1 /*
   2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #ifndef CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  27 #define CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  28 
  29 #include "asm/assembler.inline.hpp"
  30 #include "code/aotCodeCache.hpp"
  31 #include "code/vmreg.hpp"
  32 #include "metaprogramming/enableIf.hpp"
  33 #include "oops/compressedOops.hpp"
  34 #include "oops/compressedKlass.hpp"
  35 #include "runtime/atomicAccess.hpp"
  36 #include "runtime/vm_version.hpp"
  37 #include "utilities/globalDefinitions.hpp"
  38 #include "utilities/macros.hpp"
  39 #include "utilities/powerOfTwo.hpp"
  40 #include "runtime/signature.hpp"
  41 
  42 
  43 class ciInlineKlass;
  44 
  45 class OopMap;
  46 struct GtestFriendToMacroAssembler;
  47 
  48 // MacroAssembler extends Assembler by frequently used macros.
  49 //
  50 // Instructions for which a 'better' code sequence exists depending
  51 // on arguments should also go in here.
  52 
  53 class MacroAssembler: public Assembler {
  54   friend class LIR_Assembler;
  55   friend struct GtestFriendToMacroAssembler;
  56 
  57  public:
  58   using Assembler::mov;
  59   using Assembler::movi;
  60 
  61  protected:
  62 
  63   // Support for VM calls
  64   //
  65   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
  66   // may customize this version by overriding it for its purposes (e.g., to save/restore
  67   // additional registers when doing a VM call).
  68   virtual void call_VM_leaf_base(
  69     address entry_point,               // the entry point
  70     int     number_of_arguments,        // the number of arguments to pop after the call
  71     Label *retaddr = nullptr
  72   );
  73 
  74   virtual void call_VM_leaf_base(
  75     address entry_point,               // the entry point
  76     int     number_of_arguments,        // the number of arguments to pop after the call
  77     Label &retaddr) {
  78     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
  79   }
  80 
  81   // This is the base routine called by the different versions of call_VM. The interpreter
  82   // may customize this version by overriding it for its purposes (e.g., to save/restore
  83   // additional registers when doing a VM call).
  84   //
  85   // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
  86   // returns the register which contains the thread upon return. If a thread register has been
  87   // specified, the return value will correspond to that register. If no last_java_sp is specified
  88   // (noreg) than rsp will be used instead.
  89   virtual void call_VM_base(           // returns the register containing the thread upon return
  90     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
  91     Register java_thread,              // the thread if computed before     ; use noreg otherwise
  92     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
  93     Label*   return_pc,                // to set up last_Java_frame; use nullptr otherwise
  94     address  entry_point,              // the entry point
  95     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
  96     bool     check_exceptions          // whether to check for pending exceptions after return
  97   );
  98 
  99   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
 100 
 101  private:
 102 
 103   enum KlassDecodeMode {
 104     KlassDecodeNone,
 105     KlassDecodeZero,
 106     KlassDecodeXor,
 107     KlassDecodeMovk,
 108     KlassDecodeFallback
 109   };
 110 
 111   static KlassDecodeMode _klass_decode_mode;
 112 
 113   // Returns above setting with asserts
 114   static KlassDecodeMode klass_decode_mode();
 115 
 116   // Calculate decoding mode based on given parameters, used for checking then ultimately setting.
 117   static KlassDecodeMode klass_decode_mode(address base, int shift, const size_t range);
 118 
 119   void emit_encode_klass_not_null(Register dst, Register src, Register tmp,
 120                                   address base, int shift, KlassDecodeMode decode_mode);
 121   void emit_decode_klass_not_null(Register dst, Register src, Register tmp,
 122                                   address base, int shift, KlassDecodeMode decode_mode);
 123  public:
 124   // Determines the decode mode best suited for the given encoding parameters.
 125   static void initialize_klass_decode_mode(address base, int shift, const size_t range);
 126 
 127  public:
 128   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 129 
 130  // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 131  // The implementation is only non-empty for the InterpreterMacroAssembler,
 132  // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 133  virtual void check_and_handle_popframe(Register java_thread);
 134  virtual void check_and_handle_earlyret(Register java_thread);
 135 
 136   void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp = rscratch1);
 137   void rt_call(address dest, Register tmp = rscratch1);
 138 
 139   // Load Effective Address
 140   void lea(Register r, const Address &a) {
 141     InstructionMark im(this);
 142     a.lea(this, r);
 143   }
 144 
 145   // Whether materializing the given address for a LDR/STR requires an
 146   // additional lea instruction.
 147   static bool legitimize_address_requires_lea(const Address &a, int size) {
 148     return a.getMode() == Address::base_plus_offset &&
 149            !Address::offset_ok_for_immed(a.offset(), exact_log2(size));
 150   }
 151 
 152   /* Sometimes we get misaligned loads and stores, usually from Unsafe
 153      accesses, and these can exceed the offset range. */
 154   Address legitimize_address(const Address &a, int size, Register scratch) {
 155     if (legitimize_address_requires_lea(a, size)) {
 156       block_comment("legitimize_address {");
 157       lea(scratch, a);
 158       block_comment("} legitimize_address");
 159       return Address(scratch);
 160     }
 161     return a;
 162   }
 163 
 164   void addmw(Address a, Register incr, Register scratch) {
 165     ldrw(scratch, a);
 166     addw(scratch, scratch, incr);
 167     strw(scratch, a);
 168   }
 169 
 170   // Add constant to memory word
 171   void addmw(Address a, int imm, Register scratch) {
 172     ldrw(scratch, a);
 173     if (imm > 0)
 174       addw(scratch, scratch, (unsigned)imm);
 175     else
 176       subw(scratch, scratch, (unsigned)-imm);
 177     strw(scratch, a);
 178   }
 179 
 180   void bind(Label& L) {
 181     Assembler::bind(L);
 182     code()->clear_last_merge_candidate();
 183     code()->set_last_label(pc());
 184   }
 185 
 186   void membar(Membar_mask_bits order_constraint);
 187 
 188   using Assembler::ldr;
 189   using Assembler::str;
 190   using Assembler::ldrw;
 191   using Assembler::strw;
 192 
 193   void ldr(Register Rx, const Address &adr);
 194   void ldrw(Register Rw, const Address &adr);
 195   void str(Register Rx, const Address &adr);
 196   void strw(Register Rx, const Address &adr);
 197 
 198   // Frame creation and destruction shared between JITs.
 199   DEBUG_ONLY(void build_frame(int framesize);)
 200   void build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills));
 201   void remove_frame(int framesize);
 202 
 203   virtual void _call_Unimplemented(address call_site) {
 204     mov(rscratch2, call_site);
 205   }
 206 
 207 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 208 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 209 // https://reviews.llvm.org/D3311
 210 
 211 #ifdef _WIN64
 212 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 213 #else
 214 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 215 #endif
 216 
 217   // aliases defined in AARCH64 spec
 218 
 219   template<class T>
 220   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }
 221 
 222   inline void cmp(Register Rd, unsigned char imm8)  { subs(zr, Rd, imm8); }
 223   inline void cmp(Register Rd, unsigned imm) = delete;
 224 
 225   template<class T>
 226   inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); }
 227 
 228   inline void cmn(Register Rd, unsigned char imm8)  { adds(zr, Rd, imm8); }
 229   inline void cmn(Register Rd, unsigned imm) = delete;
 230 
 231   void cset(Register Rd, Assembler::Condition cond) {
 232     csinc(Rd, zr, zr, ~cond);
 233   }
 234   void csetw(Register Rd, Assembler::Condition cond) {
 235     csincw(Rd, zr, zr, ~cond);
 236   }
 237 
 238   void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
 239     csneg(Rd, Rn, Rn, ~cond);
 240   }
 241   void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
 242     csnegw(Rd, Rn, Rn, ~cond);
 243   }
 244 
 245   inline void movw(Register Rd, Register Rn) {
 246     if (Rd == sp || Rn == sp) {
 247       Assembler::addw(Rd, Rn, 0U);
 248     } else {
 249       orrw(Rd, zr, Rn);
 250     }
 251   }
 252   inline void mov(Register Rd, Register Rn) {
 253     assert(Rd != r31_sp && Rn != r31_sp, "should be");
 254     if (Rd == Rn) {
 255     } else if (Rd == sp || Rn == sp) {
 256       Assembler::add(Rd, Rn, 0U);
 257     } else {
 258       orr(Rd, zr, Rn);
 259     }
 260   }
 261 
 262   inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
 263   inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
 264 
 265   inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
 266   inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
 267 
 268   inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
 269   inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
 270 
 271   inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 272     bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 273   }
 274   inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 275     bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 276   }
 277 
 278   inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 279     bfmw(Rd, Rn, lsb, (lsb + width - 1));
 280   }
 281   inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 282     bfm(Rd, Rn, lsb , (lsb + width - 1));
 283   }
 284 
 285   inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 286     sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 287   }
 288   inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 289     sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 290   }
 291 
 292   inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 293     sbfmw(Rd, Rn, lsb, (lsb + width - 1));
 294   }
 295   inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 296     sbfm(Rd, Rn, lsb , (lsb + width - 1));
 297   }
 298 
 299   inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 300     ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 301   }
 302   inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 303     ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 304   }
 305 
 306   inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 307     ubfmw(Rd, Rn, lsb, (lsb + width - 1));
 308   }
 309   inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 310     ubfm(Rd, Rn, lsb , (lsb + width - 1));
 311   }
 312 
 313   inline void asrw(Register Rd, Register Rn, unsigned imm) {
 314     sbfmw(Rd, Rn, imm, 31);
 315   }
 316 
 317   inline void asr(Register Rd, Register Rn, unsigned imm) {
 318     sbfm(Rd, Rn, imm, 63);
 319   }
 320 
 321   inline void lslw(Register Rd, Register Rn, unsigned imm) {
 322     if (imm > 0 || Rd != Rn) {
 323       ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
 324     }
 325   }
 326 
 327   inline void lsl(Register Rd, Register Rn, unsigned imm) {
 328     if (imm > 0 || Rd != Rn) {
 329       ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
 330     }
 331   }
 332 
 333   inline void lsrw(Register Rd, Register Rn, unsigned imm) {
 334     if (imm > 0 || Rd != Rn) {
 335       ubfmw(Rd, Rn, imm, 31);
 336     }
 337   }
 338 
 339   inline void lsr(Register Rd, Register Rn, unsigned imm) {
 340     if (imm > 0 || Rd != Rn) {
 341       ubfm(Rd, Rn, imm, 63);
 342     }
 343   }
 344 
 345   inline void rorw(Register Rd, Register Rn, unsigned imm) {
 346     extrw(Rd, Rn, Rn, imm);
 347   }
 348 
 349   inline void ror(Register Rd, Register Rn, unsigned imm) {
 350     extr(Rd, Rn, Rn, imm);
 351   }
 352 
 353   inline void rolw(Register Rd, Register Rn, unsigned imm) {
 354     extrw(Rd, Rn, Rn, (32 - imm));
 355   }
 356 
 357   inline void rol(Register Rd, Register Rn, unsigned imm) {
 358     extr(Rd, Rn, Rn, (64 - imm));
 359   }
 360 
 361   using Assembler::rax1;
 362   using Assembler::eor3;
 363 
 364   inline void rax1(Register Rd, Register Rn, Register Rm) {
 365     eor(Rd, Rn, Rm, ROR, 63); // Rd = Rn ^ rol(Rm, 1)
 366   }
 367 
 368   inline void eor3(Register Rd, Register Rn, Register Rm, Register Rk) {
 369     assert(Rd != Rn, "Use tmp register");
 370     eor(Rd, Rm, Rk);
 371     eor(Rd, Rd, Rn);
 372   }
 373 
 374   inline void sxtbw(Register Rd, Register Rn) {
 375     sbfmw(Rd, Rn, 0, 7);
 376   }
 377   inline void sxthw(Register Rd, Register Rn) {
 378     sbfmw(Rd, Rn, 0, 15);
 379   }
 380   inline void sxtb(Register Rd, Register Rn) {
 381     sbfm(Rd, Rn, 0, 7);
 382   }
 383   inline void sxth(Register Rd, Register Rn) {
 384     sbfm(Rd, Rn, 0, 15);
 385   }
 386   inline void sxtw(Register Rd, Register Rn) {
 387     sbfm(Rd, Rn, 0, 31);
 388   }
 389 
 390   inline void uxtbw(Register Rd, Register Rn) {
 391     ubfmw(Rd, Rn, 0, 7);
 392   }
 393   inline void uxthw(Register Rd, Register Rn) {
 394     ubfmw(Rd, Rn, 0, 15);
 395   }
 396   inline void uxtb(Register Rd, Register Rn) {
 397     ubfm(Rd, Rn, 0, 7);
 398   }
 399   inline void uxth(Register Rd, Register Rn) {
 400     ubfm(Rd, Rn, 0, 15);
 401   }
 402   inline void uxtw(Register Rd, Register Rn) {
 403     ubfm(Rd, Rn, 0, 31);
 404   }
 405 
 406   inline void cmnw(Register Rn, Register Rm) {
 407     addsw(zr, Rn, Rm);
 408   }
 409   inline void cmn(Register Rn, Register Rm) {
 410     adds(zr, Rn, Rm);
 411   }
 412 
 413   inline void cmpw(Register Rn, Register Rm) {
 414     subsw(zr, Rn, Rm);
 415   }
 416   inline void cmp(Register Rn, Register Rm) {
 417     subs(zr, Rn, Rm);
 418   }
 419 
 420   inline void negw(Register Rd, Register Rn) {
 421     subw(Rd, zr, Rn);
 422   }
 423 
 424   inline void neg(Register Rd, Register Rn) {
 425     sub(Rd, zr, Rn);
 426   }
 427 
 428   inline void negsw(Register Rd, Register Rn) {
 429     subsw(Rd, zr, Rn);
 430   }
 431 
 432   inline void negs(Register Rd, Register Rn) {
 433     subs(Rd, zr, Rn);
 434   }
 435 
 436   inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 437     addsw(zr, Rn, Rm, kind, shift);
 438   }
 439   inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 440     adds(zr, Rn, Rm, kind, shift);
 441   }
 442 
 443   inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 444     subsw(zr, Rn, Rm, kind, shift);
 445   }
 446   inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 447     subs(zr, Rn, Rm, kind, shift);
 448   }
 449 
 450   inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 451     subw(Rd, zr, Rn, kind, shift);
 452   }
 453 
 454   inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 455     sub(Rd, zr, Rn, kind, shift);
 456   }
 457 
 458   inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 459     subsw(Rd, zr, Rn, kind, shift);
 460   }
 461 
 462   inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 463     subs(Rd, zr, Rn, kind, shift);
 464   }
 465 
 466   inline void mnegw(Register Rd, Register Rn, Register Rm) {
 467     msubw(Rd, Rn, Rm, zr);
 468   }
 469   inline void mneg(Register Rd, Register Rn, Register Rm) {
 470     msub(Rd, Rn, Rm, zr);
 471   }
 472 
 473   inline void mulw(Register Rd, Register Rn, Register Rm) {
 474     maddw(Rd, Rn, Rm, zr);
 475   }
 476   inline void mul(Register Rd, Register Rn, Register Rm) {
 477     madd(Rd, Rn, Rm, zr);
 478   }
 479 
 480   inline void smnegl(Register Rd, Register Rn, Register Rm) {
 481     smsubl(Rd, Rn, Rm, zr);
 482   }
 483   inline void smull(Register Rd, Register Rn, Register Rm) {
 484     smaddl(Rd, Rn, Rm, zr);
 485   }
 486 
 487   inline void umnegl(Register Rd, Register Rn, Register Rm) {
 488     umsubl(Rd, Rn, Rm, zr);
 489   }
 490   inline void umull(Register Rd, Register Rn, Register Rm) {
 491     umaddl(Rd, Rn, Rm, zr);
 492   }
 493 
 494 #define WRAP(INSN)                                                            \
 495   void INSN(Register Rd, Register Rn, Register Rm, Register Ra) {             \
 496     if (VM_Version::supports_a53mac() && Ra != zr)                            \
 497       nop();                                                                  \
 498     Assembler::INSN(Rd, Rn, Rm, Ra);                                          \
 499   }
 500 
 501   WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
 502   WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
 503 #undef WRAP
 504 
 505   using Assembler::andw, Assembler::andr;
 506   void andw(Register Rd, Register Rn, uint64_t imm) {
 507     if (operand_valid_for_logical_immediate(/*is32*/true, imm)) {
 508       Assembler::andw(Rd, Rn, imm);
 509     } else {
 510       assert(Rd != Rn, "must be");
 511       movw(Rd, imm);
 512       andw(Rd, Rn, Rd);
 513     }
 514   }
 515   void andr(Register Rd, Register Rn, uint64_t imm) {
 516     if (operand_valid_for_logical_immediate(/*is32*/false, imm)) {
 517       Assembler::andr(Rd, Rn, imm);
 518     } else {
 519       assert(Rd != Rn, "must be");
 520       mov(Rd, imm);
 521       andr(Rd, Rn, Rd);
 522     }
 523   }
 524 
 525   // macro assembly operations needed for aarch64
 526 
 527 public:
 528 
 529   enum FpPushPopMode {
 530     PushPopFull,
 531     PushPopSVE,
 532     PushPopNeon,
 533     PushPopFp
 534   };
 535 
 536   // first two private routines for loading 32 bit or 64 bit constants
 537 private:
 538 
 539   void mov_immediate64(Register dst, uint64_t imm64);
 540   void mov_immediate32(Register dst, uint32_t imm32);
 541 
 542   void mov(Register dst, Address a);
 543 
 544 public:
 545 
 546   int push(RegSet regset, Register stack);
 547   int pop(RegSet regset, Register stack);
 548 
 549   int push_fp(FloatRegSet regset, Register stack, FpPushPopMode mode = PushPopFull);
 550   int pop_fp(FloatRegSet regset, Register stack, FpPushPopMode mode = PushPopFull);
 551 
 552   static RegSet call_clobbered_gp_registers();
 553 
 554   int push_p(PRegSet regset, Register stack);
 555   int pop_p(PRegSet regset, Register stack);
 556 
 557   // Push and pop everything that might be clobbered by a native
 558   // runtime call except rscratch1 and rscratch2.  (They are always
 559   // scratch, so we don't have to protect them.)  Only save the lower
 560   // 64 bits of each vector register. Additional registers can be excluded
 561   // in a passed RegSet.
 562   void push_call_clobbered_registers_except(RegSet exclude);
 563   void pop_call_clobbered_registers_except(RegSet exclude);
 564 
 565   void push_call_clobbered_registers() {
 566     push_call_clobbered_registers_except(RegSet());
 567   }
 568   void pop_call_clobbered_registers() {
 569     pop_call_clobbered_registers_except(RegSet());
 570   }
 571 
 572 
 573   // now mov instructions for loading absolute addresses and 32 or
 574   // 64 bit integers
 575 
 576   inline void mov(Register dst, address addr)             { mov_immediate64(dst, (uint64_t)addr); }
 577 
 578   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 579   inline void mov(Register dst, T o)                      { mov_immediate64(dst, (uint64_t)o); }
 580 
 581   inline void movw(Register dst, uint32_t imm32)          { mov_immediate32(dst, imm32); }
 582 
 583   void mov(Register dst, RegisterOrConstant src) {
 584     if (src.is_register())
 585       mov(dst, src.as_register());
 586     else
 587       mov(dst, src.as_constant());
 588   }
 589 
 590   void movptr(Register r, uintptr_t imm64);
 591 
 592   void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64);
 593 
 594   void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
 595     orr(Vd, T, Vn, Vn);
 596   }
 597 
 598   void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) {
 599     fcvtsh(tmp, src);
 600     smov(dst, tmp, H, 0);
 601   }
 602 
 603   void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) {
 604     mov(tmp, H, 0, src);
 605     fcvths(dst, tmp);
 606   }
 607 
 608   // Generalized Test Bit And Branch, including a "far" variety which
 609   // spans more than 32KiB.
 610   void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) {
 611     assert(cond == EQ || cond == NE, "must be");
 612 
 613     if (isfar)
 614       cond = ~cond;
 615 
 616     void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
 617     if (cond == Assembler::EQ)
 618       branch = &Assembler::tbz;
 619     else
 620       branch = &Assembler::tbnz;
 621 
 622     if (isfar) {
 623       Label L;
 624       (this->*branch)(Rt, bitpos, L);
 625       b(dest);
 626       bind(L);
 627     } else {
 628       (this->*branch)(Rt, bitpos, dest);
 629     }
 630   }
 631 
 632   // macro instructions for accessing and updating floating point
 633   // status register
 634   //
 635   // FPSR : op1 == 011
 636   //        CRn == 0100
 637   //        CRm == 0100
 638   //        op2 == 001
 639 
 640   inline void get_fpsr(Register reg)
 641   {
 642     mrs(0b11, 0b0100, 0b0100, 0b001, reg);
 643   }
 644 
 645   inline void set_fpsr(Register reg)
 646   {
 647     msr(0b011, 0b0100, 0b0100, 0b001, reg);
 648   }
 649 
 650   inline void clear_fpsr()
 651   {
 652     msr(0b011, 0b0100, 0b0100, 0b001, zr);
 653   }
 654 
 655   // FPCR : op1 == 011
 656   //        CRn == 0100
 657   //        CRm == 0100
 658   //        op2 == 000
 659 
 660   inline void get_fpcr(Register reg) {
 661     mrs(0b11, 0b0100, 0b0100, 0b000, reg);
 662   }
 663 
 664   inline void set_fpcr(Register reg) {
 665     msr(0b011, 0b0100, 0b0100, 0b000, reg);
 666   }
 667 
 668   // DCZID_EL0: op1 == 011
 669   //            CRn == 0000
 670   //            CRm == 0000
 671   //            op2 == 111
 672   inline void get_dczid_el0(Register reg)
 673   {
 674     mrs(0b011, 0b0000, 0b0000, 0b111, reg);
 675   }
 676 
 677   // CTR_EL0:   op1 == 011
 678   //            CRn == 0000
 679   //            CRm == 0000
 680   //            op2 == 001
 681   inline void get_ctr_el0(Register reg)
 682   {
 683     mrs(0b011, 0b0000, 0b0000, 0b001, reg);
 684   }
 685 
 686   inline void get_nzcv(Register reg) {
 687     mrs(0b011, 0b0100, 0b0010, 0b000, reg);
 688   }
 689 
 690   inline void set_nzcv(Register reg) {
 691     msr(0b011, 0b0100, 0b0010, 0b000, reg);
 692   }
 693 
 694   // CNTVCTSS_EL0:   op1 == 011
 695   //                 CRn == 1110
 696   //                 CRm == 0000
 697   //                 op2 == 110
 698   inline void get_cntvctss_el0(Register reg) {
 699     mrs(0b011, 0b1110, 0b0000, 0b110, reg);
 700   }
 701 
 702   // idiv variant which deals with MINLONG as dividend and -1 as divisor
 703   int corrected_idivl(Register result, Register ra, Register rb,
 704                       bool want_remainder, Register tmp = rscratch1);
 705   int corrected_idivq(Register result, Register ra, Register rb,
 706                       bool want_remainder, Register tmp = rscratch1);
 707 
 708   // Support for null-checks
 709   //
 710   // Generates code that causes a null OS exception if the content of reg is null.
 711   // If the accessed location is M[reg + offset] and the offset is known, provide the
 712   // offset. No explicit code generation is needed if the offset is within a certain
 713   // range (0 <= offset <= page_size).
 714 
 715   virtual void null_check(Register reg, int offset = -1);
 716   static bool needs_explicit_null_check(intptr_t offset);
 717   static bool uses_implicit_null_check(void* address);
 718 
 719   // markWord tests, kills markWord reg
 720   void test_markword_is_inline_type(Register markword, Label& is_inline_type);
 721 
 722   // inlineKlass queries, kills temp_reg
 723   void test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null = true);
 724 
 725   void test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free);
 726   void test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free);
 727   void test_field_is_flat(Register flags, Register temp_reg, Label& is_flat);
 728 
 729   // Check oops for special arrays, i.e. flat arrays and/or null-free arrays
 730   void test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label);
 731   void test_flat_array_oop(Register klass, Register temp_reg, Label& is_flat_array);
 732   void test_non_flat_array_oop(Register oop, Register temp_reg, Label&is_non_flat_array);
 733   void test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array);
 734   void test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array);
 735 
 736   // Check array klass layout helper for flat or null-free arrays...
 737   void test_flat_array_layout(Register lh, Label& is_flat_array);
 738 
 739   static address target_addr_for_insn(address insn_addr);
 740 
 741   // Required platform-specific helpers for Label::patch_instructions.
 742   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 743   static int pd_patch_instruction_size(address branch, address target);
 744   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 745     pd_patch_instruction_size(branch, target);
 746   }
 747   static address pd_call_destination(address branch) {
 748     return target_addr_for_insn(branch);
 749   }
 750 #ifndef PRODUCT
 751   static void pd_print_patched_instruction(address branch);
 752 #endif
 753 
 754   static int patch_oop(address insn_addr, address o);
 755 
 756   // Return whether code is emitted to a scratch blob.
 757   virtual bool in_scratch_emit_size() {
 758     return false;
 759   }
 760   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
 761   static int max_trampoline_stub_size();
 762   void emit_static_call_stub();
 763   static int static_call_stub_size();
 764 
 765   // The following 4 methods return the offset of the appropriate move instruction
 766 
 767   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 768   int load_unsigned_byte(Register dst, Address src);
 769   int load_unsigned_short(Register dst, Address src);
 770 
 771   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 772   int load_signed_byte(Register dst, Address src);
 773   int load_signed_short(Register dst, Address src);
 774 
 775   int load_signed_byte32(Register dst, Address src);
 776   int load_signed_short32(Register dst, Address src);
 777 
 778   // Support for sign-extension (hi:lo = extend_sign(lo))
 779   void extend_sign(Register hi, Register lo);
 780 
 781   // Clean up a subword typed value to the representation in compliance with JVMS ยง2.3
 782   void narrow_subword_type(Register reg, BasicType bt);
 783 
 784   // Load and store values by size and signed-ness
 785   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 786   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 787 
 788   // Support for inc/dec with optimal instruction selection depending on value
 789 
 790   // x86_64 aliases an unqualified register/address increment and
 791   // decrement to call incrementq and decrementq but also supports
 792   // explicitly sized calls to incrementq/decrementq or
 793   // incrementl/decrementl
 794 
 795   // for aarch64 the proper convention would be to use
 796   // increment/decrement for 64 bit operations and
 797   // incrementw/decrementw for 32 bit operations. so when porting
 798   // x86_64 code we can leave calls to increment/decrement as is,
 799   // replace incrementq/decrementq with increment/decrement and
 800   // replace incrementl/decrementl with incrementw/decrementw.
 801 
 802   // n.b. increment/decrement calls with an Address destination will
 803   // need to use a scratch register to load the value to be
 804   // incremented. increment/decrement calls which add or subtract a
 805   // constant value greater than 2^24 will need to use a 2nd scratch
 806   // register to hold the constant. so, a register increment/decrement
 807   // may trash rscratch2 and an address increment/decrement trash
 808   // rscratch and rscratch2
 809 
 810   void decrementw(Address dst, int value = 1);
 811   void decrementw(Register reg, int value = 1);
 812 
 813   void decrement(Register reg, int value = 1);
 814   void decrement(Address dst, int value = 1);
 815 
 816   void incrementw(Address dst, int value = 1, Register result = rscratch1);
 817   void incrementw(Register reg, int value = 1);
 818 
 819   void increment(Register reg, int value = 1);
 820   void increment(Address dst, int value = 1, Register result = rscratch1);
 821 
 822 
 823   // Alignment
 824   void align(int modulus);
 825   void align(int modulus, int target);
 826 
 827   // nop
 828   void post_call_nop();
 829 
 830   // Stack frame creation/removal
 831   void enter(bool strip_ret_addr = false);
 832   void leave();
 833 
 834   // ROP Protection
 835   void protect_return_address();
 836   void protect_return_address(Register return_reg);
 837   void authenticate_return_address();
 838   void authenticate_return_address(Register return_reg);
 839   void strip_return_address();
 840   void check_return_address(Register return_reg=lr) PRODUCT_RETURN;
 841 
 842   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
 843   // The pointer will be loaded into the thread register.
 844   void get_thread(Register thread);
 845 
 846   // support for argument shuffling
 847   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 848   void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 849   void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 850   void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 851   void object_move(
 852                    OopMap* map,
 853                    int oop_handle_offset,
 854                    int framesize_in_slots,
 855                    VMRegPair src,
 856                    VMRegPair dst,
 857                    bool is_receiver,
 858                    int* receiver_offset);
 859 
 860 
 861   // Support for VM calls
 862   //
 863   // It is imperative that all calls into the VM are handled via the call_VM macros.
 864   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 865   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 866 
 867 
 868   void call_VM(Register oop_result,
 869                address entry_point,
 870                bool check_exceptions = true);
 871   void call_VM(Register oop_result,
 872                address entry_point,
 873                Register arg_1,
 874                bool check_exceptions = true);
 875   void call_VM(Register oop_result,
 876                address entry_point,
 877                Register arg_1, Register arg_2,
 878                bool check_exceptions = true);
 879   void call_VM(Register oop_result,
 880                address entry_point,
 881                Register arg_1, Register arg_2, Register arg_3,
 882                bool check_exceptions = true);
 883 
 884   // Overloadings with last_Java_sp
 885   void call_VM(Register oop_result,
 886                Register last_java_sp,
 887                address entry_point,
 888                int number_of_arguments = 0,
 889                bool check_exceptions = true);
 890   void call_VM(Register oop_result,
 891                Register last_java_sp,
 892                address entry_point,
 893                Register arg_1, bool
 894                check_exceptions = true);
 895   void call_VM(Register oop_result,
 896                Register last_java_sp,
 897                address entry_point,
 898                Register arg_1, Register arg_2,
 899                bool check_exceptions = true);
 900   void call_VM(Register oop_result,
 901                Register last_java_sp,
 902                address entry_point,
 903                Register arg_1, Register arg_2, Register arg_3,
 904                bool check_exceptions = true);
 905 
 906   void get_vm_result_oop(Register oop_result, Register thread);
 907   void get_vm_result_metadata(Register metadata_result, Register thread);
 908 
 909   // These always tightly bind to MacroAssembler::call_VM_base
 910   // bypassing the virtual implementation
 911   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 912   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
 913   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 914   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);
 915   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);
 916 
 917   void call_VM_leaf(address entry_point,
 918                     int number_of_arguments = 0);
 919   void call_VM_leaf(address entry_point,
 920                     Register arg_1);
 921   void call_VM_leaf(address entry_point,
 922                     Register arg_1, Register arg_2);
 923   void call_VM_leaf(address entry_point,
 924                     Register arg_1, Register arg_2, Register arg_3);
 925 
 926   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 927   // bypassing the virtual implementation
 928   void super_call_VM_leaf(address entry_point);
 929   void super_call_VM_leaf(address entry_point, Register arg_1);
 930   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 931   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 932   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 933 
 934   // last Java Frame (fills frame anchor)
 935   void set_last_Java_frame(Register last_java_sp,
 936                            Register last_java_fp,
 937                            address last_java_pc,
 938                            Register scratch);
 939 
 940   void set_last_Java_frame(Register last_java_sp,
 941                            Register last_java_fp,
 942                            Label &last_java_pc,
 943                            Register scratch);
 944 
 945   void set_last_Java_frame(Register last_java_sp,
 946                            Register last_java_fp,
 947                            Register last_java_pc,
 948                            Register scratch);
 949 
 950   void reset_last_Java_frame(Register thread);
 951 
 952   // thread in the default location (rthread)
 953   void reset_last_Java_frame(bool clear_fp);
 954 
 955   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 956   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 957 
 958   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 959   void c2bool(Register x);
 960 
 961   void load_method_holder_cld(Register rresult, Register rmethod);
 962   void load_method_holder(Register holder, Register method);
 963 
 964   // oop manipulations
 965   void load_metadata(Register dst, Register src);
 966 
 967   void load_narrow_klass_compact(Register dst, Register src);
 968   void load_narrow_klass(Register dst, Register src);
 969   void load_klass(Register dst, Register src, Register tmp);
 970   void store_klass(Register dst, Register src, Register tmp);
 971   void cmp_klass(Register obj, Register klass, Register tmp, Register tmp2);
 972   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 973 
 974   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 975   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 976   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 977 
 978   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 979                       Register tmp1, Register tmp2);
 980 
 981   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 982                        Register tmp1, Register tmp2, Register tmp3);
 983 
 984   void flat_field_copy(DecoratorSet decorators, Register src, Register dst, Register inline_layout_info);
 985 
 986   // inline type data payload offsets...
 987   void payload_offset(Register inline_klass, Register offset);
 988   void payload_address(Register oop, Register data, Register inline_klass);
 989 
 990   void load_heap_oop(Register dst, Address src, Register tmp1,
 991                      Register tmp2, DecoratorSet decorators = 0);
 992 
 993   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 994                               Register tmp2, DecoratorSet decorators = 0);
 995   void store_heap_oop(Address dst, Register val, Register tmp1,
 996                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 997 
 998   // currently unimplemented
 999   // Used for storing null. All other oop constants should be
1000   // stored using routines that take a jobject.
1001   void store_heap_oop_null(Address dst);
1002 
1003   void load_prototype_header(Register dst, Register src);
1004 
1005   void store_klass_gap(Register dst, Register src);
1006 
1007   // This dummy is to prevent a call to store_heap_oop from
1008   // converting a zero (like null) into a Register by giving
1009   // the compiler two choices it can't resolve
1010 
1011   void store_heap_oop(Address dst, void* dummy);
1012 
1013   void encode_heap_oop(Register d, Register s);
1014   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
1015   void decode_heap_oop(Register d, Register s);
1016   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
1017   void encode_heap_oop_not_null(Register r);
1018   void decode_heap_oop_not_null(Register r);
1019   void encode_heap_oop_not_null(Register dst, Register src);
1020   void decode_heap_oop_not_null(Register dst, Register src);
1021 
1022   void set_narrow_oop(Register dst, jobject obj);
1023 
1024   void encode_klass_not_null(Register dst, Register src, Register tmp);
1025   void decode_klass_not_null(Register dst, Register src, Register tmp);
1026 
1027   void set_narrow_klass(Register dst, Klass* k);
1028 
1029   // if heap base register is used - reinit it with the correct value
1030   void reinit_heapbase();
1031 
1032   DEBUG_ONLY(void verify_heapbase(const char* msg);)
1033 
1034   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
1035                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1036   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
1037                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1038 
1039   void push_cont_fastpath(Register java_thread = rthread);
1040   void pop_cont_fastpath(Register java_thread = rthread);
1041 
1042   // Round up to a power of two
1043   void round_to(Register reg, int modulus);
1044 
1045   // java.lang.Math::round intrinsics
1046   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
1047   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
1048 
1049   // allocation
1050 
1051   void tlab_allocate(
1052     Register obj,                      // result: pointer to object after successful allocation
1053     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1054     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1055     Register t1,                       // temp register
1056     Register t2,                       // temp register
1057     Label&   slow_case                 // continuation point if fast allocation fails
1058   );
1059   void verify_tlab();
1060 
1061   void inline_layout_info(Register holder_klass, Register index, Register layout_info);
1062 
1063   // interface method calling
1064   void lookup_interface_method(Register recv_klass,
1065                                Register intf_klass,
1066                                RegisterOrConstant itable_index,
1067                                Register method_result,
1068                                Register scan_temp,
1069                                Label& no_such_interface,
1070                    bool return_method = true);
1071 
1072   void lookup_interface_method_stub(Register recv_klass,
1073                                     Register holder_klass,
1074                                     Register resolved_klass,
1075                                     Register method_result,
1076                                     Register temp_reg,
1077                                     Register temp_reg2,
1078                                     int itable_index,
1079                                     Label& L_no_such_interface);
1080 
1081   // virtual method calling
1082   // n.b. x86 allows RegisterOrConstant for vtable_index
1083   void lookup_virtual_method(Register recv_klass,
1084                              RegisterOrConstant vtable_index,
1085                              Register method_result);
1086 
1087   // Test sub_klass against super_klass, with fast and slow paths.
1088 
1089   // The fast path produces a tri-state answer: yes / no / maybe-slow.
1090   // One of the three labels can be null, meaning take the fall-through.
1091   // If super_check_offset is -1, the value is loaded up from super_klass.
1092   // No registers are killed, except temp_reg.
1093   void check_klass_subtype_fast_path(Register sub_klass,
1094                                      Register super_klass,
1095                                      Register temp_reg,
1096                                      Label* L_success,
1097                                      Label* L_failure,
1098                                      Label* L_slow_path,
1099                                      Register super_check_offset = noreg);
1100 
1101   // The rest of the type check; must be wired to a corresponding fast path.
1102   // It does not repeat the fast path logic, so don't use it standalone.
1103   // The temp_reg and temp2_reg can be noreg, if no temps are available.
1104   // Updates the sub's secondary super cache as necessary.
1105   // If set_cond_codes, condition codes will be Z on success, NZ on failure.
1106   void check_klass_subtype_slow_path(Register sub_klass,
1107                                      Register super_klass,
1108                                      Register temp_reg,
1109                                      Register temp2_reg,
1110                                      Label* L_success,
1111                                      Label* L_failure,
1112                                      bool set_cond_codes = false);
1113 
1114   void check_klass_subtype_slow_path_linear(Register sub_klass,
1115                                             Register super_klass,
1116                                             Register temp_reg,
1117                                             Register temp2_reg,
1118                                             Label* L_success,
1119                                             Label* L_failure,
1120                                             bool set_cond_codes = false);
1121 
1122   void check_klass_subtype_slow_path_table(Register sub_klass,
1123                                            Register super_klass,
1124                                            Register temp_reg,
1125                                            Register temp2_reg,
1126                                            Register temp3_reg,
1127                                            Register result_reg,
1128                                            FloatRegister vtemp_reg,
1129                                            Label* L_success,
1130                                            Label* L_failure,
1131                                            bool set_cond_codes = false);
1132 
1133   // If r is valid, return r.
1134   // If r is invalid, remove a register r2 from available_regs, add r2
1135   // to regs_to_push, then return r2.
1136   Register allocate_if_noreg(const Register r,
1137                              RegSetIterator<Register> &available_regs,
1138                              RegSet &regs_to_push);
1139 
1140   // Secondary subtype checking
1141   void lookup_secondary_supers_table_var(Register sub_klass,
1142                                          Register r_super_klass,
1143                                          Register temp1,
1144                                          Register temp2,
1145                                          Register temp3,
1146                                          FloatRegister vtemp,
1147                                          Register result,
1148                                          Label *L_success);
1149 
1150 
1151   // As above, but with a constant super_klass.
1152   // The result is in Register result, not the condition codes.
1153   bool lookup_secondary_supers_table_const(Register r_sub_klass,
1154                                            Register r_super_klass,
1155                                            Register temp1,
1156                                            Register temp2,
1157                                            Register temp3,
1158                                            FloatRegister vtemp,
1159                                            Register result,
1160                                            u1 super_klass_slot,
1161                                            bool stub_is_near = false);
1162 
1163   void verify_secondary_supers_table(Register r_sub_klass,
1164                                      Register r_super_klass,
1165                                      Register temp1,
1166                                      Register temp2,
1167                                      Register result);
1168 
1169   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
1170                                                Register r_array_base,
1171                                                Register r_array_index,
1172                                                Register r_bitmap,
1173                                                Register temp1,
1174                                                Register result,
1175                                                bool is_stub = true);
1176 
1177   // Simplified, combined version, good for typical uses.
1178   // Falls through on failure.
1179   void check_klass_subtype(Register sub_klass,
1180                            Register super_klass,
1181                            Register temp_reg,
1182                            Label& L_success);
1183 
1184   void clinit_barrier(Register klass,
1185                       Register thread,
1186                       Label* L_fast_path = nullptr,
1187                       Label* L_slow_path = nullptr);
1188 
1189   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
1190 
1191   void profile_receiver_type(Register recv, Register mdp, int mdp_offset);
1192 
1193   void verify_sve_vector_length(Register tmp = rscratch1);
1194   void reinitialize_ptrue() {
1195     if (UseSVE > 0) {
1196       sve_ptrue(ptrue, B);
1197     }
1198   }
1199   void verify_ptrue();
1200 
1201   // Debugging
1202 
1203   // only if +VerifyOops
1204   void _verify_oop(Register reg, const char* s, const char* file, int line);
1205   void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
1206 
1207   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
1208     if (VerifyOops) {
1209       _verify_oop(reg, s, file, line);
1210     }
1211   }
1212   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
1213     if (VerifyOops) {
1214       _verify_oop_addr(reg, s, file, line);
1215     }
1216   }
1217 
1218 // TODO: verify method and klass metadata (compare against vptr?)
1219   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1220   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1221 
1222 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
1223 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
1224 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
1225 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1226 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1227 
1228   // Restore cpu control state after JNI call
1229   void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2);
1230 
1231   // prints msg, dumps registers and stops execution
1232   void stop(const char* msg);
1233 
1234   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1235 
1236   void untested()                                { stop("untested"); }
1237 
1238   void unimplemented(const char* what = "");
1239 
1240   void should_not_reach_here()                   { stop("should not reach here"); }
1241 
1242   void _assert_asm(Condition cc, const char* msg);
1243 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg)
1244 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg))
1245 
1246   // Stack overflow checking
1247   void bang_stack_with_offset(int offset) {
1248     // stack grows down, caller passes positive offset
1249     assert(offset > 0, "must bang with negative offset");
1250     sub(rscratch2, sp, offset);
1251     str(zr, Address(rscratch2));
1252   }
1253 
1254   // Writes to stack successive pages until offset reached to check for
1255   // stack overflow + shadow pages.  Also, clobbers tmp
1256   void bang_stack_size(Register size, Register tmp);
1257 
1258   // Check for reserved stack access in method being exited (for JIT)
1259   void reserved_stack_check();
1260 
1261   // Arithmetics
1262 
1263   // Clobber: rscratch1, rscratch2
1264   void addptr(const Address &dst, int32_t src);
1265 
1266   // Clobber: rscratch1
1267   void cmpptr(Register src1, Address src2);
1268 
1269   void cmpoop(Register obj1, Register obj2);
1270 
1271   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1272   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1273   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1274   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1275 
1276   void atomic_xchg(Register prev, Register newv, Register addr);
1277   void atomic_xchgw(Register prev, Register newv, Register addr);
1278   void atomic_xchgl(Register prev, Register newv, Register addr);
1279   void atomic_xchglw(Register prev, Register newv, Register addr);
1280   void atomic_xchgal(Register prev, Register newv, Register addr);
1281   void atomic_xchgalw(Register prev, Register newv, Register addr);
1282 
1283   void orptr(Address adr, RegisterOrConstant src) {
1284     ldr(rscratch1, adr);
1285     if (src.is_register())
1286       orr(rscratch1, rscratch1, src.as_register());
1287     else
1288       orr(rscratch1, rscratch1, src.as_constant());
1289     str(rscratch1, adr);
1290   }
1291 
1292 private:
1293   // A generic CAS; success or failure is in the EQ flag.
1294   // Clobbers rscratch1
1295   void cmpxchg(Register addr, Register expected, Register new_val,
1296                enum operand_size size, enum atomic_memory_order order,
1297                bool weak, Register result);
1298 
1299 public:
1300   void cmpxchg(Register addr, Register expected, Register new_val,
1301                enum operand_size size, enum atomic_memory_order order,
1302                Register result = noreg) {
1303     cmpxchg(addr, expected, new_val, size, order, /* weak */ false, result);
1304   }
1305 
1306   void cmpxchg_weak(Register addr, Register expected, Register new_val,
1307                     enum operand_size size, enum atomic_memory_order order,
1308                     Register result = noreg) {
1309     cmpxchg(addr, expected, new_val, size, order, /* weak */ true, result);
1310   }
1311 
1312 #ifdef ASSERT
1313   // Template short-hand support to clean-up after a failed call to trampoline
1314   // call generation (see trampoline_call() below),  when a set of Labels must
1315   // be reset (before returning).
1316   template<typename Label, typename... More>
1317   void reset_labels(Label &lbl, More&... more) {
1318     lbl.reset(); reset_labels(more...);
1319   }
1320   template<typename Label>
1321   void reset_labels(Label &lbl) {
1322     lbl.reset();
1323   }
1324 #endif
1325 
1326 private:
1327   void compare_eq(Register rn, Register rm, enum operand_size size);
1328 
1329 public:
1330   // AArch64 OpenJDK uses four different types of calls:
1331   //   - direct call: bl pc_relative_offset
1332   //     This is the shortest and the fastest, but the offset has the range:
1333   //     +/-128MB for the release build, +/-2MB for the debug build.
1334   //
1335   //   - far call: adrp reg, pc_relative_offset; add; bl reg
1336   //     This is longer than a direct call. The offset has
1337   //     the range +/-4GB. As the code cache size is limited to 4GB,
1338   //     far calls can reach anywhere in the code cache. If a jump is
1339   //     needed rather than a call, a far jump 'b reg' can be used instead.
1340   //     All instructions are embedded at a call site.
1341   //
1342   //   - trampoline call:
1343   //     This is only available in C1/C2-generated code (nmethod). It is a combination
1344   //     of a direct call, which is used if the destination of a call is in range,
1345   //     and a register-indirect call. It has the advantages of reaching anywhere in
1346   //     the AArch64 address space and being patchable at runtime when the generated
1347   //     code is being executed by other threads.
1348   //
1349   //     [Main code section]
1350   //       bl trampoline
1351   //     [Stub code section]
1352   //     trampoline:
1353   //       ldr reg, pc + 8
1354   //       br reg
1355   //       <64-bit destination address>
1356   //
1357   //     If the destination is in range when the generated code is moved to the code
1358   //     cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline
1359   //     is not used.
1360   //     The optimization does not remove the trampoline from the stub section.
1361   //     This is necessary because the trampoline may well be redirected later when
1362   //     code is patched, and the new destination may not be reachable by a simple BR
1363   //     instruction.
1364   //
1365   //   - indirect call: move reg, address; blr reg
1366   //     This too can reach anywhere in the address space, but it cannot be
1367   //     patched while code is running, so it must only be modified at a safepoint.
1368   //     This form of call is most suitable for targets at fixed addresses, which
1369   //     will never be patched.
1370   //
1371   // The patching we do conforms to the "Concurrent modification and
1372   // execution of instructions" section of the Arm Architectural
1373   // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC,
1374   // or SVC instructions to be modified while another thread is
1375   // executing them.
1376   //
1377   // To patch a trampoline call when the BL can't reach, we first modify
1378   // the 64-bit destination address in the trampoline, then modify the
1379   // BL to point to the trampoline, then flush the instruction cache to
1380   // broadcast the change to all executing threads. See
1381   // NativeCall::set_destination_mt_safe for the details.
1382   //
1383   // There is a benign race in that the other thread might observe the
1384   // modified BL before it observes the modified 64-bit destination
1385   // address. That does not matter because the destination method has been
1386   // invalidated, so there will be a trap at its start.
1387   // For this to work, the destination address in the trampoline is
1388   // always updated, even if we're not using the trampoline.
1389 
1390   // Emit a direct call if the entry address will always be in range,
1391   // otherwise a trampoline call.
1392   // Supported entry.rspec():
1393   // - relocInfo::runtime_call_type
1394   // - relocInfo::opt_virtual_call_type
1395   // - relocInfo::static_call_type
1396   // - relocInfo::virtual_call_type
1397   //
1398   // Return: the call PC or null if CodeCache is full.
1399   // Clobbers: rscratch1
1400   address trampoline_call(Address entry);
1401 
1402   static bool far_branches() {
1403     return ReservedCodeCacheSize > branch_range;
1404   }
1405   // Check if the static call stub branch needs a far jump.
1406   static bool codestub_branch_needs_far_jump() {
1407     if (AOTCodeCache::is_on_for_dump()) {
1408       // To calculate static_call_stub_size correctly.
1409       return true;
1410     }
1411     return far_branches();
1412   }
1413   // Check if a branch to the given address needs a far jump.
1414   static bool target_needs_far_branch(address addr);
1415 
1416   // Emit a direct call/jump if the entry address will always be in range,
1417   // otherwise a far call/jump.
1418   // The address must be inside the code cache.
1419   // Supported entry.rspec():
1420   // - relocInfo::external_word_type
1421   // - relocInfo::runtime_call_type
1422   // - relocInfo::none
1423   // In the case of a far call/jump, the entry address is put in the tmp register.
1424   // The tmp register is invalidated.
1425   //
1426   void far_call(Address entry, Register tmp = rscratch1);
1427   // Far_jump returns the amount of the emitted code.
1428   int far_jump(Address entry, Register tmp = rscratch1);
1429 
1430   // Emit the CompiledIC call idiom
1431   address ic_call(address entry, jint method_index = 0);
1432   static int ic_check_size();
1433   int ic_check(int end_alignment);
1434 
1435 public:
1436 
1437   // Data
1438 
1439   void mov_metadata(Register dst, Metadata* obj);
1440   Address allocate_metadata_address(Metadata* obj);
1441   Address constant_oop_address(jobject obj);
1442 
1443   void movoop(Register dst, jobject obj);
1444 
1445   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1446   void kernel_crc32(Register crc, Register buf, Register len,
1447         Register table0, Register table1, Register table2, Register table3,
1448         Register tmp, Register tmp2, Register tmp3);
1449   // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic.
1450   void kernel_crc32c(Register crc, Register buf, Register len,
1451         Register table0, Register table1, Register table2, Register table3,
1452         Register tmp, Register tmp2, Register tmp3);
1453 
1454   // Stack push and pop individual 64 bit registers
1455   void push(Register src);
1456   void pop(Register dst);
1457 
1458   void repne_scan(Register addr, Register value, Register count,
1459                   Register scratch);
1460   void repne_scanw(Register addr, Register value, Register count,
1461                    Register scratch);
1462 
1463   typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1464   typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1465 
1466   // If a constant does not fit in an immediate field, generate some
1467   // number of MOV instructions and then perform the operation
1468   void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
1469                              add_sub_imm_insn insn1,
1470                              add_sub_reg_insn insn2, bool is32);
1471   // Separate vsn which sets the flags
1472   void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
1473                                add_sub_imm_insn insn1,
1474                                add_sub_reg_insn insn2, bool is32);
1475 
1476 #define WRAP(INSN, is32)                                                \
1477   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1478     wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1479   }                                                                     \
1480                                                                         \
1481   void INSN(Register Rd, Register Rn, Register Rm,                      \
1482              enum shift_kind kind, unsigned shift = 0) {                \
1483     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1484   }                                                                     \
1485                                                                         \
1486   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1487     Assembler::INSN(Rd, Rn, Rm);                                        \
1488   }                                                                     \
1489                                                                         \
1490   void INSN(Register Rd, Register Rn, Register Rm,                      \
1491            ext::operation option, int amount = 0) {                     \
1492     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1493   }
1494 
1495   WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true)
1496 
1497 #undef WRAP
1498 #define WRAP(INSN, is32)                                                \
1499   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1500     wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1501   }                                                                     \
1502                                                                         \
1503   void INSN(Register Rd, Register Rn, Register Rm,                      \
1504              enum shift_kind kind, unsigned shift = 0) {                \
1505     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1506   }                                                                     \
1507                                                                         \
1508   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1509     Assembler::INSN(Rd, Rn, Rm);                                        \
1510   }                                                                     \
1511                                                                         \
1512   void INSN(Register Rd, Register Rn, Register Rm,                      \
1513            ext::operation option, int amount = 0) {                     \
1514     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1515   }
1516 
1517   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1518 
1519   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1520   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1521   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1522   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1523 
1524   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1525 
1526   void verified_entry(Compile* C, int sp_inc);
1527 
1528   // Inline type specific methods
1529   #include "asm/macroAssembler_common.hpp"
1530 
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   // Load a constant address in the AOT Runtime Constants area
1560   void load_aotrc_address(Register reg, address a);
1561 
1562   // Prolog generator routines to support switch between x86 code and
1563   // generated ARM code
1564 
1565   // routine to generate an x86 prolog for a stub function which
1566   // bootstraps into the generated ARM code which directly follows the
1567   // stub
1568   //
1569 
1570   public:
1571 
1572   address read_polling_page(Register r, relocInfo::relocType rtype);
1573   void get_polling_page(Register dest, relocInfo::relocType rtype);
1574 
1575   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1576   void update_byte_crc32(Register crc, Register val, Register table);
1577   void update_word_crc32(Register crc, Register v, Register tmp,
1578         Register table0, Register table1, Register table2, Register table3,
1579         bool upper = false);
1580 
1581   address count_positives(Register ary1, Register len, Register result);
1582 
1583   address arrays_equals(Register a1, Register a2, Register result, Register cnt1,
1584                         Register tmp1, Register tmp2, Register tmp3, int elem_size);
1585 
1586 // Ensure that the inline code and the stub use the same registers.
1587 #define ARRAYS_HASHCODE_REGISTERS \
1588   do {                      \
1589     assert(result == r0  && \
1590            ary    == r1  && \
1591            cnt    == r2  && \
1592            vdata0 == v3  && \
1593            vdata1 == v2  && \
1594            vdata2 == v1  && \
1595            vdata3 == v0  && \
1596            vmul0  == v4  && \
1597            vmul1  == v5  && \
1598            vmul2  == v6  && \
1599            vmul3  == v7  && \
1600            vpow   == v12 && \
1601            vpowm  == v13, "registers must match aarch64.ad"); \
1602   } while (0)
1603 
1604   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1605 
1606   void fill_words(Register base, Register cnt, Register value);
1607   void fill_words(Register base, uint64_t cnt, Register value);
1608 
1609   address zero_words(Register base, uint64_t cnt);
1610   address zero_words(Register ptr, Register cnt);
1611   void zero_dcache_blocks(Register base, Register cnt);
1612 
1613   static const int zero_words_block_size;
1614 
1615   address byte_array_inflate(Register src, Register dst, Register len,
1616                              FloatRegister vtmp1, FloatRegister vtmp2,
1617                              FloatRegister vtmp3, Register tmp4);
1618 
1619   void char_array_compress(Register src, Register dst, Register len,
1620                            Register res,
1621                            FloatRegister vtmp0, FloatRegister vtmp1,
1622                            FloatRegister vtmp2, FloatRegister vtmp3,
1623                            FloatRegister vtmp4, FloatRegister vtmp5);
1624 
1625   void encode_iso_array(Register src, Register dst,
1626                         Register len, Register res, bool ascii,
1627                         FloatRegister vtmp0, FloatRegister vtmp1,
1628                         FloatRegister vtmp2, FloatRegister vtmp3,
1629                         FloatRegister vtmp4, FloatRegister vtmp5);
1630 
1631   void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi,
1632       address pio2, address dsin_coef, address dcos_coef);
1633  private:
1634   // begin trigonometric functions support block
1635   void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2);
1636   void generate__kernel_rem_pio2(address two_over_pi, address pio2);
1637   void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef);
1638   void generate_kernel_cos(FloatRegister x, address dcos_coef);
1639   // end trigonometric functions support block
1640   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1641                        Register src1, Register src2);
1642   void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1643     add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1644   }
1645   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1646                              Register y, Register y_idx, Register z,
1647                              Register carry, Register product,
1648                              Register idx, Register kdx);
1649   void multiply_128_x_128_loop(Register y, Register z,
1650                                Register carry, Register carry2,
1651                                Register idx, Register jdx,
1652                                Register yz_idx1, Register yz_idx2,
1653                                Register tmp, Register tmp3, Register tmp4,
1654                                Register tmp7, Register product_hi);
1655   void kernel_crc32_using_crypto_pmull(Register crc, Register buf,
1656         Register len, Register tmp0, Register tmp1, Register tmp2,
1657         Register tmp3);
1658   void kernel_crc32_using_crc32(Register crc, Register buf,
1659         Register len, Register tmp0, Register tmp1, Register tmp2,
1660         Register tmp3);
1661   void kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
1662         Register len, Register tmp0, Register tmp1, Register tmp2,
1663         Register tmp3);
1664   void kernel_crc32c_using_crc32c(Register crc, Register buf,
1665         Register len, Register tmp0, Register tmp1, Register tmp2,
1666         Register tmp3);
1667   void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
1668         Register len, Register tmp0, Register tmp1, Register tmp2,
1669         size_t table_offset);
1670 
1671   void ghash_modmul (FloatRegister result,
1672                      FloatRegister result_lo, FloatRegister result_hi, FloatRegister b,
1673                      FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p,
1674                      FloatRegister t1, FloatRegister t2, FloatRegister t3);
1675   void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
1676 public:
1677   void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1678                        Register tmp0, Register tmp1, Register tmp2, Register tmp3,
1679                        Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1680   void mul_add(Register out, Register in, Register offs, Register len, Register k);
1681   void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,
1682                       FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1683                       FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1684   void ghash_multiply_wide(int index,
1685                            FloatRegister result_lo, FloatRegister result_hi,
1686                            FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1687                            FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1688   void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi,
1689                     FloatRegister p, FloatRegister z, FloatRegister t1);
1690   void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi,
1691                     FloatRegister p, FloatRegister z, FloatRegister t1);
1692   void ghash_processBlocks_wide(Label& p, Register state, Register subkeyH,
1693                                 Register data, Register blocks, int unrolls);
1694 
1695 
1696   void aesenc_loadkeys(Register key, Register keylen);
1697   void aesecb_encrypt(Register from, Register to, Register keylen,
1698                       FloatRegister data = v0, int unrolls = 1);
1699   void aesecb_decrypt(Register from, Register to, Register key, Register keylen);
1700   void aes_round(FloatRegister input, FloatRegister subkey);
1701 
1702   // ChaCha20 functions support block
1703   void cc20_qr_add4(FloatRegister (&addFirst)[4],
1704           FloatRegister (&addSecond)[4]);
1705   void cc20_qr_xor4(FloatRegister (&firstElem)[4],
1706           FloatRegister (&secondElem)[4], FloatRegister (&result)[4]);
1707   void cc20_qr_lrot4(FloatRegister (&sourceReg)[4],
1708           FloatRegister (&destReg)[4], int bits, FloatRegister table);
1709   void cc20_set_qr_registers(FloatRegister (&vectorSet)[4],
1710           const FloatRegister (&stateVectors)[16], int idx1, int idx2,
1711           int idx3, int idx4);
1712 
1713   // Rotate using ORR (for identity) or USHR + SLI.
1714   void neon_vector_rotate(FloatRegister dst, SIMD_Arrangement T,
1715                           FloatRegister src, int shift_amount);
1716 
1717   // Place an ISB after code may have been modified due to a safepoint.
1718   void safepoint_isb();
1719 
1720 private:
1721   // Return the effective address r + (r1 << ext) + offset.
1722   // Uses rscratch2.
1723   Address offsetted_address(Register r, Register r1, Address::extend ext,
1724                             int offset, int size);
1725 
1726 private:
1727   // Returns an address on the stack which is reachable with a ldr/str of size
1728   // Uses rscratch2 if the address is not directly reachable
1729   Address spill_address(int size, int offset, Register tmp=rscratch2);
1730   Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2);
1731 
1732   bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const;
1733 
1734   // Check whether two loads/stores can be merged into ldp/stp.
1735   bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const;
1736 
1737   // Merge current load/store with previous load/store into ldp/stp.
1738   void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1739 
1740   // Try to merge two loads/stores into ldp/stp. If success, returns true else false.
1741   bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1742 
1743 public:
1744   void spill(Register Rx, bool is64, int offset) {
1745     if (is64) {
1746       str(Rx, spill_address(8, offset));
1747     } else {
1748       strw(Rx, spill_address(4, offset));
1749     }
1750   }
1751   void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1752     str(Vx, T, spill_address(1 << (int)T, offset));
1753   }
1754 
1755   void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1756     sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1757   }
1758   void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1759     sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1760   }
1761 
1762   void unspill(Register Rx, bool is64, int offset) {
1763     if (is64) {
1764       ldr(Rx, spill_address(8, offset));
1765     } else {
1766       ldrw(Rx, spill_address(4, offset));
1767     }
1768   }
1769   void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1770     ldr(Vx, T, spill_address(1 << (int)T, offset));
1771   }
1772 
1773   void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1774     sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1775   }
1776   void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1777     sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1778   }
1779 
1780   void spill_copy128(int src_offset, int dst_offset,
1781                      Register tmp1=rscratch1, Register tmp2=rscratch2) {
1782     if (src_offset < 512 && (src_offset & 7) == 0 &&
1783         dst_offset < 512 && (dst_offset & 7) == 0) {
1784       ldp(tmp1, tmp2, Address(sp, src_offset));
1785       stp(tmp1, tmp2, Address(sp, dst_offset));
1786     } else {
1787       unspill(tmp1, true, src_offset);
1788       spill(tmp1, true, dst_offset);
1789       unspill(tmp1, true, src_offset+8);
1790       spill(tmp1, true, dst_offset+8);
1791     }
1792   }
1793   void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset,
1794                                             int sve_vec_reg_size_in_bytes) {
1795     assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size");
1796     for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) {
1797       spill_copy128(src_offset, dst_offset);
1798       src_offset += 16;
1799       dst_offset += 16;
1800     }
1801   }
1802   void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset,
1803                                                int sve_predicate_reg_size_in_bytes) {
1804     sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset));
1805     sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset));
1806     reinitialize_ptrue();
1807   }
1808   void cache_wb(Address line);
1809   void cache_wbsync(bool is_pre);
1810 
1811   // Code for java.lang.Thread::onSpinWait() intrinsic.
1812   void spin_wait();
1813   void spin_wait_wfet(int delay_ns);
1814 
1815   void fast_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow);
1816   void fast_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1817 
1818 private:
1819   // Check the current thread doesn't need a cross modify fence.
1820   void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
1821   void try_to_replace_prev_vector_copy_with_movprfx(FloatRegister dst);
1822 
1823 public:
1824   void maybe_movprfx(FloatRegister dst, FloatRegister src) {
1825     if (dst != src) {
1826       sve_movprfx(dst, src);
1827     }
1828   }
1829 
1830 // Wrappers for SVE explicit destructive instructions, overriding the
1831 // same-signature Assembler entry points to enable movprfx fusion optimization.
1832 //
1833 // Implicit destructive instructions (e.g. predicated unary ops like sve_abs/
1834 // sve_neg/sve_not, whose ISA encoding allows Zd != Zn but whose use as a Java
1835 // Vector API masked operation requires pass-through of the first source) are
1836 // not covered here. For those, the .ad file is responsible for emitting
1837 // movprfx explicitly via maybe_movprfx() before the destructive op.
1838 #define SVE_DESTRUCTIVE_BINARY_INS(NAME)                                       \
1839   using Assembler::NAME;                                                       \
1840   void NAME(FloatRegister Zd, SIMD_RegVariant T, PRegister Pg,                 \
1841             FloatRegister Zm) {                                                \
1842     if (Zd != Zm) {                                                            \
1843       try_to_replace_prev_vector_copy_with_movprfx(Zd);                        \
1844     }                                                                          \
1845     Assembler::NAME(Zd, T, Pg, Zm);                                            \
1846   }
1847 
1848 #define SVE_DESTRUCTIVE_BINARY_5(I1, I2, I3, I4, I5)                           \
1849   SVE_DESTRUCTIVE_BINARY_INS(I1); SVE_DESTRUCTIVE_BINARY_INS(I2);              \
1850   SVE_DESTRUCTIVE_BINARY_INS(I3); SVE_DESTRUCTIVE_BINARY_INS(I4);              \
1851   SVE_DESTRUCTIVE_BINARY_INS(I5);
1852 
1853   SVE_DESTRUCTIVE_BINARY_5(sve_add,  sve_and,   sve_asr,   sve_bic,   sve_eor)
1854   SVE_DESTRUCTIVE_BINARY_5(sve_fabd, sve_fadd,  sve_fdiv,  sve_fmax,  sve_fmin)
1855   SVE_DESTRUCTIVE_BINARY_5(sve_fmul, sve_fsub,  sve_lsl,   sve_lsr,   sve_mul)
1856   SVE_DESTRUCTIVE_BINARY_5(sve_orr,  sve_smax,  sve_smin,  sve_sqadd, sve_sqsub)
1857   SVE_DESTRUCTIVE_BINARY_5(sve_sub,  sve_uqadd, sve_uqsub, sve_umax,  sve_umin)
1858   SVE_DESTRUCTIVE_BINARY_INS(sve_sdiv);
1859   SVE_DESTRUCTIVE_BINARY_INS(sve_udiv);
1860 
1861 #undef SVE_DESTRUCTIVE_BINARY_INS
1862 #undef SVE_DESTRUCTIVE_BINARY_5
1863 
1864 #define SVE_DESTRUCTIVE_SHIFT_IMM_INS(NAME)                                    \
1865   void NAME(FloatRegister Zd, SIMD_RegVariant T, PRegister Pg, int shift) {    \
1866     try_to_replace_prev_vector_copy_with_movprfx(Zd);                          \
1867     Assembler::NAME(Zd, T, Pg, shift);                                         \
1868   }
1869 
1870   SVE_DESTRUCTIVE_SHIFT_IMM_INS(sve_asr);
1871   SVE_DESTRUCTIVE_SHIFT_IMM_INS(sve_lsl);
1872   SVE_DESTRUCTIVE_SHIFT_IMM_INS(sve_lsr);
1873 
1874 #undef SVE_DESTRUCTIVE_SHIFT_IMM_INS
1875 
1876 #define SVE_DESTRUCTIVE_UNPRED_IMM_INS(NAME, IMM_TYPE)                         \
1877   void NAME(FloatRegister Zd, SIMD_RegVariant T, IMM_TYPE imm) {               \
1878     try_to_replace_prev_vector_copy_with_movprfx(Zd);                          \
1879     Assembler::NAME(Zd, T, imm);                                               \
1880   }
1881 
1882   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_add, unsigned);
1883   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_sub, unsigned);
1884   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_and, uint64_t);
1885   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_eor, uint64_t);
1886   SVE_DESTRUCTIVE_UNPRED_IMM_INS(sve_orr, uint64_t);
1887 
1888 #undef SVE_DESTRUCTIVE_UNPRED_IMM_INS
1889 
1890 #define SVE_DESTRUCTIVE_TERNARY_INS(NAME)                                      \
1891   using Assembler::NAME;                                                       \
1892   void NAME(FloatRegister Zd, SIMD_RegVariant T, PRegister Pg,                 \
1893             FloatRegister Zn, FloatRegister Zm) {                              \
1894     if (Zd != Zn && Zd != Zm) {                                                \
1895       try_to_replace_prev_vector_copy_with_movprfx(Zd);                        \
1896     }                                                                          \
1897     Assembler::NAME(Zd, T, Pg, Zn, Zm);                                        \
1898   }
1899 
1900   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmad);
1901   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmla);
1902   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmls);
1903   SVE_DESTRUCTIVE_TERNARY_INS(sve_fmsb);
1904   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmad);
1905   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmla);
1906   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmls);
1907   SVE_DESTRUCTIVE_TERNARY_INS(sve_fnmsb);
1908   SVE_DESTRUCTIVE_TERNARY_INS(sve_mla);
1909   SVE_DESTRUCTIVE_TERNARY_INS(sve_mls);
1910 
1911 #undef SVE_DESTRUCTIVE_TERNARY_INS
1912 
1913 #define SVE_DESTRUCTIVE_TERNARY_UNPRED_INS(NAME)                               \
1914   using Assembler::NAME;                                                       \
1915   void NAME(FloatRegister Zd, FloatRegister Zm, FloatRegister Zk) {            \
1916     if (Zd != Zm && Zd != Zk) {                                                \
1917       try_to_replace_prev_vector_copy_with_movprfx(Zd);                        \
1918     }                                                                          \
1919     Assembler::NAME(Zd, Zm, Zk);                                               \
1920   }
1921 
1922   SVE_DESTRUCTIVE_TERNARY_UNPRED_INS(sve_bsl);
1923   SVE_DESTRUCTIVE_TERNARY_UNPRED_INS(sve_eor3);
1924 
1925 #undef SVE_DESTRUCTIVE_TERNARY_UNPRED_INS
1926 };
1927 
1928 #ifdef ASSERT
1929 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1930 #endif
1931 
1932 struct tableswitch {
1933   Register _reg;
1934   int _insn_index; jint _first_key; jint _last_key;
1935   Label _after;
1936   Label _branches;
1937 };
1938 
1939 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP