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/macros.hpp"
  37 #include "utilities/powerOfTwo.hpp"
  38 #include "runtime/signature.hpp"
  39 
  40 
  41 class ciInlineKlass;
  42 
  43 class OopMap;
  44 
  45 // MacroAssembler extends Assembler by frequently used macros.
  46 //
  47 // Instructions for which a 'better' code sequence exists depending
  48 // on arguments should also go in here.
  49 
  50 class MacroAssembler: public Assembler {
  51   friend class LIR_Assembler;
  52 
  53  public:
  54   using Assembler::mov;
  55   using Assembler::movi;
  56 
  57  protected:
  58 
  59   // Support for VM calls
  60   //
  61   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
  62   // may customize this version by overriding it for its purposes (e.g., to save/restore
  63   // additional registers when doing a VM call).
  64   virtual void call_VM_leaf_base(
  65     address entry_point,               // the entry point
  66     int     number_of_arguments,        // the number of arguments to pop after the call
  67     Label *retaddr = nullptr
  68   );
  69 
  70   virtual void call_VM_leaf_base(
  71     address entry_point,               // the entry point
  72     int     number_of_arguments,        // the number of arguments to pop after the call
  73     Label &retaddr) {
  74     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
  75   }
  76 
  77   // This is the base routine called by the different versions of call_VM. The interpreter
  78   // may customize this version by overriding it for its purposes (e.g., to save/restore
  79   // additional registers when doing a VM call).
  80   //
  81   // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
  82   // returns the register which contains the thread upon return. If a thread register has been
  83   // specified, the return value will correspond to that register. If no last_java_sp is specified
  84   // (noreg) than rsp will be used instead.
  85   virtual void call_VM_base(           // returns the register containing the thread upon return
  86     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
  87     Register java_thread,              // the thread if computed before     ; use noreg otherwise
  88     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
  89     address  entry_point,              // the entry point
  90     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
  91     bool     check_exceptions          // whether to check for pending exceptions after return
  92   );
  93 
  94   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
  95 
  96   enum KlassDecodeMode {
  97     KlassDecodeNone,
  98     KlassDecodeZero,
  99     KlassDecodeXor,
 100     KlassDecodeMovk
 101   };
 102 
 103   // Calculate decoding mode based on given parameters, used for checking then ultimately setting.
 104   static KlassDecodeMode klass_decode_mode(address base, int shift, const size_t range);
 105 
 106  private:
 107   static KlassDecodeMode _klass_decode_mode;
 108 
 109   // Returns above setting with asserts
 110   static KlassDecodeMode klass_decode_mode();
 111 
 112  public:
 113   // Checks the decode mode and returns false if not compatible with preferred decoding mode.
 114   static bool check_klass_decode_mode(address base, int shift, const size_t range);
 115 
 116   // Sets the decode mode and returns false if cannot be set.
 117   static bool set_klass_decode_mode(address base, int shift, const size_t range);
 118 
 119  public:
 120   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 121 
 122  // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 123  // The implementation is only non-empty for the InterpreterMacroAssembler,
 124  // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 125  virtual void check_and_handle_popframe(Register java_thread);
 126  virtual void check_and_handle_earlyret(Register java_thread);
 127 
 128   void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp = rscratch1);
 129   void rt_call(address dest, Register tmp = rscratch1);
 130 
 131   // Load Effective Address
 132   void lea(Register r, const Address &a) {
 133     InstructionMark im(this);
 134     a.lea(this, r);
 135   }
 136 
 137   // Whether materializing the given address for a LDR/STR requires an
 138   // additional lea instruction.
 139   static bool legitimize_address_requires_lea(const Address &a, int size) {
 140     return a.getMode() == Address::base_plus_offset &&
 141            !Address::offset_ok_for_immed(a.offset(), exact_log2(size));
 142   }
 143 
 144   /* Sometimes we get misaligned loads and stores, usually from Unsafe
 145      accesses, and these can exceed the offset range. */
 146   Address legitimize_address(const Address &a, int size, Register scratch) {
 147     if (legitimize_address_requires_lea(a, size)) {
 148       block_comment("legitimize_address {");
 149       lea(scratch, a);
 150       block_comment("} legitimize_address");
 151       return Address(scratch);
 152     }
 153     return a;
 154   }
 155 
 156   void addmw(Address a, Register incr, Register scratch) {
 157     ldrw(scratch, a);
 158     addw(scratch, scratch, incr);
 159     strw(scratch, a);
 160   }
 161 
 162   // Add constant to memory word
 163   void addmw(Address a, int imm, Register scratch) {
 164     ldrw(scratch, a);
 165     if (imm > 0)
 166       addw(scratch, scratch, (unsigned)imm);
 167     else
 168       subw(scratch, scratch, (unsigned)-imm);
 169     strw(scratch, a);
 170   }
 171 
 172   void bind(Label& L) {
 173     Assembler::bind(L);
 174     code()->clear_last_insn();
 175     code()->set_last_label(pc());
 176   }
 177 
 178   void membar(Membar_mask_bits order_constraint);
 179 
 180   using Assembler::ldr;
 181   using Assembler::str;
 182   using Assembler::ldrw;
 183   using Assembler::strw;
 184 
 185   void ldr(Register Rx, const Address &adr);
 186   void ldrw(Register Rw, const Address &adr);
 187   void str(Register Rx, const Address &adr);
 188   void strw(Register Rx, const Address &adr);
 189 
 190   // Frame creation and destruction shared between JITs.
 191   void build_frame(int framesize);
 192   void remove_frame(int framesize);
 193 
 194   virtual void _call_Unimplemented(address call_site) {
 195     mov(rscratch2, call_site);
 196   }
 197 
 198 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 199 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 200 // https://reviews.llvm.org/D3311
 201 
 202 #ifdef _WIN64
 203 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 204 #else
 205 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 206 #endif
 207 
 208   // aliases defined in AARCH64 spec
 209 
 210   template<class T>
 211   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }
 212 
 213   inline void cmp(Register Rd, unsigned char imm8)  { subs(zr, Rd, imm8); }
 214   inline void cmp(Register Rd, unsigned imm) = delete;
 215 
 216   template<class T>
 217   inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); }
 218 
 219   inline void cmn(Register Rd, unsigned char imm8)  { adds(zr, Rd, imm8); }
 220   inline void cmn(Register Rd, unsigned imm) = delete;
 221 
 222   void cset(Register Rd, Assembler::Condition cond) {
 223     csinc(Rd, zr, zr, ~cond);
 224   }
 225   void csetw(Register Rd, Assembler::Condition cond) {
 226     csincw(Rd, zr, zr, ~cond);
 227   }
 228 
 229   void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
 230     csneg(Rd, Rn, Rn, ~cond);
 231   }
 232   void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
 233     csnegw(Rd, Rn, Rn, ~cond);
 234   }
 235 
 236   inline void movw(Register Rd, Register Rn) {
 237     if (Rd == sp || Rn == sp) {
 238       Assembler::addw(Rd, Rn, 0U);
 239     } else {
 240       orrw(Rd, zr, Rn);
 241     }
 242   }
 243   inline void mov(Register Rd, Register Rn) {
 244     assert(Rd != r31_sp && Rn != r31_sp, "should be");
 245     if (Rd == Rn) {
 246     } else if (Rd == sp || Rn == sp) {
 247       Assembler::add(Rd, Rn, 0U);
 248     } else {
 249       orr(Rd, zr, Rn);
 250     }
 251   }
 252 
 253   inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
 254   inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
 255 
 256   inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
 257   inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
 258 
 259   inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
 260   inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
 261 
 262   inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 263     bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 264   }
 265   inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 266     bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 267   }
 268 
 269   inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 270     bfmw(Rd, Rn, lsb, (lsb + width - 1));
 271   }
 272   inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 273     bfm(Rd, Rn, lsb , (lsb + width - 1));
 274   }
 275 
 276   inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 277     sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 278   }
 279   inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 280     sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 281   }
 282 
 283   inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 284     sbfmw(Rd, Rn, lsb, (lsb + width - 1));
 285   }
 286   inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 287     sbfm(Rd, Rn, lsb , (lsb + width - 1));
 288   }
 289 
 290   inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 291     ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 292   }
 293   inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 294     ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 295   }
 296 
 297   inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 298     ubfmw(Rd, Rn, lsb, (lsb + width - 1));
 299   }
 300   inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 301     ubfm(Rd, Rn, lsb , (lsb + width - 1));
 302   }
 303 
 304   inline void asrw(Register Rd, Register Rn, unsigned imm) {
 305     sbfmw(Rd, Rn, imm, 31);
 306   }
 307 
 308   inline void asr(Register Rd, Register Rn, unsigned imm) {
 309     sbfm(Rd, Rn, imm, 63);
 310   }
 311 
 312   inline void lslw(Register Rd, Register Rn, unsigned imm) {
 313     ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
 314   }
 315 
 316   inline void lsl(Register Rd, Register Rn, unsigned imm) {
 317     ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
 318   }
 319 
 320   inline void lsrw(Register Rd, Register Rn, unsigned imm) {
 321     ubfmw(Rd, Rn, imm, 31);
 322   }
 323 
 324   inline void lsr(Register Rd, Register Rn, unsigned imm) {
 325     ubfm(Rd, Rn, imm, 63);
 326   }
 327 
 328   inline void rorw(Register Rd, Register Rn, unsigned imm) {
 329     extrw(Rd, Rn, Rn, imm);
 330   }
 331 
 332   inline void ror(Register Rd, Register Rn, unsigned imm) {
 333     extr(Rd, Rn, Rn, imm);
 334   }
 335 
 336   inline void rolw(Register Rd, Register Rn, unsigned imm) {
 337     extrw(Rd, Rn, Rn, (32 - imm));
 338   }
 339 
 340   inline void rol(Register Rd, Register Rn, unsigned imm) {
 341     extr(Rd, Rn, Rn, (64 - imm));
 342   }
 343 
 344   using Assembler::rax1;
 345   using Assembler::eor3;
 346 
 347   inline void rax1(Register Rd, Register Rn, Register Rm) {
 348     eor(Rd, Rn, Rm, ROR, 63); // Rd = Rn ^ rol(Rm, 1)
 349   }
 350 
 351   inline void eor3(Register Rd, Register Rn, Register Rm, Register Rk) {
 352     assert(Rd != Rn, "Use tmp register");
 353     eor(Rd, Rm, Rk);
 354     eor(Rd, Rd, Rn);
 355   }
 356 
 357   inline void sxtbw(Register Rd, Register Rn) {
 358     sbfmw(Rd, Rn, 0, 7);
 359   }
 360   inline void sxthw(Register Rd, Register Rn) {
 361     sbfmw(Rd, Rn, 0, 15);
 362   }
 363   inline void sxtb(Register Rd, Register Rn) {
 364     sbfm(Rd, Rn, 0, 7);
 365   }
 366   inline void sxth(Register Rd, Register Rn) {
 367     sbfm(Rd, Rn, 0, 15);
 368   }
 369   inline void sxtw(Register Rd, Register Rn) {
 370     sbfm(Rd, Rn, 0, 31);
 371   }
 372 
 373   inline void uxtbw(Register Rd, Register Rn) {
 374     ubfmw(Rd, Rn, 0, 7);
 375   }
 376   inline void uxthw(Register Rd, Register Rn) {
 377     ubfmw(Rd, Rn, 0, 15);
 378   }
 379   inline void uxtb(Register Rd, Register Rn) {
 380     ubfm(Rd, Rn, 0, 7);
 381   }
 382   inline void uxth(Register Rd, Register Rn) {
 383     ubfm(Rd, Rn, 0, 15);
 384   }
 385   inline void uxtw(Register Rd, Register Rn) {
 386     ubfm(Rd, Rn, 0, 31);
 387   }
 388 
 389   inline void cmnw(Register Rn, Register Rm) {
 390     addsw(zr, Rn, Rm);
 391   }
 392   inline void cmn(Register Rn, Register Rm) {
 393     adds(zr, Rn, Rm);
 394   }
 395 
 396   inline void cmpw(Register Rn, Register Rm) {
 397     subsw(zr, Rn, Rm);
 398   }
 399   inline void cmp(Register Rn, Register Rm) {
 400     subs(zr, Rn, Rm);
 401   }
 402 
 403   inline void negw(Register Rd, Register Rn) {
 404     subw(Rd, zr, Rn);
 405   }
 406 
 407   inline void neg(Register Rd, Register Rn) {
 408     sub(Rd, zr, Rn);
 409   }
 410 
 411   inline void negsw(Register Rd, Register Rn) {
 412     subsw(Rd, zr, Rn);
 413   }
 414 
 415   inline void negs(Register Rd, Register Rn) {
 416     subs(Rd, zr, Rn);
 417   }
 418 
 419   inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 420     addsw(zr, Rn, Rm, kind, shift);
 421   }
 422   inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 423     adds(zr, Rn, Rm, kind, shift);
 424   }
 425 
 426   inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 427     subsw(zr, Rn, Rm, kind, shift);
 428   }
 429   inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 430     subs(zr, Rn, Rm, kind, shift);
 431   }
 432 
 433   inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 434     subw(Rd, zr, Rn, kind, shift);
 435   }
 436 
 437   inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 438     sub(Rd, zr, Rn, kind, shift);
 439   }
 440 
 441   inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 442     subsw(Rd, zr, Rn, kind, shift);
 443   }
 444 
 445   inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 446     subs(Rd, zr, Rn, kind, shift);
 447   }
 448 
 449   inline void mnegw(Register Rd, Register Rn, Register Rm) {
 450     msubw(Rd, Rn, Rm, zr);
 451   }
 452   inline void mneg(Register Rd, Register Rn, Register Rm) {
 453     msub(Rd, Rn, Rm, zr);
 454   }
 455 
 456   inline void mulw(Register Rd, Register Rn, Register Rm) {
 457     maddw(Rd, Rn, Rm, zr);
 458   }
 459   inline void mul(Register Rd, Register Rn, Register Rm) {
 460     madd(Rd, Rn, Rm, zr);
 461   }
 462 
 463   inline void smnegl(Register Rd, Register Rn, Register Rm) {
 464     smsubl(Rd, Rn, Rm, zr);
 465   }
 466   inline void smull(Register Rd, Register Rn, Register Rm) {
 467     smaddl(Rd, Rn, Rm, zr);
 468   }
 469 
 470   inline void umnegl(Register Rd, Register Rn, Register Rm) {
 471     umsubl(Rd, Rn, Rm, zr);
 472   }
 473   inline void umull(Register Rd, Register Rn, Register Rm) {
 474     umaddl(Rd, Rn, Rm, zr);
 475   }
 476 
 477 #define WRAP(INSN)                                                            \
 478   void INSN(Register Rd, Register Rn, Register Rm, Register Ra) {             \
 479     if (VM_Version::supports_a53mac() && Ra != zr)                            \
 480       nop();                                                                  \
 481     Assembler::INSN(Rd, Rn, Rm, Ra);                                          \
 482   }
 483 
 484   WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
 485   WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
 486 #undef WRAP
 487 
 488 
 489   // macro assembly operations needed for aarch64
 490 
 491 public:
 492 
 493   enum FpPushPopMode {
 494     PushPopFull,
 495     PushPopSVE,
 496     PushPopNeon,
 497     PushPopFp
 498   };
 499 
 500   // first two private routines for loading 32 bit or 64 bit constants
 501 private:
 502 
 503   void mov_immediate64(Register dst, uint64_t imm64);
 504   void mov_immediate32(Register dst, uint32_t imm32);
 505 
 506   int push(unsigned int bitset, Register stack);
 507   int pop(unsigned int bitset, Register stack);
 508 
 509   int push_fp(unsigned int bitset, Register stack, FpPushPopMode mode);
 510   int pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode);
 511 
 512   int push_p(unsigned int bitset, Register stack);
 513   int pop_p(unsigned int bitset, Register stack);
 514 
 515   void mov(Register dst, Address a);
 516 
 517 public:
 518 
 519   void push(RegSet regs, Register stack) { if (regs.bits()) push(regs.bits(), stack); }
 520   void pop(RegSet regs, Register stack) { if (regs.bits()) pop(regs.bits(), stack); }
 521 
 522   void push_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) push_fp(regs.bits(), stack, mode); }
 523   void pop_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) pop_fp(regs.bits(), stack, mode); }
 524 
 525   static RegSet call_clobbered_gp_registers();
 526 
 527   void push_p(PRegSet regs, Register stack) { if (regs.bits()) push_p(regs.bits(), stack); }
 528   void pop_p(PRegSet regs, Register stack) { if (regs.bits()) pop_p(regs.bits(), stack); }
 529 
 530   // Push and pop everything that might be clobbered by a native
 531   // runtime call except rscratch1 and rscratch2.  (They are always
 532   // scratch, so we don't have to protect them.)  Only save the lower
 533   // 64 bits of each vector register. Additional registers can be excluded
 534   // in a passed RegSet.
 535   void push_call_clobbered_registers_except(RegSet exclude);
 536   void pop_call_clobbered_registers_except(RegSet exclude);
 537 
 538   void push_call_clobbered_registers() {
 539     push_call_clobbered_registers_except(RegSet());
 540   }
 541   void pop_call_clobbered_registers() {
 542     pop_call_clobbered_registers_except(RegSet());
 543   }
 544 
 545 
 546   // now mov instructions for loading absolute addresses and 32 or
 547   // 64 bit integers
 548 
 549   inline void mov(Register dst, address addr)             { mov_immediate64(dst, (uint64_t)addr); }
 550 
 551   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 552   inline void mov(Register dst, T o)                      { mov_immediate64(dst, (uint64_t)o); }
 553 
 554   inline void movw(Register dst, uint32_t imm32)          { mov_immediate32(dst, imm32); }
 555 
 556   void mov(Register dst, RegisterOrConstant src) {
 557     if (src.is_register())
 558       mov(dst, src.as_register());
 559     else
 560       mov(dst, src.as_constant());
 561   }
 562 
 563   void movptr(Register r, uintptr_t imm64);
 564 
 565   void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64);
 566 
 567   void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
 568     orr(Vd, T, Vn, Vn);
 569   }
 570 
 571   void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) {
 572     fcvtsh(tmp, src);
 573     smov(dst, tmp, H, 0);
 574   }
 575 
 576   void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) {
 577     mov(tmp, H, 0, src);
 578     fcvths(dst, tmp);
 579   }
 580 
 581   // Generalized Test Bit And Branch, including a "far" variety which
 582   // spans more than 32KiB.
 583   void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) {
 584     assert(cond == EQ || cond == NE, "must be");
 585 
 586     if (isfar)
 587       cond = ~cond;
 588 
 589     void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
 590     if (cond == Assembler::EQ)
 591       branch = &Assembler::tbz;
 592     else
 593       branch = &Assembler::tbnz;
 594 
 595     if (isfar) {
 596       Label L;
 597       (this->*branch)(Rt, bitpos, L);
 598       b(dest);
 599       bind(L);
 600     } else {
 601       (this->*branch)(Rt, bitpos, dest);
 602     }
 603   }
 604 
 605   // macro instructions for accessing and updating floating point
 606   // status register
 607   //
 608   // FPSR : op1 == 011
 609   //        CRn == 0100
 610   //        CRm == 0100
 611   //        op2 == 001
 612 
 613   inline void get_fpsr(Register reg)
 614   {
 615     mrs(0b11, 0b0100, 0b0100, 0b001, reg);
 616   }
 617 
 618   inline void set_fpsr(Register reg)
 619   {
 620     msr(0b011, 0b0100, 0b0100, 0b001, reg);
 621   }
 622 
 623   inline void clear_fpsr()
 624   {
 625     msr(0b011, 0b0100, 0b0100, 0b001, zr);
 626   }
 627 
 628   // FPCR : op1 == 011
 629   //        CRn == 0100
 630   //        CRm == 0100
 631   //        op2 == 000
 632 
 633   inline void get_fpcr(Register reg) {
 634     mrs(0b11, 0b0100, 0b0100, 0b000, reg);
 635   }
 636 
 637   inline void set_fpcr(Register reg) {
 638     msr(0b011, 0b0100, 0b0100, 0b000, reg);
 639   }
 640 
 641   // DCZID_EL0: op1 == 011
 642   //            CRn == 0000
 643   //            CRm == 0000
 644   //            op2 == 111
 645   inline void get_dczid_el0(Register reg)
 646   {
 647     mrs(0b011, 0b0000, 0b0000, 0b111, reg);
 648   }
 649 
 650   // CTR_EL0:   op1 == 011
 651   //            CRn == 0000
 652   //            CRm == 0000
 653   //            op2 == 001
 654   inline void get_ctr_el0(Register reg)
 655   {
 656     mrs(0b011, 0b0000, 0b0000, 0b001, reg);
 657   }
 658 
 659   inline void get_nzcv(Register reg) {
 660     mrs(0b011, 0b0100, 0b0010, 0b000, reg);
 661   }
 662 
 663   inline void set_nzcv(Register reg) {
 664     msr(0b011, 0b0100, 0b0010, 0b000, reg);
 665   }
 666 
 667   // idiv variant which deals with MINLONG as dividend and -1 as divisor
 668   int corrected_idivl(Register result, Register ra, Register rb,
 669                       bool want_remainder, Register tmp = rscratch1);
 670   int corrected_idivq(Register result, Register ra, Register rb,
 671                       bool want_remainder, Register tmp = rscratch1);
 672 
 673   // Support for null-checks
 674   //
 675   // Generates code that causes a null OS exception if the content of reg is null.
 676   // If the accessed location is M[reg + offset] and the offset is known, provide the
 677   // offset. No explicit code generation is needed if the offset is within a certain
 678   // range (0 <= offset <= page_size).
 679 
 680   virtual void null_check(Register reg, int offset = -1);
 681   static bool needs_explicit_null_check(intptr_t offset);
 682   static bool uses_implicit_null_check(void* address);
 683 
 684   // markWord tests, kills markWord reg
 685   void test_markword_is_inline_type(Register markword, Label& is_inline_type);
 686 
 687   // inlineKlass queries, kills temp_reg
 688   void test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null = true);
 689 
 690   void test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free);
 691   void test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free);
 692   void test_field_is_flat(Register flags, Register temp_reg, Label& is_flat);
 693   void test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker);
 694 
 695   // Check oops for special arrays, i.e. flat arrays and/or null-free arrays
 696   void test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label);
 697   void test_flat_array_oop(Register klass, Register temp_reg, Label& is_flat_array);
 698   void test_non_flat_array_oop(Register oop, Register temp_reg, Label&is_non_flat_array);
 699   void test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array);
 700   void test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array);
 701 
 702   // Check array klass layout helper for flat or null-free arrays...
 703   void test_flat_array_layout(Register lh, Label& is_flat_array);
 704   void test_non_flat_array_layout(Register lh, Label& is_non_flat_array);
 705 
 706   static address target_addr_for_insn(address insn_addr);
 707   static address target_addr_for_insn_or_null(address insn_addr);
 708 
 709   // Required platform-specific helpers for Label::patch_instructions.
 710   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 711   static int pd_patch_instruction_size(address branch, address target);
 712   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 713     pd_patch_instruction_size(branch, target);
 714   }
 715   static address pd_call_destination(address branch) {
 716     return target_addr_for_insn(branch);
 717   }
 718 #ifndef PRODUCT
 719   static void pd_print_patched_instruction(address branch);
 720 #endif
 721 
 722   static int patch_oop(address insn_addr, address o);
 723   static int patch_narrow_klass(address insn_addr, narrowKlass n);
 724 
 725   // Return whether code is emitted to a scratch blob.
 726   virtual bool in_scratch_emit_size() {
 727     return false;
 728   }
 729   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
 730   static int max_trampoline_stub_size();
 731   void emit_static_call_stub();
 732   static int static_call_stub_size();
 733 
 734   // The following 4 methods return the offset of the appropriate move instruction
 735 
 736   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 737   int load_unsigned_byte(Register dst, Address src);
 738   int load_unsigned_short(Register dst, Address src);
 739 
 740   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 741   int load_signed_byte(Register dst, Address src);
 742   int load_signed_short(Register dst, Address src);
 743 
 744   int load_signed_byte32(Register dst, Address src);
 745   int load_signed_short32(Register dst, Address src);
 746 
 747   // Support for sign-extension (hi:lo = extend_sign(lo))
 748   void extend_sign(Register hi, Register lo);
 749 
 750   // Load and store values by size and signed-ness
 751   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 752   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 753 
 754   // Support for inc/dec with optimal instruction selection depending on value
 755 
 756   // x86_64 aliases an unqualified register/address increment and
 757   // decrement to call incrementq and decrementq but also supports
 758   // explicitly sized calls to incrementq/decrementq or
 759   // incrementl/decrementl
 760 
 761   // for aarch64 the proper convention would be to use
 762   // increment/decrement for 64 bit operations and
 763   // incrementw/decrementw for 32 bit operations. so when porting
 764   // x86_64 code we can leave calls to increment/decrement as is,
 765   // replace incrementq/decrementq with increment/decrement and
 766   // replace incrementl/decrementl with incrementw/decrementw.
 767 
 768   // n.b. increment/decrement calls with an Address destination will
 769   // need to use a scratch register to load the value to be
 770   // incremented. increment/decrement calls which add or subtract a
 771   // constant value greater than 2^12 will need to use a 2nd scratch
 772   // register to hold the constant. so, a register increment/decrement
 773   // may trash rscratch2 and an address increment/decrement trash
 774   // rscratch and rscratch2
 775 
 776   void decrementw(Address dst, int value = 1);
 777   void decrementw(Register reg, int value = 1);
 778 
 779   void decrement(Register reg, int value = 1);
 780   void decrement(Address dst, int value = 1);
 781 
 782   void incrementw(Address dst, int value = 1);
 783   void incrementw(Register reg, int value = 1);
 784 
 785   void increment(Register reg, int value = 1);
 786   void increment(Address dst, int value = 1);
 787 
 788 
 789   // Alignment
 790   void align(int modulus);
 791   void align(int modulus, int target);
 792 
 793   // nop
 794   void post_call_nop();
 795 
 796   // Stack frame creation/removal
 797   void enter(bool strip_ret_addr = false);
 798   void leave();
 799 
 800   // ROP Protection
 801   void protect_return_address();
 802   void protect_return_address(Register return_reg);
 803   void authenticate_return_address();
 804   void authenticate_return_address(Register return_reg);
 805   void strip_return_address();
 806   void check_return_address(Register return_reg=lr) PRODUCT_RETURN;
 807 
 808   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
 809   // The pointer will be loaded into the thread register.
 810   void get_thread(Register thread);
 811 
 812   // support for argument shuffling
 813   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 814   void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 815   void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 816   void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 817   void object_move(
 818                    OopMap* map,
 819                    int oop_handle_offset,
 820                    int framesize_in_slots,
 821                    VMRegPair src,
 822                    VMRegPair dst,
 823                    bool is_receiver,
 824                    int* receiver_offset);
 825 
 826 
 827   // Support for VM calls
 828   //
 829   // It is imperative that all calls into the VM are handled via the call_VM macros.
 830   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 831   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 832 
 833 
 834   void call_VM(Register oop_result,
 835                address entry_point,
 836                bool check_exceptions = true);
 837   void call_VM(Register oop_result,
 838                address entry_point,
 839                Register arg_1,
 840                bool check_exceptions = true);
 841   void call_VM(Register oop_result,
 842                address entry_point,
 843                Register arg_1, Register arg_2,
 844                bool check_exceptions = true);
 845   void call_VM(Register oop_result,
 846                address entry_point,
 847                Register arg_1, Register arg_2, Register arg_3,
 848                bool check_exceptions = true);
 849 
 850   // Overloadings with last_Java_sp
 851   void call_VM(Register oop_result,
 852                Register last_java_sp,
 853                address entry_point,
 854                int number_of_arguments = 0,
 855                bool check_exceptions = true);
 856   void call_VM(Register oop_result,
 857                Register last_java_sp,
 858                address entry_point,
 859                Register arg_1, bool
 860                check_exceptions = true);
 861   void call_VM(Register oop_result,
 862                Register last_java_sp,
 863                address entry_point,
 864                Register arg_1, Register arg_2,
 865                bool check_exceptions = true);
 866   void call_VM(Register oop_result,
 867                Register last_java_sp,
 868                address entry_point,
 869                Register arg_1, Register arg_2, Register arg_3,
 870                bool check_exceptions = true);
 871 
 872   void get_vm_result_oop(Register oop_result, Register thread);
 873   void get_vm_result_metadata(Register metadata_result, Register thread);
 874 
 875   // These always tightly bind to MacroAssembler::call_VM_base
 876   // bypassing the virtual implementation
 877   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 878   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
 879   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 880   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
 881   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true);
 882 
 883   void call_VM_leaf(address entry_point,
 884                     int number_of_arguments = 0);
 885   void call_VM_leaf(address entry_point,
 886                     Register arg_1);
 887   void call_VM_leaf(address entry_point,
 888                     Register arg_1, Register arg_2);
 889   void call_VM_leaf(address entry_point,
 890                     Register arg_1, Register arg_2, Register arg_3);
 891 
 892   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 893   // bypassing the virtual implementation
 894   void super_call_VM_leaf(address entry_point);
 895   void super_call_VM_leaf(address entry_point, Register arg_1);
 896   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 897   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 898   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 899 
 900   // last Java Frame (fills frame anchor)
 901   void set_last_Java_frame(Register last_java_sp,
 902                            Register last_java_fp,
 903                            address last_java_pc,
 904                            Register scratch);
 905 
 906   void set_last_Java_frame(Register last_java_sp,
 907                            Register last_java_fp,
 908                            Label &last_java_pc,
 909                            Register scratch);
 910 
 911   void set_last_Java_frame(Register last_java_sp,
 912                            Register last_java_fp,
 913                            Register last_java_pc,
 914                            Register scratch);
 915 
 916   void reset_last_Java_frame(Register thread);
 917 
 918   // thread in the default location (rthread)
 919   void reset_last_Java_frame(bool clear_fp);
 920 
 921   // Stores
 922   void store_check(Register obj);                // store check for obj - register is destroyed afterwards
 923   void store_check(Register obj, Address dst);   // same as above, dst is exact store location (reg. is destroyed)
 924 
 925   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 926   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 927 
 928   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 929   void c2bool(Register x);
 930 
 931   void load_method_holder_cld(Register rresult, Register rmethod);
 932   void load_method_holder(Register holder, Register method);
 933 
 934   // oop manipulations
 935   void load_metadata(Register dst, Register src);
 936 
 937   void load_narrow_klass_compact(Register dst, Register src);
 938   void load_klass(Register dst, Register src);
 939   void store_klass(Register dst, Register src);
 940   void cmp_klass(Register obj, Register klass, Register tmp);
 941   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 942 
 943   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 944   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 945   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 946 
 947   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 948                       Register tmp1, Register tmp2);
 949 
 950   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 951                        Register tmp1, Register tmp2, Register tmp3);
 952 
 953   void flat_field_copy(DecoratorSet decorators, Register src, Register dst, Register inline_layout_info);
 954 
 955   // inline type data payload offsets...
 956   void payload_offset(Register inline_klass, Register offset);
 957   void payload_address(Register oop, Register data, Register inline_klass);
 958   // get data payload ptr a flat value array at index, kills rcx and index
 959   void data_for_value_array_index(Register array, Register array_klass,
 960                                   Register index, Register data);
 961 
 962   void load_heap_oop(Register dst, Address src, Register tmp1,
 963                      Register tmp2, DecoratorSet decorators = 0);
 964 
 965   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 966                               Register tmp2, DecoratorSet decorators = 0);
 967   void store_heap_oop(Address dst, Register val, Register tmp1,
 968                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 969 
 970   // currently unimplemented
 971   // Used for storing null. All other oop constants should be
 972   // stored using routines that take a jobject.
 973   void store_heap_oop_null(Address dst);
 974 
 975   void load_prototype_header(Register dst, Register src);
 976 
 977   void store_klass_gap(Register dst, Register src);
 978 
 979   // This dummy is to prevent a call to store_heap_oop from
 980   // converting a zero (like null) into a Register by giving
 981   // the compiler two choices it can't resolve
 982 
 983   void store_heap_oop(Address dst, void* dummy);
 984 
 985   void encode_heap_oop(Register d, Register s);
 986   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
 987   void decode_heap_oop(Register d, Register s);
 988   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 989   void encode_heap_oop_not_null(Register r);
 990   void decode_heap_oop_not_null(Register r);
 991   void encode_heap_oop_not_null(Register dst, Register src);
 992   void decode_heap_oop_not_null(Register dst, Register src);
 993 
 994   void set_narrow_oop(Register dst, jobject obj);
 995 
 996   void decode_klass_not_null_for_aot(Register dst, Register src);
 997   void encode_klass_not_null_for_aot(Register dst, Register src);
 998   void encode_klass_not_null(Register r);
 999   void decode_klass_not_null(Register r);
1000   void encode_klass_not_null(Register dst, Register src);
1001   void decode_klass_not_null(Register dst, Register src);
1002 
1003   void set_narrow_klass(Register dst, Klass* k);
1004 
1005   // if heap base register is used - reinit it with the correct value
1006   void reinit_heapbase();
1007 
1008   DEBUG_ONLY(void verify_heapbase(const char* msg);)
1009 
1010   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
1011                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1012   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
1013                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1014 
1015   void push_cont_fastpath(Register java_thread = rthread);
1016   void pop_cont_fastpath(Register java_thread = rthread);
1017 
1018   // Round up to a power of two
1019   void round_to(Register reg, int modulus);
1020 
1021   // java.lang.Math::round intrinsics
1022   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
1023   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
1024 
1025   // allocation
1026 
1027   // Object / value buffer allocation...
1028   // Allocate instance of klass, assumes klass initialized by caller
1029   // new_obj prefers to be rax
1030   // Kills t1 and t2, perserves klass, return allocation in new_obj (rsi on LP64)
1031   void allocate_instance(Register klass, Register new_obj,
1032                          Register t1, Register t2,
1033                          bool clear_fields, Label& alloc_failed);
1034 
1035   void tlab_allocate(
1036     Register obj,                      // result: pointer to object after successful allocation
1037     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1038     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1039     Register t1,                       // temp register
1040     Register t2,                       // temp register
1041     Label&   slow_case                 // continuation point if fast allocation fails
1042   );
1043   void verify_tlab();
1044 
1045   // For field "index" within "klass", return inline_klass ...
1046   void get_inline_type_field_klass(Register klass, Register index, Register inline_klass);
1047   void inline_layout_info(Register holder_klass, Register index, Register layout_info);
1048 
1049 
1050   // interface method calling
1051   void lookup_interface_method(Register recv_klass,
1052                                Register intf_klass,
1053                                RegisterOrConstant itable_index,
1054                                Register method_result,
1055                                Register scan_temp,
1056                                Label& no_such_interface,
1057                    bool return_method = true);
1058 
1059   void lookup_interface_method_stub(Register recv_klass,
1060                                     Register holder_klass,
1061                                     Register resolved_klass,
1062                                     Register method_result,
1063                                     Register temp_reg,
1064                                     Register temp_reg2,
1065                                     int itable_index,
1066                                     Label& L_no_such_interface);
1067 
1068   // virtual method calling
1069   // n.b. x86 allows RegisterOrConstant for vtable_index
1070   void lookup_virtual_method(Register recv_klass,
1071                              RegisterOrConstant vtable_index,
1072                              Register method_result);
1073 
1074   // Test sub_klass against super_klass, with fast and slow paths.
1075 
1076   // The fast path produces a tri-state answer: yes / no / maybe-slow.
1077   // One of the three labels can be null, meaning take the fall-through.
1078   // If super_check_offset is -1, the value is loaded up from super_klass.
1079   // No registers are killed, except temp_reg.
1080   void check_klass_subtype_fast_path(Register sub_klass,
1081                                      Register super_klass,
1082                                      Register temp_reg,
1083                                      Label* L_success,
1084                                      Label* L_failure,
1085                                      Label* L_slow_path,
1086                                      Register super_check_offset = noreg);
1087 
1088   // The rest of the type check; must be wired to a corresponding fast path.
1089   // It does not repeat the fast path logic, so don't use it standalone.
1090   // The temp_reg and temp2_reg can be noreg, if no temps are available.
1091   // Updates the sub's secondary super cache as necessary.
1092   // If set_cond_codes, condition codes will be Z on success, NZ on failure.
1093   void check_klass_subtype_slow_path(Register sub_klass,
1094                                      Register super_klass,
1095                                      Register temp_reg,
1096                                      Register temp2_reg,
1097                                      Label* L_success,
1098                                      Label* L_failure,
1099                                      bool set_cond_codes = false);
1100 
1101   void check_klass_subtype_slow_path_linear(Register sub_klass,
1102                                             Register super_klass,
1103                                             Register temp_reg,
1104                                             Register temp2_reg,
1105                                             Label* L_success,
1106                                             Label* L_failure,
1107                                             bool set_cond_codes = false);
1108 
1109   void check_klass_subtype_slow_path_table(Register sub_klass,
1110                                            Register super_klass,
1111                                            Register temp_reg,
1112                                            Register temp2_reg,
1113                                            Register temp3_reg,
1114                                            Register result_reg,
1115                                            FloatRegister vtemp_reg,
1116                                            Label* L_success,
1117                                            Label* L_failure,
1118                                            bool set_cond_codes = false);
1119 
1120   // If r is valid, return r.
1121   // If r is invalid, remove a register r2 from available_regs, add r2
1122   // to regs_to_push, then return r2.
1123   Register allocate_if_noreg(const Register r,
1124                              RegSetIterator<Register> &available_regs,
1125                              RegSet &regs_to_push);
1126 
1127   // Secondary subtype checking
1128   void lookup_secondary_supers_table_var(Register sub_klass,
1129                                          Register r_super_klass,
1130                                          Register temp1,
1131                                          Register temp2,
1132                                          Register temp3,
1133                                          FloatRegister vtemp,
1134                                          Register result,
1135                                          Label *L_success);
1136 
1137 
1138   // As above, but with a constant super_klass.
1139   // The result is in Register result, not the condition codes.
1140   bool lookup_secondary_supers_table_const(Register r_sub_klass,
1141                                            Register r_super_klass,
1142                                            Register temp1,
1143                                            Register temp2,
1144                                            Register temp3,
1145                                            FloatRegister vtemp,
1146                                            Register result,
1147                                            u1 super_klass_slot,
1148                                            bool stub_is_near = false);
1149 
1150   void verify_secondary_supers_table(Register r_sub_klass,
1151                                      Register r_super_klass,
1152                                      Register temp1,
1153                                      Register temp2,
1154                                      Register result);
1155 
1156   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
1157                                                Register r_array_base,
1158                                                Register r_array_index,
1159                                                Register r_bitmap,
1160                                                Register temp1,
1161                                                Register result,
1162                                                bool is_stub = true);
1163 
1164   // Simplified, combined version, good for typical uses.
1165   // Falls through on failure.
1166   void check_klass_subtype(Register sub_klass,
1167                            Register super_klass,
1168                            Register temp_reg,
1169                            Label& L_success);
1170 
1171   void clinit_barrier(Register klass,
1172                       Register thread,
1173                       Label* L_fast_path = nullptr,
1174                       Label* L_slow_path = nullptr);
1175 
1176   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
1177 
1178   void verify_sve_vector_length(Register tmp = rscratch1);
1179   void reinitialize_ptrue() {
1180     if (UseSVE > 0) {
1181       sve_ptrue(ptrue, B);
1182     }
1183   }
1184   void verify_ptrue();
1185 
1186   // Debugging
1187 
1188   // only if +VerifyOops
1189   void _verify_oop(Register reg, const char* s, const char* file, int line);
1190   void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
1191 
1192   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
1193     if (VerifyOops) {
1194       _verify_oop(reg, s, file, line);
1195     }
1196   }
1197   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
1198     if (VerifyOops) {
1199       _verify_oop_addr(reg, s, file, line);
1200     }
1201   }
1202 
1203 // TODO: verify method and klass metadata (compare against vptr?)
1204   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1205   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1206 
1207 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
1208 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
1209 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
1210 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1211 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1212 
1213   // Restore cpu control state after JNI call
1214   void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2);
1215 
1216   // prints msg, dumps registers and stops execution
1217   void stop(const char* msg);
1218 
1219   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1220 
1221   void untested()                                { stop("untested"); }
1222 
1223   void unimplemented(const char* what = "");
1224 
1225   void should_not_reach_here()                   { stop("should not reach here"); }
1226 
1227   void _assert_asm(Condition cc, const char* msg);
1228 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg)
1229 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg))
1230 
1231   // Stack overflow checking
1232   void bang_stack_with_offset(int offset) {
1233     // stack grows down, caller passes positive offset
1234     assert(offset > 0, "must bang with negative offset");
1235     sub(rscratch2, sp, offset);
1236     str(zr, Address(rscratch2));
1237   }
1238 
1239   // Writes to stack successive pages until offset reached to check for
1240   // stack overflow + shadow pages.  Also, clobbers tmp
1241   void bang_stack_size(Register size, Register tmp);
1242 
1243   // Check for reserved stack access in method being exited (for JIT)
1244   void reserved_stack_check();
1245 
1246   // Arithmetics
1247 
1248   // Clobber: rscratch1, rscratch2
1249   void addptr(const Address &dst, int32_t src);
1250 
1251   // Clobber: rscratch1
1252   void cmpptr(Register src1, Address src2);
1253 
1254   void cmpoop(Register obj1, Register obj2);
1255 
1256   // Various forms of CAS
1257 
1258   void cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp,
1259                           Label &succeed, Label *fail);
1260   void cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
1261                   Label &succeed, Label *fail);
1262 
1263   void cmpxchgw(Register oldv, Register newv, Register addr, Register tmp,
1264                   Label &succeed, Label *fail);
1265 
1266   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1267   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1268   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1269   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1270 
1271   void atomic_xchg(Register prev, Register newv, Register addr);
1272   void atomic_xchgw(Register prev, Register newv, Register addr);
1273   void atomic_xchgl(Register prev, Register newv, Register addr);
1274   void atomic_xchglw(Register prev, Register newv, Register addr);
1275   void atomic_xchgal(Register prev, Register newv, Register addr);
1276   void atomic_xchgalw(Register prev, Register newv, Register addr);
1277 
1278   void orptr(Address adr, RegisterOrConstant src) {
1279     ldr(rscratch1, adr);
1280     if (src.is_register())
1281       orr(rscratch1, rscratch1, src.as_register());
1282     else
1283       orr(rscratch1, rscratch1, src.as_constant());
1284     str(rscratch1, adr);
1285   }
1286 
1287   // A generic CAS; success or failure is in the EQ flag.
1288   // Clobbers rscratch1
1289   void cmpxchg(Register addr, Register expected, Register new_val,
1290                enum operand_size size,
1291                bool acquire, bool release, bool weak,
1292                Register result);
1293 
1294 #ifdef ASSERT
1295   // Template short-hand support to clean-up after a failed call to trampoline
1296   // call generation (see trampoline_call() below),  when a set of Labels must
1297   // be reset (before returning).
1298   template<typename Label, typename... More>
1299   void reset_labels(Label &lbl, More&... more) {
1300     lbl.reset(); reset_labels(more...);
1301   }
1302   template<typename Label>
1303   void reset_labels(Label &lbl) {
1304     lbl.reset();
1305   }
1306 #endif
1307 
1308 private:
1309   void compare_eq(Register rn, Register rm, enum operand_size size);
1310 
1311 public:
1312   // AArch64 OpenJDK uses four different types of calls:
1313   //   - direct call: bl pc_relative_offset
1314   //     This is the shortest and the fastest, but the offset has the range:
1315   //     +/-128MB for the release build, +/-2MB for the debug build.
1316   //
1317   //   - far call: adrp reg, pc_relative_offset; add; bl reg
1318   //     This is longer than a direct call. The offset has
1319   //     the range +/-4GB. As the code cache size is limited to 4GB,
1320   //     far calls can reach anywhere in the code cache. If a jump is
1321   //     needed rather than a call, a far jump 'b reg' can be used instead.
1322   //     All instructions are embedded at a call site.
1323   //
1324   //   - trampoline call:
1325   //     This is only available in C1/C2-generated code (nmethod). It is a combination
1326   //     of a direct call, which is used if the destination of a call is in range,
1327   //     and a register-indirect call. It has the advantages of reaching anywhere in
1328   //     the AArch64 address space and being patchable at runtime when the generated
1329   //     code is being executed by other threads.
1330   //
1331   //     [Main code section]
1332   //       bl trampoline
1333   //     [Stub code section]
1334   //     trampoline:
1335   //       ldr reg, pc + 8
1336   //       br reg
1337   //       <64-bit destination address>
1338   //
1339   //     If the destination is in range when the generated code is moved to the code
1340   //     cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline
1341   //     is not used.
1342   //     The optimization does not remove the trampoline from the stub section.
1343   //     This is necessary because the trampoline may well be redirected later when
1344   //     code is patched, and the new destination may not be reachable by a simple BR
1345   //     instruction.
1346   //
1347   //   - indirect call: move reg, address; blr reg
1348   //     This too can reach anywhere in the address space, but it cannot be
1349   //     patched while code is running, so it must only be modified at a safepoint.
1350   //     This form of call is most suitable for targets at fixed addresses, which
1351   //     will never be patched.
1352   //
1353   // The patching we do conforms to the "Concurrent modification and
1354   // execution of instructions" section of the Arm Architectural
1355   // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC,
1356   // or SVC instructions to be modified while another thread is
1357   // executing them.
1358   //
1359   // To patch a trampoline call when the BL can't reach, we first modify
1360   // the 64-bit destination address in the trampoline, then modify the
1361   // BL to point to the trampoline, then flush the instruction cache to
1362   // broadcast the change to all executing threads. See
1363   // NativeCall::set_destination_mt_safe for the details.
1364   //
1365   // There is a benign race in that the other thread might observe the
1366   // modified BL before it observes the modified 64-bit destination
1367   // address. That does not matter because the destination method has been
1368   // invalidated, so there will be a trap at its start.
1369   // For this to work, the destination address in the trampoline is
1370   // always updated, even if we're not using the trampoline.
1371 
1372   // Emit a direct call if the entry address will always be in range,
1373   // otherwise a trampoline call.
1374   // Supported entry.rspec():
1375   // - relocInfo::runtime_call_type
1376   // - relocInfo::opt_virtual_call_type
1377   // - relocInfo::static_call_type
1378   // - relocInfo::virtual_call_type
1379   //
1380   // Return: the call PC or null if CodeCache is full.
1381   // Clobbers: rscratch1
1382   address trampoline_call(Address entry);
1383 
1384   static bool far_branches() {
1385     return ReservedCodeCacheSize > branch_range;
1386   }
1387 
1388   // Check if branches to the non nmethod section require a far jump
1389   static bool codestub_branch_needs_far_jump() {
1390     if (AOTCodeCache::is_on_for_dump()) {
1391       // To calculate far_codestub_branch_size correctly.
1392       return true;
1393     }
1394     return CodeCache::max_distance_to_non_nmethod() > branch_range;
1395   }
1396 
1397   // Emit a direct call/jump if the entry address will always be in range,
1398   // otherwise a far call/jump.
1399   // The address must be inside the code cache.
1400   // Supported entry.rspec():
1401   // - relocInfo::external_word_type
1402   // - relocInfo::runtime_call_type
1403   // - relocInfo::none
1404   // In the case of a far call/jump, the entry address is put in the tmp register.
1405   // The tmp register is invalidated.
1406   //
1407   // Far_jump returns the amount of the emitted code.
1408   void far_call(Address entry, Register tmp = rscratch1);
1409   int far_jump(Address entry, Register tmp = rscratch1);
1410 
1411   static int far_codestub_branch_size() {
1412     if (codestub_branch_needs_far_jump()) {
1413       return 3 * 4;  // adrp, add, br
1414     } else {
1415       return 4;
1416     }
1417   }
1418 
1419   // Emit the CompiledIC call idiom
1420   address ic_call(address entry, jint method_index = 0);
1421   static int ic_check_size();
1422   int ic_check(int end_alignment);
1423 
1424 public:
1425 
1426   // Data
1427 
1428   void mov_metadata(Register dst, Metadata* obj);
1429   Address allocate_metadata_address(Metadata* obj);
1430   Address constant_oop_address(jobject obj);
1431 
1432   void movoop(Register dst, jobject obj);
1433 
1434   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1435   void kernel_crc32(Register crc, Register buf, Register len,
1436         Register table0, Register table1, Register table2, Register table3,
1437         Register tmp, Register tmp2, Register tmp3);
1438   // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic.
1439   void kernel_crc32c(Register crc, Register buf, Register len,
1440         Register table0, Register table1, Register table2, Register table3,
1441         Register tmp, Register tmp2, Register tmp3);
1442 
1443   // Stack push and pop individual 64 bit registers
1444   void push(Register src);
1445   void pop(Register dst);
1446 
1447   void repne_scan(Register addr, Register value, Register count,
1448                   Register scratch);
1449   void repne_scanw(Register addr, Register value, Register count,
1450                    Register scratch);
1451 
1452   typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1453   typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1454 
1455   // If a constant does not fit in an immediate field, generate some
1456   // number of MOV instructions and then perform the operation
1457   void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
1458                              add_sub_imm_insn insn1,
1459                              add_sub_reg_insn insn2, bool is32);
1460   // Separate vsn which sets the flags
1461   void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
1462                                add_sub_imm_insn insn1,
1463                                add_sub_reg_insn insn2, bool is32);
1464 
1465 #define WRAP(INSN, is32)                                                \
1466   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1467     wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1468   }                                                                     \
1469                                                                         \
1470   void INSN(Register Rd, Register Rn, Register Rm,                      \
1471              enum shift_kind kind, unsigned shift = 0) {                \
1472     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1473   }                                                                     \
1474                                                                         \
1475   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1476     Assembler::INSN(Rd, Rn, Rm);                                        \
1477   }                                                                     \
1478                                                                         \
1479   void INSN(Register Rd, Register Rn, Register Rm,                      \
1480            ext::operation option, int amount = 0) {                     \
1481     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1482   }
1483 
1484   WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true)
1485 
1486 #undef WRAP
1487 #define WRAP(INSN, is32)                                                \
1488   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1489     wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1490   }                                                                     \
1491                                                                         \
1492   void INSN(Register Rd, Register Rn, Register Rm,                      \
1493              enum shift_kind kind, unsigned shift = 0) {                \
1494     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1495   }                                                                     \
1496                                                                         \
1497   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1498     Assembler::INSN(Rd, Rn, Rm);                                        \
1499   }                                                                     \
1500                                                                         \
1501   void INSN(Register Rd, Register Rn, Register Rm,                      \
1502            ext::operation option, int amount = 0) {                     \
1503     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1504   }
1505 
1506   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1507 
1508   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1509   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1510   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1511   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1512 
1513   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1514 
1515   void verified_entry(Compile* C, int sp_inc);
1516 
1517   // Inline type specific methods
1518   #include "asm/macroAssembler_common.hpp"
1519 
1520   int store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter = true);
1521   bool move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]);
1522   bool unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
1523                             VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
1524                             RegState reg_state[]);
1525   bool pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
1526                           VMRegPair* from, int from_count, int& from_index, VMReg to,
1527                           RegState reg_state[], Register val_array);
1528   int extend_stack_for_inline_args(int args_on_stack);
1529   void remove_frame(int initial_framesize, bool needs_stack_repair);
1530   VMReg spill_reg_for(VMReg reg);
1531   void save_stack_increment(int sp_inc, int frame_size);
1532 
1533   void tableswitch(Register index, jint lowbound, jint highbound,
1534                    Label &jumptable, Label &jumptable_end, int stride = 1) {
1535     adr(rscratch1, jumptable);
1536     subsw(rscratch2, index, lowbound);
1537     subsw(zr, rscratch2, highbound - lowbound);
1538     br(Assembler::HS, jumptable_end);
1539     add(rscratch1, rscratch1, rscratch2,
1540         ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1541     br(rscratch1);
1542   }
1543 
1544   // Form an address from base + offset in Rd.  Rd may or may not
1545   // actually be used: you must use the Address that is returned.  It
1546   // is up to you to ensure that the shift provided matches the size
1547   // of your data.
1548   Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1549 
1550   // Return true iff an address is within the 48-bit AArch64 address
1551   // space.
1552   bool is_valid_AArch64_address(address a) {
1553     return ((uint64_t)a >> 48) == 0;
1554   }
1555 
1556   // Load the base of the cardtable byte map into reg.
1557   void load_byte_map_base(Register reg);
1558 
1559   // Prolog generator routines to support switch between x86 code and
1560   // generated ARM code
1561 
1562   // routine to generate an x86 prolog for a stub function which
1563   // bootstraps into the generated ARM code which directly follows the
1564   // stub
1565   //
1566 
1567   public:
1568 
1569   address read_polling_page(Register r, relocInfo::relocType rtype);
1570   void get_polling_page(Register dest, relocInfo::relocType rtype);
1571 
1572   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1573   void update_byte_crc32(Register crc, Register val, Register table);
1574   void update_word_crc32(Register crc, Register v, Register tmp,
1575         Register table0, Register table1, Register table2, Register table3,
1576         bool upper = false);
1577 
1578   address count_positives(Register ary1, Register len, Register result);
1579 
1580   address arrays_equals(Register a1, Register a2, Register result, Register cnt1,
1581                         Register tmp1, Register tmp2, Register tmp3, int elem_size);
1582 
1583 // Ensure that the inline code and the stub use the same registers.
1584 #define ARRAYS_HASHCODE_REGISTERS \
1585   do {                      \
1586     assert(result == r0  && \
1587            ary    == r1  && \
1588            cnt    == r2  && \
1589            vdata0 == v3  && \
1590            vdata1 == v2  && \
1591            vdata2 == v1  && \
1592            vdata3 == v0  && \
1593            vmul0  == v4  && \
1594            vmul1  == v5  && \
1595            vmul2  == v6  && \
1596            vmul3  == v7  && \
1597            vpow   == v12 && \
1598            vpowm  == v13, "registers must match aarch64.ad"); \
1599   } while (0)
1600 
1601   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1602 
1603   void fill_words(Register base, Register cnt, Register value);
1604   void fill_words(Register base, uint64_t cnt, Register value);
1605 
1606   address zero_words(Register base, uint64_t cnt);
1607   address zero_words(Register ptr, Register cnt);
1608   void zero_dcache_blocks(Register base, Register cnt);
1609 
1610   static const int zero_words_block_size;
1611 
1612   address byte_array_inflate(Register src, Register dst, Register len,
1613                              FloatRegister vtmp1, FloatRegister vtmp2,
1614                              FloatRegister vtmp3, Register tmp4);
1615 
1616   void char_array_compress(Register src, Register dst, Register len,
1617                            Register res,
1618                            FloatRegister vtmp0, FloatRegister vtmp1,
1619                            FloatRegister vtmp2, FloatRegister vtmp3,
1620                            FloatRegister vtmp4, FloatRegister vtmp5);
1621 
1622   void encode_iso_array(Register src, Register dst,
1623                         Register len, Register res, bool ascii,
1624                         FloatRegister vtmp0, FloatRegister vtmp1,
1625                         FloatRegister vtmp2, FloatRegister vtmp3,
1626                         FloatRegister vtmp4, FloatRegister vtmp5);
1627 
1628   void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi,
1629       address pio2, address dsin_coef, address dcos_coef);
1630  private:
1631   // begin trigonometric functions support block
1632   void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2);
1633   void generate__kernel_rem_pio2(address two_over_pi, address pio2);
1634   void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef);
1635   void generate_kernel_cos(FloatRegister x, address dcos_coef);
1636   // end trigonometric functions support block
1637   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1638                        Register src1, Register src2);
1639   void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1640     add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1641   }
1642   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1643                              Register y, Register y_idx, Register z,
1644                              Register carry, Register product,
1645                              Register idx, Register kdx);
1646   void multiply_128_x_128_loop(Register y, Register z,
1647                                Register carry, Register carry2,
1648                                Register idx, Register jdx,
1649                                Register yz_idx1, Register yz_idx2,
1650                                Register tmp, Register tmp3, Register tmp4,
1651                                Register tmp7, Register product_hi);
1652   void kernel_crc32_using_crypto_pmull(Register crc, Register buf,
1653         Register len, Register tmp0, Register tmp1, Register tmp2,
1654         Register tmp3);
1655   void kernel_crc32_using_crc32(Register crc, Register buf,
1656         Register len, Register tmp0, Register tmp1, Register tmp2,
1657         Register tmp3);
1658   void kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
1659         Register len, Register tmp0, Register tmp1, Register tmp2,
1660         Register tmp3);
1661   void kernel_crc32c_using_crc32c(Register crc, Register buf,
1662         Register len, Register tmp0, Register tmp1, Register tmp2,
1663         Register tmp3);
1664   void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
1665         Register len, Register tmp0, Register tmp1, Register tmp2,
1666         size_t table_offset);
1667 
1668   void ghash_modmul (FloatRegister result,
1669                      FloatRegister result_lo, FloatRegister result_hi, FloatRegister b,
1670                      FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p,
1671                      FloatRegister t1, FloatRegister t2, FloatRegister t3);
1672   void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
1673 public:
1674   void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1675                        Register tmp0, Register tmp1, Register tmp2, Register tmp3,
1676                        Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1677   void mul_add(Register out, Register in, Register offs, Register len, Register k);
1678   void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,
1679                       FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1680                       FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1681   void ghash_multiply_wide(int index,
1682                            FloatRegister result_lo, FloatRegister result_hi,
1683                            FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1684                            FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1685   void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi,
1686                     FloatRegister p, FloatRegister z, FloatRegister t1);
1687   void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi,
1688                     FloatRegister p, FloatRegister z, FloatRegister t1);
1689   void ghash_processBlocks_wide(Label& p, Register state, Register subkeyH,
1690                                 Register data, Register blocks, int unrolls);
1691 
1692 
1693   void aesenc_loadkeys(Register key, Register keylen);
1694   void aesecb_encrypt(Register from, Register to, Register keylen,
1695                       FloatRegister data = v0, int unrolls = 1);
1696   void aesecb_decrypt(Register from, Register to, Register key, Register keylen);
1697   void aes_round(FloatRegister input, FloatRegister subkey);
1698 
1699   // ChaCha20 functions support block
1700   void cc20_qr_add4(FloatRegister (&addFirst)[4],
1701           FloatRegister (&addSecond)[4]);
1702   void cc20_qr_xor4(FloatRegister (&firstElem)[4],
1703           FloatRegister (&secondElem)[4], FloatRegister (&result)[4]);
1704   void cc20_qr_lrot4(FloatRegister (&sourceReg)[4],
1705           FloatRegister (&destReg)[4], int bits, FloatRegister table);
1706   void cc20_set_qr_registers(FloatRegister (&vectorSet)[4],
1707           const FloatRegister (&stateVectors)[16], int idx1, int idx2,
1708           int idx3, int idx4);
1709 
1710   // Place an ISB after code may have been modified due to a safepoint.
1711   void safepoint_isb();
1712 
1713 private:
1714   // Return the effective address r + (r1 << ext) + offset.
1715   // Uses rscratch2.
1716   Address offsetted_address(Register r, Register r1, Address::extend ext,
1717                             int offset, int size);
1718 
1719 private:
1720   // Returns an address on the stack which is reachable with a ldr/str of size
1721   // Uses rscratch2 if the address is not directly reachable
1722   Address spill_address(int size, int offset, Register tmp=rscratch2);
1723   Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2);
1724 
1725   bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const;
1726 
1727   // Check whether two loads/stores can be merged into ldp/stp.
1728   bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const;
1729 
1730   // Merge current load/store with previous load/store into ldp/stp.
1731   void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1732 
1733   // Try to merge two loads/stores into ldp/stp. If success, returns true else false.
1734   bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1735 
1736 public:
1737   void spill(Register Rx, bool is64, int offset) {
1738     if (is64) {
1739       str(Rx, spill_address(8, offset));
1740     } else {
1741       strw(Rx, spill_address(4, offset));
1742     }
1743   }
1744   void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1745     str(Vx, T, spill_address(1 << (int)T, offset));
1746   }
1747 
1748   void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1749     sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1750   }
1751   void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1752     sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1753   }
1754 
1755   void unspill(Register Rx, bool is64, int offset) {
1756     if (is64) {
1757       ldr(Rx, spill_address(8, offset));
1758     } else {
1759       ldrw(Rx, spill_address(4, offset));
1760     }
1761   }
1762   void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1763     ldr(Vx, T, spill_address(1 << (int)T, offset));
1764   }
1765 
1766   void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1767     sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1768   }
1769   void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1770     sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1771   }
1772 
1773   void spill_copy128(int src_offset, int dst_offset,
1774                      Register tmp1=rscratch1, Register tmp2=rscratch2) {
1775     if (src_offset < 512 && (src_offset & 7) == 0 &&
1776         dst_offset < 512 && (dst_offset & 7) == 0) {
1777       ldp(tmp1, tmp2, Address(sp, src_offset));
1778       stp(tmp1, tmp2, Address(sp, dst_offset));
1779     } else {
1780       unspill(tmp1, true, src_offset);
1781       spill(tmp1, true, dst_offset);
1782       unspill(tmp1, true, src_offset+8);
1783       spill(tmp1, true, dst_offset+8);
1784     }
1785   }
1786   void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset,
1787                                             int sve_vec_reg_size_in_bytes) {
1788     assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size");
1789     for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) {
1790       spill_copy128(src_offset, dst_offset);
1791       src_offset += 16;
1792       dst_offset += 16;
1793     }
1794   }
1795   void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset,
1796                                                int sve_predicate_reg_size_in_bytes) {
1797     sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset));
1798     sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset));
1799     reinitialize_ptrue();
1800   }
1801   void cache_wb(Address line);
1802   void cache_wbsync(bool is_pre);
1803 
1804   // Code for java.lang.Thread::onSpinWait() intrinsic.
1805   void spin_wait();
1806 
1807   void lightweight_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow);
1808   void lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1809 
1810 private:
1811   // Check the current thread doesn't need a cross modify fence.
1812   void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
1813 
1814 };
1815 
1816 #ifdef ASSERT
1817 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1818 #endif
1819 
1820 struct tableswitch {
1821   Register _reg;
1822   int _insn_index; jint _first_key; jint _last_key;
1823   Label _after;
1824   Label _branches;
1825 };
1826 
1827 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP