1 /*
2 * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #ifndef CPU_SPARC_VM_MACROASSEMBLER_SPARC_HPP
26 #define CPU_SPARC_VM_MACROASSEMBLER_SPARC_HPP
27
28 #include "asm/assembler.hpp"
29 #include "utilities/macros.hpp"
30
31 // <sys/trap.h> promises that the system will not use traps 16-31
32 #define ST_RESERVED_FOR_USER_0 0x10
33
34 class BiasedLockingCounters;
35
36
37 // Register aliases for parts of the system:
38
39 // 64 bit values can be kept in g1-g5, o1-o5 and o7 and all 64 bits are safe
40 // across context switches in V8+ ABI. Of course, there are no 64 bit regs
41 // in V8 ABI. All 64 bits are preserved in V9 ABI for all registers.
42
43 // g2-g4 are scratch registers called "application globals". Their
44 // meaning is reserved to the "compilation system"--which means us!
45 // They are are not supposed to be touched by ordinary C code, although
46 // highly-optimized C code might steal them for temps. They are safe
47 // across thread switches, and the ABI requires that they be safe
48 // across function calls.
49 //
50 // g1 and g3 are touched by more modules. V8 allows g1 to be clobbered
51 // across func calls, and V8+ also allows g5 to be clobbered across
52 // func calls. Also, g1 and g5 can get touched while doing shared
53 // library loading.
54 //
55 // We must not touch g7 (it is the thread-self register) and g6 is
56 // reserved for certain tools. g0, of course, is always zero.
57 //
58 // (Sources: SunSoft Compilers Group, thread library engineers.)
59
60 // %%%% The interpreter should be revisited to reduce global scratch regs.
61
62 // This global always holds the current JavaThread pointer:
63
64 REGISTER_DECLARATION(Register, G2_thread , G2);
65 REGISTER_DECLARATION(Register, G6_heapbase , G6);
66
67 // The following globals are part of the Java calling convention:
68
69 REGISTER_DECLARATION(Register, G5_method , G5);
70 REGISTER_DECLARATION(Register, G5_megamorphic_method , G5_method);
71 REGISTER_DECLARATION(Register, G5_inline_cache_reg , G5_method);
72
73 // The following globals are used for the new C1 & interpreter calling convention:
74 REGISTER_DECLARATION(Register, Gargs , G4); // pointing to the last argument
75
76 // This local is used to preserve G2_thread in the interpreter and in stubs:
77 REGISTER_DECLARATION(Register, L7_thread_cache , L7);
78
79 // These globals are used as scratch registers in the interpreter:
80
81 REGISTER_DECLARATION(Register, Gframe_size , G1); // SAME REG as G1_scratch
82 REGISTER_DECLARATION(Register, G1_scratch , G1); // also SAME
83 REGISTER_DECLARATION(Register, G3_scratch , G3);
84 REGISTER_DECLARATION(Register, G4_scratch , G4);
85
86 // These globals are used as short-lived scratch registers in the compiler:
87
88 REGISTER_DECLARATION(Register, Gtemp , G5);
89
90 // JSR 292 fixed register usages:
91 REGISTER_DECLARATION(Register, G5_method_type , G5);
92 REGISTER_DECLARATION(Register, G3_method_handle , G3);
93 REGISTER_DECLARATION(Register, L7_mh_SP_save , L7);
94
95 // The compiler requires that G5_megamorphic_method is G5_inline_cache_klass,
96 // because a single patchable "set" instruction (NativeMovConstReg,
97 // or NativeMovConstPatching for compiler1) instruction
98 // serves to set up either quantity, depending on whether the compiled
99 // call site is an inline cache or is megamorphic. See the function
100 // CompiledIC::set_to_megamorphic.
101 //
102 // If a inline cache targets an interpreted method, then the
103 // G5 register will be used twice during the call. First,
104 // the call site will be patched to load a compiledICHolder
105 // into G5. (This is an ordered pair of ic_klass, method.)
106 // The c2i adapter will first check the ic_klass, then load
107 // G5_method with the method part of the pair just before
108 // jumping into the interpreter.
109 //
110 // Note that G5_method is only the method-self for the interpreter,
111 // and is logically unrelated to G5_megamorphic_method.
112 //
113 // Invariants on G2_thread (the JavaThread pointer):
114 // - it should not be used for any other purpose anywhere
115 // - it must be re-initialized by StubRoutines::call_stub()
116 // - it must be preserved around every use of call_VM
117
118 // We can consider using g2/g3/g4 to cache more values than the
119 // JavaThread, such as the card-marking base or perhaps pointers into
120 // Eden. It's something of a waste to use them as scratch temporaries,
121 // since they are not supposed to be volatile. (Of course, if we find
122 // that Java doesn't benefit from application globals, then we can just
123 // use them as ordinary temporaries.)
124 //
125 // Since g1 and g5 (and/or g6) are the volatile (caller-save) registers,
126 // it makes sense to use them routinely for procedure linkage,
127 // whenever the On registers are not applicable. Examples: G5_method,
128 // G5_inline_cache_klass, and a double handful of miscellaneous compiler
129 // stubs. This means that compiler stubs, etc., should be kept to a
130 // maximum of two or three G-register arguments.
131
132
133 // stub frames
134
135 REGISTER_DECLARATION(Register, Lentry_args , L0); // pointer to args passed to callee (interpreter) not stub itself
136
137 // Interpreter frames
138
139 #ifdef CC_INTERP
140 REGISTER_DECLARATION(Register, Lstate , L0); // interpreter state object pointer
141 REGISTER_DECLARATION(Register, L1_scratch , L1); // scratch
142 REGISTER_DECLARATION(Register, Lmirror , L1); // mirror (for native methods only)
143 REGISTER_DECLARATION(Register, L2_scratch , L2);
144 REGISTER_DECLARATION(Register, L3_scratch , L3);
145 REGISTER_DECLARATION(Register, L4_scratch , L4);
146 REGISTER_DECLARATION(Register, Lscratch , L5); // C1 uses
147 REGISTER_DECLARATION(Register, Lscratch2 , L6); // C1 uses
148 REGISTER_DECLARATION(Register, L7_scratch , L7); // constant pool cache
149 REGISTER_DECLARATION(Register, O5_savedSP , O5);
150 REGISTER_DECLARATION(Register, I5_savedSP , I5); // Saved SP before bumping for locals. This is simply
151 // a copy SP, so in 64-bit it's a biased value. The bias
152 // is added and removed as needed in the frame code.
153 // Interface to signature handler
154 REGISTER_DECLARATION(Register, Llocals , L7); // pointer to locals for signature handler
155 REGISTER_DECLARATION(Register, Lmethod , L6); // Method* when calling signature handler
156
157 #else
158 REGISTER_DECLARATION(Register, Lesp , L0); // expression stack pointer
159 REGISTER_DECLARATION(Register, Lbcp , L1); // pointer to next bytecode
160 REGISTER_DECLARATION(Register, Lmethod , L2);
161 REGISTER_DECLARATION(Register, Llocals , L3);
162 REGISTER_DECLARATION(Register, Largs , L3); // pointer to locals for signature handler
163 // must match Llocals in asm interpreter
164 REGISTER_DECLARATION(Register, Lmonitors , L4);
165 REGISTER_DECLARATION(Register, Lbyte_code , L5);
166 // When calling out from the interpreter we record SP so that we can remove any extra stack
167 // space allocated during adapter transitions. This register is only live from the point
168 // of the call until we return.
169 REGISTER_DECLARATION(Register, Llast_SP , L5);
170 REGISTER_DECLARATION(Register, Lscratch , L5);
171 REGISTER_DECLARATION(Register, Lscratch2 , L6);
172 REGISTER_DECLARATION(Register, LcpoolCache , L6); // constant pool cache
173
174 REGISTER_DECLARATION(Register, O5_savedSP , O5);
175 REGISTER_DECLARATION(Register, I5_savedSP , I5); // Saved SP before bumping for locals. This is simply
176 // a copy SP, so in 64-bit it's a biased value. The bias
177 // is added and removed as needed in the frame code.
178 REGISTER_DECLARATION(Register, IdispatchTables , I4); // Base address of the bytecode dispatch tables
179 REGISTER_DECLARATION(Register, IdispatchAddress , I3); // Register which saves the dispatch address for each bytecode
180 REGISTER_DECLARATION(Register, ImethodDataPtr , I2); // Pointer to the current method data
181 #endif /* CC_INTERP */
182
183 // NOTE: Lscratch2 and LcpoolCache point to the same registers in
184 // the interpreter code. If Lscratch2 needs to be used for some
185 // purpose than LcpoolCache should be restore after that for
186 // the interpreter to work right
187 // (These assignments must be compatible with L7_thread_cache; see above.)
188
189 // Lbcp points into the middle of the method object.
190
191 // Exception processing
192 // These registers are passed into exception handlers.
193 // All exception handlers require the exception object being thrown.
194 // In addition, an nmethod's exception handler must be passed
195 // the address of the call site within the nmethod, to allow
196 // proper selection of the applicable catch block.
197 // (Interpreter frames use their own bcp() for this purpose.)
198 //
199 // The Oissuing_pc value is not always needed. When jumping to a
200 // handler that is known to be interpreted, the Oissuing_pc value can be
201 // omitted. An actual catch block in compiled code receives (from its
202 // nmethod's exception handler) the thrown exception in the Oexception,
203 // but it doesn't need the Oissuing_pc.
204 //
205 // If an exception handler (either interpreted or compiled)
206 // discovers there is no applicable catch block, it updates
207 // the Oissuing_pc to the continuation PC of its own caller,
208 // pops back to that caller's stack frame, and executes that
209 // caller's exception handler. Obviously, this process will
210 // iterate until the control stack is popped back to a method
211 // containing an applicable catch block. A key invariant is
212 // that the Oissuing_pc value is always a value local to
213 // the method whose exception handler is currently executing.
214 //
215 // Note: The issuing PC value is __not__ a raw return address (I7 value).
216 // It is a "return pc", the address __following__ the call.
217 // Raw return addresses are converted to issuing PCs by frame::pc(),
218 // or by stubs. Issuing PCs can be used directly with PC range tables.
219 //
220 REGISTER_DECLARATION(Register, Oexception , O0); // exception being thrown
221 REGISTER_DECLARATION(Register, Oissuing_pc , O1); // where the exception is coming from
222
223
224 // These must occur after the declarations above
225 #ifndef DONT_USE_REGISTER_DEFINES
226
227 #define Gthread AS_REGISTER(Register, Gthread)
228 #define Gmethod AS_REGISTER(Register, Gmethod)
229 #define Gmegamorphic_method AS_REGISTER(Register, Gmegamorphic_method)
230 #define Ginline_cache_reg AS_REGISTER(Register, Ginline_cache_reg)
231 #define Gargs AS_REGISTER(Register, Gargs)
232 #define Lthread_cache AS_REGISTER(Register, Lthread_cache)
233 #define Gframe_size AS_REGISTER(Register, Gframe_size)
234 #define Gtemp AS_REGISTER(Register, Gtemp)
235
236 #ifdef CC_INTERP
237 #define Lstate AS_REGISTER(Register, Lstate)
238 #define Lesp AS_REGISTER(Register, Lesp)
239 #define L1_scratch AS_REGISTER(Register, L1_scratch)
240 #define Lmirror AS_REGISTER(Register, Lmirror)
241 #define L2_scratch AS_REGISTER(Register, L2_scratch)
242 #define L3_scratch AS_REGISTER(Register, L3_scratch)
243 #define L4_scratch AS_REGISTER(Register, L4_scratch)
244 #define Lscratch AS_REGISTER(Register, Lscratch)
245 #define Lscratch2 AS_REGISTER(Register, Lscratch2)
246 #define L7_scratch AS_REGISTER(Register, L7_scratch)
247 #define Ostate AS_REGISTER(Register, Ostate)
248 #else
249 #define Lesp AS_REGISTER(Register, Lesp)
250 #define Lbcp AS_REGISTER(Register, Lbcp)
251 #define Lmethod AS_REGISTER(Register, Lmethod)
252 #define Llocals AS_REGISTER(Register, Llocals)
253 #define Lmonitors AS_REGISTER(Register, Lmonitors)
254 #define Lbyte_code AS_REGISTER(Register, Lbyte_code)
255 #define Lscratch AS_REGISTER(Register, Lscratch)
256 #define Lscratch2 AS_REGISTER(Register, Lscratch2)
257 #define LcpoolCache AS_REGISTER(Register, LcpoolCache)
258 #endif /* ! CC_INTERP */
259
260 #define Lentry_args AS_REGISTER(Register, Lentry_args)
261 #define I5_savedSP AS_REGISTER(Register, I5_savedSP)
262 #define O5_savedSP AS_REGISTER(Register, O5_savedSP)
263 #define IdispatchAddress AS_REGISTER(Register, IdispatchAddress)
264 #define ImethodDataPtr AS_REGISTER(Register, ImethodDataPtr)
265 #define IdispatchTables AS_REGISTER(Register, IdispatchTables)
266
267 #define Oexception AS_REGISTER(Register, Oexception)
268 #define Oissuing_pc AS_REGISTER(Register, Oissuing_pc)
269
270 #endif
271
272
273 // Address is an abstraction used to represent a memory location.
274 //
275 // Note: A register location is represented via a Register, not
276 // via an address for efficiency & simplicity reasons.
277
278 class Address VALUE_OBJ_CLASS_SPEC {
279 private:
280 Register _base; // Base register.
281 RegisterOrConstant _index_or_disp; // Index register or constant displacement.
282 RelocationHolder _rspec;
283
284 public:
285 Address() : _base(noreg), _index_or_disp(noreg) {}
286
287 Address(Register base, RegisterOrConstant index_or_disp)
288 : _base(base),
289 _index_or_disp(index_or_disp) {
290 }
291
292 Address(Register base, Register index)
293 : _base(base),
294 _index_or_disp(index) {
295 }
296
297 Address(Register base, int disp)
298 : _base(base),
299 _index_or_disp(disp) {
300 }
301
302 #ifdef ASSERT
303 // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
304 Address(Register base, ByteSize disp)
305 : _base(base),
306 _index_or_disp(in_bytes(disp)) {
307 }
308 #endif
309
310 // accessors
311 Register base() const { return _base; }
312 Register index() const { return _index_or_disp.as_register(); }
313 int disp() const { return _index_or_disp.as_constant(); }
314
315 bool has_index() const { return _index_or_disp.is_register(); }
316 bool has_disp() const { return _index_or_disp.is_constant(); }
317
318 bool uses(Register reg) const { return base() == reg || (has_index() && index() == reg); }
319
320 const relocInfo::relocType rtype() { return _rspec.type(); }
321 const RelocationHolder& rspec() { return _rspec; }
322
323 RelocationHolder rspec(int offset) const {
324 return offset == 0 ? _rspec : _rspec.plus(offset);
325 }
326
327 inline bool is_simm13(int offset = 0); // check disp+offset for overflow
328
329 Address plus_disp(int plusdisp) const { // bump disp by a small amount
330 assert(_index_or_disp.is_constant(), "must have a displacement");
331 Address a(base(), disp() + plusdisp);
332 return a;
333 }
334 bool is_same_address(Address a) const {
335 // disregard _rspec
336 return base() == a.base() && (has_index() ? index() == a.index() : disp() == a.disp());
337 }
338
339 Address after_save() const {
340 Address a = (*this);
341 a._base = a._base->after_save();
342 return a;
343 }
344
345 Address after_restore() const {
346 Address a = (*this);
347 a._base = a._base->after_restore();
348 return a;
349 }
350
351 // Convert the raw encoding form into the form expected by the
352 // constructor for Address.
353 static Address make_raw(int base, int index, int scale, int disp, relocInfo::relocType disp_reloc);
354
355 friend class Assembler;
356 };
357
358
359 class AddressLiteral VALUE_OBJ_CLASS_SPEC {
360 private:
361 address _address;
362 RelocationHolder _rspec;
363
364 RelocationHolder rspec_from_rtype(relocInfo::relocType rtype, address addr) {
365 switch (rtype) {
366 case relocInfo::external_word_type:
367 return external_word_Relocation::spec(addr);
368 case relocInfo::internal_word_type:
369 return internal_word_Relocation::spec(addr);
370 #ifdef _LP64
371 case relocInfo::opt_virtual_call_type:
372 return opt_virtual_call_Relocation::spec();
373 case relocInfo::static_call_type:
374 return static_call_Relocation::spec();
375 case relocInfo::runtime_call_type:
376 return runtime_call_Relocation::spec();
377 #endif
378 case relocInfo::none:
379 return RelocationHolder();
380 default:
381 ShouldNotReachHere();
382 return RelocationHolder();
383 }
384 }
385
386 protected:
387 // creation
388 AddressLiteral() : _address(NULL), _rspec(NULL) {}
389
390 public:
391 AddressLiteral(address addr, RelocationHolder const& rspec)
392 : _address(addr),
393 _rspec(rspec) {}
394
395 // Some constructors to avoid casting at the call site.
396 AddressLiteral(jobject obj, RelocationHolder const& rspec)
397 : _address((address) obj),
398 _rspec(rspec) {}
399
400 AddressLiteral(intptr_t value, RelocationHolder const& rspec)
401 : _address((address) value),
402 _rspec(rspec) {}
403
404 AddressLiteral(address addr, relocInfo::relocType rtype = relocInfo::none)
405 : _address((address) addr),
406 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
407
408 // Some constructors to avoid casting at the call site.
409 AddressLiteral(address* addr, relocInfo::relocType rtype = relocInfo::none)
410 : _address((address) addr),
411 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
412
413 AddressLiteral(bool* addr, relocInfo::relocType rtype = relocInfo::none)
414 : _address((address) addr),
415 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
416
417 AddressLiteral(const bool* addr, relocInfo::relocType rtype = relocInfo::none)
418 : _address((address) addr),
419 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
420
421 AddressLiteral(signed char* addr, relocInfo::relocType rtype = relocInfo::none)
422 : _address((address) addr),
423 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
424
425 AddressLiteral(int* addr, relocInfo::relocType rtype = relocInfo::none)
426 : _address((address) addr),
427 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
428
429 AddressLiteral(intptr_t addr, relocInfo::relocType rtype = relocInfo::none)
430 : _address((address) addr),
431 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
432
433 #ifdef _LP64
434 // 32-bit complains about a multiple declaration for int*.
435 AddressLiteral(intptr_t* addr, relocInfo::relocType rtype = relocInfo::none)
436 : _address((address) addr),
437 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
438 #endif
439
440 AddressLiteral(Metadata* addr, relocInfo::relocType rtype = relocInfo::none)
441 : _address((address) addr),
442 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
443
444 AddressLiteral(Metadata** addr, relocInfo::relocType rtype = relocInfo::none)
445 : _address((address) addr),
446 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
447
448 AddressLiteral(float* addr, relocInfo::relocType rtype = relocInfo::none)
449 : _address((address) addr),
450 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
451
452 AddressLiteral(double* addr, relocInfo::relocType rtype = relocInfo::none)
453 : _address((address) addr),
454 _rspec(rspec_from_rtype(rtype, (address) addr)) {}
455
456 intptr_t value() const { return (intptr_t) _address; }
457 int low10() const;
458
459 const relocInfo::relocType rtype() const { return _rspec.type(); }
460 const RelocationHolder& rspec() const { return _rspec; }
461
462 RelocationHolder rspec(int offset) const {
463 return offset == 0 ? _rspec : _rspec.plus(offset);
464 }
465 };
466
467 // Convenience classes
468 class ExternalAddress: public AddressLiteral {
469 private:
470 static relocInfo::relocType reloc_for_target(address target) {
471 // Sometimes ExternalAddress is used for values which aren't
472 // exactly addresses, like the card table base.
473 // external_word_type can't be used for values in the first page
474 // so just skip the reloc in that case.
475 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none;
476 }
477
478 public:
479 ExternalAddress(address target) : AddressLiteral(target, reloc_for_target( target)) {}
480 ExternalAddress(Metadata** target) : AddressLiteral(target, reloc_for_target((address) target)) {}
481 };
482
483 inline Address RegisterImpl::address_in_saved_window() const {
484 return (Address(SP, (sp_offset_in_saved_window() * wordSize) + STACK_BIAS));
485 }
486
487
488
489 // Argument is an abstraction used to represent an outgoing
490 // actual argument or an incoming formal parameter, whether
491 // it resides in memory or in a register, in a manner consistent
492 // with the SPARC Application Binary Interface, or ABI. This is
493 // often referred to as the native or C calling convention.
494
495 class Argument VALUE_OBJ_CLASS_SPEC {
496 private:
497 int _number;
498 bool _is_in;
499
500 public:
501 #ifdef _LP64
502 enum {
503 n_register_parameters = 6, // only 6 registers may contain integer parameters
504 n_float_register_parameters = 16 // Can have up to 16 floating registers
505 };
506 #else
507 enum {
508 n_register_parameters = 6 // only 6 registers may contain integer parameters
509 };
510 #endif
511
512 // creation
513 Argument(int number, bool is_in) : _number(number), _is_in(is_in) {}
514
515 int number() const { return _number; }
516 bool is_in() const { return _is_in; }
517 bool is_out() const { return !is_in(); }
518
519 Argument successor() const { return Argument(number() + 1, is_in()); }
520 Argument as_in() const { return Argument(number(), true ); }
521 Argument as_out() const { return Argument(number(), false); }
522
523 // locating register-based arguments:
524 bool is_register() const { return _number < n_register_parameters; }
525
526 #ifdef _LP64
527 // locating Floating Point register-based arguments:
528 bool is_float_register() const { return _number < n_float_register_parameters; }
529
530 FloatRegister as_float_register() const {
531 assert(is_float_register(), "must be a register argument");
532 return as_FloatRegister(( number() *2 ) + 1);
533 }
534 FloatRegister as_double_register() const {
535 assert(is_float_register(), "must be a register argument");
536 return as_FloatRegister(( number() *2 ));
537 }
538 #endif
539
540 Register as_register() const {
541 assert(is_register(), "must be a register argument");
542 return is_in() ? as_iRegister(number()) : as_oRegister(number());
543 }
544
545 // locating memory-based arguments
546 Address as_address() const {
547 assert(!is_register(), "must be a memory argument");
548 return address_in_frame();
549 }
550
551 // When applied to a register-based argument, give the corresponding address
552 // into the 6-word area "into which callee may store register arguments"
553 // (This is a different place than the corresponding register-save area location.)
554 Address address_in_frame() const;
555
556 // debugging
557 const char* name() const;
558
559 friend class Assembler;
560 };
561
562
563 class RegistersForDebugging : public StackObj {
564 public:
565 intptr_t i[8], l[8], o[8], g[8];
566 float f[32];
567 double d[32];
568
569 void print(outputStream* s);
570
571 static int i_offset(int j) { return offset_of(RegistersForDebugging, i[j]); }
572 static int l_offset(int j) { return offset_of(RegistersForDebugging, l[j]); }
573 static int o_offset(int j) { return offset_of(RegistersForDebugging, o[j]); }
574 static int g_offset(int j) { return offset_of(RegistersForDebugging, g[j]); }
575 static int f_offset(int j) { return offset_of(RegistersForDebugging, f[j]); }
576 static int d_offset(int j) { return offset_of(RegistersForDebugging, d[j / 2]); }
577
578 // gen asm code to save regs
579 static void save_registers(MacroAssembler* a);
580
581 // restore global registers in case C code disturbed them
582 static void restore_registers(MacroAssembler* a, Register r);
583 };
584
585
586 // MacroAssembler extends Assembler by a few frequently used macros.
587 //
588 // Most of the standard SPARC synthetic ops are defined here.
589 // Instructions for which a 'better' code sequence exists depending
590 // on arguments should also go in here.
591
592 #define JMP2(r1, r2) jmp(r1, r2, __FILE__, __LINE__)
593 #define JMP(r1, off) jmp(r1, off, __FILE__, __LINE__)
594 #define JUMP(a, temp, off) jump(a, temp, off, __FILE__, __LINE__)
595 #define JUMPL(a, temp, d, off) jumpl(a, temp, d, off, __FILE__, __LINE__)
596
597
598 class MacroAssembler : public Assembler {
599 // code patchers need various routines like inv_wdisp()
600 friend class NativeInstruction;
601 friend class NativeGeneralJump;
602 friend class Relocation;
603 friend class Label;
604
605 protected:
606 static int patched_branch(int dest_pos, int inst, int inst_pos);
607 static int branch_destination(int inst, int pos);
608
609 // Support for VM calls
610 // This is the base routine called by the different versions of call_VM_leaf. The interpreter
611 // may customize this version by overriding it for its purposes (e.g., to save/restore
612 // additional registers when doing a VM call).
613 #ifdef CC_INTERP
614 #define VIRTUAL
615 #else
616 #define VIRTUAL virtual
617 #endif
618
619 VIRTUAL void call_VM_leaf_base(Register thread_cache, address entry_point, int number_of_arguments);
620
621 //
622 // It is imperative that all calls into the VM are handled via the call_VM macros.
623 // They make sure that the stack linkage is setup correctly. call_VM's correspond
624 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
625 //
626 // This is the base routine called by the different versions of call_VM. The interpreter
627 // may customize this version by overriding it for its purposes (e.g., to save/restore
628 // additional registers when doing a VM call).
629 //
630 // A non-volatile java_thread_cache register should be specified so
631 // that the G2_thread value can be preserved across the call.
632 // (If java_thread_cache is noreg, then a slow get_thread call
633 // will re-initialize the G2_thread.) call_VM_base returns the register that contains the
634 // thread.
635 //
636 // If no last_java_sp is specified (noreg) than SP will be used instead.
637
638 virtual void call_VM_base(
639 Register oop_result, // where an oop-result ends up if any; use noreg otherwise
640 Register java_thread_cache, // the thread if computed before ; use noreg otherwise
641 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise
642 address entry_point, // the entry point
643 int number_of_arguments, // the number of arguments (w/o thread) to pop after call
644 bool check_exception=true // flag which indicates if exception should be checked
645 );
646
647 // This routine should emit JVMTI PopFrame and ForceEarlyReturn handling code.
648 // The implementation is only non-empty for the InterpreterMacroAssembler,
649 // as only the interpreter handles and ForceEarlyReturn PopFrame requests.
650 virtual void check_and_handle_popframe(Register scratch_reg);
651 virtual void check_and_handle_earlyret(Register scratch_reg);
652
653 public:
654 MacroAssembler(CodeBuffer* code) : Assembler(code) {}
655
656 // Support for NULL-checks
657 //
658 // Generates code that causes a NULL OS exception if the content of reg is NULL.
659 // If the accessed location is M[reg + offset] and the offset is known, provide the
660 // offset. No explicit code generation is needed if the offset is within a certain
661 // range (0 <= offset <= page_size).
662 //
663 // %%%%%% Currently not done for SPARC
664
665 void null_check(Register reg, int offset = -1);
666 static bool needs_explicit_null_check(intptr_t offset);
667
668 // support for delayed instructions
669 MacroAssembler* delayed() { Assembler::delayed(); return this; }
670
671 // branches that use right instruction for v8 vs. v9
672 inline void br( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
673 inline void br( Condition c, bool a, Predict p, Label& L );
674
675 inline void fb( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
676 inline void fb( Condition c, bool a, Predict p, Label& L );
677
678 // compares register with zero (32 bit) and branches (V9 and V8 instructions)
679 void cmp_zero_and_br( Condition c, Register s1, Label& L, bool a = false, Predict p = pn );
680 // Compares a pointer register with zero and branches on (not)null.
681 // Does a test & branch on 32-bit systems and a register-branch on 64-bit.
682 void br_null ( Register s1, bool a, Predict p, Label& L );
683 void br_notnull( Register s1, bool a, Predict p, Label& L );
684
685 //
686 // Compare registers and branch with nop in delay slot or cbcond without delay slot.
687 //
688 // ATTENTION: use these instructions with caution because cbcond instruction
689 // has very short distance: 512 instructions (2Kbyte).
690
691 // Compare integer (32 bit) values (icc only).
692 void cmp_and_br_short(Register s1, Register s2, Condition c, Predict p, Label& L);
693 void cmp_and_br_short(Register s1, int simm13a, Condition c, Predict p, Label& L);
694 // Platform depending version for pointer compare (icc on !LP64 and xcc on LP64).
695 void cmp_and_brx_short(Register s1, Register s2, Condition c, Predict p, Label& L);
696 void cmp_and_brx_short(Register s1, int simm13a, Condition c, Predict p, Label& L);
697
698 // Short branch version for compares a pointer pwith zero.
699 void br_null_short ( Register s1, Predict p, Label& L );
700 void br_notnull_short( Register s1, Predict p, Label& L );
701
702 // unconditional short branch
703 void ba_short(Label& L);
704
705 inline void bp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
706 inline void bp( Condition c, bool a, CC cc, Predict p, Label& L );
707
708 // Branch that tests xcc in LP64 and icc in !LP64
709 inline void brx( Condition c, bool a, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
710 inline void brx( Condition c, bool a, Predict p, Label& L );
711
712 // unconditional branch
713 inline void ba( Label& L );
714
715 // Branch that tests fp condition codes
716 inline void fbp( Condition c, bool a, CC cc, Predict p, address d, relocInfo::relocType rt = relocInfo::none );
717 inline void fbp( Condition c, bool a, CC cc, Predict p, Label& L );
718
719 // get PC the best way
720 inline int get_pc( Register d );
721
722 // Sparc shorthands(pp 85, V8 manual, pp 289 V9 manual)
723 inline void cmp( Register s1, Register s2 );
724 inline void cmp( Register s1, int simm13a );
725
726 inline void jmp( Register s1, Register s2 );
727 inline void jmp( Register s1, int simm13a, RelocationHolder const& rspec = RelocationHolder() );
728
729 // Check if the call target is out of wdisp30 range (relative to the code cache)
730 static inline bool is_far_target(address d);
731 inline void call( address d, relocInfo::relocType rt = relocInfo::runtime_call_type );
732 inline void call( address d, RelocationHolder const& rspec);
733
734 inline void call( Label& L, relocInfo::relocType rt = relocInfo::runtime_call_type );
735 inline void call( Label& L, RelocationHolder const& rspec);
736
737 inline void callr( Register s1, Register s2 );
738 inline void callr( Register s1, int simm13a, RelocationHolder const& rspec = RelocationHolder() );
739
740 // Emits nothing on V8
741 inline void iprefetch( address d, relocInfo::relocType rt = relocInfo::none );
742 inline void iprefetch( Label& L);
743
744 inline void tst( Register s );
745
746 inline void ret( bool trace = TraceJumps );
747 inline void retl( bool trace = TraceJumps );
748
749 // Required platform-specific helpers for Label::patch_instructions.
750 // They _shadow_ the declarations in AbstractAssembler, which are undefined.
751 void pd_patch_instruction(address branch, address target);
752
753 // sethi Macro handles optimizations and relocations
754 private:
755 void internal_sethi(const AddressLiteral& addrlit, Register d, bool ForceRelocatable);
756 public:
757 void sethi(const AddressLiteral& addrlit, Register d);
758 void patchable_sethi(const AddressLiteral& addrlit, Register d);
759
760 // compute the number of instructions for a sethi/set
761 static int insts_for_sethi( address a, bool worst_case = false );
762 static int worst_case_insts_for_set();
763
764 // set may be either setsw or setuw (high 32 bits may be zero or sign)
765 private:
766 void internal_set(const AddressLiteral& al, Register d, bool ForceRelocatable);
767 static int insts_for_internal_set(intptr_t value);
768 public:
769 void set(const AddressLiteral& addrlit, Register d);
770 void set(intptr_t value, Register d);
771 void set(address addr, Register d, RelocationHolder const& rspec);
772 static int insts_for_set(intptr_t value) { return insts_for_internal_set(value); }
773
774 void patchable_set(const AddressLiteral& addrlit, Register d);
775 void patchable_set(intptr_t value, Register d);
776 void set64(jlong value, Register d, Register tmp);
777 static int insts_for_set64(jlong value);
778
779 // sign-extend 32 to 64
780 inline void signx( Register s, Register d );
781 inline void signx( Register d );
782
783 inline void not1( Register s, Register d );
784 inline void not1( Register d );
785
786 inline void neg( Register s, Register d );
787 inline void neg( Register d );
788
789 inline void cas( Register s1, Register s2, Register d);
790 inline void casx( Register s1, Register s2, Register d);
791 // Functions for isolating 64 bit atomic swaps for LP64
792 // cas_ptr will perform cas for 32 bit VM's and casx for 64 bit VM's
793 inline void cas_ptr( Register s1, Register s2, Register d);
794
795 // Functions for isolating 64 bit shifts for LP64
796 inline void sll_ptr( Register s1, Register s2, Register d );
797 inline void sll_ptr( Register s1, int imm6a, Register d );
798 inline void sll_ptr( Register s1, RegisterOrConstant s2, Register d );
799 inline void srl_ptr( Register s1, Register s2, Register d );
800 inline void srl_ptr( Register s1, int imm6a, Register d );
801
802 // little-endian
803 inline void casl( Register s1, Register s2, Register d);
804 inline void casxl( Register s1, Register s2, Register d);
805
806 inline void inc( Register d, int const13 = 1 );
807 inline void inccc( Register d, int const13 = 1 );
808
809 inline void dec( Register d, int const13 = 1 );
810 inline void deccc( Register d, int const13 = 1 );
811
812 using Assembler::add;
813 inline void add(Register s1, int simm13a, Register d, relocInfo::relocType rtype);
814 inline void add(Register s1, int simm13a, Register d, RelocationHolder const& rspec);
815 inline void add(Register s1, RegisterOrConstant s2, Register d, int offset = 0);
816 inline void add(const Address& a, Register d, int offset = 0);
817
818 using Assembler::andn;
819 inline void andn( Register s1, RegisterOrConstant s2, Register d);
820
821 inline void btst( Register s1, Register s2 );
822 inline void btst( int simm13a, Register s );
823
824 inline void bset( Register s1, Register s2 );
825 inline void bset( int simm13a, Register s );
826
827 inline void bclr( Register s1, Register s2 );
828 inline void bclr( int simm13a, Register s );
829
830 inline void btog( Register s1, Register s2 );
831 inline void btog( int simm13a, Register s );
832
833 inline void clr( Register d );
834
835 inline void clrb( Register s1, Register s2);
836 inline void clrh( Register s1, Register s2);
837 inline void clr( Register s1, Register s2);
838 inline void clrx( Register s1, Register s2);
839
840 inline void clrb( Register s1, int simm13a);
841 inline void clrh( Register s1, int simm13a);
842 inline void clr( Register s1, int simm13a);
843 inline void clrx( Register s1, int simm13a);
844
845 // copy & clear upper word
846 inline void clruw( Register s, Register d );
847 // clear upper word
848 inline void clruwu( Register d );
849
850 using Assembler::ldsb;
851 using Assembler::ldsh;
852 using Assembler::ldsw;
853 using Assembler::ldub;
854 using Assembler::lduh;
855 using Assembler::lduw;
856 using Assembler::ldx;
857 using Assembler::ldd;
858
859 #ifdef ASSERT
860 // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
861 inline void ld(Register s1, ByteSize simm13a, Register d);
862 #endif
863
864 inline void ld(Register s1, Register s2, Register d);
865 inline void ld(Register s1, int simm13a, Register d);
866
867 inline void ldsb(const Address& a, Register d, int offset = 0);
868 inline void ldsh(const Address& a, Register d, int offset = 0);
869 inline void ldsw(const Address& a, Register d, int offset = 0);
870 inline void ldub(const Address& a, Register d, int offset = 0);
871 inline void lduh(const Address& a, Register d, int offset = 0);
872 inline void lduw(const Address& a, Register d, int offset = 0);
873 inline void ldx( const Address& a, Register d, int offset = 0);
874 inline void ld( const Address& a, Register d, int offset = 0);
875 inline void ldd( const Address& a, Register d, int offset = 0);
876
877 inline void ldub(Register s1, RegisterOrConstant s2, Register d );
878 inline void ldsb(Register s1, RegisterOrConstant s2, Register d );
879 inline void lduh(Register s1, RegisterOrConstant s2, Register d );
880 inline void ldsh(Register s1, RegisterOrConstant s2, Register d );
881 inline void lduw(Register s1, RegisterOrConstant s2, Register d );
882 inline void ldsw(Register s1, RegisterOrConstant s2, Register d );
883 inline void ldx( Register s1, RegisterOrConstant s2, Register d );
884 inline void ld( Register s1, RegisterOrConstant s2, Register d );
885 inline void ldd( Register s1, RegisterOrConstant s2, Register d );
886
887 using Assembler::ldf;
888 inline void ldf(FloatRegisterImpl::Width w, Register s1, RegisterOrConstant s2, FloatRegister d);
889 inline void ldf(FloatRegisterImpl::Width w, const Address& a, FloatRegister d, int offset = 0);
890
891 // little-endian
892 inline void lduwl(Register s1, Register s2, Register d);
893 inline void ldswl(Register s1, Register s2, Register d);
894 inline void ldxl( Register s1, Register s2, Register d);
895 inline void ldfl(FloatRegisterImpl::Width w, Register s1, Register s2, FloatRegister d);
896
897 // membar psuedo instruction. takes into account target memory model.
898 inline void membar( Assembler::Membar_mask_bits const7a );
899
900 // returns if membar generates anything.
901 inline bool membar_has_effect( Assembler::Membar_mask_bits const7a );
902
903 // mov pseudo instructions
904 inline void mov( Register s, Register d);
905
906 inline void mov_or_nop( Register s, Register d);
907
908 inline void mov( int simm13a, Register d);
909
910 using Assembler::prefetch;
911 inline void prefetch(const Address& a, PrefetchFcn F, int offset = 0);
912
913 using Assembler::stb;
914 using Assembler::sth;
915 using Assembler::stw;
916 using Assembler::stx;
917 using Assembler::std;
918
919 #ifdef ASSERT
920 // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
921 inline void st(Register d, Register s1, ByteSize simm13a);
922 #endif
923
924 inline void st(Register d, Register s1, Register s2);
925 inline void st(Register d, Register s1, int simm13a);
926
927 inline void stb(Register d, const Address& a, int offset = 0 );
928 inline void sth(Register d, const Address& a, int offset = 0 );
929 inline void stw(Register d, const Address& a, int offset = 0 );
930 inline void stx(Register d, const Address& a, int offset = 0 );
931 inline void st( Register d, const Address& a, int offset = 0 );
932 inline void std(Register d, const Address& a, int offset = 0 );
933
934 inline void stb(Register d, Register s1, RegisterOrConstant s2 );
935 inline void sth(Register d, Register s1, RegisterOrConstant s2 );
936 inline void stw(Register d, Register s1, RegisterOrConstant s2 );
937 inline void stx(Register d, Register s1, RegisterOrConstant s2 );
938 inline void std(Register d, Register s1, RegisterOrConstant s2 );
939 inline void st( Register d, Register s1, RegisterOrConstant s2 );
940
941 using Assembler::stf;
942 inline void stf(FloatRegisterImpl::Width w, FloatRegister d, Register s1, RegisterOrConstant s2);
943 inline void stf(FloatRegisterImpl::Width w, FloatRegister d, const Address& a, int offset = 0);
944
945 // Note: offset is added to s2.
946 using Assembler::sub;
947 inline void sub(Register s1, RegisterOrConstant s2, Register d, int offset = 0);
948
949 using Assembler::swap;
950 inline void swap(const Address& a, Register d, int offset = 0);
951
952 // address pseudos: make these names unlike instruction names to avoid confusion
953 inline intptr_t load_pc_address( Register reg, int bytes_to_skip );
954 inline void load_contents(const AddressLiteral& addrlit, Register d, int offset = 0);
955 inline void load_bool_contents(const AddressLiteral& addrlit, Register d, int offset = 0);
956 inline void load_ptr_contents(const AddressLiteral& addrlit, Register d, int offset = 0);
957 inline void store_contents(Register s, const AddressLiteral& addrlit, Register temp, int offset = 0);
958 inline void store_ptr_contents(Register s, const AddressLiteral& addrlit, Register temp, int offset = 0);
959 inline void jumpl_to(const AddressLiteral& addrlit, Register temp, Register d, int offset = 0);
960 inline void jump_to(const AddressLiteral& addrlit, Register temp, int offset = 0);
961 inline void jump_indirect_to(Address& a, Register temp, int ld_offset = 0, int jmp_offset = 0);
962
963 // ring buffer traceable jumps
964
965 void jmp2( Register r1, Register r2, const char* file, int line );
966 void jmp ( Register r1, int offset, const char* file, int line );
967
968 void jumpl(const AddressLiteral& addrlit, Register temp, Register d, int offset, const char* file, int line);
969 void jump (const AddressLiteral& addrlit, Register temp, int offset, const char* file, int line);
970
971
972 // argument pseudos:
973
974 inline void load_argument( Argument& a, Register d );
975 inline void store_argument( Register s, Argument& a );
976 inline void store_ptr_argument( Register s, Argument& a );
977 inline void store_float_argument( FloatRegister s, Argument& a );
978 inline void store_double_argument( FloatRegister s, Argument& a );
979 inline void store_long_argument( Register s, Argument& a );
980
981 // handy macros:
982
983 inline void round_to( Register r, int modulus );
984
985 // --------------------------------------------------
986
987 // Functions for isolating 64 bit loads for LP64
988 // ld_ptr will perform ld for 32 bit VM's and ldx for 64 bit VM's
989 // st_ptr will perform st for 32 bit VM's and stx for 64 bit VM's
990 inline void ld_ptr(Register s1, Register s2, Register d);
991 inline void ld_ptr(Register s1, int simm13a, Register d);
992 inline void ld_ptr(Register s1, RegisterOrConstant s2, Register d);
993 inline void ld_ptr(const Address& a, Register d, int offset = 0);
994 inline void st_ptr(Register d, Register s1, Register s2);
995 inline void st_ptr(Register d, Register s1, int simm13a);
996 inline void st_ptr(Register d, Register s1, RegisterOrConstant s2);
997 inline void st_ptr(Register d, const Address& a, int offset = 0);
998
999 #ifdef ASSERT
1000 // ByteSize is only a class when ASSERT is defined, otherwise it's an int.
1001 inline void ld_ptr(Register s1, ByteSize simm13a, Register d);
1002 inline void st_ptr(Register d, Register s1, ByteSize simm13a);
1003 #endif
1004
1005 // ld_long will perform ldd for 32 bit VM's and ldx for 64 bit VM's
1006 // st_long will perform std for 32 bit VM's and stx for 64 bit VM's
1007 inline void ld_long(Register s1, Register s2, Register d);
1008 inline void ld_long(Register s1, int simm13a, Register d);
1009 inline void ld_long(Register s1, RegisterOrConstant s2, Register d);
1010 inline void ld_long(const Address& a, Register d, int offset = 0);
1011 inline void st_long(Register d, Register s1, Register s2);
1012 inline void st_long(Register d, Register s1, int simm13a);
1013 inline void st_long(Register d, Register s1, RegisterOrConstant s2);
1014 inline void st_long(Register d, const Address& a, int offset = 0);
1015
1016 // Helpers for address formation.
1017 // - They emit only a move if s2 is a constant zero.
1018 // - If dest is a constant and either s1 or s2 is a register, the temp argument is required and becomes the result.
1019 // - If dest is a register and either s1 or s2 is a non-simm13 constant, the temp argument is required and used to materialize the constant.
1020 RegisterOrConstant regcon_andn_ptr(RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp = noreg);
1021 RegisterOrConstant regcon_inc_ptr( RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp = noreg);
1022 RegisterOrConstant regcon_sll_ptr( RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp = noreg);
1023
1024 RegisterOrConstant ensure_simm13_or_reg(RegisterOrConstant src, Register temp) {
1025 if (is_simm13(src.constant_or_zero()))
1026 return src; // register or short constant
1027 guarantee(temp != noreg, "constant offset overflow");
1028 set(src.as_constant(), temp);
1029 return temp;
1030 }
1031
1032 // --------------------------------------------------
1033
1034 public:
1035 // traps as per trap.h (SPARC ABI?)
1036
1037 void breakpoint_trap();
1038 void breakpoint_trap(Condition c, CC cc);
1039
1040 // Support for serializing memory accesses between threads
1041 void serialize_memory(Register thread, Register tmp1, Register tmp2);
1042
1043 // Stack frame creation/removal
1044 void enter();
1045 void leave();
1046
1047 // Manipulation of C++ bools
1048 // These are idioms to flag the need for care with accessing bools but on
1049 // this platform we assume byte size
1050
1051 inline void stbool(Register d, const Address& a);
1052 inline void ldbool(const Address& a, Register d);
1053 inline void movbool( bool boolconst, Register d);
1054
1055 // klass oop manipulations if compressed
1056 void load_klass(Register src_oop, Register klass);
1057 void store_klass(Register klass, Register dst_oop);
1058 void store_klass_gap(Register s, Register dst_oop);
1059
1060 // oop manipulations
1061 void load_heap_oop(const Address& s, Register d);
1062 void load_heap_oop(Register s1, Register s2, Register d);
1063 void load_heap_oop(Register s1, int simm13a, Register d);
1064 void load_heap_oop(Register s1, RegisterOrConstant s2, Register d);
1065 void store_heap_oop(Register d, Register s1, Register s2);
1066 void store_heap_oop(Register d, Register s1, int simm13a);
1067 void store_heap_oop(Register d, const Address& a, int offset = 0);
1068
1069 void encode_heap_oop(Register src, Register dst);
1070 void encode_heap_oop(Register r) {
1071 encode_heap_oop(r, r);
1072 }
1073 void decode_heap_oop(Register src, Register dst);
1074 void decode_heap_oop(Register r) {
1075 decode_heap_oop(r, r);
1076 }
1077 void encode_heap_oop_not_null(Register r);
1078 void decode_heap_oop_not_null(Register r);
1079 void encode_heap_oop_not_null(Register src, Register dst);
1080 void decode_heap_oop_not_null(Register src, Register dst);
1081
1082 void encode_klass_not_null(Register r);
1083 void decode_klass_not_null(Register r);
1084 void encode_klass_not_null(Register src, Register dst);
1085 void decode_klass_not_null(Register src, Register dst);
1086
1087 // Support for managing the JavaThread pointer (i.e.; the reference to
1088 // thread-local information).
1089 void get_thread(); // load G2_thread
1090 void verify_thread(); // verify G2_thread contents
1091 void save_thread (const Register threache); // save to cache
1092 void restore_thread(const Register thread_cache); // restore from cache
1093
1094 // Support for last Java frame (but use call_VM instead where possible)
1095 void set_last_Java_frame(Register last_java_sp, Register last_Java_pc);
1096 void reset_last_Java_frame(void);
1097
1098 // Call into the VM.
1099 // Passes the thread pointer (in O0) as a prepended argument.
1100 // Makes sure oop return values are visible to the GC.
1101 void call_VM(Register oop_result, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
1102 void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
1103 void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
1104 void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
1105
1106 // these overloadings are not presently used on SPARC:
1107 void call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
1108 void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
1109 void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
1110 void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
1111
1112 void call_VM_leaf(Register thread_cache, address entry_point, int number_of_arguments = 0);
1113 void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1);
1114 void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2);
1115 void call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2, Register arg_3);
1116
1117 void get_vm_result (Register oop_result);
1118 void get_vm_result_2(Register metadata_result);
1119
1120 // vm result is currently getting hijacked to for oop preservation
1121 void set_vm_result(Register oop_result);
1122
1123 // Emit the CompiledIC call idiom
1124 void ic_call(address entry, bool emit_delay = true, jint method_index = 0);
1125
1126 // if call_VM_base was called with check_exceptions=false, then call
1127 // check_and_forward_exception to handle exceptions when it is safe
1128 void check_and_forward_exception(Register scratch_reg);
1129
1130 // Write to card table for - register is destroyed afterwards.
1131 void card_table_write(jbyte* byte_map_base, Register tmp, Register obj);
1132
1133 void card_write_barrier_post(Register store_addr, Register new_val, Register tmp);
1134
1135 #if INCLUDE_ALL_GCS
1136 // General G1 pre-barrier generator.
1137 void g1_write_barrier_pre(Register obj, Register index, int offset, Register pre_val, Register tmp, bool preserve_o_regs);
1138
1139 // General G1 post-barrier generator
1140 void g1_write_barrier_post(Register store_addr, Register new_val, Register tmp);
1141 #endif // INCLUDE_ALL_GCS
1142
1143 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
1144 void push_fTOS();
1145
1146 // pops double TOS element from CPU stack and pushes on FPU stack
1147 void pop_fTOS();
1148
1149 void empty_FPU_stack();
1150
1151 void push_IU_state();
1152 void pop_IU_state();
1153
1154 void push_FPU_state();
1155 void pop_FPU_state();
1156
1157 void push_CPU_state();
1158 void pop_CPU_state();
1159
1160 // Returns the byte size of the instructions generated by decode_klass_not_null().
1161 static int instr_size_for_decode_klass_not_null();
1162
1163 // if heap base register is used - reinit it with the correct value
1164 void reinit_heapbase();
1165
1166 // Debugging
1167 void _verify_oop(Register reg, const char * msg, const char * file, int line);
1168 void _verify_oop_addr(Address addr, const char * msg, const char * file, int line);
1169
1170 // TODO: verify_method and klass metadata (compare against vptr?)
1171 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1172 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1173
1174 #define verify_oop(reg) _verify_oop(reg, "broken oop " #reg, __FILE__, __LINE__)
1175 #define verify_oop_addr(addr) _verify_oop_addr(addr, "broken oop addr ", __FILE__, __LINE__)
1176 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1177 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1178
1179 // only if +VerifyOops
1180 void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
1181 // only if +VerifyFPU
1182 void stop(const char* msg); // prints msg, dumps registers and stops execution
1183 void warn(const char* msg); // prints msg, but don't stop
1184 void untested(const char* what = "");
1185 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); }
1186 void should_not_reach_here() { stop("should not reach here"); }
1187 void print_CPU_state();
1188
1189 // oops in code
1190 AddressLiteral allocate_oop_address(jobject obj); // allocate_index
1191 AddressLiteral constant_oop_address(jobject obj); // find_index
1192 inline void set_oop (jobject obj, Register d); // uses allocate_oop_address
1193 inline void set_oop_constant (jobject obj, Register d); // uses constant_oop_address
1194 inline void set_oop (const AddressLiteral& obj_addr, Register d); // same as load_address
1195
1196 // metadata in code that we have to keep track of
1197 AddressLiteral allocate_metadata_address(Metadata* obj); // allocate_index
1198 AddressLiteral constant_metadata_address(Metadata* obj); // find_index
1199 inline void set_metadata (Metadata* obj, Register d); // uses allocate_metadata_address
1200 inline void set_metadata_constant (Metadata* obj, Register d); // uses constant_metadata_address
1201 inline void set_metadata (const AddressLiteral& obj_addr, Register d); // same as load_address
1202
1203 void set_narrow_oop( jobject obj, Register d );
1204 void set_narrow_klass( Klass* k, Register d );
1205
1206 // nop padding
1207 void align(int modulus);
1208
1209 // declare a safepoint
1210 void safepoint();
1211
1212 // factor out part of stop into subroutine to save space
1213 void stop_subroutine();
1214 // factor out part of verify_oop into subroutine to save space
1215 void verify_oop_subroutine();
1216
1217 // side-door communication with signalHandler in os_solaris.cpp
1218 static address _verify_oop_implicit_branch[3];
1219
1220 int total_frame_size_in_bytes(int extraWords);
1221
1222 // used when extraWords known statically
1223 void save_frame(int extraWords = 0);
1224 void save_frame_c1(int size_in_bytes);
1225 // make a frame, and simultaneously pass up one or two register value
1226 // into the new register window
1227 void save_frame_and_mov(int extraWords, Register s1, Register d1, Register s2 = Register(), Register d2 = Register());
1228
1229 // give no. (outgoing) params, calc # of words will need on frame
1230 void calc_mem_param_words(Register Rparam_words, Register Rresult);
1231
1232 // used to calculate frame size dynamically
1233 // result is in bytes and must be negated for save inst
1234 void calc_frame_size(Register extraWords, Register resultReg);
1235
1236 // calc and also save
1237 void calc_frame_size_and_save(Register extraWords, Register resultReg);
1238
1239 static void debug(char* msg, RegistersForDebugging* outWindow);
1240
1241 // implementations of bytecodes used by both interpreter and compiler
1242
1243 void lcmp( Register Ra_hi, Register Ra_low,
1244 Register Rb_hi, Register Rb_low,
1245 Register Rresult);
1246
1247 void lneg( Register Rhi, Register Rlow );
1248
1249 void lshl( Register Rin_high, Register Rin_low, Register Rcount,
1250 Register Rout_high, Register Rout_low, Register Rtemp );
1251
1252 void lshr( Register Rin_high, Register Rin_low, Register Rcount,
1253 Register Rout_high, Register Rout_low, Register Rtemp );
1254
1255 void lushr( Register Rin_high, Register Rin_low, Register Rcount,
1256 Register Rout_high, Register Rout_low, Register Rtemp );
1257
1258 #ifdef _LP64
1259 void lcmp( Register Ra, Register Rb, Register Rresult);
1260 #endif
1261
1262 // Load and store values by size and signed-ness
1263 void load_sized_value( Address src, Register dst, size_t size_in_bytes, bool is_signed);
1264 void store_sized_value(Register src, Address dst, size_t size_in_bytes);
1265
1266 void float_cmp( bool is_float, int unordered_result,
1267 FloatRegister Fa, FloatRegister Fb,
1268 Register Rresult);
1269
1270 void save_all_globals_into_locals();
1271 void restore_globals_from_locals();
1272
1273 // These set the icc condition code to equal if the lock succeeded
1274 // and notEqual if it failed and requires a slow case
1275 void compiler_lock_object(Register Roop, Register Rmark, Register Rbox,
1276 Register Rscratch,
1277 BiasedLockingCounters* counters = NULL,
1278 bool try_bias = UseBiasedLocking);
1279 void compiler_unlock_object(Register Roop, Register Rmark, Register Rbox,
1280 Register Rscratch,
1281 bool try_bias = UseBiasedLocking);
1282
1283 // Biased locking support
1284 // Upon entry, lock_reg must point to the lock record on the stack,
1285 // obj_reg must contain the target object, and mark_reg must contain
1286 // the target object's header.
1287 // Destroys mark_reg if an attempt is made to bias an anonymously
1288 // biased lock. In this case a failure will go either to the slow
1289 // case or fall through with the notEqual condition code set with
1290 // the expectation that the slow case in the runtime will be called.
1291 // In the fall-through case where the CAS-based lock is done,
1292 // mark_reg is not destroyed.
1293 void biased_locking_enter(Register obj_reg, Register mark_reg, Register temp_reg,
1294 Label& done, Label* slow_case = NULL,
1295 BiasedLockingCounters* counters = NULL);
1296 // Upon entry, the base register of mark_addr must contain the oop.
1297 // Destroys temp_reg.
1298
1299 // If allow_delay_slot_filling is set to true, the next instruction
1300 // emitted after this one will go in an annulled delay slot if the
1301 // biased locking exit case failed.
1302 void biased_locking_exit(Address mark_addr, Register temp_reg, Label& done, bool allow_delay_slot_filling = false);
1303
1304 // allocation
1305 void eden_allocate(
1306 Register obj, // result: pointer to object after successful allocation
1307 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
1308 int con_size_in_bytes, // object size in bytes if known at compile time
1309 Register t1, // temp register
1310 Register t2, // temp register
1311 Label& slow_case // continuation point if fast allocation fails
1312 );
1313 void tlab_allocate(
1314 Register obj, // result: pointer to object after successful allocation
1315 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
1316 int con_size_in_bytes, // object size in bytes if known at compile time
1317 Register t1, // temp register
1318 Label& slow_case // continuation point if fast allocation fails
1319 );
1320 void tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case);
1321 void incr_allocated_bytes(RegisterOrConstant size_in_bytes,
1322 Register t1, Register t2);
1323
1324 // interface method calling
1325 void lookup_interface_method(Register recv_klass,
1326 Register intf_klass,
1327 RegisterOrConstant itable_index,
1328 Register method_result,
1329 Register temp_reg, Register temp2_reg,
1330 Label& no_such_interface);
1331
1332 // virtual method calling
1333 void lookup_virtual_method(Register recv_klass,
1334 RegisterOrConstant vtable_index,
1335 Register method_result);
1336
1337 // Test sub_klass against super_klass, with fast and slow paths.
1338
1339 // The fast path produces a tri-state answer: yes / no / maybe-slow.
1340 // One of the three labels can be NULL, meaning take the fall-through.
1341 // If super_check_offset is -1, the value is loaded up from super_klass.
1342 // No registers are killed, except temp_reg and temp2_reg.
1343 // If super_check_offset is not -1, temp2_reg is not used and can be noreg.
1344 void check_klass_subtype_fast_path(Register sub_klass,
1345 Register super_klass,
1346 Register temp_reg,
1347 Register temp2_reg,
1348 Label* L_success,
1349 Label* L_failure,
1350 Label* L_slow_path,
1351 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
1352
1353 // The rest of the type check; must be wired to a corresponding fast path.
1354 // It does not repeat the fast path logic, so don't use it standalone.
1355 // The temp_reg can be noreg, if no temps are available.
1356 // It can also be sub_klass or super_klass, meaning it's OK to kill that one.
1357 // Updates the sub's secondary super cache as necessary.
1358 void check_klass_subtype_slow_path(Register sub_klass,
1359 Register super_klass,
1360 Register temp_reg,
1361 Register temp2_reg,
1362 Register temp3_reg,
1363 Register temp4_reg,
1364 Label* L_success,
1365 Label* L_failure);
1366
1367 // Simplified, combined version, good for typical uses.
1368 // Falls through on failure.
1369 void check_klass_subtype(Register sub_klass,
1370 Register super_klass,
1371 Register temp_reg,
1372 Register temp2_reg,
1373 Label& L_success);
1374
1375 // method handles (JSR 292)
1376 // offset relative to Gargs of argument at tos[arg_slot].
1377 // (arg_slot == 0 means the last argument, not the first).
1378 RegisterOrConstant argument_offset(RegisterOrConstant arg_slot,
1379 Register temp_reg,
1380 int extra_slot_offset = 0);
1381 // Address of Gargs and argument_offset.
1382 Address argument_address(RegisterOrConstant arg_slot,
1383 Register temp_reg = noreg,
1384 int extra_slot_offset = 0);
1385
1386 // Stack overflow checking
1387
1388 // Note: this clobbers G3_scratch
1389 inline void bang_stack_with_offset(int offset);
1390
1391 // Writes to stack successive pages until offset reached to check for
1392 // stack overflow + shadow pages. Clobbers tsp and scratch registers.
1393 void bang_stack_size(Register Rsize, Register Rtsp, Register Rscratch);
1394
1395 // Check for reserved stack access in method being exited (for JIT)
1396 void reserved_stack_check();
1397
1398 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, Register tmp, int offset);
1399
1400 void verify_tlab();
1401
1402 Condition negate_condition(Condition cond);
1403
1404 // Helper functions for statistics gathering.
1405 // Conditionally (non-atomically) increments passed counter address, preserving condition codes.
1406 void cond_inc(Condition cond, address counter_addr, Register Rtemp1, Register Rtemp2);
1407 // Unconditional increment.
1408 void inc_counter(address counter_addr, Register Rtmp1, Register Rtmp2);
1409 void inc_counter(int* counter_addr, Register Rtmp1, Register Rtmp2);
1410
1411 #ifdef COMPILER2
1412 // Compress char[] to byte[] by compressing 16 bytes at once. Return 0 on failure.
1413 void string_compress_16(Register src, Register dst, Register cnt, Register result,
1414 Register tmp1, Register tmp2, Register tmp3, Register tmp4,
1415 FloatRegister ftmp1, FloatRegister ftmp2, FloatRegister ftmp3, Label& Ldone);
1416
1417 // Compress char[] to byte[]. Return 0 on failure.
1418 void string_compress(Register src, Register dst, Register cnt, Register tmp, Register result, Label& Ldone);
1419
1420 // Inflate byte[] to char[] by inflating 16 bytes at once.
1421 void string_inflate_16(Register src, Register dst, Register cnt, Register tmp,
1422 FloatRegister ftmp1, FloatRegister ftmp2, FloatRegister ftmp3, FloatRegister ftmp4, Label& Ldone);
1423
1424 // Inflate byte[] to char[].
1425 void string_inflate(Register src, Register dst, Register cnt, Register tmp, Label& Ldone);
1426
1427 void string_compare(Register str1, Register str2,
1428 Register cnt1, Register cnt2,
1429 Register tmp1, Register tmp2,
1430 Register result, int ae);
1431
1432 void array_equals(bool is_array_equ, Register ary1, Register ary2,
1433 Register limit, Register tmp, Register result, bool is_byte);
1434 #endif
1435
1436 // Use BIS for zeroing
1437 void bis_zeroing(Register to, Register count, Register temp, Label& Ldone);
1438
1439 // Update CRC-32[C] with a byte value according to constants in table
1440 void update_byte_crc32(Register crc, Register val, Register table);
1441
1442 // Reverse byte order of lower 32 bits, assuming upper 32 bits all zeros
1443 void reverse_bytes_32(Register src, Register dst, Register tmp);
1444 void movitof_revbytes(Register src, FloatRegister dst, Register tmp1, Register tmp2);
1445 void movftoi_revbytes(FloatRegister src, Register dst, Register tmp1, Register tmp2);
1446
1447 // CRC32 code for java.util.zip.CRC32::updateBytes0() instrinsic.
1448 void kernel_crc32(Register crc, Register buf, Register len, Register table);
1449 // Fold 128-bit data chunk
1450 void fold_128bit_crc32(Register xcrc_hi, Register xcrc_lo, Register xK_hi, Register xK_lo, Register xtmp_hi, Register xtmp_lo, Register buf, int offset);
1451 void fold_128bit_crc32(Register xcrc_hi, Register xcrc_lo, Register xK_hi, Register xK_lo, Register xtmp_hi, Register xtmp_lo, Register xbuf_hi, Register xbuf_lo);
1452 // Fold 8-bit data
1453 void fold_8bit_crc32(Register xcrc, Register table, Register xtmp, Register tmp);
1454 void fold_8bit_crc32(Register crc, Register table, Register tmp);
1455
1456 #undef VIRTUAL
1457 };
1458
1459 /**
1460 * class SkipIfEqual:
1461 *
1462 * Instantiating this class will result in assembly code being output that will
1463 * jump around any code emitted between the creation of the instance and it's
1464 * automatic destruction at the end of a scope block, depending on the value of
1465 * the flag passed to the constructor, which will be checked at run-time.
1466 */
1467 class SkipIfEqual : public StackObj {
1468 private:
1469 MacroAssembler* _masm;
1470 Label _label;
1471
1472 public:
1473 // 'temp' is a temp register that this object can use (and trash)
1474 SkipIfEqual(MacroAssembler*, Register temp,
1475 const bool* flag_addr, Assembler::Condition condition);
1476 ~SkipIfEqual();
1477 };
1478
1479 #endif // CPU_SPARC_VM_MACROASSEMBLER_SPARC_HPP