1 /*
   2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/assembler.hpp"
  27 #include "asm/assembler.inline.hpp"
  28 #include "ci/ciEnv.hpp"
  29 #include "ci/ciInlineKlass.hpp"
  30 #include "code/compiledIC.hpp"
  31 #include "compiler/compileTask.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/barrierSet.hpp"
  35 #include "gc/shared/barrierSetAssembler.hpp"
  36 #include "gc/shared/cardTableBarrierSet.hpp"
  37 #include "gc/shared/cardTable.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/tlab_globals.hpp"
  40 #include "interpreter/bytecodeHistogram.hpp"
  41 #include "interpreter/interpreter.hpp"
  42 #include "interpreter/interpreterRuntime.hpp"
  43 #include "jvm.h"
  44 #include "memory/resourceArea.hpp"
  45 #include "memory/universe.hpp"
  46 #include "nativeInst_aarch64.hpp"
  47 #include "oops/accessDecorators.hpp"
  48 #include "oops/compressedKlass.inline.hpp"
  49 #include "oops/compressedOops.inline.hpp"
  50 #include "oops/klass.inline.hpp"
  51 #include "oops/resolvedFieldEntry.hpp"
  52 #include "runtime/arguments.hpp"
  53 #include "runtime/continuation.hpp"
  54 #include "runtime/globals.hpp"
  55 #include "runtime/icache.hpp"
  56 #include "runtime/interfaceSupport.inline.hpp"
  57 #include "runtime/javaThread.hpp"
  58 #include "runtime/jniHandles.inline.hpp"
  59 #include "runtime/sharedRuntime.hpp"
  60 #include "runtime/signature_cc.hpp"
  61 #include "runtime/stubRoutines.hpp"
  62 #include "utilities/globalDefinitions.hpp"
  63 #include "utilities/powerOfTwo.hpp"
  64 #include "vmreg_aarch64.inline.hpp"
  65 #ifdef COMPILER1
  66 #include "c1/c1_LIRAssembler.hpp"
  67 #endif
  68 #ifdef COMPILER2
  69 #include "oops/oop.hpp"
  70 #include "opto/compile.hpp"
  71 #include "opto/node.hpp"
  72 #include "opto/output.hpp"
  73 #endif
  74 
  75 #include <sys/types.h>
  76 
  77 #ifdef PRODUCT
  78 #define BLOCK_COMMENT(str) /* nothing */
  79 #else
  80 #define BLOCK_COMMENT(str) block_comment(str)
  81 #endif
  82 #define STOP(str) stop(str);
  83 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  84 
  85 #ifdef ASSERT
  86 extern "C" void disnm(intptr_t p);
  87 #endif
  88 // Target-dependent relocation processing
  89 //
  90 // Instruction sequences whose target may need to be retrieved or
  91 // patched are distinguished by their leading instruction, sorting
  92 // them into three main instruction groups and related subgroups.
  93 //
  94 // 1) Branch, Exception and System (insn count = 1)
  95 //    1a) Unconditional branch (immediate):
  96 //      b/bl imm19
  97 //    1b) Compare & branch (immediate):
  98 //      cbz/cbnz Rt imm19
  99 //    1c) Test & branch (immediate):
 100 //      tbz/tbnz Rt imm14
 101 //    1d) Conditional branch (immediate):
 102 //      b.cond imm19
 103 //
 104 // 2) Loads and Stores (insn count = 1)
 105 //    2a) Load register literal:
 106 //      ldr Rt imm19
 107 //
 108 // 3) Data Processing Immediate (insn count = 2 or 3)
 109 //    3a) PC-rel. addressing
 110 //      adr/adrp Rx imm21; ldr/str Ry Rx  #imm12
 111 //      adr/adrp Rx imm21; add Ry Rx  #imm12
 112 //      adr/adrp Rx imm21; movk Rx #imm16<<32; ldr/str Ry, [Rx, #offset_in_page]
 113 //      adr/adrp Rx imm21
 114 //      adr/adrp Rx imm21; movk Rx #imm16<<32
 115 //      adr/adrp Rx imm21; movk Rx #imm16<<32; add Ry, Rx, #offset_in_page
 116 //      The latter form can only happen when the target is an
 117 //      ExternalAddress, and (by definition) ExternalAddresses don't
 118 //      move. Because of that property, there is never any need to
 119 //      patch the last of the three instructions. However,
 120 //      MacroAssembler::target_addr_for_insn takes all three
 121 //      instructions into account and returns the correct address.
 122 //    3b) Move wide (immediate)
 123 //      movz Rx #imm16; movk Rx #imm16 << 16; movk Rx #imm16 << 32;
 124 //
 125 // A switch on a subset of the instruction's bits provides an
 126 // efficient dispatch to these subcases.
 127 //
 128 // insn[28:26] -> main group ('x' == don't care)
 129 //   00x -> UNALLOCATED
 130 //   100 -> Data Processing Immediate
 131 //   101 -> Branch, Exception and System
 132 //   x1x -> Loads and Stores
 133 //
 134 // insn[30:25] -> subgroup ('_' == group, 'x' == don't care).
 135 // n.b. in some cases extra bits need to be checked to verify the
 136 // instruction is as expected
 137 //
 138 // 1) ... xx101x Branch, Exception and System
 139 //   1a)  00___x Unconditional branch (immediate)
 140 //   1b)  01___0 Compare & branch (immediate)
 141 //   1c)  01___1 Test & branch (immediate)
 142 //   1d)  10___0 Conditional branch (immediate)
 143 //        other  Should not happen
 144 //
 145 // 2) ... xxx1x0 Loads and Stores
 146 //   2a)  xx1__00 Load/Store register (insn[28] == 1 && insn[24] == 0)
 147 //   2aa) x01__00 Load register literal (i.e. requires insn[29] == 0)
 148 //                strictly should be 64 bit non-FP/SIMD i.e.
 149 //       0101_000 (i.e. requires insn[31:24] == 01011000)
 150 //
 151 // 3) ... xx100x Data Processing Immediate
 152 //   3a)  xx___00 PC-rel. addressing (n.b. requires insn[24] == 0)
 153 //   3b)  xx___101 Move wide (immediate) (n.b. requires insn[24:23] == 01)
 154 //                 strictly should be 64 bit movz #imm16<<0
 155 //       110___10100 (i.e. requires insn[31:21] == 11010010100)
 156 //
 157 
 158 static uint32_t insn_at(address insn_addr, int n) {
 159   return ((uint32_t*)insn_addr)[n];
 160 }
 161 
 162 template<typename T>
 163 class RelocActions : public AllStatic {
 164 
 165 public:
 166 
 167   static int ALWAYSINLINE run(address insn_addr, address &target) {
 168     int instructions = 1;
 169     uint32_t insn = insn_at(insn_addr, 0);
 170 
 171     uint32_t dispatch = Instruction_aarch64::extract(insn, 30, 25);
 172     switch(dispatch) {
 173       case 0b001010:
 174       case 0b001011: {
 175         instructions = T::unconditionalBranch(insn_addr, target);
 176         break;
 177       }
 178       case 0b101010:   // Conditional branch (immediate)
 179       case 0b011010: { // Compare & branch (immediate)
 180         instructions = T::conditionalBranch(insn_addr, target);
 181         break;
 182       }
 183       case 0b011011: {
 184         instructions = T::testAndBranch(insn_addr, target);
 185         break;
 186       }
 187       case 0b001100:
 188       case 0b001110:
 189       case 0b011100:
 190       case 0b011110:
 191       case 0b101100:
 192       case 0b101110:
 193       case 0b111100:
 194       case 0b111110: {
 195         // load/store
 196         if ((Instruction_aarch64::extract(insn, 29, 24) & 0b111011) == 0b011000) {
 197           // Load register (literal)
 198           instructions = T::loadStore(insn_addr, target);
 199           break;
 200         } else {
 201           // nothing to do
 202           assert(target == nullptr, "did not expect to relocate target for polling page load");
 203         }
 204         break;
 205       }
 206       case 0b001000:
 207       case 0b011000:
 208       case 0b101000:
 209       case 0b111000: {
 210         // adr/adrp
 211         assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 212         int shift = Instruction_aarch64::extract(insn, 31, 31);
 213         if (shift) {
 214           uint32_t insn2 = insn_at(insn_addr, 1);
 215           if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 216               Instruction_aarch64::extract(insn, 4, 0) ==
 217               Instruction_aarch64::extract(insn2, 9, 5)) {
 218             instructions = T::adrp(insn_addr, target, T::adrpMem);
 219           } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 220                      Instruction_aarch64::extract(insn, 4, 0) ==
 221                      Instruction_aarch64::extract(insn2, 4, 0)) {
 222             instructions = T::adrp(insn_addr, target, T::adrpAdd);
 223           } else if (Instruction_aarch64::extract(insn2, 31, 21) == 0b11110010110 &&
 224                      Instruction_aarch64::extract(insn, 4, 0) ==
 225                      Instruction_aarch64::extract(insn2, 4, 0)) {
 226             instructions = T::adrp(insn_addr, target, T::adrpMovk);
 227           } else {
 228             ShouldNotReachHere();
 229           }
 230         } else {
 231           instructions = T::adr(insn_addr, target);
 232         }
 233         break;
 234       }
 235       case 0b001001:
 236       case 0b011001:
 237       case 0b101001:
 238       case 0b111001: {
 239         instructions = T::immediate(insn_addr, target);
 240         break;
 241       }
 242       default: {
 243         ShouldNotReachHere();
 244       }
 245     }
 246 
 247     T::verify(insn_addr, target);
 248     return instructions * NativeInstruction::instruction_size;
 249   }
 250 };
 251 
 252 class Patcher : public AllStatic {
 253 public:
 254   static int unconditionalBranch(address insn_addr, address &target) {
 255     intptr_t offset = (target - insn_addr) >> 2;
 256     Instruction_aarch64::spatch(insn_addr, 25, 0, offset);
 257     return 1;
 258   }
 259   static int conditionalBranch(address insn_addr, address &target) {
 260     intptr_t offset = (target - insn_addr) >> 2;
 261     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 262     return 1;
 263   }
 264   static int testAndBranch(address insn_addr, address &target) {
 265     intptr_t offset = (target - insn_addr) >> 2;
 266     Instruction_aarch64::spatch(insn_addr, 18, 5, offset);
 267     return 1;
 268   }
 269   static int loadStore(address insn_addr, address &target) {
 270     intptr_t offset = (target - insn_addr) >> 2;
 271     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 272     return 1;
 273   }
 274   static int adr(address insn_addr, address &target) {
 275 #ifdef ASSERT
 276     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 277 #endif
 278     // PC-rel. addressing
 279     ptrdiff_t offset = target - insn_addr;
 280     int offset_lo = offset & 3;
 281     offset >>= 2;
 282     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 283     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 284     return 1;
 285   }
 286   template<typename U>
 287   static int adrp(address insn_addr, address &target, U inner) {
 288     int instructions = 1;
 289 #ifdef ASSERT
 290     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 28, 24) == 0b10000, "must be");
 291 #endif
 292     ptrdiff_t offset = target - insn_addr;
 293     instructions = 2;
 294     precond(inner != nullptr);
 295     // Give the inner reloc a chance to modify the target.
 296     address adjusted_target = target;
 297     instructions = inner(insn_addr, adjusted_target);
 298     uintptr_t pc_page = (uintptr_t)insn_addr >> 12;
 299     uintptr_t adr_page = (uintptr_t)adjusted_target >> 12;
 300     offset = adr_page - pc_page;
 301     int offset_lo = offset & 3;
 302     offset >>= 2;
 303     Instruction_aarch64::spatch(insn_addr, 23, 5, offset);
 304     Instruction_aarch64::patch(insn_addr, 30, 29, offset_lo);
 305     return instructions;
 306   }
 307   static int adrpMem(address insn_addr, address &target) {
 308     uintptr_t dest = (uintptr_t)target;
 309     int offset_lo = dest & 0xfff;
 310     uint32_t insn2 = insn_at(insn_addr, 1);
 311     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 312     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo >> size);
 313     guarantee(((dest >> size) << size) == dest, "misaligned target");
 314     return 2;
 315   }
 316   static int adrpAdd(address insn_addr, address &target) {
 317     uintptr_t dest = (uintptr_t)target;
 318     int offset_lo = dest & 0xfff;
 319     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 21, 10, offset_lo);
 320     return 2;
 321   }
 322   static int adrpMovk(address insn_addr, address &target) {
 323     uintptr_t dest = uintptr_t(target);
 324     Instruction_aarch64::patch(insn_addr + sizeof (uint32_t), 20, 5, (uintptr_t)target >> 32);
 325     dest = (dest & 0xffffffffULL) | (uintptr_t(insn_addr) & 0xffff00000000ULL);
 326     target = address(dest);
 327     return 2;
 328   }
 329   static int immediate(address insn_addr, address &target) {
 330     assert(Instruction_aarch64::extract(insn_at(insn_addr, 0), 31, 21) == 0b11010010100, "must be");
 331     uint64_t dest = (uint64_t)target;
 332     // Move wide constant
 333     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 334     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 335     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 336     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 337     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 338     return 3;
 339   }
 340   static void verify(address insn_addr, address &target) {
 341 #ifdef ASSERT
 342     address address_is = MacroAssembler::target_addr_for_insn(insn_addr);
 343     if (!(address_is == target)) {
 344       tty->print_cr("%p at %p should be %p", address_is, insn_addr, target);
 345       disnm((intptr_t)insn_addr);
 346       assert(address_is == target, "should be");
 347     }
 348 #endif
 349   }
 350 };
 351 
 352 // If insn1 and insn2 use the same register to form an address, either
 353 // by an offsetted LDR or a simple ADD, return the offset. If the
 354 // second instruction is an LDR, the offset may be scaled.
 355 static bool offset_for(uint32_t insn1, uint32_t insn2, ptrdiff_t &byte_offset) {
 356   if (Instruction_aarch64::extract(insn2, 29, 24) == 0b111001 &&
 357       Instruction_aarch64::extract(insn1, 4, 0) ==
 358       Instruction_aarch64::extract(insn2, 9, 5)) {
 359     // Load/store register (unsigned immediate)
 360     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 361     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 362     byte_offset <<= size;
 363     return true;
 364   } else if (Instruction_aarch64::extract(insn2, 31, 22) == 0b1001000100 &&
 365              Instruction_aarch64::extract(insn1, 4, 0) ==
 366              Instruction_aarch64::extract(insn2, 4, 0)) {
 367     // add (immediate)
 368     byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 369     return true;
 370   }
 371   return false;
 372 }
 373 
 374 class AArch64Decoder : public AllStatic {
 375 public:
 376 
 377   static int loadStore(address insn_addr, address &target) {
 378     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 379     target = insn_addr + (offset << 2);
 380     return 1;
 381   }
 382   static int unconditionalBranch(address insn_addr, address &target) {
 383     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 25, 0);
 384     target = insn_addr + (offset << 2);
 385     return 1;
 386   }
 387   static int conditionalBranch(address insn_addr, address &target) {
 388     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 23, 5);
 389     target = address(((uint64_t)insn_addr + (offset << 2)));
 390     return 1;
 391   }
 392   static int testAndBranch(address insn_addr, address &target) {
 393     intptr_t offset = Instruction_aarch64::sextract(insn_at(insn_addr, 0), 18, 5);
 394     target = address(((uint64_t)insn_addr + (offset << 2)));
 395     return 1;
 396   }
 397   static int adr(address insn_addr, address &target) {
 398     // PC-rel. addressing
 399     uint32_t insn = insn_at(insn_addr, 0);
 400     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 401     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 402     target = address((uint64_t)insn_addr + offset);
 403     return 1;
 404   }
 405   template<typename U>
 406   static int adrp(address insn_addr, address &target, U inner) {
 407     uint32_t insn = insn_at(insn_addr, 0);
 408     assert(Instruction_aarch64::extract(insn, 28, 24) == 0b10000, "must be");
 409     intptr_t offset = Instruction_aarch64::extract(insn, 30, 29);
 410     offset |= Instruction_aarch64::sextract(insn, 23, 5) << 2;
 411     int shift = 12;
 412     offset <<= shift;
 413     uint64_t target_page = ((uint64_t)insn_addr) + offset;
 414     target_page &= ((uint64_t)-1) << shift;
 415     target = address(target_page);
 416     precond(inner != nullptr);
 417     inner(insn_addr, target);
 418     return 2;
 419   }
 420   static int adrpMem(address insn_addr, address &target) {
 421     uint32_t insn2 = insn_at(insn_addr, 1);
 422     // Load/store register (unsigned immediate)
 423     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 424     uint32_t size = Instruction_aarch64::extract(insn2, 31, 30);
 425     byte_offset <<= size;
 426     target += byte_offset;
 427     return 2;
 428   }
 429   static int adrpAdd(address insn_addr, address &target) {
 430     uint32_t insn2 = insn_at(insn_addr, 1);
 431     // add (immediate)
 432     ptrdiff_t byte_offset = Instruction_aarch64::extract(insn2, 21, 10);
 433     target += byte_offset;
 434     return 2;
 435   }
 436   static int adrpMovk(address insn_addr, address &target) {
 437     uint32_t insn2 = insn_at(insn_addr, 1);
 438     uint64_t dest = uint64_t(target);
 439     dest = (dest & 0xffff0000ffffffff) |
 440       ((uint64_t)Instruction_aarch64::extract(insn2, 20, 5) << 32);
 441     target = address(dest);
 442 
 443     // We know the destination 4k page. Maybe we have a third
 444     // instruction.
 445     uint32_t insn = insn_at(insn_addr, 0);
 446     uint32_t insn3 = insn_at(insn_addr, 2);
 447     ptrdiff_t byte_offset;
 448     if (offset_for(insn, insn3, byte_offset)) {
 449       target += byte_offset;
 450       return 3;
 451     } else {
 452       return 2;
 453     }
 454   }
 455   static int immediate(address insn_addr, address &target) {
 456     uint32_t *insns = (uint32_t *)insn_addr;
 457     assert(Instruction_aarch64::extract(insns[0], 31, 21) == 0b11010010100, "must be");
 458     // Move wide constant: movz, movk, movk.  See movptr().
 459     assert(nativeInstruction_at(insns+1)->is_movk(), "wrong insns in patch");
 460     assert(nativeInstruction_at(insns+2)->is_movk(), "wrong insns in patch");
 461     target = address(uint64_t(Instruction_aarch64::extract(insns[0], 20, 5))
 462                   + (uint64_t(Instruction_aarch64::extract(insns[1], 20, 5)) << 16)
 463                   + (uint64_t(Instruction_aarch64::extract(insns[2], 20, 5)) << 32));
 464     assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 465     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 466     return 3;
 467   }
 468   static void verify(address insn_addr, address &target) {
 469   }
 470 };
 471 
 472 address MacroAssembler::target_addr_for_insn(address insn_addr) {
 473   address target;
 474   RelocActions<AArch64Decoder>::run(insn_addr, target);
 475   return target;
 476 }
 477 
 478 // Patch any kind of instruction; there may be several instructions.
 479 // Return the total length (in bytes) of the instructions.
 480 int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
 481   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 482   return RelocActions<Patcher>::run(insn_addr, target);
 483 }
 484 
 485 int MacroAssembler::patch_oop(address insn_addr, address o) {
 486   int instructions;
 487   unsigned insn = *(unsigned*)insn_addr;
 488   assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 489 
 490   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 491 
 492   // OOPs are either narrow (32 bits) or wide (48 bits).  We encode
 493   // narrow OOPs by setting the upper 16 bits in the first
 494   // instruction.
 495   if (Instruction_aarch64::extract(insn, 31, 21) == 0b11010010101) {
 496     // Move narrow OOP
 497     uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
 498     Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 499     Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 500     instructions = 2;
 501   } else {
 502     // Move wide OOP
 503     assert(nativeInstruction_at(insn_addr+8)->is_movk(), "wrong insns in patch");
 504     uintptr_t dest = (uintptr_t)o;
 505     Instruction_aarch64::patch(insn_addr, 20, 5, dest & 0xffff);
 506     Instruction_aarch64::patch(insn_addr+4, 20, 5, (dest >>= 16) & 0xffff);
 507     Instruction_aarch64::patch(insn_addr+8, 20, 5, (dest >>= 16) & 0xffff);
 508     instructions = 3;
 509   }
 510   return instructions * NativeInstruction::instruction_size;
 511 }
 512 
 513 int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
 514   // Metadata pointers are either narrow (32 bits) or wide (48 bits).
 515   // We encode narrow ones by setting the upper 16 bits in the first
 516   // instruction.
 517   NativeInstruction *insn = nativeInstruction_at(insn_addr);
 518   assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
 519          nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
 520 
 521   MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
 522 
 523   Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
 524   Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
 525   return 2 * NativeInstruction::instruction_size;
 526 }
 527 
 528 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp) {
 529   ldr(tmp, Address(rthread, JavaThread::polling_word_offset()));
 530   if (at_return) {
 531     // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
 532     // we may safely use the sp instead to perform the stack watermark check.
 533     cmp(in_nmethod ? sp : rfp, tmp);
 534     br(Assembler::HI, slow_path);
 535   } else {
 536     tbnz(tmp, log2i_exact(SafepointMechanism::poll_bit()), slow_path);
 537   }
 538 }
 539 
 540 void MacroAssembler::rt_call(address dest, Register tmp) {
 541   CodeBlob *cb = CodeCache::find_blob(dest);
 542   if (cb) {
 543     far_call(RuntimeAddress(dest));
 544   } else {
 545     lea(tmp, RuntimeAddress(dest));
 546     blr(tmp);
 547   }
 548 }
 549 
 550 void MacroAssembler::push_cont_fastpath(Register java_thread) {
 551   if (!Continuations::enabled()) return;
 552   Label done;
 553   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 554   cmp(sp, rscratch1);
 555   br(Assembler::LS, done);
 556   mov(rscratch1, sp); // we can't use sp as the source in str
 557   str(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 558   bind(done);
 559 }
 560 
 561 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
 562   if (!Continuations::enabled()) return;
 563   Label done;
 564   ldr(rscratch1, Address(java_thread, JavaThread::cont_fastpath_offset()));
 565   cmp(sp, rscratch1);
 566   br(Assembler::LO, done);
 567   str(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
 568   bind(done);
 569 }
 570 
 571 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 572   // we must set sp to zero to clear frame
 573   str(zr, Address(rthread, JavaThread::last_Java_sp_offset()));
 574 
 575   // must clear fp, so that compiled frames are not confused; it is
 576   // possible that we need it only for debugging
 577   if (clear_fp) {
 578     str(zr, Address(rthread, JavaThread::last_Java_fp_offset()));
 579   }
 580 
 581   // Always clear the pc because it could have been set by make_walkable()
 582   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
 583 }
 584 
 585 // Calls to C land
 586 //
 587 // When entering C land, the rfp, & resp of the last Java frame have to be recorded
 588 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
 589 // has to be reset to 0. This is required to allow proper stack traversal.
 590 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 591                                          Register last_java_fp,
 592                                          Register last_java_pc,
 593                                          Register scratch) {
 594 
 595   if (last_java_pc->is_valid()) {
 596       str(last_java_pc, Address(rthread,
 597                                 JavaThread::frame_anchor_offset()
 598                                 + JavaFrameAnchor::last_Java_pc_offset()));
 599     }
 600 
 601   // determine last_java_sp register
 602   if (last_java_sp == sp) {
 603     mov(scratch, sp);
 604     last_java_sp = scratch;
 605   } else if (!last_java_sp->is_valid()) {
 606     last_java_sp = esp;
 607   }
 608 
 609   // last_java_fp is optional
 610   if (last_java_fp->is_valid()) {
 611     str(last_java_fp, Address(rthread, JavaThread::last_Java_fp_offset()));
 612   }
 613 
 614   // We must set sp last.
 615   str(last_java_sp, Address(rthread, JavaThread::last_Java_sp_offset()));
 616 }
 617 
 618 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 619                                          Register last_java_fp,
 620                                          address  last_java_pc,
 621                                          Register scratch) {
 622   assert(last_java_pc != nullptr, "must provide a valid PC");
 623 
 624   adr(scratch, last_java_pc);
 625   str(scratch, Address(rthread,
 626                        JavaThread::frame_anchor_offset()
 627                        + JavaFrameAnchor::last_Java_pc_offset()));
 628 
 629   set_last_Java_frame(last_java_sp, last_java_fp, noreg, scratch);
 630 }
 631 
 632 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
 633                                          Register last_java_fp,
 634                                          Label &L,
 635                                          Register scratch) {
 636   if (L.is_bound()) {
 637     set_last_Java_frame(last_java_sp, last_java_fp, target(L), scratch);
 638   } else {
 639     InstructionMark im(this);
 640     L.add_patch_at(code(), locator());
 641     set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, scratch);
 642   }
 643 }
 644 
 645 static inline bool target_needs_far_branch(address addr) {
 646   if (AOTCodeCache::is_on_for_dump()) {
 647     return true;
 648   }
 649   // codecache size <= 128M
 650   if (!MacroAssembler::far_branches()) {
 651     return false;
 652   }
 653   // codecache size > 240M
 654   if (MacroAssembler::codestub_branch_needs_far_jump()) {
 655     return true;
 656   }
 657   // codecache size: 128M..240M
 658   return !CodeCache::is_non_nmethod(addr);
 659 }
 660 
 661 void MacroAssembler::far_call(Address entry, Register tmp) {
 662   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 663   assert(CodeCache::find_blob(entry.target()) != nullptr,
 664          "destination of far call not found in code cache");
 665   assert(entry.rspec().type() == relocInfo::external_word_type
 666          || entry.rspec().type() == relocInfo::runtime_call_type
 667          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 668   if (target_needs_far_branch(entry.target())) {
 669     uint64_t offset;
 670     // We can use ADRP here because we know that the total size of
 671     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 672     adrp(tmp, entry, offset);
 673     add(tmp, tmp, offset);
 674     blr(tmp);
 675   } else {
 676     bl(entry);
 677   }
 678 }
 679 
 680 int MacroAssembler::far_jump(Address entry, Register tmp) {
 681   assert(ReservedCodeCacheSize < 4*G, "branch out of range");
 682   assert(CodeCache::find_blob(entry.target()) != nullptr,
 683          "destination of far call not found in code cache");
 684   assert(entry.rspec().type() == relocInfo::external_word_type
 685          || entry.rspec().type() == relocInfo::runtime_call_type
 686          || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
 687   address start = pc();
 688   if (target_needs_far_branch(entry.target())) {
 689     uint64_t offset;
 690     // We can use ADRP here because we know that the total size of
 691     // the code cache cannot exceed 2Gb (ADRP limit is 4GB).
 692     adrp(tmp, entry, offset);
 693     add(tmp, tmp, offset);
 694     br(tmp);
 695   } else {
 696     b(entry);
 697   }
 698   return pc() - start;
 699 }
 700 
 701 void MacroAssembler::reserved_stack_check() {
 702     // testing if reserved zone needs to be enabled
 703     Label no_reserved_zone_enabling;
 704 
 705     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 706     cmp(sp, rscratch1);
 707     br(Assembler::LO, no_reserved_zone_enabling);
 708 
 709     enter();   // LR and FP are live.
 710     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone)));
 711     mov(c_rarg0, rthread);
 712     blr(rscratch1);
 713     leave();
 714 
 715     // We have already removed our own frame.
 716     // throw_delayed_StackOverflowError will think that it's been
 717     // called by our caller.
 718     lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
 719     br(rscratch1);
 720     should_not_reach_here();
 721 
 722     bind(no_reserved_zone_enabling);
 723 }
 724 
 725 static void pass_arg0(MacroAssembler* masm, Register arg) {
 726   if (c_rarg0 != arg ) {
 727     masm->mov(c_rarg0, arg);
 728   }
 729 }
 730 
 731 static void pass_arg1(MacroAssembler* masm, Register arg) {
 732   if (c_rarg1 != arg ) {
 733     masm->mov(c_rarg1, arg);
 734   }
 735 }
 736 
 737 static void pass_arg2(MacroAssembler* masm, Register arg) {
 738   if (c_rarg2 != arg ) {
 739     masm->mov(c_rarg2, arg);
 740   }
 741 }
 742 
 743 static void pass_arg3(MacroAssembler* masm, Register arg) {
 744   if (c_rarg3 != arg ) {
 745     masm->mov(c_rarg3, arg);
 746   }
 747 }
 748 
 749 void MacroAssembler::call_VM_base(Register oop_result,
 750                                   Register java_thread,
 751                                   Register last_java_sp,
 752                                   Label*   return_pc,
 753                                   address  entry_point,
 754                                   int      number_of_arguments,
 755                                   bool     check_exceptions) {
 756    // determine java_thread register
 757   if (!java_thread->is_valid()) {
 758     java_thread = rthread;
 759   }
 760 
 761   // determine last_java_sp register
 762   if (!last_java_sp->is_valid()) {
 763     last_java_sp = esp;
 764   }
 765 
 766   // debugging support
 767   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
 768   assert(java_thread == rthread, "unexpected register");
 769 #ifdef ASSERT
 770   // TraceBytecodes does not use r12 but saves it over the call, so don't verify
 771   // if ((UseCompressedOops || UseCompressedClassPointers) && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
 772 #endif // ASSERT
 773 
 774   assert(java_thread != oop_result  , "cannot use the same register for java_thread & oop_result");
 775   assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
 776 
 777   // push java thread (becomes first argument of C function)
 778 
 779   mov(c_rarg0, java_thread);
 780 
 781   // set last Java frame before call
 782   assert(last_java_sp != rfp, "can't use rfp");
 783 
 784   Label l;
 785   set_last_Java_frame(last_java_sp, rfp, return_pc != nullptr ? *return_pc : l, rscratch1);
 786 
 787   // do the call, remove parameters
 788   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
 789 
 790   // lr could be poisoned with PAC signature during throw_pending_exception
 791   // if it was tail-call optimized by compiler, since lr is not callee-saved
 792   // reload it with proper value
 793   adr(lr, l);
 794 
 795   // reset last Java frame
 796   // Only interpreter should have to clear fp
 797   reset_last_Java_frame(true);
 798 
 799    // C++ interp handles this in the interpreter
 800   check_and_handle_popframe(java_thread);
 801   check_and_handle_earlyret(java_thread);
 802 
 803   if (check_exceptions) {
 804     // check for pending exceptions (java_thread is set upon return)
 805     ldr(rscratch1, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
 806     Label ok;
 807     cbz(rscratch1, ok);
 808     lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
 809     br(rscratch1);
 810     bind(ok);
 811   }
 812 
 813   // get oop result if there is one and reset the value in the thread
 814   if (oop_result->is_valid()) {
 815     get_vm_result_oop(oop_result, java_thread);
 816   }
 817 }
 818 
 819 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
 820   call_VM_base(oop_result, noreg, noreg, nullptr, entry_point, number_of_arguments, check_exceptions);
 821 }
 822 
 823 // Check the entry target is always reachable from any branch.
 824 static bool is_always_within_branch_range(Address entry) {
 825   if (AOTCodeCache::is_on_for_dump()) {
 826     return false;
 827   }
 828   const address target = entry.target();
 829 
 830   if (!CodeCache::contains(target)) {
 831     // We always use trampolines for callees outside CodeCache.
 832     assert(entry.rspec().type() == relocInfo::runtime_call_type, "non-runtime call of an external target");
 833     return false;
 834   }
 835 
 836   if (!MacroAssembler::far_branches()) {
 837     return true;
 838   }
 839 
 840   if (entry.rspec().type() == relocInfo::runtime_call_type) {
 841     // Runtime calls are calls of a non-compiled method (stubs, adapters).
 842     // Non-compiled methods stay forever in CodeCache.
 843     // We check whether the longest possible branch is within the branch range.
 844     assert(CodeCache::find_blob(target) != nullptr &&
 845           !CodeCache::find_blob(target)->is_nmethod(),
 846           "runtime call of compiled method");
 847     const address right_longest_branch_start = CodeCache::high_bound() - NativeInstruction::instruction_size;
 848     const address left_longest_branch_start = CodeCache::low_bound();
 849     const bool is_reachable = Assembler::reachable_from_branch_at(left_longest_branch_start, target) &&
 850                               Assembler::reachable_from_branch_at(right_longest_branch_start, target);
 851     return is_reachable;
 852   }
 853 
 854   return false;
 855 }
 856 
 857 // Maybe emit a call via a trampoline. If the code cache is small
 858 // trampolines won't be emitted.
 859 address MacroAssembler::trampoline_call(Address entry) {
 860   assert(entry.rspec().type() == relocInfo::runtime_call_type
 861          || entry.rspec().type() == relocInfo::opt_virtual_call_type
 862          || entry.rspec().type() == relocInfo::static_call_type
 863          || entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
 864 
 865   address target = entry.target();
 866 
 867   if (!is_always_within_branch_range(entry)) {
 868     if (!in_scratch_emit_size()) {
 869       // We don't want to emit a trampoline if C2 is generating dummy
 870       // code during its branch shortening phase.
 871       if (entry.rspec().type() == relocInfo::runtime_call_type) {
 872         assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
 873         code()->share_trampoline_for(entry.target(), offset());
 874       } else {
 875         address stub = emit_trampoline_stub(offset(), target);
 876         if (stub == nullptr) {
 877           postcond(pc() == badAddress);
 878           return nullptr; // CodeCache is full
 879         }
 880       }
 881     }
 882     target = pc();
 883   }
 884 
 885   address call_pc = pc();
 886   relocate(entry.rspec());
 887   bl(target);
 888 
 889   postcond(pc() != badAddress);
 890   return call_pc;
 891 }
 892 
 893 // Emit a trampoline stub for a call to a target which is too far away.
 894 //
 895 // code sequences:
 896 //
 897 // call-site:
 898 //   branch-and-link to <destination> or <trampoline stub>
 899 //
 900 // Related trampoline stub for this call site in the stub section:
 901 //   load the call target from the constant pool
 902 //   branch (LR still points to the call site above)
 903 
 904 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
 905                                              address dest) {
 906   // Max stub size: alignment nop, TrampolineStub.
 907   address stub = start_a_stub(max_trampoline_stub_size());
 908   if (stub == nullptr) {
 909     return nullptr;  // CodeBuffer::expand failed
 910   }
 911 
 912   // Create a trampoline stub relocation which relates this trampoline stub
 913   // with the call instruction at insts_call_instruction_offset in the
 914   // instructions code-section.
 915   align(wordSize);
 916   relocate(trampoline_stub_Relocation::spec(code()->insts()->start()
 917                                             + insts_call_instruction_offset));
 918   const int stub_start_offset = offset();
 919 
 920   // Now, create the trampoline stub's code:
 921   // - load the call
 922   // - call
 923   Label target;
 924   ldr(rscratch1, target);
 925   br(rscratch1);
 926   bind(target);
 927   assert(offset() - stub_start_offset == NativeCallTrampolineStub::data_offset,
 928          "should be");
 929   emit_int64((int64_t)dest);
 930 
 931   const address stub_start_addr = addr_at(stub_start_offset);
 932 
 933   assert(is_NativeCallTrampolineStub_at(stub_start_addr), "doesn't look like a trampoline");
 934 
 935   end_a_stub();
 936   return stub_start_addr;
 937 }
 938 
 939 int MacroAssembler::max_trampoline_stub_size() {
 940   // Max stub size: alignment nop, TrampolineStub.
 941   return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
 942 }
 943 
 944 void MacroAssembler::emit_static_call_stub() {
 945   // CompiledDirectCall::set_to_interpreted knows the
 946   // exact layout of this stub.
 947 
 948   isb();
 949   mov_metadata(rmethod, nullptr);
 950 
 951   // Jump to the entry point of the c2i stub.
 952   if (codestub_branch_needs_far_jump()) {
 953     movptr(rscratch1, 0);
 954     br(rscratch1);
 955   } else {
 956     b(pc());
 957   }
 958 }
 959 
 960 int MacroAssembler::static_call_stub_size() {
 961   if (!codestub_branch_needs_far_jump()) {
 962     // isb; movk; movz; movz; b
 963     return 5 * NativeInstruction::instruction_size;
 964   }
 965   // isb; movk; movz; movz; movk; movz; movz; br
 966   return 8 * NativeInstruction::instruction_size;
 967 }
 968 
 969 void MacroAssembler::c2bool(Register x) {
 970   // implements x == 0 ? 0 : 1
 971   // note: must only look at least-significant byte of x
 972   //       since C-style booleans are stored in one byte
 973   //       only! (was bug)
 974   tst(x, 0xff);
 975   cset(x, Assembler::NE);
 976 }
 977 
 978 address MacroAssembler::ic_call(address entry, jint method_index) {
 979   RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
 980   movptr(rscratch2, (intptr_t)Universe::non_oop_word());
 981   return trampoline_call(Address(entry, rh));
 982 }
 983 
 984 int MacroAssembler::ic_check_size() {
 985   int extra_instructions = UseCompactObjectHeaders ? 1 : 0;
 986   if (target_needs_far_branch(CAST_FROM_FN_PTR(address, SharedRuntime::get_ic_miss_stub()))) {
 987     return NativeInstruction::instruction_size * (7 + extra_instructions);
 988   } else {
 989     return NativeInstruction::instruction_size * (5 + extra_instructions);
 990   }
 991 }
 992 
 993 int MacroAssembler::ic_check(int end_alignment) {
 994   Register receiver = j_rarg0;
 995   Register data = rscratch2;
 996   Register tmp1 = rscratch1;
 997   Register tmp2 = r10;
 998 
 999   // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
1000   // before the inline cache check, so we don't have to execute any nop instructions when dispatching
1001   // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
1002   // before the inline cache check here, and not after
1003   align(end_alignment, offset() + ic_check_size());
1004 
1005   int uep_offset = offset();
1006 
1007   if (UseCompactObjectHeaders) {
1008     load_narrow_klass_compact(tmp1, receiver);
1009     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1010     cmpw(tmp1, tmp2);
1011   } else if (UseCompressedClassPointers) {
1012     ldrw(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1013     ldrw(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1014     cmpw(tmp1, tmp2);
1015   } else {
1016     ldr(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
1017     ldr(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
1018     cmp(tmp1, tmp2);
1019   }
1020 
1021   Label dont;
1022   br(Assembler::EQ, dont);
1023   far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1024   bind(dont);
1025   assert((offset() % end_alignment) == 0, "Misaligned verified entry point");
1026 
1027   return uep_offset;
1028 }
1029 
1030 // Implementation of call_VM versions
1031 
1032 void MacroAssembler::call_VM(Register oop_result,
1033                              address entry_point,
1034                              bool check_exceptions) {
1035   call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1036 }
1037 
1038 void MacroAssembler::call_VM(Register oop_result,
1039                              address entry_point,
1040                              Register arg_1,
1041                              bool check_exceptions) {
1042   pass_arg1(this, arg_1);
1043   call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1044 }
1045 
1046 void MacroAssembler::call_VM(Register oop_result,
1047                              address entry_point,
1048                              Register arg_1,
1049                              Register arg_2,
1050                              bool check_exceptions) {
1051   assert_different_registers(arg_1, c_rarg2);
1052   pass_arg2(this, arg_2);
1053   pass_arg1(this, arg_1);
1054   call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1055 }
1056 
1057 void MacroAssembler::call_VM(Register oop_result,
1058                              address entry_point,
1059                              Register arg_1,
1060                              Register arg_2,
1061                              Register arg_3,
1062                              bool check_exceptions) {
1063   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1064   assert_different_registers(arg_2, c_rarg3);
1065   pass_arg3(this, arg_3);
1066 
1067   pass_arg2(this, arg_2);
1068 
1069   pass_arg1(this, arg_1);
1070   call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1071 }
1072 
1073 void MacroAssembler::call_VM(Register oop_result,
1074                              Register last_java_sp,
1075                              address entry_point,
1076                              int number_of_arguments,
1077                              bool check_exceptions) {
1078   call_VM_base(oop_result, rthread, last_java_sp, nullptr, entry_point, number_of_arguments, check_exceptions);
1079 }
1080 
1081 void MacroAssembler::call_VM(Register oop_result,
1082                              Register last_java_sp,
1083                              address entry_point,
1084                              Register arg_1,
1085                              bool check_exceptions) {
1086   pass_arg1(this, arg_1);
1087   call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1088 }
1089 
1090 void MacroAssembler::call_VM(Register oop_result,
1091                              Register last_java_sp,
1092                              address entry_point,
1093                              Register arg_1,
1094                              Register arg_2,
1095                              bool check_exceptions) {
1096 
1097   assert_different_registers(arg_1, c_rarg2);
1098   pass_arg2(this, arg_2);
1099   pass_arg1(this, arg_1);
1100   call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1101 }
1102 
1103 void MacroAssembler::call_VM(Register oop_result,
1104                              Register last_java_sp,
1105                              address entry_point,
1106                              Register arg_1,
1107                              Register arg_2,
1108                              Register arg_3,
1109                              bool check_exceptions) {
1110   assert_different_registers(arg_1, c_rarg2, c_rarg3);
1111   assert_different_registers(arg_2, c_rarg3);
1112   pass_arg3(this, arg_3);
1113   pass_arg2(this, arg_2);
1114   pass_arg1(this, arg_1);
1115   call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1116 }
1117 
1118 
1119 void MacroAssembler::get_vm_result_oop(Register oop_result, Register java_thread) {
1120   ldr(oop_result, Address(java_thread, JavaThread::vm_result_oop_offset()));
1121   str(zr, Address(java_thread, JavaThread::vm_result_oop_offset()));
1122   verify_oop_msg(oop_result, "broken oop in call_VM_base");
1123 }
1124 
1125 void MacroAssembler::get_vm_result_metadata(Register metadata_result, Register java_thread) {
1126   ldr(metadata_result, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1127   str(zr, Address(java_thread, JavaThread::vm_result_metadata_offset()));
1128 }
1129 
1130 void MacroAssembler::align(int modulus) {
1131   align(modulus, offset());
1132 }
1133 
1134 // Ensure that the code at target bytes offset from the current offset() is aligned
1135 // according to modulus.
1136 void MacroAssembler::align(int modulus, int target) {
1137   int delta = target - offset();
1138   while ((offset() + delta) % modulus != 0) nop();
1139 }
1140 
1141 void MacroAssembler::post_call_nop() {
1142   if (!Continuations::enabled()) {
1143     return;
1144   }
1145   InstructionMark im(this);
1146   relocate(post_call_nop_Relocation::spec());
1147   InlineSkippedInstructionsCounter skipCounter(this);
1148   nop();
1149   movk(zr, 0);
1150   movk(zr, 0);
1151 }
1152 
1153 // these are no-ops overridden by InterpreterMacroAssembler
1154 
1155 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { }
1156 
1157 void MacroAssembler::check_and_handle_popframe(Register java_thread) { }
1158 
1159 // Look up the method for a megamorphic invokeinterface call.
1160 // The target method is determined by <intf_klass, itable_index>.
1161 // The receiver klass is in recv_klass.
1162 // On success, the result will be in method_result, and execution falls through.
1163 // On failure, execution transfers to the given label.
1164 void MacroAssembler::lookup_interface_method(Register recv_klass,
1165                                              Register intf_klass,
1166                                              RegisterOrConstant itable_index,
1167                                              Register method_result,
1168                                              Register scan_temp,
1169                                              Label& L_no_such_interface,
1170                          bool return_method) {
1171   assert_different_registers(recv_klass, intf_klass, scan_temp);
1172   assert_different_registers(method_result, intf_klass, scan_temp);
1173   assert(recv_klass != method_result || !return_method,
1174      "recv_klass can be destroyed when method isn't needed");
1175   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
1176          "caller must use same register for non-constant itable index as for method");
1177 
1178   // Compute start of first itableOffsetEntry (which is at the end of the vtable)
1179   int vtable_base = in_bytes(Klass::vtable_start_offset());
1180   int itentry_off = in_bytes(itableMethodEntry::method_offset());
1181   int scan_step   = itableOffsetEntry::size() * wordSize;
1182   int vte_size    = vtableEntry::size_in_bytes();
1183   assert(vte_size == wordSize, "else adjust times_vte_scale");
1184 
1185   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1186 
1187   // Could store the aligned, prescaled offset in the klass.
1188   // lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
1189   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(3)));
1190   add(scan_temp, scan_temp, vtable_base);
1191 
1192   if (return_method) {
1193     // Adjust recv_klass by scaled itable_index, so we can free itable_index.
1194     assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1195     // lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
1196     lea(recv_klass, Address(recv_klass, itable_index, Address::lsl(3)));
1197     if (itentry_off)
1198       add(recv_klass, recv_klass, itentry_off);
1199   }
1200 
1201   // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
1202   //   if (scan->interface() == intf) {
1203   //     result = (klass + scan->offset() + itable_index);
1204   //   }
1205   // }
1206   Label search, found_method;
1207 
1208   ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1209   cmp(intf_klass, method_result);
1210   br(Assembler::EQ, found_method);
1211   bind(search);
1212   // Check that the previous entry is non-null.  A null entry means that
1213   // the receiver class doesn't implement the interface, and wasn't the
1214   // same as when the caller was compiled.
1215   cbz(method_result, L_no_such_interface);
1216   if (itableOffsetEntry::interface_offset() != 0) {
1217     add(scan_temp, scan_temp, scan_step);
1218     ldr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
1219   } else {
1220     ldr(method_result, Address(pre(scan_temp, scan_step)));
1221   }
1222   cmp(intf_klass, method_result);
1223   br(Assembler::NE, search);
1224 
1225   bind(found_method);
1226 
1227   // Got a hit.
1228   if (return_method) {
1229     ldrw(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
1230     ldr(method_result, Address(recv_klass, scan_temp, Address::uxtw(0)));
1231   }
1232 }
1233 
1234 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
1235 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
1236 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
1237 // The target method is determined by <holder_klass, itable_index>.
1238 // The receiver klass is in recv_klass.
1239 // On success, the result will be in method_result, and execution falls through.
1240 // On failure, execution transfers to the given label.
1241 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
1242                                                   Register holder_klass,
1243                                                   Register resolved_klass,
1244                                                   Register method_result,
1245                                                   Register temp_itbl_klass,
1246                                                   Register scan_temp,
1247                                                   int itable_index,
1248                                                   Label& L_no_such_interface) {
1249   // 'method_result' is only used as output register at the very end of this method.
1250   // Until then we can reuse it as 'holder_offset'.
1251   Register holder_offset = method_result;
1252   assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
1253 
1254   int vtable_start_offset = in_bytes(Klass::vtable_start_offset());
1255   int itable_offset_entry_size = itableOffsetEntry::size() * wordSize;
1256   int ioffset = in_bytes(itableOffsetEntry::interface_offset());
1257   int ooffset = in_bytes(itableOffsetEntry::offset_offset());
1258 
1259   Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
1260 
1261   ldrw(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
1262   add(recv_klass, recv_klass, vtable_start_offset + ioffset);
1263   // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset() + sizeof(vtableEntry) * recv_klass->_vtable_len;
1264   // temp_itbl_klass = itable[0]._interface;
1265   int vtblEntrySize = vtableEntry::size_in_bytes();
1266   assert(vtblEntrySize == wordSize, "ldr lsl shift amount must be 3");
1267   ldr(temp_itbl_klass, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1268   mov(holder_offset, zr);
1269   // scan_temp = &(itable[0]._interface)
1270   lea(scan_temp, Address(recv_klass, scan_temp, Address::lsl(exact_log2(vtblEntrySize))));
1271 
1272   // Initial checks:
1273   //   - if (holder_klass != resolved_klass), go to "scan for resolved"
1274   //   - if (itable[0] == holder_klass), shortcut to "holder found"
1275   //   - if (itable[0] == 0), no such interface
1276   cmp(resolved_klass, holder_klass);
1277   br(Assembler::NE, L_loop_search_resolved_entry);
1278   cmp(holder_klass, temp_itbl_klass);
1279   br(Assembler::EQ, L_holder_found);
1280   cbz(temp_itbl_klass, L_no_such_interface);
1281 
1282   // Loop: Look for holder_klass record in itable
1283   //   do {
1284   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1285   //     if (temp_itbl_klass == holder_klass) {
1286   //       goto L_holder_found; // Found!
1287   //     }
1288   //   } while (temp_itbl_klass != 0);
1289   //   goto L_no_such_interface // Not found.
1290   Label L_search_holder;
1291   bind(L_search_holder);
1292     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1293     cmp(holder_klass, temp_itbl_klass);
1294     br(Assembler::EQ, L_holder_found);
1295     cbnz(temp_itbl_klass, L_search_holder);
1296 
1297   b(L_no_such_interface);
1298 
1299   // Loop: Look for resolved_class record in itable
1300   //   while (true) {
1301   //     temp_itbl_klass = *(scan_temp += itable_offset_entry_size);
1302   //     if (temp_itbl_klass == 0) {
1303   //       goto L_no_such_interface;
1304   //     }
1305   //     if (temp_itbl_klass == resolved_klass) {
1306   //        goto L_resolved_found;  // Found!
1307   //     }
1308   //     if (temp_itbl_klass == holder_klass) {
1309   //        holder_offset = scan_temp;
1310   //     }
1311   //   }
1312   //
1313   Label L_loop_search_resolved;
1314   bind(L_loop_search_resolved);
1315     ldr(temp_itbl_klass, Address(pre(scan_temp, itable_offset_entry_size)));
1316   bind(L_loop_search_resolved_entry);
1317     cbz(temp_itbl_klass, L_no_such_interface);
1318     cmp(resolved_klass, temp_itbl_klass);
1319     br(Assembler::EQ, L_resolved_found);
1320     cmp(holder_klass, temp_itbl_klass);
1321     br(Assembler::NE, L_loop_search_resolved);
1322     mov(holder_offset, scan_temp);
1323     b(L_loop_search_resolved);
1324 
1325   // See if we already have a holder klass. If not, go and scan for it.
1326   bind(L_resolved_found);
1327   cbz(holder_offset, L_search_holder);
1328   mov(scan_temp, holder_offset);
1329 
1330   // Finally, scan_temp contains holder_klass vtable offset
1331   bind(L_holder_found);
1332   ldrw(method_result, Address(scan_temp, ooffset - ioffset));
1333   add(recv_klass, recv_klass, itable_index * wordSize + in_bytes(itableMethodEntry::method_offset())
1334     - vtable_start_offset - ioffset); // substract offsets to restore the original value of recv_klass
1335   ldr(method_result, Address(recv_klass, method_result, Address::uxtw(0)));
1336 }
1337 
1338 // virtual method calling
1339 void MacroAssembler::lookup_virtual_method(Register recv_klass,
1340                                            RegisterOrConstant vtable_index,
1341                                            Register method_result) {
1342   assert(vtableEntry::size() * wordSize == 8,
1343          "adjust the scaling in the code below");
1344   int64_t vtable_offset_in_bytes = in_bytes(Klass::vtable_start_offset() + vtableEntry::method_offset());
1345 
1346   if (vtable_index.is_register()) {
1347     lea(method_result, Address(recv_klass,
1348                                vtable_index.as_register(),
1349                                Address::lsl(LogBytesPerWord)));
1350     ldr(method_result, Address(method_result, vtable_offset_in_bytes));
1351   } else {
1352     vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
1353     ldr(method_result,
1354         form_address(rscratch1, recv_klass, vtable_offset_in_bytes, 0));
1355   }
1356 }
1357 
1358 void MacroAssembler::check_klass_subtype(Register sub_klass,
1359                            Register super_klass,
1360                            Register temp_reg,
1361                            Label& L_success) {
1362   Label L_failure;
1363   check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg,        &L_success, &L_failure, nullptr);
1364   check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
1365   bind(L_failure);
1366 }
1367 
1368 
1369 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
1370                                                    Register super_klass,
1371                                                    Register temp_reg,
1372                                                    Label* L_success,
1373                                                    Label* L_failure,
1374                                                    Label* L_slow_path,
1375                                                    Register super_check_offset) {
1376   assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset);
1377   bool must_load_sco = ! super_check_offset->is_valid();
1378   if (must_load_sco) {
1379     assert(temp_reg != noreg, "supply either a temp or a register offset");
1380   }
1381 
1382   Label L_fallthrough;
1383   int label_nulls = 0;
1384   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1385   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1386   if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
1387   assert(label_nulls <= 1, "at most one null in the batch");
1388 
1389   int sco_offset = in_bytes(Klass::super_check_offset_offset());
1390   Address super_check_offset_addr(super_klass, sco_offset);
1391 
1392   // Hacked jmp, which may only be used just before L_fallthrough.
1393 #define final_jmp(label)                                                \
1394   if (&(label) == &L_fallthrough) { /*do nothing*/ }                    \
1395   else                            b(label)                /*omit semi*/
1396 
1397   // If the pointers are equal, we are done (e.g., String[] elements).
1398   // This self-check enables sharing of secondary supertype arrays among
1399   // non-primary types such as array-of-interface.  Otherwise, each such
1400   // type would need its own customized SSA.
1401   // We move this check to the front of the fast path because many
1402   // type checks are in fact trivially successful in this manner,
1403   // so we get a nicely predicted branch right at the start of the check.
1404   cmp(sub_klass, super_klass);
1405   br(Assembler::EQ, *L_success);
1406 
1407   // Check the supertype display:
1408   if (must_load_sco) {
1409     ldrw(temp_reg, super_check_offset_addr);
1410     super_check_offset = temp_reg;
1411   }
1412 
1413   Address super_check_addr(sub_klass, super_check_offset);
1414   ldr(rscratch1, super_check_addr);
1415   cmp(super_klass, rscratch1); // load displayed supertype
1416   br(Assembler::EQ, *L_success);
1417 
1418   // This check has worked decisively for primary supers.
1419   // Secondary supers are sought in the super_cache ('super_cache_addr').
1420   // (Secondary supers are interfaces and very deeply nested subtypes.)
1421   // This works in the same check above because of a tricky aliasing
1422   // between the super_cache and the primary super display elements.
1423   // (The 'super_check_addr' can address either, as the case requires.)
1424   // Note that the cache is updated below if it does not help us find
1425   // what we need immediately.
1426   // So if it was a primary super, we can just fail immediately.
1427   // Otherwise, it's the slow path for us (no success at this point).
1428 
1429   sub(rscratch1, super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
1430   if (L_failure == &L_fallthrough) {
1431     cbz(rscratch1, *L_slow_path);
1432   } else {
1433     cbnz(rscratch1, *L_failure);
1434     final_jmp(*L_slow_path);
1435   }
1436 
1437   bind(L_fallthrough);
1438 
1439 #undef final_jmp
1440 }
1441 
1442 // These two are taken from x86, but they look generally useful
1443 
1444 // scans count pointer sized words at [addr] for occurrence of value,
1445 // generic
1446 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
1447                                 Register scratch) {
1448   Label Lloop, Lexit;
1449   cbz(count, Lexit);
1450   bind(Lloop);
1451   ldr(scratch, post(addr, wordSize));
1452   cmp(value, scratch);
1453   br(EQ, Lexit);
1454   sub(count, count, 1);
1455   cbnz(count, Lloop);
1456   bind(Lexit);
1457 }
1458 
1459 // scans count 4 byte words at [addr] for occurrence of value,
1460 // generic
1461 void MacroAssembler::repne_scanw(Register addr, Register value, Register count,
1462                                 Register scratch) {
1463   Label Lloop, Lexit;
1464   cbz(count, Lexit);
1465   bind(Lloop);
1466   ldrw(scratch, post(addr, wordSize));
1467   cmpw(value, scratch);
1468   br(EQ, Lexit);
1469   sub(count, count, 1);
1470   cbnz(count, Lloop);
1471   bind(Lexit);
1472 }
1473 
1474 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
1475                                                           Register super_klass,
1476                                                           Register temp_reg,
1477                                                           Register temp2_reg,
1478                                                           Label* L_success,
1479                                                           Label* L_failure,
1480                                                           bool set_cond_codes) {
1481   // NB! Callers may assume that, when temp2_reg is a valid register,
1482   // this code sets it to a nonzero value.
1483 
1484   assert_different_registers(sub_klass, super_klass, temp_reg);
1485   if (temp2_reg != noreg)
1486     assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1487 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
1488 
1489   Label L_fallthrough;
1490   int label_nulls = 0;
1491   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1492   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1493   assert(label_nulls <= 1, "at most one null in the batch");
1494 
1495   // a couple of useful fields in sub_klass:
1496   int ss_offset = in_bytes(Klass::secondary_supers_offset());
1497   int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1498   Address secondary_supers_addr(sub_klass, ss_offset);
1499   Address super_cache_addr(     sub_klass, sc_offset);
1500 
1501   BLOCK_COMMENT("check_klass_subtype_slow_path");
1502 
1503   // Do a linear scan of the secondary super-klass chain.
1504   // This code is rarely used, so simplicity is a virtue here.
1505   // The repne_scan instruction uses fixed registers, which we must spill.
1506   // Don't worry too much about pre-existing connections with the input regs.
1507 
1508   assert(sub_klass != r0, "killed reg"); // killed by mov(r0, super)
1509   assert(sub_klass != r2, "killed reg"); // killed by lea(r2, &pst_counter)
1510 
1511   RegSet pushed_registers;
1512   if (!IS_A_TEMP(r2))    pushed_registers += r2;
1513   if (!IS_A_TEMP(r5))    pushed_registers += r5;
1514 
1515   if (super_klass != r0) {
1516     if (!IS_A_TEMP(r0))   pushed_registers += r0;
1517   }
1518 
1519   push(pushed_registers, sp);
1520 
1521   // Get super_klass value into r0 (even if it was in r5 or r2).
1522   if (super_klass != r0) {
1523     mov(r0, super_klass);
1524   }
1525 
1526 #ifndef PRODUCT
1527   incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
1528 #endif //PRODUCT
1529 
1530   // We will consult the secondary-super array.
1531   ldr(r5, secondary_supers_addr);
1532   // Load the array length.
1533   ldrw(r2, Address(r5, Array<Klass*>::length_offset_in_bytes()));
1534   // Skip to start of data.
1535   add(r5, r5, Array<Klass*>::base_offset_in_bytes());
1536 
1537   cmp(sp, zr); // Clear Z flag; SP is never zero
1538   // Scan R2 words at [R5] for an occurrence of R0.
1539   // Set NZ/Z based on last compare.
1540   repne_scan(r5, r0, r2, rscratch1);
1541 
1542   // Unspill the temp. registers:
1543   pop(pushed_registers, sp);
1544 
1545   br(Assembler::NE, *L_failure);
1546 
1547   // Success.  Cache the super we found and proceed in triumph.
1548 
1549   if (UseSecondarySupersCache) {
1550     str(super_klass, super_cache_addr);
1551   }
1552 
1553   if (L_success != &L_fallthrough) {
1554     b(*L_success);
1555   }
1556 
1557 #undef IS_A_TEMP
1558 
1559   bind(L_fallthrough);
1560 }
1561 
1562 // If Register r is invalid, remove a new register from
1563 // available_regs, and add new register to regs_to_push.
1564 Register MacroAssembler::allocate_if_noreg(Register r,
1565                                   RegSetIterator<Register> &available_regs,
1566                                   RegSet &regs_to_push) {
1567   if (!r->is_valid()) {
1568     r = *available_regs++;
1569     regs_to_push += r;
1570   }
1571   return r;
1572 }
1573 
1574 // check_klass_subtype_slow_path_table() looks for super_klass in the
1575 // hash table belonging to super_klass, branching to L_success or
1576 // L_failure as appropriate. This is essentially a shim which
1577 // allocates registers as necessary then calls
1578 // lookup_secondary_supers_table() to do the work. Any of the temp
1579 // regs may be noreg, in which case this logic will chooses some
1580 // registers push and pop them from the stack.
1581 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
1582                                                          Register super_klass,
1583                                                          Register temp_reg,
1584                                                          Register temp2_reg,
1585                                                          Register temp3_reg,
1586                                                          Register result_reg,
1587                                                          FloatRegister vtemp,
1588                                                          Label* L_success,
1589                                                          Label* L_failure,
1590                                                          bool set_cond_codes) {
1591   RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
1592 
1593   assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, rscratch1);
1594 
1595   Label L_fallthrough;
1596   int label_nulls = 0;
1597   if (L_success == nullptr)   { L_success   = &L_fallthrough; label_nulls++; }
1598   if (L_failure == nullptr)   { L_failure   = &L_fallthrough; label_nulls++; }
1599   assert(label_nulls <= 1, "at most one null in the batch");
1600 
1601   BLOCK_COMMENT("check_klass_subtype_slow_path");
1602 
1603   RegSetIterator<Register> available_regs
1604     = (RegSet::range(r0, r15) - temps - sub_klass - super_klass).begin();
1605 
1606   RegSet pushed_regs;
1607 
1608   temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
1609   temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
1610   temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
1611   result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
1612 
1613   push(pushed_regs, sp);
1614 
1615   lookup_secondary_supers_table_var(sub_klass,
1616                                     super_klass,
1617                                     temp_reg, temp2_reg, temp3_reg, vtemp, result_reg,
1618                                     nullptr);
1619   cmp(result_reg, zr);
1620 
1621   // Unspill the temp. registers:
1622   pop(pushed_regs, sp);
1623 
1624   // NB! Callers may assume that, when set_cond_codes is true, this
1625   // code sets temp2_reg to a nonzero value.
1626   if (set_cond_codes) {
1627     mov(temp2_reg, 1);
1628   }
1629 
1630   br(Assembler::NE, *L_failure);
1631 
1632   if (L_success != &L_fallthrough) {
1633     b(*L_success);
1634   }
1635 
1636   bind(L_fallthrough);
1637 }
1638 
1639 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
1640                                                    Register super_klass,
1641                                                    Register temp_reg,
1642                                                    Register temp2_reg,
1643                                                    Label* L_success,
1644                                                    Label* L_failure,
1645                                                    bool set_cond_codes) {
1646   if (UseSecondarySupersTable) {
1647     check_klass_subtype_slow_path_table
1648       (sub_klass, super_klass, temp_reg, temp2_reg, /*temp3*/noreg, /*result*/noreg,
1649        /*vtemp*/fnoreg,
1650        L_success, L_failure, set_cond_codes);
1651   } else {
1652     check_klass_subtype_slow_path_linear
1653       (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, set_cond_codes);
1654   }
1655 }
1656 
1657 
1658 // Ensure that the inline code and the stub are using the same registers.
1659 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS                    \
1660 do {                                                               \
1661   assert(r_super_klass  == r0                                   && \
1662          r_array_base   == r1                                   && \
1663          r_array_length == r2                                   && \
1664          (r_array_index == r3        || r_array_index == noreg) && \
1665          (r_sub_klass   == r4        || r_sub_klass   == noreg) && \
1666          (r_bitmap      == rscratch2 || r_bitmap      == noreg) && \
1667          (result        == r5        || result        == noreg), "registers must match aarch64.ad"); \
1668 } while(0)
1669 
1670 bool MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
1671                                                          Register r_super_klass,
1672                                                          Register temp1,
1673                                                          Register temp2,
1674                                                          Register temp3,
1675                                                          FloatRegister vtemp,
1676                                                          Register result,
1677                                                          u1 super_klass_slot,
1678                                                          bool stub_is_near) {
1679   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1680 
1681   Label L_fallthrough;
1682 
1683   BLOCK_COMMENT("lookup_secondary_supers_table {");
1684 
1685   const Register
1686     r_array_base   = temp1, // r1
1687     r_array_length = temp2, // r2
1688     r_array_index  = temp3, // r3
1689     r_bitmap       = rscratch2;
1690 
1691   LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1692 
1693   u1 bit = super_klass_slot;
1694 
1695   // Make sure that result is nonzero if the TBZ below misses.
1696   mov(result, 1);
1697 
1698   // We're going to need the bitmap in a vector reg and in a core reg,
1699   // so load both now.
1700   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1701   if (bit != 0) {
1702     ldrd(vtemp, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1703   }
1704   // First check the bitmap to see if super_klass might be present. If
1705   // the bit is zero, we are certain that super_klass is not one of
1706   // the secondary supers.
1707   tbz(r_bitmap, bit, L_fallthrough);
1708 
1709   // Get the first array index that can contain super_klass into r_array_index.
1710   if (bit != 0) {
1711     shld(vtemp, vtemp, Klass::SECONDARY_SUPERS_TABLE_MASK - bit);
1712     cnt(vtemp, T8B, vtemp);
1713     addv(vtemp, T8B, vtemp);
1714     fmovd(r_array_index, vtemp);
1715   } else {
1716     mov(r_array_index, (u1)1);
1717   }
1718   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1719 
1720   // We will consult the secondary-super array.
1721   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1722 
1723   // The value i in r_array_index is >= 1, so even though r_array_base
1724   // points to the length, we don't need to adjust it to point to the
1725   // data.
1726   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1727   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1728 
1729   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1730   eor(result, result, r_super_klass);
1731   cbz(result, L_fallthrough); // Found a match
1732 
1733   // Is there another entry to check? Consult the bitmap.
1734   tbz(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK, L_fallthrough);
1735 
1736   // Linear probe.
1737   if (bit != 0) {
1738     ror(r_bitmap, r_bitmap, bit);
1739   }
1740 
1741   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1742   // The next slot to be inspected, by the stub we're about to call,
1743   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1744   // have been checked.
1745   Address stub = RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub());
1746   if (stub_is_near) {
1747     bl(stub);
1748   } else {
1749     address call = trampoline_call(stub);
1750     if (call == nullptr) {
1751       return false; // trampoline allocation failed
1752     }
1753   }
1754 
1755   BLOCK_COMMENT("} lookup_secondary_supers_table");
1756 
1757   bind(L_fallthrough);
1758 
1759   if (VerifySecondarySupers) {
1760     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1761                                   temp1, temp2, result);      // r1, r2, r5
1762   }
1763   return true;
1764 }
1765 
1766 // At runtime, return 0 in result if r_super_klass is a superclass of
1767 // r_sub_klass, otherwise return nonzero. Use this version of
1768 // lookup_secondary_supers_table() if you don't know ahead of time
1769 // which superclass will be searched for. Used by interpreter and
1770 // runtime stubs. It is larger and has somewhat greater latency than
1771 // the version above, which takes a constant super_klass_slot.
1772 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
1773                                                        Register r_super_klass,
1774                                                        Register temp1,
1775                                                        Register temp2,
1776                                                        Register temp3,
1777                                                        FloatRegister vtemp,
1778                                                        Register result,
1779                                                        Label *L_success) {
1780   assert_different_registers(r_sub_klass, temp1, temp2, temp3, result, rscratch1, rscratch2);
1781 
1782   Label L_fallthrough;
1783 
1784   BLOCK_COMMENT("lookup_secondary_supers_table {");
1785 
1786   const Register
1787     r_array_index = temp3,
1788     slot          = rscratch1,
1789     r_bitmap      = rscratch2;
1790 
1791   ldrb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
1792 
1793   // Make sure that result is nonzero if the test below misses.
1794   mov(result, 1);
1795 
1796   ldr(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
1797 
1798   // First check the bitmap to see if super_klass might be present. If
1799   // the bit is zero, we are certain that super_klass is not one of
1800   // the secondary supers.
1801 
1802   // This next instruction is equivalent to:
1803   // mov(tmp_reg, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1804   // sub(temp2, tmp_reg, slot);
1805   eor(temp2, slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1));
1806   lslv(temp2, r_bitmap, temp2);
1807   tbz(temp2, Klass::SECONDARY_SUPERS_TABLE_SIZE - 1, L_fallthrough);
1808 
1809   bool must_save_v0 = (vtemp == fnoreg);
1810   if (must_save_v0) {
1811     // temp1 and result are free, so use them to preserve vtemp
1812     vtemp = v0;
1813     mov(temp1,  vtemp, D, 0);
1814     mov(result, vtemp, D, 1);
1815   }
1816 
1817   // Get the first array index that can contain super_klass into r_array_index.
1818   mov(vtemp, D, 0, temp2);
1819   cnt(vtemp, T8B, vtemp);
1820   addv(vtemp, T8B, vtemp);
1821   mov(r_array_index, vtemp, D, 0);
1822 
1823   if (must_save_v0) {
1824     mov(vtemp, D, 0, temp1 );
1825     mov(vtemp, D, 1, result);
1826   }
1827 
1828   // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
1829 
1830   const Register
1831     r_array_base   = temp1,
1832     r_array_length = temp2;
1833 
1834   // The value i in r_array_index is >= 1, so even though r_array_base
1835   // points to the length, we don't need to adjust it to point to the
1836   // data.
1837   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
1838   assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
1839 
1840   // We will consult the secondary-super array.
1841   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1842 
1843   ldr(result, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1844   eor(result, result, r_super_klass);
1845   cbz(result, L_success ? *L_success : L_fallthrough); // Found a match
1846 
1847   // Is there another entry to check? Consult the bitmap.
1848   rorv(r_bitmap, r_bitmap, slot);
1849   // rol(r_bitmap, r_bitmap, 1);
1850   tbz(r_bitmap, 1, L_fallthrough);
1851 
1852   // The slot we just inspected is at secondary_supers[r_array_index - 1].
1853   // The next slot to be inspected, by the logic we're about to call,
1854   // is secondary_supers[r_array_index]. Bits 0 and 1 in the bitmap
1855   // have been checked.
1856   lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index,
1857                                           r_bitmap, r_array_length, result, /*is_stub*/false);
1858 
1859   BLOCK_COMMENT("} lookup_secondary_supers_table");
1860 
1861   bind(L_fallthrough);
1862 
1863   if (VerifySecondarySupers) {
1864     verify_secondary_supers_table(r_sub_klass, r_super_klass, // r4, r0
1865                                   temp1, temp2, result);      // r1, r2, r5
1866   }
1867 
1868   if (L_success) {
1869     cbz(result, *L_success);
1870   }
1871 }
1872 
1873 // Called by code generated by check_klass_subtype_slow_path
1874 // above. This is called when there is a collision in the hashed
1875 // lookup in the secondary supers array.
1876 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
1877                                                              Register r_array_base,
1878                                                              Register r_array_index,
1879                                                              Register r_bitmap,
1880                                                              Register temp1,
1881                                                              Register result,
1882                                                              bool is_stub) {
1883   assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, result, rscratch1);
1884 
1885   const Register
1886     r_array_length = temp1,
1887     r_sub_klass    = noreg; // unused
1888 
1889   if (is_stub) {
1890     LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
1891   }
1892 
1893   Label L_fallthrough, L_huge;
1894 
1895   // Load the array length.
1896   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1897   // And adjust the array base to point to the data.
1898   // NB! Effectively increments current slot index by 1.
1899   assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
1900   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1901 
1902   // The bitmap is full to bursting.
1903   // Implicit invariant: BITMAP_FULL implies (length > 0)
1904   assert(Klass::SECONDARY_SUPERS_BITMAP_FULL == ~uintx(0), "");
1905   cmpw(r_array_length, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 2));
1906   br(GT, L_huge);
1907 
1908   // NB! Our caller has checked bits 0 and 1 in the bitmap. The
1909   // current slot (at secondary_supers[r_array_index]) has not yet
1910   // been inspected, and r_array_index may be out of bounds if we
1911   // wrapped around the end of the array.
1912 
1913   { // This is conventional linear probing, but instead of terminating
1914     // when a null entry is found in the table, we maintain a bitmap
1915     // in which a 0 indicates missing entries.
1916     // As long as the bitmap is not completely full,
1917     // array_length == popcount(bitmap). The array_length check above
1918     // guarantees there are 0s in the bitmap, so the loop eventually
1919     // terminates.
1920     Label L_loop;
1921     bind(L_loop);
1922 
1923     // Check for wraparound.
1924     cmp(r_array_index, r_array_length);
1925     csel(r_array_index, zr, r_array_index, GE);
1926 
1927     ldr(rscratch1, Address(r_array_base, r_array_index, Address::lsl(LogBytesPerWord)));
1928     eor(result, rscratch1, r_super_klass);
1929     cbz(result, L_fallthrough);
1930 
1931     tbz(r_bitmap, 2, L_fallthrough); // look-ahead check (Bit 2); result is non-zero
1932 
1933     ror(r_bitmap, r_bitmap, 1);
1934     add(r_array_index, r_array_index, 1);
1935     b(L_loop);
1936   }
1937 
1938   { // Degenerate case: more than 64 secondary supers.
1939     // FIXME: We could do something smarter here, maybe a vectorized
1940     // comparison or a binary search, but is that worth any added
1941     // complexity?
1942     bind(L_huge);
1943     cmp(sp, zr); // Clear Z flag; SP is never zero
1944     repne_scan(r_array_base, r_super_klass, r_array_length, rscratch1);
1945     cset(result, NE); // result == 0 iff we got a match.
1946   }
1947 
1948   bind(L_fallthrough);
1949 }
1950 
1951 // Make sure that the hashed lookup and a linear scan agree.
1952 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
1953                                                    Register r_super_klass,
1954                                                    Register temp1,
1955                                                    Register temp2,
1956                                                    Register result) {
1957   assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, result, rscratch1);
1958 
1959   const Register
1960     r_array_base   = temp1,
1961     r_array_length = temp2;
1962 
1963   BLOCK_COMMENT("verify_secondary_supers_table {");
1964 
1965   // We will consult the secondary-super array.
1966   ldr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
1967 
1968   // Load the array length.
1969   ldrw(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
1970   // And adjust the array base to point to the data.
1971   add(r_array_base, r_array_base, Array<Klass*>::base_offset_in_bytes());
1972 
1973   cmp(sp, zr); // Clear Z flag; SP is never zero
1974   // Scan R2 words at [R5] for an occurrence of R0.
1975   // Set NZ/Z based on last compare.
1976   repne_scan(/*addr*/r_array_base, /*value*/r_super_klass, /*count*/r_array_length, rscratch2);
1977   // rscratch1 == 0 iff we got a match.
1978   cset(rscratch1, NE);
1979 
1980   Label passed;
1981   cmp(result, zr);
1982   cset(result, NE); // normalize result to 0/1 for comparison
1983 
1984   cmp(rscratch1, result);
1985   br(EQ, passed);
1986   {
1987     mov(r0, r_super_klass);         // r0 <- r0
1988     mov(r1, r_sub_klass);           // r1 <- r4
1989     mov(r2, /*expected*/rscratch1); // r2 <- r8
1990     mov(r3, result);                // r3 <- r5
1991     mov(r4, (address)("mismatch")); // r4 <- const
1992     rt_call(CAST_FROM_FN_PTR(address, Klass::on_secondary_supers_verification_failure), rscratch2);
1993     should_not_reach_here();
1994   }
1995   bind(passed);
1996 
1997   BLOCK_COMMENT("} verify_secondary_supers_table");
1998 }
1999 
2000 void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
2001   assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
2002   assert_different_registers(klass, rthread, scratch);
2003 
2004   Label L_fallthrough, L_tmp;
2005   if (L_fast_path == nullptr) {
2006     L_fast_path = &L_fallthrough;
2007   } else if (L_slow_path == nullptr) {
2008     L_slow_path = &L_fallthrough;
2009   }
2010   // Fast path check: class is fully initialized
2011   lea(scratch, Address(klass, InstanceKlass::init_state_offset()));
2012   ldarb(scratch, scratch);
2013   cmp(scratch, InstanceKlass::fully_initialized);
2014   br(Assembler::EQ, *L_fast_path);
2015 
2016   // Fast path check: current thread is initializer thread
2017   ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
2018   cmp(rthread, scratch);
2019 
2020   if (L_slow_path == &L_fallthrough) {
2021     br(Assembler::EQ, *L_fast_path);
2022     bind(*L_slow_path);
2023   } else if (L_fast_path == &L_fallthrough) {
2024     br(Assembler::NE, *L_slow_path);
2025     bind(*L_fast_path);
2026   } else {
2027     Unimplemented();
2028   }
2029 }
2030 
2031 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
2032   if (!VerifyOops || VerifyAdapterSharing) {
2033     // Below address of the code string confuses VerifyAdapterSharing
2034     // because it may differ between otherwise equivalent adapters.
2035     return;
2036   }
2037 
2038   // Pass register number to verify_oop_subroutine
2039   const char* b = nullptr;
2040   {
2041     ResourceMark rm;
2042     stringStream ss;
2043     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
2044     b = code_string(ss.as_string());
2045   }
2046   BLOCK_COMMENT("verify_oop {");
2047 
2048   strip_return_address(); // This might happen within a stack frame.
2049   protect_return_address();
2050   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2051   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2052 
2053   mov(r0, reg);
2054   movptr(rscratch1, (uintptr_t)(address)b);
2055 
2056   // call indirectly to solve generation ordering problem
2057   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2058   ldr(rscratch2, Address(rscratch2));
2059   blr(rscratch2);
2060 
2061   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2062   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2063   authenticate_return_address();
2064 
2065   BLOCK_COMMENT("} verify_oop");
2066 }
2067 
2068 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
2069   if (!VerifyOops || VerifyAdapterSharing) {
2070     // Below address of the code string confuses VerifyAdapterSharing
2071     // because it may differ between otherwise equivalent adapters.
2072     return;
2073   }
2074 
2075   const char* b = nullptr;
2076   {
2077     ResourceMark rm;
2078     stringStream ss;
2079     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
2080     b = code_string(ss.as_string());
2081   }
2082   BLOCK_COMMENT("verify_oop_addr {");
2083 
2084   strip_return_address(); // This might happen within a stack frame.
2085   protect_return_address();
2086   stp(r0, rscratch1, Address(pre(sp, -2 * wordSize)));
2087   stp(rscratch2, lr, Address(pre(sp, -2 * wordSize)));
2088 
2089   // addr may contain sp so we will have to adjust it based on the
2090   // pushes that we just did.
2091   if (addr.uses(sp)) {
2092     lea(r0, addr);
2093     ldr(r0, Address(r0, 4 * wordSize));
2094   } else {
2095     ldr(r0, addr);
2096   }
2097   movptr(rscratch1, (uintptr_t)(address)b);
2098 
2099   // call indirectly to solve generation ordering problem
2100   lea(rscratch2, RuntimeAddress(StubRoutines::verify_oop_subroutine_entry_address()));
2101   ldr(rscratch2, Address(rscratch2));
2102   blr(rscratch2);
2103 
2104   ldp(rscratch2, lr, Address(post(sp, 2 * wordSize)));
2105   ldp(r0, rscratch1, Address(post(sp, 2 * wordSize)));
2106   authenticate_return_address();
2107 
2108   BLOCK_COMMENT("} verify_oop_addr");
2109 }
2110 
2111 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
2112                                          int extra_slot_offset) {
2113   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
2114   int stackElementSize = Interpreter::stackElementSize;
2115   int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
2116 #ifdef ASSERT
2117   int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
2118   assert(offset1 - offset == stackElementSize, "correct arithmetic");
2119 #endif
2120   if (arg_slot.is_constant()) {
2121     return Address(esp, arg_slot.as_constant() * stackElementSize
2122                    + offset);
2123   } else {
2124     add(rscratch1, esp, arg_slot.as_register(),
2125         ext::uxtx, exact_log2(stackElementSize));
2126     return Address(rscratch1, offset);
2127   }
2128 }
2129 
2130 // Handle the receiver type profile update given the "recv" klass.
2131 //
2132 // Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
2133 // If there are no matching or claimable receiver entries in RD, updates
2134 // the polymorphic counter.
2135 //
2136 // This code expected to run by either the interpreter or JIT-ed code, without
2137 // extra synchronization. For safety, receiver cells are claimed atomically, which
2138 // avoids grossly misrepresenting the profiles under concurrent updates. For speed,
2139 // counter updates are not atomic.
2140 //
2141 void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
2142   assert_different_registers(recv, mdp, rscratch1, rscratch2);
2143 
2144   int base_receiver_offset   = in_bytes(ReceiverTypeData::receiver_offset(0));
2145   int end_receiver_offset    = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
2146   int poly_count_offset      = in_bytes(CounterData::count_offset());
2147   int receiver_step          = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
2148   int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
2149 
2150   // Adjust for MDP offsets.
2151   base_receiver_offset += mdp_offset;
2152   end_receiver_offset  += mdp_offset;
2153   poly_count_offset    += mdp_offset;
2154 
2155 #ifdef ASSERT
2156   // We are about to walk the MDO slots without asking for offsets.
2157   // Check that our math hits all the right spots.
2158   for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
2159     int real_recv_offset  = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
2160     int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
2161     int offset = base_receiver_offset + receiver_step*c;
2162     int count_offset = offset + receiver_to_count_step;
2163     assert(offset == real_recv_offset, "receiver slot math");
2164     assert(count_offset == real_count_offset, "receiver count math");
2165   }
2166   int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
2167   assert(poly_count_offset == real_poly_count_offset, "poly counter math");
2168 #endif
2169 
2170   // Corner case: no profile table. Increment poly counter and exit.
2171   if (ReceiverTypeData::row_limit() == 0) {
2172     increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
2173     return;
2174   }
2175 
2176   Register offset = rscratch2;
2177 
2178   Label L_loop_search_receiver, L_loop_search_empty;
2179   Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
2180 
2181   // The code here recognizes three major cases:
2182   //   A. Fastest: receiver found in the table
2183   //   B. Fast: no receiver in the table, and the table is full
2184   //   C. Slow: no receiver in the table, free slots in the table
2185   //
2186   // The case A performance is most important, as perfectly-behaved code would end up
2187   // there, especially with larger TypeProfileWidth. The case B performance is
2188   // important as well, this is where bulk of code would land for normally megamorphic
2189   // cases. The case C performance is not essential, its job is to deal with installation
2190   // races, we optimize for code density instead. Case C needs to make sure that receiver
2191   // rows are only claimed once. This makes sure we never overwrite a row for another
2192   // receiver and never duplicate the receivers in the list, making profile type-accurate.
2193   //
2194   // It is very tempting to handle these cases in a single loop, and claim the first slot
2195   // without checking the rest of the table. But, profiling code should tolerate free slots
2196   // in the table, as class unloading can clear them. After such cleanup, the receiver
2197   // we need might be _after_ the free slot. Therefore, we need to let at least full scan
2198   // to complete, before trying to install new slots. Splitting the code in several tight
2199   // loops also helpfully optimizes for cases A and B.
2200   //
2201   // This code is effectively:
2202   //
2203   // restart:
2204   //   // Fastest: receiver is already installed
2205   //   for (i = 0; i < receiver_count(); i++) {
2206   //     if (receiver(i) == recv) goto found_recv(i);
2207   //   }
2208   //
2209   //   // Fast: no receiver, but profile is full
2210   //   for (i = 0; i < receiver_count(); i++) {
2211   //     if (receiver(i) == null) goto found_null(i);
2212   //   }
2213   //   goto polymorphic
2214   //
2215   //   // Slow: try to install receiver
2216   // found_null(i):
2217   //   CAS(&receiver(i), null, recv);
2218   //   goto restart
2219   //
2220   // polymorphic:
2221   //   count++;
2222   //   return
2223   //
2224   // found_recv(i):
2225   //   *receiver_count(i)++
2226   //
2227 
2228   bind(L_restart);
2229 
2230   // Fastest: receiver is already installed
2231   mov(offset, base_receiver_offset);
2232   bind(L_loop_search_receiver);
2233     ldr(rscratch1, Address(mdp, offset));
2234     cmp(rscratch1, recv);
2235     br(Assembler::EQ, L_found_recv);
2236   add(offset, offset, receiver_step);
2237   sub(rscratch1, offset, end_receiver_offset);
2238   cbnz(rscratch1, L_loop_search_receiver);
2239 
2240   // Fast: no receiver, but profile is full
2241   mov(offset, base_receiver_offset);
2242   bind(L_loop_search_empty);
2243     ldr(rscratch1, Address(mdp, offset));
2244     cbz(rscratch1, L_found_empty);
2245   add(offset, offset, receiver_step);
2246   sub(rscratch1, offset, end_receiver_offset);
2247   cbnz(rscratch1, L_loop_search_empty);
2248   b(L_polymorphic);
2249 
2250   // Slow: try to install receiver
2251   bind(L_found_empty);
2252 
2253   // Atomically swing receiver slot: null -> recv.
2254   //
2255   // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
2256   // is used to hold the destination address. This is safe because the
2257   // offset is no longer needed after the address is computed.
2258 
2259   lea(rscratch2, Address(mdp, offset));
2260   cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
2261           /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
2262 
2263   // CAS success means the slot now has the receiver we want. CAS failure means
2264   // something had claimed the slot concurrently: it can be the same receiver we want,
2265   // or something else. Since this is a slow path, we can optimize for code density,
2266   // and just restart the search from the beginning.
2267   b(L_restart);
2268 
2269   // Counter updates:
2270 
2271   // Increment polymorphic counter instead of receiver slot.
2272   bind(L_polymorphic);
2273   mov(offset, poly_count_offset);
2274   b(L_count_update);
2275 
2276   // Found a receiver, convert its slot offset to corresponding count offset.
2277   bind(L_found_recv);
2278   add(offset, offset, receiver_to_count_step);
2279 
2280   bind(L_count_update);
2281   increment(Address(mdp, offset), DataLayout::counter_increment);
2282 }
2283 
2284 
2285 void MacroAssembler::call_VM_leaf_base(address entry_point,
2286                                        int number_of_arguments,
2287                                        Label *retaddr) {
2288   Label E, L;
2289 
2290   stp(rscratch1, rmethod, Address(pre(sp, -2 * wordSize)));
2291 
2292   mov(rscratch1, RuntimeAddress(entry_point));
2293   blr(rscratch1);
2294   if (retaddr)
2295     bind(*retaddr);
2296 
2297   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
2298 }
2299 
2300 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
2301   call_VM_leaf_base(entry_point, number_of_arguments);
2302 }
2303 
2304 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
2305   pass_arg0(this, arg_0);
2306   call_VM_leaf_base(entry_point, 1);
2307 }
2308 
2309 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2310   assert_different_registers(arg_1, c_rarg0);
2311   pass_arg0(this, arg_0);
2312   pass_arg1(this, arg_1);
2313   call_VM_leaf_base(entry_point, 2);
2314 }
2315 
2316 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
2317                                   Register arg_1, Register arg_2) {
2318   assert_different_registers(arg_1, c_rarg0);
2319   assert_different_registers(arg_2, c_rarg0, c_rarg1);
2320   pass_arg0(this, arg_0);
2321   pass_arg1(this, arg_1);
2322   pass_arg2(this, arg_2);
2323   call_VM_leaf_base(entry_point, 3);
2324 }
2325 
2326 void MacroAssembler::super_call_VM_leaf(address entry_point) {
2327   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2328 }
2329 
2330 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
2331   pass_arg0(this, arg_0);
2332   MacroAssembler::call_VM_leaf_base(entry_point, 1);
2333 }
2334 
2335 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
2336 
2337   assert_different_registers(arg_0, c_rarg1);
2338   pass_arg1(this, arg_1);
2339   pass_arg0(this, arg_0);
2340   MacroAssembler::call_VM_leaf_base(entry_point, 2);
2341 }
2342 
2343 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
2344   assert_different_registers(arg_0, c_rarg1, c_rarg2);
2345   assert_different_registers(arg_1, c_rarg2);
2346   pass_arg2(this, arg_2);
2347   pass_arg1(this, arg_1);
2348   pass_arg0(this, arg_0);
2349   MacroAssembler::call_VM_leaf_base(entry_point, 3);
2350 }
2351 
2352 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
2353   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
2354   assert_different_registers(arg_1, c_rarg2, c_rarg3);
2355   assert_different_registers(arg_2, c_rarg3);
2356   pass_arg3(this, arg_3);
2357   pass_arg2(this, arg_2);
2358   pass_arg1(this, arg_1);
2359   pass_arg0(this, arg_0);
2360   MacroAssembler::call_VM_leaf_base(entry_point, 4);
2361 }
2362 
2363 void MacroAssembler::null_check(Register reg, int offset) {
2364   if (needs_explicit_null_check(offset)) {
2365     // provoke OS null exception if reg is null by
2366     // accessing M[reg] w/o changing any registers
2367     // NOTE: this is plenty to provoke a segv
2368     ldr(zr, Address(reg));
2369   } else {
2370     // nothing to do, (later) access of M[reg + offset]
2371     // will provoke OS null exception if reg is null
2372   }
2373 }
2374 
2375 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
2376   assert_different_registers(markword, rscratch2);
2377   mov(rscratch2, markWord::inline_type_mask_in_place);
2378   andr(markword, markword, rscratch2);
2379   mov(rscratch2, markWord::inline_type_pattern);
2380   cmp(markword, rscratch2);
2381   br(Assembler::EQ, is_inline_type);
2382 }
2383 
2384 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
2385   assert_different_registers(tmp, rscratch1);
2386   if (can_be_null) {
2387     cbz(object, not_inline_type);
2388   }
2389   const int is_inline_type_mask = markWord::inline_type_pattern;
2390   ldr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
2391   mov(rscratch1, is_inline_type_mask);
2392   andr(tmp, tmp, rscratch1);
2393   cmp(tmp, rscratch1);
2394   br(Assembler::NE, not_inline_type);
2395 }
2396 
2397 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
2398   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2399   tbnz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, is_null_free_inline_type);
2400 }
2401 
2402 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
2403   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2404   tbz(flags, ResolvedFieldEntry::is_null_free_inline_type_shift, not_null_free_inline_type);
2405 }
2406 
2407 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
2408   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2409   tbnz(flags, ResolvedFieldEntry::is_flat_shift, is_flat);
2410 }
2411 
2412 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
2413   assert(temp_reg == noreg, "not needed"); // keep signature uniform with x86
2414   tbnz(flags, ResolvedFieldEntry::has_null_marker_shift, has_null_marker);
2415 }
2416 
2417 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
2418   Label test_mark_word;
2419   // load mark word
2420   ldr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
2421   // check displaced
2422   tst(temp_reg, markWord::unlocked_value);
2423   br(Assembler::NE, test_mark_word);
2424   // slow path use klass prototype
2425   load_prototype_header(temp_reg, oop);
2426 
2427   bind(test_mark_word);
2428   andr(temp_reg, temp_reg, test_bit);
2429   if (jmp_set) {
2430     cbnz(temp_reg, jmp_label);
2431   } else {
2432     cbz(temp_reg, jmp_label);
2433   }
2434 }
2435 
2436 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array) {
2437   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
2438 }
2439 
2440 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
2441                                                   Label&is_non_flat_array) {
2442   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
2443 }
2444 
2445 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array) {
2446   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
2447 }
2448 
2449 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
2450   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
2451 }
2452 
2453 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
2454   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2455   br(Assembler::NE, is_flat_array);
2456 }
2457 
2458 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
2459   tst(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
2460   br(Assembler::EQ, is_non_flat_array);
2461 }
2462 
2463 // MacroAssembler protected routines needed to implement
2464 // public methods
2465 
2466 void MacroAssembler::mov(Register r, Address dest) {
2467   code_section()->relocate(pc(), dest.rspec());
2468   uint64_t imm64 = (uint64_t)dest.target();
2469   movptr(r, imm64);
2470 }
2471 
2472 // Move a constant pointer into r.  In AArch64 mode the virtual
2473 // address space is 48 bits in size, so we only need three
2474 // instructions to create a patchable instruction sequence that can
2475 // reach anywhere.
2476 void MacroAssembler::movptr(Register r, uintptr_t imm64) {
2477 #ifndef PRODUCT
2478   {
2479     char buffer[64];
2480     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, (uint64_t)imm64);
2481     block_comment(buffer);
2482   }
2483 #endif
2484   assert(imm64 < (1ull << 48), "48-bit overflow in address constant");
2485   movz(r, imm64 & 0xffff);
2486   imm64 >>= 16;
2487   movk(r, imm64 & 0xffff, 16);
2488   imm64 >>= 16;
2489   movk(r, imm64 & 0xffff, 32);
2490 }
2491 
2492 // Macro to mov replicated immediate to vector register.
2493 // imm64: only the lower 8/16/32 bits are considered for B/H/S type. That is,
2494 //        the upper 56/48/32 bits must be zeros for B/H/S type.
2495 // Vd will get the following values for different arrangements in T
2496 //   imm64 == hex 000000gh  T8B:  Vd = ghghghghghghghgh
2497 //   imm64 == hex 000000gh  T16B: Vd = ghghghghghghghghghghghghghghghgh
2498 //   imm64 == hex 0000efgh  T4H:  Vd = efghefghefghefgh
2499 //   imm64 == hex 0000efgh  T8H:  Vd = efghefghefghefghefghefghefghefgh
2500 //   imm64 == hex abcdefgh  T2S:  Vd = abcdefghabcdefgh
2501 //   imm64 == hex abcdefgh  T4S:  Vd = abcdefghabcdefghabcdefghabcdefgh
2502 //   imm64 == hex abcdefgh  T1D:  Vd = 00000000abcdefgh
2503 //   imm64 == hex abcdefgh  T2D:  Vd = 00000000abcdefgh00000000abcdefgh
2504 // Clobbers rscratch1
2505 void MacroAssembler::mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64) {
2506   assert(T != T1Q, "unsupported");
2507   if (T == T1D || T == T2D) {
2508     int imm = operand_valid_for_movi_immediate(imm64, T);
2509     if (-1 != imm) {
2510       movi(Vd, T, imm);
2511     } else {
2512       mov(rscratch1, imm64);
2513       dup(Vd, T, rscratch1);
2514     }
2515     return;
2516   }
2517 
2518 #ifdef ASSERT
2519   if (T == T8B || T == T16B) assert((imm64 & ~0xff) == 0, "extraneous bits (T8B/T16B)");
2520   if (T == T4H || T == T8H) assert((imm64  & ~0xffff) == 0, "extraneous bits (T4H/T8H)");
2521   if (T == T2S || T == T4S) assert((imm64  & ~0xffffffff) == 0, "extraneous bits (T2S/T4S)");
2522 #endif
2523   int shift = operand_valid_for_movi_immediate(imm64, T);
2524   uint32_t imm32 = imm64 & 0xffffffffULL;
2525   if (shift >= 0) {
2526     movi(Vd, T, (imm32 >> shift) & 0xff, shift);
2527   } else {
2528     movw(rscratch1, imm32);
2529     dup(Vd, T, rscratch1);
2530   }
2531 }
2532 
2533 void MacroAssembler::mov_immediate64(Register dst, uint64_t imm64)
2534 {
2535 #ifndef PRODUCT
2536   {
2537     char buffer[64];
2538     os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX64, imm64);
2539     block_comment(buffer);
2540   }
2541 #endif
2542   if (operand_valid_for_logical_immediate(false, imm64)) {
2543     orr(dst, zr, imm64);
2544   } else {
2545     // we can use a combination of MOVZ or MOVN with
2546     // MOVK to build up the constant
2547     uint64_t imm_h[4];
2548     int zero_count = 0;
2549     int neg_count = 0;
2550     int i;
2551     for (i = 0; i < 4; i++) {
2552       imm_h[i] = ((imm64 >> (i * 16)) & 0xffffL);
2553       if (imm_h[i] == 0) {
2554         zero_count++;
2555       } else if (imm_h[i] == 0xffffL) {
2556         neg_count++;
2557       }
2558     }
2559     if (zero_count == 4) {
2560       // one MOVZ will do
2561       movz(dst, 0);
2562     } else if (neg_count == 4) {
2563       // one MOVN will do
2564       movn(dst, 0);
2565     } else if (zero_count == 3) {
2566       for (i = 0; i < 4; i++) {
2567         if (imm_h[i] != 0L) {
2568           movz(dst, (uint32_t)imm_h[i], (i << 4));
2569           break;
2570         }
2571       }
2572     } else if (neg_count == 3) {
2573       // one MOVN will do
2574       for (int i = 0; i < 4; i++) {
2575         if (imm_h[i] != 0xffffL) {
2576           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2577           break;
2578         }
2579       }
2580     } else if (zero_count == 2) {
2581       // one MOVZ and one MOVK will do
2582       for (i = 0; i < 3; i++) {
2583         if (imm_h[i] != 0L) {
2584           movz(dst, (uint32_t)imm_h[i], (i << 4));
2585           i++;
2586           break;
2587         }
2588       }
2589       for (;i < 4; i++) {
2590         if (imm_h[i] != 0L) {
2591           movk(dst, (uint32_t)imm_h[i], (i << 4));
2592         }
2593       }
2594     } else if (neg_count == 2) {
2595       // one MOVN and one MOVK will do
2596       for (i = 0; i < 4; i++) {
2597         if (imm_h[i] != 0xffffL) {
2598           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2599           i++;
2600           break;
2601         }
2602       }
2603       for (;i < 4; i++) {
2604         if (imm_h[i] != 0xffffL) {
2605           movk(dst, (uint32_t)imm_h[i], (i << 4));
2606         }
2607       }
2608     } else if (zero_count == 1) {
2609       // one MOVZ and two MOVKs will do
2610       for (i = 0; i < 4; i++) {
2611         if (imm_h[i] != 0L) {
2612           movz(dst, (uint32_t)imm_h[i], (i << 4));
2613           i++;
2614           break;
2615         }
2616       }
2617       for (;i < 4; i++) {
2618         if (imm_h[i] != 0x0L) {
2619           movk(dst, (uint32_t)imm_h[i], (i << 4));
2620         }
2621       }
2622     } else if (neg_count == 1) {
2623       // one MOVN and two MOVKs will do
2624       for (i = 0; i < 4; i++) {
2625         if (imm_h[i] != 0xffffL) {
2626           movn(dst, (uint32_t)imm_h[i] ^ 0xffffL, (i << 4));
2627           i++;
2628           break;
2629         }
2630       }
2631       for (;i < 4; i++) {
2632         if (imm_h[i] != 0xffffL) {
2633           movk(dst, (uint32_t)imm_h[i], (i << 4));
2634         }
2635       }
2636     } else {
2637       // use a MOVZ and 3 MOVKs (makes it easier to debug)
2638       movz(dst, (uint32_t)imm_h[0], 0);
2639       for (i = 1; i < 4; i++) {
2640         movk(dst, (uint32_t)imm_h[i], (i << 4));
2641       }
2642     }
2643   }
2644 }
2645 
2646 void MacroAssembler::mov_immediate32(Register dst, uint32_t imm32)
2647 {
2648 #ifndef PRODUCT
2649     {
2650       char buffer[64];
2651       os::snprintf_checked(buffer, sizeof(buffer), "0x%" PRIX32, imm32);
2652       block_comment(buffer);
2653     }
2654 #endif
2655   if (operand_valid_for_logical_immediate(true, imm32)) {
2656     orrw(dst, zr, imm32);
2657   } else {
2658     // we can use MOVZ, MOVN or two calls to MOVK to build up the
2659     // constant
2660     uint32_t imm_h[2];
2661     imm_h[0] = imm32 & 0xffff;
2662     imm_h[1] = ((imm32 >> 16) & 0xffff);
2663     if (imm_h[0] == 0) {
2664       movzw(dst, imm_h[1], 16);
2665     } else if (imm_h[0] == 0xffff) {
2666       movnw(dst, imm_h[1] ^ 0xffff, 16);
2667     } else if (imm_h[1] == 0) {
2668       movzw(dst, imm_h[0], 0);
2669     } else if (imm_h[1] == 0xffff) {
2670       movnw(dst, imm_h[0] ^ 0xffff, 0);
2671     } else {
2672       // use a MOVZ and MOVK (makes it easier to debug)
2673       movzw(dst, imm_h[0], 0);
2674       movkw(dst, imm_h[1], 16);
2675     }
2676   }
2677 }
2678 
2679 // Form an address from base + offset in Rd.  Rd may or may
2680 // not actually be used: you must use the Address that is returned.
2681 // It is up to you to ensure that the shift provided matches the size
2682 // of your data.
2683 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset, int shift) {
2684   if (Address::offset_ok_for_immed(byte_offset, shift))
2685     // It fits; no need for any heroics
2686     return Address(base, byte_offset);
2687 
2688   // Don't do anything clever with negative or misaligned offsets
2689   unsigned mask = (1 << shift) - 1;
2690   if (byte_offset < 0 || byte_offset & mask) {
2691     mov(Rd, byte_offset);
2692     add(Rd, base, Rd);
2693     return Address(Rd);
2694   }
2695 
2696   // See if we can do this with two 12-bit offsets
2697   {
2698     uint64_t word_offset = byte_offset >> shift;
2699     uint64_t masked_offset = word_offset & 0xfff000;
2700     if (Address::offset_ok_for_immed(word_offset - masked_offset, 0)
2701         && Assembler::operand_valid_for_add_sub_immediate(masked_offset << shift)) {
2702       add(Rd, base, masked_offset << shift);
2703       word_offset -= masked_offset;
2704       return Address(Rd, word_offset << shift);
2705     }
2706   }
2707 
2708   // Do it the hard way
2709   mov(Rd, byte_offset);
2710   add(Rd, base, Rd);
2711   return Address(Rd);
2712 }
2713 
2714 int MacroAssembler::corrected_idivl(Register result, Register ra, Register rb,
2715                                     bool want_remainder, Register scratch)
2716 {
2717   // Full implementation of Java idiv and irem.  The function
2718   // returns the (pc) offset of the div instruction - may be needed
2719   // for implicit exceptions.
2720   //
2721   // constraint : ra/rb =/= scratch
2722   //         normal case
2723   //
2724   // input : ra: dividend
2725   //         rb: divisor
2726   //
2727   // result: either
2728   //         quotient  (= ra idiv rb)
2729   //         remainder (= ra irem rb)
2730 
2731   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2732 
2733   int idivl_offset = offset();
2734   if (! want_remainder) {
2735     sdivw(result, ra, rb);
2736   } else {
2737     sdivw(scratch, ra, rb);
2738     Assembler::msubw(result, scratch, rb, ra);
2739   }
2740 
2741   return idivl_offset;
2742 }
2743 
2744 int MacroAssembler::corrected_idivq(Register result, Register ra, Register rb,
2745                                     bool want_remainder, Register scratch)
2746 {
2747   // Full implementation of Java ldiv and lrem.  The function
2748   // returns the (pc) offset of the div instruction - may be needed
2749   // for implicit exceptions.
2750   //
2751   // constraint : ra/rb =/= scratch
2752   //         normal case
2753   //
2754   // input : ra: dividend
2755   //         rb: divisor
2756   //
2757   // result: either
2758   //         quotient  (= ra idiv rb)
2759   //         remainder (= ra irem rb)
2760 
2761   assert(ra != scratch && rb != scratch, "reg cannot be scratch");
2762 
2763   int idivq_offset = offset();
2764   if (! want_remainder) {
2765     sdiv(result, ra, rb);
2766   } else {
2767     sdiv(scratch, ra, rb);
2768     Assembler::msub(result, scratch, rb, ra);
2769   }
2770 
2771   return idivq_offset;
2772 }
2773 
2774 void MacroAssembler::membar(Membar_mask_bits order_constraint) {
2775   address prev = pc() - NativeMembar::instruction_size;
2776   address last = code()->last_insn();
2777   if (last != nullptr && nativeInstruction_at(last)->is_Membar() && prev == last) {
2778     NativeMembar *bar = NativeMembar_at(prev);
2779     if (AlwaysMergeDMB) {
2780       bar->set_kind(bar->get_kind() | order_constraint);
2781       BLOCK_COMMENT("merged membar(always)");
2782       return;
2783     }
2784     // Don't promote DMB ST|DMB LD to DMB (a full barrier) because
2785     // doing so would introduce a StoreLoad which the caller did not
2786     // intend
2787     if (bar->get_kind() == order_constraint
2788         || bar->get_kind() == AnyAny
2789         || order_constraint == AnyAny) {
2790       // We are merging two memory barrier instructions.  On AArch64 we
2791       // can do this simply by ORing them together.
2792       bar->set_kind(bar->get_kind() | order_constraint);
2793       BLOCK_COMMENT("merged membar");
2794       return;
2795     } else {
2796       // A special case like "DMB ST;DMB LD;DMB ST", the last DMB can be skipped
2797       // We need check the last 2 instructions
2798       address prev2 = prev - NativeMembar::instruction_size;
2799       if (last != code()->last_label() && nativeInstruction_at(prev2)->is_Membar()) {
2800         NativeMembar *bar2 = NativeMembar_at(prev2);
2801         assert(bar2->get_kind() == order_constraint, "it should be merged before");
2802         BLOCK_COMMENT("merged membar(elided)");
2803         return;
2804       }
2805     }
2806   }
2807   code()->set_last_insn(pc());
2808   dmb(Assembler::barrier(order_constraint));
2809 }
2810 
2811 bool MacroAssembler::try_merge_ldst(Register rt, const Address &adr, size_t size_in_bytes, bool is_store) {
2812   if (ldst_can_merge(rt, adr, size_in_bytes, is_store)) {
2813     merge_ldst(rt, adr, size_in_bytes, is_store);
2814     code()->clear_last_insn();
2815     return true;
2816   } else {
2817     assert(size_in_bytes == 8 || size_in_bytes == 4, "only 8 bytes or 4 bytes load/store is supported.");
2818     const uint64_t mask = size_in_bytes - 1;
2819     if (adr.getMode() == Address::base_plus_offset &&
2820         (adr.offset() & mask) == 0) { // only supports base_plus_offset.
2821       code()->set_last_insn(pc());
2822     }
2823     return false;
2824   }
2825 }
2826 
2827 void MacroAssembler::ldr(Register Rx, const Address &adr) {
2828   // We always try to merge two adjacent loads into one ldp.
2829   if (!try_merge_ldst(Rx, adr, 8, false)) {
2830     Assembler::ldr(Rx, adr);
2831   }
2832 }
2833 
2834 void MacroAssembler::ldrw(Register Rw, const Address &adr) {
2835   // We always try to merge two adjacent loads into one ldp.
2836   if (!try_merge_ldst(Rw, adr, 4, false)) {
2837     Assembler::ldrw(Rw, adr);
2838   }
2839 }
2840 
2841 void MacroAssembler::str(Register Rx, const Address &adr) {
2842   // We always try to merge two adjacent stores into one stp.
2843   if (!try_merge_ldst(Rx, adr, 8, true)) {
2844     Assembler::str(Rx, adr);
2845   }
2846 }
2847 
2848 void MacroAssembler::strw(Register Rw, const Address &adr) {
2849   // We always try to merge two adjacent stores into one stp.
2850   if (!try_merge_ldst(Rw, adr, 4, true)) {
2851     Assembler::strw(Rw, adr);
2852   }
2853 }
2854 
2855 // MacroAssembler routines found actually to be needed
2856 
2857 void MacroAssembler::push(Register src)
2858 {
2859   str(src, Address(pre(esp, -1 * wordSize)));
2860 }
2861 
2862 void MacroAssembler::pop(Register dst)
2863 {
2864   ldr(dst, Address(post(esp, 1 * wordSize)));
2865 }
2866 
2867 // Note: load_unsigned_short used to be called load_unsigned_word.
2868 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2869   int off = offset();
2870   ldrh(dst, src);
2871   return off;
2872 }
2873 
2874 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2875   int off = offset();
2876   ldrb(dst, src);
2877   return off;
2878 }
2879 
2880 int MacroAssembler::load_signed_short(Register dst, Address src) {
2881   int off = offset();
2882   ldrsh(dst, src);
2883   return off;
2884 }
2885 
2886 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2887   int off = offset();
2888   ldrsb(dst, src);
2889   return off;
2890 }
2891 
2892 int MacroAssembler::load_signed_short32(Register dst, Address src) {
2893   int off = offset();
2894   ldrshw(dst, src);
2895   return off;
2896 }
2897 
2898 int MacroAssembler::load_signed_byte32(Register dst, Address src) {
2899   int off = offset();
2900   ldrsbw(dst, src);
2901   return off;
2902 }
2903 
2904 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
2905   switch (size_in_bytes) {
2906   case  8:  ldr(dst, src); break;
2907   case  4:  ldrw(dst, src); break;
2908   case  2:  is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2909   case  1:  is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2910   default:  ShouldNotReachHere();
2911   }
2912 }
2913 
2914 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
2915   switch (size_in_bytes) {
2916   case  8:  str(src, dst); break;
2917   case  4:  strw(src, dst); break;
2918   case  2:  strh(src, dst); break;
2919   case  1:  strb(src, dst); break;
2920   default:  ShouldNotReachHere();
2921   }
2922 }
2923 
2924 void MacroAssembler::decrementw(Register reg, int value)
2925 {
2926   if (value < 0)  { incrementw(reg, -value);      return; }
2927   if (value == 0) {                               return; }
2928   if (value < (1 << 12)) { subw(reg, reg, value); return; }
2929   /* else */ {
2930     guarantee(reg != rscratch2, "invalid dst for register decrement");
2931     movw(rscratch2, (unsigned)value);
2932     subw(reg, reg, rscratch2);
2933   }
2934 }
2935 
2936 void MacroAssembler::decrement(Register reg, int value)
2937 {
2938   if (value < 0)  { increment(reg, -value);      return; }
2939   if (value == 0) {                              return; }
2940   if (value < (1 << 12)) { sub(reg, reg, value); return; }
2941   /* else */ {
2942     assert(reg != rscratch2, "invalid dst for register decrement");
2943     mov(rscratch2, (uint64_t)value);
2944     sub(reg, reg, rscratch2);
2945   }
2946 }
2947 
2948 void MacroAssembler::decrementw(Address dst, int value)
2949 {
2950   assert(!dst.uses(rscratch1), "invalid dst for address decrement");
2951   if (dst.getMode() == Address::literal) {
2952     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2953     lea(rscratch2, dst);
2954     dst = Address(rscratch2);
2955   }
2956   ldrw(rscratch1, dst);
2957   decrementw(rscratch1, value);
2958   strw(rscratch1, dst);
2959 }
2960 
2961 void MacroAssembler::decrement(Address dst, int value)
2962 {
2963   assert(!dst.uses(rscratch1), "invalid address for decrement");
2964   if (dst.getMode() == Address::literal) {
2965     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
2966     lea(rscratch2, dst);
2967     dst = Address(rscratch2);
2968   }
2969   ldr(rscratch1, dst);
2970   decrement(rscratch1, value);
2971   str(rscratch1, dst);
2972 }
2973 
2974 void MacroAssembler::incrementw(Register reg, int value)
2975 {
2976   if (value < 0)  { decrementw(reg, -value);      return; }
2977   if (value == 0) {                               return; }
2978   if (value < (1 << 12)) { addw(reg, reg, value); return; }
2979   /* else */ {
2980     assert(reg != rscratch2, "invalid dst for register increment");
2981     movw(rscratch2, (unsigned)value);
2982     addw(reg, reg, rscratch2);
2983   }
2984 }
2985 
2986 void MacroAssembler::increment(Register reg, int value)
2987 {
2988   if (value < 0)  { decrement(reg, -value);      return; }
2989   if (value == 0) {                              return; }
2990   if (value < (1 << 12)) { add(reg, reg, value); return; }
2991   /* else */ {
2992     assert(reg != rscratch2, "invalid dst for register increment");
2993     movw(rscratch2, (unsigned)value);
2994     add(reg, reg, rscratch2);
2995   }
2996 }
2997 
2998 void MacroAssembler::incrementw(Address dst, int value)
2999 {
3000   assert(!dst.uses(rscratch1), "invalid dst for address increment");
3001   if (dst.getMode() == Address::literal) {
3002     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
3003     lea(rscratch2, dst);
3004     dst = Address(rscratch2);
3005   }
3006   ldrw(rscratch1, dst);
3007   incrementw(rscratch1, value);
3008   strw(rscratch1, dst);
3009 }
3010 
3011 void MacroAssembler::increment(Address dst, int value)
3012 {
3013   assert(!dst.uses(rscratch1), "invalid dst for address increment");
3014   if (dst.getMode() == Address::literal) {
3015     assert(abs(value) < (1 << 12), "invalid value and address mode combination");
3016     lea(rscratch2, dst);
3017     dst = Address(rscratch2);
3018   }
3019   ldr(rscratch1, dst);
3020   increment(rscratch1, value);
3021   str(rscratch1, dst);
3022 }
3023 
3024 // Push lots of registers in the bit set supplied.  Don't push sp.
3025 // Return the number of words pushed
3026 int MacroAssembler::push(unsigned int bitset, Register stack) {
3027   int words_pushed = 0;
3028 
3029   // Scan bitset to accumulate register pairs
3030   unsigned char regs[32];
3031   int count = 0;
3032   for (int reg = 0; reg <= 30; reg++) {
3033     if (1 & bitset)
3034       regs[count++] = reg;
3035     bitset >>= 1;
3036   }
3037   regs[count++] = zr->raw_encoding();
3038   count &= ~1;  // Only push an even number of regs
3039 
3040   if (count) {
3041     stp(as_Register(regs[0]), as_Register(regs[1]),
3042        Address(pre(stack, -count * wordSize)));
3043     words_pushed += 2;
3044   }
3045   for (int i = 2; i < count; i += 2) {
3046     stp(as_Register(regs[i]), as_Register(regs[i+1]),
3047        Address(stack, i * wordSize));
3048     words_pushed += 2;
3049   }
3050 
3051   assert(words_pushed == count, "oops, pushed != count");
3052 
3053   return count;
3054 }
3055 
3056 int MacroAssembler::pop(unsigned int bitset, Register stack) {
3057   int words_pushed = 0;
3058 
3059   // Scan bitset to accumulate register pairs
3060   unsigned char regs[32];
3061   int count = 0;
3062   for (int reg = 0; reg <= 30; reg++) {
3063     if (1 & bitset)
3064       regs[count++] = reg;
3065     bitset >>= 1;
3066   }
3067   regs[count++] = zr->raw_encoding();
3068   count &= ~1;
3069 
3070   for (int i = 2; i < count; i += 2) {
3071     ldp(as_Register(regs[i]), as_Register(regs[i+1]),
3072        Address(stack, i * wordSize));
3073     words_pushed += 2;
3074   }
3075   if (count) {
3076     ldp(as_Register(regs[0]), as_Register(regs[1]),
3077        Address(post(stack, count * wordSize)));
3078     words_pushed += 2;
3079   }
3080 
3081   assert(words_pushed == count, "oops, pushed != count");
3082 
3083   return count;
3084 }
3085 
3086 // Push lots of registers in the bit set supplied.  Don't push sp.
3087 // Return the number of dwords pushed
3088 int MacroAssembler::push_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3089   int words_pushed = 0;
3090   bool use_sve = false;
3091   int sve_vector_size_in_bytes = 0;
3092 
3093 #ifdef COMPILER2
3094   use_sve = Matcher::supports_scalable_vector();
3095   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3096 #endif
3097 
3098   // Scan bitset to accumulate register pairs
3099   unsigned char regs[32];
3100   int count = 0;
3101   for (int reg = 0; reg <= 31; reg++) {
3102     if (1 & bitset)
3103       regs[count++] = reg;
3104     bitset >>= 1;
3105   }
3106 
3107   if (count == 0) {
3108     return 0;
3109   }
3110 
3111   if (mode == PushPopFull) {
3112     if (use_sve && sve_vector_size_in_bytes > 16) {
3113       mode = PushPopSVE;
3114     } else {
3115       mode = PushPopNeon;
3116     }
3117   }
3118 
3119 #ifndef PRODUCT
3120   {
3121     char buffer[48];
3122     if (mode == PushPopSVE) {
3123       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d SVE registers", count);
3124     } else if (mode == PushPopNeon) {
3125       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d Neon registers", count);
3126     } else {
3127       os::snprintf_checked(buffer, sizeof(buffer), "push_fp: %d fp registers", count);
3128     }
3129     block_comment(buffer);
3130   }
3131 #endif
3132 
3133   if (mode == PushPopSVE) {
3134     sub(stack, stack, sve_vector_size_in_bytes * count);
3135     for (int i = 0; i < count; i++) {
3136       sve_str(as_FloatRegister(regs[i]), Address(stack, i));
3137     }
3138     return count * sve_vector_size_in_bytes / 8;
3139   }
3140 
3141   if (mode == PushPopNeon) {
3142     if (count == 1) {
3143       strq(as_FloatRegister(regs[0]), Address(pre(stack, -wordSize * 2)));
3144       return 2;
3145     }
3146 
3147     bool odd = (count & 1) == 1;
3148     int push_slots = count + (odd ? 1 : 0);
3149 
3150     // Always pushing full 128 bit registers.
3151     stpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize * 2)));
3152     words_pushed += 2;
3153 
3154     for (int i = 2; i + 1 < count; i += 2) {
3155       stpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3156       words_pushed += 2;
3157     }
3158 
3159     if (odd) {
3160       strq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3161       words_pushed++;
3162     }
3163 
3164     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3165     return count * 2;
3166   }
3167 
3168   if (mode == PushPopFp) {
3169     bool odd = (count & 1) == 1;
3170     int push_slots = count + (odd ? 1 : 0);
3171 
3172     if (count == 1) {
3173       // Stack pointer must be 16 bytes aligned
3174       strd(as_FloatRegister(regs[0]), Address(pre(stack, -push_slots * wordSize)));
3175       return 1;
3176     }
3177 
3178     stpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(pre(stack, -push_slots * wordSize)));
3179     words_pushed += 2;
3180 
3181     for (int i = 2; i + 1 < count; i += 2) {
3182       stpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3183       words_pushed += 2;
3184     }
3185 
3186     if (odd) {
3187       // Stack pointer must be 16 bytes aligned
3188       strd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3189       words_pushed++;
3190     }
3191 
3192     assert(words_pushed == count, "oops, pushed != count");
3193 
3194     return count;
3195   }
3196 
3197   return 0;
3198 }
3199 
3200 // Return the number of dwords popped
3201 int MacroAssembler::pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode) {
3202   int words_pushed = 0;
3203   bool use_sve = false;
3204   int sve_vector_size_in_bytes = 0;
3205 
3206 #ifdef COMPILER2
3207   use_sve = Matcher::supports_scalable_vector();
3208   sve_vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
3209 #endif
3210   // Scan bitset to accumulate register pairs
3211   unsigned char regs[32];
3212   int count = 0;
3213   for (int reg = 0; reg <= 31; reg++) {
3214     if (1 & bitset)
3215       regs[count++] = reg;
3216     bitset >>= 1;
3217   }
3218 
3219   if (count == 0) {
3220     return 0;
3221   }
3222 
3223   if (mode == PushPopFull) {
3224     if (use_sve && sve_vector_size_in_bytes > 16) {
3225       mode = PushPopSVE;
3226     } else {
3227       mode = PushPopNeon;
3228     }
3229   }
3230 
3231 #ifndef PRODUCT
3232   {
3233     char buffer[48];
3234     if (mode == PushPopSVE) {
3235       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d SVE registers", count);
3236     } else if (mode == PushPopNeon) {
3237       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d Neon registers", count);
3238     } else {
3239       os::snprintf_checked(buffer, sizeof(buffer), "pop_fp: %d fp registers", count);
3240     }
3241     block_comment(buffer);
3242   }
3243 #endif
3244 
3245   if (mode == PushPopSVE) {
3246     for (int i = count - 1; i >= 0; i--) {
3247       sve_ldr(as_FloatRegister(regs[i]), Address(stack, i));
3248     }
3249     add(stack, stack, sve_vector_size_in_bytes * count);
3250     return count * sve_vector_size_in_bytes / 8;
3251   }
3252 
3253   if (mode == PushPopNeon) {
3254     if (count == 1) {
3255       ldrq(as_FloatRegister(regs[0]), Address(post(stack, wordSize * 2)));
3256       return 2;
3257     }
3258 
3259     bool odd = (count & 1) == 1;
3260     int push_slots = count + (odd ? 1 : 0);
3261 
3262     if (odd) {
3263       ldrq(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize * 2));
3264       words_pushed++;
3265     }
3266 
3267     for (int i = 2; i + 1 < count; i += 2) {
3268       ldpq(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize * 2));
3269       words_pushed += 2;
3270     }
3271 
3272     ldpq(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize * 2)));
3273     words_pushed += 2;
3274 
3275     assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
3276 
3277     return count * 2;
3278   }
3279 
3280   if (mode == PushPopFp) {
3281     bool odd = (count & 1) == 1;
3282     int push_slots = count + (odd ? 1 : 0);
3283 
3284     if (count == 1) {
3285       ldrd(as_FloatRegister(regs[0]), Address(post(stack, push_slots * wordSize)));
3286       return 1;
3287     }
3288 
3289     if (odd) {
3290       ldrd(as_FloatRegister(regs[count - 1]), Address(stack, (count - 1) * wordSize));
3291       words_pushed++;
3292     }
3293 
3294     for (int i = 2; i + 1 < count; i += 2) {
3295       ldpd(as_FloatRegister(regs[i]), as_FloatRegister(regs[i+1]), Address(stack, i * wordSize));
3296       words_pushed += 2;
3297     }
3298 
3299     ldpd(as_FloatRegister(regs[0]), as_FloatRegister(regs[1]), Address(post(stack, push_slots * wordSize)));
3300     words_pushed += 2;
3301 
3302     assert(words_pushed == count, "oops, pushed != count");
3303 
3304     return count;
3305   }
3306 
3307   return 0;
3308 }
3309 
3310 // Return the number of dwords pushed
3311 int MacroAssembler::push_p(unsigned int bitset, Register stack) {
3312   bool use_sve = false;
3313   int sve_predicate_size_in_slots = 0;
3314 
3315 #ifdef COMPILER2
3316   use_sve = Matcher::supports_scalable_vector();
3317   if (use_sve) {
3318     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3319   }
3320 #endif
3321 
3322   if (!use_sve) {
3323     return 0;
3324   }
3325 
3326   unsigned char regs[PRegister::number_of_registers];
3327   int count = 0;
3328   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3329     if (1 & bitset)
3330       regs[count++] = reg;
3331     bitset >>= 1;
3332   }
3333 
3334   if (count == 0) {
3335     return 0;
3336   }
3337 
3338   int total_push_bytes = align_up(sve_predicate_size_in_slots *
3339                                   VMRegImpl::stack_slot_size * count, 16);
3340   sub(stack, stack, total_push_bytes);
3341   for (int i = 0; i < count; i++) {
3342     sve_str(as_PRegister(regs[i]), Address(stack, i));
3343   }
3344   return total_push_bytes / 8;
3345 }
3346 
3347 // Return the number of dwords popped
3348 int MacroAssembler::pop_p(unsigned int bitset, Register stack) {
3349   bool use_sve = false;
3350   int sve_predicate_size_in_slots = 0;
3351 
3352 #ifdef COMPILER2
3353   use_sve = Matcher::supports_scalable_vector();
3354   if (use_sve) {
3355     sve_predicate_size_in_slots = Matcher::scalable_predicate_reg_slots();
3356   }
3357 #endif
3358 
3359   if (!use_sve) {
3360     return 0;
3361   }
3362 
3363   unsigned char regs[PRegister::number_of_registers];
3364   int count = 0;
3365   for (int reg = 0; reg < PRegister::number_of_registers; reg++) {
3366     if (1 & bitset)
3367       regs[count++] = reg;
3368     bitset >>= 1;
3369   }
3370 
3371   if (count == 0) {
3372     return 0;
3373   }
3374 
3375   int total_pop_bytes = align_up(sve_predicate_size_in_slots *
3376                                  VMRegImpl::stack_slot_size * count, 16);
3377   for (int i = count - 1; i >= 0; i--) {
3378     sve_ldr(as_PRegister(regs[i]), Address(stack, i));
3379   }
3380   add(stack, stack, total_pop_bytes);
3381   return total_pop_bytes / 8;
3382 }
3383 
3384 #ifdef ASSERT
3385 void MacroAssembler::verify_heapbase(const char* msg) {
3386 #if 0
3387   assert (UseCompressedOops || UseCompressedClassPointers, "should be compressed");
3388   assert (Universe::heap() != nullptr, "java heap should be initialized");
3389   if (!UseCompressedOops || Universe::ptr_base() == nullptr) {
3390     // rheapbase is allocated as general register
3391     return;
3392   }
3393   if (CheckCompressedOops) {
3394     Label ok;
3395     push(1 << rscratch1->encoding(), sp); // cmpptr trashes rscratch1
3396     cmpptr(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3397     br(Assembler::EQ, ok);
3398     stop(msg);
3399     bind(ok);
3400     pop(1 << rscratch1->encoding(), sp);
3401   }
3402 #endif
3403 }
3404 #endif
3405 
3406 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
3407   assert_different_registers(value, tmp1, tmp2);
3408   Label done, tagged, weak_tagged;
3409 
3410   cbz(value, done);           // Use null as-is.
3411   tst(value, JNIHandles::tag_mask); // Test for tag.
3412   br(Assembler::NE, tagged);
3413 
3414   // Resolve local handle
3415   access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
3416   verify_oop(value);
3417   b(done);
3418 
3419   bind(tagged);
3420   STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
3421   tbnz(value, 0, weak_tagged);    // Test for weak tag.
3422 
3423   // Resolve global handle
3424   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3425   verify_oop(value);
3426   b(done);
3427 
3428   bind(weak_tagged);
3429   // Resolve jweak.
3430   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3431                  value, Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
3432   verify_oop(value);
3433 
3434   bind(done);
3435 }
3436 
3437 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
3438   assert_different_registers(value, tmp1, tmp2);
3439   Label done;
3440 
3441   cbz(value, done);           // Use null as-is.
3442 
3443 #ifdef ASSERT
3444   {
3445     STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
3446     Label valid_global_tag;
3447     tbnz(value, 1, valid_global_tag); // Test for global tag
3448     stop("non global jobject using resolve_global_jobject");
3449     bind(valid_global_tag);
3450   }
3451 #endif
3452 
3453   // Resolve global handle
3454   access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
3455   verify_oop(value);
3456 
3457   bind(done);
3458 }
3459 
3460 void MacroAssembler::stop(const char* msg) {
3461   // Skip AOT caching C strings in scratch buffer.
3462   const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
3463   BLOCK_COMMENT(str);
3464   // load msg into r0 so we can access it from the signal handler
3465   // ExternalAddress enables saving and restoring via the code cache
3466   lea(c_rarg0, ExternalAddress((address) str));
3467   dcps1(0xdeae);
3468 }
3469 
3470 void MacroAssembler::unimplemented(const char* what) {
3471   const char* buf = nullptr;
3472   {
3473     ResourceMark rm;
3474     stringStream ss;
3475     ss.print("unimplemented: %s", what);
3476     buf = code_string(ss.as_string());
3477   }
3478   stop(buf);
3479 }
3480 
3481 void MacroAssembler::_assert_asm(Assembler::Condition cc, const char* msg) {
3482 #ifdef ASSERT
3483   Label OK;
3484   br(cc, OK);
3485   stop(msg);
3486   bind(OK);
3487 #endif
3488 }
3489 
3490 // If a constant does not fit in an immediate field, generate some
3491 // number of MOV instructions and then perform the operation.
3492 void MacroAssembler::wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
3493                                            add_sub_imm_insn insn1,
3494                                            add_sub_reg_insn insn2,
3495                                            bool is32) {
3496   assert(Rd != zr, "Rd = zr and not setting flags?");
3497   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3498   if (fits) {
3499     (this->*insn1)(Rd, Rn, imm);
3500   } else {
3501     if (g_uabs(imm) < (1 << 24)) {
3502        (this->*insn1)(Rd, Rn, imm & -(1 << 12));
3503        (this->*insn1)(Rd, Rd, imm & ((1 << 12)-1));
3504     } else {
3505        assert_different_registers(Rd, Rn);
3506        mov(Rd, imm);
3507        (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3508     }
3509   }
3510 }
3511 
3512 // Separate vsn which sets the flags. Optimisations are more restricted
3513 // because we must set the flags correctly.
3514 void MacroAssembler::wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
3515                                              add_sub_imm_insn insn1,
3516                                              add_sub_reg_insn insn2,
3517                                              bool is32) {
3518   bool fits = operand_valid_for_add_sub_immediate(is32 ? (int32_t)imm : imm);
3519   if (fits) {
3520     (this->*insn1)(Rd, Rn, imm);
3521   } else {
3522     assert_different_registers(Rd, Rn);
3523     assert(Rd != zr, "overflow in immediate operand");
3524     mov(Rd, imm);
3525     (this->*insn2)(Rd, Rn, Rd, LSL, 0);
3526   }
3527 }
3528 
3529 
3530 void MacroAssembler::add(Register Rd, Register Rn, RegisterOrConstant increment) {
3531   if (increment.is_register()) {
3532     add(Rd, Rn, increment.as_register());
3533   } else {
3534     add(Rd, Rn, increment.as_constant());
3535   }
3536 }
3537 
3538 void MacroAssembler::addw(Register Rd, Register Rn, RegisterOrConstant increment) {
3539   if (increment.is_register()) {
3540     addw(Rd, Rn, increment.as_register());
3541   } else {
3542     addw(Rd, Rn, increment.as_constant());
3543   }
3544 }
3545 
3546 void MacroAssembler::sub(Register Rd, Register Rn, RegisterOrConstant decrement) {
3547   if (decrement.is_register()) {
3548     sub(Rd, Rn, decrement.as_register());
3549   } else {
3550     sub(Rd, Rn, decrement.as_constant());
3551   }
3552 }
3553 
3554 void MacroAssembler::subw(Register Rd, Register Rn, RegisterOrConstant decrement) {
3555   if (decrement.is_register()) {
3556     subw(Rd, Rn, decrement.as_register());
3557   } else {
3558     subw(Rd, Rn, decrement.as_constant());
3559   }
3560 }
3561 
3562 void MacroAssembler::reinit_heapbase()
3563 {
3564   if (UseCompressedOops) {
3565     if (Universe::is_fully_initialized()) {
3566       mov(rheapbase, CompressedOops::base());
3567     } else {
3568       lea(rheapbase, ExternalAddress(CompressedOops::base_addr()));
3569       ldr(rheapbase, Address(rheapbase));
3570     }
3571   }
3572 }
3573 
3574 // A generic CAS; success or failure is in the EQ flag.  A weak CAS
3575 // doesn't retry and may fail spuriously.  If the oldval is wanted,
3576 // Pass a register for the result, otherwise pass noreg.
3577 
3578 // Clobbers rscratch1
3579 void MacroAssembler::cmpxchg(Register addr, Register expected,
3580                              Register new_val,
3581                              enum operand_size size,
3582                              bool acquire, bool release,
3583                              bool weak,
3584                              Register result) {
3585   if (result == noreg)  result = rscratch1;
3586   BLOCK_COMMENT("cmpxchg {");
3587   if (UseLSE) {
3588     mov(result, expected);
3589     lse_cas(result, new_val, addr, size, acquire, release, /*not_pair*/ true);
3590     compare_eq(result, expected, size);
3591 #ifdef ASSERT
3592     // Poison rscratch1 which is written on !UseLSE branch
3593     mov(rscratch1, 0x1f1f1f1f1f1f1f1f);
3594 #endif
3595   } else {
3596     Label retry_load, done;
3597     prfm(Address(addr), PSTL1STRM);
3598     bind(retry_load);
3599     load_exclusive(result, addr, size, acquire);
3600     compare_eq(result, expected, size);
3601     br(Assembler::NE, done);
3602     store_exclusive(rscratch1, new_val, addr, size, release);
3603     if (weak) {
3604       cmpw(rscratch1, 0u);  // If the store fails, return NE to our caller.
3605     } else {
3606       cbnzw(rscratch1, retry_load);
3607     }
3608     bind(done);
3609   }
3610   BLOCK_COMMENT("} cmpxchg");
3611 }
3612 
3613 // A generic comparison. Only compares for equality, clobbers rscratch1.
3614 void MacroAssembler::compare_eq(Register rm, Register rn, enum operand_size size) {
3615   if (size == xword) {
3616     cmp(rm, rn);
3617   } else if (size == word) {
3618     cmpw(rm, rn);
3619   } else if (size == halfword) {
3620     eorw(rscratch1, rm, rn);
3621     ands(zr, rscratch1, 0xffff);
3622   } else if (size == byte) {
3623     eorw(rscratch1, rm, rn);
3624     ands(zr, rscratch1, 0xff);
3625   } else {
3626     ShouldNotReachHere();
3627   }
3628 }
3629 
3630 
3631 static bool different(Register a, RegisterOrConstant b, Register c) {
3632   if (b.is_constant())
3633     return a != c;
3634   else
3635     return a != b.as_register() && a != c && b.as_register() != c;
3636 }
3637 
3638 #define ATOMIC_OP(NAME, LDXR, OP, IOP, AOP, STXR, sz)                   \
3639 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3640   if (UseLSE) {                                                         \
3641     prev = prev->is_valid() ? prev : zr;                                \
3642     if (incr.is_register()) {                                           \
3643       AOP(sz, incr.as_register(), prev, addr);                          \
3644     } else {                                                            \
3645       mov(rscratch2, incr.as_constant());                               \
3646       AOP(sz, rscratch2, prev, addr);                                   \
3647     }                                                                   \
3648     return;                                                             \
3649   }                                                                     \
3650   Register result = rscratch2;                                          \
3651   if (prev->is_valid())                                                 \
3652     result = different(prev, incr, addr) ? prev : rscratch2;            \
3653                                                                         \
3654   Label retry_load;                                                     \
3655   prfm(Address(addr), PSTL1STRM);                                       \
3656   bind(retry_load);                                                     \
3657   LDXR(result, addr);                                                   \
3658   OP(rscratch1, result, incr);                                          \
3659   STXR(rscratch2, rscratch1, addr);                                     \
3660   cbnzw(rscratch2, retry_load);                                         \
3661   if (prev->is_valid() && prev != result) {                             \
3662     IOP(prev, rscratch1, incr);                                         \
3663   }                                                                     \
3664 }
3665 
3666 ATOMIC_OP(add, ldxr, add, sub, ldadd, stxr, Assembler::xword)
3667 ATOMIC_OP(addw, ldxrw, addw, subw, ldadd, stxrw, Assembler::word)
3668 ATOMIC_OP(addal, ldaxr, add, sub, ldaddal, stlxr, Assembler::xword)
3669 ATOMIC_OP(addalw, ldaxrw, addw, subw, ldaddal, stlxrw, Assembler::word)
3670 
3671 #undef ATOMIC_OP
3672 
3673 #define ATOMIC_XCHG(OP, AOP, LDXR, STXR, sz)                            \
3674 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3675   if (UseLSE) {                                                         \
3676     prev = prev->is_valid() ? prev : zr;                                \
3677     AOP(sz, newv, prev, addr);                                          \
3678     return;                                                             \
3679   }                                                                     \
3680   Register result = rscratch2;                                          \
3681   if (prev->is_valid())                                                 \
3682     result = different(prev, newv, addr) ? prev : rscratch2;            \
3683                                                                         \
3684   Label retry_load;                                                     \
3685   prfm(Address(addr), PSTL1STRM);                                       \
3686   bind(retry_load);                                                     \
3687   LDXR(result, addr);                                                   \
3688   STXR(rscratch1, newv, addr);                                          \
3689   cbnzw(rscratch1, retry_load);                                         \
3690   if (prev->is_valid() && prev != result)                               \
3691     mov(prev, result);                                                  \
3692 }
3693 
3694 ATOMIC_XCHG(xchg, swp, ldxr, stxr, Assembler::xword)
3695 ATOMIC_XCHG(xchgw, swp, ldxrw, stxrw, Assembler::word)
3696 ATOMIC_XCHG(xchgl, swpl, ldxr, stlxr, Assembler::xword)
3697 ATOMIC_XCHG(xchglw, swpl, ldxrw, stlxrw, Assembler::word)
3698 ATOMIC_XCHG(xchgal, swpal, ldaxr, stlxr, Assembler::xword)
3699 ATOMIC_XCHG(xchgalw, swpal, ldaxrw, stlxrw, Assembler::word)
3700 
3701 #undef ATOMIC_XCHG
3702 
3703 #ifndef PRODUCT
3704 extern "C" void findpc(intptr_t x);
3705 #endif
3706 
3707 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
3708 {
3709   // In order to get locks to work, we need to fake a in_VM state
3710   if (ShowMessageBoxOnError) {
3711     JavaThread* thread = JavaThread::current();
3712     thread->set_thread_state(_thread_in_vm);
3713 #ifndef PRODUCT
3714     if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
3715       ttyLocker ttyl;
3716       BytecodeCounter::print();
3717     }
3718 #endif
3719     if (os::message_box(msg, "Execution stopped, print registers?")) {
3720       ttyLocker ttyl;
3721       tty->print_cr(" pc = 0x%016" PRIx64, pc);
3722 #ifndef PRODUCT
3723       tty->cr();
3724       findpc(pc);
3725       tty->cr();
3726 #endif
3727       tty->print_cr(" r0 = 0x%016" PRIx64, regs[0]);
3728       tty->print_cr(" r1 = 0x%016" PRIx64, regs[1]);
3729       tty->print_cr(" r2 = 0x%016" PRIx64, regs[2]);
3730       tty->print_cr(" r3 = 0x%016" PRIx64, regs[3]);
3731       tty->print_cr(" r4 = 0x%016" PRIx64, regs[4]);
3732       tty->print_cr(" r5 = 0x%016" PRIx64, regs[5]);
3733       tty->print_cr(" r6 = 0x%016" PRIx64, regs[6]);
3734       tty->print_cr(" r7 = 0x%016" PRIx64, regs[7]);
3735       tty->print_cr(" r8 = 0x%016" PRIx64, regs[8]);
3736       tty->print_cr(" r9 = 0x%016" PRIx64, regs[9]);
3737       tty->print_cr("r10 = 0x%016" PRIx64, regs[10]);
3738       tty->print_cr("r11 = 0x%016" PRIx64, regs[11]);
3739       tty->print_cr("r12 = 0x%016" PRIx64, regs[12]);
3740       tty->print_cr("r13 = 0x%016" PRIx64, regs[13]);
3741       tty->print_cr("r14 = 0x%016" PRIx64, regs[14]);
3742       tty->print_cr("r15 = 0x%016" PRIx64, regs[15]);
3743       tty->print_cr("r16 = 0x%016" PRIx64, regs[16]);
3744       tty->print_cr("r17 = 0x%016" PRIx64, regs[17]);
3745       tty->print_cr("r18 = 0x%016" PRIx64, regs[18]);
3746       tty->print_cr("r19 = 0x%016" PRIx64, regs[19]);
3747       tty->print_cr("r20 = 0x%016" PRIx64, regs[20]);
3748       tty->print_cr("r21 = 0x%016" PRIx64, regs[21]);
3749       tty->print_cr("r22 = 0x%016" PRIx64, regs[22]);
3750       tty->print_cr("r23 = 0x%016" PRIx64, regs[23]);
3751       tty->print_cr("r24 = 0x%016" PRIx64, regs[24]);
3752       tty->print_cr("r25 = 0x%016" PRIx64, regs[25]);
3753       tty->print_cr("r26 = 0x%016" PRIx64, regs[26]);
3754       tty->print_cr("r27 = 0x%016" PRIx64, regs[27]);
3755       tty->print_cr("r28 = 0x%016" PRIx64, regs[28]);
3756       tty->print_cr("r30 = 0x%016" PRIx64, regs[30]);
3757       tty->print_cr("r31 = 0x%016" PRIx64, regs[31]);
3758       BREAKPOINT;
3759     }
3760   }
3761   fatal("DEBUG MESSAGE: %s", msg);
3762 }
3763 
3764 RegSet MacroAssembler::call_clobbered_gp_registers() {
3765   RegSet regs = RegSet::range(r0, r17) - RegSet::of(rscratch1, rscratch2);
3766 #ifndef R18_RESERVED
3767   regs += r18_tls;
3768 #endif
3769   return regs;
3770 }
3771 
3772 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
3773   int step = 4 * wordSize;
3774   push(call_clobbered_gp_registers() - exclude, sp);
3775   sub(sp, sp, step);
3776   mov(rscratch1, -step);
3777   // Push v0-v7, v16-v31.
3778   for (int i = 31; i>= 4; i -= 4) {
3779     if (i <= v7->encoding() || i >= v16->encoding())
3780       st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
3781           as_FloatRegister(i), T1D, Address(post(sp, rscratch1)));
3782   }
3783   st1(as_FloatRegister(0), as_FloatRegister(1), as_FloatRegister(2),
3784       as_FloatRegister(3), T1D, Address(sp));
3785 }
3786 
3787 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
3788   for (int i = 0; i < 32; i += 4) {
3789     if (i <= v7->encoding() || i >= v16->encoding())
3790       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3791           as_FloatRegister(i+3), T1D, Address(post(sp, 4 * wordSize)));
3792   }
3793 
3794   reinitialize_ptrue();
3795 
3796   pop(call_clobbered_gp_registers() - exclude, sp);
3797 }
3798 
3799 void MacroAssembler::push_CPU_state(bool save_vectors, bool use_sve,
3800                                     int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3801   push(RegSet::range(r0, r29), sp); // integer registers except lr & sp
3802   if (save_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3803     sub(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3804     for (int i = 0; i < FloatRegister::number_of_registers; i++) {
3805       sve_str(as_FloatRegister(i), Address(sp, i));
3806     }
3807   } else {
3808     int step = (save_vectors ? 8 : 4) * wordSize;
3809     mov(rscratch1, -step);
3810     sub(sp, sp, step);
3811     for (int i = 28; i >= 4; i -= 4) {
3812       st1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3813           as_FloatRegister(i+3), save_vectors ? T2D : T1D, Address(post(sp, rscratch1)));
3814     }
3815     st1(v0, v1, v2, v3, save_vectors ? T2D : T1D, sp);
3816   }
3817   if (save_vectors && use_sve && total_predicate_in_bytes > 0) {
3818     sub(sp, sp, total_predicate_in_bytes);
3819     for (int i = 0; i < PRegister::number_of_registers; i++) {
3820       sve_str(as_PRegister(i), Address(sp, i));
3821     }
3822   }
3823 }
3824 
3825 void MacroAssembler::pop_CPU_state(bool restore_vectors, bool use_sve,
3826                                    int sve_vector_size_in_bytes, int total_predicate_in_bytes) {
3827   if (restore_vectors && use_sve && total_predicate_in_bytes > 0) {
3828     for (int i = PRegister::number_of_registers - 1; i >= 0; i--) {
3829       sve_ldr(as_PRegister(i), Address(sp, i));
3830     }
3831     add(sp, sp, total_predicate_in_bytes);
3832   }
3833   if (restore_vectors && use_sve && sve_vector_size_in_bytes > 16) {
3834     for (int i = FloatRegister::number_of_registers - 1; i >= 0; i--) {
3835       sve_ldr(as_FloatRegister(i), Address(sp, i));
3836     }
3837     add(sp, sp, sve_vector_size_in_bytes * FloatRegister::number_of_registers);
3838   } else {
3839     int step = (restore_vectors ? 8 : 4) * wordSize;
3840     for (int i = 0; i <= 28; i += 4)
3841       ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
3842           as_FloatRegister(i+3), restore_vectors ? T2D : T1D, Address(post(sp, step)));
3843   }
3844 
3845   // We may use predicate registers and rely on ptrue with SVE,
3846   // regardless of wide vector (> 8 bytes) used or not.
3847   if (use_sve) {
3848     reinitialize_ptrue();
3849   }
3850 
3851   // integer registers except lr & sp
3852   pop(RegSet::range(r0, r17), sp);
3853 #ifdef R18_RESERVED
3854   ldp(zr, r19, Address(post(sp, 2 * wordSize)));
3855   pop(RegSet::range(r20, r29), sp);
3856 #else
3857   pop(RegSet::range(r18_tls, r29), sp);
3858 #endif
3859 }
3860 
3861 /**
3862  * Helpers for multiply_to_len().
3863  */
3864 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
3865                                      Register src1, Register src2) {
3866   adds(dest_lo, dest_lo, src1);
3867   adc(dest_hi, dest_hi, zr);
3868   adds(dest_lo, dest_lo, src2);
3869   adc(final_dest_hi, dest_hi, zr);
3870 }
3871 
3872 // Generate an address from (r + r1 extend offset).  "size" is the
3873 // size of the operand.  The result may be in rscratch2.
3874 Address MacroAssembler::offsetted_address(Register r, Register r1,
3875                                           Address::extend ext, int offset, int size) {
3876   if (offset || (ext.shift() % size != 0)) {
3877     lea(rscratch2, Address(r, r1, ext));
3878     return Address(rscratch2, offset);
3879   } else {
3880     return Address(r, r1, ext);
3881   }
3882 }
3883 
3884 Address MacroAssembler::spill_address(int size, int offset, Register tmp)
3885 {
3886   assert(offset >= 0, "spill to negative address?");
3887   // Offset reachable ?
3888   //   Not aligned - 9 bits signed offset
3889   //   Aligned - 12 bits unsigned offset shifted
3890   Register base = sp;
3891   if ((offset & (size-1)) && offset >= (1<<8)) {
3892     add(tmp, base, offset & ((1<<12)-1));
3893     base = tmp;
3894     offset &= -1u<<12;
3895   }
3896 
3897   if (offset >= (1<<12) * size) {
3898     add(tmp, base, offset & (((1<<12)-1)<<12));
3899     base = tmp;
3900     offset &= ~(((1<<12)-1)<<12);
3901   }
3902 
3903   return Address(base, offset);
3904 }
3905 
3906 Address MacroAssembler::sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp) {
3907   assert(offset >= 0, "spill to negative address?");
3908 
3909   Register base = sp;
3910 
3911   // An immediate offset in the range 0 to 255 which is multiplied
3912   // by the current vector or predicate register size in bytes.
3913   if (offset % sve_reg_size_in_bytes == 0 && offset < ((1<<8)*sve_reg_size_in_bytes)) {
3914     return Address(base, offset / sve_reg_size_in_bytes);
3915   }
3916 
3917   add(tmp, base, offset);
3918   return Address(tmp);
3919 }
3920 
3921 // Checks whether offset is aligned.
3922 // Returns true if it is, else false.
3923 bool MacroAssembler::merge_alignment_check(Register base,
3924                                            size_t size,
3925                                            int64_t cur_offset,
3926                                            int64_t prev_offset) const {
3927   if (AvoidUnalignedAccesses) {
3928     if (base == sp) {
3929       // Checks whether low offset if aligned to pair of registers.
3930       int64_t pair_mask = size * 2 - 1;
3931       int64_t offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3932       return (offset & pair_mask) == 0;
3933     } else { // If base is not sp, we can't guarantee the access is aligned.
3934       return false;
3935     }
3936   } else {
3937     int64_t mask = size - 1;
3938     // Load/store pair instruction only supports element size aligned offset.
3939     return (cur_offset & mask) == 0 && (prev_offset & mask) == 0;
3940   }
3941 }
3942 
3943 // Checks whether current and previous loads/stores can be merged.
3944 // Returns true if it can be merged, else false.
3945 bool MacroAssembler::ldst_can_merge(Register rt,
3946                                     const Address &adr,
3947                                     size_t cur_size_in_bytes,
3948                                     bool is_store) const {
3949   address prev = pc() - NativeInstruction::instruction_size;
3950   address last = code()->last_insn();
3951 
3952   if (last == nullptr || !nativeInstruction_at(last)->is_Imm_LdSt()) {
3953     return false;
3954   }
3955 
3956   if (adr.getMode() != Address::base_plus_offset || prev != last) {
3957     return false;
3958   }
3959 
3960   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
3961   size_t prev_size_in_bytes = prev_ldst->size_in_bytes();
3962 
3963   assert(prev_size_in_bytes == 4 || prev_size_in_bytes == 8, "only supports 64/32bit merging.");
3964   assert(cur_size_in_bytes == 4 || cur_size_in_bytes == 8, "only supports 64/32bit merging.");
3965 
3966   if (cur_size_in_bytes != prev_size_in_bytes || is_store != prev_ldst->is_store()) {
3967     return false;
3968   }
3969 
3970   int64_t max_offset = 63 * prev_size_in_bytes;
3971   int64_t min_offset = -64 * prev_size_in_bytes;
3972 
3973   assert(prev_ldst->is_not_pre_post_index(), "pre-index or post-index is not supported to be merged.");
3974 
3975   // Only same base can be merged.
3976   if (adr.base() != prev_ldst->base()) {
3977     return false;
3978   }
3979 
3980   int64_t cur_offset = adr.offset();
3981   int64_t prev_offset = prev_ldst->offset();
3982   size_t diff = abs(cur_offset - prev_offset);
3983   if (diff != prev_size_in_bytes) {
3984     return false;
3985   }
3986 
3987   // Following cases can not be merged:
3988   // ldr x2, [x2, #8]
3989   // ldr x3, [x2, #16]
3990   // or:
3991   // ldr x2, [x3, #8]
3992   // ldr x2, [x3, #16]
3993   // If t1 and t2 is the same in "ldp t1, t2, [xn, #imm]", we'll get SIGILL.
3994   if (!is_store && (adr.base() == prev_ldst->target() || rt == prev_ldst->target())) {
3995     return false;
3996   }
3997 
3998   int64_t low_offset = prev_offset > cur_offset ? cur_offset : prev_offset;
3999   // Offset range must be in ldp/stp instruction's range.
4000   if (low_offset > max_offset || low_offset < min_offset) {
4001     return false;
4002   }
4003 
4004   if (merge_alignment_check(adr.base(), prev_size_in_bytes, cur_offset, prev_offset)) {
4005     return true;
4006   }
4007 
4008   return false;
4009 }
4010 
4011 // Merge current load/store with previous load/store into ldp/stp.
4012 void MacroAssembler::merge_ldst(Register rt,
4013                                 const Address &adr,
4014                                 size_t cur_size_in_bytes,
4015                                 bool is_store) {
4016 
4017   assert(ldst_can_merge(rt, adr, cur_size_in_bytes, is_store) == true, "cur and prev must be able to be merged.");
4018 
4019   Register rt_low, rt_high;
4020   address prev = pc() - NativeInstruction::instruction_size;
4021   NativeLdSt* prev_ldst = NativeLdSt_at(prev);
4022 
4023   int64_t offset;
4024 
4025   if (adr.offset() < prev_ldst->offset()) {
4026     offset = adr.offset();
4027     rt_low = rt;
4028     rt_high = prev_ldst->target();
4029   } else {
4030     offset = prev_ldst->offset();
4031     rt_low = prev_ldst->target();
4032     rt_high = rt;
4033   }
4034 
4035   Address adr_p = Address(prev_ldst->base(), offset);
4036   // Overwrite previous generated binary.
4037   code_section()->set_end(prev);
4038 
4039   const size_t sz = prev_ldst->size_in_bytes();
4040   assert(sz == 8 || sz == 4, "only supports 64/32bit merging.");
4041   if (!is_store) {
4042     BLOCK_COMMENT("merged ldr pair");
4043     if (sz == 8) {
4044       ldp(rt_low, rt_high, adr_p);
4045     } else {
4046       ldpw(rt_low, rt_high, adr_p);
4047     }
4048   } else {
4049     BLOCK_COMMENT("merged str pair");
4050     if (sz == 8) {
4051       stp(rt_low, rt_high, adr_p);
4052     } else {
4053       stpw(rt_low, rt_high, adr_p);
4054     }
4055   }
4056 }
4057 
4058 /**
4059  * Multiply 64 bit by 64 bit first loop.
4060  */
4061 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
4062                                            Register y, Register y_idx, Register z,
4063                                            Register carry, Register product,
4064                                            Register idx, Register kdx) {
4065   //
4066   //  jlong carry, x[], y[], z[];
4067   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4068   //    huge_128 product = y[idx] * x[xstart] + carry;
4069   //    z[kdx] = (jlong)product;
4070   //    carry  = (jlong)(product >>> 64);
4071   //  }
4072   //  z[xstart] = carry;
4073   //
4074 
4075   Label L_first_loop, L_first_loop_exit;
4076   Label L_one_x, L_one_y, L_multiply;
4077 
4078   subsw(xstart, xstart, 1);
4079   br(Assembler::MI, L_one_x);
4080 
4081   lea(rscratch1, Address(x, xstart, Address::lsl(LogBytesPerInt)));
4082   ldr(x_xstart, Address(rscratch1));
4083   ror(x_xstart, x_xstart, 32); // convert big-endian to little-endian
4084 
4085   bind(L_first_loop);
4086   subsw(idx, idx, 1);
4087   br(Assembler::MI, L_first_loop_exit);
4088   subsw(idx, idx, 1);
4089   br(Assembler::MI, L_one_y);
4090   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4091   ldr(y_idx, Address(rscratch1));
4092   ror(y_idx, y_idx, 32); // convert big-endian to little-endian
4093   bind(L_multiply);
4094 
4095   // AArch64 has a multiply-accumulate instruction that we can't use
4096   // here because it has no way to process carries, so we have to use
4097   // separate add and adc instructions.  Bah.
4098   umulh(rscratch1, x_xstart, y_idx); // x_xstart * y_idx -> rscratch1:product
4099   mul(product, x_xstart, y_idx);
4100   adds(product, product, carry);
4101   adc(carry, rscratch1, zr);   // x_xstart * y_idx + carry -> carry:product
4102 
4103   subw(kdx, kdx, 2);
4104   ror(product, product, 32); // back to big-endian
4105   str(product, offsetted_address(z, kdx, Address::uxtw(LogBytesPerInt), 0, BytesPerLong));
4106 
4107   b(L_first_loop);
4108 
4109   bind(L_one_y);
4110   ldrw(y_idx, Address(y,  0));
4111   b(L_multiply);
4112 
4113   bind(L_one_x);
4114   ldrw(x_xstart, Address(x,  0));
4115   b(L_first_loop);
4116 
4117   bind(L_first_loop_exit);
4118 }
4119 
4120 /**
4121  * Multiply 128 bit by 128. Unrolled inner loop.
4122  *
4123  */
4124 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4125                                              Register carry, Register carry2,
4126                                              Register idx, Register jdx,
4127                                              Register yz_idx1, Register yz_idx2,
4128                                              Register tmp, Register tmp3, Register tmp4,
4129                                              Register tmp6, Register product_hi) {
4130 
4131   //   jlong carry, x[], y[], z[];
4132   //   int kdx = ystart+1;
4133   //   for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4134   //     huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4135   //     jlong carry2  = (jlong)(tmp3 >>> 64);
4136   //     huge_128 tmp4 = (y[idx]   * product_hi) + z[kdx+idx] + carry2;
4137   //     carry  = (jlong)(tmp4 >>> 64);
4138   //     z[kdx+idx+1] = (jlong)tmp3;
4139   //     z[kdx+idx] = (jlong)tmp4;
4140   //   }
4141   //   idx += 2;
4142   //   if (idx > 0) {
4143   //     yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4144   //     z[kdx+idx] = (jlong)yz_idx1;
4145   //     carry  = (jlong)(yz_idx1 >>> 64);
4146   //   }
4147   //
4148 
4149   Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4150 
4151   lsrw(jdx, idx, 2);
4152 
4153   bind(L_third_loop);
4154 
4155   subsw(jdx, jdx, 1);
4156   br(Assembler::MI, L_third_loop_exit);
4157   subw(idx, idx, 4);
4158 
4159   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4160 
4161   ldp(yz_idx2, yz_idx1, Address(rscratch1, 0));
4162 
4163   lea(tmp6, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4164 
4165   ror(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4166   ror(yz_idx2, yz_idx2, 32);
4167 
4168   ldp(rscratch2, rscratch1, Address(tmp6, 0));
4169 
4170   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4171   umulh(tmp4, product_hi, yz_idx1);
4172 
4173   ror(rscratch1, rscratch1, 32); // convert big-endian to little-endian
4174   ror(rscratch2, rscratch2, 32);
4175 
4176   mul(tmp, product_hi, yz_idx2);   //  yz_idx2 * product_hi -> carry2:tmp
4177   umulh(carry2, product_hi, yz_idx2);
4178 
4179   // propagate sum of both multiplications into carry:tmp4:tmp3
4180   adds(tmp3, tmp3, carry);
4181   adc(tmp4, tmp4, zr);
4182   adds(tmp3, tmp3, rscratch1);
4183   adcs(tmp4, tmp4, tmp);
4184   adc(carry, carry2, zr);
4185   adds(tmp4, tmp4, rscratch2);
4186   adc(carry, carry, zr);
4187 
4188   ror(tmp3, tmp3, 32); // convert little-endian to big-endian
4189   ror(tmp4, tmp4, 32);
4190   stp(tmp4, tmp3, Address(tmp6, 0));
4191 
4192   b(L_third_loop);
4193   bind (L_third_loop_exit);
4194 
4195   andw (idx, idx, 0x3);
4196   cbz(idx, L_post_third_loop_done);
4197 
4198   Label L_check_1;
4199   subsw(idx, idx, 2);
4200   br(Assembler::MI, L_check_1);
4201 
4202   lea(rscratch1, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4203   ldr(yz_idx1, Address(rscratch1, 0));
4204   ror(yz_idx1, yz_idx1, 32);
4205   mul(tmp3, product_hi, yz_idx1);  //  yz_idx1 * product_hi -> tmp4:tmp3
4206   umulh(tmp4, product_hi, yz_idx1);
4207   lea(rscratch1, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4208   ldr(yz_idx2, Address(rscratch1, 0));
4209   ror(yz_idx2, yz_idx2, 32);
4210 
4211   add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2);
4212 
4213   ror(tmp3, tmp3, 32);
4214   str(tmp3, Address(rscratch1, 0));
4215 
4216   bind (L_check_1);
4217 
4218   andw (idx, idx, 0x1);
4219   subsw(idx, idx, 1);
4220   br(Assembler::MI, L_post_third_loop_done);
4221   ldrw(tmp4, Address(y, idx, Address::uxtw(LogBytesPerInt)));
4222   mul(tmp3, tmp4, product_hi);  //  tmp4 * product_hi -> carry2:tmp3
4223   umulh(carry2, tmp4, product_hi);
4224   ldrw(tmp4, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4225 
4226   add2_with_carry(carry2, tmp3, tmp4, carry);
4227 
4228   strw(tmp3, Address(z, idx, Address::uxtw(LogBytesPerInt)));
4229   extr(carry, carry2, tmp3, 32);
4230 
4231   bind(L_post_third_loop_done);
4232 }
4233 
4234 /**
4235  * Code for BigInteger::multiplyToLen() intrinsic.
4236  *
4237  * r0: x
4238  * r1: xlen
4239  * r2: y
4240  * r3: ylen
4241  * r4:  z
4242  * r5: tmp0
4243  * r10: tmp1
4244  * r11: tmp2
4245  * r12: tmp3
4246  * r13: tmp4
4247  * r14: tmp5
4248  * r15: tmp6
4249  * r16: tmp7
4250  *
4251  */
4252 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4253                                      Register z, Register tmp0,
4254                                      Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4255                                      Register tmp5, Register tmp6, Register product_hi) {
4256 
4257   assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, product_hi);
4258 
4259   const Register idx = tmp1;
4260   const Register kdx = tmp2;
4261   const Register xstart = tmp3;
4262 
4263   const Register y_idx = tmp4;
4264   const Register carry = tmp5;
4265   const Register product  = xlen;
4266   const Register x_xstart = tmp0;
4267 
4268   // First Loop.
4269   //
4270   //  final static long LONG_MASK = 0xffffffffL;
4271   //  int xstart = xlen - 1;
4272   //  int ystart = ylen - 1;
4273   //  long carry = 0;
4274   //  for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
4275   //    long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
4276   //    z[kdx] = (int)product;
4277   //    carry = product >>> 32;
4278   //  }
4279   //  z[xstart] = (int)carry;
4280   //
4281 
4282   movw(idx, ylen);       // idx = ylen;
4283   addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4284   mov(carry, zr);        // carry = 0;
4285 
4286   Label L_done;
4287 
4288   movw(xstart, xlen);
4289   subsw(xstart, xstart, 1);
4290   br(Assembler::MI, L_done);
4291 
4292   multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4293 
4294   Label L_second_loop;
4295   cbzw(kdx, L_second_loop);
4296 
4297   Label L_carry;
4298   subw(kdx, kdx, 1);
4299   cbzw(kdx, L_carry);
4300 
4301   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4302   lsr(carry, carry, 32);
4303   subw(kdx, kdx, 1);
4304 
4305   bind(L_carry);
4306   strw(carry, Address(z, kdx, Address::uxtw(LogBytesPerInt)));
4307 
4308   // Second and third (nested) loops.
4309   //
4310   // for (int i = xstart-1; i >= 0; i--) { // Second loop
4311   //   carry = 0;
4312   //   for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4313   //     long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4314   //                    (z[k] & LONG_MASK) + carry;
4315   //     z[k] = (int)product;
4316   //     carry = product >>> 32;
4317   //   }
4318   //   z[i] = (int)carry;
4319   // }
4320   //
4321   // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4322 
4323   const Register jdx = tmp1;
4324 
4325   bind(L_second_loop);
4326   mov(carry, zr);                // carry = 0;
4327   movw(jdx, ylen);               // j = ystart+1
4328 
4329   subsw(xstart, xstart, 1);      // i = xstart-1;
4330   br(Assembler::MI, L_done);
4331 
4332   str(z, Address(pre(sp, -4 * wordSize)));
4333 
4334   Label L_last_x;
4335   lea(z, offsetted_address(z, xstart, Address::uxtw(LogBytesPerInt), 4, BytesPerInt)); // z = z + k - j
4336   subsw(xstart, xstart, 1);       // i = xstart-1;
4337   br(Assembler::MI, L_last_x);
4338 
4339   lea(rscratch1, Address(x, xstart, Address::uxtw(LogBytesPerInt)));
4340   ldr(product_hi, Address(rscratch1));
4341   ror(product_hi, product_hi, 32);  // convert big-endian to little-endian
4342 
4343   Label L_third_loop_prologue;
4344   bind(L_third_loop_prologue);
4345 
4346   str(ylen, Address(sp, wordSize));
4347   stp(x, xstart, Address(sp, 2 * wordSize));
4348   multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4349                           tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4350   ldp(z, ylen, Address(post(sp, 2 * wordSize)));
4351   ldp(x, xlen, Address(post(sp, 2 * wordSize)));   // copy old xstart -> xlen
4352 
4353   addw(tmp3, xlen, 1);
4354   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4355   subsw(tmp3, tmp3, 1);
4356   br(Assembler::MI, L_done);
4357 
4358   lsr(carry, carry, 32);
4359   strw(carry, Address(z, tmp3, Address::uxtw(LogBytesPerInt)));
4360   b(L_second_loop);
4361 
4362   // Next infrequent code is moved outside loops.
4363   bind(L_last_x);
4364   ldrw(product_hi, Address(x,  0));
4365   b(L_third_loop_prologue);
4366 
4367   bind(L_done);
4368 }
4369 
4370 // Code for BigInteger::mulAdd intrinsic
4371 // out     = r0
4372 // in      = r1
4373 // offset  = r2  (already out.length-offset)
4374 // len     = r3
4375 // k       = r4
4376 //
4377 // pseudo code from java implementation:
4378 // carry = 0;
4379 // offset = out.length-offset - 1;
4380 // for (int j=len-1; j >= 0; j--) {
4381 //     product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4382 //     out[offset--] = (int)product;
4383 //     carry = product >>> 32;
4384 // }
4385 // return (int)carry;
4386 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4387       Register len, Register k) {
4388     Label LOOP, END;
4389     // pre-loop
4390     cmp(len, zr); // cmp, not cbz/cbnz: to use condition twice => less branches
4391     csel(out, zr, out, Assembler::EQ);
4392     br(Assembler::EQ, END);
4393     add(in, in, len, LSL, 2); // in[j+1] address
4394     add(offset, out, offset, LSL, 2); // out[offset + 1] address
4395     mov(out, zr); // used to keep carry now
4396     BIND(LOOP);
4397     ldrw(rscratch1, Address(pre(in, -4)));
4398     madd(rscratch1, rscratch1, k, out);
4399     ldrw(rscratch2, Address(pre(offset, -4)));
4400     add(rscratch1, rscratch1, rscratch2);
4401     strw(rscratch1, Address(offset));
4402     lsr(out, rscratch1, 32);
4403     subs(len, len, 1);
4404     br(Assembler::NE, LOOP);
4405     BIND(END);
4406 }
4407 
4408 /**
4409  * Emits code to update CRC-32 with a byte value according to constants in table
4410  *
4411  * @param [in,out]crc   Register containing the crc.
4412  * @param [in]val       Register containing the byte to fold into the CRC.
4413  * @param [in]table     Register containing the table of crc constants.
4414  *
4415  * uint32_t crc;
4416  * val = crc_table[(val ^ crc) & 0xFF];
4417  * crc = val ^ (crc >> 8);
4418  *
4419  */
4420 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
4421   eor(val, val, crc);
4422   andr(val, val, 0xff);
4423   ldrw(val, Address(table, val, Address::lsl(2)));
4424   eor(crc, val, crc, Assembler::LSR, 8);
4425 }
4426 
4427 /**
4428  * Emits code to update CRC-32 with a 32-bit value according to tables 0 to 3
4429  *
4430  * @param [in,out]crc   Register containing the crc.
4431  * @param [in]v         Register containing the 32-bit to fold into the CRC.
4432  * @param [in]table0    Register containing table 0 of crc constants.
4433  * @param [in]table1    Register containing table 1 of crc constants.
4434  * @param [in]table2    Register containing table 2 of crc constants.
4435  * @param [in]table3    Register containing table 3 of crc constants.
4436  *
4437  * uint32_t crc;
4438  *   v = crc ^ v
4439  *   crc = table3[v&0xff]^table2[(v>>8)&0xff]^table1[(v>>16)&0xff]^table0[v>>24]
4440  *
4441  */
4442 void MacroAssembler::update_word_crc32(Register crc, Register v, Register tmp,
4443         Register table0, Register table1, Register table2, Register table3,
4444         bool upper) {
4445   eor(v, crc, v, upper ? LSR:LSL, upper ? 32:0);
4446   uxtb(tmp, v);
4447   ldrw(crc, Address(table3, tmp, Address::lsl(2)));
4448   ubfx(tmp, v, 8, 8);
4449   ldrw(tmp, Address(table2, tmp, Address::lsl(2)));
4450   eor(crc, crc, tmp);
4451   ubfx(tmp, v, 16, 8);
4452   ldrw(tmp, Address(table1, tmp, Address::lsl(2)));
4453   eor(crc, crc, tmp);
4454   ubfx(tmp, v, 24, 8);
4455   ldrw(tmp, Address(table0, tmp, Address::lsl(2)));
4456   eor(crc, crc, tmp);
4457 }
4458 
4459 void MacroAssembler::kernel_crc32_using_crypto_pmull(Register crc, Register buf,
4460         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4461     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4462     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4463 
4464     subs(tmp0, len, 384);
4465     mvnw(crc, crc);
4466     br(Assembler::GE, CRC_by128_pre);
4467   BIND(CRC_less128);
4468     subs(len, len, 32);
4469     br(Assembler::GE, CRC_by32_loop);
4470   BIND(CRC_less32);
4471     adds(len, len, 32 - 4);
4472     br(Assembler::GE, CRC_by4_loop);
4473     adds(len, len, 4);
4474     br(Assembler::GT, CRC_by1_loop);
4475     b(L_exit);
4476 
4477   BIND(CRC_by32_loop);
4478     ldp(tmp0, tmp1, Address(buf));
4479     crc32x(crc, crc, tmp0);
4480     ldp(tmp2, tmp3, Address(buf, 16));
4481     crc32x(crc, crc, tmp1);
4482     add(buf, buf, 32);
4483     crc32x(crc, crc, tmp2);
4484     subs(len, len, 32);
4485     crc32x(crc, crc, tmp3);
4486     br(Assembler::GE, CRC_by32_loop);
4487     cmn(len, (u1)32);
4488     br(Assembler::NE, CRC_less32);
4489     b(L_exit);
4490 
4491   BIND(CRC_by4_loop);
4492     ldrw(tmp0, Address(post(buf, 4)));
4493     subs(len, len, 4);
4494     crc32w(crc, crc, tmp0);
4495     br(Assembler::GE, CRC_by4_loop);
4496     adds(len, len, 4);
4497     br(Assembler::LE, L_exit);
4498   BIND(CRC_by1_loop);
4499     ldrb(tmp0, Address(post(buf, 1)));
4500     subs(len, len, 1);
4501     crc32b(crc, crc, tmp0);
4502     br(Assembler::GT, CRC_by1_loop);
4503     b(L_exit);
4504 
4505   BIND(CRC_by128_pre);
4506     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4507       4*256*sizeof(juint) + 8*sizeof(juint));
4508     mov(crc, 0);
4509     crc32x(crc, crc, tmp0);
4510     crc32x(crc, crc, tmp1);
4511 
4512     cbnz(len, CRC_less128);
4513 
4514   BIND(L_exit);
4515     mvnw(crc, crc);
4516 }
4517 
4518 void MacroAssembler::kernel_crc32_using_crc32(Register crc, Register buf,
4519         Register len, Register tmp0, Register tmp1, Register tmp2,
4520         Register tmp3) {
4521     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4522     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4523 
4524     mvnw(crc, crc);
4525 
4526     subs(len, len, 128);
4527     br(Assembler::GE, CRC_by64_pre);
4528   BIND(CRC_less64);
4529     adds(len, len, 128-32);
4530     br(Assembler::GE, CRC_by32_loop);
4531   BIND(CRC_less32);
4532     adds(len, len, 32-4);
4533     br(Assembler::GE, CRC_by4_loop);
4534     adds(len, len, 4);
4535     br(Assembler::GT, CRC_by1_loop);
4536     b(L_exit);
4537 
4538   BIND(CRC_by32_loop);
4539     ldp(tmp0, tmp1, Address(post(buf, 16)));
4540     subs(len, len, 32);
4541     crc32x(crc, crc, tmp0);
4542     ldr(tmp2, Address(post(buf, 8)));
4543     crc32x(crc, crc, tmp1);
4544     ldr(tmp3, Address(post(buf, 8)));
4545     crc32x(crc, crc, tmp2);
4546     crc32x(crc, crc, tmp3);
4547     br(Assembler::GE, CRC_by32_loop);
4548     cmn(len, (u1)32);
4549     br(Assembler::NE, CRC_less32);
4550     b(L_exit);
4551 
4552   BIND(CRC_by4_loop);
4553     ldrw(tmp0, Address(post(buf, 4)));
4554     subs(len, len, 4);
4555     crc32w(crc, crc, tmp0);
4556     br(Assembler::GE, CRC_by4_loop);
4557     adds(len, len, 4);
4558     br(Assembler::LE, L_exit);
4559   BIND(CRC_by1_loop);
4560     ldrb(tmp0, Address(post(buf, 1)));
4561     subs(len, len, 1);
4562     crc32b(crc, crc, tmp0);
4563     br(Assembler::GT, CRC_by1_loop);
4564     b(L_exit);
4565 
4566   BIND(CRC_by64_pre);
4567     sub(buf, buf, 8);
4568     ldp(tmp0, tmp1, Address(buf, 8));
4569     crc32x(crc, crc, tmp0);
4570     ldr(tmp2, Address(buf, 24));
4571     crc32x(crc, crc, tmp1);
4572     ldr(tmp3, Address(buf, 32));
4573     crc32x(crc, crc, tmp2);
4574     ldr(tmp0, Address(buf, 40));
4575     crc32x(crc, crc, tmp3);
4576     ldr(tmp1, Address(buf, 48));
4577     crc32x(crc, crc, tmp0);
4578     ldr(tmp2, Address(buf, 56));
4579     crc32x(crc, crc, tmp1);
4580     ldr(tmp3, Address(pre(buf, 64)));
4581 
4582     b(CRC_by64_loop);
4583 
4584     align(CodeEntryAlignment);
4585   BIND(CRC_by64_loop);
4586     subs(len, len, 64);
4587     crc32x(crc, crc, tmp2);
4588     ldr(tmp0, Address(buf, 8));
4589     crc32x(crc, crc, tmp3);
4590     ldr(tmp1, Address(buf, 16));
4591     crc32x(crc, crc, tmp0);
4592     ldr(tmp2, Address(buf, 24));
4593     crc32x(crc, crc, tmp1);
4594     ldr(tmp3, Address(buf, 32));
4595     crc32x(crc, crc, tmp2);
4596     ldr(tmp0, Address(buf, 40));
4597     crc32x(crc, crc, tmp3);
4598     ldr(tmp1, Address(buf, 48));
4599     crc32x(crc, crc, tmp0);
4600     ldr(tmp2, Address(buf, 56));
4601     crc32x(crc, crc, tmp1);
4602     ldr(tmp3, Address(pre(buf, 64)));
4603     br(Assembler::GE, CRC_by64_loop);
4604 
4605     // post-loop
4606     crc32x(crc, crc, tmp2);
4607     crc32x(crc, crc, tmp3);
4608 
4609     sub(len, len, 64);
4610     add(buf, buf, 8);
4611     cmn(len, (u1)128);
4612     br(Assembler::NE, CRC_less64);
4613   BIND(L_exit);
4614     mvnw(crc, crc);
4615 }
4616 
4617 /**
4618  * @param crc   register containing existing CRC (32-bit)
4619  * @param buf   register pointing to input byte buffer (byte*)
4620  * @param len   register containing number of bytes
4621  * @param table register that will contain address of CRC table
4622  * @param tmp   scratch register
4623  */
4624 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len,
4625         Register table0, Register table1, Register table2, Register table3,
4626         Register tmp, Register tmp2, Register tmp3) {
4627   Label L_by16, L_by16_loop, L_by4, L_by4_loop, L_by1, L_by1_loop, L_exit;
4628 
4629   if (UseCryptoPmullForCRC32) {
4630       kernel_crc32_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4631       return;
4632   }
4633 
4634   if (UseCRC32) {
4635       kernel_crc32_using_crc32(crc, buf, len, table0, table1, table2, table3);
4636       return;
4637   }
4638 
4639     mvnw(crc, crc);
4640 
4641     {
4642       uint64_t offset;
4643       adrp(table0, ExternalAddress(StubRoutines::crc_table_addr()), offset);
4644       add(table0, table0, offset);
4645     }
4646     add(table1, table0, 1*256*sizeof(juint));
4647     add(table2, table0, 2*256*sizeof(juint));
4648     add(table3, table0, 3*256*sizeof(juint));
4649 
4650     { // Neon code start
4651       cmp(len, (u1)64);
4652       br(Assembler::LT, L_by16);
4653       eor(v16, T16B, v16, v16);
4654 
4655     Label L_fold;
4656 
4657       add(tmp, table0, 4*256*sizeof(juint)); // Point at the Neon constants
4658 
4659       ld1(v0, v1, T2D, post(buf, 32));
4660       ld1r(v4, T2D, post(tmp, 8));
4661       ld1r(v5, T2D, post(tmp, 8));
4662       ld1r(v6, T2D, post(tmp, 8));
4663       ld1r(v7, T2D, post(tmp, 8));
4664       mov(v16, S, 0, crc);
4665 
4666       eor(v0, T16B, v0, v16);
4667       sub(len, len, 64);
4668 
4669     BIND(L_fold);
4670       pmull(v22, T8H, v0, v5, T8B);
4671       pmull(v20, T8H, v0, v7, T8B);
4672       pmull(v23, T8H, v0, v4, T8B);
4673       pmull(v21, T8H, v0, v6, T8B);
4674 
4675       pmull2(v18, T8H, v0, v5, T16B);
4676       pmull2(v16, T8H, v0, v7, T16B);
4677       pmull2(v19, T8H, v0, v4, T16B);
4678       pmull2(v17, T8H, v0, v6, T16B);
4679 
4680       uzp1(v24, T8H, v20, v22);
4681       uzp2(v25, T8H, v20, v22);
4682       eor(v20, T16B, v24, v25);
4683 
4684       uzp1(v26, T8H, v16, v18);
4685       uzp2(v27, T8H, v16, v18);
4686       eor(v16, T16B, v26, v27);
4687 
4688       ushll2(v22, T4S, v20, T8H, 8);
4689       ushll(v20, T4S, v20, T4H, 8);
4690 
4691       ushll2(v18, T4S, v16, T8H, 8);
4692       ushll(v16, T4S, v16, T4H, 8);
4693 
4694       eor(v22, T16B, v23, v22);
4695       eor(v18, T16B, v19, v18);
4696       eor(v20, T16B, v21, v20);
4697       eor(v16, T16B, v17, v16);
4698 
4699       uzp1(v17, T2D, v16, v20);
4700       uzp2(v21, T2D, v16, v20);
4701       eor(v17, T16B, v17, v21);
4702 
4703       ushll2(v20, T2D, v17, T4S, 16);
4704       ushll(v16, T2D, v17, T2S, 16);
4705 
4706       eor(v20, T16B, v20, v22);
4707       eor(v16, T16B, v16, v18);
4708 
4709       uzp1(v17, T2D, v20, v16);
4710       uzp2(v21, T2D, v20, v16);
4711       eor(v28, T16B, v17, v21);
4712 
4713       pmull(v22, T8H, v1, v5, T8B);
4714       pmull(v20, T8H, v1, v7, T8B);
4715       pmull(v23, T8H, v1, v4, T8B);
4716       pmull(v21, T8H, v1, v6, T8B);
4717 
4718       pmull2(v18, T8H, v1, v5, T16B);
4719       pmull2(v16, T8H, v1, v7, T16B);
4720       pmull2(v19, T8H, v1, v4, T16B);
4721       pmull2(v17, T8H, v1, v6, T16B);
4722 
4723       ld1(v0, v1, T2D, post(buf, 32));
4724 
4725       uzp1(v24, T8H, v20, v22);
4726       uzp2(v25, T8H, v20, v22);
4727       eor(v20, T16B, v24, v25);
4728 
4729       uzp1(v26, T8H, v16, v18);
4730       uzp2(v27, T8H, v16, v18);
4731       eor(v16, T16B, v26, v27);
4732 
4733       ushll2(v22, T4S, v20, T8H, 8);
4734       ushll(v20, T4S, v20, T4H, 8);
4735 
4736       ushll2(v18, T4S, v16, T8H, 8);
4737       ushll(v16, T4S, v16, T4H, 8);
4738 
4739       eor(v22, T16B, v23, v22);
4740       eor(v18, T16B, v19, v18);
4741       eor(v20, T16B, v21, v20);
4742       eor(v16, T16B, v17, v16);
4743 
4744       uzp1(v17, T2D, v16, v20);
4745       uzp2(v21, T2D, v16, v20);
4746       eor(v16, T16B, v17, v21);
4747 
4748       ushll2(v20, T2D, v16, T4S, 16);
4749       ushll(v16, T2D, v16, T2S, 16);
4750 
4751       eor(v20, T16B, v22, v20);
4752       eor(v16, T16B, v16, v18);
4753 
4754       uzp1(v17, T2D, v20, v16);
4755       uzp2(v21, T2D, v20, v16);
4756       eor(v20, T16B, v17, v21);
4757 
4758       shl(v16, T2D, v28, 1);
4759       shl(v17, T2D, v20, 1);
4760 
4761       eor(v0, T16B, v0, v16);
4762       eor(v1, T16B, v1, v17);
4763 
4764       subs(len, len, 32);
4765       br(Assembler::GE, L_fold);
4766 
4767       mov(crc, 0);
4768       mov(tmp, v0, D, 0);
4769       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4770       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4771       mov(tmp, v0, D, 1);
4772       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4773       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4774       mov(tmp, v1, D, 0);
4775       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4776       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4777       mov(tmp, v1, D, 1);
4778       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4779       update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4780 
4781       add(len, len, 32);
4782     } // Neon code end
4783 
4784   BIND(L_by16);
4785     subs(len, len, 16);
4786     br(Assembler::GE, L_by16_loop);
4787     adds(len, len, 16-4);
4788     br(Assembler::GE, L_by4_loop);
4789     adds(len, len, 4);
4790     br(Assembler::GT, L_by1_loop);
4791     b(L_exit);
4792 
4793   BIND(L_by4_loop);
4794     ldrw(tmp, Address(post(buf, 4)));
4795     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3);
4796     subs(len, len, 4);
4797     br(Assembler::GE, L_by4_loop);
4798     adds(len, len, 4);
4799     br(Assembler::LE, L_exit);
4800   BIND(L_by1_loop);
4801     subs(len, len, 1);
4802     ldrb(tmp, Address(post(buf, 1)));
4803     update_byte_crc32(crc, tmp, table0);
4804     br(Assembler::GT, L_by1_loop);
4805     b(L_exit);
4806 
4807     align(CodeEntryAlignment);
4808   BIND(L_by16_loop);
4809     subs(len, len, 16);
4810     ldp(tmp, tmp3, Address(post(buf, 16)));
4811     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, false);
4812     update_word_crc32(crc, tmp, tmp2, table0, table1, table2, table3, true);
4813     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, false);
4814     update_word_crc32(crc, tmp3, tmp2, table0, table1, table2, table3, true);
4815     br(Assembler::GE, L_by16_loop);
4816     adds(len, len, 16-4);
4817     br(Assembler::GE, L_by4_loop);
4818     adds(len, len, 4);
4819     br(Assembler::GT, L_by1_loop);
4820   BIND(L_exit);
4821     mvnw(crc, crc);
4822 }
4823 
4824 void MacroAssembler::kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
4825         Register len, Register tmp0, Register tmp1, Register tmp2, Register tmp3) {
4826     Label CRC_by4_loop, CRC_by1_loop, CRC_less128, CRC_by128_pre, CRC_by32_loop, CRC_less32, L_exit;
4827     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4828 
4829     subs(tmp0, len, 384);
4830     br(Assembler::GE, CRC_by128_pre);
4831   BIND(CRC_less128);
4832     subs(len, len, 32);
4833     br(Assembler::GE, CRC_by32_loop);
4834   BIND(CRC_less32);
4835     adds(len, len, 32 - 4);
4836     br(Assembler::GE, CRC_by4_loop);
4837     adds(len, len, 4);
4838     br(Assembler::GT, CRC_by1_loop);
4839     b(L_exit);
4840 
4841   BIND(CRC_by32_loop);
4842     ldp(tmp0, tmp1, Address(buf));
4843     crc32cx(crc, crc, tmp0);
4844     ldr(tmp2, Address(buf, 16));
4845     crc32cx(crc, crc, tmp1);
4846     ldr(tmp3, Address(buf, 24));
4847     crc32cx(crc, crc, tmp2);
4848     add(buf, buf, 32);
4849     subs(len, len, 32);
4850     crc32cx(crc, crc, tmp3);
4851     br(Assembler::GE, CRC_by32_loop);
4852     cmn(len, (u1)32);
4853     br(Assembler::NE, CRC_less32);
4854     b(L_exit);
4855 
4856   BIND(CRC_by4_loop);
4857     ldrw(tmp0, Address(post(buf, 4)));
4858     subs(len, len, 4);
4859     crc32cw(crc, crc, tmp0);
4860     br(Assembler::GE, CRC_by4_loop);
4861     adds(len, len, 4);
4862     br(Assembler::LE, L_exit);
4863   BIND(CRC_by1_loop);
4864     ldrb(tmp0, Address(post(buf, 1)));
4865     subs(len, len, 1);
4866     crc32cb(crc, crc, tmp0);
4867     br(Assembler::GT, CRC_by1_loop);
4868     b(L_exit);
4869 
4870   BIND(CRC_by128_pre);
4871     kernel_crc32_common_fold_using_crypto_pmull(crc, buf, len, tmp0, tmp1, tmp2,
4872       4*256*sizeof(juint) + 8*sizeof(juint) + 0x50);
4873     mov(crc, 0);
4874     crc32cx(crc, crc, tmp0);
4875     crc32cx(crc, crc, tmp1);
4876 
4877     cbnz(len, CRC_less128);
4878 
4879   BIND(L_exit);
4880 }
4881 
4882 void MacroAssembler::kernel_crc32c_using_crc32c(Register crc, Register buf,
4883         Register len, Register tmp0, Register tmp1, Register tmp2,
4884         Register tmp3) {
4885     Label CRC_by64_loop, CRC_by4_loop, CRC_by1_loop, CRC_less64, CRC_by64_pre, CRC_by32_loop, CRC_less32, L_exit;
4886     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2, tmp3);
4887 
4888     subs(len, len, 128);
4889     br(Assembler::GE, CRC_by64_pre);
4890   BIND(CRC_less64);
4891     adds(len, len, 128-32);
4892     br(Assembler::GE, CRC_by32_loop);
4893   BIND(CRC_less32);
4894     adds(len, len, 32-4);
4895     br(Assembler::GE, CRC_by4_loop);
4896     adds(len, len, 4);
4897     br(Assembler::GT, CRC_by1_loop);
4898     b(L_exit);
4899 
4900   BIND(CRC_by32_loop);
4901     ldp(tmp0, tmp1, Address(post(buf, 16)));
4902     subs(len, len, 32);
4903     crc32cx(crc, crc, tmp0);
4904     ldr(tmp2, Address(post(buf, 8)));
4905     crc32cx(crc, crc, tmp1);
4906     ldr(tmp3, Address(post(buf, 8)));
4907     crc32cx(crc, crc, tmp2);
4908     crc32cx(crc, crc, tmp3);
4909     br(Assembler::GE, CRC_by32_loop);
4910     cmn(len, (u1)32);
4911     br(Assembler::NE, CRC_less32);
4912     b(L_exit);
4913 
4914   BIND(CRC_by4_loop);
4915     ldrw(tmp0, Address(post(buf, 4)));
4916     subs(len, len, 4);
4917     crc32cw(crc, crc, tmp0);
4918     br(Assembler::GE, CRC_by4_loop);
4919     adds(len, len, 4);
4920     br(Assembler::LE, L_exit);
4921   BIND(CRC_by1_loop);
4922     ldrb(tmp0, Address(post(buf, 1)));
4923     subs(len, len, 1);
4924     crc32cb(crc, crc, tmp0);
4925     br(Assembler::GT, CRC_by1_loop);
4926     b(L_exit);
4927 
4928   BIND(CRC_by64_pre);
4929     sub(buf, buf, 8);
4930     ldp(tmp0, tmp1, Address(buf, 8));
4931     crc32cx(crc, crc, tmp0);
4932     ldr(tmp2, Address(buf, 24));
4933     crc32cx(crc, crc, tmp1);
4934     ldr(tmp3, Address(buf, 32));
4935     crc32cx(crc, crc, tmp2);
4936     ldr(tmp0, Address(buf, 40));
4937     crc32cx(crc, crc, tmp3);
4938     ldr(tmp1, Address(buf, 48));
4939     crc32cx(crc, crc, tmp0);
4940     ldr(tmp2, Address(buf, 56));
4941     crc32cx(crc, crc, tmp1);
4942     ldr(tmp3, Address(pre(buf, 64)));
4943 
4944     b(CRC_by64_loop);
4945 
4946     align(CodeEntryAlignment);
4947   BIND(CRC_by64_loop);
4948     subs(len, len, 64);
4949     crc32cx(crc, crc, tmp2);
4950     ldr(tmp0, Address(buf, 8));
4951     crc32cx(crc, crc, tmp3);
4952     ldr(tmp1, Address(buf, 16));
4953     crc32cx(crc, crc, tmp0);
4954     ldr(tmp2, Address(buf, 24));
4955     crc32cx(crc, crc, tmp1);
4956     ldr(tmp3, Address(buf, 32));
4957     crc32cx(crc, crc, tmp2);
4958     ldr(tmp0, Address(buf, 40));
4959     crc32cx(crc, crc, tmp3);
4960     ldr(tmp1, Address(buf, 48));
4961     crc32cx(crc, crc, tmp0);
4962     ldr(tmp2, Address(buf, 56));
4963     crc32cx(crc, crc, tmp1);
4964     ldr(tmp3, Address(pre(buf, 64)));
4965     br(Assembler::GE, CRC_by64_loop);
4966 
4967     // post-loop
4968     crc32cx(crc, crc, tmp2);
4969     crc32cx(crc, crc, tmp3);
4970 
4971     sub(len, len, 64);
4972     add(buf, buf, 8);
4973     cmn(len, (u1)128);
4974     br(Assembler::NE, CRC_less64);
4975   BIND(L_exit);
4976 }
4977 
4978 /**
4979  * @param crc   register containing existing CRC (32-bit)
4980  * @param buf   register pointing to input byte buffer (byte*)
4981  * @param len   register containing number of bytes
4982  * @param table register that will contain address of CRC table
4983  * @param tmp   scratch register
4984  */
4985 void MacroAssembler::kernel_crc32c(Register crc, Register buf, Register len,
4986         Register table0, Register table1, Register table2, Register table3,
4987         Register tmp, Register tmp2, Register tmp3) {
4988   if (UseCryptoPmullForCRC32) {
4989     kernel_crc32c_using_crypto_pmull(crc, buf, len, table0, table1, table2, table3);
4990   } else {
4991     kernel_crc32c_using_crc32c(crc, buf, len, table0, table1, table2, table3);
4992   }
4993 }
4994 
4995 void MacroAssembler::kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
4996         Register len, Register tmp0, Register tmp1, Register tmp2, size_t table_offset) {
4997     Label CRC_by128_loop;
4998     assert_different_registers(crc, buf, len, tmp0, tmp1, tmp2);
4999 
5000     sub(len, len, 256);
5001     Register table = tmp0;
5002     {
5003       uint64_t offset;
5004       adrp(table, ExternalAddress(StubRoutines::crc_table_addr()), offset);
5005       add(table, table, offset);
5006     }
5007     add(table, table, table_offset);
5008 
5009     // Registers v0..v7 are used as data registers.
5010     // Registers v16..v31 are used as tmp registers.
5011     sub(buf, buf, 0x10);
5012     ldrq(v0, Address(buf, 0x10));
5013     ldrq(v1, Address(buf, 0x20));
5014     ldrq(v2, Address(buf, 0x30));
5015     ldrq(v3, Address(buf, 0x40));
5016     ldrq(v4, Address(buf, 0x50));
5017     ldrq(v5, Address(buf, 0x60));
5018     ldrq(v6, Address(buf, 0x70));
5019     ldrq(v7, Address(pre(buf, 0x80)));
5020 
5021     movi(v31, T4S, 0);
5022     mov(v31, S, 0, crc);
5023     eor(v0, T16B, v0, v31);
5024 
5025     // Register v16 contains constants from the crc table.
5026     ldrq(v16, Address(table));
5027     b(CRC_by128_loop);
5028 
5029     align(OptoLoopAlignment);
5030   BIND(CRC_by128_loop);
5031     pmull (v17,  T1Q, v0, v16, T1D);
5032     pmull2(v18, T1Q, v0, v16, T2D);
5033     ldrq(v0, Address(buf, 0x10));
5034     eor3(v0, T16B, v17,  v18, v0);
5035 
5036     pmull (v19, T1Q, v1, v16, T1D);
5037     pmull2(v20, T1Q, v1, v16, T2D);
5038     ldrq(v1, Address(buf, 0x20));
5039     eor3(v1, T16B, v19, v20, v1);
5040 
5041     pmull (v21, T1Q, v2, v16, T1D);
5042     pmull2(v22, T1Q, v2, v16, T2D);
5043     ldrq(v2, Address(buf, 0x30));
5044     eor3(v2, T16B, v21, v22, v2);
5045 
5046     pmull (v23, T1Q, v3, v16, T1D);
5047     pmull2(v24, T1Q, v3, v16, T2D);
5048     ldrq(v3, Address(buf, 0x40));
5049     eor3(v3, T16B, v23, v24, v3);
5050 
5051     pmull (v25, T1Q, v4, v16, T1D);
5052     pmull2(v26, T1Q, v4, v16, T2D);
5053     ldrq(v4, Address(buf, 0x50));
5054     eor3(v4, T16B, v25, v26, v4);
5055 
5056     pmull (v27, T1Q, v5, v16, T1D);
5057     pmull2(v28, T1Q, v5, v16, T2D);
5058     ldrq(v5, Address(buf, 0x60));
5059     eor3(v5, T16B, v27, v28, v5);
5060 
5061     pmull (v29, T1Q, v6, v16, T1D);
5062     pmull2(v30, T1Q, v6, v16, T2D);
5063     ldrq(v6, Address(buf, 0x70));
5064     eor3(v6, T16B, v29, v30, v6);
5065 
5066     // Reuse registers v23, v24.
5067     // Using them won't block the first instruction of the next iteration.
5068     pmull (v23, T1Q, v7, v16, T1D);
5069     pmull2(v24, T1Q, v7, v16, T2D);
5070     ldrq(v7, Address(pre(buf, 0x80)));
5071     eor3(v7, T16B, v23, v24, v7);
5072 
5073     subs(len, len, 0x80);
5074     br(Assembler::GE, CRC_by128_loop);
5075 
5076     // fold into 512 bits
5077     // Use v31 for constants because v16 can be still in use.
5078     ldrq(v31, Address(table, 0x10));
5079 
5080     pmull (v17,  T1Q, v0, v31, T1D);
5081     pmull2(v18, T1Q, v0, v31, T2D);
5082     eor3(v0, T16B, v17, v18, v4);
5083 
5084     pmull (v19, T1Q, v1, v31, T1D);
5085     pmull2(v20, T1Q, v1, v31, T2D);
5086     eor3(v1, T16B, v19, v20, v5);
5087 
5088     pmull (v21, T1Q, v2, v31, T1D);
5089     pmull2(v22, T1Q, v2, v31, T2D);
5090     eor3(v2, T16B, v21, v22, v6);
5091 
5092     pmull (v23, T1Q, v3, v31, T1D);
5093     pmull2(v24, T1Q, v3, v31, T2D);
5094     eor3(v3, T16B, v23, v24, v7);
5095 
5096     // fold into 128 bits
5097     // Use v17 for constants because v31 can be still in use.
5098     ldrq(v17, Address(table, 0x20));
5099     pmull (v25, T1Q, v0, v17, T1D);
5100     pmull2(v26, T1Q, v0, v17, T2D);
5101     eor3(v3, T16B, v3, v25, v26);
5102 
5103     // Use v18 for constants because v17 can be still in use.
5104     ldrq(v18, Address(table, 0x30));
5105     pmull (v27, T1Q, v1, v18, T1D);
5106     pmull2(v28, T1Q, v1, v18, T2D);
5107     eor3(v3, T16B, v3, v27, v28);
5108 
5109     // Use v19 for constants because v18 can be still in use.
5110     ldrq(v19, Address(table, 0x40));
5111     pmull (v29, T1Q, v2, v19, T1D);
5112     pmull2(v30, T1Q, v2, v19, T2D);
5113     eor3(v0, T16B, v3, v29, v30);
5114 
5115     add(len, len, 0x80);
5116     add(buf, buf, 0x10);
5117 
5118     mov(tmp0, v0, D, 0);
5119     mov(tmp1, v0, D, 1);
5120 }
5121 
5122 void MacroAssembler::addptr(const Address &dst, int32_t src) {
5123   Address adr;
5124   switch(dst.getMode()) {
5125   case Address::base_plus_offset:
5126     // This is the expected mode, although we allow all the other
5127     // forms below.
5128     adr = form_address(rscratch2, dst.base(), dst.offset(), LogBytesPerWord);
5129     break;
5130   default:
5131     lea(rscratch2, dst);
5132     adr = Address(rscratch2);
5133     break;
5134   }
5135   ldr(rscratch1, adr);
5136   add(rscratch1, rscratch1, src);
5137   str(rscratch1, adr);
5138 }
5139 
5140 void MacroAssembler::cmpptr(Register src1, Address src2) {
5141   uint64_t offset;
5142   adrp(rscratch1, src2, offset);
5143   ldr(rscratch1, Address(rscratch1, offset));
5144   cmp(src1, rscratch1);
5145 }
5146 
5147 void MacroAssembler::cmpoop(Register obj1, Register obj2) {
5148   cmp(obj1, obj2);
5149 }
5150 
5151 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5152   load_method_holder(rresult, rmethod);
5153   ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5154 }
5155 
5156 void MacroAssembler::load_method_holder(Register holder, Register method) {
5157   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
5158   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
5159   ldr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
5160 }
5161 
5162 void MacroAssembler::load_metadata(Register dst, Register src) {
5163   if (UseCompactObjectHeaders) {
5164     load_narrow_klass_compact(dst, src);
5165   } else if (UseCompressedClassPointers) {
5166     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5167   } else {
5168     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5169   }
5170 }
5171 
5172 // Loads the obj's Klass* into dst.
5173 // Preserves all registers (incl src, rscratch1 and rscratch2).
5174 // Input:
5175 // src - the oop we want to load the klass from.
5176 // dst - output narrow klass.
5177 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5178   assert(UseCompactObjectHeaders, "expects UseCompactObjectHeaders");
5179   ldr(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5180   lsr(dst, dst, markWord::klass_shift);
5181 }
5182 
5183 void MacroAssembler::load_klass(Register dst, Register src) {
5184   if (UseCompactObjectHeaders) {
5185     load_narrow_klass_compact(dst, src);
5186     decode_klass_not_null(dst);
5187   } else if (UseCompressedClassPointers) {
5188     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5189     decode_klass_not_null(dst);
5190   } else {
5191     ldr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5192   }
5193 }
5194 
5195 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp1, Register tmp2) {
5196   if (RestoreMXCSROnJNICalls) {
5197     Label OK;
5198     get_fpcr(tmp1);
5199     mov(tmp2, tmp1);
5200     // Set FPCR to the state we need. We do want Round to Nearest. We
5201     // don't want non-IEEE rounding modes or floating-point traps.
5202     bfi(tmp1, zr, 22, 4); // Clear DN, FZ, and Rmode
5203     bfi(tmp1, zr, 8, 5);  // Clear exception-control bits (8-12)
5204     bfi(tmp1, zr, 0, 2);  // Clear AH:FIZ
5205     eor(tmp2, tmp1, tmp2);
5206     cbz(tmp2, OK);        // Only reset FPCR if it's wrong
5207     set_fpcr(tmp1);
5208     bind(OK);
5209   }
5210 }
5211 
5212 // ((OopHandle)result).resolve();
5213 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
5214   // OopHandle::resolve is an indirection.
5215   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
5216 }
5217 
5218 // ((WeakHandle)result).resolve();
5219 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
5220   assert_different_registers(result, tmp1, tmp2);
5221   Label resolved;
5222 
5223   // A null weak handle resolves to null.
5224   cbz(result, resolved);
5225 
5226   // Only 64 bit platforms support GCs that require a tmp register
5227   // WeakHandle::resolve is an indirection like jweak.
5228   access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5229                  result, Address(result), tmp1, tmp2);
5230   bind(resolved);
5231 }
5232 
5233 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
5234   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5235   ldr(dst, Address(rmethod, Method::const_offset()));
5236   ldr(dst, Address(dst, ConstMethod::constants_offset()));
5237   ldr(dst, Address(dst, ConstantPool::pool_holder_offset()));
5238   ldr(dst, Address(dst, mirror_offset));
5239   resolve_oop_handle(dst, tmp1, tmp2);
5240 }
5241 
5242 void MacroAssembler::cmp_klass(Register obj, Register klass, Register tmp) {
5243   assert_different_registers(obj, klass, tmp);
5244   if (UseCompressedClassPointers) {
5245     if (UseCompactObjectHeaders) {
5246       load_narrow_klass_compact(tmp, obj);
5247     } else {
5248       ldrw(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5249     }
5250     if (CompressedKlassPointers::base() == nullptr) {
5251       cmp(klass, tmp, LSL, CompressedKlassPointers::shift());
5252       return;
5253     } else if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0
5254                && CompressedKlassPointers::shift() == 0) {
5255       // Only the bottom 32 bits matter
5256       cmpw(klass, tmp);
5257       return;
5258     }
5259     decode_klass_not_null(tmp);
5260   } else {
5261     ldr(tmp, Address(obj, oopDesc::klass_offset_in_bytes()));
5262   }
5263   cmp(klass, tmp);
5264 }
5265 
5266 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5267   if (UseCompactObjectHeaders) {
5268     load_narrow_klass_compact(tmp1, obj1);
5269     load_narrow_klass_compact(tmp2,  obj2);
5270     cmpw(tmp1, tmp2);
5271   } else if (UseCompressedClassPointers) {
5272     ldrw(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5273     ldrw(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5274     cmpw(tmp1, tmp2);
5275   } else {
5276     ldr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5277     ldr(tmp2, Address(obj2, oopDesc::klass_offset_in_bytes()));
5278     cmp(tmp1, tmp2);
5279   }
5280 }
5281 
5282 void MacroAssembler::load_prototype_header(Register dst, Register src) {
5283   load_klass(dst, src);
5284   ldr(dst, Address(dst, Klass::prototype_header_offset()));
5285 }
5286 
5287 void MacroAssembler::store_klass(Register dst, Register src) {
5288   // FIXME: Should this be a store release?  concurrent gcs assumes
5289   // klass length is valid if klass field is not null.
5290   assert(!UseCompactObjectHeaders, "not with compact headers");
5291   if (UseCompressedClassPointers) {
5292     encode_klass_not_null(src);
5293     strw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5294   } else {
5295     str(src, Address(dst, oopDesc::klass_offset_in_bytes()));
5296   }
5297 }
5298 
5299 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5300   assert(!UseCompactObjectHeaders, "not with compact headers");
5301   if (UseCompressedClassPointers) {
5302     // Store to klass gap in destination
5303     strw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
5304   }
5305 }
5306 
5307 // Algorithm must match CompressedOops::encode.
5308 void MacroAssembler::encode_heap_oop(Register d, Register s) {
5309 #ifdef ASSERT
5310   verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5311 #endif
5312   verify_oop_msg(s, "broken oop in encode_heap_oop");
5313   if (CompressedOops::base() == nullptr) {
5314     if (CompressedOops::shift() != 0) {
5315       assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5316       lsr(d, s, LogMinObjAlignmentInBytes);
5317     } else {
5318       mov(d, s);
5319     }
5320   } else {
5321     subs(d, s, rheapbase);
5322     csel(d, d, zr, Assembler::HS);
5323     lsr(d, d, LogMinObjAlignmentInBytes);
5324 
5325     /*  Old algorithm: is this any worse?
5326     Label nonnull;
5327     cbnz(r, nonnull);
5328     sub(r, r, rheapbase);
5329     bind(nonnull);
5330     lsr(r, r, LogMinObjAlignmentInBytes);
5331     */
5332   }
5333 }
5334 
5335 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5336 #ifdef ASSERT
5337   verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5338   if (CheckCompressedOops) {
5339     Label ok;
5340     cbnz(r, ok);
5341     stop("null oop passed to encode_heap_oop_not_null");
5342     bind(ok);
5343   }
5344 #endif
5345   verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5346   if (CompressedOops::base() != nullptr) {
5347     sub(r, r, rheapbase);
5348   }
5349   if (CompressedOops::shift() != 0) {
5350     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5351     lsr(r, r, LogMinObjAlignmentInBytes);
5352   }
5353 }
5354 
5355 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5356 #ifdef ASSERT
5357   verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5358   if (CheckCompressedOops) {
5359     Label ok;
5360     cbnz(src, ok);
5361     stop("null oop passed to encode_heap_oop_not_null2");
5362     bind(ok);
5363   }
5364 #endif
5365   verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5366 
5367   Register data = src;
5368   if (CompressedOops::base() != nullptr) {
5369     sub(dst, src, rheapbase);
5370     data = dst;
5371   }
5372   if (CompressedOops::shift() != 0) {
5373     assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5374     lsr(dst, data, LogMinObjAlignmentInBytes);
5375     data = dst;
5376   }
5377   if (data == src)
5378     mov(dst, src);
5379 }
5380 
5381 void  MacroAssembler::decode_heap_oop(Register d, Register s) {
5382 #ifdef ASSERT
5383   verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5384 #endif
5385   if (CompressedOops::base() == nullptr) {
5386     if (CompressedOops::shift() != 0) {
5387       lsl(d, s, CompressedOops::shift());
5388     } else if (d != s) {
5389       mov(d, s);
5390     }
5391   } else {
5392     Label done;
5393     if (d != s)
5394       mov(d, s);
5395     cbz(s, done);
5396     add(d, rheapbase, s, Assembler::LSL, LogMinObjAlignmentInBytes);
5397     bind(done);
5398   }
5399   verify_oop_msg(d, "broken oop in decode_heap_oop");
5400 }
5401 
5402 void  MacroAssembler::decode_heap_oop_not_null(Register r) {
5403   assert (UseCompressedOops, "should only be used for compressed headers");
5404   assert (Universe::heap() != nullptr, "java heap should be initialized");
5405   // Cannot assert, unverified entry point counts instructions (see .ad file)
5406   // vtableStubs also counts instructions in pd_code_size_limit.
5407   // Also do not verify_oop as this is called by verify_oop.
5408   if (CompressedOops::shift() != 0) {
5409     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5410     if (CompressedOops::base() != nullptr) {
5411       add(r, rheapbase, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5412     } else {
5413       add(r, zr, r, Assembler::LSL, LogMinObjAlignmentInBytes);
5414     }
5415   } else {
5416     assert (CompressedOops::base() == nullptr, "sanity");
5417   }
5418 }
5419 
5420 void  MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5421   assert (UseCompressedOops, "should only be used for compressed headers");
5422   assert (Universe::heap() != nullptr, "java heap should be initialized");
5423   // Cannot assert, unverified entry point counts instructions (see .ad file)
5424   // vtableStubs also counts instructions in pd_code_size_limit.
5425   // Also do not verify_oop as this is called by verify_oop.
5426   if (CompressedOops::shift() != 0) {
5427     assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5428     if (CompressedOops::base() != nullptr) {
5429       add(dst, rheapbase, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5430     } else {
5431       add(dst, zr, src, Assembler::LSL, LogMinObjAlignmentInBytes);
5432     }
5433   } else {
5434     assert (CompressedOops::base() == nullptr, "sanity");
5435     if (dst != src) {
5436       mov(dst, src);
5437     }
5438   }
5439 }
5440 
5441 MacroAssembler::KlassDecodeMode MacroAssembler::_klass_decode_mode(KlassDecodeNone);
5442 
5443 MacroAssembler::KlassDecodeMode MacroAssembler::klass_decode_mode() {
5444   assert(Metaspace::initialized(), "metaspace not initialized yet");
5445   assert(_klass_decode_mode != KlassDecodeNone, "should be initialized");
5446   return _klass_decode_mode;
5447 }
5448 
5449 MacroAssembler::KlassDecodeMode  MacroAssembler::klass_decode_mode(address base, int shift, const size_t range) {
5450   assert(UseCompressedClassPointers, "not using compressed class pointers");
5451 
5452   // KlassDecodeMode shouldn't be set already.
5453   assert(_klass_decode_mode == KlassDecodeNone, "set once");
5454 
5455   if (base == nullptr) {
5456     return KlassDecodeZero;
5457   }
5458 
5459   if (operand_valid_for_logical_immediate(
5460         /*is32*/false, (uint64_t)base)) {
5461     const uint64_t range_mask = right_n_bits(log2i_ceil(range));
5462     if (((uint64_t)base & range_mask) == 0) {
5463       return KlassDecodeXor;
5464     }
5465   }
5466 
5467   const uint64_t shifted_base =
5468     (uint64_t)base >> shift;
5469   if ((shifted_base & 0xffff0000ffffffff) == 0) {
5470     return KlassDecodeMovk;
5471   }
5472 
5473   // No valid encoding.
5474   return KlassDecodeNone;
5475 }
5476 
5477 // Check if one of the above decoding modes will work for given base, shift and range.
5478 bool MacroAssembler::check_klass_decode_mode(address base, int shift, const size_t range) {
5479   return klass_decode_mode(base, shift, range) != KlassDecodeNone;
5480 }
5481 
5482 bool MacroAssembler::set_klass_decode_mode(address base, int shift, const size_t range) {
5483   _klass_decode_mode = klass_decode_mode(base, shift, range);
5484   return _klass_decode_mode != KlassDecodeNone;
5485 }
5486 
5487 static Register pick_different_tmp(Register dst, Register src) {
5488   auto tmps = RegSet::of(r0, r1, r2) - RegSet::of(src, dst);
5489   return *tmps.begin();
5490 }
5491 
5492 void MacroAssembler::encode_klass_not_null_for_aot(Register dst, Register src) {
5493   // we have to load the klass base from the AOT constants area but
5494   // not the shift because it is not allowed to change
5495   int shift = CompressedKlassPointers::shift();
5496   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5497   if (dst != src) {
5498     // we can load the base into dst, subtract it formthe src and shift down
5499     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5500     ldr(dst, dst);
5501     sub(dst, src, dst);
5502     lsr(dst, dst, shift);
5503   } else {
5504     // we need an extra register in order to load the coop base
5505     Register tmp = pick_different_tmp(dst, src);
5506     RegSet regs = RegSet::of(tmp);
5507     push(regs, sp);
5508     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5509     ldr(tmp, tmp);
5510     sub(dst, src, tmp);
5511     lsr(dst, dst, shift);
5512     pop(regs, sp);
5513   }
5514 }
5515 
5516 void MacroAssembler::encode_klass_not_null(Register dst, Register src) {
5517   if (AOTCodeCache::is_on_for_dump()) {
5518     encode_klass_not_null_for_aot(dst, src);
5519     return;
5520   }
5521 
5522   switch (klass_decode_mode()) {
5523   case KlassDecodeZero:
5524     if (CompressedKlassPointers::shift() != 0) {
5525       lsr(dst, src, CompressedKlassPointers::shift());
5526     } else {
5527       if (dst != src) mov(dst, src);
5528     }
5529     break;
5530 
5531   case KlassDecodeXor:
5532     if (CompressedKlassPointers::shift() != 0) {
5533       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5534       lsr(dst, dst, CompressedKlassPointers::shift());
5535     } else {
5536       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5537     }
5538     break;
5539 
5540   case KlassDecodeMovk:
5541     if (CompressedKlassPointers::shift() != 0) {
5542       ubfx(dst, src, CompressedKlassPointers::shift(), 32);
5543     } else {
5544       movw(dst, src);
5545     }
5546     break;
5547 
5548   case KlassDecodeNone:
5549     ShouldNotReachHere();
5550     break;
5551   }
5552 }
5553 
5554 void MacroAssembler::encode_klass_not_null(Register r) {
5555   encode_klass_not_null(r, r);
5556 }
5557 
5558 void MacroAssembler::decode_klass_not_null_for_aot(Register dst, Register src) {
5559   // we have to load the klass base from the AOT constants area but
5560   // not the shift because it is not allowed to change
5561   int shift = CompressedKlassPointers::shift();
5562   assert(shift >= 0 && shift <= CompressedKlassPointers::max_shift(), "unexpected compressed klass shift!");
5563   if (dst != src) {
5564     // we can load the base into dst then add the offset with a suitable shift
5565     lea(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5566     ldr(dst, dst);
5567     add(dst, dst, src, LSL,  shift);
5568   } else {
5569     // we need an extra register in order to load the coop base
5570     Register tmp = pick_different_tmp(dst, src);
5571     RegSet regs = RegSet::of(tmp);
5572     push(regs, sp);
5573     lea(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5574     ldr(tmp, tmp);
5575     add(dst, tmp,  src, LSL,  shift);
5576     pop(regs, sp);
5577   }
5578 }
5579 
5580 void  MacroAssembler::decode_klass_not_null(Register dst, Register src) {
5581   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5582 
5583   if (AOTCodeCache::is_on_for_dump()) {
5584     decode_klass_not_null_for_aot(dst, src);
5585     return;
5586   }
5587 
5588   switch (klass_decode_mode()) {
5589   case KlassDecodeZero:
5590     if (CompressedKlassPointers::shift() != 0) {
5591       lsl(dst, src, CompressedKlassPointers::shift());
5592     } else {
5593       if (dst != src) mov(dst, src);
5594     }
5595     break;
5596 
5597   case KlassDecodeXor:
5598     if (CompressedKlassPointers::shift() != 0) {
5599       lsl(dst, src, CompressedKlassPointers::shift());
5600       eor(dst, dst, (uint64_t)CompressedKlassPointers::base());
5601     } else {
5602       eor(dst, src, (uint64_t)CompressedKlassPointers::base());
5603     }
5604     break;
5605 
5606   case KlassDecodeMovk: {
5607     const uint64_t shifted_base =
5608       (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5609 
5610     if (dst != src) movw(dst, src);
5611     movk(dst, shifted_base >> 32, 32);
5612 
5613     if (CompressedKlassPointers::shift() != 0) {
5614       lsl(dst, dst, CompressedKlassPointers::shift());
5615     }
5616 
5617     break;
5618   }
5619 
5620   case KlassDecodeNone:
5621     ShouldNotReachHere();
5622     break;
5623   }
5624 }
5625 
5626 void  MacroAssembler::decode_klass_not_null(Register r) {
5627   decode_klass_not_null(r, r);
5628 }
5629 
5630 void  MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5631 #ifdef ASSERT
5632   {
5633     ThreadInVMfromUnknown tiv;
5634     assert (UseCompressedOops, "should only be used for compressed oops");
5635     assert (Universe::heap() != nullptr, "java heap should be initialized");
5636     assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5637     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5638   }
5639 #endif
5640   int oop_index = oop_recorder()->find_index(obj);
5641   InstructionMark im(this);
5642   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5643   code_section()->relocate(inst_mark(), rspec);
5644   movz(dst, 0xDEAD, 16);
5645   movk(dst, 0xBEEF);
5646 }
5647 
5648 void  MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5649   assert (UseCompressedClassPointers, "should only be used for compressed headers");
5650   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5651   int index = oop_recorder()->find_index(k);
5652 
5653   InstructionMark im(this);
5654   RelocationHolder rspec = metadata_Relocation::spec(index);
5655   code_section()->relocate(inst_mark(), rspec);
5656   narrowKlass nk = CompressedKlassPointers::encode(k);
5657   movz(dst, (nk >> 16), 16);
5658   movk(dst, nk & 0xffff);
5659 }
5660 
5661 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
5662                                     Register dst, Address src,
5663                                     Register tmp1, Register tmp2) {
5664   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5665   decorators = AccessInternal::decorator_fixup(decorators, type);
5666   bool as_raw = (decorators & AS_RAW) != 0;
5667   if (as_raw) {
5668     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
5669   } else {
5670     bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
5671   }
5672 }
5673 
5674 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
5675                                      Address dst, Register val,
5676                                      Register tmp1, Register tmp2, Register tmp3) {
5677   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5678   decorators = AccessInternal::decorator_fixup(decorators, type);
5679   bool as_raw = (decorators & AS_RAW) != 0;
5680   if (as_raw) {
5681     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5682   } else {
5683     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5684   }
5685 }
5686 
5687 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
5688                                      Register inline_layout_info) {
5689   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5690   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
5691 }
5692 
5693 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
5694   ldr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
5695   ldrw(offset, Address(offset, InlineKlass::payload_offset_offset()));
5696 }
5697 
5698 void MacroAssembler::payload_address(Register oop, Register data, Register inline_klass) {
5699   // ((address) (void*) o) + vk->payload_offset();
5700   Register offset = (data == oop) ? rscratch1 : data;
5701   payload_offset(inline_klass, offset);
5702   if (data == oop) {
5703     add(data, data, offset);
5704   } else {
5705     lea(data, Address(oop, offset));
5706   }
5707 }
5708 
5709 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
5710                                                 Register index, Register data) {
5711   assert_different_registers(array, array_klass, index);
5712   assert_different_registers(rscratch1, array, index);
5713 
5714   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
5715   ldrw(rscratch1, Address(array_klass, Klass::layout_helper_offset()));
5716 
5717   // Klass::layout_helper_log2_element_size(lh)
5718   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
5719   lsr(rscratch1, rscratch1, Klass::_lh_log2_element_size_shift);
5720   andr(rscratch1, rscratch1, Klass::_lh_log2_element_size_mask);
5721   lslv(index, index, rscratch1);
5722 
5723   add(data, array, index);
5724   add(data, data, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT));
5725 }
5726 
5727 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5728                                    Register tmp2, DecoratorSet decorators) {
5729   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
5730 }
5731 
5732 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5733                                             Register tmp2, DecoratorSet decorators) {
5734   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, tmp2);
5735 }
5736 
5737 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5738                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
5739   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5740 }
5741 
5742 // Used for storing nulls.
5743 void MacroAssembler::store_heap_oop_null(Address dst) {
5744   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5745 }
5746 
5747 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {
5748   assert(oop_recorder() != nullptr, "this assembler needs a Recorder");
5749   int index = oop_recorder()->allocate_metadata_index(obj);
5750   RelocationHolder rspec = metadata_Relocation::spec(index);
5751   return Address((address)obj, rspec);
5752 }
5753 
5754 // Move an oop into a register.
5755 void MacroAssembler::movoop(Register dst, jobject obj) {
5756   int oop_index;
5757   if (obj == nullptr) {
5758     oop_index = oop_recorder()->allocate_oop_index(obj);
5759   } else {
5760 #ifdef ASSERT
5761     {
5762       ThreadInVMfromUnknown tiv;
5763       assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
5764     }
5765 #endif
5766     oop_index = oop_recorder()->find_index(obj);
5767   }
5768   RelocationHolder rspec = oop_Relocation::spec(oop_index);
5769 
5770   if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
5771     mov(dst, Address((address)obj, rspec));
5772   } else {
5773     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
5774     ldr(dst, Address(dummy, rspec));
5775   }
5776 }
5777 
5778 // Move a metadata address into a register.
5779 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
5780   int oop_index;
5781   if (obj == nullptr) {
5782     oop_index = oop_recorder()->allocate_metadata_index(obj);
5783   } else {
5784     oop_index = oop_recorder()->find_index(obj);
5785   }
5786   RelocationHolder rspec = metadata_Relocation::spec(oop_index);
5787   mov(dst, Address((address)obj, rspec));
5788 }
5789 
5790 Address MacroAssembler::constant_oop_address(jobject obj) {
5791 #ifdef ASSERT
5792   {
5793     ThreadInVMfromUnknown tiv;
5794     assert(oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5795     assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "not an oop");
5796   }
5797 #endif
5798   int oop_index = oop_recorder()->find_index(obj);
5799   return Address((address)obj, oop_Relocation::spec(oop_index));
5800 }
5801 
5802 // Object / value buffer allocation...
5803 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
5804                                        Register t1, Register t2,
5805                                        bool clear_fields, Label& alloc_failed)
5806 {
5807   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
5808   Register layout_size = t1;
5809   assert(new_obj == r0, "needs to be r0");
5810   assert_different_registers(klass, new_obj, t1, t2);
5811 
5812   // get instance_size in InstanceKlass (scaled to a count of bytes)
5813   ldrw(layout_size, Address(klass, Klass::layout_helper_offset()));
5814   // test to see if it is malformed in some way
5815   tst(layout_size, Klass::_lh_instance_slow_path_bit);
5816   br(Assembler::NE, slow_case_no_pop);
5817 
5818   // Allocate the instance:
5819   //  If TLAB is enabled:
5820   //    Try to allocate in the TLAB.
5821   //    If fails, go to the slow path.
5822   //    Initialize the allocation.
5823   //    Exit.
5824   //
5825   //  Go to slow path.
5826 
5827   if (UseTLAB) {
5828     push(klass);
5829     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
5830     if (ZeroTLAB || (!clear_fields)) {
5831       // the fields have been already cleared
5832       b(initialize_header);
5833     } else {
5834       // initialize both the header and fields
5835       b(initialize_object);
5836     }
5837 
5838     if (clear_fields) {
5839       // The object is initialized before the header.  If the object size is
5840       // zero, go directly to the header initialization.
5841       bind(initialize_object);
5842       int header_size = oopDesc::header_size() * HeapWordSize;
5843       assert(is_aligned(header_size, BytesPerLong), "oop header size must be 8-byte-aligned");
5844       subs(layout_size, layout_size, header_size);
5845       br(Assembler::EQ, initialize_header);
5846 
5847       // Initialize topmost object field, divide size by 8, check if odd and
5848       // test if zero.
5849 
5850   #ifdef ASSERT
5851       // make sure instance_size was multiple of 8
5852       Label L;
5853       tst(layout_size, 7);
5854       br(Assembler::EQ, L);
5855       stop("object size is not multiple of 8 - adjust this code");
5856       bind(L);
5857       // must be > 0, no extra check needed here
5858   #endif
5859 
5860       lsr(layout_size, layout_size, LogBytesPerLong);
5861 
5862       // initialize remaining object fields: instance_size was a multiple of 8
5863       {
5864         Label loop;
5865         Register base = t2;
5866 
5867         bind(loop);
5868         add(rscratch1, new_obj, layout_size, Assembler::LSL, LogBytesPerLong);
5869         str(zr, Address(rscratch1, header_size - 1*oopSize));
5870         subs(layout_size, layout_size, 1);
5871         br(Assembler::NE, loop);
5872       }
5873     } // clear_fields
5874 
5875     // initialize object header only.
5876     bind(initialize_header);
5877     pop(klass);
5878     Register mark_word = t2;
5879     if (UseCompactObjectHeaders || Arguments::is_valhalla_enabled()) {
5880       ldr(mark_word, Address(klass, Klass::prototype_header_offset()));
5881       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5882     } else {
5883       mov(mark_word, (intptr_t)markWord::prototype().value());
5884       str(mark_word, Address(new_obj, oopDesc::mark_offset_in_bytes()));
5885     }
5886     if (!UseCompactObjectHeaders) {
5887       store_klass_gap(new_obj, zr);  // zero klass gap for compressed oops
5888       mov(t2, klass);                // preserve klass
5889       store_klass(new_obj, t2);      // src klass reg is potentially compressed
5890     }
5891     b(done);
5892   }
5893 
5894   if (UseTLAB) {
5895     bind(slow_case);
5896     pop(klass);
5897   }
5898   bind(slow_case_no_pop);
5899   b(alloc_failed);
5900 
5901   bind(done);
5902 }
5903 
5904 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
5905 void MacroAssembler::tlab_allocate(Register obj,
5906                                    Register var_size_in_bytes,
5907                                    int con_size_in_bytes,
5908                                    Register t1,
5909                                    Register t2,
5910                                    Label& slow_case) {
5911   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
5912   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
5913 }
5914 
5915 void MacroAssembler::verify_tlab() {
5916 #ifdef ASSERT
5917   if (UseTLAB && VerifyOops) {
5918     Label next, ok;
5919 
5920     stp(rscratch2, rscratch1, Address(pre(sp, -16)));
5921 
5922     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5923     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_start_offset())));
5924     cmp(rscratch2, rscratch1);
5925     br(Assembler::HS, next);
5926     STOP("assert(top >= start)");
5927     should_not_reach_here();
5928 
5929     bind(next);
5930     ldr(rscratch2, Address(rthread, in_bytes(JavaThread::tlab_end_offset())));
5931     ldr(rscratch1, Address(rthread, in_bytes(JavaThread::tlab_top_offset())));
5932     cmp(rscratch2, rscratch1);
5933     br(Assembler::HS, ok);
5934     STOP("assert(top <= end)");
5935     should_not_reach_here();
5936 
5937     bind(ok);
5938     ldp(rscratch2, rscratch1, Address(post(sp, 16)));
5939   }
5940 #endif
5941 }
5942 
5943 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
5944   assert_different_registers(holder_klass, index, layout_info);
5945   InlineLayoutInfo array[2];
5946   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
5947   if (is_power_of_2(size)) {
5948     lsl(index, index, log2i_exact(size)); // Scale index by power of 2
5949   } else {
5950     mov(layout_info, size);
5951     mul(index, index, layout_info); // Scale the index to be the entry index * array_element_size
5952   }
5953   ldr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
5954   add(layout_info, layout_info, Array<InlineLayoutInfo>::base_offset_in_bytes());
5955   lea(layout_info, Address(layout_info, index));
5956 }
5957 
5958 // Writes to stack successive pages until offset reached to check for
5959 // stack overflow + shadow pages.  This clobbers tmp.
5960 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5961   assert_different_registers(tmp, size, rscratch1);
5962   mov(tmp, sp);
5963   // Bang stack for total size given plus shadow page size.
5964   // Bang one page at a time because large size can bang beyond yellow and
5965   // red zones.
5966   Label loop;
5967   mov(rscratch1, (int)os::vm_page_size());
5968   bind(loop);
5969   lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5970   subsw(size, size, rscratch1);
5971   str(size, Address(tmp));
5972   br(Assembler::GT, loop);
5973 
5974   // Bang down shadow pages too.
5975   // At this point, (tmp-0) is the last address touched, so don't
5976   // touch it again.  (It was touched as (tmp-pagesize) but then tmp
5977   // was post-decremented.)  Skip this address by starting at i=1, and
5978   // touch a few more pages below.  N.B.  It is important to touch all
5979   // the way down to and including i=StackShadowPages.
5980   for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
5981     // this could be any sized move but this is can be a debugging crumb
5982     // so the bigger the better.
5983     lea(tmp, Address(tmp, -(int)os::vm_page_size()));
5984     str(size, Address(tmp));
5985   }
5986 }
5987 
5988 // Move the address of the polling page into dest.
5989 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
5990   ldr(dest, Address(rthread, JavaThread::polling_page_offset()));
5991 }
5992 
5993 // Read the polling page.  The address of the polling page must
5994 // already be in r.
5995 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
5996   address mark;
5997   {
5998     InstructionMark im(this);
5999     code_section()->relocate(inst_mark(), rtype);
6000     ldrw(zr, Address(r, 0));
6001     mark = inst_mark();
6002   }
6003   verify_cross_modify_fence_not_required();
6004   return mark;
6005 }
6006 
6007 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
6008   uint64_t low_page = (uint64_t)CodeCache::low_bound() >> 12;
6009   uint64_t high_page = (uint64_t)(CodeCache::high_bound()-1) >> 12;
6010   uint64_t dest_page = (uint64_t)dest.target() >> 12;
6011   int64_t offset_low = dest_page - low_page;
6012   int64_t offset_high = dest_page - high_page;
6013 
6014   assert(is_valid_AArch64_address(dest.target()), "bad address");
6015   assert(dest.getMode() == Address::literal, "ADRP must be applied to a literal address");
6016 
6017   InstructionMark im(this);
6018   code_section()->relocate(inst_mark(), dest.rspec());
6019   // 8143067: Ensure that the adrp can reach the dest from anywhere within
6020   // the code cache so that if it is relocated we know it will still reach
6021   if (offset_high >= -(1<<20) && offset_low < (1<<20)) {
6022     _adrp(reg1, dest.target());
6023   } else {
6024     uint64_t target = (uint64_t)dest.target();
6025     uint64_t adrp_target
6026       = (target & 0xffffffffULL) | ((uint64_t)pc() & 0xffff00000000ULL);
6027 
6028     _adrp(reg1, (address)adrp_target);
6029     movk(reg1, target >> 32, 32);
6030   }
6031   byte_offset = (uint64_t)dest.target() & 0xfff;
6032 }
6033 
6034 void MacroAssembler::load_byte_map_base(Register reg) {
6035   CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
6036 
6037   // Strictly speaking the card table base isn't an address at all, and it might
6038   // even be negative. It is thus materialised as a constant.
6039   mov(reg, (uint64_t)ctbs->card_table_base_const());
6040 }
6041 
6042 #ifdef ASSERT
6043 void MacroAssembler::build_frame(int framesize) {
6044   build_frame(framesize, false);
6045 }
6046 #endif
6047 
6048 void MacroAssembler::build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills)) {
6049   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6050   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6051   protect_return_address();
6052   if (framesize < ((1 << 9) + 2 * wordSize)) {
6053     sub(sp, sp, framesize);
6054     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
6055       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
6056       stp(rscratch1, rscratch1, Address(sp, framesize - 2 * wordSize));
6057     } else {
6058       stp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6059     }
6060     if (PreserveFramePointer) add(rfp, sp, framesize - 2 * wordSize);
6061   } else {
6062     if (DEBUG_ONLY(zap_rfp_lr_spills ||) false) {
6063       mov_immediate64(rscratch1, ((uint64_t)badRegWordVal) << 32 | (uint64_t)badRegWordVal);
6064       stp(rscratch1, rscratch1, Address(pre(sp, -2 * wordSize)));
6065     } else {
6066       stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
6067     }
6068     if (PreserveFramePointer) mov(rfp, sp);
6069     if (framesize < ((1 << 12) + 2 * wordSize))
6070       sub(sp, sp, framesize - 2 * wordSize);
6071     else {
6072       mov(rscratch1, framesize - 2 * wordSize);
6073       sub(sp, sp, rscratch1);
6074     }
6075   }
6076   verify_cross_modify_fence_not_required();
6077 }
6078 
6079 void MacroAssembler::remove_frame(int framesize) {
6080   assert(framesize >= 2 * wordSize, "framesize must include space for FP/LR");
6081   assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
6082   if (framesize < ((1 << 9) + 2 * wordSize)) {
6083     ldp(rfp, lr, Address(sp, framesize - 2 * wordSize));
6084     add(sp, sp, framesize);
6085   } else {
6086     if (framesize < ((1 << 12) + 2 * wordSize))
6087       add(sp, sp, framesize - 2 * wordSize);
6088     else {
6089       mov(rscratch1, framesize - 2 * wordSize);
6090       add(sp, sp, rscratch1);
6091     }
6092     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6093   }
6094   authenticate_return_address();
6095 }
6096 
6097 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
6098   if (needs_stack_repair) {
6099     // The method has a scalarized entry point (where fields of value object arguments
6100     // are passed through registers and stack), and a non-scalarized entry point (where
6101     // value object arguments are given as oops). The non-scalarized entry point will
6102     // first load each field of value object arguments and store them in registers and on
6103     // the stack in a way compatible with the scalarized entry point. To do so, some extra
6104     // stack space might be reserved (if argument registers are not enough). On leaving the
6105     // method, this space must be freed.
6106     //
6107     // In case we used the non-scalarized entry point the stack looks like this:
6108     //
6109     // | Arguments from caller     |
6110     // |---------------------------|  <-- caller's SP
6111     // | Saved LR #1               |
6112     // | Saved FP #1               |
6113     // |---------------------------|
6114     // | Extension space for       |
6115     // |   inline arg (un)packing  |
6116     // |---------------------------|  <-- start of this method's frame
6117     // | Saved LR #2               |
6118     // | Saved FP #2               |
6119     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6120     // | sp_inc                    |
6121     // | method locals             |
6122     // |---------------------------|  <-- SP
6123     //
6124     // There are two copies of FP and LR on the stack. They will be identical at
6125     // first, but that can change.
6126     // If the caller has been deoptimized, LR #1 will be patched to point at the
6127     // deopt blob, and LR #2 will still point into the old method.
6128     // If the saved FP (x29) was not used as the frame pointer, but to store an
6129     // oop, the GC will be aware only of FP #1 as the spilled location of x29 and
6130     // will fix only this one. Overall, FP/LR #2 are not reliable and are simply
6131     // needed to add space between the extension space and the locals, as there
6132     // would be between the real arguments and the locals if we don't need to
6133     // do unpacking (from the scalarized entry point).
6134     //
6135     // When restoring, one must then load FP #1 into x29, and LR #1 into x30,
6136     // while keeping in mind that from the scalarized entry point, there will be
6137     // only one copy of each. Indeed, in the case we used the scalarized calling
6138     // convention, the stack looks like this:
6139     //
6140     // | Arguments from caller     |
6141     // |---------------------------|  <-- caller's SP / start of this method's frame
6142     // | Saved LR                  |
6143     // | Saved FP                  |
6144     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
6145     // | sp_inc                    |
6146     // | method locals             |
6147     // |---------------------------|  <-- SP
6148     //
6149     // The sp_inc stack slot holds the total size of the frame including the
6150     // extension space minus two words for the saved FP and LR. That is how to
6151     // find FP/LR #1. This size is expressed in bytes. Be careful when using it
6152     // from C++ in pointer arithmetic; you might need to divide it by wordSize.
6153     //
6154     // One can find sp_inc since the start the method's frame is SP + initial_framesize.
6155 
6156     int sp_inc_offset = initial_framesize - 3 * wordSize;  // Immediately below saved LR and FP
6157 
6158     ldr(rscratch1, Address(sp, sp_inc_offset));
6159     add(sp, sp, rscratch1);
6160     ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
6161   } else {
6162     remove_frame(initial_framesize);
6163   }
6164 }
6165 
6166 void MacroAssembler::save_stack_increment(int sp_inc, int frame_size) {
6167   int real_frame_size = frame_size + sp_inc;
6168   assert(sp_inc == 0 || sp_inc > 2*wordSize, "invalid sp_inc value");
6169   assert(real_frame_size >= 2*wordSize, "frame size must include FP/LR space");
6170   assert((real_frame_size & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
6171 
6172   int sp_inc_offset = frame_size - 3 * wordSize;  // Immediately below saved LR and FP
6173 
6174   // Subtract two words for the saved FP and LR as these will be popped
6175   // separately. See remove_frame above.
6176   mov(rscratch1, real_frame_size - 2*wordSize);
6177   str(rscratch1, Address(sp, sp_inc_offset));
6178 }
6179 
6180 // This method counts leading positive bytes (highest bit not set) in provided byte array
6181 address MacroAssembler::count_positives(Register ary1, Register len, Register result) {
6182     // Simple and most common case of aligned small array which is not at the
6183     // end of memory page is placed here. All other cases are in stub.
6184     Label LOOP, END, STUB, STUB_LONG, SET_RESULT, DONE;
6185     const uint64_t UPPER_BIT_MASK=0x8080808080808080;
6186     assert_different_registers(ary1, len, result);
6187 
6188     mov(result, len);
6189     cmpw(len, 0);
6190     br(LE, DONE);
6191     cmpw(len, 4 * wordSize);
6192     br(GE, STUB_LONG); // size > 32 then go to stub
6193 
6194     int shift = 64 - exact_log2(os::vm_page_size());
6195     lsl(rscratch1, ary1, shift);
6196     mov(rscratch2, (size_t)(4 * wordSize) << shift);
6197     adds(rscratch2, rscratch1, rscratch2);  // At end of page?
6198     br(CS, STUB); // at the end of page then go to stub
6199     subs(len, len, wordSize);
6200     br(LT, END);
6201 
6202   BIND(LOOP);
6203     ldr(rscratch1, Address(post(ary1, wordSize)));
6204     tst(rscratch1, UPPER_BIT_MASK);
6205     br(NE, SET_RESULT);
6206     subs(len, len, wordSize);
6207     br(GE, LOOP);
6208     cmpw(len, -wordSize);
6209     br(EQ, DONE);
6210 
6211   BIND(END);
6212     ldr(rscratch1, Address(ary1));
6213     sub(rscratch2, zr, len, LSL, 3); // LSL 3 is to get bits from bytes
6214     lslv(rscratch1, rscratch1, rscratch2);
6215     tst(rscratch1, UPPER_BIT_MASK);
6216     br(NE, SET_RESULT);
6217     b(DONE);
6218 
6219   BIND(STUB);
6220     RuntimeAddress count_pos = RuntimeAddress(StubRoutines::aarch64::count_positives());
6221     assert(count_pos.target() != nullptr, "count_positives stub has not been generated");
6222     address tpc1 = trampoline_call(count_pos);
6223     if (tpc1 == nullptr) {
6224       DEBUG_ONLY(reset_labels(STUB_LONG, SET_RESULT, DONE));
6225       postcond(pc() == badAddress);
6226       return nullptr;
6227     }
6228     b(DONE);
6229 
6230   BIND(STUB_LONG);
6231     RuntimeAddress count_pos_long = RuntimeAddress(StubRoutines::aarch64::count_positives_long());
6232     assert(count_pos_long.target() != nullptr, "count_positives_long stub has not been generated");
6233     address tpc2 = trampoline_call(count_pos_long);
6234     if (tpc2 == nullptr) {
6235       DEBUG_ONLY(reset_labels(SET_RESULT, DONE));
6236       postcond(pc() == badAddress);
6237       return nullptr;
6238     }
6239     b(DONE);
6240 
6241   BIND(SET_RESULT);
6242 
6243     add(len, len, wordSize);
6244     sub(result, result, len);
6245 
6246   BIND(DONE);
6247   postcond(pc() != badAddress);
6248   return pc();
6249 }
6250 
6251 // Clobbers: rscratch1, rscratch2, rflags
6252 // May also clobber v0-v7 when (!UseSimpleArrayEquals && UseSIMDForArrayEquals)
6253 address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
6254                                       Register tmp4, Register tmp5, Register result,
6255                                       Register cnt1, int elem_size) {
6256   Label DONE, SAME;
6257   Register tmp1 = rscratch1;
6258   Register tmp2 = rscratch2;
6259   int elem_per_word = wordSize/elem_size;
6260   int log_elem_size = exact_log2(elem_size);
6261   int klass_offset  = arrayOopDesc::klass_offset_in_bytes();
6262   int length_offset = arrayOopDesc::length_offset_in_bytes();
6263   int base_offset
6264     = arrayOopDesc::base_offset_in_bytes(elem_size == 2 ? T_CHAR : T_BYTE);
6265   // When the length offset is not aligned to 8 bytes,
6266   // then we align it down. This is valid because the new
6267   // offset will always be the klass which is the same
6268   // for type arrays.
6269   int start_offset = align_down(length_offset, BytesPerWord);
6270   int extra_length = base_offset - start_offset;
6271   assert(start_offset == length_offset || start_offset == klass_offset,
6272          "start offset must be 8-byte-aligned or be the klass offset");
6273   assert(base_offset != start_offset, "must include the length field");
6274   extra_length = extra_length / elem_size; // We count in elements, not bytes.
6275   int stubBytesThreshold = 3 * 64 + (UseSIMDForArrayEquals ? 0 : 16);
6276 
6277   assert(elem_size == 1 || elem_size == 2, "must be char or byte");
6278   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6279 
6280 #ifndef PRODUCT
6281   {
6282     const char kind = (elem_size == 2) ? 'U' : 'L';
6283     char comment[64];
6284     os::snprintf_checked(comment, sizeof comment, "array_equals%c{", kind);
6285     BLOCK_COMMENT(comment);
6286   }
6287 #endif
6288 
6289   // if (a1 == a2)
6290   //     return true;
6291   cmpoop(a1, a2); // May have read barriers for a1 and a2.
6292   br(EQ, SAME);
6293 
6294   if (UseSimpleArrayEquals) {
6295     Label NEXT_WORD, SHORT, TAIL03, TAIL01, A_MIGHT_BE_NULL, A_IS_NOT_NULL;
6296     // if (a1 == nullptr || a2 == nullptr)
6297     //     return false;
6298     // a1 & a2 == 0 means (some-pointer is null) or
6299     // (very-rare-or-even-probably-impossible-pointer-values)
6300     // so, we can save one branch in most cases
6301     tst(a1, a2);
6302     mov(result, false);
6303     br(EQ, A_MIGHT_BE_NULL);
6304     // if (a1.length != a2.length)
6305     //      return false;
6306     bind(A_IS_NOT_NULL);
6307     ldrw(cnt1, Address(a1, length_offset));
6308     ldrw(tmp5, Address(a2, length_offset));
6309     cmp(cnt1, tmp5);
6310     br(NE, DONE); // If lengths differ, return false
6311     // Increase loop counter by diff between base- and actual start-offset.
6312     addw(cnt1, cnt1, extra_length);
6313     lea(a1, Address(a1, start_offset));
6314     lea(a2, Address(a2, start_offset));
6315     // Check for short strings, i.e. smaller than wordSize.
6316     subs(cnt1, cnt1, elem_per_word);
6317     br(Assembler::LT, SHORT);
6318     // Main 8 byte comparison loop.
6319     bind(NEXT_WORD); {
6320       ldr(tmp1, Address(post(a1, wordSize)));
6321       ldr(tmp2, Address(post(a2, wordSize)));
6322       subs(cnt1, cnt1, elem_per_word);
6323       eor(tmp5, tmp1, tmp2);
6324       cbnz(tmp5, DONE);
6325     } br(GT, NEXT_WORD);
6326     // Last longword.  In the case where length == 4 we compare the
6327     // same longword twice, but that's still faster than another
6328     // conditional branch.
6329     // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6330     // length == 4.
6331     if (log_elem_size > 0)
6332       lsl(cnt1, cnt1, log_elem_size);
6333     ldr(tmp3, Address(a1, cnt1));
6334     ldr(tmp4, Address(a2, cnt1));
6335     eor(tmp5, tmp3, tmp4);
6336     cbnz(tmp5, DONE);
6337     b(SAME);
6338     bind(A_MIGHT_BE_NULL);
6339     // in case both a1 and a2 are not-null, proceed with loads
6340     cbz(a1, DONE);
6341     cbz(a2, DONE);
6342     b(A_IS_NOT_NULL);
6343     bind(SHORT);
6344 
6345     tbz(cnt1, 2 - log_elem_size, TAIL03); // 0-7 bytes left.
6346     {
6347       ldrw(tmp1, Address(post(a1, 4)));
6348       ldrw(tmp2, Address(post(a2, 4)));
6349       eorw(tmp5, tmp1, tmp2);
6350       cbnzw(tmp5, DONE);
6351     }
6352     bind(TAIL03);
6353     tbz(cnt1, 1 - log_elem_size, TAIL01); // 0-3 bytes left.
6354     {
6355       ldrh(tmp3, Address(post(a1, 2)));
6356       ldrh(tmp4, Address(post(a2, 2)));
6357       eorw(tmp5, tmp3, tmp4);
6358       cbnzw(tmp5, DONE);
6359     }
6360     bind(TAIL01);
6361     if (elem_size == 1) { // Only needed when comparing byte arrays.
6362       tbz(cnt1, 0, SAME); // 0-1 bytes left.
6363       {
6364         ldrb(tmp1, a1);
6365         ldrb(tmp2, a2);
6366         eorw(tmp5, tmp1, tmp2);
6367         cbnzw(tmp5, DONE);
6368       }
6369     }
6370   } else {
6371     Label NEXT_DWORD, SHORT, TAIL, TAIL2, STUB,
6372         CSET_EQ, LAST_CHECK;
6373     mov(result, false);
6374     cbz(a1, DONE);
6375     ldrw(cnt1, Address(a1, length_offset));
6376     cbz(a2, DONE);
6377     ldrw(tmp5, Address(a2, length_offset));
6378     cmp(cnt1, tmp5);
6379     br(NE, DONE); // If lengths differ, return false
6380     // Increase loop counter by diff between base- and actual start-offset.
6381     addw(cnt1, cnt1, extra_length);
6382 
6383     // on most CPUs a2 is still "locked"(surprisingly) in ldrw and it's
6384     // faster to perform another branch before comparing a1 and a2
6385     cmp(cnt1, (u1)elem_per_word);
6386     br(LE, SHORT); // short or same
6387     ldr(tmp3, Address(pre(a1, start_offset)));
6388     subs(zr, cnt1, stubBytesThreshold);
6389     br(GE, STUB);
6390     ldr(tmp4, Address(pre(a2, start_offset)));
6391     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6392 
6393     // Main 16 byte comparison loop with 2 exits
6394     bind(NEXT_DWORD); {
6395       ldr(tmp1, Address(pre(a1, wordSize)));
6396       ldr(tmp2, Address(pre(a2, wordSize)));
6397       subs(cnt1, cnt1, 2 * elem_per_word);
6398       br(LE, TAIL);
6399       eor(tmp4, tmp3, tmp4);
6400       cbnz(tmp4, DONE);
6401       ldr(tmp3, Address(pre(a1, wordSize)));
6402       ldr(tmp4, Address(pre(a2, wordSize)));
6403       cmp(cnt1, (u1)elem_per_word);
6404       br(LE, TAIL2);
6405       cmp(tmp1, tmp2);
6406     } br(EQ, NEXT_DWORD);
6407     b(DONE);
6408 
6409     bind(TAIL);
6410     eor(tmp4, tmp3, tmp4);
6411     eor(tmp2, tmp1, tmp2);
6412     lslv(tmp2, tmp2, tmp5);
6413     orr(tmp5, tmp4, tmp2);
6414     cmp(tmp5, zr);
6415     b(CSET_EQ);
6416 
6417     bind(TAIL2);
6418     eor(tmp2, tmp1, tmp2);
6419     cbnz(tmp2, DONE);
6420     b(LAST_CHECK);
6421 
6422     bind(STUB);
6423     ldr(tmp4, Address(pre(a2, start_offset)));
6424     if (elem_size == 2) { // convert to byte counter
6425       lsl(cnt1, cnt1, 1);
6426     }
6427     eor(tmp5, tmp3, tmp4);
6428     cbnz(tmp5, DONE);
6429     RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_array_equals());
6430     assert(stub.target() != nullptr, "array_equals_long stub has not been generated");
6431     address tpc = trampoline_call(stub);
6432     if (tpc == nullptr) {
6433       DEBUG_ONLY(reset_labels(SHORT, LAST_CHECK, CSET_EQ, SAME, DONE));
6434       postcond(pc() == badAddress);
6435       return nullptr;
6436     }
6437     b(DONE);
6438 
6439     // (a1 != null && a2 == null) || (a1 != null && a2 != null && a1 == a2)
6440     // so, if a2 == null => return false(0), else return true, so we can return a2
6441     mov(result, a2);
6442     b(DONE);
6443     bind(SHORT);
6444     sub(tmp5, zr, cnt1, LSL, 3 + log_elem_size);
6445     ldr(tmp3, Address(a1, start_offset));
6446     ldr(tmp4, Address(a2, start_offset));
6447     bind(LAST_CHECK);
6448     eor(tmp4, tmp3, tmp4);
6449     lslv(tmp5, tmp4, tmp5);
6450     cmp(tmp5, zr);
6451     bind(CSET_EQ);
6452     cset(result, EQ);
6453     b(DONE);
6454   }
6455 
6456   bind(SAME);
6457   mov(result, true);
6458   // That's it.
6459   bind(DONE);
6460 
6461   BLOCK_COMMENT("} array_equals");
6462   postcond(pc() != badAddress);
6463   return pc();
6464 }
6465 
6466 // Compare Strings
6467 
6468 // For Strings we're passed the address of the first characters in a1
6469 // and a2 and the length in cnt1.
6470 // There are two implementations.  For arrays >= 8 bytes, all
6471 // comparisons (including the final one, which may overlap) are
6472 // performed 8 bytes at a time.  For strings < 8 bytes, we compare a
6473 // halfword, then a short, and then a byte.
6474 
6475 void MacroAssembler::string_equals(Register a1, Register a2,
6476                                    Register result, Register cnt1)
6477 {
6478   Label SAME, DONE, SHORT, NEXT_WORD;
6479   Register tmp1 = rscratch1;
6480   Register tmp2 = rscratch2;
6481 
6482   assert_different_registers(a1, a2, result, cnt1, rscratch1, rscratch2);
6483 
6484 #ifndef PRODUCT
6485   {
6486     char comment[64];
6487     os::snprintf_checked(comment, sizeof comment, "{string_equalsL");
6488     BLOCK_COMMENT(comment);
6489   }
6490 #endif
6491 
6492   mov(result, false);
6493 
6494   // Check for short strings, i.e. smaller than wordSize.
6495   subs(cnt1, cnt1, wordSize);
6496   br(Assembler::LT, SHORT);
6497   // Main 8 byte comparison loop.
6498   bind(NEXT_WORD); {
6499     ldr(tmp1, Address(post(a1, wordSize)));
6500     ldr(tmp2, Address(post(a2, wordSize)));
6501     subs(cnt1, cnt1, wordSize);
6502     eor(tmp1, tmp1, tmp2);
6503     cbnz(tmp1, DONE);
6504   } br(GT, NEXT_WORD);
6505   // Last longword.  In the case where length == 4 we compare the
6506   // same longword twice, but that's still faster than another
6507   // conditional branch.
6508   // cnt1 could be 0, -1, -2, -3, -4 for chars; -4 only happens when
6509   // length == 4.
6510   ldr(tmp1, Address(a1, cnt1));
6511   ldr(tmp2, Address(a2, cnt1));
6512   eor(tmp2, tmp1, tmp2);
6513   cbnz(tmp2, DONE);
6514   b(SAME);
6515 
6516   bind(SHORT);
6517   Label TAIL03, TAIL01;
6518 
6519   tbz(cnt1, 2, TAIL03); // 0-7 bytes left.
6520   {
6521     ldrw(tmp1, Address(post(a1, 4)));
6522     ldrw(tmp2, Address(post(a2, 4)));
6523     eorw(tmp1, tmp1, tmp2);
6524     cbnzw(tmp1, DONE);
6525   }
6526   bind(TAIL03);
6527   tbz(cnt1, 1, TAIL01); // 0-3 bytes left.
6528   {
6529     ldrh(tmp1, Address(post(a1, 2)));
6530     ldrh(tmp2, Address(post(a2, 2)));
6531     eorw(tmp1, tmp1, tmp2);
6532     cbnzw(tmp1, DONE);
6533   }
6534   bind(TAIL01);
6535   tbz(cnt1, 0, SAME); // 0-1 bytes left.
6536     {
6537     ldrb(tmp1, a1);
6538     ldrb(tmp2, a2);
6539     eorw(tmp1, tmp1, tmp2);
6540     cbnzw(tmp1, DONE);
6541   }
6542   // Arrays are equal.
6543   bind(SAME);
6544   mov(result, true);
6545 
6546   // That's it.
6547   bind(DONE);
6548   BLOCK_COMMENT("} string_equals");
6549 }
6550 
6551 
6552 // The size of the blocks erased by the zero_blocks stub.  We must
6553 // handle anything smaller than this ourselves in zero_words().
6554 const int MacroAssembler::zero_words_block_size = 8;
6555 
6556 // zero_words() is used by C2 ClearArray patterns and by
6557 // C1_MacroAssembler.  It is as small as possible, handling small word
6558 // counts locally and delegating anything larger to the zero_blocks
6559 // stub.  It is expanded many times in compiled code, so it is
6560 // important to keep it short.
6561 
6562 // ptr:   Address of a buffer to be zeroed.
6563 // cnt:   Count in HeapWords.
6564 //
6565 // ptr, cnt, rscratch1, and rscratch2 are clobbered.
6566 address MacroAssembler::zero_words(Register ptr, Register cnt)
6567 {
6568   assert(is_power_of_2(zero_words_block_size), "adjust this");
6569 
6570   BLOCK_COMMENT("zero_words {");
6571   assert(ptr == r10 && cnt == r11, "mismatch in register usage");
6572   RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6573   assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6574 
6575   subs(rscratch1, cnt, zero_words_block_size);
6576   Label around;
6577   br(LO, around);
6578   {
6579     RuntimeAddress zero_blocks = RuntimeAddress(StubRoutines::aarch64::zero_blocks());
6580     assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
6581     // Make sure this is a C2 compilation. C1 allocates space only for
6582     // trampoline stubs generated by Call LIR ops, and in any case it
6583     // makes sense for a C1 compilation task to proceed as quickly as
6584     // possible.
6585     CompileTask* task;
6586     if (StubRoutines::aarch64::complete()
6587         && Thread::current()->is_Compiler_thread()
6588         && (task = ciEnv::current()->task())
6589         && is_c2_compile(task->comp_level())) {
6590       address tpc = trampoline_call(zero_blocks);
6591       if (tpc == nullptr) {
6592         DEBUG_ONLY(reset_labels(around));
6593         return nullptr;
6594       }
6595     } else {
6596       far_call(zero_blocks);
6597     }
6598   }
6599   bind(around);
6600 
6601   // We have a few words left to do. zero_blocks has adjusted r10 and r11
6602   // for us.
6603   for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
6604     Label l;
6605     tbz(cnt, exact_log2(i), l);
6606     for (int j = 0; j < i; j += 2) {
6607       stp(zr, zr, post(ptr, 2 * BytesPerWord));
6608     }
6609     bind(l);
6610   }
6611   {
6612     Label l;
6613     tbz(cnt, 0, l);
6614     str(zr, Address(ptr));
6615     bind(l);
6616   }
6617 
6618   BLOCK_COMMENT("} zero_words");
6619   return pc();
6620 }
6621 
6622 // base:         Address of a buffer to be zeroed, 8 bytes aligned.
6623 // cnt:          Immediate count in HeapWords.
6624 //
6625 // r10, r11, rscratch1, and rscratch2 are clobbered.
6626 address MacroAssembler::zero_words(Register base, uint64_t cnt)
6627 {
6628   assert(wordSize <= BlockZeroingLowLimit,
6629             "increase BlockZeroingLowLimit");
6630   address result = nullptr;
6631   if (cnt <= (uint64_t)BlockZeroingLowLimit / BytesPerWord) {
6632 #ifndef PRODUCT
6633     {
6634       char buf[64];
6635       os::snprintf_checked(buf, sizeof buf, "zero_words (count = %" PRIu64 ") {", cnt);
6636       BLOCK_COMMENT(buf);
6637     }
6638 #endif
6639     if (cnt >= 16) {
6640       uint64_t loops = cnt/16;
6641       if (loops > 1) {
6642         mov(rscratch2, loops - 1);
6643       }
6644       {
6645         Label loop;
6646         bind(loop);
6647         for (int i = 0; i < 16; i += 2) {
6648           stp(zr, zr, Address(base, i * BytesPerWord));
6649         }
6650         add(base, base, 16 * BytesPerWord);
6651         if (loops > 1) {
6652           subs(rscratch2, rscratch2, 1);
6653           br(GE, loop);
6654         }
6655       }
6656     }
6657     cnt %= 16;
6658     int i = cnt & 1;  // store any odd word to start
6659     if (i) str(zr, Address(base));
6660     for (; i < (int)cnt; i += 2) {
6661       stp(zr, zr, Address(base, i * wordSize));
6662     }
6663     BLOCK_COMMENT("} zero_words");
6664     result = pc();
6665   } else {
6666     mov(r10, base); mov(r11, cnt);
6667     result = zero_words(r10, r11);
6668   }
6669   return result;
6670 }
6671 
6672 // Zero blocks of memory by using DC ZVA.
6673 //
6674 // Aligns the base address first sufficiently for DC ZVA, then uses
6675 // DC ZVA repeatedly for every full block.  cnt is the size to be
6676 // zeroed in HeapWords.  Returns the count of words left to be zeroed
6677 // in cnt.
6678 //
6679 // NOTE: This is intended to be used in the zero_blocks() stub.  If
6680 // you want to use it elsewhere, note that cnt must be >= 2*zva_length.
6681 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt) {
6682   Register tmp = rscratch1;
6683   Register tmp2 = rscratch2;
6684   int zva_length = VM_Version::zva_length();
6685   Label initial_table_end, loop_zva;
6686   Label fini;
6687 
6688   // Base must be 16 byte aligned. If not just return and let caller handle it
6689   tst(base, 0x0f);
6690   br(Assembler::NE, fini);
6691   // Align base with ZVA length.
6692   neg(tmp, base);
6693   andr(tmp, tmp, zva_length - 1);
6694 
6695   // tmp: the number of bytes to be filled to align the base with ZVA length.
6696   add(base, base, tmp);
6697   sub(cnt, cnt, tmp, Assembler::ASR, 3);
6698   adr(tmp2, initial_table_end);
6699   sub(tmp2, tmp2, tmp, Assembler::LSR, 2);
6700   br(tmp2);
6701 
6702   for (int i = -zva_length + 16; i < 0; i += 16)
6703     stp(zr, zr, Address(base, i));
6704   bind(initial_table_end);
6705 
6706   sub(cnt, cnt, zva_length >> 3);
6707   bind(loop_zva);
6708   dc(Assembler::ZVA, base);
6709   subs(cnt, cnt, zva_length >> 3);
6710   add(base, base, zva_length);
6711   br(Assembler::GE, loop_zva);
6712   add(cnt, cnt, zva_length >> 3); // count not zeroed by DC ZVA
6713   bind(fini);
6714 }
6715 
6716 // base:   Address of a buffer to be filled, 8 bytes aligned.
6717 // cnt:    Count in 8-byte unit.
6718 // value:  Value to be filled with.
6719 // base will point to the end of the buffer after filling.
6720 void MacroAssembler::fill_words(Register base, Register cnt, Register value)
6721 {
6722 //  Algorithm:
6723 //
6724 //    if (cnt == 0) {
6725 //      return;
6726 //    }
6727 //    if ((p & 8) != 0) {
6728 //      *p++ = v;
6729 //    }
6730 //
6731 //    scratch1 = cnt & 14;
6732 //    cnt -= scratch1;
6733 //    p += scratch1;
6734 //    switch (scratch1 / 2) {
6735 //      do {
6736 //        cnt -= 16;
6737 //          p[-16] = v;
6738 //          p[-15] = v;
6739 //        case 7:
6740 //          p[-14] = v;
6741 //          p[-13] = v;
6742 //        case 6:
6743 //          p[-12] = v;
6744 //          p[-11] = v;
6745 //          // ...
6746 //        case 1:
6747 //          p[-2] = v;
6748 //          p[-1] = v;
6749 //        case 0:
6750 //          p += 16;
6751 //      } while (cnt);
6752 //    }
6753 //    if ((cnt & 1) == 1) {
6754 //      *p++ = v;
6755 //    }
6756 
6757   assert_different_registers(base, cnt, value, rscratch1, rscratch2);
6758 
6759   Label fini, skip, entry, loop;
6760   const int unroll = 8; // Number of stp instructions we'll unroll
6761 
6762   cbz(cnt, fini);
6763   tbz(base, 3, skip);
6764   str(value, Address(post(base, 8)));
6765   sub(cnt, cnt, 1);
6766   bind(skip);
6767 
6768   andr(rscratch1, cnt, (unroll-1) * 2);
6769   sub(cnt, cnt, rscratch1);
6770   add(base, base, rscratch1, Assembler::LSL, 3);
6771   adr(rscratch2, entry);
6772   sub(rscratch2, rscratch2, rscratch1, Assembler::LSL, 1);
6773   br(rscratch2);
6774 
6775   bind(loop);
6776   add(base, base, unroll * 16);
6777   for (int i = -unroll; i < 0; i++)
6778     stp(value, value, Address(base, i * 16));
6779   bind(entry);
6780   subs(cnt, cnt, unroll * 2);
6781   br(Assembler::GE, loop);
6782 
6783   tbz(cnt, 0, fini);
6784   str(value, Address(post(base, 8)));
6785   bind(fini);
6786 }
6787 
6788 // Intrinsic for
6789 //
6790 // - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6791 //   Encodes char[] to byte[] in ISO-8859-1
6792 //
6793 // - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6794 //   Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
6795 //
6796 // - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6797 //   Encodes char[] to byte[] in ASCII
6798 //
6799 // This version always returns the number of characters copied, and does not
6800 // clobber the 'len' register. A successful copy will complete with the post-
6801 // condition: 'res' == 'len', while an unsuccessful copy will exit with the
6802 // post-condition: 0 <= 'res' < 'len'.
6803 //
6804 // NOTE: Attempts to use 'ld2' (and 'umaxv' in the ISO part) has proven to
6805 //       degrade performance (on Ampere Altra - Neoverse N1), to an extent
6806 //       beyond the acceptable, even though the footprint would be smaller.
6807 //       Using 'umaxv' in the ASCII-case comes with a small penalty but does
6808 //       avoid additional bloat.
6809 //
6810 // Clobbers: src, dst, res, rscratch1, rscratch2, rflags
6811 void MacroAssembler::encode_iso_array(Register src, Register dst,
6812                                       Register len, Register res, bool ascii,
6813                                       FloatRegister vtmp0, FloatRegister vtmp1,
6814                                       FloatRegister vtmp2, FloatRegister vtmp3,
6815                                       FloatRegister vtmp4, FloatRegister vtmp5)
6816 {
6817   Register cnt = res;
6818   Register max = rscratch1;
6819   Register chk = rscratch2;
6820 
6821   prfm(Address(src), PLDL1STRM);
6822   movw(cnt, len);
6823 
6824 #define ASCII(insn) do { if (ascii) { insn; } } while (0)
6825 
6826   Label LOOP_32, DONE_32, FAIL_32;
6827 
6828   BIND(LOOP_32);
6829   {
6830     cmpw(cnt, 32);
6831     br(LT, DONE_32);
6832     ld1(vtmp0, vtmp1, vtmp2, vtmp3, T8H, Address(post(src, 64)));
6833     // Extract lower bytes.
6834     FloatRegister vlo0 = vtmp4;
6835     FloatRegister vlo1 = vtmp5;
6836     uzp1(vlo0, T16B, vtmp0, vtmp1);
6837     uzp1(vlo1, T16B, vtmp2, vtmp3);
6838     // Merge bits...
6839     orr(vtmp0, T16B, vtmp0, vtmp1);
6840     orr(vtmp2, T16B, vtmp2, vtmp3);
6841     // Extract merged upper bytes.
6842     FloatRegister vhix = vtmp0;
6843     uzp2(vhix, T16B, vtmp0, vtmp2);
6844     // ISO-check on hi-parts (all zero).
6845     //                          ASCII-check on lo-parts (no sign).
6846     FloatRegister vlox = vtmp1; // Merge lower bytes.
6847                                 ASCII(orr(vlox, T16B, vlo0, vlo1));
6848     umov(chk, vhix, D, 1);      ASCII(cm(LT, vlox, T16B, vlox));
6849     fmovd(max, vhix);           ASCII(umaxv(vlox, T16B, vlox));
6850     orr(chk, chk, max);         ASCII(umov(max, vlox, B, 0));
6851                                 ASCII(orr(chk, chk, max));
6852     cbnz(chk, FAIL_32);
6853     subw(cnt, cnt, 32);
6854     st1(vlo0, vlo1, T16B, Address(post(dst, 32)));
6855     b(LOOP_32);
6856   }
6857   BIND(FAIL_32);
6858   sub(src, src, 64);
6859   BIND(DONE_32);
6860 
6861   Label LOOP_8, SKIP_8;
6862 
6863   BIND(LOOP_8);
6864   {
6865     cmpw(cnt, 8);
6866     br(LT, SKIP_8);
6867     FloatRegister vhi = vtmp0;
6868     FloatRegister vlo = vtmp1;
6869     ld1(vtmp3, T8H, src);
6870     uzp1(vlo, T16B, vtmp3, vtmp3);
6871     uzp2(vhi, T16B, vtmp3, vtmp3);
6872     // ISO-check on hi-parts (all zero).
6873     //                          ASCII-check on lo-parts (no sign).
6874                                 ASCII(cm(LT, vtmp2, T16B, vlo));
6875     fmovd(chk, vhi);            ASCII(umaxv(vtmp2, T16B, vtmp2));
6876                                 ASCII(umov(max, vtmp2, B, 0));
6877                                 ASCII(orr(chk, chk, max));
6878     cbnz(chk, SKIP_8);
6879 
6880     strd(vlo, Address(post(dst, 8)));
6881     subw(cnt, cnt, 8);
6882     add(src, src, 16);
6883     b(LOOP_8);
6884   }
6885   BIND(SKIP_8);
6886 
6887 #undef ASCII
6888 
6889   Label LOOP, DONE;
6890 
6891   cbz(cnt, DONE);
6892   BIND(LOOP);
6893   {
6894     Register chr = rscratch1;
6895     ldrh(chr, Address(post(src, 2)));
6896     tst(chr, ascii ? 0xff80 : 0xff00);
6897     br(NE, DONE);
6898     strb(chr, Address(post(dst, 1)));
6899     subs(cnt, cnt, 1);
6900     br(GT, LOOP);
6901   }
6902   BIND(DONE);
6903   // Return index where we stopped.
6904   subw(res, len, cnt);
6905 }
6906 
6907 // Inflate byte[] array to char[].
6908 // Clobbers: src, dst, len, rflags, rscratch1, v0-v6
6909 address MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
6910                                            FloatRegister vtmp1, FloatRegister vtmp2,
6911                                            FloatRegister vtmp3, Register tmp4) {
6912   Label big, done, after_init, to_stub;
6913 
6914   assert_different_registers(src, dst, len, tmp4, rscratch1);
6915 
6916   fmovd(vtmp1, 0.0);
6917   lsrw(tmp4, len, 3);
6918   bind(after_init);
6919   cbnzw(tmp4, big);
6920   // Short string: less than 8 bytes.
6921   {
6922     Label loop, tiny;
6923 
6924     cmpw(len, 4);
6925     br(LT, tiny);
6926     // Use SIMD to do 4 bytes.
6927     ldrs(vtmp2, post(src, 4));
6928     zip1(vtmp3, T8B, vtmp2, vtmp1);
6929     subw(len, len, 4);
6930     strd(vtmp3, post(dst, 8));
6931 
6932     cbzw(len, done);
6933 
6934     // Do the remaining bytes by steam.
6935     bind(loop);
6936     ldrb(tmp4, post(src, 1));
6937     strh(tmp4, post(dst, 2));
6938     subw(len, len, 1);
6939 
6940     bind(tiny);
6941     cbnz(len, loop);
6942 
6943     b(done);
6944   }
6945 
6946   if (SoftwarePrefetchHintDistance >= 0) {
6947     bind(to_stub);
6948       RuntimeAddress stub = RuntimeAddress(StubRoutines::aarch64::large_byte_array_inflate());
6949       assert(stub.target() != nullptr, "large_byte_array_inflate stub has not been generated");
6950       address tpc = trampoline_call(stub);
6951       if (tpc == nullptr) {
6952         DEBUG_ONLY(reset_labels(big, done));
6953         postcond(pc() == badAddress);
6954         return nullptr;
6955       }
6956       b(after_init);
6957   }
6958 
6959   // Unpack the bytes 8 at a time.
6960   bind(big);
6961   {
6962     Label loop, around, loop_last, loop_start;
6963 
6964     if (SoftwarePrefetchHintDistance >= 0) {
6965       const int large_loop_threshold = (64 + 16)/8;
6966       ldrd(vtmp2, post(src, 8));
6967       andw(len, len, 7);
6968       cmp(tmp4, (u1)large_loop_threshold);
6969       br(GE, to_stub);
6970       b(loop_start);
6971 
6972       bind(loop);
6973       ldrd(vtmp2, post(src, 8));
6974       bind(loop_start);
6975       subs(tmp4, tmp4, 1);
6976       br(EQ, loop_last);
6977       zip1(vtmp2, T16B, vtmp2, vtmp1);
6978       ldrd(vtmp3, post(src, 8));
6979       st1(vtmp2, T8H, post(dst, 16));
6980       subs(tmp4, tmp4, 1);
6981       zip1(vtmp3, T16B, vtmp3, vtmp1);
6982       st1(vtmp3, T8H, post(dst, 16));
6983       br(NE, loop);
6984       b(around);
6985       bind(loop_last);
6986       zip1(vtmp2, T16B, vtmp2, vtmp1);
6987       st1(vtmp2, T8H, post(dst, 16));
6988       bind(around);
6989       cbz(len, done);
6990     } else {
6991       andw(len, len, 7);
6992       bind(loop);
6993       ldrd(vtmp2, post(src, 8));
6994       sub(tmp4, tmp4, 1);
6995       zip1(vtmp3, T16B, vtmp2, vtmp1);
6996       st1(vtmp3, T8H, post(dst, 16));
6997       cbnz(tmp4, loop);
6998     }
6999   }
7000 
7001   // Do the tail of up to 8 bytes.
7002   add(src, src, len);
7003   ldrd(vtmp3, Address(src, -8));
7004   add(dst, dst, len, ext::uxtw, 1);
7005   zip1(vtmp3, T16B, vtmp3, vtmp1);
7006   strq(vtmp3, Address(dst, -16));
7007 
7008   bind(done);
7009   postcond(pc() != badAddress);
7010   return pc();
7011 }
7012 
7013 // Compress char[] array to byte[].
7014 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
7015 // Return the array length if every element in array can be encoded,
7016 // otherwise, the index of first non-latin1 (> 0xff) character.
7017 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
7018                                          Register res,
7019                                          FloatRegister tmp0, FloatRegister tmp1,
7020                                          FloatRegister tmp2, FloatRegister tmp3,
7021                                          FloatRegister tmp4, FloatRegister tmp5) {
7022   encode_iso_array(src, dst, len, res, false, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
7023 }
7024 
7025 // java.math.round(double a)
7026 // Returns the closest long to the argument, with ties rounding to
7027 // positive infinity.  This requires some fiddling for corner
7028 // cases. We take care to avoid double rounding in e.g. (jlong)(a + 0.5).
7029 void MacroAssembler::java_round_double(Register dst, FloatRegister src,
7030                                        FloatRegister ftmp) {
7031   Label DONE;
7032   BLOCK_COMMENT("java_round_double: { ");
7033   fmovd(rscratch1, src);
7034   // Use RoundToNearestTiesAway unless src small and -ve.
7035   fcvtasd(dst, src);
7036   // Test if src >= 0 || abs(src) >= 0x1.0p52
7037   eor(rscratch1, rscratch1, UCONST64(1) << 63); // flip sign bit
7038   mov(rscratch2, julong_cast(0x1.0p52));
7039   cmp(rscratch1, rscratch2);
7040   br(HS, DONE); {
7041     // src < 0 && abs(src) < 0x1.0p52
7042     // src may have a fractional part, so add 0.5
7043     fmovd(ftmp, 0.5);
7044     faddd(ftmp, src, ftmp);
7045     // Convert double to jlong, use RoundTowardsNegative
7046     fcvtmsd(dst, ftmp);
7047   }
7048   bind(DONE);
7049   BLOCK_COMMENT("} java_round_double");
7050 }
7051 
7052 void MacroAssembler::java_round_float(Register dst, FloatRegister src,
7053                                       FloatRegister ftmp) {
7054   Label DONE;
7055   BLOCK_COMMENT("java_round_float: { ");
7056   fmovs(rscratch1, src);
7057   // Use RoundToNearestTiesAway unless src small and -ve.
7058   fcvtassw(dst, src);
7059   // Test if src >= 0 || abs(src) >= 0x1.0p23
7060   eor(rscratch1, rscratch1, 0x80000000); // flip sign bit
7061   mov(rscratch2, jint_cast(0x1.0p23f));
7062   cmp(rscratch1, rscratch2);
7063   br(HS, DONE); {
7064     // src < 0 && |src| < 0x1.0p23
7065     // src may have a fractional part, so add 0.5
7066     fmovs(ftmp, 0.5f);
7067     fadds(ftmp, src, ftmp);
7068     // Convert float to jint, use RoundTowardsNegative
7069     fcvtmssw(dst, ftmp);
7070   }
7071   bind(DONE);
7072   BLOCK_COMMENT("} java_round_float");
7073 }
7074 
7075 // get_thread() can be called anywhere inside generated code so we
7076 // need to save whatever non-callee save context might get clobbered
7077 // by the call to JavaThread::aarch64_get_thread_helper() or, indeed,
7078 // the call setup code.
7079 //
7080 // On Linux, aarch64_get_thread_helper() clobbers only r0, r1, and flags.
7081 // On other systems, the helper is a usual C function.
7082 //
7083 void MacroAssembler::get_thread(Register dst) {
7084   RegSet saved_regs =
7085     LINUX_ONLY(RegSet::range(r0, r1)  + lr - dst)
7086     NOT_LINUX (RegSet::range(r0, r17) + lr - dst);
7087 
7088   protect_return_address();
7089   push(saved_regs, sp);
7090 
7091   mov(lr, ExternalAddress(CAST_FROM_FN_PTR(address, JavaThread::aarch64_get_thread_helper)));
7092   blr(lr);
7093   if (dst != c_rarg0) {
7094     mov(dst, c_rarg0);
7095   }
7096 
7097   pop(saved_regs, sp);
7098   authenticate_return_address();
7099 }
7100 
7101 #ifdef COMPILER2
7102 // C2 compiled method's prolog code
7103 // Moved here from aarch64.ad to support Valhalla code below
7104 void MacroAssembler::verified_entry(Compile* C, int sp_inc) {
7105   if (C->clinit_barrier_on_entry()) {
7106     assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
7107 
7108     Label L_skip_barrier;
7109 
7110     mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
7111     clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
7112     far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
7113     bind(L_skip_barrier);
7114   }
7115 
7116   if (C->max_vector_size() > 0) {
7117     reinitialize_ptrue();
7118   }
7119 
7120   int bangsize = C->output()->bang_size_in_bytes();
7121   if (C->output()->need_stack_bang(bangsize))
7122     generate_stack_overflow_check(bangsize);
7123 
7124   // n.b. frame size includes space for return pc and rfp
7125   const long framesize = C->output()->frame_size_in_bytes();
7126   build_frame(framesize DEBUG_ONLY(COMMA sp_inc != 0));
7127 
7128   if (C->needs_stack_repair()) {
7129     save_stack_increment(sp_inc, framesize);
7130   }
7131 
7132   if (VerifyStackAtCalls) {
7133     Unimplemented();
7134   }
7135 }
7136 #endif // COMPILER2
7137 
7138 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
7139   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
7140   // An inline type might be returned. If fields are in registers we
7141   // need to allocate an inline type instance and initialize it with
7142   // the value of the fields.
7143   Label skip;
7144   // We only need a new buffered inline type if a new one is not returned
7145   tbz(r0, 0, skip);
7146   int call_offset = -1;
7147 
7148   // Be careful not to clobber r1-7 which hold returned fields
7149   // Also do not use callee-saved registers as these may be live in the interpreter
7150   Register tmp1 = r13, tmp2 = r14, klass = r15, r0_preserved = r12;
7151 
7152   // The following code is similar to allocate_instance but has some slight differences,
7153   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
7154   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
7155   Label slow_case;
7156   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
7157   mov(r0_preserved, r0); // save r0 for slow_case since *_allocate may corrupt it when allocation failed
7158 
7159   if (vk != nullptr) {
7160     // Called from C1, where the return type is statically known.
7161     movptr(klass, (intptr_t)vk->get_InlineKlass());
7162     jint lh = vk->layout_helper();
7163     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
7164     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
7165       tlab_allocate(r0, noreg, lh, tmp1, tmp2, slow_case);
7166     } else {
7167       b(slow_case);
7168     }
7169   } else {
7170     // Call from interpreter. R0 contains ((the InlineKlass* of the return type) | 0x01)
7171     andr(klass, r0, -2);
7172     if (UseTLAB) {
7173       ldrw(tmp2, Address(klass, Klass::layout_helper_offset()));
7174       tst(tmp2, Klass::_lh_instance_slow_path_bit);
7175       br(Assembler::NE, slow_case);
7176       tlab_allocate(r0, tmp2, 0, tmp1, tmp2, slow_case);
7177     } else {
7178       b(slow_case);
7179     }
7180   }
7181   if (UseTLAB) {
7182     // 2. Initialize buffered inline instance header
7183     Register buffer_obj = r0;
7184     if (UseCompactObjectHeaders) {
7185       ldr(rscratch1, Address(klass, Klass::prototype_header_offset()));
7186       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7187     } else {
7188       mov(rscratch1, (intptr_t)markWord::inline_type_prototype().value());
7189       str(rscratch1, Address(buffer_obj, oopDesc::mark_offset_in_bytes()));
7190       store_klass_gap(buffer_obj, zr);
7191       if (vk == nullptr) {
7192         // store_klass corrupts klass, so save it for later use (interpreter case only).
7193         mov(tmp1, klass);
7194       }
7195       store_klass(buffer_obj, klass);
7196       klass = tmp1;
7197     }
7198     // 3. Initialize its fields with an inline class specific handler
7199     if (vk != nullptr) {
7200       far_call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
7201     } else {
7202       ldr(tmp1, Address(klass, InlineKlass::adr_members_offset()));
7203       ldr(tmp1, Address(tmp1, InlineKlass::pack_handler_offset()));
7204       blr(tmp1);
7205     }
7206 
7207     membar(Assembler::StoreStore);
7208     b(skip);
7209   } else {
7210     // Must have already branched to slow_case above.
7211     DEBUG_ONLY(should_not_reach_here());
7212   }
7213   bind(slow_case);
7214   // We failed to allocate a new inline type, fall back to a runtime
7215   // call. Some oop field may be live in some registers but we can't
7216   // tell. That runtime call will take care of preserving them
7217   // across a GC if there's one.
7218   mov(r0, r0_preserved);
7219 
7220   if (from_interpreter) {
7221     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
7222   } else {
7223     far_call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
7224     call_offset = offset();
7225   }
7226   membar(Assembler::StoreStore);
7227 
7228   bind(skip);
7229   return call_offset;
7230 }
7231 
7232 // Move a value between registers/stack slots and update the reg_state
7233 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
7234   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
7235   if (reg_state[to->value()] == reg_written) {
7236     return true; // Already written
7237   }
7238 
7239   if (from != to && bt != T_VOID) {
7240     if (reg_state[to->value()] == reg_readonly) {
7241       return false; // Not yet writable
7242     }
7243     if (from->is_reg()) {
7244       if (to->is_reg()) {
7245         if (from->is_Register() && to->is_Register()) {
7246           mov(to->as_Register(), from->as_Register());
7247         } else if (from->is_FloatRegister() && to->is_FloatRegister()) {
7248           fmovd(to->as_FloatRegister(), from->as_FloatRegister());
7249         } else {
7250           ShouldNotReachHere();
7251         }
7252       } else {
7253         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7254         Address to_addr = Address(sp, st_off);
7255         if (from->is_FloatRegister()) {
7256           if (bt == T_DOUBLE) {
7257              strd(from->as_FloatRegister(), to_addr);
7258           } else {
7259              assert(bt == T_FLOAT, "must be float");
7260              strs(from->as_FloatRegister(), to_addr);
7261           }
7262         } else {
7263           str(from->as_Register(), to_addr);
7264         }
7265       }
7266     } else {
7267       Address from_addr = Address(sp, from->reg2stack() * VMRegImpl::stack_slot_size);
7268       if (to->is_reg()) {
7269         if (to->is_FloatRegister()) {
7270           if (bt == T_DOUBLE) {
7271             ldrd(to->as_FloatRegister(), from_addr);
7272           } else {
7273             assert(bt == T_FLOAT, "must be float");
7274             ldrs(to->as_FloatRegister(), from_addr);
7275           }
7276         } else {
7277           ldr(to->as_Register(), from_addr);
7278         }
7279       } else {
7280         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size;
7281         ldr(rscratch1, from_addr);
7282         str(rscratch1, Address(sp, st_off));
7283       }
7284     }
7285   }
7286 
7287   // Update register states
7288   reg_state[from->value()] = reg_writable;
7289   reg_state[to->value()] = reg_written;
7290   return true;
7291 }
7292 
7293 // Calculate the extra stack space required for packing or unpacking inline
7294 // args and adjust the stack pointer
7295 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
7296   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
7297   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
7298   assert(sp_inc > 0, "sanity");
7299 
7300   // Save a copy of the FP and LR here for deoptimization patching and frame walking
7301   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7302 
7303   // Adjust the stack pointer. This will be repaired on return by MacroAssembler::remove_frame
7304   if (sp_inc < (1 << 9)) {
7305     sub(sp, sp, sp_inc);   // Fits in an immediate
7306   } else {
7307     mov(rscratch1, sp_inc);
7308     sub(sp, sp, rscratch1);
7309   }
7310 
7311   return sp_inc + 2 * wordSize;  // Account for the FP/LR space
7312 }
7313 
7314 // Read all fields from an inline type oop and store the values in registers/stack slots
7315 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
7316                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
7317                                           RegState reg_state[]) {
7318   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
7319   assert(from->is_valid(), "source must be valid");
7320   bool progress = false;
7321 #ifdef ASSERT
7322   const int start_offset = offset();
7323 #endif
7324 
7325   Label L_null, L_notNull;
7326   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
7327   Register tmp1 = r10;
7328   Register tmp2 = r11;
7329 
7330 #ifndef ASSERT
7331   RegSet clobbered_gp_regs = MacroAssembler::call_clobbered_gp_registers();
7332   assert(clobbered_gp_regs.contains(tmp1), "tmp1 must be saved explicitly if it's not a clobber");
7333   assert(clobbered_gp_regs.contains(tmp2), "tmp2 must be saved explicitly if it's not a clobber");
7334   assert(clobbered_gp_regs.contains(r14), "r14 must be saved explicitly if it's not a clobber");
7335 #endif
7336 
7337   Register fromReg = noreg;
7338   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
7339   bool done = true;
7340   bool mark_done = true;
7341   VMReg toReg;
7342   BasicType bt;
7343   // Check if argument requires a null check
7344   bool null_check = false;
7345   VMReg nullCheckReg;
7346   while (stream.next(nullCheckReg, bt)) {
7347     if (sig->at(stream.sig_index())._offset == -1) {
7348       null_check = true;
7349       break;
7350     }
7351   }
7352   stream.reset(sig_index, to_index);
7353   while (stream.next(toReg, bt)) {
7354     assert(toReg->is_valid(), "destination must be valid");
7355     int idx = (int)toReg->value();
7356     if (reg_state[idx] == reg_readonly) {
7357       if (idx != from->value()) {
7358         mark_done = false;
7359       }
7360       done = false;
7361       continue;
7362     } else if (reg_state[idx] == reg_written) {
7363       continue;
7364     }
7365     assert(reg_state[idx] == reg_writable, "must be writable");
7366     reg_state[idx] = reg_written;
7367     progress = true;
7368 
7369     if (fromReg == noreg) {
7370       if (from->is_reg()) {
7371         fromReg = from->as_Register();
7372       } else {
7373         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size;
7374         ldr(tmp1, Address(sp, st_off));
7375         fromReg = tmp1;
7376       }
7377       if (null_check) {
7378         // Nullable inline type argument, emit null check
7379         cbz(fromReg, L_null);
7380       }
7381     }
7382     int off = sig->at(stream.sig_index())._offset;
7383     if (off == -1) {
7384       assert(null_check, "Missing null check at");
7385       if (toReg->is_stack()) {
7386         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7387         mov(tmp2, 1);
7388         str(tmp2, Address(sp, st_off));
7389       } else {
7390         mov(toReg->as_Register(), 1);
7391       }
7392       continue;
7393     }
7394     assert(off > 0, "offset in object should be positive");
7395     Address fromAddr = Address(fromReg, off);
7396     if (!toReg->is_FloatRegister()) {
7397       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
7398       if (is_reference_type(bt)) {
7399         load_heap_oop(dst, fromAddr, rscratch1, rscratch2);
7400       } else {
7401         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
7402         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
7403       }
7404       if (toReg->is_stack()) {
7405         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7406         str(dst, Address(sp, st_off));
7407       }
7408     } else if (bt == T_DOUBLE) {
7409       ldrd(toReg->as_FloatRegister(), fromAddr);
7410     } else {
7411       assert(bt == T_FLOAT, "must be float");
7412       ldrs(toReg->as_FloatRegister(), fromAddr);
7413     }
7414   }
7415   if (progress && null_check) {
7416     if (done) {
7417       b(L_notNull);
7418       bind(L_null);
7419       // Set null marker to zero to signal that the argument is null.
7420       // Also set all fields to zero since the runtime requires a canonical
7421       // representation of a flat null.
7422       stream.reset(sig_index, to_index);
7423       while (stream.next(toReg, bt)) {
7424         if (toReg->is_stack()) {
7425           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size;
7426           str(zr, Address(sp, st_off));
7427         } else if (toReg->is_FloatRegister()) {
7428           mov(toReg->as_FloatRegister(), T2S, 0);
7429         } else {
7430           mov(toReg->as_Register(), zr);
7431         }
7432       }
7433       bind(L_notNull);
7434     } else {
7435       bind(L_null);
7436     }
7437   }
7438 
7439   sig_index = stream.sig_index();
7440   to_index = stream.regs_index();
7441 
7442   if (mark_done && reg_state[from->value()] != reg_written) {
7443     // This is okay because no one else will write to that slot
7444     reg_state[from->value()] = reg_writable;
7445   }
7446   from_index--;
7447   assert(progress || (start_offset == offset()), "should not emit code");
7448   return done;
7449 }
7450 
7451 // Pack fields back into an inline type oop
7452 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
7453                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
7454                                         RegState reg_state[], Register val_array) {
7455   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
7456   assert(to->is_valid(), "destination must be valid");
7457 
7458   if (reg_state[to->value()] == reg_written) {
7459     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7460     return true; // Already written
7461   }
7462 
7463   // The GC barrier expanded by store_heap_oop below may call into the
7464   // runtime so use callee-saved registers for any values that need to be
7465   // preserved. The GC barrier assembler should take care of saving the
7466   // Java argument registers.
7467   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
7468   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
7469   Register val_obj_tmp = r21;
7470   Register from_reg_tmp = r22;
7471   Register tmp1 = r14;
7472   Register tmp2 = r13;
7473   Register tmp3 = r12;
7474   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
7475 
7476   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
7477 
7478   if (reg_state[to->value()] == reg_readonly) {
7479     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
7480       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
7481       return false; // Not yet writable
7482     }
7483     val_obj = val_obj_tmp;
7484   }
7485 
7486   int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
7487   load_heap_oop(val_obj, Address(val_array, index), tmp1, tmp2);
7488 
7489   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
7490   VMReg fromReg;
7491   BasicType bt;
7492   Label L_null;
7493   while (stream.next(fromReg, bt)) {
7494     assert(fromReg->is_valid(), "source must be valid");
7495     reg_state[fromReg->value()] = reg_writable;
7496 
7497     int off = sig->at(stream.sig_index())._offset;
7498     if (off == -1) {
7499       // Nullable inline type argument, emit null check
7500       Label L_notNull;
7501       if (fromReg->is_stack()) {
7502         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7503         ldrb(tmp2, Address(sp, ld_off));
7504         cbnz(tmp2, L_notNull);
7505       } else {
7506         cbnz(fromReg->as_Register(), L_notNull);
7507       }
7508       mov(val_obj, 0);
7509       b(L_null);
7510       bind(L_notNull);
7511       continue;
7512     }
7513 
7514     assert(off > 0, "offset in object should be positive");
7515     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
7516 
7517     // Pack the scalarized field into the value object.
7518     Address dst(val_obj, off);
7519     if (!fromReg->is_FloatRegister()) {
7520       Register src;
7521       if (fromReg->is_stack()) {
7522         src = from_reg_tmp;
7523         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size;
7524         load_sized_value(src, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
7525       } else {
7526         src = fromReg->as_Register();
7527       }
7528       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
7529       if (is_reference_type(bt)) {
7530         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
7531         mov(tmp3, val_obj);
7532         Address dst_with_tmp3(tmp3, off);
7533         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
7534       } else {
7535         store_sized_value(dst, src, size_in_bytes);
7536       }
7537     } else if (bt == T_DOUBLE) {
7538       strd(fromReg->as_FloatRegister(), dst);
7539     } else {
7540       assert(bt == T_FLOAT, "must be float");
7541       strs(fromReg->as_FloatRegister(), dst);
7542     }
7543   }
7544   bind(L_null);
7545   sig_index = stream.sig_index();
7546   from_index = stream.regs_index();
7547 
7548   assert(reg_state[to->value()] == reg_writable, "must have already been read");
7549   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
7550   assert(success, "to register must be writeable");
7551   return true;
7552 }
7553 
7554 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
7555   return (reg->is_FloatRegister()) ? v8->as_VMReg() : r14->as_VMReg();
7556 }
7557 
7558 void MacroAssembler::cache_wb(Address line) {
7559   assert(line.getMode() == Address::base_plus_offset, "mode should be base_plus_offset");
7560   assert(line.index() == noreg, "index should be noreg");
7561   assert(line.offset() == 0, "offset should be 0");
7562   // would like to assert this
7563   // assert(line._ext.shift == 0, "shift should be zero");
7564   if (VM_Version::supports_dcpop()) {
7565     // writeback using clear virtual address to point of persistence
7566     dc(Assembler::CVAP, line.base());
7567   } else {
7568     // no need to generate anything as Unsafe.writebackMemory should
7569     // never invoke this stub
7570   }
7571 }
7572 
7573 void MacroAssembler::cache_wbsync(bool is_pre) {
7574   // we only need a barrier post sync
7575   if (!is_pre) {
7576     membar(Assembler::AnyAny);
7577   }
7578 }
7579 
7580 void MacroAssembler::verify_sve_vector_length(Register tmp) {
7581   if (!UseSVE || VM_Version::get_max_supported_sve_vector_length() == FloatRegister::sve_vl_min) {
7582     return;
7583   }
7584   // Make sure that native code does not change SVE vector length.
7585   Label verify_ok;
7586   movw(tmp, zr);
7587   sve_inc(tmp, B);
7588   subsw(zr, tmp, VM_Version::get_initial_sve_vector_length());
7589   br(EQ, verify_ok);
7590   stop("Error: SVE vector length has changed since jvm startup");
7591   bind(verify_ok);
7592 }
7593 
7594 void MacroAssembler::verify_ptrue() {
7595   Label verify_ok;
7596   if (!UseSVE) {
7597     return;
7598   }
7599   sve_cntp(rscratch1, B, ptrue, ptrue); // get true elements count.
7600   sve_dec(rscratch1, B);
7601   cbz(rscratch1, verify_ok);
7602   stop("Error: the preserved predicate register (p7) elements are not all true");
7603   bind(verify_ok);
7604 }
7605 
7606 void MacroAssembler::safepoint_isb() {
7607   isb();
7608 #ifndef PRODUCT
7609   if (VerifyCrossModifyFence) {
7610     // Clear the thread state.
7611     strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7612   }
7613 #endif
7614 }
7615 
7616 #ifndef PRODUCT
7617 void MacroAssembler::verify_cross_modify_fence_not_required() {
7618   if (VerifyCrossModifyFence) {
7619     // Check if thread needs a cross modify fence.
7620     ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
7621     Label fence_not_required;
7622     cbz(rscratch1, fence_not_required);
7623     // If it does then fail.
7624     lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure)));
7625     mov(c_rarg0, rthread);
7626     blr(rscratch1);
7627     bind(fence_not_required);
7628   }
7629 }
7630 #endif
7631 
7632 void MacroAssembler::spin_wait() {
7633   block_comment("spin_wait {");
7634   for (int i = 0; i < VM_Version::spin_wait_desc().inst_count(); ++i) {
7635     switch (VM_Version::spin_wait_desc().inst()) {
7636       case SpinWait::NOP:
7637         nop();
7638         break;
7639       case SpinWait::ISB:
7640         isb();
7641         break;
7642       case SpinWait::YIELD:
7643         yield();
7644         break;
7645       case SpinWait::SB:
7646         assert(VM_Version::supports_sb(), "current CPU does not support SB instruction");
7647         sb();
7648         break;
7649       default:
7650         ShouldNotReachHere();
7651     }
7652   }
7653   block_comment("}");
7654 }
7655 
7656 // Stack frame creation/removal
7657 
7658 void MacroAssembler::enter(bool strip_ret_addr) {
7659   if (strip_ret_addr) {
7660     // Addresses can only be signed once. If there are multiple nested frames being created
7661     // in the same function, then the return address needs stripping first.
7662     strip_return_address();
7663   }
7664   protect_return_address();
7665   stp(rfp, lr, Address(pre(sp, -2 * wordSize)));
7666   mov(rfp, sp);
7667 }
7668 
7669 void MacroAssembler::leave() {
7670   mov(sp, rfp);
7671   ldp(rfp, lr, Address(post(sp, 2 * wordSize)));
7672   authenticate_return_address();
7673 }
7674 
7675 // ROP Protection
7676 // Use the AArch64 PAC feature to add ROP protection for generated code. Use whenever creating/
7677 // destroying stack frames or whenever directly loading/storing the LR to memory.
7678 // If ROP protection is not set then these functions are no-ops.
7679 // For more details on PAC see pauth_aarch64.hpp.
7680 
7681 // Sign the LR. Use during construction of a stack frame, before storing the LR to memory.
7682 // Uses value zero as the modifier.
7683 //
7684 void MacroAssembler::protect_return_address() {
7685   if (VM_Version::use_rop_protection()) {
7686     check_return_address();
7687     paciaz();
7688   }
7689 }
7690 
7691 // Sign the return value in the given register. Use before updating the LR in the existing stack
7692 // frame for the current function.
7693 // Uses value zero as the modifier.
7694 //
7695 void MacroAssembler::protect_return_address(Register return_reg) {
7696   if (VM_Version::use_rop_protection()) {
7697     check_return_address(return_reg);
7698     paciza(return_reg);
7699   }
7700 }
7701 
7702 // Authenticate the LR. Use before function return, after restoring FP and loading LR from memory.
7703 // Uses value zero as the modifier.
7704 //
7705 void MacroAssembler::authenticate_return_address() {
7706   if (VM_Version::use_rop_protection()) {
7707     autiaz();
7708     check_return_address();
7709   }
7710 }
7711 
7712 // Authenticate the return value in the given register. Use before updating the LR in the existing
7713 // stack frame for the current function.
7714 // Uses value zero as the modifier.
7715 //
7716 void MacroAssembler::authenticate_return_address(Register return_reg) {
7717   if (VM_Version::use_rop_protection()) {
7718     autiza(return_reg);
7719     check_return_address(return_reg);
7720   }
7721 }
7722 
7723 // Strip any PAC data from LR without performing any authentication. Use with caution - only if
7724 // there is no guaranteed way of authenticating the LR.
7725 //
7726 void MacroAssembler::strip_return_address() {
7727   if (VM_Version::use_rop_protection()) {
7728     xpaclri();
7729   }
7730 }
7731 
7732 #ifndef PRODUCT
7733 // PAC failures can be difficult to debug. After an authentication failure, a segfault will only
7734 // occur when the pointer is used - ie when the program returns to the invalid LR. At this point
7735 // it is difficult to debug back to the callee function.
7736 // This function simply loads from the address in the given register.
7737 // Use directly after authentication to catch authentication failures.
7738 // Also use before signing to check that the pointer is valid and hasn't already been signed.
7739 //
7740 void MacroAssembler::check_return_address(Register return_reg) {
7741   if (VM_Version::use_rop_protection()) {
7742     ldr(zr, Address(return_reg));
7743   }
7744 }
7745 #endif
7746 
7747 // The java_calling_convention describes stack locations as ideal slots on
7748 // a frame with no abi restrictions. Since we must observe abi restrictions
7749 // (like the placement of the register window) the slots must be biased by
7750 // the following value.
7751 static int reg2offset_in(VMReg r) {
7752   // Account for saved rfp and lr
7753   // This should really be in_preserve_stack_slots
7754   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
7755 }
7756 
7757 static int reg2offset_out(VMReg r) {
7758   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
7759 }
7760 
7761 // On 64bit we will store integer like items to the stack as
7762 // 64bits items (AArch64 ABI) even though java would only store
7763 // 32bits for a parameter. On 32bit it will simply be 32bits
7764 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
7765 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
7766   if (src.first()->is_stack()) {
7767     if (dst.first()->is_stack()) {
7768       // stack to stack
7769       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7770       str(tmp, Address(sp, reg2offset_out(dst.first())));
7771     } else {
7772       // stack to reg
7773       ldrsw(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7774     }
7775   } else if (dst.first()->is_stack()) {
7776     // reg to stack
7777     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7778   } else {
7779     if (dst.first() != src.first()) {
7780       sxtw(dst.first()->as_Register(), src.first()->as_Register());
7781     }
7782   }
7783 }
7784 
7785 // An oop arg. Must pass a handle not the oop itself
7786 void MacroAssembler::object_move(
7787                         OopMap* map,
7788                         int oop_handle_offset,
7789                         int framesize_in_slots,
7790                         VMRegPair src,
7791                         VMRegPair dst,
7792                         bool is_receiver,
7793                         int* receiver_offset) {
7794 
7795   // must pass a handle. First figure out the location we use as a handle
7796 
7797   Register rHandle = dst.first()->is_stack() ? rscratch2 : dst.first()->as_Register();
7798 
7799   // See if oop is null if it is we need no handle
7800 
7801   if (src.first()->is_stack()) {
7802 
7803     // Oop is already on the stack as an argument
7804     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
7805     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
7806     if (is_receiver) {
7807       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
7808     }
7809 
7810     ldr(rscratch1, Address(rfp, reg2offset_in(src.first())));
7811     lea(rHandle, Address(rfp, reg2offset_in(src.first())));
7812     // conditionally move a null
7813     cmp(rscratch1, zr);
7814     csel(rHandle, zr, rHandle, Assembler::EQ);
7815   } else {
7816 
7817     // Oop is in an a register we must store it to the space we reserve
7818     // on the stack for oop_handles and pass a handle if oop is non-null
7819 
7820     const Register rOop = src.first()->as_Register();
7821     int oop_slot;
7822     if (rOop == j_rarg0)
7823       oop_slot = 0;
7824     else if (rOop == j_rarg1)
7825       oop_slot = 1;
7826     else if (rOop == j_rarg2)
7827       oop_slot = 2;
7828     else if (rOop == j_rarg3)
7829       oop_slot = 3;
7830     else if (rOop == j_rarg4)
7831       oop_slot = 4;
7832     else if (rOop == j_rarg5)
7833       oop_slot = 5;
7834     else if (rOop == j_rarg6)
7835       oop_slot = 6;
7836     else {
7837       assert(rOop == j_rarg7, "wrong register");
7838       oop_slot = 7;
7839     }
7840 
7841     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
7842     int offset = oop_slot*VMRegImpl::stack_slot_size;
7843 
7844     map->set_oop(VMRegImpl::stack2reg(oop_slot));
7845     // Store oop in handle area, may be null
7846     str(rOop, Address(sp, offset));
7847     if (is_receiver) {
7848       *receiver_offset = offset;
7849     }
7850 
7851     cmp(rOop, zr);
7852     lea(rHandle, Address(sp, offset));
7853     // conditionally move a null
7854     csel(rHandle, zr, rHandle, Assembler::EQ);
7855   }
7856 
7857   // If arg is on the stack then place it otherwise it is already in correct reg.
7858   if (dst.first()->is_stack()) {
7859     str(rHandle, Address(sp, reg2offset_out(dst.first())));
7860   }
7861 }
7862 
7863 // A float arg may have to do float reg int reg conversion
7864 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
7865  if (src.first()->is_stack()) {
7866     if (dst.first()->is_stack()) {
7867       ldrw(tmp, Address(rfp, reg2offset_in(src.first())));
7868       strw(tmp, Address(sp, reg2offset_out(dst.first())));
7869     } else {
7870       ldrs(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7871     }
7872   } else if (src.first() != dst.first()) {
7873     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7874       fmovs(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7875     else
7876       strs(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7877   }
7878 }
7879 
7880 // A long move
7881 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
7882   if (src.first()->is_stack()) {
7883     if (dst.first()->is_stack()) {
7884       // stack to stack
7885       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7886       str(tmp, Address(sp, reg2offset_out(dst.first())));
7887     } else {
7888       // stack to reg
7889       ldr(dst.first()->as_Register(), Address(rfp, reg2offset_in(src.first())));
7890     }
7891   } else if (dst.first()->is_stack()) {
7892     // reg to stack
7893     // Do we really have to sign extend???
7894     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
7895     str(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
7896   } else {
7897     if (dst.first() != src.first()) {
7898       mov(dst.first()->as_Register(), src.first()->as_Register());
7899     }
7900   }
7901 }
7902 
7903 
7904 // A double move
7905 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
7906  if (src.first()->is_stack()) {
7907     if (dst.first()->is_stack()) {
7908       ldr(tmp, Address(rfp, reg2offset_in(src.first())));
7909       str(tmp, Address(sp, reg2offset_out(dst.first())));
7910     } else {
7911       ldrd(dst.first()->as_FloatRegister(), Address(rfp, reg2offset_in(src.first())));
7912     }
7913   } else if (src.first() != dst.first()) {
7914     if (src.is_single_phys_reg() && dst.is_single_phys_reg())
7915       fmovd(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
7916     else
7917       strd(src.first()->as_FloatRegister(), Address(sp, reg2offset_out(dst.first())));
7918   }
7919 }
7920 
7921 // Implements fast-locking.
7922 //
7923 //  - obj: the object to be locked
7924 //  - t1, t2, t3: temporary registers, will be destroyed
7925 //  - slow: branched to if locking fails, absolute offset may larger than 32KB (imm14 encoding).
7926 void MacroAssembler::fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow) {
7927   assert_different_registers(basic_lock, obj, t1, t2, t3, rscratch1);
7928 
7929   Label push;
7930   const Register top = t1;
7931   const Register mark = t2;
7932   const Register t = t3;
7933 
7934   // Preload the markWord. It is important that this is the first
7935   // instruction emitted as it is part of C1's null check semantics.
7936   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
7937 
7938   if (UseObjectMonitorTable) {
7939     // Clear cache in case fast locking succeeds or we need to take the slow-path.
7940     str(zr, Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))));
7941   }
7942 
7943   if (DiagnoseSyncOnValueBasedClasses != 0) {
7944     load_klass(t1, obj);
7945     ldrb(t1, Address(t1, Klass::misc_flags_offset()));
7946     tst(t1, KlassFlags::_misc_is_value_based_class);
7947     br(Assembler::NE, slow);
7948   }
7949 
7950   // Check if the lock-stack is full.
7951   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7952   cmpw(top, (unsigned)LockStack::end_offset());
7953   br(Assembler::GE, slow);
7954 
7955   // Check for recursion.
7956   subw(t, top, oopSize);
7957   ldr(t, Address(rthread, t));
7958   cmp(obj, t);
7959   br(Assembler::EQ, push);
7960 
7961   // Check header for monitor (0b10).
7962   tst(mark, markWord::monitor_value);
7963   br(Assembler::NE, slow);
7964 
7965   // Try to lock. Transition lock bits 0b01 => 0b00
7966   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
7967   orr(mark, mark, markWord::unlocked_value);
7968   // Mask inline_type bit such that we go to the slow path if object is an inline type
7969   andr(mark, mark, ~((int) markWord::inline_type_bit_in_place));
7970 
7971   eor(t, mark, markWord::unlocked_value);
7972   cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::xword,
7973           /*acquire*/ true, /*release*/ false, /*weak*/ false, noreg);
7974   br(Assembler::NE, slow);
7975 
7976   bind(push);
7977   // After successful lock, push object on lock-stack.
7978   str(obj, Address(rthread, top));
7979   addw(top, top, oopSize);
7980   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
7981 }
7982 
7983 // Implements fast-unlocking.
7984 //
7985 // - obj: the object to be unlocked
7986 // - t1, t2, t3: temporary registers
7987 // - slow: branched to if unlocking fails, absolute offset may larger than 32KB (imm14 encoding).
7988 void MacroAssembler::fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow) {
7989   // cmpxchg clobbers rscratch1.
7990   assert_different_registers(obj, t1, t2, t3, rscratch1);
7991 
7992 #ifdef ASSERT
7993   {
7994     // Check for lock-stack underflow.
7995     Label stack_ok;
7996     ldrw(t1, Address(rthread, JavaThread::lock_stack_top_offset()));
7997     cmpw(t1, (unsigned)LockStack::start_offset());
7998     br(Assembler::GE, stack_ok);
7999     STOP("Lock-stack underflow");
8000     bind(stack_ok);
8001   }
8002 #endif
8003 
8004   Label unlocked, push_and_slow;
8005   const Register top = t1;
8006   const Register mark = t2;
8007   const Register t = t3;
8008 
8009   // Check if obj is top of lock-stack.
8010   ldrw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8011   subw(top, top, oopSize);
8012   ldr(t, Address(rthread, top));
8013   cmp(obj, t);
8014   br(Assembler::NE, slow);
8015 
8016   // Pop lock-stack.
8017   DEBUG_ONLY(str(zr, Address(rthread, top));)
8018   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8019 
8020   // Check if recursive.
8021   subw(t, top, oopSize);
8022   ldr(t, Address(rthread, t));
8023   cmp(obj, t);
8024   br(Assembler::EQ, unlocked);
8025 
8026   // Not recursive. Check header for monitor (0b10).
8027   ldr(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
8028   tbnz(mark, log2i_exact(markWord::monitor_value), push_and_slow);
8029 
8030 #ifdef ASSERT
8031   // Check header not unlocked (0b01).
8032   Label not_unlocked;
8033   tbz(mark, log2i_exact(markWord::unlocked_value), not_unlocked);
8034   stop("fast_unlock already unlocked");
8035   bind(not_unlocked);
8036 #endif
8037 
8038   // Try to unlock. Transition lock bits 0b00 => 0b01
8039   assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
8040   orr(t, mark, markWord::unlocked_value);
8041   cmpxchg(obj, mark, t, Assembler::xword,
8042           /*acquire*/ false, /*release*/ true, /*weak*/ false, noreg);
8043   br(Assembler::EQ, unlocked);
8044 
8045   bind(push_and_slow);
8046   // Restore lock-stack and handle the unlock in runtime.
8047   DEBUG_ONLY(str(obj, Address(rthread, top));)
8048   addw(top, top, oopSize);
8049   strw(top, Address(rthread, JavaThread::lock_stack_top_offset()));
8050   b(slow);
8051 
8052   bind(unlocked);
8053 }
--- EOF ---