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/atomicAccess.hpp"
  36 #include "runtime/vm_version.hpp"
  37 #include "utilities/globalDefinitions.hpp"
  38 #include "utilities/macros.hpp"
  39 #include "utilities/powerOfTwo.hpp"
  40 #include "runtime/signature.hpp"
  41 
  42 
  43 class ciInlineKlass;
  44 
  45 class OopMap;
  46 
  47 // MacroAssembler extends Assembler by frequently used macros.
  48 //
  49 // Instructions for which a 'better' code sequence exists depending
  50 // on arguments should also go in here.
  51 
  52 class MacroAssembler: public Assembler {
  53   friend class LIR_Assembler;
  54 
  55  public:
  56   using Assembler::mov;
  57   using Assembler::movi;
  58 
  59  protected:
  60 
  61   // Support for VM calls
  62   //
  63   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
  64   // may customize this version by overriding it for its purposes (e.g., to save/restore
  65   // additional registers when doing a VM call).
  66   virtual void call_VM_leaf_base(
  67     address entry_point,               // the entry point
  68     int     number_of_arguments,        // the number of arguments to pop after the call
  69     Label *retaddr = nullptr
  70   );
  71 
  72   virtual void call_VM_leaf_base(
  73     address entry_point,               // the entry point
  74     int     number_of_arguments,        // the number of arguments to pop after the call
  75     Label &retaddr) {
  76     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
  77   }
  78 
  79   // This is the base routine called by the different versions of call_VM. The interpreter
  80   // may customize this version by overriding it for its purposes (e.g., to save/restore
  81   // additional registers when doing a VM call).
  82   //
  83   // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
  84   // returns the register which contains the thread upon return. If a thread register has been
  85   // specified, the return value will correspond to that register. If no last_java_sp is specified
  86   // (noreg) than rsp will be used instead.
  87   virtual void call_VM_base(           // returns the register containing the thread upon return
  88     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
  89     Register java_thread,              // the thread if computed before     ; use noreg otherwise
  90     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
  91     Label*   return_pc,                // to set up last_Java_frame; use nullptr otherwise
  92     address  entry_point,              // the entry point
  93     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
  94     bool     check_exceptions          // whether to check for pending exceptions after return
  95   );
  96 
  97   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
  98 
  99   enum KlassDecodeMode {
 100     KlassDecodeNone,
 101     KlassDecodeZero,
 102     KlassDecodeXor,
 103     KlassDecodeMovk
 104   };
 105 
 106   // Calculate decoding mode based on given parameters, used for checking then ultimately setting.
 107   static KlassDecodeMode klass_decode_mode(address base, int shift, const size_t range);
 108 
 109  private:
 110   static KlassDecodeMode _klass_decode_mode;
 111 
 112   // Returns above setting with asserts
 113   static KlassDecodeMode klass_decode_mode();
 114 
 115  public:
 116   // Checks the decode mode and returns false if not compatible with preferred decoding mode.
 117   static bool check_klass_decode_mode(address base, int shift, const size_t range);
 118 
 119   // Sets the decode mode and returns false if cannot be set.
 120   static bool set_klass_decode_mode(address base, int shift, const size_t range);
 121 
 122  public:
 123   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 124 
 125  // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 126  // The implementation is only non-empty for the InterpreterMacroAssembler,
 127  // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 128  virtual void check_and_handle_popframe(Register java_thread);
 129  virtual void check_and_handle_earlyret(Register java_thread);
 130 
 131   void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp = rscratch1);
 132   void rt_call(address dest, Register tmp = rscratch1);
 133 
 134   // Load Effective Address
 135   void lea(Register r, const Address &a) {
 136     InstructionMark im(this);
 137     a.lea(this, r);
 138   }
 139 
 140   // Whether materializing the given address for a LDR/STR requires an
 141   // additional lea instruction.
 142   static bool legitimize_address_requires_lea(const Address &a, int size) {
 143     return a.getMode() == Address::base_plus_offset &&
 144            !Address::offset_ok_for_immed(a.offset(), exact_log2(size));
 145   }
 146 
 147   /* Sometimes we get misaligned loads and stores, usually from Unsafe
 148      accesses, and these can exceed the offset range. */
 149   Address legitimize_address(const Address &a, int size, Register scratch) {
 150     if (legitimize_address_requires_lea(a, size)) {
 151       block_comment("legitimize_address {");
 152       lea(scratch, a);
 153       block_comment("} legitimize_address");
 154       return Address(scratch);
 155     }
 156     return a;
 157   }
 158 
 159   void addmw(Address a, Register incr, Register scratch) {
 160     ldrw(scratch, a);
 161     addw(scratch, scratch, incr);
 162     strw(scratch, a);
 163   }
 164 
 165   // Add constant to memory word
 166   void addmw(Address a, int imm, Register scratch) {
 167     ldrw(scratch, a);
 168     if (imm > 0)
 169       addw(scratch, scratch, (unsigned)imm);
 170     else
 171       subw(scratch, scratch, (unsigned)-imm);
 172     strw(scratch, a);
 173   }
 174 
 175   void bind(Label& L) {
 176     Assembler::bind(L);
 177     code()->clear_last_merge_candidate();
 178     code()->set_last_label(pc());
 179   }
 180 
 181   void membar(Membar_mask_bits order_constraint);
 182 
 183   using Assembler::ldr;
 184   using Assembler::str;
 185   using Assembler::ldrw;
 186   using Assembler::strw;
 187 
 188   void ldr(Register Rx, const Address &adr);
 189   void ldrw(Register Rw, const Address &adr);
 190   void str(Register Rx, const Address &adr);
 191   void strw(Register Rx, const Address &adr);
 192 
 193   // Frame creation and destruction shared between JITs.
 194   DEBUG_ONLY(void build_frame(int framesize);)
 195   void build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills));
 196   void remove_frame(int framesize);
 197 
 198   virtual void _call_Unimplemented(address call_site) {
 199     mov(rscratch2, call_site);
 200   }
 201 
 202 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 203 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 204 // https://reviews.llvm.org/D3311
 205 
 206 #ifdef _WIN64
 207 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 208 #else
 209 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 210 #endif
 211 
 212   // aliases defined in AARCH64 spec
 213 
 214   template<class T>
 215   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }
 216 
 217   inline void cmp(Register Rd, unsigned char imm8)  { subs(zr, Rd, imm8); }
 218   inline void cmp(Register Rd, unsigned imm) = delete;
 219 
 220   template<class T>
 221   inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); }
 222 
 223   inline void cmn(Register Rd, unsigned char imm8)  { adds(zr, Rd, imm8); }
 224   inline void cmn(Register Rd, unsigned imm) = delete;
 225 
 226   void cset(Register Rd, Assembler::Condition cond) {
 227     csinc(Rd, zr, zr, ~cond);
 228   }
 229   void csetw(Register Rd, Assembler::Condition cond) {
 230     csincw(Rd, zr, zr, ~cond);
 231   }
 232 
 233   void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
 234     csneg(Rd, Rn, Rn, ~cond);
 235   }
 236   void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
 237     csnegw(Rd, Rn, Rn, ~cond);
 238   }
 239 
 240   inline void movw(Register Rd, Register Rn) {
 241     if (Rd == sp || Rn == sp) {
 242       Assembler::addw(Rd, Rn, 0U);
 243     } else {
 244       orrw(Rd, zr, Rn);
 245     }
 246   }
 247   inline void mov(Register Rd, Register Rn) {
 248     assert(Rd != r31_sp && Rn != r31_sp, "should be");
 249     if (Rd == Rn) {
 250     } else if (Rd == sp || Rn == sp) {
 251       Assembler::add(Rd, Rn, 0U);
 252     } else {
 253       orr(Rd, zr, Rn);
 254     }
 255   }
 256 
 257   inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
 258   inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
 259 
 260   inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
 261   inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
 262 
 263   inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
 264   inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
 265 
 266   inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 267     bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 268   }
 269   inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 270     bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 271   }
 272 
 273   inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 274     bfmw(Rd, Rn, lsb, (lsb + width - 1));
 275   }
 276   inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 277     bfm(Rd, Rn, lsb , (lsb + width - 1));
 278   }
 279 
 280   inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 281     sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 282   }
 283   inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 284     sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 285   }
 286 
 287   inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 288     sbfmw(Rd, Rn, lsb, (lsb + width - 1));
 289   }
 290   inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 291     sbfm(Rd, Rn, lsb , (lsb + width - 1));
 292   }
 293 
 294   inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 295     ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 296   }
 297   inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 298     ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 299   }
 300 
 301   inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 302     ubfmw(Rd, Rn, lsb, (lsb + width - 1));
 303   }
 304   inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 305     ubfm(Rd, Rn, lsb , (lsb + width - 1));
 306   }
 307 
 308   inline void asrw(Register Rd, Register Rn, unsigned imm) {
 309     sbfmw(Rd, Rn, imm, 31);
 310   }
 311 
 312   inline void asr(Register Rd, Register Rn, unsigned imm) {
 313     sbfm(Rd, Rn, imm, 63);
 314   }
 315 
 316   inline void lslw(Register Rd, Register Rn, unsigned imm) {
 317     ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
 318   }
 319 
 320   inline void lsl(Register Rd, Register Rn, unsigned imm) {
 321     ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
 322   }
 323 
 324   inline void lsrw(Register Rd, Register Rn, unsigned imm) {
 325     ubfmw(Rd, Rn, imm, 31);
 326   }
 327 
 328   inline void lsr(Register Rd, Register Rn, unsigned imm) {
 329     ubfm(Rd, Rn, imm, 63);
 330   }
 331 
 332   inline void rorw(Register Rd, Register Rn, unsigned imm) {
 333     extrw(Rd, Rn, Rn, imm);
 334   }
 335 
 336   inline void ror(Register Rd, Register Rn, unsigned imm) {
 337     extr(Rd, Rn, Rn, imm);
 338   }
 339 
 340   inline void rolw(Register Rd, Register Rn, unsigned imm) {
 341     extrw(Rd, Rn, Rn, (32 - imm));
 342   }
 343 
 344   inline void rol(Register Rd, Register Rn, unsigned imm) {
 345     extr(Rd, Rn, Rn, (64 - imm));
 346   }
 347 
 348   using Assembler::rax1;
 349   using Assembler::eor3;
 350 
 351   inline void rax1(Register Rd, Register Rn, Register Rm) {
 352     eor(Rd, Rn, Rm, ROR, 63); // Rd = Rn ^ rol(Rm, 1)
 353   }
 354 
 355   inline void eor3(Register Rd, Register Rn, Register Rm, Register Rk) {
 356     assert(Rd != Rn, "Use tmp register");
 357     eor(Rd, Rm, Rk);
 358     eor(Rd, Rd, Rn);
 359   }
 360 
 361   inline void sxtbw(Register Rd, Register Rn) {
 362     sbfmw(Rd, Rn, 0, 7);
 363   }
 364   inline void sxthw(Register Rd, Register Rn) {
 365     sbfmw(Rd, Rn, 0, 15);
 366   }
 367   inline void sxtb(Register Rd, Register Rn) {
 368     sbfm(Rd, Rn, 0, 7);
 369   }
 370   inline void sxth(Register Rd, Register Rn) {
 371     sbfm(Rd, Rn, 0, 15);
 372   }
 373   inline void sxtw(Register Rd, Register Rn) {
 374     sbfm(Rd, Rn, 0, 31);
 375   }
 376 
 377   inline void uxtbw(Register Rd, Register Rn) {
 378     ubfmw(Rd, Rn, 0, 7);
 379   }
 380   inline void uxthw(Register Rd, Register Rn) {
 381     ubfmw(Rd, Rn, 0, 15);
 382   }
 383   inline void uxtb(Register Rd, Register Rn) {
 384     ubfm(Rd, Rn, 0, 7);
 385   }
 386   inline void uxth(Register Rd, Register Rn) {
 387     ubfm(Rd, Rn, 0, 15);
 388   }
 389   inline void uxtw(Register Rd, Register Rn) {
 390     ubfm(Rd, Rn, 0, 31);
 391   }
 392 
 393   inline void cmnw(Register Rn, Register Rm) {
 394     addsw(zr, Rn, Rm);
 395   }
 396   inline void cmn(Register Rn, Register Rm) {
 397     adds(zr, Rn, Rm);
 398   }
 399 
 400   inline void cmpw(Register Rn, Register Rm) {
 401     subsw(zr, Rn, Rm);
 402   }
 403   inline void cmp(Register Rn, Register Rm) {
 404     subs(zr, Rn, Rm);
 405   }
 406 
 407   inline void negw(Register Rd, Register Rn) {
 408     subw(Rd, zr, Rn);
 409   }
 410 
 411   inline void neg(Register Rd, Register Rn) {
 412     sub(Rd, zr, Rn);
 413   }
 414 
 415   inline void negsw(Register Rd, Register Rn) {
 416     subsw(Rd, zr, Rn);
 417   }
 418 
 419   inline void negs(Register Rd, Register Rn) {
 420     subs(Rd, zr, Rn);
 421   }
 422 
 423   inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 424     addsw(zr, Rn, Rm, kind, shift);
 425   }
 426   inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 427     adds(zr, Rn, Rm, kind, shift);
 428   }
 429 
 430   inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 431     subsw(zr, Rn, Rm, kind, shift);
 432   }
 433   inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 434     subs(zr, Rn, Rm, kind, shift);
 435   }
 436 
 437   inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 438     subw(Rd, zr, Rn, kind, shift);
 439   }
 440 
 441   inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 442     sub(Rd, zr, Rn, kind, shift);
 443   }
 444 
 445   inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 446     subsw(Rd, zr, Rn, kind, shift);
 447   }
 448 
 449   inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 450     subs(Rd, zr, Rn, kind, shift);
 451   }
 452 
 453   inline void mnegw(Register Rd, Register Rn, Register Rm) {
 454     msubw(Rd, Rn, Rm, zr);
 455   }
 456   inline void mneg(Register Rd, Register Rn, Register Rm) {
 457     msub(Rd, Rn, Rm, zr);
 458   }
 459 
 460   inline void mulw(Register Rd, Register Rn, Register Rm) {
 461     maddw(Rd, Rn, Rm, zr);
 462   }
 463   inline void mul(Register Rd, Register Rn, Register Rm) {
 464     madd(Rd, Rn, Rm, zr);
 465   }
 466 
 467   inline void smnegl(Register Rd, Register Rn, Register Rm) {
 468     smsubl(Rd, Rn, Rm, zr);
 469   }
 470   inline void smull(Register Rd, Register Rn, Register Rm) {
 471     smaddl(Rd, Rn, Rm, zr);
 472   }
 473 
 474   inline void umnegl(Register Rd, Register Rn, Register Rm) {
 475     umsubl(Rd, Rn, Rm, zr);
 476   }
 477   inline void umull(Register Rd, Register Rn, Register Rm) {
 478     umaddl(Rd, Rn, Rm, zr);
 479   }
 480 
 481 #define WRAP(INSN)                                                            \
 482   void INSN(Register Rd, Register Rn, Register Rm, Register Ra) {             \
 483     if (VM_Version::supports_a53mac() && Ra != zr)                            \
 484       nop();                                                                  \
 485     Assembler::INSN(Rd, Rn, Rm, Ra);                                          \
 486   }
 487 
 488   WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
 489   WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
 490 #undef WRAP
 491 
 492   using Assembler::andw, Assembler::andr;
 493   void andw(Register Rd, Register Rn, uint64_t imm) {
 494     if (operand_valid_for_logical_immediate(/*is32*/true, imm)) {
 495       Assembler::andw(Rd, Rn, imm);
 496     } else {
 497       assert(Rd != Rn, "must be");
 498       movw(Rd, imm);
 499       andw(Rd, Rn, Rd);
 500     }
 501   }
 502   void andr(Register Rd, Register Rn, uint64_t imm) {
 503     if (operand_valid_for_logical_immediate(/*is32*/false, imm)) {
 504       Assembler::andr(Rd, Rn, imm);
 505     } else {
 506       assert(Rd != Rn, "must be");
 507       mov(Rd, imm);
 508       andr(Rd, Rn, Rd);
 509     }
 510   }
 511 
 512   // macro assembly operations needed for aarch64
 513 
 514 public:
 515 
 516   enum FpPushPopMode {
 517     PushPopFull,
 518     PushPopSVE,
 519     PushPopNeon,
 520     PushPopFp
 521   };
 522 
 523   // first two private routines for loading 32 bit or 64 bit constants
 524 private:
 525 
 526   void mov_immediate64(Register dst, uint64_t imm64);
 527   void mov_immediate32(Register dst, uint32_t imm32);
 528 
 529   void mov(Register dst, Address a);
 530 
 531 public:
 532 
 533   int push(RegSet regset, Register stack);
 534   int pop(RegSet regset, Register stack);
 535 
 536   int push_fp(FloatRegSet regset, Register stack, FpPushPopMode mode = PushPopFull);
 537   int pop_fp(FloatRegSet regset, Register stack, FpPushPopMode mode = PushPopFull);
 538 
 539   static RegSet call_clobbered_gp_registers();
 540 
 541   int push_p(PRegSet regset, Register stack);
 542   int pop_p(PRegSet regset, Register stack);
 543 
 544   // Push and pop everything that might be clobbered by a native
 545   // runtime call except rscratch1 and rscratch2.  (They are always
 546   // scratch, so we don't have to protect them.)  Only save the lower
 547   // 64 bits of each vector register. Additional registers can be excluded
 548   // in a passed RegSet.
 549   void push_call_clobbered_registers_except(RegSet exclude);
 550   void pop_call_clobbered_registers_except(RegSet exclude);
 551 
 552   void push_call_clobbered_registers() {
 553     push_call_clobbered_registers_except(RegSet());
 554   }
 555   void pop_call_clobbered_registers() {
 556     pop_call_clobbered_registers_except(RegSet());
 557   }
 558 
 559 
 560   // now mov instructions for loading absolute addresses and 32 or
 561   // 64 bit integers
 562 
 563   inline void mov(Register dst, address addr)             { mov_immediate64(dst, (uint64_t)addr); }
 564 
 565   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 566   inline void mov(Register dst, T o)                      { mov_immediate64(dst, (uint64_t)o); }
 567 
 568   inline void movw(Register dst, uint32_t imm32)          { mov_immediate32(dst, imm32); }
 569 
 570   void mov(Register dst, RegisterOrConstant src) {
 571     if (src.is_register())
 572       mov(dst, src.as_register());
 573     else
 574       mov(dst, src.as_constant());
 575   }
 576 
 577   void movptr(Register r, uintptr_t imm64);
 578 
 579   void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64);
 580 
 581   void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
 582     orr(Vd, T, Vn, Vn);
 583   }
 584 
 585   void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) {
 586     fcvtsh(tmp, src);
 587     smov(dst, tmp, H, 0);
 588   }
 589 
 590   void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) {
 591     mov(tmp, H, 0, src);
 592     fcvths(dst, tmp);
 593   }
 594 
 595   // Generalized Test Bit And Branch, including a "far" variety which
 596   // spans more than 32KiB.
 597   void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) {
 598     assert(cond == EQ || cond == NE, "must be");
 599 
 600     if (isfar)
 601       cond = ~cond;
 602 
 603     void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
 604     if (cond == Assembler::EQ)
 605       branch = &Assembler::tbz;
 606     else
 607       branch = &Assembler::tbnz;
 608 
 609     if (isfar) {
 610       Label L;
 611       (this->*branch)(Rt, bitpos, L);
 612       b(dest);
 613       bind(L);
 614     } else {
 615       (this->*branch)(Rt, bitpos, dest);
 616     }
 617   }
 618 
 619   // macro instructions for accessing and updating floating point
 620   // status register
 621   //
 622   // FPSR : op1 == 011
 623   //        CRn == 0100
 624   //        CRm == 0100
 625   //        op2 == 001
 626 
 627   inline void get_fpsr(Register reg)
 628   {
 629     mrs(0b11, 0b0100, 0b0100, 0b001, reg);
 630   }
 631 
 632   inline void set_fpsr(Register reg)
 633   {
 634     msr(0b011, 0b0100, 0b0100, 0b001, reg);
 635   }
 636 
 637   inline void clear_fpsr()
 638   {
 639     msr(0b011, 0b0100, 0b0100, 0b001, zr);
 640   }
 641 
 642   // FPCR : op1 == 011
 643   //        CRn == 0100
 644   //        CRm == 0100
 645   //        op2 == 000
 646 
 647   inline void get_fpcr(Register reg) {
 648     mrs(0b11, 0b0100, 0b0100, 0b000, reg);
 649   }
 650 
 651   inline void set_fpcr(Register reg) {
 652     msr(0b011, 0b0100, 0b0100, 0b000, reg);
 653   }
 654 
 655   // DCZID_EL0: op1 == 011
 656   //            CRn == 0000
 657   //            CRm == 0000
 658   //            op2 == 111
 659   inline void get_dczid_el0(Register reg)
 660   {
 661     mrs(0b011, 0b0000, 0b0000, 0b111, reg);
 662   }
 663 
 664   // CTR_EL0:   op1 == 011
 665   //            CRn == 0000
 666   //            CRm == 0000
 667   //            op2 == 001
 668   inline void get_ctr_el0(Register reg)
 669   {
 670     mrs(0b011, 0b0000, 0b0000, 0b001, reg);
 671   }
 672 
 673   inline void get_nzcv(Register reg) {
 674     mrs(0b011, 0b0100, 0b0010, 0b000, reg);
 675   }
 676 
 677   inline void set_nzcv(Register reg) {
 678     msr(0b011, 0b0100, 0b0010, 0b000, reg);
 679   }
 680 
 681   // CNTVCTSS_EL0:   op1 == 011
 682   //                 CRn == 1110
 683   //                 CRm == 0000
 684   //                 op2 == 110
 685   inline void get_cntvctss_el0(Register reg) {
 686     mrs(0b011, 0b1110, 0b0000, 0b110, reg);
 687   }
 688 
 689   // idiv variant which deals with MINLONG as dividend and -1 as divisor
 690   int corrected_idivl(Register result, Register ra, Register rb,
 691                       bool want_remainder, Register tmp = rscratch1);
 692   int corrected_idivq(Register result, Register ra, Register rb,
 693                       bool want_remainder, Register tmp = rscratch1);
 694 
 695   // Support for null-checks
 696   //
 697   // Generates code that causes a null OS exception if the content of reg is null.
 698   // If the accessed location is M[reg + offset] and the offset is known, provide the
 699   // offset. No explicit code generation is needed if the offset is within a certain
 700   // range (0 <= offset <= page_size).
 701 
 702   virtual void null_check(Register reg, int offset = -1);
 703   static bool needs_explicit_null_check(intptr_t offset);
 704   static bool uses_implicit_null_check(void* address);
 705 
 706   // markWord tests, kills markWord reg
 707   void test_markword_is_inline_type(Register markword, Label& is_inline_type);
 708 
 709   // inlineKlass queries, kills temp_reg
 710   void test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null = true);
 711 
 712   void test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free);
 713   void test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free);
 714   void test_field_is_flat(Register flags, Register temp_reg, Label& is_flat);
 715 
 716   // Check oops for special arrays, i.e. flat arrays and/or null-free arrays
 717   void test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label);
 718   void test_flat_array_oop(Register klass, Register temp_reg, Label& is_flat_array);
 719   void test_non_flat_array_oop(Register oop, Register temp_reg, Label&is_non_flat_array);
 720   void test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array);
 721   void test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array);
 722 
 723   // Check array klass layout helper for flat or null-free arrays...
 724   void test_flat_array_layout(Register lh, Label& is_flat_array);
 725 
 726   static address target_addr_for_insn(address insn_addr);
 727 
 728   // Required platform-specific helpers for Label::patch_instructions.
 729   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 730   static int pd_patch_instruction_size(address branch, address target);
 731   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 732     pd_patch_instruction_size(branch, target);
 733   }
 734   static address pd_call_destination(address branch) {
 735     return target_addr_for_insn(branch);
 736   }
 737 #ifndef PRODUCT
 738   static void pd_print_patched_instruction(address branch);
 739 #endif
 740 
 741   static int patch_oop(address insn_addr, address o);
 742 
 743   // Return whether code is emitted to a scratch blob.
 744   virtual bool in_scratch_emit_size() {
 745     return false;
 746   }
 747   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
 748   static int max_trampoline_stub_size();
 749   void emit_static_call_stub();
 750   static int static_call_stub_size();
 751 
 752   // The following 4 methods return the offset of the appropriate move instruction
 753 
 754   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 755   int load_unsigned_byte(Register dst, Address src);
 756   int load_unsigned_short(Register dst, Address src);
 757 
 758   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 759   int load_signed_byte(Register dst, Address src);
 760   int load_signed_short(Register dst, Address src);
 761 
 762   int load_signed_byte32(Register dst, Address src);
 763   int load_signed_short32(Register dst, Address src);
 764 
 765   // Support for sign-extension (hi:lo = extend_sign(lo))
 766   void extend_sign(Register hi, Register lo);
 767 
 768   // Clean up a subword typed value to the representation in compliance with JVMS ยง2.3
 769   void narrow_subword_type(Register reg, BasicType bt);
 770 
 771   // Load and store values by size and signed-ness
 772   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 773   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 774 
 775   // Support for inc/dec with optimal instruction selection depending on value
 776 
 777   // x86_64 aliases an unqualified register/address increment and
 778   // decrement to call incrementq and decrementq but also supports
 779   // explicitly sized calls to incrementq/decrementq or
 780   // incrementl/decrementl
 781 
 782   // for aarch64 the proper convention would be to use
 783   // increment/decrement for 64 bit operations and
 784   // incrementw/decrementw for 32 bit operations. so when porting
 785   // x86_64 code we can leave calls to increment/decrement as is,
 786   // replace incrementq/decrementq with increment/decrement and
 787   // replace incrementl/decrementl with incrementw/decrementw.
 788 
 789   // n.b. increment/decrement calls with an Address destination will
 790   // need to use a scratch register to load the value to be
 791   // incremented. increment/decrement calls which add or subtract a
 792   // constant value greater than 2^24 will need to use a 2nd scratch
 793   // register to hold the constant. so, a register increment/decrement
 794   // may trash rscratch2 and an address increment/decrement trash
 795   // rscratch and rscratch2
 796 
 797   void decrementw(Address dst, int value = 1);
 798   void decrementw(Register reg, int value = 1);
 799 
 800   void decrement(Register reg, int value = 1);
 801   void decrement(Address dst, int value = 1);
 802 
 803   void incrementw(Address dst, int value = 1, Register result = rscratch1);
 804   void incrementw(Register reg, int value = 1);
 805 
 806   void increment(Register reg, int value = 1);
 807   void increment(Address dst, int value = 1, Register result = rscratch1);
 808 
 809 
 810   // Alignment
 811   void align(int modulus);
 812   void align(int modulus, int target);
 813 
 814   // nop
 815   void post_call_nop();
 816 
 817   // Stack frame creation/removal
 818   void enter(bool strip_ret_addr = false);
 819   void leave();
 820 
 821   // ROP Protection
 822   void protect_return_address();
 823   void protect_return_address(Register return_reg);
 824   void authenticate_return_address();
 825   void authenticate_return_address(Register return_reg);
 826   void strip_return_address();
 827   void check_return_address(Register return_reg=lr) PRODUCT_RETURN;
 828 
 829   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
 830   // The pointer will be loaded into the thread register.
 831   void get_thread(Register thread);
 832 
 833   // support for argument shuffling
 834   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 835   void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 836   void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 837   void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 838   void object_move(
 839                    OopMap* map,
 840                    int oop_handle_offset,
 841                    int framesize_in_slots,
 842                    VMRegPair src,
 843                    VMRegPair dst,
 844                    bool is_receiver,
 845                    int* receiver_offset);
 846 
 847 
 848   // Support for VM calls
 849   //
 850   // It is imperative that all calls into the VM are handled via the call_VM macros.
 851   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 852   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 853 
 854 
 855   void call_VM(Register oop_result,
 856                address entry_point,
 857                bool check_exceptions = true);
 858   void call_VM(Register oop_result,
 859                address entry_point,
 860                Register arg_1,
 861                bool check_exceptions = true);
 862   void call_VM(Register oop_result,
 863                address entry_point,
 864                Register arg_1, Register arg_2,
 865                bool check_exceptions = true);
 866   void call_VM(Register oop_result,
 867                address entry_point,
 868                Register arg_1, Register arg_2, Register arg_3,
 869                bool check_exceptions = true);
 870 
 871   // Overloadings with last_Java_sp
 872   void call_VM(Register oop_result,
 873                Register last_java_sp,
 874                address entry_point,
 875                int number_of_arguments = 0,
 876                bool check_exceptions = true);
 877   void call_VM(Register oop_result,
 878                Register last_java_sp,
 879                address entry_point,
 880                Register arg_1, bool
 881                check_exceptions = true);
 882   void call_VM(Register oop_result,
 883                Register last_java_sp,
 884                address entry_point,
 885                Register arg_1, Register arg_2,
 886                bool check_exceptions = true);
 887   void call_VM(Register oop_result,
 888                Register last_java_sp,
 889                address entry_point,
 890                Register arg_1, Register arg_2, Register arg_3,
 891                bool check_exceptions = true);
 892 
 893   void get_vm_result_oop(Register oop_result, Register thread);
 894   void get_vm_result_metadata(Register metadata_result, Register thread);
 895 
 896   // These always tightly bind to MacroAssembler::call_VM_base
 897   // bypassing the virtual implementation
 898   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 899   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
 900   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 901   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);
 902   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);
 903 
 904   void call_VM_leaf(address entry_point,
 905                     int number_of_arguments = 0);
 906   void call_VM_leaf(address entry_point,
 907                     Register arg_1);
 908   void call_VM_leaf(address entry_point,
 909                     Register arg_1, Register arg_2);
 910   void call_VM_leaf(address entry_point,
 911                     Register arg_1, Register arg_2, Register arg_3);
 912 
 913   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 914   // bypassing the virtual implementation
 915   void super_call_VM_leaf(address entry_point);
 916   void super_call_VM_leaf(address entry_point, Register arg_1);
 917   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 918   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 919   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 920 
 921   // last Java Frame (fills frame anchor)
 922   void set_last_Java_frame(Register last_java_sp,
 923                            Register last_java_fp,
 924                            address last_java_pc,
 925                            Register scratch);
 926 
 927   void set_last_Java_frame(Register last_java_sp,
 928                            Register last_java_fp,
 929                            Label &last_java_pc,
 930                            Register scratch);
 931 
 932   void set_last_Java_frame(Register last_java_sp,
 933                            Register last_java_fp,
 934                            Register last_java_pc,
 935                            Register scratch);
 936 
 937   void reset_last_Java_frame(Register thread);
 938 
 939   // thread in the default location (rthread)
 940   void reset_last_Java_frame(bool clear_fp);
 941 
 942   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 943   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 944 
 945   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 946   void c2bool(Register x);
 947 
 948   void load_method_holder_cld(Register rresult, Register rmethod);
 949   void load_method_holder(Register holder, Register method);
 950 
 951   // oop manipulations
 952   void load_metadata(Register dst, Register src);
 953 
 954   void load_narrow_klass_compact(Register dst, Register src);
 955   void load_klass(Register dst, Register src);
 956   void store_klass(Register dst, Register src);
 957   void cmp_klass(Register obj, Register klass, Register tmp);
 958   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 959 
 960   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 961   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 962   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 963 
 964   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 965                       Register tmp1, Register tmp2);
 966 
 967   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 968                        Register tmp1, Register tmp2, Register tmp3);
 969 
 970   void flat_field_copy(DecoratorSet decorators, Register src, Register dst, Register inline_layout_info);
 971 
 972   // inline type data payload offsets...
 973   void payload_offset(Register inline_klass, Register offset);
 974   void payload_address(Register oop, Register data, Register inline_klass);
 975 
 976   void load_heap_oop(Register dst, Address src, Register tmp1,
 977                      Register tmp2, DecoratorSet decorators = 0);
 978 
 979   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 980                               Register tmp2, DecoratorSet decorators = 0);
 981   void store_heap_oop(Address dst, Register val, Register tmp1,
 982                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 983 
 984   // currently unimplemented
 985   // Used for storing null. All other oop constants should be
 986   // stored using routines that take a jobject.
 987   void store_heap_oop_null(Address dst);
 988 
 989   void load_prototype_header(Register dst, Register src);
 990 
 991   void store_klass_gap(Register dst, Register src);
 992 
 993   // This dummy is to prevent a call to store_heap_oop from
 994   // converting a zero (like null) into a Register by giving
 995   // the compiler two choices it can't resolve
 996 
 997   void store_heap_oop(Address dst, void* dummy);
 998 
 999   void encode_heap_oop(Register d, Register s);
1000   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
1001   void decode_heap_oop(Register d, Register s);
1002   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
1003   void encode_heap_oop_not_null(Register r);
1004   void decode_heap_oop_not_null(Register r);
1005   void encode_heap_oop_not_null(Register dst, Register src);
1006   void decode_heap_oop_not_null(Register dst, Register src);
1007 
1008   void set_narrow_oop(Register dst, jobject obj);
1009 
1010   void decode_klass_not_null_for_aot(Register dst, Register src);
1011   void encode_klass_not_null_for_aot(Register dst, Register src);
1012   void encode_klass_not_null(Register r);
1013   void decode_klass_not_null(Register r);
1014   void encode_klass_not_null(Register dst, Register src);
1015   void decode_klass_not_null(Register dst, Register src);
1016 
1017   void set_narrow_klass(Register dst, Klass* k);
1018 
1019   // if heap base register is used - reinit it with the correct value
1020   void reinit_heapbase();
1021 
1022   DEBUG_ONLY(void verify_heapbase(const char* msg);)
1023 
1024   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
1025                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1026   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
1027                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1028 
1029   void push_cont_fastpath(Register java_thread = rthread);
1030   void pop_cont_fastpath(Register java_thread = rthread);
1031 
1032   // Round up to a power of two
1033   void round_to(Register reg, int modulus);
1034 
1035   // java.lang.Math::round intrinsics
1036   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
1037   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
1038 
1039   // allocation
1040 
1041   void tlab_allocate(
1042     Register obj,                      // result: pointer to object after successful allocation
1043     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1044     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1045     Register t1,                       // temp register
1046     Register t2,                       // temp register
1047     Label&   slow_case                 // continuation point if fast allocation fails
1048   );
1049   void verify_tlab();
1050 
1051   void inline_layout_info(Register holder_klass, Register index, Register layout_info);
1052 
1053   // interface method calling
1054   void lookup_interface_method(Register recv_klass,
1055                                Register intf_klass,
1056                                RegisterOrConstant itable_index,
1057                                Register method_result,
1058                                Register scan_temp,
1059                                Label& no_such_interface,
1060                    bool return_method = true);
1061 
1062   void lookup_interface_method_stub(Register recv_klass,
1063                                     Register holder_klass,
1064                                     Register resolved_klass,
1065                                     Register method_result,
1066                                     Register temp_reg,
1067                                     Register temp_reg2,
1068                                     int itable_index,
1069                                     Label& L_no_such_interface);
1070 
1071   // virtual method calling
1072   // n.b. x86 allows RegisterOrConstant for vtable_index
1073   void lookup_virtual_method(Register recv_klass,
1074                              RegisterOrConstant vtable_index,
1075                              Register method_result);
1076 
1077   // Test sub_klass against super_klass, with fast and slow paths.
1078 
1079   // The fast path produces a tri-state answer: yes / no / maybe-slow.
1080   // One of the three labels can be null, meaning take the fall-through.
1081   // If super_check_offset is -1, the value is loaded up from super_klass.
1082   // No registers are killed, except temp_reg.
1083   void check_klass_subtype_fast_path(Register sub_klass,
1084                                      Register super_klass,
1085                                      Register temp_reg,
1086                                      Label* L_success,
1087                                      Label* L_failure,
1088                                      Label* L_slow_path,
1089                                      Register super_check_offset = noreg);
1090 
1091   // The rest of the type check; must be wired to a corresponding fast path.
1092   // It does not repeat the fast path logic, so don't use it standalone.
1093   // The temp_reg and temp2_reg can be noreg, if no temps are available.
1094   // Updates the sub's secondary super cache as necessary.
1095   // If set_cond_codes, condition codes will be Z on success, NZ on failure.
1096   void check_klass_subtype_slow_path(Register sub_klass,
1097                                      Register super_klass,
1098                                      Register temp_reg,
1099                                      Register temp2_reg,
1100                                      Label* L_success,
1101                                      Label* L_failure,
1102                                      bool set_cond_codes = false);
1103 
1104   void check_klass_subtype_slow_path_linear(Register sub_klass,
1105                                             Register super_klass,
1106                                             Register temp_reg,
1107                                             Register temp2_reg,
1108                                             Label* L_success,
1109                                             Label* L_failure,
1110                                             bool set_cond_codes = false);
1111 
1112   void check_klass_subtype_slow_path_table(Register sub_klass,
1113                                            Register super_klass,
1114                                            Register temp_reg,
1115                                            Register temp2_reg,
1116                                            Register temp3_reg,
1117                                            Register result_reg,
1118                                            FloatRegister vtemp_reg,
1119                                            Label* L_success,
1120                                            Label* L_failure,
1121                                            bool set_cond_codes = false);
1122 
1123   // If r is valid, return r.
1124   // If r is invalid, remove a register r2 from available_regs, add r2
1125   // to regs_to_push, then return r2.
1126   Register allocate_if_noreg(const Register r,
1127                              RegSetIterator<Register> &available_regs,
1128                              RegSet &regs_to_push);
1129 
1130   // Secondary subtype checking
1131   void lookup_secondary_supers_table_var(Register sub_klass,
1132                                          Register r_super_klass,
1133                                          Register temp1,
1134                                          Register temp2,
1135                                          Register temp3,
1136                                          FloatRegister vtemp,
1137                                          Register result,
1138                                          Label *L_success);
1139 
1140 
1141   // As above, but with a constant super_klass.
1142   // The result is in Register result, not the condition codes.
1143   bool lookup_secondary_supers_table_const(Register r_sub_klass,
1144                                            Register r_super_klass,
1145                                            Register temp1,
1146                                            Register temp2,
1147                                            Register temp3,
1148                                            FloatRegister vtemp,
1149                                            Register result,
1150                                            u1 super_klass_slot,
1151                                            bool stub_is_near = false);
1152 
1153   void verify_secondary_supers_table(Register r_sub_klass,
1154                                      Register r_super_klass,
1155                                      Register temp1,
1156                                      Register temp2,
1157                                      Register result);
1158 
1159   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
1160                                                Register r_array_base,
1161                                                Register r_array_index,
1162                                                Register r_bitmap,
1163                                                Register temp1,
1164                                                Register result,
1165                                                bool is_stub = true);
1166 
1167   // Simplified, combined version, good for typical uses.
1168   // Falls through on failure.
1169   void check_klass_subtype(Register sub_klass,
1170                            Register super_klass,
1171                            Register temp_reg,
1172                            Label& L_success);
1173 
1174   void clinit_barrier(Register klass,
1175                       Register thread,
1176                       Label* L_fast_path = nullptr,
1177                       Label* L_slow_path = nullptr);
1178 
1179   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
1180 
1181   void profile_receiver_type(Register recv, Register mdp, int mdp_offset);
1182 
1183   void verify_sve_vector_length(Register tmp = rscratch1);
1184   void reinitialize_ptrue() {
1185     if (UseSVE > 0) {
1186       sve_ptrue(ptrue, B);
1187     }
1188   }
1189   void verify_ptrue();
1190 
1191   // Debugging
1192 
1193   // only if +VerifyOops
1194   void _verify_oop(Register reg, const char* s, const char* file, int line);
1195   void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
1196 
1197   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
1198     if (VerifyOops) {
1199       _verify_oop(reg, s, file, line);
1200     }
1201   }
1202   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
1203     if (VerifyOops) {
1204       _verify_oop_addr(reg, s, file, line);
1205     }
1206   }
1207 
1208 // TODO: verify method and klass metadata (compare against vptr?)
1209   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1210   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1211 
1212 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
1213 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
1214 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
1215 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1216 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1217 
1218   // Restore cpu control state after JNI call
1219   void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2);
1220 
1221   // prints msg, dumps registers and stops execution
1222   void stop(const char* msg);
1223 
1224   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1225 
1226   void untested()                                { stop("untested"); }
1227 
1228   void unimplemented(const char* what = "");
1229 
1230   void should_not_reach_here()                   { stop("should not reach here"); }
1231 
1232   void _assert_asm(Condition cc, const char* msg);
1233 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg)
1234 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg))
1235 
1236   // Stack overflow checking
1237   void bang_stack_with_offset(int offset) {
1238     // stack grows down, caller passes positive offset
1239     assert(offset > 0, "must bang with negative offset");
1240     sub(rscratch2, sp, offset);
1241     str(zr, Address(rscratch2));
1242   }
1243 
1244   // Writes to stack successive pages until offset reached to check for
1245   // stack overflow + shadow pages.  Also, clobbers tmp
1246   void bang_stack_size(Register size, Register tmp);
1247 
1248   // Check for reserved stack access in method being exited (for JIT)
1249   void reserved_stack_check();
1250 
1251   // Arithmetics
1252 
1253   // Clobber: rscratch1, rscratch2
1254   void addptr(const Address &dst, int32_t src);
1255 
1256   // Clobber: rscratch1
1257   void cmpptr(Register src1, Address src2);
1258 
1259   void cmpoop(Register obj1, Register obj2);
1260 
1261   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1262   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1263   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1264   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1265 
1266   void atomic_xchg(Register prev, Register newv, Register addr);
1267   void atomic_xchgw(Register prev, Register newv, Register addr);
1268   void atomic_xchgl(Register prev, Register newv, Register addr);
1269   void atomic_xchglw(Register prev, Register newv, Register addr);
1270   void atomic_xchgal(Register prev, Register newv, Register addr);
1271   void atomic_xchgalw(Register prev, Register newv, Register addr);
1272 
1273   void orptr(Address adr, RegisterOrConstant src) {
1274     ldr(rscratch1, adr);
1275     if (src.is_register())
1276       orr(rscratch1, rscratch1, src.as_register());
1277     else
1278       orr(rscratch1, rscratch1, src.as_constant());
1279     str(rscratch1, adr);
1280   }
1281 
1282 private:
1283   // A generic CAS; success or failure is in the EQ flag.
1284   // Clobbers rscratch1
1285   void cmpxchg(Register addr, Register expected, Register new_val,
1286                enum operand_size size, enum atomic_memory_order order,
1287                bool weak, Register result);
1288 
1289 public:
1290   void cmpxchg(Register addr, Register expected, Register new_val,
1291                enum operand_size size, enum atomic_memory_order order,
1292                Register result = noreg) {
1293     cmpxchg(addr, expected, new_val, size, order, /* weak */ false, result);
1294   }
1295 
1296   void cmpxchg_weak(Register addr, Register expected, Register new_val,
1297                     enum operand_size size, enum atomic_memory_order order,
1298                     Register result = noreg) {
1299     cmpxchg(addr, expected, new_val, size, order, /* weak */ true, result);
1300   }
1301 
1302 #ifdef ASSERT
1303   // Template short-hand support to clean-up after a failed call to trampoline
1304   // call generation (see trampoline_call() below),  when a set of Labels must
1305   // be reset (before returning).
1306   template<typename Label, typename... More>
1307   void reset_labels(Label &lbl, More&... more) {
1308     lbl.reset(); reset_labels(more...);
1309   }
1310   template<typename Label>
1311   void reset_labels(Label &lbl) {
1312     lbl.reset();
1313   }
1314 #endif
1315 
1316 private:
1317   void compare_eq(Register rn, Register rm, enum operand_size size);
1318 
1319 public:
1320   // AArch64 OpenJDK uses four different types of calls:
1321   //   - direct call: bl pc_relative_offset
1322   //     This is the shortest and the fastest, but the offset has the range:
1323   //     +/-128MB for the release build, +/-2MB for the debug build.
1324   //
1325   //   - far call: adrp reg, pc_relative_offset; add; bl reg
1326   //     This is longer than a direct call. The offset has
1327   //     the range +/-4GB. As the code cache size is limited to 4GB,
1328   //     far calls can reach anywhere in the code cache. If a jump is
1329   //     needed rather than a call, a far jump 'b reg' can be used instead.
1330   //     All instructions are embedded at a call site.
1331   //
1332   //   - trampoline call:
1333   //     This is only available in C1/C2-generated code (nmethod). It is a combination
1334   //     of a direct call, which is used if the destination of a call is in range,
1335   //     and a register-indirect call. It has the advantages of reaching anywhere in
1336   //     the AArch64 address space and being patchable at runtime when the generated
1337   //     code is being executed by other threads.
1338   //
1339   //     [Main code section]
1340   //       bl trampoline
1341   //     [Stub code section]
1342   //     trampoline:
1343   //       ldr reg, pc + 8
1344   //       br reg
1345   //       <64-bit destination address>
1346   //
1347   //     If the destination is in range when the generated code is moved to the code
1348   //     cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline
1349   //     is not used.
1350   //     The optimization does not remove the trampoline from the stub section.
1351   //     This is necessary because the trampoline may well be redirected later when
1352   //     code is patched, and the new destination may not be reachable by a simple BR
1353   //     instruction.
1354   //
1355   //   - indirect call: move reg, address; blr reg
1356   //     This too can reach anywhere in the address space, but it cannot be
1357   //     patched while code is running, so it must only be modified at a safepoint.
1358   //     This form of call is most suitable for targets at fixed addresses, which
1359   //     will never be patched.
1360   //
1361   // The patching we do conforms to the "Concurrent modification and
1362   // execution of instructions" section of the Arm Architectural
1363   // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC,
1364   // or SVC instructions to be modified while another thread is
1365   // executing them.
1366   //
1367   // To patch a trampoline call when the BL can't reach, we first modify
1368   // the 64-bit destination address in the trampoline, then modify the
1369   // BL to point to the trampoline, then flush the instruction cache to
1370   // broadcast the change to all executing threads. See
1371   // NativeCall::set_destination_mt_safe for the details.
1372   //
1373   // There is a benign race in that the other thread might observe the
1374   // modified BL before it observes the modified 64-bit destination
1375   // address. That does not matter because the destination method has been
1376   // invalidated, so there will be a trap at its start.
1377   // For this to work, the destination address in the trampoline is
1378   // always updated, even if we're not using the trampoline.
1379 
1380   // Emit a direct call if the entry address will always be in range,
1381   // otherwise a trampoline call.
1382   // Supported entry.rspec():
1383   // - relocInfo::runtime_call_type
1384   // - relocInfo::opt_virtual_call_type
1385   // - relocInfo::static_call_type
1386   // - relocInfo::virtual_call_type
1387   //
1388   // Return: the call PC or null if CodeCache is full.
1389   // Clobbers: rscratch1
1390   address trampoline_call(Address entry);
1391 
1392   static bool far_branches() {
1393     return ReservedCodeCacheSize > branch_range;
1394   }
1395 
1396   // Check if branches to the non nmethod section require a far jump
1397   static bool codestub_branch_needs_far_jump() {
1398     if (AOTCodeCache::is_on_for_dump()) {
1399       // To calculate far_codestub_branch_size correctly.
1400       return true;
1401     }
1402     return CodeCache::max_distance_to_non_nmethod() > branch_range;
1403   }
1404 
1405   // Emit a direct call/jump if the entry address will always be in range,
1406   // otherwise a far call/jump.
1407   // The address must be inside the code cache.
1408   // Supported entry.rspec():
1409   // - relocInfo::external_word_type
1410   // - relocInfo::runtime_call_type
1411   // - relocInfo::none
1412   // In the case of a far call/jump, the entry address is put in the tmp register.
1413   // The tmp register is invalidated.
1414   //
1415   // Far_jump returns the amount of the emitted code.
1416   void far_call(Address entry, Register tmp = rscratch1);
1417   int far_jump(Address entry, Register tmp = rscratch1);
1418 
1419   static int far_codestub_branch_size() {
1420     if (codestub_branch_needs_far_jump()) {
1421       return 3 * 4;  // adrp, add, br
1422     } else {
1423       return 4;
1424     }
1425   }
1426 
1427   // Emit the CompiledIC call idiom
1428   address ic_call(address entry, jint method_index = 0);
1429   static int ic_check_size();
1430   int ic_check(int end_alignment);
1431 
1432 public:
1433 
1434   // Data
1435 
1436   void mov_metadata(Register dst, Metadata* obj);
1437   Address allocate_metadata_address(Metadata* obj);
1438   Address constant_oop_address(jobject obj);
1439 
1440   void movoop(Register dst, jobject obj);
1441 
1442   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1443   void kernel_crc32(Register crc, Register buf, Register len,
1444         Register table0, Register table1, Register table2, Register table3,
1445         Register tmp, Register tmp2, Register tmp3);
1446   // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic.
1447   void kernel_crc32c(Register crc, Register buf, Register len,
1448         Register table0, Register table1, Register table2, Register table3,
1449         Register tmp, Register tmp2, Register tmp3);
1450 
1451   // Stack push and pop individual 64 bit registers
1452   void push(Register src);
1453   void pop(Register dst);
1454 
1455   void repne_scan(Register addr, Register value, Register count,
1456                   Register scratch);
1457   void repne_scanw(Register addr, Register value, Register count,
1458                    Register scratch);
1459 
1460   typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1461   typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1462 
1463   // If a constant does not fit in an immediate field, generate some
1464   // number of MOV instructions and then perform the operation
1465   void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
1466                              add_sub_imm_insn insn1,
1467                              add_sub_reg_insn insn2, bool is32);
1468   // Separate vsn which sets the flags
1469   void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
1470                                add_sub_imm_insn insn1,
1471                                add_sub_reg_insn insn2, bool is32);
1472 
1473 #define WRAP(INSN, is32)                                                \
1474   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1475     wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1476   }                                                                     \
1477                                                                         \
1478   void INSN(Register Rd, Register Rn, Register Rm,                      \
1479              enum shift_kind kind, unsigned shift = 0) {                \
1480     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1481   }                                                                     \
1482                                                                         \
1483   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1484     Assembler::INSN(Rd, Rn, Rm);                                        \
1485   }                                                                     \
1486                                                                         \
1487   void INSN(Register Rd, Register Rn, Register Rm,                      \
1488            ext::operation option, int amount = 0) {                     \
1489     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1490   }
1491 
1492   WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true)
1493 
1494 #undef WRAP
1495 #define WRAP(INSN, is32)                                                \
1496   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1497     wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1498   }                                                                     \
1499                                                                         \
1500   void INSN(Register Rd, Register Rn, Register Rm,                      \
1501              enum shift_kind kind, unsigned shift = 0) {                \
1502     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1503   }                                                                     \
1504                                                                         \
1505   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1506     Assembler::INSN(Rd, Rn, Rm);                                        \
1507   }                                                                     \
1508                                                                         \
1509   void INSN(Register Rd, Register Rn, Register Rm,                      \
1510            ext::operation option, int amount = 0) {                     \
1511     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1512   }
1513 
1514   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1515 
1516   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1517   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1518   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1519   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1520 
1521   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1522 
1523   void verified_entry(Compile* C, int sp_inc);
1524 
1525   // Inline type specific methods
1526   #include "asm/macroAssembler_common.hpp"
1527 
1528   void save_stack_increment(int sp_inc, int frame_size);
1529 
1530   void tableswitch(Register index, jint lowbound, jint highbound,
1531                    Label &jumptable, Label &jumptable_end, int stride = 1) {
1532     adr(rscratch1, jumptable);
1533     subsw(rscratch2, index, lowbound);
1534     subsw(zr, rscratch2, highbound - lowbound);
1535     br(Assembler::HS, jumptable_end);
1536     add(rscratch1, rscratch1, rscratch2,
1537         ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1538     br(rscratch1);
1539   }
1540 
1541   // Form an address from base + offset in Rd.  Rd may or may not
1542   // actually be used: you must use the Address that is returned.  It
1543   // is up to you to ensure that the shift provided matches the size
1544   // of your data.
1545   Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1546 
1547   // Return true iff an address is within the 48-bit AArch64 address
1548   // space.
1549   bool is_valid_AArch64_address(address a) {
1550     return ((uint64_t)a >> 48) == 0;
1551   }
1552 
1553   // Load the base of the cardtable byte map into reg.
1554   void load_byte_map_base(Register reg);
1555 
1556   // Load a constant address in the AOT Runtime Constants area
1557   void load_aotrc_address(Register reg, address a);
1558 
1559   // Prolog generator routines to support switch between x86 code and
1560   // generated ARM code
1561 
1562   // routine to generate an x86 prolog for a stub function which
1563   // bootstraps into the generated ARM code which directly follows the
1564   // stub
1565   //
1566 
1567   public:
1568 
1569   address read_polling_page(Register r, relocInfo::relocType rtype);
1570   void get_polling_page(Register dest, relocInfo::relocType rtype);
1571 
1572   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1573   void update_byte_crc32(Register crc, Register val, Register table);
1574   void update_word_crc32(Register crc, Register v, Register tmp,
1575         Register table0, Register table1, Register table2, Register table3,
1576         bool upper = false);
1577 
1578   address count_positives(Register ary1, Register len, Register result);
1579 
1580   address arrays_equals(Register a1, Register a2, Register result, Register cnt1,
1581                         Register tmp1, Register tmp2, Register tmp3, int elem_size);
1582 
1583 // Ensure that the inline code and the stub use the same registers.
1584 #define ARRAYS_HASHCODE_REGISTERS \
1585   do {                      \
1586     assert(result == r0  && \
1587            ary    == r1  && \
1588            cnt    == r2  && \
1589            vdata0 == v3  && \
1590            vdata1 == v2  && \
1591            vdata2 == v1  && \
1592            vdata3 == v0  && \
1593            vmul0  == v4  && \
1594            vmul1  == v5  && \
1595            vmul2  == v6  && \
1596            vmul3  == v7  && \
1597            vpow   == v12 && \
1598            vpowm  == v13, "registers must match aarch64.ad"); \
1599   } while (0)
1600 
1601   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1602 
1603   void fill_words(Register base, Register cnt, Register value);
1604   void fill_words(Register base, uint64_t cnt, Register value);
1605 
1606   address zero_words(Register base, uint64_t cnt);
1607   address zero_words(Register ptr, Register cnt);
1608   void zero_dcache_blocks(Register base, Register cnt);
1609 
1610   static const int zero_words_block_size;
1611 
1612   address byte_array_inflate(Register src, Register dst, Register len,
1613                              FloatRegister vtmp1, FloatRegister vtmp2,
1614                              FloatRegister vtmp3, Register tmp4);
1615 
1616   void char_array_compress(Register src, Register dst, Register len,
1617                            Register res,
1618                            FloatRegister vtmp0, FloatRegister vtmp1,
1619                            FloatRegister vtmp2, FloatRegister vtmp3,
1620                            FloatRegister vtmp4, FloatRegister vtmp5);
1621 
1622   void encode_iso_array(Register src, Register dst,
1623                         Register len, Register res, bool ascii,
1624                         FloatRegister vtmp0, FloatRegister vtmp1,
1625                         FloatRegister vtmp2, FloatRegister vtmp3,
1626                         FloatRegister vtmp4, FloatRegister vtmp5);
1627 
1628   void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi,
1629       address pio2, address dsin_coef, address dcos_coef);
1630  private:
1631   // begin trigonometric functions support block
1632   void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2);
1633   void generate__kernel_rem_pio2(address two_over_pi, address pio2);
1634   void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef);
1635   void generate_kernel_cos(FloatRegister x, address dcos_coef);
1636   // end trigonometric functions support block
1637   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1638                        Register src1, Register src2);
1639   void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1640     add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1641   }
1642   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1643                              Register y, Register y_idx, Register z,
1644                              Register carry, Register product,
1645                              Register idx, Register kdx);
1646   void multiply_128_x_128_loop(Register y, Register z,
1647                                Register carry, Register carry2,
1648                                Register idx, Register jdx,
1649                                Register yz_idx1, Register yz_idx2,
1650                                Register tmp, Register tmp3, Register tmp4,
1651                                Register tmp7, Register product_hi);
1652   void kernel_crc32_using_crypto_pmull(Register crc, Register buf,
1653         Register len, Register tmp0, Register tmp1, Register tmp2,
1654         Register tmp3);
1655   void kernel_crc32_using_crc32(Register crc, Register buf,
1656         Register len, Register tmp0, Register tmp1, Register tmp2,
1657         Register tmp3);
1658   void kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
1659         Register len, Register tmp0, Register tmp1, Register tmp2,
1660         Register tmp3);
1661   void kernel_crc32c_using_crc32c(Register crc, Register buf,
1662         Register len, Register tmp0, Register tmp1, Register tmp2,
1663         Register tmp3);
1664   void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
1665         Register len, Register tmp0, Register tmp1, Register tmp2,
1666         size_t table_offset);
1667 
1668   void ghash_modmul (FloatRegister result,
1669                      FloatRegister result_lo, FloatRegister result_hi, FloatRegister b,
1670                      FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p,
1671                      FloatRegister t1, FloatRegister t2, FloatRegister t3);
1672   void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
1673 public:
1674   void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1675                        Register tmp0, Register tmp1, Register tmp2, Register tmp3,
1676                        Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1677   void mul_add(Register out, Register in, Register offs, Register len, Register k);
1678   void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,
1679                       FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1680                       FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1681   void ghash_multiply_wide(int index,
1682                            FloatRegister result_lo, FloatRegister result_hi,
1683                            FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1684                            FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1685   void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi,
1686                     FloatRegister p, FloatRegister z, FloatRegister t1);
1687   void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi,
1688                     FloatRegister p, FloatRegister z, FloatRegister t1);
1689   void ghash_processBlocks_wide(Label& p, Register state, Register subkeyH,
1690                                 Register data, Register blocks, int unrolls);
1691 
1692 
1693   void aesenc_loadkeys(Register key, Register keylen);
1694   void aesecb_encrypt(Register from, Register to, Register keylen,
1695                       FloatRegister data = v0, int unrolls = 1);
1696   void aesecb_decrypt(Register from, Register to, Register key, Register keylen);
1697   void aes_round(FloatRegister input, FloatRegister subkey);
1698 
1699   // ChaCha20 functions support block
1700   void cc20_qr_add4(FloatRegister (&addFirst)[4],
1701           FloatRegister (&addSecond)[4]);
1702   void cc20_qr_xor4(FloatRegister (&firstElem)[4],
1703           FloatRegister (&secondElem)[4], FloatRegister (&result)[4]);
1704   void cc20_qr_lrot4(FloatRegister (&sourceReg)[4],
1705           FloatRegister (&destReg)[4], int bits, FloatRegister table);
1706   void cc20_set_qr_registers(FloatRegister (&vectorSet)[4],
1707           const FloatRegister (&stateVectors)[16], int idx1, int idx2,
1708           int idx3, int idx4);
1709 
1710   // Rotate using ORR (for identity) or USHR + SLI.
1711   void neon_vector_rotate(FloatRegister dst, SIMD_Arrangement T,
1712                           FloatRegister src, int shift_amount);
1713 
1714   // Place an ISB after code may have been modified due to a safepoint.
1715   void safepoint_isb();
1716 
1717 private:
1718   // Return the effective address r + (r1 << ext) + offset.
1719   // Uses rscratch2.
1720   Address offsetted_address(Register r, Register r1, Address::extend ext,
1721                             int offset, int size);
1722 
1723 private:
1724   // Returns an address on the stack which is reachable with a ldr/str of size
1725   // Uses rscratch2 if the address is not directly reachable
1726   Address spill_address(int size, int offset, Register tmp=rscratch2);
1727   Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2);
1728 
1729   bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const;
1730 
1731   // Check whether two loads/stores can be merged into ldp/stp.
1732   bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const;
1733 
1734   // Merge current load/store with previous load/store into ldp/stp.
1735   void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1736 
1737   // Try to merge two loads/stores into ldp/stp. If success, returns true else false.
1738   bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1739 
1740 public:
1741   void spill(Register Rx, bool is64, int offset) {
1742     if (is64) {
1743       str(Rx, spill_address(8, offset));
1744     } else {
1745       strw(Rx, spill_address(4, offset));
1746     }
1747   }
1748   void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1749     str(Vx, T, spill_address(1 << (int)T, offset));
1750   }
1751 
1752   void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1753     sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1754   }
1755   void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1756     sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1757   }
1758 
1759   void unspill(Register Rx, bool is64, int offset) {
1760     if (is64) {
1761       ldr(Rx, spill_address(8, offset));
1762     } else {
1763       ldrw(Rx, spill_address(4, offset));
1764     }
1765   }
1766   void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1767     ldr(Vx, T, spill_address(1 << (int)T, offset));
1768   }
1769 
1770   void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1771     sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1772   }
1773   void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1774     sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1775   }
1776 
1777   void spill_copy128(int src_offset, int dst_offset,
1778                      Register tmp1=rscratch1, Register tmp2=rscratch2) {
1779     if (src_offset < 512 && (src_offset & 7) == 0 &&
1780         dst_offset < 512 && (dst_offset & 7) == 0) {
1781       ldp(tmp1, tmp2, Address(sp, src_offset));
1782       stp(tmp1, tmp2, Address(sp, dst_offset));
1783     } else {
1784       unspill(tmp1, true, src_offset);
1785       spill(tmp1, true, dst_offset);
1786       unspill(tmp1, true, src_offset+8);
1787       spill(tmp1, true, dst_offset+8);
1788     }
1789   }
1790   void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset,
1791                                             int sve_vec_reg_size_in_bytes) {
1792     assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size");
1793     for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) {
1794       spill_copy128(src_offset, dst_offset);
1795       src_offset += 16;
1796       dst_offset += 16;
1797     }
1798   }
1799   void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset,
1800                                                int sve_predicate_reg_size_in_bytes) {
1801     sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset));
1802     sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset));
1803     reinitialize_ptrue();
1804   }
1805   void cache_wb(Address line);
1806   void cache_wbsync(bool is_pre);
1807 
1808   // Code for java.lang.Thread::onSpinWait() intrinsic.
1809   void spin_wait();
1810   void spin_wait_wfet(int delay_ns);
1811 
1812   void fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow);
1813   void fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1814 
1815 private:
1816   // Check the current thread doesn't need a cross modify fence.
1817   void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
1818   void try_to_replace_prev_vector_copy_with_movprfx(FloatRegister dst);
1819 
1820 public:
1821   void maybe_movprfx(FloatRegister dst, FloatRegister src) {
1822     if (dst != src) {
1823       sve_movprfx(dst, src);
1824     }
1825   }
1826 
1827 // Wrappers for SVE explicit destructive instructions, overriding the
1828 // same-signature Assembler entry points to enable movprfx fusion optimization.
1829 //
1830 // Implicit destructive instructions (e.g. predicated unary ops like sve_abs/
1831 // sve_neg/sve_not, whose ISA encoding allows Zd != Zn but whose use as a Java
1832 // Vector API masked operation requires pass-through of the first source) are
1833 // not covered here. For those, the .ad file is responsible for emitting
1834 // movprfx explicitly via maybe_movprfx() before the destructive op.
1835 #define SVE_DESTRUCTIVE_BINARY_INS(NAME)                                       \
1836   using Assembler::NAME;                                                       \
1837   void NAME(FloatRegister Zd, SIMD_RegVariant T, PRegister Pg,                 \
1838             FloatRegister Zm) {                                                \
1839     if (Zd != Zm) {                                                            \
1840       try_to_replace_prev_vector_copy_with_movprfx(Zd);                        \
1841     }                                                                          \
1842     Assembler::NAME(Zd, T, Pg, Zm);                                            \
1843   }
1844 
1845 #define SVE_DESTRUCTIVE_BINARY_5(I1, I2, I3, I4, I5)                           \
1846   SVE_DESTRUCTIVE_BINARY_INS(I1); SVE_DESTRUCTIVE_BINARY_INS(I2);              \
1847   SVE_DESTRUCTIVE_BINARY_INS(I3); SVE_DESTRUCTIVE_BINARY_INS(I4);              \
1848   SVE_DESTRUCTIVE_BINARY_INS(I5);
1849 
1850   SVE_DESTRUCTIVE_BINARY_5(sve_add,  sve_and,   sve_asr,   sve_bic,   sve_eor)
1851   SVE_DESTRUCTIVE_BINARY_5(sve_fabd, sve_fadd,  sve_fdiv,  sve_fmax,  sve_fmin)
1852   SVE_DESTRUCTIVE_BINARY_5(sve_fmul, sve_fsub,  sve_lsl,   sve_lsr,   sve_mul)
1853   SVE_DESTRUCTIVE_BINARY_5(sve_orr,  sve_smax,  sve_smin,  sve_sqadd, sve_sqsub)
1854   SVE_DESTRUCTIVE_BINARY_5(sve_sub,  sve_uqadd, sve_uqsub, sve_umax,  sve_umin)
1855 
1856 #undef SVE_DESTRUCTIVE_BINARY_INS
1857 #undef SVE_DESTRUCTIVE_BINARY_5
1858 
1859 #define SVE_DESTRUCTIVE_SHIFT_IMM_INS(NAME)                                    \
1860   void NAME(FloatRegister Zd, SIMD_RegVariant T, PRegister Pg, int shift) {    \
1861     try_to_replace_prev_vector_copy_with_movprfx(Zd);                          \
1862     Assembler::NAME(Zd, T, Pg, shift);                                         \
1863   }
1864 
1865   SVE_DESTRUCTIVE_SHIFT_IMM_INS(sve_asr);
1866   SVE_DESTRUCTIVE_SHIFT_IMM_INS(sve_lsl);
1867   SVE_DESTRUCTIVE_SHIFT_IMM_INS(sve_lsr);
1868 
1869 #undef SVE_DESTRUCTIVE_SHIFT_IMM_INS
1870 
1871 #define SVE_DESTRUCTIVE_UNPRED_IMM_INS(NAME, IMM_TYPE)                         \
1872   void NAME(FloatRegister Zd, SIMD_RegVariant T, IMM_TYPE imm) {               \
1873     try_to_replace_prev_vector_copy_with_movprfx(Zd);                          \
1874     Assembler::NAME(Zd, T, imm);                                               \
1875   }
1876 
1877   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_add, unsigned);
1878   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_sub, unsigned);
1879   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_and, uint64_t);
1880   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_eor, uint64_t);
1881   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_orr, uint64_t);
1882 
1883 #undef SVE_DESTRUCTIVE_UNPRED_IMM_INS
1884 
1885 #define SVE_DESTRUCTIVE_TERNARY_INS(NAME)                                      \
1886   using Assembler::NAME;                                                       \
1887   void NAME(FloatRegister Zd, SIMD_RegVariant T, PRegister Pg,                 \
1888             FloatRegister Zn, FloatRegister Zm) {                              \
1889     if (Zd != Zn && Zd != Zm) {                                                \
1890       try_to_replace_prev_vector_copy_with_movprfx(Zd);                        \
1891     }                                                                          \
1892     Assembler::NAME(Zd, T, Pg, Zn, Zm);                                        \
1893   }
1894 
1895   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmad);
1896   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmla);
1897   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmls);
1898   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmsb);
1899   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmad);
1900   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmla);
1901   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmls);
1902   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmsb);
1903   SVE_DESTRUCTIVE_TERNARY_INS(sve_mla);
1904   SVE_DESTRUCTIVE_TERNARY_INS(sve_mls);
1905 
1906 #undef SVE_DESTRUCTIVE_TERNARY_INS
1907 
1908 #define SVE_DESTRUCTIVE_TERNARY_UNPRED_INS(NAME)                               \
1909   using Assembler::NAME;                                                       \
1910   void NAME(FloatRegister Zd, FloatRegister Zm, FloatRegister Zk) {            \
1911     if (Zd != Zm && Zd != Zk) {                                                \
1912       try_to_replace_prev_vector_copy_with_movprfx(Zd);                        \
1913     }                                                                          \
1914     Assembler::NAME(Zd, Zm, Zk);                                               \
1915   }
1916 
1917   SVE_DESTRUCTIVE_TERNARY_UNPRED_INS(sve_bsl);
1918   SVE_DESTRUCTIVE_TERNARY_UNPRED_INS(sve_eor3);
1919 
1920 #undef SVE_DESTRUCTIVE_TERNARY_UNPRED_INS
1921 };
1922 
1923 #ifdef ASSERT
1924 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1925 #endif
1926 
1927 struct tableswitch {
1928   Register _reg;
1929   int _insn_index; jint _first_key; jint _last_key;
1930   Label _after;
1931   Label _branches;
1932 };
1933 
1934 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP