1 /*
   2  * Copyright (c) 1997, 2025, 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/powerOfTwo.hpp"
  37 
  38 class OopMap;
  39 
  40 // MacroAssembler extends Assembler by frequently used macros.
  41 //
  42 // Instructions for which a 'better' code sequence exists depending
  43 // on arguments should also go in here.
  44 
  45 class MacroAssembler: public Assembler {
  46   friend class LIR_Assembler;
  47 
  48  public:
  49   using Assembler::mov;
  50   using Assembler::movi;
  51 
  52  protected:
  53 
  54   // Support for VM calls
  55   //
  56   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
  57   // may customize this version by overriding it for its purposes (e.g., to save/restore
  58   // additional registers when doing a VM call).
  59   virtual void call_VM_leaf_base(
  60     address entry_point,               // the entry point
  61     int     number_of_arguments,        // the number of arguments to pop after the call
  62     Label *retaddr = nullptr
  63   );
  64 
  65   virtual void call_VM_leaf_base(
  66     address entry_point,               // the entry point
  67     int     number_of_arguments,        // the number of arguments to pop after the call
  68     Label &retaddr) {
  69     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
  70   }
  71 
  72   // This is the base routine called by the different versions of call_VM. The interpreter
  73   // may customize this version by overriding it for its purposes (e.g., to save/restore
  74   // additional registers when doing a VM call).
  75   //
  76   // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
  77   // returns the register which contains the thread upon return. If a thread register has been
  78   // specified, the return value will correspond to that register. If no last_java_sp is specified
  79   // (noreg) than rsp will be used instead.
  80   virtual void call_VM_base(           // returns the register containing the thread upon return
  81     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
  82     Register java_thread,              // the thread if computed before     ; use noreg otherwise
  83     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
  84     Label*   return_pc,                // to set up last_Java_frame; use nullptr otherwise
  85     address  entry_point,              // the entry point
  86     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
  87     bool     check_exceptions          // whether to check for pending exceptions after return
  88   );
  89 
  90   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
  91 
  92   enum KlassDecodeMode {
  93     KlassDecodeNone,
  94     KlassDecodeZero,
  95     KlassDecodeXor,
  96     KlassDecodeMovk
  97   };
  98 
  99   // Calculate decoding mode based on given parameters, used for checking then ultimately setting.
 100   static KlassDecodeMode klass_decode_mode(address base, int shift, const size_t range);
 101 
 102  private:
 103   static KlassDecodeMode _klass_decode_mode;
 104 
 105   // Returns above setting with asserts
 106   static KlassDecodeMode klass_decode_mode();
 107 
 108  public:
 109   // Checks the decode mode and returns false if not compatible with preferred decoding mode.
 110   static bool check_klass_decode_mode(address base, int shift, const size_t range);
 111 
 112   // Sets the decode mode and returns false if cannot be set.
 113   static bool set_klass_decode_mode(address base, int shift, const size_t range);
 114 
 115  public:
 116   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 117 
 118  // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 119  // The implementation is only non-empty for the InterpreterMacroAssembler,
 120  // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 121  virtual void check_and_handle_popframe(Register java_thread);
 122  virtual void check_and_handle_earlyret(Register java_thread);
 123 
 124   void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp = rscratch1);
 125   void rt_call(address dest, Register tmp = rscratch1);
 126 
 127   // Load Effective Address
 128   void lea(Register r, const Address &a) {
 129     InstructionMark im(this);
 130     a.lea(this, r);
 131   }
 132 
 133   // Whether materializing the given address for a LDR/STR requires an
 134   // additional lea instruction.
 135   static bool legitimize_address_requires_lea(const Address &a, int size) {
 136     return a.getMode() == Address::base_plus_offset &&
 137            !Address::offset_ok_for_immed(a.offset(), exact_log2(size));
 138   }
 139 
 140   /* Sometimes we get misaligned loads and stores, usually from Unsafe
 141      accesses, and these can exceed the offset range. */
 142   Address legitimize_address(const Address &a, int size, Register scratch) {
 143     if (legitimize_address_requires_lea(a, size)) {
 144       block_comment("legitimize_address {");
 145       lea(scratch, a);
 146       block_comment("} legitimize_address");
 147       return Address(scratch);
 148     }
 149     return a;
 150   }
 151 
 152   void addmw(Address a, Register incr, Register scratch) {
 153     ldrw(scratch, a);
 154     addw(scratch, scratch, incr);
 155     strw(scratch, a);
 156   }
 157 
 158   // Add constant to memory word
 159   void addmw(Address a, int imm, Register scratch) {
 160     ldrw(scratch, a);
 161     if (imm > 0)
 162       addw(scratch, scratch, (unsigned)imm);
 163     else
 164       subw(scratch, scratch, (unsigned)-imm);
 165     strw(scratch, a);
 166   }
 167 
 168   void bind(Label& L) {
 169     Assembler::bind(L);
 170     code()->clear_last_insn();
 171     code()->set_last_label(pc());
 172   }
 173 
 174   void membar(Membar_mask_bits order_constraint);
 175 
 176   using Assembler::ldr;
 177   using Assembler::str;
 178   using Assembler::ldrw;
 179   using Assembler::strw;
 180 
 181   void ldr(Register Rx, const Address &adr);
 182   void ldrw(Register Rw, const Address &adr);
 183   void str(Register Rx, const Address &adr);
 184   void strw(Register Rx, const Address &adr);
 185 
 186   // Frame creation and destruction shared between JITs.
 187   void build_frame(int framesize);
 188   void remove_frame(int framesize);
 189 
 190   virtual void _call_Unimplemented(address call_site) {
 191     mov(rscratch2, call_site);
 192   }
 193 
 194 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 195 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 196 // https://reviews.llvm.org/D3311
 197 
 198 #ifdef _WIN64
 199 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 200 #else
 201 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 202 #endif
 203 
 204   // aliases defined in AARCH64 spec
 205 
 206   template<class T>
 207   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }
 208 
 209   inline void cmp(Register Rd, unsigned char imm8)  { subs(zr, Rd, imm8); }
 210   inline void cmp(Register Rd, unsigned imm) = delete;
 211 
 212   template<class T>
 213   inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); }
 214 
 215   inline void cmn(Register Rd, unsigned char imm8)  { adds(zr, Rd, imm8); }
 216   inline void cmn(Register Rd, unsigned imm) = delete;
 217 
 218   void cset(Register Rd, Assembler::Condition cond) {
 219     csinc(Rd, zr, zr, ~cond);
 220   }
 221   void csetw(Register Rd, Assembler::Condition cond) {
 222     csincw(Rd, zr, zr, ~cond);
 223   }
 224 
 225   void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
 226     csneg(Rd, Rn, Rn, ~cond);
 227   }
 228   void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
 229     csnegw(Rd, Rn, Rn, ~cond);
 230   }
 231 
 232   inline void movw(Register Rd, Register Rn) {
 233     if (Rd == sp || Rn == sp) {
 234       Assembler::addw(Rd, Rn, 0U);
 235     } else {
 236       orrw(Rd, zr, Rn);
 237     }
 238   }
 239   inline void mov(Register Rd, Register Rn) {
 240     assert(Rd != r31_sp && Rn != r31_sp, "should be");
 241     if (Rd == Rn) {
 242     } else if (Rd == sp || Rn == sp) {
 243       Assembler::add(Rd, Rn, 0U);
 244     } else {
 245       orr(Rd, zr, Rn);
 246     }
 247   }
 248 
 249   inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
 250   inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
 251 
 252   inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
 253   inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
 254 
 255   inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
 256   inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
 257 
 258   inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 259     bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 260   }
 261   inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 262     bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 263   }
 264 
 265   inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 266     bfmw(Rd, Rn, lsb, (lsb + width - 1));
 267   }
 268   inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 269     bfm(Rd, Rn, lsb , (lsb + width - 1));
 270   }
 271 
 272   inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 273     sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 274   }
 275   inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 276     sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 277   }
 278 
 279   inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 280     sbfmw(Rd, Rn, lsb, (lsb + width - 1));
 281   }
 282   inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 283     sbfm(Rd, Rn, lsb , (lsb + width - 1));
 284   }
 285 
 286   inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 287     ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 288   }
 289   inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 290     ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 291   }
 292 
 293   inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 294     ubfmw(Rd, Rn, lsb, (lsb + width - 1));
 295   }
 296   inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 297     ubfm(Rd, Rn, lsb , (lsb + width - 1));
 298   }
 299 
 300   inline void asrw(Register Rd, Register Rn, unsigned imm) {
 301     sbfmw(Rd, Rn, imm, 31);
 302   }
 303 
 304   inline void asr(Register Rd, Register Rn, unsigned imm) {
 305     sbfm(Rd, Rn, imm, 63);
 306   }
 307 
 308   inline void lslw(Register Rd, Register Rn, unsigned imm) {
 309     ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
 310   }
 311 
 312   inline void lsl(Register Rd, Register Rn, unsigned imm) {
 313     ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
 314   }
 315 
 316   inline void lsrw(Register Rd, Register Rn, unsigned imm) {
 317     ubfmw(Rd, Rn, imm, 31);
 318   }
 319 
 320   inline void lsr(Register Rd, Register Rn, unsigned imm) {
 321     ubfm(Rd, Rn, imm, 63);
 322   }
 323 
 324   inline void rorw(Register Rd, Register Rn, unsigned imm) {
 325     extrw(Rd, Rn, Rn, imm);
 326   }
 327 
 328   inline void ror(Register Rd, Register Rn, unsigned imm) {
 329     extr(Rd, Rn, Rn, imm);
 330   }
 331 
 332   inline void rolw(Register Rd, Register Rn, unsigned imm) {
 333     extrw(Rd, Rn, Rn, (32 - imm));
 334   }
 335 
 336   inline void rol(Register Rd, Register Rn, unsigned imm) {
 337     extr(Rd, Rn, Rn, (64 - imm));
 338   }
 339 
 340   using Assembler::rax1;
 341   using Assembler::eor3;
 342 
 343   inline void rax1(Register Rd, Register Rn, Register Rm) {
 344     eor(Rd, Rn, Rm, ROR, 63); // Rd = Rn ^ rol(Rm, 1)
 345   }
 346 
 347   inline void eor3(Register Rd, Register Rn, Register Rm, Register Rk) {
 348     assert(Rd != Rn, "Use tmp register");
 349     eor(Rd, Rm, Rk);
 350     eor(Rd, Rd, Rn);
 351   }
 352 
 353   inline void sxtbw(Register Rd, Register Rn) {
 354     sbfmw(Rd, Rn, 0, 7);
 355   }
 356   inline void sxthw(Register Rd, Register Rn) {
 357     sbfmw(Rd, Rn, 0, 15);
 358   }
 359   inline void sxtb(Register Rd, Register Rn) {
 360     sbfm(Rd, Rn, 0, 7);
 361   }
 362   inline void sxth(Register Rd, Register Rn) {
 363     sbfm(Rd, Rn, 0, 15);
 364   }
 365   inline void sxtw(Register Rd, Register Rn) {
 366     sbfm(Rd, Rn, 0, 31);
 367   }
 368 
 369   inline void uxtbw(Register Rd, Register Rn) {
 370     ubfmw(Rd, Rn, 0, 7);
 371   }
 372   inline void uxthw(Register Rd, Register Rn) {
 373     ubfmw(Rd, Rn, 0, 15);
 374   }
 375   inline void uxtb(Register Rd, Register Rn) {
 376     ubfm(Rd, Rn, 0, 7);
 377   }
 378   inline void uxth(Register Rd, Register Rn) {
 379     ubfm(Rd, Rn, 0, 15);
 380   }
 381   inline void uxtw(Register Rd, Register Rn) {
 382     ubfm(Rd, Rn, 0, 31);
 383   }
 384 
 385   inline void cmnw(Register Rn, Register Rm) {
 386     addsw(zr, Rn, Rm);
 387   }
 388   inline void cmn(Register Rn, Register Rm) {
 389     adds(zr, Rn, Rm);
 390   }
 391 
 392   inline void cmpw(Register Rn, Register Rm) {
 393     subsw(zr, Rn, Rm);
 394   }
 395   inline void cmp(Register Rn, Register Rm) {
 396     subs(zr, Rn, Rm);
 397   }
 398 
 399   inline void negw(Register Rd, Register Rn) {
 400     subw(Rd, zr, Rn);
 401   }
 402 
 403   inline void neg(Register Rd, Register Rn) {
 404     sub(Rd, zr, Rn);
 405   }
 406 
 407   inline void negsw(Register Rd, Register Rn) {
 408     subsw(Rd, zr, Rn);
 409   }
 410 
 411   inline void negs(Register Rd, Register Rn) {
 412     subs(Rd, zr, Rn);
 413   }
 414 
 415   inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 416     addsw(zr, Rn, Rm, kind, shift);
 417   }
 418   inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 419     adds(zr, Rn, Rm, kind, shift);
 420   }
 421 
 422   inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 423     subsw(zr, Rn, Rm, kind, shift);
 424   }
 425   inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 426     subs(zr, Rn, Rm, kind, shift);
 427   }
 428 
 429   inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 430     subw(Rd, zr, Rn, kind, shift);
 431   }
 432 
 433   inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 434     sub(Rd, zr, Rn, kind, shift);
 435   }
 436 
 437   inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 438     subsw(Rd, zr, Rn, kind, shift);
 439   }
 440 
 441   inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 442     subs(Rd, zr, Rn, kind, shift);
 443   }
 444 
 445   inline void mnegw(Register Rd, Register Rn, Register Rm) {
 446     msubw(Rd, Rn, Rm, zr);
 447   }
 448   inline void mneg(Register Rd, Register Rn, Register Rm) {
 449     msub(Rd, Rn, Rm, zr);
 450   }
 451 
 452   inline void mulw(Register Rd, Register Rn, Register Rm) {
 453     maddw(Rd, Rn, Rm, zr);
 454   }
 455   inline void mul(Register Rd, Register Rn, Register Rm) {
 456     madd(Rd, Rn, Rm, zr);
 457   }
 458 
 459   inline void smnegl(Register Rd, Register Rn, Register Rm) {
 460     smsubl(Rd, Rn, Rm, zr);
 461   }
 462   inline void smull(Register Rd, Register Rn, Register Rm) {
 463     smaddl(Rd, Rn, Rm, zr);
 464   }
 465 
 466   inline void umnegl(Register Rd, Register Rn, Register Rm) {
 467     umsubl(Rd, Rn, Rm, zr);
 468   }
 469   inline void umull(Register Rd, Register Rn, Register Rm) {
 470     umaddl(Rd, Rn, Rm, zr);
 471   }
 472 
 473 #define WRAP(INSN)                                                            \
 474   void INSN(Register Rd, Register Rn, Register Rm, Register Ra) {             \
 475     if (VM_Version::supports_a53mac() && Ra != zr)                            \
 476       nop();                                                                  \
 477     Assembler::INSN(Rd, Rn, Rm, Ra);                                          \
 478   }
 479 
 480   WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
 481   WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
 482 #undef WRAP
 483 
 484 
 485   // macro assembly operations needed for aarch64
 486 
 487 public:
 488 
 489   enum FpPushPopMode {
 490     PushPopFull,
 491     PushPopSVE,
 492     PushPopNeon,
 493     PushPopFp
 494   };
 495 
 496   // first two private routines for loading 32 bit or 64 bit constants
 497 private:
 498 
 499   void mov_immediate64(Register dst, uint64_t imm64);
 500   void mov_immediate32(Register dst, uint32_t imm32);
 501 
 502   int push(unsigned int bitset, Register stack);
 503   int pop(unsigned int bitset, Register stack);
 504 
 505   int push_fp(unsigned int bitset, Register stack, FpPushPopMode mode);
 506   int pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode);
 507 
 508   int push_p(unsigned int bitset, Register stack);
 509   int pop_p(unsigned int bitset, Register stack);
 510 
 511   void mov(Register dst, Address a);
 512 
 513 public:
 514 
 515   void push(RegSet regs, Register stack) { if (regs.bits()) push(regs.bits(), stack); }
 516   void pop(RegSet regs, Register stack) { if (regs.bits()) pop(regs.bits(), stack); }
 517 
 518   void push_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) push_fp(regs.bits(), stack, mode); }
 519   void pop_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) pop_fp(regs.bits(), stack, mode); }
 520 
 521   static RegSet call_clobbered_gp_registers();
 522 
 523   void push_p(PRegSet regs, Register stack) { if (regs.bits()) push_p(regs.bits(), stack); }
 524   void pop_p(PRegSet regs, Register stack) { if (regs.bits()) pop_p(regs.bits(), stack); }
 525 
 526   // Push and pop everything that might be clobbered by a native
 527   // runtime call except rscratch1 and rscratch2.  (They are always
 528   // scratch, so we don't have to protect them.)  Only save the lower
 529   // 64 bits of each vector register. Additional registers can be excluded
 530   // in a passed RegSet.
 531   void push_call_clobbered_registers_except(RegSet exclude);
 532   void pop_call_clobbered_registers_except(RegSet exclude);
 533 
 534   void push_call_clobbered_registers() {
 535     push_call_clobbered_registers_except(RegSet());
 536   }
 537   void pop_call_clobbered_registers() {
 538     pop_call_clobbered_registers_except(RegSet());
 539   }
 540 
 541 
 542   // now mov instructions for loading absolute addresses and 32 or
 543   // 64 bit integers
 544 
 545   inline void mov(Register dst, address addr)             { mov_immediate64(dst, (uint64_t)addr); }
 546 
 547   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 548   inline void mov(Register dst, T o)                      { mov_immediate64(dst, (uint64_t)o); }
 549 
 550   inline void movw(Register dst, uint32_t imm32)          { mov_immediate32(dst, imm32); }
 551 
 552   void mov(Register dst, RegisterOrConstant src) {
 553     if (src.is_register())
 554       mov(dst, src.as_register());
 555     else
 556       mov(dst, src.as_constant());
 557   }
 558 
 559   void movptr(Register r, uintptr_t imm64);
 560 
 561   void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64);
 562 
 563   void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
 564     orr(Vd, T, Vn, Vn);
 565   }
 566 
 567   void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) {
 568     fcvtsh(tmp, src);
 569     smov(dst, tmp, H, 0);
 570   }
 571 
 572   void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) {
 573     mov(tmp, H, 0, src);
 574     fcvths(dst, tmp);
 575   }
 576 
 577   // Generalized Test Bit And Branch, including a "far" variety which
 578   // spans more than 32KiB.
 579   void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) {
 580     assert(cond == EQ || cond == NE, "must be");
 581 
 582     if (isfar)
 583       cond = ~cond;
 584 
 585     void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
 586     if (cond == Assembler::EQ)
 587       branch = &Assembler::tbz;
 588     else
 589       branch = &Assembler::tbnz;
 590 
 591     if (isfar) {
 592       Label L;
 593       (this->*branch)(Rt, bitpos, L);
 594       b(dest);
 595       bind(L);
 596     } else {
 597       (this->*branch)(Rt, bitpos, dest);
 598     }
 599   }
 600 
 601   // macro instructions for accessing and updating floating point
 602   // status register
 603   //
 604   // FPSR : op1 == 011
 605   //        CRn == 0100
 606   //        CRm == 0100
 607   //        op2 == 001
 608 
 609   inline void get_fpsr(Register reg)
 610   {
 611     mrs(0b11, 0b0100, 0b0100, 0b001, reg);
 612   }
 613 
 614   inline void set_fpsr(Register reg)
 615   {
 616     msr(0b011, 0b0100, 0b0100, 0b001, reg);
 617   }
 618 
 619   inline void clear_fpsr()
 620   {
 621     msr(0b011, 0b0100, 0b0100, 0b001, zr);
 622   }
 623 
 624   // FPCR : op1 == 011
 625   //        CRn == 0100
 626   //        CRm == 0100
 627   //        op2 == 000
 628 
 629   inline void get_fpcr(Register reg) {
 630     mrs(0b11, 0b0100, 0b0100, 0b000, reg);
 631   }
 632 
 633   inline void set_fpcr(Register reg) {
 634     msr(0b011, 0b0100, 0b0100, 0b000, reg);
 635   }
 636 
 637   // DCZID_EL0: op1 == 011
 638   //            CRn == 0000
 639   //            CRm == 0000
 640   //            op2 == 111
 641   inline void get_dczid_el0(Register reg)
 642   {
 643     mrs(0b011, 0b0000, 0b0000, 0b111, reg);
 644   }
 645 
 646   // CTR_EL0:   op1 == 011
 647   //            CRn == 0000
 648   //            CRm == 0000
 649   //            op2 == 001
 650   inline void get_ctr_el0(Register reg)
 651   {
 652     mrs(0b011, 0b0000, 0b0000, 0b001, reg);
 653   }
 654 
 655   inline void get_nzcv(Register reg) {
 656     mrs(0b011, 0b0100, 0b0010, 0b000, reg);
 657   }
 658 
 659   inline void set_nzcv(Register reg) {
 660     msr(0b011, 0b0100, 0b0010, 0b000, 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   static address target_addr_for_insn_or_null(address insn_addr);
 682 
 683   // Required platform-specific helpers for Label::patch_instructions.
 684   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 685   static int pd_patch_instruction_size(address branch, address target);
 686   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 687     pd_patch_instruction_size(branch, target);
 688   }
 689   static address pd_call_destination(address branch) {
 690     return target_addr_for_insn(branch);
 691   }
 692 #ifndef PRODUCT
 693   static void pd_print_patched_instruction(address branch);
 694 #endif
 695 
 696   static int patch_oop(address insn_addr, address o);
 697   static int patch_narrow_klass(address insn_addr, narrowKlass n);
 698 
 699   // Return whether code is emitted to a scratch blob.
 700   virtual bool in_scratch_emit_size() {
 701     return false;
 702   }
 703   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
 704   static int max_trampoline_stub_size();
 705   void emit_static_call_stub();
 706   static int static_call_stub_size();
 707 
 708   // The following 4 methods return the offset of the appropriate move instruction
 709 
 710   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 711   int load_unsigned_byte(Register dst, Address src);
 712   int load_unsigned_short(Register dst, Address src);
 713 
 714   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 715   int load_signed_byte(Register dst, Address src);
 716   int load_signed_short(Register dst, Address src);
 717 
 718   int load_signed_byte32(Register dst, Address src);
 719   int load_signed_short32(Register dst, Address src);
 720 
 721   // Support for sign-extension (hi:lo = extend_sign(lo))
 722   void extend_sign(Register hi, Register lo);
 723 
 724   // Load and store values by size and signed-ness
 725   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 726   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 727 
 728   // Support for inc/dec with optimal instruction selection depending on value
 729 
 730   // x86_64 aliases an unqualified register/address increment and
 731   // decrement to call incrementq and decrementq but also supports
 732   // explicitly sized calls to incrementq/decrementq or
 733   // incrementl/decrementl
 734 
 735   // for aarch64 the proper convention would be to use
 736   // increment/decrement for 64 bit operations and
 737   // incrementw/decrementw for 32 bit operations. so when porting
 738   // x86_64 code we can leave calls to increment/decrement as is,
 739   // replace incrementq/decrementq with increment/decrement and
 740   // replace incrementl/decrementl with incrementw/decrementw.
 741 
 742   // n.b. increment/decrement calls with an Address destination will
 743   // need to use a scratch register to load the value to be
 744   // incremented. increment/decrement calls which add or subtract a
 745   // constant value greater than 2^12 will need to use a 2nd scratch
 746   // register to hold the constant. so, a register increment/decrement
 747   // may trash rscratch2 and an address increment/decrement trash
 748   // rscratch and rscratch2
 749 
 750   void decrementw(Address dst, int value = 1);
 751   void decrementw(Register reg, int value = 1);
 752 
 753   void decrement(Register reg, int value = 1);
 754   void decrement(Address dst, int value = 1);
 755 
 756   void incrementw(Address dst, int value = 1);
 757   void incrementw(Register reg, int value = 1);
 758 
 759   void increment(Register reg, int value = 1);
 760   void increment(Address dst, int value = 1);
 761 
 762 
 763   // Alignment
 764   void align(int modulus);
 765   void align(int modulus, int target);
 766 
 767   // nop
 768   void post_call_nop();
 769 
 770   // Stack frame creation/removal
 771   void enter(bool strip_ret_addr = false);
 772   void leave();
 773 
 774   // ROP Protection
 775   void protect_return_address();
 776   void protect_return_address(Register return_reg);
 777   void authenticate_return_address();
 778   void authenticate_return_address(Register return_reg);
 779   void strip_return_address();
 780   void check_return_address(Register return_reg=lr) PRODUCT_RETURN;
 781 
 782   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
 783   // The pointer will be loaded into the thread register.
 784   void get_thread(Register thread);
 785 
 786   // support for argument shuffling
 787   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 788   void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 789   void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 790   void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 791   void object_move(
 792                    OopMap* map,
 793                    int oop_handle_offset,
 794                    int framesize_in_slots,
 795                    VMRegPair src,
 796                    VMRegPair dst,
 797                    bool is_receiver,
 798                    int* receiver_offset);
 799 
 800 
 801   // Support for VM calls
 802   //
 803   // It is imperative that all calls into the VM are handled via the call_VM macros.
 804   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 805   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 806 
 807 
 808   void call_VM(Register oop_result,
 809                address entry_point,
 810                bool check_exceptions = true);
 811   void call_VM(Register oop_result,
 812                address entry_point,
 813                Register arg_1,
 814                bool check_exceptions = true);
 815   void call_VM(Register oop_result,
 816                address entry_point,
 817                Register arg_1, Register arg_2,
 818                bool check_exceptions = true);
 819   void call_VM(Register oop_result,
 820                address entry_point,
 821                Register arg_1, Register arg_2, Register arg_3,
 822                bool check_exceptions = true);
 823 
 824   // Overloadings with last_Java_sp
 825   void call_VM(Register oop_result,
 826                Register last_java_sp,
 827                address entry_point,
 828                int number_of_arguments = 0,
 829                bool check_exceptions = true);
 830   void call_VM(Register oop_result,
 831                Register last_java_sp,
 832                address entry_point,
 833                Register arg_1, bool
 834                check_exceptions = true);
 835   void call_VM(Register oop_result,
 836                Register last_java_sp,
 837                address entry_point,
 838                Register arg_1, Register arg_2,
 839                bool check_exceptions = true);
 840   void call_VM(Register oop_result,
 841                Register last_java_sp,
 842                address entry_point,
 843                Register arg_1, Register arg_2, Register arg_3,
 844                bool check_exceptions = true);
 845 
 846   void get_vm_result_oop(Register oop_result, Register thread);
 847   void get_vm_result_metadata(Register metadata_result, Register thread);
 848 
 849   // These always tightly bind to MacroAssembler::call_VM_base
 850   // bypassing the virtual implementation
 851   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 852   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
 853   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, 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, Register arg_3, 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, Register arg_4, bool check_exceptions = true);
 856 
 857   void call_VM_leaf(address entry_point,
 858                     int number_of_arguments = 0);
 859   void call_VM_leaf(address entry_point,
 860                     Register arg_1);
 861   void call_VM_leaf(address entry_point,
 862                     Register arg_1, Register arg_2);
 863   void call_VM_leaf(address entry_point,
 864                     Register arg_1, Register arg_2, Register arg_3);
 865 
 866   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 867   // bypassing the virtual implementation
 868   void super_call_VM_leaf(address entry_point);
 869   void super_call_VM_leaf(address entry_point, Register arg_1);
 870   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 871   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 872   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 873 
 874   // last Java Frame (fills frame anchor)
 875   void set_last_Java_frame(Register last_java_sp,
 876                            Register last_java_fp,
 877                            address last_java_pc,
 878                            Register scratch);
 879 
 880   void set_last_Java_frame(Register last_java_sp,
 881                            Register last_java_fp,
 882                            Label &last_java_pc,
 883                            Register scratch);
 884 
 885   void set_last_Java_frame(Register last_java_sp,
 886                            Register last_java_fp,
 887                            Register last_java_pc,
 888                            Register scratch);
 889 
 890   void reset_last_Java_frame(Register thread);
 891 
 892   // thread in the default location (rthread)
 893   void reset_last_Java_frame(bool clear_fp);
 894 
 895   // Stores
 896   void store_check(Register obj);                // store check for obj - register is destroyed afterwards
 897   void store_check(Register obj, Address dst);   // same as above, dst is exact store location (reg. is destroyed)
 898 
 899   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 900   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 901 
 902   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 903   void c2bool(Register x);
 904 
 905   void load_method_holder_cld(Register rresult, Register rmethod);
 906   void load_method_holder(Register holder, Register method);
 907 
 908   // oop manipulations
 909   void load_narrow_klass_compact(Register dst, Register src);
 910   void load_klass(Register dst, Register src);
 911   void store_klass(Register dst, Register src);
 912   void cmp_klass(Register obj, Register klass, Register tmp);
 913   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 914 
 915   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 916   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 917   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 918 
 919   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 920                       Register tmp1, Register tmp2);
 921 
 922   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 923                        Register tmp1, Register tmp2, Register tmp3);
 924 
 925   void load_heap_oop(Register dst, Address src, Register tmp1,
 926                      Register tmp2, DecoratorSet decorators = 0);
 927 
 928   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 929                               Register tmp2, DecoratorSet decorators = 0);
 930   void store_heap_oop(Address dst, Register val, Register tmp1,
 931                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 932 
 933   // currently unimplemented
 934   // Used for storing null. All other oop constants should be
 935   // stored using routines that take a jobject.
 936   void store_heap_oop_null(Address dst);
 937 
 938   void store_klass_gap(Register dst, Register src);
 939 
 940   // This dummy is to prevent a call to store_heap_oop from
 941   // converting a zero (like null) into a Register by giving
 942   // the compiler two choices it can't resolve
 943 
 944   void store_heap_oop(Address dst, void* dummy);
 945 
 946   void encode_heap_oop(Register d, Register s);
 947   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
 948   void decode_heap_oop(Register d, Register s);
 949   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 950   void encode_heap_oop_not_null(Register r);
 951   void decode_heap_oop_not_null(Register r);
 952   void encode_heap_oop_not_null(Register dst, Register src);
 953   void decode_heap_oop_not_null(Register dst, Register src);
 954 
 955   void set_narrow_oop(Register dst, jobject obj);
 956 
 957   void decode_klass_not_null_for_aot(Register dst, Register src);
 958   void encode_klass_not_null_for_aot(Register dst, Register src);
 959   void encode_klass_not_null(Register r);
 960   void decode_klass_not_null(Register r);
 961   void encode_klass_not_null(Register dst, Register src);
 962   void decode_klass_not_null(Register dst, Register src);
 963 
 964   void set_narrow_klass(Register dst, Klass* k);
 965 
 966   // if heap base register is used - reinit it with the correct value
 967   void reinit_heapbase();
 968 
 969   DEBUG_ONLY(void verify_heapbase(const char* msg);)
 970 
 971   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
 972                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
 973   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
 974                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
 975 
 976   void push_cont_fastpath(Register java_thread = rthread);
 977   void pop_cont_fastpath(Register java_thread = rthread);
 978 
 979   // Round up to a power of two
 980   void round_to(Register reg, int modulus);
 981 
 982   // java.lang.Math::round intrinsics
 983   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
 984   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
 985 
 986   // allocation
 987   void tlab_allocate(
 988     Register obj,                      // result: pointer to object after successful allocation
 989     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
 990     int      con_size_in_bytes,        // object size in bytes if   known at compile time
 991     Register t1,                       // temp register
 992     Register t2,                       // temp register
 993     Label&   slow_case                 // continuation point if fast allocation fails
 994   );
 995   void verify_tlab();
 996 
 997   // interface method calling
 998   void lookup_interface_method(Register recv_klass,
 999                                Register intf_klass,
1000                                RegisterOrConstant itable_index,
1001                                Register method_result,
1002                                Register scan_temp,
1003                                Label& no_such_interface,
1004                    bool return_method = true);
1005 
1006   void lookup_interface_method_stub(Register recv_klass,
1007                                     Register holder_klass,
1008                                     Register resolved_klass,
1009                                     Register method_result,
1010                                     Register temp_reg,
1011                                     Register temp_reg2,
1012                                     int itable_index,
1013                                     Label& L_no_such_interface);
1014 
1015   // virtual method calling
1016   // n.b. x86 allows RegisterOrConstant for vtable_index
1017   void lookup_virtual_method(Register recv_klass,
1018                              RegisterOrConstant vtable_index,
1019                              Register method_result);
1020 
1021   // Test sub_klass against super_klass, with fast and slow paths.
1022 
1023   // The fast path produces a tri-state answer: yes / no / maybe-slow.
1024   // One of the three labels can be null, meaning take the fall-through.
1025   // If super_check_offset is -1, the value is loaded up from super_klass.
1026   // No registers are killed, except temp_reg.
1027   void check_klass_subtype_fast_path(Register sub_klass,
1028                                      Register super_klass,
1029                                      Register temp_reg,
1030                                      Label* L_success,
1031                                      Label* L_failure,
1032                                      Label* L_slow_path,
1033                                      Register super_check_offset = noreg);
1034 
1035   // The rest of the type check; must be wired to a corresponding fast path.
1036   // It does not repeat the fast path logic, so don't use it standalone.
1037   // The temp_reg and temp2_reg can be noreg, if no temps are available.
1038   // Updates the sub's secondary super cache as necessary.
1039   // If set_cond_codes, condition codes will be Z on success, NZ on failure.
1040   void check_klass_subtype_slow_path(Register sub_klass,
1041                                      Register super_klass,
1042                                      Register temp_reg,
1043                                      Register temp2_reg,
1044                                      Label* L_success,
1045                                      Label* L_failure,
1046                                      bool set_cond_codes = false);
1047 
1048   void check_klass_subtype_slow_path_linear(Register sub_klass,
1049                                             Register super_klass,
1050                                             Register temp_reg,
1051                                             Register temp2_reg,
1052                                             Label* L_success,
1053                                             Label* L_failure,
1054                                             bool set_cond_codes = false);
1055 
1056   void check_klass_subtype_slow_path_table(Register sub_klass,
1057                                            Register super_klass,
1058                                            Register temp_reg,
1059                                            Register temp2_reg,
1060                                            Register temp3_reg,
1061                                            Register result_reg,
1062                                            FloatRegister vtemp_reg,
1063                                            Label* L_success,
1064                                            Label* L_failure,
1065                                            bool set_cond_codes = false);
1066 
1067   // If r is valid, return r.
1068   // If r is invalid, remove a register r2 from available_regs, add r2
1069   // to regs_to_push, then return r2.
1070   Register allocate_if_noreg(const Register r,
1071                              RegSetIterator<Register> &available_regs,
1072                              RegSet &regs_to_push);
1073 
1074   // Secondary subtype checking
1075   void lookup_secondary_supers_table_var(Register sub_klass,
1076                                          Register r_super_klass,
1077                                          Register temp1,
1078                                          Register temp2,
1079                                          Register temp3,
1080                                          FloatRegister vtemp,
1081                                          Register result,
1082                                          Label *L_success);
1083 
1084 
1085   // As above, but with a constant super_klass.
1086   // The result is in Register result, not the condition codes.
1087   bool lookup_secondary_supers_table_const(Register r_sub_klass,
1088                                            Register r_super_klass,
1089                                            Register temp1,
1090                                            Register temp2,
1091                                            Register temp3,
1092                                            FloatRegister vtemp,
1093                                            Register result,
1094                                            u1 super_klass_slot,
1095                                            bool stub_is_near = false);
1096 
1097   void verify_secondary_supers_table(Register r_sub_klass,
1098                                      Register r_super_klass,
1099                                      Register temp1,
1100                                      Register temp2,
1101                                      Register result);
1102 
1103   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
1104                                                Register r_array_base,
1105                                                Register r_array_index,
1106                                                Register r_bitmap,
1107                                                Register temp1,
1108                                                Register result,
1109                                                bool is_stub = true);
1110 
1111   // Simplified, combined version, good for typical uses.
1112   // Falls through on failure.
1113   void check_klass_subtype(Register sub_klass,
1114                            Register super_klass,
1115                            Register temp_reg,
1116                            Label& L_success);
1117 
1118   void clinit_barrier(Register klass,
1119                       Register thread,
1120                       Label* L_fast_path = nullptr,
1121                       Label* L_slow_path = nullptr);
1122 
1123   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
1124 
1125   void verify_sve_vector_length(Register tmp = rscratch1);
1126   void reinitialize_ptrue() {
1127     if (UseSVE > 0) {
1128       sve_ptrue(ptrue, B);
1129     }
1130   }
1131   void verify_ptrue();
1132 
1133   // Debugging
1134 
1135   // only if +VerifyOops
1136   void _verify_oop(Register reg, const char* s, const char* file, int line);
1137   void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
1138 
1139   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
1140     if (VerifyOops) {
1141       _verify_oop(reg, s, file, line);
1142     }
1143   }
1144   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
1145     if (VerifyOops) {
1146       _verify_oop_addr(reg, s, file, line);
1147     }
1148   }
1149 
1150 // TODO: verify method and klass metadata (compare against vptr?)
1151   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1152   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1153 
1154 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
1155 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
1156 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
1157 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1158 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1159 
1160   // Restore cpu control state after JNI call
1161   void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2);
1162 
1163   // prints msg, dumps registers and stops execution
1164   void stop(const char* msg);
1165 
1166   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1167 
1168   void untested()                                { stop("untested"); }
1169 
1170   void unimplemented(const char* what = "");
1171 
1172   void should_not_reach_here()                   { stop("should not reach here"); }
1173 
1174   void _assert_asm(Condition cc, const char* msg);
1175 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg)
1176 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg))
1177 
1178   // Stack overflow checking
1179   void bang_stack_with_offset(int offset) {
1180     // stack grows down, caller passes positive offset
1181     assert(offset > 0, "must bang with negative offset");
1182     sub(rscratch2, sp, offset);
1183     str(zr, Address(rscratch2));
1184   }
1185 
1186   // Writes to stack successive pages until offset reached to check for
1187   // stack overflow + shadow pages.  Also, clobbers tmp
1188   void bang_stack_size(Register size, Register tmp);
1189 
1190   // Check for reserved stack access in method being exited (for JIT)
1191   void reserved_stack_check();
1192 
1193   // Arithmetics
1194 
1195   // Clobber: rscratch1, rscratch2
1196   void addptr(const Address &dst, int32_t src);
1197 
1198   // Clobber: rscratch1
1199   void cmpptr(Register src1, Address src2);
1200 
1201   void cmpoop(Register obj1, Register obj2);
1202 
1203   // Various forms of CAS
1204 
1205   void cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp,
1206                           Label &succeed, Label *fail);
1207   void cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
1208                   Label &succeed, Label *fail);
1209 
1210   void cmpxchgw(Register oldv, Register newv, Register addr, Register tmp,
1211                   Label &succeed, Label *fail);
1212 
1213   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1214   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1215   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1216   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1217 
1218   void atomic_xchg(Register prev, Register newv, Register addr);
1219   void atomic_xchgw(Register prev, Register newv, Register addr);
1220   void atomic_xchgl(Register prev, Register newv, Register addr);
1221   void atomic_xchglw(Register prev, Register newv, Register addr);
1222   void atomic_xchgal(Register prev, Register newv, Register addr);
1223   void atomic_xchgalw(Register prev, Register newv, Register addr);
1224 
1225   void orptr(Address adr, RegisterOrConstant src) {
1226     ldr(rscratch1, adr);
1227     if (src.is_register())
1228       orr(rscratch1, rscratch1, src.as_register());
1229     else
1230       orr(rscratch1, rscratch1, src.as_constant());
1231     str(rscratch1, adr);
1232   }
1233 
1234   // A generic CAS; success or failure is in the EQ flag.
1235   // Clobbers rscratch1
1236   void cmpxchg(Register addr, Register expected, Register new_val,
1237                enum operand_size size,
1238                bool acquire, bool release, bool weak,
1239                Register result);
1240 
1241 #ifdef ASSERT
1242   // Template short-hand support to clean-up after a failed call to trampoline
1243   // call generation (see trampoline_call() below),  when a set of Labels must
1244   // be reset (before returning).
1245   template<typename Label, typename... More>
1246   void reset_labels(Label &lbl, More&... more) {
1247     lbl.reset(); reset_labels(more...);
1248   }
1249   template<typename Label>
1250   void reset_labels(Label &lbl) {
1251     lbl.reset();
1252   }
1253 #endif
1254 
1255 private:
1256   void compare_eq(Register rn, Register rm, enum operand_size size);
1257 
1258 public:
1259   // AArch64 OpenJDK uses four different types of calls:
1260   //   - direct call: bl pc_relative_offset
1261   //     This is the shortest and the fastest, but the offset has the range:
1262   //     +/-128MB for the release build, +/-2MB for the debug build.
1263   //
1264   //   - far call: adrp reg, pc_relative_offset; add; bl reg
1265   //     This is longer than a direct call. The offset has
1266   //     the range +/-4GB. As the code cache size is limited to 4GB,
1267   //     far calls can reach anywhere in the code cache. If a jump is
1268   //     needed rather than a call, a far jump 'b reg' can be used instead.
1269   //     All instructions are embedded at a call site.
1270   //
1271   //   - trampoline call:
1272   //     This is only available in C1/C2-generated code (nmethod). It is a combination
1273   //     of a direct call, which is used if the destination of a call is in range,
1274   //     and a register-indirect call. It has the advantages of reaching anywhere in
1275   //     the AArch64 address space and being patchable at runtime when the generated
1276   //     code is being executed by other threads.
1277   //
1278   //     [Main code section]
1279   //       bl trampoline
1280   //     [Stub code section]
1281   //     trampoline:
1282   //       ldr reg, pc + 8
1283   //       br reg
1284   //       <64-bit destination address>
1285   //
1286   //     If the destination is in range when the generated code is moved to the code
1287   //     cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline
1288   //     is not used.
1289   //     The optimization does not remove the trampoline from the stub section.
1290   //     This is necessary because the trampoline may well be redirected later when
1291   //     code is patched, and the new destination may not be reachable by a simple BR
1292   //     instruction.
1293   //
1294   //   - indirect call: move reg, address; blr reg
1295   //     This too can reach anywhere in the address space, but it cannot be
1296   //     patched while code is running, so it must only be modified at a safepoint.
1297   //     This form of call is most suitable for targets at fixed addresses, which
1298   //     will never be patched.
1299   //
1300   // The patching we do conforms to the "Concurrent modification and
1301   // execution of instructions" section of the Arm Architectural
1302   // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC,
1303   // or SVC instructions to be modified while another thread is
1304   // executing them.
1305   //
1306   // To patch a trampoline call when the BL can't reach, we first modify
1307   // the 64-bit destination address in the trampoline, then modify the
1308   // BL to point to the trampoline, then flush the instruction cache to
1309   // broadcast the change to all executing threads. See
1310   // NativeCall::set_destination_mt_safe for the details.
1311   //
1312   // There is a benign race in that the other thread might observe the
1313   // modified BL before it observes the modified 64-bit destination
1314   // address. That does not matter because the destination method has been
1315   // invalidated, so there will be a trap at its start.
1316   // For this to work, the destination address in the trampoline is
1317   // always updated, even if we're not using the trampoline.
1318 
1319   // Emit a direct call if the entry address will always be in range,
1320   // otherwise a trampoline call.
1321   // Supported entry.rspec():
1322   // - relocInfo::runtime_call_type
1323   // - relocInfo::opt_virtual_call_type
1324   // - relocInfo::static_call_type
1325   // - relocInfo::virtual_call_type
1326   //
1327   // Return: the call PC or null if CodeCache is full.
1328   // Clobbers: rscratch1
1329   address trampoline_call(Address entry);
1330 
1331   static bool far_branches() {
1332     return ReservedCodeCacheSize > branch_range;
1333   }
1334 
1335   // Check if branches to the non nmethod section require a far jump
1336   static bool codestub_branch_needs_far_jump() {
1337     if (AOTCodeCache::is_on_for_dump()) {
1338       // To calculate far_codestub_branch_size correctly.
1339       return true;
1340     }
1341     return CodeCache::max_distance_to_non_nmethod() > branch_range;
1342   }
1343 
1344   // Emit a direct call/jump if the entry address will always be in range,
1345   // otherwise a far call/jump.
1346   // The address must be inside the code cache.
1347   // Supported entry.rspec():
1348   // - relocInfo::external_word_type
1349   // - relocInfo::runtime_call_type
1350   // - relocInfo::none
1351   // In the case of a far call/jump, the entry address is put in the tmp register.
1352   // The tmp register is invalidated.
1353   //
1354   // Far_jump returns the amount of the emitted code.
1355   void far_call(Address entry, Register tmp = rscratch1);
1356   int far_jump(Address entry, Register tmp = rscratch1);
1357 
1358   static int far_codestub_branch_size() {
1359     if (codestub_branch_needs_far_jump()) {
1360       return 3 * 4;  // adrp, add, br
1361     } else {
1362       return 4;
1363     }
1364   }
1365 
1366   // Emit the CompiledIC call idiom
1367   address ic_call(address entry, jint method_index = 0);
1368   static int ic_check_size();
1369   int ic_check(int end_alignment);
1370 
1371 public:
1372 
1373   // Data
1374 
1375   void mov_metadata(Register dst, Metadata* obj);
1376   Address allocate_metadata_address(Metadata* obj);
1377   Address constant_oop_address(jobject obj);
1378 
1379   void movoop(Register dst, jobject obj);
1380 
1381   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1382   void kernel_crc32(Register crc, Register buf, Register len,
1383         Register table0, Register table1, Register table2, Register table3,
1384         Register tmp, Register tmp2, Register tmp3);
1385   // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic.
1386   void kernel_crc32c(Register crc, Register buf, Register len,
1387         Register table0, Register table1, Register table2, Register table3,
1388         Register tmp, Register tmp2, Register tmp3);
1389 
1390   // Stack push and pop individual 64 bit registers
1391   void push(Register src);
1392   void pop(Register dst);
1393 
1394   void repne_scan(Register addr, Register value, Register count,
1395                   Register scratch);
1396   void repne_scanw(Register addr, Register value, Register count,
1397                    Register scratch);
1398 
1399   typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1400   typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1401 
1402   // If a constant does not fit in an immediate field, generate some
1403   // number of MOV instructions and then perform the operation
1404   void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
1405                              add_sub_imm_insn insn1,
1406                              add_sub_reg_insn insn2, bool is32);
1407   // Separate vsn which sets the flags
1408   void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
1409                                add_sub_imm_insn insn1,
1410                                add_sub_reg_insn insn2, bool is32);
1411 
1412 #define WRAP(INSN, is32)                                                \
1413   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1414     wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1415   }                                                                     \
1416                                                                         \
1417   void INSN(Register Rd, Register Rn, Register Rm,                      \
1418              enum shift_kind kind, unsigned shift = 0) {                \
1419     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1420   }                                                                     \
1421                                                                         \
1422   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1423     Assembler::INSN(Rd, Rn, Rm);                                        \
1424   }                                                                     \
1425                                                                         \
1426   void INSN(Register Rd, Register Rn, Register Rm,                      \
1427            ext::operation option, int amount = 0) {                     \
1428     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1429   }
1430 
1431   WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true)
1432 
1433 #undef WRAP
1434 #define WRAP(INSN, is32)                                                \
1435   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1436     wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1437   }                                                                     \
1438                                                                         \
1439   void INSN(Register Rd, Register Rn, Register Rm,                      \
1440              enum shift_kind kind, unsigned shift = 0) {                \
1441     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1442   }                                                                     \
1443                                                                         \
1444   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1445     Assembler::INSN(Rd, Rn, Rm);                                        \
1446   }                                                                     \
1447                                                                         \
1448   void INSN(Register Rd, Register Rn, Register Rm,                      \
1449            ext::operation option, int amount = 0) {                     \
1450     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1451   }
1452 
1453   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1454 
1455   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1456   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1457   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1458   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1459 
1460   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1461 
1462   void tableswitch(Register index, jint lowbound, jint highbound,
1463                    Label &jumptable, Label &jumptable_end, int stride = 1) {
1464     adr(rscratch1, jumptable);
1465     subsw(rscratch2, index, lowbound);
1466     subsw(zr, rscratch2, highbound - lowbound);
1467     br(Assembler::HS, jumptable_end);
1468     add(rscratch1, rscratch1, rscratch2,
1469         ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1470     br(rscratch1);
1471   }
1472 
1473   // Form an address from base + offset in Rd.  Rd may or may not
1474   // actually be used: you must use the Address that is returned.  It
1475   // is up to you to ensure that the shift provided matches the size
1476   // of your data.
1477   Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1478 
1479   // Return true iff an address is within the 48-bit AArch64 address
1480   // space.
1481   bool is_valid_AArch64_address(address a) {
1482     return ((uint64_t)a >> 48) == 0;
1483   }
1484 
1485   // Load the base of the cardtable byte map into reg.
1486   void load_byte_map_base(Register reg);
1487 
1488   // Prolog generator routines to support switch between x86 code and
1489   // generated ARM code
1490 
1491   // routine to generate an x86 prolog for a stub function which
1492   // bootstraps into the generated ARM code which directly follows the
1493   // stub
1494   //
1495 
1496   public:
1497 
1498   address read_polling_page(Register r, relocInfo::relocType rtype);
1499   void get_polling_page(Register dest, relocInfo::relocType rtype);
1500 
1501   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1502   void update_byte_crc32(Register crc, Register val, Register table);
1503   void update_word_crc32(Register crc, Register v, Register tmp,
1504         Register table0, Register table1, Register table2, Register table3,
1505         bool upper = false);
1506 
1507   address count_positives(Register ary1, Register len, Register result);
1508 
1509   address arrays_equals(Register a1, Register a2, Register result, Register cnt1,
1510                         Register tmp1, Register tmp2, Register tmp3, int elem_size);
1511 
1512 // Ensure that the inline code and the stub use the same registers.
1513 #define ARRAYS_HASHCODE_REGISTERS \
1514   do {                      \
1515     assert(result == r0  && \
1516            ary    == r1  && \
1517            cnt    == r2  && \
1518            vdata0 == v3  && \
1519            vdata1 == v2  && \
1520            vdata2 == v1  && \
1521            vdata3 == v0  && \
1522            vmul0  == v4  && \
1523            vmul1  == v5  && \
1524            vmul2  == v6  && \
1525            vmul3  == v7  && \
1526            vpow   == v12 && \
1527            vpowm  == v13, "registers must match aarch64.ad"); \
1528   } while (0)
1529 
1530   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1531 
1532   void fill_words(Register base, Register cnt, Register value);
1533   address zero_words(Register base, uint64_t cnt);
1534   address zero_words(Register ptr, Register cnt);
1535   void zero_dcache_blocks(Register base, Register cnt);
1536 
1537   static const int zero_words_block_size;
1538 
1539   address byte_array_inflate(Register src, Register dst, Register len,
1540                              FloatRegister vtmp1, FloatRegister vtmp2,
1541                              FloatRegister vtmp3, Register tmp4);
1542 
1543   void char_array_compress(Register src, Register dst, Register len,
1544                            Register res,
1545                            FloatRegister vtmp0, FloatRegister vtmp1,
1546                            FloatRegister vtmp2, FloatRegister vtmp3,
1547                            FloatRegister vtmp4, FloatRegister vtmp5);
1548 
1549   void encode_iso_array(Register src, Register dst,
1550                         Register len, Register res, bool ascii,
1551                         FloatRegister vtmp0, FloatRegister vtmp1,
1552                         FloatRegister vtmp2, FloatRegister vtmp3,
1553                         FloatRegister vtmp4, FloatRegister vtmp5);
1554 
1555   void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi,
1556       address pio2, address dsin_coef, address dcos_coef);
1557  private:
1558   // begin trigonometric functions support block
1559   void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2);
1560   void generate__kernel_rem_pio2(address two_over_pi, address pio2);
1561   void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef);
1562   void generate_kernel_cos(FloatRegister x, address dcos_coef);
1563   // end trigonometric functions support block
1564   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1565                        Register src1, Register src2);
1566   void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1567     add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1568   }
1569   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1570                              Register y, Register y_idx, Register z,
1571                              Register carry, Register product,
1572                              Register idx, Register kdx);
1573   void multiply_128_x_128_loop(Register y, Register z,
1574                                Register carry, Register carry2,
1575                                Register idx, Register jdx,
1576                                Register yz_idx1, Register yz_idx2,
1577                                Register tmp, Register tmp3, Register tmp4,
1578                                Register tmp7, Register product_hi);
1579   void kernel_crc32_using_crypto_pmull(Register crc, Register buf,
1580         Register len, Register tmp0, Register tmp1, Register tmp2,
1581         Register tmp3);
1582   void kernel_crc32_using_crc32(Register crc, Register buf,
1583         Register len, Register tmp0, Register tmp1, Register tmp2,
1584         Register tmp3);
1585   void kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
1586         Register len, Register tmp0, Register tmp1, Register tmp2,
1587         Register tmp3);
1588   void kernel_crc32c_using_crc32c(Register crc, Register buf,
1589         Register len, Register tmp0, Register tmp1, Register tmp2,
1590         Register tmp3);
1591   void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
1592         Register len, Register tmp0, Register tmp1, Register tmp2,
1593         size_t table_offset);
1594 
1595   void ghash_modmul (FloatRegister result,
1596                      FloatRegister result_lo, FloatRegister result_hi, FloatRegister b,
1597                      FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p,
1598                      FloatRegister t1, FloatRegister t2, FloatRegister t3);
1599   void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
1600 public:
1601   void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1602                        Register tmp0, Register tmp1, Register tmp2, Register tmp3,
1603                        Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1604   void mul_add(Register out, Register in, Register offs, Register len, Register k);
1605   void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,
1606                       FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1607                       FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1608   void ghash_multiply_wide(int index,
1609                            FloatRegister result_lo, FloatRegister result_hi,
1610                            FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1611                            FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1612   void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi,
1613                     FloatRegister p, FloatRegister z, FloatRegister t1);
1614   void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi,
1615                     FloatRegister p, FloatRegister z, FloatRegister t1);
1616   void ghash_processBlocks_wide(Label& p, Register state, Register subkeyH,
1617                                 Register data, Register blocks, int unrolls);
1618 
1619 
1620   void aesenc_loadkeys(Register key, Register keylen);
1621   void aesecb_encrypt(Register from, Register to, Register keylen,
1622                       FloatRegister data = v0, int unrolls = 1);
1623   void aesecb_decrypt(Register from, Register to, Register key, Register keylen);
1624   void aes_round(FloatRegister input, FloatRegister subkey);
1625 
1626   // ChaCha20 functions support block
1627   void cc20_qr_add4(FloatRegister (&addFirst)[4],
1628           FloatRegister (&addSecond)[4]);
1629   void cc20_qr_xor4(FloatRegister (&firstElem)[4],
1630           FloatRegister (&secondElem)[4], FloatRegister (&result)[4]);
1631   void cc20_qr_lrot4(FloatRegister (&sourceReg)[4],
1632           FloatRegister (&destReg)[4], int bits, FloatRegister table);
1633   void cc20_set_qr_registers(FloatRegister (&vectorSet)[4],
1634           const FloatRegister (&stateVectors)[16], int idx1, int idx2,
1635           int idx3, int idx4);
1636 
1637   // Place an ISB after code may have been modified due to a safepoint.
1638   void safepoint_isb();
1639 
1640 private:
1641   // Return the effective address r + (r1 << ext) + offset.
1642   // Uses rscratch2.
1643   Address offsetted_address(Register r, Register r1, Address::extend ext,
1644                             int offset, int size);
1645 
1646 private:
1647   // Returns an address on the stack which is reachable with a ldr/str of size
1648   // Uses rscratch2 if the address is not directly reachable
1649   Address spill_address(int size, int offset, Register tmp=rscratch2);
1650   Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2);
1651 
1652   bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const;
1653 
1654   // Check whether two loads/stores can be merged into ldp/stp.
1655   bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const;
1656 
1657   // Merge current load/store with previous load/store into ldp/stp.
1658   void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1659 
1660   // Try to merge two loads/stores into ldp/stp. If success, returns true else false.
1661   bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1662 
1663 public:
1664   void spill(Register Rx, bool is64, int offset) {
1665     if (is64) {
1666       str(Rx, spill_address(8, offset));
1667     } else {
1668       strw(Rx, spill_address(4, offset));
1669     }
1670   }
1671   void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1672     str(Vx, T, spill_address(1 << (int)T, offset));
1673   }
1674 
1675   void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1676     sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1677   }
1678   void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1679     sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1680   }
1681 
1682   void unspill(Register Rx, bool is64, int offset) {
1683     if (is64) {
1684       ldr(Rx, spill_address(8, offset));
1685     } else {
1686       ldrw(Rx, spill_address(4, offset));
1687     }
1688   }
1689   void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1690     ldr(Vx, T, spill_address(1 << (int)T, offset));
1691   }
1692 
1693   void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1694     sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1695   }
1696   void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1697     sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1698   }
1699 
1700   void spill_copy128(int src_offset, int dst_offset,
1701                      Register tmp1=rscratch1, Register tmp2=rscratch2) {
1702     if (src_offset < 512 && (src_offset & 7) == 0 &&
1703         dst_offset < 512 && (dst_offset & 7) == 0) {
1704       ldp(tmp1, tmp2, Address(sp, src_offset));
1705       stp(tmp1, tmp2, Address(sp, dst_offset));
1706     } else {
1707       unspill(tmp1, true, src_offset);
1708       spill(tmp1, true, dst_offset);
1709       unspill(tmp1, true, src_offset+8);
1710       spill(tmp1, true, dst_offset+8);
1711     }
1712   }
1713   void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset,
1714                                             int sve_vec_reg_size_in_bytes) {
1715     assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size");
1716     for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) {
1717       spill_copy128(src_offset, dst_offset);
1718       src_offset += 16;
1719       dst_offset += 16;
1720     }
1721   }
1722   void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset,
1723                                                int sve_predicate_reg_size_in_bytes) {
1724     sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset));
1725     sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset));
1726     reinitialize_ptrue();
1727   }
1728   void cache_wb(Address line);
1729   void cache_wbsync(bool is_pre);
1730 
1731   // Code for java.lang.Thread::onSpinWait() intrinsic.
1732   void spin_wait();
1733 
1734   void lightweight_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow);
1735   void lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1736 
1737 private:
1738   // Check the current thread doesn't need a cross modify fence.
1739   void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
1740 
1741 };
1742 
1743 #ifdef ASSERT
1744 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1745 #endif
1746 
1747 struct tableswitch {
1748   Register _reg;
1749   int _insn_index; jint _first_key; jint _last_key;
1750   Label _after;
1751   Label _branches;
1752 };
1753 
1754 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP