmap.
1718   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1719 
1720   // Linear probe.
1721   if (bit != 0) {
1722     ror(r_bitmap, r_bitmap, bit);
1723   }
1724 
1725   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1726   // The next slot to be inspected, by the stub we're about to call,
1727   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1728   // have been checked.
1729   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1730   if (stub_is_near) {
1731     bl(stub);
1732   } else {
1733     address call = trampoline_call(stub);
1734     if (call == nullptr) {
1735       return false; // trampoline allocation failed
1736     }
1737   }
1738 
1739   BLOCK_COMMENT("} lookup_secondary_supers_table");
1740 
1741   bind(L_fallthrough);
1742 
1743   if (VerifySecondarySupers) {
1744     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1745                                   temp1, temp2, result);      // r1, r2, r5
1746   }
1747   return true;
1748 }
1749 
1750 // At runtime, return 0 in result if r_super_klass is a superclass of
1751 // r_sub_klass, otherwise return nonzero. Use this version of
1752 // lookup_secondary_supers_table() if you don't know ahead of time
1753 // which superclass will be searched for. Used by interpreter and
1754 // runtime stubs. It is larger and has somewhat greater latency than
1755 // the version above, which takes a constant super_klass_slot.
1756 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1757                                                        Register r_super_klass,
1758                                                        Register temp1,
1759                                                        Register temp2,
1760                                                        Register temp3,
1761                                                        FloatRegister vtemp,
1762                                                        Register result,
1763                                                        Label *L_success) {
1764   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1765 
1766   Label L_fallthrough;
1767 
1768   BLOCK_COMMENT("lookup_secondary_supers_table {");
1769 
1770   const Register
1771     r_array_index = temp3,
1772     slot          = rscratch1,
1773     r_bitmap      = rscratch2;
1774 
1775   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1776 
1777   // Make sure that result is nonzero if the test below misses.
1778   mov(result, 1);
1779 
1780   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1781 
1782   // First check the bitmap to see if super_klass might be present. If
1783   // the bit is zero, we are certain that super_klass is not one of
1784   // the secondary supers.
1785 
1786   // This next instruction is equivalent to:
1787   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1788   // sub(temp2, tmp_reg, slot);
1789   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1790   lslv(temp2, r_bitmap, temp2);
1791   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1792 
1793   bool must_save_v0 = (vtemp == fnoreg);
1794   if (must_save_v0) {
1795     // temp1 and result are free, so use them to preserve vtemp
1796     vtemp = v0;
1797     mov(temp1,  vtemp, D, 0);
1798     mov(result, vtemp, D, 1);
1799   }
1800 
1801   // Get the first array index that can contain super_klass into r_array_index.
1802   mov(vtemp, D, 0, temp2);
1803   cnt(vtemp, T8B, vtemp);
1804   addv(vtemp, T8B, vtemp);
1805   mov(r_array_index, vtemp, D, 0);
1806 
1807   if (must_save_v0) {
1808     mov(vtemp, D, 0, temp1 );
1809     mov(vtemp, D, 1, result);
1810   }
1811 
1812   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1813 
1814   const Register
1815     r_array_base   = temp1,
1816     r_array_length = temp2;
1817 
1818   // The value i in r_array_index is >= 1, so even though r_array_base
1819   // points to the length, we don't need to adjust it to point to the
1820   // data.
1821   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1822   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1823 
1824   // We will consult the secondary-super array.
1825   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1826 
1827   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1828   eor(result, result, r_super_klass);
1829   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1830 
1831   // Is there another entry to check? Consult the bitmap.
1832   rorv(r_bitmap, r_bitmap, slot);
1833   // rol(r_bitmap, r_bitmap, 1);
1834   tbz(r_bitmap, 1, L_fallthrough);
1835 
1836   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1837   // The next slot to be inspected, by the logic we're about to call,
1838   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1839   // have been checked.
1840   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1841                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1842 
1843   BLOCK_COMMENT("} lookup_secondary_supers_table");
1844 
1845   bind(L_fallthrough);
1846 
1847   if (VerifySecondarySupers) {
1848     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1849                                   temp1, temp2, result);      // r1, r2, r5
1850   }
1851 
1852   if (L_success) {
1853     cbz(result, *L_success);
1854   }
1855 }
1856 
1857 // Called by code generated by check_klass_subtype_slow_path
1858 // above. This is called when there is a collision in the hashed
1859 // lookup in the secondary supers array.
1860 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1861                                                              Register r_array_base,
1862                                                              Register r_array_index,
1863                                                              Register r_bitmap,
1864                                                              Register temp1,
1865                                                              Register result,
1866                                                              bool is_stub) {
1867   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1868 
1869   const Register
1870     r_array_length = temp1,
1871     r_sub_klass    = noreg; // unused
1872 
1873   if (is_stub) {
1874     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1875   }
1876 
1877   Label L_fallthrough, L_huge;
1878 
1879   // Load the array length.
1880   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1881   // And adjust the array base to point to the data.
1882   // NB! Effectively increments current slot index by 1.
1883   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1884   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1885 
1886   // The bitmap is full to bursting.
1887   // Implicit invariant: BITMAP_FULL implies (length > 0)
1888   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1889   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1890   br(GT, L_huge);
1891 
1892   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1893   // current slot (at secondary_supers[r_array_index]) has not yet
1894   // been inspected, and r_array_index may be out of bounds if we
1895   // wrapped around the end of the array.
1896 
1897   { // This is conventional linear probing, but instead of terminating
1898     // when a null entry is found in the table, we maintain a bitmap
1899     // in which a 0 indicates missing entries.
1900     // As long as the bitmap is not completely full,
1901     // array_length == popcount(bitmap). The array_length check above
1902     // guarantees there are 0s in the bitmap, so the loop eventually
1903     // terminates.
1904     Label L_loop;
1905     bind(L_loop);
1906 
1907     // Check for wraparound.
1908     cmp(r_array_index, r_array_length);
1909     csel(r_array_index, zr, r_array_index, GE);
1910 
1911     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1912     eor(result, rscratch1, r_super_klass);
1913     cbz(result, L_fallthrough);
1914 
1915     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1916 
1917     ror(r_bitmap, r_bitmap, 1);
1918     add(r_array_index, r_array_index, 1);
1919     b(L_loop);
1920   }
1921 
1922   { // Degenerate case: more than 64 secondary supers.
1923     // FIXME: We could do something smarter here, maybe a vectorized
1924     // comparison or a binary search, but is that worth any added
1925     // complexity?
1926     bind(L_huge);
1927     cmp(sp, zr); // Clear Z flag; SP is never zero
1928     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1929     cset(result, NE); // result == 0 iff we got a match.
1930   }
1931 
1932   bind(L_fallthrough);
1933 }
1934 
1935 // Make sure that the hashed lookup and a linear scan agree.
1936 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1937                                                    Register r_super_klass,
1938                                                    Register temp1,
1939                                                    Register temp2,
1940                                                    Register result) {
1941   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1942 
1943   const Register
1944     r_array_base   = temp1,
1945     r_array_length = temp2;
1946 
1947   BLOCK_COMMENT("verify_secondary_supers_table {");
1948 
1949   // We will consult the secondary-super array.
1950   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1951 
1952   // Load the array length.
1953   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1954   // And adjust the array base to point to the data.
1955   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1956 
1957   cmp(sp, zr); // Clear Z flag; SP is never zero
1958   // Scan R2 words at [R5] for an occurrence of R0.
1959   // Set NZ/Z based on last compare.
1960   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1961   // rscratch1 == 0 iff we got a match.
1962   cset(rscratch1, NE);
1963 
1964   Label passed;
1965   cmp(result, zr);
1966   cset(result, NE); // normalize result to 0/1 for comparison
1967 
1968   cmp(rscratch1, result);
1969   br(EQ, passed);
1970   {
1971     mov(r0, r_super_klass);         // r0 <- r0
1972     mov(r1, r_sub_klass);           // r1 <- r4
1973     mov(r2, /*expected*/rscratch1); // r2 <- r8
1974     mov(r3, result);                // r3 <- r5
1975     mov(r4, (address)("mismatch")); // r4 <- const
1976     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
1977     should_not_reach_here();
1978   }
1979   bind(passed);
1980 
1981   BLOCK_COMMENT("} verify_secondary_supers_table");
1982 }
1983 
1984 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
1985   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
1986   assert_different_registers(klass, rthread, scratch);
1987 
1988   Label L_fallthrough, L_tmp;
1989   if (L_fast_path == nullptr) {
1990     L_fast_path = &L_fallthrough;
1991   } else if (L_slow_path == nullptr) {
1992     L_slow_path = &L_fallthrough;
1993   }
1994   // Fast path check: class is fully initialized
1995   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
1996   ldarb(scratch, scratch);
1997   cmp(scratch, InstanceKlass::fully_initialized);
1998   br(Assembler::EQ, *L_fast_path);
1999 
2000   // Fast path check: current thread is initializer thread
2001   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2002   cmp(rthread, scratch);
2003 
2004   if (L_slow_path == &L_fallthrough) {
2005     br(Assembler::EQ, *L_fast_path);
2006     bind(*L_slow_path);
2007   } else if (L_fast_path == &L_fallthrough) {
2008     br(Assembler::NE, *L_slow_path);
2009     bind(*L_fast_path);
2010   } else {
2011     Unimplemented();
2012   }
2013 }
2014 
2015 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2016   if (!VerifyOops || VerifyAdapterSharing) {
2017     // Below address of the code string confuses VerifyAdapterSharing
2018     // because it may differ between otherwise equivalent adapters.
2019     return;
2020   }
2021 
2022   // Pass register number to verify_oop_subroutine
2023   const char* b = nullptr;
2024   {
2025     ResourceMark rm;
2026     stringStream ss;
2027     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2028     b = code_string(ss.as_string());
2029   }
2030   BLOCK_COMMENT("verify_oop {");
2031 
2032   strip_return_address(); // This might happen within a stack frame.
2033   protect_return_address();
2034   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2035   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2036 
2037   mov(r0, reg);
2038   movptr(rscratch1, (uintptr_t)(address)b);
2039 
2040   // call indirectly to solve generation ordering problem
2041   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2042   ldr(rscratch2, Address(rscratch2));
2043   blr(rscratch2);
2044 
2045   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2046   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2047   authenticate_return_address();
2048 
2049   BLOCK_COMMENT("} verify_oop");
2050 }
2051 
2052 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2053   if (!VerifyOops || VerifyAdapterSharing) {
2054     // Below address of the code string confuses VerifyAdapterSharing
2055     // because it may differ between otherwise equivalent adapters.
2056     return;
2057   }
2058 
2059   const char* b = nullptr;
2060   {
2061     ResourceMark rm;
2062     stringStream ss;
2063     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2064     b = code_string(ss.as_string());
2065   }
2066   BLOCK_COMMENT("verify_oop_addr {");
2067 
2068   strip_return_address(); // This might happen within a stack frame.
2069   protect_return_address();
2070   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2071   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2072 
2073   // addr may contain sp so we will have to adjust it based on the
2074   // pushes that we just did.
2075   if (addr.uses(sp)) {
2076     lea(r0, addr);
2077     ldr(r0, Address(r0, 4 * wordSize));
2078   } else {
2079     ldr(r0, addr);
2080   }
2081   movptr(rscratch1, (uintptr_t)(address)b);
2082 
2083   // call indirectly to solve generation ordering problem
2084   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2085   ldr(rscratch2, Address(rscratch2));
2086   blr(rscratch2);
2087 
2088   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2089   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2090   authenticate_return_address();
2091 
2092   BLOCK_COMMENT("} verify_oop_addr");
2093 }
2094 
2095 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2096                                          int extra_slot_offset) {
2097   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2098   int stackElementSize = Interpreter::stackElementSize;
2099   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2100 #ifdef ASSERT
2101   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2102   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2103 #endif
2104   if (arg_slot.is_constant()) {
2105     return Address(esp, arg_slot.as_constant() * stackElementSize
2106                    + offset);
2107   } else {
2108     add(rscratch1, esp, arg_slot.as_register(),
2109         ext::uxtx, exact_log2(stackElementSize));
2110     return Address(rscratch1, offset);
2111   }
2112 }
2113 
2114 // Handle the receiver type profile update given the "recv" klass.
2115 //
2116 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2117 // If there are no matching or claimable receiver entries in RD, updates
2118 // the polymorphic counter.
2119 //
2120 // This code expected to run by either the interpreter or JIT-ed code, without
2121 // extra synchronization. For safety, receiver cells are claimed atomically, which
2122 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2123 // counter updates are not atomic.
2124 //
2125 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2126   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2127 
2128   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2129   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2130   int poly_count_offset      = in_bytes(CounterData::count_offset());
2131   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2132   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2133 
2134   // Adjust for MDP offsets.
2135   base_receiver_offset += mdp_offset;
2136   end_receiver_offset  += mdp_offset;
2137   poly_count_offset    += mdp_offset;
2138 
2139 #ifdef ASSERT
2140   // We are about to walk the MDO slots without asking for offsets.
2141   // Check that our math hits all the right spots.
2142   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2143     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2144     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2145     int offset = base_receiver_offset + receiver_step*c;
2146     int count_offset = offset + receiver_to_count_step;
2147     assert(offset == real_recv_offset, "receiver slot math");
2148     assert(count_offset == real_count_offset, "receiver count math");
2149   }
2150   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2151   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2152 #endif
2153 
2154   // Corner case: no profile table. Increment poly counter and exit.
2155   if (ReceiverTypeData::row_limit() == 0) {
2156     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2157     return;
2158   }
2159 
2160   Register offset = rscratch2;
2161 
2162   Label L_loop_search_receiver, L_loop_search_empty;
2163   Label L_restart, L_found_recv, L_found_empty, L_count_update;
2164 
2165   // The code here recognizes three major cases:
2166   //   A. Fastest: receiver found in the table
2167   //   B. Fast: no receiver in the table, and the table is full
2168   //   C. Slow: no receiver in the table, free slots in the table
2169   //
2170   // The case A performance is most important, as perfectly-behaved code would end up
2171   // there, especially with larger TypeProfileWidth. The case B performance is
2172   // important as well, this is where bulk of code would land for normally megamorphic
2173   // cases. The case C performance is not essential, its job is to deal with installation
2174   // races, we optimize for code density instead. Case C needs to make sure that receiver
2175   // rows are only claimed once. This makes sure we never overwrite a row for another
2176   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2177   //
2178   // It is very tempting to handle these cases in a single loop, and claim the first slot
2179   // without checking the rest of the table. But, profiling code should tolerate free slots
2180   // in the table, as class unloading can clear them. After such cleanup, the receiver
2181   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2182   // to complete, before trying to install new slots. Splitting the code in several tight
2183   // loops also helpfully optimizes for cases A and B.
2184   //
2185   // This code is effectively:
2186   //
2187   // restart:
2188   //   // Fastest: receiver is already installed
2189   //   for (i = 0; i < receiver_count(); i++) {
2190   //     if (receiver(i) == recv) goto found_recv(i);
2191   //   }
2192   //
2193   //   // Fast: no receiver, but profile is not full
2194   //   for (i = 0; i < receiver_count(); i++) {
2195   //     if (receiver(i) == null) goto found_null(i);
2196   //   }
2197   //
2198   //   // Slow: profile is full, polymorphic case
2199   //   count++;
2200   //   return
2201   //
2202   //   // Slow: try to install receiver
2203   // found_null(i):
2204   //   CAS(&receiver(i), null, recv);
2205   //   goto restart
2206   //
2207   // found_recv(i):
2208   //   *receiver_count(i)++
2209   //
2210 
2211   bind(L_restart);
2212 
2213   // Fastest: receiver is already installed
2214   mov(offset, base_receiver_offset);
2215   bind(L_loop_search_receiver);
2216     ldr(rscratch1, Address(mdp, offset));
2217     cmp(rscratch1, recv);
2218     br(Assembler::EQ, L_found_recv);
2219   add(offset, offset, receiver_step);
2220   sub(rscratch1, offset, end_receiver_offset);
2221   cbnz(rscratch1, L_loop_search_receiver);
2222 
2223   // Fast: no receiver, but profile is not full
2224   mov(offset, base_receiver_offset);
2225   bind(L_loop_search_empty);
2226     ldr(rscratch1, Address(mdp, offset));
2227     cbz(rscratch1, L_found_empty);
2228   add(offset, offset, receiver_step);
2229   sub(rscratch1, offset, end_receiver_offset);
2230   cbnz(rscratch1, L_loop_search_empty);
2231 
2232   // Slow: Receiver is not found and table is full.
2233   // Increment polymorphic counter instead of receiver slot.
2234   mov(offset, poly_count_offset);
2235   b(L_count_update);
2236 
2237   // Slowest: try to install receiver
2238   bind(L_found_empty);
2239 
2240   // Atomically swing receiver slot: null -> recv.
2241   //
2242   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2243   // is used to hold the destination address. This is safe because the
2244   // offset is no longer needed after the address is computed.
2245 
2246   lea(rscratch2, Address(mdp, offset));
2247   cmpxchg_weak(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword, memory_order_relaxed);
2248 
2249   // CAS success means the slot now has the receiver we want. CAS failure means
2250   // something had claimed the slot concurrently: it can be the same receiver we want,
2251   // or something else. Since this is a slow path, we can optimize for code density,
2252   // and just restart the search from the beginning.
2253   b(L_restart);
2254 
2255   // Found a receiver, convert its slot offset to corresponding count offset.
2256   bind(L_found_recv);
2257   add(offset, offset, receiver_to_count_step);
2258 
2259   // Finally, update the counter
2260   bind(L_count_update);
2261   increment(Address(mdp, offset), DataLayout::counter_increment);
2262 }
2263 
2264 
2265 void MacroAssembler::call_VM_leaf_base(address entry_point,
2266                                        int number_of_arguments,
2267                                        Label *retaddr) {
2268   Label E, L;
2269 
2270   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2271 
2272   mov(rscratch1, RuntimeAddress(entry_point));
2273   blr(rscratch1);
2274   if (retaddr)
2275     bind(*retaddr);
2276 
2277   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2278 }
2279 
2280 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2281   call_VM_leaf_base(entry_point, number_of_arguments);
2282 }
2283 
2284 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2285   pass_arg0(this, arg_0);
2286   call_VM_leaf_base(entry_point, 1);
2287 }
2288 
2289 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2290   assert_different_registers(arg_1, c_rarg0);
2291   pass_arg0(this, arg_0);
2292   pass_arg1(this, arg_1);
2293   call_VM_leaf_base(entry_point, 2);
2294 }
2295 
2296 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2297                                   Register arg_1, Register arg_2) {
2298   assert_different_registers(arg_1, c_rarg0);
2299   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2300   pass_arg0(this, arg_0);
2301   pass_arg1(this, arg_1);
2302   pass_arg2(this, arg_2);
2303   call_VM_leaf_base(entry_point, 3);
2304 }
2305 
2306 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2307   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2308 }
2309 
2310 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2311   pass_arg0(this, arg_0);
2312   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2313 }
2314 
2315 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2316 
2317   assert_different_registers(arg_0, c_rarg1);
2318   pass_arg1(this, arg_1);
2319   pass_arg0(this, arg_0);
2320   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2321 }
2322 
2323 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2324   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2325   assert_different_registers(arg_1, c_rarg2);
2326   pass_arg2(this, arg_2);
2327   pass_arg1(this, arg_1);
2328   pass_arg0(this, arg_0);
2329   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2330 }
2331 
2332 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2333   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2334   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2335   assert_different_registers(arg_2, c_rarg3);
2336   pass_arg3(this, arg_3);
2337   pass_arg2(this, arg_2);
2338   pass_arg1(this, arg_1);
2339   pass_arg0(this, arg_0);
2340   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2341 }
2342 
2343 void MacroAssembler::null_check(Register reg, int offset) {
2344   if (needs_explicit_null_check(offset)) {
2345     // provoke OS null exception if reg is null by
2346     // accessing M[reg] w/o changing any registers
2347     // NOTE: this is plenty to provoke a segv
2348     ldr(zr, Address(reg));
2349   } else {
2350     // nothing to do, (later) access of M[reg + offset]
2351     // will provoke OS null exception if reg is null
2352   }
2353 }
2354 
2355 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2356   assert_different_registers(markword, rscratch2);
2357   mov(rscratch2, markWord::inline_type_pattern_mask);
2358   andr(markword, markword, rscratch2);
2359   mov(rscratch2, markWord::inline_type_pattern);
2360   cmp(markword, rscratch2);
2361   br(Assembler::EQ, is_inline_type);
2362 }
2363 
2364 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2365   assert_different_registers(tmp, rscratch1);
2366   if (can_be_null) {
2367     cbz(object, not_inline_type);
2368   }
2369   const int is_inline_type_mask = markWord::inline_type_pattern;
2370   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2371   mov(rscratch1, is_inline_type_mask);
2372   andr(tmp, tmp, rscratch1);
2373   cmp(tmp, rscratch1);
2374   br(Assembler::NE, not_inline_type);
2375 }
2376 
2377 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2378   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2379   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2380 }
2381 
2382 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2383   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2384   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2385 }
2386 
2387 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2388   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2389   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2390 }
2391 
2392 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2393   // load mark word
2394   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2395   if (!UseObjectMonitorTable) {
2396     Label test_mark_word;
2397     // check displaced
2398     tst(temp_reg, markWord::unlocked_value);
2399     br(Assembler::NE, test_mark_word);
2400     // slow path use klass prototype
2401     load_prototype_header(temp_reg, oop);
2402 
2403     bind(test_mark_word);
2404   }
2405   andr(temp_reg, temp_reg, test_bit);
2406   if (jmp_set) {
2407     cbnz(temp_reg, jmp_label);
2408   } else {
2409     cbz(temp_reg, jmp_label);
2410   }
2411 }
2412 
2413 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2414   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2415 }
2416 
2417 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2418                                                   Label&is_non_flat_array) {
2419   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2420 }
2421 
2422 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2423   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2424 }
2425 
2426 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2427   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2428 }
2429 
2430 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2431   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2432   br(Assembler::NE, is_flat_array);
2433 }
2434 
2435 // MacroAssembler protected routines needed to implement
2436 // public methods
2437 
2438 void MacroAssembler::mov(Register r, Address dest) {
2439   code_section()->relocate(pc(), dest.rspec());
2440   uint64_t imm64 = (uint64_t)dest.target();
2441   movptr(r, imm64);
2442 }
2443 
2444 // Move a constant pointer into r.  In AArch64 mode the virtual
2445 // address space is 48 bits in size, so we only need three
2446 // instructions to create a patchable instruction sequence that can
2447 // reach anywhere.
2448 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2449 #ifndef PRODUCT
2450   {
2451     char buffer[64];
2452     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2453     block_comment(buffer);
2454   }
2455 #endif
2456   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2457   movz(r, imm64 & 0xffff);
2458   imm64 >>= 16;
2459   movk(r, imm64 & 0xffff, 16);
2460   imm64 >>= 16;
2461   movk(r, imm64 & 0xffff, 32);
2462 }
2463 
2464 // Macro to mov replicated immediate to vector register.
2465 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2466 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2467 // Vd will get the following values for different arrangements in T
2468 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2469 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2470 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2471 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2472 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2473 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2474 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2475 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2476 // Clobbers rscratch1
2477 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2478   assert(T != T1Q, "unsupported");
2479   if (T == T1D || T == T2D) {
2480     int imm = operand_valid_for_movi_immediate(imm64, T);
2481     if (-1 != imm) {
2482       movi(Vd, T, imm);
2483     } else {
2484       mov(rscratch1, imm64);
2485       dup(Vd, T, rscratch1);
2486     }
2487     return;
2488   }
2489 
2490 #ifdef ASSERT
2491   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2492   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2493   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2494 #endif
2495   int shift = operand_valid_for_movi_immediate(imm64, T);
2496   uint32_t imm32 = imm64 & 0xffffffffULL;
2497   if (shift >= 0) {
2498     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2499   } else {
2500     movw(rscratch1, imm32);
2501     dup(Vd, T, rscratch1);
2502   }
2503 }
2504 
2505 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2506 {
2507 #ifndef PRODUCT
2508   {
2509     char buffer[64];
2510     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2511     block_comment(buffer);
2512   }
2513 #endif
2514   if (operand_valid_for_logical_immediate(false, imm64)) {
2515     orr(dst, zr, imm64);
2516   } else {
2517     // we can use a combination of MOVZ or MOVN with
2518     // MOVK to build up the constant
2519     uint64_t imm_h[4];
2520     int zero_count = 0;
2521     int neg_count = 0;
2522     int i;
2523     for (i = 0; i < 4; i++) {
2524       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2525       if (imm_h[i] == 0) {
2526         zero_count++;
2527       } else if (imm_h[i] == 0xffffL) {
2528         neg_count++;
2529       }
2530     }
2531     if (zero_count == 4) {
2532       // one MOVZ will do
2533       movz(dst, 0);
2534     } else if (neg_count == 4) {
2535       // one MOVN will do
2536       movn(dst, 0);
2537     } else if (zero_count == 3) {
2538       for (i = 0; i < 4; i++) {
2539         if (imm_h[i] != 0L) {
2540           movz(dst, (uint32_t)imm_h[i], (i << 4));
2541           break;
2542         }
2543       }
2544     } else if (neg_count == 3) {
2545       // one MOVN will do
2546       for (int i = 0; i < 4; i++) {
2547         if (imm_h[i] != 0xffffL) {
2548           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2549           break;
2550         }
2551       }
2552     } else if (zero_count == 2) {
2553       // one MOVZ and one MOVK will do
2554       for (i = 0; i < 3; i++) {
2555         if (imm_h[i] != 0L) {
2556           movz(dst, (uint32_t)imm_h[i], (i << 4));
2557           i++;
2558           break;
2559         }
2560       }
2561       for (;i < 4; i++) {
2562         if (imm_h[i] != 0L) {
2563           movk(dst, (uint32_t)imm_h[i], (i << 4));
2564         }
2565       }
2566     } else if (neg_count == 2) {
2567       // one MOVN and one MOVK will do
2568       for (i = 0; i < 4; i++) {
2569         if (imm_h[i] != 0xffffL) {
2570           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2571           i++;
2572           break;
2573         }
2574       }
2575       for (;i < 4; i++) {
2576         if (imm_h[i] != 0xffffL) {
2577           movk(dst, (uint32_t)imm_h[i], (i << 4));
2578         }
2579       }
2580     } else if (zero_count == 1) {
2581       // one MOVZ and two MOVKs will do
2582       for (i = 0; i < 4; i++) {
2583         if (imm_h[i] != 0L) {
2584           movz(dst, (uint32_t)imm_h[i], (i << 4));
2585           i++;
2586           break;
2587         }
2588       }
2589       for (;i < 4; i++) {
2590         if (imm_h[i] != 0x0L) {
2591           movk(dst, (uint32_t)imm_h[i], (i << 4));
2592         }
2593       }
2594     } else if (neg_count == 1) {
2595       // one MOVN and two MOVKs will do
2596       for (i = 0; i < 4; i++) {
2597         if (imm_h[i] != 0xffffL) {
2598           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2599           i++;
2600           break;
2601         }
2602       }
2603       for (;i < 4; i++) {
2604         if (imm_h[i] != 0xffffL) {
2605           movk(dst, (uint32_t)imm_h[i], (i << 4));
2606         }
2607       }
2608     } else {
2609       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2610       movz(dst, (uint32_t)imm_h[0], 0);
2611       for (i = 1; i < 4; i++) {
2612         movk(dst, (uint32_t)imm_h[i], (i << 4));
2613       }
2614     }
2615   }
2616 }
2617 
2618 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2619 {
2620 #ifndef PRODUCT
2621     {
2622       char buffer[64];
2623       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2624       block_comment(buffer);
2625     }
2626 #endif
2627   if (operand_valid_for_logical_immediate(true, imm32)) {
2628     orrw(dst, zr, imm32);
2629   } else {
2630     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2631     // constant
2632     uint32_t imm_h[2];
2633     imm_h[0] = imm32 & 0xffff;
2634     imm_h[1] = ((imm32 >> 16) & 0xffff);
2635     if (imm_h[0] == 0) {
2636       movzw(dst, imm_h[1], 16);
2637     } else if (imm_h[0] == 0xffff) {
2638       movnw(dst, imm_h[1] ^ 0xffff, 16);
2639     } else if (imm_h[1] == 0) {
2640       movzw(dst, imm_h[0], 0);
2641     } else if (imm_h[1] == 0xffff) {
2642       movnw(dst, imm_h[0] ^ 0xffff, 0);
2643     } else {
2644       // use a MOVZ and MOVK (makes it easier to debug)
2645       movzw(dst, imm_h[0], 0);
2646       movkw(dst, imm_h[1], 16);
2647     }
2648   }
2649 }
2650 
2651 // Form an address from base + offset in Rd.  Rd may or may
2652 // not actually be used: you must use the Address that is returned.
2653 // It is up to you to ensure that the shift provided matches the size
2654 // of your data.
2655 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2656   if (Address::offset_ok_for_immed(byte_offset, shift))
2657     // It fits; no need for any heroics
2658     return Address(base, byte_offset);
2659 
2660   // Don't do anything clever with negative or misaligned offsets
2661   unsigned mask = (1 << shift) - 1;
2662   if (byte_offset < 0 || byte_offset & mask) {
2663     mov(Rd, byte_offset);
2664     add(Rd, base, Rd);
2665     return Address(Rd);
2666   }
2667 
2668   // See if we can do this with two 12-bit offsets
2669   {
2670     uint64_t word_offset = byte_offset >> shift;
2671     uint64_t masked_offset = word_offset & 0xfff000;
2672     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2673         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2674       add(Rd, base, masked_offset << shift);
2675       word_offset -= masked_offset;
2676       return Address(Rd, word_offset << shift);
2677     }
2678   }
2679 
2680   // Do it the hard way
2681   mov(Rd, byte_offset);
2682   add(Rd, base, Rd);
2683   return Address(Rd);
2684 }
2685 
2686 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2687                                     bool want_remainder, Register scratch)
2688 {
2689   // Full implementation of Java idiv and irem.  The function
2690   // returns the (pc) offset of the div instruction - may be needed
2691   // for implicit exceptions.
2692   //
2693   // constraint : ra/rb =/= scratch
2694   //         normal case
2695   //
2696   // input : ra: dividend
2697   //         rb: divisor
2698   //
2699   // result: either
2700   //         quotient  (= ra idiv rb)
2701   //         remainder (= ra irem rb)
2702 
2703   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2704 
2705   int idivl_offset = offset();
2706   if (! want_remainder) {
2707     sdivw(result, ra, rb);
2708   } else {
2709     sdivw(scratch, ra, rb);
2710     Assembler::msubw(result, scratch, rb, ra);
2711   }
2712 
2713   return idivl_offset;
2714 }
2715 
2716 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2717                                     bool want_remainder, Register scratch)
2718 {
2719   // Full implementation of Java ldiv and lrem.  The function
2720   // returns the (pc) offset of the div instruction - may be needed
2721   // for implicit exceptions.
2722   //
2723   // constraint : ra/rb =/= scratch
2724   //         normal case
2725   //
2726   // input : ra: dividend
2727   //         rb: divisor
2728   //
2729   // result: either
2730   //         quotient  (= ra idiv rb)
2731   //         remainder (= ra irem rb)
2732 
2733   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2734 
2735   int idivq_offset = offset();
2736   if (! want_remainder) {
2737     sdiv(result, ra, rb);
2738   } else {
2739     sdiv(scratch, ra, rb);
2740     Assembler::msub(result, scratch, rb, ra);
2741   }
2742 
2743   return idivq_offset;
2744 }
2745 
2746 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2747   address prev = pc() - NativeMembar::instruction_size;
2748   address last = code()->last_merge_candidate();
2749   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2750     NativeMembar *bar = NativeMembar_at(prev);
2751     if (AlwaysMergeDMB) {
2752       bar->set_kind(bar->get_kind() | order_constraint);
2753       BLOCK_COMMENT("merged membar(always)");
2754       return;
2755     }
2756     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2757     // doing so would introduce a StoreLoad which the caller did not
2758     // intend
2759     if (bar->get_kind() == order_constraint
2760         || bar->get_kind() == AnyAny
2761         || order_constraint == AnyAny) {
2762       // We are merging two memory barrier instructions.  On AArch64 we
2763       // can do this simply by ORing them together.
2764       bar->set_kind(bar->get_kind() | order_constraint);
2765       BLOCK_COMMENT("merged membar");
2766       return;
2767     } else {
2768       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped.
2769       // We need to check the second-to-last instruction, only if it is inside
2770       // the current code section.
2771       address prev2 = prev - NativeMembar::instruction_size;
2772       if (prev2 >= begin() && last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2773         NativeMembar *bar2 = NativeMembar_at(prev2);
2774         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2775         BLOCK_COMMENT("merged membar(elided)");
2776         return;
2777       }
2778     }
2779   }
2780   code()->set_last_merge_candidate(pc());
2781   dmb(Assembler::barrier(order_constraint));
2782 }
2783 
2784 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2785   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2786     merge_ldst(rt, adr, size_in_bytes, is_store);
2787     code()->clear_last_merge_candidate();
2788     return true;
2789   } else {
2790     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2791     const uint64_t mask = size_in_bytes - 1;
2792     if (adr.getMode() == Address::base_plus_offset &&
2793         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2794       code()->set_last_merge_candidate(pc());
2795     }
2796     return false;
2797   }
2798 }
2799 
2800 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2801   // We always try to merge two adjacent loads into one ldp.
2802   if (!try_merge_ldst(Rx, adr, 8, false)) {
2803     Assembler::ldr(Rx, adr);
2804   }
2805 }
2806 
2807 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2808   // We always try to merge two adjacent loads into one ldp.
2809   if (!try_merge_ldst(Rw, adr, 4, false)) {
2810     Assembler::ldrw(Rw, adr);
2811   }
2812 }
2813 
2814 void MacroAssembler::str(Register Rx, const Address &adr) {
2815   // We always try to merge two adjacent stores into one stp.
2816   if (!try_merge_ldst(Rx, adr, 8, true)) {
2817     Assembler::str(Rx, adr);
2818   }
2819 }
2820 
2821 void MacroAssembler::strw(Register Rw, const Address &adr) {
2822   // We always try to merge two adjacent stores into one stp.
2823   if (!try_merge_ldst(Rw, adr, 4, true)) {
2824     Assembler::strw(Rw, adr);
2825   }
2826 }
2827 
2828 // MacroAssembler routines found actually to be needed
2829 
2830 void MacroAssembler::push(Register src)
2831 {
2832   str(src, Address(pre(esp, -1 * wordSize)));
2833 }
2834 
2835 void MacroAssembler::pop(Register dst)
2836 {
2837   ldr(dst, Address(post(esp, 1 * wordSize)));
2838 }
2839 
2840 // Note: load_unsigned_short used to be called load_unsigned_word.
2841 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2842   int off = offset();
2843   ldrh(dst, src);
2844   return off;
2845 }
2846 
2847 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2848   int off = offset();
2849   ldrb(dst, src);
2850   return off;
2851 }
2852 
2853 int MacroAssembler::load_signed_short(Register dst, Address src) {
2854   int off = offset();
2855   ldrsh(dst, src);
2856   return off;
2857 }
2858 
2859 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2860   int off = offset();
2861   ldrsb(dst, src);
2862   return off;
2863 }
2864 
2865 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2866   int off = offset();
2867   ldrshw(dst, src);
2868   return off;
2869 }
2870 
2871 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2872   int off = offset();
2873   ldrsbw(dst, src);
2874   return off;
2875 }
2876 
2877 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2878   switch (size_in_bytes) {
2879   case  8:  ldr(dst, src); break;
2880   case  4:  ldrw(dst, src); break;
2881   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2882   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2883   default:  ShouldNotReachHere();
2884   }
2885 }
2886 
2887 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2888   switch (size_in_bytes) {
2889   case  8:  str(src, dst); break;
2890   case  4:  strw(src, dst); break;
2891   case  2:  strh(src, dst); break;
2892   case  1:  strb(src, dst); break;
2893   default:  ShouldNotReachHere();
2894   }
2895 }
2896 
2897 void MacroAssembler::narrow_subword_type(Register reg, BasicType bt) {
2898   assert(is_subword_type(bt), "required");
2899   switch (bt) {
2900   case T_BOOLEAN: andw(reg, reg, 1); break;
2901   case T_BYTE:    sxtbw(reg, reg); break;
2902   case T_CHAR:    uxthw(reg, reg); break;
2903   case T_SHORT:   sxthw(reg, reg); break;
2904   default:        ShouldNotReachHere();
2905   }
2906 }
2907 
2908 void MacroAssembler::decrementw(Register reg, int value)
2909 {
2910   if (value < 0)  { incrementw(reg, -value);      return; }
2911   if (value == 0) {                               return; }
2912   if (value < (1 << 24)) { subw(reg, reg, value); return; }
2913   /* else */ {
2914     guarantee(reg != rscratch2, "invalid dst for register decrement");
2915     movw(rscratch2, (unsigned)value);
2916     subw(reg, reg, rscratch2);
2917   }
2918 }
2919 
2920 void MacroAssembler::decrement(Register reg, int value)
2921 {
2922   if (value < 0)  { increment(reg, -value);      return; }
2923   if (value == 0) {                              return; }
2924   if (value < (1 << 24)) { sub(reg, reg, value); return; }
2925   /* else */ {
2926     assert(reg != rscratch2, "invalid dst for register decrement");
2927     mov(rscratch2, (uint64_t)value);
2928     sub(reg, reg, rscratch2);
2929   }
2930 }
2931 
2932 void MacroAssembler::decrementw(Address dst, int value)
2933 {
2934   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2935   if (dst.getMode() == Address::literal) {
2936     assert(abs(value) < (1 << 24), "invalid value and address mode combination");
2937     lea(rscratch2, dst);
2938     dst = Address(rscratch2);
2939   }
2940   ldrw(rscratch1, dst);
2941   decrementw(rscratch1, value);
2942   strw(rscratch1, dst);
2943 }
2944 
2945 void MacroAssembler::decrement(Address dst, int value)
2946 {
2947   assert(!dst.uses(rscratch1), "invalid address for decrement");
2948   if (dst.getMode() == Address::literal) {
2949     assert(abs(value) < (1 << 24), "invalid value and address mode combination");
2950     lea(rscratch2, dst);
2951     dst = Address(rscratch2);
2952   }
2953   ldr(rscratch1, dst);
2954   decrement(rscratch1, value);
2955   str(rscratch1, dst);
2956 }
2957 
2958 void MacroAssembler::incrementw(Register reg, int value)
2959 {
2960   if (value < 0)  { decrementw(reg, -value);      return; }
2961   if (value == 0) {                               return; }
2962   if (value < (1 << 24)) { addw(reg, reg, value); return; }
2963   /* else */ {
2964     assert(reg != rscratch2, "invalid dst for register increment");
2965     movw(rscratch2, (unsigned)value);
2966     addw(reg, reg, rscratch2);
2967   }
2968 }
2969 
2970 void MacroAssembler::increment(Register reg, int value)
2971 {
2972   if (value < 0)  { decrement(reg, -value);      return; }
2973   if (value == 0) {                              return; }
2974   if (value < (1 << 24)) { add(reg, reg, value); return; }
2975   /* else */ {
2976     assert(reg != rscratch2, "invalid dst for register increment");
2977     movw(rscratch2, (unsigned)value);
2978     add(reg, reg, rscratch2);
2979   }
2980 }
2981 
2982 void MacroAssembler::incrementw(Address dst, int value, Register result)
2983 {
2984   assert(!dst.uses(result), "invalid dst for address increment");
2985   assert(result->is_valid(), "must be");
2986   assert_different_registers(result, rscratch2);
2987   if (dst.getMode() == Address::literal) {
2988     assert(abs(value) < (1 << 24), "invalid value and address mode combination");
2989     lea(rscratch2, dst);
2990     dst = Address(rscratch2);
2991   }
2992   ldrw(result, dst);
2993   incrementw(result, value);
2994   strw(result, dst);
2995 }
2996 
2997 void MacroAssembler::increment(Address dst, int value, Register result)
2998 {
2999   assert(!dst.uses(result), "invalid dst for address increment");
3000   assert(result->is_valid(), "must be");
3001   assert_different_registers(result, rscratch2);
3002   if (dst.getMode() == Address::literal) {
3003     assert(abs(value) < (1 << 24), "invalid value and address mode combination");
3004     lea(rscratch2, dst);
3005     dst = Address(rscratch2);
3006   }
3007   ldr(result, dst);
3008   increment(result, value);
3009   str(result, dst);
3010 }
3011 
3012 // Push lots of registers in the bit set supplied.  Don't push sp.
3013 // Return the number of words pushed
3014 int MacroAssembler::push(RegSet regset, Register stack) {
3015   if (regset.bits() == 0) {
3016     return 0;
3017   }
3018   auto bitset = integer_cast<unsigned int>(regset.bits());
3019   int words_pushed = 0;
3020 
3021   // Scan bitset to accumulate register pairs
3022   unsigned char regs[32];
3023   int count = 0;
3024   for (int reg = 0; reg <= 30; reg++) {
3025     if (1 & bitset)
3026       regs[count++] = reg;
3027     bitset >>= 1;
3028   }
3029   regs[count++] = zr->raw_encoding();
3030   count &= ~1;  // Only push an even number of regs
3031 
3032   if (count) {
3033     stp(as_Register(regs[0]), as_Register(regs[1]),
3034        Address(pre(stack, -count * wordSize)));
3035     words_pushed += 2;
3036   }
3037   for (int i = 2; i < count; i += 2) {
3038     stp(as_Register(regs[i]), as_Register(regs[i+1]),
3039        Address(stack, i * wordSize));
3040     words_pushed += 2;
3041   }
3042 
3043   assert(words_pushed == count, "oops, pushed != count");
3044 
3045   return count;
3046 }
3047 
3048 int MacroAssembler::pop(RegSet regset, Register stack) {
3049   if (regset.bits() == 0) {
3050     return 0;
3051   }
3052   auto bitset = integer_cast<unsigned int>(regset.bits());
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(FloatRegSet regset, Register stack, FpPushPopMode mode) {
3085   if (regset.bits() == 0) {
3086     return 0;
3087   }
3088   auto bitset = integer_cast<unsigned int>(regset.bits());
3089   int words_pushed = 0;
3090   bool use_sve = false;
3091   int sve_vector_size_in_bytes = 0;
3092 
3093 #ifdef COMPILER2
3094   use_sve = Matcher::supports_scalable_vector();
3095   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3096 #endif
3097 
3098   // Scan bitset to accumulate register pairs
3099   unsigned char regs[32];
3100   int count = 0;
3101   for (int reg = 0; reg <= 31; reg++) {
3102     if (1 & bitset)
3103       regs[count++] = reg;
3104     bitset >>= 1;
3105   }
3106 
3107   if (count == 0) {
3108     return 0;
3109   }
3110 
3111   if (mode == PushPopFull) {
3112     if (use_sve && sve_vector_size_in_bytes > 16) {
3113       mode = PushPopSVE;
3114     } else {
3115       mode = PushPopNeon;
3116     }
3117   }
3118 
3119 #ifndef PRODUCT
3120   {
3121     char buffer[48];
3122     if (mode == PushPopSVE) {
3123       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3124     } else if (mode == PushPopNeon) {
3125       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3126     } else {
3127       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3128     }
3129     block_comment(buffer);
3130   }
3131 #endif
3132 
3133   if (mode == PushPopSVE) {
3134     sub(stack, stack, sve_vector_size_in_bytes * count);
3135     for (int i = 0; i < count; i++) {
3136       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3137     }
3138     return count * sve_vector_size_in_bytes / 8;
3139   }
3140 
3141   if (mode == PushPopNeon) {
3142     if (count == 1) {
3143       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3144       return 2;
3145     }
3146 
3147     bool odd = (count & 1) == 1;
3148     int push_slots = count + (odd ? 1 : 0);
3149 
3150     // Always pushing full 128 bit registers.
3151     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3152     words_pushed += 2;
3153 
3154     for (int i = 2; i + 1 < count; i += 2) {
3155       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3156       words_pushed += 2;
3157     }
3158 
3159     if (odd) {
3160       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3161       words_pushed++;
3162     }
3163 
3164     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3165     return count * 2;
3166   }
3167 
3168   if (mode == PushPopFp) {
3169     bool odd = (count & 1) == 1;
3170     int push_slots = count + (odd ? 1 : 0);
3171 
3172     if (count == 1) {
3173       // Stack pointer must be 16 bytes aligned
3174       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3175       return 1;
3176     }
3177 
3178     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3179     words_pushed += 2;
3180 
3181     for (int i = 2; i + 1 < count; i += 2) {
3182       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3183       words_pushed += 2;
3184     }
3185 
3186     if (odd) {
3187       // Stack pointer must be 16 bytes aligned
3188       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3189       words_pushed++;
3190     }
3191 
3192     assert(words_pushed == count, "oops, pushed != count");
3193 
3194     return count;
3195   }
3196 
3197   return 0;
3198 }
3199 
3200 // Return the number of dwords popped
3201 int MacroAssembler::pop_fp(FloatRegSet regset, Register stack, FpPushPopMode mode) {
3202   if (regset.bits() == 0) {
3203     return 0;
3204   }
3205   auto bitset = integer_cast<unsigned int>(regset.bits());
3206   int words_pushed = 0;
3207   bool use_sve = false;
3208   int sve_vector_size_in_bytes = 0;
3209 
3210 #ifdef COMPILER2
3211   use_sve = Matcher::supports_scalable_vector();
3212   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3213 #endif
3214   // Scan bitset to accumulate register pairs
3215   unsigned char regs[32];
3216   int count = 0;
3217   for (int reg = 0; reg <= 31; reg++) {
3218     if (1 & bitset)
3219       regs[count++] = reg;
3220     bitset >>= 1;
3221   }
3222 
3223   if (count == 0) {
3224     return 0;
3225   }
3226 
3227   if (mode == PushPopFull) {
3228     if (use_sve && sve_vector_size_in_bytes > 16) {
3229       mode = PushPopSVE;
3230     } else {
3231       mode = PushPopNeon;
3232     }
3233   }
3234 
3235 #ifndef PRODUCT
3236   {
3237     char buffer[48];
3238     if (mode == PushPopSVE) {
3239       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3240     } else if (mode == PushPopNeon) {
3241       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3242     } else {
3243       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3244     }
3245     block_comment(buffer);
3246   }
3247 #endif
3248 
3249   if (mode == PushPopSVE) {
3250     for (int i = count - 1; i >= 0; i--) {
3251       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3252     }
3253     add(stack, stack, sve_vector_size_in_bytes * count);
3254     return count * sve_vector_size_in_bytes / 8;
3255   }
3256 
3257   if (mode == PushPopNeon) {
3258     if (count == 1) {
3259       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3260       return 2;
3261     }
3262 
3263     bool odd = (count & 1) == 1;
3264     int push_slots = count + (odd ? 1 : 0);
3265 
3266     if (odd) {
3267       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3268       words_pushed++;
3269     }
3270 
3271     for (int i = 2; i + 1 < count; i += 2) {
3272       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3273       words_pushed += 2;
3274     }
3275 
3276     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3277     words_pushed += 2;
3278 
3279     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3280 
3281     return count * 2;
3282   }
3283 
3284   if (mode == PushPopFp) {
3285     bool odd = (count & 1) == 1;
3286     int push_slots = count + (odd ? 1 : 0);
3287 
3288     if (count == 1) {
3289       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3290       return 1;
3291     }
3292 
3293     if (odd) {
3294       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3295       words_pushed++;
3296     }
3297 
3298     for (int i = 2; i + 1 < count; i += 2) {
3299       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3300       words_pushed += 2;
3301     }
3302 
3303     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3304     words_pushed += 2;
3305 
3306     assert(words_pushed == count, "oops, pushed != count");
3307 
3308     return count;
3309   }
3310 
3311   return 0;
3312 }
3313 
3314 // Return the number of dwords pushed
3315 int MacroAssembler::push_p(PRegSet regset, Register stack) {
3316   if (regset.bits() == 0) {
3317     return 0;
3318   }
3319   auto bitset = integer_cast<unsigned int>(regset.bits());
3320   bool use_sve = false;
3321   int sve_predicate_size_in_slots = 0;
3322 
3323 #ifdef COMPILER2
3324   use_sve = Matcher::supports_scalable_vector();
3325   if (use_sve) {
3326     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3327   }
3328 #endif
3329 
3330   if (!use_sve) {
3331     return 0;
3332   }
3333 
3334   unsigned char regs[PRegister::number_of_registers];
3335   int count = 0;
3336   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3337     if (1 & bitset)
3338       regs[count++] = reg;
3339     bitset >>= 1;
3340   }
3341 
3342   if (count == 0) {
3343     return 0;
3344   }
3345 
3346   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3347                                   VMRegImpl::stack_slot_size * count, 16);
3348   sub(stack, stack, total_push_bytes);
3349   for (int i = 0; i < count; i++) {
3350     sve_str(as_PRegister(regs[i]), Address(stack, i));
3351   }
3352   return total_push_bytes / 8;
3353 }
3354 
3355 // Return the number of dwords popped
3356 int MacroAssembler::pop_p(PRegSet regset, Register stack) {
3357   if (regset.bits() == 0) {
3358     return 0;
3359   }
3360   auto bitset = integer_cast<unsigned int>(regset.bits());
3361   bool use_sve = false;
3362   int sve_predicate_size_in_slots = 0;
3363 
3364 #ifdef COMPILER2
3365   use_sve = Matcher::supports_scalable_vector();
3366   if (use_sve) {
3367     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3368   }
3369 #endif
3370 
3371   if (!use_sve) {
3372     return 0;
3373   }
3374 
3375   unsigned char regs[PRegister::number_of_registers];
3376   int count = 0;
3377   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3378     if (1 & bitset)
3379       regs[count++] = reg;
3380     bitset >>= 1;
3381   }
3382 
3383   if (count == 0) {
3384     return 0;
3385   }
3386 
3387   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3388                                  VMRegImpl::stack_slot_size * count, 16);
3389   for (int i = count - 1; i >= 0; i--) {
3390     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3391   }
3392   add(stack, stack, total_pop_bytes);
3393   return total_pop_bytes / 8;
3394 }
3395 
3396 #ifdef ASSERT
3397 void MacroAssembler::verify_heapbase(const char* msg) {
3398 #if 0
3399   assert (Universe::heap() != nullptr, "java heap should be initialized");
3400   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3401     // rheapbase is allocated as general register
3402     return;
3403   }
3404   if (CheckCompressedOops) {
3405     Label ok;
3406     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3407     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3408     br(Assembler::EQ, ok);
3409     stop(msg);
3410     bind(ok);
3411     pop(1 << rscratch1->encoding(), sp);
3412   }
3413 #endif
3414 }
3415 #endif
3416 
3417 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3418   assert_different_registers(value, tmp1, tmp2);
3419   Label done, tagged, weak_tagged;
3420 
3421   cbz(value, done);           // Use null as-is.
3422   tst(value, JNIHandles::tag_mask); // Test for tag.
3423   br(Assembler::NE, tagged);
3424 
3425   // Resolve local handle
3426   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3427   verify_oop(value);
3428   b(done);
3429 
3430   bind(tagged);
3431   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3432   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3433 
3434   // Resolve global handle
3435   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3436   verify_oop(value);
3437   b(done);
3438 
3439   bind(weak_tagged);
3440   // Resolve jweak.
3441   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3442                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3443   verify_oop(value);
3444 
3445   bind(done);
3446 }
3447 
3448 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3449   assert_different_registers(value, tmp1, tmp2);
3450   Label done;
3451 
3452   cbz(value, done);           // Use null as-is.
3453 
3454 #ifdef ASSERT
3455   {
3456     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3457     Label valid_global_tag;
3458     tbnz(value, 1, valid_global_tag); // Test for global tag
3459     stop("non global jobject using resolve_global_jobject");
3460     bind(valid_global_tag);
3461   }
3462 #endif
3463 
3464   // Resolve global handle
3465   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3466   verify_oop(value);
3467 
3468   bind(done);
3469 }
3470 
3471 void MacroAssembler::stop(const char* msg) {
3472   // Skip AOT caching C strings in scratch buffer.
3473   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3474   BLOCK_COMMENT(str);
3475   // load msg into r0 so we can access it from the signal handler
3476   // ExternalAddress enables saving and restoring via the code cache
3477   lea(c_rarg0, ExternalAddress((address) str));
3478   dcps1(0xdeae);
3479 }
3480 
3481 void MacroAssembler::unimplemented(const char* what) {
3482   const char* buf = nullptr;
3483   {
3484     ResourceMark rm;
3485     stringStream ss;
3486     ss.print("unimplemented: %s", what);
3487     buf = code_string(ss.as_string());
3488   }
3489   stop(buf);
3490 }
3491 
3492 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3493 #ifdef ASSERT
3494   Label OK;
3495   br(cc, OK);
3496   stop(msg);
3497   bind(OK);
3498 #endif
3499 }
3500 
3501 // If a constant does not fit in an immediate field, generate some
3502 // number of MOV instructions and then perform the operation.
3503 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3504                                            add_sub_imm_insn insn1,
3505                                            add_sub_reg_insn insn2,
3506                                            bool is32) {
3507   assert(Rd != zr, "Rd = zr and not setting flags?");
3508   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3509   if (fits) {
3510     (this->*insn1)(Rd, Rn, imm);
3511   } else {
3512     if (g_uabs(imm) < (1 << 24)) {
3513        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3514        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3515     } else {
3516        assert_different_registers(Rd, Rn);
3517        mov(Rd, imm);
3518        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3519     }
3520   }
3521 }
3522 
3523 // Separate vsn which sets the flags. Optimisations are more restricted
3524 // because we must set the flags correctly.
3525 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3526                                              add_sub_imm_insn insn1,
3527                                              add_sub_reg_insn insn2,
3528                                              bool is32) {
3529   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3530   if (fits) {
3531     (this->*insn1)(Rd, Rn, imm);
3532   } else {
3533     assert_different_registers(Rd, Rn);
3534     assert(Rd != zr, "overflow in immediate operand");
3535     mov(Rd, imm);
3536     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3537   }
3538 }
3539 
3540 
3541 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3542   if (increment.is_register()) {
3543     add(Rd, Rn, increment.as_register());
3544   } else {
3545     add(Rd, Rn, increment.as_constant());
3546   }
3547 }
3548 
3549 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3550   if (increment.is_register()) {
3551     addw(Rd, Rn, increment.as_register());
3552   } else {
3553     addw(Rd, Rn, increment.as_constant());
3554   }
3555 }
3556 
3557 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3558   if (decrement.is_register()) {
3559     sub(Rd, Rn, decrement.as_register());
3560   } else {
3561     sub(Rd, Rn, decrement.as_constant());
3562   }
3563 }
3564 
3565 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3566   if (decrement.is_register()) {
3567     subw(Rd, Rn, decrement.as_register());
3568   } else {
3569     subw(Rd, Rn, decrement.as_constant());
3570   }
3571 }
3572 
3573 void MacroAssembler::reinit_heapbase()
3574 {
3575   if (UseCompressedOops) {
3576     if (Universe::is_fully_initialized() && !AOTCodeCache::is_on_for_dump()) {
3577       mov(rheapbase, CompressedOops::base());
3578     } else {
3579       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3580       ldr(rheapbase, Address(rheapbase));
3581     }
3582   }
3583 }
3584 
3585 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3586 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3587 // Pass a register for the result, otherwise pass noreg.
3588 
3589 // Clobbers rscratch1
3590 void MacroAssembler::cmpxchg(Register addr, Register expected,
3591                              Register new_val,
3592                              enum operand_size size,
3593                              enum atomic_memory_order order,
3594                              bool weak,
3595                              Register result) {
3596   bool acquire, release;
3597 
3598   switch (order) {
3599     case memory_order_relaxed:
3600       acquire = false;
3601       release = false;
3602       break;
3603     case memory_order_acquire:
3604       acquire = true;
3605       release = false;
3606       break;
3607     case memory_order_release:
3608       acquire = false;
3609       release = true;
3610       break;
3611     case memory_order_acq_rel:
3612     case memory_order_seq_cst:
3613       acquire = true;
3614       release = true;
3615       break;
3616     default:
3617       ShouldNotReachHere();
3618   }
3619 
3620   if (result == noreg)  result = rscratch1;
3621   BLOCK_COMMENT("cmpxchg {");
3622   if (UseLSE) {
3623     mov(result, expected);
3624     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3625     compare_eq(result, expected, size);
3626 #ifdef ASSERT
3627     // Poison rscratch1 which is written on !UseLSE branch
3628     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3629 #endif
3630   } else {
3631     Label retry_load, done;
3632     prfm(Address(addr), PSTL1STRM);
3633     bind(retry_load);
3634     load_exclusive(result, addr, size, acquire);
3635     compare_eq(result, expected, size);
3636     br(Assembler::NE, done);
3637     store_exclusive(rscratch1, new_val, addr, size, release);
3638     if (weak) {
3639       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3640     } else {
3641       cbnzw(rscratch1, retry_load);
3642     }
3643     bind(done);
3644   }
3645   BLOCK_COMMENT("} cmpxchg");
3646 }
3647 
3648 // A generic comparison. Only compares for equality, clobbers rscratch1.
3649 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3650   if (size == xword) {
3651     cmp(rm, rn);
3652   } else if (size == word) {
3653     cmpw(rm, rn);
3654   } else if (size == halfword) {
3655     eorw(rscratch1, rm, rn);
3656     ands(zr, rscratch1, 0xffff);
3657   } else if (size == byte) {
3658     eorw(rscratch1, rm, rn);
3659     ands(zr, rscratch1, 0xff);
3660   } else {
3661     ShouldNotReachHere();
3662   }
3663 }
3664 
3665 
3666 static bool different(Register a, RegisterOrConstant b, Register c) {
3667   if (b.is_constant())
3668     return a != c;
3669   else
3670     return a != b.as_register() && a != c && b.as_register() != c;
3671 }
3672 
3673 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3674 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3675   if (UseLSE) {                                                         \
3676     prev = prev->is_valid() ? prev : zr;                                \
3677     if (incr.is_register()) {                                           \
3678       AOP(sz, incr.as_register(), prev, addr);                          \
3679     } else {                                                            \
3680       mov(rscratch2, incr.as_constant());                               \
3681       AOP(sz, rscratch2, prev, addr);                                   \
3682     }                                                                   \
3683     return;                                                             \
3684   }                                                                     \
3685   Register result = rscratch2;                                          \
3686   if (prev->is_valid())                                                 \
3687     result = different(prev, incr, addr) ? prev : rscratch2;            \
3688                                                                         \
3689   Label retry_load;                                                     \
3690   prfm(Address(addr), PSTL1STRM);                                       \
3691   bind(retry_load);                                                     \
3692   LDXR(result, addr);                                                   \
3693   OP(rscratch1, result, incr);                                          \
3694   STXR(rscratch2, rscratch1, addr);                                     \
3695   cbnzw(rscratch2, retry_load);                                         \
3696   if (prev->is_valid() && prev != result) {                             \
3697     IOP(prev, rscratch1, incr);                                         \
3698   }                                                                     \
3699 }
3700 
3701 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3702 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3703 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3704 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3705 
3706 #undef ATOMIC_OP
3707 
3708 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3709 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3710   if (UseLSE) {                                                         \
3711     prev = prev->is_valid() ? prev : zr;                                \
3712     AOP(sz, newv, prev, addr);                                          \
3713     return;                                                             \
3714   }                                                                     \
3715   Register result = rscratch2;                                          \
3716   if (prev->is_valid())                                                 \
3717     result = different(prev, newv, addr) ? prev : rscratch2;            \
3718                                                                         \
3719   Label retry_load;                                                     \
3720   prfm(Address(addr), PSTL1STRM);                                       \
3721   bind(retry_load);                                                     \
3722   LDXR(result, addr);                                                   \
3723   STXR(rscratch1, newv, addr);                                          \
3724   cbnzw(rscratch1, retry_load);                                         \
3725   if (prev->is_valid() && prev != result)                               \
3726     mov(prev, result);                                                  \
3727 }
3728 
3729 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3730 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3731 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3732 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3733 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3734 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3735 
3736 #undef ATOMIC_XCHG
3737 
3738 #ifndef PRODUCT
3739 extern "C" void findpc(intptr_t x);
3740 #endif
3741 
3742 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3743 {
3744   // In order to get locks to work, we need to fake a in_VM state
3745   if (ShowMessageBoxOnError) {
3746     JavaThread* thread = JavaThread::current();
3747     thread->set_thread_state(_thread_in_vm);
3748 #ifndef PRODUCT
3749     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3750       ttyLocker ttyl;
3751       BytecodeCounter::print();
3752     }
3753 #endif
3754     if (os::message_box(msg, "Execution stopped, print registers?")) {
3755       ttyLocker ttyl;
3756       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3757 #ifndef PRODUCT
3758       tty->cr();
3759       findpc(pc);
3760       tty->cr();
3761 #endif
3762       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3763       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3764       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3765       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3766       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3767       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3768       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3769       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3770       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3771       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3772       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3773       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3774       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3775       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3776       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3777       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3778       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3779       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3780       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3781       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3782       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3783       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3784       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3785       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3786       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3787       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3788       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3789       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3790       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3791       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3792       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3793       BREAKPOINT;
3794     }
3795   }
3796   fatal("DEBUG MESSAGE: %s", msg);
3797 }
3798 
3799 RegSet MacroAssembler::call_clobbered_gp_registers() {
3800   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3801 #ifndef R18_RESERVED
3802   regs += r18_tls;
3803 #endif
3804   return regs;
3805 }
3806 
3807 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3808   int step = 4 * wordSize;
3809   push(call_clobbered_gp_registers() - exclude, sp);
3810   sub(sp, sp, step);
3811   mov(rscratch1, -step);
3812   // Push v0-v7, v16-v31.
3813   for (int i = 31; i>= 4; i -= 4) {
3814     if (i <= v7->encoding() || i >= v16->encoding())
3815       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3816           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3817   }
3818   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3819       as_FloatRegister(3), T1D, Address(sp));
3820 }
3821 
3822 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3823   for (int i = 0; i < 32; i += 4) {
3824     if (i <= v7->encoding() || i >= v16->encoding())
3825       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3826           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3827   }
3828 
3829   reinitialize_ptrue();
3830 
3831   pop(call_clobbered_gp_registers() - exclude, sp);
3832 }
3833 
3834 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3835                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3836   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3837   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3838     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3839     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3840       sve_str(as_FloatRegister(i), Address(sp, i));
3841     }
3842   } else {
3843     int step = (save_vectors ? 8 : 4) * wordSize;
3844     mov(rscratch1, -step);
3845     sub(sp, sp, step);
3846     for (int i = 28; i >= 4; i -= 4) {
3847       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3848           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3849     }
3850     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3851   }
3852   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3853     sub(sp, sp, total_predicate_in_bytes);
3854     for (int i = 0; i < PRegister::number_of_registers; i++) {
3855       sve_str(as_PRegister(i), Address(sp, i));
3856     }
3857   }
3858 }
3859 
3860 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3861                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3862   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3863     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3864       sve_ldr(as_PRegister(i), Address(sp, i));
3865     }
3866     add(sp, sp, total_predicate_in_bytes);
3867   }
3868   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3869     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3870       sve_ldr(as_FloatRegister(i), Address(sp, i));
3871     }
3872     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3873   } else {
3874     int step = (restore_vectors ? 8 : 4) * wordSize;
3875     for (int i = 0; i <= 28; i += 4)
3876       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3877           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3878   }
3879 
3880   // We may use predicate registers and rely on ptrue with SVE,
3881   // regardless of wide vector (> 8 bytes) used or not.
3882   if (use_sve) {
3883     reinitialize_ptrue();
3884   }
3885 
3886   // integer registers except lr & sp
3887   pop(RegSet::range(r0, r17), sp);
3888 #ifdef R18_RESERVED
3889   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3890   pop(RegSet::range(r20, r29), sp);
3891 #else
3892   pop(RegSet::range(r18_tls, r29), sp);
3893 #endif
3894 }
3895 
3896 /**
3897  * Helpers for multiply_to_len().
3898  */
3899 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3900                                      Register src1, Register src2) {
3901   adds(dest_lo, dest_lo, src1);
3902   adc(dest_hi, dest_hi, zr);
3903   adds(dest_lo, dest_lo, src2);
3904   adc(final_dest_hi, dest_hi, zr);
3905 }
3906 
3907 // Generate an address from (r + r1 extend offset).  "size" is the
3908 // size of the operand.  The result may be in rscratch2.
3909 Address MacroAssembler::offsetted_address(Register r, Register r1,
3910                                           Address::extend ext, int offset, int size) {
3911   if (offset || (ext.shift() % size != 0)) {
3912     lea(rscratch2, Address(r, r1, ext));
3913     return Address(rscratch2, offset);
3914   } else {
3915     return Address(r, r1, ext);
3916   }
3917 }
3918 
3919 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3920 {
3921   assert(offset >= 0, "spill to negative address?");
3922   // Offset reachable ?
3923   //   Not aligned - 9 bits signed offset
3924   //   Aligned - 12 bits unsigned offset shifted
3925   Register base = sp;
3926   if ((offset & (size-1)) && offset >= (1<<8)) {
3927     add(tmp, base, offset & ((1<<12)-1));
3928     base = tmp;
3929     offset &= -1u<<12;
3930   }
3931 
3932   if (offset >= (1<<12) * size) {
3933     add(tmp, base, offset & (((1<<12)-1)<<12));
3934     base = tmp;
3935     offset &= ~(((1<<12)-1)<<12);
3936   }
3937 
3938   return Address(base, offset);
3939 }
3940 
3941 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3942   assert(offset >= 0, "spill to negative address?");
3943 
3944   Register base = sp;
3945 
3946   // An immediate offset in the range 0 to 255 which is multiplied
3947   // by the current vector or predicate register size in bytes.
3948   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3949     return Address(base, offset / sve_reg_size_in_bytes);
3950   }
3951 
3952   add(tmp, base, offset);
3953   return Address(tmp);
3954 }
3955 
3956 // Checks whether offset is aligned.
3957 // Returns true if it is, else false.
3958 bool MacroAssembler::merge_alignment_check(Register base,
3959                                            size_t size,
3960                                            int64_t cur_offset,
3961                                            int64_t prev_offset) const {
3962   if (AvoidUnalignedAccesses) {
3963     if (base == sp) {
3964       // Checks whether low offset if aligned to pair of registers.
3965       int64_t pair_mask = size * 2 - 1;
3966       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3967       return (offset & pair_mask) == 0;
3968     } else { // If base is not sp, we can't guarantee the access is aligned.
3969       return false;
3970     }
3971   } else {
3972     int64_t mask = size - 1;
3973     // Load/store pair instruction only supports element size aligned offset.
3974     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3975   }
3976 }
3977 
3978 // Checks whether current and previous loads/stores can be merged.
3979 // Returns true if it can be merged, else false.
3980 bool MacroAssembler::ldst_can_merge(Register rt,
3981                                     const Address &adr,
3982                                     size_t cur_size_in_bytes,
3983                                     bool is_store) const {
3984   address prev = pc() - NativeInstruction::instruction_size;
3985   address last = code()->last_merge_candidate();
3986 
3987   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3988     return false;
3989   }
3990 
3991   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3992     return false;
3993   }
3994 
3995   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3996   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3997 
3998   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3999   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
4000 
4001   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
4002     return false;
4003   }
4004 
4005   int64_t max_offset = 63 * prev_size_in_bytes;
4006   int64_t min_offset = -64 * prev_size_in_bytes;
4007 
4008   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
4009 
4010   // Only same base can be merged.
4011   if (adr.base() != prev_ldst->base()) {
4012     return false;
4013   }
4014 
4015   int64_t cur_offset = adr.offset();
4016   int64_t prev_offset = prev_ldst->offset();
4017   size_t diff = abs(cur_offset - prev_offset);
4018   if (diff != prev_size_in_bytes) {
4019     return false;
4020   }
4021 
4022   // Following cases can not be merged:
4023   // ldr x2, [x2, #8]
4024   // ldr x3, [x2, #16]
4025   // or:
4026   // ldr x2, [x3, #8]
4027   // ldr x2, [x3, #16]
4028   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
4029   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
4030     return false;
4031   }
4032 
4033   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
4034   // Offset range must be in ldp/stp instruction's range.
4035   if (low_offset > max_offset || low_offset < min_offset) {
4036     return false;
4037   }
4038 
4039   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
4040     return true;
4041   }
4042 
4043   return false;
4044 }
4045 
4046 // Merge current load/store with previous load/store into ldp/stp.
4047 void MacroAssembler::merge_ldst(Register rt,
4048                                 const Address &adr,
4049                                 size_t cur_size_in_bytes,
4050                                 bool is_store) {
4051 
4052   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
4053 
4054   Register rt_low, rt_high;
4055   address prev = pc() - NativeInstruction::instruction_size;
4056   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
4057 
4058   int64_t offset;
4059 
4060   if (adr.offset() < prev_ldst->offset()) {
4061     offset = adr.offset();
4062     rt_low = rt;
4063     rt_high = prev_ldst->target();
4064   } else {
4065     offset = prev_ldst->offset();
4066     rt_low = prev_ldst->target();
4067     rt_high = rt;
4068   }
4069 
4070   Address adr_p = Address(prev_ldst->base(), offset);
4071   // Overwrite previous generated binary.
4072   code_section()->set_end(prev);
4073 
4074   const size_t sz = prev_ldst->size_in_bytes();
4075   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
4076   if (!is_store) {
4077     BLOCK_COMMENT("merged ldr pair");
4078     if (sz == 8) {
4079       ldp(rt_low, rt_high, adr_p);
4080     } else {
4081       ldpw(rt_low, rt_high, adr_p);
4082     }
4083   } else {
4084     BLOCK_COMMENT("merged str pair");
4085     if (sz == 8) {
4086       stp(rt_low, rt_high, adr_p);
4087     } else {
4088       stpw(rt_low, rt_high, adr_p);
4089     }
4090   }
4091 }
4092 
4093 /**
4094  * Multiply 64 bit by 64 bit first loop.
4095  */
4096 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
4097                                            Register y, Register y_idx, Register z,
4098                                            Register carry, Register product,
4099                                            Register idx, Register kdx) {
4100   //
4101   //  jlong carry, x[], y[], z[];
4102   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4103   //    huge_128 product = y[idx] * x[xstart] + carry;
4104   //    z[kdx] = (jlong)product;
4105   //    carry  = (jlong)(product >>> 64);
4106   //  }
4107   //  z[xstart] = carry;
4108   //
4109 
4110   Label L_first_loop, L_first_loop_exit;
4111   Label L_one_x, L_one_y, L_multiply;
4112 
4113   subsw(xstart, xstart, 1);
4114   br(Assembler::MI, L_one_x);
4115 
4116   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
4117   ldr(x_xstart, Address(rscratch1));
4118   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
4119 
4120   bind(L_first_loop);
4121   subsw(idx, idx, 1);
4122   br(Assembler::MI, L_first_loop_exit);
4123   subsw(idx, idx, 1);
4124   br(Assembler::MI, L_one_y);
4125   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4126   ldr(y_idx, Address(rscratch1));
4127   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
4128   bind(L_multiply);
4129 
4130   // AArch64 has a multiply-accumulate instruction that we can't use
4131   // here because it has no way to process carries, so we have to use
4132   // separate add and adc instructions.  Bah.
4133   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
4134   mul(product, x_xstart, y_idx);
4135   adds(product, product, carry);
4136   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
4137 
4138   subw(kdx, kdx, 2);
4139   ror(product, product, 32); // back to big-endian
4140   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4141 
4142   b(L_first_loop);
4143 
4144   bind(L_one_y);
4145   ldrw(y_idx, Address(y,  0));
4146   b(L_multiply);
4147 
4148   bind(L_one_x);
4149   ldrw(x_xstart, Address(x,  0));
4150   b(L_first_loop);
4151 
4152   bind(L_first_loop_exit);
4153 }
4154 
4155 /**
4156  * Multiply 128 bit by 128. Unrolled inner loop.
4157  *
4158  */
4159 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4160                                              Register carry, Register carry2,
4161                                              Register idx, Register jdx,
4162                                              Register yz_idx1, Register yz_idx2,
4163                                              Register tmp, Register tmp3, Register tmp4,
4164                                              Register tmp6, Register product_hi) {
4165 
4166   //   jlong carry, x[], y[], z[];
4167   //   int kdx = ystart+1;
4168   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4169   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4170   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4171   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4172   //     carry  = (jlong)(tmp4 >>> 64);
4173   //     z[kdx+idx+1] = (jlong)tmp3;
4174   //     z[kdx+idx] = (jlong)tmp4;
4175   //   }
4176   //   idx += 2;
4177   //   if (idx > 0) {
4178   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4179   //     z[kdx+idx] = (jlong)yz_idx1;
4180   //     carry  = (jlong)(yz_idx1 >>> 64);
4181   //   }
4182   //
4183 
4184   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4185 
4186   lsrw(jdx, idx, 2);
4187 
4188   bind(L_third_loop);
4189 
4190   subsw(jdx, jdx, 1);
4191   br(Assembler::MI, L_third_loop_exit);
4192   subw(idx, idx, 4);
4193 
4194   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4195 
4196   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4197 
4198   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4199 
4200   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4201   ror(yz_idx2, yz_idx2, 32);
4202 
4203   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4204 
4205   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4206   umulh(tmp4, product_hi, yz_idx1);
4207 
4208   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4209   ror(rscratch2, rscratch2, 32);
4210 
4211   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4212   umulh(carry2, product_hi, yz_idx2);
4213 
4214   // propagate sum of both multiplications into carry:tmp4:tmp3
4215   adds(tmp3, tmp3, carry);
4216   adc(tmp4, tmp4, zr);
4217   adds(tmp3, tmp3, rscratch1);
4218   adcs(tmp4, tmp4, tmp);
4219   adc(carry, carry2, zr);
4220   adds(tmp4, tmp4, rscratch2);
4221   adc(carry, carry, zr);
4222 
4223   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4224   ror(tmp4, tmp4, 32);
4225   stp(tmp4, tmp3, Address(tmp6, 0));
4226 
4227   b(L_third_loop);
4228   bind (L_third_loop_exit);
4229 
4230   andw (idx, idx, 0x3);
4231   cbz(idx, L_post_third_loop_done);
4232 
4233   Label L_check_1;
4234   subsw(idx, idx, 2);
4235   br(Assembler::MI, L_check_1);
4236 
4237   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4238   ldr(yz_idx1, Address(rscratch1, 0));
4239   ror(yz_idx1, yz_idx1, 32);
4240   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4241   umulh(tmp4, product_hi, yz_idx1);
4242   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4243   ldr(yz_idx2, Address(rscratch1, 0));
4244   ror(yz_idx2, yz_idx2, 32);
4245 
4246   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4247 
4248   ror(tmp3, tmp3, 32);
4249   str(tmp3, Address(rscratch1, 0));
4250 
4251   bind (L_check_1);
4252 
4253   andw (idx, idx, 0x1);
4254   subsw(idx, idx, 1);
4255   br(Assembler::MI, L_post_third_loop_done);
4256   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4257   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4258   umulh(carry2, tmp4, product_hi);
4259   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4260 
4261   add2_with_carry(carry2, tmp3, tmp4, carry);
4262 
4263   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4264   extr(carry, carry2, tmp3, 32);
4265 
4266   bind(L_post_third_loop_done);
4267 }
4268 
4269 /**
4270  * Code for BigInteger::multiplyToLen() intrinsic.
4271  *
4272  * r0: x
4273  * r1: xlen
4274  * r2: y
4275  * r3: ylen
4276  * r4:  z
4277  * r5: tmp0
4278  * r10: tmp1
4279  * r11: tmp2
4280  * r12: tmp3
4281  * r13: tmp4
4282  * r14: tmp5
4283  * r15: tmp6
4284  * r16: tmp7
4285  *
4286  */
4287 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4288                                      Register z, Register tmp0,
4289                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4290                                      Register tmp5, Register tmp6, Register product_hi) {
4291 
4292   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4293 
4294   const Register idx = tmp1;
4295   const Register kdx = tmp2;
4296   const Register xstart = tmp3;
4297 
4298   const Register y_idx = tmp4;
4299   const Register carry = tmp5;
4300   const Register product  = xlen;
4301   const Register x_xstart = tmp0;
4302 
4303   // First Loop.
4304   //
4305   //  final static long LONG_MASK = 0xffffffffL;
4306   //  int xstart = xlen - 1;
4307   //  int ystart = ylen - 1;
4308   //  long carry = 0;
4309   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4310   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4311   //    z[kdx] = (int)product;
4312   //    carry = product >>> 32;
4313   //  }
4314   //  z[xstart] = (int)carry;
4315   //
4316 
4317   movw(idx, ylen);       // idx = ylen;
4318   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4319   mov(carry, zr);        // carry = 0;
4320 
4321   Label L_done;
4322 
4323   movw(xstart, xlen);
4324   subsw(xstart, xstart, 1);
4325   br(Assembler::MI, L_done);
4326 
4327   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4328 
4329   Label L_second_loop;
4330   cbzw(kdx, L_second_loop);
4331 
4332   Label L_carry;
4333   subw(kdx, kdx, 1);
4334   cbzw(kdx, L_carry);
4335 
4336   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4337   lsr(carry, carry, 32);
4338   subw(kdx, kdx, 1);
4339 
4340   bind(L_carry);
4341   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4342 
4343   // Second and third (nested) loops.
4344   //
4345   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4346   //   carry = 0;
4347   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4348   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4349   //                    (z[k] & LONG_MASK) + carry;
4350   //     z[k] = (int)product;
4351   //     carry = product >>> 32;
4352   //   }
4353   //   z[i] = (int)carry;
4354   // }
4355   //
4356   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4357 
4358   const Register jdx = tmp1;
4359 
4360   bind(L_second_loop);
4361   mov(carry, zr);                // carry = 0;
4362   movw(jdx, ylen);               // j = ystart+1
4363 
4364   subsw(xstart, xstart, 1);      // i = xstart-1;
4365   br(Assembler::MI, L_done);
4366 
4367   str(z, Address(pre(sp, -4 * wordSize)));
4368 
4369   Label L_last_x;
4370   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4371   subsw(xstart, xstart, 1);       // i = xstart-1;
4372   br(Assembler::MI, L_last_x);
4373 
4374   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4375   ldr(product_hi, Address(rscratch1));
4376   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4377 
4378   Label L_third_loop_prologue;
4379   bind(L_third_loop_prologue);
4380 
4381   str(ylen, Address(sp, wordSize));
4382   stp(x, xstart, Address(sp, 2 * wordSize));
4383   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4384                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4385   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4386   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4387 
4388   addw(tmp3, xlen, 1);
4389   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4390   subsw(tmp3, tmp3, 1);
4391   br(Assembler::MI, L_done);
4392 
4393   lsr(carry, carry, 32);
4394   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4395   b(L_second_loop);
4396 
4397   // Next infrequent code is moved outside loops.
4398   bind(L_last_x);
4399   ldrw(product_hi, Address(x,  0));
4400   b(L_third_loop_prologue);
4401 
4402   bind(L_done);
4403 }
4404 
4405 // Code for BigInteger::mulAdd intrinsic
4406 // out     = r0
4407 // in      = r1
4408 // offset  = r2  (already out.length-offset)
4409 // len     = r3
4410 // k       = r4
4411 //
4412 // pseudo code from java implementation:
4413 // carry = 0;
4414 // offset = out.length-offset - 1;
4415 // for (int j=len-1; j >= 0; j--) {
4416 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4417 //     out[offset--] = (int)product;
4418 //     carry = product >>> 32;
4419 // }
4420 // return (int)carry;
4421 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4422       Register len, Register k) {
4423     Label LOOP, END;
4424     // pre-loop
4425     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4426     csel(out, zr, out, Assembler::EQ);
4427     br(Assembler::EQ, END);
4428     add(in, in, len, LSL, 2); // in[j+1] address
4429     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4430     mov(out, zr); // used to keep carry now
4431     BIND(LOOP);
4432     ldrw(rscratch1, Address(pre(in, -4)));
4433     madd(rscratch1, rscratch1, k, out);
4434     ldrw(rscratch2, Address(pre(offset, -4)));
4435     add(rscratch1, rscratch1, rscratch2);
4436     strw(rscratch1, Address(offset));
4437     lsr(out, rscratch1, 32);
4438     subs(len, len, 1);
4439     br(Assembler::NE, LOOP);
4440     BIND(END);
4441 }
4442 
4443 /**
4444  * Emits code to update CRC-32 with a byte value according to constants in table
4445  *
4446  * @param [in,out]crc   Register containing the crc.
4447  * @param [in]val       Register containing the byte to fold into the CRC.
4448  * @param [in]table     Register containing the table of crc constants.
4449  *
4450  * uint32_t crc;
4451  * val = crc_table[(val ^ crc) & 0xFF];
4452  * crc = val ^ (crc >> 8);
4453  *
4454  */
4455 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4456   eor(val, val, crc);
4457   andr(val, val, 0xff);
4458   ldrw(val, Address(table, val, Address::lsl(2)));
4459   eor(crc, val, crc, Assembler::LSR, 8);
4460 }
4461 
4462 /**
4463  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4464  *
4465  * @param [in,out]crc   Register containing the crc.
4466  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4467  * @param [in]table0    Register containing table 0 of crc constants.
4468  * @param [in]table1    Register containing table 1 of crc constants.
4469  * @param [in]table2    Register containing table 2 of crc constants.
4470  * @param [in]table3    Register containing table 3 of crc constants.
4471  *
4472  * uint32_t crc;
4473  *   v = crc ^ v
4474  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4475  *
4476  */
4477 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4478         Register table0, Register table1, Register table2, Register table3,
4479         bool upper) {
4480   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4481   uxtb(tmp, v);
4482   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4483   ubfx(tmp, v, 8, 8);
4484   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4485   eor(crc, crc, tmp);
4486   ubfx(tmp, v, 16, 8);
4487   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4488   eor(crc, crc, tmp);
4489   ubfx(tmp, v, 24, 8);
4490   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4491   eor(crc, crc, tmp);
4492 }
4493 
4494 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4495         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4496     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4497     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4498 
4499     subs(tmp0, len, 384);
4500     mvnw(crc, crc);
4501     br(Assembler::GE, CRC_by128_pre);
4502   BIND(CRC_less128);
4503     subs(len, len, 32);
4504     br(Assembler::GE, CRC_by32_loop);
4505   BIND(CRC_less32);
4506     adds(len, len, 32 - 4);
4507     br(Assembler::GE, CRC_by4_loop);
4508     adds(len, len, 4);
4509     br(Assembler::GT, CRC_by1_loop);
4510     b(L_exit);
4511 
4512   BIND(CRC_by32_loop);
4513     ldp(tmp0, tmp1, Address(buf));
4514     crc32x(crc, crc, tmp0);
4515     ldp(tmp2, tmp3, Address(buf, 16));
4516     crc32x(crc, crc, tmp1);
4517     add(buf, buf, 32);
4518     crc32x(crc, crc, tmp2);
4519     subs(len, len, 32);
4520     crc32x(crc, crc, tmp3);
4521     br(Assembler::GE, CRC_by32_loop);
4522     cmn(len, (u1)32);
4523     br(Assembler::NE, CRC_less32);
4524     b(L_exit);
4525 
4526   BIND(CRC_by4_loop);
4527     ldrw(tmp0, Address(post(buf, 4)));
4528     subs(len, len, 4);
4529     crc32w(crc, crc, tmp0);
4530     br(Assembler::GE, CRC_by4_loop);
4531     adds(len, len, 4);
4532     br(Assembler::LE, L_exit);
4533   BIND(CRC_by1_loop);
4534     ldrb(tmp0, Address(post(buf, 1)));
4535     subs(len, len, 1);
4536     crc32b(crc, crc, tmp0);
4537     br(Assembler::GT, CRC_by1_loop);
4538     b(L_exit);
4539 
4540   BIND(CRC_by128_pre);
4541     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4542       4*256*sizeof(juint) + 8*sizeof(juint));
4543     mov(crc, 0);
4544     crc32x(crc, crc, tmp0);
4545     crc32x(crc, crc, tmp1);
4546 
4547     cbnz(len, CRC_less128);
4548 
4549   BIND(L_exit);
4550     mvnw(crc, crc);
4551 }
4552 
4553 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4554         Register len, Register tmp0, Register tmp1, Register tmp2,
4555         Register tmp3) {
4556     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4557     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4558 
4559     mvnw(crc, crc);
4560 
4561     subs(len, len, 128);
4562     br(Assembler::GE, CRC_by64_pre);
4563   BIND(CRC_less64);
4564     adds(len, len, 128-32);
4565     br(Assembler::GE, CRC_by32_loop);
4566   BIND(CRC_less32);
4567     adds(len, len, 32-4);
4568     br(Assembler::GE, CRC_by4_loop);
4569     adds(len, len, 4);
4570     br(Assembler::GT, CRC_by1_loop);
4571     b(L_exit);
4572 
4573   BIND(CRC_by32_loop);
4574     ldp(tmp0, tmp1, Address(post(buf, 16)));
4575     subs(len, len, 32);
4576     crc32x(crc, crc, tmp0);
4577     ldr(tmp2, Address(post(buf, 8)));
4578     crc32x(crc, crc, tmp1);
4579     ldr(tmp3, Address(post(buf, 8)));
4580     crc32x(crc, crc, tmp2);
4581     crc32x(crc, crc, tmp3);
4582     br(Assembler::GE, CRC_by32_loop);
4583     cmn(len, (u1)32);
4584     br(Assembler::NE, CRC_less32);
4585     b(L_exit);
4586 
4587   BIND(CRC_by4_loop);
4588     ldrw(tmp0, Address(post(buf, 4)));
4589     subs(len, len, 4);
4590     crc32w(crc, crc, tmp0);
4591     br(Assembler::GE, CRC_by4_loop);
4592     adds(len, len, 4);
4593     br(Assembler::LE, L_exit);
4594   BIND(CRC_by1_loop);
4595     ldrb(tmp0, Address(post(buf, 1)));
4596     subs(len, len, 1);
4597     crc32b(crc, crc, tmp0);
4598     br(Assembler::GT, CRC_by1_loop);
4599     b(L_exit);
4600 
4601   BIND(CRC_by64_pre);
4602     sub(buf, buf, 8);
4603     ldp(tmp0, tmp1, Address(buf, 8));
4604     crc32x(crc, crc, tmp0);
4605     ldr(tmp2, Address(buf, 24));
4606     crc32x(crc, crc, tmp1);
4607     ldr(tmp3, Address(buf, 32));
4608     crc32x(crc, crc, tmp2);
4609     ldr(tmp0, Address(buf, 40));
4610     crc32x(crc, crc, tmp3);
4611     ldr(tmp1, Address(buf, 48));
4612     crc32x(crc, crc, tmp0);
4613     ldr(tmp2, Address(buf, 56));
4614     crc32x(crc, crc, tmp1);
4615     ldr(tmp3, Address(pre(buf, 64)));
4616 
4617     b(CRC_by64_loop);
4618 
4619     align(CodeEntryAlignment);
4620   BIND(CRC_by64_loop);
4621     subs(len, len, 64);
4622     crc32x(crc, crc, tmp2);
4623     ldr(tmp0, Address(buf, 8));
4624     crc32x(crc, crc, tmp3);
4625     ldr(tmp1, Address(buf, 16));
4626     crc32x(crc, crc, tmp0);
4627     ldr(tmp2, Address(buf, 24));
4628     crc32x(crc, crc, tmp1);
4629     ldr(tmp3, Address(buf, 32));
4630     crc32x(crc, crc, tmp2);
4631     ldr(tmp0, Address(buf, 40));
4632     crc32x(crc, crc, tmp3);
4633     ldr(tmp1, Address(buf, 48));
4634     crc32x(crc, crc, tmp0);
4635     ldr(tmp2, Address(buf, 56));
4636     crc32x(crc, crc, tmp1);
4637     ldr(tmp3, Address(pre(buf, 64)));
4638     br(Assembler::GE, CRC_by64_loop);
4639 
4640     // post-loop
4641     crc32x(crc, crc, tmp2);
4642     crc32x(crc, crc, tmp3);
4643 
4644     sub(len, len, 64);
4645     add(buf, buf, 8);
4646     cmn(len, (u1)128);
4647     br(Assembler::NE, CRC_less64);
4648   BIND(L_exit);
4649     mvnw(crc, crc);
4650 }
4651 
4652 /**
4653  * @param crc   register containing existing CRC (32-bit)
4654  * @param buf   register pointing to input byte buffer (byte*)
4655  * @param len   register containing number of bytes
4656  * @param table register that will contain address of CRC table
4657  * @param tmp   scratch register
4658  */
4659 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4660         Register table0, Register table1, Register table2, Register table3,
4661         Register tmp, Register tmp2, Register tmp3) {
4662   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4663 
4664   if (UseCryptoPmullForCRC32) {
4665       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4666       return;
4667   }
4668 
4669   if (UseCRC32) {
4670       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4671       return;
4672   }
4673 
4674     mvnw(crc, crc);
4675 
4676     {
4677       uint64_t offset;
4678       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4679       add(table0, table0, offset);
4680     }
4681     add(table1, table0, 1*256*sizeof(juint));
4682     add(table2, table0, 2*256*sizeof(juint));
4683     add(table3, table0, 3*256*sizeof(juint));
4684 
4685     { // Neon code start
4686       cmp(len, (u1)64);
4687       br(Assembler::LT, L_by16);
4688       eor(v16, T16B, v16, v16);
4689 
4690     Label L_fold;
4691 
4692       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4693 
4694       ld1(v0, v1, T2D, post(buf, 32));
4695       ld1r(v4, T2D, post(tmp, 8));
4696       ld1r(v5, T2D, post(tmp, 8));
4697       ld1r(v6, T2D, post(tmp, 8));
4698       ld1r(v7, T2D, post(tmp, 8));
4699       mov(v16, S, 0, crc);
4700 
4701       eor(v0, T16B, v0, v16);
4702       sub(len, len, 64);
4703 
4704     BIND(L_fold);
4705       pmull(v22, T8H, v0, v5, T8B);
4706       pmull(v20, T8H, v0, v7, T8B);
4707       pmull(v23, T8H, v0, v4, T8B);
4708       pmull(v21, T8H, v0, v6, T8B);
4709 
4710       pmull2(v18, T8H, v0, v5, T16B);
4711       pmull2(v16, T8H, v0, v7, T16B);
4712       pmull2(v19, T8H, v0, v4, T16B);
4713       pmull2(v17, T8H, v0, v6, T16B);
4714 
4715       uzp1(v24, T8H, v20, v22);
4716       uzp2(v25, T8H, v20, v22);
4717       eor(v20, T16B, v24, v25);
4718 
4719       uzp1(v26, T8H, v16, v18);
4720       uzp2(v27, T8H, v16, v18);
4721       eor(v16, T16B, v26, v27);
4722 
4723       ushll2(v22, T4S, v20, T8H, 8);
4724       ushll(v20, T4S, v20, T4H, 8);
4725 
4726       ushll2(v18, T4S, v16, T8H, 8);
4727       ushll(v16, T4S, v16, T4H, 8);
4728 
4729       eor(v22, T16B, v23, v22);
4730       eor(v18, T16B, v19, v18);
4731       eor(v20, T16B, v21, v20);
4732       eor(v16, T16B, v17, v16);
4733 
4734       uzp1(v17, T2D, v16, v20);
4735       uzp2(v21, T2D, v16, v20);
4736       eor(v17, T16B, v17, v21);
4737 
4738       ushll2(v20, T2D, v17, T4S, 16);
4739       ushll(v16, T2D, v17, T2S, 16);
4740 
4741       eor(v20, T16B, v20, v22);
4742       eor(v16, T16B, v16, v18);
4743 
4744       uzp1(v17, T2D, v20, v16);
4745       uzp2(v21, T2D, v20, v16);
4746       eor(v28, T16B, v17, v21);
4747 
4748       pmull(v22, T8H, v1, v5, T8B);
4749       pmull(v20, T8H, v1, v7, T8B);
4750       pmull(v23, T8H, v1, v4, T8B);
4751       pmull(v21, T8H, v1, v6, T8B);
4752 
4753       pmull2(v18, T8H, v1, v5, T16B);
4754       pmull2(v16, T8H, v1, v7, T16B);
4755       pmull2(v19, T8H, v1, v4, T16B);
4756       pmull2(v17, T8H, v1, v6, T16B);
4757 
4758       ld1(v0, v1, T2D, post(buf, 32));
4759 
4760       uzp1(v24, T8H, v20, v22);
4761       uzp2(v25, T8H, v20, v22);
4762       eor(v20, T16B, v24, v25);
4763 
4764       uzp1(v26, T8H, v16, v18);
4765       uzp2(v27, T8H, v16, v18);
4766       eor(v16, T16B, v26, v27);
4767 
4768       ushll2(v22, T4S, v20, T8H, 8);
4769       ushll(v20, T4S, v20, T4H, 8);
4770 
4771       ushll2(v18, T4S, v16, T8H, 8);
4772       ushll(v16, T4S, v16, T4H, 8);
4773 
4774       eor(v22, T16B, v23, v22);
4775       eor(v18, T16B, v19, v18);
4776       eor(v20, T16B, v21, v20);
4777       eor(v16, T16B, v17, v16);
4778 
4779       uzp1(v17, T2D, v16, v20);
4780       uzp2(v21, T2D, v16, v20);
4781       eor(v16, T16B, v17, v21);
4782 
4783       ushll2(v20, T2D, v16, T4S, 16);
4784       ushll(v16, T2D, v16, T2S, 16);
4785 
4786       eor(v20, T16B, v22, v20);
4787       eor(v16, T16B, v16, v18);
4788 
4789       uzp1(v17, T2D, v20, v16);
4790       uzp2(v21, T2D, v20, v16);
4791       eor(v20, T16B, v17, v21);
4792 
4793       shl(v16, T2D, v28, 1);
4794       shl(v17, T2D, v20, 1);
4795 
4796       eor(v0, T16B, v0, v16);
4797       eor(v1, T16B, v1, v17);
4798 
4799       subs(len, len, 32);
4800       br(Assembler::GE, L_fold);
4801 
4802       mov(crc, 0);
4803       mov(tmp, v0, D, 0);
4804       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4805       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4806       mov(tmp, v0, D, 1);
4807       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4808       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4809       mov(tmp, v1, D, 0);
4810       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4811       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4812       mov(tmp, v1, D, 1);
4813       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4814       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4815 
4816       add(len, len, 32);
4817     } // Neon code end
4818 
4819   BIND(L_by16);
4820     subs(len, len, 16);
4821     br(Assembler::GE, L_by16_loop);
4822     adds(len, len, 16-4);
4823     br(Assembler::GE, L_by4_loop);
4824     adds(len, len, 4);
4825     br(Assembler::GT, L_by1_loop);
4826     b(L_exit);
4827 
4828   BIND(L_by4_loop);
4829     ldrw(tmp, Address(post(buf, 4)));
4830     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4831     subs(len, len, 4);
4832     br(Assembler::GE, L_by4_loop);
4833     adds(len, len, 4);
4834     br(Assembler::LE, L_exit);
4835   BIND(L_by1_loop);
4836     subs(len, len, 1);
4837     ldrb(tmp, Address(post(buf, 1)));
4838     update_byte_crc32(crc, tmp, table0);
4839     br(Assembler::GT, L_by1_loop);
4840     b(L_exit);
4841 
4842     align(CodeEntryAlignment);
4843   BIND(L_by16_loop);
4844     subs(len, len, 16);
4845     ldp(tmp, tmp3, Address(post(buf, 16)));
4846     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4847     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4848     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4849     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4850     br(Assembler::GE, L_by16_loop);
4851     adds(len, len, 16-4);
4852     br(Assembler::GE, L_by4_loop);
4853     adds(len, len, 4);
4854     br(Assembler::GT, L_by1_loop);
4855   BIND(L_exit);
4856     mvnw(crc, crc);
4857 }
4858 
4859 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4860         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4861     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4862     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4863 
4864     subs(tmp0, len, 384);
4865     br(Assembler::GE, CRC_by128_pre);
4866   BIND(CRC_less128);
4867     subs(len, len, 32);
4868     br(Assembler::GE, CRC_by32_loop);
4869   BIND(CRC_less32);
4870     adds(len, len, 32 - 4);
4871     br(Assembler::GE, CRC_by4_loop);
4872     adds(len, len, 4);
4873     br(Assembler::GT, CRC_by1_loop);
4874     b(L_exit);
4875 
4876   BIND(CRC_by32_loop);
4877     ldp(tmp0, tmp1, Address(buf));
4878     crc32cx(crc, crc, tmp0);
4879     ldr(tmp2, Address(buf, 16));
4880     crc32cx(crc, crc, tmp1);
4881     ldr(tmp3, Address(buf, 24));
4882     crc32cx(crc, crc, tmp2);
4883     add(buf, buf, 32);
4884     subs(len, len, 32);
4885     crc32cx(crc, crc, tmp3);
4886     br(Assembler::GE, CRC_by32_loop);
4887     cmn(len, (u1)32);
4888     br(Assembler::NE, CRC_less32);
4889     b(L_exit);
4890 
4891   BIND(CRC_by4_loop);
4892     ldrw(tmp0, Address(post(buf, 4)));
4893     subs(len, len, 4);
4894     crc32cw(crc, crc, tmp0);
4895     br(Assembler::GE, CRC_by4_loop);
4896     adds(len, len, 4);
4897     br(Assembler::LE, L_exit);
4898   BIND(CRC_by1_loop);
4899     ldrb(tmp0, Address(post(buf, 1)));
4900     subs(len, len, 1);
4901     crc32cb(crc, crc, tmp0);
4902     br(Assembler::GT, CRC_by1_loop);
4903     b(L_exit);
4904 
4905   BIND(CRC_by128_pre);
4906     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4907       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4908     mov(crc, 0);
4909     crc32cx(crc, crc, tmp0);
4910     crc32cx(crc, crc, tmp1);
4911 
4912     cbnz(len, CRC_less128);
4913 
4914   BIND(L_exit);
4915 }
4916 
4917 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4918         Register len, Register tmp0, Register tmp1, Register tmp2,
4919         Register tmp3) {
4920     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4921     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4922 
4923     subs(len, len, 128);
4924     br(Assembler::GE, CRC_by64_pre);
4925   BIND(CRC_less64);
4926     adds(len, len, 128-32);
4927     br(Assembler::GE, CRC_by32_loop);
4928   BIND(CRC_less32);
4929     adds(len, len, 32-4);
4930     br(Assembler::GE, CRC_by4_loop);
4931     adds(len, len, 4);
4932     br(Assembler::GT, CRC_by1_loop);
4933     b(L_exit);
4934 
4935   BIND(CRC_by32_loop);
4936     ldp(tmp0, tmp1, Address(post(buf, 16)));
4937     subs(len, len, 32);
4938     crc32cx(crc, crc, tmp0);
4939     ldr(tmp2, Address(post(buf, 8)));
4940     crc32cx(crc, crc, tmp1);
4941     ldr(tmp3, Address(post(buf, 8)));
4942     crc32cx(crc, crc, tmp2);
4943     crc32cx(crc, crc, tmp3);
4944     br(Assembler::GE, CRC_by32_loop);
4945     cmn(len, (u1)32);
4946     br(Assembler::NE, CRC_less32);
4947     b(L_exit);
4948 
4949   BIND(CRC_by4_loop);
4950     ldrw(tmp0, Address(post(buf, 4)));
4951     subs(len, len, 4);
4952     crc32cw(crc, crc, tmp0);
4953     br(Assembler::GE, CRC_by4_loop);
4954     adds(len, len, 4);
4955     br(Assembler::LE, L_exit);
4956   BIND(CRC_by1_loop);
4957     ldrb(tmp0, Address(post(buf, 1)));
4958     subs(len, len, 1);
4959     crc32cb(crc, crc, tmp0);
4960     br(Assembler::GT, CRC_by1_loop);
4961     b(L_exit);
4962 
4963   BIND(CRC_by64_pre);
4964     sub(buf, buf, 8);
4965     ldp(tmp0, tmp1, Address(buf, 8));
4966     crc32cx(crc, crc, tmp0);
4967     ldr(tmp2, Address(buf, 24));
4968     crc32cx(crc, crc, tmp1);
4969     ldr(tmp3, Address(buf, 32));
4970     crc32cx(crc, crc, tmp2);
4971     ldr(tmp0, Address(buf, 40));
4972     crc32cx(crc, crc, tmp3);
4973     ldr(tmp1, Address(buf, 48));
4974     crc32cx(crc, crc, tmp0);
4975     ldr(tmp2, Address(buf, 56));
4976     crc32cx(crc, crc, tmp1);
4977     ldr(tmp3, Address(pre(buf, 64)));
4978 
4979     b(CRC_by64_loop);
4980 
4981     align(CodeEntryAlignment);
4982   BIND(CRC_by64_loop);
4983     subs(len, len, 64);
4984     crc32cx(crc, crc, tmp2);
4985     ldr(tmp0, Address(buf, 8));
4986     crc32cx(crc, crc, tmp3);
4987     ldr(tmp1, Address(buf, 16));
4988     crc32cx(crc, crc, tmp0);
4989     ldr(tmp2, Address(buf, 24));
4990     crc32cx(crc, crc, tmp1);
4991     ldr(tmp3, Address(buf, 32));
4992     crc32cx(crc, crc, tmp2);
4993     ldr(tmp0, Address(buf, 40));
4994     crc32cx(crc, crc, tmp3);
4995     ldr(tmp1, Address(buf, 48));
4996     crc32cx(crc, crc, tmp0);
4997     ldr(tmp2, Address(buf, 56));
4998     crc32cx(crc, crc, tmp1);
4999     ldr(tmp3, Address(pre(buf, 64)));
5000     br(Assembler::GE, CRC_by64_loop);
5001 
5002     // post-loop
5003     crc32cx(crc, crc, tmp2);
5004     crc32cx(crc, crc, tmp3);
5005 
5006     sub(len, len, 64);
5007     add(buf, buf, 8);
5008     cmn(len, (u1)128);
5009     br(Assembler::NE, CRC_less64);
5010   BIND(L_exit);
5011 }
5012 
5013 /**
5014  * @param crc   register containing existing CRC (32-bit)
5015  * @param buf   register pointing to input byte buffer (byte*)
5016  * @param len   register containing number of bytes
5017  * @param table register that will contain address of CRC table
5018  * @param tmp   scratch register
5019  */
5020 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
5021         Register table0, Register table1, Register table2, Register table3,
5022         Register tmp, Register tmp2, Register tmp3) {
5023   if (UseCryptoPmullForCRC32) {
5024     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
5025   } else {
5026     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
5027   }
5028 }
5029 
5030 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
5031         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
5032     Label CRC_by128_loop;
5033     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
5034 
5035     sub(len, len, 256);
5036     Register table = tmp0;
5037     {
5038       uint64_t offset;
5039       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
5040       add(table, table, offset);
5041     }
5042     add(table, table, table_offset);
5043 
5044     // Registers v0..v7 are used as data registers.
5045     // Registers v16..v31 are used as tmp registers.
5046     sub(buf, buf, 0x10);
5047     ldrq(v0, Address(buf, 0x10));
5048     ldrq(v1, Address(buf, 0x20));
5049     ldrq(v2, Address(buf, 0x30));
5050     ldrq(v3, Address(buf, 0x40));
5051     ldrq(v4, Address(buf, 0x50));
5052     ldrq(v5, Address(buf, 0x60));
5053     ldrq(v6, Address(buf, 0x70));
5054     ldrq(v7, Address(pre(buf, 0x80)));
5055 
5056     movi(v31, T4S, 0);
5057     mov(v31, S, 0, crc);
5058     eor(v0, T16B, v0, v31);
5059 
5060     // Register v16 contains constants from the crc table.
5061     ldrq(v16, Address(table));
5062     b(CRC_by128_loop);
5063 
5064     align(OptoLoopAlignment);
5065   BIND(CRC_by128_loop);
5066     pmull (v17,  T1Q, v0, v16, T1D);
5067     pmull2(v18, T1Q, v0, v16, T2D);
5068     ldrq(v0, Address(buf, 0x10));
5069     eor3(v0, T16B, v17,  v18, v0);
5070 
5071     pmull (v19, T1Q, v1, v16, T1D);
5072     pmull2(v20, T1Q, v1, v16, T2D);
5073     ldrq(v1, Address(buf, 0x20));
5074     eor3(v1, T16B, v19, v20, v1);
5075 
5076     pmull (v21, T1Q, v2, v16, T1D);
5077     pmull2(v22, T1Q, v2, v16, T2D);
5078     ldrq(v2, Address(buf, 0x30));
5079     eor3(v2, T16B, v21, v22, v2);
5080 
5081     pmull (v23, T1Q, v3, v16, T1D);
5082     pmull2(v24, T1Q, v3, v16, T2D);
5083     ldrq(v3, Address(buf, 0x40));
5084     eor3(v3, T16B, v23, v24, v3);
5085 
5086     pmull (v25, T1Q, v4, v16, T1D);
5087     pmull2(v26, T1Q, v4, v16, T2D);
5088     ldrq(v4, Address(buf, 0x50));
5089     eor3(v4, T16B, v25, v26, v4);
5090 
5091     pmull (v27, T1Q, v5, v16, T1D);
5092     pmull2(v28, T1Q, v5, v16, T2D);
5093     ldrq(v5, Address(buf, 0x60));
5094     eor3(v5, T16B, v27, v28, v5);
5095 
5096     pmull (v29, T1Q, v6, v16, T1D);
5097     pmull2(v30, T1Q, v6, v16, T2D);
5098     ldrq(v6, Address(buf, 0x70));
5099     eor3(v6, T16B, v29, v30, v6);
5100 
5101     // Reuse registers v23, v24.
5102     // Using them won't block the first instruction of the next iteration.
5103     pmull (v23, T1Q, v7, v16, T1D);
5104     pmull2(v24, T1Q, v7, v16, T2D);
5105     ldrq(v7, Address(pre(buf, 0x80)));
5106     eor3(v7, T16B, v23, v24, v7);
5107 
5108     subs(len, len, 0x80);
5109     br(Assembler::GE, CRC_by128_loop);
5110 
5111     // fold into 512 bits
5112     // Use v31 for constants because v16 can be still in use.
5113     ldrq(v31, Address(table, 0x10));
5114 
5115     pmull (v17,  T1Q, v0, v31, T1D);
5116     pmull2(v18, T1Q, v0, v31, T2D);
5117     eor3(v0, T16B, v17, v18, v4);
5118 
5119     pmull (v19, T1Q, v1, v31, T1D);
5120     pmull2(v20, T1Q, v1, v31, T2D);
5121     eor3(v1, T16B, v19, v20, v5);
5122 
5123     pmull (v21, T1Q, v2, v31, T1D);
5124     pmull2(v22, T1Q, v2, v31, T2D);
5125     eor3(v2, T16B, v21, v22, v6);
5126 
5127     pmull (v23, T1Q, v3, v31, T1D);
5128     pmull2(v24, T1Q, v3, v31, T2D);
5129     eor3(v3, T16B, v23, v24, v7);
5130 
5131     // fold into 128 bits
5132     // Use v17 for constants because v31 can be still in use.
5133     ldrq(v17, Address(table, 0x20));
5134     pmull (v25, T1Q, v0, v17, T1D);
5135     pmull2(v26, T1Q, v0, v17, T2D);
5136     eor3(v3, T16B, v3, v25, v26);
5137 
5138     // Use v18 for constants because v17 can be still in use.
5139     ldrq(v18, Address(table, 0x30));
5140     pmull (v27, T1Q, v1, v18, T1D);
5141     pmull2(v28, T1Q, v1, v18, T2D);
5142     eor3(v3, T16B, v3, v27, v28);
5143 
5144     // Use v19 for constants because v18 can be still in use.
5145     ldrq(v19, Address(table, 0x40));
5146     pmull (v29, T1Q, v2, v19, T1D);
5147     pmull2(v30, T1Q, v2, v19, T2D);
5148     eor3(v0, T16B, v3, v29, v30);
5149 
5150     add(len, len, 0x80);
5151     add(buf, buf, 0x10);
5152 
5153     mov(tmp0, v0, D, 0);
5154     mov(tmp1, v0, D, 1);
5155 }
5156 
5157 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5158   Address adr;
5159   switch(dst.getMode()) {
5160   case Address::base_plus_offset:
5161     // This is the expected mode, although we allow all the other
5162     // forms below.
5163     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5164     break;
5165   default:
5166     lea(rscratch2, dst);
5167     adr = Address(rscratch2);
5168     break;
5169   }
5170   ldr(rscratch1, adr);
5171   add(rscratch1, rscratch1, src);
5172   str(rscratch1, adr);
5173 }
5174 
5175 void MacroAssembler::cmpptr(Register src1, Address src2) {
5176   uint64_t offset;
5177   adrp(rscratch1, src2, offset);
5178   ldr(rscratch1, Address(rscratch1, offset));
5179   cmp(src1, rscratch1);
5180 }
5181 
5182 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5183   cmp(obj1, obj2);
5184 }
5185 
5186 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5187   load_method_holder(rresult, rmethod);
5188   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5189 }
5190 
5191 void MacroAssembler::load_method_holder(Register holder, Register method) {
5192   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5193   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5194   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5195 }
5196 
5197 void MacroAssembler::load_metadata(Register dst, Register src) {
5198   if (UseCompactObjectHeaders) {
5199     load_narrow_klass_compact(dst, src);
5200   } else {
5201     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5202   }
5203 }
5204 
5205 // Loads the obj's narrow Klass from a compact object header (+COH) into dst.
5206 // Preserves all registers (incl src, rscratch1 and rscratch2).
5207 // Input:
5208 // src - the oop we want to load the klass from.
5209 // dst - output narrow klass.
5210 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5211   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5212   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5213   lsr(dst, dst, markWord::klass_shift);
5214 }
5215 
5216 // Loads the obj's narrow Klass from any header (compact or not) into dst.
5217 void MacroAssembler::load_narrow_klass(Register dst, Register src) {
5218   if (UseCompactObjectHeaders) {
5219     load_narrow_klass_compact(dst, src);
5220   } else {
5221     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5222   }
5223 }
5224 
5225 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
5226   load_narrow_klass(dst, src);
5227   decode_klass_not_null(dst, dst, tmp);
5228 }
5229 
5230 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5231   if (RestoreMXCSROnJNICalls) {
5232     Label OK;
5233     get_fpcr(tmp1);
5234     mov(tmp2, tmp1);
5235     // Set FPCR to the state we need. We do want Round to Nearest. We
5236     // don't want non-IEEE rounding modes or floating-point traps.
5237     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5238     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5239     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5240     eor(tmp2, tmp1, tmp2);
5241     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5242     set_fpcr(tmp1);
5243     bind(OK);
5244   }
5245 }
5246 
5247 // ((OopHandle)result).resolve();
5248 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5249   // OopHandle::resolve is an indirection.
5250   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5251 }
5252 
5253 // ((WeakHandle)result).resolve();
5254 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5255   assert_different_registers(result, tmp1, tmp2);
5256   Label resolved;
5257 
5258   // A null weak handle resolves to null.
5259   cbz(result, resolved);
5260 
5261   // Only 64 bit platforms support GCs that require a tmp register
5262   // WeakHandle::resolve is an indirection like jweak.
5263   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5264                  result, Address(result), tmp1, tmp2);
5265   bind(resolved);
5266 }
5267 
5268 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5269   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5270   ldr(dst, Address(rmethod, Method::const_offset()));
5271   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5272   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5273   ldr(dst, Address(dst, mirror_offset));
5274   resolve_oop_handle(dst, tmp1, tmp2);
5275 }
5276 
5277 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp, Register tmp2) {
5278   assert_different_registers(obj, klass, tmp, tmp2);
5279   if (UseCompactObjectHeaders) {
5280     load_narrow_klass_compact(tmp, obj);
5281   } else {
5282     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5283   }
5284   if (CompressedKlassPointers::base() == nullptr) {
5285     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5286     return;
5287   } else if (!AOTCodeCache::is_on_for_dump() &&
5288              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5289              && CompressedKlassPointers::shift() == 0) {
5290     // Only the bottom 32 bits matter
5291     cmpw(klass, tmp);
5292     return;
5293   }
5294   decode_klass_not_null(tmp, tmp, tmp2);
5295   cmp(klass, tmp);
5296 }
5297 
5298 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5299   if (UseCompactObjectHeaders) {
5300     load_narrow_klass_compact(tmp1, obj1);
5301     load_narrow_klass_compact(tmp2, obj2);
5302   } else {
5303     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5304     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5305   }
5306   cmpw(tmp1, tmp2);
5307 }
5308 
5309 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5310   Register tmp = (dst == rscratch1) ? rscratch2 : rscratch1;
5311   load_klass(dst, src, tmp);
5312   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5313 }
5314 
5315 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
5316   // FIXME: Should this be a store release?  concurrent gcs assumes
5317   // klass length is valid if klass field is not null.
5318   assert(!UseCompactObjectHeaders, "not with compact headers");
5319   encode_klass_not_null(src, src, tmp);
5320   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5321 }
5322 
5323 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5324   assert(!UseCompactObjectHeaders, "not with compact headers");
5325   // Store to klass gap in destination
5326   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5327 }
5328 
5329 // Algorithm must match CompressedOops::encode.
5330 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5331 #ifdef ASSERT
5332   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5333 #endif
5334   verify_oop_msg(s, "broken oop in encode_heap_oop");
5335   if (CompressedOops::base() == nullptr) {
5336     if (CompressedOops::shift() != 0) {
5337       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5338       lsr(d, s, LogMinObjAlignmentInBytes);
5339     } else {
5340       mov(d, s);
5341     }
5342   } else {
5343     subs(d, s, rheapbase);
5344     csel(d, d, zr, Assembler::HS);
5345     lsr(d, d, LogMinObjAlignmentInBytes);
5346 
5347     /*  Old algorithm: is this any worse?
5348     Label nonnull;
5349     cbnz(r, nonnull);
5350     sub(r, r, rheapbase);
5351     bind(nonnull);
5352     lsr(r, r, LogMinObjAlignmentInBytes);
5353     */
5354   }
5355 }
5356 
5357 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5358 #ifdef ASSERT
5359   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5360   if (CheckCompressedOops) {
5361     Label ok;
5362     cbnz(r, ok);
5363     stop("null oop passed to encode_heap_oop_not_null");
5364     bind(ok);
5365   }
5366 #endif
5367   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5368   if (CompressedOops::base() != nullptr) {
5369     sub(r, r, rheapbase);
5370   }
5371   if (CompressedOops::shift() != 0) {
5372     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5373     lsr(r, r, LogMinObjAlignmentInBytes);
5374   }
5375 }
5376 
5377 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5378 #ifdef ASSERT
5379   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5380   if (CheckCompressedOops) {
5381     Label ok;
5382     cbnz(src, ok);
5383     stop("null oop passed to encode_heap_oop_not_null2");
5384     bind(ok);
5385   }
5386 #endif
5387   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5388 
5389   Register data = src;
5390   if (CompressedOops::base() != nullptr) {
5391     sub(dst, src, rheapbase);
5392     data = dst;
5393   }
5394   if (CompressedOops::shift() != 0) {
5395     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5396     lsr(dst, data, LogMinObjAlignmentInBytes);
5397     data = dst;
5398   }
5399   if (data == src)
5400     mov(dst, src);
5401 }
5402 
5403 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5404 #ifdef ASSERT
5405   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5406 #endif
5407   if (CompressedOops::base() == nullptr) {
5408     if (CompressedOops::shift() != 0) {
5409       lsl(d, s, CompressedOops::shift());
5410     } else if (d != s) {
5411       mov(d, s);
5412     }
5413   } else {
5414     Label done;
5415     if (d != s)
5416       mov(d, s);
5417     cbz(s, done);
5418     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5419     bind(done);
5420   }
5421   verify_oop_msg(d, "broken oop in decode_heap_oop");
5422 }
5423 
5424 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5425   assert (UseCompressedOops, "should only be used for compressed headers");
5426   assert (Universe::heap() != nullptr, "java heap should be initialized");
5427   // Cannot assert, unverified entry point counts instructions (see .ad file)
5428   // vtableStubs also counts instructions in pd_code_size_limit.
5429   // Also do not verify_oop as this is called by verify_oop.
5430   if (CompressedOops::shift() != 0) {
5431     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5432     if (CompressedOops::base() != nullptr) {
5433       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5434     } else {
5435       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5436     }
5437   } else {
5438     assert (CompressedOops::base() == nullptr, "sanity");
5439   }
5440 }
5441 
5442 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5443   assert (UseCompressedOops, "should only be used for compressed headers");
5444   assert (Universe::heap() != nullptr, "java heap should be initialized");
5445   // Cannot assert, unverified entry point counts instructions (see .ad file)
5446   // vtableStubs also counts instructions in pd_code_size_limit.
5447   // Also do not verify_oop as this is called by verify_oop.
5448   if (CompressedOops::shift() != 0) {
5449     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5450     if (CompressedOops::base() != nullptr) {
5451       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5452     } else {
5453       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5454     }
5455   } else {
5456     assert (CompressedOops::base() == nullptr, "sanity");
5457     if (dst != src) {
5458       mov(dst, src);
5459     }
5460   }
5461 }
5462 
5463 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5464 
5465 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5466   assert(Metaspace::initialized(), "metaspace not initialized yet");
5467   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5468   return _klass_decode_mode;
5469 }
5470 
5471 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5472 
5473   if (base == nullptr) {
5474     return KlassDecodeZero;
5475   }
5476 
5477   if (operand_valid_for_logical_immediate(
5478         /*is32*/false, (uint64_t)base)) {
5479     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5480     if (((uint64_t)base & range_mask) == 0) {
5481       return KlassDecodeXor;
5482     }
5483   }
5484 
5485   const uint64_t shifted_base =
5486     (uint64_t)base >> shift;
5487   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5488     return KlassDecodeMovk;
5489   }
5490 
5491   return KlassDecodeFallback;
5492 }
5493 
5494 void MacroAssembler::initialize_klass_decode_mode(address base, int shift, const size_t range) {
5495   // KlassDecodeMode shouldn't be set already.
5496   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5497   _klass_decode_mode = klass_decode_mode(base, shift, range);
5498   log_info(metaspace)("Klass Decode Mode: %d", (int)_klass_decode_mode);
5499 }
5500 
5501 void MacroAssembler::encode_klass_not_null(Register dst, Register src, Register tmp) {
5502   emit_encode_klass_not_null(dst, src, tmp, CompressedKlassPointers::base(),
5503                              CompressedKlassPointers::shift(), klass_decode_mode());
5504 }
5505 
5506 void MacroAssembler::emit_encode_klass_not_null(Register dst, Register src, Register tmp,
5507                                                 address base, int shift, KlassDecodeMode decode_mode) {
5508 
5509   assert_different_registers(tmp, src);
5510   assert(tmp != noreg, "valid tmp required");
5511 
5512   if (AOTCodeCache::is_on_for_dump()) {
5513     // We are generating code during AOT buildup that will run in *future* processes
5514     // with likely different encoding settings. Therefore, we have to load the
5515     // encoding base dynamically, we cannot just bake it in as immediate.
5516     // Note that we only need to do this for base. The encoding shift would be the
5517     // same between build time and runtime: the standard precomputed shift.
5518     assert(shift == ArchiveBuilder::precomputed_narrow_klass_shift(), "unexpected compressed klass shift!");
5519     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5520     ldr(tmp, tmp);
5521     sub(dst, src, tmp);
5522     lsr(dst, dst, shift);
5523     return;
5524   }
5525 
5526   switch (decode_mode) {
5527   case KlassDecodeZero:
5528     lsr(dst, src, shift);
5529     break;
5530 
5531   case KlassDecodeXor:
5532     eor(dst, src, (uint64_t)base);
5533     lsr(dst, dst, shift);
5534     break;
5535 
5536   case KlassDecodeMovk:
5537     if (shift != 0) {
5538       ubfx(dst, src, shift, 32);
5539     } else {
5540       movw(dst, src);
5541     }
5542     break;
5543 
5544   case KlassDecodeFallback: {
5545     mov(tmp, base);
5546     sub(dst, src, tmp);
5547     lsr(dst, dst, shift);
5548     break;
5549   }
5550 
5551   case KlassDecodeNone:
5552     ShouldNotReachHere();
5553     break;
5554   }
5555 
5556 #ifdef ASSERT
5557   if (tmp != dst) {
5558     mov(tmp, 0xdead);
5559   }
5560 #endif // ASSERT
5561 
5562 }
5563 
5564 void  MacroAssembler::decode_klass_not_null(Register dst, Register src, Register tmp) {
5565   emit_decode_klass_not_null(dst, src, tmp,
5566                              CompressedKlassPointers::base(),
5567                              CompressedKlassPointers::shift(),
5568                              klass_decode_mode());
5569 }
5570 
5571 void MacroAssembler::emit_decode_klass_not_null(Register dst, Register src, Register tmp,
5572                                                 address base, int shift, KlassDecodeMode decode_mode) {
5573 
5574   assert_different_registers(tmp, src);
5575   assert(tmp != noreg, "valid tmp required");
5576 
5577   if (AOTCodeCache::is_on_for_dump()) {
5578     // We are generating code during AOT buildup that will run in *future* processes
5579     // with likely different encoding settings. Therefore, we have to load the
5580     // encoding base dynamically, we cannot just bake it in as immediate.
5581     // Note that we only need to do this for base. The encoding shift would be the
5582     // same between build time and runtime: the standard precomputed shift.
5583     assert(shift == ArchiveBuilder::precomputed_narrow_klass_shift(), "unexpected compressed klass shift!");
5584     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5585     ldr(tmp, tmp);
5586     add(dst, tmp,  src, LSL,  shift);
5587     return;
5588   }
5589 
5590   switch (decode_mode) {
5591   case KlassDecodeZero: // 0-1 instructions
5592     lsl(dst, src, shift);
5593     break;
5594 
5595   case KlassDecodeXor: // 1-2 instructions
5596     lsl(dst, src, shift);
5597     eor(dst, dst, (uint64_t)base);
5598     break;
5599 
5600   case KlassDecodeMovk: { // 1-3 instructions
5601     const uint64_t shifted_base =
5602       (uint64_t)base >> shift;
5603 
5604     if (dst != src) movw(dst, src);
5605     movk(dst, shifted_base >> 32, 32);
5606     lsl(dst, dst, shift);
5607     break;
5608   }
5609 
5610   case KlassDecodeFallback: { // 3-4 instructions
5611     mov(tmp, base);
5612     add(dst, tmp, src, LSL, shift);
5613     break;
5614   }
5615 
5616   case KlassDecodeNone:
5617     ShouldNotReachHere();
5618     break;
5619   }
5620 
5621 #ifdef ASSERT
5622   // Always clobber tmp
5623   if (tmp != dst) {
5624     mov(tmp, 0xdead);
5625   }
5626 #endif // ASSERT
5627 
5628 }
5629 
5630 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5631 #ifdef ASSERT
5632   {
5633     ThreadInVMfromUnknown tiv;
5634     assert (UseCompressedOops, "should only be used for compressed oops");
5635     assert (Universe::heap() != nullptr, "java heap should be initialized");
5636     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5637     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5638   }
5639 #endif
5640   int oop_index = oop_recorder()->find_index(obj);
5641   InstructionMark im(this);
5642   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5643   code_section()->relocate(inst_mark(), rspec);
5644   movz(dst, 0xDEAD, 16);
5645   movk(dst, 0xBEEF);
5646 }
5647 
5648 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5649   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5650   int index = oop_recorder()->find_index(k);
5651 
5652   InstructionMark im(this);
5653   RelocationHolder rspec = metadata_Relocation::spec(index);
5654   code_section()->relocate(inst_mark(), rspec);
5655   narrowKlass nk = CompressedKlassPointers::encode(k);
5656   movz(dst, (nk >> 16), 16);
5657   movk(dst, nk & 0xffff);
5658 }
5659 
5660 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5661                                     Register dst, Address src,
5662                                     Register tmp1, Register tmp2) {
5663   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5664   decorators = AccessInternal::decorator_fixup(decorators, type);
5665   bool as_raw = (decorators & AS_RAW) != 0;
5666   if (as_raw) {
5667     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5668   } else {
5669     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5670   }
5671 }
5672 
5673 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5674                                      Address dst, Register val,
5675                                      Register tmp1, Register tmp2, Register tmp3) {
5676   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5677   decorators = AccessInternal::decorator_fixup(decorators, type);
5678   bool as_raw = (decorators & AS_RAW) != 0;
5679   if (as_raw) {
5680     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5681   } else {
5682     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5683   }
5684 }
5685 
5686 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5687                                      Register inline_layout_info) {
5688   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5689   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5690 }
5691 
5692 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5693   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5694   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5695 }
5696 
5697 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5698   // ((address) (void*) o) + vk->payload_offset();
5699   Register offset = (data == oop) ? rscratch1 : data;
5700   payload_offset(inline_klass, offset);
5701   if (data == oop) {
5702     add(data, data, offset);
5703   } else {
5704     lea(data, Address(oop, offset));
5705   }
5706 }
5707 
5708 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5709                                    Register tmp2, DecoratorSet decorators) {
5710   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5711 }
5712 
5713 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5714                                             Register tmp2, DecoratorSet decorators) {
5715   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5716 }
5717 
5718 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5719                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5720   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5721 }
5722 
5723 // Used for storing nulls.
5724 void MacroAssembler::store_heap_oop_null(Address dst) {
5725   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5726 }
5727 
5728 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5729   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5730   int index = oop_recorder()->allocate_metadata_index(obj);
5731   RelocationHolder rspec = metadata_Relocation::spec(index);
5732   return Address((address)obj, rspec);
5733 }
5734 
5735 // Move an oop into a register.
5736 void MacroAssembler::movoop(Register dst, jobject obj) {
5737   int oop_index;
5738   if (obj == nullptr) {
5739     oop_index = oop_recorder()->allocate_oop_index(obj);
5740   } else {
5741 #ifdef ASSERT
5742     {
5743       ThreadInVMfromUnknown tiv;
5744       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5745     }
5746 #endif
5747     oop_index = oop_recorder()->find_index(obj);
5748   }
5749   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5750 
5751   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5752     mov(dst, Address((address)obj, rspec));
5753   } else {
5754     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5755     ldr(dst, Address(dummy, rspec));
5756   }
5757 }
5758 
5759 // Move a metadata address into a register.
5760 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5761   int oop_index;
5762   if (obj == nullptr) {
5763     oop_index = oop_recorder()->allocate_metadata_index(obj);
5764   } else {
5765     oop_index = oop_recorder()->find_index(obj);
5766   }
5767   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5768   mov(dst, Address((address)obj, rspec));
5769 }
5770 
5771 Address MacroAssembler::constant_oop_address(jobject obj) {
5772 #ifdef ASSERT
5773   {
5774     ThreadInVMfromUnknown tiv;
5775     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5776     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5777   }
5778 #endif
5779   int oop_index = oop_recorder()->find_index(obj);
5780   return Address((address)obj, oop_Relocation::spec(oop_index));
5781 }
5782 
5783 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5784 void MacroAssembler::tlab_allocate(Register obj,
5785                                    Register var_size_in_bytes,
5786                                    int con_size_in_bytes,
5787                                    Register t1,
5788                                    Register t2,
5789                                    Label& slow_case) {
5790   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5791   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5792 }
5793 
5794 void MacroAssembler::verify_tlab() {
5795 #ifdef ASSERT
5796   if (UseTLAB && VerifyOops) {
5797     Label next, ok;
5798 
5799     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5800 
5801     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5802     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5803     cmp(rscratch2, rscratch1);
5804     br(Assembler::HS, next);
5805     STOP("assert(top >= start)");
5806     should_not_reach_here();
5807 
5808     bind(next);
5809     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5810     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5811     cmp(rscratch2, rscratch1);
5812     br(Assembler::HS, ok);
5813     STOP("assert(top <= end)");
5814     should_not_reach_here();
5815 
5816     bind(ok);
5817     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5818   }
5819 #endif
5820 }
5821 
5822 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5823   assert_different_registers(holder_klass, index, layout_info);
5824   InlineLayoutInfo array[2];
5825   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5826   if (is_power_of_2(size)) {
5827     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5828   } else {
5829     mov(layout_info, size);
5830     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5831   }
5832   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5833   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5834   lea(layout_info, Address(layout_info, index));
5835 }
5836 
5837 // Writes to stack successive pages until offset reached to check for
5838 // stack overflow + shadow pages.  This clobbers tmp.
5839 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5840   assert_different_registers(tmp, size, rscratch1);
5841   mov(tmp, sp);
5842   // Bang stack for total size given plus shadow page size.
5843   // Bang one page at a time because large size can bang beyond yellow and
5844   // red zones.
5845   Label loop;
5846   mov(rscratch1, (int)os::vm_page_size());
5847   bind(loop);
5848   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5849   subsw(size, size, rscratch1);
5850   str(size, Address(tmp));
5851   br(Assembler::GT, loop);
5852 
5853   // Bang down shadow pages too.
5854   // At this point, (tmp-0) is the last address touched, so don't
5855   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5856   // was post-decremented.)  Skip this address by starting at i=1, and
5857   // touch a few more pages below.  N.B.  It is important to touch all
5858   // the way down to and including i=StackShadowPages.
5859   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5860     // this could be any sized move but this is can be a debugging crumb
5861     // so the bigger the better.
5862     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5863     str(size, Address(tmp));
5864   }
5865 }
5866 
5867 // Move the address of the polling page into dest.
5868 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5869   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5870 }
5871 
5872 // Read the polling page.  The address of the polling page must
5873 // already be in r.
5874 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5875   address mark;
5876   {
5877     InstructionMark im(this);
5878     code_section()->relocate(inst_mark(), rtype);
5879     ldrw(zr, Address(r, 0));
5880     mark = inst_mark();
5881   }
5882   verify_cross_modify_fence_not_required();
5883   return mark;
5884 }
5885 
5886 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5887   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5888   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5889   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5890   int64_t offset_low = dest_page - low_page;
5891   int64_t offset_high = dest_page - high_page;
5892 
5893   assert(is_valid_AArch64_address(dest.target()), "bad address");
5894   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5895 
5896   InstructionMark im(this);
5897   code_section()->relocate(inst_mark(), dest.rspec());
5898   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5899   // the code cache so that if it is relocated we know it will still reach
5900   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5901     _adrp(reg1, dest.target());
5902   } else {
5903     uint64_t target = (uint64_t)dest.target();
5904     uint64_t adrp_target
5905       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5906 
5907     _adrp(reg1, (address)adrp_target);
5908     movk(reg1, target >> 32, 32);
5909   }
5910   byte_offset = (uint64_t)dest.target() & 0xfff;
5911 }
5912 
5913 void MacroAssembler::load_byte_map_base(Register reg) {
5914 #if INCLUDE_CDS
5915   if (AOTCodeCache::is_on_for_dump()) {
5916     address byte_map_base_adr = AOTRuntimeConstants::card_table_base_address();
5917     lea(reg, ExternalAddress(byte_map_base_adr));
5918     ldr(reg, Address(reg));
5919     return;
5920   }
5921 #endif
5922   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
5923 
5924   // Strictly speaking the card table base isn't an address at all, and it might
5925   // even be negative. It is thus materialised as a constant.
5926   mov(reg, (uint64_t)ctbs->card_table_base_const());
5927 }
5928 
5929 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5930 #if INCLUDE_CDS
5931   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5932   if (AOTCodeCache::is_on_for_dump()) {
5933     // all aotrc field addresses should be registered in the AOTCodeCache address table
5934     lea(reg, ExternalAddress(a));
5935   } else {
5936     mov(reg, (uint64_t)a);
5937   }
5938 #else
5939   ShouldNotReachHere();
5940 #endif
5941 }
5942 
5943 #ifdef ASSERT
5944 void MacroAssembler::build_frame(int framesize) {
5945   build_frame(framesize, false);
5946 }
5947 #endif
5948 
5949 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
5950   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5951   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5952   protect_return_address();
5953   if (framesize < ((1 << 9) + 2 * wordSize)) {
5954     sub(sp, sp, framesize);
5955     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5956       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5957       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
5958     } else {
5959       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5960     }
5961     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5962   } else {
5963     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
5964       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
5965       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
5966     } else {
5967       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5968     }
5969     if (PreserveFramePointer) mov(rfp, sp);
5970     if (framesize < ((1 << 12) + 2 * wordSize))
5971       sub(sp, sp, framesize - 2 * wordSize);
5972     else {
5973       mov(rscratch1, framesize - 2 * wordSize);
5974       sub(sp, sp, rscratch1);
5975     }
5976   }
5977   verify_cross_modify_fence_not_required();
5978 }
5979 
5980 void MacroAssembler::remove_frame(int framesize) {
5981   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5982   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5983   if (framesize < ((1 << 9) + 2 * wordSize)) {
5984     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5985     add(sp, sp, framesize);
5986   } else {
5987     if (framesize < ((1 << 12) + 2 * wordSize))
5988       add(sp, sp, framesize - 2 * wordSize);
5989     else {
5990       mov(rscratch1, framesize - 2 * wordSize);
5991       add(sp, sp, rscratch1);
5992     }
5993     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5994   }
5995   authenticate_return_address();
5996 }
5997 
5998 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
5999   if (needs_stack_repair) {
6000     // The method has a scalarized entry point (where fields of value object arguments
6001     // are passed through registers and stack), and a non-scalarized entry point (where
6002     // value object arguments are given as oops). The non-scalarized entry point will
6003     // first load each field of value object arguments and store them in registers and on
6004     // the stack in a way compatible with the scalarized entry point. To do so, some extra
6005     // stack space might be reserved (if argument registers are not enough). On leaving the
6006     // method, this space must be freed.
6007     //
6008     // In case we used the non-scalarized entry point the stack looks like this:
6009     //
6010     // | Arguments from caller     |
6011     // |---------------------------|  <-- caller's SP
6012     // | Saved LR #1               |
6013     // | Saved FP #1               |
6014     // |---------------------------|
6015     // | Extension space for       |
6016     // |   inline arg (un)packing  |
6017     // |---------------------------|  <-- start of this method's frame
6018     // | Saved LR #2               |
6019     // | Saved FP #2               |
6020     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6021     // | sp_inc                    |
6022     // | method locals             |
6023     // |---------------------------|  <-- SP
6024     //
6025     // There are two copies of FP and LR on the stack. They will be identical at
6026     // first, but that can change.
6027     // If the caller has been deoptimized, LR #1 will be patched to point at the
6028     // deopt blob, and LR #2 will still point into the old method.
6029     // If the saved FP (x29) was not used as the frame pointer, but to store an
6030     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
6031     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
6032     // needed to add space between the extension space and the locals, as there
6033     // would be between the real arguments and the locals if we don't need to
6034     // do unpacking (from the scalarized entry point).
6035     //
6036     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
6037     // while keeping in mind that from the scalarized entry point, there will be
6038     // only one copy of each. Indeed, in the case we used the scalarized calling
6039     // convention, the stack looks like this:
6040     //
6041     // | Arguments from caller     |
6042     // |---------------------------|  <-- caller's SP / start of this method's frame
6043     // | Saved LR                  |
6044     // | Saved FP                  |
6045     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6046     // | sp_inc                    |
6047     // | method locals             |
6048     // |---------------------------|  <-- SP
6049     //
6050     // The sp_inc stack slot holds the total size of the frame including the
6051     // extension space minus two words for the saved FP and LR. That is how to
6052     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
6053     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
6054     //
6055     // One can find sp_inc since the start the method's frame is SP + initial_framesize.
6056 
6057     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6058 
6059     ldr(rscratch1, Address(sp, sp_inc_offset));
6060     add(sp, sp, rscratch1);
6061     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6062   } else {
6063     remove_frame(initial_framesize);
6064   }
6065 }
6066 
6067 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6068   int real_frame_size = frame_size + sp_inc;
6069   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6070   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6071   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6072 
6073   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6074 
6075   // Subtract two words for the saved FP and LR as these will be popped
6076   // separately. See remove_frame above.
6077   mov(rscratch1, real_frame_size - 2*wordSize);
6078   str(rscratch1, Address(sp, sp_inc_offset));
6079 }
6080 
6081 // This method counts leading positive bytes (highest bit not set) in provided byte array
6082 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6083     // Simple and most common case of aligned small array which is not at the
6084     // end of memory page is placed here. All other cases are in stub.
6085     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6086     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6087     assert_different_registers(ary1, len, result);
6088 
6089     mov(result, len);
6090     cmpw(len, 0);
6091     br(LE, DONE);
6092     cmpw(len, 4 * wordSize);
6093     br(GE, STUB_LONG); // size > 32 then go to stub
6094 
6095     int shift = 64 - exact_log2(os::vm_page_size());
6096     lsl(rscratch1, ary1, shift);
6097     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6098     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6099     br(CS, STUB); // at the end of page then go to stub
6100     subs(len, len, wordSize);
6101     br(LT, END);
6102 
6103   BIND(LOOP);
6104     ldr(rscratch1, Address(post(ary1, wordSize)));
6105     tst(rscratch1, UPPER_BIT_MASK);
6106     br(NE, SET_RESULT);
6107     subs(len, len, wordSize);
6108     br(GE, LOOP);
6109     cmpw(len, -wordSize);
6110     br(EQ, DONE);
6111 
6112   BIND(END);
6113     ldr(rscratch1, Address(ary1));
6114     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
6115     lslv(rscratch1, rscratch1, rscratch2);
6116     tst(rscratch1, UPPER_BIT_MASK);
6117     br(NE, SET_RESULT);
6118     b(DONE);
6119 
6120   BIND(STUB);
6121     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
6122     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
6123     address tpc1 = trampoline_call(count_pos);
6124     if (tpc1 == nullptr) {
6125       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
6126       postcond(pc() == badAddress);
6127       return nullptr;
6128     }
6129     b(DONE);
6130 
6131   BIND(STUB_LONG);
6132     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
6133     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
6134     address tpc2 = trampoline_call(count_pos_long);
6135     if (tpc2 == nullptr) {
6136       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
6137       postcond(pc() == badAddress);
6138       return nullptr;
6139     }
6140     b(DONE);
6141 
6142   BIND(SET_RESULT);
6143 
6144     add(len, len, wordSize);
6145     sub(result, result, len);
6146 
6147   BIND(DONE);
6148   postcond(pc() != badAddress);
6149   return pc();
6150 }
6151 
6152 // Clobbers: rscratch1, rscratch2, rflags
6153 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
6154 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
6155                                       Register tmp4, Register tmp5, Register result,
6156                                       Register cnt1, int elem_size) {
6157   Label DONE, SAME;
6158   Register tmp1 = rscratch1;
6159   Register tmp2 = rscratch2;
6160   int elem_per_word = wordSize/elem_size;
6161   int log_elem_size = exact_log2(elem_size);
6162   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
6163   int length_offset = arrayOopDesc::length_offset_in_bytes();
6164   int base_offset
6165     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
6166   // When the length offset is not aligned to 8 bytes,
6167   // then we align it down. This is valid because the new
6168   // offset will always be the klass which is the same
6169   // for type arrays.
6170   int start_offset = align_down(length_offset, BytesPerWord);
6171   int extra_length = base_offset - start_offset;
6172   assert(start_offset == length_offset || start_offset == klass_offset,
6173          "start offset must be 8-byte-aligned or be the klass offset");
6174   assert(base_offset != start_offset, "must include the length field");
6175   extra_length = extra_length / elem_size; // We count in elements, not bytes.
6176   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
6177 
6178   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
6179   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6180 
6181 #ifndef PRODUCT
6182   {
6183     const char kind = (elem_size == 2) ? 'U' : 'L';
6184     char comment[64];
6185     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
6186     BLOCK_COMMENT(comment);
6187   }
6188 #endif
6189 
6190   // if (a1 == a2)
6191   //     return true;
6192   cmpoop(a1, a2); // May have read barriers for a1 and a2.
6193   br(EQ, SAME);
6194 
6195   if (UseSimpleArrayEquals) {
6196     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
6197     // if (a1 == nullptr || a2 == nullptr)
6198     //     return false;
6199     // a1 & a2 == 0 means (some-pointer is null) or
6200     // (very-rare-or-even-probably-impossible-pointer-values)
6201     // so, we can save one branch in most cases
6202     tst(a1, a2);
6203     mov(result, false);
6204     br(EQ, A_MIGHT_BE_NULL);
6205     // if (a1.length != a2.length)
6206     //      return false;
6207     bind(A_IS_NOT_NULL);
6208     ldrw(cnt1, Address(a1, length_offset));
6209     ldrw(tmp5, Address(a2, length_offset));
6210     cmp(cnt1, tmp5);
6211     br(NE, DONE); // If lengths differ, return false
6212     // Increase loop counter by diff between base- and actual start-offset.
6213     addw(cnt1, cnt1, extra_length);
6214     lea(a1, Address(a1, start_offset));
6215     lea(a2, Address(a2, start_offset));
6216     // Check for short strings, i.e. smaller than wordSize.
6217     subs(cnt1, cnt1, elem_per_word);
6218     br(Assembler::LT, SHORT);
6219     // Main 8 byte comparison loop.
6220     bind(NEXT_WORD); {
6221       ldr(tmp1, Address(post(a1, wordSize)));
6222       ldr(tmp2, Address(post(a2, wordSize)));
6223       subs(cnt1, cnt1, elem_per_word);
6224       eor(tmp5, tmp1, tmp2);
6225       cbnz(tmp5, DONE);
6226     } br(GT, NEXT_WORD);
6227     // Last longword.  In the case where length == 4 we compare the
6228     // same longword twice, but that's still faster than another
6229     // conditional branch.
6230     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6231     // length == 4.
6232     if (log_elem_size > 0)
6233       lsl(cnt1, cnt1, log_elem_size);
6234     ldr(tmp3, Address(a1, cnt1));
6235     ldr(tmp4, Address(a2, cnt1));
6236     eor(tmp5, tmp3, tmp4);
6237     cbnz(tmp5, DONE);
6238     b(SAME);
6239     bind(A_MIGHT_BE_NULL);
6240     // in case both a1 and a2 are not-null, proceed with loads
6241     cbz(a1, DONE);
6242     cbz(a2, DONE);
6243     b(A_IS_NOT_NULL);
6244     bind(SHORT);
6245 
6246     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
6247     {
6248       ldrw(tmp1, Address(post(a1, 4)));
6249       ldrw(tmp2, Address(post(a2, 4)));
6250       eorw(tmp5, tmp1, tmp2);
6251       cbnzw(tmp5, DONE);
6252     }
6253     bind(TAIL03);
6254     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6255     {
6256       ldrh(tmp3, Address(post(a1, 2)));
6257       ldrh(tmp4, Address(post(a2, 2)));
6258       eorw(tmp5, tmp3, tmp4);
6259       cbnzw(tmp5, DONE);
6260     }
6261     bind(TAIL01);
6262     if (elem_size == 1) { // Only needed when comparing byte arrays.
6263       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6264       {
6265         ldrb(tmp1, a1);
6266         ldrb(tmp2, a2);
6267         eorw(tmp5, tmp1, tmp2);
6268         cbnzw(tmp5, DONE);
6269       }
6270     }
6271   } else {
6272     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6273         CSET_EQ, LAST_CHECK;
6274     mov(result, false);
6275     cbz(a1, DONE);
6276     ldrw(cnt1, Address(a1, length_offset));
6277     cbz(a2, DONE);
6278     ldrw(tmp5, Address(a2, length_offset));
6279     cmp(cnt1, tmp5);
6280     br(NE, DONE); // If lengths differ, return false
6281     // Increase loop counter by diff between base- and actual start-offset.
6282     addw(cnt1, cnt1, extra_length);
6283 
6284     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6285     // faster to perform another branch before comparing a1 and a2
6286     cmp(cnt1, (u1)elem_per_word);
6287     br(LE, SHORT); // short or same
6288     ldr(tmp3, Address(pre(a1, start_offset)));
6289     subs(zr, cnt1, stubBytesThreshold);
6290     br(GE, STUB);
6291     ldr(tmp4, Address(pre(a2, start_offset)));
6292     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6293 
6294     // Main 16 byte comparison loop with 2 exits
6295     bind(NEXT_DWORD); {
6296       ldr(tmp1, Address(pre(a1, wordSize)));
6297       ldr(tmp2, Address(pre(a2, wordSize)));
6298       subs(cnt1, cnt1, 2 * elem_per_word);
6299       br(LE, TAIL);
6300       eor(tmp4, tmp3, tmp4);
6301       cbnz(tmp4, DONE);
6302       ldr(tmp3, Address(pre(a1, wordSize)));
6303       ldr(tmp4, Address(pre(a2, wordSize)));
6304       cmp(cnt1, (u1)elem_per_word);
6305       br(LE, TAIL2);
6306       cmp(tmp1, tmp2);
6307     } br(EQ, NEXT_DWORD);
6308     b(DONE);
6309 
6310     bind(TAIL);
6311     eor(tmp4, tmp3, tmp4);
6312     eor(tmp2, tmp1, tmp2);
6313     lslv(tmp2, tmp2, tmp5);
6314     orr(tmp5, tmp4, tmp2);
6315     cmp(tmp5, zr);
6316     b(CSET_EQ);
6317 
6318     bind(TAIL2);
6319     eor(tmp2, tmp1, tmp2);
6320     cbnz(tmp2, DONE);
6321     b(LAST_CHECK);
6322 
6323     bind(STUB);
6324     ldr(tmp4, Address(pre(a2, start_offset)));
6325     if (elem_size == 2) { // convert to byte counter
6326       lsl(cnt1, cnt1, 1);
6327     }
6328     eor(tmp5, tmp3, tmp4);
6329     cbnz(tmp5, DONE);
6330     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6331     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6332     address tpc = trampoline_call(stub);
6333     if (tpc == nullptr) {
6334       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6335       postcond(pc() == badAddress);
6336       return nullptr;
6337     }
6338     b(DONE);
6339 
6340     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6341     // so, if a2 == null => return false(0), else return true, so we can return a2
6342     mov(result, a2);
6343     b(DONE);
6344     bind(SHORT);
6345     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6346     ldr(tmp3, Address(a1, start_offset));
6347     ldr(tmp4, Address(a2, start_offset));
6348     bind(LAST_CHECK);
6349     eor(tmp4, tmp3, tmp4);
6350     lslv(tmp5, tmp4, tmp5);
6351     cmp(tmp5, zr);
6352     bind(CSET_EQ);
6353     cset(result, EQ);
6354     b(DONE);
6355   }
6356 
6357   bind(SAME);
6358   mov(result, true);
6359   // That's it.
6360   bind(DONE);
6361 
6362   BLOCK_COMMENT("} array_equals");
6363   postcond(pc() != badAddress);
6364   return pc();
6365 }
6366 
6367 // Compare Strings
6368 
6369 // For Strings we're passed the address of the first characters in a1
6370 // and a2 and the length in cnt1.
6371 // There are two implementations.  For arrays >= 8 bytes, all
6372 // comparisons (including the final one, which may overlap) are
6373 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6374 // halfword, then a short, and then a byte.
6375 
6376 void MacroAssembler::string_equals(Register a1, Register a2,
6377                                    Register result, Register cnt1)
6378 {
6379   Label SAME, DONE, SHORT, NEXT_WORD;
6380   Register tmp1 = rscratch1;
6381   Register tmp2 = rscratch2;
6382 
6383   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6384 
6385 #ifndef PRODUCT
6386   {
6387     char comment[64];
6388     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6389     BLOCK_COMMENT(comment);
6390   }
6391 #endif
6392 
6393   mov(result, false);
6394 
6395   // Check for short strings, i.e. smaller than wordSize.
6396   subs(cnt1, cnt1, wordSize);
6397   br(Assembler::LT, SHORT);
6398   // Main 8 byte comparison loop.
6399   bind(NEXT_WORD); {
6400     ldr(tmp1, Address(post(a1, wordSize)));
6401     ldr(tmp2, Address(post(a2, wordSize)));
6402     subs(cnt1, cnt1, wordSize);
6403     eor(tmp1, tmp1, tmp2);
6404     cbnz(tmp1, DONE);
6405   } br(GT, NEXT_WORD);
6406   // Last longword.  In the case where length == 4 we compare the
6407   // same longword twice, but that's still faster than another
6408   // conditional branch.
6409   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6410   // length == 4.
6411   ldr(tmp1, Address(a1, cnt1));
6412   ldr(tmp2, Address(a2, cnt1));
6413   eor(tmp2, tmp1, tmp2);
6414   cbnz(tmp2, DONE);
6415   b(SAME);
6416 
6417   bind(SHORT);
6418   Label TAIL03, TAIL01;
6419 
6420   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6421   {
6422     ldrw(tmp1, Address(post(a1, 4)));
6423     ldrw(tmp2, Address(post(a2, 4)));
6424     eorw(tmp1, tmp1, tmp2);
6425     cbnzw(tmp1, DONE);
6426   }
6427   bind(TAIL03);
6428   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6429   {
6430     ldrh(tmp1, Address(post(a1, 2)));
6431     ldrh(tmp2, Address(post(a2, 2)));
6432     eorw(tmp1, tmp1, tmp2);
6433     cbnzw(tmp1, DONE);
6434   }
6435   bind(TAIL01);
6436   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6437     {
6438     ldrb(tmp1, a1);
6439     ldrb(tmp2, a2);
6440     eorw(tmp1, tmp1, tmp2);
6441     cbnzw(tmp1, DONE);
6442   }
6443   // Arrays are equal.
6444   bind(SAME);
6445   mov(result, true);
6446 
6447   // That's it.
6448   bind(DONE);
6449   BLOCK_COMMENT("} string_equals");
6450 }
6451 
6452 
6453 // The size of the blocks erased by the zero_blocks stub.  We must
6454 // handle anything smaller than this ourselves in zero_words().
6455 const int MacroAssembler::zero_words_block_size = 8;
6456 
6457 // zero_words() is used by C2 ClearArray patterns and by
6458 // C1_MacroAssembler.  It is as small as possible, handling small word
6459 // counts locally and delegating anything larger to the zero_blocks
6460 // stub.  It is expanded many times in compiled code, so it is
6461 // important to keep it short.
6462 
6463 // ptr:   Address of a buffer to be zeroed.
6464 // cnt:   Count in HeapWords.
6465 //
6466 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6467 address MacroAssembler::zero_words(Register ptr, Register cnt)
6468 {
6469   assert(is_power_of_2(zero_words_block_size), "adjust this");
6470 
6471   BLOCK_COMMENT("zero_words {");
6472   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6473   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6474   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6475 
6476   subs(rscratch1, cnt, zero_words_block_size);
6477   Label around;
6478   br(LO, around);
6479   {
6480     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6481     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6482     // Make sure this is a C2 compilation. C1 allocates space only for
6483     // trampoline stubs generated by Call LIR ops, and in any case it
6484     // makes sense for a C1 compilation task to proceed as quickly as
6485     // possible.
6486     CompileTask* task;
6487     if (StubRoutines::aarch64::complete()
6488         && Thread::current()->is_Compiler_thread()
6489         && (task = ciEnv::current()->task())
6490         && is_c2_compile(task->comp_level())) {
6491       address tpc = trampoline_call(zero_blocks);
6492       if (tpc == nullptr) {
6493         DEBUG_ONLY(reset_labels(around));
6494         return nullptr;
6495       }
6496     } else {
6497       far_call(zero_blocks);
6498     }
6499   }
6500   bind(around);
6501 
6502   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6503   // for us.
6504   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6505     Label l;
6506     tbz(cnt, exact_log2(i), l);
6507     for (int j = 0; j < i; j += 2) {
6508       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6509     }
6510     bind(l);
6511   }
6512   {
6513     Label l;
6514     tbz(cnt, 0, l);
6515     str(zr, Address(ptr));
6516     bind(l);
6517   }
6518 
6519   BLOCK_COMMENT("} zero_words");
6520   return pc();
6521 }
6522 
6523 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6524 // cnt:          Immediate count in HeapWords.
6525 //
6526 // r10, r11, rscratch1, and rscratch2 are clobbered.
6527 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6528 {
6529   assert(wordSize <= BlockZeroingLowLimit,
6530             "increase BlockZeroingLowLimit");
6531   address result = nullptr;
6532   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6533 #ifndef PRODUCT
6534     {
6535       char buf[64];
6536       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6537       BLOCK_COMMENT(buf);
6538     }
6539 #endif
6540     if (cnt >= 16) {
6541       uint64_t loops = cnt/16;
6542       if (loops > 1) {
6543         mov(rscratch2, loops - 1);
6544       }
6545       {
6546         Label loop;
6547         bind(loop);
6548         for (int i = 0; i < 16; i += 2) {
6549           stp(zr, zr, Address(base, i * BytesPerWord));
6550         }
6551         add(base, base, 16 * BytesPerWord);
6552         if (loops > 1) {
6553           subs(rscratch2, rscratch2, 1);
6554           br(GE, loop);
6555         }
6556       }
6557     }
6558     cnt %= 16;
6559     int i = cnt & 1;  // store any odd word to start
6560     if (i) str(zr, Address(base));
6561     for (; i < (int)cnt; i += 2) {
6562       stp(zr, zr, Address(base, i * wordSize));
6563     }
6564     BLOCK_COMMENT("} zero_words");
6565     result = pc();
6566   } else {
6567     mov(r10, base); mov(r11, cnt);
6568     result = zero_words(r10, r11);
6569   }
6570   return result;
6571 }
6572 
6573 // Zero blocks of memory by using DC ZVA.
6574 //
6575 // Aligns the base address first sufficiently for DC ZVA, then uses
6576 // DC ZVA repeatedly for every full block.  cnt is the size to be
6577 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6578 // in cnt.
6579 //
6580 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6581 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6582 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6583   Register tmp = rscratch1;
6584   Register tmp2 = rscratch2;
6585   int zva_length = VM_Version::zva_length();
6586   Label initial_table_end, loop_zva;
6587   Label fini;
6588 
6589   // Base must be 16 byte aligned. If not just return and let caller handle it
6590   tst(base, 0x0f);
6591   br(Assembler::NE, fini);
6592   // Align base with ZVA length.
6593   neg(tmp, base);
6594   andr(tmp, tmp, zva_length - 1);
6595 
6596   // tmp: the number of bytes to be filled to align the base with ZVA length.
6597   add(base, base, tmp);
6598   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6599   adr(tmp2, initial_table_end);
6600   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6601   br(tmp2);
6602 
6603   for (int i = -zva_length + 16; i < 0; i += 16)
6604     stp(zr, zr, Address(base, i));
6605   bind(initial_table_end);
6606 
6607   sub(cnt, cnt, zva_length >> 3);
6608   bind(loop_zva);
6609   dc(Assembler::ZVA, base);
6610   subs(cnt, cnt, zva_length >> 3);
6611   add(base, base, zva_length);
6612   br(Assembler::GE, loop_zva);
6613   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6614   bind(fini);
6615 }
6616 
6617 // base:   Address of a buffer to be filled, 8 bytes aligned.
6618 // cnt:    Count in 8-byte unit.
6619 // value:  Value to be filled with.
6620 // base will point to the end of the buffer after filling.
6621 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6622 {
6623 //  Algorithm:
6624 //
6625 //    if (cnt == 0) {
6626 //      return;
6627 //    }
6628 //    if ((p & 8) != 0) {
6629 //      *p++ = v;
6630 //    }
6631 //
6632 //    scratch1 = cnt & 14;
6633 //    cnt -= scratch1;
6634 //    p += scratch1;
6635 //    switch (scratch1 / 2) {
6636 //      do {
6637 //        cnt -= 16;
6638 //          p[-16] = v;
6639 //          p[-15] = v;
6640 //        case 7:
6641 //          p[-14] = v;
6642 //          p[-13] = v;
6643 //        case 6:
6644 //          p[-12] = v;
6645 //          p[-11] = v;
6646 //          // ...
6647 //        case 1:
6648 //          p[-2] = v;
6649 //          p[-1] = v;
6650 //        case 0:
6651 //          p += 16;
6652 //      } while (cnt);
6653 //    }
6654 //    if ((cnt & 1) == 1) {
6655 //      *p++ = v;
6656 //    }
6657 
6658   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6659 
6660   Label fini, skip, entry, loop;
6661   const int unroll = 8; // Number of stp instructions we'll unroll
6662 
6663   cbz(cnt, fini);
6664   tbz(base, 3, skip);
6665   str(value, Address(post(base, 8)));
6666   sub(cnt, cnt, 1);
6667   bind(skip);
6668 
6669   andr(rscratch1, cnt, (unroll-1) * 2);
6670   sub(cnt, cnt, rscratch1);
6671   add(base, base, rscratch1, Assembler::LSL, 3);
6672   adr(rscratch2, entry);
6673   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6674   br(rscratch2);
6675 
6676   bind(loop);
6677   add(base, base, unroll * 16);
6678   for (int i = -unroll; i < 0; i++)
6679     stp(value, value, Address(base, i * 16));
6680   bind(entry);
6681   subs(cnt, cnt, unroll * 2);
6682   br(Assembler::GE, loop);
6683 
6684   tbz(cnt, 0, fini);
6685   str(value, Address(post(base, 8)));
6686   bind(fini);
6687 }
6688 
6689 // Intrinsic for
6690 //
6691 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6692 //   Encodes char[] to byte[] in ISO-8859-1
6693 //
6694 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6695 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6696 //
6697 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6698 //   Encodes char[] to byte[] in ASCII
6699 //
6700 // This version always returns the number of characters copied, and does not
6701 // clobber the 'len' register. A successful copy will complete with the post-
6702 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6703 // post-condition: 0 <= 'res' < 'len'.
6704 //
6705 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6706 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6707 //       beyond the acceptable, even though the footprint would be smaller.
6708 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6709 //       avoid additional bloat.
6710 //
6711 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6712 void MacroAssembler::encode_iso_array(Register src, Register dst,
6713                                       Register len, Register res, bool ascii,
6714                                       FloatRegister vtmp0, FloatRegister vtmp1,
6715                                       FloatRegister vtmp2, FloatRegister vtmp3,
6716                                       FloatRegister vtmp4, FloatRegister vtmp5)
6717 {
6718   Register cnt = res;
6719   Register max = rscratch1;
6720   Register chk = rscratch2;
6721 
6722   prfm(Address(src), PLDL1STRM);
6723   movw(cnt, len);
6724 
6725 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6726 
6727   Label LOOP_32, DONE_32, FAIL_32;
6728 
6729   BIND(LOOP_32);
6730   {
6731     cmpw(cnt, 32);
6732     br(LT, DONE_32);
6733     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6734     // Extract lower bytes.
6735     FloatRegister vlo0 = vtmp4;
6736     FloatRegister vlo1 = vtmp5;
6737     uzp1(vlo0, T16B, vtmp0, vtmp1);
6738     uzp1(vlo1, T16B, vtmp2, vtmp3);
6739     // Merge bits...
6740     orr(vtmp0, T16B, vtmp0, vtmp1);
6741     orr(vtmp2, T16B, vtmp2, vtmp3);
6742     // Extract merged upper bytes.
6743     FloatRegister vhix = vtmp0;
6744     uzp2(vhix, T16B, vtmp0, vtmp2);
6745     // ISO-check on hi-parts (all zero).
6746     //                          ASCII-check on lo-parts (no sign).
6747     FloatRegister vlox = vtmp1; // Merge lower bytes.
6748                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6749     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6750     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6751     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6752                                 ASCII(orr(chk, chk, max));
6753     cbnz(chk, FAIL_32);
6754     subw(cnt, cnt, 32);
6755     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6756     b(LOOP_32);
6757   }
6758   BIND(FAIL_32);
6759   sub(src, src, 64);
6760   BIND(DONE_32);
6761 
6762   Label LOOP_8, SKIP_8;
6763 
6764   BIND(LOOP_8);
6765   {
6766     cmpw(cnt, 8);
6767     br(LT, SKIP_8);
6768     FloatRegister vhi = vtmp0;
6769     FloatRegister vlo = vtmp1;
6770     ld1(vtmp3, T8H, src);
6771     uzp1(vlo, T16B, vtmp3, vtmp3);
6772     uzp2(vhi, T16B, vtmp3, vtmp3);
6773     // ISO-check on hi-parts (all zero).
6774     //                          ASCII-check on lo-parts (no sign).
6775                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6776     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6777                                 ASCII(umov(max, vtmp2, B, 0));
6778                                 ASCII(orr(chk, chk, max));
6779     cbnz(chk, SKIP_8);
6780 
6781     strd(vlo, Address(post(dst, 8)));
6782     subw(cnt, cnt, 8);
6783     add(src, src, 16);
6784     b(LOOP_8);
6785   }
6786   BIND(SKIP_8);
6787 
6788 #undef ASCII
6789 
6790   Label LOOP, DONE;
6791 
6792   cbz(cnt, DONE);
6793   BIND(LOOP);
6794   {
6795     Register chr = rscratch1;
6796     ldrh(chr, Address(post(src, 2)));
6797     tst(chr, ascii ? 0xff80 : 0xff00);
6798     br(NE, DONE);
6799     strb(chr, Address(post(dst, 1)));
6800     subs(cnt, cnt, 1);
6801     br(GT, LOOP);
6802   }
6803   BIND(DONE);
6804   // Return index where we stopped.
6805   subw(res, len, cnt);
6806 }
6807 
6808 // Inflate byte[] array to char[].
6809 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6810 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6811                                            FloatRegister vtmp1, FloatRegister vtmp2,
6812                                            FloatRegister vtmp3, Register tmp4) {
6813   Label big, done, after_init, to_stub;
6814 
6815   assert_different_registers(src, dst, len, tmp4, rscratch1);
6816 
6817   fmovd(vtmp1, 0.0);
6818   lsrw(tmp4, len, 3);
6819   bind(after_init);
6820   cbnzw(tmp4, big);
6821   // Short string: less than 8 bytes.
6822   {
6823     Label loop, tiny;
6824 
6825     cmpw(len, 4);
6826     br(LT, tiny);
6827     // Use SIMD to do 4 bytes.
6828     ldrs(vtmp2, post(src, 4));
6829     zip1(vtmp3, T8B, vtmp2, vtmp1);
6830     subw(len, len, 4);
6831     strd(vtmp3, post(dst, 8));
6832 
6833     cbzw(len, done);
6834 
6835     // Do the remaining bytes by steam.
6836     bind(loop);
6837     ldrb(tmp4, post(src, 1));
6838     strh(tmp4, post(dst, 2));
6839     subw(len, len, 1);
6840 
6841     bind(tiny);
6842     cbnz(len, loop);
6843 
6844     b(done);
6845   }
6846 
6847   if (SoftwarePrefetchHintDistance >= 0) {
6848     bind(to_stub);
6849       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6850       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6851       address tpc = trampoline_call(stub);
6852       if (tpc == nullptr) {
6853         DEBUG_ONLY(reset_labels(big, done));
6854         postcond(pc() == badAddress);
6855         return nullptr;
6856       }
6857       b(after_init);
6858   }
6859 
6860   // Unpack the bytes 8 at a time.
6861   bind(big);
6862   {
6863     Label loop, around, loop_last, loop_start;
6864 
6865     if (SoftwarePrefetchHintDistance >= 0) {
6866       const int large_loop_threshold = (64 + 16)/8;
6867       ldrd(vtmp2, post(src, 8));
6868       andw(len, len, 7);
6869       cmp(tmp4, (u1)large_loop_threshold);
6870       br(GE, to_stub);
6871       b(loop_start);
6872 
6873       bind(loop);
6874       ldrd(vtmp2, post(src, 8));
6875       bind(loop_start);
6876       subs(tmp4, tmp4, 1);
6877       br(EQ, loop_last);
6878       zip1(vtmp2, T16B, vtmp2, vtmp1);
6879       ldrd(vtmp3, post(src, 8));
6880       st1(vtmp2, T8H, post(dst, 16));
6881       subs(tmp4, tmp4, 1);
6882       zip1(vtmp3, T16B, vtmp3, vtmp1);
6883       st1(vtmp3, T8H, post(dst, 16));
6884       br(NE, loop);
6885       b(around);
6886       bind(loop_last);
6887       zip1(vtmp2, T16B, vtmp2, vtmp1);
6888       st1(vtmp2, T8H, post(dst, 16));
6889       bind(around);
6890       cbz(len, done);
6891     } else {
6892       andw(len, len, 7);
6893       bind(loop);
6894       ldrd(vtmp2, post(src, 8));
6895       sub(tmp4, tmp4, 1);
6896       zip1(vtmp3, T16B, vtmp2, vtmp1);
6897       st1(vtmp3, T8H, post(dst, 16));
6898       cbnz(tmp4, loop);
6899     }
6900   }
6901 
6902   // Do the tail of up to 8 bytes.
6903   add(src, src, len);
6904   ldrd(vtmp3, Address(src, -8));
6905   add(dst, dst, len, ext::uxtw, 1);
6906   zip1(vtmp3, T16B, vtmp3, vtmp1);
6907   strq(vtmp3, Address(dst, -16));
6908 
6909   bind(done);
6910   postcond(pc() != badAddress);
6911   return pc();
6912 }
6913 
6914 // Compress char[] array to byte[].
6915 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6916 // Return the array length if every element in array can be encoded,
6917 // otherwise, the index of first non-latin1 (> 0xff) character.
6918 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6919                                          Register res,
6920                                          FloatRegister tmp0, FloatRegister tmp1,
6921                                          FloatRegister tmp2, FloatRegister tmp3,
6922                                          FloatRegister tmp4, FloatRegister tmp5) {
6923   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6924 }
6925 
6926 // java.math.round(double a)
6927 // Returns the closest long to the argument, with ties rounding to
6928 // positive infinity.  This requires some fiddling for corner
6929 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6930 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6931                                        FloatRegister ftmp) {
6932   Label DONE;
6933   BLOCK_COMMENT("java_round_double: { ");
6934   fmovd(rscratch1, src);
6935   // Use RoundToNearestTiesAway unless src small and -ve.
6936   fcvtasd(dst, src);
6937   // Test if src >= 0 || abs(src) >= 0x1.0p52
6938   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6939   mov(rscratch2, julong_cast(0x1.0p52));
6940   cmp(rscratch1, rscratch2);
6941   br(HS, DONE); {
6942     // src < 0 && abs(src) < 0x1.0p52
6943     // src may have a fractional part, so add 0.5
6944     fmovd(ftmp, 0.5);
6945     faddd(ftmp, src, ftmp);
6946     // Convert double to jlong, use RoundTowardsNegative
6947     fcvtmsd(dst, ftmp);
6948   }
6949   bind(DONE);
6950   BLOCK_COMMENT("} java_round_double");
6951 }
6952 
6953 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6954                                       FloatRegister ftmp) {
6955   Label DONE;
6956   BLOCK_COMMENT("java_round_float: { ");
6957   fmovs(rscratch1, src);
6958   // Use RoundToNearestTiesAway unless src small and -ve.
6959   fcvtassw(dst, src);
6960   // Test if src >= 0 || abs(src) >= 0x1.0p23
6961   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6962   mov(rscratch2, jint_cast(0x1.0p23f));
6963   cmp(rscratch1, rscratch2);
6964   br(HS, DONE); {
6965     // src < 0 && |src| < 0x1.0p23
6966     // src may have a fractional part, so add 0.5
6967     fmovs(ftmp, 0.5f);
6968     fadds(ftmp, src, ftmp);
6969     // Convert float to jint, use RoundTowardsNegative
6970     fcvtmssw(dst, ftmp);
6971   }
6972   bind(DONE);
6973   BLOCK_COMMENT("} java_round_float");
6974 }
6975 
6976 // get_thread() can be called anywhere inside generated code so we
6977 // need to save whatever non-callee save context might get clobbered
6978 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6979 // the call setup code.
6980 //
6981 // On Linux and Windows, aarch64_get_thread_helper() is implemented in
6982 // assembly and clobbers only r0, r1, and flags.
6983 // On other systems, the helper is a usual C function.
6984 //
6985 void MacroAssembler::get_thread(Register dst) {
6986   RegSet saved_regs =
6987     BSD_ONLY(RegSet::range(r0, r17) + lr - dst)
6988     NOT_BSD (RegSet::range(r0, r1)  + lr - dst);
6989 
6990   protect_return_address();
6991   push(saved_regs, sp);
6992 
6993   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6994   blr(lr);
6995   if (dst != c_rarg0) {
6996     mov(dst, c_rarg0);
6997   }
6998 
6999   pop(saved_regs, sp);
7000   authenticate_return_address();
7001 }
7002 
7003 #ifdef COMPILER2
7004 // C2 compiled method's prolog code
7005 // Moved here from aarch64.ad to support Valhalla code below
7006 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
7007   if (C->clinit_barrier_on_entry()) {
7008     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
7009 
7010     Label L_skip_barrier;
7011 
7012     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7013     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7014     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7015     bind(L_skip_barrier);
7016   }
7017 
7018   if (C->max_vector_size() > 0) {
7019     reinitialize_ptrue();
7020   }
7021 
7022   int bangsize = C->output()->bang_size_in_bytes();
7023   if (C->output()->need_stack_bang(bangsize))
7024     generate_stack_overflow_check(bangsize);
7025 
7026   // n.b. frame size includes space for return pc and rfp
7027   const long framesize = C->output()->frame_size_in_bytes();
7028   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
7029 
7030   if (C->needs_stack_repair()) {
7031     save_stack_increment(sp_inc, framesize);
7032   }
7033 
7034   if (VerifyStackAtCalls) {
7035     Unimplemented();
7036   }
7037 }
7038 #endif // COMPILER2
7039 
7040 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7041   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7042   // An inline type might be returned. If fields are in registers we
7043   // need to allocate an inline type instance and initialize it with
7044   // the value of the fields.
7045   Label skip;
7046   // We only need a new buffered inline type if a new one is not returned
7047   tbz(r0, 0, skip);
7048   int call_offset = -1;
7049 
7050   // Be careful not to clobber r1-7 which hold returned fields
7051   // Also do not use callee-saved registers as these may be live in the interpreter
7052   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7053 
7054   // The following code is similar to the instance allocation code in TemplateTable::_new
7055   //  but has some slight differences,
7056   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7057   // allocating is not necessary if vk != nullptr, etc.
7058   Label slow_case;
7059   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7060   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7061 
7062   if (vk != nullptr) {
7063     // Called from C1, where the return type is statically known.
7064     movptr(klass, (intptr_t)vk->get_InlineKlass());
7065     jint lh = vk->layout_helper();
7066     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7067     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7068       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7069     } else {
7070       b(slow_case);
7071     }
7072   } else {
7073     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7074     andr(klass, r0, -2);
7075     if (UseTLAB) {
7076       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7077       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7078       br(Assembler::NE, slow_case);
7079       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7080     } else {
7081       b(slow_case);
7082     }
7083   }
7084   if (UseTLAB) {
7085     // 2. Initialize buffered inline instance header
7086     Register buffer_obj = r0;
7087     if (UseCompactObjectHeaders) {
7088       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7089       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7090     } else {
7091       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7092       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7093       store_klass_gap(buffer_obj, zr);
7094       if (vk == nullptr) {
7095         // store_klass corrupts klass, so save it for later use (interpreter case only).
7096         mov(tmp1, klass);
7097       }
7098       store_klass(buffer_obj, klass, rscratch1);
7099       klass = tmp1;
7100     }
7101     // 3. Initialize its fields with an inline class specific handler
7102     if (vk != nullptr) {
7103       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7104     } else {
7105       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7106       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7107       blr(tmp1);
7108     }
7109 
7110     membar(Assembler::StoreStore);
7111     b(skip);
7112   } else {
7113     // Must have already branched to slow_case above.
7114     DEBUG_ONLY(should_not_reach_here());
7115   }
7116   bind(slow_case);
7117   // We failed to allocate a new inline type, fall back to a runtime
7118   // call. Some oop field may be live in some registers but we can't
7119   // tell. That runtime call will take care of preserving them
7120   // across a GC if there's one.
7121   mov(r0, r0_preserved);
7122 
7123   if (from_interpreter) {
7124     super_call_VM_leaf(SharedRuntime::store_inline_type_fields_to_buf_entry());
7125   } else {
7126     far_call(RuntimeAddress(SharedRuntime::store_inline_type_fields_to_buf_entry()));
7127     call_offset = offset();
7128   }
7129   membar(Assembler::StoreStore);
7130 
7131   bind(skip);
7132   return call_offset;
7133 }
7134 
7135 // Move a value between registers/stack slots and update the reg_state
7136 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7137   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7138   if (reg_state[to->value()] == reg_written) {
7139     return true; // Already written
7140   }
7141 
7142   if (from != to && bt != T_VOID) {
7143     if (reg_state[to->value()] == reg_readonly) {
7144       return false; // Not yet writable
7145     }
7146     if (from->is_reg()) {
7147       if (to->is_reg()) {
7148         if (from->is_Register() && to->is_Register()) {
7149           mov(to->as_Register(), from->as_Register());
7150         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7151           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7152         } else {
7153           ShouldNotReachHere();
7154         }
7155       } else {
7156         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7157         Address to_addr = Address(sp, st_off);
7158         if (from->is_FloatRegister()) {
7159           if (bt == T_DOUBLE) {
7160              strd(from->as_FloatRegister(), to_addr);
7161           } else {
7162              assert(bt == T_FLOAT, "must be float");
7163              strs(from->as_FloatRegister(), to_addr);
7164           }
7165         } else {
7166           str(from->as_Register(), to_addr);
7167         }
7168       }
7169     } else {
7170       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7171       if (to->is_reg()) {
7172         if (to->is_FloatRegister()) {
7173           if (bt == T_DOUBLE) {
7174             ldrd(to->as_FloatRegister(), from_addr);
7175           } else {
7176             assert(bt == T_FLOAT, "must be float");
7177             ldrs(to->as_FloatRegister(), from_addr);
7178           }
7179         } else {
7180           ldr(to->as_Register(), from_addr);
7181         }
7182       } else {
7183         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7184         ldr(rscratch1, from_addr);
7185         str(rscratch1, Address(sp, st_off));
7186       }
7187     }
7188   }
7189 
7190   // Update register states
7191   reg_state[from->value()] = reg_writable;
7192   reg_state[to->value()] = reg_written;
7193   return true;
7194 }
7195 
7196 // Calculate the extra stack space required for packing or unpacking inline
7197 // args and adjust the stack pointer
7198 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7199   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7200   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7201   assert(sp_inc > 0, "sanity");
7202 
7203   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7204   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7205 
7206   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7207   if (sp_inc < (1 << 9)) {
7208     sub(sp, sp, sp_inc);   // Fits in an immediate
7209   } else {
7210     mov(rscratch1, sp_inc);
7211     sub(sp, sp, rscratch1);
7212   }
7213 
7214   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7215 }
7216 
7217 // Read all fields from an inline type oop and store the values in registers/stack slots
7218 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7219                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7220                                           RegState reg_state[]) {
7221   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7222   assert(from->is_valid(), "source must be valid");
7223   bool progress = false;
7224 #ifdef ASSERT
7225   const int start_offset = offset();
7226 #endif
7227 
7228   Label L_null, L_notNull;
7229   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7230   Register tmp1 = r10;
7231   Register tmp2 = r11;
7232 
7233 #ifdef ASSERT
7234   RegSet clobbered_gp_regs = MacroAssembler::call_clobbered_gp_registers();
7235   assert(clobbered_gp_regs.contains(tmp1), "tmp1 must be saved explicitly if it's not a clobber");
7236   assert(clobbered_gp_regs.contains(tmp2), "tmp2 must be saved explicitly if it's not a clobber");
7237   assert(clobbered_gp_regs.contains(r14), "r14 must be saved explicitly if it's not a clobber");
7238 #endif
7239 
7240   Register fromReg = noreg;
7241   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
7242   bool done = true;
7243   bool mark_done = true;
7244   VMReg toReg;
7245   BasicType bt;
7246   // Check if argument requires a null check
7247   bool null_check = false;
7248   VMReg nullCheckReg;
7249   while (stream.next(nullCheckReg, bt)) {
7250     if (sig->at(stream.sig_index())._offset == -1) {
7251       null_check = true;
7252       break;
7253     }
7254   }
7255   stream.reset(sig_index, to_index);
7256   while (stream.next(toReg, bt)) {
7257     assert(toReg->is_valid(), "destination must be valid");
7258     int idx = (int)toReg->value();
7259     if (reg_state[idx] == reg_readonly) {
7260       if (idx != from->value()) {
7261         mark_done = false;
7262       }
7263       done = false;
7264       continue;
7265     } else if (reg_state[idx] == reg_written) {
7266       continue;
7267     }
7268     assert(reg_state[idx] == reg_writable, "must be writable");
7269     reg_state[idx] = reg_written;
7270     progress = true;
7271 
7272     if (fromReg == noreg) {
7273       if (from->is_reg()) {
7274         fromReg = from->as_Register();
7275       } else {
7276         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7277         ldr(tmp1, Address(sp, st_off));
7278         fromReg = tmp1;
7279       }
7280       if (null_check) {
7281         // Nullable inline type argument, emit null check
7282         cbz(fromReg, L_null);
7283       }
7284     }
7285     int off = sig->at(stream.sig_index())._offset;
7286     if (off == -1) {
7287       assert(null_check, "Missing null check at");
7288       if (toReg->is_stack()) {
7289         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7290         mov(tmp2, 1);
7291         str(tmp2, Address(sp, st_off));
7292       } else {
7293         mov(toReg->as_Register(), 1);
7294       }
7295       continue;
7296     }
7297     if (sig->at(stream.sig_index())._vt_oop) {
7298       if (toReg->is_stack()) {
7299         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7300         str(fromReg, Address(sp, st_off));
7301       } else {
7302         mov(toReg->as_Register(), fromReg);
7303       }
7304       continue;
7305     }
7306     assert(off > 0, "offset in object should be positive");
7307     Address fromAddr = Address(fromReg, off);
7308     if (!toReg->is_FloatRegister()) {
7309       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7310       if (is_reference_type(bt)) {
7311         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7312       } else {
7313         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7314         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7315       }
7316       if (toReg->is_stack()) {
7317         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7318         str(dst, Address(sp, st_off));
7319       }
7320     } else if (bt == T_DOUBLE) {
7321       ldrd(toReg->as_FloatRegister(), fromAddr);
7322     } else {
7323       assert(bt == T_FLOAT, "must be float");
7324       ldrs(toReg->as_FloatRegister(), fromAddr);
7325     }
7326   }
7327   if (progress && null_check) {
7328     if (done) {
7329       b(L_notNull);
7330       bind(L_null);
7331       // Set null marker to zero to signal that the argument is null.
7332       // Also set all fields to zero since the runtime requires a canonical
7333       // representation of a flat null.
7334       stream.reset(sig_index, to_index);
7335       while (stream.next(toReg, bt)) {
7336         if (toReg->is_stack()) {
7337           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7338           str(zr, Address(sp, st_off));
7339         } else if (toReg->is_FloatRegister()) {
7340           mov(toReg->as_FloatRegister(), T2S, 0);
7341         } else {
7342           mov(toReg->as_Register(), zr);
7343         }
7344       }
7345       bind(L_notNull);
7346     } else {
7347       bind(L_null);
7348     }
7349   }
7350 
7351   sig_index = stream.sig_index();
7352   to_index = stream.regs_index();
7353 
7354   if (mark_done && reg_state[from->value()] != reg_written) {
7355     // This is okay because no one else will write to that slot
7356     reg_state[from->value()] = reg_writable;
7357   }
7358   from_index--;
7359   assert(progress || (start_offset == offset()), "should not emit code");
7360   return done;
7361 }
7362 
7363 // Pack fields back into an inline type oop
7364 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7365                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7366                                         RegState reg_state[], Register val_array) {
7367   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7368   assert(to->is_valid(), "destination must be valid");
7369 
7370   if (reg_state[to->value()] == reg_written) {
7371     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7372     return true; // Already written
7373   }
7374 
7375   // The GC barrier expanded by store_heap_oop below may call into the
7376   // runtime so use callee-saved registers for any values that need to be
7377   // preserved. The GC barrier assembler should take care of saving the
7378   // Java argument registers.
7379   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7380   Register val_obj_tmp = r21;
7381   Register from_reg_tmp = r22;
7382   Register tmp1 = r14;
7383   Register tmp2 = r13;
7384   Register tmp3 = r12;
7385   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7386 
7387   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7388 
7389   if (reg_state[to->value()] == reg_readonly) {
7390     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7391       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7392       return false; // Not yet writable
7393     }
7394     val_obj = val_obj_tmp;
7395   }
7396 
7397   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7398   VMReg fromReg;
7399   BasicType bt;
7400   Label L_null;
7401   while (stream.next(fromReg, bt)) {
7402     assert(fromReg->is_valid(), "source must be valid");
7403     reg_state[fromReg->value()] = reg_writable;
7404 
7405     int off = sig->at(stream.sig_index())._offset;
7406     if (off == -1) {
7407       // Nullable inline type argument, emit null check
7408       Label L_notNull;
7409       if (fromReg->is_stack()) {
7410         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7411         ldrb(tmp2, Address(sp, ld_off));
7412         cbnz(tmp2, L_notNull);
7413       } else {
7414         cbnz(fromReg->as_Register(), L_notNull);
7415       }
7416       mov(val_obj, 0);
7417       b(L_null);
7418       bind(L_notNull);
7419       continue;
7420     }
7421     if (sig->at(stream.sig_index())._vt_oop) {
7422       if (fromReg->is_stack()) {
7423         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7424         ldr(val_obj, Address(sp, ld_off));
7425       } else {
7426         mov(val_obj, fromReg->as_Register());
7427       }
7428       cbnz(val_obj, L_null);
7429       // get the buffer from the just allocated pool of buffers
7430       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7431       load_heap_oop(val_obj, Address(val_array, index), rscratch1, rscratch2);
7432       continue;
7433     }
7434 
7435     assert(off > 0, "offset in object should be positive");
7436     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7437 
7438     // Pack the scalarized field into the value object.
7439     Address dst(val_obj, off);
7440     if (!fromReg->is_FloatRegister()) {
7441       Register src;
7442       if (fromReg->is_stack()) {
7443         src = from_reg_tmp;
7444         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7445         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7446       } else {
7447         src = fromReg->as_Register();
7448       }
7449       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7450       if (is_reference_type(bt)) {
7451         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
7452         mov(tmp3, val_obj);
7453         Address dst_with_tmp3(tmp3, off);
7454         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7455       } else {
7456         store_sized_value(dst, src, size_in_bytes);
7457       }
7458     } else if (bt == T_DOUBLE) {
7459       strd(fromReg->as_FloatRegister(), dst);
7460     } else {
7461       assert(bt == T_FLOAT, "must be float");
7462       strs(fromReg->as_FloatRegister(), dst);
7463     }
7464   }
7465   bind(L_null);
7466   sig_index = stream.sig_index();
7467   from_index = stream.regs_index();
7468 
7469   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7470   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7471   assert(success, "to register must be writable");
7472   return true;
7473 }
7474 
7475 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7476   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7477 }
7478 
7479 void MacroAssembler::cache_wb(Address line) {
7480   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7481   assert(line.index() == noreg, "index should be noreg");
7482   assert(line.offset() == 0, "offset should be 0");
7483   // would like to assert this
7484   // assert(line._ext.shift == 0, "shift should be zero");
7485   if (VM_Version::supports_dcpop()) {
7486     // writeback using clear virtual address to point of persistence
7487     dc(Assembler::CVAP, line.base());
7488   } else {
7489     // no need to generate anything as Unsafe.writebackMemory should
7490     // never invoke this stub
7491   }
7492 }
7493 
7494 void MacroAssembler::cache_wbsync(bool is_pre) {
7495   // we only need a barrier post sync
7496   if (!is_pre) {
7497     membar(Assembler::AnyAny);
7498   }
7499 }
7500 
7501 void MacroAssembler::verify_sve_vector_length(Register tmp) {
7502   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
7503     return;
7504   }
7505   // Make sure that native code does not change SVE vector length.
7506   Label verify_ok;
7507   movw(tmp, zr);
7508   sve_inc(tmp, B);
7509   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
7510   br(EQ, verify_ok);
7511   stop("Error: SVE vector length has changed since jvm startup");
7512   bind(verify_ok);
7513 }
7514 
7515 void MacroAssembler::verify_ptrue() {
7516   Label verify_ok;
7517   if (!UseSVE) {
7518     return;
7519   }
7520   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
7521   sve_dec(rscratch1, B);
7522   cbz(rscratch1, verify_ok);
7523   stop("Error: the preserved predicate register (p7) elements are not all true");
7524   bind(verify_ok);
7525 }
7526 
7527 void MacroAssembler::safepoint_isb() {
7528   isb();
7529 #ifndef PRODUCT
7530   if (VerifyCrossModifyFence) {
7531     // Clear the thread state.
7532     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7533   }
7534 #endif
7535 }
7536 
7537 #ifndef PRODUCT
7538 void MacroAssembler::verify_cross_modify_fence_not_required() {
7539   if (VerifyCrossModifyFence) {
7540     // Check if thread needs a cross modify fence.
7541     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7542     Label fence_not_required;
7543     cbz(rscratch1, fence_not_required);
7544     // If it does then fail.
7545     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
7546     mov(c_rarg0, rthread);
7547     blr(rscratch1);
7548     bind(fence_not_required);
7549   }
7550 }
7551 #endif
7552 
7553 void MacroAssembler::spin_wait() {
7554   block_comment("spin_wait {");
7555   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
7556     switch (VM_Version::spin_wait_desc().inst()) {
7557       case SpinWait::NOP:
7558         nop();
7559         break;
7560       case SpinWait::ISB:
7561         isb();
7562         break;
7563       case SpinWait::YIELD:
7564         yield();
7565         break;
7566       case SpinWait::SB:
7567         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
7568         sb();
7569         break;
7570       case SpinWait::WFET:
7571         spin_wait_wfet(VM_Version::spin_wait_desc().delay());
7572         break;
7573       default:
7574         ShouldNotReachHere();
7575     }
7576   }
7577   block_comment("}");
7578 }
7579 
7580 void MacroAssembler::spin_wait_wfet(int delay_ns) {
7581   // The sequence assumes CNTFRQ_EL0 is fixed to 1GHz. The assumption is valid
7582   // starting from Armv8.6, according to the "D12.1.2 The system counter" of the
7583   // Arm Architecture Reference Manual for A-profile architecture version M.a.a.
7584   // This is sufficient because FEAT_WFXT is introduced from Armv8.6.
7585   Register target = rscratch1;
7586   Register current = rscratch2;
7587   get_cntvctss_el0(current);
7588   add(target, current, delay_ns);
7589 
7590   Label L_wait_loop;
7591   bind(L_wait_loop);
7592 
7593   wfet(target);
7594   get_cntvctss_el0(current);
7595 
7596   cmp(current, target);
7597   br(LT, L_wait_loop);
7598 
7599   sb();
7600 }
7601 
7602 // Stack frame creation/removal
7603 
7604 void MacroAssembler::enter(bool strip_ret_addr) {
7605   if (strip_ret_addr) {
7606     // Addresses can only be signed once. If there are multiple nested frames being created
7607     // in the same function, then the return address needs stripping first.
7608     strip_return_address();
7609   }
7610   protect_return_address();
7611   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7612   mov(rfp, sp);
7613 }
7614 
7615 void MacroAssembler::leave() {
7616   mov(sp, rfp);
7617   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
7618   authenticate_return_address();
7619 }
7620 
7621 // ROP Protection
7622 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
7623 // destroying stack frames or whenever directly loading/storing the LR to memory.
7624 // If ROP protection is not set then these functions are no-ops.
7625 // For more details on PAC see pauth_aarch64.hpp.
7626 
7627 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
7628 // Uses value zero as the modifier.
7629 //
7630 void MacroAssembler::protect_return_address() {
7631   if (VM_Version::use_rop_protection()) {
7632     check_return_address();
7633     paciaz();
7634   }
7635 }
7636 
7637 // Sign the return value in the given register. Use before updating the LR in the existing stack
7638 // frame for the current function.
7639 // Uses value zero as the modifier.
7640 //
7641 void MacroAssembler::protect_return_address(Register return_reg) {
7642   if (VM_Version::use_rop_protection()) {
7643     check_return_address(return_reg);
7644     paciza(return_reg);
7645   }
7646 }
7647 
7648 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
7649 // Uses value zero as the modifier.
7650 //
7651 void MacroAssembler::authenticate_return_address() {
7652   if (VM_Version::use_rop_protection()) {
7653     autiaz();
7654     check_return_address();
7655   }
7656 }
7657 
7658 // Authenticate the return value in the given register. Use before updating the LR in the existing
7659 // stack frame for the current function.
7660 // Uses value zero as the modifier.
7661 //
7662 void MacroAssembler::authenticate_return_address(Register return_reg) {
7663   if (VM_Version::use_rop_protection()) {
7664     autiza(return_reg);
7665     check_return_address(return_reg);
7666   }
7667 }
7668 
7669 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
7670 // there is no guaranteed way of authenticating the LR.
7671 //
7672 void MacroAssembler::strip_return_address() {
7673   if (VM_Version::use_rop_protection()) {
7674     xpaclri();
7675   }
7676 }
7677 
7678 #ifndef PRODUCT
7679 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
7680 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
7681 // it is difficult to debug back to the callee function.
7682 // This function simply loads from the address in the given register.
7683 // Use directly after authentication to catch authentication failures.
7684 // Also use before signing to check that the pointer is valid and hasn't already been signed.
7685 //
7686 void MacroAssembler::check_return_address(Register return_reg) {
7687   if (VM_Version::use_rop_protection()) {
7688     ldr(zr, Address(return_reg));
7689   }
7690 }
7691 #endif
7692 
7693 // The java_calling_convention describes stack locations as ideal slots on
7694 // a frame with no abi restrictions. Since we must observe abi restrictions
7695 // (like the placement of the register window) the slots must be biased by
7696 // the following value.
7697 static int reg2offset_in(VMReg r) {
7698   // Account for saved rfp and lr
7699   // This should really be in_preserve_stack_slots
7700   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
7701 }
7702 
7703 static int reg2offset_out(VMReg r) {
7704   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
7705 }
7706 
7707 // On 64bit we will store integer like items to the stack as
7708 // 64bits items (AArch64 ABI) even though java would only store
7709 // 32bits for a parameter. On 32bit it will simply be 32bits
7710 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
7711 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
7712   if (src.first()->is_stack()) {
7713     if (dst.first()->is_stack()) {
7714       // stack to stack
7715       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7716       str(tmp, Address(sp, reg2offset_out(dst.first())));
7717     } else {
7718       // stack to reg
7719       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7720     }
7721   } else if (dst.first()->is_stack()) {
7722     // reg to stack
7723     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7724   } else {
7725     if (dst.first() != src.first()) {
7726       sxtw(dst.first()->as_Register(), src.first()->as_Register());
7727     }
7728   }
7729 }
7730 
7731 // An oop arg. Must pass a handle not the oop itself
7732 void MacroAssembler::object_move(
7733                         OopMap* map,
7734                         int oop_handle_offset,
7735                         int framesize_in_slots,
7736                         VMRegPair src,
7737                         VMRegPair dst,
7738                         bool is_receiver,
7739                         int* receiver_offset) {
7740 
7741   // must pass a handle. First figure out the location we use as a handle
7742 
7743   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
7744 
7745   // See if oop is null if it is we need no handle
7746 
7747   if (src.first()->is_stack()) {
7748 
7749     // Oop is already on the stack as an argument
7750     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
7751     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
7752     if (is_receiver) {
7753       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
7754     }
7755 
7756     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7757     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7758     // conditionally move a null
7759     cmp(rscratch1, zr);
7760     csel(rHandle, zr, rHandle, Assembler::EQ);
7761   } else {
7762 
7763     // Oop is in an a register we must store it to the space we reserve
7764     // on the stack for oop_handles and pass a handle if oop is non-null
7765 
7766     const Register rOop = src.first()->as_Register();
7767     int oop_slot;
7768     if (rOop == j_rarg0)
7769       oop_slot = 0;
7770     else if (rOop == j_rarg1)
7771       oop_slot = 1;
7772     else if (rOop == j_rarg2)
7773       oop_slot = 2;
7774     else if (rOop == j_rarg3)
7775       oop_slot = 3;
7776     else if (rOop == j_rarg4)
7777       oop_slot = 4;
7778     else if (rOop == j_rarg5)
7779       oop_slot = 5;
7780     else if (rOop == j_rarg6)
7781       oop_slot = 6;
7782     else {
7783       assert(rOop == j_rarg7, "wrong register");
7784       oop_slot = 7;
7785     }
7786 
7787     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7788     int offset = oop_slot*VMRegImpl::stack_slot_size;
7789 
7790     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7791     // Store oop in handle area, may be null
7792     str(rOop, Address(sp, offset));
7793     if (is_receiver) {
7794       *receiver_offset = offset;
7795     }
7796 
7797     cmp(rOop, zr);
7798     lea(rHandle, Address(sp, offset));
7799     // conditionally move a null
7800     csel(rHandle, zr, rHandle, Assembler::EQ);
7801   }
7802 
7803   // If arg is on the stack then place it otherwise it is already in correct reg.
7804   if (dst.first()->is_stack()) {
7805     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7806   }
7807 }
7808 
7809 // A float arg may have to do float reg int reg conversion
7810 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7811  if (src.first()->is_stack()) {
7812     if (dst.first()->is_stack()) {
7813       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7814       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7815     } else {
7816       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7817     }
7818   } else if (src.first() != dst.first()) {
7819     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7820       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7821     else
7822       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7823   }
7824 }
7825 
7826 // A long move
7827 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7828   if (src.first()->is_stack()) {
7829     if (dst.first()->is_stack()) {
7830       // stack to stack
7831       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7832       str(tmp, Address(sp, reg2offset_out(dst.first())));
7833     } else {
7834       // stack to reg
7835       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7836     }
7837   } else if (dst.first()->is_stack()) {
7838     // reg to stack
7839     // Do we really have to sign extend???
7840     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7841     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7842   } else {
7843     if (dst.first() != src.first()) {
7844       mov(dst.first()->as_Register(), src.first()->as_Register());
7845     }
7846   }
7847 }
7848 
7849 
7850 // A double move
7851 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7852  if (src.first()->is_stack()) {
7853     if (dst.first()->is_stack()) {
7854       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7855       str(tmp, Address(sp, reg2offset_out(dst.first())));
7856     } else {
7857       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7858     }
7859   } else if (src.first() != dst.first()) {
7860     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7861       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7862     else
7863       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7864   }
7865 }
7866 
7867 // Implements fast-locking.
7868 //
7869 //  - obj: the object to be locked
7870 //  - t1, t2, t3: temporary registers, will be destroyed
7871 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7872 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7873   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7874 
7875   Label push;
7876   const Register top = t1;
7877   const Register mark = t2;
7878   const Register t = t3;
7879 
7880   // Preload the markWord. It is important that this is the first
7881   // instruction emitted as it is part of C1's null check semantics.
7882   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7883 
7884   if (UseObjectMonitorTable) {
7885     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7886     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7887   }
7888 
7889   if (DiagnoseSyncOnValueBasedClasses != 0) {
7890     load_klass(t1, obj, rscratch1);
7891     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7892     tst(t1, KlassFlags::_misc_is_value_based_class);
7893     br(Assembler::NE, slow);
7894   }
7895 
7896   // Check if the lock-stack is full.
7897   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7898   cmpw(top, (unsigned)LockStack::end_offset());
7899   br(Assembler::GE, slow);
7900 
7901   // Check for recursion.
7902   subw(t, top, oopSize);
7903   ldr(t, Address(rthread, t));
7904   cmp(obj, t);
7905   br(Assembler::EQ, push);
7906 
7907   // Check header for monitor (0b10).
7908   tst(mark, markWord::monitor_value);
7909   br(Assembler::NE, slow);
7910 
7911   // Try to lock. Transition lock bits 0b01 => 0b00
7912   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7913   orr(mark, mark, markWord::unlocked_value);
7914   // Mask inline_type bit such that we go to the slow path if object is an inline type
7915   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7916 
7917   eor(t, mark, markWord::unlocked_value);
7918   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword, memory_order_acquire);
7919   br(Assembler::NE, slow);
7920 
7921   bind(push);
7922   // After successful lock, push object on lock-stack.
7923   str(obj, Address(rthread, top));
7924   addw(top, top, oopSize);
7925   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7926 }
7927 
7928 // Implements fast-unlocking.
7929 //
7930 // - obj: the object to be unlocked
7931 // - t1, t2, t3: temporary registers
7932 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7933 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7934   // cmpxchg clobbers rscratch1.
7935   assert_different_registers(obj, t1, t2, t3, rscratch1);
7936 
7937 #ifdef ASSERT
7938   {
7939     // Check for lock-stack underflow.
7940     Label stack_ok;
7941     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7942     cmpw(t1, (unsigned)LockStack::start_offset());
7943     br(Assembler::GE, stack_ok);
7944     STOP("Lock-stack underflow");
7945     bind(stack_ok);
7946   }
7947 #endif
7948 
7949   Label unlocked, push_and_slow;
7950   const Register top = t1;
7951   const Register mark = t2;
7952   const Register t = t3;
7953 
7954   // Check if obj is top of lock-stack.
7955   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7956   subw(top, top, oopSize);
7957   ldr(t, Address(rthread, top));
7958   cmp(obj, t);
7959   br(Assembler::NE, slow);
7960 
7961   // Pop lock-stack.
7962   DEBUG_ONLY(str(zr, Address(rthread, top));)
7963   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7964 
7965   // Check if recursive.
7966   subw(t, top, oopSize);
7967   ldr(t, Address(rthread, t));
7968   cmp(obj, t);
7969   br(Assembler::EQ, unlocked);
7970 
7971   // Not recursive. Check header for monitor (0b10).
7972   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7973   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7974 
7975 #ifdef ASSERT
7976   // Check header not unlocked (0b01).
7977   Label not_unlocked;
7978   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7979   stop("fast_unlock already unlocked");
7980   bind(not_unlocked);
7981 #endif
7982 
7983   // Try to unlock. Transition lock bits 0b00 => 0b01
7984   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7985   orr(t, mark, markWord::unlocked_value);
7986   cmpxchg(obj, mark, t, Assembler::xword, memory_order_release);
7987   br(Assembler::EQ, unlocked);
7988 
7989   bind(push_and_slow);
7990   // Restore lock-stack and handle the unlock in runtime.
7991   DEBUG_ONLY(str(obj, Address(rthread, top));)
7992   addw(top, top, oopSize);
7993   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7994   b(slow);
7995 
7996   bind(unlocked);
7997 }
7998 
7999 // Rotate using USHR and SLI instructions (or copy, if rotate count is zero)
8000 void MacroAssembler::neon_vector_rotate(FloatRegister dst, SIMD_Arrangement T,
8001                                         FloatRegister src, int shift_amount) {
8002   assert(src != dst, "did not expect src and dst to be the same register");
8003 
8004   int esize = BitsPerByte << (T / 2);
8005   int lshift = shift_amount & (esize - 1);
8006 
8007   if (lshift == 0) {
8008     // T & 1 == 0 => 64-bit arrangements, else 128-bit arrangements
8009     orr(dst, (T & 1) == 0 ? T8B : T16B, src, src);
8010   } else {
8011     ushr(dst, T, src, esize - lshift);
8012     sli(dst, T, src, lshift);
8013   }
8014 }
8015 
8016 void MacroAssembler::try_to_replace_prev_vector_copy_with_movprfx(FloatRegister dst) {
8017   if (code_section()->is_empty()) {
8018     return;
8019   }
8020 
8021   address prev = pc() - NativeInstruction::instruction_size;
8022   uint32_t insn = nativeInstruction_at(prev)->encoding();
8023   if (!NativeInstruction::is_neon_vector_mov_alias(insn) &&
8024       !NativeInstruction::is_sve_vector_mov_alias(insn)) {
8025     return;
8026   }
8027 
8028   // The destructive instruction must reuse the mov alias destination.
8029   uint32_t rd = Instruction_aarch64::extract(insn, 4, 0);
8030   if (rd != (uint32_t)dst->encoding()) {
8031     return;
8032   }
8033 
8034   uint32_t rn = Instruction_aarch64::extract(insn, 9, 5);
8035   Instruction_aarch64::patch(prev, 31, 0,
8036                              NativeInstruction::encode_sve_movprfx(rd, rn));
8037 }