1 /*
   2  * Copyright (c) 2002, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2012, 2026 SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #ifndef CPU_PPC_MACROASSEMBLER_PPC_HPP
  27 #define CPU_PPC_MACROASSEMBLER_PPC_HPP
  28 
  29 #include "asm/assembler.hpp"
  30 #include "oops/accessDecorators.hpp"
  31 #include "utilities/macros.hpp"
  32 
  33 // MacroAssembler extends Assembler by a few frequently used macros.
  34 
  35 class ciTypeArray;
  36 class OopMap;
  37 class ciInlineKlass;
  38 class SigEntry;
  39 class VMRegPair;
  40 
  41 class MacroAssembler: public Assembler {
  42  public:
  43   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
  44 
  45   // Indicates whether and, if so, which registers must be preserved when calling runtime code.
  46   enum PreservationLevel {
  47     PRESERVATION_NONE,
  48     PRESERVATION_FRAME_LR,
  49     PRESERVATION_FRAME_LR_GP_REGS,
  50     PRESERVATION_FRAME_LR_GP_FP_REGS
  51   };
  52 
  53   //
  54   // Optimized instruction emitters
  55   //
  56 
  57   inline static int largeoffset_si16_si16_hi(int si31) { return (si31 + (1<<15)) >> 16; }
  58   inline static int largeoffset_si16_si16_lo(int si31) { return si31 - (((si31 + (1<<15)) >> 16) << 16); }
  59 
  60   // load d = *[a+si31]
  61   // Emits several instructions if the offset is not encodable in one instruction.
  62   void ld_largeoffset_unchecked(Register d, int si31, Register a, int emit_filler_nop);
  63   void ld_largeoffset          (Register d, int si31, Register a, int emit_filler_nop);
  64   inline static bool is_ld_largeoffset(address a);
  65   inline static int get_ld_largeoffset_offset(address a);
  66 
  67   inline void round_to(Register r, int modulus);
  68 
  69   // Load/store with type given by parameter.
  70   void load_sized_value( Register dst, RegisterOrConstant offs, Register base, size_t size_in_bytes, bool is_signed);
  71   void store_sized_value(Register dst, RegisterOrConstant offs, Register base, size_t size_in_bytes);
  72 
  73   // Move register if destination register and target register are different
  74   inline void mr_if_needed(Register rd, Register rs, bool allow_invalid = false);
  75   inline void fmr_if_needed(FloatRegister rd, FloatRegister rs);
  76 
  77   // Memory barriers.
  78   inline void membar(int bits);
  79   inline void release();
  80   inline void acquire();
  81   inline void fence();
  82 
  83   // nop padding
  84   void align(int modulus, int max = 252, int rem = 0);
  85 
  86   // Align prefix opcode to make sure it's not on the last word of a
  87   // 64-byte block.
  88   //
  89   // Note: do not call align_prefix() in a .ad file (e.g. ppc.ad).  Instead
  90   // add ins_alignment(2) to the instruct definition and implement the
  91   // compute_padding() method of the instruct node to use
  92   // compute_prefix_padding().  See loadConI32Node::compute_padding() in
  93   // ppc.ad for an example.
  94   void align_prefix();
  95 
  96   //
  97   // Constants, loading constants, TOC support
  98   //
  99 
 100   // Address of the global TOC.
 101   inline static address global_toc();
 102   // Offset of given address to the global TOC.
 103   inline static int offset_to_global_toc(const address addr);
 104 
 105   // Address of TOC of the current method.
 106   inline address method_toc();
 107   // Offset of given address to TOC of the current method.
 108   inline int offset_to_method_toc(const address addr);
 109 
 110   // Global TOC.
 111   void calculate_address_from_global_toc(Register dst, address addr,
 112                                          bool hi16 = true, bool lo16 = true,
 113                                          bool add_relocation = true, bool emit_dummy_addr = false,
 114                                          bool add_addr_to_reloc = true);
 115   void calculate_address_from_global_toc(Register dst, Label& addr,
 116                                          bool hi16 = true, bool lo16 = true,
 117                                          bool add_relocation = true, bool emit_dummy_addr = false) {
 118     calculate_address_from_global_toc(dst, target(addr), hi16, lo16, add_relocation, emit_dummy_addr, false);
 119   }
 120   inline void calculate_address_from_global_toc_hi16only(Register dst, address addr) {
 121     calculate_address_from_global_toc(dst, addr, true, false);
 122   };
 123   inline void calculate_address_from_global_toc_lo16only(Register dst, address addr) {
 124     calculate_address_from_global_toc(dst, addr, false, true);
 125   };
 126 
 127   inline static bool is_calculate_address_from_global_toc_at(address a, address bound);
 128   // Returns address of first instruction in sequence.
 129   static address patch_calculate_address_from_global_toc_at(address a, address bound, address addr);
 130   static address get_address_of_calculate_address_from_global_toc_at(address a, address addr);
 131 
 132 #ifdef _LP64
 133   // Patch narrow oop constant.
 134   inline static bool is_set_narrow_oop(address a, address bound);
 135   // Returns address of first instruction in sequence.
 136   static address patch_set_narrow_oop(address a, address bound, narrowOop data);
 137   static narrowOop get_narrow_oop(address a, address bound);
 138 #endif
 139 
 140   inline static bool is_load_const_at(address a);
 141 
 142   // Emits an oop const to the constant pool, loads the constant, and
 143   // sets a relocation info with address current_pc.
 144   // Returns true if successful.
 145   bool load_const_from_method_toc(Register dst, AddressLiteral& a, Register toc, bool fixed_size = false);
 146 
 147   static bool is_load_const_from_method_toc_at(address a);
 148   static int get_offset_of_load_const_from_method_toc_at(address a);
 149 
 150   // Get the 64 bit constant from a `load_const' sequence.
 151   static long get_const(address load_const);
 152 
 153   // Patch the 64 bit constant of a `load_const' sequence. This is a
 154   // low level procedure. It neither flushes the instruction cache nor
 155   // is it atomic.
 156   static void patch_const(address load_const, long x);
 157 
 158   // Metadata in code that we have to keep track of.
 159   AddressLiteral allocate_metadata_address(Metadata* obj); // allocate_index
 160   AddressLiteral constant_metadata_address(Metadata* obj); // find_index
 161   // Oops used directly in compiled code are stored in the constant pool,
 162   // and loaded from there.
 163   // Allocate new entry for oop in constant pool. Generate relocation.
 164   AddressLiteral allocate_oop_address(jobject obj);
 165   // Find oop obj in constant pool. Return relocation with it's index.
 166   AddressLiteral constant_oop_address(jobject obj);
 167 
 168   // Find oop in constant pool and emit instructions to load it.
 169   // Uses constant_oop_address.
 170   inline void set_oop_constant(jobject obj, Register d);
 171   // Same as load_address.
 172   inline void set_oop         (AddressLiteral obj_addr, Register d);
 173 
 174   //
 175   // branch, jump
 176   //
 177   // set dst to -1, 0, +1 as follows: if CR0bi is "greater than", dst is set to 1,
 178   // if CR0bi is "equal", dst is set to 0, otherwise it's set to -1.
 179   void inline set_cmp3(Register dst);
 180   // set dst to (treat_unordered_like_less ? -1 : +1)
 181   void inline set_cmpu3(Register dst, bool treat_unordered_like_less);
 182   // Branch-free implementation to convert !=0 to 1.
 183   void inline normalize_bool(Register dst, Register temp = R0, bool is_64bit = false);
 184   // Convert between half precision float encoded into a short and a float in a FloatRegister.
 185   void inline f2hf(Register dst, FloatRegister src, FloatRegister tmp);
 186   void inline hf2f(FloatRegister dst, Register src);
 187 
 188   inline void pd_patch_instruction(address branch, address target, const char* file, int line);
 189   NOT_PRODUCT(static void pd_print_patched_instruction(address branch);)
 190 
 191   // Conditional far branch for destinations encodable in 24+2 bits.
 192   // Same interface as bc, e.g. no inverse boint-field.
 193   enum {
 194     bc_far_optimize_not         = 0,
 195     bc_far_optimize_on_relocate = 1
 196   };
 197   // optimize: flag for telling the conditional far branch to optimize
 198   //           itself when relocated.
 199   void bc_far(int boint, int biint, Label& dest, int optimize);
 200   void bc_far_optimized(int boint, int biint, Label& dest); // 1 or 2 instructions
 201   // Relocation of conditional far branches.
 202   static bool    is_bc_far_at(address instruction_addr);
 203   static address get_dest_of_bc_far_at(address instruction_addr);
 204   static void    set_dest_of_bc_far_at(address instruction_addr, address dest);
 205  private:
 206   static bool inline is_bc_far_variant1_at(address instruction_addr);
 207   static bool inline is_bc_far_variant2_at(address instruction_addr);
 208   static bool inline is_bc_far_variant3_at(address instruction_addr);
 209  public:
 210 
 211   // Convenience bc_far versions.
 212   inline void blt_far(ConditionRegister crx, Label& L, int optimize);
 213   inline void bgt_far(ConditionRegister crx, Label& L, int optimize);
 214   inline void beq_far(ConditionRegister crx, Label& L, int optimize);
 215   inline void bso_far(ConditionRegister crx, Label& L, int optimize);
 216   inline void bge_far(ConditionRegister crx, Label& L, int optimize);
 217   inline void ble_far(ConditionRegister crx, Label& L, int optimize);
 218   inline void bne_far(ConditionRegister crx, Label& L, int optimize);
 219   inline void bns_far(ConditionRegister crx, Label& L, int optimize);
 220 
 221   // Emit, identify and patch a NOT mt-safe patchable 64 bit absolute call/jump.
 222  private:
 223   enum {
 224     bxx64_patchable_instruction_count = (2/*load_codecache_const*/ + 3/*5load_const*/ + 1/*mtctr*/ + 1/*bctrl*/),
 225     bxx64_patchable_size              = bxx64_patchable_instruction_count * BytesPerInstWord,
 226     bxx64_patchable_ret_addr_offset   = bxx64_patchable_size
 227   };
 228   void bxx64_patchable(address target, relocInfo::relocType rt, bool link);
 229   static bool is_bxx64_patchable_at(            address instruction_addr, bool link);
 230   // Does the instruction use a pc-relative encoding of the destination?
 231   static bool is_bxx64_patchable_pcrelative_at( address instruction_addr, bool link);
 232   static bool is_bxx64_patchable_variant1_at(   address instruction_addr, bool link);
 233   // Load destination relative to global toc.
 234   static bool is_bxx64_patchable_variant1b_at(  address instruction_addr, bool link);
 235   static bool is_bxx64_patchable_variant2_at(   address instruction_addr, bool link);
 236   static void set_dest_of_bxx64_patchable_at(   address instruction_addr, address target, bool link);
 237   static address get_dest_of_bxx64_patchable_at(address instruction_addr, bool link);
 238 
 239  public:
 240   // call
 241   enum {
 242     bl64_patchable_instruction_count = bxx64_patchable_instruction_count,
 243     bl64_patchable_size              = bxx64_patchable_size,
 244     bl64_patchable_ret_addr_offset   = bxx64_patchable_ret_addr_offset
 245   };
 246   inline void bl64_patchable(address target, relocInfo::relocType rt) {
 247     bxx64_patchable(target, rt, /*link=*/true);
 248   }
 249   inline static bool is_bl64_patchable_at(address instruction_addr) {
 250     return is_bxx64_patchable_at(instruction_addr, /*link=*/true);
 251   }
 252   inline static bool is_bl64_patchable_pcrelative_at(address instruction_addr) {
 253     return is_bxx64_patchable_pcrelative_at(instruction_addr, /*link=*/true);
 254   }
 255   inline static void set_dest_of_bl64_patchable_at(address instruction_addr, address target) {
 256     set_dest_of_bxx64_patchable_at(instruction_addr, target, /*link=*/true);
 257   }
 258   inline static address get_dest_of_bl64_patchable_at(address instruction_addr) {
 259     return get_dest_of_bxx64_patchable_at(instruction_addr, /*link=*/true);
 260   }
 261   // jump
 262   enum {
 263     b64_patchable_instruction_count = bxx64_patchable_instruction_count,
 264     b64_patchable_size              = bxx64_patchable_size,
 265   };
 266   inline void b64_patchable(address target, relocInfo::relocType rt) {
 267     bxx64_patchable(target, rt, /*link=*/false);
 268   }
 269   inline static bool is_b64_patchable_at(address instruction_addr) {
 270     return is_bxx64_patchable_at(instruction_addr, /*link=*/false);
 271   }
 272   inline static bool is_b64_patchable_pcrelative_at(address instruction_addr) {
 273     return is_bxx64_patchable_pcrelative_at(instruction_addr, /*link=*/false);
 274   }
 275   inline static void set_dest_of_b64_patchable_at(address instruction_addr, address target) {
 276     set_dest_of_bxx64_patchable_at(instruction_addr, target, /*link=*/false);
 277   }
 278   inline static address get_dest_of_b64_patchable_at(address instruction_addr) {
 279     return get_dest_of_bxx64_patchable_at(instruction_addr, /*link=*/false);
 280   }
 281 
 282   //
 283   // Support for frame handling
 284   //
 285 
 286   // some ABI-related functions
 287 
 288   // Clobbers all volatile, (non-floating-point) general-purpose registers for debugging purposes.
 289   // This is especially useful for making calls to the JRT in places in which this hasn't been done before;
 290   // e.g. with the introduction of LRBs (load reference barriers) for concurrent garbage collection.
 291   void clobber_volatile_gprs(Register excluded_register = noreg) NOT_DEBUG_RETURN;
 292   // Load bad values into registers that are nonvolatile according to the ABI except R16_thread and R29_TOC.
 293   // This is done after vthread preemption and before vthread resume.
 294   void clobber_nonvolatile_registers() NOT_DEBUG_RETURN;
 295   void clobber_carg_stack_slots(Register tmp);
 296 
 297   int  save_nonvolatile_registers_size(bool include_fp_regs, bool include_vector_regs) {
 298     int size = (32 - 14) * 8; // GP regs
 299     if (include_fp_regs) size += (32 - 14) * 8;
 300     if (include_vector_regs) size += (32 - 20) * 16;
 301     return size;
 302   }
 303   void save_nonvolatile_registers(   Register dst_base, int offset, bool include_fp_regs, bool include_vector_regs);
 304   void restore_nonvolatile_registers(Register src_base, int offset, bool include_fp_regs, bool include_vector_regs);
 305 
 306   enum {
 307     num_volatile_gp_regs = 11,
 308     num_volatile_fp_regs = 14,
 309     num_volatile_regs = num_volatile_gp_regs + num_volatile_fp_regs
 310   };
 311 
 312   void save_volatile_gprs(   Register dst_base, int offset,
 313                              bool include_fp_regs = true, bool include_R3_RET_reg = true);
 314   void restore_volatile_gprs(Register src_base, int offset,
 315                              bool include_fp_regs = true, bool include_R3_RET_reg = true);
 316   void save_LR(Register tmp);
 317   void restore_LR(Register tmp);
 318   void save_LR_CR(Register tmp);     // tmp contains LR on return.
 319   void restore_LR_CR(Register tmp);
 320 
 321   // Get current PC using bl-next-instruction trick.
 322   address get_PC_trash_LR(Register result);
 323 
 324   // Resize current frame either relatively wrt to current SP or absolute.
 325   void resize_frame(Register offset, Register tmp);
 326   void resize_frame(int      offset, Register tmp);
 327   void resize_frame_absolute(Register addr, Register tmp1, Register tmp2);
 328 
 329   // Push a frame of size bytes.
 330   void push_frame(Register bytes, Register tmp);
 331 
 332   // Push a frame of size `bytes'. No abi space provided.
 333   void push_frame(unsigned int bytes, Register tmp);
 334 
 335   // Push a frame of size `bytes' plus native_abi_reg_args on top.
 336   void push_frame_reg_args(unsigned int bytes, Register tmp);
 337 
 338   // pop current C frame
 339   void pop_frame();
 340 
 341   //
 342   // Calls
 343   //
 344 
 345  private:
 346   address _last_calls_return_pc;
 347 
 348 #if defined(ABI_ELFv2)
 349   // Generic version of a call to C function.
 350   // Updates and returns _last_calls_return_pc.
 351   address branch_to(Register function_entry, bool and_link);
 352 #else
 353   // Generic version of a call to C function via a function descriptor
 354   // with variable support for C calling conventions (TOC, ENV, etc.).
 355   // updates and returns _last_calls_return_pc.
 356   address branch_to(Register function_descriptor, bool and_link, bool save_toc_before_call,
 357                     bool restore_toc_after_call, bool load_toc_of_callee, bool load_env_of_callee);
 358 #endif
 359 
 360  public:
 361 
 362   // Get the pc where the last call will return to. returns _last_calls_return_pc.
 363   inline address last_calls_return_pc();
 364 
 365 #if defined(ABI_ELFv2)
 366   // Call a C function via a function descriptor and use full C
 367   // calling conventions. Updates and returns _last_calls_return_pc.
 368   address call_c(Register function_entry);
 369   // For tail calls: only branch, don't link, so callee returns to caller of this function.
 370   address call_c_and_return_to_caller(Register function_entry);
 371   address call_c(address function_entry, relocInfo::relocType rt = relocInfo::none);
 372 #else
 373   // Call a C function via a function descriptor and use full C
 374   // calling conventions. Updates and returns _last_calls_return_pc.
 375   address call_c(Register function_descriptor);
 376   // For tail calls: only branch, don't link, so callee returns to caller of this function.
 377   address call_c_and_return_to_caller(Register function_descriptor);
 378   address call_c(const FunctionDescriptor* function_descriptor, relocInfo::relocType rt);
 379   address call_c(address function_entry, relocInfo::relocType rt = relocInfo::none) {
 380     return call_c((const FunctionDescriptor*)function_entry, rt);
 381   }
 382   address call_c_using_toc(const FunctionDescriptor* function_descriptor, relocInfo::relocType rt,
 383                            Register toc);
 384 #endif
 385 
 386   // CompiledIC call
 387   bool ic_call(Register Rmethod_toc,
 388                address target,
 389                jint method_index = 0,
 390                bool scratch_emit = false,
 391                bool fixed_size = false);
 392   static int ic_check_size();
 393   int ic_check(int end_alignment);
 394 
 395   enum { trampoline_stub_size = 6 * 4 };
 396   address trampoline_call(AddressLiteral target,
 397                           Register Rmethod_toc = noreg,
 398                           bool scratch_emit = false);
 399 
 400  protected:
 401 
 402   // It is imperative that all calls into the VM are handled via the
 403   // call_VM macros. They make sure that the stack linkage is setup
 404   // correctly. call_VM's correspond to ENTRY/ENTRY_X entry points
 405   // while call_VM_leaf's correspond to LEAF entry points.
 406   //
 407   // This is the base routine called by the different versions of
 408   // call_VM. The interpreter may customize this version by overriding
 409   // it for its purposes (e.g., to save/restore additional registers
 410   // when doing a VM call).
 411   //
 412   // If no last_java_sp is specified (noreg) then SP will be used instead.
 413   virtual void call_VM_base(
 414      // where an oop-result ends up if any; use noreg otherwise
 415     Register        oop_result,
 416     // to set up last_Java_frame in stubs; use noreg otherwise
 417     Register        last_java_sp,
 418     // the entry point
 419     address         entry_point,
 420     // flag which indicates if exception should be checked
 421     bool            check_exception = true,
 422     Label* last_java_pc = nullptr
 423   );
 424 
 425   // Support for VM calls. This is the base routine called by the
 426   // different versions of call_VM_leaf. The interpreter may customize
 427   // this version by overriding it for its purposes (e.g., to
 428   // save/restore additional registers when doing a VM call).
 429   void call_VM_leaf_base(address entry_point);
 430 
 431  public:
 432   // Call into the VM.
 433   // Passes the thread pointer (in R3_ARG1) as a prepended argument.
 434   // Makes sure oop return values are visible to the GC.
 435   void call_VM(Register oop_result, address entry_point, bool check_exceptions = true, Label* last_java_pc = nullptr);
 436   void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
 437   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 438   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg3, bool check_exceptions = true);
 439   void call_VM_leaf(address entry_point);
 440   void call_VM_leaf(address entry_point, Register arg_1);
 441   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 442   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 443 
 444   // Call a stub function via a function descriptor, but don't save
 445   // TOC before call, don't setup TOC and ENV for call, and don't
 446   // restore TOC after call. Updates and returns _last_calls_return_pc.
 447   inline address call_stub(Register function_entry);
 448   inline void call_stub_and_return_to(Register function_entry, Register return_pc);
 449 
 450   void post_call_nop();
 451   static bool is_post_call_nop(int instr_bits) {
 452     const uint32_t nineth_bit = opp_u_field(1, 9, 9);
 453     const uint32_t opcode_mask = 0b111110 << OPCODE_SHIFT;
 454     const uint32_t pcn_mask = opcode_mask | nineth_bit;
 455     return (instr_bits & pcn_mask) == (Assembler::CMPLI_OPCODE | nineth_bit);
 456   }
 457 
 458   //
 459   // Java utilities
 460   //
 461 
 462   // Read from the polling page, its address is already in a register.
 463   inline void load_from_polling_page(Register polling_page_address, int offset = 0);
 464   // Check whether instruction is a read access to the polling page
 465   // which was emitted by load_from_polling_page(..).
 466   static bool is_load_from_polling_page(int instruction, void* ucontext/*may be nullptr*/,
 467                                         address* polling_address_ptr = nullptr);
 468 
 469   // Support for null-checks
 470   //
 471   // Generates code that causes a null OS exception if the content of reg is null.
 472   // If the accessed location is M[reg + offset] and the offset is known, provide the
 473   // offset. No explicit code generation is needed if the offset is within a certain
 474   // range (0 <= offset <= page_size).
 475 
 476   // Stack overflow checking
 477   void bang_stack_with_offset(int offset);
 478 
 479   // If instruction is a stack bang of the form ld, stdu, or
 480   // stdux, return the banged address. Otherwise, return 0.
 481   static address get_stack_bang_address(int instruction, void* ucontext);
 482 
 483   // Check for reserved stack access in method being exited. If the reserved
 484   // stack area was accessed, protect it again and throw StackOverflowError.
 485   void reserved_stack_check(Register return_pc);
 486 
 487   // Atomics
 488   // CmpxchgX sets condition register to cmpX(current, compare).
 489   // (flag == ne) => (dest_current_value != compare_value), (!swapped)
 490   // (flag == eq) => (dest_current_value == compare_value), ( swapped)
 491   static inline bool cmpxchgx_hint_acquire_lock()  { return true; }
 492   // The stxcx will probably not be succeeded by a releasing store.
 493   static inline bool cmpxchgx_hint_release_lock()  { return false; }
 494   static inline bool cmpxchgx_hint_atomic_update() { return false; }
 495 
 496   // Cmpxchg semantics
 497   enum {
 498     MemBarNone = 0,
 499     MemBarRel  = 1,
 500     MemBarAcq  = 2,
 501     MemBarFenceAfter = 4 // use powers of 2
 502   };
 503  private:
 504   // Helper functions for word/sub-word atomics.
 505   void atomic_get_and_modify_generic(Register dest_current_value, Register exchange_value,
 506                                      Register addr_base, Register tmp1, Register tmp2, Register tmp3,
 507                                      bool cmpxchgx_hint, bool is_add, int size);
 508   void cmpxchg_loop_body(ConditionRegister flag, Register dest_current_value,
 509                          RegisterOrConstant compare_value, Register exchange_value,
 510                          Register addr_base,Label &retry, Label &failed, bool cmpxchgx_hint, int size);
 511   void cmpxchg_generic(ConditionRegister flag, Register dest_current_value,
 512                        RegisterOrConstant compare_value, Register exchange_value,
 513                        Register addr_base, int semantics, bool cmpxchgx_hint, Register int_flag_success,
 514                        Label* failed_ext, bool contention_hint, bool weak, int size);
 515  public:
 516   // Temps and addr_base are killed if processor does not support Power 8 instructions.
 517   // Result will be sign extended.
 518   void getandsetb(Register dest_current_value, Register exchange_value, Register addr_base,
 519                   Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {
 520     atomic_get_and_modify_generic(dest_current_value, exchange_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, false, 1);
 521   }
 522   // Temps and addr_base are killed if processor does not support Power 8 instructions.
 523   // Result will be sign extended.
 524   void getandseth(Register dest_current_value, Register exchange_value, Register addr_base,
 525                   Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {
 526     atomic_get_and_modify_generic(dest_current_value, exchange_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, false, 2);
 527   }
 528   void getandsetw(Register dest_current_value, Register exchange_value, Register addr_base,
 529                   bool cmpxchgx_hint) {
 530     atomic_get_and_modify_generic(dest_current_value, exchange_value, addr_base, noreg, noreg, noreg, cmpxchgx_hint, false, 4);
 531   }
 532   void getandsetd(Register dest_current_value, Register exchange_value, Register addr_base,
 533                   bool cmpxchgx_hint);
 534   // tmp2/3 and addr_base are killed if processor does not support Power 8 instructions (tmp1 is always needed).
 535   // Result will be sign extended.
 536   void getandaddb(Register dest_current_value, Register inc_value, Register addr_base,
 537                   Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {
 538     atomic_get_and_modify_generic(dest_current_value, inc_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, true, 1);
 539   }
 540   // tmp2/3 and addr_base are killed if processor does not support Power 8 instructions (tmp1 is always needed).
 541   // Result will be sign extended.
 542   void getandaddh(Register dest_current_value, Register inc_value, Register addr_base,
 543                   Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {
 544     atomic_get_and_modify_generic(dest_current_value, inc_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, true, 2);
 545   }
 546   void getandaddw(Register dest_current_value, Register inc_value, Register addr_base,
 547                   Register tmp1, bool cmpxchgx_hint) {
 548     atomic_get_and_modify_generic(dest_current_value, inc_value, addr_base, tmp1, noreg, noreg, cmpxchgx_hint, true, 4);
 549   }
 550   void getandaddd(Register dest_current_value, Register exchange_value, Register addr_base,
 551                   Register tmp, bool cmpxchgx_hint);
 552   // Temps, addr_base and exchange_value are killed if processor does not support Power 8 instructions.
 553   // compare_value must be at least 32 bit sign extended. Result will be sign extended.
 554   void cmpxchgb(ConditionRegister flag, Register dest_current_value,
 555                 RegisterOrConstant compare_value, Register exchange_value,
 556                 Register addr_base, int semantics, bool cmpxchgx_hint = false,
 557                 Register int_flag_success = noreg, Label* failed = nullptr,
 558                 bool contention_hint = false, bool weak = false) {
 559     cmpxchg_generic(flag, dest_current_value, compare_value, exchange_value, addr_base, semantics,
 560                     cmpxchgx_hint, int_flag_success, failed, contention_hint, weak, 1);
 561   }
 562   // Temps, addr_base and exchange_value are killed if processor does not support Power 8 instructions.
 563   // compare_value must be at least 32 bit sign extended. Result will be sign extended.
 564   void cmpxchgh(ConditionRegister flag, Register dest_current_value,
 565                 RegisterOrConstant compare_value, Register exchange_value,
 566                 Register addr_base, int semantics, bool cmpxchgx_hint = false,
 567                 Register int_flag_success = noreg, Label* failed = nullptr,
 568                 bool contention_hint = false, bool weak = false) {
 569     cmpxchg_generic(flag, dest_current_value, compare_value, exchange_value, addr_base,
 570                     semantics, cmpxchgx_hint, int_flag_success, failed, contention_hint, weak, 2);
 571   }
 572   void cmpxchgw(ConditionRegister flag, Register dest_current_value,
 573                 RegisterOrConstant compare_value, Register exchange_value,
 574                 Register addr_base,
 575                 int semantics, bool cmpxchgx_hint = false, Register int_flag_success = noreg,
 576                 Label* failed = nullptr, bool contention_hint = false, bool weak = false) {
 577     cmpxchg_generic(flag, dest_current_value, compare_value, exchange_value, addr_base,
 578                     semantics, cmpxchgx_hint, int_flag_success, failed, contention_hint, weak, 4);
 579   }
 580   void cmpxchgd(ConditionRegister flag, Register dest_current_value,
 581                 RegisterOrConstant compare_value, Register exchange_value,
 582                 Register addr_base,
 583                 int semantics, bool cmpxchgx_hint = false, Register int_flag_success = noreg,
 584                 Label* failed = nullptr, bool contention_hint = false, bool weak = false);
 585 
 586   // interface method calling
 587   void lookup_interface_method(Register recv_klass,
 588                                Register intf_klass,
 589                                RegisterOrConstant itable_index,
 590                                Register method_result,
 591                                Register temp_reg, Register temp2_reg,
 592                                Label& no_such_interface,
 593                                bool return_method = true);
 594 
 595   // virtual method calling
 596   void lookup_virtual_method(Register recv_klass,
 597                              RegisterOrConstant vtable_index,
 598                              Register method_result);
 599 
 600   // Test sub_klass against super_klass, with fast and slow paths.
 601 
 602   // The fast path produces a tri-state answer: yes / no / maybe-slow.
 603   // One of the three labels can be null, meaning take the fall-through.
 604   // If super_check_offset is -1, the value is loaded up from super_klass.
 605   // No registers are killed, except temp_reg and temp2_reg.
 606   // If super_check_offset is not -1, temp2_reg is not used and can be noreg.
 607   void check_klass_subtype_fast_path(Register sub_klass,
 608                                      Register super_klass,
 609                                      Register temp1_reg,
 610                                      Register temp2_reg,
 611                                      Label* L_success,
 612                                      Label* L_failure,
 613                                      Label* L_slow_path = nullptr, // default fall through
 614                                      RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
 615 
 616   // The rest of the type check; must be wired to a corresponding fast path.
 617   // It does not repeat the fast path logic, so don't use it standalone.
 618   // The temp_reg can be noreg, if no temps are available.
 619   // It can also be sub_klass or super_klass, meaning it's OK to kill that one.
 620   // Updates the sub's secondary super cache as necessary.
 621   void check_klass_subtype_slow_path_linear(Register sub_klass,
 622                                             Register super_klass,
 623                                             Register temp1_reg,
 624                                             Register temp2_reg,
 625                                             Label* L_success = nullptr,
 626                                             Register result_reg = noreg);
 627 
 628   void check_klass_subtype_slow_path_table(Register sub_klass,
 629                                            Register super_klass,
 630                                            Register temp1_reg,
 631                                            Register temp2_reg,
 632                                            Label* L_success = nullptr,
 633                                            Register result_reg = noreg);
 634 
 635   void check_klass_subtype_slow_path(Register sub_klass,
 636                                      Register super_klass,
 637                                      Register temp1_reg,
 638                                      Register temp2_reg,
 639                                      Label* L_success = nullptr,
 640                                      Register result_reg = noreg);
 641 
 642   void lookup_secondary_supers_table_var(Register sub_klass,
 643                                          Register r_super_klass,
 644                                          Register temp1,
 645                                          Register temp2,
 646                                          Register temp3,
 647                                          Register temp4,
 648                                          Register result);
 649 
 650   // If r is valid, return r.
 651   // If r is invalid, remove a register r2 from available_regs, add r2
 652   // to regs_to_push, then return r2.
 653   Register allocate_if_noreg(const Register r,
 654                              RegSetIterator<Register> &available_regs,
 655                              RegSet &regs_to_push);
 656 
 657   // Frameless register spills (negative offset from SP)
 658   void push_set(RegSet set);
 659   void pop_set(RegSet set);
 660 
 661   // Simplified, combined version, good for typical uses.
 662   // Falls through on failure.
 663   void check_klass_subtype(Register sub_klass,
 664                            Register super_klass,
 665                            Register temp1_reg,
 666                            Register temp2_reg,
 667                            Label& L_success);
 668 
 669   void repne_scan(Register addr, Register value, Register count, Register scratch);
 670 
 671   // As above, but with a constant super_klass.
 672   // The result is in Register result, not the condition codes.
 673   void lookup_secondary_supers_table_const(Register r_sub_klass,
 674                                            Register r_super_klass,
 675                                            Register temp1,
 676                                            Register temp2,
 677                                            Register temp3,
 678                                            Register temp4,
 679                                            Register result,
 680                                            u1 super_klass_slot);
 681 
 682   void verify_secondary_supers_table(Register r_sub_klass,
 683                                      Register r_super_klass,
 684                                      Register result,
 685                                      Register temp1,
 686                                      Register temp2,
 687                                      Register temp3);
 688 
 689   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
 690                                                Register r_array_base,
 691                                                Register r_array_index,
 692                                                Register r_bitmap,
 693                                                Register result,
 694                                                Register temp1);
 695 
 696   void clinit_barrier(Register klass,
 697                       Register thread,
 698                       Label* L_fast_path = nullptr,
 699                       Label* L_slow_path = nullptr);
 700 
 701   // Method handle support (JSR 292).
 702   RegisterOrConstant argument_offset(RegisterOrConstant arg_slot, Register temp_reg, int extra_slot_offset = 0);
 703 
 704   void push_cont_fastpath();
 705   void pop_cont_fastpath();
 706   void atomically_flip_locked_state(bool is_unlock, Register obj, Register tmp, Label& failed, int semantics);
 707   void fast_lock(Register box, Register obj, Register t1, Register t2, Label& slow);
 708   void fast_unlock(Register obj, Register t1, Label& slow);
 709 
 710   // allocation (for C1)
 711   void tlab_allocate(
 712     Register obj,                      // result: pointer to object after successful allocation
 713     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
 714     int      con_size_in_bytes,        // object size in bytes if   known at compile time
 715     Register t1,                       // temp register
 716     Label&   slow_case                 // continuation point if fast allocation fails
 717   );
 718 
 719   void compiler_fast_lock_object(ConditionRegister flag, Register oop, Register box,
 720                                  Register tmp1, Register tmp2, Register tmp3);
 721 
 722   void compiler_fast_unlock_object(ConditionRegister flag, Register oop, Register box,
 723                                    Register tmp1, Register tmp2, Register tmp3);
 724 
 725   // Check if safepoint requested and if so branch
 726   void safepoint_poll(Label& slow_path, Register temp, bool at_return, bool in_nmethod);
 727   void jump_to_polling_page_return_handler_blob(int safepoint_offset, bool fixed_size = false);
 728 
 729   void resolve_jobject(Register value, Register tmp1, Register tmp2,
 730                        MacroAssembler::PreservationLevel preservation_level);
 731   void resolve_global_jobject(Register value, Register tmp1, Register tmp2,
 732                               MacroAssembler::PreservationLevel preservation_level);
 733 
 734   // Support for managing the JavaThread pointer (i.e.; the reference to
 735   // thread-local information).
 736 
 737   // Support for last Java frame (but use call_VM instead where possible):
 738   // access R16_thread->last_Java_sp.
 739   void set_last_Java_frame(Register last_java_sp, Register last_Java_pc);
 740   void reset_last_Java_frame(bool check_last_java_sp = true);
 741   void set_top_ijava_frame_at_SP_as_last_Java_frame(Register sp, Register tmp1, Label* jpc = nullptr);
 742 
 743   // Read vm result from thread: oop_result = R16_thread->result;
 744   void get_vm_result_oop(Register oop_result);
 745   void get_vm_result_metadata(Register metadata_result);
 746 
 747   static bool needs_explicit_null_check(intptr_t offset);
 748   static bool uses_implicit_null_check(void* address);
 749 
 750   // Trap-instruction-based checks.
 751   // Range checks can be distinguished from zero checks as they check 32 bit,
 752   // zero checks all 64 bits (tw, td).
 753   inline void trap_null_check(Register a, trap_to_bits cmp = traptoEqual);
 754   static bool is_trap_null_check(int x) {
 755     return is_tdi(x, traptoEqual,               -1/*any reg*/, 0) ||
 756            is_tdi(x, traptoGreaterThanUnsigned, -1/*any reg*/, 0);
 757   }
 758 
 759   inline void trap_ic_miss_check(Register a, Register b);
 760   static bool is_trap_ic_miss_check(int x) {
 761     return is_td(x, traptoGreaterThanUnsigned | traptoLessThanUnsigned, -1/*any reg*/, -1/*any reg*/);
 762   }
 763 
 764   // Implicit or explicit null check, jumps to static address exception_entry.
 765   inline void null_check_throw(Register a, int offset, Register temp_reg, address exception_entry);
 766   inline void null_check(Register a, int offset, Label *Lis_null); // implicit only if Lis_null not provided
 767 
 768   // Access heap oop, handle encoding and GC barriers.
 769   // Some GC barriers call C so use needs_frame = true if an extra frame is needed at the current call site.
 770   inline void access_store_at(BasicType type, DecoratorSet decorators,
 771                               Register base, RegisterOrConstant ind_or_offs, Register val,
 772                               Register tmp1, Register tmp2, Register tmp3,
 773                               MacroAssembler::PreservationLevel preservation_level);
 774   inline void access_load_at(BasicType type, DecoratorSet decorators,
 775                              Register base, RegisterOrConstant ind_or_offs, Register dst,
 776                              Register tmp1, Register tmp2,
 777                              MacroAssembler::PreservationLevel preservation_level, Label *L_handle_null = nullptr);
 778 
 779  public:
 780   // Specify tmp1 for better code in certain compressed oops cases. Specify Label to bail out on null oop.
 781   // tmp1, tmp2 and needs_frame are used with decorators ON_PHANTOM_OOP_REF or ON_WEAK_OOP_REF.
 782   inline void load_heap_oop(Register d, RegisterOrConstant offs, Register s1,
 783                             Register tmp1, Register tmp2,
 784                             MacroAssembler::PreservationLevel preservation_level,
 785                             DecoratorSet decorators = 0, Label *L_handle_null = nullptr);
 786 
 787   inline void store_heap_oop(Register d, RegisterOrConstant offs, Register s1,
 788                              Register tmp1, Register tmp2, Register tmp3,
 789                              MacroAssembler::PreservationLevel preservation_level, DecoratorSet decorators = 0);
 790 
 791   // Encode/decode heap oop. Oop may not be null, else en/decoding goes wrong.
 792   // src == d allowed.
 793   inline Register encode_heap_oop_not_null(Register d, Register src = noreg);
 794   inline Register decode_heap_oop_not_null(Register d, Register src = noreg);
 795 
 796   // Null allowed.
 797   inline Register encode_heap_oop(Register d, Register src); // Prefer null check in GC barrier!
 798   inline void decode_heap_oop(Register d);
 799 
 800   // Load/Store klass oop from klass field. Compress.
 801   void load_klass_no_decode(Register dst, Register src);
 802   void load_klass(Register dst, Register src);
 803   void load_narrow_klass_compact(Register dst, Register src);
 804   void cmp_klass(ConditionRegister dst, Register obj, Register klass, Register tmp, Register tmp2);
 805   void cmp_klasses_from_objects(ConditionRegister dst, Register obj1, Register obj2, Register tmp1, Register tmp2);
 806   void load_klass_check_null(Register dst, Register src, Label* is_null = nullptr);
 807   void store_klass(Register dst_oop, Register klass, Register tmp = R0);
 808   void store_klass_gap(Register dst_oop, Register val = noreg); // Will store 0 if val not specified.
 809 
 810   void resolve_oop_handle(Register result, Register tmp1, Register tmp2,
 811                           MacroAssembler::PreservationLevel preservation_level);
 812   void resolve_weak_handle(Register result, Register tmp1, Register tmp2,
 813                            MacroAssembler::PreservationLevel preservation_level);
 814   void load_method_holder(Register holder, Register method);
 815 
 816   void decode_klass_not_null(Register dst, Register src = noreg);
 817   Register encode_klass_not_null(Register dst, Register src = noreg);
 818 
 819   // SIGTRAP-based range checks for arrays.
 820   inline void trap_range_check_l(Register a, Register b);
 821   inline void trap_range_check_l(Register a, int si16);
 822   static bool is_trap_range_check_l(int x) {
 823     return (is_tw (x, traptoLessThanUnsigned, -1/*any reg*/, -1/*any reg*/) ||
 824             is_twi(x, traptoLessThanUnsigned, -1/*any reg*/)                  );
 825   }
 826   inline void trap_range_check_le(Register a, int si16);
 827   static bool is_trap_range_check_le(int x) {
 828     return is_twi(x, traptoEqual | traptoLessThanUnsigned, -1/*any reg*/);
 829   }
 830   inline void trap_range_check_g(Register a, int si16);
 831   static bool is_trap_range_check_g(int x) {
 832     return is_twi(x, traptoGreaterThanUnsigned, -1/*any reg*/);
 833   }
 834   inline void trap_range_check_ge(Register a, Register b);
 835   inline void trap_range_check_ge(Register a, int si16);
 836   static bool is_trap_range_check_ge(int x) {
 837     return (is_tw (x, traptoEqual | traptoGreaterThanUnsigned, -1/*any reg*/, -1/*any reg*/) ||
 838             is_twi(x, traptoEqual | traptoGreaterThanUnsigned, -1/*any reg*/)                  );
 839   }
 840   static bool is_trap_range_check(int x) {
 841     return is_trap_range_check_l(x) || is_trap_range_check_le(x) ||
 842            is_trap_range_check_g(x) || is_trap_range_check_ge(x);
 843   }
 844 
 845   void clear_memory_unrolled(Register base_ptr, int cnt_dwords, Register tmp = R0, int offset = 0);
 846   void clear_memory_constlen(Register base_ptr, int cnt_dwords, Register tmp = R0);
 847   void clear_memory_doubleword(Register base_ptr, Register cnt_dwords, Register tmp = R0, long const_cnt = -1);
 848 
 849   // Emitters for BigInteger.multiplyToLen intrinsic.
 850   inline void multiply64(Register dest_hi, Register dest_lo,
 851                          Register x, Register y);
 852   void add2_with_carry(Register dest_hi, Register dest_lo,
 853                        Register src1, Register src2);
 854   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 855                              Register y, Register y_idx, Register z,
 856                              Register carry, Register product_high, Register product,
 857                              Register idx, Register kdx, Register tmp);
 858   void multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 859                               Register yz_idx, Register idx, Register carry,
 860                               Register product_high, Register product, Register tmp,
 861                               int offset);
 862   void multiply_128_x_128_loop(Register x_xstart,
 863                                Register y, Register z,
 864                                Register yz_idx, Register idx, Register carry,
 865                                Register product_high, Register product,
 866                                Register carry2, Register tmp);
 867   void muladd(Register out, Register in, Register offset, Register len, Register k,
 868               Register tmp1, Register tmp2, Register carry);
 869   void multiply_to_len(Register x, Register xlen,
 870                        Register y, Register ylen,
 871                        Register z,
 872                        Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
 873                        Register tmp6, Register tmp7, Register tmp8, Register tmp9, Register tmp10,
 874                        Register tmp11, Register tmp12, Register tmp13);
 875 
 876   // non-atomic 64-bit memory increment by simm16
 877   void increment_mem64(Register base, RegisterOrConstant ind_or_offs, int val, Register tmp);
 878 
 879   // Bytecode profiling (tmp2 = noreg is allowed, but then recv is killed)
 880   void profile_receiver_type(Register recv, Register mdp, int mdp_offset, Register tmp1, Register tmp2);
 881 
 882   // Emitters for CRC32 calculation.
 883   // A note on invertCRC:
 884   //   Unfortunately, internal representation of crc differs between CRC32 and CRC32C.
 885   //   CRC32 holds it's current crc value in the externally visible representation.
 886   //   CRC32C holds it's current crc value in internal format, ready for updating.
 887   //   Thus, the crc value must be bit-flipped before updating it in the CRC32 case.
 888   //   In the CRC32C case, it must be bit-flipped when it is given to the outside world (getValue()).
 889   //   The bool invertCRC parameter indicates whether bit-flipping is required before updates.
 890   void load_reverse_32(Register dst, Register src);
 891   int  crc32_table_columns(Register table, Register tc0, Register tc1, Register tc2, Register tc3);
 892   void fold_byte_crc32(Register crc, Register val, Register table, Register tmp);
 893   void update_byte_crc32(Register crc, Register val, Register table);
 894   void update_byteLoop_crc32(Register crc, Register buf, Register len, Register table,
 895                              Register data, bool loopAlignment);
 896   void update_1word_crc32(Register crc, Register buf, Register table, int bufDisp, int bufInc,
 897                           Register t0,  Register t1,  Register t2,  Register t3,
 898                           Register tc0, Register tc1, Register tc2, Register tc3);
 899   void kernel_crc32_vpmsum(Register crc, Register buf, Register len, Register constants,
 900                            Register t0, Register t1, Register t2, Register t3, Register t4,
 901                            Register t5, Register t6, bool invertCRC);
 902   void kernel_crc32_vpmsum_aligned(Register crc, Register buf, Register len, Register constants,
 903                                    Register t0, Register t1, Register t2, Register t3, Register t4,
 904                                    Register t5, Register t6);
 905   // Version which internally decides what to use.
 906   void crc32(Register crc, Register buf, Register len, Register t0, Register t1, Register t2,
 907              Register t3, Register t4, Register t5, Register t6, Register t7, bool is_crc32c);
 908 
 909   void kernel_crc32_singleByteReg(Register crc, Register val, Register table,
 910                                   bool invertCRC);
 911 
 912   // SHA-2 auxiliary functions and public interfaces
 913  private:
 914   void sha256_deque(const VectorRegister src,
 915       const VectorRegister dst1, const VectorRegister dst2, const VectorRegister dst3);
 916   void sha256_load_h_vec(const VectorRegister a, const VectorRegister e, const Register hptr);
 917   void sha256_round(const VectorRegister* hs, const int total_hs, int& h_cnt, const VectorRegister kpw);
 918   void sha256_load_w_plus_k_vec(const Register buf_in, const VectorRegister* ws,
 919       const int total_ws, const Register k, const VectorRegister* kpws,
 920       const int total_kpws);
 921   void sha256_calc_4w(const VectorRegister w0, const VectorRegister w1,
 922       const VectorRegister w2, const VectorRegister w3, const VectorRegister kpw0,
 923       const VectorRegister kpw1, const VectorRegister kpw2, const VectorRegister kpw3,
 924       const Register j, const Register k);
 925   void sha256_update_sha_state(const VectorRegister a, const VectorRegister b,
 926       const VectorRegister c, const VectorRegister d, const VectorRegister e,
 927       const VectorRegister f, const VectorRegister g, const VectorRegister h,
 928       const Register hptr);
 929 
 930   void sha512_load_w_vec(const Register buf_in, const VectorRegister* ws, const int total_ws);
 931   void sha512_update_sha_state(const Register state, const VectorRegister* hs, const int total_hs);
 932   void sha512_round(const VectorRegister* hs, const int total_hs, int& h_cnt, const VectorRegister kpw);
 933   void sha512_load_h_vec(const Register state, const VectorRegister* hs, const int total_hs);
 934   void sha512_calc_2w(const VectorRegister w0, const VectorRegister w1,
 935       const VectorRegister w2, const VectorRegister w3,
 936       const VectorRegister w4, const VectorRegister w5,
 937       const VectorRegister w6, const VectorRegister w7,
 938       const VectorRegister kpw0, const VectorRegister kpw1, const Register j,
 939       const VectorRegister vRb, const Register k);
 940 
 941  public:
 942   void sha256(bool multi_block);
 943   void sha512(bool multi_block);
 944 
 945   void cache_wb(Address line);
 946   void cache_wbsync(bool is_presync);
 947 
 948   //
 949   // Debugging
 950   //
 951 
 952   // assert on cr0
 953   enum AsmAssertCond {
 954     eq,
 955     ne,
 956     ge,
 957     gt,
 958     lt,
 959     le
 960   };
 961   void asm_assert(AsmAssertCond cond, const char* msg) PRODUCT_RETURN;
 962   void asm_assert_eq(const char* msg) { asm_assert(eq, msg); }
 963   void asm_assert_ne(const char* msg) { asm_assert(ne, msg); }
 964 
 965  private:
 966   void asm_assert_mems_zero(AsmAssertCond cond, int size, int mem_offset, Register mem_base,
 967                             const char* msg) NOT_DEBUG_RETURN;
 968 
 969  public:
 970 
 971   void asm_assert_mem8_is_zero(int mem_offset, Register mem_base, const char* msg) {
 972     asm_assert_mems_zero(eq,  8, mem_offset, mem_base, msg);
 973   }
 974   void asm_assert_mem8_isnot_zero(int mem_offset, Register mem_base, const char* msg) {
 975     asm_assert_mems_zero(ne, 8, mem_offset, mem_base, msg);
 976   }
 977 
 978   // Calls verify_oop. If UseCompressedOops is on, decodes the oop.
 979   // Preserves reg.
 980   void verify_coop(Register reg, const char*);
 981   // Emit code to verify that reg contains a valid oop if +VerifyOops is set.
 982   void verify_oop(Register reg, const char* s = "broken oop");
 983   void verify_oop_addr(RegisterOrConstant offs, Register base, const char* s = "contains broken oop");
 984 
 985   // TODO: verify method and klass metadata (compare against vptr?)
 986   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
 987   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line) {}
 988 
 989   // Convenience method returning function entry. For the ELFv1 case
 990   // creates function descriptor at the current address and returns
 991   // the pointer to it. For the ELFv2 case returns the current address.
 992   inline address function_entry();
 993 
 994 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
 995 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
 996 
 997  private:
 998   void stop(int type, const char* msg);
 999 
1000  public:
1001   enum {
1002     stop_stop               = 0,
1003     stop_untested           = 1,
1004     stop_unimplemented      = 2,
1005     stop_shouldnotreachhere = 3,
1006     stop_msg_present        = -0x8000
1007   };
1008 
1009   // Prints msg, dumps registers and stops execution.
1010   void stop                 (const char* msg = nullptr) { stop(stop_stop,               msg); }
1011   void untested             (const char* msg = nullptr) { stop(stop_untested,           msg); }
1012   void unimplemented        (const char* msg = nullptr) { stop(stop_unimplemented,      msg); }
1013   void should_not_reach_here(const char* msg = nullptr) { stop(stop_shouldnotreachhere, msg); }
1014 
1015   void zap_from_to(Register low, int before, Register high, int after, Register val, Register addr) PRODUCT_RETURN;
1016 
1017   // Inline type specific methods
1018   #include "asm/macroAssembler_common.hpp"
1019 };
1020 
1021 #endif // CPU_PPC_MACROASSEMBLER_PPC_HPP