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 "runtime/signature.hpp"
  32 #include "utilities/macros.hpp"
  33 
  34 // MacroAssembler extends Assembler by a few frequently used macros.
  35 
  36 class ciTypeArray;
  37 class OopMap;
  38 class ciInlineKlass;
  39 class SigEntry;
  40 class VMRegPair;
  41 
  42 class MacroAssembler: public Assembler {
  43  public:
  44   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
  45 
  46   // Indicates whether and, if so, which registers must be preserved when calling runtime code.
  47   enum PreservationLevel {
  48     PRESERVATION_NONE,
  49     PRESERVATION_FRAME_LR,
  50     PRESERVATION_FRAME_LR_GP_REGS,
  51     PRESERVATION_FRAME_LR_GP_FP_REGS
  52   };
  53 
  54   //
  55   // Optimized instruction emitters
  56   //
  57 
  58   inline static int largeoffset_si16_si16_hi(int si31) { return (si31 + (1<<15)) >> 16; }
  59   inline static int largeoffset_si16_si16_lo(int si31) { return si31 - (((si31 + (1<<15)) >> 16) << 16); }
  60 
  61   // load d = *[a+si31]
  62   // Emits several instructions if the offset is not encodable in one instruction.
  63   void ld_largeoffset_unchecked(Register d, int si31, Register a, int emit_filler_nop);
  64   void ld_largeoffset          (Register d, int si31, Register a, int emit_filler_nop);
  65   inline static bool is_ld_largeoffset(address a);
  66   inline static int get_ld_largeoffset_offset(address a);
  67 
  68   inline void round_to(Register r, int modulus);
  69 
  70   // Load/store with type given by parameter.
  71   void load_sized_value( Register dst, RegisterOrConstant offs, Register base, size_t size_in_bytes, bool is_signed);
  72   void store_sized_value(Register dst, RegisterOrConstant offs, Register base, size_t size_in_bytes);
  73 
  74   // Move register if destination register and target register are different
  75   inline void mr_if_needed(Register rd, Register rs, bool allow_invalid = false);
  76   inline void fmr_if_needed(FloatRegister rd, FloatRegister rs);
  77 
  78   // Memory barriers.
  79   inline void membar(int bits);
  80   inline void release();
  81   inline void acquire();
  82   inline void fence();
  83 
  84   // nop padding
  85   void align(int modulus, int max = 252, int rem = 0);
  86 
  87   // Align prefix opcode to make sure it's not on the last word of a
  88   // 64-byte block.
  89   //
  90   // Note: do not call align_prefix() in a .ad file (e.g. ppc.ad).  Instead
  91   // add ins_alignment(2) to the instruct definition and implement the
  92   // compute_padding() method of the instruct node to use
  93   // compute_prefix_padding().  See loadConI32Node::compute_padding() in
  94   // ppc.ad for an example.
  95   void align_prefix();
  96 
  97   //
  98   // Constants, loading constants, TOC support
  99   //
 100 
 101   // Address of the global TOC.
 102   inline static address global_toc();
 103   // Offset of given address to the global TOC.
 104   inline static int offset_to_global_toc(const address addr);
 105 
 106   // Address of TOC of the current method.
 107   inline address method_toc();
 108   // Offset of given address to TOC of the current method.
 109   inline int offset_to_method_toc(const address addr);
 110 
 111   // Global TOC.
 112   void calculate_address_from_global_toc(Register dst, address addr,
 113                                          bool hi16 = true, bool lo16 = true,
 114                                          bool add_relocation = true, bool emit_dummy_addr = false,
 115                                          bool add_addr_to_reloc = true);
 116   void calculate_address_from_global_toc(Register dst, Label& addr,
 117                                          bool hi16 = true, bool lo16 = true,
 118                                          bool add_relocation = true, bool emit_dummy_addr = false) {
 119     calculate_address_from_global_toc(dst, target(addr), hi16, lo16, add_relocation, emit_dummy_addr, false);
 120   }
 121   inline void calculate_address_from_global_toc_hi16only(Register dst, address addr) {
 122     calculate_address_from_global_toc(dst, addr, true, false);
 123   };
 124   inline void calculate_address_from_global_toc_lo16only(Register dst, address addr) {
 125     calculate_address_from_global_toc(dst, addr, false, true);
 126   };
 127 
 128   inline static bool is_calculate_address_from_global_toc_at(address a, address bound);
 129   // Returns address of first instruction in sequence.
 130   static address patch_calculate_address_from_global_toc_at(address a, address bound, address addr);
 131   static address get_address_of_calculate_address_from_global_toc_at(address a, address addr);
 132 
 133 #ifdef _LP64
 134   // Patch narrow oop constant.
 135   inline static bool is_set_narrow_oop(address a, address bound);
 136   // Returns address of first instruction in sequence.
 137   static address patch_set_narrow_oop(address a, address bound, narrowOop data);
 138   static narrowOop get_narrow_oop(address a, address bound);
 139 #endif
 140 
 141   inline static bool is_load_const_at(address a);
 142 
 143   // Emits an oop const to the constant pool, loads the constant, and
 144   // sets a relocation info with address current_pc.
 145   // Returns true if successful.
 146   bool load_const_from_method_toc(Register dst, AddressLiteral& a, Register toc, bool fixed_size = false);
 147 
 148   static bool is_load_const_from_method_toc_at(address a);
 149   static int get_offset_of_load_const_from_method_toc_at(address a);
 150 
 151   // Get the 64 bit constant from a `load_const' sequence.
 152   static long get_const(address load_const);
 153 
 154   // Patch the 64 bit constant of a `load_const' sequence. This is a
 155   // low level procedure. It neither flushes the instruction cache nor
 156   // is it atomic.
 157   static void patch_const(address load_const, long x);
 158 
 159   // Metadata in code that we have to keep track of.
 160   AddressLiteral allocate_metadata_address(Metadata* obj); // allocate_index
 161   AddressLiteral constant_metadata_address(Metadata* obj); // find_index
 162   // Oops used directly in compiled code are stored in the constant pool,
 163   // and loaded from there.
 164   // Allocate new entry for oop in constant pool. Generate relocation.
 165   AddressLiteral allocate_oop_address(jobject obj);
 166   // Find oop obj in constant pool. Return relocation with it's index.
 167   AddressLiteral constant_oop_address(jobject obj);
 168 
 169   // Find oop in constant pool and emit instructions to load it.
 170   // Uses constant_oop_address.
 171   inline void set_oop_constant(jobject obj, Register d);
 172   // Same as load_address.
 173   inline void set_oop         (AddressLiteral obj_addr, Register d);
 174 
 175   //
 176   // branch, jump
 177   //
 178   // set dst to -1, 0, +1 as follows: if CR0bi is "greater than", dst is set to 1,
 179   // if CR0bi is "equal", dst is set to 0, otherwise it's set to -1.
 180   void inline set_cmp3(Register dst);
 181   // set dst to (treat_unordered_like_less ? -1 : +1)
 182   void inline set_cmpu3(Register dst, bool treat_unordered_like_less);
 183   // Branch-free implementation to convert !=0 to 1.
 184   void inline normalize_bool(Register dst, Register temp = R0, bool is_64bit = false);
 185   // Convert between half precision float encoded into a short and a float in a FloatRegister.
 186   void inline f2hf(Register dst, FloatRegister src, FloatRegister tmp);
 187   void inline hf2f(FloatRegister dst, Register src);
 188 
 189   inline void pd_patch_instruction(address branch, address target, const char* file, int line);
 190   NOT_PRODUCT(static void pd_print_patched_instruction(address branch);)
 191 
 192   // Conditional far branch for destinations encodable in 24+2 bits.
 193   // Same interface as bc, e.g. no inverse boint-field.
 194   enum {
 195     bc_far_optimize_not         = 0,
 196     bc_far_optimize_on_relocate = 1
 197   };
 198   // optimize: flag for telling the conditional far branch to optimize
 199   //           itself when relocated.
 200   void bc_far(int boint, int biint, Label& dest, int optimize);
 201   void bc_far_optimized(int boint, int biint, Label& dest); // 1 or 2 instructions
 202   // Relocation of conditional far branches.
 203   static bool    is_bc_far_at(address instruction_addr);
 204   static address get_dest_of_bc_far_at(address instruction_addr);
 205   static void    set_dest_of_bc_far_at(address instruction_addr, address dest);
 206  private:
 207   static bool inline is_bc_far_variant1_at(address instruction_addr);
 208   static bool inline is_bc_far_variant2_at(address instruction_addr);
 209   static bool inline is_bc_far_variant3_at(address instruction_addr);
 210  public:
 211 
 212   // Convenience bc_far versions.
 213   inline void blt_far(ConditionRegister crx, Label& L, int optimize);
 214   inline void bgt_far(ConditionRegister crx, Label& L, int optimize);
 215   inline void beq_far(ConditionRegister crx, Label& L, int optimize);
 216   inline void bso_far(ConditionRegister crx, Label& L, int optimize);
 217   inline void bge_far(ConditionRegister crx, Label& L, int optimize);
 218   inline void ble_far(ConditionRegister crx, Label& L, int optimize);
 219   inline void bne_far(ConditionRegister crx, Label& L, int optimize);
 220   inline void bns_far(ConditionRegister crx, Label& L, int optimize);
 221 
 222   // Emit, identify and patch a NOT mt-safe patchable 64 bit absolute call/jump.
 223  private:
 224   enum {
 225     bxx64_patchable_instruction_count = (2/*load_codecache_const*/ + 3/*5load_const*/ + 1/*mtctr*/ + 1/*bctrl*/),
 226     bxx64_patchable_size              = bxx64_patchable_instruction_count * BytesPerInstWord,
 227     bxx64_patchable_ret_addr_offset   = bxx64_patchable_size
 228   };
 229   void bxx64_patchable(address target, relocInfo::relocType rt, bool link);
 230   static bool is_bxx64_patchable_at(            address instruction_addr, bool link);
 231   // Does the instruction use a pc-relative encoding of the destination?
 232   static bool is_bxx64_patchable_pcrelative_at( address instruction_addr, bool link);
 233   static bool is_bxx64_patchable_variant1_at(   address instruction_addr, bool link);
 234   // Load destination relative to global toc.
 235   static bool is_bxx64_patchable_variant1b_at(  address instruction_addr, bool link);
 236   static bool is_bxx64_patchable_variant2_at(   address instruction_addr, bool link);
 237   static void set_dest_of_bxx64_patchable_at(   address instruction_addr, address target, bool link);
 238   static address get_dest_of_bxx64_patchable_at(address instruction_addr, bool link);
 239 
 240  public:
 241   // call
 242   enum {
 243     bl64_patchable_instruction_count = bxx64_patchable_instruction_count,
 244     bl64_patchable_size              = bxx64_patchable_size,
 245     bl64_patchable_ret_addr_offset   = bxx64_patchable_ret_addr_offset
 246   };
 247   inline void bl64_patchable(address target, relocInfo::relocType rt) {
 248     bxx64_patchable(target, rt, /*link=*/true);
 249   }
 250   inline static bool is_bl64_patchable_at(address instruction_addr) {
 251     return is_bxx64_patchable_at(instruction_addr, /*link=*/true);
 252   }
 253   inline static bool is_bl64_patchable_pcrelative_at(address instruction_addr) {
 254     return is_bxx64_patchable_pcrelative_at(instruction_addr, /*link=*/true);
 255   }
 256   inline static void set_dest_of_bl64_patchable_at(address instruction_addr, address target) {
 257     set_dest_of_bxx64_patchable_at(instruction_addr, target, /*link=*/true);
 258   }
 259   inline static address get_dest_of_bl64_patchable_at(address instruction_addr) {
 260     return get_dest_of_bxx64_patchable_at(instruction_addr, /*link=*/true);
 261   }
 262   // jump
 263   enum {
 264     b64_patchable_instruction_count = bxx64_patchable_instruction_count,
 265     b64_patchable_size              = bxx64_patchable_size,
 266   };
 267   inline void b64_patchable(address target, relocInfo::relocType rt) {
 268     bxx64_patchable(target, rt, /*link=*/false);
 269   }
 270   inline static bool is_b64_patchable_at(address instruction_addr) {
 271     return is_bxx64_patchable_at(instruction_addr, /*link=*/false);
 272   }
 273   inline static bool is_b64_patchable_pcrelative_at(address instruction_addr) {
 274     return is_bxx64_patchable_pcrelative_at(instruction_addr, /*link=*/false);
 275   }
 276   inline static void set_dest_of_b64_patchable_at(address instruction_addr, address target) {
 277     set_dest_of_bxx64_patchable_at(instruction_addr, target, /*link=*/false);
 278   }
 279   inline static address get_dest_of_b64_patchable_at(address instruction_addr) {
 280     return get_dest_of_bxx64_patchable_at(instruction_addr, /*link=*/false);
 281   }
 282 
 283   //
 284   // Support for frame handling
 285   //
 286 
 287   // some ABI-related functions
 288 
 289   // Clobbers all volatile, (non-floating-point) general-purpose registers for debugging purposes.
 290   // This is especially useful for making calls to the JRT in places in which this hasn't been done before;
 291   // e.g. with the introduction of LRBs (load reference barriers) for concurrent garbage collection.
 292   void clobber_volatile_gprs(Register excluded_register = noreg) NOT_DEBUG_RETURN;
 293   // Load bad values into registers that are nonvolatile according to the ABI except R16_thread and R29_TOC.
 294   // This is done after vthread preemption and before vthread resume.
 295   void clobber_nonvolatile_registers() NOT_DEBUG_RETURN;
 296   void clobber_carg_stack_slots(Register tmp);
 297 
 298   int  save_nonvolatile_registers_size(bool include_fp_regs, bool include_vector_regs) {
 299     int size = (32 - 14) * 8; // GP regs
 300     if (include_fp_regs) size += (32 - 14) * 8;
 301     if (include_vector_regs) size += (32 - 20) * 16;
 302     return size;
 303   }
 304   void save_nonvolatile_registers(   Register dst_base, int offset, bool include_fp_regs, bool include_vector_regs);
 305   void restore_nonvolatile_registers(Register src_base, int offset, bool include_fp_regs, bool include_vector_regs);
 306 
 307   enum {
 308     num_volatile_gp_regs = 11,
 309     num_volatile_fp_regs = 14,
 310     num_volatile_regs = num_volatile_gp_regs + num_volatile_fp_regs
 311   };
 312 
 313   void save_volatile_gprs(   Register dst_base, int offset,
 314                              bool include_fp_regs = true, bool include_R3_RET_reg = true);
 315   void restore_volatile_gprs(Register src_base, int offset,
 316                              bool include_fp_regs = true, bool include_R3_RET_reg = true);
 317   void save_LR(Register tmp);
 318   void restore_LR(Register tmp);
 319   void save_LR_CR(Register tmp);     // tmp contains LR on return.
 320   void restore_LR_CR(Register tmp);
 321 
 322   // Get current PC using bl-next-instruction trick.
 323   address get_PC_trash_LR(Register result);
 324 
 325   // Resize current frame either relatively wrt to current SP or absolute.
 326   void resize_frame(Register offset, Register tmp);
 327   void resize_frame(int      offset, Register tmp);
 328   void resize_frame_absolute(Register addr, Register tmp1, Register tmp2);
 329 
 330   // Push a frame of size bytes.
 331   void push_frame(Register bytes, Register tmp);
 332 
 333   // Push a frame of size `bytes'. No abi space provided.
 334   void push_frame(unsigned int bytes, Register tmp);
 335 
 336   // Push a frame of size `bytes' plus native_abi_reg_args on top.
 337   void push_frame_reg_args(unsigned int bytes, Register tmp);
 338 
 339   // pop current C frame
 340   void pop_frame();
 341 
 342   //
 343   // Calls
 344   //
 345 
 346  private:
 347   address _last_calls_return_pc;
 348 
 349 #if defined(ABI_ELFv2)
 350   // Generic version of a call to C function.
 351   // Updates and returns _last_calls_return_pc.
 352   address branch_to(Register function_entry, bool and_link);
 353 #else
 354   // Generic version of a call to C function via a function descriptor
 355   // with variable support for C calling conventions (TOC, ENV, etc.).
 356   // updates and returns _last_calls_return_pc.
 357   address branch_to(Register function_descriptor, bool and_link, bool save_toc_before_call,
 358                     bool restore_toc_after_call, bool load_toc_of_callee, bool load_env_of_callee);
 359 #endif
 360 
 361  public:
 362 
 363   // Get the pc where the last call will return to. returns _last_calls_return_pc.
 364   inline address last_calls_return_pc();
 365 
 366 #if defined(ABI_ELFv2)
 367   // Call a C function via a function descriptor and use full C
 368   // calling conventions. Updates and returns _last_calls_return_pc.
 369   address call_c(Register function_entry);
 370   // For tail calls: only branch, don't link, so callee returns to caller of this function.
 371   address call_c_and_return_to_caller(Register function_entry);
 372   address call_c(address function_entry, relocInfo::relocType rt = relocInfo::none);
 373 #else
 374   // Call a C function via a function descriptor and use full C
 375   // calling conventions. Updates and returns _last_calls_return_pc.
 376   address call_c(Register function_descriptor);
 377   // For tail calls: only branch, don't link, so callee returns to caller of this function.
 378   address call_c_and_return_to_caller(Register function_descriptor);
 379   address call_c(const FunctionDescriptor* function_descriptor, relocInfo::relocType rt);
 380   address call_c(address function_entry, relocInfo::relocType rt = relocInfo::none) {
 381     return call_c((const FunctionDescriptor*)function_entry, rt);
 382   }
 383   address call_c_using_toc(const FunctionDescriptor* function_descriptor, relocInfo::relocType rt,
 384                            Register toc);
 385 #endif
 386 
 387   // CompiledIC call
 388   bool ic_call(Register Rmethod_toc,
 389                address target,
 390                jint method_index = 0,
 391                bool scratch_emit = false,
 392                bool fixed_size = false);
 393   static int ic_check_size();
 394   int ic_check(int end_alignment);
 395 
 396   enum { trampoline_stub_size = 6 * 4 };
 397   address trampoline_call(AddressLiteral target,
 398                           Register Rmethod_toc = noreg,
 399                           bool scratch_emit = false);
 400 
 401   // Inline type specific methods
 402 #include "asm/macroAssembler_common.hpp"
 403 
 404   void save_stack_increment(int sp_inc, int frame_size);
 405 
 406  protected:
 407 
 408   // It is imperative that all calls into the VM are handled via the
 409   // call_VM macros. They make sure that the stack linkage is setup
 410   // correctly. call_VM's correspond to ENTRY/ENTRY_X entry points
 411   // while call_VM_leaf's correspond to LEAF entry points.
 412   //
 413   // This is the base routine called by the different versions of
 414   // call_VM. The interpreter may customize this version by overriding
 415   // it for its purposes (e.g., to save/restore additional registers
 416   // when doing a VM call).
 417   //
 418   // If no last_java_sp is specified (noreg) then SP will be used instead.
 419   virtual void call_VM_base(
 420      // where an oop-result ends up if any; use noreg otherwise
 421     Register        oop_result,
 422     // to set up last_Java_frame in stubs; use noreg otherwise
 423     Register        last_java_sp,
 424     // the entry point
 425     address         entry_point,
 426     // flag which indicates if exception should be checked
 427     bool            check_exception = true,
 428     Label* last_java_pc = nullptr
 429   );
 430 
 431   // Support for VM calls. This is the base routine called by the
 432   // different versions of call_VM_leaf. The interpreter may customize
 433   // this version by overriding it for its purposes (e.g., to
 434   // save/restore additional registers when doing a VM call).
 435   void call_VM_leaf_base(address entry_point);
 436 
 437  public:
 438   // Call into the VM.
 439   // Passes the thread pointer (in R3_ARG1) as a prepended argument.
 440   // Makes sure oop return values are visible to the GC.
 441   void call_VM(Register oop_result, address entry_point, bool check_exceptions = true, Label* last_java_pc = nullptr);
 442   void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
 443   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 444   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg3, bool check_exceptions = true);
 445   void call_VM_leaf(address entry_point);
 446   void call_VM_leaf(address entry_point, Register arg_1);
 447   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 448   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 449 
 450   // Call a stub function via a function descriptor, but don't save
 451   // TOC before call, don't setup TOC and ENV for call, and don't
 452   // restore TOC after call. Updates and returns _last_calls_return_pc.
 453   inline address call_stub(Register function_entry);
 454   inline void call_stub_and_return_to(Register function_entry, Register return_pc);
 455 
 456   void post_call_nop();
 457   static bool is_post_call_nop(int instr_bits) {
 458     const uint32_t nineth_bit = opp_u_field(1, 9, 9);
 459     const uint32_t opcode_mask = 0b111110 << OPCODE_SHIFT;
 460     const uint32_t pcn_mask = opcode_mask | nineth_bit;
 461     return (instr_bits & pcn_mask) == (Assembler::CMPLI_OPCODE | nineth_bit);
 462   }
 463 
 464   //
 465   // Java utilities
 466   //
 467 
 468   // Read from the polling page, its address is already in a register.
 469   inline void load_from_polling_page(Register polling_page_address, int offset = 0);
 470   // Check whether instruction is a read access to the polling page
 471   // which was emitted by load_from_polling_page(..).
 472   static bool is_load_from_polling_page(int instruction, void* ucontext/*may be nullptr*/,
 473                                         address* polling_address_ptr = nullptr);
 474 
 475   // Support for null-checks
 476   //
 477   // Generates code that causes a null OS exception if the content of reg is null.
 478   // If the accessed location is M[reg + offset] and the offset is known, provide the
 479   // offset. No explicit code generation is needed if the offset is within a certain
 480   // range (0 <= offset <= page_size).
 481 
 482   // Stack overflow checking
 483   void bang_stack_with_offset(int offset);
 484 
 485   // If instruction is a stack bang of the form ld, stdu, or
 486   // stdux, return the banged address. Otherwise, return 0.
 487   static address get_stack_bang_address(int instruction, void* ucontext);
 488 
 489   // Check for reserved stack access in method being exited. If the reserved
 490   // stack area was accessed, protect it again and throw StackOverflowError.
 491   void reserved_stack_check(Register return_pc);
 492 
 493   // Atomics
 494   // CmpxchgX sets condition register to cmpX(current, compare).
 495   // (flag == ne) => (dest_current_value != compare_value), (!swapped)
 496   // (flag == eq) => (dest_current_value == compare_value), ( swapped)
 497   static inline bool cmpxchgx_hint_acquire_lock()  { return true; }
 498   // The stxcx will probably not be succeeded by a releasing store.
 499   static inline bool cmpxchgx_hint_release_lock()  { return false; }
 500   static inline bool cmpxchgx_hint_atomic_update() { return false; }
 501 
 502   // Cmpxchg semantics
 503   enum {
 504     MemBarNone = 0,
 505     MemBarRel  = 1,
 506     MemBarAcq  = 2,
 507     MemBarFenceAfter = 4 // use powers of 2
 508   };
 509  private:
 510   // Helper functions for word/sub-word atomics.
 511   void atomic_get_and_modify_generic(Register dest_current_value, Register exchange_value,
 512                                      Register addr_base, Register tmp1, Register tmp2, Register tmp3,
 513                                      bool cmpxchgx_hint, bool is_add, int size);
 514   void cmpxchg_loop_body(ConditionRegister flag, Register dest_current_value,
 515                          RegisterOrConstant compare_value, Register exchange_value,
 516                          Register addr_base,Label &retry, Label &failed, bool cmpxchgx_hint, int size);
 517   void cmpxchg_generic(ConditionRegister flag, Register dest_current_value,
 518                        RegisterOrConstant compare_value, Register exchange_value,
 519                        Register addr_base, int semantics, bool cmpxchgx_hint, Register int_flag_success,
 520                        Label* failed_ext, bool contention_hint, bool weak, int size);
 521  public:
 522   // Temps and addr_base are killed if processor does not support Power 8 instructions.
 523   // Result will be sign extended.
 524   void getandsetb(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, 1);
 527   }
 528   // Temps and addr_base are killed if processor does not support Power 8 instructions.
 529   // Result will be sign extended.
 530   void getandseth(Register dest_current_value, Register exchange_value, Register addr_base,
 531                   Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {
 532     atomic_get_and_modify_generic(dest_current_value, exchange_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, false, 2);
 533   }
 534   void getandsetw(Register dest_current_value, Register exchange_value, Register addr_base,
 535                   bool cmpxchgx_hint) {
 536     atomic_get_and_modify_generic(dest_current_value, exchange_value, addr_base, noreg, noreg, noreg, cmpxchgx_hint, false, 4);
 537   }
 538   void getandsetd(Register dest_current_value, Register exchange_value, Register addr_base,
 539                   bool cmpxchgx_hint);
 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 getandaddb(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, 1);
 545   }
 546   // tmp2/3 and addr_base are killed if processor does not support Power 8 instructions (tmp1 is always needed).
 547   // Result will be sign extended.
 548   void getandaddh(Register dest_current_value, Register inc_value, Register addr_base,
 549                   Register tmp1, Register tmp2, Register tmp3, bool cmpxchgx_hint) {
 550     atomic_get_and_modify_generic(dest_current_value, inc_value, addr_base, tmp1, tmp2, tmp3, cmpxchgx_hint, true, 2);
 551   }
 552   void getandaddw(Register dest_current_value, Register inc_value, Register addr_base,
 553                   Register tmp1, bool cmpxchgx_hint) {
 554     atomic_get_and_modify_generic(dest_current_value, inc_value, addr_base, tmp1, noreg, noreg, cmpxchgx_hint, true, 4);
 555   }
 556   void getandaddd(Register dest_current_value, Register exchange_value, Register addr_base,
 557                   Register tmp, bool cmpxchgx_hint);
 558   // Temps, addr_base and exchange_value are killed if processor does not support Power 8 instructions.
 559   // compare_value must be at least 32 bit sign extended. Result will be sign extended.
 560   void cmpxchgb(ConditionRegister flag, Register dest_current_value,
 561                 RegisterOrConstant compare_value, Register exchange_value,
 562                 Register addr_base, int semantics, bool cmpxchgx_hint = false,
 563                 Register int_flag_success = noreg, Label* failed = nullptr,
 564                 bool contention_hint = false, bool weak = false) {
 565     cmpxchg_generic(flag, dest_current_value, compare_value, exchange_value, addr_base, semantics,
 566                     cmpxchgx_hint, int_flag_success, failed, contention_hint, weak, 1);
 567   }
 568   // Temps, addr_base and exchange_value are killed if processor does not support Power 8 instructions.
 569   // compare_value must be at least 32 bit sign extended. Result will be sign extended.
 570   void cmpxchgh(ConditionRegister flag, Register dest_current_value,
 571                 RegisterOrConstant compare_value, Register exchange_value,
 572                 Register addr_base, int semantics, bool cmpxchgx_hint = false,
 573                 Register int_flag_success = noreg, Label* failed = nullptr,
 574                 bool contention_hint = false, bool weak = false) {
 575     cmpxchg_generic(flag, dest_current_value, compare_value, exchange_value, addr_base,
 576                     semantics, cmpxchgx_hint, int_flag_success, failed, contention_hint, weak, 2);
 577   }
 578   void cmpxchgw(ConditionRegister flag, Register dest_current_value,
 579                 RegisterOrConstant compare_value, Register exchange_value,
 580                 Register addr_base,
 581                 int semantics, bool cmpxchgx_hint = false, Register int_flag_success = noreg,
 582                 Label* failed = nullptr, bool contention_hint = false, bool weak = false) {
 583     cmpxchg_generic(flag, dest_current_value, compare_value, exchange_value, addr_base,
 584                     semantics, cmpxchgx_hint, int_flag_success, failed, contention_hint, weak, 4);
 585   }
 586   void cmpxchgd(ConditionRegister flag, Register dest_current_value,
 587                 RegisterOrConstant compare_value, Register exchange_value,
 588                 Register addr_base,
 589                 int semantics, bool cmpxchgx_hint = false, Register int_flag_success = noreg,
 590                 Label* failed = nullptr, bool contention_hint = false, bool weak = false);
 591 
 592   // interface method calling
 593   void lookup_interface_method(Register recv_klass,
 594                                Register intf_klass,
 595                                RegisterOrConstant itable_index,
 596                                Register method_result,
 597                                Register temp_reg, Register temp2_reg,
 598                                Label& no_such_interface,
 599                                bool return_method = true);
 600 
 601   // virtual method calling
 602   void lookup_virtual_method(Register recv_klass,
 603                              RegisterOrConstant vtable_index,
 604                              Register method_result);
 605 
 606   // Test sub_klass against super_klass, with fast and slow paths.
 607 
 608   // The fast path produces a tri-state answer: yes / no / maybe-slow.
 609   // One of the three labels can be null, meaning take the fall-through.
 610   // If super_check_offset is -1, the value is loaded up from super_klass.
 611   // No registers are killed, except temp_reg and temp2_reg.
 612   // If super_check_offset is not -1, temp2_reg is not used and can be noreg.
 613   void check_klass_subtype_fast_path(Register sub_klass,
 614                                      Register super_klass,
 615                                      Register temp1_reg,
 616                                      Register temp2_reg,
 617                                      Label* L_success,
 618                                      Label* L_failure,
 619                                      Label* L_slow_path = nullptr, // default fall through
 620                                      RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
 621 
 622   // The rest of the type check; must be wired to a corresponding fast path.
 623   // It does not repeat the fast path logic, so don't use it standalone.
 624   // The temp_reg can be noreg, if no temps are available.
 625   // It can also be sub_klass or super_klass, meaning it's OK to kill that one.
 626   // Updates the sub's secondary super cache as necessary.
 627   void check_klass_subtype_slow_path_linear(Register sub_klass,
 628                                             Register super_klass,
 629                                             Register temp1_reg,
 630                                             Register temp2_reg,
 631                                             Label* L_success = nullptr,
 632                                             Register result_reg = noreg);
 633 
 634   void check_klass_subtype_slow_path_table(Register sub_klass,
 635                                            Register super_klass,
 636                                            Register temp1_reg,
 637                                            Register temp2_reg,
 638                                            Label* L_success = nullptr,
 639                                            Register result_reg = noreg);
 640 
 641   void check_klass_subtype_slow_path(Register sub_klass,
 642                                      Register super_klass,
 643                                      Register temp1_reg,
 644                                      Register temp2_reg,
 645                                      Label* L_success = nullptr,
 646                                      Register result_reg = noreg);
 647 
 648   void lookup_secondary_supers_table_var(Register sub_klass,
 649                                          Register r_super_klass,
 650                                          Register temp1,
 651                                          Register temp2,
 652                                          Register temp3,
 653                                          Register temp4,
 654                                          Register result);
 655 
 656   // If r is valid, return r.
 657   // If r is invalid, remove a register r2 from available_regs, add r2
 658   // to regs_to_push, then return r2.
 659   Register allocate_if_noreg(const Register r,
 660                              RegSetIterator<Register> &available_regs,
 661                              RegSet &regs_to_push);
 662 
 663   // Frameless register spills (negative offset from SP)
 664   void push_set(RegSet set);
 665   void pop_set(RegSet set);
 666 
 667   // Simplified, combined version, good for typical uses.
 668   // Falls through on failure.
 669   void check_klass_subtype(Register sub_klass,
 670                            Register super_klass,
 671                            Register temp1_reg,
 672                            Register temp2_reg,
 673                            Label& L_success);
 674 
 675   void repne_scan(Register addr, Register value, Register count, Register scratch);
 676 
 677   // As above, but with a constant super_klass.
 678   // The result is in Register result, not the condition codes.
 679   void lookup_secondary_supers_table_const(Register r_sub_klass,
 680                                            Register r_super_klass,
 681                                            Register temp1,
 682                                            Register temp2,
 683                                            Register temp3,
 684                                            Register temp4,
 685                                            Register result,
 686                                            u1 super_klass_slot);
 687 
 688   void verify_secondary_supers_table(Register r_sub_klass,
 689                                      Register r_super_klass,
 690                                      Register result,
 691                                      Register temp1,
 692                                      Register temp2,
 693                                      Register temp3);
 694 
 695   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
 696                                                Register r_array_base,
 697                                                Register r_array_index,
 698                                                Register r_bitmap,
 699                                                Register result,
 700                                                Register temp1);
 701 
 702   void clinit_barrier(Register klass,
 703                       Register thread,
 704                       Label* L_fast_path = nullptr,
 705                       Label* L_slow_path = nullptr);
 706 
 707   // Method handle support (JSR 292).
 708   RegisterOrConstant argument_offset(RegisterOrConstant arg_slot, Register temp_reg, int extra_slot_offset = 0);
 709 
 710   void push_cont_fastpath();
 711   void pop_cont_fastpath();
 712   void atomically_flip_locked_state(bool is_unlock, Register obj, Register tmp, Label& failed, int semantics);
 713   void fast_lock(Register box, Register obj, Register t1, Register t2, Label& slow);
 714   void fast_unlock(Register obj, Register t1, Label& slow);
 715 
 716   // allocation (for C1)
 717   void tlab_allocate(
 718     Register obj,                      // result: pointer to object after successful allocation
 719     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
 720     int      con_size_in_bytes,        // object size in bytes if   known at compile time
 721     Register t1,                       // temp register
 722     Label&   slow_case                 // continuation point if fast allocation fails
 723   );
 724 
 725   void compiler_fast_lock_object(ConditionRegister flag, Register oop, Register box,
 726                                  Register tmp1, Register tmp2, Register tmp3);
 727 
 728   void compiler_fast_unlock_object(ConditionRegister flag, Register oop, Register box,
 729                                    Register tmp1, Register tmp2, Register tmp3);
 730 
 731   // Check if safepoint requested and if so branch
 732   void safepoint_poll(Label& slow_path, Register temp, bool at_return, bool in_nmethod);
 733   void jump_to_polling_page_return_handler_blob(int safepoint_offset, bool fixed_size = false);
 734 
 735   void resolve_jobject(Register value, Register tmp1, Register tmp2,
 736                        MacroAssembler::PreservationLevel preservation_level);
 737   void resolve_global_jobject(Register value, Register tmp1, Register tmp2,
 738                               MacroAssembler::PreservationLevel preservation_level);
 739 
 740   // Support for managing the JavaThread pointer (i.e.; the reference to
 741   // thread-local information).
 742 
 743   // Support for last Java frame (but use call_VM instead where possible):
 744   // access R16_thread->last_Java_sp.
 745   void set_last_Java_frame(Register last_java_sp, Register last_Java_pc);
 746   void reset_last_Java_frame(bool check_last_java_sp = true);
 747   void set_top_ijava_frame_at_SP_as_last_Java_frame(Register sp, Register tmp1, Label* jpc = nullptr);
 748 
 749   // Read vm result from thread: oop_result = R16_thread->result;
 750   void get_vm_result_oop(Register oop_result);
 751   void get_vm_result_metadata(Register metadata_result);
 752 
 753   static bool needs_explicit_null_check(intptr_t offset);
 754   static bool uses_implicit_null_check(void* address);
 755 
 756   // Trap-instruction-based checks.
 757   // Range checks can be distinguished from zero checks as they check 32 bit,
 758   // zero checks all 64 bits (tw, td).
 759   inline void trap_null_check(Register a, trap_to_bits cmp = traptoEqual);
 760   static bool is_trap_null_check(int x) {
 761     return is_tdi(x, traptoEqual,               -1/*any reg*/, 0) ||
 762            is_tdi(x, traptoGreaterThanUnsigned, -1/*any reg*/, 0);
 763   }
 764 
 765   inline void trap_ic_miss_check(Register a, Register b);
 766   static bool is_trap_ic_miss_check(int x) {
 767     return is_td(x, traptoGreaterThanUnsigned | traptoLessThanUnsigned, -1/*any reg*/, -1/*any reg*/);
 768   }
 769 
 770   // Implicit or explicit null check, jumps to static address exception_entry.
 771   inline void null_check_throw(Register a, int offset, Register temp_reg, address exception_entry);
 772   inline void null_check(Register a, int offset, Label *Lis_null); // implicit only if Lis_null not provided
 773 
 774   // Access heap oop, handle encoding and GC barriers.
 775   // Some GC barriers call C so use needs_frame = true if an extra frame is needed at the current call site.
 776   inline void access_store_at(BasicType type, DecoratorSet decorators,
 777                               Register base, RegisterOrConstant ind_or_offs, Register val,
 778                               Register tmp1, Register tmp2, Register tmp3,
 779                               MacroAssembler::PreservationLevel preservation_level);
 780   inline void access_load_at(BasicType type, DecoratorSet decorators,
 781                              Register base, RegisterOrConstant ind_or_offs, Register dst,
 782                              Register tmp1, Register tmp2,
 783                              MacroAssembler::PreservationLevel preservation_level, Label *L_handle_null = nullptr);
 784 
 785  public:
 786   // Specify tmp1 for better code in certain compressed oops cases. Specify Label to bail out on null oop.
 787   // tmp1, tmp2 and needs_frame are used with decorators ON_PHANTOM_OOP_REF or ON_WEAK_OOP_REF.
 788   inline void load_heap_oop(Register d, RegisterOrConstant offs, Register s1,
 789                             Register tmp1, Register tmp2,
 790                             MacroAssembler::PreservationLevel preservation_level,
 791                             DecoratorSet decorators = 0, Label *L_handle_null = nullptr);
 792 
 793   inline void store_heap_oop(Register d, RegisterOrConstant offs, Register s1,
 794                              Register tmp1, Register tmp2, Register tmp3,
 795                              MacroAssembler::PreservationLevel preservation_level, DecoratorSet decorators = 0);
 796 
 797   // Encode/decode heap oop. Oop may not be null, else en/decoding goes wrong.
 798   // src == d allowed.
 799   inline Register encode_heap_oop_not_null(Register d, Register src = noreg);
 800   inline Register decode_heap_oop_not_null(Register d, Register src = noreg);
 801 
 802   // Null allowed.
 803   inline Register encode_heap_oop(Register d, Register src); // Prefer null check in GC barrier!
 804   inline void decode_heap_oop(Register d);
 805 
 806   // Load/Store klass oop from klass field. Compress.
 807   void load_klass_no_decode(Register dst, Register src);
 808   void load_klass(Register dst, Register src);
 809   void load_narrow_klass_compact(Register dst, Register src);
 810   void cmp_klass(ConditionRegister dst, Register obj, Register klass, Register tmp, Register tmp2);
 811   void cmp_klasses_from_objects(ConditionRegister dst, Register obj1, Register obj2, Register tmp1, Register tmp2);
 812   void load_klass_check_null(Register dst, Register src, Label* is_null = nullptr);
 813   void store_klass(Register dst_oop, Register klass, Register tmp = R0);
 814   void store_klass_gap(Register dst_oop, Register val = noreg); // Will store 0 if val not specified.
 815 
 816   void resolve_oop_handle(Register result, Register tmp1, Register tmp2,
 817                           MacroAssembler::PreservationLevel preservation_level);
 818   void resolve_weak_handle(Register result, Register tmp1, Register tmp2,
 819                            MacroAssembler::PreservationLevel preservation_level);
 820   void load_method_holder(Register holder, Register method);
 821 
 822   void decode_klass_not_null(Register dst, Register src = noreg);
 823   Register encode_klass_not_null(Register dst, Register src = noreg);
 824 
 825   // markWord tests, kills markWord reg
 826   void test_markword_is_inline_type(Register markword, Label& is_inline_type);
 827 
 828   // inlineKlass queries, kills temp_reg
 829   void test_oop_is_not_inline_type(Register object, Label& not_inline_type, bool can_be_null = true);
 830 
 831   void test_field_is_null_free_inline_type(Register flags, Label& is_null_free);
 832   void test_field_is_not_null_free_inline_type(Register flags, Label& not_null_free);
 833   void test_field_is_flat(Register flags, Label& is_flat);
 834 
 835   // Check oops for special arrays, i.e. flat arrays and/or null-free arrays
 836   void test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label, bool maybe_far = false);
 837   void test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array, bool maybe_far = false);
 838   void test_non_flat_array_oop(Register oop, Register temp_reg, Label& is_non_flat_array);
 839   void test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array, bool maybe_far = false);
 840   void test_non_null_free_array_oop(Register oop, Register temp_reg, Label& is_non_null_free_array);
 841 
 842   // Check array klass layout helper for flat or null-free arrays...
 843   void test_flat_array_layout(Register lh, Label& is_flat_array);
 844 
 845   void load_metadata(Register dst, Register src);
 846 
 847   void flat_field_copy(DecoratorSet decorators, Register src, Register dst, Register inline_layout_info);
 848 
 849   void load_prototype_header(Register dst, Register src);
 850 
 851   void inline_layout_info(Register holder_klass, Register index, Register layout_info);
 852 
 853   // inline type data payload offsets...
 854   void payload_offset(Register inline_klass, Register offset);
 855   void payload_address(Register oop, Register data, Register inline_klass, Register t1);
 856 
 857   // SIGTRAP-based range checks for arrays.
 858   inline void trap_range_check_l(Register a, Register b);
 859   inline void trap_range_check_l(Register a, int si16);
 860   static bool is_trap_range_check_l(int x) {
 861     return (is_tw (x, traptoLessThanUnsigned, -1/*any reg*/, -1/*any reg*/) ||
 862             is_twi(x, traptoLessThanUnsigned, -1/*any reg*/)                  );
 863   }
 864   inline void trap_range_check_le(Register a, int si16);
 865   static bool is_trap_range_check_le(int x) {
 866     return is_twi(x, traptoEqual | traptoLessThanUnsigned, -1/*any reg*/);
 867   }
 868   inline void trap_range_check_g(Register a, int si16);
 869   static bool is_trap_range_check_g(int x) {
 870     return is_twi(x, traptoGreaterThanUnsigned, -1/*any reg*/);
 871   }
 872   inline void trap_range_check_ge(Register a, Register b);
 873   inline void trap_range_check_ge(Register a, int si16);
 874   static bool is_trap_range_check_ge(int x) {
 875     return (is_tw (x, traptoEqual | traptoGreaterThanUnsigned, -1/*any reg*/, -1/*any reg*/) ||
 876             is_twi(x, traptoEqual | traptoGreaterThanUnsigned, -1/*any reg*/)                  );
 877   }
 878   static bool is_trap_range_check(int x) {
 879     return is_trap_range_check_l(x) || is_trap_range_check_le(x) ||
 880            is_trap_range_check_g(x) || is_trap_range_check_ge(x);
 881   }
 882 
 883   void clear_memory_unrolled(Register base_ptr, int cnt_dwords, Register tmp = R0, int offset = 0);
 884   void clear_memory_constlen(Register base_ptr, int cnt_dwords, Register tmp = R0);
 885   void clear_memory_doubleword(Register base_ptr, Register cnt_dwords, Register tmp = R0, long const_cnt = -1);
 886   void fill_words(Register base, Register cnt, Register value);
 887 
 888   // Emitters for BigInteger.multiplyToLen intrinsic.
 889   inline void multiply64(Register dest_hi, Register dest_lo,
 890                          Register x, Register y);
 891   void add2_with_carry(Register dest_hi, Register dest_lo,
 892                        Register src1, Register src2);
 893   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
 894                              Register y, Register y_idx, Register z,
 895                              Register carry, Register product_high, Register product,
 896                              Register idx, Register kdx, Register tmp);
 897   void multiply_add_128_x_128(Register x_xstart, Register y, Register z,
 898                               Register yz_idx, Register idx, Register carry,
 899                               Register product_high, Register product, Register tmp,
 900                               int offset);
 901   void multiply_128_x_128_loop(Register x_xstart,
 902                                Register y, Register z,
 903                                Register yz_idx, Register idx, Register carry,
 904                                Register product_high, Register product,
 905                                Register carry2, Register tmp);
 906   void muladd(Register out, Register in, Register offset, Register len, Register k,
 907               Register tmp1, Register tmp2, Register carry);
 908   void multiply_to_len(Register x, Register xlen,
 909                        Register y, Register ylen,
 910                        Register z,
 911                        Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
 912                        Register tmp6, Register tmp7, Register tmp8, Register tmp9, Register tmp10,
 913                        Register tmp11, Register tmp12, Register tmp13);
 914 
 915   // non-atomic 64-bit memory increment by simm16
 916   void increment_mem64(Register base, RegisterOrConstant ind_or_offs, int val, Register tmp);
 917 
 918   // Bytecode profiling (tmp2 = noreg is allowed, but then recv is killed)
 919   void profile_receiver_type(Register recv, Register mdp, int mdp_offset, Register tmp1, Register tmp2);
 920 
 921   // Emitters for CRC32 calculation.
 922   // A note on invertCRC:
 923   //   Unfortunately, internal representation of crc differs between CRC32 and CRC32C.
 924   //   CRC32 holds it's current crc value in the externally visible representation.
 925   //   CRC32C holds it's current crc value in internal format, ready for updating.
 926   //   Thus, the crc value must be bit-flipped before updating it in the CRC32 case.
 927   //   In the CRC32C case, it must be bit-flipped when it is given to the outside world (getValue()).
 928   //   The bool invertCRC parameter indicates whether bit-flipping is required before updates.
 929   void load_reverse_32(Register dst, Register src);
 930   int  crc32_table_columns(Register table, Register tc0, Register tc1, Register tc2, Register tc3);
 931   void fold_byte_crc32(Register crc, Register val, Register table, Register tmp);
 932   void update_byte_crc32(Register crc, Register val, Register table);
 933   void update_byteLoop_crc32(Register crc, Register buf, Register len, Register table,
 934                              Register data, bool loopAlignment);
 935   void update_1word_crc32(Register crc, Register buf, Register table, int bufDisp, int bufInc,
 936                           Register t0,  Register t1,  Register t2,  Register t3,
 937                           Register tc0, Register tc1, Register tc2, Register tc3);
 938   void kernel_crc32_vpmsum(Register crc, Register buf, Register len, Register constants,
 939                            Register t0, Register t1, Register t2, Register t3, Register t4,
 940                            Register t5, Register t6, bool invertCRC);
 941   void kernel_crc32_vpmsum_aligned(Register crc, Register buf, Register len, Register constants,
 942                                    Register t0, Register t1, Register t2, Register t3, Register t4,
 943                                    Register t5, Register t6);
 944   // Version which internally decides what to use.
 945   void crc32(Register crc, Register buf, Register len, Register t0, Register t1, Register t2,
 946              Register t3, Register t4, Register t5, Register t6, Register t7, bool is_crc32c);
 947 
 948   void kernel_crc32_singleByteReg(Register crc, Register val, Register table,
 949                                   bool invertCRC);
 950 
 951   // SHA-2 auxiliary functions and public interfaces
 952  private:
 953   void sha256_deque(const VectorRegister src,
 954       const VectorRegister dst1, const VectorRegister dst2, const VectorRegister dst3);
 955   void sha256_load_h_vec(const VectorRegister a, const VectorRegister e, const Register hptr);
 956   void sha256_round(const VectorRegister* hs, const int total_hs, int& h_cnt, const VectorRegister kpw);
 957   void sha256_load_w_plus_k_vec(const Register buf_in, const VectorRegister* ws,
 958       const int total_ws, const Register k, const VectorRegister* kpws,
 959       const int total_kpws);
 960   void sha256_calc_4w(const VectorRegister w0, const VectorRegister w1,
 961       const VectorRegister w2, const VectorRegister w3, const VectorRegister kpw0,
 962       const VectorRegister kpw1, const VectorRegister kpw2, const VectorRegister kpw3,
 963       const Register j, const Register k);
 964   void sha256_update_sha_state(const VectorRegister a, const VectorRegister b,
 965       const VectorRegister c, const VectorRegister d, const VectorRegister e,
 966       const VectorRegister f, const VectorRegister g, const VectorRegister h,
 967       const Register hptr);
 968 
 969   void sha512_load_w_vec(const Register buf_in, const VectorRegister* ws, const int total_ws);
 970   void sha512_update_sha_state(const Register state, const VectorRegister* hs, const int total_hs);
 971   void sha512_round(const VectorRegister* hs, const int total_hs, int& h_cnt, const VectorRegister kpw);
 972   void sha512_load_h_vec(const Register state, const VectorRegister* hs, const int total_hs);
 973   void sha512_calc_2w(const VectorRegister w0, const VectorRegister w1,
 974       const VectorRegister w2, const VectorRegister w3,
 975       const VectorRegister w4, const VectorRegister w5,
 976       const VectorRegister w6, const VectorRegister w7,
 977       const VectorRegister kpw0, const VectorRegister kpw1, const Register j,
 978       const VectorRegister vRb, const Register k);
 979 
 980  public:
 981   void sha256(bool multi_block);
 982   void sha512(bool multi_block);
 983 
 984   void cache_wb(Address line);
 985   void cache_wbsync(bool is_presync);
 986 
 987   //
 988   // Debugging
 989   //
 990 
 991   // assert on cr0
 992   enum AsmAssertCond {
 993     eq,
 994     ne,
 995     ge,
 996     gt,
 997     lt,
 998     le
 999   };
1000   void asm_assert(AsmAssertCond cond, const char* msg) PRODUCT_RETURN;
1001   void asm_assert_eq(const char* msg) { asm_assert(eq, msg); }
1002   void asm_assert_ne(const char* msg) { asm_assert(ne, msg); }
1003 
1004  private:
1005   void asm_assert_mems_zero(AsmAssertCond cond, int size, int mem_offset, Register mem_base,
1006                             const char* msg) NOT_DEBUG_RETURN;
1007 
1008  public:
1009 
1010   void asm_assert_mem8_is_zero(int mem_offset, Register mem_base, const char* msg) {
1011     asm_assert_mems_zero(eq,  8, mem_offset, mem_base, msg);
1012   }
1013   void asm_assert_mem8_isnot_zero(int mem_offset, Register mem_base, const char* msg) {
1014     asm_assert_mems_zero(ne, 8, mem_offset, mem_base, msg);
1015   }
1016 
1017   // Calls verify_oop. If UseCompressedOops is on, decodes the oop.
1018   // Preserves reg.
1019   void verify_coop(Register reg, const char*);
1020   // Emit code to verify that reg contains a valid oop if +VerifyOops is set.
1021   void verify_oop(Register reg, const char* s = "broken oop");
1022   void verify_oop_addr(RegisterOrConstant offs, Register base, const char* s = "contains broken oop");
1023 
1024   // TODO: verify method and klass metadata (compare against vptr?)
1025   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1026   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line) {}
1027 
1028   // Convenience method returning function entry. For the ELFv1 case
1029   // creates function descriptor at the current address and returns
1030   // the pointer to it. For the ELFv2 case returns the current address.
1031   inline address function_entry();
1032 
1033 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1034 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1035 
1036  private:
1037   void stop(int type, const char* msg);
1038 
1039  public:
1040   enum {
1041     stop_stop               = 0,
1042     stop_untested           = 1,
1043     stop_unimplemented      = 2,
1044     stop_shouldnotreachhere = 3,
1045     stop_msg_present        = -0x8000
1046   };
1047 
1048   // Prints msg, dumps registers and stops execution.
1049   void stop                 (const char* msg = nullptr) { stop(stop_stop,               msg); }
1050   void untested             (const char* msg = nullptr) { stop(stop_untested,           msg); }
1051   void unimplemented        (const char* msg = nullptr) { stop(stop_unimplemented,      msg); }
1052   void should_not_reach_here(const char* msg = nullptr) { stop(stop_shouldnotreachhere, msg); }
1053 
1054   void zap_from_to(Register low, int before, Register high, int after, Register val, Register addr) PRODUCT_RETURN;
1055 
1056   // Inline type specific methods
1057   #include "asm/macroAssembler_common.hpp"
1058 };
1059 
1060 #endif // CPU_PPC_MACROASSEMBLER_PPC_HPP