1 /*
   2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "ci/ciUtilities.hpp"
  30 #include "code/compiledIC.hpp"
  31 #include "compiler/compileTask.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/barrierSet.hpp"
  35 #include "gc/shared/barrierSetAssembler.hpp"
  36 #include "gc/shared/cardTableBarrierSet.hpp"
  37 #include "gc/shared/cardTable.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/tlab_globals.hpp"
  40 #include "interpreter/bytecodeHistogram.hpp"
  41 #include "interpreter/interpreter.hpp"
  42 #include "interpreter/interpreterRuntime.hpp"
  43 #include "jvm.h"
  44 #include "memory/resourceArea.hpp"
  45 #include "memory/universe.hpp"
  46 #include "nativeInst_aarch64.hpp"
  47 #include "oops/accessDecorators.hpp"
  48 #include "oops/compressedKlass.inline.hpp"
  49 #include "oops/compressedOops.inline.hpp"
  50 #include "oops/klass.inline.hpp"
  51 #include "runtime/continuation.hpp"
  52 #include "runtime/icache.hpp"
  53 #include "runtime/interfaceSupport.inline.hpp"
  54 #include "runtime/javaThread.hpp"
  55 #include "runtime/jniHandles.inline.hpp"
  56 #include "runtime/sharedRuntime.hpp"
  57 #include "runtime/stubRoutines.hpp"
  58 #include "utilities/globalDefinitions.hpp"
  59 #include "utilities/powerOfTwo.hpp"
  60 #ifdef COMPILER1
  61 #include "c1/c1_LIRAssembler.hpp"
  62 #endif
  63 #ifdef COMPILER2
  64 #include "oops/oop.hpp"
  65 #include "opto/compile.hpp"
  66 #include "opto/node.hpp"
  67 #include "opto/output.hpp"
  68 #endif
  69 
  70 #include <sys/types.h>
  71 
  72 #ifdef PRODUCT
  73 #define BLOCK_COMMENT(str) /* nothing */
  74 #else
  75 #define BLOCK_COMMENT(str) block_comment(str)
  76 #endif
  77 #define STOP(str) stop(str);
  78 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  79 
  80 #ifdef ASSERT
  81 extern "C" void disnm(intptr_t p);
  82 #endif
  83 // Target-dependent relocation processing
  84 //
  85 // Instruction sequences whose target may need to be retrieved or
  86 // patched are distinguished by their leading instruction, sorting
  87 // them into three main instruction groups and related subgroups.
  88 //
  89 // 1) Branch, Exception and System (insn count = 1)
  90 //    1a) Unconditional branch (immediate):
  91 //      b/bl imm19
  92 //    1b) Compare & branch (immediate):
  93 //      cbz/cbnz Rt imm19
  94 //    1c) Test & branch (immediate):
  95 //      tbz/tbnz Rt imm14
  96 //    1d) Conditional branch (immediate):
  97 //      b.cond imm19
  98 //
  99 // 2) Loads and Stores (insn count = 1)
 100 //    2a) Load register literal:
 101 //      ldr Rt imm19
 102 //
 103 // 3) Data Processing Immediate (insn count = 2 or 3)
 104 //    3a) PC-rel. addressing
 105 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 106 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 107 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 108 //      adr/adrp Rx imm21
 109 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 110 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 111 //      The latter form can only happen when the target is an
 112 //      ExternalAddress, and (by definition) ExternalAddresses don't
 113 //      move. Because of that property, there is never any need to
 114 //      patch the last of the three instructions. However,
 115 //      MacroAssembler::target_addr_for_insn takes all three
 116 //      instructions into account and returns the correct address.
 117 //    3b) Move wide (immediate)
 118 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 119 //
 120 // A switch on a subset of the instruction's bits provides an
 121 // efficient dispatch to these subcases.
 122 //
 123 // insn[28:26] -> main group ('x' == don't care)
 124 //   00x -> UNALLOCATED
 125 //   100 -> Data Processing Immediate
 126 //   101 -> Branch, Exception and System
 127 //   x1x -> Loads and Stores
 128 //
 129 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 130 // n.b. in some cases extra bits need to be checked to verify the
 131 // instruction is as expected
 132 //
 133 // 1) ... xx101x Branch, Exception and System
 134 //   1a)  00___x Unconditional branch (immediate)
 135 //   1b)  01___0 Compare & branch (immediate)
 136 //   1c)  01___1 Test & branch (immediate)
 137 //   1d)  10___0 Conditional branch (immediate)
 138 //        other  Should not happen
 139 //
 140 // 2) ... xxx1x0 Loads and Stores
 141 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 142 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 143 //                strictly should be 64 bit non-FP/SIMD i.e.
 144 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 145 //
 146 // 3) ... xx100x Data Processing Immediate
 147 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 148 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 149 //                 strictly should be 64 bit movz #imm16<<0
 150 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 151 //
 152 
 153 static uint32_t insn_at(address insn_addr, int n) {
 154   return ((uint32_t*)insn_addr)[n];
 155 }
 156 
 157 template<typename T>
 158 class RelocActions : public AllStatic {
 159 
 160 public:
 161 
 162   static int ALWAYSINLINE run(address insn_addr, address &target) {
 163     int instructions = 1;
 164     uint32_t insn = insn_at(insn_addr, 0);
 165 
 166     uint32_t dispatch = Instruction_aarch64::extract(insn, 30, 25);
 167     switch(dispatch) {
 168       case 0b001010:
 169       case 0b001011: {
 170         instructions = T::unconditionalBranch(insn_addr, target);
 171         break;
 172       }
 173       case 0b101010:   // Conditional branch (immediate)
 174       case 0b011010: { // Compare & branch (immediate)
 175         instructions = T::conditionalBranch(insn_addr, target);
 176         break;
 177       }
 178       case 0b011011: {
 179         instructions = T::testAndBranch(insn_addr, target);
 180         break;
 181       }
 182       case 0b001100:
 183       case 0b001110:
 184       case 0b011100:
 185       case 0b011110:
 186       case 0b101100:
 187       case 0b101110:
 188       case 0b111100:
 189       case 0b111110: {
 190         // load/store
 191         if ((Instruction_aarch64::extract(insn, 29, 24) & 0b111011) == 0b011000) {
 192           // Load register (literal)
 193           instructions = T::loadStore(insn_addr, target);
 194           break;
 195         } else {
 196           // nothing to do
 197           assert(target == nullptr, "did not expect to relocate target for polling page load");
 198         }
 199         break;
 200       }
 201       case 0b001000:
 202       case 0b011000:
 203       case 0b101000:
 204       case 0b111000: {
 205         // adr/adrp
 206         assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 207         int shift = Instruction_aarch64::extract(insn, 31, 31);
 208         if (shift) {
 209           uint32_t insn2 = insn_at(insn_addr, 1);
 210           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 211               Instruction_aarch64::extract(insn, 4, 0) ==
 212               Instruction_aarch64::extract(insn2, 9, 5)) {
 213             instructions = T::adrp(insn_addr, target, T::adrpMem);
 214           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 215                      Instruction_aarch64::extract(insn, 4, 0) ==
 216                      Instruction_aarch64::extract(insn2, 4, 0)) {
 217             instructions = T::adrp(insn_addr, target, T::adrpAdd);
 218           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 219                      Instruction_aarch64::extract(insn, 4, 0) ==
 220                      Instruction_aarch64::extract(insn2, 4, 0)) {
 221             instructions = T::adrp(insn_addr, target, T::adrpMovk);
 222           } else {
 223             ShouldNotReachHere();
 224           }
 225         } else {
 226           instructions = T::adr(insn_addr, target);
 227         }
 228         break;
 229       }
 230       case 0b001001:
 231       case 0b011001:
 232       case 0b101001:
 233       case 0b111001: {
 234         instructions = T::immediate(insn_addr, target);
 235         break;
 236       }
 237       default: {
 238         ShouldNotReachHere();
 239       }
 240     }
 241 
 242     T::verify(insn_addr, target);
 243     return instructions * NativeInstruction::instruction_size;
 244   }
 245 };
 246 
 247 class Patcher : public AllStatic {
 248 public:
 249   static int unconditionalBranch(address insn_addr, address &target) {
 250     intptr_t offset = (target - insn_addr) >> 2;
 251     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 252     return 1;
 253   }
 254   static int conditionalBranch(address insn_addr, address &target) {
 255     intptr_t offset = (target - insn_addr) >> 2;
 256     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 257     return 1;
 258   }
 259   static int testAndBranch(address insn_addr, address &target) {
 260     intptr_t offset = (target - insn_addr) >> 2;
 261     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 262     return 1;
 263   }
 264   static int loadStore(address insn_addr, address &target) {
 265     intptr_t offset = (target - insn_addr) >> 2;
 266     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 267     return 1;
 268   }
 269   static int adr(address insn_addr, address &target) {
 270 #ifdef ASSERT
 271     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 272 #endif
 273     // PC-rel. addressing
 274     ptrdiff_t offset = target - insn_addr;
 275     int offset_lo = offset & 3;
 276     offset >>= 2;
 277     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 278     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 279     return 1;
 280   }
 281   template<typename U>
 282   static int adrp(address insn_addr, address &target, U inner) {
 283     int instructions = 1;
 284 #ifdef ASSERT
 285     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 286 #endif
 287     ptrdiff_t offset = target - insn_addr;
 288     instructions = 2;
 289     precond(inner != nullptr);
 290     // Give the inner reloc a chance to modify the target.
 291     address adjusted_target = target;
 292     instructions = inner(insn_addr, adjusted_target);
 293     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 294     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 295     offset = adr_page - pc_page;
 296     int offset_lo = offset & 3;
 297     offset >>= 2;
 298     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 299     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 300     return instructions;
 301   }
 302   static int adrpMem(address insn_addr, address &target) {
 303     uintptr_t dest = (uintptr_t)target;
 304     int offset_lo = dest & 0xfff;
 305     uint32_t insn2 = insn_at(insn_addr, 1);
 306     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 307     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 308     guarantee(((dest >> size) << size) == dest, "misaligned target");
 309     return 2;
 310   }
 311   static int adrpAdd(address insn_addr, address &target) {
 312     uintptr_t dest = (uintptr_t)target;
 313     int offset_lo = dest & 0xfff;
 314     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 315     return 2;
 316   }
 317   static int adrpMovk(address insn_addr, address &target) {
 318     uintptr_t dest = uintptr_t(target);
 319     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 320     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 321     target = address(dest);
 322     return 2;
 323   }
 324   static int immediate(address insn_addr, address &target) {
 325     // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 326     // We encode narrow ones by setting the upper 16 bits in the first
 327     // instruction.
 328     if (Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010101) {
 329       assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 330       narrowKlass nk = CompressedKlassPointers::encode((Klass*)target);
 331       Instruction_aarch64::patch(insn_addr, 20, 5, nk >> 16);
 332       Instruction_aarch64::patch(insn_addr+4, 20, 5, nk & 0xffff);
 333       return 2;
 334     }
 335     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010100, "must be");
 336     uint64_t dest = (uint64_t)target;
 337     // Move wide constant
 338     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 339     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 340     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 341     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 342     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 343     return 3;
 344   }
 345   static void verify(address insn_addr, address &target) {
 346 #ifdef ASSERT
 347     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 348     if (!(address_is == target)) {
 349       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 350       disnm((intptr_t)insn_addr);
 351       assert(address_is == target, "should be");
 352     }
 353 #endif
 354   }
 355 };
 356 
 357 // If insn1 and insn2 use the same register to form an address, either
 358 // by an offsetted LDR or a simple ADD, return the offset. If the
 359 // second instruction is an LDR, the offset may be scaled.
 360 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 361   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 362       Instruction_aarch64::extract(insn1, 4, 0) ==
 363       Instruction_aarch64::extract(insn2, 9, 5)) {
 364     // Load/store register (unsigned immediate)
 365     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 366     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 367     byte_offset <<= size;
 368     return true;
 369   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 370              Instruction_aarch64::extract(insn1, 4, 0) ==
 371              Instruction_aarch64::extract(insn2, 4, 0)) {
 372     // add (immediate)
 373     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 374     return true;
 375   }
 376   return false;
 377 }
 378 
 379 class AArch64Decoder : public AllStatic {
 380 public:
 381 
 382   static int loadStore(address insn_addr, address &target) {
 383     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 384     target = insn_addr + (offset << 2);
 385     return 1;
 386   }
 387   static int unconditionalBranch(address insn_addr, address &target) {
 388     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 25, 0);
 389     target = insn_addr + (offset << 2);
 390     return 1;
 391   }
 392   static int conditionalBranch(address insn_addr, address &target) {
 393     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 394     target = address(((uint64_t)insn_addr + (offset << 2)));
 395     return 1;
 396   }
 397   static int testAndBranch(address insn_addr, address &target) {
 398     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 18, 5);
 399     target = address(((uint64_t)insn_addr + (offset << 2)));
 400     return 1;
 401   }
 402   static int adr(address insn_addr, address &target) {
 403     // PC-rel. addressing
 404     uint32_t insn = insn_at(insn_addr, 0);
 405     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 406     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 407     target = address((uint64_t)insn_addr + offset);
 408     return 1;
 409   }
 410   template<typename U>
 411   static int adrp(address insn_addr, address &target, U inner) {
 412     uint32_t insn = insn_at(insn_addr, 0);
 413     assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 414     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 415     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 416     int shift = 12;
 417     offset <<= shift;
 418     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 419     target_page &= ((uint64_t)-1) << shift;
 420     target = address(target_page);
 421     precond(inner != nullptr);
 422     inner(insn_addr, target);
 423     return 2;
 424   }
 425   static int adrpMem(address insn_addr, address &target) {
 426     uint32_t insn2 = insn_at(insn_addr, 1);
 427     // Load/store register (unsigned immediate)
 428     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 429     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 430     byte_offset <<= size;
 431     target += byte_offset;
 432     return 2;
 433   }
 434   static int adrpAdd(address insn_addr, address &target) {
 435     uint32_t insn2 = insn_at(insn_addr, 1);
 436     // add (immediate)
 437     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 438     target += byte_offset;
 439     return 2;
 440   }
 441   static int adrpMovk(address insn_addr, address &target) {
 442     uint32_t insn2 = insn_at(insn_addr, 1);
 443     uint64_t dest = uint64_t(target);
 444     dest = (dest & 0xffff0000ffffffff) |
 445       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 446     target = address(dest);
 447 
 448     // We know the destination 4k page. Maybe we have a third
 449     // instruction.
 450     uint32_t insn = insn_at(insn_addr, 0);
 451     uint32_t insn3 = insn_at(insn_addr, 2);
 452     ptrdiff_t byte_offset;
 453     if (offset_for(insn, insn3, byte_offset)) {
 454       target += byte_offset;
 455       return 3;
 456     } else {
 457       return 2;
 458     }
 459   }
 460   static int immediate(address insn_addr, address &target) {
 461     uint32_t *insns = (uint32_t *)insn_addr;
 462     // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 463     // We encode narrow ones by setting the upper 16 bits in the first
 464     // instruction.
 465     if (Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010101) {
 466       assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 467       narrowKlass nk = (narrowKlass)((uint32_t(Instruction_aarch64::extract(insns[0], 20, 5)) << 16)
 468                                    +  uint32_t(Instruction_aarch64::extract(insns[1], 20, 5)));
 469       target = (address)CompressedKlassPointers::decode(nk);
 470       return 2;
 471     }
 472     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 473     // Move wide constant: movz, movk, movk.  See movptr().
 474     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 475     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 476     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 477                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 478                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 479     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 480     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 481     return 3;
 482   }
 483   static void verify(address insn_addr, address &target) {
 484   }
 485 };
 486 
 487 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 488   address target;
 489   RelocActions<AArch64Decoder>::run(insn_addr, target);
 490   return target;
 491 }
 492 
 493 // Patch any kind of instruction; there may be several instructions.
 494 // Return the total length (in bytes) of the instructions.
 495 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 496   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 497   return RelocActions<Patcher>::run(insn_addr, target);
 498 }
 499 
 500 int MacroAssembler::patch_oop(address insn_addr, address o) {
 501   int instructions;
 502   unsigned insn = *(unsigned*)insn_addr;
 503   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 504 
 505   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 506 
 507   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 508   // narrow OOPs by setting the upper 16 bits in the first
 509   // instruction.
 510   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 511     // Move narrow OOP
 512     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 513     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 514     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 515     instructions = 2;
 516   } else {
 517     // Move wide OOP
 518     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 519     uintptr_t dest = (uintptr_t)o;
 520     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 521     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 522     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 523     instructions = 3;
 524   }
 525   return instructions * NativeInstruction::instruction_size;
 526 }
 527 
 528 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 529   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 530   // We encode narrow ones by setting the upper 16 bits in the first
 531   // instruction.
 532   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 533   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 534          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 535 
 536   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 537 
 538   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 539   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 540   return 2 * NativeInstruction::instruction_size;
 541 }
 542 
 543 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 544   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 545   if (at_return) {
 546     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 547     // we may safely use the sp instead to perform the stack watermark check.
 548     cmp(in_nmethod ? sp : rfp, tmp);
 549     br(Assembler::HI, slow_path);
 550   } else {
 551     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 552   }
 553 }
 554 
 555 void MacroAssembler::rt_call(address dest, Register tmp) {
 556   CodeBlob *cb = CodeCache::find_blob(dest);
 557   if (cb) {
 558     far_call(RuntimeAddress(dest));
 559   } else {
 560     lea(tmp, RuntimeAddress(dest));
 561     blr(tmp);
 562   }
 563 }
 564 
 565 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 566   if (!Continuations::enabled()) return;
 567   Label done;
 568   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 569   cmp(sp, rscratch1);
 570   br(Assembler::LS, done);
 571   mov(rscratch1, sp); // we can't use sp as the source in str
 572   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 573   bind(done);
 574 }
 575 
 576 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 577   if (!Continuations::enabled()) return;
 578   Label done;
 579   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 580   cmp(sp, rscratch1);
 581   br(Assembler::LO, done);
 582   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 583   bind(done);
 584 }
 585 
 586 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 587   // we must set sp to zero to clear frame
 588   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 589 
 590   // must clear fp, so that compiled frames are not confused; it is
 591   // possible that we need it only for debugging
 592   if (clear_fp) {
 593     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 594   }
 595 
 596   // Always clear the pc because it could have been set by make_walkable()
 597   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 598 }
 599 
 600 // Calls to C land
 601 //
 602 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 603 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 604 // has to be reset to 0. This is required to allow proper stack traversal.
 605 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 606                                          Register last_java_fp,
 607                                          Register last_java_pc,
 608                                          Register scratch) {
 609 
 610   if (last_java_pc->is_valid()) {
 611       str(last_java_pc, Address(rthread,
 612                                 JavaThread::frame_anchor_offset()
 613                                 + JavaFrameAnchor::last_Java_pc_offset()));
 614     }
 615 
 616   // determine last_java_sp register
 617   if (last_java_sp == sp) {
 618     mov(scratch, sp);
 619     last_java_sp = scratch;
 620   } else if (!last_java_sp->is_valid()) {
 621     last_java_sp = esp;
 622   }
 623 
 624   // last_java_fp is optional
 625   if (last_java_fp->is_valid()) {
 626     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 627   }
 628 
 629   // We must set sp last.
 630   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 631 }
 632 
 633 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 634                                          Register last_java_fp,
 635                                          address  last_java_pc,
 636                                          Register scratch) {
 637   assert(last_java_pc != nullptr, "must provide a valid PC");
 638 
 639   adr(scratch, last_java_pc);
 640   str(scratch, Address(rthread,
 641                        JavaThread::frame_anchor_offset()
 642                        + JavaFrameAnchor::last_Java_pc_offset()));
 643 
 644   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 645 }
 646 
 647 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 648                                          Register last_java_fp,
 649                                          Label &L,
 650                                          Register scratch) {
 651   if (L.is_bound()) {
 652     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 653   } else {
 654     InstructionMark im(this);
 655     L.add_patch_at(code(), locator());
 656     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 657   }
 658 }
 659 
 660 static inline bool target_needs_far_branch(address addr) {
 661   if (AOTCodeCache::is_on_for_dump()) {
 662     return true;
 663   }
 664   // codecache size <= 128M
 665   if (!MacroAssembler::far_branches()) {
 666     return false;
 667   }
 668   // codecache size > 240M
 669   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 670     return true;
 671   }
 672   // codecache size: 128M..240M
 673   return !CodeCache::is_non_nmethod(addr);
 674 }
 675 
 676 void MacroAssembler::far_call(Address entry, Register tmp) {
 677   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 678   assert(CodeCache::find_blob(entry.target()) != nullptr,
 679          "destination of far call not found in code cache");
 680   assert(entry.rspec().type() == relocInfo::external_word_type
 681          || entry.rspec().type() == relocInfo::runtime_call_type
 682          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 683   if (target_needs_far_branch(entry.target())) {
 684     uint64_t offset;
 685     // We can use ADRP here because we know that the total size of
 686     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 687     adrp(tmp, entry, offset);
 688     add(tmp, tmp, offset);
 689     blr(tmp);
 690   } else {
 691     bl(entry);
 692   }
 693 }
 694 
 695 int MacroAssembler::far_jump(Address entry, Register tmp) {
 696   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 697   assert(CodeCache::find_blob(entry.target()) != nullptr,
 698          "destination of far call not found in code cache");
 699   assert(entry.rspec().type() == relocInfo::external_word_type
 700          || entry.rspec().type() == relocInfo::runtime_call_type
 701          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 702   address start = pc();
 703   if (target_needs_far_branch(entry.target())) {
 704     uint64_t offset;
 705     // We can use ADRP here because we know that the total size of
 706     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 707     adrp(tmp, entry, offset);
 708     add(tmp, tmp, offset);
 709     br(tmp);
 710   } else {
 711     b(entry);
 712   }
 713   return pc() - start;
 714 }
 715 
 716 void MacroAssembler::reserved_stack_check() {
 717     // testing if reserved zone needs to be enabled
 718     Label no_reserved_zone_enabling;
 719 
 720     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 721     cmp(sp, rscratch1);
 722     br(Assembler::LO, no_reserved_zone_enabling);
 723 
 724     enter();   // LR and FP are live.
 725     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 726     mov(c_rarg0, rthread);
 727     blr(rscratch1);
 728     leave();
 729 
 730     // We have already removed our own frame.
 731     // throw_delayed_StackOverflowError will think that it's been
 732     // called by our caller.
 733     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 734     br(rscratch1);
 735     should_not_reach_here();
 736 
 737     bind(no_reserved_zone_enabling);
 738 }
 739 
 740 static void pass_arg0(MacroAssembler* masm, Register arg) {
 741   if (c_rarg0 != arg ) {
 742     masm->mov(c_rarg0, arg);
 743   }
 744 }
 745 
 746 static void pass_arg1(MacroAssembler* masm, Register arg) {
 747   if (c_rarg1 != arg ) {
 748     masm->mov(c_rarg1, arg);
 749   }
 750 }
 751 
 752 static void pass_arg2(MacroAssembler* masm, Register arg) {
 753   if (c_rarg2 != arg ) {
 754     masm->mov(c_rarg2, arg);
 755   }
 756 }
 757 
 758 static void pass_arg3(MacroAssembler* masm, Register arg) {
 759   if (c_rarg3 != arg ) {
 760     masm->mov(c_rarg3, arg);
 761   }
 762 }
 763 
 764 void MacroAssembler::call_VM_base(Register oop_result,
 765                                   Register java_thread,
 766                                   Register last_java_sp,
 767                                   Label*   return_pc,
 768                                   address  entry_point,
 769                                   int      number_of_arguments,
 770                                   bool     check_exceptions) {
 771    // determine java_thread register
 772   if (!java_thread->is_valid()) {
 773     java_thread = rthread;
 774   }
 775 
 776   // determine last_java_sp register
 777   if (!last_java_sp->is_valid()) {
 778     last_java_sp = esp;
 779   }
 780 
 781   // debugging support
 782   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 783   assert(java_thread == rthread, "unexpected register");
 784 #ifdef ASSERT
 785   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 786   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 787 #endif // ASSERT
 788 
 789   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 790   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 791 
 792   // push java thread (becomes first argument of C function)
 793 
 794   mov(c_rarg0, java_thread);
 795 
 796   // set last Java frame before call
 797   assert(last_java_sp != rfp, "can't use rfp");
 798 
 799   Label l;
 800   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 801 
 802   // do the call, remove parameters
 803   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 804 
 805   // lr could be poisoned with PAC signature during throw_pending_exception
 806   // if it was tail-call optimized by compiler, since lr is not callee-saved
 807   // reload it with proper value
 808   adr(lr, l);
 809 
 810   // reset last Java frame
 811   // Only interpreter should have to clear fp
 812   reset_last_Java_frame(true);
 813 
 814    // C++ interp handles this in the interpreter
 815   check_and_handle_popframe(java_thread);
 816   check_and_handle_earlyret(java_thread);
 817 
 818   if (check_exceptions) {
 819     // check for pending exceptions (java_thread is set upon return)
 820     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 821     Label ok;
 822     cbz(rscratch1, ok);
 823     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 824     br(rscratch1);
 825     bind(ok);
 826   }
 827 
 828   // get oop result if there is one and reset the value in the thread
 829   if (oop_result->is_valid()) {
 830     get_vm_result_oop(oop_result, java_thread);
 831   }
 832 }
 833 
 834 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 835   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 836 }
 837 
 838 // Check the entry target is always reachable from any branch.
 839 static bool is_always_within_branch_range(Address entry) {
 840   if (AOTCodeCache::is_on_for_dump()) {
 841     return false;
 842   }
 843   const address target = entry.target();
 844 
 845   if (!CodeCache::contains(target)) {
 846     // We always use trampolines for callees outside CodeCache.
 847     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 848     return false;
 849   }
 850 
 851   if (!MacroAssembler::far_branches()) {
 852     return true;
 853   }
 854 
 855   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 856     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 857     // Non-compiled methods stay forever in CodeCache.
 858     // We check whether the longest possible branch is within the branch range.
 859     assert(CodeCache::find_blob(target) != nullptr &&
 860           !CodeCache::find_blob(target)->is_nmethod(),
 861           "runtime call of compiled method");
 862     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 863     const address left_longest_branch_start = CodeCache::low_bound();
 864     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 865                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 866     return is_reachable;
 867   }
 868 
 869   return false;
 870 }
 871 
 872 // Maybe emit a call via a trampoline. If the code cache is small
 873 // trampolines won't be emitted.
 874 address MacroAssembler::trampoline_call(Address entry) {
 875   assert(entry.rspec().type() == relocInfo::runtime_call_type
 876          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 877          || entry.rspec().type() == relocInfo::static_call_type
 878          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 879 
 880   address target = entry.target();
 881 
 882   if (!is_always_within_branch_range(entry)) {
 883     if (!in_scratch_emit_size()) {
 884       // We don't want to emit a trampoline if C2 is generating dummy
 885       // code during its branch shortening phase.
 886       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 887         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 888         code()->share_trampoline_for(entry.target(), offset());
 889       } else {
 890         address stub = emit_trampoline_stub(offset(), target);
 891         if (stub == nullptr) {
 892           postcond(pc() == badAddress);
 893           return nullptr; // CodeCache is full
 894         }
 895       }
 896     }
 897     target = pc();
 898   }
 899 
 900   address call_pc = pc();
 901   relocate(entry.rspec());
 902   bl(target);
 903 
 904   postcond(pc() != badAddress);
 905   return call_pc;
 906 }
 907 
 908 // Emit a trampoline stub for a call to a target which is too far away.
 909 //
 910 // code sequences:
 911 //
 912 // call-site:
 913 //   branch-and-link to <destination> or <trampoline stub>
 914 //
 915 // Related trampoline stub for this call site in the stub section:
 916 //   load the call target from the constant pool
 917 //   branch (LR still points to the call site above)
 918 
 919 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 920                                              address dest) {
 921   // Max stub size: alignment nop, TrampolineStub.
 922   address stub = start_a_stub(max_trampoline_stub_size());
 923   if (stub == nullptr) {
 924     return nullptr;  // CodeBuffer::expand failed
 925   }
 926 
 927   // Create a trampoline stub relocation which relates this trampoline stub
 928   // with the call instruction at insts_call_instruction_offset in the
 929   // instructions code-section.
 930   align(wordSize);
 931   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 932                                             + insts_call_instruction_offset));
 933   const int stub_start_offset = offset();
 934 
 935   // Now, create the trampoline stub's code:
 936   // - load the call
 937   // - call
 938   Label target;
 939   ldr(rscratch1, target);
 940   br(rscratch1);
 941   bind(target);
 942   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 943          "should be");
 944   emit_int64((int64_t)dest);
 945 
 946   const address stub_start_addr = addr_at(stub_start_offset);
 947 
 948   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 949 
 950   end_a_stub();
 951   return stub_start_addr;
 952 }
 953 
 954 int MacroAssembler::max_trampoline_stub_size() {
 955   // Max stub size: alignment nop, TrampolineStub.
 956   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 957 }
 958 
 959 void MacroAssembler::emit_static_call_stub() {
 960   // CompiledDirectCall::set_to_interpreted knows the
 961   // exact layout of this stub.
 962 
 963   isb();
 964   mov_metadata(rmethod, nullptr);
 965 
 966   // Jump to the entry point of the c2i stub.
 967   if (codestub_branch_needs_far_jump()) {
 968     movptr(rscratch1, 0);
 969     br(rscratch1);
 970   } else {
 971     b(pc());
 972   }
 973 }
 974 
 975 int MacroAssembler::max_static_call_stub_size() {
 976   // isb; movk; movz; movz; movk; movz; movz; br
 977   return 8 * NativeInstruction::instruction_size;
 978 }
 979 
 980 void MacroAssembler::c2bool(Register x) {
 981   // implements x == 0 ? 0 : 1
 982   // note: must only look at least-significant byte of x
 983   //       since C-style booleans are stored in one byte
 984   //       only! (was bug)
 985   tst(x, 0xff);
 986   cset(x, Assembler::NE);
 987 }
 988 
 989 address MacroAssembler::ic_call(address entry, jint method_index) {
 990   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 991   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 992   return trampoline_call(Address(entry, rh));
 993 }
 994 
 995 int MacroAssembler::ic_check_size() {
 996   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 997   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 998     return NativeInstruction::instruction_size * (7 + extra_instructions);
 999   } else {
1000     return NativeInstruction::instruction_size * (5 + extra_instructions);
1001   }
1002 }
1003 
1004 int MacroAssembler::ic_check(int end_alignment) {
1005   Register receiver = j_rarg0;
1006   Register data = rscratch2;
1007   Register tmp1 = rscratch1;
1008   Register tmp2 = r10;
1009 
1010   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
1011   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
1012   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1013   // before the inline cache check here, and not after
1014   align(end_alignment, offset() + ic_check_size());
1015 
1016   int uep_offset = offset();
1017 
1018   if (UseCompactObjectHeaders) {
1019     load_narrow_klass_compact(tmp1, receiver);
1020     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1021     cmpw(tmp1, tmp2);
1022   } else if (UseCompressedClassPointers) {
1023     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1024     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1025     cmpw(tmp1, tmp2);
1026   } else {
1027     ldr(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1028     ldr(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1029     cmp(tmp1, tmp2);
1030   }
1031 
1032   Label dont;
1033   br(Assembler::EQ, dont);
1034   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1035   bind(dont);
1036   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1037 
1038   return uep_offset;
1039 }
1040 
1041 // Implementation of call_VM versions
1042 
1043 void MacroAssembler::call_VM(Register oop_result,
1044                              address entry_point,
1045                              bool check_exceptions) {
1046   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1047 }
1048 
1049 void MacroAssembler::call_VM(Register oop_result,
1050                              address entry_point,
1051                              Register arg_1,
1052                              bool check_exceptions) {
1053   pass_arg1(this, arg_1);
1054   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1055 }
1056 
1057 void MacroAssembler::call_VM(Register oop_result,
1058                              address entry_point,
1059                              Register arg_1,
1060                              Register arg_2,
1061                              bool check_exceptions) {
1062   assert_different_registers(arg_1, c_rarg2);
1063   pass_arg2(this, arg_2);
1064   pass_arg1(this, arg_1);
1065   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1066 }
1067 
1068 void MacroAssembler::call_VM(Register oop_result,
1069                              address entry_point,
1070                              Register arg_1,
1071                              Register arg_2,
1072                              Register arg_3,
1073                              bool check_exceptions) {
1074   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1075   assert_different_registers(arg_2, c_rarg3);
1076   pass_arg3(this, arg_3);
1077 
1078   pass_arg2(this, arg_2);
1079 
1080   pass_arg1(this, arg_1);
1081   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1082 }
1083 
1084 void MacroAssembler::call_VM(Register oop_result,
1085                              Register last_java_sp,
1086                              address entry_point,
1087                              int number_of_arguments,
1088                              bool check_exceptions) {
1089   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1090 }
1091 
1092 void MacroAssembler::call_VM(Register oop_result,
1093                              Register last_java_sp,
1094                              address entry_point,
1095                              Register arg_1,
1096                              bool check_exceptions) {
1097   pass_arg1(this, arg_1);
1098   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1099 }
1100 
1101 void MacroAssembler::call_VM(Register oop_result,
1102                              Register last_java_sp,
1103                              address entry_point,
1104                              Register arg_1,
1105                              Register arg_2,
1106                              bool check_exceptions) {
1107 
1108   assert_different_registers(arg_1, c_rarg2);
1109   pass_arg2(this, arg_2);
1110   pass_arg1(this, arg_1);
1111   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1112 }
1113 
1114 void MacroAssembler::call_VM(Register oop_result,
1115                              Register last_java_sp,
1116                              address entry_point,
1117                              Register arg_1,
1118                              Register arg_2,
1119                              Register arg_3,
1120                              bool check_exceptions) {
1121   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1122   assert_different_registers(arg_2, c_rarg3);
1123   pass_arg3(this, arg_3);
1124   pass_arg2(this, arg_2);
1125   pass_arg1(this, arg_1);
1126   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1127 }
1128 
1129 
1130 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1131   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1132   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1133   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1134 }
1135 
1136 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1137   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1138   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1139 }
1140 
1141 void MacroAssembler::align(int modulus) {
1142   align(modulus, offset());
1143 }
1144 
1145 // Ensure that the code at target bytes offset from the current offset() is aligned
1146 // according to modulus.
1147 void MacroAssembler::align(int modulus, int target) {
1148   int delta = target - offset();
1149   while ((offset() + delta) % modulus != 0) nop();
1150 }
1151 
1152 void MacroAssembler::post_call_nop() {
1153   if (!Continuations::enabled()) {
1154     return;
1155   }
1156   InstructionMark im(this);
1157   relocate(post_call_nop_Relocation::spec());
1158   InlineSkippedInstructionsCounter skipCounter(this);
1159   nop();
1160   movk(zr, 0);
1161   movk(zr, 0);
1162 }
1163 
1164 // these are no-ops overridden by InterpreterMacroAssembler
1165 
1166 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1167 
1168 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1169 
1170 // Look up the method for a megamorphic invokeinterface call.
1171 // The target method is determined by <intf_klass, itable_index>.
1172 // The receiver klass is in recv_klass.
1173 // On success, the result will be in method_result, and execution falls through.
1174 // On failure, execution transfers to the given label.
1175 void MacroAssembler::lookup_interface_method(Register recv_klass,
1176                                              Register intf_klass,
1177                                              RegisterOrConstant itable_index,
1178                                              Register method_result,
1179                                              Register scan_temp,
1180                                              Label& L_no_such_interface,
1181                          bool return_method) {
1182   assert_different_registers(recv_klass, intf_klass, scan_temp);
1183   assert_different_registers(method_result, intf_klass, scan_temp);
1184   assert(recv_klass != method_result || !return_method,
1185      "recv_klass can be destroyed when method isn't needed");
1186   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1187          "caller must use same register for non-constant itable index as for method");
1188 
1189   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1190   int vtable_base = in_bytes(Klass::vtable_start_offset());
1191   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1192   int scan_step   = itableOffsetEntry::size() * wordSize;
1193   int vte_size    = vtableEntry::size_in_bytes();
1194   assert(vte_size == wordSize, "else adjust times_vte_scale");
1195 
1196   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1197 
1198   // Could store the aligned, prescaled offset in the klass.
1199   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1200   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1201   add(scan_temp, scan_temp, vtable_base);
1202 
1203   if (return_method) {
1204     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1205     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1206     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1207     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1208     if (itentry_off)
1209       add(recv_klass, recv_klass, itentry_off);
1210   }
1211 
1212   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1213   //   if (scan->interface() == intf) {
1214   //     result = (klass + scan->offset() + itable_index);
1215   //   }
1216   // }
1217   Label search, found_method;
1218 
1219   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1220   cmp(intf_klass, method_result);
1221   br(Assembler::EQ, found_method);
1222   bind(search);
1223   // Check that the previous entry is non-null.  A null entry means that
1224   // the receiver class doesn't implement the interface, and wasn't the
1225   // same as when the caller was compiled.
1226   cbz(method_result, L_no_such_interface);
1227   if (itableOffsetEntry::interface_offset() != 0) {
1228     add(scan_temp, scan_temp, scan_step);
1229     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1230   } else {
1231     ldr(method_result, Address(pre(scan_temp, scan_step)));
1232   }
1233   cmp(intf_klass, method_result);
1234   br(Assembler::NE, search);
1235 
1236   bind(found_method);
1237 
1238   // Got a hit.
1239   if (return_method) {
1240     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1241     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1242   }
1243 }
1244 
1245 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1246 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1247 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1248 // The target method is determined by <holder_klass, itable_index>.
1249 // The receiver klass is in recv_klass.
1250 // On success, the result will be in method_result, and execution falls through.
1251 // On failure, execution transfers to the given label.
1252 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1253                                                   Register holder_klass,
1254                                                   Register resolved_klass,
1255                                                   Register method_result,
1256                                                   Register temp_itbl_klass,
1257                                                   Register scan_temp,
1258                                                   int itable_index,
1259                                                   Label& L_no_such_interface) {
1260   // 'method_result' is only used as output register at the very end of this method.
1261   // Until then we can reuse it as 'holder_offset'.
1262   Register holder_offset = method_result;
1263   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1264 
1265   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1266   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1267   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1268   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1269 
1270   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1271 
1272   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1273   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1274   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1275   // temp_itbl_klass = itable[0]._interface;
1276   int vtblEntrySize = vtableEntry::size_in_bytes();
1277   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1278   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1279   mov(holder_offset, zr);
1280   // scan_temp = &(itable[0]._interface)
1281   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1282 
1283   // Initial checks:
1284   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1285   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1286   //   - if (itable[0] == 0), no such interface
1287   cmp(resolved_klass, holder_klass);
1288   br(Assembler::NE, L_loop_search_resolved_entry);
1289   cmp(holder_klass, temp_itbl_klass);
1290   br(Assembler::EQ, L_holder_found);
1291   cbz(temp_itbl_klass, L_no_such_interface);
1292 
1293   // Loop: Look for holder_klass record in itable
1294   //   do {
1295   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1296   //     if (temp_itbl_klass == holder_klass) {
1297   //       goto L_holder_found; // Found!
1298   //     }
1299   //   } while (temp_itbl_klass != 0);
1300   //   goto L_no_such_interface // Not found.
1301   Label L_search_holder;
1302   bind(L_search_holder);
1303     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1304     cmp(holder_klass, temp_itbl_klass);
1305     br(Assembler::EQ, L_holder_found);
1306     cbnz(temp_itbl_klass, L_search_holder);
1307 
1308   b(L_no_such_interface);
1309 
1310   // Loop: Look for resolved_class record in itable
1311   //   while (true) {
1312   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1313   //     if (temp_itbl_klass == 0) {
1314   //       goto L_no_such_interface;
1315   //     }
1316   //     if (temp_itbl_klass == resolved_klass) {
1317   //        goto L_resolved_found;  // Found!
1318   //     }
1319   //     if (temp_itbl_klass == holder_klass) {
1320   //        holder_offset = scan_temp;
1321   //     }
1322   //   }
1323   //
1324   Label L_loop_search_resolved;
1325   bind(L_loop_search_resolved);
1326     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1327   bind(L_loop_search_resolved_entry);
1328     cbz(temp_itbl_klass, L_no_such_interface);
1329     cmp(resolved_klass, temp_itbl_klass);
1330     br(Assembler::EQ, L_resolved_found);
1331     cmp(holder_klass, temp_itbl_klass);
1332     br(Assembler::NE, L_loop_search_resolved);
1333     mov(holder_offset, scan_temp);
1334     b(L_loop_search_resolved);
1335 
1336   // See if we already have a holder klass. If not, go and scan for it.
1337   bind(L_resolved_found);
1338   cbz(holder_offset, L_search_holder);
1339   mov(scan_temp, holder_offset);
1340 
1341   // Finally, scan_temp contains holder_klass vtable offset
1342   bind(L_holder_found);
1343   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1344   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1345     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1346   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1347 }
1348 
1349 // virtual method calling
1350 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1351                                            RegisterOrConstant vtable_index,
1352                                            Register method_result) {
1353   assert(vtableEntry::size() * wordSize == 8,
1354          "adjust the scaling in the code below");
1355   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1356 
1357   if (vtable_index.is_register()) {
1358     lea(method_result, Address(recv_klass,
1359                                vtable_index.as_register(),
1360                                Address::lsl(LogBytesPerWord)));
1361     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1362   } else {
1363     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1364     ldr(method_result,
1365         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1366   }
1367 }
1368 
1369 void MacroAssembler::check_klass_subtype(Register sub_klass,
1370                            Register super_klass,
1371                            Register temp_reg,
1372                            Label& L_success) {
1373   Label L_failure;
1374   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1375   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1376   bind(L_failure);
1377 }
1378 
1379 
1380 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1381                                                    Register super_klass,
1382                                                    Register temp_reg,
1383                                                    Label* L_success,
1384                                                    Label* L_failure,
1385                                                    Label* L_slow_path,
1386                                                    Register super_check_offset) {
1387   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1388   bool must_load_sco = ! super_check_offset->is_valid();
1389   if (must_load_sco) {
1390     assert(temp_reg != noreg, "supply either a temp or a register offset");
1391   }
1392 
1393   Label L_fallthrough;
1394   int label_nulls = 0;
1395   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1396   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1397   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1398   assert(label_nulls <= 1, "at most one null in the batch");
1399 
1400   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1401   Address super_check_offset_addr(super_klass, sco_offset);
1402 
1403   // Hacked jmp, which may only be used just before L_fallthrough.
1404 #define final_jmp(label)                                                \
1405   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1406   else                            b(label)                /*omit semi*/
1407 
1408   // If the pointers are equal, we are done (e.g., String[] elements).
1409   // This self-check enables sharing of secondary supertype arrays among
1410   // non-primary types such as array-of-interface.  Otherwise, each such
1411   // type would need its own customized SSA.
1412   // We move this check to the front of the fast path because many
1413   // type checks are in fact trivially successful in this manner,
1414   // so we get a nicely predicted branch right at the start of the check.
1415   cmp(sub_klass, super_klass);
1416   br(Assembler::EQ, *L_success);
1417 
1418   // Check the supertype display:
1419   if (must_load_sco) {
1420     ldrw(temp_reg, super_check_offset_addr);
1421     super_check_offset = temp_reg;
1422   }
1423 
1424   Address super_check_addr(sub_klass, super_check_offset);
1425   ldr(rscratch1, super_check_addr);
1426   cmp(super_klass, rscratch1); // load displayed supertype
1427   br(Assembler::EQ, *L_success);
1428 
1429   // This check has worked decisively for primary supers.
1430   // Secondary supers are sought in the super_cache ('super_cache_addr').
1431   // (Secondary supers are interfaces and very deeply nested subtypes.)
1432   // This works in the same check above because of a tricky aliasing
1433   // between the super_cache and the primary super display elements.
1434   // (The 'super_check_addr' can address either, as the case requires.)
1435   // Note that the cache is updated below if it does not help us find
1436   // what we need immediately.
1437   // So if it was a primary super, we can just fail immediately.
1438   // Otherwise, it's the slow path for us (no success at this point).
1439 
1440   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1441   if (L_failure == &L_fallthrough) {
1442     cbz(rscratch1, *L_slow_path);
1443   } else {
1444     cbnz(rscratch1, *L_failure);
1445     final_jmp(*L_slow_path);
1446   }
1447 
1448   bind(L_fallthrough);
1449 
1450 #undef final_jmp
1451 }
1452 
1453 // These two are taken from x86, but they look generally useful
1454 
1455 // scans count pointer sized words at [addr] for occurrence of value,
1456 // generic
1457 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1458                                 Register scratch) {
1459   Label Lloop, Lexit;
1460   cbz(count, Lexit);
1461   bind(Lloop);
1462   ldr(scratch, post(addr, wordSize));
1463   cmp(value, scratch);
1464   br(EQ, Lexit);
1465   sub(count, count, 1);
1466   cbnz(count, Lloop);
1467   bind(Lexit);
1468 }
1469 
1470 // scans count 4 byte words at [addr] for occurrence of value,
1471 // generic
1472 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1473                                 Register scratch) {
1474   Label Lloop, Lexit;
1475   cbz(count, Lexit);
1476   bind(Lloop);
1477   ldrw(scratch, post(addr, wordSize));
1478   cmpw(value, scratch);
1479   br(EQ, Lexit);
1480   sub(count, count, 1);
1481   cbnz(count, Lloop);
1482   bind(Lexit);
1483 }
1484 
1485 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1486                                                           Register super_klass,
1487                                                           Register temp_reg,
1488                                                           Register temp2_reg,
1489                                                           Label* L_success,
1490                                                           Label* L_failure,
1491                                                           bool set_cond_codes) {
1492   // NB! Callers may assume that, when temp2_reg is a valid register,
1493   // this code sets it to a nonzero value.
1494 
1495   assert_different_registers(sub_klass, super_klass, temp_reg);
1496   if (temp2_reg != noreg)
1497     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1498 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1499 
1500   Label L_fallthrough;
1501   int label_nulls = 0;
1502   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1503   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1504   assert(label_nulls <= 1, "at most one null in the batch");
1505 
1506   // a couple of useful fields in sub_klass:
1507   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1508   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1509   Address secondary_supers_addr(sub_klass, ss_offset);
1510   Address super_cache_addr(     sub_klass, sc_offset);
1511 
1512   BLOCK_COMMENT("check_klass_subtype_slow_path");
1513 
1514   // Do a linear scan of the secondary super-klass chain.
1515   // This code is rarely used, so simplicity is a virtue here.
1516   // The repne_scan instruction uses fixed registers, which we must spill.
1517   // Don't worry too much about pre-existing connections with the input regs.
1518 
1519   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1520   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1521 
1522   RegSet pushed_registers;
1523   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1524   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1525 
1526   if (super_klass != r0) {
1527     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1528   }
1529 
1530   push(pushed_registers, sp);
1531 
1532   // Get super_klass value into r0 (even if it was in r5 or r2).
1533   if (super_klass != r0) {
1534     mov(r0, super_klass);
1535   }
1536 
1537 #ifndef PRODUCT
1538   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1539 #endif //PRODUCT
1540 
1541   // We will consult the secondary-super array.
1542   ldr(r5, secondary_supers_addr);
1543   // Load the array length.
1544   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1545   // Skip to start of data.
1546   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1547 
1548   cmp(sp, zr); // Clear Z flag; SP is never zero
1549   // Scan R2 words at [R5] for an occurrence of R0.
1550   // Set NZ/Z based on last compare.
1551   repne_scan(r5, r0, r2, rscratch1);
1552 
1553   // Unspill the temp. registers:
1554   pop(pushed_registers, sp);
1555 
1556   br(Assembler::NE, *L_failure);
1557 
1558   // Success.  Cache the super we found and proceed in triumph.
1559 
1560   if (UseSecondarySupersCache) {
1561     str(super_klass, super_cache_addr);
1562   }
1563 
1564   if (L_success != &L_fallthrough) {
1565     b(*L_success);
1566   }
1567 
1568 #undef IS_A_TEMP
1569 
1570   bind(L_fallthrough);
1571 }
1572 
1573 // If Register r is invalid, remove a new register from
1574 // available_regs, and add new register to regs_to_push.
1575 Register MacroAssembler::allocate_if_noreg(Register r,
1576                                   RegSetIterator<Register> &available_regs,
1577                                   RegSet &regs_to_push) {
1578   if (!r->is_valid()) {
1579     r = *available_regs++;
1580     regs_to_push += r;
1581   }
1582   return r;
1583 }
1584 
1585 // check_klass_subtype_slow_path_table() looks for super_klass in the
1586 // hash table belonging to super_klass, branching to L_success or
1587 // L_failure as appropriate. This is essentially a shim which
1588 // allocates registers as necessary then calls
1589 // lookup_secondary_supers_table() to do the work. Any of the temp
1590 // regs may be noreg, in which case this logic will chooses some
1591 // registers push and pop them from the stack.
1592 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1593                                                          Register super_klass,
1594                                                          Register temp_reg,
1595                                                          Register temp2_reg,
1596                                                          Register temp3_reg,
1597                                                          Register result_reg,
1598                                                          FloatRegister vtemp,
1599                                                          Label* L_success,
1600                                                          Label* L_failure,
1601                                                          bool set_cond_codes) {
1602   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1603 
1604   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1605 
1606   Label L_fallthrough;
1607   int label_nulls = 0;
1608   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1609   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1610   assert(label_nulls <= 1, "at most one null in the batch");
1611 
1612   BLOCK_COMMENT("check_klass_subtype_slow_path");
1613 
1614   RegSetIterator<Register> available_regs
1615     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1616 
1617   RegSet pushed_regs;
1618 
1619   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1620   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1621   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1622   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1623 
1624   push(pushed_regs, sp);
1625 
1626   lookup_secondary_supers_table_var(sub_klass,
1627                                     super_klass,
1628                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1629                                     nullptr);
1630   cmp(result_reg, zr);
1631 
1632   // Unspill the temp. registers:
1633   pop(pushed_regs, sp);
1634 
1635   // NB! Callers may assume that, when set_cond_codes is true, this
1636   // code sets temp2_reg to a nonzero value.
1637   if (set_cond_codes) {
1638     mov(temp2_reg, 1);
1639   }
1640 
1641   br(Assembler::NE, *L_failure);
1642 
1643   if (L_success != &L_fallthrough) {
1644     b(*L_success);
1645   }
1646 
1647   bind(L_fallthrough);
1648 }
1649 
1650 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1651                                                    Register super_klass,
1652                                                    Register temp_reg,
1653                                                    Register temp2_reg,
1654                                                    Label* L_success,
1655                                                    Label* L_failure,
1656                                                    bool set_cond_codes) {
1657   if (UseSecondarySupersTable) {
1658     check_klass_subtype_slow_path_table
1659       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1660        /*vtemp*/fnoreg,
1661        L_success, L_failure, set_cond_codes);
1662   } else {
1663     check_klass_subtype_slow_path_linear
1664       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1665   }
1666 }
1667 
1668 
1669 // Ensure that the inline code and the stub are using the same registers.
1670 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1671 do {                                                               \
1672   assert(r_super_klass  == r0                                   && \
1673          r_array_base   == r1                                   && \
1674          r_array_length == r2                                   && \
1675          (r_array_index == r3        || r_array_index == noreg) && \
1676          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1677          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1678          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1679 } while(0)
1680 
1681 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1682                                                          Register r_super_klass,
1683                                                          Register temp1,
1684                                                          Register temp2,
1685                                                          Register temp3,
1686                                                          FloatRegister vtemp,
1687                                                          Register result,
1688                                                          u1 super_klass_slot,
1689                                                          bool stub_is_near) {
1690   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1691 
1692   Label L_fallthrough;
1693 
1694   BLOCK_COMMENT("lookup_secondary_supers_table {");
1695 
1696   const Register
1697     r_array_base   = temp1, // r1
1698     r_array_length = temp2, // r2
1699     r_array_index  = temp3, // r3
1700     r_bitmap       = rscratch2;
1701 
1702   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1703 
1704   u1 bit = super_klass_slot;
1705 
1706   // Make sure that result is nonzero if the TBZ below misses.
1707   mov(result, 1);
1708 
1709   // We're going to need the bitmap in a vector reg and in a core reg,
1710   // so load both now.
1711   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1712   if (bit != 0) {
1713     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1714   }
1715   // First check the bitmap to see if super_klass might be present. If
1716   // the bit is zero, we are certain that super_klass is not one of
1717   // the secondary supers.
1718   tbz(r_bitmap, bit, L_fallthrough);
1719 
1720   // Get the first array index that can contain super_klass into r_array_index.
1721   if (bit != 0) {
1722     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1723     cnt(vtemp, T8B, vtemp);
1724     addv(vtemp, T8B, vtemp);
1725     fmovd(r_array_index, vtemp);
1726   } else {
1727     mov(r_array_index, (u1)1);
1728   }
1729   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1730 
1731   // We will consult the secondary-super array.
1732   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1733 
1734   // The value i in r_array_index is >= 1, so even though r_array_base
1735   // points to the length, we don't need to adjust it to point to the
1736   // data.
1737   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1738   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1739 
1740   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1741   eor(result, result, r_super_klass);
1742   cbz(result, L_fallthrough); // Found a match
1743 
1744   // Is there another entry to check? Consult the bitmap.
1745   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1746 
1747   // Linear probe.
1748   if (bit != 0) {
1749     ror(r_bitmap, r_bitmap, bit);
1750   }
1751 
1752   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1753   // The next slot to be inspected, by the stub we're about to call,
1754   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1755   // have been checked.
1756   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1757   if (stub_is_near) {
1758     bl(stub);
1759   } else {
1760     address call = trampoline_call(stub);
1761     if (call == nullptr) {
1762       return false; // trampoline allocation failed
1763     }
1764   }
1765 
1766   BLOCK_COMMENT("} lookup_secondary_supers_table");
1767 
1768   bind(L_fallthrough);
1769 
1770   if (VerifySecondarySupers) {
1771     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1772                                   temp1, temp2, result);      // r1, r2, r5
1773   }
1774   return true;
1775 }
1776 
1777 // At runtime, return 0 in result if r_super_klass is a superclass of
1778 // r_sub_klass, otherwise return nonzero. Use this version of
1779 // lookup_secondary_supers_table() if you don't know ahead of time
1780 // which superclass will be searched for. Used by interpreter and
1781 // runtime stubs. It is larger and has somewhat greater latency than
1782 // the version above, which takes a constant super_klass_slot.
1783 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1784                                                        Register r_super_klass,
1785                                                        Register temp1,
1786                                                        Register temp2,
1787                                                        Register temp3,
1788                                                        FloatRegister vtemp,
1789                                                        Register result,
1790                                                        Label *L_success) {
1791   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1792 
1793   Label L_fallthrough;
1794 
1795   BLOCK_COMMENT("lookup_secondary_supers_table {");
1796 
1797   const Register
1798     r_array_index = temp3,
1799     slot          = rscratch1,
1800     r_bitmap      = rscratch2;
1801 
1802   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1803 
1804   // Make sure that result is nonzero if the test below misses.
1805   mov(result, 1);
1806 
1807   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1808 
1809   // First check the bitmap to see if super_klass might be present. If
1810   // the bit is zero, we are certain that super_klass is not one of
1811   // the secondary supers.
1812 
1813   // This next instruction is equivalent to:
1814   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1815   // sub(temp2, tmp_reg, slot);
1816   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1817   lslv(temp2, r_bitmap, temp2);
1818   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1819 
1820   bool must_save_v0 = (vtemp == fnoreg);
1821   if (must_save_v0) {
1822     // temp1 and result are free, so use them to preserve vtemp
1823     vtemp = v0;
1824     mov(temp1,  vtemp, D, 0);
1825     mov(result, vtemp, D, 1);
1826   }
1827 
1828   // Get the first array index that can contain super_klass into r_array_index.
1829   mov(vtemp, D, 0, temp2);
1830   cnt(vtemp, T8B, vtemp);
1831   addv(vtemp, T8B, vtemp);
1832   mov(r_array_index, vtemp, D, 0);
1833 
1834   if (must_save_v0) {
1835     mov(vtemp, D, 0, temp1 );
1836     mov(vtemp, D, 1, result);
1837   }
1838 
1839   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1840 
1841   const Register
1842     r_array_base   = temp1,
1843     r_array_length = temp2;
1844 
1845   // The value i in r_array_index is >= 1, so even though r_array_base
1846   // points to the length, we don't need to adjust it to point to the
1847   // data.
1848   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1849   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1850 
1851   // We will consult the secondary-super array.
1852   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1853 
1854   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1855   eor(result, result, r_super_klass);
1856   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1857 
1858   // Is there another entry to check? Consult the bitmap.
1859   rorv(r_bitmap, r_bitmap, slot);
1860   // rol(r_bitmap, r_bitmap, 1);
1861   tbz(r_bitmap, 1, L_fallthrough);
1862 
1863   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1864   // The next slot to be inspected, by the logic we're about to call,
1865   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1866   // have been checked.
1867   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1868                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1869 
1870   BLOCK_COMMENT("} lookup_secondary_supers_table");
1871 
1872   bind(L_fallthrough);
1873 
1874   if (VerifySecondarySupers) {
1875     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1876                                   temp1, temp2, result);      // r1, r2, r5
1877   }
1878 
1879   if (L_success) {
1880     cbz(result, *L_success);
1881   }
1882 }
1883 
1884 // Called by code generated by check_klass_subtype_slow_path
1885 // above. This is called when there is a collision in the hashed
1886 // lookup in the secondary supers array.
1887 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1888                                                              Register r_array_base,
1889                                                              Register r_array_index,
1890                                                              Register r_bitmap,
1891                                                              Register temp1,
1892                                                              Register result,
1893                                                              bool is_stub) {
1894   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1895 
1896   const Register
1897     r_array_length = temp1,
1898     r_sub_klass    = noreg; // unused
1899 
1900   if (is_stub) {
1901     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1902   }
1903 
1904   Label L_fallthrough, L_huge;
1905 
1906   // Load the array length.
1907   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1908   // And adjust the array base to point to the data.
1909   // NB! Effectively increments current slot index by 1.
1910   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1911   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1912 
1913   // The bitmap is full to bursting.
1914   // Implicit invariant: BITMAP_FULL implies (length > 0)
1915   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1916   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1917   br(GT, L_huge);
1918 
1919   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1920   // current slot (at secondary_supers[r_array_index]) has not yet
1921   // been inspected, and r_array_index may be out of bounds if we
1922   // wrapped around the end of the array.
1923 
1924   { // This is conventional linear probing, but instead of terminating
1925     // when a null entry is found in the table, we maintain a bitmap
1926     // in which a 0 indicates missing entries.
1927     // As long as the bitmap is not completely full,
1928     // array_length == popcount(bitmap). The array_length check above
1929     // guarantees there are 0s in the bitmap, so the loop eventually
1930     // terminates.
1931     Label L_loop;
1932     bind(L_loop);
1933 
1934     // Check for wraparound.
1935     cmp(r_array_index, r_array_length);
1936     csel(r_array_index, zr, r_array_index, GE);
1937 
1938     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1939     eor(result, rscratch1, r_super_klass);
1940     cbz(result, L_fallthrough);
1941 
1942     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1943 
1944     ror(r_bitmap, r_bitmap, 1);
1945     add(r_array_index, r_array_index, 1);
1946     b(L_loop);
1947   }
1948 
1949   { // Degenerate case: more than 64 secondary supers.
1950     // FIXME: We could do something smarter here, maybe a vectorized
1951     // comparison or a binary search, but is that worth any added
1952     // complexity?
1953     bind(L_huge);
1954     cmp(sp, zr); // Clear Z flag; SP is never zero
1955     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1956     cset(result, NE); // result == 0 iff we got a match.
1957   }
1958 
1959   bind(L_fallthrough);
1960 }
1961 
1962 // Make sure that the hashed lookup and a linear scan agree.
1963 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1964                                                    Register r_super_klass,
1965                                                    Register temp1,
1966                                                    Register temp2,
1967                                                    Register result) {
1968   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1969 
1970   const Register
1971     r_array_base   = temp1,
1972     r_array_length = temp2;
1973 
1974   BLOCK_COMMENT("verify_secondary_supers_table {");
1975 
1976   // We will consult the secondary-super array.
1977   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1978 
1979   // Load the array length.
1980   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1981   // And adjust the array base to point to the data.
1982   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1983 
1984   cmp(sp, zr); // Clear Z flag; SP is never zero
1985   // Scan R2 words at [R5] for an occurrence of R0.
1986   // Set NZ/Z based on last compare.
1987   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1988   // rscratch1 == 0 iff we got a match.
1989   cset(rscratch1, NE);
1990 
1991   Label passed;
1992   cmp(result, zr);
1993   cset(result, NE); // normalize result to 0/1 for comparison
1994 
1995   cmp(rscratch1, result);
1996   br(EQ, passed);
1997   {
1998     mov(r0, r_super_klass);         // r0 <- r0
1999     mov(r1, r_sub_klass);           // r1 <- r4
2000     mov(r2, /*expected*/rscratch1); // r2 <- r8
2001     mov(r3, result);                // r3 <- r5
2002     mov(r4, (address)("mismatch")); // r4 <- const
2003     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
2004     should_not_reach_here();
2005   }
2006   bind(passed);
2007 
2008   BLOCK_COMMENT("} verify_secondary_supers_table");
2009 }
2010 
2011 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
2012   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
2013   assert_different_registers(klass, rthread, scratch);
2014 
2015   Label L_fallthrough, L_tmp;
2016   if (L_fast_path == nullptr) {
2017     L_fast_path = &L_fallthrough;
2018   } else if (L_slow_path == nullptr) {
2019     L_slow_path = &L_fallthrough;
2020   }
2021   // Fast path check: class is fully initialized
2022   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2023   ldarb(scratch, scratch);
2024   cmp(scratch, InstanceKlass::fully_initialized);
2025   br(Assembler::EQ, *L_fast_path);
2026 
2027   // Fast path check: current thread is initializer thread
2028   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2029   cmp(rthread, scratch);
2030 
2031   if (L_slow_path == &L_fallthrough) {
2032     br(Assembler::EQ, *L_fast_path);
2033     bind(*L_slow_path);
2034   } else if (L_fast_path == &L_fallthrough) {
2035     br(Assembler::NE, *L_slow_path);
2036     bind(*L_fast_path);
2037   } else {
2038     Unimplemented();
2039   }
2040 }
2041 
2042 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2043   if (!VerifyOops) return;
2044 
2045   // Pass register number to verify_oop_subroutine
2046   const char* b = nullptr;
2047   {
2048     ResourceMark rm;
2049     stringStream ss;
2050     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2051     b = code_string(ss.as_string());
2052   }
2053   BLOCK_COMMENT("verify_oop {");
2054 
2055   strip_return_address(); // This might happen within a stack frame.
2056   protect_return_address();
2057   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2058   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2059 
2060   mov(r0, reg);
2061   movptr(rscratch1, (uintptr_t)(address)b);
2062 
2063   // call indirectly to solve generation ordering problem
2064   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2065   ldr(rscratch2, Address(rscratch2));
2066   blr(rscratch2);
2067 
2068   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2069   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2070   authenticate_return_address();
2071 
2072   BLOCK_COMMENT("} verify_oop");
2073 }
2074 
2075 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2076   if (!VerifyOops) return;
2077 
2078   const char* b = nullptr;
2079   {
2080     ResourceMark rm;
2081     stringStream ss;
2082     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2083     b = code_string(ss.as_string());
2084   }
2085   BLOCK_COMMENT("verify_oop_addr {");
2086 
2087   strip_return_address(); // This might happen within a stack frame.
2088   protect_return_address();
2089   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2090   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2091 
2092   // addr may contain sp so we will have to adjust it based on the
2093   // pushes that we just did.
2094   if (addr.uses(sp)) {
2095     lea(r0, addr);
2096     ldr(r0, Address(r0, 4 * wordSize));
2097   } else {
2098     ldr(r0, addr);
2099   }
2100   movptr(rscratch1, (uintptr_t)(address)b);
2101 
2102   // call indirectly to solve generation ordering problem
2103   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2104   ldr(rscratch2, Address(rscratch2));
2105   blr(rscratch2);
2106 
2107   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2108   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2109   authenticate_return_address();
2110 
2111   BLOCK_COMMENT("} verify_oop_addr");
2112 }
2113 
2114 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2115                                          int extra_slot_offset) {
2116   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2117   int stackElementSize = Interpreter::stackElementSize;
2118   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2119 #ifdef ASSERT
2120   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2121   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2122 #endif
2123   if (arg_slot.is_constant()) {
2124     return Address(esp, arg_slot.as_constant() * stackElementSize
2125                    + offset);
2126   } else {
2127     add(rscratch1, esp, arg_slot.as_register(),
2128         ext::uxtx, exact_log2(stackElementSize));
2129     return Address(rscratch1, offset);
2130   }
2131 }
2132 
2133 // Handle the receiver type profile update given the "recv" klass.
2134 //
2135 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2136 // If there are no matching or claimable receiver entries in RD, updates
2137 // the polymorphic counter.
2138 //
2139 // This code expected to run by either the interpreter or JIT-ed code, without
2140 // extra synchronization. For safety, receiver cells are claimed atomically, which
2141 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2142 // counter updates are not atomic.
2143 //
2144 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2145   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2146 
2147   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2148   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2149   int poly_count_offset      = in_bytes(CounterData::count_offset());
2150   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2151   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2152 
2153   // Adjust for MDP offsets.
2154   base_receiver_offset += mdp_offset;
2155   end_receiver_offset  += mdp_offset;
2156   poly_count_offset    += mdp_offset;
2157 
2158 #ifdef ASSERT
2159   // We are about to walk the MDO slots without asking for offsets.
2160   // Check that our math hits all the right spots.
2161   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2162     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2163     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2164     int offset = base_receiver_offset + receiver_step*c;
2165     int count_offset = offset + receiver_to_count_step;
2166     assert(offset == real_recv_offset, "receiver slot math");
2167     assert(count_offset == real_count_offset, "receiver count math");
2168   }
2169   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2170   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2171 #endif
2172 
2173   // Corner case: no profile table. Increment poly counter and exit.
2174   if (ReceiverTypeData::row_limit() == 0) {
2175     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2176     return;
2177   }
2178 
2179   Register offset = rscratch2;
2180 
2181   Label L_loop_search_receiver, L_loop_search_empty;
2182   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
2183 
2184   // The code here recognizes three major cases:
2185   //   A. Fastest: receiver found in the table
2186   //   B. Fast: no receiver in the table, and the table is full
2187   //   C. Slow: no receiver in the table, free slots in the table
2188   //
2189   // The case A performance is most important, as perfectly-behaved code would end up
2190   // there, especially with larger TypeProfileWidth. The case B performance is
2191   // important as well, this is where bulk of code would land for normally megamorphic
2192   // cases. The case C performance is not essential, its job is to deal with installation
2193   // races, we optimize for code density instead. Case C needs to make sure that receiver
2194   // rows are only claimed once. This makes sure we never overwrite a row for another
2195   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2196   //
2197   // It is very tempting to handle these cases in a single loop, and claim the first slot
2198   // without checking the rest of the table. But, profiling code should tolerate free slots
2199   // in the table, as class unloading can clear them. After such cleanup, the receiver
2200   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2201   // to complete, before trying to install new slots. Splitting the code in several tight
2202   // loops also helpfully optimizes for cases A and B.
2203   //
2204   // This code is effectively:
2205   //
2206   // restart:
2207   //   // Fastest: receiver is already installed
2208   //   for (i = 0; i < receiver_count(); i++) {
2209   //     if (receiver(i) == recv) goto found_recv(i);
2210   //   }
2211   //
2212   //   // Fast: no receiver, but profile is full
2213   //   for (i = 0; i < receiver_count(); i++) {
2214   //     if (receiver(i) == null) goto found_null(i);
2215   //   }
2216   //   goto polymorphic
2217   //
2218   //   // Slow: try to install receiver
2219   // found_null(i):
2220   //   CAS(&receiver(i), null, recv);
2221   //   goto restart
2222   //
2223   // polymorphic:
2224   //   count++;
2225   //   return
2226   //
2227   // found_recv(i):
2228   //   *receiver_count(i)++
2229   //
2230 
2231   bind(L_restart);
2232 
2233   // Fastest: receiver is already installed
2234   mov(offset, base_receiver_offset);
2235   bind(L_loop_search_receiver);
2236     ldr(rscratch1, Address(mdp, offset));
2237     cmp(rscratch1, recv);
2238     br(Assembler::EQ, L_found_recv);
2239   add(offset, offset, receiver_step);
2240   sub(rscratch1, offset, end_receiver_offset);
2241   cbnz(rscratch1, L_loop_search_receiver);
2242 
2243   // Fast: no receiver, but profile is full
2244   mov(offset, base_receiver_offset);
2245   bind(L_loop_search_empty);
2246     ldr(rscratch1, Address(mdp, offset));
2247     cbz(rscratch1, L_found_empty);
2248   add(offset, offset, receiver_step);
2249   sub(rscratch1, offset, end_receiver_offset);
2250   cbnz(rscratch1, L_loop_search_empty);
2251   b(L_polymorphic);
2252 
2253   // Slow: try to install receiver
2254   bind(L_found_empty);
2255 
2256   // Atomically swing receiver slot: null -> recv.
2257   //
2258   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2259   // is used to hold the destination address. This is safe because the
2260   // offset is no longer needed after the address is computed.
2261 
2262   lea(rscratch2, Address(mdp, offset));
2263   cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
2264           /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
2265 
2266   // CAS success means the slot now has the receiver we want. CAS failure means
2267   // something had claimed the slot concurrently: it can be the same receiver we want,
2268   // or something else. Since this is a slow path, we can optimize for code density,
2269   // and just restart the search from the beginning.
2270   b(L_restart);
2271 
2272   // Counter updates:
2273 
2274   // Increment polymorphic counter instead of receiver slot.
2275   bind(L_polymorphic);
2276   mov(offset, poly_count_offset);
2277   b(L_count_update);
2278 
2279   // Found a receiver, convert its slot offset to corresponding count offset.
2280   bind(L_found_recv);
2281   add(offset, offset, receiver_to_count_step);
2282 
2283   bind(L_count_update);
2284   increment(Address(mdp, offset), DataLayout::counter_increment);
2285 }
2286 
2287 
2288 void MacroAssembler::call_VM_leaf_base(address entry_point,
2289                                        int number_of_arguments,
2290                                        Label *retaddr) {
2291   Label E, L;
2292 
2293   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2294 
2295   mov(rscratch1, RuntimeAddress(entry_point));
2296   blr(rscratch1);
2297   if (retaddr)
2298     bind(*retaddr);
2299 
2300   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2301 }
2302 
2303 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2304   call_VM_leaf_base(entry_point, number_of_arguments);
2305 }
2306 
2307 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2308   pass_arg0(this, arg_0);
2309   call_VM_leaf_base(entry_point, 1);
2310 }
2311 
2312 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2313   assert_different_registers(arg_1, c_rarg0);
2314   pass_arg0(this, arg_0);
2315   pass_arg1(this, arg_1);
2316   call_VM_leaf_base(entry_point, 2);
2317 }
2318 
2319 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2320                                   Register arg_1, Register arg_2) {
2321   assert_different_registers(arg_1, c_rarg0);
2322   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2323   pass_arg0(this, arg_0);
2324   pass_arg1(this, arg_1);
2325   pass_arg2(this, arg_2);
2326   call_VM_leaf_base(entry_point, 3);
2327 }
2328 
2329 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2330   pass_arg0(this, arg_0);
2331   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2332 }
2333 
2334 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2335 
2336   assert_different_registers(arg_0, c_rarg1);
2337   pass_arg1(this, arg_1);
2338   pass_arg0(this, arg_0);
2339   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2340 }
2341 
2342 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2343   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2344   assert_different_registers(arg_1, c_rarg2);
2345   pass_arg2(this, arg_2);
2346   pass_arg1(this, arg_1);
2347   pass_arg0(this, arg_0);
2348   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2349 }
2350 
2351 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2352   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2353   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2354   assert_different_registers(arg_2, c_rarg3);
2355   pass_arg3(this, arg_3);
2356   pass_arg2(this, arg_2);
2357   pass_arg1(this, arg_1);
2358   pass_arg0(this, arg_0);
2359   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2360 }
2361 
2362 void MacroAssembler::null_check(Register reg, int offset) {
2363   if (needs_explicit_null_check(offset)) {
2364     // provoke OS null exception if reg is null by
2365     // accessing M[reg] w/o changing any registers
2366     // NOTE: this is plenty to provoke a segv
2367     ldr(zr, Address(reg));
2368   } else {
2369     // nothing to do, (later) access of M[reg + offset]
2370     // will provoke OS null exception if reg is null
2371   }
2372 }
2373 
2374 // MacroAssembler protected routines needed to implement
2375 // public methods
2376 
2377 void MacroAssembler::mov(Register r, Address dest) {
2378   code_section()->relocate(pc(), dest.rspec());
2379   uint64_t imm64 = (uint64_t)dest.target();
2380   movptr(r, imm64);
2381 }
2382 
2383 // Move a constant pointer into r.  In AArch64 mode the virtual
2384 // address space is 48 bits in size, so we only need three
2385 // instructions to create a patchable instruction sequence that can
2386 // reach anywhere.
2387 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2388 #ifndef PRODUCT
2389   {
2390     char buffer[64];
2391     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2392     block_comment(buffer);
2393   }
2394 #endif
2395   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2396   movz(r, imm64 & 0xffff);
2397   imm64 >>= 16;
2398   movk(r, imm64 & 0xffff, 16);
2399   imm64 >>= 16;
2400   movk(r, imm64 & 0xffff, 32);
2401 }
2402 
2403 // Macro to mov replicated immediate to vector register.
2404 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2405 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2406 // Vd will get the following values for different arrangements in T
2407 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2408 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2409 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2410 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2411 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2412 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2413 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2414 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2415 // Clobbers rscratch1
2416 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2417   assert(T != T1Q, "unsupported");
2418   if (T == T1D || T == T2D) {
2419     int imm = operand_valid_for_movi_immediate(imm64, T);
2420     if (-1 != imm) {
2421       movi(Vd, T, imm);
2422     } else {
2423       mov(rscratch1, imm64);
2424       dup(Vd, T, rscratch1);
2425     }
2426     return;
2427   }
2428 
2429 #ifdef ASSERT
2430   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2431   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2432   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2433 #endif
2434   int shift = operand_valid_for_movi_immediate(imm64, T);
2435   uint32_t imm32 = imm64 & 0xffffffffULL;
2436   if (shift >= 0) {
2437     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2438   } else {
2439     movw(rscratch1, imm32);
2440     dup(Vd, T, rscratch1);
2441   }
2442 }
2443 
2444 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2445 {
2446 #ifndef PRODUCT
2447   {
2448     char buffer[64];
2449     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2450     block_comment(buffer);
2451   }
2452 #endif
2453   if (operand_valid_for_logical_immediate(false, imm64)) {
2454     orr(dst, zr, imm64);
2455   } else {
2456     // we can use a combination of MOVZ or MOVN with
2457     // MOVK to build up the constant
2458     uint64_t imm_h[4];
2459     int zero_count = 0;
2460     int neg_count = 0;
2461     int i;
2462     for (i = 0; i < 4; i++) {
2463       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2464       if (imm_h[i] == 0) {
2465         zero_count++;
2466       } else if (imm_h[i] == 0xffffL) {
2467         neg_count++;
2468       }
2469     }
2470     if (zero_count == 4) {
2471       // one MOVZ will do
2472       movz(dst, 0);
2473     } else if (neg_count == 4) {
2474       // one MOVN will do
2475       movn(dst, 0);
2476     } else if (zero_count == 3) {
2477       for (i = 0; i < 4; i++) {
2478         if (imm_h[i] != 0L) {
2479           movz(dst, (uint32_t)imm_h[i], (i << 4));
2480           break;
2481         }
2482       }
2483     } else if (neg_count == 3) {
2484       // one MOVN will do
2485       for (int i = 0; i < 4; i++) {
2486         if (imm_h[i] != 0xffffL) {
2487           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2488           break;
2489         }
2490       }
2491     } else if (zero_count == 2) {
2492       // one MOVZ and one MOVK will do
2493       for (i = 0; i < 3; i++) {
2494         if (imm_h[i] != 0L) {
2495           movz(dst, (uint32_t)imm_h[i], (i << 4));
2496           i++;
2497           break;
2498         }
2499       }
2500       for (;i < 4; i++) {
2501         if (imm_h[i] != 0L) {
2502           movk(dst, (uint32_t)imm_h[i], (i << 4));
2503         }
2504       }
2505     } else if (neg_count == 2) {
2506       // one MOVN and one MOVK will do
2507       for (i = 0; i < 4; i++) {
2508         if (imm_h[i] != 0xffffL) {
2509           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2510           i++;
2511           break;
2512         }
2513       }
2514       for (;i < 4; i++) {
2515         if (imm_h[i] != 0xffffL) {
2516           movk(dst, (uint32_t)imm_h[i], (i << 4));
2517         }
2518       }
2519     } else if (zero_count == 1) {
2520       // one MOVZ and two MOVKs will do
2521       for (i = 0; i < 4; i++) {
2522         if (imm_h[i] != 0L) {
2523           movz(dst, (uint32_t)imm_h[i], (i << 4));
2524           i++;
2525           break;
2526         }
2527       }
2528       for (;i < 4; i++) {
2529         if (imm_h[i] != 0x0L) {
2530           movk(dst, (uint32_t)imm_h[i], (i << 4));
2531         }
2532       }
2533     } else if (neg_count == 1) {
2534       // one MOVN and two MOVKs will do
2535       for (i = 0; i < 4; i++) {
2536         if (imm_h[i] != 0xffffL) {
2537           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2538           i++;
2539           break;
2540         }
2541       }
2542       for (;i < 4; i++) {
2543         if (imm_h[i] != 0xffffL) {
2544           movk(dst, (uint32_t)imm_h[i], (i << 4));
2545         }
2546       }
2547     } else {
2548       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2549       movz(dst, (uint32_t)imm_h[0], 0);
2550       for (i = 1; i < 4; i++) {
2551         movk(dst, (uint32_t)imm_h[i], (i << 4));
2552       }
2553     }
2554   }
2555 }
2556 
2557 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2558 {
2559 #ifndef PRODUCT
2560     {
2561       char buffer[64];
2562       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2563       block_comment(buffer);
2564     }
2565 #endif
2566   if (operand_valid_for_logical_immediate(true, imm32)) {
2567     orrw(dst, zr, imm32);
2568   } else {
2569     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2570     // constant
2571     uint32_t imm_h[2];
2572     imm_h[0] = imm32 & 0xffff;
2573     imm_h[1] = ((imm32 >> 16) & 0xffff);
2574     if (imm_h[0] == 0) {
2575       movzw(dst, imm_h[1], 16);
2576     } else if (imm_h[0] == 0xffff) {
2577       movnw(dst, imm_h[1] ^ 0xffff, 16);
2578     } else if (imm_h[1] == 0) {
2579       movzw(dst, imm_h[0], 0);
2580     } else if (imm_h[1] == 0xffff) {
2581       movnw(dst, imm_h[0] ^ 0xffff, 0);
2582     } else {
2583       // use a MOVZ and MOVK (makes it easier to debug)
2584       movzw(dst, imm_h[0], 0);
2585       movkw(dst, imm_h[1], 16);
2586     }
2587   }
2588 }
2589 
2590 // Form an address from base + offset in Rd.  Rd may or may
2591 // not actually be used: you must use the Address that is returned.
2592 // It is up to you to ensure that the shift provided matches the size
2593 // of your data.
2594 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2595   if (Address::offset_ok_for_immed(byte_offset, shift))
2596     // It fits; no need for any heroics
2597     return Address(base, byte_offset);
2598 
2599   // Don't do anything clever with negative or misaligned offsets
2600   unsigned mask = (1 << shift) - 1;
2601   if (byte_offset < 0 || byte_offset & mask) {
2602     mov(Rd, byte_offset);
2603     add(Rd, base, Rd);
2604     return Address(Rd);
2605   }
2606 
2607   // See if we can do this with two 12-bit offsets
2608   {
2609     uint64_t word_offset = byte_offset >> shift;
2610     uint64_t masked_offset = word_offset & 0xfff000;
2611     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2612         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2613       add(Rd, base, masked_offset << shift);
2614       word_offset -= masked_offset;
2615       return Address(Rd, word_offset << shift);
2616     }
2617   }
2618 
2619   // Do it the hard way
2620   mov(Rd, byte_offset);
2621   add(Rd, base, Rd);
2622   return Address(Rd);
2623 }
2624 
2625 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2626                                     bool want_remainder, Register scratch)
2627 {
2628   // Full implementation of Java idiv and irem.  The function
2629   // returns the (pc) offset of the div instruction - may be needed
2630   // for implicit exceptions.
2631   //
2632   // constraint : ra/rb =/= scratch
2633   //         normal case
2634   //
2635   // input : ra: dividend
2636   //         rb: divisor
2637   //
2638   // result: either
2639   //         quotient  (= ra idiv rb)
2640   //         remainder (= ra irem rb)
2641 
2642   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2643 
2644   int idivl_offset = offset();
2645   if (! want_remainder) {
2646     sdivw(result, ra, rb);
2647   } else {
2648     sdivw(scratch, ra, rb);
2649     Assembler::msubw(result, scratch, rb, ra);
2650   }
2651 
2652   return idivl_offset;
2653 }
2654 
2655 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2656                                     bool want_remainder, Register scratch)
2657 {
2658   // Full implementation of Java ldiv and lrem.  The function
2659   // returns the (pc) offset of the div instruction - may be needed
2660   // for implicit exceptions.
2661   //
2662   // constraint : ra/rb =/= scratch
2663   //         normal case
2664   //
2665   // input : ra: dividend
2666   //         rb: divisor
2667   //
2668   // result: either
2669   //         quotient  (= ra idiv rb)
2670   //         remainder (= ra irem rb)
2671 
2672   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2673 
2674   int idivq_offset = offset();
2675   if (! want_remainder) {
2676     sdiv(result, ra, rb);
2677   } else {
2678     sdiv(scratch, ra, rb);
2679     Assembler::msub(result, scratch, rb, ra);
2680   }
2681 
2682   return idivq_offset;
2683 }
2684 
2685 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2686   address prev = pc() - NativeMembar::instruction_size;
2687   address last = code()->last_insn();
2688   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2689     NativeMembar *bar = NativeMembar_at(prev);
2690     if (AlwaysMergeDMB) {
2691       bar->set_kind(bar->get_kind() | order_constraint);
2692       BLOCK_COMMENT("merged membar(always)");
2693       return;
2694     }
2695     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2696     // doing so would introduce a StoreLoad which the caller did not
2697     // intend
2698     if (bar->get_kind() == order_constraint
2699         || bar->get_kind() == AnyAny
2700         || order_constraint == AnyAny) {
2701       // We are merging two memory barrier instructions.  On AArch64 we
2702       // can do this simply by ORing them together.
2703       bar->set_kind(bar->get_kind() | order_constraint);
2704       BLOCK_COMMENT("merged membar");
2705       return;
2706     } else {
2707       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2708       // We need check the last 2 instructions
2709       address prev2 = prev - NativeMembar::instruction_size;
2710       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2711         NativeMembar *bar2 = NativeMembar_at(prev2);
2712         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2713         BLOCK_COMMENT("merged membar(elided)");
2714         return;
2715       }
2716     }
2717   }
2718   code()->set_last_insn(pc());
2719   dmb(Assembler::barrier(order_constraint));
2720 }
2721 
2722 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2723   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2724     merge_ldst(rt, adr, size_in_bytes, is_store);
2725     code()->clear_last_insn();
2726     return true;
2727   } else {
2728     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2729     const uint64_t mask = size_in_bytes - 1;
2730     if (adr.getMode() == Address::base_plus_offset &&
2731         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2732       code()->set_last_insn(pc());
2733     }
2734     return false;
2735   }
2736 }
2737 
2738 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2739   // We always try to merge two adjacent loads into one ldp.
2740   if (!try_merge_ldst(Rx, adr, 8, false)) {
2741     Assembler::ldr(Rx, adr);
2742   }
2743 }
2744 
2745 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2746   // We always try to merge two adjacent loads into one ldp.
2747   if (!try_merge_ldst(Rw, adr, 4, false)) {
2748     Assembler::ldrw(Rw, adr);
2749   }
2750 }
2751 
2752 void MacroAssembler::str(Register Rx, const Address &adr) {
2753   // We always try to merge two adjacent stores into one stp.
2754   if (!try_merge_ldst(Rx, adr, 8, true)) {
2755     Assembler::str(Rx, adr);
2756   }
2757 }
2758 
2759 void MacroAssembler::strw(Register Rw, const Address &adr) {
2760   // We always try to merge two adjacent stores into one stp.
2761   if (!try_merge_ldst(Rw, adr, 4, true)) {
2762     Assembler::strw(Rw, adr);
2763   }
2764 }
2765 
2766 // MacroAssembler routines found actually to be needed
2767 
2768 void MacroAssembler::push(Register src)
2769 {
2770   str(src, Address(pre(esp, -1 * wordSize)));
2771 }
2772 
2773 void MacroAssembler::pop(Register dst)
2774 {
2775   ldr(dst, Address(post(esp, 1 * wordSize)));
2776 }
2777 
2778 // Note: load_unsigned_short used to be called load_unsigned_word.
2779 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2780   int off = offset();
2781   ldrh(dst, src);
2782   return off;
2783 }
2784 
2785 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2786   int off = offset();
2787   ldrb(dst, src);
2788   return off;
2789 }
2790 
2791 int MacroAssembler::load_signed_short(Register dst, Address src) {
2792   int off = offset();
2793   ldrsh(dst, src);
2794   return off;
2795 }
2796 
2797 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2798   int off = offset();
2799   ldrsb(dst, src);
2800   return off;
2801 }
2802 
2803 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2804   int off = offset();
2805   ldrshw(dst, src);
2806   return off;
2807 }
2808 
2809 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2810   int off = offset();
2811   ldrsbw(dst, src);
2812   return off;
2813 }
2814 
2815 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2816   switch (size_in_bytes) {
2817   case  8:  ldr(dst, src); break;
2818   case  4:  ldrw(dst, src); break;
2819   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2820   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2821   default:  ShouldNotReachHere();
2822   }
2823 }
2824 
2825 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2826   switch (size_in_bytes) {
2827   case  8:  str(src, dst); break;
2828   case  4:  strw(src, dst); break;
2829   case  2:  strh(src, dst); break;
2830   case  1:  strb(src, dst); break;
2831   default:  ShouldNotReachHere();
2832   }
2833 }
2834 
2835 void MacroAssembler::decrementw(Register reg, int value)
2836 {
2837   if (value < 0)  { incrementw(reg, -value);      return; }
2838   if (value == 0) {                               return; }
2839   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2840   /* else */ {
2841     guarantee(reg != rscratch2, "invalid dst for register decrement");
2842     movw(rscratch2, (unsigned)value);
2843     subw(reg, reg, rscratch2);
2844   }
2845 }
2846 
2847 void MacroAssembler::decrement(Register reg, int value)
2848 {
2849   if (value < 0)  { increment(reg, -value);      return; }
2850   if (value == 0) {                              return; }
2851   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2852   /* else */ {
2853     assert(reg != rscratch2, "invalid dst for register decrement");
2854     mov(rscratch2, (uint64_t)value);
2855     sub(reg, reg, rscratch2);
2856   }
2857 }
2858 
2859 void MacroAssembler::decrementw(Address dst, int value)
2860 {
2861   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2862   if (dst.getMode() == Address::literal) {
2863     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2864     lea(rscratch2, dst);
2865     dst = Address(rscratch2);
2866   }
2867   ldrw(rscratch1, dst);
2868   decrementw(rscratch1, value);
2869   strw(rscratch1, dst);
2870 }
2871 
2872 void MacroAssembler::decrement(Address dst, int value)
2873 {
2874   assert(!dst.uses(rscratch1), "invalid address for decrement");
2875   if (dst.getMode() == Address::literal) {
2876     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2877     lea(rscratch2, dst);
2878     dst = Address(rscratch2);
2879   }
2880   ldr(rscratch1, dst);
2881   decrement(rscratch1, value);
2882   str(rscratch1, dst);
2883 }
2884 
2885 void MacroAssembler::incrementw(Register reg, int value)
2886 {
2887   if (value < 0)  { decrementw(reg, -value);      return; }
2888   if (value == 0) {                               return; }
2889   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2890   /* else */ {
2891     assert(reg != rscratch2, "invalid dst for register increment");
2892     movw(rscratch2, (unsigned)value);
2893     addw(reg, reg, rscratch2);
2894   }
2895 }
2896 
2897 void MacroAssembler::increment(Register reg, int value)
2898 {
2899   if (value < 0)  { decrement(reg, -value);      return; }
2900   if (value == 0) {                              return; }
2901   if (value < (1 << 12)) { add(reg, reg, value); return; }
2902   /* else */ {
2903     assert(reg != rscratch2, "invalid dst for register increment");
2904     movw(rscratch2, (unsigned)value);
2905     add(reg, reg, rscratch2);
2906   }
2907 }
2908 
2909 void MacroAssembler::incrementw(Address dst, int value)
2910 {
2911   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2912   if (dst.getMode() == Address::literal) {
2913     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2914     lea(rscratch2, dst);
2915     dst = Address(rscratch2);
2916   }
2917   ldrw(rscratch1, dst);
2918   incrementw(rscratch1, value);
2919   strw(rscratch1, dst);
2920 }
2921 
2922 void MacroAssembler::increment(Address dst, int value)
2923 {
2924   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2925   if (dst.getMode() == Address::literal) {
2926     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2927     lea(rscratch2, dst);
2928     dst = Address(rscratch2);
2929   }
2930   ldr(rscratch1, dst);
2931   increment(rscratch1, value);
2932   str(rscratch1, dst);
2933 }
2934 
2935 // Push lots of registers in the bit set supplied.  Don't push sp.
2936 // Return the number of words pushed
2937 int MacroAssembler::push(unsigned int bitset, Register stack) {
2938   int words_pushed = 0;
2939 
2940   // Scan bitset to accumulate register pairs
2941   unsigned char regs[32];
2942   int count = 0;
2943   for (int reg = 0; reg <= 30; reg++) {
2944     if (1 & bitset)
2945       regs[count++] = reg;
2946     bitset >>= 1;
2947   }
2948   regs[count++] = zr->raw_encoding();
2949   count &= ~1;  // Only push an even number of regs
2950 
2951   if (count) {
2952     stp(as_Register(regs[0]), as_Register(regs[1]),
2953        Address(pre(stack, -count * wordSize)));
2954     words_pushed += 2;
2955   }
2956   for (int i = 2; i < count; i += 2) {
2957     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2958        Address(stack, i * wordSize));
2959     words_pushed += 2;
2960   }
2961 
2962   assert(words_pushed == count, "oops, pushed != count");
2963 
2964   return count;
2965 }
2966 
2967 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2968   int words_pushed = 0;
2969 
2970   // Scan bitset to accumulate register pairs
2971   unsigned char regs[32];
2972   int count = 0;
2973   for (int reg = 0; reg <= 30; reg++) {
2974     if (1 & bitset)
2975       regs[count++] = reg;
2976     bitset >>= 1;
2977   }
2978   regs[count++] = zr->raw_encoding();
2979   count &= ~1;
2980 
2981   for (int i = 2; i < count; i += 2) {
2982     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2983        Address(stack, i * wordSize));
2984     words_pushed += 2;
2985   }
2986   if (count) {
2987     ldp(as_Register(regs[0]), as_Register(regs[1]),
2988        Address(post(stack, count * wordSize)));
2989     words_pushed += 2;
2990   }
2991 
2992   assert(words_pushed == count, "oops, pushed != count");
2993 
2994   return count;
2995 }
2996 
2997 // Push lots of registers in the bit set supplied.  Don't push sp.
2998 // Return the number of dwords pushed
2999 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3000   int words_pushed = 0;
3001   bool use_sve = false;
3002   int sve_vector_size_in_bytes = 0;
3003 
3004 #ifdef COMPILER2
3005   use_sve = Matcher::supports_scalable_vector();
3006   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3007 #endif
3008 
3009   // Scan bitset to accumulate register pairs
3010   unsigned char regs[32];
3011   int count = 0;
3012   for (int reg = 0; reg <= 31; reg++) {
3013     if (1 & bitset)
3014       regs[count++] = reg;
3015     bitset >>= 1;
3016   }
3017 
3018   if (count == 0) {
3019     return 0;
3020   }
3021 
3022   if (mode == PushPopFull) {
3023     if (use_sve && sve_vector_size_in_bytes > 16) {
3024       mode = PushPopSVE;
3025     } else {
3026       mode = PushPopNeon;
3027     }
3028   }
3029 
3030 #ifndef PRODUCT
3031   {
3032     char buffer[48];
3033     if (mode == PushPopSVE) {
3034       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3035     } else if (mode == PushPopNeon) {
3036       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3037     } else {
3038       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3039     }
3040     block_comment(buffer);
3041   }
3042 #endif
3043 
3044   if (mode == PushPopSVE) {
3045     sub(stack, stack, sve_vector_size_in_bytes * count);
3046     for (int i = 0; i < count; i++) {
3047       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3048     }
3049     return count * sve_vector_size_in_bytes / 8;
3050   }
3051 
3052   if (mode == PushPopNeon) {
3053     if (count == 1) {
3054       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3055       return 2;
3056     }
3057 
3058     bool odd = (count & 1) == 1;
3059     int push_slots = count + (odd ? 1 : 0);
3060 
3061     // Always pushing full 128 bit registers.
3062     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3063     words_pushed += 2;
3064 
3065     for (int i = 2; i + 1 < count; i += 2) {
3066       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3067       words_pushed += 2;
3068     }
3069 
3070     if (odd) {
3071       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3072       words_pushed++;
3073     }
3074 
3075     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3076     return count * 2;
3077   }
3078 
3079   if (mode == PushPopFp) {
3080     bool odd = (count & 1) == 1;
3081     int push_slots = count + (odd ? 1 : 0);
3082 
3083     if (count == 1) {
3084       // Stack pointer must be 16 bytes aligned
3085       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3086       return 1;
3087     }
3088 
3089     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3090     words_pushed += 2;
3091 
3092     for (int i = 2; i + 1 < count; i += 2) {
3093       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3094       words_pushed += 2;
3095     }
3096 
3097     if (odd) {
3098       // Stack pointer must be 16 bytes aligned
3099       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3100       words_pushed++;
3101     }
3102 
3103     assert(words_pushed == count, "oops, pushed != count");
3104 
3105     return count;
3106   }
3107 
3108   return 0;
3109 }
3110 
3111 // Return the number of dwords popped
3112 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3113   int words_pushed = 0;
3114   bool use_sve = false;
3115   int sve_vector_size_in_bytes = 0;
3116 
3117 #ifdef COMPILER2
3118   use_sve = Matcher::supports_scalable_vector();
3119   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3120 #endif
3121   // Scan bitset to accumulate register pairs
3122   unsigned char regs[32];
3123   int count = 0;
3124   for (int reg = 0; reg <= 31; reg++) {
3125     if (1 & bitset)
3126       regs[count++] = reg;
3127     bitset >>= 1;
3128   }
3129 
3130   if (count == 0) {
3131     return 0;
3132   }
3133 
3134   if (mode == PushPopFull) {
3135     if (use_sve && sve_vector_size_in_bytes > 16) {
3136       mode = PushPopSVE;
3137     } else {
3138       mode = PushPopNeon;
3139     }
3140   }
3141 
3142 #ifndef PRODUCT
3143   {
3144     char buffer[48];
3145     if (mode == PushPopSVE) {
3146       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3147     } else if (mode == PushPopNeon) {
3148       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3149     } else {
3150       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3151     }
3152     block_comment(buffer);
3153   }
3154 #endif
3155 
3156   if (mode == PushPopSVE) {
3157     for (int i = count - 1; i >= 0; i--) {
3158       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3159     }
3160     add(stack, stack, sve_vector_size_in_bytes * count);
3161     return count * sve_vector_size_in_bytes / 8;
3162   }
3163 
3164   if (mode == PushPopNeon) {
3165     if (count == 1) {
3166       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3167       return 2;
3168     }
3169 
3170     bool odd = (count & 1) == 1;
3171     int push_slots = count + (odd ? 1 : 0);
3172 
3173     if (odd) {
3174       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3175       words_pushed++;
3176     }
3177 
3178     for (int i = 2; i + 1 < count; i += 2) {
3179       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3180       words_pushed += 2;
3181     }
3182 
3183     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3184     words_pushed += 2;
3185 
3186     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3187 
3188     return count * 2;
3189   }
3190 
3191   if (mode == PushPopFp) {
3192     bool odd = (count & 1) == 1;
3193     int push_slots = count + (odd ? 1 : 0);
3194 
3195     if (count == 1) {
3196       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3197       return 1;
3198     }
3199 
3200     if (odd) {
3201       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3202       words_pushed++;
3203     }
3204 
3205     for (int i = 2; i + 1 < count; i += 2) {
3206       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3207       words_pushed += 2;
3208     }
3209 
3210     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3211     words_pushed += 2;
3212 
3213     assert(words_pushed == count, "oops, pushed != count");
3214 
3215     return count;
3216   }
3217 
3218   return 0;
3219 }
3220 
3221 // Return the number of dwords pushed
3222 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3223   bool use_sve = false;
3224   int sve_predicate_size_in_slots = 0;
3225 
3226 #ifdef COMPILER2
3227   use_sve = Matcher::supports_scalable_vector();
3228   if (use_sve) {
3229     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3230   }
3231 #endif
3232 
3233   if (!use_sve) {
3234     return 0;
3235   }
3236 
3237   unsigned char regs[PRegister::number_of_registers];
3238   int count = 0;
3239   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3240     if (1 & bitset)
3241       regs[count++] = reg;
3242     bitset >>= 1;
3243   }
3244 
3245   if (count == 0) {
3246     return 0;
3247   }
3248 
3249   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3250                                   VMRegImpl::stack_slot_size * count, 16);
3251   sub(stack, stack, total_push_bytes);
3252   for (int i = 0; i < count; i++) {
3253     sve_str(as_PRegister(regs[i]), Address(stack, i));
3254   }
3255   return total_push_bytes / 8;
3256 }
3257 
3258 // Return the number of dwords popped
3259 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3260   bool use_sve = false;
3261   int sve_predicate_size_in_slots = 0;
3262 
3263 #ifdef COMPILER2
3264   use_sve = Matcher::supports_scalable_vector();
3265   if (use_sve) {
3266     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3267   }
3268 #endif
3269 
3270   if (!use_sve) {
3271     return 0;
3272   }
3273 
3274   unsigned char regs[PRegister::number_of_registers];
3275   int count = 0;
3276   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3277     if (1 & bitset)
3278       regs[count++] = reg;
3279     bitset >>= 1;
3280   }
3281 
3282   if (count == 0) {
3283     return 0;
3284   }
3285 
3286   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3287                                  VMRegImpl::stack_slot_size * count, 16);
3288   for (int i = count - 1; i >= 0; i--) {
3289     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3290   }
3291   add(stack, stack, total_pop_bytes);
3292   return total_pop_bytes / 8;
3293 }
3294 
3295 #ifdef ASSERT
3296 void MacroAssembler::verify_heapbase(const char* msg) {
3297 #if 0
3298   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3299   assert (Universe::heap() != nullptr, "java heap should be initialized");
3300   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3301     // rheapbase is allocated as general register
3302     return;
3303   }
3304   if (CheckCompressedOops) {
3305     Label ok;
3306     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3307     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3308     br(Assembler::EQ, ok);
3309     stop(msg);
3310     bind(ok);
3311     pop(1 << rscratch1->encoding(), sp);
3312   }
3313 #endif
3314 }
3315 #endif
3316 
3317 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3318   assert_different_registers(value, tmp1, tmp2);
3319   Label done, tagged, weak_tagged;
3320 
3321   cbz(value, done);           // Use null as-is.
3322   tst(value, JNIHandles::tag_mask); // Test for tag.
3323   br(Assembler::NE, tagged);
3324 
3325   // Resolve local handle
3326   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3327   verify_oop(value);
3328   b(done);
3329 
3330   bind(tagged);
3331   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3332   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3333 
3334   // Resolve global handle
3335   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3336   verify_oop(value);
3337   b(done);
3338 
3339   bind(weak_tagged);
3340   // Resolve jweak.
3341   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3342                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3343   verify_oop(value);
3344 
3345   bind(done);
3346 }
3347 
3348 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3349   assert_different_registers(value, tmp1, tmp2);
3350   Label done;
3351 
3352   cbz(value, done);           // Use null as-is.
3353 
3354 #ifdef ASSERT
3355   {
3356     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3357     Label valid_global_tag;
3358     tbnz(value, 1, valid_global_tag); // Test for global tag
3359     stop("non global jobject using resolve_global_jobject");
3360     bind(valid_global_tag);
3361   }
3362 #endif
3363 
3364   // Resolve global handle
3365   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3366   verify_oop(value);
3367 
3368   bind(done);
3369 }
3370 
3371 void MacroAssembler::stop(const char* msg) {
3372   // Skip AOT caching C strings in scratch buffer.
3373   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3374   BLOCK_COMMENT(str);
3375   // load msg into r0 so we can access it from the signal handler
3376   // ExternalAddress enables saving and restoring via the code cache
3377   lea(c_rarg0, ExternalAddress((address) str));
3378   dcps1(0xdeae);
3379 }
3380 
3381 void MacroAssembler::unimplemented(const char* what) {
3382   const char* buf = nullptr;
3383   {
3384     ResourceMark rm;
3385     stringStream ss;
3386     ss.print("unimplemented: %s", what);
3387     buf = code_string(ss.as_string());
3388   }
3389   stop(buf);
3390 }
3391 
3392 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3393 #ifdef ASSERT
3394   Label OK;
3395   br(cc, OK);
3396   stop(msg);
3397   bind(OK);
3398 #endif
3399 }
3400 
3401 // If a constant does not fit in an immediate field, generate some
3402 // number of MOV instructions and then perform the operation.
3403 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3404                                            add_sub_imm_insn insn1,
3405                                            add_sub_reg_insn insn2,
3406                                            bool is32) {
3407   assert(Rd != zr, "Rd = zr and not setting flags?");
3408   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3409   if (fits) {
3410     (this->*insn1)(Rd, Rn, imm);
3411   } else {
3412     if (g_uabs(imm) < (1 << 24)) {
3413        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3414        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3415     } else {
3416        assert_different_registers(Rd, Rn);
3417        mov(Rd, imm);
3418        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3419     }
3420   }
3421 }
3422 
3423 // Separate vsn which sets the flags. Optimisations are more restricted
3424 // because we must set the flags correctly.
3425 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3426                                              add_sub_imm_insn insn1,
3427                                              add_sub_reg_insn insn2,
3428                                              bool is32) {
3429   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3430   if (fits) {
3431     (this->*insn1)(Rd, Rn, imm);
3432   } else {
3433     assert_different_registers(Rd, Rn);
3434     assert(Rd != zr, "overflow in immediate operand");
3435     mov(Rd, imm);
3436     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3437   }
3438 }
3439 
3440 
3441 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3442   if (increment.is_register()) {
3443     add(Rd, Rn, increment.as_register());
3444   } else {
3445     add(Rd, Rn, increment.as_constant());
3446   }
3447 }
3448 
3449 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3450   if (increment.is_register()) {
3451     addw(Rd, Rn, increment.as_register());
3452   } else {
3453     addw(Rd, Rn, increment.as_constant());
3454   }
3455 }
3456 
3457 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3458   if (decrement.is_register()) {
3459     sub(Rd, Rn, decrement.as_register());
3460   } else {
3461     sub(Rd, Rn, decrement.as_constant());
3462   }
3463 }
3464 
3465 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3466   if (decrement.is_register()) {
3467     subw(Rd, Rn, decrement.as_register());
3468   } else {
3469     subw(Rd, Rn, decrement.as_constant());
3470   }
3471 }
3472 
3473 void MacroAssembler::reinit_heapbase()
3474 {
3475   if (UseCompressedOops) {
3476     if (Universe::is_fully_initialized() && !AOTCodeCache::is_on_for_dump()) {
3477       mov(rheapbase, CompressedOops::base());
3478     } else {
3479       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3480       ldr(rheapbase, Address(rheapbase));
3481     }
3482   }
3483 }
3484 
3485 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3486 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3487 // Pass a register for the result, otherwise pass noreg.
3488 
3489 // Clobbers rscratch1
3490 void MacroAssembler::cmpxchg(Register addr, Register expected,
3491                              Register new_val,
3492                              enum operand_size size,
3493                              bool acquire, bool release,
3494                              bool weak,
3495                              Register result) {
3496   if (result == noreg)  result = rscratch1;
3497   BLOCK_COMMENT("cmpxchg {");
3498   if (UseLSE) {
3499     mov(result, expected);
3500     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3501     compare_eq(result, expected, size);
3502 #ifdef ASSERT
3503     // Poison rscratch1 which is written on !UseLSE branch
3504     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3505 #endif
3506   } else {
3507     Label retry_load, done;
3508     prfm(Address(addr), PSTL1STRM);
3509     bind(retry_load);
3510     load_exclusive(result, addr, size, acquire);
3511     compare_eq(result, expected, size);
3512     br(Assembler::NE, done);
3513     store_exclusive(rscratch1, new_val, addr, size, release);
3514     if (weak) {
3515       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3516     } else {
3517       cbnzw(rscratch1, retry_load);
3518     }
3519     bind(done);
3520   }
3521   BLOCK_COMMENT("} cmpxchg");
3522 }
3523 
3524 // A generic comparison. Only compares for equality, clobbers rscratch1.
3525 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3526   if (size == xword) {
3527     cmp(rm, rn);
3528   } else if (size == word) {
3529     cmpw(rm, rn);
3530   } else if (size == halfword) {
3531     eorw(rscratch1, rm, rn);
3532     ands(zr, rscratch1, 0xffff);
3533   } else if (size == byte) {
3534     eorw(rscratch1, rm, rn);
3535     ands(zr, rscratch1, 0xff);
3536   } else {
3537     ShouldNotReachHere();
3538   }
3539 }
3540 
3541 
3542 static bool different(Register a, RegisterOrConstant b, Register c) {
3543   if (b.is_constant())
3544     return a != c;
3545   else
3546     return a != b.as_register() && a != c && b.as_register() != c;
3547 }
3548 
3549 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3550 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3551   if (UseLSE) {                                                         \
3552     prev = prev->is_valid() ? prev : zr;                                \
3553     if (incr.is_register()) {                                           \
3554       AOP(sz, incr.as_register(), prev, addr);                          \
3555     } else {                                                            \
3556       mov(rscratch2, incr.as_constant());                               \
3557       AOP(sz, rscratch2, prev, addr);                                   \
3558     }                                                                   \
3559     return;                                                             \
3560   }                                                                     \
3561   Register result = rscratch2;                                          \
3562   if (prev->is_valid())                                                 \
3563     result = different(prev, incr, addr) ? prev : rscratch2;            \
3564                                                                         \
3565   Label retry_load;                                                     \
3566   prfm(Address(addr), PSTL1STRM);                                       \
3567   bind(retry_load);                                                     \
3568   LDXR(result, addr);                                                   \
3569   OP(rscratch1, result, incr);                                          \
3570   STXR(rscratch2, rscratch1, addr);                                     \
3571   cbnzw(rscratch2, retry_load);                                         \
3572   if (prev->is_valid() && prev != result) {                             \
3573     IOP(prev, rscratch1, incr);                                         \
3574   }                                                                     \
3575 }
3576 
3577 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3578 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3579 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3580 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3581 
3582 #undef ATOMIC_OP
3583 
3584 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3585 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3586   if (UseLSE) {                                                         \
3587     prev = prev->is_valid() ? prev : zr;                                \
3588     AOP(sz, newv, prev, addr);                                          \
3589     return;                                                             \
3590   }                                                                     \
3591   Register result = rscratch2;                                          \
3592   if (prev->is_valid())                                                 \
3593     result = different(prev, newv, addr) ? prev : rscratch2;            \
3594                                                                         \
3595   Label retry_load;                                                     \
3596   prfm(Address(addr), PSTL1STRM);                                       \
3597   bind(retry_load);                                                     \
3598   LDXR(result, addr);                                                   \
3599   STXR(rscratch1, newv, addr);                                          \
3600   cbnzw(rscratch1, retry_load);                                         \
3601   if (prev->is_valid() && prev != result)                               \
3602     mov(prev, result);                                                  \
3603 }
3604 
3605 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3606 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3607 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3608 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3609 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3610 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3611 
3612 #undef ATOMIC_XCHG
3613 
3614 #ifndef PRODUCT
3615 extern "C" void findpc(intptr_t x);
3616 #endif
3617 
3618 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3619 {
3620   // In order to get locks to work, we need to fake a in_VM state
3621   if (ShowMessageBoxOnError) {
3622     JavaThread* thread = JavaThread::current();
3623     thread->set_thread_state(_thread_in_vm);
3624 #ifndef PRODUCT
3625     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3626       ttyLocker ttyl;
3627       BytecodeCounter::print();
3628     }
3629 #endif
3630     if (os::message_box(msg, "Execution stopped, print registers?")) {
3631       ttyLocker ttyl;
3632       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3633 #ifndef PRODUCT
3634       tty->cr();
3635       findpc(pc);
3636       tty->cr();
3637 #endif
3638       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3639       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3640       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3641       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3642       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3643       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3644       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3645       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3646       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3647       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3648       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3649       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3650       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3651       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3652       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3653       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3654       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3655       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3656       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3657       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3658       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3659       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3660       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3661       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3662       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3663       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3664       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3665       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3666       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3667       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3668       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3669       BREAKPOINT;
3670     }
3671   }
3672   fatal("DEBUG MESSAGE: %s", msg);
3673 }
3674 
3675 RegSet MacroAssembler::call_clobbered_gp_registers() {
3676   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3677 #ifndef R18_RESERVED
3678   regs += r18_tls;
3679 #endif
3680   return regs;
3681 }
3682 
3683 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3684   int step = 4 * wordSize;
3685   push(call_clobbered_gp_registers() - exclude, sp);
3686   sub(sp, sp, step);
3687   mov(rscratch1, -step);
3688   // Push v0-v7, v16-v31.
3689   for (int i = 31; i>= 4; i -= 4) {
3690     if (i <= v7->encoding() || i >= v16->encoding())
3691       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3692           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3693   }
3694   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3695       as_FloatRegister(3), T1D, Address(sp));
3696 }
3697 
3698 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3699   for (int i = 0; i < 32; i += 4) {
3700     if (i <= v7->encoding() || i >= v16->encoding())
3701       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3702           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3703   }
3704 
3705   reinitialize_ptrue();
3706 
3707   pop(call_clobbered_gp_registers() - exclude, sp);
3708 }
3709 
3710 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3711                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3712   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3713   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3714     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3715     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3716       sve_str(as_FloatRegister(i), Address(sp, i));
3717     }
3718   } else {
3719     int step = (save_vectors ? 8 : 4) * wordSize;
3720     mov(rscratch1, -step);
3721     sub(sp, sp, step);
3722     for (int i = 28; i >= 4; i -= 4) {
3723       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3724           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3725     }
3726     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3727   }
3728   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3729     sub(sp, sp, total_predicate_in_bytes);
3730     for (int i = 0; i < PRegister::number_of_registers; i++) {
3731       sve_str(as_PRegister(i), Address(sp, i));
3732     }
3733   }
3734 }
3735 
3736 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3737                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3738   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3739     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3740       sve_ldr(as_PRegister(i), Address(sp, i));
3741     }
3742     add(sp, sp, total_predicate_in_bytes);
3743   }
3744   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3745     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3746       sve_ldr(as_FloatRegister(i), Address(sp, i));
3747     }
3748     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3749   } else {
3750     int step = (restore_vectors ? 8 : 4) * wordSize;
3751     for (int i = 0; i <= 28; i += 4)
3752       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3753           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3754   }
3755 
3756   // We may use predicate registers and rely on ptrue with SVE,
3757   // regardless of wide vector (> 8 bytes) used or not.
3758   if (use_sve) {
3759     reinitialize_ptrue();
3760   }
3761 
3762   // integer registers except lr & sp
3763   pop(RegSet::range(r0, r17), sp);
3764 #ifdef R18_RESERVED
3765   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3766   pop(RegSet::range(r20, r29), sp);
3767 #else
3768   pop(RegSet::range(r18_tls, r29), sp);
3769 #endif
3770 }
3771 
3772 /**
3773  * Helpers for multiply_to_len().
3774  */
3775 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3776                                      Register src1, Register src2) {
3777   adds(dest_lo, dest_lo, src1);
3778   adc(dest_hi, dest_hi, zr);
3779   adds(dest_lo, dest_lo, src2);
3780   adc(final_dest_hi, dest_hi, zr);
3781 }
3782 
3783 // Generate an address from (r + r1 extend offset).  "size" is the
3784 // size of the operand.  The result may be in rscratch2.
3785 Address MacroAssembler::offsetted_address(Register r, Register r1,
3786                                           Address::extend ext, int offset, int size) {
3787   if (offset || (ext.shift() % size != 0)) {
3788     lea(rscratch2, Address(r, r1, ext));
3789     return Address(rscratch2, offset);
3790   } else {
3791     return Address(r, r1, ext);
3792   }
3793 }
3794 
3795 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3796 {
3797   assert(offset >= 0, "spill to negative address?");
3798   // Offset reachable ?
3799   //   Not aligned - 9 bits signed offset
3800   //   Aligned - 12 bits unsigned offset shifted
3801   Register base = sp;
3802   if ((offset & (size-1)) && offset >= (1<<8)) {
3803     add(tmp, base, offset & ((1<<12)-1));
3804     base = tmp;
3805     offset &= -1u<<12;
3806   }
3807 
3808   if (offset >= (1<<12) * size) {
3809     add(tmp, base, offset & (((1<<12)-1)<<12));
3810     base = tmp;
3811     offset &= ~(((1<<12)-1)<<12);
3812   }
3813 
3814   return Address(base, offset);
3815 }
3816 
3817 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3818   assert(offset >= 0, "spill to negative address?");
3819 
3820   Register base = sp;
3821 
3822   // An immediate offset in the range 0 to 255 which is multiplied
3823   // by the current vector or predicate register size in bytes.
3824   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3825     return Address(base, offset / sve_reg_size_in_bytes);
3826   }
3827 
3828   add(tmp, base, offset);
3829   return Address(tmp);
3830 }
3831 
3832 // Checks whether offset is aligned.
3833 // Returns true if it is, else false.
3834 bool MacroAssembler::merge_alignment_check(Register base,
3835                                            size_t size,
3836                                            int64_t cur_offset,
3837                                            int64_t prev_offset) const {
3838   if (AvoidUnalignedAccesses) {
3839     if (base == sp) {
3840       // Checks whether low offset if aligned to pair of registers.
3841       int64_t pair_mask = size * 2 - 1;
3842       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3843       return (offset & pair_mask) == 0;
3844     } else { // If base is not sp, we can't guarantee the access is aligned.
3845       return false;
3846     }
3847   } else {
3848     int64_t mask = size - 1;
3849     // Load/store pair instruction only supports element size aligned offset.
3850     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3851   }
3852 }
3853 
3854 // Checks whether current and previous loads/stores can be merged.
3855 // Returns true if it can be merged, else false.
3856 bool MacroAssembler::ldst_can_merge(Register rt,
3857                                     const Address &adr,
3858                                     size_t cur_size_in_bytes,
3859                                     bool is_store) const {
3860   address prev = pc() - NativeInstruction::instruction_size;
3861   address last = code()->last_insn();
3862 
3863   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3864     return false;
3865   }
3866 
3867   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3868     return false;
3869   }
3870 
3871   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3872   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3873 
3874   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3875   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3876 
3877   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3878     return false;
3879   }
3880 
3881   int64_t max_offset = 63 * prev_size_in_bytes;
3882   int64_t min_offset = -64 * prev_size_in_bytes;
3883 
3884   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3885 
3886   // Only same base can be merged.
3887   if (adr.base() != prev_ldst->base()) {
3888     return false;
3889   }
3890 
3891   int64_t cur_offset = adr.offset();
3892   int64_t prev_offset = prev_ldst->offset();
3893   size_t diff = abs(cur_offset - prev_offset);
3894   if (diff != prev_size_in_bytes) {
3895     return false;
3896   }
3897 
3898   // Following cases can not be merged:
3899   // ldr x2, [x2, #8]
3900   // ldr x3, [x2, #16]
3901   // or:
3902   // ldr x2, [x3, #8]
3903   // ldr x2, [x3, #16]
3904   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3905   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3906     return false;
3907   }
3908 
3909   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3910   // Offset range must be in ldp/stp instruction's range.
3911   if (low_offset > max_offset || low_offset < min_offset) {
3912     return false;
3913   }
3914 
3915   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3916     return true;
3917   }
3918 
3919   return false;
3920 }
3921 
3922 // Merge current load/store with previous load/store into ldp/stp.
3923 void MacroAssembler::merge_ldst(Register rt,
3924                                 const Address &adr,
3925                                 size_t cur_size_in_bytes,
3926                                 bool is_store) {
3927 
3928   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3929 
3930   Register rt_low, rt_high;
3931   address prev = pc() - NativeInstruction::instruction_size;
3932   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3933 
3934   int64_t offset;
3935 
3936   if (adr.offset() < prev_ldst->offset()) {
3937     offset = adr.offset();
3938     rt_low = rt;
3939     rt_high = prev_ldst->target();
3940   } else {
3941     offset = prev_ldst->offset();
3942     rt_low = prev_ldst->target();
3943     rt_high = rt;
3944   }
3945 
3946   Address adr_p = Address(prev_ldst->base(), offset);
3947   // Overwrite previous generated binary.
3948   code_section()->set_end(prev);
3949 
3950   const size_t sz = prev_ldst->size_in_bytes();
3951   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3952   if (!is_store) {
3953     BLOCK_COMMENT("merged ldr pair");
3954     if (sz == 8) {
3955       ldp(rt_low, rt_high, adr_p);
3956     } else {
3957       ldpw(rt_low, rt_high, adr_p);
3958     }
3959   } else {
3960     BLOCK_COMMENT("merged str pair");
3961     if (sz == 8) {
3962       stp(rt_low, rt_high, adr_p);
3963     } else {
3964       stpw(rt_low, rt_high, adr_p);
3965     }
3966   }
3967 }
3968 
3969 /**
3970  * Multiply 64 bit by 64 bit first loop.
3971  */
3972 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3973                                            Register y, Register y_idx, Register z,
3974                                            Register carry, Register product,
3975                                            Register idx, Register kdx) {
3976   //
3977   //  jlong carry, x[], y[], z[];
3978   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3979   //    huge_128 product = y[idx] * x[xstart] + carry;
3980   //    z[kdx] = (jlong)product;
3981   //    carry  = (jlong)(product >>> 64);
3982   //  }
3983   //  z[xstart] = carry;
3984   //
3985 
3986   Label L_first_loop, L_first_loop_exit;
3987   Label L_one_x, L_one_y, L_multiply;
3988 
3989   subsw(xstart, xstart, 1);
3990   br(Assembler::MI, L_one_x);
3991 
3992   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3993   ldr(x_xstart, Address(rscratch1));
3994   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3995 
3996   bind(L_first_loop);
3997   subsw(idx, idx, 1);
3998   br(Assembler::MI, L_first_loop_exit);
3999   subsw(idx, idx, 1);
4000   br(Assembler::MI, L_one_y);
4001   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4002   ldr(y_idx, Address(rscratch1));
4003   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
4004   bind(L_multiply);
4005 
4006   // AArch64 has a multiply-accumulate instruction that we can't use
4007   // here because it has no way to process carries, so we have to use
4008   // separate add and adc instructions.  Bah.
4009   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
4010   mul(product, x_xstart, y_idx);
4011   adds(product, product, carry);
4012   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
4013 
4014   subw(kdx, kdx, 2);
4015   ror(product, product, 32); // back to big-endian
4016   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4017 
4018   b(L_first_loop);
4019 
4020   bind(L_one_y);
4021   ldrw(y_idx, Address(y,  0));
4022   b(L_multiply);
4023 
4024   bind(L_one_x);
4025   ldrw(x_xstart, Address(x,  0));
4026   b(L_first_loop);
4027 
4028   bind(L_first_loop_exit);
4029 }
4030 
4031 /**
4032  * Multiply 128 bit by 128. Unrolled inner loop.
4033  *
4034  */
4035 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4036                                              Register carry, Register carry2,
4037                                              Register idx, Register jdx,
4038                                              Register yz_idx1, Register yz_idx2,
4039                                              Register tmp, Register tmp3, Register tmp4,
4040                                              Register tmp6, Register product_hi) {
4041 
4042   //   jlong carry, x[], y[], z[];
4043   //   int kdx = ystart+1;
4044   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4045   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4046   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4047   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4048   //     carry  = (jlong)(tmp4 >>> 64);
4049   //     z[kdx+idx+1] = (jlong)tmp3;
4050   //     z[kdx+idx] = (jlong)tmp4;
4051   //   }
4052   //   idx += 2;
4053   //   if (idx > 0) {
4054   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4055   //     z[kdx+idx] = (jlong)yz_idx1;
4056   //     carry  = (jlong)(yz_idx1 >>> 64);
4057   //   }
4058   //
4059 
4060   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4061 
4062   lsrw(jdx, idx, 2);
4063 
4064   bind(L_third_loop);
4065 
4066   subsw(jdx, jdx, 1);
4067   br(Assembler::MI, L_third_loop_exit);
4068   subw(idx, idx, 4);
4069 
4070   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4071 
4072   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4073 
4074   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4075 
4076   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4077   ror(yz_idx2, yz_idx2, 32);
4078 
4079   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4080 
4081   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4082   umulh(tmp4, product_hi, yz_idx1);
4083 
4084   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4085   ror(rscratch2, rscratch2, 32);
4086 
4087   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4088   umulh(carry2, product_hi, yz_idx2);
4089 
4090   // propagate sum of both multiplications into carry:tmp4:tmp3
4091   adds(tmp3, tmp3, carry);
4092   adc(tmp4, tmp4, zr);
4093   adds(tmp3, tmp3, rscratch1);
4094   adcs(tmp4, tmp4, tmp);
4095   adc(carry, carry2, zr);
4096   adds(tmp4, tmp4, rscratch2);
4097   adc(carry, carry, zr);
4098 
4099   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4100   ror(tmp4, tmp4, 32);
4101   stp(tmp4, tmp3, Address(tmp6, 0));
4102 
4103   b(L_third_loop);
4104   bind (L_third_loop_exit);
4105 
4106   andw (idx, idx, 0x3);
4107   cbz(idx, L_post_third_loop_done);
4108 
4109   Label L_check_1;
4110   subsw(idx, idx, 2);
4111   br(Assembler::MI, L_check_1);
4112 
4113   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4114   ldr(yz_idx1, Address(rscratch1, 0));
4115   ror(yz_idx1, yz_idx1, 32);
4116   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4117   umulh(tmp4, product_hi, yz_idx1);
4118   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4119   ldr(yz_idx2, Address(rscratch1, 0));
4120   ror(yz_idx2, yz_idx2, 32);
4121 
4122   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4123 
4124   ror(tmp3, tmp3, 32);
4125   str(tmp3, Address(rscratch1, 0));
4126 
4127   bind (L_check_1);
4128 
4129   andw (idx, idx, 0x1);
4130   subsw(idx, idx, 1);
4131   br(Assembler::MI, L_post_third_loop_done);
4132   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4133   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4134   umulh(carry2, tmp4, product_hi);
4135   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4136 
4137   add2_with_carry(carry2, tmp3, tmp4, carry);
4138 
4139   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4140   extr(carry, carry2, tmp3, 32);
4141 
4142   bind(L_post_third_loop_done);
4143 }
4144 
4145 /**
4146  * Code for BigInteger::multiplyToLen() intrinsic.
4147  *
4148  * r0: x
4149  * r1: xlen
4150  * r2: y
4151  * r3: ylen
4152  * r4:  z
4153  * r5: tmp0
4154  * r10: tmp1
4155  * r11: tmp2
4156  * r12: tmp3
4157  * r13: tmp4
4158  * r14: tmp5
4159  * r15: tmp6
4160  * r16: tmp7
4161  *
4162  */
4163 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4164                                      Register z, Register tmp0,
4165                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4166                                      Register tmp5, Register tmp6, Register product_hi) {
4167 
4168   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4169 
4170   const Register idx = tmp1;
4171   const Register kdx = tmp2;
4172   const Register xstart = tmp3;
4173 
4174   const Register y_idx = tmp4;
4175   const Register carry = tmp5;
4176   const Register product  = xlen;
4177   const Register x_xstart = tmp0;
4178 
4179   // First Loop.
4180   //
4181   //  final static long LONG_MASK = 0xffffffffL;
4182   //  int xstart = xlen - 1;
4183   //  int ystart = ylen - 1;
4184   //  long carry = 0;
4185   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4186   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4187   //    z[kdx] = (int)product;
4188   //    carry = product >>> 32;
4189   //  }
4190   //  z[xstart] = (int)carry;
4191   //
4192 
4193   movw(idx, ylen);       // idx = ylen;
4194   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4195   mov(carry, zr);        // carry = 0;
4196 
4197   Label L_done;
4198 
4199   movw(xstart, xlen);
4200   subsw(xstart, xstart, 1);
4201   br(Assembler::MI, L_done);
4202 
4203   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4204 
4205   Label L_second_loop;
4206   cbzw(kdx, L_second_loop);
4207 
4208   Label L_carry;
4209   subw(kdx, kdx, 1);
4210   cbzw(kdx, L_carry);
4211 
4212   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4213   lsr(carry, carry, 32);
4214   subw(kdx, kdx, 1);
4215 
4216   bind(L_carry);
4217   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4218 
4219   // Second and third (nested) loops.
4220   //
4221   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4222   //   carry = 0;
4223   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4224   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4225   //                    (z[k] & LONG_MASK) + carry;
4226   //     z[k] = (int)product;
4227   //     carry = product >>> 32;
4228   //   }
4229   //   z[i] = (int)carry;
4230   // }
4231   //
4232   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4233 
4234   const Register jdx = tmp1;
4235 
4236   bind(L_second_loop);
4237   mov(carry, zr);                // carry = 0;
4238   movw(jdx, ylen);               // j = ystart+1
4239 
4240   subsw(xstart, xstart, 1);      // i = xstart-1;
4241   br(Assembler::MI, L_done);
4242 
4243   str(z, Address(pre(sp, -4 * wordSize)));
4244 
4245   Label L_last_x;
4246   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4247   subsw(xstart, xstart, 1);       // i = xstart-1;
4248   br(Assembler::MI, L_last_x);
4249 
4250   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4251   ldr(product_hi, Address(rscratch1));
4252   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4253 
4254   Label L_third_loop_prologue;
4255   bind(L_third_loop_prologue);
4256 
4257   str(ylen, Address(sp, wordSize));
4258   stp(x, xstart, Address(sp, 2 * wordSize));
4259   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4260                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4261   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4262   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4263 
4264   addw(tmp3, xlen, 1);
4265   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4266   subsw(tmp3, tmp3, 1);
4267   br(Assembler::MI, L_done);
4268 
4269   lsr(carry, carry, 32);
4270   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4271   b(L_second_loop);
4272 
4273   // Next infrequent code is moved outside loops.
4274   bind(L_last_x);
4275   ldrw(product_hi, Address(x,  0));
4276   b(L_third_loop_prologue);
4277 
4278   bind(L_done);
4279 }
4280 
4281 // Code for BigInteger::mulAdd intrinsic
4282 // out     = r0
4283 // in      = r1
4284 // offset  = r2  (already out.length-offset)
4285 // len     = r3
4286 // k       = r4
4287 //
4288 // pseudo code from java implementation:
4289 // carry = 0;
4290 // offset = out.length-offset - 1;
4291 // for (int j=len-1; j >= 0; j--) {
4292 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4293 //     out[offset--] = (int)product;
4294 //     carry = product >>> 32;
4295 // }
4296 // return (int)carry;
4297 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4298       Register len, Register k) {
4299     Label LOOP, END;
4300     // pre-loop
4301     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4302     csel(out, zr, out, Assembler::EQ);
4303     br(Assembler::EQ, END);
4304     add(in, in, len, LSL, 2); // in[j+1] address
4305     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4306     mov(out, zr); // used to keep carry now
4307     BIND(LOOP);
4308     ldrw(rscratch1, Address(pre(in, -4)));
4309     madd(rscratch1, rscratch1, k, out);
4310     ldrw(rscratch2, Address(pre(offset, -4)));
4311     add(rscratch1, rscratch1, rscratch2);
4312     strw(rscratch1, Address(offset));
4313     lsr(out, rscratch1, 32);
4314     subs(len, len, 1);
4315     br(Assembler::NE, LOOP);
4316     BIND(END);
4317 }
4318 
4319 /**
4320  * Emits code to update CRC-32 with a byte value according to constants in table
4321  *
4322  * @param [in,out]crc   Register containing the crc.
4323  * @param [in]val       Register containing the byte to fold into the CRC.
4324  * @param [in]table     Register containing the table of crc constants.
4325  *
4326  * uint32_t crc;
4327  * val = crc_table[(val ^ crc) & 0xFF];
4328  * crc = val ^ (crc >> 8);
4329  *
4330  */
4331 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4332   eor(val, val, crc);
4333   andr(val, val, 0xff);
4334   ldrw(val, Address(table, val, Address::lsl(2)));
4335   eor(crc, val, crc, Assembler::LSR, 8);
4336 }
4337 
4338 /**
4339  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4340  *
4341  * @param [in,out]crc   Register containing the crc.
4342  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4343  * @param [in]table0    Register containing table 0 of crc constants.
4344  * @param [in]table1    Register containing table 1 of crc constants.
4345  * @param [in]table2    Register containing table 2 of crc constants.
4346  * @param [in]table3    Register containing table 3 of crc constants.
4347  *
4348  * uint32_t crc;
4349  *   v = crc ^ v
4350  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4351  *
4352  */
4353 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4354         Register table0, Register table1, Register table2, Register table3,
4355         bool upper) {
4356   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4357   uxtb(tmp, v);
4358   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4359   ubfx(tmp, v, 8, 8);
4360   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4361   eor(crc, crc, tmp);
4362   ubfx(tmp, v, 16, 8);
4363   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4364   eor(crc, crc, tmp);
4365   ubfx(tmp, v, 24, 8);
4366   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4367   eor(crc, crc, tmp);
4368 }
4369 
4370 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4371         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4372     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4373     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4374 
4375     subs(tmp0, len, 384);
4376     mvnw(crc, crc);
4377     br(Assembler::GE, CRC_by128_pre);
4378   BIND(CRC_less128);
4379     subs(len, len, 32);
4380     br(Assembler::GE, CRC_by32_loop);
4381   BIND(CRC_less32);
4382     adds(len, len, 32 - 4);
4383     br(Assembler::GE, CRC_by4_loop);
4384     adds(len, len, 4);
4385     br(Assembler::GT, CRC_by1_loop);
4386     b(L_exit);
4387 
4388   BIND(CRC_by32_loop);
4389     ldp(tmp0, tmp1, Address(buf));
4390     crc32x(crc, crc, tmp0);
4391     ldp(tmp2, tmp3, Address(buf, 16));
4392     crc32x(crc, crc, tmp1);
4393     add(buf, buf, 32);
4394     crc32x(crc, crc, tmp2);
4395     subs(len, len, 32);
4396     crc32x(crc, crc, tmp3);
4397     br(Assembler::GE, CRC_by32_loop);
4398     cmn(len, (u1)32);
4399     br(Assembler::NE, CRC_less32);
4400     b(L_exit);
4401 
4402   BIND(CRC_by4_loop);
4403     ldrw(tmp0, Address(post(buf, 4)));
4404     subs(len, len, 4);
4405     crc32w(crc, crc, tmp0);
4406     br(Assembler::GE, CRC_by4_loop);
4407     adds(len, len, 4);
4408     br(Assembler::LE, L_exit);
4409   BIND(CRC_by1_loop);
4410     ldrb(tmp0, Address(post(buf, 1)));
4411     subs(len, len, 1);
4412     crc32b(crc, crc, tmp0);
4413     br(Assembler::GT, CRC_by1_loop);
4414     b(L_exit);
4415 
4416   BIND(CRC_by128_pre);
4417     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4418       4*256*sizeof(juint) + 8*sizeof(juint));
4419     mov(crc, 0);
4420     crc32x(crc, crc, tmp0);
4421     crc32x(crc, crc, tmp1);
4422 
4423     cbnz(len, CRC_less128);
4424 
4425   BIND(L_exit);
4426     mvnw(crc, crc);
4427 }
4428 
4429 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4430         Register len, Register tmp0, Register tmp1, Register tmp2,
4431         Register tmp3) {
4432     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4433     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4434 
4435     mvnw(crc, crc);
4436 
4437     subs(len, len, 128);
4438     br(Assembler::GE, CRC_by64_pre);
4439   BIND(CRC_less64);
4440     adds(len, len, 128-32);
4441     br(Assembler::GE, CRC_by32_loop);
4442   BIND(CRC_less32);
4443     adds(len, len, 32-4);
4444     br(Assembler::GE, CRC_by4_loop);
4445     adds(len, len, 4);
4446     br(Assembler::GT, CRC_by1_loop);
4447     b(L_exit);
4448 
4449   BIND(CRC_by32_loop);
4450     ldp(tmp0, tmp1, Address(post(buf, 16)));
4451     subs(len, len, 32);
4452     crc32x(crc, crc, tmp0);
4453     ldr(tmp2, Address(post(buf, 8)));
4454     crc32x(crc, crc, tmp1);
4455     ldr(tmp3, Address(post(buf, 8)));
4456     crc32x(crc, crc, tmp2);
4457     crc32x(crc, crc, tmp3);
4458     br(Assembler::GE, CRC_by32_loop);
4459     cmn(len, (u1)32);
4460     br(Assembler::NE, CRC_less32);
4461     b(L_exit);
4462 
4463   BIND(CRC_by4_loop);
4464     ldrw(tmp0, Address(post(buf, 4)));
4465     subs(len, len, 4);
4466     crc32w(crc, crc, tmp0);
4467     br(Assembler::GE, CRC_by4_loop);
4468     adds(len, len, 4);
4469     br(Assembler::LE, L_exit);
4470   BIND(CRC_by1_loop);
4471     ldrb(tmp0, Address(post(buf, 1)));
4472     subs(len, len, 1);
4473     crc32b(crc, crc, tmp0);
4474     br(Assembler::GT, CRC_by1_loop);
4475     b(L_exit);
4476 
4477   BIND(CRC_by64_pre);
4478     sub(buf, buf, 8);
4479     ldp(tmp0, tmp1, Address(buf, 8));
4480     crc32x(crc, crc, tmp0);
4481     ldr(tmp2, Address(buf, 24));
4482     crc32x(crc, crc, tmp1);
4483     ldr(tmp3, Address(buf, 32));
4484     crc32x(crc, crc, tmp2);
4485     ldr(tmp0, Address(buf, 40));
4486     crc32x(crc, crc, tmp3);
4487     ldr(tmp1, Address(buf, 48));
4488     crc32x(crc, crc, tmp0);
4489     ldr(tmp2, Address(buf, 56));
4490     crc32x(crc, crc, tmp1);
4491     ldr(tmp3, Address(pre(buf, 64)));
4492 
4493     b(CRC_by64_loop);
4494 
4495     align(CodeEntryAlignment);
4496   BIND(CRC_by64_loop);
4497     subs(len, len, 64);
4498     crc32x(crc, crc, tmp2);
4499     ldr(tmp0, Address(buf, 8));
4500     crc32x(crc, crc, tmp3);
4501     ldr(tmp1, Address(buf, 16));
4502     crc32x(crc, crc, tmp0);
4503     ldr(tmp2, Address(buf, 24));
4504     crc32x(crc, crc, tmp1);
4505     ldr(tmp3, Address(buf, 32));
4506     crc32x(crc, crc, tmp2);
4507     ldr(tmp0, Address(buf, 40));
4508     crc32x(crc, crc, tmp3);
4509     ldr(tmp1, Address(buf, 48));
4510     crc32x(crc, crc, tmp0);
4511     ldr(tmp2, Address(buf, 56));
4512     crc32x(crc, crc, tmp1);
4513     ldr(tmp3, Address(pre(buf, 64)));
4514     br(Assembler::GE, CRC_by64_loop);
4515 
4516     // post-loop
4517     crc32x(crc, crc, tmp2);
4518     crc32x(crc, crc, tmp3);
4519 
4520     sub(len, len, 64);
4521     add(buf, buf, 8);
4522     cmn(len, (u1)128);
4523     br(Assembler::NE, CRC_less64);
4524   BIND(L_exit);
4525     mvnw(crc, crc);
4526 }
4527 
4528 /**
4529  * @param crc   register containing existing CRC (32-bit)
4530  * @param buf   register pointing to input byte buffer (byte*)
4531  * @param len   register containing number of bytes
4532  * @param table register that will contain address of CRC table
4533  * @param tmp   scratch register
4534  */
4535 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4536         Register table0, Register table1, Register table2, Register table3,
4537         Register tmp, Register tmp2, Register tmp3) {
4538   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4539 
4540   if (UseCryptoPmullForCRC32) {
4541       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4542       return;
4543   }
4544 
4545   if (UseCRC32) {
4546       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4547       return;
4548   }
4549 
4550     mvnw(crc, crc);
4551 
4552     {
4553       uint64_t offset;
4554       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4555       add(table0, table0, offset);
4556     }
4557     add(table1, table0, 1*256*sizeof(juint));
4558     add(table2, table0, 2*256*sizeof(juint));
4559     add(table3, table0, 3*256*sizeof(juint));
4560 
4561     { // Neon code start
4562       cmp(len, (u1)64);
4563       br(Assembler::LT, L_by16);
4564       eor(v16, T16B, v16, v16);
4565 
4566     Label L_fold;
4567 
4568       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4569 
4570       ld1(v0, v1, T2D, post(buf, 32));
4571       ld1r(v4, T2D, post(tmp, 8));
4572       ld1r(v5, T2D, post(tmp, 8));
4573       ld1r(v6, T2D, post(tmp, 8));
4574       ld1r(v7, T2D, post(tmp, 8));
4575       mov(v16, S, 0, crc);
4576 
4577       eor(v0, T16B, v0, v16);
4578       sub(len, len, 64);
4579 
4580     BIND(L_fold);
4581       pmull(v22, T8H, v0, v5, T8B);
4582       pmull(v20, T8H, v0, v7, T8B);
4583       pmull(v23, T8H, v0, v4, T8B);
4584       pmull(v21, T8H, v0, v6, T8B);
4585 
4586       pmull2(v18, T8H, v0, v5, T16B);
4587       pmull2(v16, T8H, v0, v7, T16B);
4588       pmull2(v19, T8H, v0, v4, T16B);
4589       pmull2(v17, T8H, v0, v6, T16B);
4590 
4591       uzp1(v24, T8H, v20, v22);
4592       uzp2(v25, T8H, v20, v22);
4593       eor(v20, T16B, v24, v25);
4594 
4595       uzp1(v26, T8H, v16, v18);
4596       uzp2(v27, T8H, v16, v18);
4597       eor(v16, T16B, v26, v27);
4598 
4599       ushll2(v22, T4S, v20, T8H, 8);
4600       ushll(v20, T4S, v20, T4H, 8);
4601 
4602       ushll2(v18, T4S, v16, T8H, 8);
4603       ushll(v16, T4S, v16, T4H, 8);
4604 
4605       eor(v22, T16B, v23, v22);
4606       eor(v18, T16B, v19, v18);
4607       eor(v20, T16B, v21, v20);
4608       eor(v16, T16B, v17, v16);
4609 
4610       uzp1(v17, T2D, v16, v20);
4611       uzp2(v21, T2D, v16, v20);
4612       eor(v17, T16B, v17, v21);
4613 
4614       ushll2(v20, T2D, v17, T4S, 16);
4615       ushll(v16, T2D, v17, T2S, 16);
4616 
4617       eor(v20, T16B, v20, v22);
4618       eor(v16, T16B, v16, v18);
4619 
4620       uzp1(v17, T2D, v20, v16);
4621       uzp2(v21, T2D, v20, v16);
4622       eor(v28, T16B, v17, v21);
4623 
4624       pmull(v22, T8H, v1, v5, T8B);
4625       pmull(v20, T8H, v1, v7, T8B);
4626       pmull(v23, T8H, v1, v4, T8B);
4627       pmull(v21, T8H, v1, v6, T8B);
4628 
4629       pmull2(v18, T8H, v1, v5, T16B);
4630       pmull2(v16, T8H, v1, v7, T16B);
4631       pmull2(v19, T8H, v1, v4, T16B);
4632       pmull2(v17, T8H, v1, v6, T16B);
4633 
4634       ld1(v0, v1, T2D, post(buf, 32));
4635 
4636       uzp1(v24, T8H, v20, v22);
4637       uzp2(v25, T8H, v20, v22);
4638       eor(v20, T16B, v24, v25);
4639 
4640       uzp1(v26, T8H, v16, v18);
4641       uzp2(v27, T8H, v16, v18);
4642       eor(v16, T16B, v26, v27);
4643 
4644       ushll2(v22, T4S, v20, T8H, 8);
4645       ushll(v20, T4S, v20, T4H, 8);
4646 
4647       ushll2(v18, T4S, v16, T8H, 8);
4648       ushll(v16, T4S, v16, T4H, 8);
4649 
4650       eor(v22, T16B, v23, v22);
4651       eor(v18, T16B, v19, v18);
4652       eor(v20, T16B, v21, v20);
4653       eor(v16, T16B, v17, v16);
4654 
4655       uzp1(v17, T2D, v16, v20);
4656       uzp2(v21, T2D, v16, v20);
4657       eor(v16, T16B, v17, v21);
4658 
4659       ushll2(v20, T2D, v16, T4S, 16);
4660       ushll(v16, T2D, v16, T2S, 16);
4661 
4662       eor(v20, T16B, v22, v20);
4663       eor(v16, T16B, v16, v18);
4664 
4665       uzp1(v17, T2D, v20, v16);
4666       uzp2(v21, T2D, v20, v16);
4667       eor(v20, T16B, v17, v21);
4668 
4669       shl(v16, T2D, v28, 1);
4670       shl(v17, T2D, v20, 1);
4671 
4672       eor(v0, T16B, v0, v16);
4673       eor(v1, T16B, v1, v17);
4674 
4675       subs(len, len, 32);
4676       br(Assembler::GE, L_fold);
4677 
4678       mov(crc, 0);
4679       mov(tmp, v0, D, 0);
4680       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4681       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4682       mov(tmp, v0, D, 1);
4683       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4684       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4685       mov(tmp, v1, D, 0);
4686       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4687       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4688       mov(tmp, v1, D, 1);
4689       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4690       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4691 
4692       add(len, len, 32);
4693     } // Neon code end
4694 
4695   BIND(L_by16);
4696     subs(len, len, 16);
4697     br(Assembler::GE, L_by16_loop);
4698     adds(len, len, 16-4);
4699     br(Assembler::GE, L_by4_loop);
4700     adds(len, len, 4);
4701     br(Assembler::GT, L_by1_loop);
4702     b(L_exit);
4703 
4704   BIND(L_by4_loop);
4705     ldrw(tmp, Address(post(buf, 4)));
4706     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4707     subs(len, len, 4);
4708     br(Assembler::GE, L_by4_loop);
4709     adds(len, len, 4);
4710     br(Assembler::LE, L_exit);
4711   BIND(L_by1_loop);
4712     subs(len, len, 1);
4713     ldrb(tmp, Address(post(buf, 1)));
4714     update_byte_crc32(crc, tmp, table0);
4715     br(Assembler::GT, L_by1_loop);
4716     b(L_exit);
4717 
4718     align(CodeEntryAlignment);
4719   BIND(L_by16_loop);
4720     subs(len, len, 16);
4721     ldp(tmp, tmp3, Address(post(buf, 16)));
4722     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4723     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4724     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4725     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4726     br(Assembler::GE, L_by16_loop);
4727     adds(len, len, 16-4);
4728     br(Assembler::GE, L_by4_loop);
4729     adds(len, len, 4);
4730     br(Assembler::GT, L_by1_loop);
4731   BIND(L_exit);
4732     mvnw(crc, crc);
4733 }
4734 
4735 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4736         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4737     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4738     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4739 
4740     subs(tmp0, len, 384);
4741     br(Assembler::GE, CRC_by128_pre);
4742   BIND(CRC_less128);
4743     subs(len, len, 32);
4744     br(Assembler::GE, CRC_by32_loop);
4745   BIND(CRC_less32);
4746     adds(len, len, 32 - 4);
4747     br(Assembler::GE, CRC_by4_loop);
4748     adds(len, len, 4);
4749     br(Assembler::GT, CRC_by1_loop);
4750     b(L_exit);
4751 
4752   BIND(CRC_by32_loop);
4753     ldp(tmp0, tmp1, Address(buf));
4754     crc32cx(crc, crc, tmp0);
4755     ldr(tmp2, Address(buf, 16));
4756     crc32cx(crc, crc, tmp1);
4757     ldr(tmp3, Address(buf, 24));
4758     crc32cx(crc, crc, tmp2);
4759     add(buf, buf, 32);
4760     subs(len, len, 32);
4761     crc32cx(crc, crc, tmp3);
4762     br(Assembler::GE, CRC_by32_loop);
4763     cmn(len, (u1)32);
4764     br(Assembler::NE, CRC_less32);
4765     b(L_exit);
4766 
4767   BIND(CRC_by4_loop);
4768     ldrw(tmp0, Address(post(buf, 4)));
4769     subs(len, len, 4);
4770     crc32cw(crc, crc, tmp0);
4771     br(Assembler::GE, CRC_by4_loop);
4772     adds(len, len, 4);
4773     br(Assembler::LE, L_exit);
4774   BIND(CRC_by1_loop);
4775     ldrb(tmp0, Address(post(buf, 1)));
4776     subs(len, len, 1);
4777     crc32cb(crc, crc, tmp0);
4778     br(Assembler::GT, CRC_by1_loop);
4779     b(L_exit);
4780 
4781   BIND(CRC_by128_pre);
4782     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4783       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4784     mov(crc, 0);
4785     crc32cx(crc, crc, tmp0);
4786     crc32cx(crc, crc, tmp1);
4787 
4788     cbnz(len, CRC_less128);
4789 
4790   BIND(L_exit);
4791 }
4792 
4793 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4794         Register len, Register tmp0, Register tmp1, Register tmp2,
4795         Register tmp3) {
4796     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4797     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4798 
4799     subs(len, len, 128);
4800     br(Assembler::GE, CRC_by64_pre);
4801   BIND(CRC_less64);
4802     adds(len, len, 128-32);
4803     br(Assembler::GE, CRC_by32_loop);
4804   BIND(CRC_less32);
4805     adds(len, len, 32-4);
4806     br(Assembler::GE, CRC_by4_loop);
4807     adds(len, len, 4);
4808     br(Assembler::GT, CRC_by1_loop);
4809     b(L_exit);
4810 
4811   BIND(CRC_by32_loop);
4812     ldp(tmp0, tmp1, Address(post(buf, 16)));
4813     subs(len, len, 32);
4814     crc32cx(crc, crc, tmp0);
4815     ldr(tmp2, Address(post(buf, 8)));
4816     crc32cx(crc, crc, tmp1);
4817     ldr(tmp3, Address(post(buf, 8)));
4818     crc32cx(crc, crc, tmp2);
4819     crc32cx(crc, crc, tmp3);
4820     br(Assembler::GE, CRC_by32_loop);
4821     cmn(len, (u1)32);
4822     br(Assembler::NE, CRC_less32);
4823     b(L_exit);
4824 
4825   BIND(CRC_by4_loop);
4826     ldrw(tmp0, Address(post(buf, 4)));
4827     subs(len, len, 4);
4828     crc32cw(crc, crc, tmp0);
4829     br(Assembler::GE, CRC_by4_loop);
4830     adds(len, len, 4);
4831     br(Assembler::LE, L_exit);
4832   BIND(CRC_by1_loop);
4833     ldrb(tmp0, Address(post(buf, 1)));
4834     subs(len, len, 1);
4835     crc32cb(crc, crc, tmp0);
4836     br(Assembler::GT, CRC_by1_loop);
4837     b(L_exit);
4838 
4839   BIND(CRC_by64_pre);
4840     sub(buf, buf, 8);
4841     ldp(tmp0, tmp1, Address(buf, 8));
4842     crc32cx(crc, crc, tmp0);
4843     ldr(tmp2, Address(buf, 24));
4844     crc32cx(crc, crc, tmp1);
4845     ldr(tmp3, Address(buf, 32));
4846     crc32cx(crc, crc, tmp2);
4847     ldr(tmp0, Address(buf, 40));
4848     crc32cx(crc, crc, tmp3);
4849     ldr(tmp1, Address(buf, 48));
4850     crc32cx(crc, crc, tmp0);
4851     ldr(tmp2, Address(buf, 56));
4852     crc32cx(crc, crc, tmp1);
4853     ldr(tmp3, Address(pre(buf, 64)));
4854 
4855     b(CRC_by64_loop);
4856 
4857     align(CodeEntryAlignment);
4858   BIND(CRC_by64_loop);
4859     subs(len, len, 64);
4860     crc32cx(crc, crc, tmp2);
4861     ldr(tmp0, Address(buf, 8));
4862     crc32cx(crc, crc, tmp3);
4863     ldr(tmp1, Address(buf, 16));
4864     crc32cx(crc, crc, tmp0);
4865     ldr(tmp2, Address(buf, 24));
4866     crc32cx(crc, crc, tmp1);
4867     ldr(tmp3, Address(buf, 32));
4868     crc32cx(crc, crc, tmp2);
4869     ldr(tmp0, Address(buf, 40));
4870     crc32cx(crc, crc, tmp3);
4871     ldr(tmp1, Address(buf, 48));
4872     crc32cx(crc, crc, tmp0);
4873     ldr(tmp2, Address(buf, 56));
4874     crc32cx(crc, crc, tmp1);
4875     ldr(tmp3, Address(pre(buf, 64)));
4876     br(Assembler::GE, CRC_by64_loop);
4877 
4878     // post-loop
4879     crc32cx(crc, crc, tmp2);
4880     crc32cx(crc, crc, tmp3);
4881 
4882     sub(len, len, 64);
4883     add(buf, buf, 8);
4884     cmn(len, (u1)128);
4885     br(Assembler::NE, CRC_less64);
4886   BIND(L_exit);
4887 }
4888 
4889 /**
4890  * @param crc   register containing existing CRC (32-bit)
4891  * @param buf   register pointing to input byte buffer (byte*)
4892  * @param len   register containing number of bytes
4893  * @param table register that will contain address of CRC table
4894  * @param tmp   scratch register
4895  */
4896 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4897         Register table0, Register table1, Register table2, Register table3,
4898         Register tmp, Register tmp2, Register tmp3) {
4899   if (UseCryptoPmullForCRC32) {
4900     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4901   } else {
4902     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4903   }
4904 }
4905 
4906 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4907         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4908     Label CRC_by128_loop;
4909     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4910 
4911     sub(len, len, 256);
4912     Register table = tmp0;
4913     {
4914       uint64_t offset;
4915       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4916       add(table, table, offset);
4917     }
4918     add(table, table, table_offset);
4919 
4920     // Registers v0..v7 are used as data registers.
4921     // Registers v16..v31 are used as tmp registers.
4922     sub(buf, buf, 0x10);
4923     ldrq(v0, Address(buf, 0x10));
4924     ldrq(v1, Address(buf, 0x20));
4925     ldrq(v2, Address(buf, 0x30));
4926     ldrq(v3, Address(buf, 0x40));
4927     ldrq(v4, Address(buf, 0x50));
4928     ldrq(v5, Address(buf, 0x60));
4929     ldrq(v6, Address(buf, 0x70));
4930     ldrq(v7, Address(pre(buf, 0x80)));
4931 
4932     movi(v31, T4S, 0);
4933     mov(v31, S, 0, crc);
4934     eor(v0, T16B, v0, v31);
4935 
4936     // Register v16 contains constants from the crc table.
4937     ldrq(v16, Address(table));
4938     b(CRC_by128_loop);
4939 
4940     align(OptoLoopAlignment);
4941   BIND(CRC_by128_loop);
4942     pmull (v17,  T1Q, v0, v16, T1D);
4943     pmull2(v18, T1Q, v0, v16, T2D);
4944     ldrq(v0, Address(buf, 0x10));
4945     eor3(v0, T16B, v17,  v18, v0);
4946 
4947     pmull (v19, T1Q, v1, v16, T1D);
4948     pmull2(v20, T1Q, v1, v16, T2D);
4949     ldrq(v1, Address(buf, 0x20));
4950     eor3(v1, T16B, v19, v20, v1);
4951 
4952     pmull (v21, T1Q, v2, v16, T1D);
4953     pmull2(v22, T1Q, v2, v16, T2D);
4954     ldrq(v2, Address(buf, 0x30));
4955     eor3(v2, T16B, v21, v22, v2);
4956 
4957     pmull (v23, T1Q, v3, v16, T1D);
4958     pmull2(v24, T1Q, v3, v16, T2D);
4959     ldrq(v3, Address(buf, 0x40));
4960     eor3(v3, T16B, v23, v24, v3);
4961 
4962     pmull (v25, T1Q, v4, v16, T1D);
4963     pmull2(v26, T1Q, v4, v16, T2D);
4964     ldrq(v4, Address(buf, 0x50));
4965     eor3(v4, T16B, v25, v26, v4);
4966 
4967     pmull (v27, T1Q, v5, v16, T1D);
4968     pmull2(v28, T1Q, v5, v16, T2D);
4969     ldrq(v5, Address(buf, 0x60));
4970     eor3(v5, T16B, v27, v28, v5);
4971 
4972     pmull (v29, T1Q, v6, v16, T1D);
4973     pmull2(v30, T1Q, v6, v16, T2D);
4974     ldrq(v6, Address(buf, 0x70));
4975     eor3(v6, T16B, v29, v30, v6);
4976 
4977     // Reuse registers v23, v24.
4978     // Using them won't block the first instruction of the next iteration.
4979     pmull (v23, T1Q, v7, v16, T1D);
4980     pmull2(v24, T1Q, v7, v16, T2D);
4981     ldrq(v7, Address(pre(buf, 0x80)));
4982     eor3(v7, T16B, v23, v24, v7);
4983 
4984     subs(len, len, 0x80);
4985     br(Assembler::GE, CRC_by128_loop);
4986 
4987     // fold into 512 bits
4988     // Use v31 for constants because v16 can be still in use.
4989     ldrq(v31, Address(table, 0x10));
4990 
4991     pmull (v17,  T1Q, v0, v31, T1D);
4992     pmull2(v18, T1Q, v0, v31, T2D);
4993     eor3(v0, T16B, v17, v18, v4);
4994 
4995     pmull (v19, T1Q, v1, v31, T1D);
4996     pmull2(v20, T1Q, v1, v31, T2D);
4997     eor3(v1, T16B, v19, v20, v5);
4998 
4999     pmull (v21, T1Q, v2, v31, T1D);
5000     pmull2(v22, T1Q, v2, v31, T2D);
5001     eor3(v2, T16B, v21, v22, v6);
5002 
5003     pmull (v23, T1Q, v3, v31, T1D);
5004     pmull2(v24, T1Q, v3, v31, T2D);
5005     eor3(v3, T16B, v23, v24, v7);
5006 
5007     // fold into 128 bits
5008     // Use v17 for constants because v31 can be still in use.
5009     ldrq(v17, Address(table, 0x20));
5010     pmull (v25, T1Q, v0, v17, T1D);
5011     pmull2(v26, T1Q, v0, v17, T2D);
5012     eor3(v3, T16B, v3, v25, v26);
5013 
5014     // Use v18 for constants because v17 can be still in use.
5015     ldrq(v18, Address(table, 0x30));
5016     pmull (v27, T1Q, v1, v18, T1D);
5017     pmull2(v28, T1Q, v1, v18, T2D);
5018     eor3(v3, T16B, v3, v27, v28);
5019 
5020     // Use v19 for constants because v18 can be still in use.
5021     ldrq(v19, Address(table, 0x40));
5022     pmull (v29, T1Q, v2, v19, T1D);
5023     pmull2(v30, T1Q, v2, v19, T2D);
5024     eor3(v0, T16B, v3, v29, v30);
5025 
5026     add(len, len, 0x80);
5027     add(buf, buf, 0x10);
5028 
5029     mov(tmp0, v0, D, 0);
5030     mov(tmp1, v0, D, 1);
5031 }
5032 
5033 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5034   Address adr;
5035   switch(dst.getMode()) {
5036   case Address::base_plus_offset:
5037     // This is the expected mode, although we allow all the other
5038     // forms below.
5039     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5040     break;
5041   default:
5042     lea(rscratch2, dst);
5043     adr = Address(rscratch2);
5044     break;
5045   }
5046   ldr(rscratch1, adr);
5047   add(rscratch1, rscratch1, src);
5048   str(rscratch1, adr);
5049 }
5050 
5051 void MacroAssembler::cmpptr(Register src1, Address src2) {
5052   uint64_t offset;
5053   adrp(rscratch1, src2, offset);
5054   ldr(rscratch1, Address(rscratch1, offset));
5055   cmp(src1, rscratch1);
5056 }
5057 
5058 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5059   cmp(obj1, obj2);
5060 }
5061 
5062 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5063   load_method_holder(rresult, rmethod);
5064   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5065 }
5066 
5067 void MacroAssembler::load_method_holder(Register holder, Register method) {
5068   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5069   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5070   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5071 }
5072 
5073 // Loads the obj's Klass* into dst.
5074 // Preserves all registers (incl src, rscratch1 and rscratch2).
5075 // Input:
5076 // src - the oop we want to load the klass from.
5077 // dst - output narrow klass.
5078 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5079   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5080   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5081   lsr(dst, dst, markWord::klass_shift);
5082 }
5083 
5084 void MacroAssembler::load_klass(Register dst, Register src) {
5085   if (UseCompactObjectHeaders) {
5086     load_narrow_klass_compact(dst, src);
5087     decode_klass_not_null(dst);
5088   } else if (UseCompressedClassPointers) {
5089     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5090     decode_klass_not_null(dst);
5091   } else {
5092     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5093   }
5094 }
5095 
5096 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5097   if (RestoreMXCSROnJNICalls) {
5098     Label OK;
5099     get_fpcr(tmp1);
5100     mov(tmp2, tmp1);
5101     // Set FPCR to the state we need. We do want Round to Nearest. We
5102     // don't want non-IEEE rounding modes or floating-point traps.
5103     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5104     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5105     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5106     eor(tmp2, tmp1, tmp2);
5107     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5108     set_fpcr(tmp1);
5109     bind(OK);
5110   }
5111 }
5112 
5113 // ((OopHandle)result).resolve();
5114 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5115   // OopHandle::resolve is an indirection.
5116   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5117 }
5118 
5119 // ((WeakHandle)result).resolve();
5120 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5121   assert_different_registers(result, tmp1, tmp2);
5122   Label resolved;
5123 
5124   // A null weak handle resolves to null.
5125   cbz(result, resolved);
5126 
5127   // Only 64 bit platforms support GCs that require a tmp register
5128   // WeakHandle::resolve is an indirection like jweak.
5129   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5130                  result, Address(result), tmp1, tmp2);
5131   bind(resolved);
5132 }
5133 
5134 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5135   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5136   ldr(dst, Address(rmethod, Method::const_offset()));
5137   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5138   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5139   ldr(dst, Address(dst, mirror_offset));
5140   resolve_oop_handle(dst, tmp1, tmp2);
5141 }
5142 
5143 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5144   assert_different_registers(obj, klass, tmp);
5145   if (UseCompressedClassPointers) {
5146     if (UseCompactObjectHeaders) {
5147       load_narrow_klass_compact(tmp, obj);
5148     } else {
5149       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5150     }
5151     if (CompressedKlassPointers::base() == nullptr) {
5152       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5153       return;
5154     } else if (!AOTCodeCache::is_on_for_dump() &&
5155                ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5156                && CompressedKlassPointers::shift() == 0) {
5157       // Only the bottom 32 bits matter
5158       cmpw(klass, tmp);
5159       return;
5160     }
5161     decode_klass_not_null(tmp);
5162   } else {
5163     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5164   }
5165   cmp(klass, tmp);
5166 }
5167 
5168 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5169   if (UseCompactObjectHeaders) {
5170     load_narrow_klass_compact(tmp1, obj1);
5171     load_narrow_klass_compact(tmp2,  obj2);
5172     cmpw(tmp1, tmp2);
5173   } else if (UseCompressedClassPointers) {
5174     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5175     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5176     cmpw(tmp1, tmp2);
5177   } else {
5178     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5179     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5180     cmp(tmp1, tmp2);
5181   }
5182 }
5183 
5184 void MacroAssembler::store_klass(Register dst, Register src) {
5185   // FIXME: Should this be a store release?  concurrent gcs assumes
5186   // klass length is valid if klass field is not null.
5187   assert(!UseCompactObjectHeaders, "not with compact headers");
5188   if (UseCompressedClassPointers) {
5189     encode_klass_not_null(src);
5190     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5191   } else {
5192     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5193   }
5194 }
5195 
5196 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5197   assert(!UseCompactObjectHeaders, "not with compact headers");
5198   if (UseCompressedClassPointers) {
5199     // Store to klass gap in destination
5200     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5201   }
5202 }
5203 
5204 // Algorithm must match CompressedOops::encode.
5205 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5206 #ifdef ASSERT
5207   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5208 #endif
5209   verify_oop_msg(s, "broken oop in encode_heap_oop");
5210   if (CompressedOops::base() == nullptr) {
5211     if (CompressedOops::shift() != 0) {
5212       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5213       lsr(d, s, LogMinObjAlignmentInBytes);
5214     } else {
5215       mov(d, s);
5216     }
5217   } else {
5218     subs(d, s, rheapbase);
5219     csel(d, d, zr, Assembler::HS);
5220     lsr(d, d, LogMinObjAlignmentInBytes);
5221 
5222     /*  Old algorithm: is this any worse?
5223     Label nonnull;
5224     cbnz(r, nonnull);
5225     sub(r, r, rheapbase);
5226     bind(nonnull);
5227     lsr(r, r, LogMinObjAlignmentInBytes);
5228     */
5229   }
5230 }
5231 
5232 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5233 #ifdef ASSERT
5234   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5235   if (CheckCompressedOops) {
5236     Label ok;
5237     cbnz(r, ok);
5238     stop("null oop passed to encode_heap_oop_not_null");
5239     bind(ok);
5240   }
5241 #endif
5242   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5243   if (CompressedOops::base() != nullptr) {
5244     sub(r, r, rheapbase);
5245   }
5246   if (CompressedOops::shift() != 0) {
5247     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5248     lsr(r, r, LogMinObjAlignmentInBytes);
5249   }
5250 }
5251 
5252 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5253 #ifdef ASSERT
5254   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5255   if (CheckCompressedOops) {
5256     Label ok;
5257     cbnz(src, ok);
5258     stop("null oop passed to encode_heap_oop_not_null2");
5259     bind(ok);
5260   }
5261 #endif
5262   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5263 
5264   Register data = src;
5265   if (CompressedOops::base() != nullptr) {
5266     sub(dst, src, rheapbase);
5267     data = dst;
5268   }
5269   if (CompressedOops::shift() != 0) {
5270     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5271     lsr(dst, data, LogMinObjAlignmentInBytes);
5272     data = dst;
5273   }
5274   if (data == src)
5275     mov(dst, src);
5276 }
5277 
5278 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5279 #ifdef ASSERT
5280   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5281 #endif
5282   if (CompressedOops::base() == nullptr) {
5283     if (CompressedOops::shift() != 0) {
5284       lsl(d, s, CompressedOops::shift());
5285     } else if (d != s) {
5286       mov(d, s);
5287     }
5288   } else {
5289     Label done;
5290     if (d != s)
5291       mov(d, s);
5292     cbz(s, done);
5293     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5294     bind(done);
5295   }
5296   verify_oop_msg(d, "broken oop in decode_heap_oop");
5297 }
5298 
5299 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5300   assert (UseCompressedOops, "should only be used for compressed headers");
5301   assert (Universe::heap() != nullptr, "java heap should be initialized");
5302   // Cannot assert, unverified entry point counts instructions (see .ad file)
5303   // vtableStubs also counts instructions in pd_code_size_limit.
5304   // Also do not verify_oop as this is called by verify_oop.
5305   if (CompressedOops::shift() != 0) {
5306     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5307     if (CompressedOops::base() != nullptr) {
5308       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5309     } else {
5310       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5311     }
5312   } else {
5313     assert (CompressedOops::base() == nullptr, "sanity");
5314   }
5315 }
5316 
5317 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5318   assert (UseCompressedOops, "should only be used for compressed headers");
5319   assert (Universe::heap() != nullptr, "java heap should be initialized");
5320   // Cannot assert, unverified entry point counts instructions (see .ad file)
5321   // vtableStubs also counts instructions in pd_code_size_limit.
5322   // Also do not verify_oop as this is called by verify_oop.
5323   if (CompressedOops::shift() != 0) {
5324     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5325     if (CompressedOops::base() != nullptr) {
5326       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5327     } else {
5328       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5329     }
5330   } else {
5331     assert (CompressedOops::base() == nullptr, "sanity");
5332     if (dst != src) {
5333       mov(dst, src);
5334     }
5335   }
5336 }
5337 
5338 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5339 
5340 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5341   assert(Metaspace::initialized(), "metaspace not initialized yet");
5342   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5343   return _klass_decode_mode;
5344 }
5345 
5346 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5347   assert(UseCompressedClassPointers, "not using compressed class pointers");
5348 
5349   // KlassDecodeMode shouldn't be set already.
5350   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5351 
5352   if (base == nullptr) {
5353     return KlassDecodeZero;
5354   }
5355 
5356   if (operand_valid_for_logical_immediate(
5357         /*is32*/false, (uint64_t)base)) {
5358     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5359     if (((uint64_t)base & range_mask) == 0) {
5360       return KlassDecodeXor;
5361     }
5362   }
5363 
5364   const uint64_t shifted_base =
5365     (uint64_t)base >> shift;
5366   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5367     return KlassDecodeMovk;
5368   }
5369 
5370   // No valid encoding.
5371   return KlassDecodeNone;
5372 }
5373 
5374 // Check if one of the above decoding modes will work for given base, shift and range.
5375 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5376   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5377 }
5378 
5379 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5380   _klass_decode_mode = klass_decode_mode(base, shift, range);
5381   return _klass_decode_mode != KlassDecodeNone;
5382 }
5383 
5384 static Register pick_different_tmp(Register dst, Register src) {
5385   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5386   return *tmps.begin();
5387 }
5388 
5389 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5390   // we have to load the klass base from the AOT constants area but
5391   // not the shift because it is not allowed to change
5392   int shift = CompressedKlassPointers::shift();
5393   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5394   if (dst != src) {
5395     // we can load the base into dst, subtract it formthe src and shift down
5396     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5397     ldr(dst, dst);
5398     sub(dst, src, dst);
5399     lsr(dst, dst, shift);
5400   } else {
5401     // we need an extra register in order to load the coop base
5402     Register tmp = pick_different_tmp(dst, src);
5403     RegSet regs = RegSet::of(tmp);
5404     push(regs, sp);
5405     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5406     ldr(tmp, tmp);
5407     sub(dst, src, tmp);
5408     lsr(dst, dst, shift);
5409     pop(regs, sp);
5410   }
5411 }
5412 
5413 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5414   if (CompressedKlassPointers::base() != nullptr && AOTCodeCache::is_on_for_dump()) {
5415     encode_klass_not_null_for_aot(dst, src);
5416     return;
5417   }
5418 
5419   switch (klass_decode_mode()) {
5420   case KlassDecodeZero:
5421     if (CompressedKlassPointers::shift() != 0) {
5422       lsr(dst, src, CompressedKlassPointers::shift());
5423     } else {
5424       if (dst != src) mov(dst, src);
5425     }
5426     break;
5427 
5428   case KlassDecodeXor:
5429     if (CompressedKlassPointers::shift() != 0) {
5430       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5431       lsr(dst, dst, CompressedKlassPointers::shift());
5432     } else {
5433       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5434     }
5435     break;
5436 
5437   case KlassDecodeMovk:
5438     if (CompressedKlassPointers::shift() != 0) {
5439       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5440     } else {
5441       movw(dst, src);
5442     }
5443     break;
5444 
5445   case KlassDecodeNone:
5446     ShouldNotReachHere();
5447     break;
5448   }
5449 }
5450 
5451 void MacroAssembler::encode_klass_not_null(Register r) {
5452   encode_klass_not_null(r, r);
5453 }
5454 
5455 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5456   // we have to load the klass base from the AOT constants area but
5457   // not the shift because it is not allowed to change
5458   int shift = CompressedKlassPointers::shift();
5459   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5460   if (dst != src) {
5461     // we can load the base into dst then add the offset with a suitable shift
5462     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5463     ldr(dst, dst);
5464     add(dst, dst, src, LSL,  shift);
5465   } else {
5466     // we need an extra register in order to load the coop base
5467     Register tmp = pick_different_tmp(dst, src);
5468     RegSet regs = RegSet::of(tmp);
5469     push(regs, sp);
5470     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5471     ldr(tmp, tmp);
5472     add(dst, tmp,  src, LSL,  shift);
5473     pop(regs, sp);
5474   }
5475 }
5476 
5477 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5478   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5479 
5480   if (CompressedKlassPointers::base() != nullptr && AOTCodeCache::is_on_for_dump()) {
5481     decode_klass_not_null_for_aot(dst, src);
5482     return;
5483   }
5484 
5485   switch (klass_decode_mode()) {
5486   case KlassDecodeZero:
5487     if (CompressedKlassPointers::shift() != 0) {
5488       lsl(dst, src, CompressedKlassPointers::shift());
5489     } else {
5490       if (dst != src) mov(dst, src);
5491     }
5492     break;
5493 
5494   case KlassDecodeXor:
5495     if (CompressedKlassPointers::shift() != 0) {
5496       lsl(dst, src, CompressedKlassPointers::shift());
5497       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5498     } else {
5499       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5500     }
5501     break;
5502 
5503   case KlassDecodeMovk: {
5504     const uint64_t shifted_base =
5505       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5506 
5507     if (dst != src) movw(dst, src);
5508     movk(dst, shifted_base >> 32, 32);
5509 
5510     if (CompressedKlassPointers::shift() != 0) {
5511       lsl(dst, dst, CompressedKlassPointers::shift());
5512     }
5513 
5514     break;
5515   }
5516 
5517   case KlassDecodeNone:
5518     ShouldNotReachHere();
5519     break;
5520   }
5521 }
5522 
5523 void  MacroAssembler::decode_klass_not_null(Register r) {
5524   decode_klass_not_null(r, r);
5525 }
5526 
5527 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5528 #ifdef ASSERT
5529   {
5530     ThreadInVMfromUnknown tiv;
5531     assert (UseCompressedOops, "should only be used for compressed oops");
5532     assert (Universe::heap() != nullptr, "java heap should be initialized");
5533     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5534     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5535   }
5536 #endif
5537   int oop_index = oop_recorder()->find_index(obj);
5538   InstructionMark im(this);
5539   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5540   code_section()->relocate(inst_mark(), rspec);
5541   movz(dst, 0xDEAD, 16);
5542   movk(dst, 0xBEEF);
5543 }
5544 
5545 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5546   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5547   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5548   int index = oop_recorder()->find_index(k);
5549 
5550   InstructionMark im(this);
5551   RelocationHolder rspec = metadata_Relocation::spec(index);
5552   code_section()->relocate(inst_mark(), rspec);
5553   narrowKlass nk = CompressedKlassPointers::encode(k);
5554   movz(dst, (nk >> 16), 16);
5555   movk(dst, nk & 0xffff);
5556 }
5557 
5558 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5559                                     Register dst, Address src,
5560                                     Register tmp1, Register tmp2) {
5561   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5562   decorators = AccessInternal::decorator_fixup(decorators, type);
5563   bool as_raw = (decorators & AS_RAW) != 0;
5564   if (as_raw) {
5565     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5566   } else {
5567     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5568   }
5569 }
5570 
5571 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5572                                      Address dst, Register val,
5573                                      Register tmp1, Register tmp2, Register tmp3) {
5574   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5575   decorators = AccessInternal::decorator_fixup(decorators, type);
5576   bool as_raw = (decorators & AS_RAW) != 0;
5577   if (as_raw) {
5578     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5579   } else {
5580     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5581   }
5582 }
5583 
5584 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5585                                    Register tmp2, DecoratorSet decorators) {
5586   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5587 }
5588 
5589 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5590                                             Register tmp2, DecoratorSet decorators) {
5591   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5592 }
5593 
5594 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5595                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5596   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5597 }
5598 
5599 // Used for storing nulls.
5600 void MacroAssembler::store_heap_oop_null(Address dst) {
5601   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5602 }
5603 
5604 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5605   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5606   int index = oop_recorder()->allocate_metadata_index(obj);
5607   RelocationHolder rspec = metadata_Relocation::spec(index);
5608   return Address((address)obj, rspec);
5609 }
5610 
5611 // Move an oop into a register.
5612 void MacroAssembler::movoop(Register dst, jobject obj) {
5613   int oop_index;
5614   if (obj == nullptr) {
5615     oop_index = oop_recorder()->allocate_oop_index(obj);
5616   } else {
5617 #ifdef ASSERT
5618     {
5619       ThreadInVMfromUnknown tiv;
5620       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5621     }
5622 #endif
5623     oop_index = oop_recorder()->find_index(obj);
5624   }
5625   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5626 
5627   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5628     mov(dst, Address((address)obj, rspec));
5629   } else {
5630     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5631     ldr(dst, Address(dummy, rspec));
5632   }
5633 }
5634 
5635 // Move a metadata address into a register.
5636 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5637   int oop_index;
5638   if (obj == nullptr) {
5639     oop_index = oop_recorder()->allocate_metadata_index(obj);
5640   } else {
5641     oop_index = oop_recorder()->find_index(obj);
5642   }
5643   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5644   mov(dst, Address((address)obj, rspec));
5645 }
5646 
5647 Address MacroAssembler::constant_oop_address(jobject obj) {
5648 #ifdef ASSERT
5649   {
5650     ThreadInVMfromUnknown tiv;
5651     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5652     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5653   }
5654 #endif
5655   int oop_index = oop_recorder()->find_index(obj);
5656   return Address((address)obj, oop_Relocation::spec(oop_index));
5657 }
5658 
5659 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5660 void MacroAssembler::tlab_allocate(Register obj,
5661                                    Register var_size_in_bytes,
5662                                    int con_size_in_bytes,
5663                                    Register t1,
5664                                    Register t2,
5665                                    Label& slow_case) {
5666   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5667   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5668 }
5669 
5670 void MacroAssembler::verify_tlab() {
5671 #ifdef ASSERT
5672   if (UseTLAB && VerifyOops) {
5673     Label next, ok;
5674 
5675     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5676 
5677     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5678     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5679     cmp(rscratch2, rscratch1);
5680     br(Assembler::HS, next);
5681     STOP("assert(top >= start)");
5682     should_not_reach_here();
5683 
5684     bind(next);
5685     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5686     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5687     cmp(rscratch2, rscratch1);
5688     br(Assembler::HS, ok);
5689     STOP("assert(top <= end)");
5690     should_not_reach_here();
5691 
5692     bind(ok);
5693     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5694   }
5695 #endif
5696 }
5697 
5698 // Writes to stack successive pages until offset reached to check for
5699 // stack overflow + shadow pages.  This clobbers tmp.
5700 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5701   assert_different_registers(tmp, size, rscratch1);
5702   mov(tmp, sp);
5703   // Bang stack for total size given plus shadow page size.
5704   // Bang one page at a time because large size can bang beyond yellow and
5705   // red zones.
5706   Label loop;
5707   mov(rscratch1, (int)os::vm_page_size());
5708   bind(loop);
5709   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5710   subsw(size, size, rscratch1);
5711   str(size, Address(tmp));
5712   br(Assembler::GT, loop);
5713 
5714   // Bang down shadow pages too.
5715   // At this point, (tmp-0) is the last address touched, so don't
5716   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5717   // was post-decremented.)  Skip this address by starting at i=1, and
5718   // touch a few more pages below.  N.B.  It is important to touch all
5719   // the way down to and including i=StackShadowPages.
5720   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5721     // this could be any sized move but this is can be a debugging crumb
5722     // so the bigger the better.
5723     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5724     str(size, Address(tmp));
5725   }
5726 }
5727 
5728 // Move the address of the polling page into dest.
5729 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5730   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5731 }
5732 
5733 // Read the polling page.  The address of the polling page must
5734 // already be in r.
5735 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5736   address mark;
5737   {
5738     InstructionMark im(this);
5739     code_section()->relocate(inst_mark(), rtype);
5740     ldrw(zr, Address(r, 0));
5741     mark = inst_mark();
5742   }
5743   verify_cross_modify_fence_not_required();
5744   return mark;
5745 }
5746 
5747 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5748   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5749   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5750   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5751   int64_t offset_low = dest_page - low_page;
5752   int64_t offset_high = dest_page - high_page;
5753 
5754   assert(is_valid_AArch64_address(dest.target()), "bad address");
5755   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5756 
5757   InstructionMark im(this);
5758   code_section()->relocate(inst_mark(), dest.rspec());
5759   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5760   // the code cache so that if it is relocated we know it will still reach
5761   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5762     _adrp(reg1, dest.target());
5763   } else {
5764     uint64_t target = (uint64_t)dest.target();
5765     uint64_t adrp_target
5766       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5767 
5768     _adrp(reg1, (address)adrp_target);
5769     movk(reg1, target >> 32, 32);
5770   }
5771   byte_offset = (uint64_t)dest.target() & 0xfff;
5772 }
5773 
5774 void MacroAssembler::load_byte_map_base(Register reg) {
5775 #if INCLUDE_CDS
5776   if (AOTCodeCache::is_on_for_dump()) {
5777     address byte_map_base_adr = AOTRuntimeConstants::card_table_base_address();
5778     lea(reg, ExternalAddress(byte_map_base_adr));
5779     ldr(reg, Address(reg));
5780     return;
5781   }
5782 #endif
5783   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
5784 
5785   // Strictly speaking the card table base isn't an address at all, and it might
5786   // even be negative. It is thus materialised as a constant.
5787   mov(reg, (uint64_t)ctbs->card_table_base_const());
5788 }
5789 
5790 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5791 #if INCLUDE_CDS
5792   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5793   if (AOTCodeCache::is_on_for_dump()) {
5794     // all aotrc field addresses should be registered in the AOTCodeCache address table
5795     lea(reg, ExternalAddress(a));
5796   } else {
5797     mov(reg, (uint64_t)a);
5798   }
5799 #else
5800   ShouldNotReachHere();
5801 #endif
5802 }
5803 
5804 void MacroAssembler::build_frame(int framesize) {
5805   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5806   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5807   protect_return_address();
5808   if (framesize < ((1 << 9) + 2 * wordSize)) {
5809     sub(sp, sp, framesize);
5810     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5811     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5812   } else {
5813     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5814     if (PreserveFramePointer) mov(rfp, sp);
5815     if (framesize < ((1 << 12) + 2 * wordSize))
5816       sub(sp, sp, framesize - 2 * wordSize);
5817     else {
5818       mov(rscratch1, framesize - 2 * wordSize);
5819       sub(sp, sp, rscratch1);
5820     }
5821   }
5822   verify_cross_modify_fence_not_required();
5823 }
5824 
5825 void MacroAssembler::remove_frame(int framesize) {
5826   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5827   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5828   if (framesize < ((1 << 9) + 2 * wordSize)) {
5829     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5830     add(sp, sp, framesize);
5831   } else {
5832     if (framesize < ((1 << 12) + 2 * wordSize))
5833       add(sp, sp, framesize - 2 * wordSize);
5834     else {
5835       mov(rscratch1, framesize - 2 * wordSize);
5836       add(sp, sp, rscratch1);
5837     }
5838     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5839   }
5840   authenticate_return_address();
5841 }
5842 
5843 
5844 // This method counts leading positive bytes (highest bit not set) in provided byte array
5845 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5846     // Simple and most common case of aligned small array which is not at the
5847     // end of memory page is placed here. All other cases are in stub.
5848     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5849     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5850     assert_different_registers(ary1, len, result);
5851 
5852     mov(result, len);
5853     cmpw(len, 0);
5854     br(LE, DONE);
5855     cmpw(len, 4 * wordSize);
5856     br(GE, STUB_LONG); // size > 32 then go to stub
5857 
5858     int shift = 64 - exact_log2(os::vm_page_size());
5859     lsl(rscratch1, ary1, shift);
5860     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5861     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5862     br(CS, STUB); // at the end of page then go to stub
5863     subs(len, len, wordSize);
5864     br(LT, END);
5865 
5866   BIND(LOOP);
5867     ldr(rscratch1, Address(post(ary1, wordSize)));
5868     tst(rscratch1, UPPER_BIT_MASK);
5869     br(NE, SET_RESULT);
5870     subs(len, len, wordSize);
5871     br(GE, LOOP);
5872     cmpw(len, -wordSize);
5873     br(EQ, DONE);
5874 
5875   BIND(END);
5876     ldr(rscratch1, Address(ary1));
5877     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5878     lslv(rscratch1, rscratch1, rscratch2);
5879     tst(rscratch1, UPPER_BIT_MASK);
5880     br(NE, SET_RESULT);
5881     b(DONE);
5882 
5883   BIND(STUB);
5884     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5885     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5886     address tpc1 = trampoline_call(count_pos);
5887     if (tpc1 == nullptr) {
5888       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5889       postcond(pc() == badAddress);
5890       return nullptr;
5891     }
5892     b(DONE);
5893 
5894   BIND(STUB_LONG);
5895     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5896     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5897     address tpc2 = trampoline_call(count_pos_long);
5898     if (tpc2 == nullptr) {
5899       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5900       postcond(pc() == badAddress);
5901       return nullptr;
5902     }
5903     b(DONE);
5904 
5905   BIND(SET_RESULT);
5906 
5907     add(len, len, wordSize);
5908     sub(result, result, len);
5909 
5910   BIND(DONE);
5911   postcond(pc() != badAddress);
5912   return pc();
5913 }
5914 
5915 // Clobbers: rscratch1, rscratch2, rflags
5916 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5917 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5918                                       Register tmp4, Register tmp5, Register result,
5919                                       Register cnt1, int elem_size) {
5920   Label DONE, SAME;
5921   Register tmp1 = rscratch1;
5922   Register tmp2 = rscratch2;
5923   int elem_per_word = wordSize/elem_size;
5924   int log_elem_size = exact_log2(elem_size);
5925   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5926   int length_offset = arrayOopDesc::length_offset_in_bytes();
5927   int base_offset
5928     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5929   // When the length offset is not aligned to 8 bytes,
5930   // then we align it down. This is valid because the new
5931   // offset will always be the klass which is the same
5932   // for type arrays.
5933   int start_offset = align_down(length_offset, BytesPerWord);
5934   int extra_length = base_offset - start_offset;
5935   assert(start_offset == length_offset || start_offset == klass_offset,
5936          "start offset must be 8-byte-aligned or be the klass offset");
5937   assert(base_offset != start_offset, "must include the length field");
5938   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5939   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5940 
5941   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5942   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5943 
5944 #ifndef PRODUCT
5945   {
5946     const char kind = (elem_size == 2) ? 'U' : 'L';
5947     char comment[64];
5948     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5949     BLOCK_COMMENT(comment);
5950   }
5951 #endif
5952 
5953   // if (a1 == a2)
5954   //     return true;
5955   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5956   br(EQ, SAME);
5957 
5958   if (UseSimpleArrayEquals) {
5959     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5960     // if (a1 == nullptr || a2 == nullptr)
5961     //     return false;
5962     // a1 & a2 == 0 means (some-pointer is null) or
5963     // (very-rare-or-even-probably-impossible-pointer-values)
5964     // so, we can save one branch in most cases
5965     tst(a1, a2);
5966     mov(result, false);
5967     br(EQ, A_MIGHT_BE_NULL);
5968     // if (a1.length != a2.length)
5969     //      return false;
5970     bind(A_IS_NOT_NULL);
5971     ldrw(cnt1, Address(a1, length_offset));
5972     ldrw(tmp5, Address(a2, length_offset));
5973     cmp(cnt1, tmp5);
5974     br(NE, DONE); // If lengths differ, return false
5975     // Increase loop counter by diff between base- and actual start-offset.
5976     addw(cnt1, cnt1, extra_length);
5977     lea(a1, Address(a1, start_offset));
5978     lea(a2, Address(a2, start_offset));
5979     // Check for short strings, i.e. smaller than wordSize.
5980     subs(cnt1, cnt1, elem_per_word);
5981     br(Assembler::LT, SHORT);
5982     // Main 8 byte comparison loop.
5983     bind(NEXT_WORD); {
5984       ldr(tmp1, Address(post(a1, wordSize)));
5985       ldr(tmp2, Address(post(a2, wordSize)));
5986       subs(cnt1, cnt1, elem_per_word);
5987       eor(tmp5, tmp1, tmp2);
5988       cbnz(tmp5, DONE);
5989     } br(GT, NEXT_WORD);
5990     // Last longword.  In the case where length == 4 we compare the
5991     // same longword twice, but that's still faster than another
5992     // conditional branch.
5993     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5994     // length == 4.
5995     if (log_elem_size > 0)
5996       lsl(cnt1, cnt1, log_elem_size);
5997     ldr(tmp3, Address(a1, cnt1));
5998     ldr(tmp4, Address(a2, cnt1));
5999     eor(tmp5, tmp3, tmp4);
6000     cbnz(tmp5, DONE);
6001     b(SAME);
6002     bind(A_MIGHT_BE_NULL);
6003     // in case both a1 and a2 are not-null, proceed with loads
6004     cbz(a1, DONE);
6005     cbz(a2, DONE);
6006     b(A_IS_NOT_NULL);
6007     bind(SHORT);
6008 
6009     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
6010     {
6011       ldrw(tmp1, Address(post(a1, 4)));
6012       ldrw(tmp2, Address(post(a2, 4)));
6013       eorw(tmp5, tmp1, tmp2);
6014       cbnzw(tmp5, DONE);
6015     }
6016     bind(TAIL03);
6017     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6018     {
6019       ldrh(tmp3, Address(post(a1, 2)));
6020       ldrh(tmp4, Address(post(a2, 2)));
6021       eorw(tmp5, tmp3, tmp4);
6022       cbnzw(tmp5, DONE);
6023     }
6024     bind(TAIL01);
6025     if (elem_size == 1) { // Only needed when comparing byte arrays.
6026       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6027       {
6028         ldrb(tmp1, a1);
6029         ldrb(tmp2, a2);
6030         eorw(tmp5, tmp1, tmp2);
6031         cbnzw(tmp5, DONE);
6032       }
6033     }
6034   } else {
6035     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6036         CSET_EQ, LAST_CHECK;
6037     mov(result, false);
6038     cbz(a1, DONE);
6039     ldrw(cnt1, Address(a1, length_offset));
6040     cbz(a2, DONE);
6041     ldrw(tmp5, Address(a2, length_offset));
6042     cmp(cnt1, tmp5);
6043     br(NE, DONE); // If lengths differ, return false
6044     // Increase loop counter by diff between base- and actual start-offset.
6045     addw(cnt1, cnt1, extra_length);
6046 
6047     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6048     // faster to perform another branch before comparing a1 and a2
6049     cmp(cnt1, (u1)elem_per_word);
6050     br(LE, SHORT); // short or same
6051     ldr(tmp3, Address(pre(a1, start_offset)));
6052     subs(zr, cnt1, stubBytesThreshold);
6053     br(GE, STUB);
6054     ldr(tmp4, Address(pre(a2, start_offset)));
6055     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6056 
6057     // Main 16 byte comparison loop with 2 exits
6058     bind(NEXT_DWORD); {
6059       ldr(tmp1, Address(pre(a1, wordSize)));
6060       ldr(tmp2, Address(pre(a2, wordSize)));
6061       subs(cnt1, cnt1, 2 * elem_per_word);
6062       br(LE, TAIL);
6063       eor(tmp4, tmp3, tmp4);
6064       cbnz(tmp4, DONE);
6065       ldr(tmp3, Address(pre(a1, wordSize)));
6066       ldr(tmp4, Address(pre(a2, wordSize)));
6067       cmp(cnt1, (u1)elem_per_word);
6068       br(LE, TAIL2);
6069       cmp(tmp1, tmp2);
6070     } br(EQ, NEXT_DWORD);
6071     b(DONE);
6072 
6073     bind(TAIL);
6074     eor(tmp4, tmp3, tmp4);
6075     eor(tmp2, tmp1, tmp2);
6076     lslv(tmp2, tmp2, tmp5);
6077     orr(tmp5, tmp4, tmp2);
6078     cmp(tmp5, zr);
6079     b(CSET_EQ);
6080 
6081     bind(TAIL2);
6082     eor(tmp2, tmp1, tmp2);
6083     cbnz(tmp2, DONE);
6084     b(LAST_CHECK);
6085 
6086     bind(STUB);
6087     ldr(tmp4, Address(pre(a2, start_offset)));
6088     if (elem_size == 2) { // convert to byte counter
6089       lsl(cnt1, cnt1, 1);
6090     }
6091     eor(tmp5, tmp3, tmp4);
6092     cbnz(tmp5, DONE);
6093     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6094     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6095     address tpc = trampoline_call(stub);
6096     if (tpc == nullptr) {
6097       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6098       postcond(pc() == badAddress);
6099       return nullptr;
6100     }
6101     b(DONE);
6102 
6103     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6104     // so, if a2 == null => return false(0), else return true, so we can return a2
6105     mov(result, a2);
6106     b(DONE);
6107     bind(SHORT);
6108     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6109     ldr(tmp3, Address(a1, start_offset));
6110     ldr(tmp4, Address(a2, start_offset));
6111     bind(LAST_CHECK);
6112     eor(tmp4, tmp3, tmp4);
6113     lslv(tmp5, tmp4, tmp5);
6114     cmp(tmp5, zr);
6115     bind(CSET_EQ);
6116     cset(result, EQ);
6117     b(DONE);
6118   }
6119 
6120   bind(SAME);
6121   mov(result, true);
6122   // That's it.
6123   bind(DONE);
6124 
6125   BLOCK_COMMENT("} array_equals");
6126   postcond(pc() != badAddress);
6127   return pc();
6128 }
6129 
6130 // Compare Strings
6131 
6132 // For Strings we're passed the address of the first characters in a1
6133 // and a2 and the length in cnt1.
6134 // There are two implementations.  For arrays >= 8 bytes, all
6135 // comparisons (including the final one, which may overlap) are
6136 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6137 // halfword, then a short, and then a byte.
6138 
6139 void MacroAssembler::string_equals(Register a1, Register a2,
6140                                    Register result, Register cnt1)
6141 {
6142   Label SAME, DONE, SHORT, NEXT_WORD;
6143   Register tmp1 = rscratch1;
6144   Register tmp2 = rscratch2;
6145 
6146   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6147 
6148 #ifndef PRODUCT
6149   {
6150     char comment[64];
6151     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6152     BLOCK_COMMENT(comment);
6153   }
6154 #endif
6155 
6156   mov(result, false);
6157 
6158   // Check for short strings, i.e. smaller than wordSize.
6159   subs(cnt1, cnt1, wordSize);
6160   br(Assembler::LT, SHORT);
6161   // Main 8 byte comparison loop.
6162   bind(NEXT_WORD); {
6163     ldr(tmp1, Address(post(a1, wordSize)));
6164     ldr(tmp2, Address(post(a2, wordSize)));
6165     subs(cnt1, cnt1, wordSize);
6166     eor(tmp1, tmp1, tmp2);
6167     cbnz(tmp1, DONE);
6168   } br(GT, NEXT_WORD);
6169   // Last longword.  In the case where length == 4 we compare the
6170   // same longword twice, but that's still faster than another
6171   // conditional branch.
6172   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6173   // length == 4.
6174   ldr(tmp1, Address(a1, cnt1));
6175   ldr(tmp2, Address(a2, cnt1));
6176   eor(tmp2, tmp1, tmp2);
6177   cbnz(tmp2, DONE);
6178   b(SAME);
6179 
6180   bind(SHORT);
6181   Label TAIL03, TAIL01;
6182 
6183   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6184   {
6185     ldrw(tmp1, Address(post(a1, 4)));
6186     ldrw(tmp2, Address(post(a2, 4)));
6187     eorw(tmp1, tmp1, tmp2);
6188     cbnzw(tmp1, DONE);
6189   }
6190   bind(TAIL03);
6191   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6192   {
6193     ldrh(tmp1, Address(post(a1, 2)));
6194     ldrh(tmp2, Address(post(a2, 2)));
6195     eorw(tmp1, tmp1, tmp2);
6196     cbnzw(tmp1, DONE);
6197   }
6198   bind(TAIL01);
6199   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6200     {
6201     ldrb(tmp1, a1);
6202     ldrb(tmp2, a2);
6203     eorw(tmp1, tmp1, tmp2);
6204     cbnzw(tmp1, DONE);
6205   }
6206   // Arrays are equal.
6207   bind(SAME);
6208   mov(result, true);
6209 
6210   // That's it.
6211   bind(DONE);
6212   BLOCK_COMMENT("} string_equals");
6213 }
6214 
6215 
6216 // The size of the blocks erased by the zero_blocks stub.  We must
6217 // handle anything smaller than this ourselves in zero_words().
6218 const int MacroAssembler::zero_words_block_size = 8;
6219 
6220 // zero_words() is used by C2 ClearArray patterns and by
6221 // C1_MacroAssembler.  It is as small as possible, handling small word
6222 // counts locally and delegating anything larger to the zero_blocks
6223 // stub.  It is expanded many times in compiled code, so it is
6224 // important to keep it short.
6225 
6226 // ptr:   Address of a buffer to be zeroed.
6227 // cnt:   Count in HeapWords.
6228 //
6229 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6230 address MacroAssembler::zero_words(Register ptr, Register cnt)
6231 {
6232   assert(is_power_of_2(zero_words_block_size), "adjust this");
6233 
6234   BLOCK_COMMENT("zero_words {");
6235   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6236   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6237   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6238 
6239   subs(rscratch1, cnt, zero_words_block_size);
6240   Label around;
6241   br(LO, around);
6242   {
6243     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6244     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6245     // Make sure this is a C2 compilation. C1 allocates space only for
6246     // trampoline stubs generated by Call LIR ops, and in any case it
6247     // makes sense for a C1 compilation task to proceed as quickly as
6248     // possible.
6249     CompileTask* task;
6250     if (StubRoutines::aarch64::complete()
6251         && Thread::current()->is_Compiler_thread()
6252         && (task = ciEnv::current()->task())
6253         && is_c2_compile(task->comp_level())) {
6254       address tpc = trampoline_call(zero_blocks);
6255       if (tpc == nullptr) {
6256         DEBUG_ONLY(reset_labels(around));
6257         return nullptr;
6258       }
6259     } else {
6260       far_call(zero_blocks);
6261     }
6262   }
6263   bind(around);
6264 
6265   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6266   // for us.
6267   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6268     Label l;
6269     tbz(cnt, exact_log2(i), l);
6270     for (int j = 0; j < i; j += 2) {
6271       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6272     }
6273     bind(l);
6274   }
6275   {
6276     Label l;
6277     tbz(cnt, 0, l);
6278     str(zr, Address(ptr));
6279     bind(l);
6280   }
6281 
6282   BLOCK_COMMENT("} zero_words");
6283   return pc();
6284 }
6285 
6286 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6287 // cnt:          Immediate count in HeapWords.
6288 //
6289 // r10, r11, rscratch1, and rscratch2 are clobbered.
6290 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6291 {
6292   assert(wordSize <= BlockZeroingLowLimit,
6293             "increase BlockZeroingLowLimit");
6294   address result = nullptr;
6295   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6296 #ifndef PRODUCT
6297     {
6298       char buf[64];
6299       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6300       BLOCK_COMMENT(buf);
6301     }
6302 #endif
6303     if (cnt >= 16) {
6304       uint64_t loops = cnt/16;
6305       if (loops > 1) {
6306         mov(rscratch2, loops - 1);
6307       }
6308       {
6309         Label loop;
6310         bind(loop);
6311         for (int i = 0; i < 16; i += 2) {
6312           stp(zr, zr, Address(base, i * BytesPerWord));
6313         }
6314         add(base, base, 16 * BytesPerWord);
6315         if (loops > 1) {
6316           subs(rscratch2, rscratch2, 1);
6317           br(GE, loop);
6318         }
6319       }
6320     }
6321     cnt %= 16;
6322     int i = cnt & 1;  // store any odd word to start
6323     if (i) str(zr, Address(base));
6324     for (; i < (int)cnt; i += 2) {
6325       stp(zr, zr, Address(base, i * wordSize));
6326     }
6327     BLOCK_COMMENT("} zero_words");
6328     result = pc();
6329   } else {
6330     mov(r10, base); mov(r11, cnt);
6331     result = zero_words(r10, r11);
6332   }
6333   return result;
6334 }
6335 
6336 // Zero blocks of memory by using DC ZVA.
6337 //
6338 // Aligns the base address first sufficiently for DC ZVA, then uses
6339 // DC ZVA repeatedly for every full block.  cnt is the size to be
6340 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6341 // in cnt.
6342 //
6343 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6344 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6345 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6346   Register tmp = rscratch1;
6347   Register tmp2 = rscratch2;
6348   int zva_length = VM_Version::zva_length();
6349   Label initial_table_end, loop_zva;
6350   Label fini;
6351 
6352   // Base must be 16 byte aligned. If not just return and let caller handle it
6353   tst(base, 0x0f);
6354   br(Assembler::NE, fini);
6355   // Align base with ZVA length.
6356   neg(tmp, base);
6357   andr(tmp, tmp, zva_length - 1);
6358 
6359   // tmp: the number of bytes to be filled to align the base with ZVA length.
6360   add(base, base, tmp);
6361   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6362   adr(tmp2, initial_table_end);
6363   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6364   br(tmp2);
6365 
6366   for (int i = -zva_length + 16; i < 0; i += 16)
6367     stp(zr, zr, Address(base, i));
6368   bind(initial_table_end);
6369 
6370   sub(cnt, cnt, zva_length >> 3);
6371   bind(loop_zva);
6372   dc(Assembler::ZVA, base);
6373   subs(cnt, cnt, zva_length >> 3);
6374   add(base, base, zva_length);
6375   br(Assembler::GE, loop_zva);
6376   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6377   bind(fini);
6378 }
6379 
6380 // base:   Address of a buffer to be filled, 8 bytes aligned.
6381 // cnt:    Count in 8-byte unit.
6382 // value:  Value to be filled with.
6383 // base will point to the end of the buffer after filling.
6384 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6385 {
6386 //  Algorithm:
6387 //
6388 //    if (cnt == 0) {
6389 //      return;
6390 //    }
6391 //    if ((p & 8) != 0) {
6392 //      *p++ = v;
6393 //    }
6394 //
6395 //    scratch1 = cnt & 14;
6396 //    cnt -= scratch1;
6397 //    p += scratch1;
6398 //    switch (scratch1 / 2) {
6399 //      do {
6400 //        cnt -= 16;
6401 //          p[-16] = v;
6402 //          p[-15] = v;
6403 //        case 7:
6404 //          p[-14] = v;
6405 //          p[-13] = v;
6406 //        case 6:
6407 //          p[-12] = v;
6408 //          p[-11] = v;
6409 //          // ...
6410 //        case 1:
6411 //          p[-2] = v;
6412 //          p[-1] = v;
6413 //        case 0:
6414 //          p += 16;
6415 //      } while (cnt);
6416 //    }
6417 //    if ((cnt & 1) == 1) {
6418 //      *p++ = v;
6419 //    }
6420 
6421   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6422 
6423   Label fini, skip, entry, loop;
6424   const int unroll = 8; // Number of stp instructions we'll unroll
6425 
6426   cbz(cnt, fini);
6427   tbz(base, 3, skip);
6428   str(value, Address(post(base, 8)));
6429   sub(cnt, cnt, 1);
6430   bind(skip);
6431 
6432   andr(rscratch1, cnt, (unroll-1) * 2);
6433   sub(cnt, cnt, rscratch1);
6434   add(base, base, rscratch1, Assembler::LSL, 3);
6435   adr(rscratch2, entry);
6436   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6437   br(rscratch2);
6438 
6439   bind(loop);
6440   add(base, base, unroll * 16);
6441   for (int i = -unroll; i < 0; i++)
6442     stp(value, value, Address(base, i * 16));
6443   bind(entry);
6444   subs(cnt, cnt, unroll * 2);
6445   br(Assembler::GE, loop);
6446 
6447   tbz(cnt, 0, fini);
6448   str(value, Address(post(base, 8)));
6449   bind(fini);
6450 }
6451 
6452 // Intrinsic for
6453 //
6454 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6455 //   Encodes char[] to byte[] in ISO-8859-1
6456 //
6457 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6458 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6459 //
6460 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6461 //   Encodes char[] to byte[] in ASCII
6462 //
6463 // This version always returns the number of characters copied, and does not
6464 // clobber the 'len' register. A successful copy will complete with the post-
6465 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6466 // post-condition: 0 <= 'res' < 'len'.
6467 //
6468 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6469 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6470 //       beyond the acceptable, even though the footprint would be smaller.
6471 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6472 //       avoid additional bloat.
6473 //
6474 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6475 void MacroAssembler::encode_iso_array(Register src, Register dst,
6476                                       Register len, Register res, bool ascii,
6477                                       FloatRegister vtmp0, FloatRegister vtmp1,
6478                                       FloatRegister vtmp2, FloatRegister vtmp3,
6479                                       FloatRegister vtmp4, FloatRegister vtmp5)
6480 {
6481   Register cnt = res;
6482   Register max = rscratch1;
6483   Register chk = rscratch2;
6484 
6485   prfm(Address(src), PLDL1STRM);
6486   movw(cnt, len);
6487 
6488 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6489 
6490   Label LOOP_32, DONE_32, FAIL_32;
6491 
6492   BIND(LOOP_32);
6493   {
6494     cmpw(cnt, 32);
6495     br(LT, DONE_32);
6496     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6497     // Extract lower bytes.
6498     FloatRegister vlo0 = vtmp4;
6499     FloatRegister vlo1 = vtmp5;
6500     uzp1(vlo0, T16B, vtmp0, vtmp1);
6501     uzp1(vlo1, T16B, vtmp2, vtmp3);
6502     // Merge bits...
6503     orr(vtmp0, T16B, vtmp0, vtmp1);
6504     orr(vtmp2, T16B, vtmp2, vtmp3);
6505     // Extract merged upper bytes.
6506     FloatRegister vhix = vtmp0;
6507     uzp2(vhix, T16B, vtmp0, vtmp2);
6508     // ISO-check on hi-parts (all zero).
6509     //                          ASCII-check on lo-parts (no sign).
6510     FloatRegister vlox = vtmp1; // Merge lower bytes.
6511                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6512     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6513     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6514     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6515                                 ASCII(orr(chk, chk, max));
6516     cbnz(chk, FAIL_32);
6517     subw(cnt, cnt, 32);
6518     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6519     b(LOOP_32);
6520   }
6521   BIND(FAIL_32);
6522   sub(src, src, 64);
6523   BIND(DONE_32);
6524 
6525   Label LOOP_8, SKIP_8;
6526 
6527   BIND(LOOP_8);
6528   {
6529     cmpw(cnt, 8);
6530     br(LT, SKIP_8);
6531     FloatRegister vhi = vtmp0;
6532     FloatRegister vlo = vtmp1;
6533     ld1(vtmp3, T8H, src);
6534     uzp1(vlo, T16B, vtmp3, vtmp3);
6535     uzp2(vhi, T16B, vtmp3, vtmp3);
6536     // ISO-check on hi-parts (all zero).
6537     //                          ASCII-check on lo-parts (no sign).
6538                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6539     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6540                                 ASCII(umov(max, vtmp2, B, 0));
6541                                 ASCII(orr(chk, chk, max));
6542     cbnz(chk, SKIP_8);
6543 
6544     strd(vlo, Address(post(dst, 8)));
6545     subw(cnt, cnt, 8);
6546     add(src, src, 16);
6547     b(LOOP_8);
6548   }
6549   BIND(SKIP_8);
6550 
6551 #undef ASCII
6552 
6553   Label LOOP, DONE;
6554 
6555   cbz(cnt, DONE);
6556   BIND(LOOP);
6557   {
6558     Register chr = rscratch1;
6559     ldrh(chr, Address(post(src, 2)));
6560     tst(chr, ascii ? 0xff80 : 0xff00);
6561     br(NE, DONE);
6562     strb(chr, Address(post(dst, 1)));
6563     subs(cnt, cnt, 1);
6564     br(GT, LOOP);
6565   }
6566   BIND(DONE);
6567   // Return index where we stopped.
6568   subw(res, len, cnt);
6569 }
6570 
6571 // Inflate byte[] array to char[].
6572 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6573 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6574                                            FloatRegister vtmp1, FloatRegister vtmp2,
6575                                            FloatRegister vtmp3, Register tmp4) {
6576   Label big, done, after_init, to_stub;
6577 
6578   assert_different_registers(src, dst, len, tmp4, rscratch1);
6579 
6580   fmovd(vtmp1, 0.0);
6581   lsrw(tmp4, len, 3);
6582   bind(after_init);
6583   cbnzw(tmp4, big);
6584   // Short string: less than 8 bytes.
6585   {
6586     Label loop, tiny;
6587 
6588     cmpw(len, 4);
6589     br(LT, tiny);
6590     // Use SIMD to do 4 bytes.
6591     ldrs(vtmp2, post(src, 4));
6592     zip1(vtmp3, T8B, vtmp2, vtmp1);
6593     subw(len, len, 4);
6594     strd(vtmp3, post(dst, 8));
6595 
6596     cbzw(len, done);
6597 
6598     // Do the remaining bytes by steam.
6599     bind(loop);
6600     ldrb(tmp4, post(src, 1));
6601     strh(tmp4, post(dst, 2));
6602     subw(len, len, 1);
6603 
6604     bind(tiny);
6605     cbnz(len, loop);
6606 
6607     b(done);
6608   }
6609 
6610   if (SoftwarePrefetchHintDistance >= 0) {
6611     bind(to_stub);
6612       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6613       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6614       address tpc = trampoline_call(stub);
6615       if (tpc == nullptr) {
6616         DEBUG_ONLY(reset_labels(big, done));
6617         postcond(pc() == badAddress);
6618         return nullptr;
6619       }
6620       b(after_init);
6621   }
6622 
6623   // Unpack the bytes 8 at a time.
6624   bind(big);
6625   {
6626     Label loop, around, loop_last, loop_start;
6627 
6628     if (SoftwarePrefetchHintDistance >= 0) {
6629       const int large_loop_threshold = (64 + 16)/8;
6630       ldrd(vtmp2, post(src, 8));
6631       andw(len, len, 7);
6632       cmp(tmp4, (u1)large_loop_threshold);
6633       br(GE, to_stub);
6634       b(loop_start);
6635 
6636       bind(loop);
6637       ldrd(vtmp2, post(src, 8));
6638       bind(loop_start);
6639       subs(tmp4, tmp4, 1);
6640       br(EQ, loop_last);
6641       zip1(vtmp2, T16B, vtmp2, vtmp1);
6642       ldrd(vtmp3, post(src, 8));
6643       st1(vtmp2, T8H, post(dst, 16));
6644       subs(tmp4, tmp4, 1);
6645       zip1(vtmp3, T16B, vtmp3, vtmp1);
6646       st1(vtmp3, T8H, post(dst, 16));
6647       br(NE, loop);
6648       b(around);
6649       bind(loop_last);
6650       zip1(vtmp2, T16B, vtmp2, vtmp1);
6651       st1(vtmp2, T8H, post(dst, 16));
6652       bind(around);
6653       cbz(len, done);
6654     } else {
6655       andw(len, len, 7);
6656       bind(loop);
6657       ldrd(vtmp2, post(src, 8));
6658       sub(tmp4, tmp4, 1);
6659       zip1(vtmp3, T16B, vtmp2, vtmp1);
6660       st1(vtmp3, T8H, post(dst, 16));
6661       cbnz(tmp4, loop);
6662     }
6663   }
6664 
6665   // Do the tail of up to 8 bytes.
6666   add(src, src, len);
6667   ldrd(vtmp3, Address(src, -8));
6668   add(dst, dst, len, ext::uxtw, 1);
6669   zip1(vtmp3, T16B, vtmp3, vtmp1);
6670   strq(vtmp3, Address(dst, -16));
6671 
6672   bind(done);
6673   postcond(pc() != badAddress);
6674   return pc();
6675 }
6676 
6677 // Compress char[] array to byte[].
6678 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6679 // Return the array length if every element in array can be encoded,
6680 // otherwise, the index of first non-latin1 (> 0xff) character.
6681 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6682                                          Register res,
6683                                          FloatRegister tmp0, FloatRegister tmp1,
6684                                          FloatRegister tmp2, FloatRegister tmp3,
6685                                          FloatRegister tmp4, FloatRegister tmp5) {
6686   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6687 }
6688 
6689 // java.math.round(double a)
6690 // Returns the closest long to the argument, with ties rounding to
6691 // positive infinity.  This requires some fiddling for corner
6692 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6693 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6694                                        FloatRegister ftmp) {
6695   Label DONE;
6696   BLOCK_COMMENT("java_round_double: { ");
6697   fmovd(rscratch1, src);
6698   // Use RoundToNearestTiesAway unless src small and -ve.
6699   fcvtasd(dst, src);
6700   // Test if src >= 0 || abs(src) >= 0x1.0p52
6701   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6702   mov(rscratch2, julong_cast(0x1.0p52));
6703   cmp(rscratch1, rscratch2);
6704   br(HS, DONE); {
6705     // src < 0 && abs(src) < 0x1.0p52
6706     // src may have a fractional part, so add 0.5
6707     fmovd(ftmp, 0.5);
6708     faddd(ftmp, src, ftmp);
6709     // Convert double to jlong, use RoundTowardsNegative
6710     fcvtmsd(dst, ftmp);
6711   }
6712   bind(DONE);
6713   BLOCK_COMMENT("} java_round_double");
6714 }
6715 
6716 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6717                                       FloatRegister ftmp) {
6718   Label DONE;
6719   BLOCK_COMMENT("java_round_float: { ");
6720   fmovs(rscratch1, src);
6721   // Use RoundToNearestTiesAway unless src small and -ve.
6722   fcvtassw(dst, src);
6723   // Test if src >= 0 || abs(src) >= 0x1.0p23
6724   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6725   mov(rscratch2, jint_cast(0x1.0p23f));
6726   cmp(rscratch1, rscratch2);
6727   br(HS, DONE); {
6728     // src < 0 && |src| < 0x1.0p23
6729     // src may have a fractional part, so add 0.5
6730     fmovs(ftmp, 0.5f);
6731     fadds(ftmp, src, ftmp);
6732     // Convert float to jint, use RoundTowardsNegative
6733     fcvtmssw(dst, ftmp);
6734   }
6735   bind(DONE);
6736   BLOCK_COMMENT("} java_round_float");
6737 }
6738 
6739 // get_thread() can be called anywhere inside generated code so we
6740 // need to save whatever non-callee save context might get clobbered
6741 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6742 // the call setup code.
6743 //
6744 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6745 // On other systems, the helper is a usual C function.
6746 //
6747 void MacroAssembler::get_thread(Register dst) {
6748   RegSet saved_regs =
6749     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6750     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6751 
6752   protect_return_address();
6753   push(saved_regs, sp);
6754 
6755   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6756   blr(lr);
6757   if (dst != c_rarg0) {
6758     mov(dst, c_rarg0);
6759   }
6760 
6761   pop(saved_regs, sp);
6762   authenticate_return_address();
6763 }
6764 
6765 void MacroAssembler::cache_wb(Address line) {
6766   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6767   assert(line.index() == noreg, "index should be noreg");
6768   assert(line.offset() == 0, "offset should be 0");
6769   // would like to assert this
6770   // assert(line._ext.shift == 0, "shift should be zero");
6771   if (VM_Version::supports_dcpop()) {
6772     // writeback using clear virtual address to point of persistence
6773     dc(Assembler::CVAP, line.base());
6774   } else {
6775     // no need to generate anything as Unsafe.writebackMemory should
6776     // never invoke this stub
6777   }
6778 }
6779 
6780 void MacroAssembler::cache_wbsync(bool is_pre) {
6781   // we only need a barrier post sync
6782   if (!is_pre) {
6783     membar(Assembler::AnyAny);
6784   }
6785 }
6786 
6787 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6788   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6789     return;
6790   }
6791   // Make sure that native code does not change SVE vector length.
6792   Label verify_ok;
6793   movw(tmp, zr);
6794   sve_inc(tmp, B);
6795   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6796   br(EQ, verify_ok);
6797   stop("Error: SVE vector length has changed since jvm startup");
6798   bind(verify_ok);
6799 }
6800 
6801 void MacroAssembler::verify_ptrue() {
6802   Label verify_ok;
6803   if (!UseSVE) {
6804     return;
6805   }
6806   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6807   sve_dec(rscratch1, B);
6808   cbz(rscratch1, verify_ok);
6809   stop("Error: the preserved predicate register (p7) elements are not all true");
6810   bind(verify_ok);
6811 }
6812 
6813 void MacroAssembler::safepoint_isb() {
6814   isb();
6815 #ifndef PRODUCT
6816   if (VerifyCrossModifyFence) {
6817     // Clear the thread state.
6818     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6819   }
6820 #endif
6821 }
6822 
6823 #ifndef PRODUCT
6824 void MacroAssembler::verify_cross_modify_fence_not_required() {
6825   if (VerifyCrossModifyFence) {
6826     // Check if thread needs a cross modify fence.
6827     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6828     Label fence_not_required;
6829     cbz(rscratch1, fence_not_required);
6830     // If it does then fail.
6831     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6832     mov(c_rarg0, rthread);
6833     blr(rscratch1);
6834     bind(fence_not_required);
6835   }
6836 }
6837 #endif
6838 
6839 void MacroAssembler::spin_wait() {
6840   block_comment("spin_wait {");
6841   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6842     switch (VM_Version::spin_wait_desc().inst()) {
6843       case SpinWait::NOP:
6844         nop();
6845         break;
6846       case SpinWait::ISB:
6847         isb();
6848         break;
6849       case SpinWait::YIELD:
6850         yield();
6851         break;
6852       case SpinWait::SB:
6853         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6854         sb();
6855         break;
6856       default:
6857         ShouldNotReachHere();
6858     }
6859   }
6860   block_comment("}");
6861 }
6862 
6863 // Stack frame creation/removal
6864 
6865 void MacroAssembler::enter(bool strip_ret_addr) {
6866   if (strip_ret_addr) {
6867     // Addresses can only be signed once. If there are multiple nested frames being created
6868     // in the same function, then the return address needs stripping first.
6869     strip_return_address();
6870   }
6871   protect_return_address();
6872   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6873   mov(rfp, sp);
6874 }
6875 
6876 void MacroAssembler::leave() {
6877   mov(sp, rfp);
6878   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6879   authenticate_return_address();
6880 }
6881 
6882 // ROP Protection
6883 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6884 // destroying stack frames or whenever directly loading/storing the LR to memory.
6885 // If ROP protection is not set then these functions are no-ops.
6886 // For more details on PAC see pauth_aarch64.hpp.
6887 
6888 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6889 // Uses value zero as the modifier.
6890 //
6891 void MacroAssembler::protect_return_address() {
6892   if (VM_Version::use_rop_protection()) {
6893     check_return_address();
6894     paciaz();
6895   }
6896 }
6897 
6898 // Sign the return value in the given register. Use before updating the LR in the existing stack
6899 // frame for the current function.
6900 // Uses value zero as the modifier.
6901 //
6902 void MacroAssembler::protect_return_address(Register return_reg) {
6903   if (VM_Version::use_rop_protection()) {
6904     check_return_address(return_reg);
6905     paciza(return_reg);
6906   }
6907 }
6908 
6909 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6910 // Uses value zero as the modifier.
6911 //
6912 void MacroAssembler::authenticate_return_address() {
6913   if (VM_Version::use_rop_protection()) {
6914     autiaz();
6915     check_return_address();
6916   }
6917 }
6918 
6919 // Authenticate the return value in the given register. Use before updating the LR in the existing
6920 // stack frame for the current function.
6921 // Uses value zero as the modifier.
6922 //
6923 void MacroAssembler::authenticate_return_address(Register return_reg) {
6924   if (VM_Version::use_rop_protection()) {
6925     autiza(return_reg);
6926     check_return_address(return_reg);
6927   }
6928 }
6929 
6930 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6931 // there is no guaranteed way of authenticating the LR.
6932 //
6933 void MacroAssembler::strip_return_address() {
6934   if (VM_Version::use_rop_protection()) {
6935     xpaclri();
6936   }
6937 }
6938 
6939 #ifndef PRODUCT
6940 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6941 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6942 // it is difficult to debug back to the callee function.
6943 // This function simply loads from the address in the given register.
6944 // Use directly after authentication to catch authentication failures.
6945 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6946 //
6947 void MacroAssembler::check_return_address(Register return_reg) {
6948   if (VM_Version::use_rop_protection()) {
6949     ldr(zr, Address(return_reg));
6950   }
6951 }
6952 #endif
6953 
6954 // The java_calling_convention describes stack locations as ideal slots on
6955 // a frame with no abi restrictions. Since we must observe abi restrictions
6956 // (like the placement of the register window) the slots must be biased by
6957 // the following value.
6958 static int reg2offset_in(VMReg r) {
6959   // Account for saved rfp and lr
6960   // This should really be in_preserve_stack_slots
6961   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6962 }
6963 
6964 static int reg2offset_out(VMReg r) {
6965   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6966 }
6967 
6968 // On 64bit we will store integer like items to the stack as
6969 // 64bits items (AArch64 ABI) even though java would only store
6970 // 32bits for a parameter. On 32bit it will simply be 32bits
6971 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6972 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6973   if (src.first()->is_stack()) {
6974     if (dst.first()->is_stack()) {
6975       // stack to stack
6976       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6977       str(tmp, Address(sp, reg2offset_out(dst.first())));
6978     } else {
6979       // stack to reg
6980       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6981     }
6982   } else if (dst.first()->is_stack()) {
6983     // reg to stack
6984     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6985   } else {
6986     if (dst.first() != src.first()) {
6987       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6988     }
6989   }
6990 }
6991 
6992 // An oop arg. Must pass a handle not the oop itself
6993 void MacroAssembler::object_move(
6994                         OopMap* map,
6995                         int oop_handle_offset,
6996                         int framesize_in_slots,
6997                         VMRegPair src,
6998                         VMRegPair dst,
6999                         bool is_receiver,
7000                         int* receiver_offset) {
7001 
7002   // must pass a handle. First figure out the location we use as a handle
7003 
7004   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
7005 
7006   // See if oop is null if it is we need no handle
7007 
7008   if (src.first()->is_stack()) {
7009 
7010     // Oop is already on the stack as an argument
7011     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
7012     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
7013     if (is_receiver) {
7014       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
7015     }
7016 
7017     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7018     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7019     // conditionally move a null
7020     cmp(rscratch1, zr);
7021     csel(rHandle, zr, rHandle, Assembler::EQ);
7022   } else {
7023 
7024     // Oop is in an a register we must store it to the space we reserve
7025     // on the stack for oop_handles and pass a handle if oop is non-null
7026 
7027     const Register rOop = src.first()->as_Register();
7028     int oop_slot;
7029     if (rOop == j_rarg0)
7030       oop_slot = 0;
7031     else if (rOop == j_rarg1)
7032       oop_slot = 1;
7033     else if (rOop == j_rarg2)
7034       oop_slot = 2;
7035     else if (rOop == j_rarg3)
7036       oop_slot = 3;
7037     else if (rOop == j_rarg4)
7038       oop_slot = 4;
7039     else if (rOop == j_rarg5)
7040       oop_slot = 5;
7041     else if (rOop == j_rarg6)
7042       oop_slot = 6;
7043     else {
7044       assert(rOop == j_rarg7, "wrong register");
7045       oop_slot = 7;
7046     }
7047 
7048     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7049     int offset = oop_slot*VMRegImpl::stack_slot_size;
7050 
7051     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7052     // Store oop in handle area, may be null
7053     str(rOop, Address(sp, offset));
7054     if (is_receiver) {
7055       *receiver_offset = offset;
7056     }
7057 
7058     cmp(rOop, zr);
7059     lea(rHandle, Address(sp, offset));
7060     // conditionally move a null
7061     csel(rHandle, zr, rHandle, Assembler::EQ);
7062   }
7063 
7064   // If arg is on the stack then place it otherwise it is already in correct reg.
7065   if (dst.first()->is_stack()) {
7066     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7067   }
7068 }
7069 
7070 // A float arg may have to do float reg int reg conversion
7071 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7072  if (src.first()->is_stack()) {
7073     if (dst.first()->is_stack()) {
7074       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7075       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7076     } else {
7077       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7078     }
7079   } else if (src.first() != dst.first()) {
7080     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7081       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7082     else
7083       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7084   }
7085 }
7086 
7087 // A long move
7088 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7089   if (src.first()->is_stack()) {
7090     if (dst.first()->is_stack()) {
7091       // stack to stack
7092       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7093       str(tmp, Address(sp, reg2offset_out(dst.first())));
7094     } else {
7095       // stack to reg
7096       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7097     }
7098   } else if (dst.first()->is_stack()) {
7099     // reg to stack
7100     // Do we really have to sign extend???
7101     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7102     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7103   } else {
7104     if (dst.first() != src.first()) {
7105       mov(dst.first()->as_Register(), src.first()->as_Register());
7106     }
7107   }
7108 }
7109 
7110 
7111 // A double move
7112 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7113  if (src.first()->is_stack()) {
7114     if (dst.first()->is_stack()) {
7115       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7116       str(tmp, Address(sp, reg2offset_out(dst.first())));
7117     } else {
7118       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7119     }
7120   } else if (src.first() != dst.first()) {
7121     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7122       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7123     else
7124       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7125   }
7126 }
7127 
7128 // Implements fast-locking.
7129 //
7130 //  - obj: the object to be locked
7131 //  - t1, t2, t3: temporary registers, will be destroyed
7132 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7133 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7134   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7135 
7136   Label push;
7137   const Register top = t1;
7138   const Register mark = t2;
7139   const Register t = t3;
7140 
7141   // Preload the markWord. It is important that this is the first
7142   // instruction emitted as it is part of C1's null check semantics.
7143   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7144 
7145   if (UseObjectMonitorTable) {
7146     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7147     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7148   }
7149 
7150   if (DiagnoseSyncOnValueBasedClasses != 0) {
7151     load_klass(t1, obj);
7152     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7153     tst(t1, KlassFlags::_misc_is_value_based_class);
7154     br(Assembler::NE, slow);
7155   }
7156 
7157   // Check if the lock-stack is full.
7158   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7159   cmpw(top, (unsigned)LockStack::end_offset());
7160   br(Assembler::GE, slow);
7161 
7162   // Check for recursion.
7163   subw(t, top, oopSize);
7164   ldr(t, Address(rthread, t));
7165   cmp(obj, t);
7166   br(Assembler::EQ, push);
7167 
7168   // Check header for monitor (0b10).
7169   tst(mark, markWord::monitor_value);
7170   br(Assembler::NE, slow);
7171 
7172   // Try to lock. Transition lock bits 0b01 => 0b00
7173   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7174   orr(mark, mark, markWord::unlocked_value);
7175   eor(t, mark, markWord::unlocked_value);
7176   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7177           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7178   br(Assembler::NE, slow);
7179 
7180   bind(push);
7181   // After successful lock, push object on lock-stack.
7182   str(obj, Address(rthread, top));
7183   addw(top, top, oopSize);
7184   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7185 }
7186 
7187 // Implements fast-unlocking.
7188 //
7189 // - obj: the object to be unlocked
7190 // - t1, t2, t3: temporary registers
7191 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7192 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7193   // cmpxchg clobbers rscratch1.
7194   assert_different_registers(obj, t1, t2, t3, rscratch1);
7195 
7196 #ifdef ASSERT
7197   {
7198     // Check for lock-stack underflow.
7199     Label stack_ok;
7200     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7201     cmpw(t1, (unsigned)LockStack::start_offset());
7202     br(Assembler::GE, stack_ok);
7203     STOP("Lock-stack underflow");
7204     bind(stack_ok);
7205   }
7206 #endif
7207 
7208   Label unlocked, push_and_slow;
7209   const Register top = t1;
7210   const Register mark = t2;
7211   const Register t = t3;
7212 
7213   // Check if obj is top of lock-stack.
7214   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7215   subw(top, top, oopSize);
7216   ldr(t, Address(rthread, top));
7217   cmp(obj, t);
7218   br(Assembler::NE, slow);
7219 
7220   // Pop lock-stack.
7221   DEBUG_ONLY(str(zr, Address(rthread, top));)
7222   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7223 
7224   // Check if recursive.
7225   subw(t, top, oopSize);
7226   ldr(t, Address(rthread, t));
7227   cmp(obj, t);
7228   br(Assembler::EQ, unlocked);
7229 
7230   // Not recursive. Check header for monitor (0b10).
7231   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7232   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7233 
7234 #ifdef ASSERT
7235   // Check header not unlocked (0b01).
7236   Label not_unlocked;
7237   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7238   stop("fast_unlock already unlocked");
7239   bind(not_unlocked);
7240 #endif
7241 
7242   // Try to unlock. Transition lock bits 0b00 => 0b01
7243   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7244   orr(t, mark, markWord::unlocked_value);
7245   cmpxchg(obj, mark, t, Assembler::xword,
7246           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7247   br(Assembler::EQ, unlocked);
7248 
7249   bind(push_and_slow);
7250   // Restore lock-stack and handle the unlock in runtime.
7251   DEBUG_ONLY(str(obj, Address(rthread, top));)
7252   addw(top, top, oopSize);
7253   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7254   b(slow);
7255 
7256   bind(unlocked);
7257 }