1 /*
   2  * Copyright (c) 2016, 2026, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016, 2024 SAP SE. All rights reserved.
   4  * Copyright (c) 2024 IBM Corporation. All rights reserved.
   5  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   6  *
   7  * This code is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License version 2 only, as
   9  * published by the Free Software Foundation.
  10  *
  11  * This code is distributed in the hope that it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14  * version 2 for more details (a copy is included in the LICENSE file that
  15  * accompanied this code).
  16  *
  17  * You should have received a copy of the GNU General Public License version
  18  * 2 along with this work; if not, write to the Free Software Foundation,
  19  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  20  *
  21  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  22  * or visit www.oracle.com if you need additional information or have any
  23  * questions.
  24  *
  25  */
  26 
  27 #ifndef CPU_S390_MACROASSEMBLER_S390_HPP
  28 #define CPU_S390_MACROASSEMBLER_S390_HPP
  29 
  30 #include "asm/assembler.hpp"
  31 #include "oops/accessDecorators.hpp"
  32 
  33 class ciInlineKlass;
  34 class SigEntry;
  35 class VMRegPair;
  36 
  37 #define MODERN_IFUN(name)  ((void (MacroAssembler::*)(Register, int64_t, Register, Register))&MacroAssembler::name)
  38 #define CLASSIC_IFUN(name) ((void (MacroAssembler::*)(Register, int64_t, Register, Register))&MacroAssembler::name)
  39 #define MODERN_FFUN(name)  ((void (MacroAssembler::*)(FloatRegister, int64_t, Register, Register))&MacroAssembler::name)
  40 #define CLASSIC_FFUN(name) ((void (MacroAssembler::*)(FloatRegister, int64_t, Register, Register))&MacroAssembler::name)
  41 
  42 class MacroAssembler: public Assembler {
  43  public:
  44   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
  45 
  46   //
  47   // Optimized instruction emitters
  48   //
  49 
  50   // Move register if destination register and target register are different.
  51   void lr_if_needed(Register rd, Register rs);
  52   void lgr_if_needed(Register rd, Register rs);
  53   void llgfr_if_needed(Register rd, Register rs);
  54   void ldr_if_needed(FloatRegister rd, FloatRegister rs);
  55 
  56   void move_reg_if_needed(Register dest, BasicType dest_type, Register src, BasicType src_type);
  57   void move_freg_if_needed(FloatRegister dest, BasicType dest_type, FloatRegister src, BasicType src_type);
  58 
  59   void freg2mem_opt(FloatRegister reg,
  60                     int64_t       disp,
  61                     Register      index,
  62                     Register      base,
  63                     void (MacroAssembler::*modern) (FloatRegister, int64_t, Register, Register),
  64                     void (MacroAssembler::*classic)(FloatRegister, int64_t, Register, Register),
  65                     Register      scratch = Z_R0);
  66   void freg2mem_opt(FloatRegister reg,
  67                     const Address &a, bool is_double = true);
  68 
  69   void mem2freg_opt(FloatRegister reg,
  70                     int64_t       disp,
  71                     Register      index,
  72                     Register      base,
  73                     void (MacroAssembler::*modern) (FloatRegister, int64_t, Register, Register),
  74                     void (MacroAssembler::*classic)(FloatRegister, int64_t, Register, Register),
  75                     Register      scratch = Z_R0);
  76   void mem2freg_opt(FloatRegister reg,
  77                     const Address &a, bool is_double = true);
  78 
  79   void reg2mem_opt(Register reg,
  80                    int64_t  disp,
  81                    Register index,
  82                    Register base,
  83                    void (MacroAssembler::*modern) (Register, int64_t, Register, Register),
  84                    void (MacroAssembler::*classic)(Register, int64_t, Register, Register),
  85                    Register scratch = Z_R0);
  86   // returns offset of the store instruction
  87   int reg2mem_opt(Register reg, const Address &a, bool is_double = true);
  88 
  89   void mem2reg_opt(Register reg,
  90                    int64_t  disp,
  91                    Register index,
  92                    Register base,
  93                    void (MacroAssembler::*modern) (Register, int64_t, Register, Register),
  94                    void (MacroAssembler::*classic)(Register, int64_t, Register, Register));
  95   void mem2reg_opt(Register reg, const Address &a, bool is_double = true);
  96   void mem2reg_signed_opt(Register reg, const Address &a);
  97 
  98   // AND immediate and set condition code, works for 64 bit immediates/operation as well.
  99    void and_imm(Register r, long mask, Register tmp = Z_R0, bool wide = false);
 100 
 101   // 1's complement, 32bit or 64bit. Optimized to exploit distinct operands facility.
 102   // Note: The condition code is neither preserved nor correctly set by this code!!!
 103   // Note: (wide == false) does not protect the high order half of the target register
 104   // from alternation. It only serves as optimization hint for 32-bit results.
 105   void not_(Register r1, Register r2 = noreg, bool wide = false);  // r1 = ~r2
 106 
 107   // Expanded support of all "rotate_then_<logicalOP>" instructions.
 108   //
 109   // Generalize and centralize rotate_then_<logicalOP> emitter.
 110   // Functional description. For details, see Principles of Operation, Chapter 7, "Rotate Then Insert..."
 111   //  - Bits  in a register are numbered left (most significant) to right (least significant), i.e. [0..63].
 112   //  - Bytes in a register are numbered left (most significant) to right (least significant), i.e. [0..7].
 113   //  - Register src is rotated to the left by (nRotate&0x3f) positions.
 114   //  - Negative values for nRotate result in a rotation to the right by abs(nRotate) positions.
 115   //  - The bits in positions [lBitPos..rBitPos] of the _ROTATED_ src operand take part in the
 116   //    logical operation performed on the contents (in those positions) of the dst operand.
 117   //  - The logical operation that is performed on the dst operand is one of
 118   //     o insert the selected bits (replacing the original contents of those bit positions)
 119   //     o and the selected bits with the corresponding bits of the dst operand
 120   //     o or  the selected bits with the corresponding bits of the dst operand
 121   //     o xor the selected bits with the corresponding bits of the dst operand
 122   //  - For clear_dst == true, the destination register is cleared before the bits are inserted.
 123   //    For clear_dst == false, only the bit positions that get data inserted from src
 124   //    are changed. All other bit positions remain unchanged.
 125   //  - For test_only == true,  the result of the logicalOP is only used to set the condition code, dst remains unchanged.
 126   //    For test_only == false, the result of the logicalOP replaces the selected bits of dst.
 127   //  - src32bit and dst32bit indicate the respective register is used as 32bit value only.
 128   //    Knowledge can simplify code generation.
 129   //
 130   // Here is an important performance note, valid for all <logicalOP>s except "insert":
 131   //   Due to the too complex nature of the operation, it cannot be done in a single cycle.
 132   //   Timing constraints require the instructions to be cracked into two micro-ops, taking
 133   //   one or two cycles each to execute. In some cases, an additional pipeline bubble might get added.
 134   //   Macroscopically, that makes up for a three- or four-cycle instruction where you would
 135   //   expect just a single cycle.
 136   //   It is thus not beneficial from a performance point of view to exploit those instructions.
 137   //   Other reasons (code compactness, register pressure, ...) might outweigh this penalty.
 138   //
 139   unsigned long create_mask(int lBitPos, int rBitPos);
 140   void rotate_then_mask(Register dst, Register src, int lBitPos, int rBitPos,
 141                         int nRotate, bool src32bit, bool dst32bit, bool oneBits);
 142   void rotate_then_insert(Register dst, Register src, int lBitPos, int rBitPos, int nRotate,
 143                           bool clear_dst);
 144   void rotate_then_and(Register dst, Register src, int lBitPos, int rBitPos, int nRotate,
 145                        bool test_only);
 146   void rotate_then_or(Register dst, Register src, int lBitPos, int rBitPos, int nRotate,
 147                       bool test_onlyt);
 148   void rotate_then_xor(Register dst, Register src, int lBitPos, int rBitPos, int nRotate,
 149                        bool test_only);
 150 
 151   void add64(Register r1, RegisterOrConstant inc);
 152 
 153   // Helper function to multiply the 64bit contents of a register by a 16bit constant.
 154   // The optimization tries to avoid the mghi instruction, since it uses the FPU for
 155   // calculation and is thus rather slow.
 156   //
 157   // There is no handling for special cases, e.g. cval==0 or cval==1.
 158   //
 159   // Returns len of generated code block.
 160   unsigned int mul_reg64_const16(Register rval, Register work, int cval);
 161 
 162   // Generic operation r1 := r2 + imm.
 163   void add2reg   (Register r1, int64_t imm, Register r2 = noreg);
 164   void add2reg_32(Register r1, int64_t imm, Register r2 = noreg);
 165 
 166   // Generic operation r := b + x + d.
 167   void add2reg_with_index(Register r, int64_t d, Register x, Register b = noreg);
 168 
 169   // Add2mem* methods for direct memory increment.
 170   void add2mem_32(const Address &a, int64_t imm, Register tmp);
 171   void add2mem_64(const Address &a, int64_t imm, Register tmp);
 172 
 173   // *((int8_t*)(dst)) |= imm8
 174   inline void or2mem_8(Address& dst, int64_t imm8);
 175 
 176   // Load values by size and signedness.
 177   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 178   void store_sized_value(Register src, Address dst, size_t size_in_bytes);
 179 
 180   // Load values with large offsets to base address.
 181  private:
 182   int  split_largeoffset(int64_t si20_offset, Register tmp, bool fixed_codelen, bool accumulate);
 183  public:
 184   void load_long_largeoffset(Register t, int64_t si20, Register a, Register tmp);
 185   void load_float_largeoffset(FloatRegister t, int64_t si20, Register a, Register tmp);
 186   void load_double_largeoffset(FloatRegister t, int64_t si20, Register a, Register tmp);
 187 
 188  private:
 189   long toc_distance();
 190  public:
 191   void load_toc(Register Rtoc);
 192   void load_long_pcrelative(Register Rdst, address dataLocation);
 193   static int load_long_pcrelative_size() { return 6; }
 194   void load_addr_pcrelative(Register Rdst, address dataLocation);
 195   static int load_addr_pcrel_size() { return 6; } // Just a LARL.
 196 
 197   // Load a value from memory and test (set CC).
 198   void load_and_test_byte    (Register dst, const Address &a);
 199   void load_and_test_short   (Register dst, const Address &a);
 200   void load_and_test_int     (Register dst, const Address &a);
 201   void load_and_test_int2long(Register dst, const Address &a);
 202   void load_and_test_long    (Register dst, const Address &a);
 203 
 204   // Test a bit in memory. Result is reflected in CC.
 205   void testbit(const Address &a, unsigned int bit);
 206   void testbit_ushort(const Address &a, unsigned int bit);
 207   // Test a bit in a register. Result is reflected in CC.
 208   void testbit(Register r, unsigned int bitPos);
 209 
 210   void prefetch_read(Address a);
 211   void prefetch_update(Address a);
 212 
 213   // Clear a register, i.e. load const zero into reg. Return len (in bytes) of
 214   // generated instruction(s).
 215   //   whole_reg: Clear 64 bits if true, 32 bits otherwise.
 216   //   set_cc: Use instruction that sets the condition code, if true.
 217   int clear_reg(Register r, bool whole_reg = true, bool set_cc = true);
 218 
 219 #ifdef ASSERT
 220   int preset_reg(Register r, unsigned long pattern, int pattern_len);
 221 #endif
 222 
 223   // Clear (store zeros) a small piece of memory.
 224   // CAUTION: Do not use this for atomic memory clearing. Use store_const() instead.
 225   //   addr: Address descriptor of memory to clear.
 226   //         Index register will not be used!
 227   //   size: Number of bytes to clear.
 228   void clear_mem(const Address& addr, unsigned size);
 229 
 230   // Move immediate values to memory. Currently supports 32 and 64 bit stores,
 231   // but may be extended to 16 bit store operation, if needed.
 232   // For details, see implementation in *.cpp file.
 233          int store_const(const Address &dest, long imm,
 234                          unsigned int lm, unsigned int lc,
 235                          Register scratch = Z_R0);
 236   inline int store_const(const Address &dest, long imm,
 237                          Register scratch = Z_R0, bool is_long = true);
 238 
 239   // Move/initialize arbitrarily large memory area. No check for destructive overlap.
 240   // Being interruptible, these instructions need a retry-loop.
 241   void move_long_ext(Register dst, Register src, unsigned int pad);
 242 
 243   void compare_long_ext(Register left, Register right, unsigned int pad);
 244   void compare_long_uni(Register left, Register right, unsigned int pad);
 245 
 246   void search_string(Register end, Register start);
 247   void search_string_uni(Register end, Register start);
 248 
 249   // Translate instructions
 250   // Being interruptible, these instructions need a retry-loop.
 251   void translate_oo(Register dst, Register src, uint mask);
 252   void translate_ot(Register dst, Register src, uint mask);
 253   void translate_to(Register dst, Register src, uint mask);
 254   void translate_tt(Register dst, Register src, uint mask);
 255 
 256   // Crypto instructions.
 257   // Being interruptible, these instructions need a retry-loop.
 258   void cksm(Register crcBuff, Register srcBuff);
 259   void km(   Register dstBuff, Register srcBuff);
 260   void kmc(  Register dstBuff, Register srcBuff);
 261   void kmctr(Register dstBuff, Register ctrBuff, Register srcBuff);
 262   void kimd(Register srcBuff);
 263   void klmd(Register srcBuff);
 264   void kmac(Register srcBuff);
 265 
 266   // nop padding
 267   void align(int modulus);
 268   void align(int modulus, int target);
 269   void align_address(int modulus);
 270 
 271   //
 272   // Constants, loading constants, TOC support
 273   //
 274 
 275   // Load generic address: d <- base(a) + index(a) + disp(a).
 276   inline void load_address(Register d, const Address &a);
 277   // Load absolute address (and try to optimize).
 278   void load_absolute_address(Register d, address addr);
 279 
 280   // Address of Z_ARG1 and argument_offset.
 281   // If temp_reg == arg_slot, arg_slot will be overwritten.
 282   Address argument_address(RegisterOrConstant arg_slot,
 283                            Register temp_reg = noreg,
 284                            int64_t extra_slot_offset = 0);
 285 
 286   // Load a narrow ptr constant (oop or klass ptr).
 287   void load_narrow_oop( Register t, narrowOop a);
 288   void load_narrow_klass(Register t, Klass* k);
 289 
 290   static bool is_load_const_32to64(address pos);
 291   static bool is_load_narrow_oop(address pos)   { return is_load_const_32to64(pos); }
 292   static bool is_load_narrow_klass(address pos) { return is_load_const_32to64(pos); }
 293 
 294   static int  load_const_32to64_size()          { return 6; }
 295   static bool load_narrow_oop_size()            { return load_const_32to64_size(); }
 296   static bool load_narrow_klass_size()          { return load_const_32to64_size(); }
 297 
 298   static int  patch_load_const_32to64(address pos, int64_t a);
 299   static int  patch_load_narrow_oop(address pos, oop o);
 300   static int  patch_load_narrow_klass(address pos, Klass* k);
 301 
 302   // cOops. CLFI exploit.
 303   void compare_immediate_narrow_oop(Register oop1, narrowOop oop2);
 304   void compare_immediate_narrow_klass(Register op1, Klass* op2);
 305   static bool is_compare_immediate32(address pos);
 306   static bool is_compare_immediate_narrow_oop(address pos);
 307   static bool is_compare_immediate_narrow_klass(address pos);
 308   static int  compare_immediate_narrow_size()       { return 6; }
 309   static int  compare_immediate_narrow_oop_size()   { return compare_immediate_narrow_size(); }
 310   static int  compare_immediate_narrow_klass_size() { return compare_immediate_narrow_size(); }
 311   static int  patch_compare_immediate_32(address pos, int64_t a);
 312   static int  patch_compare_immediate_narrow_oop(address pos, oop o);
 313   static int  patch_compare_immediate_narrow_klass(address pos, Klass* k);
 314 
 315   // Load a 32bit constant into a 64bit register.
 316   void load_const_32to64(Register t, int64_t x, bool sign_extend=true);
 317   // Load a 64 bit constant.
 318          void load_const(Register t, long a);
 319   inline void load_const(Register t, void* a);
 320   inline void load_const(Register t, Label& L);
 321   inline void load_const(Register t, const AddressLiteral& a);
 322   // Get the 64 bit constant from a `load_const' sequence.
 323   static long get_const(address load_const);
 324   // Patch the 64 bit constant of a `load_const' sequence. This is a low level
 325   // procedure. It neither flushes the instruction cache nor is it atomic.
 326   static void patch_const(address load_const, long x);
 327   static int load_const_size() { return 12; }
 328 
 329   // Turn a char into boolean. NOTE: destroys r.
 330   void c2bool(Register r, Register t = Z_R0);
 331 
 332   // Optimized version of load_const for constants that do not need to be
 333   // loaded by a sequence of instructions of fixed length and that do not
 334   // need to be patched.
 335   int load_const_optimized_rtn_len(Register t, long x, bool emit);
 336   inline void load_const_optimized(Register t, long x);
 337   inline void load_const_optimized(Register t, void* a);
 338   inline void load_const_optimized(Register t, Label& L);
 339   inline void load_const_optimized(Register t, const AddressLiteral& a);
 340 
 341  public:
 342 
 343   //----------------------------------------------------------
 344   //            oops in code             -------------
 345   //  including compressed oops support  -------------
 346   //----------------------------------------------------------
 347 
 348   // Metadata in code that we have to keep track of.
 349   AddressLiteral allocate_metadata_address(Metadata* obj); // allocate_index
 350   AddressLiteral constant_metadata_address(Metadata* obj); // find_index
 351 
 352   // allocate_index
 353   AddressLiteral allocate_oop_address(jobject obj);
 354   // find_index
 355   AddressLiteral constant_oop_address(jobject obj);
 356   // Uses allocate_oop_address.
 357   inline void set_oop         (jobject obj, Register d);
 358   // Uses constant_oop_address.
 359   inline void set_oop_constant(jobject obj, Register d);
 360   // Uses constant_metadata_address.
 361   inline bool set_metadata_constant(Metadata* md, Register d);
 362 
 363   //
 364   // branch, jump
 365   //
 366 
 367   // Use one generic function for all branch patches.
 368   static unsigned long patched_branch(address dest_pos, unsigned long inst, address inst_pos);
 369 
 370   void pd_patch_instruction(address branch, address target, const char* file, int line);
 371 
 372   // Extract relative address from "relative" instructions.
 373   static long get_pcrel_offset(unsigned long inst);
 374   static long get_pcrel_offset(address pc);
 375   static address get_target_addr_pcrel(address pc);
 376 
 377   static inline bool is_call_pcrelative_short(unsigned long inst);
 378   static inline bool is_call_pcrelative_long(unsigned long inst);
 379   static inline bool is_branch_pcrelative_short(unsigned long inst);
 380   static inline bool is_branch_pcrelative_long(unsigned long inst);
 381   static inline bool is_compareandbranch_pcrelative_short(unsigned long inst);
 382   static inline bool is_branchoncount_pcrelative_short(unsigned long inst);
 383   static inline bool is_branchonindex32_pcrelative_short(unsigned long inst);
 384   static inline bool is_branchonindex64_pcrelative_short(unsigned long inst);
 385   static inline bool is_branchonindex_pcrelative_short(unsigned long inst);
 386   static inline bool is_branch_pcrelative16(unsigned long inst);
 387   static inline bool is_branch_pcrelative32(unsigned long inst);
 388   static inline bool is_branch_pcrelative(unsigned long inst);
 389   static inline bool is_load_pcrelative_long(unsigned long inst);
 390   static inline bool is_misc_pcrelative_long(unsigned long inst);
 391   static inline bool is_pcrelative_short(unsigned long inst);
 392   static inline bool is_pcrelative_long(unsigned long inst);
 393   // PCrelative TOC access. Variants with address argument.
 394   static inline bool is_load_pcrelative_long(address iLoc);
 395   static inline bool is_pcrelative_short(address iLoc);
 396   static inline bool is_pcrelative_long(address iLoc);
 397 
 398   static inline bool is_pcrelative_instruction(address iloc);
 399   static inline bool is_load_addr_pcrel(address a);
 400 
 401   static void patch_target_addr_pcrel(address pc, address con);
 402   static void patch_addr_pcrel(address pc, address con) {
 403     patch_target_addr_pcrel(pc, con); // Just delegate. This is only for nativeInst_s390.cpp.
 404   }
 405 
 406   //---------------------------------------------------------
 407   //  Some macros for more comfortable assembler programming.
 408   //---------------------------------------------------------
 409 
 410   // NOTE: pass NearLabel T to signal that the branch target T will be bound to a near address.
 411 
 412   void compare32_and_branch(Register r1, RegisterOrConstant x2, branch_condition cond, Label& target);
 413   void compareU32_and_branch(Register r1, RegisterOrConstant x2, branch_condition cond, Label& target);
 414   void compare64_and_branch(Register r1, RegisterOrConstant x2, branch_condition cond, Label& target);
 415   void compareU64_and_branch(Register r1, RegisterOrConstant x2, branch_condition cond, Label& target);
 416 
 417   void branch_optimized(Assembler::branch_condition cond, address branch_target);
 418   void branch_optimized(Assembler::branch_condition cond, Label&  branch_target);
 419   void compare_and_branch_optimized(Register r1,
 420                                     Register r2,
 421                                     Assembler::branch_condition cond,
 422                                     address  branch_addr,
 423                                     bool     len64,
 424                                     bool     has_sign);
 425   void compare_and_branch_optimized(Register r1,
 426                                     jlong    x2,
 427                                     Assembler::branch_condition cond,
 428                                     Label&   branch_target,
 429                                     bool     len64,
 430                                     bool     has_sign);
 431   void compare_and_branch_optimized(Register r1,
 432                                     Register r2,
 433                                     Assembler::branch_condition cond,
 434                                     Label&   branch_target,
 435                                     bool     len64,
 436                                     bool     has_sign);
 437 
 438   //
 439   // Support for frame handling
 440   //
 441   // Specify the register that should be stored as the return pc in the
 442   // current frame (default is R14).
 443   inline void save_return_pc(Register pc = Z_R14);
 444   inline void restore_return_pc();
 445 
 446   // Get current PC.
 447   address get_PC(Register result);
 448 
 449   // Get current PC + offset. Offset given in bytes, must be even!
 450   address get_PC(Register result, int64_t offset);
 451 
 452   // Get size of instruction at pc (which must point to valid code).
 453   void instr_size(Register size, Register pc);
 454 
 455   // Accessing, and in particular modifying, a stack location is only safe if
 456   // the stack pointer (Z_SP) is set such that the accessed stack location is
 457   // in the reserved range.
 458   //
 459   // From a performance point of view, it is desirable not to change the SP
 460   // first and then immediately use it to access the freshly reserved space.
 461   // That opens a small gap, though. If, just after storing some value (the
 462   // frame pointer) into the to-be-reserved space, an interrupt is caught,
 463   // the handler might use the space beyond Z_SP for it's own purpose.
 464   // If that happens, the stored value might get altered.
 465 
 466   // Resize current frame either relatively wrt to current SP or absolute.
 467   void resize_frame_sub(Register offset, Register fp, bool load_fp=true);
 468   void resize_frame_abs_with_offset(Register newSP, Register fp, int offset, bool load_fp);
 469   void resize_frame_absolute(Register addr, Register fp, bool load_fp);
 470   void resize_frame(RegisterOrConstant offset, Register fp, bool load_fp=true);
 471 
 472   // Push a frame of size bytes, if copy_sp is false, old_sp must already
 473   // contain a copy of Z_SP.
 474   void push_frame(Register bytes, Register old_sp, bool copy_sp = true, bool bytes_with_inverted_sign = false);
 475 
 476   // Push a frame of size `bytes'. no abi space provided.
 477   // Don't rely on register locking, instead pass a scratch register
 478   // (Z_R0 by default).
 479   // CAUTION! passing registers >= Z_R2 may produce bad results on
 480   // old CPUs!
 481   unsigned int push_frame(unsigned int bytes, Register scratch = Z_R0);
 482 
 483   // Push a frame of size `bytes' with abi160 on top.
 484   unsigned int push_frame_abi160(unsigned int bytes);
 485 
 486   // Pop current C frame.
 487   void pop_frame();
 488   // Pop current C frame and restore return PC register (Z_R14).
 489   void pop_frame_restore_retPC(int frame_size_in_bytes);
 490 
 491 #ifdef ASSERT
 492   void clobber_volatile_registers(Register excluded_register[], int n);
 493 #endif // ASSERT
 494 
 495   //
 496   // Calls
 497   //
 498 
 499  private:
 500   address _last_calls_return_pc;
 501 
 502  public:
 503   // Support for VM calls. This is the base routine called by the
 504   // different versions of call_VM_leaf. The interpreter may customize
 505   // this version by overriding it for its purposes (e.g., to
 506   // save/restore additional registers when doing a VM call).
 507   void call_VM_leaf_base(address entry_point);
 508   void call_VM_leaf_base(address entry_point, bool allow_relocation);
 509 
 510   // It is imperative that all calls into the VM are handled via the
 511   // call_VM macros. They make sure that the stack linkage is setup
 512   // correctly. Call_VM's correspond to ENTRY/ENTRY_X entry points
 513   // while call_VM_leaf's correspond to LEAF entry points.
 514   //
 515   // This is the base routine called by the different versions of
 516   // call_VM. The interpreter may customize this version by overriding
 517   // it for its purposes (e.g., to save/restore additional registers
 518   // when doing a VM call).
 519 
 520   // If no last_java_sp is specified (noreg) then SP will be used instead.
 521 
 522   virtual void call_VM_base(
 523     Register        oop_result,        // Where an oop-result ends up if any; use noreg otherwise.
 524     Register        last_java_sp,      // To set up last_Java_frame in stubs; use noreg otherwise.
 525     address         entry_point,       // The entry point.
 526     bool            check_exception);  // Flag which indicates if exception should be checked.
 527   virtual void call_VM_base(
 528     Register        oop_result,       // Where an oop-result ends up if any; use noreg otherwise.
 529     Register        last_java_sp,     // To set up last_Java_frame in stubs; use noreg otherwise.
 530     address         entry_point,      // The entry point.
 531     bool            allow_relocation, // Flag to request generation of relocatable code.
 532     bool            check_exception); // Flag which indicates if exception should be checked.
 533 
 534   // Call into the VM.
 535   // Passes the thread pointer (in Z_ARG1) as a prepended argument.
 536   // Makes sure oop return values are visible to the GC.
 537   void call_VM(Register oop_result, address entry_point, bool check_exceptions = true);
 538   void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
 539   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 540   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2,
 541                Register arg_3, bool check_exceptions = true);
 542 
 543   void call_VM_static(Register oop_result, address entry_point, bool check_exceptions = true);
 544   void call_VM_static(Register oop_result, address entry_point, Register arg_1, Register arg_2,
 545                       Register arg_3, bool check_exceptions = true);
 546 
 547   // Overloaded with last_java_sp.
 548   void call_VM(Register oop_result, Register last_java_sp, address entry_point, bool check_exceptions = true);
 549   void call_VM(Register oop_result, Register last_java_sp, address entry_point,
 550                Register arg_1, bool check_exceptions = true);
 551   void call_VM(Register oop_result, Register last_java_sp, address entry_point,
 552                Register arg_1, Register arg_2, bool check_exceptions = true);
 553   void call_VM(Register oop_result, Register last_java_sp, address entry_point,
 554                Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
 555 
 556   void call_VM_leaf(address entry_point);
 557   void call_VM_leaf(address entry_point, Register arg_1);
 558   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 559   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 560 
 561   // Really static VM leaf call (never patched).
 562   void call_VM_leaf_static(address entry_point);
 563   void call_VM_leaf_static(address entry_point, Register arg_1);
 564   void call_VM_leaf_static(address entry_point, Register arg_1, Register arg_2);
 565   void call_VM_leaf_static(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 566 
 567   // Call a C function via its function entry. Updates and returns _last_calls_return_pc.
 568   inline address call(Register function_entry);
 569   inline address call_c(Register function_entry);
 570          address call_c(address function_entry);
 571   // Variant for really static (non-relocatable) calls which are never patched.
 572          address call_c_static(address function_entry);
 573   // TOC or pc-relative call + emits a runtime_call relocation.
 574          address call_c_opt(address function_entry);
 575 
 576   inline address call_stub(Register function_entry);
 577   inline address call_stub(address  function_entry);
 578 
 579   // Get the pc where the last call will return to. Returns _last_calls_return_pc.
 580   inline address last_calls_return_pc();
 581 
 582   static int ic_check_size();
 583   int ic_check(int end_alignment);
 584 
 585  private:
 586   static bool is_call_far_patchable_variant0_at(address instruction_addr); // Dynamic TOC: load target addr from CP and call.
 587   static bool is_call_far_patchable_variant2_at(address instruction_addr); // PC-relative call, prefixed with NOPs.
 588 
 589 
 590  public:
 591   bool           call_far_patchable(address target, int64_t toc_offset);
 592   static bool    is_call_far_patchable_at(address inst_start);             // All supported forms of patchable calls.
 593   static bool    is_call_far_patchable_pcrelative_at(address inst_start);  // Pc-relative call with leading nops.
 594   static bool    is_call_far_pcrelative(address instruction_addr);         // Pure far pc-relative call, with one leading size adjustment nop.
 595   static void    set_dest_of_call_far_patchable_at(address inst_start, address target, int64_t toc_offset);
 596   static address get_dest_of_call_far_patchable_at(address inst_start, address toc_start);
 597 
 598   void align_call_far_patchable(address pc);
 599 
 600   // PCrelative TOC access.
 601 
 602   // This value is independent of code position - constant for the lifetime of the VM.
 603   static int call_far_patchable_size() {
 604     return load_const_from_toc_size() + call_byregister_size();
 605   }
 606 
 607   static int call_far_patchable_ret_addr_offset() { return call_far_patchable_size(); }
 608 
 609   static bool call_far_patchable_requires_alignment_nop(address pc) {
 610     int size = call_far_patchable_size();
 611     return ((intptr_t)(pc + size) & 0x03L) != 0;
 612   }
 613 
 614   // END OF PCrelative TOC access.
 615 
 616   static int jump_byregister_size()          { return 2; }
 617   static int jump_pcrelative_size()          { return 4; }
 618   static int jump_far_pcrelative_size()      { return 6; }
 619   static int call_byregister_size()          { return 2; }
 620   static int call_pcrelative_size()          { return 4; }
 621   static int call_far_pcrelative_size()      { return 2 + 6; } // Prepend each BRASL with a nop.
 622   static int call_far_pcrelative_size_raw()  { return 6; }     // Prepend each BRASL with a nop.
 623 
 624   //
 625   // Java utilities
 626   //
 627 
 628   // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 629   // The implementation is only non-empty for the InterpreterMacroAssembler,
 630   // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 631   virtual void check_and_handle_popframe(Register java_thread);
 632   virtual void check_and_handle_earlyret(Register java_thread);
 633 
 634   // Polling page support.
 635   enum poll_mask {
 636     mask_stackbang = 0xde, // 222 (dec)
 637     mask_safepoint = 0x6f, // 111 (dec)
 638     mask_profiling = 0xba  // 186 (dec)
 639   };
 640 
 641   // Read from the polling page.
 642   void load_from_polling_page(Register polling_page_address, int64_t offset = 0);
 643 
 644   // Check if given instruction is a read from the polling page
 645   // as emitted by load_from_polling_page.
 646   static bool is_load_from_polling_page(address instr_loc);
 647   // Extract poll address from instruction and ucontext.
 648   static address get_poll_address(address instr_loc, void* ucontext);
 649   // Extract poll register from instruction.
 650   static uint get_poll_register(address instr_loc);
 651 
 652   // Check if safepoint requested and if so branch
 653   void safepoint_poll(Label& slow_path, Register temp_reg);
 654 
 655   // Stack overflow checking
 656   void bang_stack_with_offset(int offset);
 657 
 658   // Check for reserved stack access in method being exited. If the reserved
 659   // stack area was accessed, protect it again and throw StackOverflowError.
 660   // Uses Z_R1.
 661   void reserved_stack_check(Register return_pc);
 662 
 663   // Atomics
 664   // -- none?
 665 
 666   void tlab_allocate(Register obj,                // Result: pointer to object after successful allocation
 667                      Register var_size_in_bytes,  // Object size in bytes if unknown at compile time; invalid otherwise.
 668                      int      con_size_in_bytes,  // Object size in bytes if   known at compile time.
 669                      Register t1,                 // temp register
 670                      Label&   slow_case);         // Continuation point if fast allocation fails.
 671 
 672   // Emitter for interface method lookup.
 673   //   input: recv_klass, intf_klass, itable_index
 674   //   output: method_result
 675   //   kills: itable_index, temp1_reg, Z_R0, Z_R1
 676   void lookup_interface_method(Register           recv_klass,
 677                                Register           intf_klass,
 678                                RegisterOrConstant itable_index,
 679                                Register           method_result,
 680                                Register           temp1_reg,
 681                                Label&             no_such_interface,
 682                                bool               return_method = true);
 683 
 684   // virtual method calling
 685   void lookup_virtual_method(Register             recv_klass,
 686                              RegisterOrConstant   vtable_index,
 687                              Register             method_result);
 688 
 689   // Factor out code to call ic_miss_handler.
 690   unsigned int call_ic_miss_handler(Label& ICM, int trapMarker, int requiredSize, Register scratch);
 691   void nmethod_UEP(Label& ic_miss);
 692 
 693   // Emitters for "partial subtype" checks.
 694 
 695   // Test sub_klass against super_klass, with fast and slow paths.
 696 
 697   // The fast path produces a tri-state answer: yes / no / maybe-slow.
 698   // One of the three labels can be null, meaning take the fall-through.
 699   // If super_check_offset is -1, the value is loaded up from super_klass.
 700   // No registers are killed, except temp_reg and temp2_reg.
 701   // If super_check_offset is not -1, temp1_reg is not used and can be noreg.
 702   void check_klass_subtype_fast_path(Register sub_klass,
 703                                      Register super_klass,
 704                                      Register temp1_reg,
 705                                      Label*   L_success,
 706                                      Label*   L_failure,
 707                                      Label*   L_slow_path,
 708                                      Register super_check_offset = noreg);
 709 
 710   // The rest of the type check; must be wired to a corresponding fast path.
 711   // It does not repeat the fast path logic, so don't use it standalone.
 712   // The temp_reg can be noreg, if no temps are available.
 713   // It can also be sub_klass or super_klass, meaning it's OK to kill that one.
 714   // Updates the sub's secondary super cache as necessary.
 715   void check_klass_subtype_slow_path(Register Rsubklass,
 716                                      Register Rsuperklas,
 717                                      Register Rarray_ptr, // tmp
 718                                      Register Rlength,    // tmp
 719                                      Label* L_success,
 720                                      Label* L_failure,
 721                                      bool set_cond_codes = false);
 722 
 723   void check_klass_subtype_slow_path_linear(Register sub_klass,
 724                                             Register super_klass,
 725                                             Register temp_reg,
 726                                             Register temp2_reg,
 727                                             Label* L_success,
 728                                             Label* L_failure,
 729                                             bool set_cond_codes = false);
 730 
 731   void check_klass_subtype_slow_path_table(Register sub_klass,
 732                                            Register super_klass,
 733                                            Register temp_reg,
 734                                            Register temp2_reg,
 735                                            Register temp3_reg,
 736                                            Register temp4_reg,
 737                                            Register result_reg,
 738                                            Label* L_success,
 739                                            Label* L_failure,
 740                                            bool set_cond_codes = false);
 741 
 742   // If r is valid, return r.
 743   // If r is invalid, remove a register r2 from available_regs, add r2
 744   // to regs_to_push, then return r2.
 745   Register allocate_if_noreg(const Register r,
 746                              RegSetIterator<Register> &available_regs,
 747                              RegSet &regs_to_push);
 748 
 749   void repne_scan(Register r_addr, Register r_value, Register r_count, Register r_scratch);
 750 
 751   // Secondary subtype checking
 752   void lookup_secondary_supers_table_var(Register sub_klass,
 753                                          Register r_super_klass,
 754                                          Register temp1,
 755                                          Register temp2,
 756                                          Register temp3,
 757                                          Register temp4,
 758                                          Register result);
 759 
 760   void lookup_secondary_supers_table_const(Register r_sub_klass,
 761                                            Register r_super_klass,
 762                                            Register r_temp1,
 763                                            Register r_temp2,
 764                                            Register r_temp3,
 765                                            Register r_temp4,
 766                                            Register r_result,
 767                                            u1 super_klass_slot);
 768 
 769   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
 770                                                Register r_array_base,
 771                                                Register r_array_index,
 772                                                Register r_bitmap,
 773                                                Register r_temp,
 774                                                Register r_result,
 775                                                bool is_stub);
 776 
 777   void verify_secondary_supers_table(Register r_sub_klass,
 778                                      Register r_super_klass,
 779                                      Register r_result /* expected */,
 780                                      Register r_temp1,
 781                                      Register r_temp2,
 782                                      Register r_temp3);
 783 
 784   // Simplified, combined version, good for typical uses.
 785   // Falls through on failure.
 786   void check_klass_subtype(Register sub_klass,
 787                            Register super_klass,
 788                            Register temp1_reg,
 789                            Register temp2_reg,
 790                            Label&   L_success);
 791 
 792   void clinit_barrier(Register klass,
 793                       Register thread,
 794                       Label* L_fast_path = nullptr,
 795                       Label* L_slow_path = nullptr);
 796 
 797   // Increment a counter at counter_address when the eq condition code is set.
 798   // Kills registers tmp1_reg and tmp2_reg and preserves the condition code.
 799   void increment_counter_eq(address counter_address, Register tmp1_reg, Register tmp2_reg);
 800 
 801   void fast_lock(Register basic_lock, Register obj, Register tmp1, Register tmp2, Label& slow);
 802   void fast_unlock(Register obj, Register tmp1, Register tmp2, Label& slow);
 803   void compiler_fast_lock_object(Register obj, Register box, Register tmp1, Register tmp2);
 804   void compiler_fast_unlock_object(Register obj, Register box, Register tmp1, Register tmp2);
 805 
 806   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 807   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 808 
 809   // Support for last Java frame (but use call_VM instead where possible).
 810  private:
 811   void set_last_Java_frame(Register last_Java_sp, Register last_Java_pc, bool allow_relocation);
 812   void reset_last_Java_frame(bool allow_relocation);
 813   void set_top_ijava_frame_at_SP_as_last_Java_frame(Register sp, Register tmp1, bool allow_relocation);
 814  public:
 815   inline void set_last_Java_frame(Register last_java_sp, Register last_Java_pc);
 816   inline void set_last_Java_frame_static(Register last_java_sp, Register last_Java_pc);
 817   inline void reset_last_Java_frame(void);
 818   inline void reset_last_Java_frame_static(void);
 819   inline void set_top_ijava_frame_at_SP_as_last_Java_frame(Register sp, Register tmp1);
 820   inline void set_top_ijava_frame_at_SP_as_last_Java_frame_static(Register sp, Register tmp1);
 821 
 822   void set_thread_state(JavaThreadState new_state);
 823 
 824   // Read vm result from thread.
 825   void get_vm_result_oop  (Register oop_result);
 826   void get_vm_result_metadata(Register result);
 827 
 828   // Vm result is currently getting hijacked to for oop preservation.
 829   void set_vm_result(Register oop_result);
 830 
 831   // Support for null-checks
 832   //
 833   // Generates code that causes a null OS exception if the content of reg is null.
 834   // If the accessed location is M[reg + offset] and the offset is known, provide the
 835   // offset. No explicit code generation is needed if the offset is within a certain
 836   // range (0 <= offset <= page_size).
 837   //
 838   // %%%%%% Currently not done for z/Architecture
 839 
 840   void null_check(Register reg, Register tmp = Z_R0, int64_t offset = -1);
 841   static bool needs_explicit_null_check(intptr_t offset);  // Implemented in shared file ?!
 842   static bool uses_implicit_null_check(void* address);
 843 
 844   // Klass oop manipulations if compressed.
 845   void encode_klass_not_null(Register dst, Register src = noreg);
 846   void decode_klass_not_null(Register dst, Register src);
 847   void decode_klass_not_null(Register dst);
 848   void load_klass(Register klass, Address mem);
 849   void load_klass(Register klass, Register src_oop);
 850   void store_klass(Register klass, Register dst_oop, Register ck = noreg); // Klass will get compressed if ck not provided.
 851   void store_klass_gap(Register s, Register dst_oop);
 852   void load_narrow_klass_compact(Register dst, Register src);
 853   // Compares the narrow Klass pointer of an object to a given narrow Klass
 854   void cmp_klass(Register klass, Register obj, Register tmp);
 855   // Compares the Klass pointer of two objects obj1 and obj2. Result is in the condition flags.
 856   // Uses tmp1 and tmp2 as temporary registers.
 857   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 858 
 859   // This function calculates the size of the code generated by
 860   //   decode_klass_not_null(register dst)
 861   // when Universe::heap() isn't null. Hence, if the instructions
 862   // it generates change, then this method needs to be updated.
 863   static int instr_size_for_decode_klass_not_null();
 864 
 865   void encode_heap_oop(Register oop);
 866   void encode_heap_oop_not_null(Register oop);
 867 
 868   static int get_oop_base_pow2_offset(uint64_t oop_base);
 869   int  get_oop_base(Register Rbase, uint64_t oop_base);
 870   int  get_oop_base_complement(Register Rbase, uint64_t oop_base);
 871   void compare_heap_oop(Register Rop1, Address mem, bool maybenull);
 872   void compare_klass_ptr(Register Rop1, int64_t disp, Register Rbase, bool maybenull);
 873 
 874   // Access heap oop, handle encoding and GC barriers.
 875   void access_store_at(BasicType type, DecoratorSet decorators,
 876                        const Address& addr, Register val,
 877                        Register tmp1, Register tmp2, Register tmp3);
 878   void access_load_at(BasicType type, DecoratorSet decorators,
 879                       const Address& addr, Register dst,
 880                       Register tmp1, Register tmp2, Label *is_null = nullptr);
 881 
 882  public:
 883   // tmp1 and tmp2 are used with decorators ON_PHANTOM_OOP_REF or ON_WEAK_OOP_REF.
 884   void load_heap_oop(Register dest, const Address &a,
 885                      Register tmp1, Register tmp2,
 886                      DecoratorSet decorators = 0, Label *is_null = nullptr);
 887   void store_heap_oop(Register Roop, const Address &a,
 888                       Register tmp1, Register tmp2, Register tmp3,
 889                       DecoratorSet decorators = 0);
 890 
 891   void oop_encoder(Register Rdst, Register Rsrc, bool maybenull,
 892                    Register Rbase = Z_R1, int pow2_offset = -1, bool only32bitValid = false);
 893   void oop_decoder(Register Rdst, Register Rsrc, bool maybenull,
 894                    Register Rbase = Z_R1, int pow2_offset = -1);
 895 
 896   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 897   void load_method_holder(Register holder, Register method);
 898 
 899   //--------------------------
 900   //---  Operations on arrays.
 901   //--------------------------
 902   unsigned int Clear_Array(Register cnt_arg, Register base_pointer_arg, Register odd_tmp_reg);
 903   unsigned int Clear_Array_Const(long cnt, Register base);
 904   unsigned int Clear_Array_Const_Big(long cnt, Register base_pointer_arg, Register odd_tmp_reg);
 905   unsigned int CopyRawMemory_AlignedDisjoint(Register src_reg, Register dst_reg,
 906                                              Register cnt_reg,
 907                                              Register tmp1_reg, Register tmp2_reg);
 908 
 909 
 910   // Emit an oop const to the constant pool and set a relocation info
 911   // with address current_pc. Return the TOC offset of the constant.
 912   int store_const_in_toc(AddressLiteral& val);
 913   int store_oop_in_toc(AddressLiteral& oop);
 914   // Emit an oop const to the constant pool via store_oop_in_toc, or
 915   // emit a scalar const to the constant pool via store_const_in_toc,
 916   // and load the constant into register dst.
 917   bool load_const_from_toc(Register dst, AddressLiteral& a, Register Rtoc = noreg);
 918   // Get CPU version dependent size of load_const sequence.
 919   // The returned value is valid only for code sequences
 920   // generated by load_const, not load_const_optimized.
 921   static int load_const_from_toc_size() {
 922     return load_long_pcrelative_size();
 923   }
 924   bool load_oop_from_toc(Register dst, AddressLiteral& a, Register Rtoc = noreg);
 925   static intptr_t get_const_from_toc(address pc);
 926   static void     set_const_in_toc(address pc, unsigned long new_data, CodeBlob *cb);
 927 
 928   // Dynamic TOC.
 929   static bool is_load_const(address a);
 930   static bool is_load_const_from_toc_pcrelative(address a);
 931   static bool is_load_const_from_toc(address a) { return is_load_const_from_toc_pcrelative(a); }
 932 
 933   // PCrelative TOC access.
 934   static bool is_call_byregister(address a) { return is_z_basr(*(short*)a); }
 935   static bool is_load_const_from_toc_call(address a);
 936   static bool is_load_const_call(address a);
 937   static int load_const_call_size() { return load_const_size() + call_byregister_size(); }
 938   static int load_const_from_toc_call_size() { return load_const_from_toc_size() + call_byregister_size(); }
 939   // Offset is +/- 2**32 -> use long.
 940   static long get_load_const_from_toc_offset(address a);
 941 
 942   // Bit operations for single register operands.
 943   inline void lshift(Register r, int places, bool doubl = true);   // <<
 944   inline void rshift(Register r, int places, bool doubl = true);   // >>
 945 
 946   //
 947   // Debugging
 948   //
 949 
 950   // Assert on CC (condition code in CPU state).
 951   void asm_assert(branch_condition cond, const char* msg, int id, bool is_static=true);
 952   void asm_assert(bool check_equal, const char* msg, int id);
 953 
 954  private:
 955   // Emit assertions.
 956   void asm_assert_mems_zero(bool check_equal, bool allow_relocation, int size, int64_t mem_offset,
 957                             Register mem_base, const char* msg, int id);
 958 
 959  public:
 960   inline void asm_assert_mem4_is_zero(int64_t mem_offset, Register mem_base, const char* msg, int id) {
 961     asm_assert_mems_zero(true,  true, 4, mem_offset, mem_base, msg, id);
 962   }
 963   inline void asm_assert_mem8_is_zero(int64_t mem_offset, Register mem_base, const char* msg, int id) {
 964     asm_assert_mems_zero(true,  true, 8, mem_offset, mem_base, msg, id);
 965   }
 966   inline void asm_assert_mem4_isnot_zero(int64_t mem_offset, Register mem_base, const char* msg, int id) {
 967     asm_assert_mems_zero(false, true, 4, mem_offset, mem_base, msg, id);
 968   }
 969   inline void asm_assert_mem8_isnot_zero(int64_t mem_offset, Register mem_base, const char* msg, int id) {
 970     asm_assert_mems_zero(false, true, 8, mem_offset, mem_base, msg, id);
 971   }
 972   inline void asm_assert_mem4_is_zero_static(int64_t mem_offset, Register mem_base, const char* msg, int id) {
 973     asm_assert_mems_zero(true,  false, 4, mem_offset, mem_base, msg, id);
 974   }
 975   inline void asm_assert_mem8_is_zero_static(int64_t mem_offset, Register mem_base, const char* msg, int id) {
 976     asm_assert_mems_zero(true,  false, 8, mem_offset, mem_base, msg, id);
 977   }
 978   inline void asm_assert_mem4_isnot_zero_static(int64_t mem_offset, Register mem_base, const char* msg, int id) {
 979     asm_assert_mems_zero(false, false, 4, mem_offset, mem_base, msg, id);
 980   }
 981   inline void asm_assert_mem8_isnot_zero_static(int64_t mem_offset, Register mem_base, const char* msg, int id) {
 982     asm_assert_mems_zero(false, false, 8, mem_offset, mem_base, msg, id);
 983   }
 984   void asm_assert_frame_size(Register expected_size, Register tmp, const char* msg, int id);
 985 
 986   // Save and restore functions: Exclude Z_R0.
 987   void save_volatile_regs(   Register dst, int offset, bool include_fp, bool include_flags);
 988   void restore_volatile_regs(Register src, int offset, bool include_fp, bool include_flags);
 989 
 990   // Only if +VerifyOops.
 991   // Kills Z_R0.
 992   void verify_oop(Register reg, const char* s = "broken oop");
 993   // Kills Z_R0, condition code.
 994   void verify_oop_addr(Address addr, const char* msg = "contains broken oop");
 995 
 996   // TODO: verify_method and klass metadata (compare against vptr?).
 997   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
 998   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line) {}
 999 
1000 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1001 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1002 
1003  private:
1004   // Generate printout in stop().
1005   static const char* stop_types[];
1006   enum {
1007     stop_stop               = 0,
1008     stop_untested           = 1,
1009     stop_unimplemented      = 2,
1010     stop_shouldnotreachhere = 3,
1011     stop_end                = 4
1012   };
1013   // Prints msg and stops execution.
1014   void    stop(int type, const char* msg, int id = 0);
1015   address stop_chain(address reentry, int type, const char* msg, int id, bool allow_relocation); // Non-relocateable code only!!
1016   void    stop_static(int type, const char* msg, int id);                                        // Non-relocateable code only!!
1017 
1018  public:
1019 
1020   // Prints msg and stops.
1021   address stop_chain(      address reentry, const char* msg = "", int id = 0) { return stop_chain(reentry, stop_stop, msg, id, true); }
1022   address stop_chain_static(address reentry, const char* msg = "", int id = 0) { return stop_chain(reentry, stop_stop, msg, id, false); }
1023   void stop_static  (const char* msg = "", int id = 0) { stop_static(stop_stop,   msg, id); }
1024   void stop         (const char* msg = "", int id = 0) { stop(stop_stop,          msg, id); }
1025   void untested     (const char* msg = "", int id = 0) { stop(stop_untested,      msg, id); }
1026   void unimplemented(const char* msg = "", int id = 0) { stop(stop_unimplemented, msg, id); }
1027   void should_not_reach_here(const char* msg = "", int id = -1) { stop(stop_shouldnotreachhere, msg, id); }
1028 
1029   // Factor out part of stop into subroutine to save space.
1030   void stop_subroutine();
1031 
1032   // Prints msg, but don't stop.
1033   void warn(const char* msg);
1034 
1035   //-----------------------------
1036   //---  basic block tracing code
1037   //-----------------------------
1038   void trace_basic_block(uint i);
1039   void init_basic_block_trace();
1040   // Number of bytes a basic block gets larger due to the tracing code macro (worst case).
1041   // Currently, worst case is 48 bytes. 64 puts us securely on the safe side.
1042   static int basic_blck_trace_blk_size_incr() { return 64; }
1043 
1044   // Write pattern 0x0101010101010101 in region [low-before, high+after].
1045   // Low and high may be the same registers. Before and after are
1046   // the numbers of 8-byte words.
1047   void zap_from_to(Register low, Register high, Register tmp1 = Z_R0, Register tmp2 = Z_R1,
1048                    int before = 0, int after = 0) PRODUCT_RETURN;
1049 
1050   // Emitters for CRC32 calculation.
1051   // A note on invertCRC:
1052   //   Unfortunately, internal representation of crc differs between CRC32 and CRC32C.
1053   //   CRC32 holds it's current crc value in the externally visible representation.
1054   //   CRC32C holds it's current crc value in internal format, ready for updating.
1055   //   Thus, the crc value must be bit-flipped before updating it in the CRC32 case.
1056   //   In the CRC32C case, it must be bit-flipped when it is given to the outside world (getValue()).
1057   //   The bool invertCRC parameter indicates whether bit-flipping is required before updates.
1058  private:
1059   void fold_byte_crc32(Register crc, Register table, Register val, Register tmp);
1060   void fold_8bit_crc32(Register crc, Register table, Register tmp);
1061   void update_byte_crc32( Register crc, Register val, Register table);
1062   void update_byteLoop_crc32(Register crc, Register buf, Register len, Register table,
1063                              Register data);
1064   void update_1word_crc32(Register crc, Register buf, Register table, int bufDisp, int bufInc,
1065                           Register t0,  Register t1,  Register t2,  Register t3);
1066  public:
1067   void kernel_crc32_singleByteReg(Register crc, Register val, Register table,
1068                                   bool invertCRC);
1069   void kernel_crc32_singleByte(Register crc, Register buf, Register len, Register table, Register tmp,
1070                                bool invertCRC);
1071   void kernel_crc32_1byte(Register crc, Register buf, Register len, Register table,
1072                           Register t0,  Register t1,  Register t2,  Register t3,
1073                           bool invertCRC);
1074   void kernel_crc32_1word(Register crc, Register buf, Register len, Register table,
1075                           Register t0,  Register t1,  Register t2,  Register t3,
1076                           bool invertCRC);
1077 
1078   // Emitters for BigInteger.multiplyToLen intrinsic
1079   // note: length of result array (zlen) is passed on the stack
1080  private:
1081   void add2_with_carry(Register dest_hi, Register dest_lo,
1082                        Register src1, Register src2);
1083   void multiply_64_x_64_loop(Register x, Register xstart,
1084                              Register x_xstart,
1085                              Register y, Register y_idx, Register z,
1086                              Register carry, Register product,
1087                              Register idx, Register kdx);
1088   void multiply_add_128_x_128(Register x_xstart, Register y, Register z,
1089                               Register yz_idx, Register idx,
1090                               Register carry, Register product, int offset);
1091   void multiply_128_x_128_loop(Register x_xstart,
1092                                Register y, Register z,
1093                                Register yz_idx, Register idx,
1094                                Register jdx,
1095                                Register carry, Register product,
1096                                Register carry2);
1097  public:
1098   void multiply_to_len(Register x, Register xlen,
1099                        Register y, Register ylen,
1100                        Register z,
1101                        Register tmp1, Register tmp2,
1102                        Register tmp3, Register tmp4, Register tmp5);
1103 
1104   // These generate optimized code for all supported s390 implementations, and are preferred for most uses.
1105   void pop_count_int(Register dst, Register src, Register tmp);
1106   void pop_count_long(Register dst, Register src, Register tmp);
1107 
1108   // For legacy (pre-z15) use, but will work on all supported s390 implementations.
1109   void pop_count_int_without_ext3(Register dst, Register src, Register tmp);
1110   void pop_count_long_without_ext3(Register dst, Register src, Register tmp);
1111 
1112   // Only for use on z15 or later s390 implementations.
1113   void pop_count_int_with_ext3(Register dst, Register src);
1114   void pop_count_long_with_ext3(Register dst, Register src);
1115 
1116   void load_on_condition_imm_32(Register dst, int64_t i2, branch_condition cc);
1117   void load_on_condition_imm_64(Register dst, int64_t i2, branch_condition cc);
1118 
1119   // Inline type specific methods
1120   #include "asm/macroAssembler_common.hpp"
1121 };
1122 
1123 #ifdef ASSERT
1124 // Return false (e.g. important for our impl. of virtual calls).
1125 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1126 #endif
1127 
1128 #endif // CPU_S390_MACROASSEMBLER_S390_HPP