Mercurial > hg > truffle
annotate src/cpu/x86/vm/assembler_x86_64.hpp @ 342:37f87013dfd8
6711316: Open source the Garbage-First garbage collector
Summary: First mercurial integration of the code for the Garbage-First garbage collector.
Reviewed-by: apetrusenko, iveresov, jmasa, sgoldman, tonyp, ysr
author | ysr |
---|---|
date | Thu, 05 Jun 2008 15:57:56 -0700 |
parents | b130b98db9cf |
children | 6aae2f9d0294 |
rev | line source |
---|---|
0 | 1 /* |
2 * Copyright 2003-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
21 * have any questions. | |
22 * | |
23 */ | |
24 | |
25 class BiasedLockingCounters; | |
26 | |
27 // Contains all the definitions needed for amd64 assembly code generation. | |
28 | |
29 #ifdef _LP64 | |
30 // Calling convention | |
31 class Argument VALUE_OBJ_CLASS_SPEC { | |
32 public: | |
33 enum { | |
34 #ifdef _WIN64 | |
35 n_int_register_parameters_c = 4, // rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...) | |
36 n_float_register_parameters_c = 4, // xmm0 - xmm3 (c_farg0, c_farg1, ... ) | |
37 #else | |
38 n_int_register_parameters_c = 6, // rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...) | |
39 n_float_register_parameters_c = 8, // xmm0 - xmm7 (c_farg0, c_farg1, ... ) | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
40 #endif // _WIN64 |
0 | 41 n_int_register_parameters_j = 6, // j_rarg0, j_rarg1, ... |
42 n_float_register_parameters_j = 8 // j_farg0, j_farg1, ... | |
43 }; | |
44 }; | |
45 | |
46 | |
47 // Symbolically name the register arguments used by the c calling convention. | |
48 // Windows is different from linux/solaris. So much for standards... | |
49 | |
50 #ifdef _WIN64 | |
51 | |
52 REGISTER_DECLARATION(Register, c_rarg0, rcx); | |
53 REGISTER_DECLARATION(Register, c_rarg1, rdx); | |
54 REGISTER_DECLARATION(Register, c_rarg2, r8); | |
55 REGISTER_DECLARATION(Register, c_rarg3, r9); | |
56 | |
57 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); | |
58 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); | |
59 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); | |
60 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); | |
61 | |
62 #else | |
63 | |
64 REGISTER_DECLARATION(Register, c_rarg0, rdi); | |
65 REGISTER_DECLARATION(Register, c_rarg1, rsi); | |
66 REGISTER_DECLARATION(Register, c_rarg2, rdx); | |
67 REGISTER_DECLARATION(Register, c_rarg3, rcx); | |
68 REGISTER_DECLARATION(Register, c_rarg4, r8); | |
69 REGISTER_DECLARATION(Register, c_rarg5, r9); | |
70 | |
71 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); | |
72 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); | |
73 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); | |
74 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); | |
75 REGISTER_DECLARATION(XMMRegister, c_farg4, xmm4); | |
76 REGISTER_DECLARATION(XMMRegister, c_farg5, xmm5); | |
77 REGISTER_DECLARATION(XMMRegister, c_farg6, xmm6); | |
78 REGISTER_DECLARATION(XMMRegister, c_farg7, xmm7); | |
79 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
80 #endif // _WIN64 |
0 | 81 |
82 // Symbolically name the register arguments used by the Java calling convention. | |
83 // We have control over the convention for java so we can do what we please. | |
84 // What pleases us is to offset the java calling convention so that when | |
85 // we call a suitable jni method the arguments are lined up and we don't | |
86 // have to do little shuffling. A suitable jni method is non-static and a | |
87 // small number of arguments (two fewer args on windows) | |
88 // | |
89 // |-------------------------------------------------------| | |
90 // | c_rarg0 c_rarg1 c_rarg2 c_rarg3 c_rarg4 c_rarg5 | | |
91 // |-------------------------------------------------------| | |
92 // | rcx rdx r8 r9 rdi* rsi* | windows (* not a c_rarg) | |
93 // | rdi rsi rdx rcx r8 r9 | solaris/linux | |
94 // |-------------------------------------------------------| | |
95 // | j_rarg5 j_rarg0 j_rarg1 j_rarg2 j_rarg3 j_rarg4 | | |
96 // |-------------------------------------------------------| | |
97 | |
98 REGISTER_DECLARATION(Register, j_rarg0, c_rarg1); | |
99 REGISTER_DECLARATION(Register, j_rarg1, c_rarg2); | |
100 REGISTER_DECLARATION(Register, j_rarg2, c_rarg3); | |
101 // Windows runs out of register args here | |
102 #ifdef _WIN64 | |
103 REGISTER_DECLARATION(Register, j_rarg3, rdi); | |
104 REGISTER_DECLARATION(Register, j_rarg4, rsi); | |
105 #else | |
106 REGISTER_DECLARATION(Register, j_rarg3, c_rarg4); | |
107 REGISTER_DECLARATION(Register, j_rarg4, c_rarg5); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
108 #endif // _WIN64 |
0 | 109 REGISTER_DECLARATION(Register, j_rarg5, c_rarg0); |
110 | |
111 REGISTER_DECLARATION(XMMRegister, j_farg0, xmm0); | |
112 REGISTER_DECLARATION(XMMRegister, j_farg1, xmm1); | |
113 REGISTER_DECLARATION(XMMRegister, j_farg2, xmm2); | |
114 REGISTER_DECLARATION(XMMRegister, j_farg3, xmm3); | |
115 REGISTER_DECLARATION(XMMRegister, j_farg4, xmm4); | |
116 REGISTER_DECLARATION(XMMRegister, j_farg5, xmm5); | |
117 REGISTER_DECLARATION(XMMRegister, j_farg6, xmm6); | |
118 REGISTER_DECLARATION(XMMRegister, j_farg7, xmm7); | |
119 | |
120 REGISTER_DECLARATION(Register, rscratch1, r10); // volatile | |
121 REGISTER_DECLARATION(Register, rscratch2, r11); // volatile | |
122 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
123 REGISTER_DECLARATION(Register, r12_heapbase, r12); // callee-saved |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
124 REGISTER_DECLARATION(Register, r15_thread, r15); // callee-saved |
0 | 125 |
126 #endif // _LP64 | |
127 | |
128 // Address is an abstraction used to represent a memory location | |
129 // using any of the amd64 addressing modes with one object. | |
130 // | |
131 // Note: A register location is represented via a Register, not | |
132 // via an address for efficiency & simplicity reasons. | |
133 | |
134 class ArrayAddress; | |
135 | |
136 class Address VALUE_OBJ_CLASS_SPEC { | |
137 public: | |
138 enum ScaleFactor { | |
139 no_scale = -1, | |
140 times_1 = 0, | |
141 times_2 = 1, | |
142 times_4 = 2, | |
143 times_8 = 3 | |
144 }; | |
145 | |
146 private: | |
147 Register _base; | |
148 Register _index; | |
149 ScaleFactor _scale; | |
150 int _disp; | |
151 RelocationHolder _rspec; | |
152 | |
153 // Easily misused constructors make them private | |
154 Address(int disp, address loc, relocInfo::relocType rtype); | |
155 Address(int disp, address loc, RelocationHolder spec); | |
156 | |
157 public: | |
158 // creation | |
159 Address() | |
160 : _base(noreg), | |
161 _index(noreg), | |
162 _scale(no_scale), | |
163 _disp(0) { | |
164 } | |
165 | |
166 // No default displacement otherwise Register can be implicitly | |
167 // converted to 0(Register) which is quite a different animal. | |
168 | |
169 Address(Register base, int disp) | |
170 : _base(base), | |
171 _index(noreg), | |
172 _scale(no_scale), | |
173 _disp(disp) { | |
174 } | |
175 | |
176 Address(Register base, Register index, ScaleFactor scale, int disp = 0) | |
177 : _base (base), | |
178 _index(index), | |
179 _scale(scale), | |
180 _disp (disp) { | |
181 assert(!index->is_valid() == (scale == Address::no_scale), | |
182 "inconsistent address"); | |
183 } | |
184 | |
185 // The following two overloads are used in connection with the | |
186 // ByteSize type (see sizes.hpp). They simplify the use of | |
187 // ByteSize'd arguments in assembly code. Note that their equivalent | |
188 // for the optimized build are the member functions with int disp | |
189 // argument since ByteSize is mapped to an int type in that case. | |
190 // | |
191 // Note: DO NOT introduce similar overloaded functions for WordSize | |
192 // arguments as in the optimized mode, both ByteSize and WordSize | |
193 // are mapped to the same type and thus the compiler cannot make a | |
194 // distinction anymore (=> compiler errors). | |
195 | |
196 #ifdef ASSERT | |
197 Address(Register base, ByteSize disp) | |
198 : _base(base), | |
199 _index(noreg), | |
200 _scale(no_scale), | |
201 _disp(in_bytes(disp)) { | |
202 } | |
203 | |
204 Address(Register base, Register index, ScaleFactor scale, ByteSize disp) | |
205 : _base(base), | |
206 _index(index), | |
207 _scale(scale), | |
208 _disp(in_bytes(disp)) { | |
209 assert(!index->is_valid() == (scale == Address::no_scale), | |
210 "inconsistent address"); | |
211 } | |
212 #endif // ASSERT | |
213 | |
214 // accessors | |
215 bool uses(Register reg) const { | |
216 return _base == reg || _index == reg; | |
217 } | |
218 | |
219 // Convert the raw encoding form into the form expected by the constructor for | |
220 // Address. An index of 4 (rsp) corresponds to having no index, so convert | |
221 // that to noreg for the Address constructor. | |
222 static Address make_raw(int base, int index, int scale, int disp); | |
223 | |
224 static Address make_array(ArrayAddress); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
225 Register base() const { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
226 return _base; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
227 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
228 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
229 Register index() const { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
230 return _index; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
231 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
232 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
233 int disp() const { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
234 return _disp; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
235 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
236 |
0 | 237 |
238 private: | |
239 bool base_needs_rex() const { | |
240 return _base != noreg && _base->encoding() >= 8; | |
241 } | |
242 | |
243 bool index_needs_rex() const { | |
244 return _index != noreg &&_index->encoding() >= 8; | |
245 } | |
246 | |
247 relocInfo::relocType reloc() const { return _rspec.type(); } | |
248 | |
249 friend class Assembler; | |
250 friend class MacroAssembler; | |
251 friend class LIR_Assembler; // base/index/scale/disp | |
252 }; | |
253 | |
254 // | |
255 // AddressLiteral has been split out from Address because operands of this type | |
256 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out | |
257 // the few instructions that need to deal with address literals are unique and the | |
258 // MacroAssembler does not have to implement every instruction in the Assembler | |
259 // in order to search for address literals that may need special handling depending | |
260 // on the instruction and the platform. As small step on the way to merging i486/amd64 | |
261 // directories. | |
262 // | |
263 class AddressLiteral VALUE_OBJ_CLASS_SPEC { | |
264 friend class ArrayAddress; | |
265 RelocationHolder _rspec; | |
266 // Typically we use AddressLiterals we want to use their rval | |
267 // However in some situations we want the lval (effect address) of the item. | |
268 // We provide a special factory for making those lvals. | |
269 bool _is_lval; | |
270 | |
271 // If the target is far we'll need to load the ea of this to | |
272 // a register to reach it. Otherwise if near we can do rip | |
273 // relative addressing. | |
274 | |
275 address _target; | |
276 | |
277 protected: | |
278 // creation | |
279 AddressLiteral() | |
280 : _is_lval(false), | |
281 _target(NULL) | |
282 {} | |
283 | |
284 public: | |
285 | |
286 | |
287 AddressLiteral(address target, relocInfo::relocType rtype); | |
288 | |
289 AddressLiteral(address target, RelocationHolder const& rspec) | |
290 : _rspec(rspec), | |
291 _is_lval(false), | |
292 _target(target) | |
293 {} | |
294 | |
295 AddressLiteral addr() { | |
296 AddressLiteral ret = *this; | |
297 ret._is_lval = true; | |
298 return ret; | |
299 } | |
300 | |
301 | |
302 private: | |
303 | |
304 address target() { return _target; } | |
305 bool is_lval() { return _is_lval; } | |
306 | |
307 relocInfo::relocType reloc() const { return _rspec.type(); } | |
308 const RelocationHolder& rspec() const { return _rspec; } | |
309 | |
310 friend class Assembler; | |
311 friend class MacroAssembler; | |
312 friend class Address; | |
313 friend class LIR_Assembler; | |
314 }; | |
315 | |
316 // Convience classes | |
317 class RuntimeAddress: public AddressLiteral { | |
318 | |
319 public: | |
320 | |
321 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {} | |
322 | |
323 }; | |
324 | |
325 class OopAddress: public AddressLiteral { | |
326 | |
327 public: | |
328 | |
329 OopAddress(address target) : AddressLiteral(target, relocInfo::oop_type){} | |
330 | |
331 }; | |
332 | |
333 class ExternalAddress: public AddressLiteral { | |
334 | |
335 public: | |
336 | |
337 ExternalAddress(address target) : AddressLiteral(target, relocInfo::external_word_type){} | |
338 | |
339 }; | |
340 | |
341 class InternalAddress: public AddressLiteral { | |
342 | |
343 public: | |
344 | |
345 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {} | |
346 | |
347 }; | |
348 | |
349 // x86 can do array addressing as a single operation since disp can be an absolute | |
350 // address but amd64 can't [e.g. array_base(rx, ry:width) ]. We create a class | |
351 // that expresses the concept but does extra magic on amd64 to get the final result | |
352 | |
353 class ArrayAddress VALUE_OBJ_CLASS_SPEC { | |
354 private: | |
355 | |
356 AddressLiteral _base; | |
357 Address _index; | |
358 | |
359 public: | |
360 | |
361 ArrayAddress() {}; | |
362 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {}; | |
363 AddressLiteral base() { return _base; } | |
364 Address index() { return _index; } | |
365 | |
366 }; | |
367 | |
368 // The amd64 Assembler: Pure assembler doing NO optimizations on | |
369 // the instruction level (e.g. mov rax, 0 is not translated into xor | |
370 // rax, rax!); i.e., what you write is what you get. The Assembler is | |
371 // generating code into a CodeBuffer. | |
372 | |
373 const int FPUStateSizeInWords = 512 / wordSize; | |
374 | |
375 class Assembler : public AbstractAssembler { | |
376 friend class AbstractAssembler; // for the non-virtual hack | |
377 friend class StubGenerator; | |
378 | |
379 | |
380 protected: | |
381 #ifdef ASSERT | |
382 void check_relocation(RelocationHolder const& rspec, int format); | |
383 #endif | |
384 | |
385 inline void emit_long64(jlong x); | |
386 | |
387 void emit_data(jint data, relocInfo::relocType rtype, int format /* = 1 */); | |
388 void emit_data(jint data, RelocationHolder const& rspec, int format /* = 1 */); | |
389 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); | |
390 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); | |
391 | |
392 // Helper functions for groups of instructions | |
393 void emit_arith_b(int op1, int op2, Register dst, int imm8); | |
394 | |
395 void emit_arith(int op1, int op2, Register dst, int imm32); | |
396 // only x86?? | |
397 void emit_arith(int op1, int op2, Register dst, jobject obj); | |
398 void emit_arith(int op1, int op2, Register dst, Register src); | |
399 | |
400 void emit_operand(Register reg, | |
401 Register base, Register index, Address::ScaleFactor scale, | |
402 int disp, | |
403 RelocationHolder const& rspec, | |
404 int rip_relative_correction = 0); | |
405 void emit_operand(Register reg, Address adr, | |
406 int rip_relative_correction = 0); | |
407 void emit_operand(XMMRegister reg, | |
408 Register base, Register index, Address::ScaleFactor scale, | |
409 int disp, | |
410 RelocationHolder const& rspec, | |
411 int rip_relative_correction = 0); | |
412 void emit_operand(XMMRegister reg, Address adr, | |
413 int rip_relative_correction = 0); | |
414 | |
415 // Immediate-to-memory forms | |
416 void emit_arith_operand(int op1, Register rm, Address adr, int imm32); | |
417 | |
418 void emit_farith(int b1, int b2, int i); | |
419 | |
420 bool reachable(AddressLiteral adr); | |
421 | |
422 // These are all easily abused and hence protected | |
423 | |
424 // Make these disappear in 64bit mode since they would never be correct | |
425 #ifndef _LP64 | |
426 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); | |
427 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); | |
428 | |
429 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); | |
430 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); | |
431 | |
432 void push_literal32(int32_t imm32, RelocationHolder const& rspec); | |
433 #endif // _LP64 | |
434 | |
435 | |
436 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); | |
437 | |
438 // These are unique in that we are ensured by the caller that the 32bit | |
439 // relative in these instructions will always be able to reach the potentially | |
440 // 64bit address described by entry. Since they can take a 64bit address they | |
441 // don't have the 32 suffix like the other instructions in this class. | |
442 void jmp_literal(address entry, RelocationHolder const& rspec); | |
443 void call_literal(address entry, RelocationHolder const& rspec); | |
444 | |
445 public: | |
446 enum Condition { // The amd64 condition codes used for conditional jumps/moves. | |
447 zero = 0x4, | |
448 notZero = 0x5, | |
449 equal = 0x4, | |
450 notEqual = 0x5, | |
451 less = 0xc, | |
452 lessEqual = 0xe, | |
453 greater = 0xf, | |
454 greaterEqual = 0xd, | |
455 below = 0x2, | |
456 belowEqual = 0x6, | |
457 above = 0x7, | |
458 aboveEqual = 0x3, | |
459 overflow = 0x0, | |
460 noOverflow = 0x1, | |
461 carrySet = 0x2, | |
462 carryClear = 0x3, | |
463 negative = 0x8, | |
464 positive = 0x9, | |
465 parity = 0xa, | |
466 noParity = 0xb | |
467 }; | |
468 | |
469 enum Prefix { | |
470 // segment overrides | |
471 // XXX remove segment prefixes | |
472 CS_segment = 0x2e, | |
473 SS_segment = 0x36, | |
474 DS_segment = 0x3e, | |
475 ES_segment = 0x26, | |
476 FS_segment = 0x64, | |
477 GS_segment = 0x65, | |
478 | |
479 REX = 0x40, | |
480 | |
481 REX_B = 0x41, | |
482 REX_X = 0x42, | |
483 REX_XB = 0x43, | |
484 REX_R = 0x44, | |
485 REX_RB = 0x45, | |
486 REX_RX = 0x46, | |
487 REX_RXB = 0x47, | |
488 | |
489 REX_W = 0x48, | |
490 | |
491 REX_WB = 0x49, | |
492 REX_WX = 0x4A, | |
493 REX_WXB = 0x4B, | |
494 REX_WR = 0x4C, | |
495 REX_WRB = 0x4D, | |
496 REX_WRX = 0x4E, | |
497 REX_WRXB = 0x4F | |
498 }; | |
499 | |
500 enum WhichOperand { | |
501 // input to locate_operand, and format code for relocations | |
502 imm64_operand = 0, // embedded 64-bit immediate operand | |
503 disp32_operand = 1, // embedded 32-bit displacement | |
504 call32_operand = 2, // embedded 32-bit self-relative displacement | |
505 _WhichOperand_limit = 3 | |
506 }; | |
507 | |
508 public: | |
509 | |
510 // Creation | |
511 Assembler(CodeBuffer* code) | |
512 : AbstractAssembler(code) { | |
513 } | |
514 | |
515 // Decoding | |
516 static address locate_operand(address inst, WhichOperand which); | |
517 static address locate_next_instruction(address inst); | |
518 | |
519 // Utilities | |
520 | |
521 static bool is_simm(int64_t x, int nbits) { return -( CONST64(1) << (nbits-1) ) <= x && x < ( CONST64(1) << (nbits-1) ); } | |
522 static bool is_simm32 (int64_t x) { return x == (int64_t)(int32_t)x; } | |
523 | |
524 | |
525 // Stack | |
526 void pushaq(); | |
527 void popaq(); | |
528 | |
529 void pushfq(); | |
530 void popfq(); | |
531 | |
532 void pushq(int imm32); | |
533 | |
534 void pushq(Register src); | |
535 void pushq(Address src); | |
536 | |
537 void popq(Register dst); | |
538 void popq(Address dst); | |
539 | |
540 // Instruction prefixes | |
541 void prefix(Prefix p); | |
542 | |
543 int prefix_and_encode(int reg_enc, bool byteinst = false); | |
544 int prefixq_and_encode(int reg_enc); | |
545 | |
546 int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false); | |
547 int prefixq_and_encode(int dst_enc, int src_enc); | |
548 | |
549 void prefix(Register reg); | |
550 void prefix(Address adr); | |
551 void prefixq(Address adr); | |
552 | |
553 void prefix(Address adr, Register reg, bool byteinst = false); | |
554 void prefixq(Address adr, Register reg); | |
555 | |
556 void prefix(Address adr, XMMRegister reg); | |
557 | |
558 // Moves | |
559 void movb(Register dst, Address src); | |
560 void movb(Address dst, int imm8); | |
561 void movb(Address dst, Register src); | |
562 | |
563 void movw(Address dst, int imm16); | |
564 void movw(Register dst, Address src); | |
565 void movw(Address dst, Register src); | |
566 | |
567 void movl(Register dst, int imm32); | |
568 void movl(Register dst, Register src); | |
569 void movl(Register dst, Address src); | |
570 void movl(Address dst, int imm32); | |
571 void movl(Address dst, Register src); | |
572 | |
573 void movq(Register dst, Register src); | |
574 void movq(Register dst, Address src); | |
575 void movq(Address dst, Register src); | |
576 // These prevent using movq from converting a zero (like NULL) into Register | |
577 // by giving the compiler two choices it can't resolve | |
578 void movq(Address dst, void* dummy); | |
579 void movq(Register dst, void* dummy); | |
580 | |
581 void mov64(Register dst, intptr_t imm64); | |
582 void mov64(Address dst, intptr_t imm64); | |
583 | |
584 void movsbl(Register dst, Address src); | |
585 void movsbl(Register dst, Register src); | |
586 void movswl(Register dst, Address src); | |
587 void movswl(Register dst, Register src); | |
588 void movslq(Register dst, Address src); | |
589 void movslq(Register dst, Register src); | |
590 | |
591 void movzbl(Register dst, Address src); | |
592 void movzbl(Register dst, Register src); | |
593 void movzwl(Register dst, Address src); | |
594 void movzwl(Register dst, Register src); | |
595 | |
596 protected: // Avoid using the next instructions directly. | |
597 // New cpus require use of movsd and movss to avoid partial register stall | |
598 // when loading from memory. But for old Opteron use movlpd instead of movsd. | |
599 // The selection is done in MacroAssembler::movdbl() and movflt(). | |
600 void movss(XMMRegister dst, XMMRegister src); | |
601 void movss(XMMRegister dst, Address src); | |
602 void movss(Address dst, XMMRegister src); | |
603 void movsd(XMMRegister dst, XMMRegister src); | |
604 void movsd(Address dst, XMMRegister src); | |
605 void movsd(XMMRegister dst, Address src); | |
606 void movlpd(XMMRegister dst, Address src); | |
607 // New cpus require use of movaps and movapd to avoid partial register stall | |
608 // when moving between registers. | |
609 void movapd(XMMRegister dst, XMMRegister src); | |
610 void movaps(XMMRegister dst, XMMRegister src); | |
611 public: | |
612 | |
613 void movdl(XMMRegister dst, Register src); | |
614 void movdl(Register dst, XMMRegister src); | |
615 void movdq(XMMRegister dst, Register src); | |
616 void movdq(Register dst, XMMRegister src); | |
617 | |
618 void cmovl(Condition cc, Register dst, Register src); | |
619 void cmovl(Condition cc, Register dst, Address src); | |
620 void cmovq(Condition cc, Register dst, Register src); | |
621 void cmovq(Condition cc, Register dst, Address src); | |
622 | |
623 // Prefetches | |
624 private: | |
625 void prefetch_prefix(Address src); | |
626 public: | |
627 void prefetcht0(Address src); | |
628 void prefetcht1(Address src); | |
629 void prefetcht2(Address src); | |
630 void prefetchnta(Address src); | |
631 void prefetchw(Address src); | |
632 | |
633 // Arithmetics | |
634 void adcl(Register dst, int imm32); | |
635 void adcl(Register dst, Address src); | |
636 void adcl(Register dst, Register src); | |
637 void adcq(Register dst, int imm32); | |
638 void adcq(Register dst, Address src); | |
639 void adcq(Register dst, Register src); | |
640 | |
641 void addl(Address dst, int imm32); | |
642 void addl(Address dst, Register src); | |
643 void addl(Register dst, int imm32); | |
644 void addl(Register dst, Address src); | |
645 void addl(Register dst, Register src); | |
646 void addq(Address dst, int imm32); | |
647 void addq(Address dst, Register src); | |
648 void addq(Register dst, int imm32); | |
649 void addq(Register dst, Address src); | |
650 void addq(Register dst, Register src); | |
651 | |
652 void andl(Register dst, int imm32); | |
653 void andl(Register dst, Address src); | |
654 void andl(Register dst, Register src); | |
655 void andq(Register dst, int imm32); | |
656 void andq(Register dst, Address src); | |
657 void andq(Register dst, Register src); | |
658 | |
659 void cmpb(Address dst, int imm8); | |
660 void cmpl(Address dst, int imm32); | |
661 void cmpl(Register dst, int imm32); | |
662 void cmpl(Register dst, Register src); | |
663 void cmpl(Register dst, Address src); | |
664 void cmpq(Address dst, int imm32); | |
665 void cmpq(Address dst, Register src); | |
666 void cmpq(Register dst, int imm32); | |
667 void cmpq(Register dst, Register src); | |
668 void cmpq(Register dst, Address src); | |
669 | |
670 void ucomiss(XMMRegister dst, XMMRegister src); | |
671 void ucomisd(XMMRegister dst, XMMRegister src); | |
672 | |
673 protected: | |
674 // Don't use next inc() and dec() methods directly. INC & DEC instructions | |
675 // could cause a partial flag stall since they don't set CF flag. | |
676 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods | |
677 // which call inc() & dec() or add() & sub() in accordance with | |
678 // the product flag UseIncDec value. | |
679 | |
680 void decl(Register dst); | |
681 void decl(Address dst); | |
682 void decq(Register dst); | |
683 void decq(Address dst); | |
684 | |
685 void incl(Register dst); | |
686 void incl(Address dst); | |
687 void incq(Register dst); | |
688 void incq(Address dst); | |
689 | |
690 public: | |
691 void idivl(Register src); | |
692 void idivq(Register src); | |
693 void cdql(); | |
694 void cdqq(); | |
695 | |
696 void imull(Register dst, Register src); | |
697 void imull(Register dst, Register src, int value); | |
698 void imulq(Register dst, Register src); | |
699 void imulq(Register dst, Register src, int value); | |
700 | |
701 void leal(Register dst, Address src); | |
702 void leaq(Register dst, Address src); | |
703 | |
704 void mull(Address src); | |
705 void mull(Register src); | |
706 | |
707 void negl(Register dst); | |
708 void negq(Register dst); | |
709 | |
710 void notl(Register dst); | |
711 void notq(Register dst); | |
712 | |
713 void orl(Address dst, int imm32); | |
714 void orl(Register dst, int imm32); | |
715 void orl(Register dst, Address src); | |
716 void orl(Register dst, Register src); | |
717 void orq(Address dst, int imm32); | |
718 void orq(Register dst, int imm32); | |
719 void orq(Register dst, Address src); | |
720 void orq(Register dst, Register src); | |
721 | |
722 void rcll(Register dst, int imm8); | |
723 void rclq(Register dst, int imm8); | |
724 | |
725 void sarl(Register dst, int imm8); | |
726 void sarl(Register dst); | |
727 void sarq(Register dst, int imm8); | |
728 void sarq(Register dst); | |
729 | |
730 void sbbl(Address dst, int imm32); | |
731 void sbbl(Register dst, int imm32); | |
732 void sbbl(Register dst, Address src); | |
733 void sbbl(Register dst, Register src); | |
734 void sbbq(Address dst, int imm32); | |
735 void sbbq(Register dst, int imm32); | |
736 void sbbq(Register dst, Address src); | |
737 void sbbq(Register dst, Register src); | |
738 | |
739 void shll(Register dst, int imm8); | |
740 void shll(Register dst); | |
741 void shlq(Register dst, int imm8); | |
742 void shlq(Register dst); | |
743 | |
744 void shrl(Register dst, int imm8); | |
745 void shrl(Register dst); | |
746 void shrq(Register dst, int imm8); | |
747 void shrq(Register dst); | |
748 | |
749 void subl(Address dst, int imm32); | |
750 void subl(Address dst, Register src); | |
751 void subl(Register dst, int imm32); | |
752 void subl(Register dst, Address src); | |
753 void subl(Register dst, Register src); | |
754 void subq(Address dst, int imm32); | |
755 void subq(Address dst, Register src); | |
756 void subq(Register dst, int imm32); | |
757 void subq(Register dst, Address src); | |
758 void subq(Register dst, Register src); | |
759 | |
760 void testb(Register dst, int imm8); | |
761 void testl(Register dst, int imm32); | |
762 void testl(Register dst, Register src); | |
763 void testq(Register dst, int imm32); | |
764 void testq(Register dst, Register src); | |
765 | |
766 void xaddl(Address dst, Register src); | |
767 void xaddq(Address dst, Register src); | |
768 | |
769 void xorl(Register dst, int imm32); | |
770 void xorl(Register dst, Address src); | |
771 void xorl(Register dst, Register src); | |
772 void xorq(Register dst, int imm32); | |
773 void xorq(Register dst, Address src); | |
774 void xorq(Register dst, Register src); | |
775 | |
776 // Miscellaneous | |
777 void bswapl(Register reg); | |
778 void bswapq(Register reg); | |
779 void lock(); | |
780 | |
781 void xchgl(Register reg, Address adr); | |
782 void xchgl(Register dst, Register src); | |
783 void xchgq(Register reg, Address adr); | |
784 void xchgq(Register dst, Register src); | |
785 | |
786 void cmpxchgl(Register reg, Address adr); | |
787 void cmpxchgq(Register reg, Address adr); | |
788 | |
789 void nop(int i = 1); | |
790 void addr_nop_4(); | |
791 void addr_nop_5(); | |
792 void addr_nop_7(); | |
793 void addr_nop_8(); | |
794 | |
795 void hlt(); | |
796 void ret(int imm16); | |
797 void smovl(); | |
798 void rep_movl(); | |
799 void rep_movq(); | |
800 void rep_set(); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
801 void repne_scanl(); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
802 void repne_scanq(); |
0 | 803 void setb(Condition cc, Register dst); |
804 | |
805 void clflush(Address adr); | |
806 | |
807 enum Membar_mask_bits { | |
808 StoreStore = 1 << 3, | |
809 LoadStore = 1 << 2, | |
810 StoreLoad = 1 << 1, | |
811 LoadLoad = 1 << 0 | |
812 }; | |
813 | |
814 // Serializes memory. | |
815 void membar(Membar_mask_bits order_constraint) { | |
816 // We only have to handle StoreLoad and LoadLoad | |
817 if (order_constraint & StoreLoad) { | |
818 // MFENCE subsumes LFENCE | |
819 mfence(); | |
820 } /* [jk] not needed currently: else if (order_constraint & LoadLoad) { | |
821 lfence(); | |
822 } */ | |
823 } | |
824 | |
825 void lfence() { | |
826 emit_byte(0x0F); | |
827 emit_byte(0xAE); | |
828 emit_byte(0xE8); | |
829 } | |
830 | |
831 void mfence() { | |
832 emit_byte(0x0F); | |
833 emit_byte(0xAE); | |
834 emit_byte(0xF0); | |
835 } | |
836 | |
837 // Identify processor type and features | |
838 void cpuid() { | |
839 emit_byte(0x0F); | |
840 emit_byte(0xA2); | |
841 } | |
842 | |
843 void cld() { emit_byte(0xfc); | |
844 } | |
845 | |
846 void std() { emit_byte(0xfd); | |
847 } | |
848 | |
849 | |
850 // Calls | |
851 | |
852 void call(Label& L, relocInfo::relocType rtype); | |
853 void call(Register reg); | |
854 void call(Address adr); | |
855 | |
856 // Jumps | |
857 | |
858 void jmp(Register reg); | |
859 void jmp(Address adr); | |
860 | |
861 // Label operations & relative jumps (PPUM Appendix D) | |
862 // unconditional jump to L | |
863 void jmp(Label& L, relocInfo::relocType rtype = relocInfo::none); | |
864 | |
865 | |
866 // Unconditional 8-bit offset jump to L. | |
867 // WARNING: be very careful using this for forward jumps. If the label is | |
868 // not bound within an 8-bit offset of this instruction, a run-time error | |
869 // will occur. | |
870 void jmpb(Label& L); | |
871 | |
872 // jcc is the generic conditional branch generator to run- time | |
873 // routines, jcc is used for branches to labels. jcc takes a branch | |
874 // opcode (cc) and a label (L) and generates either a backward | |
875 // branch or a forward branch and links it to the label fixup | |
876 // chain. Usage: | |
877 // | |
878 // Label L; // unbound label | |
879 // jcc(cc, L); // forward branch to unbound label | |
880 // bind(L); // bind label to the current pc | |
881 // jcc(cc, L); // backward branch to bound label | |
882 // bind(L); // illegal: a label may be bound only once | |
883 // | |
884 // Note: The same Label can be used for forward and backward branches | |
885 // but it may be bound only once. | |
886 | |
887 void jcc(Condition cc, Label& L, | |
888 relocInfo::relocType rtype = relocInfo::none); | |
889 | |
890 // Conditional jump to a 8-bit offset to L. | |
891 // WARNING: be very careful using this for forward jumps. If the label is | |
892 // not bound within an 8-bit offset of this instruction, a run-time error | |
893 // will occur. | |
894 void jccb(Condition cc, Label& L); | |
895 | |
896 // Floating-point operations | |
897 | |
898 void fxsave(Address dst); | |
899 void fxrstor(Address src); | |
900 void ldmxcsr(Address src); | |
901 void stmxcsr(Address dst); | |
902 | |
903 void addss(XMMRegister dst, XMMRegister src); | |
904 void addss(XMMRegister dst, Address src); | |
905 void subss(XMMRegister dst, XMMRegister src); | |
906 void subss(XMMRegister dst, Address src); | |
907 void mulss(XMMRegister dst, XMMRegister src); | |
908 void mulss(XMMRegister dst, Address src); | |
909 void divss(XMMRegister dst, XMMRegister src); | |
910 void divss(XMMRegister dst, Address src); | |
911 void addsd(XMMRegister dst, XMMRegister src); | |
912 void addsd(XMMRegister dst, Address src); | |
913 void subsd(XMMRegister dst, XMMRegister src); | |
914 void subsd(XMMRegister dst, Address src); | |
915 void mulsd(XMMRegister dst, XMMRegister src); | |
916 void mulsd(XMMRegister dst, Address src); | |
917 void divsd(XMMRegister dst, XMMRegister src); | |
918 void divsd(XMMRegister dst, Address src); | |
919 | |
920 // We only need the double form | |
921 void sqrtsd(XMMRegister dst, XMMRegister src); | |
922 void sqrtsd(XMMRegister dst, Address src); | |
923 | |
924 void xorps(XMMRegister dst, XMMRegister src); | |
925 void xorps(XMMRegister dst, Address src); | |
926 void xorpd(XMMRegister dst, XMMRegister src); | |
927 void xorpd(XMMRegister dst, Address src); | |
928 | |
929 void cvtsi2ssl(XMMRegister dst, Register src); | |
930 void cvtsi2ssq(XMMRegister dst, Register src); | |
931 void cvtsi2sdl(XMMRegister dst, Register src); | |
932 void cvtsi2sdq(XMMRegister dst, Register src); | |
933 void cvttss2sil(Register dst, XMMRegister src); // truncates | |
934 void cvttss2siq(Register dst, XMMRegister src); // truncates | |
935 void cvttsd2sil(Register dst, XMMRegister src); // truncates | |
936 void cvttsd2siq(Register dst, XMMRegister src); // truncates | |
937 void cvtss2sd(XMMRegister dst, XMMRegister src); | |
938 void cvtsd2ss(XMMRegister dst, XMMRegister src); | |
71 | 939 void cvtdq2pd(XMMRegister dst, XMMRegister src); |
940 void cvtdq2ps(XMMRegister dst, XMMRegister src); | |
0 | 941 |
942 void pxor(XMMRegister dst, Address src); // Xor Packed Byte Integer Values | |
943 void pxor(XMMRegister dst, XMMRegister src); // Xor Packed Byte Integer Values | |
944 | |
945 void movdqa(XMMRegister dst, Address src); // Move Aligned Double Quadword | |
946 void movdqa(XMMRegister dst, XMMRegister src); | |
947 void movdqa(Address dst, XMMRegister src); | |
948 | |
949 void movq(XMMRegister dst, Address src); | |
950 void movq(Address dst, XMMRegister src); | |
951 | |
952 void pshufd(XMMRegister dst, XMMRegister src, int mode); // Shuffle Packed Doublewords | |
953 void pshufd(XMMRegister dst, Address src, int mode); | |
954 void pshuflw(XMMRegister dst, XMMRegister src, int mode); // Shuffle Packed Low Words | |
955 void pshuflw(XMMRegister dst, Address src, int mode); | |
956 | |
957 void psrlq(XMMRegister dst, int shift); // Shift Right Logical Quadword Immediate | |
958 | |
959 void punpcklbw(XMMRegister dst, XMMRegister src); // Interleave Low Bytes | |
960 void punpcklbw(XMMRegister dst, Address src); | |
961 }; | |
962 | |
963 | |
964 // MacroAssembler extends Assembler by frequently used macros. | |
965 // | |
966 // Instructions for which a 'better' code sequence exists depending | |
967 // on arguments should also go in here. | |
968 | |
969 class MacroAssembler : public Assembler { | |
970 friend class LIR_Assembler; | |
971 protected: | |
972 | |
973 Address as_Address(AddressLiteral adr); | |
974 Address as_Address(ArrayAddress adr); | |
975 | |
976 // Support for VM calls | |
977 // | |
978 // This is the base routine called by the different versions of | |
979 // call_VM_leaf. The interpreter may customize this version by | |
980 // overriding it for its purposes (e.g., to save/restore additional | |
981 // registers when doing a VM call). | |
982 | |
983 virtual void call_VM_leaf_base( | |
984 address entry_point, // the entry point | |
985 int number_of_arguments // the number of arguments to | |
986 // pop after the call | |
987 ); | |
988 | |
989 // This is the base routine called by the different versions of | |
990 // call_VM. The interpreter may customize this version by overriding | |
991 // it for its purposes (e.g., to save/restore additional registers | |
992 // when doing a VM call). | |
993 // | |
994 // If no java_thread register is specified (noreg) than rdi will be | |
995 // used instead. call_VM_base returns the register which contains | |
996 // the thread upon return. If a thread register has been specified, | |
997 // the return value will correspond to that register. If no | |
998 // last_java_sp is specified (noreg) than rsp will be used instead. | |
999 virtual void call_VM_base( // returns the register | |
1000 // containing the thread upon | |
1001 // return | |
1002 Register oop_result, // where an oop-result ends up | |
1003 // if any; use noreg otherwise | |
1004 Register java_thread, // the thread if computed | |
1005 // before ; use noreg otherwise | |
1006 Register last_java_sp, // to set up last_Java_frame in | |
1007 // stubs; use noreg otherwise | |
1008 address entry_point, // the entry point | |
1009 int number_of_arguments, // the number of arguments (w/o | |
1010 // thread) to pop after the | |
1011 // call | |
1012 bool check_exceptions // whether to check for pending | |
1013 // exceptions after return | |
1014 ); | |
1015 | |
1016 // This routines should emit JVMTI PopFrame handling and ForceEarlyReturn code. | |
1017 // The implementation is only non-empty for the InterpreterMacroAssembler, | |
1018 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. | |
1019 virtual void check_and_handle_popframe(Register java_thread); | |
1020 virtual void check_and_handle_earlyret(Register java_thread); | |
1021 | |
1022 void call_VM_helper(Register oop_result, | |
1023 address entry_point, | |
1024 int number_of_arguments, | |
1025 bool check_exceptions = true); | |
1026 | |
1027 public: | |
1028 MacroAssembler(CodeBuffer* code) : Assembler(code) {} | |
1029 | |
1030 // Support for NULL-checks | |
1031 // | |
1032 // Generates code that causes a NULL OS exception if the content of | |
1033 // reg is NULL. If the accessed location is M[reg + offset] and the | |
1034 // offset is known, provide the offset. No explicit code generation | |
1035 // is needed if the offset is within a certain range (0 <= offset <= | |
1036 // page_size). | |
1037 void null_check(Register reg, int offset = -1); | |
1038 static bool needs_explicit_null_check(int offset); | |
1039 | |
1040 // Required platform-specific helpers for Label::patch_instructions. | |
1041 // They _shadow_ the declarations in AbstractAssembler, which are undefined. | |
1042 void pd_patch_instruction(address branch, address target); | |
1043 #ifndef PRODUCT | |
1044 static void pd_print_patched_instruction(address branch); | |
1045 #endif | |
1046 | |
1047 | |
1048 // The following 4 methods return the offset of the appropriate move | |
1049 // instruction. Note: these are 32 bit instructions | |
1050 | |
1051 // Support for fast byte/word loading with zero extension (depending | |
1052 // on particular CPU) | |
1053 int load_unsigned_byte(Register dst, Address src); | |
1054 int load_unsigned_word(Register dst, Address src); | |
1055 | |
1056 // Support for fast byte/word loading with sign extension (depending | |
1057 // on particular CPU) | |
1058 int load_signed_byte(Register dst, Address src); | |
1059 int load_signed_word(Register dst, Address src); | |
1060 | |
1061 // Support for inc/dec with optimal instruction selection depending | |
1062 // on value | |
1063 void incrementl(Register reg, int value = 1); | |
1064 void decrementl(Register reg, int value = 1); | |
1065 void incrementq(Register reg, int value = 1); | |
1066 void decrementq(Register reg, int value = 1); | |
1067 | |
1068 void incrementl(Address dst, int value = 1); | |
1069 void decrementl(Address dst, int value = 1); | |
1070 void incrementq(Address dst, int value = 1); | |
1071 void decrementq(Address dst, int value = 1); | |
1072 | |
1073 // Support optimal SSE move instructions. | |
1074 void movflt(XMMRegister dst, XMMRegister src) { | |
1075 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; } | |
1076 else { movss (dst, src); return; } | |
1077 } | |
1078 | |
1079 void movflt(XMMRegister dst, Address src) { movss(dst, src); } | |
1080 | |
1081 void movflt(XMMRegister dst, AddressLiteral src); | |
1082 | |
1083 void movflt(Address dst, XMMRegister src) { movss(dst, src); } | |
1084 | |
1085 void movdbl(XMMRegister dst, XMMRegister src) { | |
1086 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; } | |
1087 else { movsd (dst, src); return; } | |
1088 } | |
1089 | |
1090 void movdbl(XMMRegister dst, AddressLiteral src); | |
1091 | |
1092 void movdbl(XMMRegister dst, Address src) { | |
1093 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; } | |
1094 else { movlpd(dst, src); return; } | |
1095 } | |
1096 | |
1097 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); } | |
1098 | |
1099 void incrementl(AddressLiteral dst); | |
1100 void incrementl(ArrayAddress dst); | |
1101 | |
1102 // Alignment | |
1103 void align(int modulus); | |
1104 | |
1105 // Misc | |
1106 void fat_nop(); // 5 byte nop | |
1107 | |
1108 | |
1109 // C++ bool manipulation | |
1110 | |
1111 void movbool(Register dst, Address src); | |
1112 void movbool(Address dst, bool boolconst); | |
1113 void movbool(Address dst, Register src); | |
1114 void testbool(Register dst); | |
1115 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1116 // oop manipulations |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1117 void load_klass(Register dst, Register src); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1118 void store_klass(Register dst, Register src); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1119 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1120 void load_heap_oop(Register dst, Address src); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1121 void store_heap_oop(Address dst, Register src); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1122 void encode_heap_oop(Register r); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1123 void decode_heap_oop(Register r); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1124 void encode_heap_oop_not_null(Register r); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1125 void decode_heap_oop_not_null(Register r); |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1126 void encode_heap_oop_not_null(Register dst, Register src); |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
1127 void decode_heap_oop_not_null(Register dst, Register src); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1128 |
0 | 1129 // Stack frame creation/removal |
1130 void enter(); | |
1131 void leave(); | |
1132 | |
1133 // Support for getting the JavaThread pointer (i.e.; a reference to | |
1134 // thread-local information) The pointer will be loaded into the | |
1135 // thread register. | |
1136 void get_thread(Register thread); | |
1137 | |
1138 void int3(); | |
1139 | |
1140 // Support for VM calls | |
1141 // | |
1142 // It is imperative that all calls into the VM are handled via the | |
1143 // call_VM macros. They make sure that the stack linkage is setup | |
1144 // correctly. call_VM's correspond to ENTRY/ENTRY_X entry points | |
1145 // while call_VM_leaf's correspond to LEAF entry points. | |
1146 void call_VM(Register oop_result, | |
1147 address entry_point, | |
1148 bool check_exceptions = true); | |
1149 void call_VM(Register oop_result, | |
1150 address entry_point, | |
1151 Register arg_1, | |
1152 bool check_exceptions = true); | |
1153 void call_VM(Register oop_result, | |
1154 address entry_point, | |
1155 Register arg_1, Register arg_2, | |
1156 bool check_exceptions = true); | |
1157 void call_VM(Register oop_result, | |
1158 address entry_point, | |
1159 Register arg_1, Register arg_2, Register arg_3, | |
1160 bool check_exceptions = true); | |
1161 | |
1162 // Overloadings with last_Java_sp | |
1163 void call_VM(Register oop_result, | |
1164 Register last_java_sp, | |
1165 address entry_point, | |
1166 int number_of_arguments = 0, | |
1167 bool check_exceptions = true); | |
1168 void call_VM(Register oop_result, | |
1169 Register last_java_sp, | |
1170 address entry_point, | |
1171 Register arg_1, bool | |
1172 check_exceptions = true); | |
1173 void call_VM(Register oop_result, | |
1174 Register last_java_sp, | |
1175 address entry_point, | |
1176 Register arg_1, Register arg_2, | |
1177 bool check_exceptions = true); | |
1178 void call_VM(Register oop_result, | |
1179 Register last_java_sp, | |
1180 address entry_point, | |
1181 Register arg_1, Register arg_2, Register arg_3, | |
1182 bool check_exceptions = true); | |
1183 | |
1184 void call_VM_leaf(address entry_point, | |
1185 int number_of_arguments = 0); | |
1186 void call_VM_leaf(address entry_point, | |
1187 Register arg_1); | |
1188 void call_VM_leaf(address entry_point, | |
1189 Register arg_1, Register arg_2); | |
1190 void call_VM_leaf(address entry_point, | |
1191 Register arg_1, Register arg_2, Register arg_3); | |
1192 | |
1193 // last Java Frame (fills frame anchor) | |
1194 void set_last_Java_frame(Register last_java_sp, | |
1195 Register last_java_fp, | |
1196 address last_java_pc); | |
1197 void reset_last_Java_frame(bool clear_fp, bool clear_pc); | |
1198 | |
1199 // Stores | |
1200 void store_check(Register obj); // store check for | |
1201 // obj - register is | |
1202 // destroyed | |
1203 // afterwards | |
1204 void store_check(Register obj, Address dst); // same as above, dst | |
1205 // is exact store | |
1206 // location (reg. is | |
1207 // destroyed) | |
1208 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1209 void g1_write_barrier_pre(Register obj, Register tmp, Register tmp2, bool tosca_live ); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1210 void g1_write_barrier_post(Register store_addr, Register new_val, Register tmp, Register tmp2); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1211 |
0 | 1212 // split store_check(Register obj) to enhance instruction interleaving |
1213 void store_check_part_1(Register obj); | |
1214 void store_check_part_2(Register obj); | |
1215 | |
1216 // C 'boolean' to Java boolean: x == 0 ? 0 : 1 | |
1217 void c2bool(Register x); | |
1218 | |
1219 // Int division/reminder for Java | |
1220 // (as idivl, but checks for special case as described in JVM spec.) | |
1221 // returns idivl instruction offset for implicit exception handling | |
1222 int corrected_idivl(Register reg); | |
1223 // Long division/reminder for Java | |
1224 // (as idivq, but checks for special case as described in JVM spec.) | |
1225 // returns idivq instruction offset for implicit exception handling | |
1226 int corrected_idivq(Register reg); | |
1227 | |
1228 // Push and pop integer/fpu/cpu state | |
1229 void push_IU_state(); | |
1230 void pop_IU_state(); | |
1231 | |
1232 void push_FPU_state(); | |
1233 void pop_FPU_state(); | |
1234 | |
1235 void push_CPU_state(); | |
1236 void pop_CPU_state(); | |
1237 | |
1238 // Sign extension | |
1239 void sign_extend_short(Register reg); | |
1240 void sign_extend_byte(Register reg); | |
1241 | |
1242 // Division by power of 2, rounding towards 0 | |
1243 void division_with_shift(Register reg, int shift_value); | |
1244 | |
1245 // Round up to a power of two | |
1246 void round_to_l(Register reg, int modulus); | |
1247 void round_to_q(Register reg, int modulus); | |
1248 | |
1249 // allocation | |
1250 void eden_allocate( | |
1251 Register obj, // result: pointer to object after | |
1252 // successful allocation | |
1253 Register var_size_in_bytes, // object size in bytes if unknown at | |
1254 // compile time; invalid otherwise | |
1255 int con_size_in_bytes, // object size in bytes if known at | |
1256 // compile time | |
1257 Register t1, // temp register | |
1258 Label& slow_case // continuation point if fast | |
1259 // allocation fails | |
1260 ); | |
1261 void tlab_allocate( | |
1262 Register obj, // result: pointer to object after | |
1263 // successful allocation | |
1264 Register var_size_in_bytes, // object size in bytes if unknown at | |
1265 // compile time; invalid otherwise | |
1266 int con_size_in_bytes, // object size in bytes if known at | |
1267 // compile time | |
1268 Register t1, // temp register | |
1269 Register t2, // temp register | |
1270 Label& slow_case // continuation point if fast | |
1271 // allocation fails | |
1272 ); | |
1273 void tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); | |
1274 | |
1275 //---- | |
1276 | |
1277 // Debugging | |
1278 | |
1279 // only if +VerifyOops | |
1280 void verify_oop(Register reg, const char* s = "broken oop"); | |
1281 void verify_oop_addr(Address addr, const char * s = "broken oop addr"); | |
1282 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1283 // if heap base register is used - reinit it with the correct value |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1284 void reinit_heapbase(); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
71
diff
changeset
|
1285 |
0 | 1286 // only if +VerifyFPU |
1287 void verify_FPU(int stack_depth, const char* s = "illegal FPU state") {} | |
1288 | |
1289 // prints msg, dumps registers and stops execution | |
1290 void stop(const char* msg); | |
1291 | |
1292 // prints message and continues | |
1293 void warn(const char* msg); | |
1294 | |
1295 static void debug(char* msg, int64_t pc, int64_t regs[]); | |
1296 | |
1297 void os_breakpoint(); | |
1298 | |
1299 void untested() | |
1300 { | |
1301 stop("untested"); | |
1302 } | |
1303 | |
1304 void unimplemented(const char* what = "") | |
1305 { | |
1306 char* b = new char[1024]; | |
1307 sprintf(b, "unimplemented: %s", what); | |
1308 stop(b); | |
1309 } | |
1310 | |
1311 void should_not_reach_here() | |
1312 { | |
1313 stop("should not reach here"); | |
1314 } | |
1315 | |
1316 // Stack overflow checking | |
1317 void bang_stack_with_offset(int offset) | |
1318 { | |
1319 // stack grows down, caller passes positive offset | |
1320 assert(offset > 0, "must bang with negative offset"); | |
1321 movl(Address(rsp, (-offset)), rax); | |
1322 } | |
1323 | |
1324 // Writes to stack successive pages until offset reached to check for | |
1325 // stack overflow + shadow pages. Also, clobbers tmp | |
1326 void bang_stack_size(Register offset, Register tmp); | |
1327 | |
1328 // Support for serializing memory accesses between threads. | |
1329 void serialize_memory(Register thread, Register tmp); | |
1330 | |
1331 void verify_tlab(); | |
1332 | |
1333 // Biased locking support | |
1334 // lock_reg and obj_reg must be loaded up with the appropriate values. | |
1335 // swap_reg must be rax and is killed. | |
1336 // tmp_reg must be supplied and is killed. | |
1337 // If swap_reg_contains_mark is true then the code assumes that the | |
1338 // mark word of the object has already been loaded into swap_reg. | |
1339 // Optional slow case is for implementations (interpreter and C1) which branch to | |
1340 // slow case directly. Leaves condition codes set for C2's Fast_Lock node. | |
1341 // Returns offset of first potentially-faulting instruction for null | |
1342 // check info (currently consumed only by C1). If | |
1343 // swap_reg_contains_mark is true then returns -1 as it is assumed | |
1344 // the calling code has already passed any potential faults. | |
1345 int biased_locking_enter(Register lock_reg, Register obj_reg, Register swap_reg, Register tmp_reg, | |
1346 bool swap_reg_contains_mark, | |
1347 Label& done, Label* slow_case = NULL, | |
1348 BiasedLockingCounters* counters = NULL); | |
1349 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done); | |
1350 | |
1351 Condition negate_condition(Condition cond); | |
1352 | |
1353 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit | |
1354 // operands. In general the names are modified to avoid hiding the instruction in Assembler | |
1355 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers | |
1356 // here in MacroAssembler. The major exception to this rule is call | |
1357 | |
1358 // Arithmetics | |
1359 | |
1360 void cmp8(AddressLiteral src1, int8_t imm32); | |
1361 | |
1362 void cmp32(AddressLiteral src1, int32_t src2); | |
1363 // compare reg - mem, or reg - &mem | |
1364 void cmp32(Register src1, AddressLiteral src2); | |
1365 | |
1366 void cmp32(Register src1, Address src2); | |
1367 | |
1368 #ifndef _LP64 | |
1369 void cmpoop(Address dst, jobject obj); | |
1370 void cmpoop(Register dst, jobject obj); | |
1371 #endif // _LP64 | |
1372 | |
1373 // NOTE src2 must be the lval. This is NOT an mem-mem compare | |
1374 void cmpptr(Address src1, AddressLiteral src2); | |
1375 | |
1376 void cmpptr(Register src1, AddressLiteral src); | |
1377 | |
1378 // will be cmpreg(?) | |
1379 void cmp64(Register src1, AddressLiteral src); | |
1380 | |
1381 void cmpxchgptr(Register reg, Address adr); | |
1382 void cmpxchgptr(Register reg, AddressLiteral adr); | |
1383 | |
1384 // Helper functions for statistics gathering. | |
1385 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes. | |
1386 void cond_inc32(Condition cond, AddressLiteral counter_addr); | |
1387 // Unconditional atomic increment. | |
1388 void atomic_incl(AddressLiteral counter_addr); | |
1389 | |
1390 | |
1391 void lea(Register dst, AddressLiteral src); | |
1392 void lea(Register dst, Address src); | |
1393 | |
1394 | |
1395 // Calls | |
1396 void call(Label& L, relocInfo::relocType rtype); | |
1397 void call(Register entry); | |
1398 void call(AddressLiteral entry); | |
1399 | |
1400 // Jumps | |
1401 | |
1402 // 32bit can do a case table jump in one instruction but we no longer allow the base | |
1403 // to be installed in the Address class | |
1404 void jump(ArrayAddress entry); | |
1405 | |
1406 void jump(AddressLiteral entry); | |
1407 void jump_cc(Condition cc, AddressLiteral dst); | |
1408 | |
1409 // Floating | |
1410 | |
1411 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); } | |
1412 void ldmxcsr(AddressLiteral src); | |
1413 | |
1414 private: | |
1415 // these are private because users should be doing movflt/movdbl | |
1416 | |
1417 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); } | |
1418 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); } | |
1419 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); } | |
1420 void movss(XMMRegister dst, AddressLiteral src); | |
1421 | |
1422 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); } | |
1423 void movlpd(XMMRegister dst, AddressLiteral src); | |
1424 | |
1425 public: | |
1426 | |
1427 | |
1428 void xorpd(XMMRegister dst, XMMRegister src) {Assembler::xorpd(dst, src); } | |
1429 void xorpd(XMMRegister dst, Address src) {Assembler::xorpd(dst, src); } | |
1430 void xorpd(XMMRegister dst, AddressLiteral src); | |
1431 | |
1432 void xorps(XMMRegister dst, XMMRegister src) {Assembler::xorps(dst, src); } | |
1433 void xorps(XMMRegister dst, Address src) {Assembler::xorps(dst, src); } | |
1434 void xorps(XMMRegister dst, AddressLiteral src); | |
1435 | |
1436 | |
1437 // Data | |
1438 | |
1439 void movoop(Register dst, jobject obj); | |
1440 void movoop(Address dst, jobject obj); | |
1441 | |
1442 void movptr(ArrayAddress dst, Register src); | |
1443 void movptr(Register dst, AddressLiteral src); | |
1444 | |
1445 void movptr(Register dst, intptr_t src); | |
1446 void movptr(Address dst, intptr_t src); | |
1447 | |
1448 void movptr(Register dst, ArrayAddress src); | |
1449 | |
1450 // to avoid hiding movl | |
1451 void mov32(AddressLiteral dst, Register src); | |
1452 void mov32(Register dst, AddressLiteral src); | |
1453 | |
1454 void pushoop(jobject obj); | |
1455 | |
1456 // Can push value or effective address | |
1457 void pushptr(AddressLiteral src); | |
1458 | |
1459 }; | |
1460 | |
1461 /** | |
1462 * class SkipIfEqual: | |
1463 * | |
1464 * Instantiating this class will result in assembly code being output that will | |
1465 * jump around any code emitted between the creation of the instance and it's | |
1466 * automatic destruction at the end of a scope block, depending on the value of | |
1467 * the flag passed to the constructor, which will be checked at run-time. | |
1468 */ | |
1469 class SkipIfEqual { | |
1470 private: | |
1471 MacroAssembler* _masm; | |
1472 Label _label; | |
1473 | |
1474 public: | |
1475 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value); | |
1476 ~SkipIfEqual(); | |
1477 }; | |
1478 | |
1479 | |
1480 #ifdef ASSERT | |
1481 inline bool AbstractAssembler::pd_check_instruction_mark() { return true; } | |
1482 #endif |