1 /*
   2  * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
   4  * Copyright (c) 2020, 2024, Huawei Technologies Co., Ltd. All rights reserved.
   5  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   6  *
   7  * This code is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License version 2 only, as
   9  * published by the Free Software Foundation.
  10  *
  11  * This code is distributed in the hope that it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14  * version 2 for more details (a copy is included in the LICENSE file that
  15  * accompanied this code).
  16  *
  17  * You should have received a copy of the GNU General Public License version
  18  * 2 along with this work; if not, write to the Free Software Foundation,
  19  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  20  *
  21  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  22  * or visit www.oracle.com if you need additional information or have any
  23  * questions.
  24  *
  25  */
  26 
  27 #ifndef CPU_RISCV_MACROASSEMBLER_RISCV_HPP
  28 #define CPU_RISCV_MACROASSEMBLER_RISCV_HPP
  29 
  30 #include "asm/assembler.inline.hpp"
  31 #include "code/vmreg.hpp"
  32 #include "metaprogramming/enableIf.hpp"
  33 #include "oops/compressedOops.hpp"
  34 #include "utilities/powerOfTwo.hpp"
  35 
  36 // MacroAssembler extends Assembler by frequently used macros.
  37 //
  38 // Instructions for which a 'better' code sequence exists depending
  39 // on arguments should also go in here.
  40 
  41 class MacroAssembler: public Assembler {
  42 
  43  public:
  44 
  45   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
  46 
  47   void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp_reg = t0);
  48 
  49   // Alignment
  50   int align(int modulus, int extra_offset = 0);
  51 
  52   static inline void assert_alignment(address pc, int alignment = MacroAssembler::instruction_size) {
  53     assert(is_aligned(pc, alignment), "bad alignment");
  54   }
  55 
  56   // nop
  57   void post_call_nop();
  58 
  59   // Stack frame creation/removal
  60   // Note that SP must be updated to the right place before saving/restoring RA and FP
  61   // because signal based thread suspend/resume could happen asynchronously.
  62   void enter() {
  63     subi(sp, sp, 2 * wordSize);
  64     sd(ra, Address(sp, wordSize));
  65     sd(fp, Address(sp));
  66     addi(fp, sp, 2 * wordSize);
  67   }
  68 
  69   void leave() {
  70     subi(sp, fp, 2 * wordSize);
  71     ld(fp, Address(sp));
  72     ld(ra, Address(sp, wordSize));
  73     addi(sp, sp, 2 * wordSize);
  74   }
  75 
  76 
  77   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
  78   // The pointer will be loaded into the thread register.
  79   void get_thread(Register thread);
  80 
  81   // Support for VM calls
  82   //
  83   // It is imperative that all calls into the VM are handled via the call_VM macros.
  84   // They make sure that the stack linkage is setup correctly. call_VM's correspond
  85   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
  86 
  87   void call_VM(Register oop_result,
  88                address entry_point,
  89                bool check_exceptions = true);
  90   void call_VM(Register oop_result,
  91                address entry_point,
  92                Register arg_1,
  93                bool check_exceptions = true);
  94   void call_VM(Register oop_result,
  95                address entry_point,
  96                Register arg_1, Register arg_2,
  97                bool check_exceptions = true);
  98   void call_VM(Register oop_result,
  99                address entry_point,
 100                Register arg_1, Register arg_2, Register arg_3,
 101                bool check_exceptions = true);
 102 
 103   // Overloadings with last_Java_sp
 104   void call_VM(Register oop_result,
 105                Register last_java_sp,
 106                address entry_point,
 107                int number_of_arguments = 0,
 108                bool check_exceptions = true);
 109   void call_VM(Register oop_result,
 110                Register last_java_sp,
 111                address entry_point,
 112                Register arg_1,
 113                bool check_exceptions = true);
 114   void call_VM(Register oop_result,
 115                Register last_java_sp,
 116                address entry_point,
 117                Register arg_1, Register arg_2,
 118                bool check_exceptions = true);
 119   void call_VM(Register oop_result,
 120                Register last_java_sp,
 121                address entry_point,
 122                Register arg_1, Register arg_2, Register arg_3,
 123                bool check_exceptions = true);
 124 
 125   void get_vm_result_oop(Register oop_result, Register java_thread);
 126   void get_vm_result_metadata(Register metadata_result, Register java_thread);
 127 
 128   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 129   // bypassing the virtual implementation
 130   void call_VM_leaf(address entry_point,
 131                     int number_of_arguments = 0);
 132   void call_VM_leaf(address entry_point,
 133                     Register arg_0);
 134   void call_VM_leaf(address entry_point,
 135                     Register arg_0, Register arg_1);
 136   void call_VM_leaf(address entry_point,
 137                     Register arg_0, Register arg_1, Register arg_2);
 138 
 139   // These always tightly bind to MacroAssembler::call_VM_base
 140   // bypassing the virtual implementation
 141   void super_call_VM_leaf(address entry_point, Register arg_0);
 142   void super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1);
 143   void super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2);
 144   void super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3);
 145 
 146   // last Java Frame (fills frame anchor)
 147   void set_last_Java_frame(Register last_java_sp, Register last_java_fp, address last_java_pc, Register tmp);
 148   void set_last_Java_frame(Register last_java_sp, Register last_java_fp, Label &last_java_pc, Register tmp);
 149   void set_last_Java_frame(Register last_java_sp, Register last_java_fp, Register last_java_pc);
 150 
 151   // thread in the default location (xthread)
 152   void reset_last_Java_frame(bool clear_fp);
 153 
 154   virtual void call_VM_leaf_base(
 155     address entry_point,                // the entry point
 156     int     number_of_arguments,        // the number of arguments to pop after the call
 157     Label*  retaddr = nullptr
 158   );
 159 
 160   virtual void call_VM_leaf_base(
 161     address entry_point,                // the entry point
 162     int     number_of_arguments,        // the number of arguments to pop after the call
 163     Label&  retaddr) {
 164     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
 165   }
 166 
 167   virtual void call_VM_base(           // returns the register containing the thread upon return
 168     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
 169     Register java_thread,              // the thread if computed before     ; use noreg otherwise
 170     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
 171     Label*   return_pc,                // to set up last_Java_frame; use nullptr otherwise
 172     address  entry_point,              // the entry point
 173     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
 174     bool     check_exceptions          // whether to check for pending exceptions after return
 175   );
 176 
 177   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions);
 178 
 179   virtual void check_and_handle_earlyret(Register java_thread);
 180   virtual void check_and_handle_popframe(Register java_thread);
 181 
 182   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 183   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 184   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 185   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 186 
 187   void movoop(Register dst, jobject obj);
 188   void mov_metadata(Register dst, Metadata* obj);
 189   void bang_stack_size(Register size, Register tmp);
 190   void set_narrow_oop(Register dst, jobject obj);
 191   void set_narrow_klass(Register dst, Klass* k);
 192 
 193   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 194   void access_load_at(BasicType type, DecoratorSet decorators, Register dst,
 195                       Address src, Register tmp1, Register tmp2);
 196   void access_store_at(BasicType type, DecoratorSet decorators, Address dst,
 197                        Register val, Register tmp1, Register tmp2, Register tmp3);
 198   void load_klass(Register dst, Register src, Register tmp = t0);
 199   void load_narrow_klass_compact(Register dst, Register src);
 200   void load_narrow_klass(Register dst, Register src);
 201   void store_klass(Register dst, Register src, Register tmp = t0);
 202   void cmp_klass_beq(Register obj, Register klass,
 203                      Register tmp1, Register tmp2,
 204                      Label &L, bool is_far = false);
 205   void cmp_klass_bne(Register obj, Register klass,
 206                      Register tmp1, Register tmp2,
 207                      Label &L, bool is_far = false);
 208 
 209   void encode_klass_not_null(Register r, Register tmp = t0);
 210   void decode_klass_not_null(Register r, Register tmp = t0);
 211   void encode_klass_not_null(Register dst, Register src, Register tmp);
 212   void decode_klass_not_null(Register dst, Register src, Register tmp);
 213   void decode_heap_oop_not_null(Register r);
 214   void decode_heap_oop_not_null(Register dst, Register src);
 215   void decode_heap_oop(Register d, Register s);
 216   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 217   void encode_heap_oop_not_null(Register r);
 218   void encode_heap_oop_not_null(Register dst, Register src);
 219   void encode_heap_oop(Register d, Register s);
 220   void encode_heap_oop(Register r) { encode_heap_oop(r, r); };
 221   void load_heap_oop(Register dst, Address src, Register tmp1,
 222                      Register tmp2, DecoratorSet decorators = 0);
 223   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 224                               Register tmp2, DecoratorSet decorators = 0);
 225   void store_heap_oop(Address dst, Register val, Register tmp1,
 226                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 227 
 228   void store_klass_gap(Register dst, Register src);
 229 
 230   // currently unimplemented
 231   // Used for storing null. All other oop constants should be
 232   // stored using routines that take a jobject.
 233   void store_heap_oop_null(Address dst);
 234 
 235   // This dummy is to prevent a call to store_heap_oop from
 236   // converting a zero (linked null) into a Register by giving
 237   // the compiler two choices it can't resolve
 238 
 239   void store_heap_oop(Address dst, void* dummy);
 240 
 241   // Support for null-checks
 242   //
 243   // Generates code that causes a null OS exception if the content of reg is null.
 244   // If the accessed location is M[reg + offset] and the offset is known, provide the
 245   // offset. No explicit code generateion is needed if the offset is within a certain
 246   // range (0 <= offset <= page_size).
 247 
 248   virtual void null_check(Register reg, int offset = -1);
 249   static bool needs_explicit_null_check(intptr_t offset);
 250   static bool uses_implicit_null_check(void* address);
 251 
 252   // interface method calling
 253   void lookup_interface_method(Register recv_klass,
 254                                Register intf_klass,
 255                                RegisterOrConstant itable_index,
 256                                Register method_result,
 257                                Register scan_tmp,
 258                                Label& no_such_interface,
 259                                bool return_method = true);
 260 
 261   void lookup_interface_method_stub(Register recv_klass,
 262                                     Register holder_klass,
 263                                     Register resolved_klass,
 264                                     Register method_result,
 265                                     Register temp_reg,
 266                                     Register temp_reg2,
 267                                     int itable_index,
 268                                     Label& L_no_such_interface);
 269 
 270   // virtual method calling
 271   // n.n. x86 allows RegisterOrConstant for vtable_index
 272   void lookup_virtual_method(Register recv_klass,
 273                              RegisterOrConstant vtable_index,
 274                              Register method_result);
 275 
 276   // Form an address from base + offset in Rd. Rd my or may not
 277   // actually be used: you must use the Address that is returned. It
 278   // is up to you to ensure that the shift provided matches the size
 279   // of your data.
 280   Address form_address(Register Rd, Register base, int64_t byte_offset);
 281 
 282   // Sometimes we get misaligned loads and stores, usually from Unsafe
 283   // accesses, and these can exceed the offset range.
 284   Address legitimize_address(Register Rd, const Address &adr) {
 285     if (adr.getMode() == Address::base_plus_offset) {
 286       if (!is_simm12(adr.offset())) {
 287         return form_address(Rd, adr.base(), adr.offset());
 288       }
 289     }
 290     return adr;
 291   }
 292 
 293   // allocation
 294   void tlab_allocate(
 295     Register obj,                   // result: pointer to object after successful allocation
 296     Register var_size_in_bytes,     // object size in bytes if unknown at compile time; invalid otherwise
 297     int      con_size_in_bytes,     // object size in bytes if   known at compile time
 298     Register tmp1,                  // temp register
 299     Register tmp2,                  // temp register
 300     Label&   slow_case,             // continuation point of fast allocation fails
 301     bool     is_far = false
 302   );
 303 
 304   // Test sub_klass against super_klass, with fast and slow paths.
 305 
 306   // The fast path produces a tri-state answer: yes / no / maybe-slow.
 307   // One of the three labels can be null, meaning take the fall-through.
 308   // If super_check_offset is -1, the value is loaded up from super_klass.
 309   // No registers are killed, except tmp_reg
 310   void check_klass_subtype_fast_path(Register sub_klass,
 311                                      Register super_klass,
 312                                      Register tmp_reg,
 313                                      Label* L_success,
 314                                      Label* L_failure,
 315                                      Label* L_slow_path,
 316                                      Register super_check_offset = noreg);
 317 
 318   // The reset of the type check; must be wired to a corresponding fast path.
 319   // It does not repeat the fast path logic, so don't use it standalone.
 320   // The tmp1_reg and tmp2_reg can be noreg, if no temps are available.
 321   // Updates the sub's secondary super cache as necessary.
 322   void check_klass_subtype_slow_path(Register sub_klass,
 323                                      Register super_klass,
 324                                      Register tmp1_reg,
 325                                      Register tmp2_reg,
 326                                      Label* L_success,
 327                                      Label* L_failure,
 328                                      bool set_cond_codes = false);
 329 
 330   void check_klass_subtype_slow_path_linear(Register sub_klass,
 331                                             Register super_klass,
 332                                             Register tmp1_reg,
 333                                             Register tmp2_reg,
 334                                             Label* L_success,
 335                                             Label* L_failure,
 336                                             bool set_cond_codes = false);
 337 
 338   void check_klass_subtype_slow_path_table(Register sub_klass,
 339                                            Register super_klass,
 340                                            Register tmp1_reg,
 341                                            Register tmp2_reg,
 342                                            Label* L_success,
 343                                            Label* L_failure,
 344                                            bool set_cond_codes = false);
 345 
 346   // If r is valid, return r.
 347   // If r is invalid, remove a register r2 from available_regs, add r2
 348   // to regs_to_push, then return r2.
 349   Register allocate_if_noreg(const Register r,
 350                              RegSetIterator<Register> &available_regs,
 351                              RegSet &regs_to_push);
 352 
 353   // Secondary subtype checking
 354   void lookup_secondary_supers_table_var(Register sub_klass,
 355                                          Register r_super_klass,
 356                                          Register result,
 357                                          Register tmp1,
 358                                          Register tmp2,
 359                                          Register tmp3,
 360                                          Register tmp4,
 361                                          Label *L_success);
 362 
 363   void population_count(Register dst, Register src, Register tmp1, Register tmp2);
 364 
 365   // As above, but with a constant super_klass.
 366   // The result is in Register result, not the condition codes.
 367   bool lookup_secondary_supers_table_const(Register r_sub_klass,
 368                                            Register r_super_klass,
 369                                            Register result,
 370                                            Register tmp1,
 371                                            Register tmp2,
 372                                            Register tmp3,
 373                                            Register tmp4,
 374                                            u1 super_klass_slot,
 375                                            bool stub_is_near = false);
 376 
 377   void verify_secondary_supers_table(Register r_sub_klass,
 378                                      Register r_super_klass,
 379                                      Register result,
 380                                      Register tmp1,
 381                                      Register tmp2,
 382                                      Register tmp3);
 383 
 384   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
 385                                                Register r_array_base,
 386                                                Register r_array_index,
 387                                                Register r_bitmap,
 388                                                Register result,
 389                                                Register tmp,
 390                                                bool is_stub = true);
 391 
 392   void check_klass_subtype(Register sub_klass,
 393                            Register super_klass,
 394                            Register tmp_reg,
 395                            Label& L_success);
 396 
 397   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
 398 
 399   void profile_receiver_type(Register recv, Register mdp, int mdp_offset);
 400 
 401   // only if +VerifyOops
 402   void _verify_oop(Register reg, const char* s, const char* file, int line);
 403   void _verify_oop_addr(Address addr, const char* s, const char* file, int line);
 404 
 405   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
 406     if (VerifyOops) {
 407       _verify_oop(reg, s, file, line);
 408     }
 409   }
 410   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
 411     if (VerifyOops) {
 412       _verify_oop_addr(reg, s, file, line);
 413     }
 414   }
 415 
 416   void _verify_method_ptr(Register reg, const char* msg, const char* file, int line) {}
 417   void _verify_klass_ptr(Register reg, const char* msg, const char* file, int line) {}
 418 
 419 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
 420 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
 421 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
 422 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
 423 #define verify_klass_ptr(reg) _verify_method_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
 424 
 425   // A more convenient access to fence for our purposes
 426   // We used four bit to indicate the read and write bits in the predecessors and successors,
 427   // and extended i for r, o for w if UseConservativeFence enabled.
 428   enum Membar_mask_bits {
 429     StoreStore = 0b0101,               // (pred = w   + succ = w)
 430     LoadStore  = 0b1001,               // (pred = r   + succ = w)
 431     StoreLoad  = 0b0110,               // (pred = w   + succ = r)
 432     LoadLoad   = 0b1010,               // (pred = r   + succ = r)
 433     AnyAny     = LoadStore | StoreLoad // (pred = rw  + succ = rw)
 434   };
 435 
 436   void membar(uint32_t order_constraint);
 437 
 438  private:
 439 
 440   static void membar_mask_to_pred_succ(uint32_t order_constraint,
 441                                        uint32_t& predecessor, uint32_t& successor) {
 442     predecessor = (order_constraint >> 2) & 0x3;
 443     successor = order_constraint & 0x3;
 444 
 445     // extend rw -> iorw:
 446     // 01(w) -> 0101(ow)
 447     // 10(r) -> 1010(ir)
 448     // 11(rw)-> 1111(iorw)
 449     if (UseConservativeFence) {
 450       predecessor |= predecessor << 2;
 451       successor   |= successor << 2;
 452     }
 453   }
 454 
 455   static int pred_succ_to_membar_mask(uint32_t predecessor, uint32_t successor) {
 456     return ((predecessor & 0x3) << 2) | (successor & 0x3);
 457   }
 458 
 459  public:
 460 
 461   void cmodx_fence();
 462 
 463   void pause() {
 464     // Zihintpause
 465     // PAUSE is encoded as a FENCE instruction with pred=W, succ=0, fm=0, rd=x0, and rs1=x0.
 466     Assembler::fence(w, 0);
 467   }
 468 
 469   // prints msg, dumps registers and stops execution
 470   void stop(const char* msg);
 471 
 472   static void debug64(char* msg, int64_t pc, int64_t regs[]);
 473 
 474   void unimplemented(const char* what = "");
 475 
 476   void should_not_reach_here() { stop("should not reach here"); }
 477 
 478   static address target_addr_for_insn(address insn_addr);
 479 
 480   // Required platform-specific helpers for Label::patch_instructions.
 481   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 482   static int pd_patch_instruction_size(address branch, address target);
 483   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 484     pd_patch_instruction_size(branch, target);
 485   }
 486   static address pd_call_destination(address branch) {
 487     return target_addr_for_insn(branch);
 488   }
 489 
 490   static int patch_oop(address insn_addr, address o);
 491 
 492   static address get_target_of_li32(address insn_addr);
 493   static int patch_imm_in_li32(address branch, int32_t target);
 494 
 495   // Return whether code is emitted to a scratch blob.
 496   virtual bool in_scratch_emit_size() {
 497     return false;
 498   }
 499 
 500   address emit_reloc_call_address_stub(int insts_call_instruction_offset, address target);
 501   static int max_reloc_call_address_stub_size();
 502 
 503   void emit_static_call_stub();
 504   static int static_call_stub_size();
 505 
 506   // The following 4 methods return the offset of the appropriate move instruction
 507 
 508   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 509   int load_unsigned_byte(Register dst, Address src);
 510   int load_unsigned_short(Register dst, Address src);
 511 
 512   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 513   int load_signed_byte(Register dst, Address src);
 514   int load_signed_short(Register dst, Address src);
 515 
 516   // Load and store values by size and signed-ness
 517   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 518   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 519 
 520   // Misaligned loads, will use the best way, according to the AvoidUnalignedAccess flag
 521   void load_short_misaligned(Register dst, Address src, Register tmp, bool is_signed, int granularity = 1);
 522   void load_int_misaligned(Register dst, Address src, Register tmp, bool is_signed, int granularity = 1);
 523   void load_long_misaligned(Register dst, Address src, Register tmp, int granularity = 1);
 524 
 525  public:
 526   // Standard pseudo instructions
 527   inline void nop() {
 528     addi(x0, x0, 0);
 529   }
 530 
 531   inline void mv(Register Rd, Register Rs) {
 532     if (Rd != Rs) {
 533       addi(Rd, Rs, 0);
 534     }
 535   }
 536 
 537   inline void notr(Register Rd, Register Rs) {
 538     if (do_compress_zcb(Rd, Rs) && (Rd == Rs)) {
 539       c_not(Rd);
 540     } else {
 541       xori(Rd, Rs, -1);
 542     }
 543   }
 544 
 545   inline void neg(Register Rd, Register Rs) {
 546     sub(Rd, x0, Rs);
 547   }
 548 
 549   inline void negw(Register Rd, Register Rs) {
 550     subw(Rd, x0, Rs);
 551   }
 552 
 553   inline void sext_w(Register Rd, Register Rs) {
 554     addiw(Rd, Rs, 0);
 555   }
 556 
 557   inline void zext_b(Register Rd, Register Rs) {
 558     if (do_compress_zcb(Rd, Rs) && (Rd == Rs)) {
 559       c_zext_b(Rd);
 560     } else {
 561       andi(Rd, Rs, 0xFF);
 562     }
 563   }
 564 
 565   inline void seqz(Register Rd, Register Rs) {
 566     sltiu(Rd, Rs, 1);
 567   }
 568 
 569   inline void snez(Register Rd, Register Rs) {
 570     sltu(Rd, x0, Rs);
 571   }
 572 
 573   inline void sltz(Register Rd, Register Rs) {
 574     slt(Rd, Rs, x0);
 575   }
 576 
 577   inline void sgtz(Register Rd, Register Rs) {
 578     slt(Rd, x0, Rs);
 579   }
 580 
 581   // Bit-manipulation extension pseudo instructions
 582   // zero extend word
 583   inline void zext_w(Register Rd, Register Rs) {
 584     assert(UseZba, "must be");
 585     if (do_compress_zcb(Rd, Rs) && (Rd == Rs)) {
 586       c_zext_w(Rd);
 587     } else {
 588       add_uw(Rd, Rs, zr);
 589     }
 590   }
 591 
 592   // Floating-point data-processing pseudo instructions
 593   inline void fmv_s(FloatRegister Rd, FloatRegister Rs) {
 594     if (Rd != Rs) {
 595       fsgnj_s(Rd, Rs, Rs);
 596     }
 597   }
 598 
 599   inline void fabs_s(FloatRegister Rd, FloatRegister Rs) {
 600     fsgnjx_s(Rd, Rs, Rs);
 601   }
 602 
 603   inline void fneg_s(FloatRegister Rd, FloatRegister Rs) {
 604     fsgnjn_s(Rd, Rs, Rs);
 605   }
 606 
 607   inline void fmv_d(FloatRegister Rd, FloatRegister Rs) {
 608     if (Rd != Rs) {
 609       fsgnj_d(Rd, Rs, Rs);
 610     }
 611   }
 612 
 613   inline void fabs_d(FloatRegister Rd, FloatRegister Rs) {
 614     fsgnjx_d(Rd, Rs, Rs);
 615   }
 616 
 617   inline void fneg_d(FloatRegister Rd, FloatRegister Rs) {
 618     fsgnjn_d(Rd, Rs, Rs);
 619   }
 620 
 621   // Control and status pseudo instructions
 622   void csrr(Register Rd, unsigned csr);         // read csr
 623   void csrw(unsigned csr, Register Rs);         // write csr
 624   void csrs(unsigned csr, Register Rs);         // set bits in csr
 625   void csrc(unsigned csr, Register Rs);         // clear bits in csr
 626   void csrwi(unsigned csr, unsigned imm);
 627   void csrsi(unsigned csr, unsigned imm);
 628   void csrci(unsigned csr, unsigned imm);
 629   void frcsr(Register Rd) { csrr(Rd, CSR_FCSR); }; // read float-point csr
 630   void fscsr(Register Rd, Register Rs);            // swap float-point csr
 631   void fscsr(Register Rs);                         // write float-point csr
 632   void frrm(Register Rd) { csrr(Rd, CSR_FRM); };   // read float-point rounding mode
 633   void fsrm(Register Rd, Register Rs);             // swap float-point rounding mode
 634   void fsrm(Register Rs);                          // write float-point rounding mode
 635   void fsrmi(Register Rd, unsigned imm);
 636   void fsrmi(unsigned imm);
 637   void frflags(Register Rd) { csrr(Rd, CSR_FFLAGS); }; // read float-point exception flags
 638   void fsflags(Register Rd, Register Rs);              // swap float-point exception flags
 639   void fsflags(Register Rs);                           // write float-point exception flags
 640   void fsflagsi(Register Rd, unsigned imm);
 641   void fsflagsi(unsigned imm);
 642   // Requires Zicntr
 643   void rdinstret(Register Rd) { csrr(Rd, CSR_INSTRET); }; // read instruction-retired counter
 644   void rdcycle(Register Rd)   { csrr(Rd, CSR_CYCLE); };   // read cycle counter
 645   void rdtime(Register Rd)    { csrr(Rd, CSR_TIME); };    // read time
 646 
 647   // Restore cpu control state after JNI call
 648   void restore_cpu_control_state_after_jni(Register tmp);
 649 
 650   // Control transfer pseudo instructions
 651   void beqz(Register Rs, const address dest);
 652   void bnez(Register Rs, const address dest);
 653   void blez(Register Rs, const address dest);
 654   void bgez(Register Rs, const address dest);
 655   void bltz(Register Rs, const address dest);
 656   void bgtz(Register Rs, const address dest);
 657 
 658   void cmov_eq(Register cmp1, Register cmp2, Register dst, Register src);
 659   void cmov_ne(Register cmp1, Register cmp2, Register dst, Register src);
 660   void cmov_le(Register cmp1, Register cmp2, Register dst, Register src);
 661   void cmov_leu(Register cmp1, Register cmp2, Register dst, Register src);
 662   void cmov_ge(Register cmp1, Register cmp2, Register dst, Register src);
 663   void cmov_geu(Register cmp1, Register cmp2, Register dst, Register src);
 664   void cmov_lt(Register cmp1, Register cmp2, Register dst, Register src);
 665   void cmov_ltu(Register cmp1, Register cmp2, Register dst, Register src);
 666   void cmov_gt(Register cmp1, Register cmp2, Register dst, Register src);
 667   void cmov_gtu(Register cmp1, Register cmp2, Register dst, Register src);
 668 
 669   void cmov_cmp_fp_eq(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 670   void cmov_cmp_fp_ne(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 671   void cmov_cmp_fp_le(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 672   void cmov_cmp_fp_ge(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 673   void cmov_cmp_fp_lt(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 674   void cmov_cmp_fp_gt(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 675 
 676   void cmov_fp_eq(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 677   void cmov_fp_ne(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 678   void cmov_fp_le(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 679   void cmov_fp_leu(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 680   void cmov_fp_ge(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 681   void cmov_fp_geu(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 682   void cmov_fp_lt(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 683   void cmov_fp_ltu(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 684   void cmov_fp_gt(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 685   void cmov_fp_gtu(Register cmp1, Register cmp2, FloatRegister dst, FloatRegister src, bool is_single);
 686 
 687   void cmov_fp_cmp_fp_eq(FloatRegister cmp1, FloatRegister cmp2, FloatRegister dst, FloatRegister src, bool cmp_single, bool cmov_single);
 688   void cmov_fp_cmp_fp_ne(FloatRegister cmp1, FloatRegister cmp2, FloatRegister dst, FloatRegister src, bool cmp_single, bool cmov_single);
 689   void cmov_fp_cmp_fp_le(FloatRegister cmp1, FloatRegister cmp2, FloatRegister dst, FloatRegister src, bool cmp_single, bool cmov_single);
 690   void cmov_fp_cmp_fp_ge(FloatRegister cmp1, FloatRegister cmp2, FloatRegister dst, FloatRegister src, bool cmp_single, bool cmov_single);
 691   void cmov_fp_cmp_fp_lt(FloatRegister cmp1, FloatRegister cmp2, FloatRegister dst, FloatRegister src, bool cmp_single, bool cmov_single);
 692   void cmov_fp_cmp_fp_gt(FloatRegister cmp1, FloatRegister cmp2, FloatRegister dst, FloatRegister src, bool cmp_single, bool cmov_single);
 693 
 694  public:
 695   // We try to follow risc-v asm menomics.
 696   // But as we don't layout a reachable GOT,
 697   // we often need to resort to movptr, li <48imm>.
 698   // https://github.com/riscv-non-isa/riscv-asm-manual/blob/main/src/asm-manual.adoc
 699 
 700   // Hotspot only use the standard calling convention using x1/ra.
 701   // The alternative calling convection using x5/t0 is not used.
 702   // Using x5 as a temp causes the CPU to mispredict returns.
 703 
 704   // JALR, return address stack updates:
 705   // | rd is x1/x5 | rs1 is x1/x5 | rd=rs1 | RAS action
 706   // | ----------- | ------------ | ------ |-------------
 707   // |     No      |      No      |   -    | None
 708   // |     No      |      Yes     |   -    | Pop
 709   // |     Yes     |      No      |   -    | Push
 710   // |     Yes     |      Yes     |   No   | Pop, then push
 711   // |     Yes     |      Yes     |   Yes  | Push
 712   //
 713   // JAL, return address stack updates:
 714   // | rd is x1/x5 | RAS action
 715   // | ----------- | ----------
 716   // |     Yes     | Push
 717   // |     No      | None
 718   //
 719   // JUMPs   uses Rd = x0/zero and Rs = x6/t1 or imm
 720   // CALLS   uses Rd = x1/ra   and Rs = x6/t1 or imm (or x1/ra*)
 721   // RETURNS uses Rd = x0/zero and Rs = x1/ra
 722   // *use of x1/ra should not normally be used, special case only.
 723 
 724   // jump: jal x0, offset
 725   // For long reach uses temp register for:
 726   // la + jr
 727   void j(const address dest, Register temp = t1);
 728   void j(const Address &dest, Register temp = t1);
 729   void j(Label &l, Register temp = noreg);
 730 
 731   // jump register: jalr x0, offset(rs)
 732   void jr(Register Rd, int32_t offset = 0);
 733 
 734   // call: la + jalr x1
 735   void call(const address dest, Register temp = t1);
 736 
 737   // jalr: jalr x1, offset(rs)
 738   void jalr(Register Rs, int32_t offset = 0);
 739 
 740   // Emit a runtime call. Only invalidates the tmp register which
 741   // is used to keep the entry address for jalr/movptr.
 742   // Uses call() for intra code cache, else movptr + jalr.
 743   // Clobebrs t1
 744   void rt_call(address dest, Register tmp = t1);
 745 
 746   // ret: jalr x0, 0(x1)
 747   inline void ret() {
 748     Assembler::jalr(x0, x1, 0);
 749   }
 750 
 751   //label
 752   void beqz(Register Rs, Label &l, bool is_far = false);
 753   void bnez(Register Rs, Label &l, bool is_far = false);
 754   void blez(Register Rs, Label &l, bool is_far = false);
 755   void bgez(Register Rs, Label &l, bool is_far = false);
 756   void bltz(Register Rs, Label &l, bool is_far = false);
 757   void bgtz(Register Rs, Label &l, bool is_far = false);
 758 
 759   void beq (Register Rs1, Register Rs2, Label &L, bool is_far = false);
 760   void bne (Register Rs1, Register Rs2, Label &L, bool is_far = false);
 761   void blt (Register Rs1, Register Rs2, Label &L, bool is_far = false);
 762   void bge (Register Rs1, Register Rs2, Label &L, bool is_far = false);
 763   void bltu(Register Rs1, Register Rs2, Label &L, bool is_far = false);
 764   void bgeu(Register Rs1, Register Rs2, Label &L, bool is_far = false);
 765 
 766   void bgt (Register Rs, Register Rt, const address dest);
 767   void ble (Register Rs, Register Rt, const address dest);
 768   void bgtu(Register Rs, Register Rt, const address dest);
 769   void bleu(Register Rs, Register Rt, const address dest);
 770 
 771   void bgt (Register Rs, Register Rt, Label &l, bool is_far = false);
 772   void ble (Register Rs, Register Rt, Label &l, bool is_far = false);
 773   void bgtu(Register Rs, Register Rt, Label &l, bool is_far = false);
 774   void bleu(Register Rs, Register Rt, Label &l, bool is_far = false);
 775 
 776 #define INSN_ENTRY_RELOC(result_type, header)                               \
 777   result_type header {                                                      \
 778     guarantee(rtype == relocInfo::internal_word_type,                       \
 779               "only internal_word_type relocs make sense here");            \
 780     relocate(InternalAddress(dest).rspec());                                \
 781     IncompressibleScope scope(this);  /* relocations */
 782 
 783 #define INSN(NAME)                                                                                       \
 784   void NAME(Register Rs1, Register Rs2, const address dest) {                                            \
 785     assert_cond(dest != nullptr);                                                                        \
 786     int64_t offset = dest - pc();                                                                        \
 787     guarantee(is_simm13(offset) && is_even(offset),                                                      \
 788               "offset is invalid: is_simm_13: %s offset: " INT64_FORMAT,                                 \
 789               BOOL_TO_STR(is_simm13(offset)), offset);                                                   \
 790     Assembler::NAME(Rs1, Rs2, offset);                                                                   \
 791   }                                                                                                      \
 792   INSN_ENTRY_RELOC(void, NAME(Register Rs1, Register Rs2, address dest, relocInfo::relocType rtype))     \
 793     NAME(Rs1, Rs2, dest);                                                                                \
 794   }
 795 
 796   INSN(beq);
 797   INSN(bne);
 798   INSN(bge);
 799   INSN(bgeu);
 800   INSN(blt);
 801   INSN(bltu);
 802 
 803 #undef INSN
 804 
 805 #undef INSN_ENTRY_RELOC
 806 
 807   void float_beq(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 808   void float_bne(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 809   void float_ble(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 810   void float_bge(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 811   void float_blt(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 812   void float_bgt(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 813 
 814   void double_beq(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 815   void double_bne(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 816   void double_ble(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 817   void double_bge(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 818   void double_blt(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 819   void double_bgt(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 820 
 821 private:
 822   // The signed 20-bit upper imm can materialize at most negative 0xF...F80000000, two G.
 823   // The following signed 12-bit imm can at max subtract 0x800, two K, from that previously loaded two G.
 824   bool is_valid_32bit_offset(int64_t x) {
 825     constexpr int64_t twoG = (2 * G);
 826     constexpr int64_t twoK = (2 * K);
 827     return x < (twoG - twoK) && x >= (-twoG - twoK);
 828   }
 829 
 830   // Ensure that the auipc can reach the destination at x from anywhere within
 831   // the code cache so that if it is relocated we know it will still reach.
 832   bool is_32bit_offset_from_codecache(int64_t x) {
 833     int64_t low  = (int64_t)CodeCache::low_bound();
 834     int64_t high = (int64_t)CodeCache::high_bound();
 835     return is_valid_32bit_offset(x - low) && is_valid_32bit_offset(x - high);
 836   }
 837 
 838 public:
 839   // Stack push and pop individual 64 bit registers
 840   void push_reg(Register Rs);
 841   void pop_reg(Register Rd);
 842 
 843   int push_reg(RegSet regset, Register stack);
 844   int pop_reg(RegSet regset, Register stack);
 845 
 846   int push_fp(FloatRegSet regset, Register stack);
 847   int pop_fp(FloatRegSet regset, Register stack);
 848 
 849 #ifdef COMPILER2
 850   int push_v(VectorRegSet regset, Register stack);
 851   int pop_v(VectorRegSet regset, Register stack);
 852 #endif // COMPILER2
 853 
 854   // Push and pop everything that might be clobbered by a native
 855   // runtime call except t0 and t1. (They are always
 856   // temporary registers, so we don't have to protect them.)
 857   // Additional registers can be excluded in a passed RegSet.
 858   void push_call_clobbered_registers_except(RegSet exclude);
 859   void pop_call_clobbered_registers_except(RegSet exclude);
 860 
 861   void push_call_clobbered_registers() {
 862     push_call_clobbered_registers_except(RegSet());
 863   }
 864   void pop_call_clobbered_registers() {
 865     pop_call_clobbered_registers_except(RegSet());
 866   }
 867 
 868   void push_CPU_state(bool save_vectors = false, int vector_size_in_bytes = 0);
 869   void pop_CPU_state(bool restore_vectors = false, int vector_size_in_bytes = 0);
 870 
 871   void push_cont_fastpath(Register java_thread = xthread);
 872   void pop_cont_fastpath(Register java_thread = xthread);
 873 
 874   // if heap base register is used - reinit it with the correct value
 875   void reinit_heapbase();
 876 
 877   void bind(Label& L) {
 878     Assembler::bind(L);
 879     // fences across basic blocks should not be merged
 880     code()->clear_last_merge_candidate();
 881   }
 882 
 883   typedef void (MacroAssembler::* compare_and_branch_insn)(Register Rs1, Register Rs2, const address dest);
 884   typedef void (MacroAssembler::* compare_and_branch_label_insn)(Register Rs1, Register Rs2, Label &L, bool is_far);
 885   typedef void (MacroAssembler::* jal_jalr_insn)(Register Rt, address dest);
 886 
 887   void wrap_label(Register r, Label &L, jal_jalr_insn insn);
 888   void wrap_label(Register r1, Register r2, Label &L,
 889                   compare_and_branch_insn insn,
 890                   compare_and_branch_label_insn neg_insn, bool is_far = false);
 891 
 892   void la(Register Rd, Label &label);
 893   void la(Register Rd, const address addr);
 894   void la(Register Rd, const address addr, int32_t &offset);
 895   void la(Register Rd, const Address &adr);
 896 
 897   void li16u(Register Rd, uint16_t imm);
 898   void li32(Register Rd, int32_t imm);
 899   void li  (Register Rd, int64_t imm);  // optimized load immediate
 900 
 901   // mv
 902   void mv(Register Rd, address addr)                  { li(Rd, (int64_t)addr); }
 903   void mv(Register Rd, address addr, int32_t &offset) {
 904     // Split address into a lower 12-bit sign-extended offset and the remainder,
 905     // so that the offset could be encoded in jalr or load/store instruction.
 906     offset = ((int32_t)(int64_t)addr << 20) >> 20;
 907     li(Rd, (int64_t)addr - offset);
 908   }
 909 
 910   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 911   inline void mv(Register Rd, T o)                    { li(Rd, (int64_t)o); }
 912 
 913   void mv(Register Rd, RegisterOrConstant src) {
 914     if (src.is_register()) {
 915       mv(Rd, src.as_register());
 916     } else {
 917       mv(Rd, src.as_constant());
 918     }
 919   }
 920 
 921   // Generates a load of a 48-bit constant which can be
 922   // patched to any 48-bit constant, i.e. address.
 923   // If common case supply additional temp register
 924   // to shorten the instruction sequence.
 925   void movptr(Register Rd, const Address &addr, Register tmp = noreg);
 926   void movptr(Register Rd, address addr, Register tmp = noreg);
 927   void movptr(Register Rd, address addr, int32_t &offset, Register tmp = noreg);
 928 
 929  private:
 930   void movptr1(Register Rd, uintptr_t addr, int32_t &offset);
 931   void movptr2(Register Rd, uintptr_t addr, int32_t &offset, Register tmp);
 932  public:
 933   // float imm move
 934   static bool can_hf_imm_load(short imm);
 935   static bool can_fp_imm_load(float imm);
 936   static bool can_dp_imm_load(double imm);
 937   void fli_h(FloatRegister Rd, short imm);
 938   void fli_s(FloatRegister Rd, float imm);
 939   void fli_d(FloatRegister Rd, double imm);
 940 
 941   // arith
 942   void add (Register Rd, Register Rn, int64_t increment, Register tmp = t0);
 943   void sub (Register Rd, Register Rn, int64_t decrement, Register tmp = t0);
 944   void addw(Register Rd, Register Rn, int64_t increment, Register tmp = t0);
 945   void subw(Register Rd, Register Rn, int64_t decrement, Register tmp = t0);
 946 
 947   void subi(Register Rd, Register Rn, int64_t decrement) {
 948     assert(is_simm12(-decrement), "Must be");
 949     addi(Rd, Rn, -decrement);
 950   }
 951 
 952   void subiw(Register Rd, Register Rn, int64_t decrement) {
 953     assert(is_simm12(-decrement), "Must be");
 954     addiw(Rd, Rn, -decrement);
 955   }
 956 
 957 #define INSN(NAME)                                               \
 958   inline void NAME(Register Rd, Register Rs1, Register Rs2) {    \
 959     Assembler::NAME(Rd, Rs1, Rs2);                               \
 960   }
 961 
 962   INSN(add);
 963   INSN(addw);
 964   INSN(sub);
 965   INSN(subw);
 966 
 967 #undef INSN
 968 
 969   // logic
 970   void andrw(Register Rd, Register Rs1, Register Rs2);
 971   void orrw(Register Rd, Register Rs1, Register Rs2);
 972   void xorrw(Register Rd, Register Rs1, Register Rs2);
 973 
 974   // logic with negate
 975   void andn(Register Rd, Register Rs1, Register Rs2);
 976   void orn(Register Rd, Register Rs1, Register Rs2);
 977 
 978   // reverse bytes
 979   void revbw(Register Rd, Register Rs, Register tmp1 = t0, Register tmp2= t1);  // reverse bytes in lower word, sign-extend
 980   void revb(Register Rd, Register Rs, Register tmp1 = t0, Register tmp2 = t1);  // reverse bytes in doubleword
 981 
 982   void ror(Register dst, Register src, Register shift, Register tmp = t0);
 983   void ror(Register dst, Register src, uint32_t shift, Register tmp = t0);
 984   void rolw(Register dst, Register src, uint32_t shift, Register tmp = t0);
 985 
 986   void orptr(Address adr, RegisterOrConstant src, Register tmp1 = t0, Register tmp2 = t1);
 987 
 988 // Load and Store Instructions
 989 #define INSN_ENTRY_RELOC(result_type, header)                               \
 990   result_type header {                                                      \
 991     guarantee(rtype == relocInfo::internal_word_type,                       \
 992               "only internal_word_type relocs make sense here");            \
 993     relocate(InternalAddress(dest).rspec());                                \
 994     IncompressibleScope scope(this);  /* relocations */
 995 
 996 #define INSN(NAME)                                                                                 \
 997   void NAME(Register Rd, address dest) {                                                           \
 998     assert_cond(dest != nullptr);                                                                  \
 999     if (CodeCache::contains(dest)) {                                                               \
1000       int64_t distance = dest - pc();                                                              \
1001       assert(is_valid_32bit_offset(distance), "Must be");                                          \
1002       auipc(Rd, (int32_t)distance + 0x800);                                                        \
1003       Assembler::NAME(Rd, Rd, ((int32_t)distance << 20) >> 20);                                    \
1004     } else {                                                                                       \
1005       int32_t offset = 0;                                                                          \
1006       movptr(Rd, dest, offset);                                                                    \
1007       Assembler::NAME(Rd, Rd, offset);                                                             \
1008     }                                                                                              \
1009   }                                                                                                \
1010   INSN_ENTRY_RELOC(void, NAME(Register Rd, address dest, relocInfo::relocType rtype))              \
1011     NAME(Rd, dest);                                                                                \
1012   }                                                                                                \
1013   void NAME(Register Rd, const Address &adr, Register temp = t0) {                                 \
1014     switch (adr.getMode()) {                                                                       \
1015       case Address::literal: {                                                                     \
1016         relocate(adr.rspec(), [&] {                                                                \
1017           NAME(Rd, adr.target());                                                                  \
1018         });                                                                                        \
1019         break;                                                                                     \
1020       }                                                                                            \
1021       case Address::base_plus_offset: {                                                            \
1022         if (is_simm12(adr.offset())) {                                                             \
1023           Assembler::NAME(Rd, adr.base(), adr.offset());                                           \
1024         } else {                                                                                   \
1025           int32_t offset = ((int32_t)adr.offset() << 20) >> 20;                                    \
1026           if (Rd == adr.base()) {                                                                  \
1027             la(temp, Address(adr.base(), adr.offset() - offset));                                  \
1028             Assembler::NAME(Rd, temp, offset);                                                     \
1029           } else {                                                                                 \
1030             la(Rd, Address(adr.base(), adr.offset() - offset));                                    \
1031             Assembler::NAME(Rd, Rd, offset);                                                       \
1032           }                                                                                        \
1033         }                                                                                          \
1034         break;                                                                                     \
1035       }                                                                                            \
1036       default:                                                                                     \
1037         ShouldNotReachHere();                                                                      \
1038     }                                                                                              \
1039   }                                                                                                \
1040   void NAME(Register Rd, Label &L) {                                                               \
1041     wrap_label(Rd, L, &MacroAssembler::NAME);                                                      \
1042   }
1043 
1044   INSN(lb);
1045   INSN(lbu);
1046   INSN(lh);
1047   INSN(lhu);
1048   INSN(lw);
1049   INSN(lwu);
1050   INSN(ld);
1051 
1052 #undef INSN
1053 
1054 #define INSN(NAME)                                                                                 \
1055   void NAME(FloatRegister Rd, address dest, Register temp = t0) {                                  \
1056     assert_cond(dest != nullptr);                                                                  \
1057     if (CodeCache::contains(dest)) {                                                               \
1058       int64_t distance = dest - pc();                                                              \
1059       assert(is_valid_32bit_offset(distance), "Must be");                                          \
1060       auipc(temp, (int32_t)distance + 0x800);                                                      \
1061       Assembler::NAME(Rd, temp, ((int32_t)distance << 20) >> 20);                                  \
1062     } else {                                                                                       \
1063       int32_t offset = 0;                                                                          \
1064       movptr(temp, dest, offset);                                                                  \
1065       Assembler::NAME(Rd, temp, offset);                                                           \
1066     }                                                                                              \
1067   }                                                                                                \
1068   INSN_ENTRY_RELOC(void, NAME(FloatRegister Rd, address dest,                                      \
1069                               relocInfo::relocType rtype, Register temp = t0))                     \
1070     NAME(Rd, dest, temp);                                                                          \
1071   }                                                                                                \
1072   void NAME(FloatRegister Rd, const Address &adr, Register temp = t0) {                            \
1073     switch (adr.getMode()) {                                                                       \
1074       case Address::literal: {                                                                     \
1075         relocate(adr.rspec(), [&] {                                                                \
1076           NAME(Rd, adr.target(), temp);                                                            \
1077         });                                                                                        \
1078         break;                                                                                     \
1079       }                                                                                            \
1080       case Address::base_plus_offset: {                                                            \
1081         if (is_simm12(adr.offset())) {                                                             \
1082           Assembler::NAME(Rd, adr.base(), adr.offset());                                           \
1083         } else {                                                                                   \
1084           int32_t offset = ((int32_t)adr.offset() << 20) >> 20;                                    \
1085           la(temp, Address(adr.base(), adr.offset() - offset));                                    \
1086           Assembler::NAME(Rd, temp, offset);                                                       \
1087         }                                                                                          \
1088         break;                                                                                     \
1089       }                                                                                            \
1090       default:                                                                                     \
1091         ShouldNotReachHere();                                                                      \
1092     }                                                                                              \
1093   }
1094 
1095   INSN(flh);
1096   INSN(flw);
1097   INSN(fld);
1098 
1099 #undef INSN
1100 
1101 #define INSN(NAME, REGISTER)                                                                       \
1102   INSN_ENTRY_RELOC(void, NAME(REGISTER Rs, address dest,                                           \
1103                               relocInfo::relocType rtype, Register temp = t0))                     \
1104     NAME(Rs, dest, temp);                                                                          \
1105   }
1106 
1107   INSN(sb,  Register);
1108   INSN(sh,  Register);
1109   INSN(sw,  Register);
1110   INSN(sd,  Register);
1111   INSN(fsw, FloatRegister);
1112   INSN(fsd, FloatRegister);
1113 
1114 #undef INSN
1115 
1116 #define INSN(NAME)                                                                                 \
1117   void NAME(Register Rs, address dest, Register temp = t0) {                                       \
1118     assert_cond(dest != nullptr);                                                                  \
1119     assert_different_registers(Rs, temp);                                                          \
1120     if (CodeCache::contains(dest)) {                                                               \
1121       int64_t distance = dest - pc();                                                              \
1122       assert(is_valid_32bit_offset(distance), "Must be");                                          \
1123       auipc(temp, (int32_t)distance + 0x800);                                                      \
1124       Assembler::NAME(Rs, temp, ((int32_t)distance << 20) >> 20);                                  \
1125     } else {                                                                                       \
1126       int32_t offset = 0;                                                                          \
1127       movptr(temp, dest, offset);                                                                  \
1128       Assembler::NAME(Rs, temp, offset);                                                           \
1129     }                                                                                              \
1130   }                                                                                                \
1131   void NAME(Register Rs, const Address &adr, Register temp = t0) {                                 \
1132     switch (adr.getMode()) {                                                                       \
1133       case Address::literal: {                                                                     \
1134         assert_different_registers(Rs, temp);                                                      \
1135         relocate(adr.rspec(), [&] {                                                                \
1136           NAME(Rs, adr.target(), temp);                                                            \
1137         });                                                                                        \
1138         break;                                                                                     \
1139       }                                                                                            \
1140       case Address::base_plus_offset: {                                                            \
1141         if (is_simm12(adr.offset())) {                                                             \
1142           Assembler::NAME(Rs, adr.base(), adr.offset());                                           \
1143         } else {                                                                                   \
1144           assert_different_registers(Rs, temp);                                                    \
1145           int32_t offset = ((int32_t)adr.offset() << 20) >> 20;                                    \
1146           la(temp, Address(adr.base(), adr.offset() - offset));                                    \
1147           Assembler::NAME(Rs, temp, offset);                                                       \
1148         }                                                                                          \
1149         break;                                                                                     \
1150       }                                                                                            \
1151       default:                                                                                     \
1152         ShouldNotReachHere();                                                                      \
1153     }                                                                                              \
1154   }
1155 
1156   INSN(sb);
1157   INSN(sh);
1158   INSN(sw);
1159   INSN(sd);
1160 
1161 #undef INSN
1162 
1163 #define INSN(NAME)                                                                                 \
1164   void NAME(FloatRegister Rs, address dest, Register temp = t0) {                                  \
1165     assert_cond(dest != nullptr);                                                                  \
1166     if (CodeCache::contains(dest)) {                                                               \
1167       int64_t distance = dest - pc();                                                              \
1168       assert(is_valid_32bit_offset(distance), "Must be");                                          \
1169       auipc(temp, (int32_t)distance + 0x800);                                                      \
1170       Assembler::NAME(Rs, temp, ((int32_t)distance << 20) >> 20);                                  \
1171     } else {                                                                                       \
1172       int32_t offset = 0;                                                                          \
1173       movptr(temp, dest, offset);                                                                  \
1174       Assembler::NAME(Rs, temp, offset);                                                           \
1175     }                                                                                              \
1176   }                                                                                                \
1177   void NAME(FloatRegister Rs, const Address &adr, Register temp = t0) {                            \
1178     switch (adr.getMode()) {                                                                       \
1179       case Address::literal: {                                                                     \
1180         relocate(adr.rspec(), [&] {                                                                \
1181           NAME(Rs, adr.target(), temp);                                                            \
1182         });                                                                                        \
1183         break;                                                                                     \
1184       }                                                                                            \
1185       case Address::base_plus_offset: {                                                            \
1186         if (is_simm12(adr.offset())) {                                                             \
1187           Assembler::NAME(Rs, adr.base(), adr.offset());                                           \
1188         } else {                                                                                   \
1189           int32_t offset = ((int32_t)adr.offset() << 20) >> 20;                                    \
1190           la(temp, Address(adr.base(), adr.offset() - offset));                                    \
1191           Assembler::NAME(Rs, temp, offset);                                                       \
1192         }                                                                                          \
1193         break;                                                                                     \
1194       }                                                                                            \
1195       default:                                                                                     \
1196         ShouldNotReachHere();                                                                      \
1197     }                                                                                              \
1198   }
1199 
1200   INSN(fsw);
1201   INSN(fsd);
1202 
1203 #undef INSN
1204 
1205 #undef INSN_ENTRY_RELOC
1206 
1207   void cmpxchg(Register addr, Register expected,
1208                Register new_val,
1209                Assembler::operand_size size,
1210                Assembler::Aqrl acquire, Assembler::Aqrl release,
1211                Register result, bool result_as_bool = false);
1212   void weak_cmpxchg(Register addr, Register expected,
1213                     Register new_val,
1214                     Assembler::operand_size size,
1215                     Assembler::Aqrl acquire, Assembler::Aqrl release,
1216                     Register result);
1217   void cmpxchg_narrow_value_helper(Register addr, Register expected, Register new_val,
1218                                    Assembler::operand_size size,
1219                                    Register shift, Register mask, Register aligned_addr);
1220   void cmpxchg_narrow_value(Register addr, Register expected,
1221                             Register new_val,
1222                             Assembler::operand_size size,
1223                             Assembler::Aqrl acquire, Assembler::Aqrl release,
1224                             Register result, bool result_as_bool,
1225                             Register tmp1, Register tmp2, Register tmp3);
1226   void weak_cmpxchg_narrow_value(Register addr, Register expected,
1227                                  Register new_val,
1228                                  Assembler::operand_size size,
1229                                  Assembler::Aqrl acquire, Assembler::Aqrl release,
1230                                  Register result,
1231                                  Register tmp1, Register tmp2, Register tmp3);
1232 
1233   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1234   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1235   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1236   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1237 
1238   void atomic_xchg(Register prev, Register newv, Register addr);
1239   void atomic_xchgw(Register prev, Register newv, Register addr);
1240   void atomic_xchgal(Register prev, Register newv, Register addr);
1241   void atomic_xchgalw(Register prev, Register newv, Register addr);
1242   void atomic_xchgwu(Register prev, Register newv, Register addr);
1243   void atomic_xchgalwu(Register prev, Register newv, Register addr);
1244 
1245   void atomic_cas(Register prev, Register newv, Register addr, Assembler::operand_size size,
1246               Assembler::Aqrl acquire = Assembler::relaxed, Assembler::Aqrl release = Assembler::relaxed);
1247 
1248   // Emit a far call/jump. Only invalidates the tmp register which
1249   // is used to keep the entry address for jalr.
1250   // The address must be inside the code cache.
1251   // Supported entry.rspec():
1252   // - relocInfo::external_word_type
1253   // - relocInfo::runtime_call_type
1254   // - relocInfo::none
1255   // Clobbers t1 default.
1256   void far_call(const Address &entry, Register tmp = t1);
1257   void far_jump(const Address &entry, Register tmp = t1);
1258 
1259   static int far_branch_size() {
1260       return 2 * MacroAssembler::instruction_size;  // auipc + jalr, see far_call() & far_jump()
1261   }
1262 
1263   void load_byte_map_base(Register reg);
1264 
1265   void bang_stack_with_offset(int offset) {
1266     // stack grows down, caller passes positive offset
1267     assert(offset > 0, "must bang with negative offset");
1268     sub(t0, sp, offset);
1269     sd(zr, Address(t0));
1270   }
1271 
1272   virtual void _call_Unimplemented(address call_site) {
1273     mv(t1, call_site);
1274   }
1275 
1276   #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
1277 
1278   // Frame creation and destruction shared between JITs.
1279   void build_frame(int framesize);
1280   void remove_frame(int framesize);
1281 
1282   void reserved_stack_check();
1283 
1284   void get_polling_page(Register dest, relocInfo::relocType rtype);
1285   void read_polling_page(Register r, int32_t offset, relocInfo::relocType rtype);
1286 
1287   // RISCV64 OpenJDK uses three different types of calls:
1288   //
1289   //   - far call: auipc reg, pc_relative_offset; jalr ra, reg, offset
1290   //     The offset has the range [-(2G + 2K), 2G - 2K). Addresses out of the
1291   //     range in the code cache requires indirect call.
1292   //     If a jump is needed rather than a call, a far jump 'jalr x0, reg, offset'
1293   //     can be used instead.
1294   //     All instructions are embedded at a call site.
1295   //
1296   //   - indirect call: movptr + jalr
1297   //     This can reach anywhere in the address space, but it cannot be patched
1298   //     while code is running, so it must only be modified at a safepoint.
1299   //     This form of call is most suitable for targets at fixed addresses,
1300   //     which will never be patched.
1301   //
1302   //   - reloc call:
1303   //     This too can reach anywhere in the address space but is only available
1304   //     in C1/C2-generated code (nmethod).
1305   //
1306   //     [Main code section]
1307   //       auipc
1308   //       ld <address_from_stub_section>
1309   //       jalr
1310   //
1311   //     [Stub section]
1312   //     address stub:
1313   //       <64-bit destination address>
1314   //
1315   //    To change the destination we simply atomically store the new
1316   //    address in the stub section.
1317   //    There is a benign race in that the other thread might observe the old
1318   //    64-bit destination address before it observes the new address. That does
1319   //    not matter because the destination method has been invalidated, so there
1320   //    will be a trap at its start.
1321 
1322   // Emit a reloc call and create a stub to hold the entry point address.
1323   // Supported entry.rspec():
1324   // - relocInfo::runtime_call_type
1325   // - relocInfo::opt_virtual_call_type
1326   // - relocInfo::static_call_type
1327   // - relocInfo::virtual_call_type
1328   //
1329   // Return: the call PC or nullptr if CodeCache is full.
1330   address reloc_call(Address entry, Register tmp = t1);
1331 
1332   address ic_call(address entry, jint method_index = 0);
1333   static int ic_check_size();
1334   int ic_check(int end_alignment = MacroAssembler::instruction_size);
1335 
1336   // Support for memory inc/dec
1337   // n.b. increment/decrement calls with an Address destination will
1338   // need to use a scratch register to load the value to be
1339   // incremented. increment/decrement calls which add or subtract a
1340   // constant value other than sign-extended 12-bit immediate will need
1341   // to use a 2nd scratch register to hold the constant. so, an address
1342   // increment/decrement may trash both t0 and t1.
1343 
1344   void increment(const Address dst, int64_t value = 1, Register tmp1 = t0, Register tmp2 = t1);
1345   void incrementw(const Address dst, int32_t value = 1, Register tmp1 = t0, Register tmp2 = t1);
1346 
1347   void decrement(const Address dst, int64_t value = 1, Register tmp1 = t0, Register tmp2 = t1);
1348   void decrementw(const Address dst, int32_t value = 1, Register tmp1 = t0, Register tmp2 = t1);
1349 
1350   void clinit_barrier(Register klass, Register tmp, Label* L_fast_path = nullptr, Label* L_slow_path = nullptr);
1351 
1352   void load_method_holder_cld(Register result, Register method);
1353   void load_method_holder(Register holder, Register method);
1354 
1355   void compute_index(Register str1, Register trailing_zeros, Register match_mask,
1356                      Register result, Register char_tmp, Register tmp,
1357                      bool haystack_isL);
1358   void compute_match_mask(Register src, Register pattern, Register match_mask,
1359                           Register mask1, Register mask2);
1360 
1361   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1362   void kernel_crc32(Register crc, Register buf, Register len,
1363         Register table0, Register table1, Register table2, Register table3,
1364         Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register tmp6);
1365   void update_word_crc32(Register crc, Register v, Register tmp1, Register tmp2, Register tmp3,
1366         Register table0, Register table1, Register table2, Register table3,
1367         bool upper);
1368   void update_byte_crc32(Register crc, Register val, Register table);
1369 
1370 #ifdef COMPILER2
1371   void vector_update_crc32(Register crc, Register buf, Register len,
1372                            Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
1373                            Register table0, Register table3);
1374   void kernel_crc32_vclmul_fold(Register crc, Register buf, Register len,
1375               Register table0, Register table1, Register table2, Register table3,
1376               Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5);
1377   void crc32_vclmul_fold_to_16_bytes_vectorsize_32(VectorRegister vx, VectorRegister vy, VectorRegister vt,
1378                             VectorRegister vtmp1, VectorRegister vtmp2, VectorRegister vtmp3, VectorRegister vtmp4);
1379   void kernel_crc32_vclmul_fold_vectorsize_32(Register crc, Register buf, Register len,
1380                                               Register vclmul_table, Register tmp1, Register tmp2);
1381   void crc32_vclmul_fold_16_bytes_vectorsize_16(VectorRegister vx, VectorRegister vt,
1382                       VectorRegister vtmp1, VectorRegister vtmp2, VectorRegister vtmp3, VectorRegister vtmp4,
1383                       Register buf, Register tmp, const int STEP);
1384   void crc32_vclmul_fold_16_bytes_vectorsize_16_2(VectorRegister vx, VectorRegister vy, VectorRegister vt,
1385                       VectorRegister vtmp1, VectorRegister vtmp2, VectorRegister vtmp3, VectorRegister vtmp4,
1386                       Register tmp);
1387   void crc32_vclmul_fold_16_bytes_vectorsize_16_3(VectorRegister vx, VectorRegister vy, VectorRegister vt,
1388                       VectorRegister vtmp1, VectorRegister vtmp2, VectorRegister vtmp3, VectorRegister vtmp4,
1389                       Register tmp);
1390   void kernel_crc32_vclmul_fold_vectorsize_16(Register crc, Register buf, Register len,
1391                                               Register vclmul_table, Register tmp1, Register tmp2);
1392 
1393   void mul_add(Register out, Register in, Register offset,
1394                Register len, Register k, Register tmp);
1395   void wide_mul(Register prod_lo, Register prod_hi, Register n, Register m);
1396   void wide_madd(Register sum_lo, Register sum_hi, Register n,
1397                  Register m, Register tmp1, Register tmp2);
1398   void cad(Register dst, Register src1, Register src2, Register carry);
1399   void cadc(Register dst, Register src1, Register src2, Register carry);
1400   void adc(Register dst, Register src1, Register src2, Register carry);
1401   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1402                        Register src1, Register src2, Register carry);
1403   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1404                              Register y, Register y_idx, Register z,
1405                              Register carry, Register product,
1406                              Register idx, Register kdx);
1407   void multiply_128_x_128_loop(Register y, Register z,
1408                                Register carry, Register carry2,
1409                                Register idx, Register jdx,
1410                                Register yz_idx1, Register yz_idx2,
1411                                Register tmp, Register tmp3, Register tmp4,
1412                                Register tmp6, Register product_hi);
1413   void multiply_to_len(Register x, Register xlen, Register y, Register ylen,
1414                        Register z, Register tmp0,
1415                        Register tmp1, Register tmp2, Register tmp3, Register tmp4,
1416                        Register tmp5, Register tmp6, Register product_hi);
1417 
1418 #endif // COMPILER2
1419 
1420   void inflate_lo32(Register Rd, Register Rs, Register tmp1 = t0, Register tmp2 = t1);
1421   void inflate_hi32(Register Rd, Register Rs, Register tmp1 = t0, Register tmp2 = t1);
1422 
1423   void ctzc_bits(Register Rd, Register Rs, bool isLL = false,
1424                  Register tmp1 = t0, Register tmp2 = t1);
1425 
1426   void zero_words(Register base, uint64_t cnt);
1427   address zero_words(Register ptr, Register cnt);
1428   void fill_words(Register base, Register cnt, Register value);
1429   void zero_memory(Register addr, Register len, Register tmp);
1430   void zero_dcache_blocks(Register base, Register cnt, Register tmp1, Register tmp2);
1431 
1432   // shift left by shamt and add
1433   void shadd(Register Rd, Register Rs1, Register Rs2, Register tmp, int shamt);
1434 
1435   // test single bit in Rs, result is set to Rd
1436   void test_bit(Register Rd, Register Rs, uint32_t bit_pos);
1437 
1438   // Here the float instructions with safe deal with some exceptions.
1439   // e.g. convert from NaN, +Inf, -Inf to int, float, double
1440   // will trigger exception, we need to deal with these situations
1441   // to get correct results.
1442   void fcvt_w_s_safe(Register dst, FloatRegister src, Register tmp = t0);
1443   void fcvt_l_s_safe(Register dst, FloatRegister src, Register tmp = t0);
1444   void fcvt_w_d_safe(Register dst, FloatRegister src, Register tmp = t0);
1445   void fcvt_l_d_safe(Register dst, FloatRegister src, Register tmp = t0);
1446 
1447   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
1448   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
1449 
1450   // Helper routine processing the slow path of NaN when converting float to float16
1451   void float_to_float16_NaN(Register dst, FloatRegister src, Register tmp1, Register tmp2);
1452 
1453   // vector load/store unit-stride instructions
1454   void vlex_v(VectorRegister vd, Register base, Assembler::SEW sew, VectorMask vm = unmasked) {
1455     switch (sew) {
1456       case Assembler::e64:
1457         vle64_v(vd, base, vm);
1458         break;
1459       case Assembler::e32:
1460         vle32_v(vd, base, vm);
1461         break;
1462       case Assembler::e16:
1463         vle16_v(vd, base, vm);
1464         break;
1465       case Assembler::e8: // fall through
1466       default:
1467         vle8_v(vd, base, vm);
1468         break;
1469     }
1470   }
1471 
1472   void vsex_v(VectorRegister store_data, Register base, Assembler::SEW sew, VectorMask vm = unmasked) {
1473     switch (sew) {
1474       case Assembler::e64:
1475         vse64_v(store_data, base, vm);
1476         break;
1477       case Assembler::e32:
1478         vse32_v(store_data, base, vm);
1479         break;
1480       case Assembler::e16:
1481         vse16_v(store_data, base, vm);
1482         break;
1483       case Assembler::e8: // fall through
1484       default:
1485         vse8_v(store_data, base, vm);
1486         break;
1487     }
1488   }
1489 
1490   // vector pseudo instructions
1491   // rotate vector register left with shift bits, 32-bit version
1492   inline void vrole32_vi(VectorRegister vd, uint32_t shift, VectorRegister tmp_vr) {
1493     vsrl_vi(tmp_vr, vd, 32 - shift);
1494     vsll_vi(vd, vd, shift);
1495     vor_vv(vd, vd, tmp_vr);
1496   }
1497 
1498   inline void vl1r_v(VectorRegister vd, Register rs) {
1499     vl1re8_v(vd, rs);
1500   }
1501 
1502   inline void vmnot_m(VectorRegister vd, VectorRegister vs) {
1503     vmnand_mm(vd, vs, vs);
1504   }
1505 
1506   inline void vncvt_x_x_w(VectorRegister vd, VectorRegister vs, VectorMask vm = unmasked) {
1507     vnsrl_wx(vd, vs, x0, vm);
1508   }
1509 
1510   inline void vneg_v(VectorRegister vd, VectorRegister vs, VectorMask vm = unmasked) {
1511     vrsub_vx(vd, vs, x0, vm);
1512   }
1513 
1514   inline void vfneg_v(VectorRegister vd, VectorRegister vs, VectorMask vm = unmasked) {
1515     vfsgnjn_vv(vd, vs, vs, vm);
1516   }
1517 
1518   inline void vfabs_v(VectorRegister vd, VectorRegister vs, VectorMask vm = unmasked) {
1519     vfsgnjx_vv(vd, vs, vs, vm);
1520   }
1521 
1522   inline void vmsgt_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1523     vmslt_vv(vd, vs1, vs2, vm);
1524   }
1525 
1526   inline void vmsgtu_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1527     vmsltu_vv(vd, vs1, vs2, vm);
1528   }
1529 
1530   inline void vmsge_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1531     vmsle_vv(vd, vs1, vs2, vm);
1532   }
1533 
1534   inline void vmsgeu_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1535     vmsleu_vv(vd, vs1, vs2, vm);
1536   }
1537 
1538   inline void vmfgt_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1539     vmflt_vv(vd, vs1, vs2, vm);
1540   }
1541 
1542   inline void vmfge_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1543     vmfle_vv(vd, vs1, vs2, vm);
1544   }
1545 
1546   inline void vmsltu_vi(VectorRegister Vd, VectorRegister Vs2, uint32_t imm, VectorMask vm = unmasked) {
1547     guarantee(imm >= 1 && imm <= 16, "imm is invalid");
1548     vmsleu_vi(Vd, Vs2, imm-1, vm);
1549   }
1550 
1551   inline void vmsgeu_vi(VectorRegister Vd, VectorRegister Vs2, uint32_t imm, VectorMask vm = unmasked) {
1552     guarantee(imm >= 1 && imm <= 16, "imm is invalid");
1553     vmsgtu_vi(Vd, Vs2, imm-1, vm);
1554   }
1555 
1556   // Copy mask register
1557   inline void vmmv_m(VectorRegister vd, VectorRegister vs) {
1558     vmand_mm(vd, vs, vs);
1559   }
1560 
1561   // Clear mask register
1562   inline void vmclr_m(VectorRegister vd) {
1563     vmxor_mm(vd, vd, vd);
1564   }
1565 
1566   // Set mask register
1567   inline void vmset_m(VectorRegister vd) {
1568     vmxnor_mm(vd, vd, vd);
1569   }
1570 
1571   inline void vnot_v(VectorRegister Vd, VectorRegister Vs, VectorMask vm = unmasked) {
1572     vxor_vi(Vd, Vs, -1, vm);
1573   }
1574 
1575   static const int zero_words_block_size;
1576 
1577   void cast_primitive_type(BasicType type, Register Rt) {
1578     switch (type) {
1579       case T_BOOLEAN:
1580         sltu(Rt, zr, Rt);
1581         break;
1582       case T_CHAR   :
1583         zext(Rt, Rt, 16);
1584         break;
1585       case T_BYTE   :
1586         sext(Rt, Rt, 8);
1587         break;
1588       case T_SHORT  :
1589         sext(Rt, Rt, 16);
1590         break;
1591       case T_INT    :
1592         sext(Rt, Rt, 32);
1593         break;
1594       case T_LONG   : /* nothing to do */        break;
1595       case T_VOID   : /* nothing to do */        break;
1596       case T_FLOAT  : /* nothing to do */        break;
1597       case T_DOUBLE : /* nothing to do */        break;
1598       default: ShouldNotReachHere();
1599     }
1600   }
1601 
1602   // float cmp with unordered_result
1603   void float_compare(Register result, FloatRegister Rs1, FloatRegister Rs2, int unordered_result);
1604   void double_compare(Register result, FloatRegister Rs1, FloatRegister Rs2, int unordered_result);
1605 
1606   // Zero/Sign-extend
1607   void zext(Register dst, Register src, int bits);
1608   void sext(Register dst, Register src, int bits);
1609 
1610 private:
1611   void cmp_x2i(Register dst, Register src1, Register src2, Register tmp, bool is_signed = true);
1612 
1613 public:
1614   // compare src1 and src2 and get -1/0/1 in dst.
1615   // if [src1 > src2], dst = 1;
1616   // if [src1 == src2], dst = 0;
1617   // if [src1 < src2], dst = -1;
1618   void cmp_l2i(Register dst, Register src1, Register src2, Register tmp = t0);
1619   void cmp_ul2i(Register dst, Register src1, Register src2, Register tmp = t0);
1620   void cmp_uw2i(Register dst, Register src1, Register src2, Register tmp = t0);
1621 
1622   // support for argument shuffling
1623   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = t0);
1624   void float_move(VMRegPair src, VMRegPair dst, Register tmp = t0);
1625   void long_move(VMRegPair src, VMRegPair dst, Register tmp = t0);
1626   void double_move(VMRegPair src, VMRegPair dst, Register tmp = t0);
1627   void object_move(OopMap* map,
1628                    int oop_handle_offset,
1629                    int framesize_in_slots,
1630                    VMRegPair src,
1631                    VMRegPair dst,
1632                    bool is_receiver,
1633                    int* receiver_offset);
1634 
1635 #ifdef ASSERT
1636   // Template short-hand support to clean-up after a failed call to trampoline
1637   // call generation (see trampoline_call() below), when a set of Labels must
1638   // be reset (before returning).
1639   template<typename Label, typename... More>
1640   void reset_labels(Label& lbl, More&... more) {
1641     lbl.reset(); reset_labels(more...);
1642   }
1643   template<typename Label>
1644   void reset_labels(Label& lbl) {
1645     lbl.reset();
1646   }
1647 #endif
1648 
1649 private:
1650 
1651   void repne_scan(Register addr, Register value, Register count, Register tmp);
1652 
1653   int bitset_to_regs(unsigned int bitset, unsigned char* regs);
1654   Address add_memory_helper(const Address dst, Register tmp);
1655 
1656   void load_reserved(Register dst, Register addr, Assembler::operand_size size, Assembler::Aqrl acquire);
1657   void store_conditional(Register dst, Register new_val, Register addr, Assembler::operand_size size, Assembler::Aqrl release);
1658 
1659 public:
1660   void fast_lock(Register basic_lock, Register obj, Register tmp1, Register tmp2, Register tmp3, Label& slow);
1661   void fast_unlock(Register obj, Register tmp1, Register tmp2, Register tmp3, Label& slow);
1662 
1663 public:
1664   enum {
1665     // movptr
1666     movptr1_instruction_size = 6 * MacroAssembler::instruction_size, // lui, addi, slli, addi, slli, addi.  See movptr1().
1667     movptr2_instruction_size = 5 * MacroAssembler::instruction_size, // lui, lui, slli, add, addi.  See movptr2().
1668     load_pc_relative_instruction_size = 2 * MacroAssembler::instruction_size // auipc, ld
1669   };
1670 
1671   static bool is_load_pc_relative_at(address branch);
1672   static bool is_li16u_at(address instr);
1673 
1674   static bool is_jal_at(address instr)        { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b1101111; }
1675   static bool is_jalr_at(address instr)       { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b1100111 && extract_funct3(instr) == 0b000; }
1676   static bool is_branch_at(address instr)     { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b1100011; }
1677   static bool is_ld_at(address instr)         { assert_cond(instr != nullptr); return is_load_at(instr) && extract_funct3(instr) == 0b011; }
1678   static bool is_load_at(address instr)       { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0000011; }
1679   static bool is_float_load_at(address instr) { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0000111; }
1680   static bool is_auipc_at(address instr)      { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0010111; }
1681   static bool is_jump_at(address instr)       { assert_cond(instr != nullptr); return is_branch_at(instr) || is_jal_at(instr) || is_jalr_at(instr); }
1682   static bool is_add_at(address instr)        { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0110011 && extract_funct3(instr) == 0b000; }
1683   static bool is_addi_at(address instr)       { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0010011 && extract_funct3(instr) == 0b000; }
1684   static bool is_addiw_at(address instr)      { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0011011 && extract_funct3(instr) == 0b000; }
1685   static bool is_addiw_to_zr_at(address instr){ assert_cond(instr != nullptr); return is_addiw_at(instr) && extract_rd(instr) == zr; }
1686   static bool is_lui_at(address instr)        { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0110111; }
1687   static bool is_lui_to_zr_at(address instr)  { assert_cond(instr != nullptr); return is_lui_at(instr) && extract_rd(instr) == zr; }
1688 
1689   static bool is_srli_at(address instr) {
1690     assert_cond(instr != nullptr);
1691     return extract_opcode(instr) == 0b0010011 &&
1692            extract_funct3(instr) == 0b101 &&
1693            Assembler::extract(((unsigned*)instr)[0], 31, 26) == 0b000000;
1694   }
1695 
1696   static bool is_slli_shift_at(address instr, uint32_t shift) {
1697     assert_cond(instr != nullptr);
1698     return (extract_opcode(instr) == 0b0010011 && // opcode field
1699             extract_funct3(instr) == 0b001 &&     // funct3 field, select the type of operation
1700             Assembler::extract(Assembler::ld_instr(instr), 25, 20) == shift);    // shamt field
1701   }
1702 
1703   static bool is_movptr1_at(address instr);
1704   static bool is_movptr2_at(address instr);
1705 
1706   static bool is_lwu_to_zr(address instr);
1707 
1708   static Register extract_rs1(address instr);
1709   static Register extract_rs2(address instr);
1710   static Register extract_rd(address instr);
1711   static uint32_t extract_opcode(address instr);
1712   static uint32_t extract_funct3(address instr);
1713 
1714   // the instruction sequence of movptr is as below:
1715   //     lui
1716   //     addi
1717   //     slli
1718   //     addi
1719   //     slli
1720   //     addi/jalr/load
1721   static bool check_movptr1_data_dependency(address instr) {
1722     address lui = instr;
1723     address addi1 = lui + MacroAssembler::instruction_size;
1724     address slli1 = addi1 + MacroAssembler::instruction_size;
1725     address addi2 = slli1 + MacroAssembler::instruction_size;
1726     address slli2 = addi2 + MacroAssembler::instruction_size;
1727     address last_instr = slli2 + MacroAssembler::instruction_size;
1728     return extract_rs1(addi1) == extract_rd(lui) &&
1729            extract_rs1(addi1) == extract_rd(addi1) &&
1730            extract_rs1(slli1) == extract_rd(addi1) &&
1731            extract_rs1(slli1) == extract_rd(slli1) &&
1732            extract_rs1(addi2) == extract_rd(slli1) &&
1733            extract_rs1(addi2) == extract_rd(addi2) &&
1734            extract_rs1(slli2) == extract_rd(addi2) &&
1735            extract_rs1(slli2) == extract_rd(slli2) &&
1736            extract_rs1(last_instr) == extract_rd(slli2);
1737   }
1738 
1739   // the instruction sequence of movptr2 is as below:
1740   //     lui
1741   //     lui
1742   //     slli
1743   //     add
1744   //     addi/jalr/load
1745   static bool check_movptr2_data_dependency(address instr) {
1746     address lui1 = instr;
1747     address lui2 = lui1 + MacroAssembler::instruction_size;
1748     address slli = lui2 + MacroAssembler::instruction_size;
1749     address add  = slli + MacroAssembler::instruction_size;
1750     address last_instr = add + MacroAssembler::instruction_size;
1751     return extract_rd(add) == extract_rd(lui2) &&
1752            extract_rs1(add) == extract_rd(lui2) &&
1753            extract_rs2(add) == extract_rd(slli) &&
1754            extract_rs1(slli) == extract_rd(lui1) &&
1755            extract_rd(slli) == extract_rd(lui1) &&
1756            extract_rs1(last_instr) == extract_rd(add);
1757   }
1758 
1759   // the instruction sequence of li16u is as below:
1760   //     lui
1761   //     srli
1762   static bool check_li16u_data_dependency(address instr) {
1763     address lui = instr;
1764     address srli = lui + MacroAssembler::instruction_size;
1765 
1766     return extract_rs1(srli) == extract_rd(lui) &&
1767            extract_rs1(srli) == extract_rd(srli);
1768   }
1769 
1770   // the instruction sequence of li32 is as below:
1771   //     lui
1772   //     addiw
1773   static bool check_li32_data_dependency(address instr) {
1774     address lui = instr;
1775     address addiw = lui + MacroAssembler::instruction_size;
1776 
1777     return extract_rs1(addiw) == extract_rd(lui) &&
1778            extract_rs1(addiw) == extract_rd(addiw);
1779   }
1780 
1781   // the instruction sequence of pc-relative is as below:
1782   //     auipc
1783   //     jalr/addi/load/float_load
1784   static bool check_pc_relative_data_dependency(address instr) {
1785     address auipc = instr;
1786     address last_instr = auipc + MacroAssembler::instruction_size;
1787 
1788     return extract_rs1(last_instr) == extract_rd(auipc);
1789   }
1790 
1791   // the instruction sequence of load_label is as below:
1792   //     auipc
1793   //     load
1794   static bool check_load_pc_relative_data_dependency(address instr) {
1795     address auipc = instr;
1796     address load = auipc + MacroAssembler::instruction_size;
1797 
1798     return extract_rd(load) == extract_rd(auipc) &&
1799            extract_rs1(load) == extract_rd(load);
1800   }
1801 
1802   static bool is_li32_at(address instr);
1803   static bool is_pc_relative_at(address branch);
1804 
1805   static bool is_membar(address addr) {
1806     return (Bytes::get_native_u4(addr) & 0x7f) == 0b1111 && extract_funct3(addr) == 0;
1807   }
1808   static uint32_t get_membar_kind(address addr);
1809   static void set_membar_kind(address addr, uint32_t order_kind);
1810 };
1811 
1812 #ifdef ASSERT
1813 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1814 #endif
1815 
1816 #endif // CPU_RISCV_MACROASSEMBLER_RISCV_HPP