1 /*
   2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "code/compiledIC.hpp"
  30 #include "compiler/compileTask.hpp"
  31 #include "compiler/disassembler.hpp"
  32 #include "compiler/oopMap.hpp"
  33 #include "gc/shared/barrierSet.hpp"
  34 #include "gc/shared/barrierSetAssembler.hpp"
  35 #include "gc/shared/cardTableBarrierSet.hpp"
  36 #include "gc/shared/cardTable.hpp"
  37 #include "gc/shared/collectedHeap.hpp"
  38 #include "gc/shared/tlab_globals.hpp"
  39 #include "interpreter/bytecodeHistogram.hpp"
  40 #include "interpreter/interpreter.hpp"
  41 #include "interpreter/interpreterRuntime.hpp"
  42 #include "jvm.h"
  43 #include "memory/resourceArea.hpp"
  44 #include "memory/universe.hpp"
  45 #include "nativeInst_aarch64.hpp"
  46 #include "oops/accessDecorators.hpp"
  47 #include "oops/compressedKlass.inline.hpp"
  48 #include "oops/compressedOops.inline.hpp"
  49 #include "oops/klass.inline.hpp"
  50 #include "runtime/continuation.hpp"
  51 #include "runtime/icache.hpp"
  52 #include "runtime/interfaceSupport.inline.hpp"
  53 #include "runtime/javaThread.hpp"
  54 #include "runtime/jniHandles.inline.hpp"
  55 #include "runtime/sharedRuntime.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "utilities/globalDefinitions.hpp"
  58 #include "utilities/powerOfTwo.hpp"
  59 #ifdef COMPILER1
  60 #include "c1/c1_LIRAssembler.hpp"
  61 #endif
  62 #ifdef COMPILER2
  63 #include "oops/oop.hpp"
  64 #include "opto/compile.hpp"
  65 #include "opto/node.hpp"
  66 #include "opto/output.hpp"
  67 #endif
  68 
  69 #include <sys/types.h>
  70 
  71 #ifdef PRODUCT
  72 #define BLOCK_COMMENT(str) /* nothing */
  73 #else
  74 #define BLOCK_COMMENT(str) block_comment(str)
  75 #endif
  76 #define STOP(str) stop(str);
  77 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  78 
  79 #ifdef ASSERT
  80 extern "C" void disnm(intptr_t p);
  81 #endif
  82 // Target-dependent relocation processing
  83 //
  84 // Instruction sequences whose target may need to be retrieved or
  85 // patched are distinguished by their leading instruction, sorting
  86 // them into three main instruction groups and related subgroups.
  87 //
  88 // 1) Branch, Exception and System (insn count = 1)
  89 //    1a) Unconditional branch (immediate):
  90 //      b/bl imm19
  91 //    1b) Compare & branch (immediate):
  92 //      cbz/cbnz Rt imm19
  93 //    1c) Test & branch (immediate):
  94 //      tbz/tbnz Rt imm14
  95 //    1d) Conditional branch (immediate):
  96 //      b.cond imm19
  97 //
  98 // 2) Loads and Stores (insn count = 1)
  99 //    2a) Load register literal:
 100 //      ldr Rt imm19
 101 //
 102 // 3) Data Processing Immediate (insn count = 2 or 3)
 103 //    3a) PC-rel. addressing
 104 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 105 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 106 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 107 //      adr/adrp Rx imm21
 108 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 109 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 110 //      The latter form can only happen when the target is an
 111 //      ExternalAddress, and (by definition) ExternalAddresses don't
 112 //      move. Because of that property, there is never any need to
 113 //      patch the last of the three instructions. However,
 114 //      MacroAssembler::target_addr_for_insn takes all three
 115 //      instructions into account and returns the correct address.
 116 //    3b) Move wide (immediate)
 117 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 118 //
 119 // A switch on a subset of the instruction's bits provides an
 120 // efficient dispatch to these subcases.
 121 //
 122 // insn[28:26] -> main group ('x' == don't care)
 123 //   00x -> UNALLOCATED
 124 //   100 -> Data Processing Immediate
 125 //   101 -> Branch, Exception and System
 126 //   x1x -> Loads and Stores
 127 //
 128 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 129 // n.b. in some cases extra bits need to be checked to verify the
 130 // instruction is as expected
 131 //
 132 // 1) ... xx101x Branch, Exception and System
 133 //   1a)  00___x Unconditional branch (immediate)
 134 //   1b)  01___0 Compare & branch (immediate)
 135 //   1c)  01___1 Test & branch (immediate)
 136 //   1d)  10___0 Conditional branch (immediate)
 137 //        other  Should not happen
 138 //
 139 // 2) ... xxx1x0 Loads and Stores
 140 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 141 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 142 //                strictly should be 64 bit non-FP/SIMD i.e.
 143 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 144 //
 145 // 3) ... xx100x Data Processing Immediate
 146 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 147 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 148 //                 strictly should be 64 bit movz #imm16<<0
 149 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 150 //
 151 
 152 static uint32_t insn_at(address insn_addr, int n) {
 153   return ((uint32_t*)insn_addr)[n];
 154 }
 155 
 156 template<typename T>
 157 class RelocActions : public AllStatic {
 158 
 159 public:
 160 
 161   static int ALWAYSINLINE run(address insn_addr, address &target) {
 162     int instructions = 1;
 163     uint32_t insn = insn_at(insn_addr, 0);
 164 
 165     uint32_t dispatch = Instruction_aarch64::extract(insn, 30, 25);
 166     switch(dispatch) {
 167       case 0b001010:
 168       case 0b001011: {
 169         instructions = T::unconditionalBranch(insn_addr, target);
 170         break;
 171       }
 172       case 0b101010:   // Conditional branch (immediate)
 173       case 0b011010: { // Compare & branch (immediate)
 174         instructions = T::conditionalBranch(insn_addr, target);
 175         break;
 176       }
 177       case 0b011011: {
 178         instructions = T::testAndBranch(insn_addr, target);
 179         break;
 180       }
 181       case 0b001100:
 182       case 0b001110:
 183       case 0b011100:
 184       case 0b011110:
 185       case 0b101100:
 186       case 0b101110:
 187       case 0b111100:
 188       case 0b111110: {
 189         // load/store
 190         if ((Instruction_aarch64::extract(insn, 29, 24) & 0b111011) == 0b011000) {
 191           // Load register (literal)
 192           instructions = T::loadStore(insn_addr, target);
 193           break;
 194         } else {
 195           // nothing to do
 196           assert(target == nullptr, "did not expect to relocate target for polling page load");
 197         }
 198         break;
 199       }
 200       case 0b001000:
 201       case 0b011000:
 202       case 0b101000:
 203       case 0b111000: {
 204         // adr/adrp
 205         assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 206         int shift = Instruction_aarch64::extract(insn, 31, 31);
 207         if (shift) {
 208           uint32_t insn2 = insn_at(insn_addr, 1);
 209           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 210               Instruction_aarch64::extract(insn, 4, 0) ==
 211               Instruction_aarch64::extract(insn2, 9, 5)) {
 212             instructions = T::adrp(insn_addr, target, T::adrpMem);
 213           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 214                      Instruction_aarch64::extract(insn, 4, 0) ==
 215                      Instruction_aarch64::extract(insn2, 4, 0)) {
 216             instructions = T::adrp(insn_addr, target, T::adrpAdd);
 217           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 218                      Instruction_aarch64::extract(insn, 4, 0) ==
 219                      Instruction_aarch64::extract(insn2, 4, 0)) {
 220             instructions = T::adrp(insn_addr, target, T::adrpMovk);
 221           } else {
 222             ShouldNotReachHere();
 223           }
 224         } else {
 225           instructions = T::adr(insn_addr, target);
 226         }
 227         break;
 228       }
 229       case 0b001001:
 230       case 0b011001:
 231       case 0b101001:
 232       case 0b111001: {
 233         instructions = T::immediate(insn_addr, target);
 234         break;
 235       }
 236       default: {
 237         ShouldNotReachHere();
 238       }
 239     }
 240 
 241     T::verify(insn_addr, target);
 242     return instructions * NativeInstruction::instruction_size;
 243   }
 244 };
 245 
 246 class Patcher : public AllStatic {
 247 public:
 248   static int unconditionalBranch(address insn_addr, address &target) {
 249     intptr_t offset = (target - insn_addr) >> 2;
 250     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 251     return 1;
 252   }
 253   static int conditionalBranch(address insn_addr, address &target) {
 254     intptr_t offset = (target - insn_addr) >> 2;
 255     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 256     return 1;
 257   }
 258   static int testAndBranch(address insn_addr, address &target) {
 259     intptr_t offset = (target - insn_addr) >> 2;
 260     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 261     return 1;
 262   }
 263   static int loadStore(address insn_addr, address &target) {
 264     intptr_t offset = (target - insn_addr) >> 2;
 265     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 266     return 1;
 267   }
 268   static int adr(address insn_addr, address &target) {
 269 #ifdef ASSERT
 270     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 271 #endif
 272     // PC-rel. addressing
 273     ptrdiff_t offset = target - insn_addr;
 274     int offset_lo = offset & 3;
 275     offset >>= 2;
 276     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 277     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 278     return 1;
 279   }
 280   template<typename U>
 281   static int adrp(address insn_addr, address &target, U inner) {
 282     int instructions = 1;
 283 #ifdef ASSERT
 284     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 285 #endif
 286     ptrdiff_t offset = target - insn_addr;
 287     instructions = 2;
 288     precond(inner != nullptr);
 289     // Give the inner reloc a chance to modify the target.
 290     address adjusted_target = target;
 291     instructions = inner(insn_addr, adjusted_target);
 292     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 293     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 294     offset = adr_page - pc_page;
 295     int offset_lo = offset & 3;
 296     offset >>= 2;
 297     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 298     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 299     return instructions;
 300   }
 301   static int adrpMem(address insn_addr, address &target) {
 302     uintptr_t dest = (uintptr_t)target;
 303     int offset_lo = dest & 0xfff;
 304     uint32_t insn2 = insn_at(insn_addr, 1);
 305     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 306     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 307     guarantee(((dest >> size) << size) == dest, "misaligned target");
 308     return 2;
 309   }
 310   static int adrpAdd(address insn_addr, address &target) {
 311     uintptr_t dest = (uintptr_t)target;
 312     int offset_lo = dest & 0xfff;
 313     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 314     return 2;
 315   }
 316   static int adrpMovk(address insn_addr, address &target) {
 317     uintptr_t dest = uintptr_t(target);
 318     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 319     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 320     target = address(dest);
 321     return 2;
 322   }
 323   static int immediate(address insn_addr, address &target) {
 324     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010100, "must be");
 325     uint64_t dest = (uint64_t)target;
 326     // Move wide constant
 327     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 328     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 329     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 330     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 331     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 332     return 3;
 333   }
 334   static void verify(address insn_addr, address &target) {
 335 #ifdef ASSERT
 336     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 337     if (!(address_is == target)) {
 338       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 339       disnm((intptr_t)insn_addr);
 340       assert(address_is == target, "should be");
 341     }
 342 #endif
 343   }
 344 };
 345 
 346 // If insn1 and insn2 use the same register to form an address, either
 347 // by an offsetted LDR or a simple ADD, return the offset. If the
 348 // second instruction is an LDR, the offset may be scaled.
 349 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 350   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 351       Instruction_aarch64::extract(insn1, 4, 0) ==
 352       Instruction_aarch64::extract(insn2, 9, 5)) {
 353     // Load/store register (unsigned immediate)
 354     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 355     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 356     byte_offset <<= size;
 357     return true;
 358   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 359              Instruction_aarch64::extract(insn1, 4, 0) ==
 360              Instruction_aarch64::extract(insn2, 4, 0)) {
 361     // add (immediate)
 362     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 363     return true;
 364   }
 365   return false;
 366 }
 367 
 368 class AArch64Decoder : public AllStatic {
 369 public:
 370 
 371   static int loadStore(address insn_addr, address &target) {
 372     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 373     target = insn_addr + (offset << 2);
 374     return 1;
 375   }
 376   static int unconditionalBranch(address insn_addr, address &target) {
 377     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 25, 0);
 378     target = insn_addr + (offset << 2);
 379     return 1;
 380   }
 381   static int conditionalBranch(address insn_addr, address &target) {
 382     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 383     target = address(((uint64_t)insn_addr + (offset << 2)));
 384     return 1;
 385   }
 386   static int testAndBranch(address insn_addr, address &target) {
 387     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 18, 5);
 388     target = address(((uint64_t)insn_addr + (offset << 2)));
 389     return 1;
 390   }
 391   static int adr(address insn_addr, address &target) {
 392     // PC-rel. addressing
 393     uint32_t insn = insn_at(insn_addr, 0);
 394     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 395     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 396     target = address((uint64_t)insn_addr + offset);
 397     return 1;
 398   }
 399   template<typename U>
 400   static int adrp(address insn_addr, address &target, U inner) {
 401     uint32_t insn = insn_at(insn_addr, 0);
 402     assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 403     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 404     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 405     int shift = 12;
 406     offset <<= shift;
 407     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 408     target_page &= ((uint64_t)-1) << shift;
 409     target = address(target_page);
 410     precond(inner != nullptr);
 411     inner(insn_addr, target);
 412     return 2;
 413   }
 414   static int adrpMem(address insn_addr, address &target) {
 415     uint32_t insn2 = insn_at(insn_addr, 1);
 416     // Load/store register (unsigned immediate)
 417     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 418     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 419     byte_offset <<= size;
 420     target += byte_offset;
 421     return 2;
 422   }
 423   static int adrpAdd(address insn_addr, address &target) {
 424     uint32_t insn2 = insn_at(insn_addr, 1);
 425     // add (immediate)
 426     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 427     target += byte_offset;
 428     return 2;
 429   }
 430   static int adrpMovk(address insn_addr, address &target) {
 431     uint32_t insn2 = insn_at(insn_addr, 1);
 432     uint64_t dest = uint64_t(target);
 433     dest = (dest & 0xffff0000ffffffff) |
 434       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 435     target = address(dest);
 436 
 437     // We know the destination 4k page. Maybe we have a third
 438     // instruction.
 439     uint32_t insn = insn_at(insn_addr, 0);
 440     uint32_t insn3 = insn_at(insn_addr, 2);
 441     ptrdiff_t byte_offset;
 442     if (offset_for(insn, insn3, byte_offset)) {
 443       target += byte_offset;
 444       return 3;
 445     } else {
 446       return 2;
 447     }
 448   }
 449   static int immediate(address insn_addr, address &target) {
 450     uint32_t *insns = (uint32_t *)insn_addr;
 451     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 452     // Move wide constant: movz, movk, movk.  See movptr().
 453     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 454     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 455     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 456                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 457                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 458     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 459     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 460     return 3;
 461   }
 462   static void verify(address insn_addr, address &target) {
 463   }
 464 };
 465 
 466 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 467   address target;
 468   RelocActions<AArch64Decoder>::run(insn_addr, target);
 469   return target;
 470 }
 471 
 472 // Patch any kind of instruction; there may be several instructions.
 473 // Return the total length (in bytes) of the instructions.
 474 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 475   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 476   return RelocActions<Patcher>::run(insn_addr, target);
 477 }
 478 
 479 int MacroAssembler::patch_oop(address insn_addr, address o) {
 480   int instructions;
 481   unsigned insn = *(unsigned*)insn_addr;
 482   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 483 
 484   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 485 
 486   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 487   // narrow OOPs by setting the upper 16 bits in the first
 488   // instruction.
 489   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 490     // Move narrow OOP
 491     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 492     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 493     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 494     instructions = 2;
 495   } else {
 496     // Move wide OOP
 497     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 498     uintptr_t dest = (uintptr_t)o;
 499     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 500     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 501     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 502     instructions = 3;
 503   }
 504   return instructions * NativeInstruction::instruction_size;
 505 }
 506 
 507 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 508   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 509   // We encode narrow ones by setting the upper 16 bits in the first
 510   // instruction.
 511   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 512   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 513          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 514 
 515   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 516 
 517   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 518   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 519   return 2 * NativeInstruction::instruction_size;
 520 }
 521 
 522 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 523   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 524   if (at_return) {
 525     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 526     // we may safely use the sp instead to perform the stack watermark check.
 527     cmp(in_nmethod ? sp : rfp, tmp);
 528     br(Assembler::HI, slow_path);
 529   } else {
 530     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 531   }
 532 }
 533 
 534 void MacroAssembler::rt_call(address dest, Register tmp) {
 535   CodeBlob *cb = CodeCache::find_blob(dest);
 536   if (cb) {
 537     far_call(RuntimeAddress(dest));
 538   } else {
 539     lea(tmp, RuntimeAddress(dest));
 540     blr(tmp);
 541   }
 542 }
 543 
 544 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 545   if (!Continuations::enabled()) return;
 546   Label done;
 547   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 548   cmp(sp, rscratch1);
 549   br(Assembler::LS, done);
 550   mov(rscratch1, sp); // we can't use sp as the source in str
 551   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 552   bind(done);
 553 }
 554 
 555 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 556   if (!Continuations::enabled()) return;
 557   Label done;
 558   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 559   cmp(sp, rscratch1);
 560   br(Assembler::LO, done);
 561   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 562   bind(done);
 563 }
 564 
 565 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 566   // we must set sp to zero to clear frame
 567   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 568 
 569   // must clear fp, so that compiled frames are not confused; it is
 570   // possible that we need it only for debugging
 571   if (clear_fp) {
 572     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 573   }
 574 
 575   // Always clear the pc because it could have been set by make_walkable()
 576   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 577 }
 578 
 579 // Calls to C land
 580 //
 581 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 582 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 583 // has to be reset to 0. This is required to allow proper stack traversal.
 584 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 585                                          Register last_java_fp,
 586                                          Register last_java_pc,
 587                                          Register scratch) {
 588 
 589   if (last_java_pc->is_valid()) {
 590       str(last_java_pc, Address(rthread,
 591                                 JavaThread::frame_anchor_offset()
 592                                 + JavaFrameAnchor::last_Java_pc_offset()));
 593     }
 594 
 595   // determine last_java_sp register
 596   if (last_java_sp == sp) {
 597     mov(scratch, sp);
 598     last_java_sp = scratch;
 599   } else if (!last_java_sp->is_valid()) {
 600     last_java_sp = esp;
 601   }
 602 
 603   // last_java_fp is optional
 604   if (last_java_fp->is_valid()) {
 605     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 606   }
 607 
 608   // We must set sp last.
 609   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 610 }
 611 
 612 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 613                                          Register last_java_fp,
 614                                          address  last_java_pc,
 615                                          Register scratch) {
 616   assert(last_java_pc != nullptr, "must provide a valid PC");
 617 
 618   adr(scratch, last_java_pc);
 619   str(scratch, Address(rthread,
 620                        JavaThread::frame_anchor_offset()
 621                        + JavaFrameAnchor::last_Java_pc_offset()));
 622 
 623   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 624 }
 625 
 626 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 627                                          Register last_java_fp,
 628                                          Label &L,
 629                                          Register scratch) {
 630   if (L.is_bound()) {
 631     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 632   } else {
 633     InstructionMark im(this);
 634     L.add_patch_at(code(), locator());
 635     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 636   }
 637 }
 638 
 639 static inline bool target_needs_far_branch(address addr) {
 640   if (AOTCodeCache::is_on_for_dump()) {
 641     return true;
 642   }
 643   // codecache size <= 128M
 644   if (!MacroAssembler::far_branches()) {
 645     return false;
 646   }
 647   // codecache size > 240M
 648   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 649     return true;
 650   }
 651   // codecache size: 128M..240M
 652   return !CodeCache::is_non_nmethod(addr);
 653 }
 654 
 655 void MacroAssembler::far_call(Address entry, Register tmp) {
 656   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 657   assert(CodeCache::find_blob(entry.target()) != nullptr,
 658          "destination of far call not found in code cache");
 659   assert(entry.rspec().type() == relocInfo::external_word_type
 660          || entry.rspec().type() == relocInfo::runtime_call_type
 661          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 662   if (target_needs_far_branch(entry.target())) {
 663     uint64_t offset;
 664     // We can use ADRP here because we know that the total size of
 665     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 666     adrp(tmp, entry, offset);
 667     add(tmp, tmp, offset);
 668     blr(tmp);
 669   } else {
 670     bl(entry);
 671   }
 672 }
 673 
 674 int MacroAssembler::far_jump(Address entry, Register tmp) {
 675   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 676   assert(CodeCache::find_blob(entry.target()) != nullptr,
 677          "destination of far call not found in code cache");
 678   assert(entry.rspec().type() == relocInfo::external_word_type
 679          || entry.rspec().type() == relocInfo::runtime_call_type
 680          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 681   address start = pc();
 682   if (target_needs_far_branch(entry.target())) {
 683     uint64_t offset;
 684     // We can use ADRP here because we know that the total size of
 685     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 686     adrp(tmp, entry, offset);
 687     add(tmp, tmp, offset);
 688     br(tmp);
 689   } else {
 690     b(entry);
 691   }
 692   return pc() - start;
 693 }
 694 
 695 void MacroAssembler::reserved_stack_check() {
 696     // testing if reserved zone needs to be enabled
 697     Label no_reserved_zone_enabling;
 698 
 699     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 700     cmp(sp, rscratch1);
 701     br(Assembler::LO, no_reserved_zone_enabling);
 702 
 703     enter();   // LR and FP are live.
 704     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 705     mov(c_rarg0, rthread);
 706     blr(rscratch1);
 707     leave();
 708 
 709     // We have already removed our own frame.
 710     // throw_delayed_StackOverflowError will think that it's been
 711     // called by our caller.
 712     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 713     br(rscratch1);
 714     should_not_reach_here();
 715 
 716     bind(no_reserved_zone_enabling);
 717 }
 718 
 719 static void pass_arg0(MacroAssembler* masm, Register arg) {
 720   if (c_rarg0 != arg ) {
 721     masm->mov(c_rarg0, arg);
 722   }
 723 }
 724 
 725 static void pass_arg1(MacroAssembler* masm, Register arg) {
 726   if (c_rarg1 != arg ) {
 727     masm->mov(c_rarg1, arg);
 728   }
 729 }
 730 
 731 static void pass_arg2(MacroAssembler* masm, Register arg) {
 732   if (c_rarg2 != arg ) {
 733     masm->mov(c_rarg2, arg);
 734   }
 735 }
 736 
 737 static void pass_arg3(MacroAssembler* masm, Register arg) {
 738   if (c_rarg3 != arg ) {
 739     masm->mov(c_rarg3, arg);
 740   }
 741 }
 742 
 743 void MacroAssembler::call_VM_base(Register oop_result,
 744                                   Register java_thread,
 745                                   Register last_java_sp,
 746                                   Label*   return_pc,
 747                                   address  entry_point,
 748                                   int      number_of_arguments,
 749                                   bool     check_exceptions) {
 750    // determine java_thread register
 751   if (!java_thread->is_valid()) {
 752     java_thread = rthread;
 753   }
 754 
 755   // determine last_java_sp register
 756   if (!last_java_sp->is_valid()) {
 757     last_java_sp = esp;
 758   }
 759 
 760   // debugging support
 761   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 762   assert(java_thread == rthread, "unexpected register");
 763 #ifdef ASSERT
 764   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 765   // if (!TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 766 #endif // ASSERT
 767 
 768   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 769   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 770 
 771   // push java thread (becomes first argument of C function)
 772 
 773   mov(c_rarg0, java_thread);
 774 
 775   // set last Java frame before call
 776   assert(last_java_sp != rfp, "can't use rfp");
 777 
 778   Label l;
 779   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 780 
 781   // do the call, remove parameters
 782   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 783 
 784   // lr could be poisoned with PAC signature during throw_pending_exception
 785   // if it was tail-call optimized by compiler, since lr is not callee-saved
 786   // reload it with proper value
 787   adr(lr, l);
 788 
 789   // reset last Java frame
 790   // Only interpreter should have to clear fp
 791   reset_last_Java_frame(true);
 792 
 793    // C++ interp handles this in the interpreter
 794   check_and_handle_popframe(java_thread);
 795   check_and_handle_earlyret(java_thread);
 796 
 797   if (check_exceptions) {
 798     // check for pending exceptions (java_thread is set upon return)
 799     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 800     Label ok;
 801     cbz(rscratch1, ok);
 802     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 803     br(rscratch1);
 804     bind(ok);
 805   }
 806 
 807   // get oop result if there is one and reset the value in the thread
 808   if (oop_result->is_valid()) {
 809     get_vm_result_oop(oop_result, java_thread);
 810   }
 811 }
 812 
 813 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 814   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 815 }
 816 
 817 // Check the entry target is always reachable from any branch.
 818 static bool is_always_within_branch_range(Address entry) {
 819   if (AOTCodeCache::is_on_for_dump()) {
 820     return false;
 821   }
 822   const address target = entry.target();
 823 
 824   if (!CodeCache::contains(target)) {
 825     // We always use trampolines for callees outside CodeCache.
 826     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 827     return false;
 828   }
 829 
 830   if (!MacroAssembler::far_branches()) {
 831     return true;
 832   }
 833 
 834   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 835     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 836     // Non-compiled methods stay forever in CodeCache.
 837     // We check whether the longest possible branch is within the branch range.
 838     assert(CodeCache::find_blob(target) != nullptr &&
 839           !CodeCache::find_blob(target)->is_nmethod(),
 840           "runtime call of compiled method");
 841     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 842     const address left_longest_branch_start = CodeCache::low_bound();
 843     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 844                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 845     return is_reachable;
 846   }
 847 
 848   return false;
 849 }
 850 
 851 // Maybe emit a call via a trampoline. If the code cache is small
 852 // trampolines won't be emitted.
 853 address MacroAssembler::trampoline_call(Address entry) {
 854   assert(entry.rspec().type() == relocInfo::runtime_call_type
 855          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 856          || entry.rspec().type() == relocInfo::static_call_type
 857          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 858 
 859   address target = entry.target();
 860 
 861   if (!is_always_within_branch_range(entry)) {
 862     if (!in_scratch_emit_size()) {
 863       // We don't want to emit a trampoline if C2 is generating dummy
 864       // code during its branch shortening phase.
 865       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 866         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 867         code()->share_trampoline_for(entry.target(), offset());
 868       } else {
 869         address stub = emit_trampoline_stub(offset(), target);
 870         if (stub == nullptr) {
 871           postcond(pc() == badAddress);
 872           return nullptr; // CodeCache is full
 873         }
 874       }
 875     }
 876     target = pc();
 877   }
 878 
 879   address call_pc = pc();
 880   relocate(entry.rspec());
 881   bl(target);
 882 
 883   postcond(pc() != badAddress);
 884   return call_pc;
 885 }
 886 
 887 // Emit a trampoline stub for a call to a target which is too far away.
 888 //
 889 // code sequences:
 890 //
 891 // call-site:
 892 //   branch-and-link to <destination> or <trampoline stub>
 893 //
 894 // Related trampoline stub for this call site in the stub section:
 895 //   load the call target from the constant pool
 896 //   branch (LR still points to the call site above)
 897 
 898 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 899                                              address dest) {
 900   // Max stub size: alignment nop, TrampolineStub.
 901   address stub = start_a_stub(max_trampoline_stub_size());
 902   if (stub == nullptr) {
 903     return nullptr;  // CodeBuffer::expand failed
 904   }
 905 
 906   // Create a trampoline stub relocation which relates this trampoline stub
 907   // with the call instruction at insts_call_instruction_offset in the
 908   // instructions code-section.
 909   align(wordSize);
 910   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 911                                             + insts_call_instruction_offset));
 912   const int stub_start_offset = offset();
 913 
 914   // Now, create the trampoline stub's code:
 915   // - load the call
 916   // - call
 917   Label target;
 918   ldr(rscratch1, target);
 919   br(rscratch1);
 920   bind(target);
 921   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 922          "should be");
 923   emit_int64((int64_t)dest);
 924 
 925   const address stub_start_addr = addr_at(stub_start_offset);
 926 
 927   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 928 
 929   end_a_stub();
 930   return stub_start_addr;
 931 }
 932 
 933 int MacroAssembler::max_trampoline_stub_size() {
 934   // Max stub size: alignment nop, TrampolineStub.
 935   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 936 }
 937 
 938 void MacroAssembler::emit_static_call_stub() {
 939   // CompiledDirectCall::set_to_interpreted knows the
 940   // exact layout of this stub.
 941 
 942   isb();
 943   mov_metadata(rmethod, nullptr);
 944 
 945   // Jump to the entry point of the c2i stub.
 946   if (codestub_branch_needs_far_jump()) {
 947     movptr(rscratch1, 0);
 948     br(rscratch1);
 949   } else {
 950     b(pc());
 951   }
 952 }
 953 
 954 int MacroAssembler::static_call_stub_size() {
 955   // During AOT production run AOT and JIT compiled code
 956   // are used at the same time. We need this size
 957   // to be the same for both types of code.
 958   if (!codestub_branch_needs_far_jump() && !AOTCodeCache::is_on_for_use()) {
 959     // isb; movk; movz; movz; b
 960     return 5 * NativeInstruction::instruction_size;
 961   }
 962   // isb; movk; movz; movz; movk; movz; movz; br
 963   return 8 * NativeInstruction::instruction_size;
 964 }
 965 
 966 void MacroAssembler::c2bool(Register x) {
 967   // implements x == 0 ? 0 : 1
 968   // note: must only look at least-significant byte of x
 969   //       since C-style booleans are stored in one byte
 970   //       only! (was bug)
 971   tst(x, 0xff);
 972   cset(x, Assembler::NE);
 973 }
 974 
 975 address MacroAssembler::ic_call(address entry, jint method_index) {
 976   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 977   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 978   return trampoline_call(Address(entry, rh));
 979 }
 980 
 981 int MacroAssembler::ic_check_size() {
 982   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 983   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 984     return NativeInstruction::instruction_size * (7 + extra_instructions);
 985   } else {
 986     return NativeInstruction::instruction_size * (5 + extra_instructions);
 987   }
 988 }
 989 
 990 int MacroAssembler::ic_check(int end_alignment) {
 991   Register receiver = j_rarg0;
 992   Register data = rscratch2;
 993   Register tmp1 = rscratch1;
 994   Register tmp2 = r10;
 995 
 996   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
 997   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
 998   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
 999   // before the inline cache check here, and not after
1000   align(end_alignment, offset() + ic_check_size());
1001 
1002   int uep_offset = offset();
1003 
1004   if (UseCompactObjectHeaders) {
1005     load_narrow_klass_compact(tmp1, receiver);
1006     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1007     cmpw(tmp1, tmp2);
1008   } else {
1009     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1010     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1011     cmpw(tmp1, tmp2);
1012   }
1013 
1014   Label dont;
1015   br(Assembler::EQ, dont);
1016   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1017   bind(dont);
1018   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1019 
1020   return uep_offset;
1021 }
1022 
1023 // Implementation of call_VM versions
1024 
1025 void MacroAssembler::call_VM(Register oop_result,
1026                              address entry_point,
1027                              bool check_exceptions) {
1028   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1029 }
1030 
1031 void MacroAssembler::call_VM(Register oop_result,
1032                              address entry_point,
1033                              Register arg_1,
1034                              bool check_exceptions) {
1035   pass_arg1(this, arg_1);
1036   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1037 }
1038 
1039 void MacroAssembler::call_VM(Register oop_result,
1040                              address entry_point,
1041                              Register arg_1,
1042                              Register arg_2,
1043                              bool check_exceptions) {
1044   assert_different_registers(arg_1, c_rarg2);
1045   pass_arg2(this, arg_2);
1046   pass_arg1(this, arg_1);
1047   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1048 }
1049 
1050 void MacroAssembler::call_VM(Register oop_result,
1051                              address entry_point,
1052                              Register arg_1,
1053                              Register arg_2,
1054                              Register arg_3,
1055                              bool check_exceptions) {
1056   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1057   assert_different_registers(arg_2, c_rarg3);
1058   pass_arg3(this, arg_3);
1059 
1060   pass_arg2(this, arg_2);
1061 
1062   pass_arg1(this, arg_1);
1063   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1064 }
1065 
1066 void MacroAssembler::call_VM(Register oop_result,
1067                              Register last_java_sp,
1068                              address entry_point,
1069                              int number_of_arguments,
1070                              bool check_exceptions) {
1071   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1072 }
1073 
1074 void MacroAssembler::call_VM(Register oop_result,
1075                              Register last_java_sp,
1076                              address entry_point,
1077                              Register arg_1,
1078                              bool check_exceptions) {
1079   pass_arg1(this, arg_1);
1080   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1081 }
1082 
1083 void MacroAssembler::call_VM(Register oop_result,
1084                              Register last_java_sp,
1085                              address entry_point,
1086                              Register arg_1,
1087                              Register arg_2,
1088                              bool check_exceptions) {
1089 
1090   assert_different_registers(arg_1, c_rarg2);
1091   pass_arg2(this, arg_2);
1092   pass_arg1(this, arg_1);
1093   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1094 }
1095 
1096 void MacroAssembler::call_VM(Register oop_result,
1097                              Register last_java_sp,
1098                              address entry_point,
1099                              Register arg_1,
1100                              Register arg_2,
1101                              Register arg_3,
1102                              bool check_exceptions) {
1103   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1104   assert_different_registers(arg_2, c_rarg3);
1105   pass_arg3(this, arg_3);
1106   pass_arg2(this, arg_2);
1107   pass_arg1(this, arg_1);
1108   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1109 }
1110 
1111 
1112 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1113   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1114   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1115   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1116 }
1117 
1118 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1119   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1120   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1121 }
1122 
1123 void MacroAssembler::align(int modulus) {
1124   align(modulus, offset());
1125 }
1126 
1127 // Ensure that the code at target bytes offset from the current offset() is aligned
1128 // according to modulus.
1129 void MacroAssembler::align(int modulus, int target) {
1130   int delta = target - offset();
1131   while ((offset() + delta) % modulus != 0) nop();
1132 }
1133 
1134 void MacroAssembler::post_call_nop() {
1135   if (!Continuations::enabled()) {
1136     return;
1137   }
1138   InstructionMark im(this);
1139   relocate(post_call_nop_Relocation::spec());
1140   InlineSkippedInstructionsCounter skipCounter(this);
1141   nop();
1142   movk(zr, 0);
1143   movk(zr, 0);
1144 }
1145 
1146 // these are no-ops overridden by InterpreterMacroAssembler
1147 
1148 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1149 
1150 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1151 
1152 // Look up the method for a megamorphic invokeinterface call.
1153 // The target method is determined by <intf_klass, itable_index>.
1154 // The receiver klass is in recv_klass.
1155 // On success, the result will be in method_result, and execution falls through.
1156 // On failure, execution transfers to the given label.
1157 void MacroAssembler::lookup_interface_method(Register recv_klass,
1158                                              Register intf_klass,
1159                                              RegisterOrConstant itable_index,
1160                                              Register method_result,
1161                                              Register scan_temp,
1162                                              Label& L_no_such_interface,
1163                          bool return_method) {
1164   assert_different_registers(recv_klass, intf_klass, scan_temp);
1165   assert_different_registers(method_result, intf_klass, scan_temp);
1166   assert(recv_klass != method_result || !return_method,
1167      "recv_klass can be destroyed when method isn't needed");
1168   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1169          "caller must use same register for non-constant itable index as for method");
1170 
1171   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1172   int vtable_base = in_bytes(Klass::vtable_start_offset());
1173   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1174   int scan_step   = itableOffsetEntry::size() * wordSize;
1175   int vte_size    = vtableEntry::size_in_bytes();
1176   assert(vte_size == wordSize, "else adjust times_vte_scale");
1177 
1178   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1179 
1180   // Could store the aligned, prescaled offset in the klass.
1181   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1182   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1183   add(scan_temp, scan_temp, vtable_base);
1184 
1185   if (return_method) {
1186     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1187     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1188     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1189     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1190     if (itentry_off)
1191       add(recv_klass, recv_klass, itentry_off);
1192   }
1193 
1194   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1195   //   if (scan->interface() == intf) {
1196   //     result = (klass + scan->offset() + itable_index);
1197   //   }
1198   // }
1199   Label search, found_method;
1200 
1201   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1202   cmp(intf_klass, method_result);
1203   br(Assembler::EQ, found_method);
1204   bind(search);
1205   // Check that the previous entry is non-null.  A null entry means that
1206   // the receiver class doesn't implement the interface, and wasn't the
1207   // same as when the caller was compiled.
1208   cbz(method_result, L_no_such_interface);
1209   if (itableOffsetEntry::interface_offset() != 0) {
1210     add(scan_temp, scan_temp, scan_step);
1211     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1212   } else {
1213     ldr(method_result, Address(pre(scan_temp, scan_step)));
1214   }
1215   cmp(intf_klass, method_result);
1216   br(Assembler::NE, search);
1217 
1218   bind(found_method);
1219 
1220   // Got a hit.
1221   if (return_method) {
1222     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1223     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1224   }
1225 }
1226 
1227 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1228 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1229 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1230 // The target method is determined by <holder_klass, itable_index>.
1231 // The receiver klass is in recv_klass.
1232 // On success, the result will be in method_result, and execution falls through.
1233 // On failure, execution transfers to the given label.
1234 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1235                                                   Register holder_klass,
1236                                                   Register resolved_klass,
1237                                                   Register method_result,
1238                                                   Register temp_itbl_klass,
1239                                                   Register scan_temp,
1240                                                   int itable_index,
1241                                                   Label& L_no_such_interface) {
1242   // 'method_result' is only used as output register at the very end of this method.
1243   // Until then we can reuse it as 'holder_offset'.
1244   Register holder_offset = method_result;
1245   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1246 
1247   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1248   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1249   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1250   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1251 
1252   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1253 
1254   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1255   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1256   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1257   // temp_itbl_klass = itable[0]._interface;
1258   int vtblEntrySize = vtableEntry::size_in_bytes();
1259   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1260   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1261   mov(holder_offset, zr);
1262   // scan_temp = &(itable[0]._interface)
1263   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1264 
1265   // Initial checks:
1266   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1267   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1268   //   - if (itable[0] == 0), no such interface
1269   cmp(resolved_klass, holder_klass);
1270   br(Assembler::NE, L_loop_search_resolved_entry);
1271   cmp(holder_klass, temp_itbl_klass);
1272   br(Assembler::EQ, L_holder_found);
1273   cbz(temp_itbl_klass, L_no_such_interface);
1274 
1275   // Loop: Look for holder_klass record in itable
1276   //   do {
1277   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1278   //     if (temp_itbl_klass == holder_klass) {
1279   //       goto L_holder_found; // Found!
1280   //     }
1281   //   } while (temp_itbl_klass != 0);
1282   //   goto L_no_such_interface // Not found.
1283   Label L_search_holder;
1284   bind(L_search_holder);
1285     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1286     cmp(holder_klass, temp_itbl_klass);
1287     br(Assembler::EQ, L_holder_found);
1288     cbnz(temp_itbl_klass, L_search_holder);
1289 
1290   b(L_no_such_interface);
1291 
1292   // Loop: Look for resolved_class record in itable
1293   //   while (true) {
1294   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1295   //     if (temp_itbl_klass == 0) {
1296   //       goto L_no_such_interface;
1297   //     }
1298   //     if (temp_itbl_klass == resolved_klass) {
1299   //        goto L_resolved_found;  // Found!
1300   //     }
1301   //     if (temp_itbl_klass == holder_klass) {
1302   //        holder_offset = scan_temp;
1303   //     }
1304   //   }
1305   //
1306   Label L_loop_search_resolved;
1307   bind(L_loop_search_resolved);
1308     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1309   bind(L_loop_search_resolved_entry);
1310     cbz(temp_itbl_klass, L_no_such_interface);
1311     cmp(resolved_klass, temp_itbl_klass);
1312     br(Assembler::EQ, L_resolved_found);
1313     cmp(holder_klass, temp_itbl_klass);
1314     br(Assembler::NE, L_loop_search_resolved);
1315     mov(holder_offset, scan_temp);
1316     b(L_loop_search_resolved);
1317 
1318   // See if we already have a holder klass. If not, go and scan for it.
1319   bind(L_resolved_found);
1320   cbz(holder_offset, L_search_holder);
1321   mov(scan_temp, holder_offset);
1322 
1323   // Finally, scan_temp contains holder_klass vtable offset
1324   bind(L_holder_found);
1325   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1326   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1327     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1328   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1329 }
1330 
1331 // virtual method calling
1332 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1333                                            RegisterOrConstant vtable_index,
1334                                            Register method_result) {
1335   assert(vtableEntry::size() * wordSize == 8,
1336          "adjust the scaling in the code below");
1337   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1338 
1339   if (vtable_index.is_register()) {
1340     lea(method_result, Address(recv_klass,
1341                                vtable_index.as_register(),
1342                                Address::lsl(LogBytesPerWord)));
1343     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1344   } else {
1345     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1346     ldr(method_result,
1347         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1348   }
1349 }
1350 
1351 void MacroAssembler::check_klass_subtype(Register sub_klass,
1352                            Register super_klass,
1353                            Register temp_reg,
1354                            Label& L_success) {
1355   Label L_failure;
1356   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1357   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1358   bind(L_failure);
1359 }
1360 
1361 
1362 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1363                                                    Register super_klass,
1364                                                    Register temp_reg,
1365                                                    Label* L_success,
1366                                                    Label* L_failure,
1367                                                    Label* L_slow_path,
1368                                                    Register super_check_offset) {
1369   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1370   bool must_load_sco = ! super_check_offset->is_valid();
1371   if (must_load_sco) {
1372     assert(temp_reg != noreg, "supply either a temp or a register offset");
1373   }
1374 
1375   Label L_fallthrough;
1376   int label_nulls = 0;
1377   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1378   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1379   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1380   assert(label_nulls <= 1, "at most one null in the batch");
1381 
1382   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1383   Address super_check_offset_addr(super_klass, sco_offset);
1384 
1385   // Hacked jmp, which may only be used just before L_fallthrough.
1386 #define final_jmp(label)                                                \
1387   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1388   else                            b(label)                /*omit semi*/
1389 
1390   // If the pointers are equal, we are done (e.g., String[] elements).
1391   // This self-check enables sharing of secondary supertype arrays among
1392   // non-primary types such as array-of-interface.  Otherwise, each such
1393   // type would need its own customized SSA.
1394   // We move this check to the front of the fast path because many
1395   // type checks are in fact trivially successful in this manner,
1396   // so we get a nicely predicted branch right at the start of the check.
1397   cmp(sub_klass, super_klass);
1398   br(Assembler::EQ, *L_success);
1399 
1400   // Check the supertype display:
1401   if (must_load_sco) {
1402     ldrw(temp_reg, super_check_offset_addr);
1403     super_check_offset = temp_reg;
1404   }
1405 
1406   Address super_check_addr(sub_klass, super_check_offset);
1407   ldr(rscratch1, super_check_addr);
1408   cmp(super_klass, rscratch1); // load displayed supertype
1409   br(Assembler::EQ, *L_success);
1410 
1411   // This check has worked decisively for primary supers.
1412   // Secondary supers are sought in the super_cache ('super_cache_addr').
1413   // (Secondary supers are interfaces and very deeply nested subtypes.)
1414   // This works in the same check above because of a tricky aliasing
1415   // between the super_cache and the primary super display elements.
1416   // (The 'super_check_addr' can address either, as the case requires.)
1417   // Note that the cache is updated below if it does not help us find
1418   // what we need immediately.
1419   // So if it was a primary super, we can just fail immediately.
1420   // Otherwise, it's the slow path for us (no success at this point).
1421 
1422   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1423   if (L_failure == &L_fallthrough) {
1424     cbz(rscratch1, *L_slow_path);
1425   } else {
1426     cbnz(rscratch1, *L_failure);
1427     final_jmp(*L_slow_path);
1428   }
1429 
1430   bind(L_fallthrough);
1431 
1432 #undef final_jmp
1433 }
1434 
1435 // These two are taken from x86, but they look generally useful
1436 
1437 // scans count pointer sized words at [addr] for occurrence of value,
1438 // generic
1439 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1440                                 Register scratch) {
1441   Label Lloop, Lexit;
1442   cbz(count, Lexit);
1443   bind(Lloop);
1444   ldr(scratch, post(addr, wordSize));
1445   cmp(value, scratch);
1446   br(EQ, Lexit);
1447   sub(count, count, 1);
1448   cbnz(count, Lloop);
1449   bind(Lexit);
1450 }
1451 
1452 // scans count 4 byte words at [addr] for occurrence of value,
1453 // generic
1454 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1455                                 Register scratch) {
1456   Label Lloop, Lexit;
1457   cbz(count, Lexit);
1458   bind(Lloop);
1459   ldrw(scratch, post(addr, wordSize));
1460   cmpw(value, scratch);
1461   br(EQ, Lexit);
1462   sub(count, count, 1);
1463   cbnz(count, Lloop);
1464   bind(Lexit);
1465 }
1466 
1467 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1468                                                           Register super_klass,
1469                                                           Register temp_reg,
1470                                                           Register temp2_reg,
1471                                                           Label* L_success,
1472                                                           Label* L_failure,
1473                                                           bool set_cond_codes) {
1474   // NB! Callers may assume that, when temp2_reg is a valid register,
1475   // this code sets it to a nonzero value.
1476 
1477   assert_different_registers(sub_klass, super_klass, temp_reg);
1478   if (temp2_reg != noreg)
1479     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1480 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1481 
1482   Label L_fallthrough;
1483   int label_nulls = 0;
1484   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1485   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1486   assert(label_nulls <= 1, "at most one null in the batch");
1487 
1488   // a couple of useful fields in sub_klass:
1489   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1490   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1491   Address secondary_supers_addr(sub_klass, ss_offset);
1492   Address super_cache_addr(     sub_klass, sc_offset);
1493 
1494   BLOCK_COMMENT("check_klass_subtype_slow_path");
1495 
1496   // Do a linear scan of the secondary super-klass chain.
1497   // This code is rarely used, so simplicity is a virtue here.
1498   // The repne_scan instruction uses fixed registers, which we must spill.
1499   // Don't worry too much about pre-existing connections with the input regs.
1500 
1501   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1502   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1503 
1504   RegSet pushed_registers;
1505   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1506   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1507 
1508   if (super_klass != r0) {
1509     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1510   }
1511 
1512   push(pushed_registers, sp);
1513 
1514   // Get super_klass value into r0 (even if it was in r5 or r2).
1515   if (super_klass != r0) {
1516     mov(r0, super_klass);
1517   }
1518 
1519 #ifndef PRODUCT
1520   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1521 #endif //PRODUCT
1522 
1523   // We will consult the secondary-super array.
1524   ldr(r5, secondary_supers_addr);
1525   // Load the array length.
1526   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1527   // Skip to start of data.
1528   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1529 
1530   cmp(sp, zr); // Clear Z flag; SP is never zero
1531   // Scan R2 words at [R5] for an occurrence of R0.
1532   // Set NZ/Z based on last compare.
1533   repne_scan(r5, r0, r2, rscratch1);
1534 
1535   // Unspill the temp. registers:
1536   pop(pushed_registers, sp);
1537 
1538   br(Assembler::NE, *L_failure);
1539 
1540   // Success.  Cache the super we found and proceed in triumph.
1541 
1542   if (UseSecondarySupersCache) {
1543     str(super_klass, super_cache_addr);
1544   }
1545 
1546   if (L_success != &L_fallthrough) {
1547     b(*L_success);
1548   }
1549 
1550 #undef IS_A_TEMP
1551 
1552   bind(L_fallthrough);
1553 }
1554 
1555 // If Register r is invalid, remove a new register from
1556 // available_regs, and add new register to regs_to_push.
1557 Register MacroAssembler::allocate_if_noreg(Register r,
1558                                   RegSetIterator<Register> &available_regs,
1559                                   RegSet &regs_to_push) {
1560   if (!r->is_valid()) {
1561     r = *available_regs++;
1562     regs_to_push += r;
1563   }
1564   return r;
1565 }
1566 
1567 // check_klass_subtype_slow_path_table() looks for super_klass in the
1568 // hash table belonging to super_klass, branching to L_success or
1569 // L_failure as appropriate. This is essentially a shim which
1570 // allocates registers as necessary then calls
1571 // lookup_secondary_supers_table() to do the work. Any of the temp
1572 // regs may be noreg, in which case this logic will chooses some
1573 // registers push and pop them from the stack.
1574 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1575                                                          Register super_klass,
1576                                                          Register temp_reg,
1577                                                          Register temp2_reg,
1578                                                          Register temp3_reg,
1579                                                          Register result_reg,
1580                                                          FloatRegister vtemp,
1581                                                          Label* L_success,
1582                                                          Label* L_failure,
1583                                                          bool set_cond_codes) {
1584   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1585 
1586   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1587 
1588   Label L_fallthrough;
1589   int label_nulls = 0;
1590   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1591   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1592   assert(label_nulls <= 1, "at most one null in the batch");
1593 
1594   BLOCK_COMMENT("check_klass_subtype_slow_path");
1595 
1596   RegSetIterator<Register> available_regs
1597     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1598 
1599   RegSet pushed_regs;
1600 
1601   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1602   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1603   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1604   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1605 
1606   push(pushed_regs, sp);
1607 
1608   lookup_secondary_supers_table_var(sub_klass,
1609                                     super_klass,
1610                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1611                                     nullptr);
1612   cmp(result_reg, zr);
1613 
1614   // Unspill the temp. registers:
1615   pop(pushed_regs, sp);
1616 
1617   // NB! Callers may assume that, when set_cond_codes is true, this
1618   // code sets temp2_reg to a nonzero value.
1619   if (set_cond_codes) {
1620     mov(temp2_reg, 1);
1621   }
1622 
1623   br(Assembler::NE, *L_failure);
1624 
1625   if (L_success != &L_fallthrough) {
1626     b(*L_success);
1627   }
1628 
1629   bind(L_fallthrough);
1630 }
1631 
1632 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1633                                                    Register super_klass,
1634                                                    Register temp_reg,
1635                                                    Register temp2_reg,
1636                                                    Label* L_success,
1637                                                    Label* L_failure,
1638                                                    bool set_cond_codes) {
1639   if (UseSecondarySupersTable) {
1640     check_klass_subtype_slow_path_table
1641       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1642        /*vtemp*/fnoreg,
1643        L_success, L_failure, set_cond_codes);
1644   } else {
1645     check_klass_subtype_slow_path_linear
1646       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1647   }
1648 }
1649 
1650 
1651 // Ensure that the inline code and the stub are using the same registers.
1652 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1653 do {                                                               \
1654   assert(r_super_klass  == r0                                   && \
1655          r_array_base   == r1                                   && \
1656          r_array_length == r2                                   && \
1657          (r_array_index == r3        || r_array_index == noreg) && \
1658          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1659          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1660          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1661 } while(0)
1662 
1663 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1664                                                          Register r_super_klass,
1665                                                          Register temp1,
1666                                                          Register temp2,
1667                                                          Register temp3,
1668                                                          FloatRegister vtemp,
1669                                                          Register result,
1670                                                          u1 super_klass_slot,
1671                                                          bool stub_is_near) {
1672   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1673 
1674   Label L_fallthrough;
1675 
1676   BLOCK_COMMENT("lookup_secondary_supers_table {");
1677 
1678   const Register
1679     r_array_base   = temp1, // r1
1680     r_array_length = temp2, // r2
1681     r_array_index  = temp3, // r3
1682     r_bitmap       = rscratch2;
1683 
1684   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1685 
1686   u1 bit = super_klass_slot;
1687 
1688   // Make sure that result is nonzero if the TBZ below misses.
1689   mov(result, 1);
1690 
1691   // We're going to need the bitmap in a vector reg and in a core reg,
1692   // so load both now.
1693   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1694   if (bit != 0) {
1695     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1696   }
1697   // First check the bitmap to see if super_klass might be present. If
1698   // the bit is zero, we are certain that super_klass is not one of
1699   // the secondary supers.
1700   tbz(r_bitmap, bit, L_fallthrough);
1701 
1702   // Get the first array index that can contain super_klass into r_array_index.
1703   if (bit != 0) {
1704     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1705     cnt(vtemp, T8B, vtemp);
1706     addv(vtemp, T8B, vtemp);
1707     fmovd(r_array_index, vtemp);
1708   } else {
1709     mov(r_array_index, (u1)1);
1710   }
1711   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1712 
1713   // We will consult the secondary-super array.
1714   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1715 
1716   // The value i in r_array_index is >= 1, so even though r_array_base
1717   // points to the length, we don't need to adjust it to point to the
1718   // data.
1719   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1720   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1721 
1722   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1723   eor(result, result, r_super_klass);
1724   cbz(result, L_fallthrough); // Found a match
1725 
1726   // Is there another entry to check? Consult the bitmap.
1727   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1728 
1729   // Linear probe.
1730   if (bit != 0) {
1731     ror(r_bitmap, r_bitmap, bit);
1732   }
1733 
1734   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1735   // The next slot to be inspected, by the stub we're about to call,
1736   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1737   // have been checked.
1738   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1739   if (stub_is_near) {
1740     bl(stub);
1741   } else {
1742     address call = trampoline_call(stub);
1743     if (call == nullptr) {
1744       return false; // trampoline allocation failed
1745     }
1746   }
1747 
1748   BLOCK_COMMENT("} lookup_secondary_supers_table");
1749 
1750   bind(L_fallthrough);
1751 
1752   if (VerifySecondarySupers) {
1753     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1754                                   temp1, temp2, result);      // r1, r2, r5
1755   }
1756   return true;
1757 }
1758 
1759 // At runtime, return 0 in result if r_super_klass is a superclass of
1760 // r_sub_klass, otherwise return nonzero. Use this version of
1761 // lookup_secondary_supers_table() if you don't know ahead of time
1762 // which superclass will be searched for. Used by interpreter and
1763 // runtime stubs. It is larger and has somewhat greater latency than
1764 // the version above, which takes a constant super_klass_slot.
1765 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1766                                                        Register r_super_klass,
1767                                                        Register temp1,
1768                                                        Register temp2,
1769                                                        Register temp3,
1770                                                        FloatRegister vtemp,
1771                                                        Register result,
1772                                                        Label *L_success) {
1773   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1774 
1775   Label L_fallthrough;
1776 
1777   BLOCK_COMMENT("lookup_secondary_supers_table {");
1778 
1779   const Register
1780     r_array_index = temp3,
1781     slot          = rscratch1,
1782     r_bitmap      = rscratch2;
1783 
1784   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1785 
1786   // Make sure that result is nonzero if the test below misses.
1787   mov(result, 1);
1788 
1789   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1790 
1791   // First check the bitmap to see if super_klass might be present. If
1792   // the bit is zero, we are certain that super_klass is not one of
1793   // the secondary supers.
1794 
1795   // This next instruction is equivalent to:
1796   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1797   // sub(temp2, tmp_reg, slot);
1798   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1799   lslv(temp2, r_bitmap, temp2);
1800   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1801 
1802   bool must_save_v0 = (vtemp == fnoreg);
1803   if (must_save_v0) {
1804     // temp1 and result are free, so use them to preserve vtemp
1805     vtemp = v0;
1806     mov(temp1,  vtemp, D, 0);
1807     mov(result, vtemp, D, 1);
1808   }
1809 
1810   // Get the first array index that can contain super_klass into r_array_index.
1811   mov(vtemp, D, 0, temp2);
1812   cnt(vtemp, T8B, vtemp);
1813   addv(vtemp, T8B, vtemp);
1814   mov(r_array_index, vtemp, D, 0);
1815 
1816   if (must_save_v0) {
1817     mov(vtemp, D, 0, temp1 );
1818     mov(vtemp, D, 1, result);
1819   }
1820 
1821   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1822 
1823   const Register
1824     r_array_base   = temp1,
1825     r_array_length = temp2;
1826 
1827   // The value i in r_array_index is >= 1, so even though r_array_base
1828   // points to the length, we don't need to adjust it to point to the
1829   // data.
1830   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1831   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1832 
1833   // We will consult the secondary-super array.
1834   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1835 
1836   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1837   eor(result, result, r_super_klass);
1838   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1839 
1840   // Is there another entry to check? Consult the bitmap.
1841   rorv(r_bitmap, r_bitmap, slot);
1842   // rol(r_bitmap, r_bitmap, 1);
1843   tbz(r_bitmap, 1, L_fallthrough);
1844 
1845   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1846   // The next slot to be inspected, by the logic we're about to call,
1847   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1848   // have been checked.
1849   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1850                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1851 
1852   BLOCK_COMMENT("} lookup_secondary_supers_table");
1853 
1854   bind(L_fallthrough);
1855 
1856   if (VerifySecondarySupers) {
1857     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1858                                   temp1, temp2, result);      // r1, r2, r5
1859   }
1860 
1861   if (L_success) {
1862     cbz(result, *L_success);
1863   }
1864 }
1865 
1866 // Called by code generated by check_klass_subtype_slow_path
1867 // above. This is called when there is a collision in the hashed
1868 // lookup in the secondary supers array.
1869 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1870                                                              Register r_array_base,
1871                                                              Register r_array_index,
1872                                                              Register r_bitmap,
1873                                                              Register temp1,
1874                                                              Register result,
1875                                                              bool is_stub) {
1876   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1877 
1878   const Register
1879     r_array_length = temp1,
1880     r_sub_klass    = noreg; // unused
1881 
1882   if (is_stub) {
1883     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1884   }
1885 
1886   Label L_fallthrough, L_huge;
1887 
1888   // Load the array length.
1889   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1890   // And adjust the array base to point to the data.
1891   // NB! Effectively increments current slot index by 1.
1892   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1893   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1894 
1895   // The bitmap is full to bursting.
1896   // Implicit invariant: BITMAP_FULL implies (length > 0)
1897   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1898   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1899   br(GT, L_huge);
1900 
1901   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1902   // current slot (at secondary_supers[r_array_index]) has not yet
1903   // been inspected, and r_array_index may be out of bounds if we
1904   // wrapped around the end of the array.
1905 
1906   { // This is conventional linear probing, but instead of terminating
1907     // when a null entry is found in the table, we maintain a bitmap
1908     // in which a 0 indicates missing entries.
1909     // As long as the bitmap is not completely full,
1910     // array_length == popcount(bitmap). The array_length check above
1911     // guarantees there are 0s in the bitmap, so the loop eventually
1912     // terminates.
1913     Label L_loop;
1914     bind(L_loop);
1915 
1916     // Check for wraparound.
1917     cmp(r_array_index, r_array_length);
1918     csel(r_array_index, zr, r_array_index, GE);
1919 
1920     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1921     eor(result, rscratch1, r_super_klass);
1922     cbz(result, L_fallthrough);
1923 
1924     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1925 
1926     ror(r_bitmap, r_bitmap, 1);
1927     add(r_array_index, r_array_index, 1);
1928     b(L_loop);
1929   }
1930 
1931   { // Degenerate case: more than 64 secondary supers.
1932     // FIXME: We could do something smarter here, maybe a vectorized
1933     // comparison or a binary search, but is that worth any added
1934     // complexity?
1935     bind(L_huge);
1936     cmp(sp, zr); // Clear Z flag; SP is never zero
1937     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1938     cset(result, NE); // result == 0 iff we got a match.
1939   }
1940 
1941   bind(L_fallthrough);
1942 }
1943 
1944 // Make sure that the hashed lookup and a linear scan agree.
1945 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1946                                                    Register r_super_klass,
1947                                                    Register temp1,
1948                                                    Register temp2,
1949                                                    Register result) {
1950   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1951 
1952   const Register
1953     r_array_base   = temp1,
1954     r_array_length = temp2;
1955 
1956   BLOCK_COMMENT("verify_secondary_supers_table {");
1957 
1958   // We will consult the secondary-super array.
1959   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1960 
1961   // Load the array length.
1962   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1963   // And adjust the array base to point to the data.
1964   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1965 
1966   cmp(sp, zr); // Clear Z flag; SP is never zero
1967   // Scan R2 words at [R5] for an occurrence of R0.
1968   // Set NZ/Z based on last compare.
1969   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1970   // rscratch1 == 0 iff we got a match.
1971   cset(rscratch1, NE);
1972 
1973   Label passed;
1974   cmp(result, zr);
1975   cset(result, NE); // normalize result to 0/1 for comparison
1976 
1977   cmp(rscratch1, result);
1978   br(EQ, passed);
1979   {
1980     mov(r0, r_super_klass);         // r0 <- r0
1981     mov(r1, r_sub_klass);           // r1 <- r4
1982     mov(r2, /*expected*/rscratch1); // r2 <- r8
1983     mov(r3, result);                // r3 <- r5
1984     mov(r4, (address)("mismatch")); // r4 <- const
1985     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
1986     should_not_reach_here();
1987   }
1988   bind(passed);
1989 
1990   BLOCK_COMMENT("} verify_secondary_supers_table");
1991 }
1992 
1993 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
1994   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
1995   assert_different_registers(klass, rthread, scratch);
1996 
1997   Label L_fallthrough, L_tmp;
1998   if (L_fast_path == nullptr) {
1999     L_fast_path = &L_fallthrough;
2000   } else if (L_slow_path == nullptr) {
2001     L_slow_path = &L_fallthrough;
2002   }
2003   // Fast path check: class is fully initialized
2004   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2005   ldarb(scratch, scratch);
2006   cmp(scratch, InstanceKlass::fully_initialized);
2007   br(Assembler::EQ, *L_fast_path);
2008 
2009   // Fast path check: current thread is initializer thread
2010   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2011   cmp(rthread, scratch);
2012 
2013   if (L_slow_path == &L_fallthrough) {
2014     br(Assembler::EQ, *L_fast_path);
2015     bind(*L_slow_path);
2016   } else if (L_fast_path == &L_fallthrough) {
2017     br(Assembler::NE, *L_slow_path);
2018     bind(*L_fast_path);
2019   } else {
2020     Unimplemented();
2021   }
2022 }
2023 
2024 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2025   if (!VerifyOops) return;
2026 
2027   // Pass register number to verify_oop_subroutine
2028   const char* b = nullptr;
2029   {
2030     ResourceMark rm;
2031     stringStream ss;
2032     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2033     b = code_string(ss.as_string());
2034   }
2035   BLOCK_COMMENT("verify_oop {");
2036 
2037   strip_return_address(); // This might happen within a stack frame.
2038   protect_return_address();
2039   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2040   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2041 
2042   mov(r0, reg);
2043   movptr(rscratch1, (uintptr_t)(address)b);
2044 
2045   // call indirectly to solve generation ordering problem
2046   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2047   ldr(rscratch2, Address(rscratch2));
2048   blr(rscratch2);
2049 
2050   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2051   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2052   authenticate_return_address();
2053 
2054   BLOCK_COMMENT("} verify_oop");
2055 }
2056 
2057 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2058   if (!VerifyOops) return;
2059 
2060   const char* b = nullptr;
2061   {
2062     ResourceMark rm;
2063     stringStream ss;
2064     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2065     b = code_string(ss.as_string());
2066   }
2067   BLOCK_COMMENT("verify_oop_addr {");
2068 
2069   strip_return_address(); // This might happen within a stack frame.
2070   protect_return_address();
2071   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2072   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2073 
2074   // addr may contain sp so we will have to adjust it based on the
2075   // pushes that we just did.
2076   if (addr.uses(sp)) {
2077     lea(r0, addr);
2078     ldr(r0, Address(r0, 4 * wordSize));
2079   } else {
2080     ldr(r0, addr);
2081   }
2082   movptr(rscratch1, (uintptr_t)(address)b);
2083 
2084   // call indirectly to solve generation ordering problem
2085   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2086   ldr(rscratch2, Address(rscratch2));
2087   blr(rscratch2);
2088 
2089   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2090   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2091   authenticate_return_address();
2092 
2093   BLOCK_COMMENT("} verify_oop_addr");
2094 }
2095 
2096 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2097                                          int extra_slot_offset) {
2098   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2099   int stackElementSize = Interpreter::stackElementSize;
2100   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2101 #ifdef ASSERT
2102   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2103   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2104 #endif
2105   if (arg_slot.is_constant()) {
2106     return Address(esp, arg_slot.as_constant() * stackElementSize
2107                    + offset);
2108   } else {
2109     add(rscratch1, esp, arg_slot.as_register(),
2110         ext::uxtx, exact_log2(stackElementSize));
2111     return Address(rscratch1, offset);
2112   }
2113 }
2114 
2115 // Handle the receiver type profile update given the "recv" klass.
2116 //
2117 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2118 // If there are no matching or claimable receiver entries in RD, updates
2119 // the polymorphic counter.
2120 //
2121 // This code expected to run by either the interpreter or JIT-ed code, without
2122 // extra synchronization. For safety, receiver cells are claimed atomically, which
2123 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2124 // counter updates are not atomic.
2125 //
2126 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2127   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2128 
2129   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2130   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2131   int poly_count_offset      = in_bytes(CounterData::count_offset());
2132   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2133   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2134 
2135   // Adjust for MDP offsets.
2136   base_receiver_offset += mdp_offset;
2137   end_receiver_offset  += mdp_offset;
2138   poly_count_offset    += mdp_offset;
2139 
2140 #ifdef ASSERT
2141   // We are about to walk the MDO slots without asking for offsets.
2142   // Check that our math hits all the right spots.
2143   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2144     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2145     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2146     int offset = base_receiver_offset + receiver_step*c;
2147     int count_offset = offset + receiver_to_count_step;
2148     assert(offset == real_recv_offset, "receiver slot math");
2149     assert(count_offset == real_count_offset, "receiver count math");
2150   }
2151   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2152   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2153 #endif
2154 
2155   // Corner case: no profile table. Increment poly counter and exit.
2156   if (ReceiverTypeData::row_limit() == 0) {
2157     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2158     return;
2159   }
2160 
2161   Register offset = rscratch2;
2162 
2163   Label L_loop_search_receiver, L_loop_search_empty;
2164   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
2165 
2166   // The code here recognizes three major cases:
2167   //   A. Fastest: receiver found in the table
2168   //   B. Fast: no receiver in the table, and the table is full
2169   //   C. Slow: no receiver in the table, free slots in the table
2170   //
2171   // The case A performance is most important, as perfectly-behaved code would end up
2172   // there, especially with larger TypeProfileWidth. The case B performance is
2173   // important as well, this is where bulk of code would land for normally megamorphic
2174   // cases. The case C performance is not essential, its job is to deal with installation
2175   // races, we optimize for code density instead. Case C needs to make sure that receiver
2176   // rows are only claimed once. This makes sure we never overwrite a row for another
2177   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2178   //
2179   // It is very tempting to handle these cases in a single loop, and claim the first slot
2180   // without checking the rest of the table. But, profiling code should tolerate free slots
2181   // in the table, as class unloading can clear them. After such cleanup, the receiver
2182   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2183   // to complete, before trying to install new slots. Splitting the code in several tight
2184   // loops also helpfully optimizes for cases A and B.
2185   //
2186   // This code is effectively:
2187   //
2188   // restart:
2189   //   // Fastest: receiver is already installed
2190   //   for (i = 0; i < receiver_count(); i++) {
2191   //     if (receiver(i) == recv) goto found_recv(i);
2192   //   }
2193   //
2194   //   // Fast: no receiver, but profile is full
2195   //   for (i = 0; i < receiver_count(); i++) {
2196   //     if (receiver(i) == null) goto found_null(i);
2197   //   }
2198   //   goto polymorphic
2199   //
2200   //   // Slow: try to install receiver
2201   // found_null(i):
2202   //   CAS(&receiver(i), null, recv);
2203   //   goto restart
2204   //
2205   // polymorphic:
2206   //   count++;
2207   //   return
2208   //
2209   // found_recv(i):
2210   //   *receiver_count(i)++
2211   //
2212 
2213   bind(L_restart);
2214 
2215   // Fastest: receiver is already installed
2216   mov(offset, base_receiver_offset);
2217   bind(L_loop_search_receiver);
2218     ldr(rscratch1, Address(mdp, offset));
2219     cmp(rscratch1, recv);
2220     br(Assembler::EQ, L_found_recv);
2221   add(offset, offset, receiver_step);
2222   sub(rscratch1, offset, end_receiver_offset);
2223   cbnz(rscratch1, L_loop_search_receiver);
2224 
2225   // Fast: no receiver, but profile is full
2226   mov(offset, base_receiver_offset);
2227   bind(L_loop_search_empty);
2228     ldr(rscratch1, Address(mdp, offset));
2229     cbz(rscratch1, L_found_empty);
2230   add(offset, offset, receiver_step);
2231   sub(rscratch1, offset, end_receiver_offset);
2232   cbnz(rscratch1, L_loop_search_empty);
2233   b(L_polymorphic);
2234 
2235   // Slow: try to install receiver
2236   bind(L_found_empty);
2237 
2238   // Atomically swing receiver slot: null -> recv.
2239   //
2240   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2241   // is used to hold the destination address. This is safe because the
2242   // offset is no longer needed after the address is computed.
2243 
2244   lea(rscratch2, Address(mdp, offset));
2245   cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
2246           /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
2247 
2248   // CAS success means the slot now has the receiver we want. CAS failure means
2249   // something had claimed the slot concurrently: it can be the same receiver we want,
2250   // or something else. Since this is a slow path, we can optimize for code density,
2251   // and just restart the search from the beginning.
2252   b(L_restart);
2253 
2254   // Counter updates:
2255 
2256   // Increment polymorphic counter instead of receiver slot.
2257   bind(L_polymorphic);
2258   mov(offset, poly_count_offset);
2259   b(L_count_update);
2260 
2261   // Found a receiver, convert its slot offset to corresponding count offset.
2262   bind(L_found_recv);
2263   add(offset, offset, receiver_to_count_step);
2264 
2265   bind(L_count_update);
2266   increment(Address(mdp, offset), DataLayout::counter_increment);
2267 }
2268 
2269 
2270 void MacroAssembler::call_VM_leaf_base(address entry_point,
2271                                        int number_of_arguments,
2272                                        Label *retaddr) {
2273   Label E, L;
2274 
2275   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2276 
2277   mov(rscratch1, RuntimeAddress(entry_point));
2278   blr(rscratch1);
2279   if (retaddr)
2280     bind(*retaddr);
2281 
2282   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2283 }
2284 
2285 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2286   call_VM_leaf_base(entry_point, number_of_arguments);
2287 }
2288 
2289 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2290   pass_arg0(this, arg_0);
2291   call_VM_leaf_base(entry_point, 1);
2292 }
2293 
2294 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2295   assert_different_registers(arg_1, c_rarg0);
2296   pass_arg0(this, arg_0);
2297   pass_arg1(this, arg_1);
2298   call_VM_leaf_base(entry_point, 2);
2299 }
2300 
2301 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2302                                   Register arg_1, Register arg_2) {
2303   assert_different_registers(arg_1, c_rarg0);
2304   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2305   pass_arg0(this, arg_0);
2306   pass_arg1(this, arg_1);
2307   pass_arg2(this, arg_2);
2308   call_VM_leaf_base(entry_point, 3);
2309 }
2310 
2311 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2312   pass_arg0(this, arg_0);
2313   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2314 }
2315 
2316 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2317 
2318   assert_different_registers(arg_0, c_rarg1);
2319   pass_arg1(this, arg_1);
2320   pass_arg0(this, arg_0);
2321   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2322 }
2323 
2324 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2325   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2326   assert_different_registers(arg_1, c_rarg2);
2327   pass_arg2(this, arg_2);
2328   pass_arg1(this, arg_1);
2329   pass_arg0(this, arg_0);
2330   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2331 }
2332 
2333 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2334   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2335   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2336   assert_different_registers(arg_2, c_rarg3);
2337   pass_arg3(this, arg_3);
2338   pass_arg2(this, arg_2);
2339   pass_arg1(this, arg_1);
2340   pass_arg0(this, arg_0);
2341   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2342 }
2343 
2344 void MacroAssembler::null_check(Register reg, int offset) {
2345   if (needs_explicit_null_check(offset)) {
2346     // provoke OS null exception if reg is null by
2347     // accessing M[reg] w/o changing any registers
2348     // NOTE: this is plenty to provoke a segv
2349     ldr(zr, Address(reg));
2350   } else {
2351     // nothing to do, (later) access of M[reg + offset]
2352     // will provoke OS null exception if reg is null
2353   }
2354 }
2355 
2356 // MacroAssembler protected routines needed to implement
2357 // public methods
2358 
2359 void MacroAssembler::mov(Register r, Address dest) {
2360   code_section()->relocate(pc(), dest.rspec());
2361   uint64_t imm64 = (uint64_t)dest.target();
2362   movptr(r, imm64);
2363 }
2364 
2365 // Move a constant pointer into r.  In AArch64 mode the virtual
2366 // address space is 48 bits in size, so we only need three
2367 // instructions to create a patchable instruction sequence that can
2368 // reach anywhere.
2369 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2370 #ifndef PRODUCT
2371   {
2372     char buffer[64];
2373     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2374     block_comment(buffer);
2375   }
2376 #endif
2377   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2378   movz(r, imm64 & 0xffff);
2379   imm64 >>= 16;
2380   movk(r, imm64 & 0xffff, 16);
2381   imm64 >>= 16;
2382   movk(r, imm64 & 0xffff, 32);
2383 }
2384 
2385 // Macro to mov replicated immediate to vector register.
2386 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2387 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2388 // Vd will get the following values for different arrangements in T
2389 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2390 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2391 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2392 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2393 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2394 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2395 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2396 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2397 // Clobbers rscratch1
2398 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2399   assert(T != T1Q, "unsupported");
2400   if (T == T1D || T == T2D) {
2401     int imm = operand_valid_for_movi_immediate(imm64, T);
2402     if (-1 != imm) {
2403       movi(Vd, T, imm);
2404     } else {
2405       mov(rscratch1, imm64);
2406       dup(Vd, T, rscratch1);
2407     }
2408     return;
2409   }
2410 
2411 #ifdef ASSERT
2412   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2413   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2414   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2415 #endif
2416   int shift = operand_valid_for_movi_immediate(imm64, T);
2417   uint32_t imm32 = imm64 & 0xffffffffULL;
2418   if (shift >= 0) {
2419     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2420   } else {
2421     movw(rscratch1, imm32);
2422     dup(Vd, T, rscratch1);
2423   }
2424 }
2425 
2426 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2427 {
2428 #ifndef PRODUCT
2429   {
2430     char buffer[64];
2431     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2432     block_comment(buffer);
2433   }
2434 #endif
2435   if (operand_valid_for_logical_immediate(false, imm64)) {
2436     orr(dst, zr, imm64);
2437   } else {
2438     // we can use a combination of MOVZ or MOVN with
2439     // MOVK to build up the constant
2440     uint64_t imm_h[4];
2441     int zero_count = 0;
2442     int neg_count = 0;
2443     int i;
2444     for (i = 0; i < 4; i++) {
2445       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2446       if (imm_h[i] == 0) {
2447         zero_count++;
2448       } else if (imm_h[i] == 0xffffL) {
2449         neg_count++;
2450       }
2451     }
2452     if (zero_count == 4) {
2453       // one MOVZ will do
2454       movz(dst, 0);
2455     } else if (neg_count == 4) {
2456       // one MOVN will do
2457       movn(dst, 0);
2458     } else if (zero_count == 3) {
2459       for (i = 0; i < 4; i++) {
2460         if (imm_h[i] != 0L) {
2461           movz(dst, (uint32_t)imm_h[i], (i << 4));
2462           break;
2463         }
2464       }
2465     } else if (neg_count == 3) {
2466       // one MOVN will do
2467       for (int i = 0; i < 4; i++) {
2468         if (imm_h[i] != 0xffffL) {
2469           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2470           break;
2471         }
2472       }
2473     } else if (zero_count == 2) {
2474       // one MOVZ and one MOVK will do
2475       for (i = 0; i < 3; i++) {
2476         if (imm_h[i] != 0L) {
2477           movz(dst, (uint32_t)imm_h[i], (i << 4));
2478           i++;
2479           break;
2480         }
2481       }
2482       for (;i < 4; i++) {
2483         if (imm_h[i] != 0L) {
2484           movk(dst, (uint32_t)imm_h[i], (i << 4));
2485         }
2486       }
2487     } else if (neg_count == 2) {
2488       // one MOVN and one MOVK will do
2489       for (i = 0; i < 4; i++) {
2490         if (imm_h[i] != 0xffffL) {
2491           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2492           i++;
2493           break;
2494         }
2495       }
2496       for (;i < 4; i++) {
2497         if (imm_h[i] != 0xffffL) {
2498           movk(dst, (uint32_t)imm_h[i], (i << 4));
2499         }
2500       }
2501     } else if (zero_count == 1) {
2502       // one MOVZ and two MOVKs will do
2503       for (i = 0; i < 4; i++) {
2504         if (imm_h[i] != 0L) {
2505           movz(dst, (uint32_t)imm_h[i], (i << 4));
2506           i++;
2507           break;
2508         }
2509       }
2510       for (;i < 4; i++) {
2511         if (imm_h[i] != 0x0L) {
2512           movk(dst, (uint32_t)imm_h[i], (i << 4));
2513         }
2514       }
2515     } else if (neg_count == 1) {
2516       // one MOVN and two MOVKs will do
2517       for (i = 0; i < 4; i++) {
2518         if (imm_h[i] != 0xffffL) {
2519           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2520           i++;
2521           break;
2522         }
2523       }
2524       for (;i < 4; i++) {
2525         if (imm_h[i] != 0xffffL) {
2526           movk(dst, (uint32_t)imm_h[i], (i << 4));
2527         }
2528       }
2529     } else {
2530       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2531       movz(dst, (uint32_t)imm_h[0], 0);
2532       for (i = 1; i < 4; i++) {
2533         movk(dst, (uint32_t)imm_h[i], (i << 4));
2534       }
2535     }
2536   }
2537 }
2538 
2539 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2540 {
2541 #ifndef PRODUCT
2542     {
2543       char buffer[64];
2544       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2545       block_comment(buffer);
2546     }
2547 #endif
2548   if (operand_valid_for_logical_immediate(true, imm32)) {
2549     orrw(dst, zr, imm32);
2550   } else {
2551     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2552     // constant
2553     uint32_t imm_h[2];
2554     imm_h[0] = imm32 & 0xffff;
2555     imm_h[1] = ((imm32 >> 16) & 0xffff);
2556     if (imm_h[0] == 0) {
2557       movzw(dst, imm_h[1], 16);
2558     } else if (imm_h[0] == 0xffff) {
2559       movnw(dst, imm_h[1] ^ 0xffff, 16);
2560     } else if (imm_h[1] == 0) {
2561       movzw(dst, imm_h[0], 0);
2562     } else if (imm_h[1] == 0xffff) {
2563       movnw(dst, imm_h[0] ^ 0xffff, 0);
2564     } else {
2565       // use a MOVZ and MOVK (makes it easier to debug)
2566       movzw(dst, imm_h[0], 0);
2567       movkw(dst, imm_h[1], 16);
2568     }
2569   }
2570 }
2571 
2572 // Form an address from base + offset in Rd.  Rd may or may
2573 // not actually be used: you must use the Address that is returned.
2574 // It is up to you to ensure that the shift provided matches the size
2575 // of your data.
2576 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2577   if (Address::offset_ok_for_immed(byte_offset, shift))
2578     // It fits; no need for any heroics
2579     return Address(base, byte_offset);
2580 
2581   // Don't do anything clever with negative or misaligned offsets
2582   unsigned mask = (1 << shift) - 1;
2583   if (byte_offset < 0 || byte_offset & mask) {
2584     mov(Rd, byte_offset);
2585     add(Rd, base, Rd);
2586     return Address(Rd);
2587   }
2588 
2589   // See if we can do this with two 12-bit offsets
2590   {
2591     uint64_t word_offset = byte_offset >> shift;
2592     uint64_t masked_offset = word_offset & 0xfff000;
2593     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2594         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2595       add(Rd, base, masked_offset << shift);
2596       word_offset -= masked_offset;
2597       return Address(Rd, word_offset << shift);
2598     }
2599   }
2600 
2601   // Do it the hard way
2602   mov(Rd, byte_offset);
2603   add(Rd, base, Rd);
2604   return Address(Rd);
2605 }
2606 
2607 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2608                                     bool want_remainder, Register scratch)
2609 {
2610   // Full implementation of Java idiv and irem.  The function
2611   // returns the (pc) offset of the div instruction - may be needed
2612   // for implicit exceptions.
2613   //
2614   // constraint : ra/rb =/= scratch
2615   //         normal case
2616   //
2617   // input : ra: dividend
2618   //         rb: divisor
2619   //
2620   // result: either
2621   //         quotient  (= ra idiv rb)
2622   //         remainder (= ra irem rb)
2623 
2624   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2625 
2626   int idivl_offset = offset();
2627   if (! want_remainder) {
2628     sdivw(result, ra, rb);
2629   } else {
2630     sdivw(scratch, ra, rb);
2631     Assembler::msubw(result, scratch, rb, ra);
2632   }
2633 
2634   return idivl_offset;
2635 }
2636 
2637 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2638                                     bool want_remainder, Register scratch)
2639 {
2640   // Full implementation of Java ldiv and lrem.  The function
2641   // returns the (pc) offset of the div instruction - may be needed
2642   // for implicit exceptions.
2643   //
2644   // constraint : ra/rb =/= scratch
2645   //         normal case
2646   //
2647   // input : ra: dividend
2648   //         rb: divisor
2649   //
2650   // result: either
2651   //         quotient  (= ra idiv rb)
2652   //         remainder (= ra irem rb)
2653 
2654   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2655 
2656   int idivq_offset = offset();
2657   if (! want_remainder) {
2658     sdiv(result, ra, rb);
2659   } else {
2660     sdiv(scratch, ra, rb);
2661     Assembler::msub(result, scratch, rb, ra);
2662   }
2663 
2664   return idivq_offset;
2665 }
2666 
2667 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2668   address prev = pc() - NativeMembar::instruction_size;
2669   address last = code()->last_insn();
2670   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2671     NativeMembar *bar = NativeMembar_at(prev);
2672     if (AlwaysMergeDMB) {
2673       bar->set_kind(bar->get_kind() | order_constraint);
2674       BLOCK_COMMENT("merged membar(always)");
2675       return;
2676     }
2677     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2678     // doing so would introduce a StoreLoad which the caller did not
2679     // intend
2680     if (bar->get_kind() == order_constraint
2681         || bar->get_kind() == AnyAny
2682         || order_constraint == AnyAny) {
2683       // We are merging two memory barrier instructions.  On AArch64 we
2684       // can do this simply by ORing them together.
2685       bar->set_kind(bar->get_kind() | order_constraint);
2686       BLOCK_COMMENT("merged membar");
2687       return;
2688     } else {
2689       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2690       // We need check the last 2 instructions
2691       address prev2 = prev - NativeMembar::instruction_size;
2692       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2693         NativeMembar *bar2 = NativeMembar_at(prev2);
2694         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2695         BLOCK_COMMENT("merged membar(elided)");
2696         return;
2697       }
2698     }
2699   }
2700   code()->set_last_insn(pc());
2701   dmb(Assembler::barrier(order_constraint));
2702 }
2703 
2704 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2705   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2706     merge_ldst(rt, adr, size_in_bytes, is_store);
2707     code()->clear_last_insn();
2708     return true;
2709   } else {
2710     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2711     const uint64_t mask = size_in_bytes - 1;
2712     if (adr.getMode() == Address::base_plus_offset &&
2713         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2714       code()->set_last_insn(pc());
2715     }
2716     return false;
2717   }
2718 }
2719 
2720 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2721   // We always try to merge two adjacent loads into one ldp.
2722   if (!try_merge_ldst(Rx, adr, 8, false)) {
2723     Assembler::ldr(Rx, adr);
2724   }
2725 }
2726 
2727 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2728   // We always try to merge two adjacent loads into one ldp.
2729   if (!try_merge_ldst(Rw, adr, 4, false)) {
2730     Assembler::ldrw(Rw, adr);
2731   }
2732 }
2733 
2734 void MacroAssembler::str(Register Rx, const Address &adr) {
2735   // We always try to merge two adjacent stores into one stp.
2736   if (!try_merge_ldst(Rx, adr, 8, true)) {
2737     Assembler::str(Rx, adr);
2738   }
2739 }
2740 
2741 void MacroAssembler::strw(Register Rw, const Address &adr) {
2742   // We always try to merge two adjacent stores into one stp.
2743   if (!try_merge_ldst(Rw, adr, 4, true)) {
2744     Assembler::strw(Rw, adr);
2745   }
2746 }
2747 
2748 // MacroAssembler routines found actually to be needed
2749 
2750 void MacroAssembler::push(Register src)
2751 {
2752   str(src, Address(pre(esp, -1 * wordSize)));
2753 }
2754 
2755 void MacroAssembler::pop(Register dst)
2756 {
2757   ldr(dst, Address(post(esp, 1 * wordSize)));
2758 }
2759 
2760 // Note: load_unsigned_short used to be called load_unsigned_word.
2761 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2762   int off = offset();
2763   ldrh(dst, src);
2764   return off;
2765 }
2766 
2767 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2768   int off = offset();
2769   ldrb(dst, src);
2770   return off;
2771 }
2772 
2773 int MacroAssembler::load_signed_short(Register dst, Address src) {
2774   int off = offset();
2775   ldrsh(dst, src);
2776   return off;
2777 }
2778 
2779 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2780   int off = offset();
2781   ldrsb(dst, src);
2782   return off;
2783 }
2784 
2785 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2786   int off = offset();
2787   ldrshw(dst, src);
2788   return off;
2789 }
2790 
2791 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2792   int off = offset();
2793   ldrsbw(dst, src);
2794   return off;
2795 }
2796 
2797 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2798   switch (size_in_bytes) {
2799   case  8:  ldr(dst, src); break;
2800   case  4:  ldrw(dst, src); break;
2801   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2802   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2803   default:  ShouldNotReachHere();
2804   }
2805 }
2806 
2807 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2808   switch (size_in_bytes) {
2809   case  8:  str(src, dst); break;
2810   case  4:  strw(src, dst); break;
2811   case  2:  strh(src, dst); break;
2812   case  1:  strb(src, dst); break;
2813   default:  ShouldNotReachHere();
2814   }
2815 }
2816 
2817 void MacroAssembler::decrementw(Register reg, int value)
2818 {
2819   if (value < 0)  { incrementw(reg, -value);      return; }
2820   if (value == 0) {                               return; }
2821   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2822   /* else */ {
2823     guarantee(reg != rscratch2, "invalid dst for register decrement");
2824     movw(rscratch2, (unsigned)value);
2825     subw(reg, reg, rscratch2);
2826   }
2827 }
2828 
2829 void MacroAssembler::decrement(Register reg, int value)
2830 {
2831   if (value < 0)  { increment(reg, -value);      return; }
2832   if (value == 0) {                              return; }
2833   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2834   /* else */ {
2835     assert(reg != rscratch2, "invalid dst for register decrement");
2836     mov(rscratch2, (uint64_t)value);
2837     sub(reg, reg, rscratch2);
2838   }
2839 }
2840 
2841 void MacroAssembler::decrementw(Address dst, int value)
2842 {
2843   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2844   if (dst.getMode() == Address::literal) {
2845     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2846     lea(rscratch2, dst);
2847     dst = Address(rscratch2);
2848   }
2849   ldrw(rscratch1, dst);
2850   decrementw(rscratch1, value);
2851   strw(rscratch1, dst);
2852 }
2853 
2854 void MacroAssembler::decrement(Address dst, int value)
2855 {
2856   assert(!dst.uses(rscratch1), "invalid address for decrement");
2857   if (dst.getMode() == Address::literal) {
2858     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2859     lea(rscratch2, dst);
2860     dst = Address(rscratch2);
2861   }
2862   ldr(rscratch1, dst);
2863   decrement(rscratch1, value);
2864   str(rscratch1, dst);
2865 }
2866 
2867 void MacroAssembler::incrementw(Register reg, int value)
2868 {
2869   if (value < 0)  { decrementw(reg, -value);      return; }
2870   if (value == 0) {                               return; }
2871   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2872   /* else */ {
2873     assert(reg != rscratch2, "invalid dst for register increment");
2874     movw(rscratch2, (unsigned)value);
2875     addw(reg, reg, rscratch2);
2876   }
2877 }
2878 
2879 void MacroAssembler::increment(Register reg, int value)
2880 {
2881   if (value < 0)  { decrement(reg, -value);      return; }
2882   if (value == 0) {                              return; }
2883   if (value < (1 << 12)) { add(reg, reg, value); return; }
2884   /* else */ {
2885     assert(reg != rscratch2, "invalid dst for register increment");
2886     movw(rscratch2, (unsigned)value);
2887     add(reg, reg, rscratch2);
2888   }
2889 }
2890 
2891 void MacroAssembler::incrementw(Address dst, int value)
2892 {
2893   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2894   if (dst.getMode() == Address::literal) {
2895     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2896     lea(rscratch2, dst);
2897     dst = Address(rscratch2);
2898   }
2899   ldrw(rscratch1, dst);
2900   incrementw(rscratch1, value);
2901   strw(rscratch1, dst);
2902 }
2903 
2904 void MacroAssembler::increment(Address dst, int value)
2905 {
2906   assert(!dst.uses(rscratch1), "invalid dst for address increment");
2907   if (dst.getMode() == Address::literal) {
2908     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2909     lea(rscratch2, dst);
2910     dst = Address(rscratch2);
2911   }
2912   ldr(rscratch1, dst);
2913   increment(rscratch1, value);
2914   str(rscratch1, dst);
2915 }
2916 
2917 // Push lots of registers in the bit set supplied.  Don't push sp.
2918 // Return the number of words pushed
2919 int MacroAssembler::push(unsigned int bitset, Register stack) {
2920   int words_pushed = 0;
2921 
2922   // Scan bitset to accumulate register pairs
2923   unsigned char regs[32];
2924   int count = 0;
2925   for (int reg = 0; reg <= 30; reg++) {
2926     if (1 & bitset)
2927       regs[count++] = reg;
2928     bitset >>= 1;
2929   }
2930   regs[count++] = zr->raw_encoding();
2931   count &= ~1;  // Only push an even number of regs
2932 
2933   if (count) {
2934     stp(as_Register(regs[0]), as_Register(regs[1]),
2935        Address(pre(stack, -count * wordSize)));
2936     words_pushed += 2;
2937   }
2938   for (int i = 2; i < count; i += 2) {
2939     stp(as_Register(regs[i]), as_Register(regs[i+1]),
2940        Address(stack, i * wordSize));
2941     words_pushed += 2;
2942   }
2943 
2944   assert(words_pushed == count, "oops, pushed != count");
2945 
2946   return count;
2947 }
2948 
2949 int MacroAssembler::pop(unsigned int bitset, Register stack) {
2950   int words_pushed = 0;
2951 
2952   // Scan bitset to accumulate register pairs
2953   unsigned char regs[32];
2954   int count = 0;
2955   for (int reg = 0; reg <= 30; reg++) {
2956     if (1 & bitset)
2957       regs[count++] = reg;
2958     bitset >>= 1;
2959   }
2960   regs[count++] = zr->raw_encoding();
2961   count &= ~1;
2962 
2963   for (int i = 2; i < count; i += 2) {
2964     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
2965        Address(stack, i * wordSize));
2966     words_pushed += 2;
2967   }
2968   if (count) {
2969     ldp(as_Register(regs[0]), as_Register(regs[1]),
2970        Address(post(stack, count * wordSize)));
2971     words_pushed += 2;
2972   }
2973 
2974   assert(words_pushed == count, "oops, pushed != count");
2975 
2976   return count;
2977 }
2978 
2979 // Push lots of registers in the bit set supplied.  Don't push sp.
2980 // Return the number of dwords pushed
2981 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
2982   int words_pushed = 0;
2983   bool use_sve = false;
2984   int sve_vector_size_in_bytes = 0;
2985 
2986 #ifdef COMPILER2
2987   use_sve = Matcher::supports_scalable_vector();
2988   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
2989 #endif
2990 
2991   // Scan bitset to accumulate register pairs
2992   unsigned char regs[32];
2993   int count = 0;
2994   for (int reg = 0; reg <= 31; reg++) {
2995     if (1 & bitset)
2996       regs[count++] = reg;
2997     bitset >>= 1;
2998   }
2999 
3000   if (count == 0) {
3001     return 0;
3002   }
3003 
3004   if (mode == PushPopFull) {
3005     if (use_sve && sve_vector_size_in_bytes > 16) {
3006       mode = PushPopSVE;
3007     } else {
3008       mode = PushPopNeon;
3009     }
3010   }
3011 
3012 #ifndef PRODUCT
3013   {
3014     char buffer[48];
3015     if (mode == PushPopSVE) {
3016       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3017     } else if (mode == PushPopNeon) {
3018       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3019     } else {
3020       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3021     }
3022     block_comment(buffer);
3023   }
3024 #endif
3025 
3026   if (mode == PushPopSVE) {
3027     sub(stack, stack, sve_vector_size_in_bytes * count);
3028     for (int i = 0; i < count; i++) {
3029       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3030     }
3031     return count * sve_vector_size_in_bytes / 8;
3032   }
3033 
3034   if (mode == PushPopNeon) {
3035     if (count == 1) {
3036       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3037       return 2;
3038     }
3039 
3040     bool odd = (count & 1) == 1;
3041     int push_slots = count + (odd ? 1 : 0);
3042 
3043     // Always pushing full 128 bit registers.
3044     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3045     words_pushed += 2;
3046 
3047     for (int i = 2; i + 1 < count; i += 2) {
3048       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3049       words_pushed += 2;
3050     }
3051 
3052     if (odd) {
3053       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3054       words_pushed++;
3055     }
3056 
3057     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3058     return count * 2;
3059   }
3060 
3061   if (mode == PushPopFp) {
3062     bool odd = (count & 1) == 1;
3063     int push_slots = count + (odd ? 1 : 0);
3064 
3065     if (count == 1) {
3066       // Stack pointer must be 16 bytes aligned
3067       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3068       return 1;
3069     }
3070 
3071     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3072     words_pushed += 2;
3073 
3074     for (int i = 2; i + 1 < count; i += 2) {
3075       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3076       words_pushed += 2;
3077     }
3078 
3079     if (odd) {
3080       // Stack pointer must be 16 bytes aligned
3081       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3082       words_pushed++;
3083     }
3084 
3085     assert(words_pushed == count, "oops, pushed != count");
3086 
3087     return count;
3088   }
3089 
3090   return 0;
3091 }
3092 
3093 // Return the number of dwords popped
3094 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3095   int words_pushed = 0;
3096   bool use_sve = false;
3097   int sve_vector_size_in_bytes = 0;
3098 
3099 #ifdef COMPILER2
3100   use_sve = Matcher::supports_scalable_vector();
3101   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3102 #endif
3103   // Scan bitset to accumulate register pairs
3104   unsigned char regs[32];
3105   int count = 0;
3106   for (int reg = 0; reg <= 31; reg++) {
3107     if (1 & bitset)
3108       regs[count++] = reg;
3109     bitset >>= 1;
3110   }
3111 
3112   if (count == 0) {
3113     return 0;
3114   }
3115 
3116   if (mode == PushPopFull) {
3117     if (use_sve && sve_vector_size_in_bytes > 16) {
3118       mode = PushPopSVE;
3119     } else {
3120       mode = PushPopNeon;
3121     }
3122   }
3123 
3124 #ifndef PRODUCT
3125   {
3126     char buffer[48];
3127     if (mode == PushPopSVE) {
3128       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3129     } else if (mode == PushPopNeon) {
3130       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3131     } else {
3132       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3133     }
3134     block_comment(buffer);
3135   }
3136 #endif
3137 
3138   if (mode == PushPopSVE) {
3139     for (int i = count - 1; i >= 0; i--) {
3140       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3141     }
3142     add(stack, stack, sve_vector_size_in_bytes * count);
3143     return count * sve_vector_size_in_bytes / 8;
3144   }
3145 
3146   if (mode == PushPopNeon) {
3147     if (count == 1) {
3148       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3149       return 2;
3150     }
3151 
3152     bool odd = (count & 1) == 1;
3153     int push_slots = count + (odd ? 1 : 0);
3154 
3155     if (odd) {
3156       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3157       words_pushed++;
3158     }
3159 
3160     for (int i = 2; i + 1 < count; i += 2) {
3161       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3162       words_pushed += 2;
3163     }
3164 
3165     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3166     words_pushed += 2;
3167 
3168     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3169 
3170     return count * 2;
3171   }
3172 
3173   if (mode == PushPopFp) {
3174     bool odd = (count & 1) == 1;
3175     int push_slots = count + (odd ? 1 : 0);
3176 
3177     if (count == 1) {
3178       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3179       return 1;
3180     }
3181 
3182     if (odd) {
3183       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3184       words_pushed++;
3185     }
3186 
3187     for (int i = 2; i + 1 < count; i += 2) {
3188       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3189       words_pushed += 2;
3190     }
3191 
3192     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3193     words_pushed += 2;
3194 
3195     assert(words_pushed == count, "oops, pushed != count");
3196 
3197     return count;
3198   }
3199 
3200   return 0;
3201 }
3202 
3203 // Return the number of dwords pushed
3204 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3205   bool use_sve = false;
3206   int sve_predicate_size_in_slots = 0;
3207 
3208 #ifdef COMPILER2
3209   use_sve = Matcher::supports_scalable_vector();
3210   if (use_sve) {
3211     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3212   }
3213 #endif
3214 
3215   if (!use_sve) {
3216     return 0;
3217   }
3218 
3219   unsigned char regs[PRegister::number_of_registers];
3220   int count = 0;
3221   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3222     if (1 & bitset)
3223       regs[count++] = reg;
3224     bitset >>= 1;
3225   }
3226 
3227   if (count == 0) {
3228     return 0;
3229   }
3230 
3231   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3232                                   VMRegImpl::stack_slot_size * count, 16);
3233   sub(stack, stack, total_push_bytes);
3234   for (int i = 0; i < count; i++) {
3235     sve_str(as_PRegister(regs[i]), Address(stack, i));
3236   }
3237   return total_push_bytes / 8;
3238 }
3239 
3240 // Return the number of dwords popped
3241 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3242   bool use_sve = false;
3243   int sve_predicate_size_in_slots = 0;
3244 
3245 #ifdef COMPILER2
3246   use_sve = Matcher::supports_scalable_vector();
3247   if (use_sve) {
3248     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3249   }
3250 #endif
3251 
3252   if (!use_sve) {
3253     return 0;
3254   }
3255 
3256   unsigned char regs[PRegister::number_of_registers];
3257   int count = 0;
3258   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3259     if (1 & bitset)
3260       regs[count++] = reg;
3261     bitset >>= 1;
3262   }
3263 
3264   if (count == 0) {
3265     return 0;
3266   }
3267 
3268   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3269                                  VMRegImpl::stack_slot_size * count, 16);
3270   for (int i = count - 1; i >= 0; i--) {
3271     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3272   }
3273   add(stack, stack, total_pop_bytes);
3274   return total_pop_bytes / 8;
3275 }
3276 
3277 #ifdef ASSERT
3278 void MacroAssembler::verify_heapbase(const char* msg) {
3279 #if 0
3280   assert (Universe::heap() != nullptr, "java heap should be initialized");
3281   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3282     // rheapbase is allocated as general register
3283     return;
3284   }
3285   if (CheckCompressedOops) {
3286     Label ok;
3287     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3288     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3289     br(Assembler::EQ, ok);
3290     stop(msg);
3291     bind(ok);
3292     pop(1 << rscratch1->encoding(), sp);
3293   }
3294 #endif
3295 }
3296 #endif
3297 
3298 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3299   assert_different_registers(value, tmp1, tmp2);
3300   Label done, tagged, weak_tagged;
3301 
3302   cbz(value, done);           // Use null as-is.
3303   tst(value, JNIHandles::tag_mask); // Test for tag.
3304   br(Assembler::NE, tagged);
3305 
3306   // Resolve local handle
3307   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3308   verify_oop(value);
3309   b(done);
3310 
3311   bind(tagged);
3312   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3313   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3314 
3315   // Resolve global handle
3316   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3317   verify_oop(value);
3318   b(done);
3319 
3320   bind(weak_tagged);
3321   // Resolve jweak.
3322   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3323                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3324   verify_oop(value);
3325 
3326   bind(done);
3327 }
3328 
3329 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3330   assert_different_registers(value, tmp1, tmp2);
3331   Label done;
3332 
3333   cbz(value, done);           // Use null as-is.
3334 
3335 #ifdef ASSERT
3336   {
3337     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3338     Label valid_global_tag;
3339     tbnz(value, 1, valid_global_tag); // Test for global tag
3340     stop("non global jobject using resolve_global_jobject");
3341     bind(valid_global_tag);
3342   }
3343 #endif
3344 
3345   // Resolve global handle
3346   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3347   verify_oop(value);
3348 
3349   bind(done);
3350 }
3351 
3352 void MacroAssembler::stop(const char* msg) {
3353   // Skip AOT caching C strings in scratch buffer.
3354   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3355   BLOCK_COMMENT(str);
3356   // load msg into r0 so we can access it from the signal handler
3357   // ExternalAddress enables saving and restoring via the code cache
3358   lea(c_rarg0, ExternalAddress((address) str));
3359   dcps1(0xdeae);
3360 }
3361 
3362 void MacroAssembler::unimplemented(const char* what) {
3363   const char* buf = nullptr;
3364   {
3365     ResourceMark rm;
3366     stringStream ss;
3367     ss.print("unimplemented: %s", what);
3368     buf = code_string(ss.as_string());
3369   }
3370   stop(buf);
3371 }
3372 
3373 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3374 #ifdef ASSERT
3375   Label OK;
3376   br(cc, OK);
3377   stop(msg);
3378   bind(OK);
3379 #endif
3380 }
3381 
3382 // If a constant does not fit in an immediate field, generate some
3383 // number of MOV instructions and then perform the operation.
3384 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3385                                            add_sub_imm_insn insn1,
3386                                            add_sub_reg_insn insn2,
3387                                            bool is32) {
3388   assert(Rd != zr, "Rd = zr and not setting flags?");
3389   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3390   if (fits) {
3391     (this->*insn1)(Rd, Rn, imm);
3392   } else {
3393     if (g_uabs(imm) < (1 << 24)) {
3394        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3395        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3396     } else {
3397        assert_different_registers(Rd, Rn);
3398        mov(Rd, imm);
3399        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3400     }
3401   }
3402 }
3403 
3404 // Separate vsn which sets the flags. Optimisations are more restricted
3405 // because we must set the flags correctly.
3406 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3407                                              add_sub_imm_insn insn1,
3408                                              add_sub_reg_insn insn2,
3409                                              bool is32) {
3410   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3411   if (fits) {
3412     (this->*insn1)(Rd, Rn, imm);
3413   } else {
3414     assert_different_registers(Rd, Rn);
3415     assert(Rd != zr, "overflow in immediate operand");
3416     mov(Rd, imm);
3417     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3418   }
3419 }
3420 
3421 
3422 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3423   if (increment.is_register()) {
3424     add(Rd, Rn, increment.as_register());
3425   } else {
3426     add(Rd, Rn, increment.as_constant());
3427   }
3428 }
3429 
3430 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3431   if (increment.is_register()) {
3432     addw(Rd, Rn, increment.as_register());
3433   } else {
3434     addw(Rd, Rn, increment.as_constant());
3435   }
3436 }
3437 
3438 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3439   if (decrement.is_register()) {
3440     sub(Rd, Rn, decrement.as_register());
3441   } else {
3442     sub(Rd, Rn, decrement.as_constant());
3443   }
3444 }
3445 
3446 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3447   if (decrement.is_register()) {
3448     subw(Rd, Rn, decrement.as_register());
3449   } else {
3450     subw(Rd, Rn, decrement.as_constant());
3451   }
3452 }
3453 
3454 void MacroAssembler::reinit_heapbase()
3455 {
3456   if (UseCompressedOops) {
3457     if (Universe::is_fully_initialized() && !AOTCodeCache::is_on_for_dump()) {
3458       mov(rheapbase, CompressedOops::base());
3459     } else {
3460       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3461       ldr(rheapbase, Address(rheapbase));
3462     }
3463   }
3464 }
3465 
3466 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3467 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3468 // Pass a register for the result, otherwise pass noreg.
3469 
3470 // Clobbers rscratch1
3471 void MacroAssembler::cmpxchg(Register addr, Register expected,
3472                              Register new_val,
3473                              enum operand_size size,
3474                              bool acquire, bool release,
3475                              bool weak,
3476                              Register result) {
3477   if (result == noreg)  result = rscratch1;
3478   BLOCK_COMMENT("cmpxchg {");
3479   if (UseLSE) {
3480     mov(result, expected);
3481     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3482     compare_eq(result, expected, size);
3483 #ifdef ASSERT
3484     // Poison rscratch1 which is written on !UseLSE branch
3485     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3486 #endif
3487   } else {
3488     Label retry_load, done;
3489     prfm(Address(addr), PSTL1STRM);
3490     bind(retry_load);
3491     load_exclusive(result, addr, size, acquire);
3492     compare_eq(result, expected, size);
3493     br(Assembler::NE, done);
3494     store_exclusive(rscratch1, new_val, addr, size, release);
3495     if (weak) {
3496       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3497     } else {
3498       cbnzw(rscratch1, retry_load);
3499     }
3500     bind(done);
3501   }
3502   BLOCK_COMMENT("} cmpxchg");
3503 }
3504 
3505 // A generic comparison. Only compares for equality, clobbers rscratch1.
3506 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3507   if (size == xword) {
3508     cmp(rm, rn);
3509   } else if (size == word) {
3510     cmpw(rm, rn);
3511   } else if (size == halfword) {
3512     eorw(rscratch1, rm, rn);
3513     ands(zr, rscratch1, 0xffff);
3514   } else if (size == byte) {
3515     eorw(rscratch1, rm, rn);
3516     ands(zr, rscratch1, 0xff);
3517   } else {
3518     ShouldNotReachHere();
3519   }
3520 }
3521 
3522 
3523 static bool different(Register a, RegisterOrConstant b, Register c) {
3524   if (b.is_constant())
3525     return a != c;
3526   else
3527     return a != b.as_register() && a != c && b.as_register() != c;
3528 }
3529 
3530 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3531 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3532   if (UseLSE) {                                                         \
3533     prev = prev->is_valid() ? prev : zr;                                \
3534     if (incr.is_register()) {                                           \
3535       AOP(sz, incr.as_register(), prev, addr);                          \
3536     } else {                                                            \
3537       mov(rscratch2, incr.as_constant());                               \
3538       AOP(sz, rscratch2, prev, addr);                                   \
3539     }                                                                   \
3540     return;                                                             \
3541   }                                                                     \
3542   Register result = rscratch2;                                          \
3543   if (prev->is_valid())                                                 \
3544     result = different(prev, incr, addr) ? prev : rscratch2;            \
3545                                                                         \
3546   Label retry_load;                                                     \
3547   prfm(Address(addr), PSTL1STRM);                                       \
3548   bind(retry_load);                                                     \
3549   LDXR(result, addr);                                                   \
3550   OP(rscratch1, result, incr);                                          \
3551   STXR(rscratch2, rscratch1, addr);                                     \
3552   cbnzw(rscratch2, retry_load);                                         \
3553   if (prev->is_valid() && prev != result) {                             \
3554     IOP(prev, rscratch1, incr);                                         \
3555   }                                                                     \
3556 }
3557 
3558 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3559 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3560 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3561 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3562 
3563 #undef ATOMIC_OP
3564 
3565 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3566 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3567   if (UseLSE) {                                                         \
3568     prev = prev->is_valid() ? prev : zr;                                \
3569     AOP(sz, newv, prev, addr);                                          \
3570     return;                                                             \
3571   }                                                                     \
3572   Register result = rscratch2;                                          \
3573   if (prev->is_valid())                                                 \
3574     result = different(prev, newv, addr) ? prev : rscratch2;            \
3575                                                                         \
3576   Label retry_load;                                                     \
3577   prfm(Address(addr), PSTL1STRM);                                       \
3578   bind(retry_load);                                                     \
3579   LDXR(result, addr);                                                   \
3580   STXR(rscratch1, newv, addr);                                          \
3581   cbnzw(rscratch1, retry_load);                                         \
3582   if (prev->is_valid() && prev != result)                               \
3583     mov(prev, result);                                                  \
3584 }
3585 
3586 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3587 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3588 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3589 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3590 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3591 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3592 
3593 #undef ATOMIC_XCHG
3594 
3595 #ifndef PRODUCT
3596 extern "C" void findpc(intptr_t x);
3597 #endif
3598 
3599 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3600 {
3601   // In order to get locks to work, we need to fake a in_VM state
3602   if (ShowMessageBoxOnError) {
3603     JavaThread* thread = JavaThread::current();
3604     thread->set_thread_state(_thread_in_vm);
3605 #ifndef PRODUCT
3606     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3607       ttyLocker ttyl;
3608       BytecodeCounter::print();
3609     }
3610 #endif
3611     if (os::message_box(msg, "Execution stopped, print registers?")) {
3612       ttyLocker ttyl;
3613       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3614 #ifndef PRODUCT
3615       tty->cr();
3616       findpc(pc);
3617       tty->cr();
3618 #endif
3619       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3620       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3621       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3622       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3623       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3624       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3625       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3626       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3627       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3628       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3629       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3630       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3631       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3632       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3633       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3634       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3635       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3636       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3637       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3638       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3639       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3640       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3641       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3642       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3643       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3644       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3645       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3646       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3647       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3648       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3649       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3650       BREAKPOINT;
3651     }
3652   }
3653   fatal("DEBUG MESSAGE: %s", msg);
3654 }
3655 
3656 RegSet MacroAssembler::call_clobbered_gp_registers() {
3657   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3658 #ifndef R18_RESERVED
3659   regs += r18_tls;
3660 #endif
3661   return regs;
3662 }
3663 
3664 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3665   int step = 4 * wordSize;
3666   push(call_clobbered_gp_registers() - exclude, sp);
3667   sub(sp, sp, step);
3668   mov(rscratch1, -step);
3669   // Push v0-v7, v16-v31.
3670   for (int i = 31; i>= 4; i -= 4) {
3671     if (i <= v7->encoding() || i >= v16->encoding())
3672       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3673           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3674   }
3675   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3676       as_FloatRegister(3), T1D, Address(sp));
3677 }
3678 
3679 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3680   for (int i = 0; i < 32; i += 4) {
3681     if (i <= v7->encoding() || i >= v16->encoding())
3682       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3683           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3684   }
3685 
3686   reinitialize_ptrue();
3687 
3688   pop(call_clobbered_gp_registers() - exclude, sp);
3689 }
3690 
3691 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3692                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3693   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3694   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3695     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3696     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3697       sve_str(as_FloatRegister(i), Address(sp, i));
3698     }
3699   } else {
3700     int step = (save_vectors ? 8 : 4) * wordSize;
3701     mov(rscratch1, -step);
3702     sub(sp, sp, step);
3703     for (int i = 28; i >= 4; i -= 4) {
3704       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3705           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3706     }
3707     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3708   }
3709   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3710     sub(sp, sp, total_predicate_in_bytes);
3711     for (int i = 0; i < PRegister::number_of_registers; i++) {
3712       sve_str(as_PRegister(i), Address(sp, i));
3713     }
3714   }
3715 }
3716 
3717 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3718                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3719   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3720     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3721       sve_ldr(as_PRegister(i), Address(sp, i));
3722     }
3723     add(sp, sp, total_predicate_in_bytes);
3724   }
3725   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3726     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3727       sve_ldr(as_FloatRegister(i), Address(sp, i));
3728     }
3729     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3730   } else {
3731     int step = (restore_vectors ? 8 : 4) * wordSize;
3732     for (int i = 0; i <= 28; i += 4)
3733       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3734           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3735   }
3736 
3737   // We may use predicate registers and rely on ptrue with SVE,
3738   // regardless of wide vector (> 8 bytes) used or not.
3739   if (use_sve) {
3740     reinitialize_ptrue();
3741   }
3742 
3743   // integer registers except lr & sp
3744   pop(RegSet::range(r0, r17), sp);
3745 #ifdef R18_RESERVED
3746   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3747   pop(RegSet::range(r20, r29), sp);
3748 #else
3749   pop(RegSet::range(r18_tls, r29), sp);
3750 #endif
3751 }
3752 
3753 /**
3754  * Helpers for multiply_to_len().
3755  */
3756 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3757                                      Register src1, Register src2) {
3758   adds(dest_lo, dest_lo, src1);
3759   adc(dest_hi, dest_hi, zr);
3760   adds(dest_lo, dest_lo, src2);
3761   adc(final_dest_hi, dest_hi, zr);
3762 }
3763 
3764 // Generate an address from (r + r1 extend offset).  "size" is the
3765 // size of the operand.  The result may be in rscratch2.
3766 Address MacroAssembler::offsetted_address(Register r, Register r1,
3767                                           Address::extend ext, int offset, int size) {
3768   if (offset || (ext.shift() % size != 0)) {
3769     lea(rscratch2, Address(r, r1, ext));
3770     return Address(rscratch2, offset);
3771   } else {
3772     return Address(r, r1, ext);
3773   }
3774 }
3775 
3776 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3777 {
3778   assert(offset >= 0, "spill to negative address?");
3779   // Offset reachable ?
3780   //   Not aligned - 9 bits signed offset
3781   //   Aligned - 12 bits unsigned offset shifted
3782   Register base = sp;
3783   if ((offset & (size-1)) && offset >= (1<<8)) {
3784     add(tmp, base, offset & ((1<<12)-1));
3785     base = tmp;
3786     offset &= -1u<<12;
3787   }
3788 
3789   if (offset >= (1<<12) * size) {
3790     add(tmp, base, offset & (((1<<12)-1)<<12));
3791     base = tmp;
3792     offset &= ~(((1<<12)-1)<<12);
3793   }
3794 
3795   return Address(base, offset);
3796 }
3797 
3798 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3799   assert(offset >= 0, "spill to negative address?");
3800 
3801   Register base = sp;
3802 
3803   // An immediate offset in the range 0 to 255 which is multiplied
3804   // by the current vector or predicate register size in bytes.
3805   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3806     return Address(base, offset / sve_reg_size_in_bytes);
3807   }
3808 
3809   add(tmp, base, offset);
3810   return Address(tmp);
3811 }
3812 
3813 // Checks whether offset is aligned.
3814 // Returns true if it is, else false.
3815 bool MacroAssembler::merge_alignment_check(Register base,
3816                                            size_t size,
3817                                            int64_t cur_offset,
3818                                            int64_t prev_offset) const {
3819   if (AvoidUnalignedAccesses) {
3820     if (base == sp) {
3821       // Checks whether low offset if aligned to pair of registers.
3822       int64_t pair_mask = size * 2 - 1;
3823       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3824       return (offset & pair_mask) == 0;
3825     } else { // If base is not sp, we can't guarantee the access is aligned.
3826       return false;
3827     }
3828   } else {
3829     int64_t mask = size - 1;
3830     // Load/store pair instruction only supports element size aligned offset.
3831     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3832   }
3833 }
3834 
3835 // Checks whether current and previous loads/stores can be merged.
3836 // Returns true if it can be merged, else false.
3837 bool MacroAssembler::ldst_can_merge(Register rt,
3838                                     const Address &adr,
3839                                     size_t cur_size_in_bytes,
3840                                     bool is_store) const {
3841   address prev = pc() - NativeInstruction::instruction_size;
3842   address last = code()->last_insn();
3843 
3844   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3845     return false;
3846   }
3847 
3848   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3849     return false;
3850   }
3851 
3852   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3853   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3854 
3855   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3856   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3857 
3858   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3859     return false;
3860   }
3861 
3862   int64_t max_offset = 63 * prev_size_in_bytes;
3863   int64_t min_offset = -64 * prev_size_in_bytes;
3864 
3865   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3866 
3867   // Only same base can be merged.
3868   if (adr.base() != prev_ldst->base()) {
3869     return false;
3870   }
3871 
3872   int64_t cur_offset = adr.offset();
3873   int64_t prev_offset = prev_ldst->offset();
3874   size_t diff = abs(cur_offset - prev_offset);
3875   if (diff != prev_size_in_bytes) {
3876     return false;
3877   }
3878 
3879   // Following cases can not be merged:
3880   // ldr x2, [x2, #8]
3881   // ldr x3, [x2, #16]
3882   // or:
3883   // ldr x2, [x3, #8]
3884   // ldr x2, [x3, #16]
3885   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3886   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3887     return false;
3888   }
3889 
3890   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3891   // Offset range must be in ldp/stp instruction's range.
3892   if (low_offset > max_offset || low_offset < min_offset) {
3893     return false;
3894   }
3895 
3896   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
3897     return true;
3898   }
3899 
3900   return false;
3901 }
3902 
3903 // Merge current load/store with previous load/store into ldp/stp.
3904 void MacroAssembler::merge_ldst(Register rt,
3905                                 const Address &adr,
3906                                 size_t cur_size_in_bytes,
3907                                 bool is_store) {
3908 
3909   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
3910 
3911   Register rt_low, rt_high;
3912   address prev = pc() - NativeInstruction::instruction_size;
3913   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3914 
3915   int64_t offset;
3916 
3917   if (adr.offset() < prev_ldst->offset()) {
3918     offset = adr.offset();
3919     rt_low = rt;
3920     rt_high = prev_ldst->target();
3921   } else {
3922     offset = prev_ldst->offset();
3923     rt_low = prev_ldst->target();
3924     rt_high = rt;
3925   }
3926 
3927   Address adr_p = Address(prev_ldst->base(), offset);
3928   // Overwrite previous generated binary.
3929   code_section()->set_end(prev);
3930 
3931   const size_t sz = prev_ldst->size_in_bytes();
3932   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
3933   if (!is_store) {
3934     BLOCK_COMMENT("merged ldr pair");
3935     if (sz == 8) {
3936       ldp(rt_low, rt_high, adr_p);
3937     } else {
3938       ldpw(rt_low, rt_high, adr_p);
3939     }
3940   } else {
3941     BLOCK_COMMENT("merged str pair");
3942     if (sz == 8) {
3943       stp(rt_low, rt_high, adr_p);
3944     } else {
3945       stpw(rt_low, rt_high, adr_p);
3946     }
3947   }
3948 }
3949 
3950 /**
3951  * Multiply 64 bit by 64 bit first loop.
3952  */
3953 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
3954                                            Register y, Register y_idx, Register z,
3955                                            Register carry, Register product,
3956                                            Register idx, Register kdx) {
3957   //
3958   //  jlong carry, x[], y[], z[];
3959   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
3960   //    huge_128 product = y[idx] * x[xstart] + carry;
3961   //    z[kdx] = (jlong)product;
3962   //    carry  = (jlong)(product >>> 64);
3963   //  }
3964   //  z[xstart] = carry;
3965   //
3966 
3967   Label L_first_loop, L_first_loop_exit;
3968   Label L_one_x, L_one_y, L_multiply;
3969 
3970   subsw(xstart, xstart, 1);
3971   br(Assembler::MI, L_one_x);
3972 
3973   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
3974   ldr(x_xstart, Address(rscratch1));
3975   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
3976 
3977   bind(L_first_loop);
3978   subsw(idx, idx, 1);
3979   br(Assembler::MI, L_first_loop_exit);
3980   subsw(idx, idx, 1);
3981   br(Assembler::MI, L_one_y);
3982   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
3983   ldr(y_idx, Address(rscratch1));
3984   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
3985   bind(L_multiply);
3986 
3987   // AArch64 has a multiply-accumulate instruction that we can't use
3988   // here because it has no way to process carries, so we have to use
3989   // separate add and adc instructions.  Bah.
3990   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
3991   mul(product, x_xstart, y_idx);
3992   adds(product, product, carry);
3993   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
3994 
3995   subw(kdx, kdx, 2);
3996   ror(product, product, 32); // back to big-endian
3997   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
3998 
3999   b(L_first_loop);
4000 
4001   bind(L_one_y);
4002   ldrw(y_idx, Address(y,  0));
4003   b(L_multiply);
4004 
4005   bind(L_one_x);
4006   ldrw(x_xstart, Address(x,  0));
4007   b(L_first_loop);
4008 
4009   bind(L_first_loop_exit);
4010 }
4011 
4012 /**
4013  * Multiply 128 bit by 128. Unrolled inner loop.
4014  *
4015  */
4016 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4017                                              Register carry, Register carry2,
4018                                              Register idx, Register jdx,
4019                                              Register yz_idx1, Register yz_idx2,
4020                                              Register tmp, Register tmp3, Register tmp4,
4021                                              Register tmp6, Register product_hi) {
4022 
4023   //   jlong carry, x[], y[], z[];
4024   //   int kdx = ystart+1;
4025   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4026   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4027   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4028   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4029   //     carry  = (jlong)(tmp4 >>> 64);
4030   //     z[kdx+idx+1] = (jlong)tmp3;
4031   //     z[kdx+idx] = (jlong)tmp4;
4032   //   }
4033   //   idx += 2;
4034   //   if (idx > 0) {
4035   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4036   //     z[kdx+idx] = (jlong)yz_idx1;
4037   //     carry  = (jlong)(yz_idx1 >>> 64);
4038   //   }
4039   //
4040 
4041   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4042 
4043   lsrw(jdx, idx, 2);
4044 
4045   bind(L_third_loop);
4046 
4047   subsw(jdx, jdx, 1);
4048   br(Assembler::MI, L_third_loop_exit);
4049   subw(idx, idx, 4);
4050 
4051   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4052 
4053   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4054 
4055   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4056 
4057   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4058   ror(yz_idx2, yz_idx2, 32);
4059 
4060   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4061 
4062   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4063   umulh(tmp4, product_hi, yz_idx1);
4064 
4065   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4066   ror(rscratch2, rscratch2, 32);
4067 
4068   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4069   umulh(carry2, product_hi, yz_idx2);
4070 
4071   // propagate sum of both multiplications into carry:tmp4:tmp3
4072   adds(tmp3, tmp3, carry);
4073   adc(tmp4, tmp4, zr);
4074   adds(tmp3, tmp3, rscratch1);
4075   adcs(tmp4, tmp4, tmp);
4076   adc(carry, carry2, zr);
4077   adds(tmp4, tmp4, rscratch2);
4078   adc(carry, carry, zr);
4079 
4080   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4081   ror(tmp4, tmp4, 32);
4082   stp(tmp4, tmp3, Address(tmp6, 0));
4083 
4084   b(L_third_loop);
4085   bind (L_third_loop_exit);
4086 
4087   andw (idx, idx, 0x3);
4088   cbz(idx, L_post_third_loop_done);
4089 
4090   Label L_check_1;
4091   subsw(idx, idx, 2);
4092   br(Assembler::MI, L_check_1);
4093 
4094   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4095   ldr(yz_idx1, Address(rscratch1, 0));
4096   ror(yz_idx1, yz_idx1, 32);
4097   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4098   umulh(tmp4, product_hi, yz_idx1);
4099   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4100   ldr(yz_idx2, Address(rscratch1, 0));
4101   ror(yz_idx2, yz_idx2, 32);
4102 
4103   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4104 
4105   ror(tmp3, tmp3, 32);
4106   str(tmp3, Address(rscratch1, 0));
4107 
4108   bind (L_check_1);
4109 
4110   andw (idx, idx, 0x1);
4111   subsw(idx, idx, 1);
4112   br(Assembler::MI, L_post_third_loop_done);
4113   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4114   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4115   umulh(carry2, tmp4, product_hi);
4116   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4117 
4118   add2_with_carry(carry2, tmp3, tmp4, carry);
4119 
4120   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4121   extr(carry, carry2, tmp3, 32);
4122 
4123   bind(L_post_third_loop_done);
4124 }
4125 
4126 /**
4127  * Code for BigInteger::multiplyToLen() intrinsic.
4128  *
4129  * r0: x
4130  * r1: xlen
4131  * r2: y
4132  * r3: ylen
4133  * r4:  z
4134  * r5: tmp0
4135  * r10: tmp1
4136  * r11: tmp2
4137  * r12: tmp3
4138  * r13: tmp4
4139  * r14: tmp5
4140  * r15: tmp6
4141  * r16: tmp7
4142  *
4143  */
4144 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4145                                      Register z, Register tmp0,
4146                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4147                                      Register tmp5, Register tmp6, Register product_hi) {
4148 
4149   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4150 
4151   const Register idx = tmp1;
4152   const Register kdx = tmp2;
4153   const Register xstart = tmp3;
4154 
4155   const Register y_idx = tmp4;
4156   const Register carry = tmp5;
4157   const Register product  = xlen;
4158   const Register x_xstart = tmp0;
4159 
4160   // First Loop.
4161   //
4162   //  final static long LONG_MASK = 0xffffffffL;
4163   //  int xstart = xlen - 1;
4164   //  int ystart = ylen - 1;
4165   //  long carry = 0;
4166   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4167   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4168   //    z[kdx] = (int)product;
4169   //    carry = product >>> 32;
4170   //  }
4171   //  z[xstart] = (int)carry;
4172   //
4173 
4174   movw(idx, ylen);       // idx = ylen;
4175   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4176   mov(carry, zr);        // carry = 0;
4177 
4178   Label L_done;
4179 
4180   movw(xstart, xlen);
4181   subsw(xstart, xstart, 1);
4182   br(Assembler::MI, L_done);
4183 
4184   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4185 
4186   Label L_second_loop;
4187   cbzw(kdx, L_second_loop);
4188 
4189   Label L_carry;
4190   subw(kdx, kdx, 1);
4191   cbzw(kdx, L_carry);
4192 
4193   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4194   lsr(carry, carry, 32);
4195   subw(kdx, kdx, 1);
4196 
4197   bind(L_carry);
4198   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4199 
4200   // Second and third (nested) loops.
4201   //
4202   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4203   //   carry = 0;
4204   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4205   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4206   //                    (z[k] & LONG_MASK) + carry;
4207   //     z[k] = (int)product;
4208   //     carry = product >>> 32;
4209   //   }
4210   //   z[i] = (int)carry;
4211   // }
4212   //
4213   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4214 
4215   const Register jdx = tmp1;
4216 
4217   bind(L_second_loop);
4218   mov(carry, zr);                // carry = 0;
4219   movw(jdx, ylen);               // j = ystart+1
4220 
4221   subsw(xstart, xstart, 1);      // i = xstart-1;
4222   br(Assembler::MI, L_done);
4223 
4224   str(z, Address(pre(sp, -4 * wordSize)));
4225 
4226   Label L_last_x;
4227   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4228   subsw(xstart, xstart, 1);       // i = xstart-1;
4229   br(Assembler::MI, L_last_x);
4230 
4231   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4232   ldr(product_hi, Address(rscratch1));
4233   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4234 
4235   Label L_third_loop_prologue;
4236   bind(L_third_loop_prologue);
4237 
4238   str(ylen, Address(sp, wordSize));
4239   stp(x, xstart, Address(sp, 2 * wordSize));
4240   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4241                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4242   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4243   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4244 
4245   addw(tmp3, xlen, 1);
4246   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4247   subsw(tmp3, tmp3, 1);
4248   br(Assembler::MI, L_done);
4249 
4250   lsr(carry, carry, 32);
4251   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4252   b(L_second_loop);
4253 
4254   // Next infrequent code is moved outside loops.
4255   bind(L_last_x);
4256   ldrw(product_hi, Address(x,  0));
4257   b(L_third_loop_prologue);
4258 
4259   bind(L_done);
4260 }
4261 
4262 // Code for BigInteger::mulAdd intrinsic
4263 // out     = r0
4264 // in      = r1
4265 // offset  = r2  (already out.length-offset)
4266 // len     = r3
4267 // k       = r4
4268 //
4269 // pseudo code from java implementation:
4270 // carry = 0;
4271 // offset = out.length-offset - 1;
4272 // for (int j=len-1; j >= 0; j--) {
4273 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4274 //     out[offset--] = (int)product;
4275 //     carry = product >>> 32;
4276 // }
4277 // return (int)carry;
4278 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4279       Register len, Register k) {
4280     Label LOOP, END;
4281     // pre-loop
4282     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4283     csel(out, zr, out, Assembler::EQ);
4284     br(Assembler::EQ, END);
4285     add(in, in, len, LSL, 2); // in[j+1] address
4286     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4287     mov(out, zr); // used to keep carry now
4288     BIND(LOOP);
4289     ldrw(rscratch1, Address(pre(in, -4)));
4290     madd(rscratch1, rscratch1, k, out);
4291     ldrw(rscratch2, Address(pre(offset, -4)));
4292     add(rscratch1, rscratch1, rscratch2);
4293     strw(rscratch1, Address(offset));
4294     lsr(out, rscratch1, 32);
4295     subs(len, len, 1);
4296     br(Assembler::NE, LOOP);
4297     BIND(END);
4298 }
4299 
4300 /**
4301  * Emits code to update CRC-32 with a byte value according to constants in table
4302  *
4303  * @param [in,out]crc   Register containing the crc.
4304  * @param [in]val       Register containing the byte to fold into the CRC.
4305  * @param [in]table     Register containing the table of crc constants.
4306  *
4307  * uint32_t crc;
4308  * val = crc_table[(val ^ crc) & 0xFF];
4309  * crc = val ^ (crc >> 8);
4310  *
4311  */
4312 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4313   eor(val, val, crc);
4314   andr(val, val, 0xff);
4315   ldrw(val, Address(table, val, Address::lsl(2)));
4316   eor(crc, val, crc, Assembler::LSR, 8);
4317 }
4318 
4319 /**
4320  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4321  *
4322  * @param [in,out]crc   Register containing the crc.
4323  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4324  * @param [in]table0    Register containing table 0 of crc constants.
4325  * @param [in]table1    Register containing table 1 of crc constants.
4326  * @param [in]table2    Register containing table 2 of crc constants.
4327  * @param [in]table3    Register containing table 3 of crc constants.
4328  *
4329  * uint32_t crc;
4330  *   v = crc ^ v
4331  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4332  *
4333  */
4334 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4335         Register table0, Register table1, Register table2, Register table3,
4336         bool upper) {
4337   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4338   uxtb(tmp, v);
4339   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4340   ubfx(tmp, v, 8, 8);
4341   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4342   eor(crc, crc, tmp);
4343   ubfx(tmp, v, 16, 8);
4344   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4345   eor(crc, crc, tmp);
4346   ubfx(tmp, v, 24, 8);
4347   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4348   eor(crc, crc, tmp);
4349 }
4350 
4351 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4352         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4353     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4354     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4355 
4356     subs(tmp0, len, 384);
4357     mvnw(crc, crc);
4358     br(Assembler::GE, CRC_by128_pre);
4359   BIND(CRC_less128);
4360     subs(len, len, 32);
4361     br(Assembler::GE, CRC_by32_loop);
4362   BIND(CRC_less32);
4363     adds(len, len, 32 - 4);
4364     br(Assembler::GE, CRC_by4_loop);
4365     adds(len, len, 4);
4366     br(Assembler::GT, CRC_by1_loop);
4367     b(L_exit);
4368 
4369   BIND(CRC_by32_loop);
4370     ldp(tmp0, tmp1, Address(buf));
4371     crc32x(crc, crc, tmp0);
4372     ldp(tmp2, tmp3, Address(buf, 16));
4373     crc32x(crc, crc, tmp1);
4374     add(buf, buf, 32);
4375     crc32x(crc, crc, tmp2);
4376     subs(len, len, 32);
4377     crc32x(crc, crc, tmp3);
4378     br(Assembler::GE, CRC_by32_loop);
4379     cmn(len, (u1)32);
4380     br(Assembler::NE, CRC_less32);
4381     b(L_exit);
4382 
4383   BIND(CRC_by4_loop);
4384     ldrw(tmp0, Address(post(buf, 4)));
4385     subs(len, len, 4);
4386     crc32w(crc, crc, tmp0);
4387     br(Assembler::GE, CRC_by4_loop);
4388     adds(len, len, 4);
4389     br(Assembler::LE, L_exit);
4390   BIND(CRC_by1_loop);
4391     ldrb(tmp0, Address(post(buf, 1)));
4392     subs(len, len, 1);
4393     crc32b(crc, crc, tmp0);
4394     br(Assembler::GT, CRC_by1_loop);
4395     b(L_exit);
4396 
4397   BIND(CRC_by128_pre);
4398     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4399       4*256*sizeof(juint) + 8*sizeof(juint));
4400     mov(crc, 0);
4401     crc32x(crc, crc, tmp0);
4402     crc32x(crc, crc, tmp1);
4403 
4404     cbnz(len, CRC_less128);
4405 
4406   BIND(L_exit);
4407     mvnw(crc, crc);
4408 }
4409 
4410 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4411         Register len, Register tmp0, Register tmp1, Register tmp2,
4412         Register tmp3) {
4413     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4414     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4415 
4416     mvnw(crc, crc);
4417 
4418     subs(len, len, 128);
4419     br(Assembler::GE, CRC_by64_pre);
4420   BIND(CRC_less64);
4421     adds(len, len, 128-32);
4422     br(Assembler::GE, CRC_by32_loop);
4423   BIND(CRC_less32);
4424     adds(len, len, 32-4);
4425     br(Assembler::GE, CRC_by4_loop);
4426     adds(len, len, 4);
4427     br(Assembler::GT, CRC_by1_loop);
4428     b(L_exit);
4429 
4430   BIND(CRC_by32_loop);
4431     ldp(tmp0, tmp1, Address(post(buf, 16)));
4432     subs(len, len, 32);
4433     crc32x(crc, crc, tmp0);
4434     ldr(tmp2, Address(post(buf, 8)));
4435     crc32x(crc, crc, tmp1);
4436     ldr(tmp3, Address(post(buf, 8)));
4437     crc32x(crc, crc, tmp2);
4438     crc32x(crc, crc, tmp3);
4439     br(Assembler::GE, CRC_by32_loop);
4440     cmn(len, (u1)32);
4441     br(Assembler::NE, CRC_less32);
4442     b(L_exit);
4443 
4444   BIND(CRC_by4_loop);
4445     ldrw(tmp0, Address(post(buf, 4)));
4446     subs(len, len, 4);
4447     crc32w(crc, crc, tmp0);
4448     br(Assembler::GE, CRC_by4_loop);
4449     adds(len, len, 4);
4450     br(Assembler::LE, L_exit);
4451   BIND(CRC_by1_loop);
4452     ldrb(tmp0, Address(post(buf, 1)));
4453     subs(len, len, 1);
4454     crc32b(crc, crc, tmp0);
4455     br(Assembler::GT, CRC_by1_loop);
4456     b(L_exit);
4457 
4458   BIND(CRC_by64_pre);
4459     sub(buf, buf, 8);
4460     ldp(tmp0, tmp1, Address(buf, 8));
4461     crc32x(crc, crc, tmp0);
4462     ldr(tmp2, Address(buf, 24));
4463     crc32x(crc, crc, tmp1);
4464     ldr(tmp3, Address(buf, 32));
4465     crc32x(crc, crc, tmp2);
4466     ldr(tmp0, Address(buf, 40));
4467     crc32x(crc, crc, tmp3);
4468     ldr(tmp1, Address(buf, 48));
4469     crc32x(crc, crc, tmp0);
4470     ldr(tmp2, Address(buf, 56));
4471     crc32x(crc, crc, tmp1);
4472     ldr(tmp3, Address(pre(buf, 64)));
4473 
4474     b(CRC_by64_loop);
4475 
4476     align(CodeEntryAlignment);
4477   BIND(CRC_by64_loop);
4478     subs(len, len, 64);
4479     crc32x(crc, crc, tmp2);
4480     ldr(tmp0, Address(buf, 8));
4481     crc32x(crc, crc, tmp3);
4482     ldr(tmp1, Address(buf, 16));
4483     crc32x(crc, crc, tmp0);
4484     ldr(tmp2, Address(buf, 24));
4485     crc32x(crc, crc, tmp1);
4486     ldr(tmp3, Address(buf, 32));
4487     crc32x(crc, crc, tmp2);
4488     ldr(tmp0, Address(buf, 40));
4489     crc32x(crc, crc, tmp3);
4490     ldr(tmp1, Address(buf, 48));
4491     crc32x(crc, crc, tmp0);
4492     ldr(tmp2, Address(buf, 56));
4493     crc32x(crc, crc, tmp1);
4494     ldr(tmp3, Address(pre(buf, 64)));
4495     br(Assembler::GE, CRC_by64_loop);
4496 
4497     // post-loop
4498     crc32x(crc, crc, tmp2);
4499     crc32x(crc, crc, tmp3);
4500 
4501     sub(len, len, 64);
4502     add(buf, buf, 8);
4503     cmn(len, (u1)128);
4504     br(Assembler::NE, CRC_less64);
4505   BIND(L_exit);
4506     mvnw(crc, crc);
4507 }
4508 
4509 /**
4510  * @param crc   register containing existing CRC (32-bit)
4511  * @param buf   register pointing to input byte buffer (byte*)
4512  * @param len   register containing number of bytes
4513  * @param table register that will contain address of CRC table
4514  * @param tmp   scratch register
4515  */
4516 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4517         Register table0, Register table1, Register table2, Register table3,
4518         Register tmp, Register tmp2, Register tmp3) {
4519   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4520 
4521   if (UseCryptoPmullForCRC32) {
4522       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4523       return;
4524   }
4525 
4526   if (UseCRC32) {
4527       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4528       return;
4529   }
4530 
4531     mvnw(crc, crc);
4532 
4533     {
4534       uint64_t offset;
4535       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4536       add(table0, table0, offset);
4537     }
4538     add(table1, table0, 1*256*sizeof(juint));
4539     add(table2, table0, 2*256*sizeof(juint));
4540     add(table3, table0, 3*256*sizeof(juint));
4541 
4542     { // Neon code start
4543       cmp(len, (u1)64);
4544       br(Assembler::LT, L_by16);
4545       eor(v16, T16B, v16, v16);
4546 
4547     Label L_fold;
4548 
4549       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4550 
4551       ld1(v0, v1, T2D, post(buf, 32));
4552       ld1r(v4, T2D, post(tmp, 8));
4553       ld1r(v5, T2D, post(tmp, 8));
4554       ld1r(v6, T2D, post(tmp, 8));
4555       ld1r(v7, T2D, post(tmp, 8));
4556       mov(v16, S, 0, crc);
4557 
4558       eor(v0, T16B, v0, v16);
4559       sub(len, len, 64);
4560 
4561     BIND(L_fold);
4562       pmull(v22, T8H, v0, v5, T8B);
4563       pmull(v20, T8H, v0, v7, T8B);
4564       pmull(v23, T8H, v0, v4, T8B);
4565       pmull(v21, T8H, v0, v6, T8B);
4566 
4567       pmull2(v18, T8H, v0, v5, T16B);
4568       pmull2(v16, T8H, v0, v7, T16B);
4569       pmull2(v19, T8H, v0, v4, T16B);
4570       pmull2(v17, T8H, v0, v6, T16B);
4571 
4572       uzp1(v24, T8H, v20, v22);
4573       uzp2(v25, T8H, v20, v22);
4574       eor(v20, T16B, v24, v25);
4575 
4576       uzp1(v26, T8H, v16, v18);
4577       uzp2(v27, T8H, v16, v18);
4578       eor(v16, T16B, v26, v27);
4579 
4580       ushll2(v22, T4S, v20, T8H, 8);
4581       ushll(v20, T4S, v20, T4H, 8);
4582 
4583       ushll2(v18, T4S, v16, T8H, 8);
4584       ushll(v16, T4S, v16, T4H, 8);
4585 
4586       eor(v22, T16B, v23, v22);
4587       eor(v18, T16B, v19, v18);
4588       eor(v20, T16B, v21, v20);
4589       eor(v16, T16B, v17, v16);
4590 
4591       uzp1(v17, T2D, v16, v20);
4592       uzp2(v21, T2D, v16, v20);
4593       eor(v17, T16B, v17, v21);
4594 
4595       ushll2(v20, T2D, v17, T4S, 16);
4596       ushll(v16, T2D, v17, T2S, 16);
4597 
4598       eor(v20, T16B, v20, v22);
4599       eor(v16, T16B, v16, v18);
4600 
4601       uzp1(v17, T2D, v20, v16);
4602       uzp2(v21, T2D, v20, v16);
4603       eor(v28, T16B, v17, v21);
4604 
4605       pmull(v22, T8H, v1, v5, T8B);
4606       pmull(v20, T8H, v1, v7, T8B);
4607       pmull(v23, T8H, v1, v4, T8B);
4608       pmull(v21, T8H, v1, v6, T8B);
4609 
4610       pmull2(v18, T8H, v1, v5, T16B);
4611       pmull2(v16, T8H, v1, v7, T16B);
4612       pmull2(v19, T8H, v1, v4, T16B);
4613       pmull2(v17, T8H, v1, v6, T16B);
4614 
4615       ld1(v0, v1, T2D, post(buf, 32));
4616 
4617       uzp1(v24, T8H, v20, v22);
4618       uzp2(v25, T8H, v20, v22);
4619       eor(v20, T16B, v24, v25);
4620 
4621       uzp1(v26, T8H, v16, v18);
4622       uzp2(v27, T8H, v16, v18);
4623       eor(v16, T16B, v26, v27);
4624 
4625       ushll2(v22, T4S, v20, T8H, 8);
4626       ushll(v20, T4S, v20, T4H, 8);
4627 
4628       ushll2(v18, T4S, v16, T8H, 8);
4629       ushll(v16, T4S, v16, T4H, 8);
4630 
4631       eor(v22, T16B, v23, v22);
4632       eor(v18, T16B, v19, v18);
4633       eor(v20, T16B, v21, v20);
4634       eor(v16, T16B, v17, v16);
4635 
4636       uzp1(v17, T2D, v16, v20);
4637       uzp2(v21, T2D, v16, v20);
4638       eor(v16, T16B, v17, v21);
4639 
4640       ushll2(v20, T2D, v16, T4S, 16);
4641       ushll(v16, T2D, v16, T2S, 16);
4642 
4643       eor(v20, T16B, v22, v20);
4644       eor(v16, T16B, v16, v18);
4645 
4646       uzp1(v17, T2D, v20, v16);
4647       uzp2(v21, T2D, v20, v16);
4648       eor(v20, T16B, v17, v21);
4649 
4650       shl(v16, T2D, v28, 1);
4651       shl(v17, T2D, v20, 1);
4652 
4653       eor(v0, T16B, v0, v16);
4654       eor(v1, T16B, v1, v17);
4655 
4656       subs(len, len, 32);
4657       br(Assembler::GE, L_fold);
4658 
4659       mov(crc, 0);
4660       mov(tmp, v0, D, 0);
4661       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4662       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4663       mov(tmp, v0, D, 1);
4664       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4665       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4666       mov(tmp, v1, D, 0);
4667       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4668       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4669       mov(tmp, v1, D, 1);
4670       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4671       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4672 
4673       add(len, len, 32);
4674     } // Neon code end
4675 
4676   BIND(L_by16);
4677     subs(len, len, 16);
4678     br(Assembler::GE, L_by16_loop);
4679     adds(len, len, 16-4);
4680     br(Assembler::GE, L_by4_loop);
4681     adds(len, len, 4);
4682     br(Assembler::GT, L_by1_loop);
4683     b(L_exit);
4684 
4685   BIND(L_by4_loop);
4686     ldrw(tmp, Address(post(buf, 4)));
4687     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4688     subs(len, len, 4);
4689     br(Assembler::GE, L_by4_loop);
4690     adds(len, len, 4);
4691     br(Assembler::LE, L_exit);
4692   BIND(L_by1_loop);
4693     subs(len, len, 1);
4694     ldrb(tmp, Address(post(buf, 1)));
4695     update_byte_crc32(crc, tmp, table0);
4696     br(Assembler::GT, L_by1_loop);
4697     b(L_exit);
4698 
4699     align(CodeEntryAlignment);
4700   BIND(L_by16_loop);
4701     subs(len, len, 16);
4702     ldp(tmp, tmp3, Address(post(buf, 16)));
4703     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4704     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4705     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4706     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4707     br(Assembler::GE, L_by16_loop);
4708     adds(len, len, 16-4);
4709     br(Assembler::GE, L_by4_loop);
4710     adds(len, len, 4);
4711     br(Assembler::GT, L_by1_loop);
4712   BIND(L_exit);
4713     mvnw(crc, crc);
4714 }
4715 
4716 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4717         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4718     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4719     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4720 
4721     subs(tmp0, len, 384);
4722     br(Assembler::GE, CRC_by128_pre);
4723   BIND(CRC_less128);
4724     subs(len, len, 32);
4725     br(Assembler::GE, CRC_by32_loop);
4726   BIND(CRC_less32);
4727     adds(len, len, 32 - 4);
4728     br(Assembler::GE, CRC_by4_loop);
4729     adds(len, len, 4);
4730     br(Assembler::GT, CRC_by1_loop);
4731     b(L_exit);
4732 
4733   BIND(CRC_by32_loop);
4734     ldp(tmp0, tmp1, Address(buf));
4735     crc32cx(crc, crc, tmp0);
4736     ldr(tmp2, Address(buf, 16));
4737     crc32cx(crc, crc, tmp1);
4738     ldr(tmp3, Address(buf, 24));
4739     crc32cx(crc, crc, tmp2);
4740     add(buf, buf, 32);
4741     subs(len, len, 32);
4742     crc32cx(crc, crc, tmp3);
4743     br(Assembler::GE, CRC_by32_loop);
4744     cmn(len, (u1)32);
4745     br(Assembler::NE, CRC_less32);
4746     b(L_exit);
4747 
4748   BIND(CRC_by4_loop);
4749     ldrw(tmp0, Address(post(buf, 4)));
4750     subs(len, len, 4);
4751     crc32cw(crc, crc, tmp0);
4752     br(Assembler::GE, CRC_by4_loop);
4753     adds(len, len, 4);
4754     br(Assembler::LE, L_exit);
4755   BIND(CRC_by1_loop);
4756     ldrb(tmp0, Address(post(buf, 1)));
4757     subs(len, len, 1);
4758     crc32cb(crc, crc, tmp0);
4759     br(Assembler::GT, CRC_by1_loop);
4760     b(L_exit);
4761 
4762   BIND(CRC_by128_pre);
4763     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4764       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4765     mov(crc, 0);
4766     crc32cx(crc, crc, tmp0);
4767     crc32cx(crc, crc, tmp1);
4768 
4769     cbnz(len, CRC_less128);
4770 
4771   BIND(L_exit);
4772 }
4773 
4774 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4775         Register len, Register tmp0, Register tmp1, Register tmp2,
4776         Register tmp3) {
4777     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4778     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4779 
4780     subs(len, len, 128);
4781     br(Assembler::GE, CRC_by64_pre);
4782   BIND(CRC_less64);
4783     adds(len, len, 128-32);
4784     br(Assembler::GE, CRC_by32_loop);
4785   BIND(CRC_less32);
4786     adds(len, len, 32-4);
4787     br(Assembler::GE, CRC_by4_loop);
4788     adds(len, len, 4);
4789     br(Assembler::GT, CRC_by1_loop);
4790     b(L_exit);
4791 
4792   BIND(CRC_by32_loop);
4793     ldp(tmp0, tmp1, Address(post(buf, 16)));
4794     subs(len, len, 32);
4795     crc32cx(crc, crc, tmp0);
4796     ldr(tmp2, Address(post(buf, 8)));
4797     crc32cx(crc, crc, tmp1);
4798     ldr(tmp3, Address(post(buf, 8)));
4799     crc32cx(crc, crc, tmp2);
4800     crc32cx(crc, crc, tmp3);
4801     br(Assembler::GE, CRC_by32_loop);
4802     cmn(len, (u1)32);
4803     br(Assembler::NE, CRC_less32);
4804     b(L_exit);
4805 
4806   BIND(CRC_by4_loop);
4807     ldrw(tmp0, Address(post(buf, 4)));
4808     subs(len, len, 4);
4809     crc32cw(crc, crc, tmp0);
4810     br(Assembler::GE, CRC_by4_loop);
4811     adds(len, len, 4);
4812     br(Assembler::LE, L_exit);
4813   BIND(CRC_by1_loop);
4814     ldrb(tmp0, Address(post(buf, 1)));
4815     subs(len, len, 1);
4816     crc32cb(crc, crc, tmp0);
4817     br(Assembler::GT, CRC_by1_loop);
4818     b(L_exit);
4819 
4820   BIND(CRC_by64_pre);
4821     sub(buf, buf, 8);
4822     ldp(tmp0, tmp1, Address(buf, 8));
4823     crc32cx(crc, crc, tmp0);
4824     ldr(tmp2, Address(buf, 24));
4825     crc32cx(crc, crc, tmp1);
4826     ldr(tmp3, Address(buf, 32));
4827     crc32cx(crc, crc, tmp2);
4828     ldr(tmp0, Address(buf, 40));
4829     crc32cx(crc, crc, tmp3);
4830     ldr(tmp1, Address(buf, 48));
4831     crc32cx(crc, crc, tmp0);
4832     ldr(tmp2, Address(buf, 56));
4833     crc32cx(crc, crc, tmp1);
4834     ldr(tmp3, Address(pre(buf, 64)));
4835 
4836     b(CRC_by64_loop);
4837 
4838     align(CodeEntryAlignment);
4839   BIND(CRC_by64_loop);
4840     subs(len, len, 64);
4841     crc32cx(crc, crc, tmp2);
4842     ldr(tmp0, Address(buf, 8));
4843     crc32cx(crc, crc, tmp3);
4844     ldr(tmp1, Address(buf, 16));
4845     crc32cx(crc, crc, tmp0);
4846     ldr(tmp2, Address(buf, 24));
4847     crc32cx(crc, crc, tmp1);
4848     ldr(tmp3, Address(buf, 32));
4849     crc32cx(crc, crc, tmp2);
4850     ldr(tmp0, Address(buf, 40));
4851     crc32cx(crc, crc, tmp3);
4852     ldr(tmp1, Address(buf, 48));
4853     crc32cx(crc, crc, tmp0);
4854     ldr(tmp2, Address(buf, 56));
4855     crc32cx(crc, crc, tmp1);
4856     ldr(tmp3, Address(pre(buf, 64)));
4857     br(Assembler::GE, CRC_by64_loop);
4858 
4859     // post-loop
4860     crc32cx(crc, crc, tmp2);
4861     crc32cx(crc, crc, tmp3);
4862 
4863     sub(len, len, 64);
4864     add(buf, buf, 8);
4865     cmn(len, (u1)128);
4866     br(Assembler::NE, CRC_less64);
4867   BIND(L_exit);
4868 }
4869 
4870 /**
4871  * @param crc   register containing existing CRC (32-bit)
4872  * @param buf   register pointing to input byte buffer (byte*)
4873  * @param len   register containing number of bytes
4874  * @param table register that will contain address of CRC table
4875  * @param tmp   scratch register
4876  */
4877 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4878         Register table0, Register table1, Register table2, Register table3,
4879         Register tmp, Register tmp2, Register tmp3) {
4880   if (UseCryptoPmullForCRC32) {
4881     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4882   } else {
4883     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4884   }
4885 }
4886 
4887 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4888         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4889     Label CRC_by128_loop;
4890     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4891 
4892     sub(len, len, 256);
4893     Register table = tmp0;
4894     {
4895       uint64_t offset;
4896       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4897       add(table, table, offset);
4898     }
4899     add(table, table, table_offset);
4900 
4901     // Registers v0..v7 are used as data registers.
4902     // Registers v16..v31 are used as tmp registers.
4903     sub(buf, buf, 0x10);
4904     ldrq(v0, Address(buf, 0x10));
4905     ldrq(v1, Address(buf, 0x20));
4906     ldrq(v2, Address(buf, 0x30));
4907     ldrq(v3, Address(buf, 0x40));
4908     ldrq(v4, Address(buf, 0x50));
4909     ldrq(v5, Address(buf, 0x60));
4910     ldrq(v6, Address(buf, 0x70));
4911     ldrq(v7, Address(pre(buf, 0x80)));
4912 
4913     movi(v31, T4S, 0);
4914     mov(v31, S, 0, crc);
4915     eor(v0, T16B, v0, v31);
4916 
4917     // Register v16 contains constants from the crc table.
4918     ldrq(v16, Address(table));
4919     b(CRC_by128_loop);
4920 
4921     align(OptoLoopAlignment);
4922   BIND(CRC_by128_loop);
4923     pmull (v17,  T1Q, v0, v16, T1D);
4924     pmull2(v18, T1Q, v0, v16, T2D);
4925     ldrq(v0, Address(buf, 0x10));
4926     eor3(v0, T16B, v17,  v18, v0);
4927 
4928     pmull (v19, T1Q, v1, v16, T1D);
4929     pmull2(v20, T1Q, v1, v16, T2D);
4930     ldrq(v1, Address(buf, 0x20));
4931     eor3(v1, T16B, v19, v20, v1);
4932 
4933     pmull (v21, T1Q, v2, v16, T1D);
4934     pmull2(v22, T1Q, v2, v16, T2D);
4935     ldrq(v2, Address(buf, 0x30));
4936     eor3(v2, T16B, v21, v22, v2);
4937 
4938     pmull (v23, T1Q, v3, v16, T1D);
4939     pmull2(v24, T1Q, v3, v16, T2D);
4940     ldrq(v3, Address(buf, 0x40));
4941     eor3(v3, T16B, v23, v24, v3);
4942 
4943     pmull (v25, T1Q, v4, v16, T1D);
4944     pmull2(v26, T1Q, v4, v16, T2D);
4945     ldrq(v4, Address(buf, 0x50));
4946     eor3(v4, T16B, v25, v26, v4);
4947 
4948     pmull (v27, T1Q, v5, v16, T1D);
4949     pmull2(v28, T1Q, v5, v16, T2D);
4950     ldrq(v5, Address(buf, 0x60));
4951     eor3(v5, T16B, v27, v28, v5);
4952 
4953     pmull (v29, T1Q, v6, v16, T1D);
4954     pmull2(v30, T1Q, v6, v16, T2D);
4955     ldrq(v6, Address(buf, 0x70));
4956     eor3(v6, T16B, v29, v30, v6);
4957 
4958     // Reuse registers v23, v24.
4959     // Using them won't block the first instruction of the next iteration.
4960     pmull (v23, T1Q, v7, v16, T1D);
4961     pmull2(v24, T1Q, v7, v16, T2D);
4962     ldrq(v7, Address(pre(buf, 0x80)));
4963     eor3(v7, T16B, v23, v24, v7);
4964 
4965     subs(len, len, 0x80);
4966     br(Assembler::GE, CRC_by128_loop);
4967 
4968     // fold into 512 bits
4969     // Use v31 for constants because v16 can be still in use.
4970     ldrq(v31, Address(table, 0x10));
4971 
4972     pmull (v17,  T1Q, v0, v31, T1D);
4973     pmull2(v18, T1Q, v0, v31, T2D);
4974     eor3(v0, T16B, v17, v18, v4);
4975 
4976     pmull (v19, T1Q, v1, v31, T1D);
4977     pmull2(v20, T1Q, v1, v31, T2D);
4978     eor3(v1, T16B, v19, v20, v5);
4979 
4980     pmull (v21, T1Q, v2, v31, T1D);
4981     pmull2(v22, T1Q, v2, v31, T2D);
4982     eor3(v2, T16B, v21, v22, v6);
4983 
4984     pmull (v23, T1Q, v3, v31, T1D);
4985     pmull2(v24, T1Q, v3, v31, T2D);
4986     eor3(v3, T16B, v23, v24, v7);
4987 
4988     // fold into 128 bits
4989     // Use v17 for constants because v31 can be still in use.
4990     ldrq(v17, Address(table, 0x20));
4991     pmull (v25, T1Q, v0, v17, T1D);
4992     pmull2(v26, T1Q, v0, v17, T2D);
4993     eor3(v3, T16B, v3, v25, v26);
4994 
4995     // Use v18 for constants because v17 can be still in use.
4996     ldrq(v18, Address(table, 0x30));
4997     pmull (v27, T1Q, v1, v18, T1D);
4998     pmull2(v28, T1Q, v1, v18, T2D);
4999     eor3(v3, T16B, v3, v27, v28);
5000 
5001     // Use v19 for constants because v18 can be still in use.
5002     ldrq(v19, Address(table, 0x40));
5003     pmull (v29, T1Q, v2, v19, T1D);
5004     pmull2(v30, T1Q, v2, v19, T2D);
5005     eor3(v0, T16B, v3, v29, v30);
5006 
5007     add(len, len, 0x80);
5008     add(buf, buf, 0x10);
5009 
5010     mov(tmp0, v0, D, 0);
5011     mov(tmp1, v0, D, 1);
5012 }
5013 
5014 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5015   Address adr;
5016   switch(dst.getMode()) {
5017   case Address::base_plus_offset:
5018     // This is the expected mode, although we allow all the other
5019     // forms below.
5020     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5021     break;
5022   default:
5023     lea(rscratch2, dst);
5024     adr = Address(rscratch2);
5025     break;
5026   }
5027   ldr(rscratch1, adr);
5028   add(rscratch1, rscratch1, src);
5029   str(rscratch1, adr);
5030 }
5031 
5032 void MacroAssembler::cmpptr(Register src1, Address src2) {
5033   uint64_t offset;
5034   adrp(rscratch1, src2, offset);
5035   ldr(rscratch1, Address(rscratch1, offset));
5036   cmp(src1, rscratch1);
5037 }
5038 
5039 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5040   cmp(obj1, obj2);
5041 }
5042 
5043 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5044   load_method_holder(rresult, rmethod);
5045   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5046 }
5047 
5048 void MacroAssembler::load_method_holder(Register holder, Register method) {
5049   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5050   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5051   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5052 }
5053 
5054 // Loads the obj's Klass* into dst.
5055 // Preserves all registers (incl src, rscratch1 and rscratch2).
5056 // Input:
5057 // src - the oop we want to load the klass from.
5058 // dst - output narrow klass.
5059 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5060   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5061   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5062   lsr(dst, dst, markWord::klass_shift);
5063 }
5064 
5065 void MacroAssembler::load_klass(Register dst, Register src) {
5066   if (UseCompactObjectHeaders) {
5067     load_narrow_klass_compact(dst, src);
5068   } else {
5069     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5070   }
5071   decode_klass_not_null(dst);
5072 }
5073 
5074 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5075   if (RestoreMXCSROnJNICalls) {
5076     Label OK;
5077     get_fpcr(tmp1);
5078     mov(tmp2, tmp1);
5079     // Set FPCR to the state we need. We do want Round to Nearest. We
5080     // don't want non-IEEE rounding modes or floating-point traps.
5081     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5082     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5083     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5084     eor(tmp2, tmp1, tmp2);
5085     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5086     set_fpcr(tmp1);
5087     bind(OK);
5088   }
5089 }
5090 
5091 // ((OopHandle)result).resolve();
5092 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5093   // OopHandle::resolve is an indirection.
5094   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5095 }
5096 
5097 // ((WeakHandle)result).resolve();
5098 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5099   assert_different_registers(result, tmp1, tmp2);
5100   Label resolved;
5101 
5102   // A null weak handle resolves to null.
5103   cbz(result, resolved);
5104 
5105   // Only 64 bit platforms support GCs that require a tmp register
5106   // WeakHandle::resolve is an indirection like jweak.
5107   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5108                  result, Address(result), tmp1, tmp2);
5109   bind(resolved);
5110 }
5111 
5112 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5113   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5114   ldr(dst, Address(rmethod, Method::const_offset()));
5115   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5116   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5117   ldr(dst, Address(dst, mirror_offset));
5118   resolve_oop_handle(dst, tmp1, tmp2);
5119 }
5120 
5121 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5122   assert_different_registers(obj, klass, tmp);
5123   if (UseCompactObjectHeaders) {
5124     load_narrow_klass_compact(tmp, obj);
5125   } else {
5126     ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5127   }
5128   if (CompressedKlassPointers::base() == nullptr) {
5129     cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5130     return;
5131   } else if (!AOTCodeCache::is_on_for_dump() &&
5132              ((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5133              && CompressedKlassPointers::shift() == 0) {
5134     // Only the bottom 32 bits matter
5135     cmpw(klass, tmp);
5136     return;
5137   }
5138   decode_klass_not_null(tmp);
5139   cmp(klass, tmp);
5140 }
5141 
5142 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5143   if (UseCompactObjectHeaders) {
5144     load_narrow_klass_compact(tmp1, obj1);
5145     load_narrow_klass_compact(tmp2,  obj2);
5146   } else {
5147     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5148     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5149   }
5150   cmpw(tmp1, tmp2);
5151 }
5152 
5153 void MacroAssembler::store_klass(Register dst, Register src) {
5154   // FIXME: Should this be a store release?  concurrent gcs assumes
5155   // klass length is valid if klass field is not null.
5156   assert(!UseCompactObjectHeaders, "not with compact headers");
5157   encode_klass_not_null(src);
5158   strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5159 }
5160 
5161 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5162   assert(!UseCompactObjectHeaders, "not with compact headers");
5163   // Store to klass gap in destination
5164   strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5165 }
5166 
5167 // Algorithm must match CompressedOops::encode.
5168 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5169 #ifdef ASSERT
5170   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5171 #endif
5172   verify_oop_msg(s, "broken oop in encode_heap_oop");
5173   if (CompressedOops::base() == nullptr) {
5174     if (CompressedOops::shift() != 0) {
5175       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5176       lsr(d, s, LogMinObjAlignmentInBytes);
5177     } else {
5178       mov(d, s);
5179     }
5180   } else {
5181     subs(d, s, rheapbase);
5182     csel(d, d, zr, Assembler::HS);
5183     lsr(d, d, LogMinObjAlignmentInBytes);
5184 
5185     /*  Old algorithm: is this any worse?
5186     Label nonnull;
5187     cbnz(r, nonnull);
5188     sub(r, r, rheapbase);
5189     bind(nonnull);
5190     lsr(r, r, LogMinObjAlignmentInBytes);
5191     */
5192   }
5193 }
5194 
5195 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5196 #ifdef ASSERT
5197   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5198   if (CheckCompressedOops) {
5199     Label ok;
5200     cbnz(r, ok);
5201     stop("null oop passed to encode_heap_oop_not_null");
5202     bind(ok);
5203   }
5204 #endif
5205   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5206   if (CompressedOops::base() != nullptr) {
5207     sub(r, r, rheapbase);
5208   }
5209   if (CompressedOops::shift() != 0) {
5210     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5211     lsr(r, r, LogMinObjAlignmentInBytes);
5212   }
5213 }
5214 
5215 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5216 #ifdef ASSERT
5217   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5218   if (CheckCompressedOops) {
5219     Label ok;
5220     cbnz(src, ok);
5221     stop("null oop passed to encode_heap_oop_not_null2");
5222     bind(ok);
5223   }
5224 #endif
5225   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5226 
5227   Register data = src;
5228   if (CompressedOops::base() != nullptr) {
5229     sub(dst, src, rheapbase);
5230     data = dst;
5231   }
5232   if (CompressedOops::shift() != 0) {
5233     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5234     lsr(dst, data, LogMinObjAlignmentInBytes);
5235     data = dst;
5236   }
5237   if (data == src)
5238     mov(dst, src);
5239 }
5240 
5241 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5242 #ifdef ASSERT
5243   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5244 #endif
5245   if (CompressedOops::base() == nullptr) {
5246     if (CompressedOops::shift() != 0) {
5247       lsl(d, s, CompressedOops::shift());
5248     } else if (d != s) {
5249       mov(d, s);
5250     }
5251   } else {
5252     Label done;
5253     if (d != s)
5254       mov(d, s);
5255     cbz(s, done);
5256     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5257     bind(done);
5258   }
5259   verify_oop_msg(d, "broken oop in decode_heap_oop");
5260 }
5261 
5262 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5263   assert (UseCompressedOops, "should only be used for compressed headers");
5264   assert (Universe::heap() != nullptr, "java heap should be initialized");
5265   // Cannot assert, unverified entry point counts instructions (see .ad file)
5266   // vtableStubs also counts instructions in pd_code_size_limit.
5267   // Also do not verify_oop as this is called by verify_oop.
5268   if (CompressedOops::shift() != 0) {
5269     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5270     if (CompressedOops::base() != nullptr) {
5271       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5272     } else {
5273       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5274     }
5275   } else {
5276     assert (CompressedOops::base() == nullptr, "sanity");
5277   }
5278 }
5279 
5280 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5281   assert (UseCompressedOops, "should only be used for compressed headers");
5282   assert (Universe::heap() != nullptr, "java heap should be initialized");
5283   // Cannot assert, unverified entry point counts instructions (see .ad file)
5284   // vtableStubs also counts instructions in pd_code_size_limit.
5285   // Also do not verify_oop as this is called by verify_oop.
5286   if (CompressedOops::shift() != 0) {
5287     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5288     if (CompressedOops::base() != nullptr) {
5289       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5290     } else {
5291       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5292     }
5293   } else {
5294     assert (CompressedOops::base() == nullptr, "sanity");
5295     if (dst != src) {
5296       mov(dst, src);
5297     }
5298   }
5299 }
5300 
5301 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5302 
5303 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5304   assert(Metaspace::initialized(), "metaspace not initialized yet");
5305   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5306   return _klass_decode_mode;
5307 }
5308 
5309 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5310   // KlassDecodeMode shouldn't be set already.
5311   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5312 
5313   if (base == nullptr) {
5314     return KlassDecodeZero;
5315   }
5316 
5317   if (operand_valid_for_logical_immediate(
5318         /*is32*/false, (uint64_t)base)) {
5319     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5320     if (((uint64_t)base & range_mask) == 0) {
5321       return KlassDecodeXor;
5322     }
5323   }
5324 
5325   const uint64_t shifted_base =
5326     (uint64_t)base >> shift;
5327   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5328     return KlassDecodeMovk;
5329   }
5330 
5331   // No valid encoding.
5332   return KlassDecodeNone;
5333 }
5334 
5335 // Check if one of the above decoding modes will work for given base, shift and range.
5336 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5337   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5338 }
5339 
5340 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5341   _klass_decode_mode = klass_decode_mode(base, shift, range);
5342   return _klass_decode_mode != KlassDecodeNone;
5343 }
5344 
5345 static Register pick_different_tmp(Register dst, Register src) {
5346   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5347   return *tmps.begin();
5348 }
5349 
5350 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5351   // we have to load the klass base from the AOT constants area but
5352   // not the shift because it is not allowed to change
5353   int shift = CompressedKlassPointers::shift();
5354   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5355   if (dst != src) {
5356     // we can load the base into dst, subtract it formthe src and shift down
5357     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5358     ldr(dst, dst);
5359     sub(dst, src, dst);
5360     lsr(dst, dst, shift);
5361   } else {
5362     // we need an extra register in order to load the coop base
5363     Register tmp = pick_different_tmp(dst, src);
5364     RegSet regs = RegSet::of(tmp);
5365     push(regs, sp);
5366     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5367     ldr(tmp, tmp);
5368     sub(dst, src, tmp);
5369     lsr(dst, dst, shift);
5370     pop(regs, sp);
5371   }
5372 }
5373 
5374 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5375   if (CompressedKlassPointers::base() != nullptr && AOTCodeCache::is_on_for_dump()) {
5376     encode_klass_not_null_for_aot(dst, src);
5377     return;
5378   }
5379 
5380   switch (klass_decode_mode()) {
5381   case KlassDecodeZero:
5382     if (CompressedKlassPointers::shift() != 0) {
5383       lsr(dst, src, CompressedKlassPointers::shift());
5384     } else {
5385       if (dst != src) mov(dst, src);
5386     }
5387     break;
5388 
5389   case KlassDecodeXor:
5390     if (CompressedKlassPointers::shift() != 0) {
5391       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5392       lsr(dst, dst, CompressedKlassPointers::shift());
5393     } else {
5394       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5395     }
5396     break;
5397 
5398   case KlassDecodeMovk:
5399     if (CompressedKlassPointers::shift() != 0) {
5400       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5401     } else {
5402       movw(dst, src);
5403     }
5404     break;
5405 
5406   case KlassDecodeNone:
5407     ShouldNotReachHere();
5408     break;
5409   }
5410 }
5411 
5412 void MacroAssembler::encode_klass_not_null(Register r) {
5413   encode_klass_not_null(r, r);
5414 }
5415 
5416 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5417   // we have to load the klass base from the AOT constants area but
5418   // not the shift because it is not allowed to change
5419   int shift = CompressedKlassPointers::shift();
5420   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5421   if (dst != src) {
5422     // we can load the base into dst then add the offset with a suitable shift
5423     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5424     ldr(dst, dst);
5425     add(dst, dst, src, LSL,  shift);
5426   } else {
5427     // we need an extra register in order to load the coop base
5428     Register tmp = pick_different_tmp(dst, src);
5429     RegSet regs = RegSet::of(tmp);
5430     push(regs, sp);
5431     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5432     ldr(tmp, tmp);
5433     add(dst, tmp,  src, LSL,  shift);
5434     pop(regs, sp);
5435   }
5436 }
5437 
5438 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5439   if (AOTCodeCache::is_on_for_dump()) {
5440     decode_klass_not_null_for_aot(dst, src);
5441     return;
5442   }
5443 
5444   switch (klass_decode_mode()) {
5445   case KlassDecodeZero:
5446     if (CompressedKlassPointers::shift() != 0) {
5447       lsl(dst, src, CompressedKlassPointers::shift());
5448     } else {
5449       if (dst != src) mov(dst, src);
5450     }
5451     break;
5452 
5453   case KlassDecodeXor:
5454     if (CompressedKlassPointers::shift() != 0) {
5455       lsl(dst, src, CompressedKlassPointers::shift());
5456       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5457     } else {
5458       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5459     }
5460     break;
5461 
5462   case KlassDecodeMovk: {
5463     const uint64_t shifted_base =
5464       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5465 
5466     if (dst != src) movw(dst, src);
5467     movk(dst, shifted_base >> 32, 32);
5468 
5469     if (CompressedKlassPointers::shift() != 0) {
5470       lsl(dst, dst, CompressedKlassPointers::shift());
5471     }
5472 
5473     break;
5474   }
5475 
5476   case KlassDecodeNone:
5477     ShouldNotReachHere();
5478     break;
5479   }
5480 }
5481 
5482 void  MacroAssembler::decode_klass_not_null(Register r) {
5483   decode_klass_not_null(r, r);
5484 }
5485 
5486 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5487 #ifdef ASSERT
5488   {
5489     ThreadInVMfromUnknown tiv;
5490     assert (UseCompressedOops, "should only be used for compressed oops");
5491     assert (Universe::heap() != nullptr, "java heap should be initialized");
5492     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5493     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5494   }
5495 #endif
5496   int oop_index = oop_recorder()->find_index(obj);
5497   InstructionMark im(this);
5498   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5499   code_section()->relocate(inst_mark(), rspec);
5500   movz(dst, 0xDEAD, 16);
5501   movk(dst, 0xBEEF);
5502 }
5503 
5504 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5505   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5506   int index = oop_recorder()->find_index(k);
5507 
5508   InstructionMark im(this);
5509   RelocationHolder rspec = metadata_Relocation::spec(index);
5510   code_section()->relocate(inst_mark(), rspec);
5511   narrowKlass nk = CompressedKlassPointers::encode(k);
5512   movz(dst, (nk >> 16), 16);
5513   movk(dst, nk & 0xffff);
5514 }
5515 
5516 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5517                                     Register dst, Address src,
5518                                     Register tmp1, Register tmp2) {
5519   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5520   decorators = AccessInternal::decorator_fixup(decorators, type);
5521   bool as_raw = (decorators & AS_RAW) != 0;
5522   if (as_raw) {
5523     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5524   } else {
5525     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5526   }
5527 }
5528 
5529 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5530                                      Address dst, Register val,
5531                                      Register tmp1, Register tmp2, Register tmp3) {
5532   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5533   decorators = AccessInternal::decorator_fixup(decorators, type);
5534   bool as_raw = (decorators & AS_RAW) != 0;
5535   if (as_raw) {
5536     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5537   } else {
5538     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5539   }
5540 }
5541 
5542 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5543                                    Register tmp2, DecoratorSet decorators) {
5544   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5545 }
5546 
5547 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5548                                             Register tmp2, DecoratorSet decorators) {
5549   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5550 }
5551 
5552 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5553                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5554   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5555 }
5556 
5557 // Used for storing nulls.
5558 void MacroAssembler::store_heap_oop_null(Address dst) {
5559   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5560 }
5561 
5562 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5563   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5564   int index = oop_recorder()->allocate_metadata_index(obj);
5565   RelocationHolder rspec = metadata_Relocation::spec(index);
5566   return Address((address)obj, rspec);
5567 }
5568 
5569 // Move an oop into a register.
5570 void MacroAssembler::movoop(Register dst, jobject obj) {
5571   int oop_index;
5572   if (obj == nullptr) {
5573     oop_index = oop_recorder()->allocate_oop_index(obj);
5574   } else {
5575 #ifdef ASSERT
5576     {
5577       ThreadInVMfromUnknown tiv;
5578       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5579     }
5580 #endif
5581     oop_index = oop_recorder()->find_index(obj);
5582   }
5583   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5584 
5585   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5586     mov(dst, Address((address)obj, rspec));
5587   } else {
5588     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5589     ldr(dst, Address(dummy, rspec));
5590   }
5591 }
5592 
5593 // Move a metadata address into a register.
5594 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5595   int oop_index;
5596   if (obj == nullptr) {
5597     oop_index = oop_recorder()->allocate_metadata_index(obj);
5598   } else {
5599     oop_index = oop_recorder()->find_index(obj);
5600   }
5601   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5602   mov(dst, Address((address)obj, rspec));
5603 }
5604 
5605 Address MacroAssembler::constant_oop_address(jobject obj) {
5606 #ifdef ASSERT
5607   {
5608     ThreadInVMfromUnknown tiv;
5609     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5610     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5611   }
5612 #endif
5613   int oop_index = oop_recorder()->find_index(obj);
5614   return Address((address)obj, oop_Relocation::spec(oop_index));
5615 }
5616 
5617 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5618 void MacroAssembler::tlab_allocate(Register obj,
5619                                    Register var_size_in_bytes,
5620                                    int con_size_in_bytes,
5621                                    Register t1,
5622                                    Register t2,
5623                                    Label& slow_case) {
5624   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5625   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5626 }
5627 
5628 void MacroAssembler::verify_tlab() {
5629 #ifdef ASSERT
5630   if (UseTLAB && VerifyOops) {
5631     Label next, ok;
5632 
5633     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5634 
5635     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5636     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5637     cmp(rscratch2, rscratch1);
5638     br(Assembler::HS, next);
5639     STOP("assert(top >= start)");
5640     should_not_reach_here();
5641 
5642     bind(next);
5643     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5644     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5645     cmp(rscratch2, rscratch1);
5646     br(Assembler::HS, ok);
5647     STOP("assert(top <= end)");
5648     should_not_reach_here();
5649 
5650     bind(ok);
5651     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5652   }
5653 #endif
5654 }
5655 
5656 // Writes to stack successive pages until offset reached to check for
5657 // stack overflow + shadow pages.  This clobbers tmp.
5658 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5659   assert_different_registers(tmp, size, rscratch1);
5660   mov(tmp, sp);
5661   // Bang stack for total size given plus shadow page size.
5662   // Bang one page at a time because large size can bang beyond yellow and
5663   // red zones.
5664   Label loop;
5665   mov(rscratch1, (int)os::vm_page_size());
5666   bind(loop);
5667   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5668   subsw(size, size, rscratch1);
5669   str(size, Address(tmp));
5670   br(Assembler::GT, loop);
5671 
5672   // Bang down shadow pages too.
5673   // At this point, (tmp-0) is the last address touched, so don't
5674   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5675   // was post-decremented.)  Skip this address by starting at i=1, and
5676   // touch a few more pages below.  N.B.  It is important to touch all
5677   // the way down to and including i=StackShadowPages.
5678   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5679     // this could be any sized move but this is can be a debugging crumb
5680     // so the bigger the better.
5681     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5682     str(size, Address(tmp));
5683   }
5684 }
5685 
5686 // Move the address of the polling page into dest.
5687 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5688   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5689 }
5690 
5691 // Read the polling page.  The address of the polling page must
5692 // already be in r.
5693 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5694   address mark;
5695   {
5696     InstructionMark im(this);
5697     code_section()->relocate(inst_mark(), rtype);
5698     ldrw(zr, Address(r, 0));
5699     mark = inst_mark();
5700   }
5701   verify_cross_modify_fence_not_required();
5702   return mark;
5703 }
5704 
5705 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
5706   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
5707   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
5708   uint64_t dest_page = (uint64_t)dest.target() >> 12;
5709   int64_t offset_low = dest_page - low_page;
5710   int64_t offset_high = dest_page - high_page;
5711 
5712   assert(is_valid_AArch64_address(dest.target()), "bad address");
5713   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
5714 
5715   InstructionMark im(this);
5716   code_section()->relocate(inst_mark(), dest.rspec());
5717   // 8143067: Ensure that the adrp can reach the dest from anywhere within
5718   // the code cache so that if it is relocated we know it will still reach
5719   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
5720     _adrp(reg1, dest.target());
5721   } else {
5722     uint64_t target = (uint64_t)dest.target();
5723     uint64_t adrp_target
5724       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
5725 
5726     _adrp(reg1, (address)adrp_target);
5727     movk(reg1, target >> 32, 32);
5728   }
5729   byte_offset = (uint64_t)dest.target() & 0xfff;
5730 }
5731 
5732 void MacroAssembler::load_byte_map_base(Register reg) {
5733 #if INCLUDE_CDS
5734   if (AOTCodeCache::is_on_for_dump()) {
5735     address byte_map_base_adr = AOTRuntimeConstants::card_table_base_address();
5736     lea(reg, ExternalAddress(byte_map_base_adr));
5737     ldr(reg, Address(reg));
5738     return;
5739   }
5740 #endif
5741   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
5742 
5743   // Strictly speaking the card table base isn't an address at all, and it might
5744   // even be negative. It is thus materialised as a constant.
5745   mov(reg, (uint64_t)ctbs->card_table_base_const());
5746 }
5747 
5748 void MacroAssembler::load_aotrc_address(Register reg, address a) {
5749 #if INCLUDE_CDS
5750   assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
5751   if (AOTCodeCache::is_on_for_dump()) {
5752     // all aotrc field addresses should be registered in the AOTCodeCache address table
5753     lea(reg, ExternalAddress(a));
5754   } else {
5755     mov(reg, (uint64_t)a);
5756   }
5757 #else
5758   ShouldNotReachHere();
5759 #endif
5760 }
5761 
5762 void MacroAssembler::build_frame(int framesize) {
5763   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5764   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5765   protect_return_address();
5766   if (framesize < ((1 << 9) + 2 * wordSize)) {
5767     sub(sp, sp, framesize);
5768     stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5769     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
5770   } else {
5771     stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
5772     if (PreserveFramePointer) mov(rfp, sp);
5773     if (framesize < ((1 << 12) + 2 * wordSize))
5774       sub(sp, sp, framesize - 2 * wordSize);
5775     else {
5776       mov(rscratch1, framesize - 2 * wordSize);
5777       sub(sp, sp, rscratch1);
5778     }
5779   }
5780   verify_cross_modify_fence_not_required();
5781 }
5782 
5783 void MacroAssembler::remove_frame(int framesize) {
5784   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
5785   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
5786   if (framesize < ((1 << 9) + 2 * wordSize)) {
5787     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
5788     add(sp, sp, framesize);
5789   } else {
5790     if (framesize < ((1 << 12) + 2 * wordSize))
5791       add(sp, sp, framesize - 2 * wordSize);
5792     else {
5793       mov(rscratch1, framesize - 2 * wordSize);
5794       add(sp, sp, rscratch1);
5795     }
5796     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
5797   }
5798   authenticate_return_address();
5799 }
5800 
5801 
5802 // This method counts leading positive bytes (highest bit not set) in provided byte array
5803 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
5804     // Simple and most common case of aligned small array which is not at the
5805     // end of memory page is placed here. All other cases are in stub.
5806     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
5807     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
5808     assert_different_registers(ary1, len, result);
5809 
5810     mov(result, len);
5811     cmpw(len, 0);
5812     br(LE, DONE);
5813     cmpw(len, 4 * wordSize);
5814     br(GE, STUB_LONG); // size > 32 then go to stub
5815 
5816     int shift = 64 - exact_log2(os::vm_page_size());
5817     lsl(rscratch1, ary1, shift);
5818     mov(rscratch2, (size_t)(4 * wordSize) << shift);
5819     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
5820     br(CS, STUB); // at the end of page then go to stub
5821     subs(len, len, wordSize);
5822     br(LT, END);
5823 
5824   BIND(LOOP);
5825     ldr(rscratch1, Address(post(ary1, wordSize)));
5826     tst(rscratch1, UPPER_BIT_MASK);
5827     br(NE, SET_RESULT);
5828     subs(len, len, wordSize);
5829     br(GE, LOOP);
5830     cmpw(len, -wordSize);
5831     br(EQ, DONE);
5832 
5833   BIND(END);
5834     ldr(rscratch1, Address(ary1));
5835     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
5836     lslv(rscratch1, rscratch1, rscratch2);
5837     tst(rscratch1, UPPER_BIT_MASK);
5838     br(NE, SET_RESULT);
5839     b(DONE);
5840 
5841   BIND(STUB);
5842     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
5843     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
5844     address tpc1 = trampoline_call(count_pos);
5845     if (tpc1 == nullptr) {
5846       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
5847       postcond(pc() == badAddress);
5848       return nullptr;
5849     }
5850     b(DONE);
5851 
5852   BIND(STUB_LONG);
5853     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
5854     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
5855     address tpc2 = trampoline_call(count_pos_long);
5856     if (tpc2 == nullptr) {
5857       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
5858       postcond(pc() == badAddress);
5859       return nullptr;
5860     }
5861     b(DONE);
5862 
5863   BIND(SET_RESULT);
5864 
5865     add(len, len, wordSize);
5866     sub(result, result, len);
5867 
5868   BIND(DONE);
5869   postcond(pc() != badAddress);
5870   return pc();
5871 }
5872 
5873 // Clobbers: rscratch1, rscratch2, rflags
5874 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
5875 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
5876                                       Register tmp4, Register tmp5, Register result,
5877                                       Register cnt1, int elem_size) {
5878   Label DONE, SAME;
5879   Register tmp1 = rscratch1;
5880   Register tmp2 = rscratch2;
5881   int elem_per_word = wordSize/elem_size;
5882   int log_elem_size = exact_log2(elem_size);
5883   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
5884   int length_offset = arrayOopDesc::length_offset_in_bytes();
5885   int base_offset
5886     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
5887   // When the length offset is not aligned to 8 bytes,
5888   // then we align it down. This is valid because the new
5889   // offset will always be the klass which is the same
5890   // for type arrays.
5891   int start_offset = align_down(length_offset, BytesPerWord);
5892   int extra_length = base_offset - start_offset;
5893   assert(start_offset == length_offset || start_offset == klass_offset,
5894          "start offset must be 8-byte-aligned or be the klass offset");
5895   assert(base_offset != start_offset, "must include the length field");
5896   extra_length = extra_length / elem_size; // We count in elements, not bytes.
5897   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
5898 
5899   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
5900   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
5901 
5902 #ifndef PRODUCT
5903   {
5904     const char kind = (elem_size == 2) ? 'U' : 'L';
5905     char comment[64];
5906     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
5907     BLOCK_COMMENT(comment);
5908   }
5909 #endif
5910 
5911   // if (a1 == a2)
5912   //     return true;
5913   cmpoop(a1, a2); // May have read barriers for a1 and a2.
5914   br(EQ, SAME);
5915 
5916   if (UseSimpleArrayEquals) {
5917     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
5918     // if (a1 == nullptr || a2 == nullptr)
5919     //     return false;
5920     // a1 & a2 == 0 means (some-pointer is null) or
5921     // (very-rare-or-even-probably-impossible-pointer-values)
5922     // so, we can save one branch in most cases
5923     tst(a1, a2);
5924     mov(result, false);
5925     br(EQ, A_MIGHT_BE_NULL);
5926     // if (a1.length != a2.length)
5927     //      return false;
5928     bind(A_IS_NOT_NULL);
5929     ldrw(cnt1, Address(a1, length_offset));
5930     ldrw(tmp5, Address(a2, length_offset));
5931     cmp(cnt1, tmp5);
5932     br(NE, DONE); // If lengths differ, return false
5933     // Increase loop counter by diff between base- and actual start-offset.
5934     addw(cnt1, cnt1, extra_length);
5935     lea(a1, Address(a1, start_offset));
5936     lea(a2, Address(a2, start_offset));
5937     // Check for short strings, i.e. smaller than wordSize.
5938     subs(cnt1, cnt1, elem_per_word);
5939     br(Assembler::LT, SHORT);
5940     // Main 8 byte comparison loop.
5941     bind(NEXT_WORD); {
5942       ldr(tmp1, Address(post(a1, wordSize)));
5943       ldr(tmp2, Address(post(a2, wordSize)));
5944       subs(cnt1, cnt1, elem_per_word);
5945       eor(tmp5, tmp1, tmp2);
5946       cbnz(tmp5, DONE);
5947     } br(GT, NEXT_WORD);
5948     // Last longword.  In the case where length == 4 we compare the
5949     // same longword twice, but that's still faster than another
5950     // conditional branch.
5951     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
5952     // length == 4.
5953     if (log_elem_size > 0)
5954       lsl(cnt1, cnt1, log_elem_size);
5955     ldr(tmp3, Address(a1, cnt1));
5956     ldr(tmp4, Address(a2, cnt1));
5957     eor(tmp5, tmp3, tmp4);
5958     cbnz(tmp5, DONE);
5959     b(SAME);
5960     bind(A_MIGHT_BE_NULL);
5961     // in case both a1 and a2 are not-null, proceed with loads
5962     cbz(a1, DONE);
5963     cbz(a2, DONE);
5964     b(A_IS_NOT_NULL);
5965     bind(SHORT);
5966 
5967     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
5968     {
5969       ldrw(tmp1, Address(post(a1, 4)));
5970       ldrw(tmp2, Address(post(a2, 4)));
5971       eorw(tmp5, tmp1, tmp2);
5972       cbnzw(tmp5, DONE);
5973     }
5974     bind(TAIL03);
5975     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
5976     {
5977       ldrh(tmp3, Address(post(a1, 2)));
5978       ldrh(tmp4, Address(post(a2, 2)));
5979       eorw(tmp5, tmp3, tmp4);
5980       cbnzw(tmp5, DONE);
5981     }
5982     bind(TAIL01);
5983     if (elem_size == 1) { // Only needed when comparing byte arrays.
5984       tbz(cnt1, 0, SAME); // 0-1 bytes left.
5985       {
5986         ldrb(tmp1, a1);
5987         ldrb(tmp2, a2);
5988         eorw(tmp5, tmp1, tmp2);
5989         cbnzw(tmp5, DONE);
5990       }
5991     }
5992   } else {
5993     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
5994         CSET_EQ, LAST_CHECK;
5995     mov(result, false);
5996     cbz(a1, DONE);
5997     ldrw(cnt1, Address(a1, length_offset));
5998     cbz(a2, DONE);
5999     ldrw(tmp5, Address(a2, length_offset));
6000     cmp(cnt1, tmp5);
6001     br(NE, DONE); // If lengths differ, return false
6002     // Increase loop counter by diff between base- and actual start-offset.
6003     addw(cnt1, cnt1, extra_length);
6004 
6005     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6006     // faster to perform another branch before comparing a1 and a2
6007     cmp(cnt1, (u1)elem_per_word);
6008     br(LE, SHORT); // short or same
6009     ldr(tmp3, Address(pre(a1, start_offset)));
6010     subs(zr, cnt1, stubBytesThreshold);
6011     br(GE, STUB);
6012     ldr(tmp4, Address(pre(a2, start_offset)));
6013     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6014 
6015     // Main 16 byte comparison loop with 2 exits
6016     bind(NEXT_DWORD); {
6017       ldr(tmp1, Address(pre(a1, wordSize)));
6018       ldr(tmp2, Address(pre(a2, wordSize)));
6019       subs(cnt1, cnt1, 2 * elem_per_word);
6020       br(LE, TAIL);
6021       eor(tmp4, tmp3, tmp4);
6022       cbnz(tmp4, DONE);
6023       ldr(tmp3, Address(pre(a1, wordSize)));
6024       ldr(tmp4, Address(pre(a2, wordSize)));
6025       cmp(cnt1, (u1)elem_per_word);
6026       br(LE, TAIL2);
6027       cmp(tmp1, tmp2);
6028     } br(EQ, NEXT_DWORD);
6029     b(DONE);
6030 
6031     bind(TAIL);
6032     eor(tmp4, tmp3, tmp4);
6033     eor(tmp2, tmp1, tmp2);
6034     lslv(tmp2, tmp2, tmp5);
6035     orr(tmp5, tmp4, tmp2);
6036     cmp(tmp5, zr);
6037     b(CSET_EQ);
6038 
6039     bind(TAIL2);
6040     eor(tmp2, tmp1, tmp2);
6041     cbnz(tmp2, DONE);
6042     b(LAST_CHECK);
6043 
6044     bind(STUB);
6045     ldr(tmp4, Address(pre(a2, start_offset)));
6046     if (elem_size == 2) { // convert to byte counter
6047       lsl(cnt1, cnt1, 1);
6048     }
6049     eor(tmp5, tmp3, tmp4);
6050     cbnz(tmp5, DONE);
6051     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6052     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6053     address tpc = trampoline_call(stub);
6054     if (tpc == nullptr) {
6055       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6056       postcond(pc() == badAddress);
6057       return nullptr;
6058     }
6059     b(DONE);
6060 
6061     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6062     // so, if a2 == null => return false(0), else return true, so we can return a2
6063     mov(result, a2);
6064     b(DONE);
6065     bind(SHORT);
6066     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6067     ldr(tmp3, Address(a1, start_offset));
6068     ldr(tmp4, Address(a2, start_offset));
6069     bind(LAST_CHECK);
6070     eor(tmp4, tmp3, tmp4);
6071     lslv(tmp5, tmp4, tmp5);
6072     cmp(tmp5, zr);
6073     bind(CSET_EQ);
6074     cset(result, EQ);
6075     b(DONE);
6076   }
6077 
6078   bind(SAME);
6079   mov(result, true);
6080   // That's it.
6081   bind(DONE);
6082 
6083   BLOCK_COMMENT("} array_equals");
6084   postcond(pc() != badAddress);
6085   return pc();
6086 }
6087 
6088 // Compare Strings
6089 
6090 // For Strings we're passed the address of the first characters in a1
6091 // and a2 and the length in cnt1.
6092 // There are two implementations.  For arrays >= 8 bytes, all
6093 // comparisons (including the final one, which may overlap) are
6094 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6095 // halfword, then a short, and then a byte.
6096 
6097 void MacroAssembler::string_equals(Register a1, Register a2,
6098                                    Register result, Register cnt1)
6099 {
6100   Label SAME, DONE, SHORT, NEXT_WORD;
6101   Register tmp1 = rscratch1;
6102   Register tmp2 = rscratch2;
6103 
6104   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6105 
6106 #ifndef PRODUCT
6107   {
6108     char comment[64];
6109     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6110     BLOCK_COMMENT(comment);
6111   }
6112 #endif
6113 
6114   mov(result, false);
6115 
6116   // Check for short strings, i.e. smaller than wordSize.
6117   subs(cnt1, cnt1, wordSize);
6118   br(Assembler::LT, SHORT);
6119   // Main 8 byte comparison loop.
6120   bind(NEXT_WORD); {
6121     ldr(tmp1, Address(post(a1, wordSize)));
6122     ldr(tmp2, Address(post(a2, wordSize)));
6123     subs(cnt1, cnt1, wordSize);
6124     eor(tmp1, tmp1, tmp2);
6125     cbnz(tmp1, DONE);
6126   } br(GT, NEXT_WORD);
6127   // Last longword.  In the case where length == 4 we compare the
6128   // same longword twice, but that's still faster than another
6129   // conditional branch.
6130   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6131   // length == 4.
6132   ldr(tmp1, Address(a1, cnt1));
6133   ldr(tmp2, Address(a2, cnt1));
6134   eor(tmp2, tmp1, tmp2);
6135   cbnz(tmp2, DONE);
6136   b(SAME);
6137 
6138   bind(SHORT);
6139   Label TAIL03, TAIL01;
6140 
6141   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6142   {
6143     ldrw(tmp1, Address(post(a1, 4)));
6144     ldrw(tmp2, Address(post(a2, 4)));
6145     eorw(tmp1, tmp1, tmp2);
6146     cbnzw(tmp1, DONE);
6147   }
6148   bind(TAIL03);
6149   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6150   {
6151     ldrh(tmp1, Address(post(a1, 2)));
6152     ldrh(tmp2, Address(post(a2, 2)));
6153     eorw(tmp1, tmp1, tmp2);
6154     cbnzw(tmp1, DONE);
6155   }
6156   bind(TAIL01);
6157   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6158     {
6159     ldrb(tmp1, a1);
6160     ldrb(tmp2, a2);
6161     eorw(tmp1, tmp1, tmp2);
6162     cbnzw(tmp1, DONE);
6163   }
6164   // Arrays are equal.
6165   bind(SAME);
6166   mov(result, true);
6167 
6168   // That's it.
6169   bind(DONE);
6170   BLOCK_COMMENT("} string_equals");
6171 }
6172 
6173 
6174 // The size of the blocks erased by the zero_blocks stub.  We must
6175 // handle anything smaller than this ourselves in zero_words().
6176 const int MacroAssembler::zero_words_block_size = 8;
6177 
6178 // zero_words() is used by C2 ClearArray patterns and by
6179 // C1_MacroAssembler.  It is as small as possible, handling small word
6180 // counts locally and delegating anything larger to the zero_blocks
6181 // stub.  It is expanded many times in compiled code, so it is
6182 // important to keep it short.
6183 
6184 // ptr:   Address of a buffer to be zeroed.
6185 // cnt:   Count in HeapWords.
6186 //
6187 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6188 address MacroAssembler::zero_words(Register ptr, Register cnt)
6189 {
6190   assert(is_power_of_2(zero_words_block_size), "adjust this");
6191 
6192   BLOCK_COMMENT("zero_words {");
6193   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6194   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6195   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6196 
6197   subs(rscratch1, cnt, zero_words_block_size);
6198   Label around;
6199   br(LO, around);
6200   {
6201     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6202     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6203     // Make sure this is a C2 compilation. C1 allocates space only for
6204     // trampoline stubs generated by Call LIR ops, and in any case it
6205     // makes sense for a C1 compilation task to proceed as quickly as
6206     // possible.
6207     CompileTask* task;
6208     if (StubRoutines::aarch64::complete()
6209         && Thread::current()->is_Compiler_thread()
6210         && (task = ciEnv::current()->task())
6211         && is_c2_compile(task->comp_level())) {
6212       address tpc = trampoline_call(zero_blocks);
6213       if (tpc == nullptr) {
6214         DEBUG_ONLY(reset_labels(around));
6215         return nullptr;
6216       }
6217     } else {
6218       far_call(zero_blocks);
6219     }
6220   }
6221   bind(around);
6222 
6223   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6224   // for us.
6225   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6226     Label l;
6227     tbz(cnt, exact_log2(i), l);
6228     for (int j = 0; j < i; j += 2) {
6229       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6230     }
6231     bind(l);
6232   }
6233   {
6234     Label l;
6235     tbz(cnt, 0, l);
6236     str(zr, Address(ptr));
6237     bind(l);
6238   }
6239 
6240   BLOCK_COMMENT("} zero_words");
6241   return pc();
6242 }
6243 
6244 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6245 // cnt:          Immediate count in HeapWords.
6246 //
6247 // r10, r11, rscratch1, and rscratch2 are clobbered.
6248 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6249 {
6250   assert(wordSize <= BlockZeroingLowLimit,
6251             "increase BlockZeroingLowLimit");
6252   address result = nullptr;
6253   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6254 #ifndef PRODUCT
6255     {
6256       char buf[64];
6257       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6258       BLOCK_COMMENT(buf);
6259     }
6260 #endif
6261     if (cnt >= 16) {
6262       uint64_t loops = cnt/16;
6263       if (loops > 1) {
6264         mov(rscratch2, loops - 1);
6265       }
6266       {
6267         Label loop;
6268         bind(loop);
6269         for (int i = 0; i < 16; i += 2) {
6270           stp(zr, zr, Address(base, i * BytesPerWord));
6271         }
6272         add(base, base, 16 * BytesPerWord);
6273         if (loops > 1) {
6274           subs(rscratch2, rscratch2, 1);
6275           br(GE, loop);
6276         }
6277       }
6278     }
6279     cnt %= 16;
6280     int i = cnt & 1;  // store any odd word to start
6281     if (i) str(zr, Address(base));
6282     for (; i < (int)cnt; i += 2) {
6283       stp(zr, zr, Address(base, i * wordSize));
6284     }
6285     BLOCK_COMMENT("} zero_words");
6286     result = pc();
6287   } else {
6288     mov(r10, base); mov(r11, cnt);
6289     result = zero_words(r10, r11);
6290   }
6291   return result;
6292 }
6293 
6294 // Zero blocks of memory by using DC ZVA.
6295 //
6296 // Aligns the base address first sufficiently for DC ZVA, then uses
6297 // DC ZVA repeatedly for every full block.  cnt is the size to be
6298 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6299 // in cnt.
6300 //
6301 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6302 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6303 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6304   Register tmp = rscratch1;
6305   Register tmp2 = rscratch2;
6306   int zva_length = VM_Version::zva_length();
6307   Label initial_table_end, loop_zva;
6308   Label fini;
6309 
6310   // Base must be 16 byte aligned. If not just return and let caller handle it
6311   tst(base, 0x0f);
6312   br(Assembler::NE, fini);
6313   // Align base with ZVA length.
6314   neg(tmp, base);
6315   andr(tmp, tmp, zva_length - 1);
6316 
6317   // tmp: the number of bytes to be filled to align the base with ZVA length.
6318   add(base, base, tmp);
6319   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6320   adr(tmp2, initial_table_end);
6321   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6322   br(tmp2);
6323 
6324   for (int i = -zva_length + 16; i < 0; i += 16)
6325     stp(zr, zr, Address(base, i));
6326   bind(initial_table_end);
6327 
6328   sub(cnt, cnt, zva_length >> 3);
6329   bind(loop_zva);
6330   dc(Assembler::ZVA, base);
6331   subs(cnt, cnt, zva_length >> 3);
6332   add(base, base, zva_length);
6333   br(Assembler::GE, loop_zva);
6334   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6335   bind(fini);
6336 }
6337 
6338 // base:   Address of a buffer to be filled, 8 bytes aligned.
6339 // cnt:    Count in 8-byte unit.
6340 // value:  Value to be filled with.
6341 // base will point to the end of the buffer after filling.
6342 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6343 {
6344 //  Algorithm:
6345 //
6346 //    if (cnt == 0) {
6347 //      return;
6348 //    }
6349 //    if ((p & 8) != 0) {
6350 //      *p++ = v;
6351 //    }
6352 //
6353 //    scratch1 = cnt & 14;
6354 //    cnt -= scratch1;
6355 //    p += scratch1;
6356 //    switch (scratch1 / 2) {
6357 //      do {
6358 //        cnt -= 16;
6359 //          p[-16] = v;
6360 //          p[-15] = v;
6361 //        case 7:
6362 //          p[-14] = v;
6363 //          p[-13] = v;
6364 //        case 6:
6365 //          p[-12] = v;
6366 //          p[-11] = v;
6367 //          // ...
6368 //        case 1:
6369 //          p[-2] = v;
6370 //          p[-1] = v;
6371 //        case 0:
6372 //          p += 16;
6373 //      } while (cnt);
6374 //    }
6375 //    if ((cnt & 1) == 1) {
6376 //      *p++ = v;
6377 //    }
6378 
6379   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6380 
6381   Label fini, skip, entry, loop;
6382   const int unroll = 8; // Number of stp instructions we'll unroll
6383 
6384   cbz(cnt, fini);
6385   tbz(base, 3, skip);
6386   str(value, Address(post(base, 8)));
6387   sub(cnt, cnt, 1);
6388   bind(skip);
6389 
6390   andr(rscratch1, cnt, (unroll-1) * 2);
6391   sub(cnt, cnt, rscratch1);
6392   add(base, base, rscratch1, Assembler::LSL, 3);
6393   adr(rscratch2, entry);
6394   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6395   br(rscratch2);
6396 
6397   bind(loop);
6398   add(base, base, unroll * 16);
6399   for (int i = -unroll; i < 0; i++)
6400     stp(value, value, Address(base, i * 16));
6401   bind(entry);
6402   subs(cnt, cnt, unroll * 2);
6403   br(Assembler::GE, loop);
6404 
6405   tbz(cnt, 0, fini);
6406   str(value, Address(post(base, 8)));
6407   bind(fini);
6408 }
6409 
6410 // Intrinsic for
6411 //
6412 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6413 //   Encodes char[] to byte[] in ISO-8859-1
6414 //
6415 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6416 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6417 //
6418 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6419 //   Encodes char[] to byte[] in ASCII
6420 //
6421 // This version always returns the number of characters copied, and does not
6422 // clobber the 'len' register. A successful copy will complete with the post-
6423 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6424 // post-condition: 0 <= 'res' < 'len'.
6425 //
6426 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6427 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6428 //       beyond the acceptable, even though the footprint would be smaller.
6429 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6430 //       avoid additional bloat.
6431 //
6432 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6433 void MacroAssembler::encode_iso_array(Register src, Register dst,
6434                                       Register len, Register res, bool ascii,
6435                                       FloatRegister vtmp0, FloatRegister vtmp1,
6436                                       FloatRegister vtmp2, FloatRegister vtmp3,
6437                                       FloatRegister vtmp4, FloatRegister vtmp5)
6438 {
6439   Register cnt = res;
6440   Register max = rscratch1;
6441   Register chk = rscratch2;
6442 
6443   prfm(Address(src), PLDL1STRM);
6444   movw(cnt, len);
6445 
6446 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6447 
6448   Label LOOP_32, DONE_32, FAIL_32;
6449 
6450   BIND(LOOP_32);
6451   {
6452     cmpw(cnt, 32);
6453     br(LT, DONE_32);
6454     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6455     // Extract lower bytes.
6456     FloatRegister vlo0 = vtmp4;
6457     FloatRegister vlo1 = vtmp5;
6458     uzp1(vlo0, T16B, vtmp0, vtmp1);
6459     uzp1(vlo1, T16B, vtmp2, vtmp3);
6460     // Merge bits...
6461     orr(vtmp0, T16B, vtmp0, vtmp1);
6462     orr(vtmp2, T16B, vtmp2, vtmp3);
6463     // Extract merged upper bytes.
6464     FloatRegister vhix = vtmp0;
6465     uzp2(vhix, T16B, vtmp0, vtmp2);
6466     // ISO-check on hi-parts (all zero).
6467     //                          ASCII-check on lo-parts (no sign).
6468     FloatRegister vlox = vtmp1; // Merge lower bytes.
6469                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6470     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6471     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6472     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6473                                 ASCII(orr(chk, chk, max));
6474     cbnz(chk, FAIL_32);
6475     subw(cnt, cnt, 32);
6476     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6477     b(LOOP_32);
6478   }
6479   BIND(FAIL_32);
6480   sub(src, src, 64);
6481   BIND(DONE_32);
6482 
6483   Label LOOP_8, SKIP_8;
6484 
6485   BIND(LOOP_8);
6486   {
6487     cmpw(cnt, 8);
6488     br(LT, SKIP_8);
6489     FloatRegister vhi = vtmp0;
6490     FloatRegister vlo = vtmp1;
6491     ld1(vtmp3, T8H, src);
6492     uzp1(vlo, T16B, vtmp3, vtmp3);
6493     uzp2(vhi, T16B, vtmp3, vtmp3);
6494     // ISO-check on hi-parts (all zero).
6495     //                          ASCII-check on lo-parts (no sign).
6496                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6497     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6498                                 ASCII(umov(max, vtmp2, B, 0));
6499                                 ASCII(orr(chk, chk, max));
6500     cbnz(chk, SKIP_8);
6501 
6502     strd(vlo, Address(post(dst, 8)));
6503     subw(cnt, cnt, 8);
6504     add(src, src, 16);
6505     b(LOOP_8);
6506   }
6507   BIND(SKIP_8);
6508 
6509 #undef ASCII
6510 
6511   Label LOOP, DONE;
6512 
6513   cbz(cnt, DONE);
6514   BIND(LOOP);
6515   {
6516     Register chr = rscratch1;
6517     ldrh(chr, Address(post(src, 2)));
6518     tst(chr, ascii ? 0xff80 : 0xff00);
6519     br(NE, DONE);
6520     strb(chr, Address(post(dst, 1)));
6521     subs(cnt, cnt, 1);
6522     br(GT, LOOP);
6523   }
6524   BIND(DONE);
6525   // Return index where we stopped.
6526   subw(res, len, cnt);
6527 }
6528 
6529 // Inflate byte[] array to char[].
6530 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6531 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6532                                            FloatRegister vtmp1, FloatRegister vtmp2,
6533                                            FloatRegister vtmp3, Register tmp4) {
6534   Label big, done, after_init, to_stub;
6535 
6536   assert_different_registers(src, dst, len, tmp4, rscratch1);
6537 
6538   fmovd(vtmp1, 0.0);
6539   lsrw(tmp4, len, 3);
6540   bind(after_init);
6541   cbnzw(tmp4, big);
6542   // Short string: less than 8 bytes.
6543   {
6544     Label loop, tiny;
6545 
6546     cmpw(len, 4);
6547     br(LT, tiny);
6548     // Use SIMD to do 4 bytes.
6549     ldrs(vtmp2, post(src, 4));
6550     zip1(vtmp3, T8B, vtmp2, vtmp1);
6551     subw(len, len, 4);
6552     strd(vtmp3, post(dst, 8));
6553 
6554     cbzw(len, done);
6555 
6556     // Do the remaining bytes by steam.
6557     bind(loop);
6558     ldrb(tmp4, post(src, 1));
6559     strh(tmp4, post(dst, 2));
6560     subw(len, len, 1);
6561 
6562     bind(tiny);
6563     cbnz(len, loop);
6564 
6565     b(done);
6566   }
6567 
6568   if (SoftwarePrefetchHintDistance >= 0) {
6569     bind(to_stub);
6570       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6571       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6572       address tpc = trampoline_call(stub);
6573       if (tpc == nullptr) {
6574         DEBUG_ONLY(reset_labels(big, done));
6575         postcond(pc() == badAddress);
6576         return nullptr;
6577       }
6578       b(after_init);
6579   }
6580 
6581   // Unpack the bytes 8 at a time.
6582   bind(big);
6583   {
6584     Label loop, around, loop_last, loop_start;
6585 
6586     if (SoftwarePrefetchHintDistance >= 0) {
6587       const int large_loop_threshold = (64 + 16)/8;
6588       ldrd(vtmp2, post(src, 8));
6589       andw(len, len, 7);
6590       cmp(tmp4, (u1)large_loop_threshold);
6591       br(GE, to_stub);
6592       b(loop_start);
6593 
6594       bind(loop);
6595       ldrd(vtmp2, post(src, 8));
6596       bind(loop_start);
6597       subs(tmp4, tmp4, 1);
6598       br(EQ, loop_last);
6599       zip1(vtmp2, T16B, vtmp2, vtmp1);
6600       ldrd(vtmp3, post(src, 8));
6601       st1(vtmp2, T8H, post(dst, 16));
6602       subs(tmp4, tmp4, 1);
6603       zip1(vtmp3, T16B, vtmp3, vtmp1);
6604       st1(vtmp3, T8H, post(dst, 16));
6605       br(NE, loop);
6606       b(around);
6607       bind(loop_last);
6608       zip1(vtmp2, T16B, vtmp2, vtmp1);
6609       st1(vtmp2, T8H, post(dst, 16));
6610       bind(around);
6611       cbz(len, done);
6612     } else {
6613       andw(len, len, 7);
6614       bind(loop);
6615       ldrd(vtmp2, post(src, 8));
6616       sub(tmp4, tmp4, 1);
6617       zip1(vtmp3, T16B, vtmp2, vtmp1);
6618       st1(vtmp3, T8H, post(dst, 16));
6619       cbnz(tmp4, loop);
6620     }
6621   }
6622 
6623   // Do the tail of up to 8 bytes.
6624   add(src, src, len);
6625   ldrd(vtmp3, Address(src, -8));
6626   add(dst, dst, len, ext::uxtw, 1);
6627   zip1(vtmp3, T16B, vtmp3, vtmp1);
6628   strq(vtmp3, Address(dst, -16));
6629 
6630   bind(done);
6631   postcond(pc() != badAddress);
6632   return pc();
6633 }
6634 
6635 // Compress char[] array to byte[].
6636 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
6637 // Return the array length if every element in array can be encoded,
6638 // otherwise, the index of first non-latin1 (> 0xff) character.
6639 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
6640                                          Register res,
6641                                          FloatRegister tmp0, FloatRegister tmp1,
6642                                          FloatRegister tmp2, FloatRegister tmp3,
6643                                          FloatRegister tmp4, FloatRegister tmp5) {
6644   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
6645 }
6646 
6647 // java.math.round(double a)
6648 // Returns the closest long to the argument, with ties rounding to
6649 // positive infinity.  This requires some fiddling for corner
6650 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
6651 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
6652                                        FloatRegister ftmp) {
6653   Label DONE;
6654   BLOCK_COMMENT("java_round_double: { ");
6655   fmovd(rscratch1, src);
6656   // Use RoundToNearestTiesAway unless src small and -ve.
6657   fcvtasd(dst, src);
6658   // Test if src >= 0 || abs(src) >= 0x1.0p52
6659   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
6660   mov(rscratch2, julong_cast(0x1.0p52));
6661   cmp(rscratch1, rscratch2);
6662   br(HS, DONE); {
6663     // src < 0 && abs(src) < 0x1.0p52
6664     // src may have a fractional part, so add 0.5
6665     fmovd(ftmp, 0.5);
6666     faddd(ftmp, src, ftmp);
6667     // Convert double to jlong, use RoundTowardsNegative
6668     fcvtmsd(dst, ftmp);
6669   }
6670   bind(DONE);
6671   BLOCK_COMMENT("} java_round_double");
6672 }
6673 
6674 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
6675                                       FloatRegister ftmp) {
6676   Label DONE;
6677   BLOCK_COMMENT("java_round_float: { ");
6678   fmovs(rscratch1, src);
6679   // Use RoundToNearestTiesAway unless src small and -ve.
6680   fcvtassw(dst, src);
6681   // Test if src >= 0 || abs(src) >= 0x1.0p23
6682   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
6683   mov(rscratch2, jint_cast(0x1.0p23f));
6684   cmp(rscratch1, rscratch2);
6685   br(HS, DONE); {
6686     // src < 0 && |src| < 0x1.0p23
6687     // src may have a fractional part, so add 0.5
6688     fmovs(ftmp, 0.5f);
6689     fadds(ftmp, src, ftmp);
6690     // Convert float to jint, use RoundTowardsNegative
6691     fcvtmssw(dst, ftmp);
6692   }
6693   bind(DONE);
6694   BLOCK_COMMENT("} java_round_float");
6695 }
6696 
6697 // get_thread() can be called anywhere inside generated code so we
6698 // need to save whatever non-callee save context might get clobbered
6699 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
6700 // the call setup code.
6701 //
6702 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
6703 // On other systems, the helper is a usual C function.
6704 //
6705 void MacroAssembler::get_thread(Register dst) {
6706   RegSet saved_regs =
6707     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
6708     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
6709 
6710   protect_return_address();
6711   push(saved_regs, sp);
6712 
6713   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
6714   blr(lr);
6715   if (dst != c_rarg0) {
6716     mov(dst, c_rarg0);
6717   }
6718 
6719   pop(saved_regs, sp);
6720   authenticate_return_address();
6721 }
6722 
6723 void MacroAssembler::cache_wb(Address line) {
6724   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
6725   assert(line.index() == noreg, "index should be noreg");
6726   assert(line.offset() == 0, "offset should be 0");
6727   // would like to assert this
6728   // assert(line._ext.shift == 0, "shift should be zero");
6729   if (VM_Version::supports_dcpop()) {
6730     // writeback using clear virtual address to point of persistence
6731     dc(Assembler::CVAP, line.base());
6732   } else {
6733     // no need to generate anything as Unsafe.writebackMemory should
6734     // never invoke this stub
6735   }
6736 }
6737 
6738 void MacroAssembler::cache_wbsync(bool is_pre) {
6739   // we only need a barrier post sync
6740   if (!is_pre) {
6741     membar(Assembler::AnyAny);
6742   }
6743 }
6744 
6745 void MacroAssembler::verify_sve_vector_length(Register tmp) {
6746   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
6747     return;
6748   }
6749   // Make sure that native code does not change SVE vector length.
6750   Label verify_ok;
6751   movw(tmp, zr);
6752   sve_inc(tmp, B);
6753   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
6754   br(EQ, verify_ok);
6755   stop("Error: SVE vector length has changed since jvm startup");
6756   bind(verify_ok);
6757 }
6758 
6759 void MacroAssembler::verify_ptrue() {
6760   Label verify_ok;
6761   if (!UseSVE) {
6762     return;
6763   }
6764   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
6765   sve_dec(rscratch1, B);
6766   cbz(rscratch1, verify_ok);
6767   stop("Error: the preserved predicate register (p7) elements are not all true");
6768   bind(verify_ok);
6769 }
6770 
6771 void MacroAssembler::safepoint_isb() {
6772   isb();
6773 #ifndef PRODUCT
6774   if (VerifyCrossModifyFence) {
6775     // Clear the thread state.
6776     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6777   }
6778 #endif
6779 }
6780 
6781 #ifndef PRODUCT
6782 void MacroAssembler::verify_cross_modify_fence_not_required() {
6783   if (VerifyCrossModifyFence) {
6784     // Check if thread needs a cross modify fence.
6785     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
6786     Label fence_not_required;
6787     cbz(rscratch1, fence_not_required);
6788     // If it does then fail.
6789     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
6790     mov(c_rarg0, rthread);
6791     blr(rscratch1);
6792     bind(fence_not_required);
6793   }
6794 }
6795 #endif
6796 
6797 void MacroAssembler::spin_wait() {
6798   block_comment("spin_wait {");
6799   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
6800     switch (VM_Version::spin_wait_desc().inst()) {
6801       case SpinWait::NOP:
6802         nop();
6803         break;
6804       case SpinWait::ISB:
6805         isb();
6806         break;
6807       case SpinWait::YIELD:
6808         yield();
6809         break;
6810       case SpinWait::SB:
6811         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
6812         sb();
6813         break;
6814       case SpinWait::WFET:
6815         spin_wait_wfet(VM_Version::spin_wait_desc().delay());
6816         break;
6817       default:
6818         ShouldNotReachHere();
6819     }
6820   }
6821   block_comment("}");
6822 }
6823 
6824 void MacroAssembler::spin_wait_wfet(int delay_ns) {
6825   // The sequence assumes CNTFRQ_EL0 is fixed to 1GHz. The assumption is valid
6826   // starting from Armv8.6, according to the "D12.1.2 The system counter" of the
6827   // Arm Architecture Reference Manual for A-profile architecture version M.a.a.
6828   // This is sufficient because FEAT_WFXT is introduced from Armv8.6.
6829   Register target = rscratch1;
6830   Register current = rscratch2;
6831   get_cntvctss_el0(current);
6832   add(target, current, delay_ns);
6833 
6834   Label L_wait_loop;
6835   bind(L_wait_loop);
6836 
6837   wfet(target);
6838   get_cntvctss_el0(current);
6839 
6840   cmp(current, target);
6841   br(LT, L_wait_loop);
6842 
6843   sb();
6844 }
6845 
6846 // Stack frame creation/removal
6847 
6848 void MacroAssembler::enter(bool strip_ret_addr) {
6849   if (strip_ret_addr) {
6850     // Addresses can only be signed once. If there are multiple nested frames being created
6851     // in the same function, then the return address needs stripping first.
6852     strip_return_address();
6853   }
6854   protect_return_address();
6855   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6856   mov(rfp, sp);
6857 }
6858 
6859 void MacroAssembler::leave() {
6860   mov(sp, rfp);
6861   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6862   authenticate_return_address();
6863 }
6864 
6865 // ROP Protection
6866 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
6867 // destroying stack frames or whenever directly loading/storing the LR to memory.
6868 // If ROP protection is not set then these functions are no-ops.
6869 // For more details on PAC see pauth_aarch64.hpp.
6870 
6871 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
6872 // Uses value zero as the modifier.
6873 //
6874 void MacroAssembler::protect_return_address() {
6875   if (VM_Version::use_rop_protection()) {
6876     check_return_address();
6877     paciaz();
6878   }
6879 }
6880 
6881 // Sign the return value in the given register. Use before updating the LR in the existing stack
6882 // frame for the current function.
6883 // Uses value zero as the modifier.
6884 //
6885 void MacroAssembler::protect_return_address(Register return_reg) {
6886   if (VM_Version::use_rop_protection()) {
6887     check_return_address(return_reg);
6888     paciza(return_reg);
6889   }
6890 }
6891 
6892 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
6893 // Uses value zero as the modifier.
6894 //
6895 void MacroAssembler::authenticate_return_address() {
6896   if (VM_Version::use_rop_protection()) {
6897     autiaz();
6898     check_return_address();
6899   }
6900 }
6901 
6902 // Authenticate the return value in the given register. Use before updating the LR in the existing
6903 // stack frame for the current function.
6904 // Uses value zero as the modifier.
6905 //
6906 void MacroAssembler::authenticate_return_address(Register return_reg) {
6907   if (VM_Version::use_rop_protection()) {
6908     autiza(return_reg);
6909     check_return_address(return_reg);
6910   }
6911 }
6912 
6913 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
6914 // there is no guaranteed way of authenticating the LR.
6915 //
6916 void MacroAssembler::strip_return_address() {
6917   if (VM_Version::use_rop_protection()) {
6918     xpaclri();
6919   }
6920 }
6921 
6922 #ifndef PRODUCT
6923 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
6924 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
6925 // it is difficult to debug back to the callee function.
6926 // This function simply loads from the address in the given register.
6927 // Use directly after authentication to catch authentication failures.
6928 // Also use before signing to check that the pointer is valid and hasn't already been signed.
6929 //
6930 void MacroAssembler::check_return_address(Register return_reg) {
6931   if (VM_Version::use_rop_protection()) {
6932     ldr(zr, Address(return_reg));
6933   }
6934 }
6935 #endif
6936 
6937 // The java_calling_convention describes stack locations as ideal slots on
6938 // a frame with no abi restrictions. Since we must observe abi restrictions
6939 // (like the placement of the register window) the slots must be biased by
6940 // the following value.
6941 static int reg2offset_in(VMReg r) {
6942   // Account for saved rfp and lr
6943   // This should really be in_preserve_stack_slots
6944   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
6945 }
6946 
6947 static int reg2offset_out(VMReg r) {
6948   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
6949 }
6950 
6951 // On 64bit we will store integer like items to the stack as
6952 // 64bits items (AArch64 ABI) even though java would only store
6953 // 32bits for a parameter. On 32bit it will simply be 32bits
6954 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
6955 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
6956   if (src.first()->is_stack()) {
6957     if (dst.first()->is_stack()) {
6958       // stack to stack
6959       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
6960       str(tmp, Address(sp, reg2offset_out(dst.first())));
6961     } else {
6962       // stack to reg
6963       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
6964     }
6965   } else if (dst.first()->is_stack()) {
6966     // reg to stack
6967     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
6968   } else {
6969     if (dst.first() != src.first()) {
6970       sxtw(dst.first()->as_Register(), src.first()->as_Register());
6971     }
6972   }
6973 }
6974 
6975 // An oop arg. Must pass a handle not the oop itself
6976 void MacroAssembler::object_move(
6977                         OopMap* map,
6978                         int oop_handle_offset,
6979                         int framesize_in_slots,
6980                         VMRegPair src,
6981                         VMRegPair dst,
6982                         bool is_receiver,
6983                         int* receiver_offset) {
6984 
6985   // must pass a handle. First figure out the location we use as a handle
6986 
6987   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
6988 
6989   // See if oop is null if it is we need no handle
6990 
6991   if (src.first()->is_stack()) {
6992 
6993     // Oop is already on the stack as an argument
6994     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
6995     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
6996     if (is_receiver) {
6997       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
6998     }
6999 
7000     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7001     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7002     // conditionally move a null
7003     cmp(rscratch1, zr);
7004     csel(rHandle, zr, rHandle, Assembler::EQ);
7005   } else {
7006 
7007     // Oop is in an a register we must store it to the space we reserve
7008     // on the stack for oop_handles and pass a handle if oop is non-null
7009 
7010     const Register rOop = src.first()->as_Register();
7011     int oop_slot;
7012     if (rOop == j_rarg0)
7013       oop_slot = 0;
7014     else if (rOop == j_rarg1)
7015       oop_slot = 1;
7016     else if (rOop == j_rarg2)
7017       oop_slot = 2;
7018     else if (rOop == j_rarg3)
7019       oop_slot = 3;
7020     else if (rOop == j_rarg4)
7021       oop_slot = 4;
7022     else if (rOop == j_rarg5)
7023       oop_slot = 5;
7024     else if (rOop == j_rarg6)
7025       oop_slot = 6;
7026     else {
7027       assert(rOop == j_rarg7, "wrong register");
7028       oop_slot = 7;
7029     }
7030 
7031     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7032     int offset = oop_slot*VMRegImpl::stack_slot_size;
7033 
7034     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7035     // Store oop in handle area, may be null
7036     str(rOop, Address(sp, offset));
7037     if (is_receiver) {
7038       *receiver_offset = offset;
7039     }
7040 
7041     cmp(rOop, zr);
7042     lea(rHandle, Address(sp, offset));
7043     // conditionally move a null
7044     csel(rHandle, zr, rHandle, Assembler::EQ);
7045   }
7046 
7047   // If arg is on the stack then place it otherwise it is already in correct reg.
7048   if (dst.first()->is_stack()) {
7049     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7050   }
7051 }
7052 
7053 // A float arg may have to do float reg int reg conversion
7054 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7055  if (src.first()->is_stack()) {
7056     if (dst.first()->is_stack()) {
7057       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7058       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7059     } else {
7060       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7061     }
7062   } else if (src.first() != dst.first()) {
7063     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7064       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7065     else
7066       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7067   }
7068 }
7069 
7070 // A long move
7071 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7072   if (src.first()->is_stack()) {
7073     if (dst.first()->is_stack()) {
7074       // stack to stack
7075       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7076       str(tmp, Address(sp, reg2offset_out(dst.first())));
7077     } else {
7078       // stack to reg
7079       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7080     }
7081   } else if (dst.first()->is_stack()) {
7082     // reg to stack
7083     // Do we really have to sign extend???
7084     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7085     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7086   } else {
7087     if (dst.first() != src.first()) {
7088       mov(dst.first()->as_Register(), src.first()->as_Register());
7089     }
7090   }
7091 }
7092 
7093 
7094 // A double move
7095 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7096  if (src.first()->is_stack()) {
7097     if (dst.first()->is_stack()) {
7098       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7099       str(tmp, Address(sp, reg2offset_out(dst.first())));
7100     } else {
7101       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7102     }
7103   } else if (src.first() != dst.first()) {
7104     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7105       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7106     else
7107       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7108   }
7109 }
7110 
7111 // Implements fast-locking.
7112 //
7113 //  - obj: the object to be locked
7114 //  - t1, t2, t3: temporary registers, will be destroyed
7115 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7116 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7117   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7118 
7119   Label push;
7120   const Register top = t1;
7121   const Register mark = t2;
7122   const Register t = t3;
7123 
7124   // Preload the markWord. It is important that this is the first
7125   // instruction emitted as it is part of C1's null check semantics.
7126   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7127 
7128   if (UseObjectMonitorTable) {
7129     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7130     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7131   }
7132 
7133   if (DiagnoseSyncOnValueBasedClasses != 0) {
7134     load_klass(t1, obj);
7135     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7136     tst(t1, KlassFlags::_misc_is_value_based_class);
7137     br(Assembler::NE, slow);
7138   }
7139 
7140   // Check if the lock-stack is full.
7141   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7142   cmpw(top, (unsigned)LockStack::end_offset());
7143   br(Assembler::GE, slow);
7144 
7145   // Check for recursion.
7146   subw(t, top, oopSize);
7147   ldr(t, Address(rthread, t));
7148   cmp(obj, t);
7149   br(Assembler::EQ, push);
7150 
7151   // Check header for monitor (0b10).
7152   tst(mark, markWord::monitor_value);
7153   br(Assembler::NE, slow);
7154 
7155   // Try to lock. Transition lock bits 0b01 => 0b00
7156   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7157   orr(mark, mark, markWord::unlocked_value);
7158   eor(t, mark, markWord::unlocked_value);
7159   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7160           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7161   br(Assembler::NE, slow);
7162 
7163   bind(push);
7164   // After successful lock, push object on lock-stack.
7165   str(obj, Address(rthread, top));
7166   addw(top, top, oopSize);
7167   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7168 }
7169 
7170 // Implements fast-unlocking.
7171 //
7172 // - obj: the object to be unlocked
7173 // - t1, t2, t3: temporary registers
7174 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7175 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7176   // cmpxchg clobbers rscratch1.
7177   assert_different_registers(obj, t1, t2, t3, rscratch1);
7178 
7179 #ifdef ASSERT
7180   {
7181     // Check for lock-stack underflow.
7182     Label stack_ok;
7183     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7184     cmpw(t1, (unsigned)LockStack::start_offset());
7185     br(Assembler::GE, stack_ok);
7186     STOP("Lock-stack underflow");
7187     bind(stack_ok);
7188   }
7189 #endif
7190 
7191   Label unlocked, push_and_slow;
7192   const Register top = t1;
7193   const Register mark = t2;
7194   const Register t = t3;
7195 
7196   // Check if obj is top of lock-stack.
7197   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7198   subw(top, top, oopSize);
7199   ldr(t, Address(rthread, top));
7200   cmp(obj, t);
7201   br(Assembler::NE, slow);
7202 
7203   // Pop lock-stack.
7204   DEBUG_ONLY(str(zr, Address(rthread, top));)
7205   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7206 
7207   // Check if recursive.
7208   subw(t, top, oopSize);
7209   ldr(t, Address(rthread, t));
7210   cmp(obj, t);
7211   br(Assembler::EQ, unlocked);
7212 
7213   // Not recursive. Check header for monitor (0b10).
7214   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7215   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
7216 
7217 #ifdef ASSERT
7218   // Check header not unlocked (0b01).
7219   Label not_unlocked;
7220   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
7221   stop("fast_unlock already unlocked");
7222   bind(not_unlocked);
7223 #endif
7224 
7225   // Try to unlock. Transition lock bits 0b00 => 0b01
7226   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7227   orr(t, mark, markWord::unlocked_value);
7228   cmpxchg(obj, mark, t, Assembler::xword,
7229           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
7230   br(Assembler::EQ, unlocked);
7231 
7232   bind(push_and_slow);
7233   // Restore lock-stack and handle the unlock in runtime.
7234   DEBUG_ONLY(str(obj, Address(rthread, top));)
7235   addw(top, top, oopSize);
7236   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7237   b(slow);
7238 
7239   bind(unlocked);
7240 }