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 #ifndef CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  27 #define CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  28 
  29 #include "asm/assembler.inline.hpp"
  30 #include "code/aotCodeCache.hpp"
  31 #include "code/vmreg.hpp"
  32 #include "metaprogramming/enableIf.hpp"
  33 #include "oops/compressedOops.hpp"
  34 #include "oops/compressedKlass.hpp"
  35 #include "runtime/vm_version.hpp"
  36 #include "utilities/globalDefinitions.hpp"
  37 #include "utilities/macros.hpp"
  38 #include "utilities/powerOfTwo.hpp"
  39 #include "runtime/signature.hpp"
  40 
  41 
  42 class ciInlineKlass;
  43 
  44 class OopMap;
  45 
  46 // MacroAssembler extends Assembler by frequently used macros.
  47 //
  48 // Instructions for which a 'better' code sequence exists depending
  49 // on arguments should also go in here.
  50 
  51 class MacroAssembler: public Assembler {
  52   friend class LIR_Assembler;
  53 
  54  public:
  55   using Assembler::mov;
  56   using Assembler::movi;
  57 
  58  protected:
  59 
  60   // Support for VM calls
  61   //
  62   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
  63   // may customize this version by overriding it for its purposes (e.g., to save/restore
  64   // additional registers when doing a VM call).
  65   virtual void call_VM_leaf_base(
  66     address entry_point,               // the entry point
  67     int     number_of_arguments,        // the number of arguments to pop after the call
  68     Label *retaddr = nullptr
  69   );
  70 
  71   virtual void call_VM_leaf_base(
  72     address entry_point,               // the entry point
  73     int     number_of_arguments,        // the number of arguments to pop after the call
  74     Label &retaddr) {
  75     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
  76   }
  77 
  78   // This is the base routine called by the different versions of call_VM. The interpreter
  79   // may customize this version by overriding it for its purposes (e.g., to save/restore
  80   // additional registers when doing a VM call).
  81   //
  82   // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
  83   // returns the register which contains the thread upon return. If a thread register has been
  84   // specified, the return value will correspond to that register. If no last_java_sp is specified
  85   // (noreg) than rsp will be used instead.
  86   virtual void call_VM_base(           // returns the register containing the thread upon return
  87     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
  88     Register java_thread,              // the thread if computed before     ; use noreg otherwise
  89     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
  90     Label*   return_pc,                // to set up last_Java_frame; use nullptr otherwise
  91     address  entry_point,              // the entry point
  92     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
  93     bool     check_exceptions          // whether to check for pending exceptions after return
  94   );
  95 
  96   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
  97 
  98   enum KlassDecodeMode {
  99     KlassDecodeNone,
 100     KlassDecodeZero,
 101     KlassDecodeXor,
 102     KlassDecodeMovk
 103   };
 104 
 105   // Calculate decoding mode based on given parameters, used for checking then ultimately setting.
 106   static KlassDecodeMode klass_decode_mode(address base, int shift, const size_t range);
 107 
 108  private:
 109   static KlassDecodeMode _klass_decode_mode;
 110 
 111   // Returns above setting with asserts
 112   static KlassDecodeMode klass_decode_mode();
 113 
 114  public:
 115   // Checks the decode mode and returns false if not compatible with preferred decoding mode.
 116   static bool check_klass_decode_mode(address base, int shift, const size_t range);
 117 
 118   // Sets the decode mode and returns false if cannot be set.
 119   static bool set_klass_decode_mode(address base, int shift, const size_t range);
 120 
 121  public:
 122   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 123 
 124  // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 125  // The implementation is only non-empty for the InterpreterMacroAssembler,
 126  // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 127  virtual void check_and_handle_popframe(Register java_thread);
 128  virtual void check_and_handle_earlyret(Register java_thread);
 129 
 130   void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp = rscratch1);
 131   void rt_call(address dest, Register tmp = rscratch1);
 132 
 133   // Load Effective Address
 134   void lea(Register r, const Address &a) {
 135     InstructionMark im(this);
 136     a.lea(this, r);
 137   }
 138 
 139   // Whether materializing the given address for a LDR/STR requires an
 140   // additional lea instruction.
 141   static bool legitimize_address_requires_lea(const Address &a, int size) {
 142     return a.getMode() == Address::base_plus_offset &&
 143            !Address::offset_ok_for_immed(a.offset(), exact_log2(size));
 144   }
 145 
 146   /* Sometimes we get misaligned loads and stores, usually from Unsafe
 147      accesses, and these can exceed the offset range. */
 148   Address legitimize_address(const Address &a, int size, Register scratch) {
 149     if (legitimize_address_requires_lea(a, size)) {
 150       block_comment("legitimize_address {");
 151       lea(scratch, a);
 152       block_comment("} legitimize_address");
 153       return Address(scratch);
 154     }
 155     return a;
 156   }
 157 
 158   void addmw(Address a, Register incr, Register scratch) {
 159     ldrw(scratch, a);
 160     addw(scratch, scratch, incr);
 161     strw(scratch, a);
 162   }
 163 
 164   // Add constant to memory word
 165   void addmw(Address a, int imm, Register scratch) {
 166     ldrw(scratch, a);
 167     if (imm > 0)
 168       addw(scratch, scratch, (unsigned)imm);
 169     else
 170       subw(scratch, scratch, (unsigned)-imm);
 171     strw(scratch, a);
 172   }
 173 
 174   void bind(Label& L) {
 175     Assembler::bind(L);
 176     code()->clear_last_merge_candidate();
 177     code()->set_last_label(pc());
 178   }
 179 
 180   void membar(Membar_mask_bits order_constraint);
 181 
 182   using Assembler::ldr;
 183   using Assembler::str;
 184   using Assembler::ldrw;
 185   using Assembler::strw;
 186 
 187   void ldr(Register Rx, const Address &adr);
 188   void ldrw(Register Rw, const Address &adr);
 189   void str(Register Rx, const Address &adr);
 190   void strw(Register Rx, const Address &adr);
 191 
 192   // Frame creation and destruction shared between JITs.
 193   DEBUG_ONLY(void build_frame(int framesize);)
 194   void build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills));
 195   void remove_frame(int framesize);
 196 
 197   virtual void _call_Unimplemented(address call_site) {
 198     mov(rscratch2, call_site);
 199   }
 200 
 201 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 202 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 203 // https://reviews.llvm.org/D3311
 204 
 205 #ifdef _WIN64
 206 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 207 #else
 208 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 209 #endif
 210 
 211   // aliases defined in AARCH64 spec
 212 
 213   template<class T>
 214   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }
 215 
 216   inline void cmp(Register Rd, unsigned char imm8)  { subs(zr, Rd, imm8); }
 217   inline void cmp(Register Rd, unsigned imm) = delete;
 218 
 219   template<class T>
 220   inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); }
 221 
 222   inline void cmn(Register Rd, unsigned char imm8)  { adds(zr, Rd, imm8); }
 223   inline void cmn(Register Rd, unsigned imm) = delete;
 224 
 225   void cset(Register Rd, Assembler::Condition cond) {
 226     csinc(Rd, zr, zr, ~cond);
 227   }
 228   void csetw(Register Rd, Assembler::Condition cond) {
 229     csincw(Rd, zr, zr, ~cond);
 230   }
 231 
 232   void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
 233     csneg(Rd, Rn, Rn, ~cond);
 234   }
 235   void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
 236     csnegw(Rd, Rn, Rn, ~cond);
 237   }
 238 
 239   inline void movw(Register Rd, Register Rn) {
 240     if (Rd == sp || Rn == sp) {
 241       Assembler::addw(Rd, Rn, 0U);
 242     } else {
 243       orrw(Rd, zr, Rn);
 244     }
 245   }
 246   inline void mov(Register Rd, Register Rn) {
 247     assert(Rd != r31_sp && Rn != r31_sp, "should be");
 248     if (Rd == Rn) {
 249     } else if (Rd == sp || Rn == sp) {
 250       Assembler::add(Rd, Rn, 0U);
 251     } else {
 252       orr(Rd, zr, Rn);
 253     }
 254   }
 255 
 256   inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
 257   inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
 258 
 259   inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
 260   inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
 261 
 262   inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
 263   inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
 264 
 265   inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 266     bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 267   }
 268   inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 269     bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 270   }
 271 
 272   inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 273     bfmw(Rd, Rn, lsb, (lsb + width - 1));
 274   }
 275   inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 276     bfm(Rd, Rn, lsb , (lsb + width - 1));
 277   }
 278 
 279   inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 280     sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 281   }
 282   inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 283     sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 284   }
 285 
 286   inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 287     sbfmw(Rd, Rn, lsb, (lsb + width - 1));
 288   }
 289   inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 290     sbfm(Rd, Rn, lsb , (lsb + width - 1));
 291   }
 292 
 293   inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 294     ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 295   }
 296   inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 297     ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 298   }
 299 
 300   inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 301     ubfmw(Rd, Rn, lsb, (lsb + width - 1));
 302   }
 303   inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 304     ubfm(Rd, Rn, lsb , (lsb + width - 1));
 305   }
 306 
 307   inline void asrw(Register Rd, Register Rn, unsigned imm) {
 308     sbfmw(Rd, Rn, imm, 31);
 309   }
 310 
 311   inline void asr(Register Rd, Register Rn, unsigned imm) {
 312     sbfm(Rd, Rn, imm, 63);
 313   }
 314 
 315   inline void lslw(Register Rd, Register Rn, unsigned imm) {
 316     ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
 317   }
 318 
 319   inline void lsl(Register Rd, Register Rn, unsigned imm) {
 320     ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
 321   }
 322 
 323   inline void lsrw(Register Rd, Register Rn, unsigned imm) {
 324     ubfmw(Rd, Rn, imm, 31);
 325   }
 326 
 327   inline void lsr(Register Rd, Register Rn, unsigned imm) {
 328     ubfm(Rd, Rn, imm, 63);
 329   }
 330 
 331   inline void rorw(Register Rd, Register Rn, unsigned imm) {
 332     extrw(Rd, Rn, Rn, imm);
 333   }
 334 
 335   inline void ror(Register Rd, Register Rn, unsigned imm) {
 336     extr(Rd, Rn, Rn, imm);
 337   }
 338 
 339   inline void rolw(Register Rd, Register Rn, unsigned imm) {
 340     extrw(Rd, Rn, Rn, (32 - imm));
 341   }
 342 
 343   inline void rol(Register Rd, Register Rn, unsigned imm) {
 344     extr(Rd, Rn, Rn, (64 - imm));
 345   }
 346 
 347   using Assembler::rax1;
 348   using Assembler::eor3;
 349 
 350   inline void rax1(Register Rd, Register Rn, Register Rm) {
 351     eor(Rd, Rn, Rm, ROR, 63); // Rd = Rn ^ rol(Rm, 1)
 352   }
 353 
 354   inline void eor3(Register Rd, Register Rn, Register Rm, Register Rk) {
 355     assert(Rd != Rn, "Use tmp register");
 356     eor(Rd, Rm, Rk);
 357     eor(Rd, Rd, Rn);
 358   }
 359 
 360   inline void sxtbw(Register Rd, Register Rn) {
 361     sbfmw(Rd, Rn, 0, 7);
 362   }
 363   inline void sxthw(Register Rd, Register Rn) {
 364     sbfmw(Rd, Rn, 0, 15);
 365   }
 366   inline void sxtb(Register Rd, Register Rn) {
 367     sbfm(Rd, Rn, 0, 7);
 368   }
 369   inline void sxth(Register Rd, Register Rn) {
 370     sbfm(Rd, Rn, 0, 15);
 371   }
 372   inline void sxtw(Register Rd, Register Rn) {
 373     sbfm(Rd, Rn, 0, 31);
 374   }
 375 
 376   inline void uxtbw(Register Rd, Register Rn) {
 377     ubfmw(Rd, Rn, 0, 7);
 378   }
 379   inline void uxthw(Register Rd, Register Rn) {
 380     ubfmw(Rd, Rn, 0, 15);
 381   }
 382   inline void uxtb(Register Rd, Register Rn) {
 383     ubfm(Rd, Rn, 0, 7);
 384   }
 385   inline void uxth(Register Rd, Register Rn) {
 386     ubfm(Rd, Rn, 0, 15);
 387   }
 388   inline void uxtw(Register Rd, Register Rn) {
 389     ubfm(Rd, Rn, 0, 31);
 390   }
 391 
 392   inline void cmnw(Register Rn, Register Rm) {
 393     addsw(zr, Rn, Rm);
 394   }
 395   inline void cmn(Register Rn, Register Rm) {
 396     adds(zr, Rn, Rm);
 397   }
 398 
 399   inline void cmpw(Register Rn, Register Rm) {
 400     subsw(zr, Rn, Rm);
 401   }
 402   inline void cmp(Register Rn, Register Rm) {
 403     subs(zr, Rn, Rm);
 404   }
 405 
 406   inline void negw(Register Rd, Register Rn) {
 407     subw(Rd, zr, Rn);
 408   }
 409 
 410   inline void neg(Register Rd, Register Rn) {
 411     sub(Rd, zr, Rn);
 412   }
 413 
 414   inline void negsw(Register Rd, Register Rn) {
 415     subsw(Rd, zr, Rn);
 416   }
 417 
 418   inline void negs(Register Rd, Register Rn) {
 419     subs(Rd, zr, Rn);
 420   }
 421 
 422   inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 423     addsw(zr, Rn, Rm, kind, shift);
 424   }
 425   inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 426     adds(zr, Rn, Rm, kind, shift);
 427   }
 428 
 429   inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 430     subsw(zr, Rn, Rm, kind, shift);
 431   }
 432   inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 433     subs(zr, Rn, Rm, kind, shift);
 434   }
 435 
 436   inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 437     subw(Rd, zr, Rn, kind, shift);
 438   }
 439 
 440   inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 441     sub(Rd, zr, Rn, kind, shift);
 442   }
 443 
 444   inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 445     subsw(Rd, zr, Rn, kind, shift);
 446   }
 447 
 448   inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 449     subs(Rd, zr, Rn, kind, shift);
 450   }
 451 
 452   inline void mnegw(Register Rd, Register Rn, Register Rm) {
 453     msubw(Rd, Rn, Rm, zr);
 454   }
 455   inline void mneg(Register Rd, Register Rn, Register Rm) {
 456     msub(Rd, Rn, Rm, zr);
 457   }
 458 
 459   inline void mulw(Register Rd, Register Rn, Register Rm) {
 460     maddw(Rd, Rn, Rm, zr);
 461   }
 462   inline void mul(Register Rd, Register Rn, Register Rm) {
 463     madd(Rd, Rn, Rm, zr);
 464   }
 465 
 466   inline void smnegl(Register Rd, Register Rn, Register Rm) {
 467     smsubl(Rd, Rn, Rm, zr);
 468   }
 469   inline void smull(Register Rd, Register Rn, Register Rm) {
 470     smaddl(Rd, Rn, Rm, zr);
 471   }
 472 
 473   inline void umnegl(Register Rd, Register Rn, Register Rm) {
 474     umsubl(Rd, Rn, Rm, zr);
 475   }
 476   inline void umull(Register Rd, Register Rn, Register Rm) {
 477     umaddl(Rd, Rn, Rm, zr);
 478   }
 479 
 480 #define WRAP(INSN)                                                            \
 481   void INSN(Register Rd, Register Rn, Register Rm, Register Ra) {             \
 482     if (VM_Version::supports_a53mac() && Ra != zr)                            \
 483       nop();                                                                  \
 484     Assembler::INSN(Rd, Rn, Rm, Ra);                                          \
 485   }
 486 
 487   WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
 488   WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
 489 #undef WRAP
 490 
 491 
 492   // macro assembly operations needed for aarch64
 493 
 494 public:
 495 
 496   enum FpPushPopMode {
 497     PushPopFull,
 498     PushPopSVE,
 499     PushPopNeon,
 500     PushPopFp
 501   };
 502 
 503   // first two private routines for loading 32 bit or 64 bit constants
 504 private:
 505 
 506   void mov_immediate64(Register dst, uint64_t imm64);
 507   void mov_immediate32(Register dst, uint32_t imm32);
 508 
 509   void mov(Register dst, Address a);
 510 
 511 public:
 512 
 513   int push(RegSet regset, Register stack);
 514   int pop(RegSet regset, Register stack);
 515 
 516   int push_fp(FloatRegSet regset, Register stack, FpPushPopMode mode = PushPopFull);
 517   int pop_fp(FloatRegSet regset, Register stack, FpPushPopMode mode = PushPopFull);
 518 
 519   static RegSet call_clobbered_gp_registers();
 520 
 521   int push_p(PRegSet regset, Register stack);
 522   int pop_p(PRegSet regset, Register stack);
 523 
 524   // Push and pop everything that might be clobbered by a native
 525   // runtime call except rscratch1 and rscratch2.  (They are always
 526   // scratch, so we don't have to protect them.)  Only save the lower
 527   // 64 bits of each vector register. Additional registers can be excluded
 528   // in a passed RegSet.
 529   void push_call_clobbered_registers_except(RegSet exclude);
 530   void pop_call_clobbered_registers_except(RegSet exclude);
 531 
 532   void push_call_clobbered_registers() {
 533     push_call_clobbered_registers_except(RegSet());
 534   }
 535   void pop_call_clobbered_registers() {
 536     pop_call_clobbered_registers_except(RegSet());
 537   }
 538 
 539 
 540   // now mov instructions for loading absolute addresses and 32 or
 541   // 64 bit integers
 542 
 543   inline void mov(Register dst, address addr)             { mov_immediate64(dst, (uint64_t)addr); }
 544 
 545   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 546   inline void mov(Register dst, T o)                      { mov_immediate64(dst, (uint64_t)o); }
 547 
 548   inline void movw(Register dst, uint32_t imm32)          { mov_immediate32(dst, imm32); }
 549 
 550   void mov(Register dst, RegisterOrConstant src) {
 551     if (src.is_register())
 552       mov(dst, src.as_register());
 553     else
 554       mov(dst, src.as_constant());
 555   }
 556 
 557   void movptr(Register r, uintptr_t imm64);
 558 
 559   void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64);
 560 
 561   void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
 562     orr(Vd, T, Vn, Vn);
 563   }
 564 
 565   void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) {
 566     fcvtsh(tmp, src);
 567     smov(dst, tmp, H, 0);
 568   }
 569 
 570   void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) {
 571     mov(tmp, H, 0, src);
 572     fcvths(dst, tmp);
 573   }
 574 
 575   // Generalized Test Bit And Branch, including a "far" variety which
 576   // spans more than 32KiB.
 577   void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) {
 578     assert(cond == EQ || cond == NE, "must be");
 579 
 580     if (isfar)
 581       cond = ~cond;
 582 
 583     void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
 584     if (cond == Assembler::EQ)
 585       branch = &Assembler::tbz;
 586     else
 587       branch = &Assembler::tbnz;
 588 
 589     if (isfar) {
 590       Label L;
 591       (this->*branch)(Rt, bitpos, L);
 592       b(dest);
 593       bind(L);
 594     } else {
 595       (this->*branch)(Rt, bitpos, dest);
 596     }
 597   }
 598 
 599   // macro instructions for accessing and updating floating point
 600   // status register
 601   //
 602   // FPSR : op1 == 011
 603   //        CRn == 0100
 604   //        CRm == 0100
 605   //        op2 == 001
 606 
 607   inline void get_fpsr(Register reg)
 608   {
 609     mrs(0b11, 0b0100, 0b0100, 0b001, reg);
 610   }
 611 
 612   inline void set_fpsr(Register reg)
 613   {
 614     msr(0b011, 0b0100, 0b0100, 0b001, reg);
 615   }
 616 
 617   inline void clear_fpsr()
 618   {
 619     msr(0b011, 0b0100, 0b0100, 0b001, zr);
 620   }
 621 
 622   // FPCR : op1 == 011
 623   //        CRn == 0100
 624   //        CRm == 0100
 625   //        op2 == 000
 626 
 627   inline void get_fpcr(Register reg) {
 628     mrs(0b11, 0b0100, 0b0100, 0b000, reg);
 629   }
 630 
 631   inline void set_fpcr(Register reg) {
 632     msr(0b011, 0b0100, 0b0100, 0b000, reg);
 633   }
 634 
 635   // DCZID_EL0: op1 == 011
 636   //            CRn == 0000
 637   //            CRm == 0000
 638   //            op2 == 111
 639   inline void get_dczid_el0(Register reg)
 640   {
 641     mrs(0b011, 0b0000, 0b0000, 0b111, reg);
 642   }
 643 
 644   // CTR_EL0:   op1 == 011
 645   //            CRn == 0000
 646   //            CRm == 0000
 647   //            op2 == 001
 648   inline void get_ctr_el0(Register reg)
 649   {
 650     mrs(0b011, 0b0000, 0b0000, 0b001, reg);
 651   }
 652 
 653   inline void get_nzcv(Register reg) {
 654     mrs(0b011, 0b0100, 0b0010, 0b000, reg);
 655   }
 656 
 657   inline void set_nzcv(Register reg) {
 658     msr(0b011, 0b0100, 0b0010, 0b000, reg);
 659   }
 660 
 661   // CNTVCTSS_EL0:   op1 == 011
 662   //                 CRn == 1110
 663   //                 CRm == 0000
 664   //                 op2 == 110
 665   inline void get_cntvctss_el0(Register reg) {
 666     mrs(0b011, 0b1110, 0b0000, 0b110, reg);
 667   }
 668 
 669   // idiv variant which deals with MINLONG as dividend and -1 as divisor
 670   int corrected_idivl(Register result, Register ra, Register rb,
 671                       bool want_remainder, Register tmp = rscratch1);
 672   int corrected_idivq(Register result, Register ra, Register rb,
 673                       bool want_remainder, Register tmp = rscratch1);
 674 
 675   // Support for null-checks
 676   //
 677   // Generates code that causes a null OS exception if the content of reg is null.
 678   // If the accessed location is M[reg + offset] and the offset is known, provide the
 679   // offset. No explicit code generation is needed if the offset is within a certain
 680   // range (0 <= offset <= page_size).
 681 
 682   virtual void null_check(Register reg, int offset = -1);
 683   static bool needs_explicit_null_check(intptr_t offset);
 684   static bool uses_implicit_null_check(void* address);
 685 
 686   // markWord tests, kills markWord reg
 687   void test_markword_is_inline_type(Register markword, Label& is_inline_type);
 688 
 689   // inlineKlass queries, kills temp_reg
 690   void test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null = true);
 691 
 692   void test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free);
 693   void test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free);
 694   void test_field_is_flat(Register flags, Register temp_reg, Label& is_flat);
 695 
 696   // Check oops for special arrays, i.e. flat arrays and/or null-free arrays
 697   void test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label);
 698   void test_flat_array_oop(Register klass, Register temp_reg, Label& is_flat_array);
 699   void test_non_flat_array_oop(Register oop, Register temp_reg, Label&is_non_flat_array);
 700   void test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array);
 701   void test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array);
 702 
 703   // Check array klass layout helper for flat or null-free arrays...
 704   void test_flat_array_layout(Register lh, Label& is_flat_array);
 705 
 706   static address target_addr_for_insn(address insn_addr);
 707 
 708   // Required platform-specific helpers for Label::patch_instructions.
 709   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 710   static int pd_patch_instruction_size(address branch, address target);
 711   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 712     pd_patch_instruction_size(branch, target);
 713   }
 714   static address pd_call_destination(address branch) {
 715     return target_addr_for_insn(branch);
 716   }
 717 #ifndef PRODUCT
 718   static void pd_print_patched_instruction(address branch);
 719 #endif
 720 
 721   static int patch_oop(address insn_addr, address o);
 722 
 723   // Return whether code is emitted to a scratch blob.
 724   virtual bool in_scratch_emit_size() {
 725     return false;
 726   }
 727   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
 728   static int max_trampoline_stub_size();
 729   void emit_static_call_stub();
 730   static int static_call_stub_size();
 731 
 732   // The following 4 methods return the offset of the appropriate move instruction
 733 
 734   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 735   int load_unsigned_byte(Register dst, Address src);
 736   int load_unsigned_short(Register dst, Address src);
 737 
 738   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 739   int load_signed_byte(Register dst, Address src);
 740   int load_signed_short(Register dst, Address src);
 741 
 742   int load_signed_byte32(Register dst, Address src);
 743   int load_signed_short32(Register dst, Address src);
 744 
 745   // Support for sign-extension (hi:lo = extend_sign(lo))
 746   void extend_sign(Register hi, Register lo);
 747 
 748   // Clean up a subword typed value to the representation in compliance with JVMS ยง2.3
 749   void narrow_subword_type(Register reg, BasicType bt);
 750 
 751   // Load and store values by size and signed-ness
 752   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 753   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 754 
 755   // Support for inc/dec with optimal instruction selection depending on value
 756 
 757   // x86_64 aliases an unqualified register/address increment and
 758   // decrement to call incrementq and decrementq but also supports
 759   // explicitly sized calls to incrementq/decrementq or
 760   // incrementl/decrementl
 761 
 762   // for aarch64 the proper convention would be to use
 763   // increment/decrement for 64 bit operations and
 764   // incrementw/decrementw for 32 bit operations. so when porting
 765   // x86_64 code we can leave calls to increment/decrement as is,
 766   // replace incrementq/decrementq with increment/decrement and
 767   // replace incrementl/decrementl with incrementw/decrementw.
 768 
 769   // n.b. increment/decrement calls with an Address destination will
 770   // need to use a scratch register to load the value to be
 771   // incremented. increment/decrement calls which add or subtract a
 772   // constant value greater than 2^12 will need to use a 2nd scratch
 773   // register to hold the constant. so, a register increment/decrement
 774   // may trash rscratch2 and an address increment/decrement trash
 775   // rscratch and rscratch2
 776 
 777   void decrementw(Address dst, int value = 1);
 778   void decrementw(Register reg, int value = 1);
 779 
 780   void decrement(Register reg, int value = 1);
 781   void decrement(Address dst, int value = 1);
 782 
 783   void incrementw(Address dst, int value = 1);
 784   void incrementw(Register reg, int value = 1);
 785 
 786   void increment(Register reg, int value = 1);
 787   void increment(Address dst, int value = 1);
 788 
 789 
 790   // Alignment
 791   void align(int modulus);
 792   void align(int modulus, int target);
 793 
 794   // nop
 795   void post_call_nop();
 796 
 797   // Stack frame creation/removal
 798   void enter(bool strip_ret_addr = false);
 799   void leave();
 800 
 801   // ROP Protection
 802   void protect_return_address();
 803   void protect_return_address(Register return_reg);
 804   void authenticate_return_address();
 805   void authenticate_return_address(Register return_reg);
 806   void strip_return_address();
 807   void check_return_address(Register return_reg=lr) PRODUCT_RETURN;
 808 
 809   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
 810   // The pointer will be loaded into the thread register.
 811   void get_thread(Register thread);
 812 
 813   // support for argument shuffling
 814   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 815   void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 816   void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 817   void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 818   void object_move(
 819                    OopMap* map,
 820                    int oop_handle_offset,
 821                    int framesize_in_slots,
 822                    VMRegPair src,
 823                    VMRegPair dst,
 824                    bool is_receiver,
 825                    int* receiver_offset);
 826 
 827 
 828   // Support for VM calls
 829   //
 830   // It is imperative that all calls into the VM are handled via the call_VM macros.
 831   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 832   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 833 
 834 
 835   void call_VM(Register oop_result,
 836                address entry_point,
 837                bool check_exceptions = true);
 838   void call_VM(Register oop_result,
 839                address entry_point,
 840                Register arg_1,
 841                bool check_exceptions = true);
 842   void call_VM(Register oop_result,
 843                address entry_point,
 844                Register arg_1, Register arg_2,
 845                bool check_exceptions = true);
 846   void call_VM(Register oop_result,
 847                address entry_point,
 848                Register arg_1, Register arg_2, Register arg_3,
 849                bool check_exceptions = true);
 850 
 851   // Overloadings with last_Java_sp
 852   void call_VM(Register oop_result,
 853                Register last_java_sp,
 854                address entry_point,
 855                int number_of_arguments = 0,
 856                bool check_exceptions = true);
 857   void call_VM(Register oop_result,
 858                Register last_java_sp,
 859                address entry_point,
 860                Register arg_1, bool
 861                check_exceptions = true);
 862   void call_VM(Register oop_result,
 863                Register last_java_sp,
 864                address entry_point,
 865                Register arg_1, Register arg_2,
 866                bool check_exceptions = true);
 867   void call_VM(Register oop_result,
 868                Register last_java_sp,
 869                address entry_point,
 870                Register arg_1, Register arg_2, Register arg_3,
 871                bool check_exceptions = true);
 872 
 873   void get_vm_result_oop(Register oop_result, Register thread);
 874   void get_vm_result_metadata(Register metadata_result, Register thread);
 875 
 876   // These always tightly bind to MacroAssembler::call_VM_base
 877   // bypassing the virtual implementation
 878   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 879   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
 880   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 881   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
 882   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true);
 883 
 884   void call_VM_leaf(address entry_point,
 885                     int number_of_arguments = 0);
 886   void call_VM_leaf(address entry_point,
 887                     Register arg_1);
 888   void call_VM_leaf(address entry_point,
 889                     Register arg_1, Register arg_2);
 890   void call_VM_leaf(address entry_point,
 891                     Register arg_1, Register arg_2, Register arg_3);
 892 
 893   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 894   // bypassing the virtual implementation
 895   void super_call_VM_leaf(address entry_point);
 896   void super_call_VM_leaf(address entry_point, Register arg_1);
 897   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 898   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 899   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 900 
 901   // last Java Frame (fills frame anchor)
 902   void set_last_Java_frame(Register last_java_sp,
 903                            Register last_java_fp,
 904                            address last_java_pc,
 905                            Register scratch);
 906 
 907   void set_last_Java_frame(Register last_java_sp,
 908                            Register last_java_fp,
 909                            Label &last_java_pc,
 910                            Register scratch);
 911 
 912   void set_last_Java_frame(Register last_java_sp,
 913                            Register last_java_fp,
 914                            Register last_java_pc,
 915                            Register scratch);
 916 
 917   void reset_last_Java_frame(Register thread);
 918 
 919   // thread in the default location (rthread)
 920   void reset_last_Java_frame(bool clear_fp);
 921 
 922   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 923   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 924 
 925   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 926   void c2bool(Register x);
 927 
 928   void load_method_holder_cld(Register rresult, Register rmethod);
 929   void load_method_holder(Register holder, Register method);
 930 
 931   // oop manipulations
 932   void load_metadata(Register dst, Register src);
 933 
 934   void load_narrow_klass_compact(Register dst, Register src);
 935   void load_klass(Register dst, Register src);
 936   void store_klass(Register dst, Register src);
 937   void cmp_klass(Register obj, Register klass, Register tmp);
 938   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 939 
 940   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 941   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 942   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 943 
 944   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 945                       Register tmp1, Register tmp2);
 946 
 947   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 948                        Register tmp1, Register tmp2, Register tmp3);
 949 
 950   void flat_field_copy(DecoratorSet decorators, Register src, Register dst, Register inline_layout_info);
 951 
 952   // inline type data payload offsets...
 953   void payload_offset(Register inline_klass, Register offset);
 954   void payload_address(Register oop, Register data, Register inline_klass);
 955 
 956   void load_heap_oop(Register dst, Address src, Register tmp1,
 957                      Register tmp2, DecoratorSet decorators = 0);
 958 
 959   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 960                               Register tmp2, DecoratorSet decorators = 0);
 961   void store_heap_oop(Address dst, Register val, Register tmp1,
 962                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 963 
 964   // currently unimplemented
 965   // Used for storing null. All other oop constants should be
 966   // stored using routines that take a jobject.
 967   void store_heap_oop_null(Address dst);
 968 
 969   void load_prototype_header(Register dst, Register src);
 970 
 971   void store_klass_gap(Register dst, Register src);
 972 
 973   // This dummy is to prevent a call to store_heap_oop from
 974   // converting a zero (like null) into a Register by giving
 975   // the compiler two choices it can't resolve
 976 
 977   void store_heap_oop(Address dst, void* dummy);
 978 
 979   void encode_heap_oop(Register d, Register s);
 980   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
 981   void decode_heap_oop(Register d, Register s);
 982   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 983   void encode_heap_oop_not_null(Register r);
 984   void decode_heap_oop_not_null(Register r);
 985   void encode_heap_oop_not_null(Register dst, Register src);
 986   void decode_heap_oop_not_null(Register dst, Register src);
 987 
 988   void set_narrow_oop(Register dst, jobject obj);
 989 
 990   void decode_klass_not_null_for_aot(Register dst, Register src);
 991   void encode_klass_not_null_for_aot(Register dst, Register src);
 992   void encode_klass_not_null(Register r);
 993   void decode_klass_not_null(Register r);
 994   void encode_klass_not_null(Register dst, Register src);
 995   void decode_klass_not_null(Register dst, Register src);
 996 
 997   void set_narrow_klass(Register dst, Klass* k);
 998 
 999   // if heap base register is used - reinit it with the correct value
1000   void reinit_heapbase();
1001 
1002   DEBUG_ONLY(void verify_heapbase(const char* msg);)
1003 
1004   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
1005                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1006   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
1007                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1008 
1009   void push_cont_fastpath(Register java_thread = rthread);
1010   void pop_cont_fastpath(Register java_thread = rthread);
1011 
1012   // Round up to a power of two
1013   void round_to(Register reg, int modulus);
1014 
1015   // java.lang.Math::round intrinsics
1016   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
1017   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
1018 
1019   // allocation
1020 
1021   void tlab_allocate(
1022     Register obj,                      // result: pointer to object after successful allocation
1023     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1024     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1025     Register t1,                       // temp register
1026     Register t2,                       // temp register
1027     Label&   slow_case                 // continuation point if fast allocation fails
1028   );
1029   void verify_tlab();
1030 
1031   void inline_layout_info(Register holder_klass, Register index, Register layout_info);
1032 
1033   // interface method calling
1034   void lookup_interface_method(Register recv_klass,
1035                                Register intf_klass,
1036                                RegisterOrConstant itable_index,
1037                                Register method_result,
1038                                Register scan_temp,
1039                                Label& no_such_interface,
1040                    bool return_method = true);
1041 
1042   void lookup_interface_method_stub(Register recv_klass,
1043                                     Register holder_klass,
1044                                     Register resolved_klass,
1045                                     Register method_result,
1046                                     Register temp_reg,
1047                                     Register temp_reg2,
1048                                     int itable_index,
1049                                     Label& L_no_such_interface);
1050 
1051   // virtual method calling
1052   // n.b. x86 allows RegisterOrConstant for vtable_index
1053   void lookup_virtual_method(Register recv_klass,
1054                              RegisterOrConstant vtable_index,
1055                              Register method_result);
1056 
1057   // Test sub_klass against super_klass, with fast and slow paths.
1058 
1059   // The fast path produces a tri-state answer: yes / no / maybe-slow.
1060   // One of the three labels can be null, meaning take the fall-through.
1061   // If super_check_offset is -1, the value is loaded up from super_klass.
1062   // No registers are killed, except temp_reg.
1063   void check_klass_subtype_fast_path(Register sub_klass,
1064                                      Register super_klass,
1065                                      Register temp_reg,
1066                                      Label* L_success,
1067                                      Label* L_failure,
1068                                      Label* L_slow_path,
1069                                      Register super_check_offset = noreg);
1070 
1071   // The rest of the type check; must be wired to a corresponding fast path.
1072   // It does not repeat the fast path logic, so don't use it standalone.
1073   // The temp_reg and temp2_reg can be noreg, if no temps are available.
1074   // Updates the sub's secondary super cache as necessary.
1075   // If set_cond_codes, condition codes will be Z on success, NZ on failure.
1076   void check_klass_subtype_slow_path(Register sub_klass,
1077                                      Register super_klass,
1078                                      Register temp_reg,
1079                                      Register temp2_reg,
1080                                      Label* L_success,
1081                                      Label* L_failure,
1082                                      bool set_cond_codes = false);
1083 
1084   void check_klass_subtype_slow_path_linear(Register sub_klass,
1085                                             Register super_klass,
1086                                             Register temp_reg,
1087                                             Register temp2_reg,
1088                                             Label* L_success,
1089                                             Label* L_failure,
1090                                             bool set_cond_codes = false);
1091 
1092   void check_klass_subtype_slow_path_table(Register sub_klass,
1093                                            Register super_klass,
1094                                            Register temp_reg,
1095                                            Register temp2_reg,
1096                                            Register temp3_reg,
1097                                            Register result_reg,
1098                                            FloatRegister vtemp_reg,
1099                                            Label* L_success,
1100                                            Label* L_failure,
1101                                            bool set_cond_codes = false);
1102 
1103   // If r is valid, return r.
1104   // If r is invalid, remove a register r2 from available_regs, add r2
1105   // to regs_to_push, then return r2.
1106   Register allocate_if_noreg(const Register r,
1107                              RegSetIterator<Register> &available_regs,
1108                              RegSet &regs_to_push);
1109 
1110   // Secondary subtype checking
1111   void lookup_secondary_supers_table_var(Register sub_klass,
1112                                          Register r_super_klass,
1113                                          Register temp1,
1114                                          Register temp2,
1115                                          Register temp3,
1116                                          FloatRegister vtemp,
1117                                          Register result,
1118                                          Label *L_success);
1119 
1120 
1121   // As above, but with a constant super_klass.
1122   // The result is in Register result, not the condition codes.
1123   bool lookup_secondary_supers_table_const(Register r_sub_klass,
1124                                            Register r_super_klass,
1125                                            Register temp1,
1126                                            Register temp2,
1127                                            Register temp3,
1128                                            FloatRegister vtemp,
1129                                            Register result,
1130                                            u1 super_klass_slot,
1131                                            bool stub_is_near = false);
1132 
1133   void verify_secondary_supers_table(Register r_sub_klass,
1134                                      Register r_super_klass,
1135                                      Register temp1,
1136                                      Register temp2,
1137                                      Register result);
1138 
1139   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
1140                                                Register r_array_base,
1141                                                Register r_array_index,
1142                                                Register r_bitmap,
1143                                                Register temp1,
1144                                                Register result,
1145                                                bool is_stub = true);
1146 
1147   // Simplified, combined version, good for typical uses.
1148   // Falls through on failure.
1149   void check_klass_subtype(Register sub_klass,
1150                            Register super_klass,
1151                            Register temp_reg,
1152                            Label& L_success);
1153 
1154   void clinit_barrier(Register klass,
1155                       Register thread,
1156                       Label* L_fast_path = nullptr,
1157                       Label* L_slow_path = nullptr);
1158 
1159   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
1160 
1161   void profile_receiver_type(Register recv, Register mdp, int mdp_offset);
1162 
1163   void verify_sve_vector_length(Register tmp = rscratch1);
1164   void reinitialize_ptrue() {
1165     if (UseSVE > 0) {
1166       sve_ptrue(ptrue, B);
1167     }
1168   }
1169   void verify_ptrue();
1170 
1171   // Debugging
1172 
1173   // only if +VerifyOops
1174   void _verify_oop(Register reg, const char* s, const char* file, int line);
1175   void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
1176 
1177   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
1178     if (VerifyOops) {
1179       _verify_oop(reg, s, file, line);
1180     }
1181   }
1182   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
1183     if (VerifyOops) {
1184       _verify_oop_addr(reg, s, file, line);
1185     }
1186   }
1187 
1188 // TODO: verify method and klass metadata (compare against vptr?)
1189   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1190   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1191 
1192 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
1193 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
1194 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
1195 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1196 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1197 
1198   // Restore cpu control state after JNI call
1199   void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2);
1200 
1201   // prints msg, dumps registers and stops execution
1202   void stop(const char* msg);
1203 
1204   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1205 
1206   void untested()                                { stop("untested"); }
1207 
1208   void unimplemented(const char* what = "");
1209 
1210   void should_not_reach_here()                   { stop("should not reach here"); }
1211 
1212   void _assert_asm(Condition cc, const char* msg);
1213 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg)
1214 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg))
1215 
1216   // Stack overflow checking
1217   void bang_stack_with_offset(int offset) {
1218     // stack grows down, caller passes positive offset
1219     assert(offset > 0, "must bang with negative offset");
1220     sub(rscratch2, sp, offset);
1221     str(zr, Address(rscratch2));
1222   }
1223 
1224   // Writes to stack successive pages until offset reached to check for
1225   // stack overflow + shadow pages.  Also, clobbers tmp
1226   void bang_stack_size(Register size, Register tmp);
1227 
1228   // Check for reserved stack access in method being exited (for JIT)
1229   void reserved_stack_check();
1230 
1231   // Arithmetics
1232 
1233   // Clobber: rscratch1, rscratch2
1234   void addptr(const Address &dst, int32_t src);
1235 
1236   // Clobber: rscratch1
1237   void cmpptr(Register src1, Address src2);
1238 
1239   void cmpoop(Register obj1, Register obj2);
1240 
1241   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1242   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1243   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1244   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1245 
1246   void atomic_xchg(Register prev, Register newv, Register addr);
1247   void atomic_xchgw(Register prev, Register newv, Register addr);
1248   void atomic_xchgl(Register prev, Register newv, Register addr);
1249   void atomic_xchglw(Register prev, Register newv, Register addr);
1250   void atomic_xchgal(Register prev, Register newv, Register addr);
1251   void atomic_xchgalw(Register prev, Register newv, Register addr);
1252 
1253   void orptr(Address adr, RegisterOrConstant src) {
1254     ldr(rscratch1, adr);
1255     if (src.is_register())
1256       orr(rscratch1, rscratch1, src.as_register());
1257     else
1258       orr(rscratch1, rscratch1, src.as_constant());
1259     str(rscratch1, adr);
1260   }
1261 
1262   // A generic CAS; success or failure is in the EQ flag.
1263   // Clobbers rscratch1
1264   void cmpxchg(Register addr, Register expected, Register new_val,
1265                enum operand_size size,
1266                bool acquire, bool release, bool weak,
1267                Register result);
1268 
1269 #ifdef ASSERT
1270   // Template short-hand support to clean-up after a failed call to trampoline
1271   // call generation (see trampoline_call() below),  when a set of Labels must
1272   // be reset (before returning).
1273   template<typename Label, typename... More>
1274   void reset_labels(Label &lbl, More&... more) {
1275     lbl.reset(); reset_labels(more...);
1276   }
1277   template<typename Label>
1278   void reset_labels(Label &lbl) {
1279     lbl.reset();
1280   }
1281 #endif
1282 
1283 private:
1284   void compare_eq(Register rn, Register rm, enum operand_size size);
1285 
1286 public:
1287   // AArch64 OpenJDK uses four different types of calls:
1288   //   - direct call: bl pc_relative_offset
1289   //     This is the shortest and the fastest, but the offset has the range:
1290   //     +/-128MB for the release build, +/-2MB for the debug build.
1291   //
1292   //   - far call: adrp reg, pc_relative_offset; add; bl reg
1293   //     This is longer than a direct call. The offset has
1294   //     the range +/-4GB. As the code cache size is limited to 4GB,
1295   //     far calls can reach anywhere in the code cache. If a jump is
1296   //     needed rather than a call, a far jump 'b reg' can be used instead.
1297   //     All instructions are embedded at a call site.
1298   //
1299   //   - trampoline call:
1300   //     This is only available in C1/C2-generated code (nmethod). It is a combination
1301   //     of a direct call, which is used if the destination of a call is in range,
1302   //     and a register-indirect call. It has the advantages of reaching anywhere in
1303   //     the AArch64 address space and being patchable at runtime when the generated
1304   //     code is being executed by other threads.
1305   //
1306   //     [Main code section]
1307   //       bl trampoline
1308   //     [Stub code section]
1309   //     trampoline:
1310   //       ldr reg, pc + 8
1311   //       br reg
1312   //       <64-bit destination address>
1313   //
1314   //     If the destination is in range when the generated code is moved to the code
1315   //     cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline
1316   //     is not used.
1317   //     The optimization does not remove the trampoline from the stub section.
1318   //     This is necessary because the trampoline may well be redirected later when
1319   //     code is patched, and the new destination may not be reachable by a simple BR
1320   //     instruction.
1321   //
1322   //   - indirect call: move reg, address; blr reg
1323   //     This too can reach anywhere in the address space, but it cannot be
1324   //     patched while code is running, so it must only be modified at a safepoint.
1325   //     This form of call is most suitable for targets at fixed addresses, which
1326   //     will never be patched.
1327   //
1328   // The patching we do conforms to the "Concurrent modification and
1329   // execution of instructions" section of the Arm Architectural
1330   // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC,
1331   // or SVC instructions to be modified while another thread is
1332   // executing them.
1333   //
1334   // To patch a trampoline call when the BL can't reach, we first modify
1335   // the 64-bit destination address in the trampoline, then modify the
1336   // BL to point to the trampoline, then flush the instruction cache to
1337   // broadcast the change to all executing threads. See
1338   // NativeCall::set_destination_mt_safe for the details.
1339   //
1340   // There is a benign race in that the other thread might observe the
1341   // modified BL before it observes the modified 64-bit destination
1342   // address. That does not matter because the destination method has been
1343   // invalidated, so there will be a trap at its start.
1344   // For this to work, the destination address in the trampoline is
1345   // always updated, even if we're not using the trampoline.
1346 
1347   // Emit a direct call if the entry address will always be in range,
1348   // otherwise a trampoline call.
1349   // Supported entry.rspec():
1350   // - relocInfo::runtime_call_type
1351   // - relocInfo::opt_virtual_call_type
1352   // - relocInfo::static_call_type
1353   // - relocInfo::virtual_call_type
1354   //
1355   // Return: the call PC or null if CodeCache is full.
1356   // Clobbers: rscratch1
1357   address trampoline_call(Address entry);
1358 
1359   static bool far_branches() {
1360     return ReservedCodeCacheSize > branch_range;
1361   }
1362 
1363   // Check if branches to the non nmethod section require a far jump
1364   static bool codestub_branch_needs_far_jump() {
1365     if (AOTCodeCache::is_on_for_dump()) {
1366       // To calculate far_codestub_branch_size correctly.
1367       return true;
1368     }
1369     return CodeCache::max_distance_to_non_nmethod() > branch_range;
1370   }
1371 
1372   // Emit a direct call/jump if the entry address will always be in range,
1373   // otherwise a far call/jump.
1374   // The address must be inside the code cache.
1375   // Supported entry.rspec():
1376   // - relocInfo::external_word_type
1377   // - relocInfo::runtime_call_type
1378   // - relocInfo::none
1379   // In the case of a far call/jump, the entry address is put in the tmp register.
1380   // The tmp register is invalidated.
1381   //
1382   // Far_jump returns the amount of the emitted code.
1383   void far_call(Address entry, Register tmp = rscratch1);
1384   int far_jump(Address entry, Register tmp = rscratch1);
1385 
1386   static int far_codestub_branch_size() {
1387     if (codestub_branch_needs_far_jump()) {
1388       return 3 * 4;  // adrp, add, br
1389     } else {
1390       return 4;
1391     }
1392   }
1393 
1394   // Emit the CompiledIC call idiom
1395   address ic_call(address entry, jint method_index = 0);
1396   static int ic_check_size();
1397   int ic_check(int end_alignment);
1398 
1399 public:
1400 
1401   // Data
1402 
1403   void mov_metadata(Register dst, Metadata* obj);
1404   Address allocate_metadata_address(Metadata* obj);
1405   Address constant_oop_address(jobject obj);
1406 
1407   void movoop(Register dst, jobject obj);
1408 
1409   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1410   void kernel_crc32(Register crc, Register buf, Register len,
1411         Register table0, Register table1, Register table2, Register table3,
1412         Register tmp, Register tmp2, Register tmp3);
1413   // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic.
1414   void kernel_crc32c(Register crc, Register buf, Register len,
1415         Register table0, Register table1, Register table2, Register table3,
1416         Register tmp, Register tmp2, Register tmp3);
1417 
1418   // Stack push and pop individual 64 bit registers
1419   void push(Register src);
1420   void pop(Register dst);
1421 
1422   void repne_scan(Register addr, Register value, Register count,
1423                   Register scratch);
1424   void repne_scanw(Register addr, Register value, Register count,
1425                    Register scratch);
1426 
1427   typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1428   typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1429 
1430   // If a constant does not fit in an immediate field, generate some
1431   // number of MOV instructions and then perform the operation
1432   void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
1433                              add_sub_imm_insn insn1,
1434                              add_sub_reg_insn insn2, bool is32);
1435   // Separate vsn which sets the flags
1436   void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
1437                                add_sub_imm_insn insn1,
1438                                add_sub_reg_insn insn2, bool is32);
1439 
1440 #define WRAP(INSN, is32)                                                \
1441   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1442     wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1443   }                                                                     \
1444                                                                         \
1445   void INSN(Register Rd, Register Rn, Register Rm,                      \
1446              enum shift_kind kind, unsigned shift = 0) {                \
1447     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1448   }                                                                     \
1449                                                                         \
1450   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1451     Assembler::INSN(Rd, Rn, Rm);                                        \
1452   }                                                                     \
1453                                                                         \
1454   void INSN(Register Rd, Register Rn, Register Rm,                      \
1455            ext::operation option, int amount = 0) {                     \
1456     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1457   }
1458 
1459   WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true)
1460 
1461 #undef WRAP
1462 #define WRAP(INSN, is32)                                                \
1463   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1464     wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1465   }                                                                     \
1466                                                                         \
1467   void INSN(Register Rd, Register Rn, Register Rm,                      \
1468              enum shift_kind kind, unsigned shift = 0) {                \
1469     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1470   }                                                                     \
1471                                                                         \
1472   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1473     Assembler::INSN(Rd, Rn, Rm);                                        \
1474   }                                                                     \
1475                                                                         \
1476   void INSN(Register Rd, Register Rn, Register Rm,                      \
1477            ext::operation option, int amount = 0) {                     \
1478     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1479   }
1480 
1481   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1482 
1483   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1484   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1485   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1486   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1487 
1488   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1489 
1490   void verified_entry(Compile* C, int sp_inc);
1491 
1492   // Inline type specific methods
1493   #include "asm/macroAssembler_common.hpp"
1494 
1495   void save_stack_increment(int sp_inc, int frame_size);
1496 
1497   void tableswitch(Register index, jint lowbound, jint highbound,
1498                    Label &jumptable, Label &jumptable_end, int stride = 1) {
1499     adr(rscratch1, jumptable);
1500     subsw(rscratch2, index, lowbound);
1501     subsw(zr, rscratch2, highbound - lowbound);
1502     br(Assembler::HS, jumptable_end);
1503     add(rscratch1, rscratch1, rscratch2,
1504         ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1505     br(rscratch1);
1506   }
1507 
1508   // Form an address from base + offset in Rd.  Rd may or may not
1509   // actually be used: you must use the Address that is returned.  It
1510   // is up to you to ensure that the shift provided matches the size
1511   // of your data.
1512   Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1513 
1514   // Return true iff an address is within the 48-bit AArch64 address
1515   // space.
1516   bool is_valid_AArch64_address(address a) {
1517     return ((uint64_t)a >> 48) == 0;
1518   }
1519 
1520   // Load the base of the cardtable byte map into reg.
1521   void load_byte_map_base(Register reg);
1522 
1523   // Load a constant address in the AOT Runtime Constants area
1524   void load_aotrc_address(Register reg, address a);
1525 
1526   // Prolog generator routines to support switch between x86 code and
1527   // generated ARM code
1528 
1529   // routine to generate an x86 prolog for a stub function which
1530   // bootstraps into the generated ARM code which directly follows the
1531   // stub
1532   //
1533 
1534   public:
1535 
1536   address read_polling_page(Register r, relocInfo::relocType rtype);
1537   void get_polling_page(Register dest, relocInfo::relocType rtype);
1538 
1539   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1540   void update_byte_crc32(Register crc, Register val, Register table);
1541   void update_word_crc32(Register crc, Register v, Register tmp,
1542         Register table0, Register table1, Register table2, Register table3,
1543         bool upper = false);
1544 
1545   address count_positives(Register ary1, Register len, Register result);
1546 
1547   address arrays_equals(Register a1, Register a2, Register result, Register cnt1,
1548                         Register tmp1, Register tmp2, Register tmp3, int elem_size);
1549 
1550 // Ensure that the inline code and the stub use the same registers.
1551 #define ARRAYS_HASHCODE_REGISTERS \
1552   do {                      \
1553     assert(result == r0  && \
1554            ary    == r1  && \
1555            cnt    == r2  && \
1556            vdata0 == v3  && \
1557            vdata1 == v2  && \
1558            vdata2 == v1  && \
1559            vdata3 == v0  && \
1560            vmul0  == v4  && \
1561            vmul1  == v5  && \
1562            vmul2  == v6  && \
1563            vmul3  == v7  && \
1564            vpow   == v12 && \
1565            vpowm  == v13, "registers must match aarch64.ad"); \
1566   } while (0)
1567 
1568   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1569 
1570   void fill_words(Register base, Register cnt, Register value);
1571   void fill_words(Register base, uint64_t cnt, Register value);
1572 
1573   address zero_words(Register base, uint64_t cnt);
1574   address zero_words(Register ptr, Register cnt);
1575   void zero_dcache_blocks(Register base, Register cnt);
1576 
1577   static const int zero_words_block_size;
1578 
1579   address byte_array_inflate(Register src, Register dst, Register len,
1580                              FloatRegister vtmp1, FloatRegister vtmp2,
1581                              FloatRegister vtmp3, Register tmp4);
1582 
1583   void char_array_compress(Register src, Register dst, Register len,
1584                            Register res,
1585                            FloatRegister vtmp0, FloatRegister vtmp1,
1586                            FloatRegister vtmp2, FloatRegister vtmp3,
1587                            FloatRegister vtmp4, FloatRegister vtmp5);
1588 
1589   void encode_iso_array(Register src, Register dst,
1590                         Register len, Register res, bool ascii,
1591                         FloatRegister vtmp0, FloatRegister vtmp1,
1592                         FloatRegister vtmp2, FloatRegister vtmp3,
1593                         FloatRegister vtmp4, FloatRegister vtmp5);
1594 
1595   void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi,
1596       address pio2, address dsin_coef, address dcos_coef);
1597  private:
1598   // begin trigonometric functions support block
1599   void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2);
1600   void generate__kernel_rem_pio2(address two_over_pi, address pio2);
1601   void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef);
1602   void generate_kernel_cos(FloatRegister x, address dcos_coef);
1603   // end trigonometric functions support block
1604   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1605                        Register src1, Register src2);
1606   void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1607     add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1608   }
1609   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1610                              Register y, Register y_idx, Register z,
1611                              Register carry, Register product,
1612                              Register idx, Register kdx);
1613   void multiply_128_x_128_loop(Register y, Register z,
1614                                Register carry, Register carry2,
1615                                Register idx, Register jdx,
1616                                Register yz_idx1, Register yz_idx2,
1617                                Register tmp, Register tmp3, Register tmp4,
1618                                Register tmp7, Register product_hi);
1619   void kernel_crc32_using_crypto_pmull(Register crc, Register buf,
1620         Register len, Register tmp0, Register tmp1, Register tmp2,
1621         Register tmp3);
1622   void kernel_crc32_using_crc32(Register crc, Register buf,
1623         Register len, Register tmp0, Register tmp1, Register tmp2,
1624         Register tmp3);
1625   void kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
1626         Register len, Register tmp0, Register tmp1, Register tmp2,
1627         Register tmp3);
1628   void kernel_crc32c_using_crc32c(Register crc, Register buf,
1629         Register len, Register tmp0, Register tmp1, Register tmp2,
1630         Register tmp3);
1631   void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
1632         Register len, Register tmp0, Register tmp1, Register tmp2,
1633         size_t table_offset);
1634 
1635   void ghash_modmul (FloatRegister result,
1636                      FloatRegister result_lo, FloatRegister result_hi, FloatRegister b,
1637                      FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p,
1638                      FloatRegister t1, FloatRegister t2, FloatRegister t3);
1639   void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
1640 public:
1641   void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1642                        Register tmp0, Register tmp1, Register tmp2, Register tmp3,
1643                        Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1644   void mul_add(Register out, Register in, Register offs, Register len, Register k);
1645   void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,
1646                       FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1647                       FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1648   void ghash_multiply_wide(int index,
1649                            FloatRegister result_lo, FloatRegister result_hi,
1650                            FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1651                            FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1652   void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi,
1653                     FloatRegister p, FloatRegister z, FloatRegister t1);
1654   void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi,
1655                     FloatRegister p, FloatRegister z, FloatRegister t1);
1656   void ghash_processBlocks_wide(Label& p, Register state, Register subkeyH,
1657                                 Register data, Register blocks, int unrolls);
1658 
1659 
1660   void aesenc_loadkeys(Register key, Register keylen);
1661   void aesecb_encrypt(Register from, Register to, Register keylen,
1662                       FloatRegister data = v0, int unrolls = 1);
1663   void aesecb_decrypt(Register from, Register to, Register key, Register keylen);
1664   void aes_round(FloatRegister input, FloatRegister subkey);
1665 
1666   // ChaCha20 functions support block
1667   void cc20_qr_add4(FloatRegister (&addFirst)[4],
1668           FloatRegister (&addSecond)[4]);
1669   void cc20_qr_xor4(FloatRegister (&firstElem)[4],
1670           FloatRegister (&secondElem)[4], FloatRegister (&result)[4]);
1671   void cc20_qr_lrot4(FloatRegister (&sourceReg)[4],
1672           FloatRegister (&destReg)[4], int bits, FloatRegister table);
1673   void cc20_set_qr_registers(FloatRegister (&vectorSet)[4],
1674           const FloatRegister (&stateVectors)[16], int idx1, int idx2,
1675           int idx3, int idx4);
1676 
1677   // Rotate using ORR (for identity) or USHR + SLI.
1678   void neon_vector_rotate(FloatRegister dst, SIMD_Arrangement T,
1679                           FloatRegister src, int shift_amount);
1680 
1681   // Place an ISB after code may have been modified due to a safepoint.
1682   void safepoint_isb();
1683 
1684 private:
1685   // Return the effective address r + (r1 << ext) + offset.
1686   // Uses rscratch2.
1687   Address offsetted_address(Register r, Register r1, Address::extend ext,
1688                             int offset, int size);
1689 
1690 private:
1691   // Returns an address on the stack which is reachable with a ldr/str of size
1692   // Uses rscratch2 if the address is not directly reachable
1693   Address spill_address(int size, int offset, Register tmp=rscratch2);
1694   Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2);
1695 
1696   bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const;
1697 
1698   // Check whether two loads/stores can be merged into ldp/stp.
1699   bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const;
1700 
1701   // Merge current load/store with previous load/store into ldp/stp.
1702   void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1703 
1704   // Try to merge two loads/stores into ldp/stp. If success, returns true else false.
1705   bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1706 
1707 public:
1708   void spill(Register Rx, bool is64, int offset) {
1709     if (is64) {
1710       str(Rx, spill_address(8, offset));
1711     } else {
1712       strw(Rx, spill_address(4, offset));
1713     }
1714   }
1715   void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1716     str(Vx, T, spill_address(1 << (int)T, offset));
1717   }
1718 
1719   void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1720     sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1721   }
1722   void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1723     sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1724   }
1725 
1726   void unspill(Register Rx, bool is64, int offset) {
1727     if (is64) {
1728       ldr(Rx, spill_address(8, offset));
1729     } else {
1730       ldrw(Rx, spill_address(4, offset));
1731     }
1732   }
1733   void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1734     ldr(Vx, T, spill_address(1 << (int)T, offset));
1735   }
1736 
1737   void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1738     sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1739   }
1740   void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1741     sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1742   }
1743 
1744   void spill_copy128(int src_offset, int dst_offset,
1745                      Register tmp1=rscratch1, Register tmp2=rscratch2) {
1746     if (src_offset < 512 && (src_offset & 7) == 0 &&
1747         dst_offset < 512 && (dst_offset & 7) == 0) {
1748       ldp(tmp1, tmp2, Address(sp, src_offset));
1749       stp(tmp1, tmp2, Address(sp, dst_offset));
1750     } else {
1751       unspill(tmp1, true, src_offset);
1752       spill(tmp1, true, dst_offset);
1753       unspill(tmp1, true, src_offset+8);
1754       spill(tmp1, true, dst_offset+8);
1755     }
1756   }
1757   void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset,
1758                                             int sve_vec_reg_size_in_bytes) {
1759     assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size");
1760     for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) {
1761       spill_copy128(src_offset, dst_offset);
1762       src_offset += 16;
1763       dst_offset += 16;
1764     }
1765   }
1766   void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset,
1767                                                int sve_predicate_reg_size_in_bytes) {
1768     sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset));
1769     sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset));
1770     reinitialize_ptrue();
1771   }
1772   void cache_wb(Address line);
1773   void cache_wbsync(bool is_pre);
1774 
1775   // Code for java.lang.Thread::onSpinWait() intrinsic.
1776   void spin_wait();
1777   void spin_wait_wfet(int delay_ns);
1778 
1779   void fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow);
1780   void fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1781 
1782 private:
1783   // Check the current thread doesn't need a cross modify fence.
1784   void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
1785   void try_to_replace_prev_vector_copy_with_movprfx(FloatRegister dst);
1786 
1787 public:
1788   void maybe_movprfx(FloatRegister dst, FloatRegister src) {
1789     if (dst != src) {
1790       sve_movprfx(dst, src);
1791     }
1792   }
1793 
1794 // Wrappers for SVE explicit destructive instructions, overriding the
1795 // same-signature Assembler entry points to enable movprfx fusion optimization.
1796 //
1797 // Implicit destructive instructions (e.g. predicated unary ops like sve_abs/
1798 // sve_neg/sve_not, whose ISA encoding allows Zd != Zn but whose use as a Java
1799 // Vector API masked operation requires pass-through of the first source) are
1800 // not covered here. For those, the .ad file is responsible for emitting
1801 // movprfx explicitly via maybe_movprfx() before the destructive op.
1802 #define SVE_DESTRUCTIVE_BINARY_INS(NAME)                                       \
1803   using Assembler::NAME;                                                       \
1804   void NAME(FloatRegister Zd, SIMD_RegVariant T, PRegister Pg,                 \
1805             FloatRegister Zm) {                                                \
1806     if (Zd != Zm) {                                                            \
1807       try_to_replace_prev_vector_copy_with_movprfx(Zd);                        \
1808     }                                                                          \
1809     Assembler::NAME(Zd, T, Pg, Zm);                                            \
1810   }
1811 
1812 #define SVE_DESTRUCTIVE_BINARY_5(I1, I2, I3, I4, I5)                           \
1813   SVE_DESTRUCTIVE_BINARY_INS(I1); SVE_DESTRUCTIVE_BINARY_INS(I2);              \
1814   SVE_DESTRUCTIVE_BINARY_INS(I3); SVE_DESTRUCTIVE_BINARY_INS(I4);              \
1815   SVE_DESTRUCTIVE_BINARY_INS(I5);
1816 
1817   SVE_DESTRUCTIVE_BINARY_5(sve_add,  sve_and,   sve_asr,   sve_bic,   sve_eor)
1818   SVE_DESTRUCTIVE_BINARY_5(sve_fabd, sve_fadd,  sve_fdiv,  sve_fmax,  sve_fmin)
1819   SVE_DESTRUCTIVE_BINARY_5(sve_fmul, sve_fsub,  sve_lsl,   sve_lsr,   sve_mul)
1820   SVE_DESTRUCTIVE_BINARY_5(sve_orr,  sve_smax,  sve_smin,  sve_sqadd, sve_sqsub)
1821   SVE_DESTRUCTIVE_BINARY_5(sve_sub,  sve_uqadd, sve_uqsub, sve_umax,  sve_umin)
1822 
1823 #undef SVE_DESTRUCTIVE_BINARY_INS
1824 #undef SVE_DESTRUCTIVE_BINARY_5
1825 
1826 #define SVE_DESTRUCTIVE_SHIFT_IMM_INS(NAME)                                    \
1827   void NAME(FloatRegister Zd, SIMD_RegVariant T, PRegister Pg, int shift) {    \
1828     try_to_replace_prev_vector_copy_with_movprfx(Zd);                          \
1829     Assembler::NAME(Zd, T, Pg, shift);                                         \
1830   }
1831 
1832   SVE_DESTRUCTIVE_SHIFT_IMM_INS(sve_asr);
1833   SVE_DESTRUCTIVE_SHIFT_IMM_INS(sve_lsl);
1834   SVE_DESTRUCTIVE_SHIFT_IMM_INS(sve_lsr);
1835 
1836 #undef SVE_DESTRUCTIVE_SHIFT_IMM_INS
1837 
1838 #define SVE_DESTRUCTIVE_UNPRED_IMM_INS(NAME, IMM_TYPE)                         \
1839   void NAME(FloatRegister Zd, SIMD_RegVariant T, IMM_TYPE imm) {               \
1840     try_to_replace_prev_vector_copy_with_movprfx(Zd);                          \
1841     Assembler::NAME(Zd, T, imm);                                               \
1842   }
1843 
1844   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_add, unsigned);
1845   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_sub, unsigned);
1846   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_and, uint64_t);
1847   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_eor, uint64_t);
1848   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_orr, uint64_t);
1849 
1850 #undef SVE_DESTRUCTIVE_UNPRED_IMM_INS
1851 
1852 #define SVE_DESTRUCTIVE_TERNARY_INS(NAME)                                      \
1853   using Assembler::NAME;                                                       \
1854   void NAME(FloatRegister Zd, SIMD_RegVariant T, PRegister Pg,                 \
1855             FloatRegister Zn, FloatRegister Zm) {                              \
1856     if (Zd != Zn && Zd != Zm) {                                                \
1857       try_to_replace_prev_vector_copy_with_movprfx(Zd);                        \
1858     }                                                                          \
1859     Assembler::NAME(Zd, T, Pg, Zn, Zm);                                        \
1860   }
1861 
1862   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmad);
1863   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmla);
1864   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmls);
1865   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmsb);
1866   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmad);
1867   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmla);
1868   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmls);
1869   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmsb);
1870   SVE_DESTRUCTIVE_TERNARY_INS(sve_mla);
1871   SVE_DESTRUCTIVE_TERNARY_INS(sve_mls);
1872 
1873 #undef SVE_DESTRUCTIVE_TERNARY_INS
1874 
1875   using Assembler::sve_eor3;
1876   void sve_eor3(FloatRegister Zd, FloatRegister Zm, FloatRegister Zk) {
1877     if (Zd != Zm && Zd != Zk) {
1878       try_to_replace_prev_vector_copy_with_movprfx(Zd);
1879     }
1880     Assembler::sve_eor3(Zd, Zm, Zk);
1881   }
1882 };
1883 
1884 #ifdef ASSERT
1885 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1886 #endif
1887 
1888 struct tableswitch {
1889   Register _reg;
1890   int _insn_index; jint _first_key; jint _last_key;
1891   Label _after;
1892   Label _branches;
1893 };
1894 
1895 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP