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