1/*
2 * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2016 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_S390_VM_MACROASSEMBLER_S390_HPP
27#define CPU_S390_VM_MACROASSEMBLER_S390_HPP
28
29#include "asm/assembler.hpp"
30
31#define MODERN_IFUN(name)  ((void (MacroAssembler::*)(Register, int64_t, Register, Register))&MacroAssembler::name)
32#define CLASSIC_IFUN(name) ((void (MacroAssembler::*)(Register, int64_t, Register, Register))&MacroAssembler::name)
33#define MODERN_FFUN(name)  ((void (MacroAssembler::*)(FloatRegister, int64_t, Register, Register))&MacroAssembler::name)
34#define CLASSIC_FFUN(name) ((void (MacroAssembler::*)(FloatRegister, int64_t, Register, Register))&MacroAssembler::name)
35
36class MacroAssembler: public Assembler {
37 public:
38  MacroAssembler(CodeBuffer* code) : Assembler(code) {}
39
40  //
41  // Optimized instruction emitters
42  //
43
44  // Move register if destination register and target register are different.
45  void lr_if_needed(Register rd, Register rs);
46  void lgr_if_needed(Register rd, Register rs);
47  void llgfr_if_needed(Register rd, Register rs);
48  void ldr_if_needed(FloatRegister rd, FloatRegister rs);
49
50  void move_reg_if_needed(Register dest, BasicType dest_type, Register src, BasicType src_type);
51  void move_freg_if_needed(FloatRegister dest, BasicType dest_type, FloatRegister src, BasicType src_type);
52
53  void freg2mem_opt(FloatRegister reg,
54                    int64_t       disp,
55                    Register      index,
56                    Register      base,
57                    void (MacroAssembler::*modern) (FloatRegister, int64_t, Register, Register),
58                    void (MacroAssembler::*classic)(FloatRegister, int64_t, Register, Register),
59                    Register      scratch = Z_R0);
60  void freg2mem_opt(FloatRegister reg,
61                    const Address &a, bool is_double = true);
62
63  void mem2freg_opt(FloatRegister reg,
64                    int64_t       disp,
65                    Register      index,
66                    Register      base,
67                    void (MacroAssembler::*modern) (FloatRegister, int64_t, Register, Register),
68                    void (MacroAssembler::*classic)(FloatRegister, int64_t, Register, Register),
69                    Register      scratch = Z_R0);
70  void mem2freg_opt(FloatRegister reg,
71                    const Address &a, bool is_double = true);
72
73  void reg2mem_opt(Register reg,
74                   int64_t  disp,
75                   Register index,
76                   Register base,
77                   void (MacroAssembler::*modern) (Register, int64_t, Register, Register),
78                   void (MacroAssembler::*classic)(Register, int64_t, Register, Register),
79                   Register scratch = Z_R0);
80  // returns offset of the store instruction
81  int reg2mem_opt(Register reg, const Address &a, bool is_double = true);
82
83  void mem2reg_opt(Register reg,
84                   int64_t  disp,
85                   Register index,
86                   Register base,
87                   void (MacroAssembler::*modern) (Register, int64_t, Register, Register),
88                   void (MacroAssembler::*classic)(Register, int64_t, Register, Register));
89  void mem2reg_opt(Register reg, const Address &a, bool is_double = true);
90  void mem2reg_signed_opt(Register reg, const Address &a);
91
92  // AND immediate and set condition code, works for 64 bit immediates/operation as well.
93   void and_imm(Register r, long mask, Register tmp = Z_R0, bool wide = false);
94
95  // 1's complement, 32bit or 64bit. Optimized to exploit distinct operands facility.
96  // Note: The condition code is neither preserved nor correctly set by this code!!!
97  // Note: (wide == false) does not protect the high order half of the target register
98  // from alternation. It only serves as optimization hint for 32-bit results.
99  void not_(Register r1, Register r2 = noreg, bool wide = false);  // r1 = ~r2
100
101  // Expanded support of all "rotate_then_<logicalOP>" instructions.
102  //
103  // Generalize and centralize rotate_then_<logicalOP> emitter.
104  // Functional description. For details, see Principles of Operation, Chapter 7, "Rotate Then Insert..."
105  //  - Bits  in a register are numbered left (most significant) to right (least significant), i.e. [0..63].
106  //  - Bytes in a register are numbered left (most significant) to right (least significant), i.e. [0..7].
107  //  - Register src is rotated to the left by (nRotate&0x3f) positions.
108  //  - Negative values for nRotate result in a rotation to the right by abs(nRotate) positions.
109  //  - The bits in positions [lBitPos..rBitPos] of the _ROTATED_ src operand take part in the
110  //    logical operation performed on the contents (in those positions) of the dst operand.
111  //  - The logical operation that is performed on the dst operand is one of
112  //     o insert the selected bits (replacing the original contents of those bit positions)
113  //     o and the selected bits with the corresponding bits of the dst operand
114  //     o or  the selected bits with the corresponding bits of the dst operand
115  //     o xor the selected bits with the corresponding bits of the dst operand
116  //  - For clear_dst == true, the destination register is cleared before the bits are inserted.
117  //    For clear_dst == false, only the bit positions that get data inserted from src
118  //    are changed. All other bit positions remain unchanged.
119  //  - For test_only == true,  the result of the logicalOP is only used to set the condition code, dst remains unchanged.
120  //    For test_only == false, the result of the logicalOP replaces the selected bits of dst.
121  //  - src32bit and dst32bit indicate the respective register is used as 32bit value only.
122  //    Knowledge can simplify code generation.
123  //
124  // Here is an important performance note, valid for all <logicalOP>s except "insert":
125  //   Due to the too complex nature of the operation, it cannot be done in a single cycle.
126  //   Timing constraints require the instructions to be cracked into two micro-ops, taking
127  //   one or two cycles each to execute. In some cases, an additional pipeline bubble might get added.
128  //   Macroscopically, that makes up for a three- or four-cycle instruction where you would
129  //   expect just a single cycle.
130  //   It is thus not beneficial from a performance point of view to exploit those instructions.
131  //   Other reasons (code compactness, register pressure, ...) might outweigh this penalty.
132  //
133  unsigned long create_mask(int lBitPos, int rBitPos);
134  void rotate_then_mask(Register dst, Register src, int lBitPos, int rBitPos,
135                        int nRotate, bool src32bit, bool dst32bit, bool oneBits);
136  void rotate_then_insert(Register dst, Register src, int lBitPos, int rBitPos, int nRotate,
137                          bool clear_dst);
138  void rotate_then_and(Register dst, Register src, int lBitPos, int rBitPos, int nRotate,
139                       bool test_only);
140  void rotate_then_or(Register dst, Register src, int lBitPos, int rBitPos, int nRotate,
141                      bool test_onlyt);
142  void rotate_then_xor(Register dst, Register src, int lBitPos, int rBitPos, int nRotate,
143                       bool test_only);
144
145  void add64(Register r1, RegisterOrConstant inc);
146
147  // Helper function to multiply the 64bit contents of a register by a 16bit constant.
148  // The optimization tries to avoid the mghi instruction, since it uses the FPU for
149  // calculation and is thus rather slow.
150  //
151  // There is no handling for special cases, e.g. cval==0 or cval==1.
152  //
153  // Returns len of generated code block.
154  unsigned int mul_reg64_const16(Register rval, Register work, int cval);
155
156  // Generic operation r1 := r2 + imm.
157  void add2reg(Register r1, int64_t imm, Register r2 = noreg);
158  // Generic operation r := b + x + d.
159  void add2reg_with_index(Register r, int64_t d, Register x, Register b = noreg);
160
161  // Add2mem* methods for direct memory increment.
162  void add2mem_32(const Address &a, int64_t imm, Register tmp);
163  void add2mem_64(const Address &a, int64_t imm, Register tmp);
164
165  // *((int8_t*)(dst)) |= imm8
166  inline void or2mem_8(Address& dst, int64_t imm8);
167
168  // Load values by size and signedness.
169  void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
170  void store_sized_value(Register src, Address dst, size_t size_in_bytes);
171
172  // Load values with large offsets to base address.
173 private:
174  int  split_largeoffset(int64_t si20_offset, Register tmp, bool fixed_codelen, bool accumulate);
175 public:
176  void load_long_largeoffset(Register t, int64_t si20, Register a, Register tmp);
177  void load_float_largeoffset(FloatRegister t, int64_t si20, Register a, Register tmp);
178  void load_double_largeoffset(FloatRegister t, int64_t si20, Register a, Register tmp);
179
180 private:
181  long toc_distance();
182 public:
183  void load_toc(Register Rtoc);
184  void load_long_pcrelative(Register Rdst, address dataLocation);
185  static int load_long_pcrelative_size() { return 6; }
186  void load_addr_pcrelative(Register Rdst, address dataLocation);
187  static int load_addr_pcrel_size() { return 6; } // Just a LARL.
188
189  // Load a value from memory and test (set CC).
190  void load_and_test_byte    (Register dst, const Address &a);
191  void load_and_test_short   (Register dst, const Address &a);
192  void load_and_test_int     (Register dst, const Address &a);
193  void load_and_test_int2long(Register dst, const Address &a);
194  void load_and_test_long    (Register dst, const Address &a);
195
196  // Test a bit in memory. Result is reflected in CC.
197  void testbit(const Address &a, unsigned int bit);
198  // Test a bit in a register. Result is reflected in CC.
199  void testbit(Register r, unsigned int bitPos);
200
201  // Clear a register, i.e. load const zero into reg. Return len (in bytes) of
202  // generated instruction(s).
203  //   whole_reg: Clear 64 bits if true, 32 bits otherwise.
204  //   set_cc: Use instruction that sets the condition code, if true.
205  int clear_reg(Register r, bool whole_reg = true, bool set_cc = true);
206
207#ifdef ASSERT
208  int preset_reg(Register r, unsigned long pattern, int pattern_len);
209#endif
210
211  // Clear (store zeros) a small piece of memory.
212  // CAUTION: Do not use this for atomic memory clearing. Use store_const() instead.
213  //   addr: Address descriptor of memory to clear.
214  //         Index register will not be used!
215  //   size: Number of bytes to clear.
216  void clear_mem(const Address& addr, unsigned size);
217
218  // Move immediate values to memory. Currently supports 32 and 64 bit stores,
219  // but may be extended to 16 bit store operation, if needed.
220  // For details, see implementation in *.cpp file.
221         int store_const(const Address &dest, long imm,
222                         unsigned int lm, unsigned int lc,
223                         Register scratch = Z_R0);
224  inline int store_const(const Address &dest, long imm,
225                         Register scratch = Z_R0, bool is_long = true);
226
227  // Move/initialize arbitrarily large memory area. No check for destructive overlap.
228  // Being interruptible, these instructions need a retry-loop.
229  void move_long_ext(Register dst, Register src, unsigned int pad);
230
231  void compare_long_ext(Register left, Register right, unsigned int pad);
232  void compare_long_uni(Register left, Register right, unsigned int pad);
233
234  void search_string(Register end, Register start);
235  void search_string_uni(Register end, Register start);
236
237  // Translate instructions
238  // Being interruptible, these instructions need a retry-loop.
239  void translate_oo(Register dst, Register src, uint mask);
240  void translate_ot(Register dst, Register src, uint mask);
241  void translate_to(Register dst, Register src, uint mask);
242  void translate_tt(Register dst, Register src, uint mask);
243
244  // Crypto instructions.
245  // Being interruptible, these instructions need a retry-loop.
246  void cksm(Register crcBuff, Register srcBuff);
247  void km( Register dstBuff, Register srcBuff);
248  void kmc(Register dstBuff, Register srcBuff);
249  void kimd(Register srcBuff);
250  void klmd(Register srcBuff);
251  void kmac(Register srcBuff);
252
253  // nop padding
254  void align(int modulus);
255  void align_address(int modulus);
256
257  //
258  // Constants, loading constants, TOC support
259  //
260  // Safepoint check factored out.
261  void generate_safepoint_check(Label& slow_path, Register scratch = noreg, bool may_relocate = true);
262
263  // Load generic address: d <- base(a) + index(a) + disp(a).
264  inline void load_address(Register d, const Address &a);
265  // Load absolute address (and try to optimize).
266  void load_absolute_address(Register d, address addr);
267
268  // Address of Z_ARG1 and argument_offset.
269  // If temp_reg == arg_slot, arg_slot will be overwritten.
270  Address argument_address(RegisterOrConstant arg_slot,
271                           Register temp_reg = noreg,
272                           int64_t extra_slot_offset = 0);
273
274  // Load a narrow ptr constant (oop or klass ptr).
275  void load_narrow_oop( Register t, narrowOop a);
276  void load_narrow_klass(Register t, Klass* k);
277
278  static bool is_load_const_32to64(address pos);
279  static bool is_load_narrow_oop(address pos)   { return is_load_const_32to64(pos); }
280  static bool is_load_narrow_klass(address pos) { return is_load_const_32to64(pos); }
281
282  static int  load_const_32to64_size()          { return 6; }
283  static bool load_narrow_oop_size()            { return load_const_32to64_size(); }
284  static bool load_narrow_klass_size()          { return load_const_32to64_size(); }
285
286  static int  patch_load_const_32to64(address pos, int64_t a);
287  static int  patch_load_narrow_oop(address pos, oop o);
288  static int  patch_load_narrow_klass(address pos, Klass* k);
289
290  // cOops. CLFI exploit.
291  void compare_immediate_narrow_oop(Register oop1, narrowOop oop2);
292  void compare_immediate_narrow_klass(Register op1, Klass* op2);
293  static bool is_compare_immediate32(address pos);
294  static bool is_compare_immediate_narrow_oop(address pos);
295  static bool is_compare_immediate_narrow_klass(address pos);
296  static int  compare_immediate_narrow_size()       { return 6; }
297  static int  compare_immediate_narrow_oop_size()   { return compare_immediate_narrow_size(); }
298  static int  compare_immediate_narrow_klass_size() { return compare_immediate_narrow_size(); }
299  static int  patch_compare_immediate_32(address pos, int64_t a);
300  static int  patch_compare_immediate_narrow_oop(address pos, oop o);
301  static int  patch_compare_immediate_narrow_klass(address pos, Klass* k);
302
303  // Load a 32bit constant into a 64bit register.
304  void load_const_32to64(Register t, int64_t x, bool sign_extend=true);
305  // Load a 64 bit constant.
306         void load_const(Register t, long a);
307  inline void load_const(Register t, void* a);
308  inline void load_const(Register t, Label& L);
309  inline void load_const(Register t, const AddressLiteral& a);
310  // Get the 64 bit constant from a `load_const' sequence.
311  static long get_const(address load_const);
312  // Patch the 64 bit constant of a `load_const' sequence. This is a low level
313  // procedure. It neither flushes the instruction cache nor is it atomic.
314  static void patch_const(address load_const, long x);
315  static int load_const_size() { return 12; }
316
317  // Turn a char into boolean. NOTE: destroys r.
318  void c2bool(Register r, Register t = Z_R0);
319
320  // Optimized version of load_const for constants that do not need to be
321  // loaded by a sequence of instructions of fixed length and that do not
322  // need to be patched.
323  int load_const_optimized_rtn_len(Register t, long x, bool emit);
324  inline void load_const_optimized(Register t, long x);
325  inline void load_const_optimized(Register t, void* a);
326  inline void load_const_optimized(Register t, Label& L);
327  inline void load_const_optimized(Register t, const AddressLiteral& a);
328
329 public:
330
331  //----------------------------------------------------------
332  //            oops in code             -------------
333  //  including compressed oops support  -------------
334  //----------------------------------------------------------
335
336  // Metadata in code that we have to keep track of.
337  AddressLiteral allocate_metadata_address(Metadata* obj); // allocate_index
338  AddressLiteral constant_metadata_address(Metadata* obj); // find_index
339
340  // allocate_index
341  AddressLiteral allocate_oop_address(jobject obj);
342  // find_index
343  AddressLiteral constant_oop_address(jobject obj);
344  // Uses allocate_oop_address.
345  inline void set_oop         (jobject obj, Register d);
346  // Uses constant_oop_address.
347  inline void set_oop_constant(jobject obj, Register d);
348  // Uses constant_metadata_address.
349  inline bool set_metadata_constant(Metadata* md, Register d);
350
351  virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr,
352                                                Register tmp,
353                                                int offset);
354  //
355  // branch, jump
356  //
357
358  // Use one generic function for all branch patches.
359  static unsigned long patched_branch(address dest_pos, unsigned long inst, address inst_pos);
360
361  void pd_patch_instruction(address branch, address target);
362
363  // Extract relative address from "relative" instructions.
364  static long get_pcrel_offset(unsigned long inst);
365  static long get_pcrel_offset(address pc);
366  static address get_target_addr_pcrel(address pc);
367
368  static inline bool is_call_pcrelative_short(unsigned long inst);
369  static inline bool is_call_pcrelative_long(unsigned long inst);
370  static inline bool is_branch_pcrelative_short(unsigned long inst);
371  static inline bool is_branch_pcrelative_long(unsigned long inst);
372  static inline bool is_compareandbranch_pcrelative_short(unsigned long inst);
373  static inline bool is_branchoncount_pcrelative_short(unsigned long inst);
374  static inline bool is_branchonindex32_pcrelative_short(unsigned long inst);
375  static inline bool is_branchonindex64_pcrelative_short(unsigned long inst);
376  static inline bool is_branchonindex_pcrelative_short(unsigned long inst);
377  static inline bool is_branch_pcrelative16(unsigned long inst);
378  static inline bool is_branch_pcrelative32(unsigned long inst);
379  static inline bool is_branch_pcrelative(unsigned long inst);
380  static inline bool is_load_pcrelative_long(unsigned long inst);
381  static inline bool is_misc_pcrelative_long(unsigned long inst);
382  static inline bool is_pcrelative_short(unsigned long inst);
383  static inline bool is_pcrelative_long(unsigned long inst);
384  // PCrelative TOC access. Variants with address argument.
385  static inline bool is_load_pcrelative_long(address iLoc);
386  static inline bool is_pcrelative_short(address iLoc);
387  static inline bool is_pcrelative_long(address iLoc);
388
389  static inline bool is_pcrelative_instruction(address iloc);
390  static inline bool is_load_addr_pcrel(address a);
391
392  static void patch_target_addr_pcrel(address pc, address con);
393  static void patch_addr_pcrel(address pc, address con) {
394    patch_target_addr_pcrel(pc, con); // Just delegate. This is only for nativeInst_s390.cpp.
395  }
396
397  //---------------------------------------------------------
398  //  Some macros for more comfortable assembler programming.
399  //---------------------------------------------------------
400
401  // NOTE: pass NearLabel T to signal that the branch target T will be bound to a near address.
402
403  void compare32_and_branch(Register r1, RegisterOrConstant x2, branch_condition cond, Label& target);
404  void compareU32_and_branch(Register r1, RegisterOrConstant x2, branch_condition cond, Label& target);
405  void compare64_and_branch(Register r1, RegisterOrConstant x2, branch_condition cond, Label& target);
406  void compareU64_and_branch(Register r1, RegisterOrConstant x2, branch_condition cond, Label& target);
407
408  void branch_optimized(Assembler::branch_condition cond, address branch_target);
409  void branch_optimized(Assembler::branch_condition cond, Label&  branch_target);
410  void compare_and_branch_optimized(Register r1,
411                                    Register r2,
412                                    Assembler::branch_condition cond,
413                                    address  branch_addr,
414                                    bool     len64,
415                                    bool     has_sign);
416  void compare_and_branch_optimized(Register r1,
417                                    jlong    x2,
418                                    Assembler::branch_condition cond,
419                                    Label&   branch_target,
420                                    bool     len64,
421                                    bool     has_sign);
422  void compare_and_branch_optimized(Register r1,
423                                    Register r2,
424                                    Assembler::branch_condition cond,
425                                    Label&   branch_target,
426                                    bool     len64,
427                                    bool     has_sign);
428
429  //
430  // Support for frame handling
431  //
432  // Specify the register that should be stored as the return pc in the
433  // current frame (default is R14).
434  inline void save_return_pc(Register pc = Z_R14);
435  inline void restore_return_pc();
436
437  // Get current PC.
438  address get_PC(Register result);
439
440  // Get current PC + offset. Offset given in bytes, must be even!
441  address get_PC(Register result, int64_t offset);
442
443  // Resize current frame either relatively wrt to current SP or absolute.
444  void resize_frame_sub(Register offset, Register fp, bool load_fp=true);
445  void resize_frame_absolute(Register addr, Register fp, bool load_fp=true);
446  void resize_frame(RegisterOrConstant offset, Register fp, bool load_fp=true);
447
448  // Push a frame of size bytes, if copy_sp is false, old_sp must already
449  // contain a copy of Z_SP.
450  void push_frame(Register bytes, Register old_sp, bool copy_sp = true, bool bytes_with_inverted_sign = false);
451
452  // Push a frame of size `bytes'. no abi space provided.
453  // Don't rely on register locking, instead pass a scratch register
454  // (Z_R0 by default).
455  // CAUTION! passing registers >= Z_R2 may produce bad results on
456  // old CPUs!
457  unsigned int push_frame(unsigned int bytes, Register scratch = Z_R0);
458
459  // Push a frame of size `bytes' with abi160 on top.
460  unsigned int push_frame_abi160(unsigned int bytes);
461
462  // Pop current C frame.
463  void pop_frame();
464
465  //
466  // Calls
467  //
468
469 private:
470  address _last_calls_return_pc;
471
472 public:
473  // Support for VM calls. This is the base routine called by the
474  // different versions of call_VM_leaf. The interpreter may customize
475  // this version by overriding it for its purposes (e.g., to
476  // save/restore additional registers when doing a VM call).
477  void call_VM_leaf_base(address entry_point);
478  void call_VM_leaf_base(address entry_point, bool allow_relocation);
479
480  // It is imperative that all calls into the VM are handled via the
481  // call_VM macros. They make sure that the stack linkage is setup
482  // correctly. Call_VM's correspond to ENTRY/ENTRY_X entry points
483  // while call_VM_leaf's correspond to LEAF entry points.
484  //
485  // This is the base routine called by the different versions of
486  // call_VM. The interpreter may customize this version by overriding
487  // it for its purposes (e.g., to save/restore additional registers
488  // when doing a VM call).
489
490  // If no last_java_sp is specified (noreg) then SP will be used instead.
491
492  virtual void call_VM_base(
493    Register        oop_result,        // Where an oop-result ends up if any; use noreg otherwise.
494    Register        last_java_sp,      // To set up last_Java_frame in stubs; use noreg otherwise.
495    address         entry_point,       // The entry point.
496    bool            check_exception);  // Flag which indicates if exception should be checked.
497  virtual void call_VM_base(
498    Register        oop_result,       // Where an oop-result ends up if any; use noreg otherwise.
499    Register        last_java_sp,     // To set up last_Java_frame in stubs; use noreg otherwise.
500    address         entry_point,      // The entry point.
501    bool            allow_relocation, // Flag to request generation of relocatable code.
502    bool            check_exception); // Flag which indicates if exception should be checked.
503
504  // Call into the VM.
505  // Passes the thread pointer (in Z_ARG1) as a prepended argument.
506  // Makes sure oop return values are visible to the GC.
507  void call_VM(Register oop_result, address entry_point, bool check_exceptions = true);
508  void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
509  void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
510  void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2,
511               Register arg_3, bool check_exceptions = true);
512
513  void call_VM_static(Register oop_result, address entry_point, bool check_exceptions = true);
514  void call_VM_static(Register oop_result, address entry_point, Register arg_1, Register arg_2,
515                      Register arg_3, bool check_exceptions = true);
516
517  // Overloaded with last_java_sp.
518  void call_VM(Register oop_result, Register last_java_sp, address entry_point, bool check_exceptions = true);
519  void call_VM(Register oop_result, Register last_java_sp, address entry_point,
520               Register arg_1, bool check_exceptions = true);
521  void call_VM(Register oop_result, Register last_java_sp, address entry_point,
522               Register arg_1, Register arg_2, bool check_exceptions = true);
523  void call_VM(Register oop_result, Register last_java_sp, address entry_point,
524               Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
525
526  void call_VM_leaf(address entry_point);
527  void call_VM_leaf(address entry_point, Register arg_1);
528  void call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
529  void call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
530
531  // Really static VM leaf call (never patched).
532  void call_VM_leaf_static(address entry_point);
533  void call_VM_leaf_static(address entry_point, Register arg_1);
534  void call_VM_leaf_static(address entry_point, Register arg_1, Register arg_2);
535  void call_VM_leaf_static(address entry_point, Register arg_1, Register arg_2, Register arg_3);
536
537  // Call a C function via its function entry. Updates and returns _last_calls_return_pc.
538  inline address call(Register function_entry);
539  inline address call_c(Register function_entry);
540         address call_c(address function_entry);
541  // Variant for really static (non-relocatable) calls which are never patched.
542         address call_c_static(address function_entry);
543  // TOC or pc-relative call + emits a runtime_call relocation.
544         address call_c_opt(address function_entry);
545
546  inline address call_stub(Register function_entry);
547  inline address call_stub(address  function_entry);
548
549  // Get the pc where the last call will return to. Returns _last_calls_return_pc.
550  inline address last_calls_return_pc();
551
552 private:
553  static bool is_call_far_patchable_variant0_at(address instruction_addr); // Dynamic TOC: load target addr from CP and call.
554  static bool is_call_far_patchable_variant2_at(address instruction_addr); // PC-relative call, prefixed with NOPs.
555
556
557 public:
558  bool           call_far_patchable(address target, int64_t toc_offset);
559  static bool    is_call_far_patchable_at(address inst_start);             // All supported forms of patchable calls.
560  static bool    is_call_far_patchable_pcrelative_at(address inst_start);  // Pc-relative call with leading nops.
561  static bool    is_call_far_pcrelative(address instruction_addr);         // Pure far pc-relative call, with one leading size adjustment nop.
562  static void    set_dest_of_call_far_patchable_at(address inst_start, address target, int64_t toc_offset);
563  static address get_dest_of_call_far_patchable_at(address inst_start, address toc_start);
564
565  void align_call_far_patchable(address pc);
566
567  // PCrelative TOC access.
568
569  // This value is independent of code position - constant for the lifetime of the VM.
570  static int call_far_patchable_size() {
571    return load_const_from_toc_size() + call_byregister_size();
572  }
573
574  static int call_far_patchable_ret_addr_offset() { return call_far_patchable_size(); }
575
576  static bool call_far_patchable_requires_alignment_nop(address pc) {
577    if (!os::is_MP()) return false;
578    int size = call_far_patchable_size();
579    return ((intptr_t)(pc + size) & 0x03L) != 0;
580  }
581
582  // END OF PCrelative TOC access.
583
584  static int jump_byregister_size()          { return 2; }
585  static int jump_pcrelative_size()          { return 4; }
586  static int jump_far_pcrelative_size()      { return 6; }
587  static int call_byregister_size()          { return 2; }
588  static int call_pcrelative_size()          { return 4; }
589  static int call_far_pcrelative_size()      { return 2 + 6; } // Prepend each BRASL with a nop.
590  static int call_far_pcrelative_size_raw()  { return 6; }     // Prepend each BRASL with a nop.
591
592  //
593  // Java utilities
594  //
595
596  // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
597  // The implementation is only non-empty for the InterpreterMacroAssembler,
598  // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
599  virtual void check_and_handle_popframe(Register java_thread);
600  virtual void check_and_handle_earlyret(Register java_thread);
601
602  // Polling page support.
603  enum poll_mask {
604    mask_stackbang = 0xde, // 222 (dec)
605    mask_safepoint = 0x6f, // 111 (dec)
606    mask_profiling = 0xba  // 186 (dec)
607  };
608
609  // Read from the polling page.
610  void load_from_polling_page(Register polling_page_address, int64_t offset = 0);
611
612  // Check if given instruction is a read from the polling page
613  // as emitted by load_from_polling_page.
614  static bool is_load_from_polling_page(address instr_loc);
615  // Extract poll address from instruction and ucontext.
616  static address get_poll_address(address instr_loc, void* ucontext);
617  // Extract poll register from instruction.
618  static uint get_poll_register(address instr_loc);
619
620  // Check if instruction is a write access to the memory serialization page
621  // realized by one of the instructions stw, stwu, stwx, or stwux.
622  static bool is_memory_serialization(int instruction, JavaThread* thread, void* ucontext);
623
624  // Support for serializing memory accesses between threads.
625  void serialize_memory(Register thread, Register tmp1, Register tmp2);
626
627  // Stack overflow checking
628  void bang_stack_with_offset(int offset);
629
630  // Check for reserved stack access in method being exited. If the reserved
631  // stack area was accessed, protect it again and throw StackOverflowError.
632  // Uses Z_R1.
633  void reserved_stack_check(Register return_pc);
634
635  // Atomics
636  // -- none?
637
638  void tlab_allocate(Register obj,                // Result: pointer to object after successful allocation
639                     Register var_size_in_bytes,  // Object size in bytes if unknown at compile time; invalid otherwise.
640                     int      con_size_in_bytes,  // Object size in bytes if   known at compile time.
641                     Register t1,                 // temp register
642                     Label&   slow_case);         // Continuation point if fast allocation fails.
643
644  // Emitter for interface method lookup.
645  //   input: recv_klass, intf_klass, itable_index
646  //   output: method_result
647  //   kills: itable_index, temp1_reg, Z_R0, Z_R1
648  void lookup_interface_method(Register           recv_klass,
649                               Register           intf_klass,
650                               RegisterOrConstant itable_index,
651                               Register           method_result,
652                               Register           temp1_reg,
653                               Register           temp2_reg,
654                               Label&             no_such_interface);
655
656  // virtual method calling
657  void lookup_virtual_method(Register             recv_klass,
658                             RegisterOrConstant   vtable_index,
659                             Register             method_result);
660
661  // Factor out code to call ic_miss_handler.
662  unsigned int call_ic_miss_handler(Label& ICM, int trapMarker, int requiredSize, Register scratch);
663  void nmethod_UEP(Label& ic_miss);
664
665  // Emitters for "partial subtype" checks.
666
667  // Test sub_klass against super_klass, with fast and slow paths.
668
669  // The fast path produces a tri-state answer: yes / no / maybe-slow.
670  // One of the three labels can be NULL, meaning take the fall-through.
671  // If super_check_offset is -1, the value is loaded up from super_klass.
672  // No registers are killed, except temp_reg and temp2_reg.
673  // If super_check_offset is not -1, temp1_reg is not used and can be noreg.
674  void check_klass_subtype_fast_path(Register sub_klass,
675                                     Register super_klass,
676                                     Register temp1_reg,
677                                     Label*   L_success,
678                                     Label*   L_failure,
679                                     Label*   L_slow_path,
680                                     RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
681
682  // The rest of the type check; must be wired to a corresponding fast path.
683  // It does not repeat the fast path logic, so don't use it standalone.
684  // The temp_reg can be noreg, if no temps are available.
685  // It can also be sub_klass or super_klass, meaning it's OK to kill that one.
686  // Updates the sub's secondary super cache as necessary.
687  void check_klass_subtype_slow_path(Register Rsubklass,
688                                     Register Rsuperklas,
689                                     Register Rarray_ptr, // tmp
690                                     Register Rlength,    // tmp
691                                     Label* L_success,
692                                     Label* L_failure);
693
694  // Simplified, combined version, good for typical uses.
695  // Falls through on failure.
696  void check_klass_subtype(Register sub_klass,
697                           Register super_klass,
698                           Register temp1_reg,
699                           Register temp2_reg,
700                           Label&   L_success);
701
702  // Increment a counter at counter_address when the eq condition code is set.
703  // Kills registers tmp1_reg and tmp2_reg and preserves the condition code.
704  void increment_counter_eq(address counter_address, Register tmp1_reg, Register tmp2_reg);
705  // Biased locking support
706  // Upon entry,obj_reg must contain the target object, and mark_reg
707  // must contain the target object's header.
708  // Destroys mark_reg if an attempt is made to bias an anonymously
709  // biased lock. In this case a failure will go either to the slow
710  // case or fall through with the notEqual condition code set with
711  // the expectation that the slow case in the runtime will be called.
712  // In the fall-through case where the CAS-based lock is done,
713  // mark_reg is not destroyed.
714  void biased_locking_enter(Register obj_reg, Register mark_reg, Register temp_reg,
715                            Register temp2_reg, Label& done, Label* slow_case = NULL);
716  // Upon entry, the base register of mark_addr must contain the oop.
717  // Destroys temp_reg.
718  // If allow_delay_slot_filling is set to true, the next instruction
719  // emitted after this one will go in an annulled delay slot if the
720  // biased locking exit case failed.
721  void biased_locking_exit(Register mark_addr, Register temp_reg, Label& done);
722
723  void compiler_fast_lock_object(Register oop, Register box, Register temp1, Register temp2, bool try_bias = UseBiasedLocking);
724  void compiler_fast_unlock_object(Register oop, Register box, Register temp1, Register temp2, bool try_bias = UseBiasedLocking);
725
726  // Write to card table for modification at store_addr - register is destroyed afterwards.
727  void card_write_barrier_post(Register store_addr, Register tmp);
728
729  void resolve_jobject(Register value, Register tmp1, Register tmp2);
730
731#if INCLUDE_ALL_GCS
732  // General G1 pre-barrier generator.
733  // Purpose: record the previous value if it is not null.
734  // All non-tmps are preserved.
735  void g1_write_barrier_pre(Register           Robj,
736                            RegisterOrConstant offset,
737                            Register           Rpre_val,        // Ideally, this is a non-volatile register.
738                            Register           Rval,            // Will be preserved.
739                            Register           Rtmp1,           // If Rpre_val is volatile, either Rtmp1
740                            Register           Rtmp2,           // or Rtmp2 has to be non-volatile.
741                            bool               pre_val_needed); // Save Rpre_val across runtime call, caller uses it.
742
743  // General G1 post-barrier generator.
744  // Purpose: Store cross-region card.
745  void g1_write_barrier_post(Register Rstore_addr,
746                             Register Rnew_val,
747                             Register Rtmp1,
748                             Register Rtmp2,
749                             Register Rtmp3);
750#endif // INCLUDE_ALL_GCS
751
752  // Support for last Java frame (but use call_VM instead where possible).
753 private:
754  void set_last_Java_frame(Register last_Java_sp, Register last_Java_pc, bool allow_relocation);
755  void reset_last_Java_frame(bool allow_relocation);
756  void set_top_ijava_frame_at_SP_as_last_Java_frame(Register sp, Register tmp1, bool allow_relocation);
757 public:
758  inline void set_last_Java_frame(Register last_java_sp, Register last_Java_pc);
759  inline void set_last_Java_frame_static(Register last_java_sp, Register last_Java_pc);
760  inline void reset_last_Java_frame(void);
761  inline void reset_last_Java_frame_static(void);
762  inline void set_top_ijava_frame_at_SP_as_last_Java_frame(Register sp, Register tmp1);
763  inline void set_top_ijava_frame_at_SP_as_last_Java_frame_static(Register sp, Register tmp1);
764
765  void set_thread_state(JavaThreadState new_state);
766
767  // Read vm result from thread.
768  void get_vm_result  (Register oop_result);
769  void get_vm_result_2(Register result);
770
771  // Vm result is currently getting hijacked to for oop preservation.
772  void set_vm_result(Register oop_result);
773
774  // Support for NULL-checks
775  //
776  // Generates code that causes a NULL OS exception if the content of reg is NULL.
777  // If the accessed location is M[reg + offset] and the offset is known, provide the
778  // offset. No explicit code generation is needed if the offset is within a certain
779  // range (0 <= offset <= page_size).
780  //
781  // %%%%%% Currently not done for z/Architecture
782
783  void null_check(Register reg, Register tmp = Z_R0, int64_t offset = -1);
784  static bool needs_explicit_null_check(intptr_t offset);  // Implemented in shared file ?!
785
786  // Klass oop manipulations if compressed.
787  void encode_klass_not_null(Register dst, Register src = noreg);
788  void decode_klass_not_null(Register dst, Register src);
789  void decode_klass_not_null(Register dst);
790  void load_klass(Register klass, Address mem);
791  void load_klass(Register klass, Register src_oop);
792  void load_prototype_header(Register Rheader, Register Rsrc_oop);
793  void store_klass(Register klass, Register dst_oop, Register ck = noreg); // Klass will get compressed if ck not provided.
794  void store_klass_gap(Register s, Register dst_oop);
795
796  // This function calculates the size of the code generated by
797  //   decode_klass_not_null(register dst)
798  // when (Universe::heap() != NULL). Hence, if the instructions
799  // it generates change, then this method needs to be updated.
800  static int instr_size_for_decode_klass_not_null();
801
802  void encode_heap_oop(Register oop);
803  void encode_heap_oop_not_null(Register oop);
804
805  static int get_oop_base_pow2_offset(uint64_t oop_base);
806  int  get_oop_base(Register Rbase, uint64_t oop_base);
807  int  get_oop_base_complement(Register Rbase, uint64_t oop_base);
808  void compare_heap_oop(Register Rop1, Address mem, bool maybeNULL);
809  void compare_klass_ptr(Register Rop1, int64_t disp, Register Rbase, bool maybeNULL);
810  void load_heap_oop(Register dest, const Address &a);
811  void load_heap_oop(Register d, int64_t si16, Register s1);
812  void load_heap_oop_not_null(Register d, int64_t si16, Register s1);
813  void store_heap_oop(Register Roop, RegisterOrConstant offset, Register base);
814  void store_heap_oop_not_null(Register Roop, RegisterOrConstant offset, Register base);
815  void store_heap_oop_null(Register zero, RegisterOrConstant offset, Register base);
816  void oop_encoder(Register Rdst, Register Rsrc, bool maybeNULL,
817                   Register Rbase = Z_R1, int pow2_offset = -1, bool only32bitValid = false);
818  void oop_decoder(Register Rdst, Register Rsrc, bool maybeNULL,
819                   Register Rbase = Z_R1, int pow2_offset = -1);
820
821  void load_mirror(Register mirror, Register method);
822
823  //--------------------------
824  //---  perations on arrays.
825  //--------------------------
826  unsigned int Clear_Array(Register cnt_arg, Register base_pointer_arg, Register src_addr, Register src_len);
827  unsigned int Clear_Array_Const(long cnt, Register base);
828  unsigned int Clear_Array_Const_Big(long cnt, Register base_pointer_arg, Register src_addr, Register src_len);
829  unsigned int CopyRawMemory_AlignedDisjoint(Register src_reg, Register dst_reg,
830                                             Register cnt_reg,
831                                             Register tmp1_reg, Register tmp2_reg);
832
833  //-------------------------------------------
834  // Special String Intrinsics Implementation.
835  //-------------------------------------------
836  // Intrinsics for CompactStrings
837  // Compress char[] to byte[]. odd_reg contains cnt. tmp3 is only needed for precise behavior in failure case. Kills dst.
838  unsigned int string_compress(Register result, Register src, Register dst, Register odd_reg,
839                               Register even_reg, Register tmp, Register tmp2 = noreg);
840
841  // Kills src.
842  unsigned int has_negatives(Register result, Register src, Register cnt,
843                             Register odd_reg, Register even_reg, Register tmp);
844
845  // Inflate byte[] to char[].
846  unsigned int string_inflate_trot(Register src, Register dst, Register cnt, Register tmp);
847  // Odd_reg contains cnt. Kills src.
848  unsigned int string_inflate(Register src, Register dst, Register odd_reg,
849                              Register even_reg, Register tmp);
850
851  unsigned int string_compare(Register str1, Register str2, Register cnt1, Register cnt2,
852                              Register odd_reg, Register even_reg, Register result, int ae);
853
854  unsigned int array_equals(bool is_array_equ, Register ary1, Register ary2, Register limit,
855                            Register odd_reg, Register even_reg, Register result, bool is_byte);
856
857  unsigned int string_indexof(Register result, Register haystack, Register haycnt,
858                              Register needle, Register needlecnt, int needlecntval,
859                              Register odd_reg, Register even_reg, int ae);
860
861  unsigned int string_indexof_char(Register result, Register haystack, Register haycnt,
862                                   Register needle, jchar needleChar, Register odd_reg, Register even_reg, bool is_byte);
863
864  // Emit an oop const to the constant pool and set a relocation info
865  // with address current_pc. Return the TOC offset of the constant.
866  int store_const_in_toc(AddressLiteral& val);
867  int store_oop_in_toc(AddressLiteral& oop);
868  // Emit an oop const to the constant pool via store_oop_in_toc, or
869  // emit a scalar const to the constant pool via store_const_in_toc,
870  // and load the constant into register dst.
871  bool load_const_from_toc(Register dst, AddressLiteral& a, Register Rtoc = noreg);
872  // Get CPU version dependent size of load_const sequence.
873  // The returned value is valid only for code sequences
874  // generated by load_const, not load_const_optimized.
875  static int load_const_from_toc_size() {
876    return load_long_pcrelative_size();
877  }
878  bool load_oop_from_toc(Register dst, AddressLiteral& a, Register Rtoc = noreg);
879  static intptr_t get_const_from_toc(address pc);
880  static void     set_const_in_toc(address pc, unsigned long new_data, CodeBlob *cb);
881
882  // Dynamic TOC.
883  static bool is_load_const(address a);
884  static bool is_load_const_from_toc_pcrelative(address a);
885  static bool is_load_const_from_toc(address a) { return is_load_const_from_toc_pcrelative(a); }
886
887  // PCrelative TOC access.
888  static bool is_call_byregister(address a) { return is_z_basr(*(short*)a); }
889  static bool is_load_const_from_toc_call(address a);
890  static bool is_load_const_call(address a);
891  static int load_const_call_size() { return load_const_size() + call_byregister_size(); }
892  static int load_const_from_toc_call_size() { return load_const_from_toc_size() + call_byregister_size(); }
893  // Offset is +/- 2**32 -> use long.
894  static long get_load_const_from_toc_offset(address a);
895
896
897  void generate_type_profiling(const Register Rdata,
898                               const Register Rreceiver_klass,
899                               const Register Rwanted_receiver_klass,
900                               const Register Rmatching_row,
901                               bool is_virtual_call);
902
903  // Bit operations for single register operands.
904  inline void lshift(Register r, int places, bool doubl = true);   // <<
905  inline void rshift(Register r, int places, bool doubl = true);   // >>
906
907  //
908  // Debugging
909  //
910
911  // Assert on CC (condition code in CPU state).
912  void asm_assert(bool check_equal, const char* msg, int id) PRODUCT_RETURN;
913  void asm_assert_low(const char *msg, int id) PRODUCT_RETURN;
914  void asm_assert_high(const char *msg, int id) PRODUCT_RETURN;
915  void asm_assert_eq(const char* msg, int id) { asm_assert(true, msg, id); }
916  void asm_assert_ne(const char* msg, int id) { asm_assert(false, msg, id); }
917
918  void asm_assert_static(bool check_equal, const char* msg, int id) PRODUCT_RETURN;
919
920 private:
921  // Emit assertions.
922  void asm_assert_mems_zero(bool check_equal, bool allow_relocation, int size, int64_t mem_offset,
923                            Register mem_base, const char* msg, int id) PRODUCT_RETURN;
924
925 public:
926  inline void asm_assert_mem4_is_zero(int64_t mem_offset, Register mem_base, const char* msg, int id) {
927    asm_assert_mems_zero(true,  true, 4, mem_offset, mem_base, msg, id);
928  }
929  inline void asm_assert_mem8_is_zero(int64_t mem_offset, Register mem_base, const char* msg, int id) {
930    asm_assert_mems_zero(true,  true, 8, mem_offset, mem_base, msg, id);
931  }
932  inline void asm_assert_mem4_isnot_zero(int64_t mem_offset, Register mem_base, const char* msg, int id) {
933    asm_assert_mems_zero(false, true, 4, mem_offset, mem_base, msg, id);
934  }
935  inline void asm_assert_mem8_isnot_zero(int64_t mem_offset, Register mem_base, const char* msg, int id) {
936    asm_assert_mems_zero(false, true, 8, mem_offset, mem_base, msg, id);
937  }
938
939  inline void asm_assert_mem4_is_zero_static(int64_t mem_offset, Register mem_base, const char* msg, int id) {
940    asm_assert_mems_zero(true,  false, 4, mem_offset, mem_base, msg, id);
941  }
942  inline void asm_assert_mem8_is_zero_static(int64_t mem_offset, Register mem_base, const char* msg, int id) {
943    asm_assert_mems_zero(true,  false, 8, mem_offset, mem_base, msg, id);
944  }
945  inline void asm_assert_mem4_isnot_zero_static(int64_t mem_offset, Register mem_base, const char* msg, int id) {
946    asm_assert_mems_zero(false, false, 4, mem_offset, mem_base, msg, id);
947  }
948  inline void asm_assert_mem8_isnot_zero_static(int64_t mem_offset, Register mem_base, const char* msg, int id) {
949    asm_assert_mems_zero(false, false, 8, mem_offset, mem_base, msg, id);
950  }
951  void asm_assert_frame_size(Register expected_size, Register tmp, const char* msg, int id) PRODUCT_RETURN;
952
953  // Verify Z_thread contents.
954  void verify_thread();
955
956  // Only if +VerifyOops.
957  void verify_oop(Register reg, const char* s = "broken oop");
958
959  // TODO: verify_method and klass metadata (compare against vptr?).
960  void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
961  void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line) {}
962
963#define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
964#define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
965
966 private:
967  // Generate printout in stop().
968  static const char* stop_types[];
969  enum {
970    stop_stop               = 0,
971    stop_untested           = 1,
972    stop_unimplemented      = 2,
973    stop_shouldnotreachhere = 3,
974    stop_end                = 4
975  };
976  // Prints msg and stops execution.
977  void    stop(int type, const char* msg, int id = 0);
978  address stop_chain(address reentry, int type, const char* msg, int id, bool allow_relocation); // Non-relocateable code only!!
979  void    stop_static(int type, const char* msg, int id);                                        // Non-relocateable code only!!
980
981 public:
982
983  // Prints msg and stops.
984  address stop_chain(      address reentry, const char* msg = "", int id = 0) { return stop_chain(reentry, stop_stop, msg, id, true); }
985  address stop_chain_static(address reentry, const char* msg = "", int id = 0) { return stop_chain(reentry, stop_stop, msg, id, false); }
986  void stop_static  (const char* msg = "", int id = 0) { stop_static(stop_stop,   msg, id); }
987  void stop         (const char* msg = "", int id = 0) { stop(stop_stop,          msg, id); }
988  void untested     (const char* msg = "", int id = 0) { stop(stop_untested,      msg, id); }
989  void unimplemented(const char* msg = "", int id = 0) { stop(stop_unimplemented, msg, id); }
990  void should_not_reach_here(const char* msg = "", int id = -1) { stop(stop_shouldnotreachhere, msg, id); }
991
992  // Factor out part of stop into subroutine to save space.
993  void stop_subroutine();
994
995  // Prints msg, but don't stop.
996  void warn(const char* msg);
997
998  //-----------------------------
999  //---  basic block tracing code
1000  //-----------------------------
1001  void trace_basic_block(uint i);
1002  void init_basic_block_trace();
1003  // Number of bytes a basic block gets larger due to the tracing code macro (worst case).
1004  // Currently, worst case is 48 bytes. 64 puts us securely on the safe side.
1005  static int basic_blck_trace_blk_size_incr() { return 64; }
1006
1007  // Write pattern 0x0101010101010101 in region [low-before, high+after].
1008  // Low and high may be the same registers. Before and after are
1009  // the numbers of 8-byte words.
1010  void zap_from_to(Register low, Register high, Register tmp1 = Z_R0, Register tmp2 = Z_R1,
1011                   int before = 0, int after = 0) PRODUCT_RETURN;
1012
1013  // Emitters for CRC32 calculation.
1014 private:
1015  void fold_byte_crc32(Register crc, Register table, Register val, Register tmp);
1016  void fold_8bit_crc32(Register crc, Register table, Register tmp);
1017  void update_byteLoop_crc32(Register crc, Register buf, Register len, Register table,
1018                             Register data, bool invertCRC);
1019  void update_1word_crc32(Register crc, Register buf, Register table, int bufDisp, int bufInc,
1020                          Register t0,  Register t1,  Register t2,  Register t3);
1021 public:
1022  void update_byte_crc32( Register crc, Register val, Register table);
1023  void kernel_crc32_singleByte(Register crc, Register buf, Register len, Register table, Register tmp);
1024  void kernel_crc32_1byte(Register crc, Register buf, Register len, Register table,
1025                          Register t0,  Register t1,  Register t2,  Register t3);
1026  void kernel_crc32_1word(Register crc, Register buf, Register len, Register table,
1027                          Register t0,  Register t1,  Register t2,  Register t3);
1028  void kernel_crc32_2word(Register crc, Register buf, Register len, Register table,
1029                          Register t0,  Register t1,  Register t2,  Register t3);
1030
1031  // Emitters for BigInteger.multiplyToLen intrinsic
1032  // note: length of result array (zlen) is passed on the stack
1033 private:
1034  void add2_with_carry(Register dest_hi, Register dest_lo,
1035                       Register src1, Register src2);
1036  void multiply_64_x_64_loop(Register x, Register xstart,
1037                             Register x_xstart,
1038                             Register y, Register y_idx, Register z,
1039                             Register carry, Register product,
1040                             Register idx, Register kdx);
1041  void multiply_add_128_x_128(Register x_xstart, Register y, Register z,
1042                              Register yz_idx, Register idx,
1043                              Register carry, Register product, int offset);
1044  void multiply_128_x_128_loop(Register x_xstart,
1045                               Register y, Register z,
1046                               Register yz_idx, Register idx,
1047                               Register jdx,
1048                               Register carry, Register product,
1049                               Register carry2);
1050 public:
1051  void multiply_to_len(Register x, Register xlen,
1052                       Register y, Register ylen,
1053                       Register z,
1054                       Register tmp1, Register tmp2,
1055                       Register tmp3, Register tmp4, Register tmp5);
1056};
1057
1058/**
1059 * class SkipIfEqual:
1060 *
1061 * Instantiating this class will result in assembly code being output that will
1062 * jump around any code emitted between the creation of the instance and it's
1063 * automatic destruction at the end of a scope block, depending on the value of
1064 * the flag passed to the constructor, which will be checked at run-time.
1065 */
1066class SkipIfEqual {
1067 private:
1068  MacroAssembler* _masm;
1069  Label _label;
1070
1071 public:
1072  SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value, Register _rscratch);
1073  ~SkipIfEqual();
1074};
1075
1076#ifdef ASSERT
1077// Return false (e.g. important for our impl. of virtual calls).
1078inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1079#endif
1080
1081#endif // CPU_S390_VM_MACROASSEMBLER_S390_HPP
1082